# THE MOLECULAR MECHANISM BEHIND SYNAPTIC TRANSMISSION

EDITED BY : Jiajie Diao and Cong Ma PUBLISHED IN : Frontiers in Molecular Neuroscience

#### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88963-692-1 DOI 10.3389/978-2-88963-692-1

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

#### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# THE MOLECULAR MECHANISM BEHIND SYNAPTIC TRANSMISSION

Topic Editors: Jiajie Diao, University of Cincinnati, United States Cong Ma, Huazhong University of Science and Technology, China

Citation: Diao, J., Ma, C., eds. (2020). The Molecular Mechanism Behind Synaptic Transmission. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-692-1

# Table of Contents

*05 High Transmembrane Voltage Raised by Close Contact Initiates Fusion Pore*

Bing Bu, Zhiqi Tian, Dechang Li and Baohua Ji

*15 Drive the Car(go)s—New Modalities to Control Cargo Trafficking in Live Cells*

Payel Mondal, John S. Khamo, Vishnu V. Krishnamurthy, Qi Cai and Kai Zhang

*25 The Structure of the Synaptic Vesicle-Plasma Membrane Interface Constrains SNARE Models of Rapid, Synchronous Exocytosis at Nerve Terminals*

Cameron B. Gundersen

*32 Melatonin Mediates Protective Effects Against Kainic Acid-Induced Neuronal Death Through Safeguarding ER Stress and Mitochondrial Disturbance*

Feixiao Xue, Cai Shi, Qingjie Chen, Weijian Hang, Liangtao Xia, Yue Wu, Sophia Z. Tao, Jie Zhou, Anbing Shi and Juan Chen


Zhenli Xie, Jiangang Long, Jiankang Liu, Zuying Chai, Xinjiang Kang and Changhe Wang


Volker Kiessling, Binyong Liang, Alex J. B. Kreutzberger and Lukas K. Tamm


Paul Heo, Joon-Bum Park, Yeon-Kyun Shin and Dae-Hyuk Kweon


David Snead, Alex L. Lai, Rachel T. Wragg, Daniel A. Parisotto, Trudy F. Ramlall, Jeremy S. Dittman, Jack H. Freed and David Eliezer *137 Evolutionary Divergence of the C-terminal Domain of Complexin Accounts for Functional Disparities Between Vertebrate and Invertebrate Complexins*

Rachel T. Wragg, Daniel A. Parisotto, Zhenlong Li, Mayu S. Terakawa, David Snead, Ishani Basu, Harel Weinstein, David Eliezer and Jeremy S. Dittman

*161 Membrane Fusion Involved in Neurotransmission: Glimpse From Electron Microscope and Molecular Simulation*

Zhiwei Yang, Lu Gou, Shuyu Chen, Na Li, Shengli Zhang and Lei Zhang


Zhenyong Wu, Sathish Thiyagarajan, Ben O'Shaughnessy and Erdem Karatekin

*191 Productive and Non-productive Pathways for Synaptotagmin 1 to Support Ca2+-Triggered Fast Exocytosis*

Jaewook Kim and Yeon-Kyun Shin

*202 Accessory and Central* a*-helices of Complexin Selectively Activate Ca2+ Triggering of Synaptic Exocytosis*

Yi Yu, Su Chen, Xiaoqiang Mo, Jihong Gong, Chenhong Li and Xiaofei Yang


Michael Telias

*Domains*

# High Transmembrane Voltage Raised by Close Contact Initiates Fusion Pore

#### Bing Bu1 †, Zhiqi Tian2 †, Dechang Li <sup>1</sup> \* and Baohua Ji <sup>1</sup> \*

*<sup>1</sup> Biomechanics and Biomaterials Laboratory, Department of Applied Mechanics, Beijing Institute of Technology, Beijing, China, <sup>2</sup> Key Laboratory of Biomedical Information Engineering of Ministry of Education, Center for Mitochondrial Biology and Medicine, School of Life Science and Technology, Xi'an Jiaotong University, Xi'an, China*

Membrane fusion lies at the heart of neuronal communication but the detailed mechanism of a critical step, fusion pore initiation, remains poorly understood. Here, through atomistic molecular dynamics simulations, a transient pore formation induced by a close contact of two apposed bilayers is firstly reported. Such a close contact gives rise to a high local transmembrane voltage that induces the transient pore formation. Through simulations on two apposed bilayers fixed at a series of given distances, the process in which two bilayers approaching to each other under the pulling force from fusion proteins for membrane fusion was mimicked. Of note, this close contact induced fusion pore formation is contrasted with previous reported electroporation under *ad hoc* applied external electric field or ionic charge in-balance. We show that the transmembrane voltage increases with the decrease of the distance between the bilayers. Below a critical distance, depending on the lipid composition, the local transmembrane voltage can be sufficiently high to induce the transient pores. The size of these pores is approximately 1∼2 nm in diameter, which is large enough to allow passing of neurotransmitters. A resealing of the membrane pores resulting from the neutralization of the transmembrane voltage by ions through the pores was then observed. We also found that the membrane tension can either prolong the lifetime of transient pores or cause them to dilate for full collapse. This result provides a possible mechanism for fusion pore formation and regulation of pathway of fusion process.

#### Edited by:

*Cong Ma, Huazhong University of Science and Technology, China*

#### Reviewed by:

*Jizhong Lou, Institute of Biophysics (CAS), China Fangfu Ye, Institute of Physics (CAS) China*

#### \*Correspondence:

*Dechang Li dcli@bit.edu.cn Baohua Ji bhji@bit.edu.cn*

*† These authors have contributed equally to this work.*

Received: *16 September 2016* Accepted: *22 November 2016* Published: *09 December 2016*

#### Citation:

*Bu B, Tian Z, Li D and Ji B (2016) High Transmembrane Voltage Raised by Close Contact Initiates Fusion Pore. Front. Mol. Neurosci. 9:136. doi: 10.3389/fnmol.2016.00136* Keywords: membrane fusion, transmembrane voltage, fusion pore formation, exocytosis, mechanical force

### INTRODUCTION

Membrane fusion, as an important and ubiquitous cellular process, lies at the heart of neuronal communication wherein neurotransmitters are quickly released from synaptic vesicles following fusion with the presynaptic membrane. During this process, an initial fusion pore is assumed to be created between the two apposed membranes, and this tiny pore may either expand to a larger one or close back, which leads to two kinds of fusion mode (Alabi and Tsien, 2013). One type of fusion mode called full fusion (FF) requires fusion pore expand to the point where the vesicle membrane flattens into the plasma membrane surface, leading to complete luminal content release(Chernomordik and Kozlov, 2003; Jackson and Chapman, 2008); the other type of fusion mode proceeds without pore expansion, where vesicles transiently fuse at the plasma membrane to release a part of their neurotransmitters (Richards, 2009) without full collapse into the plasma membrane known as "kiss-and-run" fusion (KR) (He and Wu, 2007; Zhang et al., 2009). Although, the experimental results on membrane fusion at synapses are mounting (He and Wu, 2007), it is difficult to directly image the fusion process at sufficient resolution, thus the molecular mechanism of the fusion initiation is still unclear.

In addition, it is also unclear how the fusion process is regulated in vivo. For instance, there is a long standing debate over FF and KR fusion for the regulating mechanisms behind the synaptic release (Marx, 2014). One of the important issues in the debate is an incomplete understanding of the structural mechanics of the membrane under the force of fusogenic proteins (Jahn et al., 2003; Rizo and Rosenmund, 2008), and how these forces act on the membrane during the fusion process. It is known that when a small membrane pore is created (Chernomordik and Kozlov, 2008), it might either close or expand (Chanturiya et al., 1997; Wong et al., 2007; Diao et al., 2012). The reversible fusion pore modality, corresponding to KR fusion, enables a rapid and economical vesicle recycling compared to FF that may require ancillary proteins to retrieve fused vesicles (He and Wu, 2007). However, how the fusion pore is regulated for the fast synaptic transmission and the exact molecular picture of the pore evolution remains obscure.

Transient pore formation has been observed in many in vivo and in vitro systems, such as liposomes from yeast vacuoles (Starai et al., 2007; Zucchi and Zick, 2011), mating yeast pairs (Aguilar et al., 2007), cell pairs mediated by influenza fusase hemagglutinin (HA) (Blumenthal and Morris, 1999; Frolov et al., 2003), and proteoliposomes reconstituted with neuronal SNAREs (SNAP [Soluble NSF attachment protein] Receptors) (Dennison et al., 2006; Lai et al., 2013; Gong et al., 2015). And there are also many proposed mechanisms for the opening of the membrane pores through in silico simulations, such as transmembrane ionic charge imbalance, external electric and mechanical forces (Müller et al., 2003; Gurtovenko et al., 2010). However, the questions of how transient pores are produced in physiological conditions, and how they are regulated remain enigmatic. Here, through full atom molecular dynamics (MD) simulations, we discovered a fast and transient membrane pore formation when the distance of two apposed bilayers was set below a threshold value. The average size of these pores was of 1∼2 nm in diameter, which is large enough for passing neurotransmitters. The initial pore could either shrink or expand depending on the magnitude of membrane tension. The high local transmembrane voltage caused by the close contact of apposed membranes was responsible for the formation of these fast and transient membrane pores. This study might thus shed promising light on the molecular mechanisms of pore formation and evolution.

### METHODS

#### MD Simulation

To simulate the pore formation at the contact of a vesicle with the synaptic membrane, we created two apposed lipid bilayers set at various distances (see **Figure 1**). The lipid bilayers were generated by Membrane Builder online service (Jo et al., 2008, 2009). The simulation system was set up following a similar procedure in our previous studies (Li et al., 2010, 2011, 2015; Lai et al., 2015; Xu et al., 2015). The system was solvated in a ∼23 × 23 × 20 nm<sup>3</sup> TIP3P (Jorgensen et al., 1983) water box, with ∼210,000 water molecules and potassium ions added to neutralize the system. All the simulations were performed using GROMACS package (Van Der Spoel et al., 2005) with CHARMM36 force field (Klauda et al., 2010). Periodic boundary condition was applied and the temperature was coupled to 300 K with V-rescale algorithm (Bussi et al., 2007). The pressure was coupled to 1 bar with Parrinello-Rahman approach (Parrinello and Rahman, 1981). The LINCS algorithm (Hess, 2008) was applied to constrain the covalent bonds with H-atoms. The time step of the simulations is 2.0 fs. The cut-off of the nonbonded interactions was set to 10 Å. The particle mesh Ewald (PME) (Essmann et al., 1995) method was used to calculate the long-range electrostatic interactions. The non-bonded pairs were updated in every 10 steps. Each MD simulation was performed independently for three replicates. The simulation system was validated by its reproducing the stalk-like structure in experiments under dehydration condition (Yang and Huang, 2002). All graphics and visualization analysis were processed using the VMD program (Humphrey et al., 1996).

Here the lipid composition was adopted as Chol:DOPC: POPE:POPS with different ratio of percentage as (20:45:20: 15%), (20:48:20:12%), (20:51:20:9%), (20:54:20:6%), and (20: 55.5:20:4.5%). Chol, DOPC, POPE, POPS are abbreviation of cholesterol, 1,2-dioleoyl-sn-glycero-3-phosphocholine, 1 palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine, and 1 palmitoyl-2-oleoyl-sn-glycero-3-phenylserine, respectively.

In the molecular dynamics simulations, we carried out simulations on two bilayers set at various membrane distances, where the distance D was chosen by reducing its value from 6 nm, in 0.5 nm steps (see Figure S1), for mimicking the approaching process of two apposed bilayer under the pulling force of fusion protein for membrane fusion. Each simulation was done at a given constant membrane distance.

The ions were uniformly distributed in the three cytosol/fluid regions, i.e., the vesicle cytosol, cellular cytosol between the vesicle and presynaptic membranes, and the interstitial fluid around the synapse, as shown in Figure S1. The ionic concentration (potassium) was about 0.04∼0.15 mol/L, depending on the lipid composition and their percentage. The density of water was 1 g/cm<sup>3</sup> . Note that for specific lipid composition, the density of water and ions was kept constant. In the simulation, we chose potassium ions as the only ions in the system without chloride ions. For double check, we also did simulations with both potassium and chloride ions. In the presence of chloride ions, we put more potassium ions to neutralize the system. Our results showed that they had similar results as those in the absence of chloride ions.

For simulating the pore evolution under membrane tension, we built membrane tension Ŵ by applying pressure on the simulation box, given by Ŵ = L<sup>Z</sup> (P<sup>Z</sup> − PLat), where L<sup>Z</sup> is the length of the simulation box in the Z direction, P<sup>Z</sup> is the pressure along the Z direction, and PLat is the pressure along lateral direction in the membrane plane (Leontiadou et al., 2004).

#### Calculation of the Transmembrane Voltage

the sake of clarity, water molecules are not explicitly shown outside the simulation box.

To calculate the transmembrane voltage in MD simulations, we adopted the procedure proposed by Tieleman by solving a one dimensional Poisson's equation (Tieleman, 2004). Firstly, the charge density along the Z direction ρ (Z) was calculated by averaging the net charges over the membrane plane. The electric potential can be obtained by solving the one dimensional Poisson's equation, i.e., 8 (Z) = − 1 ε R Z 0 Z ′ R 0 ρ Z ′′ dZ′′dZ′ + 8(0), where 8(0) was chosen as zero at the axis of symmetry of the system at Z = 0. The difference of the electric potential through the membrane gives the transmembrane voltage.

#### Theoretical Model of Electric Potential

In our model, we used five layers to consider the five regions at the contact between the vesicle and presynaptic membrane: the cytosol in vesicle, R1, the vesicle membrane Rm1, the cellular cytosol between vesicle and presynaptic membrane, R2, the presynaptic membrane Rm2, and the interstitial fluid around synapse, R<sup>3</sup> (see **Figure 2**). Since we considered the local contact region between vesicle and presynapse, we assumed that the distribution of charges was uniform in each layer. For a neutralized system, the charge density should satisfy

$$
\rho\_1 H\_L + \rho\_3 H\_L + \rho\_2 D + \rho\_{m1} H\_m + \rho\_{m2} H\_m = 0 \tag{1}
$$

where ρ1, ρ2, ρ3, ρm1, and ρm<sup>2</sup> are charge density in layers R1, R2, R3, Rm1, and Rm2, respectively. H<sup>L</sup> is the thickness of layers R<sup>1</sup> and R<sup>3</sup> (here we adopted that the thickness of the two domains is identical), H<sup>m</sup> is the thickness of membrane, and D is the distance between the two apposed membranes. In this situation, the electric potential satisfies the one dimensional Poisson-Boltzmann equation,

$$\frac{d^2\Phi(Z)}{dZ^2} = -\frac{\rho(Z)}{\varepsilon} \tag{2}$$

where 8(Z) is the electric potential along the Z direction, ρ(Z) is the charge density and ε the dielectric constant. We set ε = 3 according to experimental measurement by Gramse et al. (2013). And we obtained the electric potential as

$$\Phi(Z) = -\frac{1}{\varepsilon} \int \limits\_{0}^{Z} \int \rho \left( Z'' \right) dZ'' dZ' \tag{3}$$

The result of the integral in Equation (3) can be found in Supplementary Material. Thus the transmembrane voltage of the vesicle membrane can be calculated as

$$
\Delta V\_1 = -\frac{1}{\varepsilon} (\frac{1}{2}\rho\_{m1}H\_m + \rho\_{m2}H\_m + \rho\_2D + \rho\_3H\_L)H\_m \tag{4}
$$

and the corresponding voltage of the presynaptic membrane is

$$
\Delta V\_2 = -\frac{1}{\varepsilon} (\frac{1}{2}\rho\_{m2}H\_m + \rho\_{m1}H\_m + \rho\_2D + \rho\_1H\_L)H\_m \tag{5}
$$

As shown in Equations (4 and 5), because ρm<sup>1</sup> and ρm<sup>2</sup> are negative, 1V<sup>1</sup> and 1V<sup>2</sup> linearly increase with the reduction of the inter-membrane distance D. If we assume ρ<sup>1</sup> = ρ3, and ρm<sup>1</sup> = ρm<sup>2</sup> = ρm, we have

$$
\Delta V = \Delta V\_1 = \Delta V\_2 = -\frac{1}{2\varepsilon} \left(\rho\_m H\_m + \rho\_2 D\right) H\_m \tag{6}
$$

### RESULTS

#### Close Contact Induces Transient Pore Formation

To understand the very early stage of membrane fusion when the vesicle and presynaptic membranes get close, we simulated two apposed membranes fixed at a series of given distances using MD simulations, mimicking the docking process pulled by SNAREs or other accessory proteins, as shown in **Figure 1**. By using a smaller distance between membranes below a critical value (discussed below), we found the formation of tiny pores on the membranes (**Figure 3**). When the two membranes were within ∼5 nm, a water line first formed across the thickness of membrane in several nanoseconds, followed by the formation of a transient pore, like a water channel, spanning the membranes (**Figure 3B**). Then the transient pore grew rapidly, causing considerable redistribution of lipid headgroups close to the channel. At the same time, ions can transport through these pore, as shown in **Figure 3B**. The duration of the ion transportation was very short (approximately 10 ns). Once the ion transportation was finished, the pores healed automatically as the results of balancing charge difference (**Figure 3C**). More details of the pore formation, ions leakage and pore healing can be found in Figure S2 and Movie S1 in Supplementary Material. To examine the physiological relevance of our simulations, the membrane distance was further reduced to mimic a higher degree of dehydration, which induced a stalk-like structure for hemifusion. This result is consistent with previous experiments of Yang and Huang (Yang and Huang, 2002) (see Figure S3) and validates our MD simulations.

To study the molecular mechanisms of the pore formation, we calculated the electric potential in our system. Our results showed that the distribution of electric potential is highly dependent on inter-membrane distance. As the distance between the two membranes decreased, the difference of electric potential across the lipid bilayer, i.e., the transmembrane voltage, rose significantly (**Figure 3D**). This transmembrane voltage formed the driving force for the pore formation, consistent with earlier reports that the transmembrane voltage caused by ionic charge imbalance between the two sides of the membrane (Gurtovenko and Vattulainen, 2005), or by external electric field (Tieleman, 2004; Sun et al., 2011), induces fusion pore and ion trafficking. Meanwhile, in previous studies, the transmembrane voltage was ad-hoc applied on the membrane (Tieleman, 2004; Gurtovenko and Vattulainen, 2005; Sun et al., 2011). Here we found that without the above ad hoc conditions, reducing the membrane distance alone was sufficient to cause a dramatic increase in transmembrane voltage capable of producing the driving force necessary for the fusion pore formation. Our simulation showed that the transmembrane voltage for pore formation produced by the close contact is in the range of 0.7∼1.2 V, consistent with the value of 0.5∼1.5 V used in electroporation experiments (Weaver and Chizmadzhev, 1996). The mean size of the pores was round 1–2 nm (see **Figures 3G–I**), which is large enough for the passing of small neurotransmitters (Zhang et al., 2009).

vesicle and the presynaptic membranes, and *Rm*1 represents interstitial fluid around synapse. *Rm*1 and *Rm*2 represents the vesicle membranes and presynaptic membrane, respectively. The charge density in domains *R*1 and *R*3 are denoted by ρ1 and ρ3, respectively; that in domain *R*2 by ρ2, and that in the domains *Rm*1 and *Rm*2 by ρ*m*1 and ρ*m*2, respectively.

#### Transmembrane Voltage vs. the Membrane Distance

To understand the mechanism of transmembrane voltage rising with the decrease of membrane distance, we employed a physical model to derive the field of electric potential in the apposed membranes as a function of the membrane distance (**Figure 2**). For a given ion distribution corresponding our MD simulations, we obtained the analytical solution of the field of electric potential and in particular, a linear relationship between the transmembrane voltage and the membrane distance for a close contact (see Methods). We found that when the membrane distance was less than ∼5 nm, the profile of the electric potential in each membrane became severely asymmetric with respect to the central line of the lipid bilayer thus producing a high transmembrane voltage (**Figure 4A**). However, after the ion leakage, the profile of the electric potential almost resumed its symmetry, resulting in a reduced transmembrane voltage (**Figures 3F**, **4A**). This prediction is consistent with observations from our MD simulations (Figure S4). Moreover, it clearly indicates that the transmembrane voltage, produced by the close contact of the apposed membranes, was the driving force for the transient pore formation.

#### Threshold Values of Membrane Distance

To find the threshold value of membrane distance for the pore formation, we performed a series of MD simulations at different membrane distances. We defined the critical distance as the

the red line is Gaussian curve. The heads of lipids are shown in yellow, and the tails of the lipids are not shown for a clear representation of the pores. The water

distance above which the pore formation will not happen in up to 20 ns in our simulation (**Figure 4B**). When the distance, D, was larger than a critical value (∼5 nm), the voltage was nearly zero because of the symmetry of the field of electric potential in the membrane. But as the distance decreased, the voltage increased, being negatively proportional to the distance (see Equation 6 in Methods). Referring to our MD simulations, a series of membrane distance D was chosen by reducing its value from 6 nm, in 0.5 nm steps, for the lipid composition Chol:DOPC:POPE:POPS (20:45:20:15%). In these settings, we found that the transmembrane voltage produced by the close contact could be up to >3 V, consistent with previous results produced by the transmembrane ionic charge imbalance (Gurtovenko and Vattulainen, 2005). We found that when the distance was larger than 5 nm, there was no transient pore formation. But when D became smaller, pore formation could be observed. In addition, the smaller the distance is, the faster the pore could form (**Figure 4B**, blue line). These results

suggest there is a threshold value for the inter-membrane distance required for the transient pore formation.

#### Effect of Lipid Composition on Fusion Pore Formation Threshold

To investigate the contribution of the lipid composition in fusion pore formation, we modified our simulations to adopt various lipid compositions by changing the percentage of POPS, which modulates the charge density on the membrane. Our results showed that the threshold value of membrane distance for pore formation was highly dependent on the percentage of POPS (**Figure 5**). For example, we found that reducing the percentage of POPS in both membranes generally reduced the threshold value due to dissipated charge density of the membrane. The lower the percentage of the POPS, the smaller the critical distance needed, and thus the more difficult the pore formation. This result also emphasizes why two membranes should be brought close together for fusion: only a membrane

molecules are shown as transparent surface.

FIGURE 4 | Electric potential and transmembrane voltage of the membranes. (A) The predicted electric potential before pore formation and after pore closing. 1*V* is the transmembrane voltage of the vesicle. (B) The transmembrane voltage and the time for pore formation, T as function of the membrane distance, D. The black solid dots are results of MD simulations denoting the voltage that enables pore formation, while the hollow dots are for those where pore formation did not occur. The black line is the theoretical prediction of transmembrane voltage according to our model (see Equation 6 in Methods). The red horizontal solid line indicates the critical transmembrane voltage (1*Vcritical*) required for pore formation. The blue solid square indicates the time needed for pore formation calculated from MD simulations. The vertical red dashed line indicates the critical distance for pore formation (4.7 nm), beyond which the time for the pore formation is infinite. Here the lipid composition was Chol:DOPC:POPE:POPS (20:45:20:15%). Chol, DOPC, POPE, POPS are abbreviation of cholesterol, 1,2-dioleoyl-sn-glycero-3-phosphocholine, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine, and 1-palmitoyl-2-oleoyl-sn-glycero-3-phenylserine, respectively. Each MD simulation was performed independently for three replicates, each for 20 ns.

distance smaller than the threshold value can produce a sufficiently high transmembrane voltage that can induce pore formation.

Note that we did not consider the asymmetry of lipid composition between the inner and outer leaflet of cell membrane (Svennerholm, 1968; Devaux, 1991). Although, the asymmetry may affect the process of fusion pore formation, it is a secondary factor in the membrane fusion in comparison with the close contact between the membranes. Heterogeneity of lipid composition may also be an influencing factor in the fusion process. For instance, phosphatidylinositol-4,5-bisphosphate (PIP2), a lipid with -5e negative charges, is an important composition that accumulates at the location of fusion site (McLaughlin and Murray, 2005; Graber et al., 2014). But the localized PIP2 mainly interacts with syntaxin proteins during the course of membrane fusion. And the function of fusion protein syntaxin is to pull the two apposed membranes to be close contact, which we had effectively considered by applying pulling force on the membranes. To test the influence of the asymmetric lipid composition and localized PIP2, we performed an additional simulation with 5% PIP2 in the cytoplasmic leaflet which make the two membranes asymmetric, as shown in Figure S5. It shows that the close contact of the two membranes containing localized PIP2 could still give rise to a high local transmembrane voltage which induced the pore formation.

#### Force Regulates the Pore Evolution

During membrane fusion, fusion proteins apply not only vertical force to pull the vesicle close to the presynaptic membrane, but also lateral force along the membrane which produces a local membrane tension. To mimic this behavior of fusion proteins, we imposed a lateral force on the membrane once the pores were initiated, to study how the local membrane tension regulates its structural evolution. We found that when the tension was small (20.6 pN/nm), the membrane pore shrank and resealed quickly in less than 20 ns. However, when the tension was increased (24.9 pN/nm), the resealing process of

the fusion pore became slower. When the tension was further increased (32.6 pN/nm) the fusion pore started to expand and the vesicle began to collapse into the membrane (**Figure 6**). In our simulations, the first two cases, where the membranes resealed, are correspond to the scenarios of fast and slow transient pore for KR fusion, while the third case corresponds to the pore dilation of FF. This result clearly indicates that the stability and evolution of the fusion pore, and the selection of fusion pathways between KR fusion and FF, could be regulated by the membrane tension.

#### DISCUSSION

This study suggests a feasible mechanism for pore formation in membrane fusion: a close contact of membranes could generate a high transmembrane voltage, which in turn produces the membrane pores. Note that the close-contact induced voltage observed in our simulation is contrasted with the ad hoc applied voltage in previous electroporation experiments (Weaver and Chizmadzhev, 1996). The opening of the fusion pore, initiated from a voltage-dependent perturbation of lipid organization at the contact of apposed bilayers, begins with a lateral parting of headgroups on two membranes for allowing water molecules to enter the hydrophobic regions (Helm et al., 1992; Fesce et al., 1994). Although, the transmembrane voltage can be produced by an ad hoc ionic charge imbalance across membranes (Gurtovenko and Vattulainen, 2005) or external electric field (Tieleman, 2004; Sun et al., 2011), how cells modulate the ionic charge imbalance or electric fields in order to produce the required voltage is not known. In this study, we showed that the transmembrane voltage can be generated by reducing the distance between two membranes, an action that could be carried out by fusion proteins, such as SNAREs or the calcium sensor, synaptotagmin, etc., (Diao et al., 2009, 2013; Kyoung et al., 2013; Lai et al., 2014). Our data agrees with a growing body of work suggesting that pore formation, stability, and evolution can be subject to regulation by cellular processes rather than a stochastic thermodynamic process (Wang et al., 2009; Risselada and Grubmüller, 2012).

In agreement with previous studies, our results further support the notion that native proteins are capable of modulating fusion pore dynamics with additional helps from other proteins producing differential layers of modulation (Alabi and Tsien, 2013; Lai et al., 2015). We found that the fusion pore dynamics appeared to be finely modulated by the type and magnitude of mechanical force exerted on membrane. A vertical force pulls the vesicle close to the plasma membrane, increasing the local voltage difference resulting in more pore formation. A further increase of the vertical force produces a closer contact for to a stalk-like structure for hemifusion (Figure S3), as observed in previous experiments (Yang and Huang, 2002; Zhao et al., 2016). In contrast, we found that the membrane tension at the contact site induced by the lateral force stabilizes pore formation and increases pore duration, while a larger force could induce pore expansion (**Figure 6**). These results suggest that the fusion dynamics could be regulated by the force of fusion proteins, and by extension, so as the rate of neurotransmitter diffusion out of the synaptic vesicle (Richards, 2009; Zhang et al., 2009; Alabi and Tsien, 2013). Since KR fusion and hemifusion are intermediate states between prime (un-fused) state and FF, the mechanical forces produced by fusion machinery proteins could select among KR, hemifusion, and FF (**Figure 7**), to modify the kinetics of neurotransmitter release which could ultimately provide a mechanism to regulate synaptic strength and achieve synaptic plasticity (He and Wu, 2007).

In conclusion, we provided a new mechanism for the initiation of fusion pore and its evolution through MD simulations of two membranes being set close together. Subsequent fusion pore initiation occurred within several nanoseconds once the membrane distance was smaller than ∼5 nm. A high transmembrane voltage induced by close membrane proximity could cause these fast-forming transient membrane pores. The fusion pores are able to conduct ions across membrane and in doing so, alleviate the voltage difference which leads to

final resealing of the membranes within a few nanoseconds. These sequential processes of pore formation, ion exchange, and pore healing in our MD simulations are consistent with the features of KR fusion observed in experiments (Zhang et al., 2009). In addition, we showed that the applied force on the membrane from native proteins plays a crucial role in modulating the stability and lifetime of fusion pores and may regulate the ultimate fusion mode to either KR fusion or FF. Therefore, both the pore formation and evolution for membrane fusion are tightly controlled by fusion proteins in physiological conditions.

### AUTHOR CONTRIBUTIONS

DL and BJ designed research; BB, ZT, and DL performed research; BB, ZT, DL, and BJ analyzed data; BB, ZT, DL, and BJ wrote the paper. All authors reviewed the manuscript.

## REFERENCES

hemifusion (D).


### FUNDING

This paper was supported by 973 programs (2015CB856304) and Natural Science Foundation of China (Grant No. 11372042, 11221202, 11532009, and 11202026).

### ACKNOWLEDGMENTS

We would like to thank Drs. Jeremy Leitz and Jiajie Diao for fruitful discussions.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2016.00136/full#supplementary-material


additive force field for lipids: validation on six lipid types. J. Phys. Chem. B. 114, 7830–7843. doi: 10.1021/jp101759q


permeability to small molecules and by lysis. Mol. Biol. Cell 22, 4635–4646. doi: 10.1091/mbc.E11-08-0680

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Bu, Tian, Li and Ji. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Drive the Car(go)s—New Modalities to Control Cargo Trafficking in Live Cells

Payel Mondal <sup>1</sup> , John S. Khamo<sup>1</sup> , Vishnu V. Krishnamurthy <sup>1</sup> , Qi Cai <sup>1</sup> and Kai Zhang1,2,3 \*

<sup>1</sup>Department of Biochemistry, University of Illinois at Urbana-Champaign, Urbana, IL, USA, <sup>2</sup>Neuroscience Program, University of Illinois at Urbana-Champaign, Urbana, IL, USA, <sup>3</sup>Center for Biophysics and Quantitative Biology, University of Illinois at Urbana-Champaign, Urbana, IL, USA

Synaptic transmission is a fundamental molecular process underlying learning and memory. Successful synaptic transmission involves coupled interaction between electrical signals (action potentials) and chemical signals (neurotransmitters). Defective synaptic transmission has been reported in a variety of neurological disorders such as Autism and Alzheimer's disease. A large variety of macromolecules and organelles are enriched near functional synapses. Although a portion of macromolecules can be produced locally at the synapse, a large number of synaptic components especially the membrane-bound receptors and peptide neurotransmitters require active transport machinery to reach their sites of action. This spatial relocation is mediated by energyconsuming, motor protein-driven cargo trafficking. Properly regulated cargo trafficking is of fundamental importance to neuronal functions, including synaptic transmission. In this review, we discuss the molecular machinery of cargo trafficking with emphasis on new experimental strategies that enable direct modulation of cargo trafficking in live cells. These strategies promise to provide insights into a quantitative understanding of cargo trafficking, which could lead to new intervention strategies for the treatment of neurological diseases.

#### Edited by:

Jiajie Diao, University of Cincinnati, USA

#### Reviewed by:

Yong Wang, University of Arkansas, USA Ruoyi Qiu, Stanford University, USA

> \*Correspondence: Kai Zhang

kaizkaiz@illinois.edu

Received: 08 November 2016 Accepted: 05 January 2017 Published: 20 January 2017

#### Citation:

Mondal P, Khamo JS, Krishnamurthy VV, Cai Q and Zhang K (2017) Drive the Car(go)s—New Modalities to Control Cargo Trafficking in Live Cells. Front. Mol. Neurosci. 10:4. doi: 10.3389/fnmol.2017.00004 Keywords: synaptic transmission, neurological disorders, cargo trafficking, motor proteins, axonal transport, optogenetics, chemically induced dimerization, photoactivatable proteins

## INTRODUCTION

The human brain has approximately 86 billion neuronal cells (Azevedo et al., 2009), each of which possesses a large number of synapses to other cells. For instance, each neocortical neuron has an average of about 7000 synapses for exchanges of information (Pakkenberg et al., 2003). Synaptic transmission, which relays information from one cell to the next via coupled events between electrical and chemical signals through synapses, plays a crucial role in learning and memory consolidation. In response to cell depolarization (electrical signals), neurotransmitters (chemical signals) are released into the synaptic cleft, bind to their postsynaptic receptors and trigger downstream signaling activities in the postsynaptic cell. Defective synaptic transmission causes a variety of neurological disorders (van Spronsen and Hoogenraad, 2010) and cognitive diseases (Lau and Zukin, 2007; Südhof, 2008).

Quantitative analysis of synapses has found that, in addition to neurotransmitters, many other molecules and organelles are enriched in the synapse. These molecules include receptors and channels, cytoskeleton, kinases and phosphatases and their regulators, GTPases and their regulators, motor proteins, scaffolding proteins, components of signaling and membrane trafficking, and mitochondria (Ziv and Garner, 2004; Sheng and Hoogenraad, 2007; Bourne and Harris, 2008; Margeta et al., 2008). Indeed, proteomic studies have revealed that more than 2000 different proteins reside in the synapse (Dieterich and Kreutz, 2016). A portion of these synaptic components are likely to be synthesized locally, given that components of translation machinery, e.g., polyribosomes, reside in dendritic shafts and spines. Indeed, it has been shown that rapid dendritic protein synthesis occurs during metabotropic glutamate receptor (mGluR)-dependent long-term depression (LTD; Huber et al., 2000). De-centralized protein synthesis has been an emerging paradigm in studying how neurons achieve specific functions and signals that occur in highly compartmentalized subcellular domains (Holt and Schuman, 2013). Many crucial synaptic components, however, cannot be produced locally in the synapse (Kennedy and Ehlers, 2006). These components include neurotrophin receptors (Ascaño et al., 2009), dense core vesicles, synaptic vesicle precursors (Goldstein et al., 2008) and neurotransmitter receptors (Kneussel and Loebrich, 2007; Shepherd and Huganir, 2007). Biogenesis of these macromolecules often occurs at a great distance from the synapse. The extremely polarized neuronal morphology excludes access of these macromolecules to synapses based on diffusion. As a result, these synaptic components require energy-consuming, motor protein-driven trafficking mechanism to reach their sites of action (Schlager and Hoogenraad, 2009). Thus, a functional synapse requires properly regulated cargo trafficking.

In this review article, we discuss basic cargo trafficking machinery with emphasis on the recently developed strategies that allow active manipulation of cargo trafficking in live cells. Here, cargo trafficking is defined as the process that involves motor-protein-driven transport along cytoskeletons, in contrast to trafficking involved in neuronal activity-regulated cargo endocytosis, exocytosis or lateral diffusion within the plasma membrane. The biotechnological advances discussed herein promise to generate new insights into the understanding of synapse building, synaptic transmission and neurological diseases.

#### BASIC COMPONENTS OF CARGO TRAFFICKING

Cargo trafficking is mediated through the interaction of the vesicle with the cytoskeletal tracks. Basic components of cargo trafficking include cargos (vehicles), motor proteins (wheels), cytoskeleton (road) and energy (fuel).

#### Cargos

In axons, cargos travel through either fast or slow axonal transport (Vallee and Bloom, 1991). Cargos that travel through fast axonal transport have an average speed of about 0.5–5 micron/s (40–400 mm/day). Synaptic vesicles and enzymes for neurotransmitter metabolism use anterograde transport (from the cell body to the axon terminal); internalized membrane receptors and neurotrophins use retrograde transport (from the axon terminal to the cell body); organelles such as mitochondria travel in both anterograde and retrograde directions through engagement of motor adaptor proteins such as trafficking kinesin protein (TRAK)/Milton (van Spronsen et al., 2013). Cargos that travel through slow axonal transport have an average speed of 0.3–8 mm/day. These cargos include ''building materials'' of neuronal cytoskeletons such as neurofilaments and microtubules, actins, spectrin and tau proteins. Both fast and slow axonal transport adopts a ''stopand-go'' pattern, i.e., cruising intersected by pausing (Brown, 2000). Intriguingly, the slow axonal transport is driven by ''fast'' motors, and the slow speed is due to prolonged pauses (Brown, 2003).

#### Cytoskeleton and Motor Proteins

Cargo trafficking depends on the interaction between cytoskeleton and motor proteins (Vale, 2003). The neuronal cytoskeleton is composed of microtubules, actin filaments and neurofilaments (Kevenaar and Hoogenraad, 2015). The main function of neurofilaments, enriched primarily in axons, is to control the axon diameter and axonal conductance (Yuan et al., 2012). Microtubules and actin filaments serve as tracks for motor proteins in axonal and dendritic shafts. On microtubules, kinesin superfamily motor proteins drive anterograde transport (Gennerich and Vale, 2009; Hirokawa et al., 2009); cytoplasmic dyneins drive retrograde transport. One type of motor protein can transport a variety of cargos. For instance, kinesin-1 transports components of cytoskeleton, mitochondria and Soluble NSF Attachment Protein Receptor (SNARE) proteins (Hirokawa and Noda, 2008). Such a specificity of transport can be achieved through splice variants of motor proteins (Cyr et al., 1991) or post translational modifications such as selective phosphorylation in kinesin light chain (Ichimura et al., 2002; Vagnoni et al., 2011). Myosins are a superfamily of motor proteins that travel along actin filaments (Mitchison and Cramer, 1996; Blanchoin et al., 2014). Recently, super-resolution microscopy showed that axonal actin is also organized in regularly spaced rings that wrap around the circumference of axons (Xu et al., 2013). This subpopulation of axonal actin is likely to provide mechanical support for the axon membrane and may not be involved in the myosin-dependent cargo trafficking.

#### Energy

Intriguingly, although mitochondria are the major organelles that provide energy to boost up molecular machineries in cells (Sheng, 2014), they may not be the energy resource for axonal transport. Instead, the energy is more likely to be supplied by ATP generated by vesicular glycolysis (Zala et al., 2013). Inhibition of ATP production from mitochondria via oligomycin, an inhibitor of mitochondrial H+-ATP-synthase, did not affect the fast axonal transport of brain-derived neurotrophic factor (BDNF). In contrast, treating cells with iodoacetate, which inhibits glyceraldehyde-3-phosphate dehydrogenase (GAPDH), the key glycolytic enzyme, significantly reduced the average velocity of BDNF (Zala et al., 2013). On the other hand, although oligomycin had no effect on vesicle transport, it blocked mitochondria trafficking, consistent with previous findings that loss of mitochondrial ATP production induces loss of mitochondrial dynamics (Kaasik et al., 2007).

#### TRACKING CARGO TRANSPORT IN LIVE CELLS

Early work used radioactive labeling to detect cargo trafficking in neurons (Lasek, 1967; Ochs et al., 1969). Although this method confirmed the existence of fast and slow axonal transport, it lacked the resolution to track individual cargos. Video-enhanced contract-differential interference contrast microscopy (Brady et al., 1982) allowed for tracking of individual cargoes, but could not differentiate their identities. Advanced fluorescence microscopy techniques, such as single-molecule fluorescence microscopy, have enabled real-time tracking of neurotrophin transport in live neuronal cells (Tani et al., 2005). However, the data acquisition time was limited to tens of seconds owing to photobleaching of the organic fluorophore such as Cy3. More photostable probes such as semiconductor nanocrystals (quantum dots) allowed for continuous tracking of cargos along neuronal processes for several minutes over hundreds of microns (Cui et al., 2007). Although typical quantum dots (about 20 nm in diameter) are much larger than organic fluorophores, they did not seem to disturb the biological activity of neurotrophin (Cui et al., 2007). Recent development of small quantum dots (9 nm in diameter) will further improve their use in live cell imaging (Cai et al., 2014). A critical difference between in vitro and live-cell cargo trafficking is that intracellular trafficking is regulated not only by motor proteins, but also by cargo-organelle and cargo-cytoskeletal interactions. Evidence shows that early endosomal interaction with microtubule intersection, other early endosomes, and endoplasmic reticulum contributes to the pausing of epidermal growth factor-containing early endosomes (Zajac et al., 2013). To effectively determine the directionality of cargo trafficking, Campenot (1977) designed the first prototype of compartmentalized culturing device that separates the cell body from the distant neurite. Improved quality has been achieved by replacing Teflon with polydimethylsiloxane (PDMS), a transparent and highly biocompatible material (Taylor et al., 2005; Mudrakola et al., 2009; Zhang et al., 2011).

### RESTORATION OF CARGO TRAFFICKING EXERTS NEUROPROTECTIVE EFFECTS

Defective cargo trafficking has been found in a variety of neurological disorders (Tischfield et al., 2011) and brain injury (Povlishock and Jenkins, 1995). For instance, huntingtinassociated protein 1 (HAP1) is highly expressed in neurons and mediates kinesin-based anterograde transport (McGuire et al., 2006). In Huntingtin disease, stronger interaction between huntingtin protein and HAP1 leads to detachment of molecular motors from BDNF-containing cargos and reduced BDNF transport (Charrin et al., 2005). Analysis of axonal transport defects in human disease has been comprehensively reviewed and will not be repeated here (Roy et al., 2005; Chevalier-Larsen and Holzbaur, 2006; De Vos et al., 2008; Morfini et al., 2009; Hirokawa et al., 2010; Hinckelmann et al., 2013). Notably, although the causality of defective cargo trafficking to neurological disorders is still under debate (Goldstein, 2012), multiple lines of research have provided evidence that restoration of axonal transport can exert neuroprotective effects (Hinckelmann et al., 2013). For instance, failed retrograde transport of nerve growth factor (NGF) from the hippocampus to the basal forebrain caused reduction in size and number of basal forebrain cholinergic neurons (BFCN) in the partial trisomy 16 (Ts65Dn) mouse model of Down's syndrome. Such defects were rescued by delivering NGF directly to the cell bodies of BFCN through intracerebroventricular administration, which bypassed defective axonal transport (Cooper et al., 2001). Reduction of the endogenous level of Tau, a microtubule-associated protein, ameliorated amyloid β-induced deficits in an Alzheimer's disease mouse model (Roberson et al., 2007). Tau reduction has also been shown to rescue defective axonal transport of mitochondria and neurotrophin receptors (Vossel et al., 2010). Modulation of tau-microtubule interactions has been proposed as a therapeutic strategy for the treatment of tauopathies (Ballatore et al., 2011). The majority of these studies used an indirect way (e.g., bypassing axonal transport or genetic modulation of microtubuleassociation protein) to rescue defective transport. It remains unknown if direct rescuing of cargo trafficking is sufficient to induce neuroprotective effects. Recent biotechnological advances have started to offer new opportunities to address this issue.

#### DIRECT CONTROL OF CARGO TRAFFICKING IN LIVE CELLS

Correct positioning of organelles plays a crucial role in signaling regulation, cell differentiation and development (van Bergeijk et al., 2016). For instance, localized positioning of endosomes contributes to polarization and local outgrowth of neuronal cells (Sadowski et al., 2009; Eva et al., 2010, 2012; Golachowska et al., 2010; Higuchi et al., 2014). Similarly, correct mitochondrial positioning helps in axon branching (Courchet et al., 2013; Spillane et al., 2013) and synaptic function (MacAskill et al., 2010; Sheng and Cai, 2012). Golgi positioning is crucial to axon specification and dendrite development (Yadav and Linstedt, 2011; Ori-McKenney et al., 2012). Active nuclear positioning ensures correct cellular function during cell division, migration and differentiation (Gundersen and Worman, 2013). Altered positioning of dynamic organelles in cells is involved in neurodegenerative disorders. For instance, perinuclear accumulation of lysosomes is increased in a cellular model of Huntington's disease (Erie et al., 2015). Taking advantages of accumulating knowledge of motor and scaffolding proteins involved in organelle transport

cryptochrome 2; CIB1, cryptochrome 2 interacting basic helix-loop-helix; BICDN, the amino terminus of bicaudal D homolog 2 (BICD2); TrkB, tropomyosin-related kinase B; fMNP, anti-TrkB functionalized superparamagnetic nanoparticle.

(Fu and Holzbaur, 2014), emerging new biotechnologies have enabled direct control of organelle trafficking in live cells with high spatiotemporal resolution and cargo specificity (**Figure 1** and **Table 1**).

## Chemically Induced Dimerization (CID)

Chemically induced dimerization (CID) uses a small molecule to induce binding between two proteins (Putyrski and Schultz, 2012; Rakhit et al., 2014; Voss et al., 2015;


TABLE 1 | Summary of current controlling mechanisms for cargo trafficking in live cells.

**Figure 1A**). A commonly used module is the rapamycin based FK506 Binding Protein (FKBP) and the FKBP Rapamycin Binding (FRB) domain of mammalian target of rapamycin (mTOR; Banaszynski et al., 2005; Inoue et al., 2005). This system has been used to recruit motor proteins (or their adapters) to peroxisomes to achieve rapamycin-induced transport along corresponding cytoskeletons (Kapitein et al., 2010). A similar scheme has also been used to position early endosomes or late endosomes by fusing the FKBP-FRB system to endosomal markers (Rab5 and Rab7) and motor proteins (Bentley et al., 2015).

#### Optochemical Control

An optochemical system utilizes a photoactivatable ligand to induce association of a pair of proteins (**Figure 1B**). One such ligand is cTMP-Htag, a synthetic, cell-permeant, small molecule comprising a Halotag ligand (a ligand for Haloenzyme) linked to photocaged trimethoprim (TMP), a ligand for Escherichia coli dihydrofolate reductase (eDHFR). A pulse of UV light uncages TMP and fully activates the dual-ligand, which crosslinks the Haloenzyme and the eDHFRfusion protein (Ballister et al., 2014). When applied in cells where eDHPR was fused to motors or motor effectors and Halotag was fused to cargos, eTMP-Htag enabled lightcontrolled crosslinking between cargos and motors (Ballister et al., 2015). This system has allowed for directional control of mitochondria or peroxisome trafficking in neurons. Other optochemical systems, such as those based on photocaged rapamycin (Karginov et al., 2011; Umeda et al., 2011), chemically modified abscisic acid (Wright et al., 2015; Zeng et al., 2015) and gibberellic acid (Schelkle et al., 2015), photoactivatable crosslinker for SNAPTag and HaloTag (Zimmermann et al., 2014), are also expected to achieve similar optochemical control.

### Optogenetic Control

Optogenetics harnesses the power of light to modulate proteinprotein interactions in live cells (**Figure 1C**). Shortly after its initial success in controlling neuronal firing (Banghart et al., 2004; Boyden et al., 2005; Deisseroth, 2011), optogenetics has been extended to control other cellular processes such as gene transcription, translation, protein splicing, protein degradation, cell differentiation and cell death. The possibility of modulating signaling pathways and cell functions with high spatiotemporal precision offers an entirely new modality to dissect molecular mechanisms governing cell fate determination (Toettcher et al., 2011; Zoltowski and Gardner, 2011; Tucker, 2012; Kim and Lin, 2013; Tischer and Weiner, 2014; Zhang and Cui, 2015). Photoactivatable proteins have been used in multiple model systems including yeast (Shimizu-Sato et al., 2002; Tyszkiewicz and Muir, 2008; Hughes et al., 2012; Strickland et al., 2012), mammalian cells (Levskaya et al., 2009; Wu et al., 2009; Yazawa et al., 2009; Kennedy et al., 2010; Toettcher et al., 2011; Idevall-Hagren et al., 2012; Mills et al., 2012; Zhou et al., 2012; Bugaj et al., 2013; Grusch et al., 2014; Kim et al., 2014; Lee et al., 2014; Taslimi et al., 2014; Zhang et al., 2014; Hughes et al., 2015; Kawano et al., 2015; Yumerefendi et al., 2016), primary neurons (Chen et al., 2013; Kakumoto and Nakata, 2013; Konermann et al., 2013), Drosophila (Boulina et al., 2013), zebrafish embryos (Liu et al., 2012; Motta-Mena et al., 2014; Buckley et al., 2016) and Xenopus embryos (Krishnamurthy et al., 2016). To control cargo trafficking, photoactivatable proteins such as the light, oxygen, voltage-peptide epitope (LOV-pep) and engineered PDZ domain (ePDZ; van Bergeijk et al., 2015) or cryptochrome 2 (CRY2) and cryptochrome 2 interacting basic helix-loop-helix (CIB1; Duan et al., 2015) were fused to cargoes and motor proteins or motor adapters (**Figure 1D**). Interestingly, directionality of transport seems to depend on the load of motor proteins. By engineering the LOV domain into the lever arm of myosin or kinesin, the directionality of these motor proteins can be reversibly modulated as reported in a recent in vitro assay (Nakamura et al., 2014).

#### Magnetic Control

Another strategy utilizes magnetic force to reverse cargo transport. Using an electromagnetic needle and antibodyfunctionalized superparamagnetic nanoparticles (fMNPs), Steketee et al. (2011) could reverse the direction of transport of TrkB-containing endosomes in retinal ganglion cells (**Figure 1E**). Manipulation of fMNP signaling endosomes by a focal magnetic field altered growth cone motility and halted neurite outgrowth (Steketee et al., 2011).

Notably, trafficking along the secretory pathway between membrane-bound cellular compartments including the endoplasmic reticulum, Golgi apparatus, endosome and plasma membrane can also be controlled via chemical, optochemical and optogenetic strategies. The general strategy involves a chemical- or light-induced activation of the targeting signal, either by uncaging a blocking motif (Abraham et al., 2016) or inducing dissociation of a mislocalized protein cluster (Rivera et al., 2000; Al-Bassam et al., 2012; Chen et al., 2013). Interested readers are encouraged to refer to the references listed in **Table 1**.

#### OUTSTANDING QUESTIONS AND FUTURE DIRECTIONS

Cargo trafficking plays a crucial role in neuronal survival, differentiation, axon pathfinding, as well as synaptogenesis and synaptic transmission. With advances in genetic and protein engineering, single-molecule fluorescence microscopy, microfluidics, CID and optogenetics, one can control cargo trafficking with superior spatiotemporal resolution and molecular specificity. Because most of controlling systems are genetically encoded, it is possible to generate novel model systems harboring light- or chemical- sensitive signaling circuits. These tools could thus provide new perspectives to address

### REFERENCES


controversies in the field of cargo trafficking in neuroscience. On the other hand, significant improvement of current technologies is needed before they can be successfully applied in tissues or multicellular organisms. For instance, single-molecule fluorescence microscopy has been mostly applied in vitro or in separated cells. Its potential in multicellular organisms has yet to be fully realized, owing to the limited penetration depth of visible light in the high-absorbing, high-scattering biological tissues. Poor penetration of visible light in biological tissues also results in invasiveness and low throughput of current optogenetic techniques, which often relies on insertion of fiber optics or microscale light emitting diodes arrays (Kim et al., 2013) in tissues for light delivery. Successful removal of these technical barriers requires a collaborative effort of researchers from multi-disciplinary fields including physics, material sciences, biochemistry and bioengineering. Shortly after the initial phase of tool development, as demonstrated in recent literature, we believe follow-up work will start to address the signaling outcomes in response to the modulated cargo trafficking. For instance, is defective cargo transport a cause or a result of misregulated neuronal functions and neurological disorders? Can we rescue defective neuronal phenotypes by direct modulation of cargo trafficking? We believe that biotechnological advances will continue pushing forward our understanding of the molecular machinery underlying neuronal survival, differentiation, repair and synaptic transmission and plasticity.

#### AUTHOR CONTRIBUTIONS

PM, JSK, VVK, QC and KZ performed literature search and wrote the initial draft. PM generated **Table 1**. QC designed and generated **Figure 1**. QC and KZ wrote the final manuscript.

#### ACKNOWLEDGMENTS

This work was supported by the University of Illinois at Urbana-Champaign. We apologize to those colleagues whose work could not be cited here owing to space limitations.


Alzheimer's disease mouse model. Science 316, 750–754. doi: 10.1126/science. 1141736


Raf/MEK/ERK pathway in PC12 cell neurite outgrowth. PLoS One 9:e92917. doi: 10.1371/journal.pone.0092917


Zoltowski, B. D., and Gardner, K. H. (2011). Tripping the light fantastic: blue-light photoreceptors as examples of environmentally modulated protein-protein interactions. Biochemistry 50, 4–16. doi: 10.1021/bi101665s

**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Mondal, Khamo, Krishnamurthy, Cai and Zhang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Structure of the Synaptic Vesicle-Plasma Membrane Interface Constrains SNARE Models of Rapid, Synchronous Exocytosis at Nerve Terminals

#### Cameron B. Gundersen \*

Department of Molecular and Medical Pharmacology, David Geffen UCLA School of Medicine, Los Angeles, CA, USA

Contemporary models of neurotransmitter release invoke direct or indirect interactions between the Ca<sup>2</sup><sup>+</sup> sensor, synaptotagmin and the incompletely zippered soluble, N-ethyl-maleimide-sensitive factor attachment protein receptor (SNARE) complex. However, recent electron microscopic (EM) investigations have raised pragmatic issues concerning the mechanism by which SNAREs trigger membrane fusion at nerve terminals. The first issue is related to the finding that the area of contact between a "fully primed" synaptic vesicle and the plasma membrane can exceed 600 nm<sup>2</sup> . Approximately four-thousands lipid molecules can inhabit this contact zone. Thus, renewed efforts will be needed to explain how the zippering of as few as two SNARE complexes mobilizes these lipids to achieve membrane fusion. The second issue emerges from the finding that "docking filaments" are sandwiched within the area of vesicle-plasma membrane contact. It is challenging to reconcile the location of these filaments with SNARE models of exocytosis. Instead, this commentary outlines how these data are more compatible with a model in which a cluster of synaptotagmins catalyzes exocytotic membrane fusion.

#### Edited by:

Jiajie Diao, University of Cincinnati, USA

#### Reviewed by:

Xiaochu Lou, University of Wisconsin-Madison, USA Volker Kiessling, University of Virginia, USA

\*Correspondence: Cameron B. Gundersen cgundersen@mednet.ucla.edu

Received: 11 December 2016 Accepted: 13 February 2017 Published: 23 February 2017

#### Citation:

Gundersen CB (2017) The Structure of the Synaptic Vesicle-Plasma Membrane Interface Constrains SNARE Models of Rapid, Synchronous Exocytosis at Nerve Terminals. Front. Mol. Neurosci. 10:48. doi: 10.3389/fnmol.2017.00048 Keywords: synaptotagmin, active zone, docking filaments, tomographic reconstruction, transmitter release mechanisms, synapse function

### INTRODUCTION

A major goal of neuroscience research is to clarify the molecular events that lead to the fast, synchronous release of neurotransmitters at chemical synapses. Seminal studies revealed that transmitter secretion is initiated via the depolarization-dependent entry of Ca2<sup>+</sup> into the nerve terminal which triggers synaptic vesicle exocytosis (Katz, 1966; Heuser, 1989). This scenario raised two important questions: first, what is the identity of the target to which Ca2<sup>+</sup> binds? and, second, how does Ca2<sup>+</sup> binding to this target promote exocytosis? The consensus answer to the first question is that synaptotagmin 1 (or 2) is the physiological Ca2+-sensor for rapid, synchronous exocytosis at most nerve terminals (Südhof, 2014). The answer to the second question remains less clear. The prevailing view is that Ca2+-bound synaptotagmin triggers exocytosis by interacting directly or indirectly with soluble, N-ethyl-maleimide-sensitive factor attachment protein receptor (SNARE) proteins (Rothman, 2014; Südhof, 2014). However, as a follow-up to a comprehensive review (Meriney et al., 2014), this commentary will emphasize that much remains to be clarified about how SNARE proteins catalyze exocytotic membrane fusion. Additionally, it will be argued that recent empirical developments favor a simpler solution in which synaptotagmin is the template for exocytotic membrane fusion (Gundersen and Umbach, 2013).

#### THE DISCOVERY OF SYNAPTOTAGMIN AND SNAREs

Systematic efforts to clone and sequence the cDNAs encoding synaptic vesicle proteins led to the finding that a previously identified constituent of synaptic vesicles, p65, had two motifs that were related to presumptive Ca2+-binding domains of protein kinase C (Perin et al., 1990). These C2 domains were later shown to bind Ca2<sup>+</sup> (Brose et al., 1992), and investigations from a number of groups ultimately led to the conclusion that synaptotagmins 1 and 2 were the principal Ca2+-sensors for synchronous exocytosis at chemical synapses (Südhof, 2014). The discovery of SNARE proteins was more convoluted. It began with the identification of soluble proteins (N-ethylmaleimide sensitive factor, or NSF, and the NSF adaptor proteins, or SNAPs) which were essential for membrane-trafficking in the Golgi apparatus. Then, because vertebrate brain had a high abundance of membrane targets for these soluble proteins, brain extracts were used in an affinity-purification scheme to identify the SNAP ''receptors'', or SNAREs. The remarkable upshot of this effort was that the SNAREs were found to include a pair of plasma membrane-associated proteins (syntaxin A/B and synaptosome-associated protein of 25 kDa, or SNAP-25) and one synaptic vesicle protein (synaptobrevin 2; Söllner et al., 1993b). The further observations that SNAREs were targets of clostridial neurotoxins (Schiavo et al., 1992) and formed a ternary complex suggested that SNAREs might constitute a molecular link between a synaptic vesicle and the plasma membrane that could be exploited to drive membrane fusion (Söllner et al., 1993a,b). However, it was the finding that SNARE proteins formed parallel, rather than anti-parallel, complexes which supplied the conceptual basis for all subsequent models of SNARE involvement in membrane fusion (Hanson et al., 1997; Sutton et al., 1998). And, with the report that SNAREs promoted liposomal fusion (Weber et al., 1998), widespread efforts focused on the mechanism by which synaptotagmin interfaces with SNAREs to regulate exocytosis.

#### EVOLVING MODELS OF SYNAPTOTAGMIN AND SNARE FUNCTION IN SYNAPTIC VESICLE EXOCYTOSIS

The crucial question to emerge from the preceding research was: ''How does synaptotagmin control SNARE-mediated membrane fusion?''. The field still lacks a clear answer for this question. This absence of a unifying model of the exocytotic cascade has spawned a large number of competing proposals. Prominent examples of exocytotic models are given in the following publications: (Jahn and Fasshauer, 2012; Kasai et al., 2012; Mohrmann and Sørensen, 2012; Fang and Lindau, 2014; Kaeser and Regehr, 2014; Südhof, 2014; Rothman, 2014; Brewer et al., 2015; Rizo and Xu, 2015; Schneggenburger and Rosenmund, 2015; Zhou et al., 2015; Lou and Shin, 2016 and for a thorough critique of SNARE models see Meriney et al., 2014). With few exceptions, these models rely on the same three assumptions: The first is that SNARE complexes of suitably docked and primed synaptic vesicles are partially ''zippered''. In other words, the coiled-coil interactions among synaptobrevin, syntaxin and SNAP-25 are arrested at an intermediate stage. The second assumption is that the completion of SNARE zippering supplies energy to drive the fusion of the vesicular and plasma membranes. The third assumption is that the Ca2+ bound state of synaptotagmin overrides the arrest of SNARE zippering to initiate the fusion process. Beyond these similarities, the reader should consult the cited references to understand how they differ in their treatment of auxiliary, SNARE-binding proteins (like, the complexins and the mammalian homologs of the nematode unc proteins, munc-13 and munc-18), and how they envision synaptotagmin relieving the arrest of SNARE zippering. However, for the purposes of this review, the most important difference among the cited models concerns their positioning of a release-ready synaptic vesicle. While some models locate the vesicle several nanometers from the plasma membrane (**Figure 1A**), others begin with the vesicular and plasma membranes in direct contact (**Figure 1B**). This difference in spatial organization has crucial implications as addressed next.

### SYNAPTIC VESICLE LOCATION IS A CRUCIAL CONSIDERATION IN MODELS OF NERVE TERMINAL EXOCYTOSIS

SNARE-centric models of exocytosis typically begin with the architecture in **Figures 1A,B**. **Figure 1A** models are attractive, because it is intuitively evident how full zippering of the SNAREs might induce the formation of a fusion ''neck'' between the juxtaposed membranes. However, the paramount objection to such models is that they are incompatible with data from the vast majority of electron microscopic (EM) studies of nerve terminals. The following citations are culled from >30 articles which used serial reconstruction or EM tomography and found no detectable separation between the membrane of ''docked'' synaptic vesicles and the plasma membrane: (Schikorski and Stevens, 1997; Harlow et al., 2001; Xu-Friedman et al., 2001; Gustafsson et al., 2002; Rizzoli and Betz, 2004; Rostaing et al., 2006; Zampighi et al., 2006; Siksou et al., 2007; Nagwaney et al., 2009; Stigloher et al., 2011; Burette et al., 2012; Holderith et al., 2012; Leitinger et al., 2012; Marra et al., 2012; Szule et al., 2012; Watanabe et al., 2013; Cole et al., 2016; Jung et al., 2016). However, in defense of **Figure 1A** models, it was prominently noted (Fernández-Busnadiego et al., 2010) that vesicle-plasma membrane contacts were very infrequent in rat synaptosomes. Nevertheless, careful perusal of this article reveals that although such contacts were rare, they were still observed in unstimulated preparations. Thus, regardless of the appeal of **Figure 1A** models, they are not supported empirically. Instead, if SNAREs drive membrane fusion, synaptic vesicles need to be positioned as in **Figure 1B**. Before critiquing **Figure 1B** models, a detour will summarize important results from two recent investigations of the synaptic vesicle-plasma membrane interface.

First, Jung et al. (2016) measured the area of contact between docked vesicles and the plasma membrane for frog nerve terminals at rest, during and after activity (reproduced in **Figure 2A**). Their data indicated that the contact area reached 650 nm<sup>2</sup> and was oval with average radii of ∼12 and ∼17 nm. Moreover, vesicles with large contact areas were depleted during synaptic activity (**Figure 2A**). This observation implied that vesicles with the largest contact areas were preferentially discharged in response to stimuli. This study also measured the thickness of the vesicular and plasma membranes away from the area of contact as well as within the contact zone. The result was that the aggregate thickness in the contact zone was twice the thickness of the individual membranes. The point here was that there was no detectable ''sandwiching'' of other material between the synaptic vesicle and the plasma membrane at their zone of contact. The other possibility was that any material that was ''sandwiched'' in this area did not measurably alter the thickness of the apposed membranes. Further implications of these results are addressed in Section ''Pros and Cons of a Synaptotagmin-Only Model of Membrane Fusion''.

A second study of vesicle-plasma membrane contacts deployed segmentation analysis of tomographic images from freeze-substituted hippocampal neurons (Cole et al., 2016). Here, the provocative finding (reproduced in **Figure 2B**) was that ''docking filaments'' traversed the interface between docked synaptic vesicles and the plasma membrane. These filaments ranged from 3 nm to 8 nm in diameter and 10–47 nm in length. Although it was concluded that these filaments were likely to include SNAREpins (a term for SNARE complexes coined by Weber et al., 1998), variation in the filament shape and distributions in the renderings indicates some level of molecular heterogeneity in the composition of these elements. Clearly, it will be important empirically to establish the identity of these filaments.

SNARE-based models of exocytosis that begin with direct vesicle-plasma membrane contact (as in **Figure 1B**) are compatible with observations from myriad groups as well as the EM data in **Figures 2A,B**. However, if the quantitative results in **Figure 2A** generalize to other nerve endings, then **Figure 1B** models confront a significant practical challenge:

FIGURE 2 | Structural features of the synaptic vesicle-plasma membrane interface. (A) These data are from Jung et al. (2016; with permission). In (A) are examples of contact areas between synaptic vesicles and the plasma membrane. (B) is a hemi-fused vesicle (scale bar: 50 nm). (C–E) are histograms of the vesicle contact areas for nerve terminals at rest, (Continued)

#### FIGURE 2 | Continued

during nerve stimulation (10 Hz for 2 min) and 1 h after stimulation. (B) This figure (from Cole et al., 2016; with permission) shows how segmentation analysis identied filaments that project into the area of contact between synaptic vesicles and the plasma membrane. The colors of the arrows in the virtual sections (A panels) correspond to the filaments in the (B) panels and the images in the (C) panels have the vesicle removed to reveal the course of the filaments. Row 4 is a fusing vesicle. Scale bar: 35 nm.

based on the data of Jung et al. (2016) one can estimate the number of lipid molecules in a 650 nm<sup>2</sup> membrane patch. By using the average cross sectional area of membrane phospholipids (0.65 nm<sup>2</sup> ; Nagle and Tristram-Nagle, 2000), and ignoring the relatively high concentration of cholesterol in the synaptic vesicle membrane (Takamori et al., 2006), the four apposed hemi-bilayers comprising the zone of vesicleplasma membrane contact harbor ∼4000 lipid molecules. At the same time, empirical studies indicate that as few as two SNARE complexes support neuronal exocytosis (Sinha et al., 2011; in contrast, explicit models requiring 6–8 SNARE complexes have been presented: Jackson, 2010; Pantano and Montecucco, 2013). Given these parameters, the drawing in **Figure 1C** illustrates the challenge facing SNARE models: there is a sea of lipid flanked by two (to scale) membrane-spanning domains contributed by synaptobrevin or syntaxin. To date, no step-by-step model explains how these SNAREs perturb the intervening lipids to induce membrane fusion.

As an alternative to the situation illustrated in **Figure 1C**, it is worthwhile considering the possibility that SNARE complexes intrude into the area of contact between synaptic vesicles and the plasma membrane. As noted above, Jung et al. (2016) found no detectable thickening of membranes at this contact zone. Because EM images of SNARE complexes reveal 4 × 14 nm filaments (Hanson et al., 1997), there should have been a demonstrable thickening of this contact region, if SNAREs were sandwiched between these membranes. The other option is that SNAREs are buried in the hydrophobic interior of the opposed membranes. To countenance this explanation, one would need to accommodate the prominent surface charge of SNARE complexes (Sutton et al., 1998) within this apolar milieu. Although such a solution appears improbable, further investigation of the vesicle-plasma interface will be needed to clarify SNARE disposition and contributions to the fusion process.

As counterpoints to the models of **Figures 1A,B**, two other proposals were recently advanced. The first was based in part on the observations of Fernández-Busnadiego et al. (2010) that synaptic vesicles seldom contacted the plasma membrane but were frequently connected to it via filaments. It was suggested that these filaments corresponded to synaptotagmin which prevented SNAREs from zippering until Ca2<sup>+</sup> entered the nerve ending (van den Bogaart et al., 2011). The primary argument against this model is the compelling evidence that releaseready synaptic vesicles directly contact the plasma membrane. The second model envisioned a ring of 16 synaptotagmins separating the vesicle from the plasma membrane and preventing full SNARE zippering (Wang et al., 2014). The concerns for this model are that the data of Jung et al. (2016) do not allow space for a synaptotagmin ring, and the filaments of Cole et al. (2016; see **Figure 2B**) are not symmetrical rings.

#### PROS AND CONS OF A "SYNAPTOTAGMIN-ONLY" MODEL OF MEMBRANE FUSION

The ''dyad hypothesis'' (Gundersen and Umbach, 2013) was advanced as an alternative to SNARE-based models of fast, synchronous exocytosis at nerve terminals. Briefly, it proposed that four synaptotagmins occupy the apical contact between a synaptic vesicle and the plasma membrane (**Figure 1D**). It further argued that Ca2<sup>+</sup> binding by the C2 domains of these synaptotagmins leads to a lateral translocation of the membranespanning domains which serve as templates for membrane fusion. Although prominent features of this model remain to be tested empirically, the following discussion indicates where the dyad model is congruent with recent observations and where further investigation of features of synaptotagmins 1 and 2 is needed.

An explicit feature of the dyad model is that protein should be found spanning the vesicle-plasma membrane interface. In this respect, it is provisionally consistent with the observations of Cole et al. (2016) that macromolecules traverse this area. Moreover, because of relatively novel structural features (discussed in Gundersen and Umbach, 2013), synaptotagmins 1 and 2 can reside at this interface without changing the thickness of the membranes. Thus, the dyad scenario is compatible with the observation (Jung et al., 2016) that there is no thickening of the membranes where vesicles are docked. However, as discussed next, two important issues need to be resolved by future experiments.

The first issue is that material equivalent to the docking filaments reported by Cole et al. (2016) has not been detected at frog motor nerve terminals (Szule et al., 2012). The origin of this discrepancy will require further evaluation. The second issue is quantitative. Based on the proposed disposition of synaptotagmins at the vesicle-plasma membrane interface, the dyad model predicted 70–80 nm<sup>2</sup> of direct contact between

#### REFERENCES


a synaptic vesicle and the plasma membrane. Even if one used a larger diameter for synaptotagmin's membrane-spanning α-helix, and extended the length of the juxta-membrane β-strand to include the seven-residue polybasic region, this area still would compute to <210 nm<sup>2</sup> . In this respect, the dyad model confronts a quantitative challenge similar to SNARE models. Namely, how does one perturb a 650 nm<sup>2</sup> area of vesicle-plasma membrane contact in a manner that is conducive to membrane fusion? To answer this query, it will be important to compare and contrast the area of direct contact between synaptic vesicles and the plasma membrane at other synapses to determine whether vesicles with >600 nm<sup>2</sup> of contact are the norm. The results of such studies should help to clarify whether fusion is driven by SNAREs (as in **Figure 1C**), synaptotagmin (as in **Figure 1D**), or via a mechanism that has yet to be proposed.

#### CONCLUSIONS

Advances in delineating the three dimensional organization and molecular composition of the synaptic vesicle-plasma membrane interface will be instrumental in distinguishing among current models of synaptic vesicle exocytosis. Although recent EM data (**Figure 2**) do not exclude SNAREs from catalyzing membrane fusion, the challenge embodied in **Figure 1C** will persist even if the area of lipid contact is halved. Instead, because synaptotagmins 1 and 2 can inhabit the synaptic vesicle-plasma membrane interface (as outlined in Gundersen and Umbach, 2013), it remains plausible that future studies will reveal a central role for synaptotagmin as a catalyst of ''fast'' membrane fusion.

#### AUTHOR CONTRIBUTIONS

This article was written by CBG.

#### FUNDING

The author currently has no extramural or intramural funding.

#### ACKNOWLEDGMENTS

Thanks to Michael Phelps for resources and Jose Rizo for comments on the manuscript.


visualized by quick-freeze/deep-etch electron microscopy. Cell 90, 523–535. doi: 10.1016/s0092-8674(00)80512-7


and facilitating synapses onto cerebellar Purkinje cells. J. Neurosci. 21, 6666–6672.


**Conflict of Interest Statement**: The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Gundersen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Melatonin Mediates Protective Effects against Kainic Acid-Induced Neuronal Death through Safeguarding ER Stress and Mitochondrial Disturbance

Feixiao Xue1,2† , Cai Shi<sup>1</sup>† , Qingjie Chen<sup>1</sup> , Weijian Hang<sup>1</sup> , Liangtao Xia<sup>1</sup> , Yue Wu<sup>1</sup> , Sophia Z. Tao<sup>3</sup> , Jie Zhou<sup>1</sup> , Anbing Shi1,4,5 \* and Juan Chen1,4 \*

<sup>1</sup> Department of Biochemistry and Molecular Biology, School of Basic Medicine and the Collaborative Innovation Center for Brain Science, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China, <sup>2</sup> Department of Clinical Laboratory, Xi'an Third Hospital, Xi'an, China, <sup>3</sup> Department of Molecular, Cellular, and Developmental Biology, University of California Santa Barbara, Santa Barbara, CA, USA, <sup>4</sup> Institute for Brain Research, Huazhong University of Science and Technology, Wuhan, China, <sup>5</sup> Key Laboratory of Neurological Disease of National Education Ministry, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China

#### Edited by:

Jiajie Diao, University of Cincinnati, USA

#### Reviewed by:

Kai Zhang, University of Illinois at Urbana-Champaign, USA Min Liu, Harvard Medical School, USA Da Xu, United States Department of Health and Human Services, USA

#### \*Correspondence:

Juan Chen chenjuanlinda69@163.com Anbing Shi ashi@hust.edu.cn †These authors have contributed equally to this work.

Received: 30 December 2016 Accepted: 13 February 2017 Published: 28 February 2017

#### Citation:

Xue F, Shi C, Chen Q, Hang W, Xia L, Wu Y, Tao SZ, Zhou J, Shi A and Chen J (2017) Melatonin Mediates Protective Effects against Kainic Acid-Induced Neuronal Death through Safeguarding ER Stress and Mitochondrial Disturbance. Front. Mol. Neurosci. 10:49. doi: 10.3389/fnmol.2017.00049 Kainic acid (KA)-induced neuronal death is linked to mitochondrial dysfunction and ER stress. Melatonin is known to protect hippocampal neurons from KA-induced apoptosis, but the exact mechanisms underlying melatonin protective effects against neuronal mitochondria disorder and ER stress remain uncertain. In this study, we investigated the sheltering roles of melatonin during KA-induced apoptosis by focusing on mitochondrial dysfunction and ER stress mediated signal pathways. KA causes mitochondrial dynamic disorder and dysfunction through calpain activation, leading to neuronal apoptosis. Ca2<sup>+</sup> chelator BAPTA-AM and calpain inhibitor calpeptin can significantly restore mitochondrial morphology and function. ER stress can also be induced by KA treatment. ER stress inhibitor 4-phenylbutyric acid (PBA) attenuates ER stress-mediated apoptosis and mitochondrial disorder. It is worth noting that calpain activation was also inhibited under PBA administration. Thus, we concluded that melatonin effectively inhibits KAinduced calpain upregulation/activation and mitochondrial deterioration by alleviating Ca2<sup>+</sup> overload and ER stress.

Keywords: melatonin, kainic acid, cell death, mitochondrial dysfunction, endoplasmic reticulum stress

## INTRODUCTION

Neurodegeneration defines the progressive loss of function and structure of neurons that ultimately leads to neuronal death (Yildiz-Unal et al., 2015). Excitotoxicity is considered to be a major factor of neuronal death in acute and chronic neurodegenerative diseases, including Alzheimer's disease (AD), Parkinson's disease (PD), Huntington's disease (HD), temporal lobe epilepsy (TLE), and amyotrophic lateral sclerosis (ALS) (Wang and Qin, 2010). Kainic acid (KA), an analog of glutamate, has been used to establish excitotoxicity models in vitro and in vivo (Sperk et al., 1983; Milatovic et al., 2001; Crespo-Biel et al., 2007; Wang et al., 2008; Li et al., 2010; Zhang and Zhu, 2011). KA associates with KA-type non-N-methyl D-aspartate (NMDA) receptors and causes

depolarization of neurons, which can result in status epilepticus, neurodegeneration, and memory loss, etc. (Osada et al., 2010). Previous studies demonstrated that KA treatment induced neuronal apoptosis in the brain, particularly affecting neurons in the hippocampal regions (Wang et al., 2005).

The overstimulation of NMDA receptors by glutamate and its analog leads to increased levels of intracellular calcium (Ca2+) (Lai et al., 2014). Calcium is one of the most important signaling molecules in neurons. Maintenance of intracellular Ca2<sup>+</sup> homeostasis is crucial for neuronal viability and functions (Bahar et al., 2016). Ca2<sup>+</sup> influx promotes the production of reactive oxygen species (ROS), releases caspase cofactors into the cytoplasm and triggers the apoptotic cascade (Manev et al., 1989). Ca2<sup>+</sup> overloading affects the mitochondrial membrane potential (MMP), which uncouples the respiratory chain and causes a decrease in ATP synthesis (Halestrap, 2009). Additionally, Ca2<sup>+</sup> overloading increases the permeability of the inner mitochondrial membrane and eventually leads to mitochondrial rupture and neuronal death (Golstein and Kroemer, 2007). Mitochondria are highly dynamic organelles, undergoing continuous fission and fusion events which are critical for the maintenance of mitochondrial functions (Nasca et al., 2016). Mitochondrial fission and fusion require outer membrane fusion proteins, mitofusin 1 and 2 (Mfn-1, Mfn-2), inner membrane fusion protein optic atrophy type 1 (OPA-1), and fission protein dynamin-related protein 1 (Drp-1) (Zhao et al., 2013). Unopposed fission leads to fragmentation, whereas unopposed fusion causes over-elongation, both of which impair mitochondrial functions (Johri and Beal, 2012). In addition to mitochondrial morphology regulation, fission and fusion processes are also essential for maintaining various aspects of mitochondrial functions (Ishihara et al., 2006). The disruption mitochondrial dynamics impair mitochondrial function and cause cell death (Ishihara et al., 2006).

The endoplasmic reticulum (ER) is another vital organelle in eukaryotic cells. The ER is responsible for various cellular activities including membrane protein synthesis and maturation, lipid biogenesis, and regulation of Ca2<sup>+</sup> levels (Bahar et al., 2016). ER stress is an adaptive response to restore ER homeostasis (Sano and Reed, 2013). However, prolonged ER stress will trigger apoptosis (Logue et al., 2013). Loss of cellular homeostasis induced Ca2<sup>+</sup> deregulation can cause ER stress-mediated apoptosis in various pathological conditions (Kruman et al., 1998; Tombal et al., 1999; Pinton et al., 2008), which further contributes to pathophysiological conditions of neurodegenerative diseases (Lindholm et al., 2006).

As a glutamate analog, KA is able to cause excessive activation of glutamate receptors (Bleakman and Lodge, 1998). Glutamate receptor over-activation induces Ca2<sup>+</sup> overloading and mitochondria functional collapse, leading to progressive neuronal death (McGeer and McGeer, 1978). KA induced Ca2<sup>+</sup> influx also leads to ER membrane disintegration, ER stress and the generation of ROS, eventually leading to neuronal apoptosis and necrosis (Schinder et al., 1996; Nicholls, 2004; Kim et al., 2016). Therefore, we speculate that KA leads to neuronal apoptosis by triggering ER stress and mitochondrial dysfunction.

Melatonin (Mel), a tryptophan metabolite, is synthesized mainly by the pineal gland. Melatonin is involved in the biological regulation of circadian rhythms, sleep, moods, reproduction, tumor growth, and aging (Jean-Louis et al., 1998; Cajochen et al., 2003; Wu and Swaab, 2005; Park et al., 2010; Prieto-Domínguez et al., 2016). A previous study showed that melatonin could inhibit traumatic brain injury induced neuronal apoptosis (Mesenge et al., 1998). It has also been demonstrated that melatonin modulates Ca2<sup>+</sup> levels and ROS production in a model of ischemia reperfusion stroke and helps to maintain the MMP in a model of oxygen and glucose deprivation (Hu et al., 2012; Gouriou et al., 2013; Li et al., 2014). All the evidence together suggests that the neuronal protective effects of melatonin might be associated with the modulation of intracellular Ca2<sup>+</sup> levels and cellular homeostasis. Thus far, the mechanisms of how melatonin protects neurons from KA-induced apoptosis remain elusive. In this study, we show that KA affects neuron viability by Ca2<sup>+</sup> overloading and mitochondrial/ER dysfunction pathways. Melatonin enacts protective effects against KAinduced neuronal apoptosis by attenuating calpain activationinduced mitochondrial dysfunction and the ER stress cascade.

#### MATERIALS AND METHODS

#### Cell Preparation

Mouse neuroblastoma N2a cells (N2a cells) were cultured with 1:1 mixture of DMEM and Opti-MEM containing 5% fetal bovine serum (Gibco, Grand Island, NY, USA) in a humified incubator aerated with 95% air and 5% CO<sup>2</sup> at 37◦C. The medium was replaced every other day, and cells were plated at an appropriate density according to each experimental scale.

In experiment 1, N2a cells were treated with KA at concentrations of 0, 25, 50, and 100 µM for 8 h. 3-[4,5 dimethylthiazol-2-yl]-2,5-diphenyl-tetrazolium bromide (MTT) assay (Solarbio, Beijing, China) and crystal violet assay (Solarbio, Beijing, China) were performed to detect viability of cells. Release of lactate dehydrogenase (LDH) was detected by using LDH Cytotoxicity Detection Kit (Dojindo, Japan). To assay melatonin effects, N2a cells were pre-treated with melatonin at concentrations of 0, 25, 50, and 100 µM for 1 h before KA was added into the medium. The MTT, crystal violet and LDH assay were performed after the KA treatment.

All animal handling and surgeries were performed in accordance with the Care Standards of Laboratory Animals (China Ministry of Health Publication, 2001). Primary hippocampus neurons were prepared from 2-day-old Sprague-Dawley rats using the described protocols, with modifications (Isaev et al., 2002). The tissue was briefly digested with 0.25% trypsin in phosphate-buffered saline (PBS) for 20 min at 37◦C followed by mechanical dissociation. Hippocampal neurons were seeded in poly-L-lysine-coated plates (120,000 cells/cm<sup>2</sup> ) and grown in neurobasal medium with B-27 serum-free supplement (Gibco, Grand Island, NY, USA), 100 U/mL penicillin, 100 g/mL streptomycin, and 2 mM L-glutamine. The cultures were maintained in a humid incubator aerated with 95% air and 5% CO<sup>2</sup> at 37◦C. The medium was changed starting from day 4 by

replacing half of the medium twice a week. Serum-free primary hippocampal cultures were utilized for the experiments after 8 days.

In experiment 2, the primary hippocampus neurons were pre-treated with or without melatonin (50 µM) for 1 h, and then stimulated with KA (50 µM) for 8 h. Melatonin (Sigma, St. Louis, MO, USA) was first dissolved in absolute ethanol at a concentration of 50 mM and diluted with culture medium to the final concentration. KA (Abcam, Cambridge, MA, USA) was first dissolved in DMSO at a concentration of 100 mM and then diluted with culture medium to the final concentration. Corresponding dilutions of ethanol and DMSO were given to the control group.

In experiment 3, the N2a cells were pre-treated with or without calpeptin (Cal, an calpain inhibitor, 20 µM, Sigma, St. Louis, MO, USA), BAPTA-AM (an Calcium chelating agent, 2.5 µM, Sigma, St. Louis, MO, USA) and sodium 4-phenylbutyrate (PBA) (Ye et al., 2014), an ER stress inhibitor, 1 mM (Sigma, St. Louis, MO, USA) for 1 h, and then treated with KA (50 µM) for 8 h.

Calpeptin and BAPTA-AM were dissolved in DMSO as stock solutions at concentrations of 20 and 5 mM, then further diluted with the cell culture medium to final concentrations of 20 and 2.5 µM. PBA was first dissolved in DMSO as stock solution at a concentration of 200 mM then diluted to 1 mM with culture medium. Corresponding dilutions of DMSO were given to the control group.

#### Animals and Treatments

Adult male C57BL/6 mice, weighing 25 ± 2 g, were supplied by the Experimental Animal Center of Tongji Medical College. All experimental procedures were approved by the Animal Care and Use Committee at the Huazhong University of Science and Technology and were performed in compliance with National Institutes of Health guidelines on the ethical use of animals. The mice were housed five per cage in a room maintained at 22 ± 2 ◦C with an alternating 12-h light–dark cycle. Food and water were available ad libitum.

Mice were divided randomly into four groups: KA-only group (KA), melatonin (Sigma, St. Louis, MO, USA) administration prior to KA group (Mel+KA), melatonin only group (Mel) and vehicle-treated control group (Con). Based on a previous study, mice were treated with an intraperitoneal (i.p.) injection of 30 mg/kg KA (Abcam, Cambridge, MA, USA) emulsified in 0.9% normal saline (Crespo-Biel et al., 2010). Melatonin was dissolved in absolute ethanol and diluted in saline to a final concentration of 2% ethanol before injection.

The mice in the Mel+KA group were given intraperitoneal injections of melatonin 20 mg/kg once, 30 min before the injection of KA on the first day and a single dose per day for a total of 3 days (Jain et al., 2013). As melatonin was dissolved in 2% ethanol, the control group mice received intraperitoneal injections of 2% ethanol at the same time with the same volume (0.1 mL). The mice in each group (n = 12) were euthanized on the 4th day after KA treatment (**Figure 2A**).

All mice were euthanized under anesthesia using 10% chloral hydrate after KA treatment, and the hippocampal tissue was harvested for further tests.

### RNA Isolation and Real-Time Polymerase Chain Reaction Quantification

Total RNA was isolated from the N2a cells by using RNase Mini Kit (Qiagen, Valencia, CA, USA) following the manufacturer's instructions. Primer sequences (Greene et al., 2015) were listed in **Table 1**.

All of the primers were synthesized by Sangon Biotech (Shanghai, China). Real-time polymerase chain reaction (PCR) for cDNA analysis was conducted at 60–95◦C for 45 cycles on a Sequence Detection System (ABI Prism 7000, Applied Biosystems, Darmstadt, Germany) following the manufacturer's instructions and using SYBR Green Reaction Master Mix (TaKaRa, Dalian, China). For each sample, VDAC-1 or GAPDH served as the housekeeping gene. Fold-change expression was calculated from the threshold cycle (Ct) values. For calculation of relative changes, gene expression measured in control tissues was taken as the baseline value.

#### Western Blotting

Total proteins were extracted by using a protein extraction kit (Pierce, IL, USA) in accordance with the manufacturer's instructions. Protein extracts were dissolved in 15% sodium dodecyl sulfate polyacrylamide gel, and then transferred to a nitrocellulose membrane at 150 mA. After being blocked with 5% non-fat skim milk [diluted with Tris-buffered saline containing 0.1% Tween 20 (TBST)] for 1 h at room temperature, the membrane containing the protein extracts was incubated overnight with primary antibody (diluted with 2% bovine serum albumin in TBST) at 4◦C. The following primary antibodies were used: anti-GAPDH (1:3000, Abcam, Santa Cruz, CA, USA); anti-GRP78 (1:2000, Abcam, Cambridge, MA, USA); anti-CHOP (1:500, Abcam, Cambridge, MA, USA); anti-Mfn-1 (1:1000, Abcam, Cambridge, MA, USA), anti-Mfn-2 (1:1000, Abcam, Cambridge, MA, USA), anti-OPA-1 (1:1000, CST, Danvers, MA, USA), anti-Drp-1 (1:1000, Santa Cruz, CA, USA), anti-calpain (1:1000, Santa Cruz, CA, USA); anti-Cyt C (1:1000, Santa Cruz, CA, USA); anti-cleaved caspase-12 (1:1000, CST, Danvers, MA, USA) and anti-cleaved caspase-3 (1:1000, CST, Danvers, MA, USA), anti-cleaved caspase-9 (1:1000, CST, Danvers, MA, USA) and anti-VDAC-1(1:1500, Abcam, Cambridge, MA, USA). On

TABLE 1 | Primer sequences for real-time PCR.


the second day, proteins were visualized using the enhanced chemiluminescence detection system (Pierce, IL, USA) after incubating with respective horseradish peroxidase-conjugated secondary antibodies (1:1000, Amersham Pharmacia Biotech, Buckinghamshire, UK) and then exposed to medical x-ray film. The intensity of the blots was quantified using a gel-image analyzer (JS380; Peiqing Science and Technology, Shanghai, China).

#### Calpain Activity Assay

fnmol-10-00049 February 25, 2017 Time: 15:46 # 4

The calpain activity was measured using Calpain Substrate II Kit (Merck Millipore, Darmstadt, Germany). Cells were washed twice with PBS and lysed on ice with extraction buffer. The total protein extracted (50 µg) was incubated with 150 µL substrate and 1 mL reaction buffer for 100 min at 37◦C. The release of 7-amino-4-methyl-coumarin (AMC) from the reaction was monitored at an emission of 440 nm using a fluorescence spectrometer. Fluorescence units were converted into AMC release using the standard curve. Activity of calpain was expressed and documented in pmol of AMC cleaved per minute per milligram of protein.

#### Time-Lapse Imaging and Mitochondrial Morphology Analyses

To observe the changes in the mitochondrial morphology in N2a cells, the KA and/or melatonin-treated N2a cells were incubated with the MitoTracker Red CMXRos probe (250 nM) (Invitrogen, Carlsbad, CA, USA) for 30 min at 37◦C. After being washed three times in cold PBS, the cells were visualized under a Nikon C2 confocal laser scanning microscope (Nikon, Tokyo, Japan) with excitation of 579 nm and emission greater than 599 nm. For morphological quantification in neurites, z-sections were merged (using maximal projection) and the entire length (from tip to tip) of MitoTracker Red labeled mitochondria of neurites was measured. In cell bodies, mitochondria length was measured in each z-section of the entire soma. Quantification of mitochondria length was performed by using ImageJ software as previously described (De Vos and Sheetz, 2007). The number of mitochondria was counted in control N2a cells (n = 50) and experimental groups (n = 40). Statistical significance was determined using one-way ANOVA analysis.

### JC-1 Dye Membrane Potential Staining

Mitochondrial membrane potential in N2a cells was measured by using MMP assay kit with JC-1 (Beyotime Biotechnology, Shanghai, China). All procedures followed the manufacture's instructions. The fluorescence intensities were measured using a fluorescence plate reader at 590 nm (red) and 529 nm (green). According to the ratio of fluorescence intensities at 590 and 529 nm, the loss of MMP was assessed and recorded.

#### TUNEL Assay

Apoptosis was assessed by detecting DNA fragmentation through TUNEL staining. TUNEL staining for apoptotic cells were performed in accordance the manufacturer's instructions (Roche Corporation, Germany). The N2a cells growing on the cover glass slide were fixed. Endogenous peroxidase activity was inhibited by immersing the sections in methanol with 0.3% H2O<sup>2</sup> for 20 min at room temperature. After washing with distilled water and 0.01 M PBS, the slides were soaked in TdT (terminal deoxynucleotidyl transferase) buffer at room temperature for 15 min and then incubated with TdT buffer containing 5 U TdT enzyme and 0.5 nmol biotinylated 16-dUTP in a humidified chamber for 1 h at 37◦C. The reaction was terminated by adding 2× sodium saline citrate (SSC). The sections were then incubated with HRP-conjugated streptavidin according to the manufacturer's instructions. Colorimetric development of the reaction was done by incubation in DAB (diaminobenzidine) solution for 5 min. For the negative control slides, either TdT enzyme or biotinylated 16-dUTP was omitted in the labeling reaction. The specimens in all groups (n = 3) were observed using an upright microscope (OLYMPUS BX53F, Japan) and in at least six random fields of high-power (×100).

#### ROS Determination

N2a cells were labeled with DCFH-DA (Sigma, St. Louis, MO, USA) (10 µM) for 30 min after KA treatment (Liu et al., 2015). ROS generation was indicated by green florescence and visualized using a fluorescence microscope (Olympus, Japan). The fluorescence in each group was assessed by flow cytometry with an excitation wavelength of 488 nm and an emission wavelength of 525 nm.

#### Transmission Electron Microscope

To determine the effects of KA and melatonin on the morphology of mitochondria, we conducted transmission electron microscopy (TEM) of N2a cells from control and experimental treatments (n = 4). N2a cells were fixed in 2.5% glutaraldehyde and post-fixed in 2% OsO<sup>4</sup> at room temperature. Cellular staining was performed at 4◦C for 2 h in 2% uranyl acetate in the dark. Samples were rinsed in sodium phosphate buffer (0.1 M, pH 7.2), dehydrated in ethanol and infiltrated overnight in Araldite. Following polymerization, embedded samples were detached from the chamber slide and glued to Araldite blocks. Serial semi-thin (1.5 mm) sections were mounted onto slides and stained with 1% toluidine blue. The selected semi-thin sections were glued (Super Glue, Loctite) to araldite blocks and detached from the glass slide. Ultrathin (0.07 mm) sections were prepared and stained with lead citrate. Finally, photomicrographs were obtained under a TEM using a digital camera (Hitachi, Japan).

## Intracellular Ca2<sup>+</sup> Measurement

Intracellular Ca2<sup>+</sup> ([Ca2+]i) was measured as previously described (Blatter and Wier, 1990). To measure the acute effect of KA on [Ca2+]i change, N2a cells were grown on cover slides and washed three times with 2 µM Fura-2 acetoxymethyl ester (Fura-2 AM) in Hanks Balanced Salt Solution (HBSS, containing 145 mM NaCl, 5 mM KCl, 1 mM MgCl2, 2 mM CaCl2, 0.75 mM Na2HPO4, 10 mM glucose, and 10 mM HEPES, pH 7.4). N2a cells were then incubated in the same solution for 30 min at 37◦C.

In calcium-free experiments, EGTA (100 µM) was substituted by CaCl2.

Before each experiment, the cover slides were mounted on a chamber positioned on the movable stage of an inverted Olympus IX-70 microscope equipped with a calcium imaging system (TILL Photonics Gmbh, Germany) and super fused by HBSS for 10 min. Fura-2 AM loaded cells were illuminated at 340 nm for 150 ms and 380 nm for 50 ms at 1 s intervals using a TILL polychrome monochromator. Fura-2 fluorescence emission was imaged at 510 nm by a cooled-CCD (TILL Photonics Gmbh, Germany) through a X-70 Fluor oil immersion lens (Olympus, Tokyo, Japan) and a 460 nm long pass barrier filter. F340/F380 fluorescence ratios were generated by Olympus software. Paired F340/F380 fluorescence ratio images were acquired every 3 s for [Ca2+]<sup>i</sup> . In the experiment, the ratio between F380 of Ca2<sup>+</sup> free solutions and F380 of high calcium medium was recorded as coefficient factor β. Intracellular Ca2<sup>+</sup> concentration ([Ca2+]i) = Kd × β × [(Ratio-Rmin)/(Rmax-Ratio)], a constant Kd is the dissociation constant of Fura-2 AM for Ca2<sup>+</sup> and was assumed to be 224 at 37◦C. Rmax is the ratio obtained after ionomycin treatment. Rmin is the ratio obtained from Ca2<sup>+</sup> free solution.

Data was obtained by evaluating the fluorescence immediately after KA stimulation with subtraction of background fluorescence and division by the fluorescence intensity before KA stimulation. For the melatonin treatment groups, melatonin (50 µM) was added to dishes 1 h before KA stimulation.

#### ATP Measurement

The concentration of ATP was determined by using ATP determination kit (Thermo Fisher Scientific, Inc., Waltham, MA, USA). All procedures followed the manufacture's instructions. Samples were incubated with ATP reaction mixture for 30 min and detected at 560 nm. A standard curve was established to calculate the concentration of samples.

#### Neuronal Loss Assay

Hippocampal tissues were embedded and sectioned at 10 µm, with three independent samples for each group. After staining with 1% toluidine blue for 15 min, the sections were dehydrated and mounted for microscopic examination. CA3 areas of the hippocampus were examined with an Olympus AX-70 microscope equipped with a motorized stage (Olympus, Astoria, NY, USA). Images were captured at 20× magnification in every section. For each independent sample, surviving neurons exhibiting normal morphology with positive blue staining Nissl bodies were included in counts.

#### Statistical Analysis

Data was expressed as mean ± standard deviation and analyzed using SPSS 10.0 statistical software (SPSS, Inc., Chicago, IL, USA). The one-way analysis of variance followed by least significant difference post hoc test was used to determine the significance of differences among groups (P < 0.05, P < 0.01, and P < 0.001).

### RESULTS

### Melatonin Remarkably Mitigates KA-Induced Apoptosis

Kainic acid is an analog of glutamate with the capability of associating with specific KA-type non-NMDA receptors. KA has been reported to induce status epilepticus, neurodegeneration and memory loss (Sperk et al., 1983). To validate the neuronal toxicity of KA, we treated N2a cells with KA at a series of concentrations (0, 25, 50, 100 µM). In MTT and crystal violet assays, KA decreased cell metabolism and cell viability in a dose-dependent manner (**Figures 1A–D**). KA application elicited the activation of caspases and the release of cytochrome C (Cyt C), which are the characteristic phenotypes of apoptosis (**Figures 1I,J**). In the assays, we noticed that application of 50 or 100 µM of KA induced significant apoptosis and toxicity. There were no detectable differences between these doses on cell viability and the release of LDH (**Figures 1E,F**).

To determine whether melatonin could inhibit the apoptosis of N2a cells, we added different concentrations (50 and 100 µM) of melatonin into cell cultures. As shown in **Figures 1I,J**, melatonin addition greatly decreased the expression of c-caspase-3 (cleaved caspase-3), c-caspase-12, c-caspase-9, and Cyt C. The inhibitory effect of melatonin on cell apoptosis was also assessed by TUNEL assay (**Figures 1G,H**). We observed that 50 µM melatonin was enough to regain cell viability and reduce LDH release effectively (**Figures 1C,D,F**). The cell viability and LDH release results between application of 50 and 100 µM melatonin were not significantly different.

To determine whether KA affects neuronal viability in vivo, we analyzed neuronal loss in KA and/or melatonin treated C57BL/6 mice. Adult male C57BL/6 mice were treated with an intraperitoneal (i.p.) injection of 30 mg/kg KA. The mice in the Mel+KA group were given intraperitoneal injections of 20 mg/kg melatonin 30 min before KA injection, and an additional dose on day 2 and day 3 (**Figure 2A**). Hippocampal tissues were stained with toluidine blue. We examined CA3 areas of the hippocampus and counted the surviving neurons with blue Nissl bodies (**Figure 2B**). Melatonin application greatly increased the density of surviving neurons (**Figure 2C**), indicating that melatonin functions autonomously in the hippocampus to promote the survival of neurons. Next, we assayed the expression of c-caspase-12, -9, -3 in hippocampal tissue and primary neurons. Similar to the N2a cells results, quantitative analysis of expression levels of c-caspase-12, -9, -3 in both hippocampal tissue and primary neurons showed that melatonin significantly reduced the levels of c-caspase-3, c-caspase-12, and c-caspase-9 in vivo (**Figures 2D,E**).

### Melatonin Ameliorates KA-Induced Mitochondrial Fragmentation and Dysfunction

Mitochondria defects are closely associated with apoptosis (Chen et al., 2007). To examine the integrity of mitochondria in KA-treated N2a cells, we first used TEM to assay the

ultrastructure of mitochondria in N2a cells (**Figure 3A**). We observed that mitochondria in the KA-treated group presented swelling, crest fracture, and disappearance (**Figure 3A**). Next, we also assessed the mitochondrial morphology changes. We found that the mitochondria displayed canonical tubular patterns under physiological conditions (control group), while KA exposed mitochondria turned into puncta-like fragments (**Figure 3B**). The average length of mitochondria was shorter

after KA treatment compared to that of the control group mitochondria (**Figure 3D**). These observations indicate the occurrence of mitochondrial fragmentation. Also, we assayed the morphology of mitochondria in KA and/or melatonintreated rat primary neurons. Consistently, melatonin ameliorates KA-induced mitochondrial fragmentation. The number and the average length of mitochondria in both KA and melatonintreated group were significantly restored compared to that of the mitochondria in the KA-only group (**Supplementary Figures S1A–C**).

In addition to the morphological impairments, we also observed a decrease in the average number of mitochondria per cell in the KA-treated group (**Figure 3C**). To examine if KA affects mitochondrial functions, we assayed the levels of ROS, ATP and the MMP in N2a cells. As shown in **Figures 3E–G**, KA treatment caused ROS accumulation, MMP collapse and a decrease in ATP production, indicating the loss of mitochondrial functions.

In an effort to characterize the protective effects of melatonin on mitochondria, we examined the subcellular pattern of mitochondria in both KA and melatonin-treated groups. Melatonin application significantly ameliorated KA-induced mitochondria morphological changes and ROS level spikes. In addition, melatonin restored ATP production and MMP in both KA and melatonin-treated groups. These observations indicate that melatonin is an effective inhibitor of KA-induced mitochondrial fragmentation and dysfunction.

### Melatonin Relieves KA-Induced Mfn-2 Degradation

As highly dynamic organelles, mitochondria undergo continuous fission and fusion events (Nicholls, 2004). Unopposed fission

FIGURE 4 | Melatonin relieves KA-induced Mfn-2 degradation. (A) Expression of OPA-1, Drp-1, Mfn-1, and Mfn-2 in KA and/or melatonin-treated N2a cells. (B) Relative protein level of OPA-1, Drp-1, Mfn-1, and Mfn-2 in KA and/or melatonin-treated N2a cells. (C) Relative mRNA level of OPA1, Drp-1, Mfn-1, and Mfn-2 in KA and/or melatonin-treated N2a cells (∗∗P < 0.01 vs. controls; ##P < 0.01 vs. the KA group; significant difference from the respective values determined by one-way analysis of variance test. n = 3).

leads to fragmentation, whereas unopposed fusion results in abnormal elongation, both of which would greatly impair mitochondrial functions (Johri and Beal, 2012). Mitochondrial fission and fusion are tightly regulated by mitochondrial fission and fusion proteins. Drp-1 is required for fission (Shen et al., 2014), while Mfn-1, Mfn-2, and OPA-1 are critical for fusion (Chen and Chan, 2009). In addition to morphology maintenance, fission and fusion processes are essential for various aspects of mitochondrial function (Ishihara et al., 2006). The disruption of mitochondrial dynamics causes its morphological and functional impairments (Ishihara et al., 2006).

Our evidence suggests that mitochondrial morphology and functions can be affected by KA treatment. To determine whether KA perturbs mitochondrial dynamic regulators' homeostasis, we examined the expression levels of proteins involved in mitochondrial fusion and fission. The protein level of Mfn-2 decreased significantly after KA treatment, while the levels of Drp-1, OPA-1, and Mfn-1 remained unchanged (**Figures 4A,B**). Next, we examined the mRNA levels of these factors. Surprisingly, the mRNA levels of all proteins remained unaffected (**Figure 4C**). As suggested by previous study (Wang et al., 2015), KA probably decreased the protein levels of Mfn-2 by prompting protein degradation but not by blocking expression. Furthermore, application of melatonin reversed the protein levels of Mfn-2 in N2a cells, which is consistent with the protective effects of melatonin on mitochondrial morphology and functions (**Figures 4A,B**). Therefore, we speculate that the inhibitory role of melatonin on mitochondrial dysfunction could be the alleviation of mitochondrial dynamic defects.

#### Melatonin Alleviates KA-Induced Increased Ca2<sup>+</sup> Levels and Calpain Activation

Ca2<sup>+</sup> homeostasis disorder was observed in a variety of diseases (Wrogemann and Pena, 1976). Upon overstimulation of NMDA receptors, Ca2<sup>+</sup> accumulates abnormally and impairs mitochondrial function, leading to a decrease in ATP production and an increased release of ROS (Halestrap, 2009). To try to understand the molecular mechanism underlying mitochondrial defect relief by melatonin treatment, we assayed intracellular Ca2<sup>+</sup> concentration and calpain activity. As shown

difference from the respective values determined by one-way analysis of variance test. n = 3).

in **Figures 5A,B**, after KA treatment, intracellular Ca2<sup>+</sup> was remarkably increased compared to the control group (∼6 fold increase). In contrast, melatonin treatment effectively decreased the levels of intracellular Ca2<sup>+</sup> in KA and melatonin-treated N2a cells (**Figures 5A,B** and **Supplementary Figures S2A,B**). Due to the Ca2<sup>+</sup> influx, the activity of calpain increased ∼3 fold in the KA-treated group. Melatonin treatment significantly decreased the abnormally elevated activity of calpain (**Figure 5C**). Together with the inhibitory effect of melatonin on Ca2<sup>+</sup> influx, our results suggest that melatonin could alleviate KA induced mitochondrial dysfunction and apoptosis through attenuating increased Ca2<sup>+</sup> levels and calpain activation.

#### Blocking KA-Induced Ca2<sup>+</sup> Increase and Calpain Activation Can Effectively Relieve Mitochondrial Defects

Our evidence suggests that the inhibitory effects on Ca2<sup>+</sup> overloading and calpain activation by melatonin could account

for the alleviation of KA-induced mitochondrial dysfunction. To determine if the mitochondrial impairments were due to Ca2<sup>+</sup> overloading and calpain activation, we used calpeptin and BAPTA-AM to inhibit calpain activity and Ca2<sup>+</sup> overloading, then examined the levels of Mfn-2 and caspase-3. We observed that both calpeptin and BAPTA-AM fully reversed the decrease in Mfn-2 and increase in cleaved caspase-3 in KA-treated cells (**Figures 6A–D**). Furthermore, applications of either calpeptin or BAPTA-AM led to significant improvements in mitochondrial functions and morphology (**Figures 6E–N**). In addition, calpeptin or BAPTA-AM treatments decreased ROS production, restored ATP and MMP levels, and improved the number and morphology of mitochondria. Together, the data indicates that the KA-induced mitochondrial defects were due to increased

Ca2<sup>+</sup> levels and calpain activation. Melatonin treatment can greatly restore mitochondrial dynamics by inhibiting aberrant Ca2<sup>+</sup> levels and calpain activation.

#### KA-Induced ER Stress Contributes to Mitochondrial Defects and Apoptosis and Melatonin Can Effectively Suppress ER Stress

Upon activation by intracellular Ca2<sup>+</sup> overloading, calpain participates in a series of physiological and pathological processes, including ER stress (Wu and Lynch, 2006). Specifically, calpain activates the expression of GRP78, CHOP, and eventually triggers apoptosis regulators including caspase-12 and caspase-3 (Chakraborti et al., 2012). After KA treatment, the levels of calpain, GRP78, CHOP and c-caspase-12 significantly increased (**Figures 7A–D**). These observations suggest that the apoptosis after KA treatment could be partially due to ER stress. Further analyses show that melatonin or PBA application effectively downregulated the increased levels of Ca2<sup>+</sup> and ER stress related proteins in the KA group (**Figures 7A–F**).

Thus far, our results suggest that melatonin could inhibit apoptosis through regulating ER stress as well. To determine if ER stress also participates in the modulation of mitochondrial dynamics, we examined the levels of Mfn-2 and caspase-3, ROS production, MMP, ATP production and mitochondria morphology in N2a cells (**Figures 7G–M**). PBA administration greatly alleviated the KA-induced degradation of Mfn-2, indicating that inhibiting ER stress contributes to the protective effects of melatonin on mitochondrial dynamics (**Figures 7G,H**). Also, the decreased ROS levels and improved ATP production/MMP/mitochondria morphology suggest that KA-induced ER stress contributes to mitochondrial dysfunction and apoptosis, and that melatonin can improve mitochondrial functions via ER stress inhibition (**Figures 7I–M**).

To examine the ER stress alleviation effects of melatonin in vivo, we assayed the expression levels of GRP78, CHOP and calpain in KA and/or melatonin-treated C57BL/6 mice. Consistent with the results in N2a cells, KA induced GRP78, CHOP and calpain overexpression were effectively reduced in vivo (**Figures 8A,B**). Increased calpain activity was also suppressed after melatonin application (**Figure 8C**).

### DISCUSSION

Previous studies indicated that excitotoxicity contributes to the neurodegenerative processes (Dong et al., 2009). Glutamate and related excitatory amino acids can induce neuronal apoptosis when administered both in vivo and in vitro (Reynolds and Hastings, 1995; Ding et al., 2015). In the present study, KA caused neuronal apoptosis and cytotoxicity in a dose-dependent manner (**Figure 1**). After KA treatment, we observed the release of Cyt C along with the activation of caspases, which act as terminal executors in the apoptosis pathway (Cohen, 1997). We observed the activation of caspase-12 and caspase-9, which participate in ER stress-mediated and mitochondria-mediated apoptosis pathways, respectively (**Figures 1**, **2**) (Garcia de la Cadena and Massieu, 2016). These results suggest that ER stress and mitochondrial damages are both responsible for KA-induced neuronal apoptosis (**Figure 9**).

Calcium homeostasis is central to various cellular functions (Berridge et al., 2000). Calcium overload will impair cellular health, resulting in massive activation of proteases, phospholipases and cell death (Pinton et al., 2008). It was reported that the excessive accumulation of cytosolic Ca2<sup>+</sup> can affect the scavenging roles of mitochondria and might lead to mitochondrial stress, including increases in mitochondrial ROS production and the collapse of mitochondrial membrane permeability (Mallilankaraman et al., 2012). ER is well-known to effectively buffer the cytosolic calcium concentration, mitochondria also participates in maintaining the cellular Ca2<sup>+</sup> homeostasis (Pinton et al., 2008). Calcium transport from the ER to mitochondria plays a significant role in regulating cellular bioenergetics, ROS production and apoptosis induction. Mitochondria-associated ER-membranes (MAMs) may play a

special role in the ER-mitochondria crosstalk (Marchi et al., 2014).

Recent report indicated that, in neurons, acute exposure to glutamate causes Parkin translocation to mitochondria in a calcium- and NMDA receptor-dependent manner. Parkin accumulates on MAMs following excitotoxicity, supporting a role of Parkin in ER-mitochondria crosstalk in mitochondrial homeostasis (Van Laar et al., 2015). Some MAMs proteins have been involved in mitochondrial dynamic fusion and fission including the mitofusin Mfn-2. Mfn-2 is enriched at the MAMs and its absence affects mitochondrial morphology and function (de Brito and Scorrano, 2008). Additionally, ER-located Mfn-2 is required for the connection with mitochondria by interacting directly with Mfn-1 or Mfn-2 on the mitochondria. The decrease in Mfn-2 could decrease Ca2<sup>+</sup> traffic to the mitochondria (Guo et al., 2007). A very recent study showed that calpain activation in response to glutamate could result in post-translational protein degradation of Mfn-2, which is a novel mechanism regulating mitochondrial fusion during glutamate excitotoxicity (Wang et al., 2015). Similarly, we observed that Mfn-2 protein levels decreased while its mRNA level remained unchanged after KA treatment (**Figure 4**), suggesting that NMDA receptor activation by various ligands will result in the post-translational degradation of Mfn-2 and mitochondrial disorders in a like manner.

Kainic acid causes Ca2<sup>+</sup> influx in N2a cells (**Figure 5**), which leads to activation of calpain and progressive neuronal death through mitochondria functional collapse (Harraz et al., 2012; Celso Constantino et al., 2014). Calpain belongs to the family of calcium-dependent non-lysosomal cysteine proteases, which is known to be involved in cytoskeleton and membrane attachments, signal transduction pathways, and apoptosis (McConkey and Orrenius, 1994; Chakraborti et al., 2012). In this study, the activity of calpain increased in response to KAinduced Ca2<sup>+</sup> accumulation. To assay whether Ca2<sup>+</sup> causes mitochondrial disorder through calpain activation, we used BAPTA-AM and calpeptin to inhibit the Ca2<sup>+</sup> overloading and calpain activation. Both BAPTA-AM and calpeptin successfully inhibited the degradation of Mfn-2 and restored the normal amount and morphology of the mitochondria (**Figure 6**). These data further demonstrated that calpain activation by Ca2<sup>+</sup> influx results in the post-translational degradation of Mfn-2 and mitochondrial impairments. Our study also suggests that ER stress plays a role in the mitochondrial disorder. ER stress inhibitor PBA decreased the expression of calpain and Ca2<sup>+</sup> concentration. PBA also alleviated the degradation of Mfn-2 and inhibited the activation of caspase-3, restoring the function and morphology of mitochondria (**Figure 7**). These results reveal that mitochondrial dynamic impairment could be partially attributed to ER stress. All these observations suggest that the mitochondrial dynamic disorder was due to the aberrant degradation of Mfn-2 by calpain upregulation and activation, and that mitochondrial dysfunction could be the result of a defect in mitochondrial fusion (**Figure 3**).

Our study also suggested that ER stress also participates in KAinduced apoptosis directly. The activation of caspase-12 indicates the activation of the ER stress pathway (**Figures 1**, **9**, caspase-12 pathway on the right side). As a proapoptotic unfolded protein response (UPR) factor, CHOP could also be activated by persistent ER stress (Malhi and Kaufman, 2011). KA treatment enhances the expression of GRP78 and CHOP, which could be activated by the UPR (Zheng et al., 2015). The simultaneous increases in expression of GRP78, CHOP and caspase-12 in the KA-treated group suggested that ER stress is involved in apoptosis directly (**Figure 7**).

Melatonin is a naturally occurring molecule with antioxidant properties (Hardeland, 2005; Reiter et al., 2009; Galano et al., 2011, 2013). Melatonin and its metabolites have the ability to scavenge ROS and reactive nitrogen species (RNS) (Galano et al., 2013). As a broad spectrum antioxidant (Zhang and Zhang, 2014; Manchester et al., 2015), melatonin has pleiotropic effects as well as neuroprotective properties (Negi et al., 2011; Miller et al., 2014). Melatonin acts against the elevation of lipid peroxidation induced by

either KA or NMDA. Melatonin also blocked both KA- and NMDA-receptor mediated neuronal damage (Kim and Kwon, 1999). In addition, melatonin attenuates morphine-induced NMDA receptor subtype NR1 expression and decreases calcium concentration via modulating PKCγ activities in the spinal cord (Song et al., 2015). Accumulating evidence emphasizes contributions of melatonin toward the maintenance of ER and mitochondrial homeostasis, which made it an increasingly interesting pharmacological agent against neurodegenerative diseases (Hu et al., 2016; Uguz et al., 2016).

At the subcellular level, melatonin was found to attenuate methamphetamine-induced translocation of mitochondrial fission proteins, cytosolic calcium overload and cell death in SH-SY5Y cells (Parameyong et al., 2015). Melatonin can also suppress methamphetamine-triggered ER stress in C6 cells (Tungkum et al., 2017).

In our study, melatonin can effectively inhibit KA-induced calpain upregulation/activation and mitochondrial deterioration by alleviating Ca2<sup>+</sup> overload and ER stress. Specifically, melatonin reduced the expression of c-caspase-9 and c-caspase-12 after KA treatment (**Figures 1F,G**), indicating the inhibitory effect on both mitochondrial dysfunction and ER stress related apoptosis pathways. Taken together, our study supported the hypothesis that melatonin has beneficial roles in countering neuronal death through blocking Ca2<sup>+</sup> overload and ER stress.

#### AUTHOR CONTRIBUTIONS

FX and CS carried out cell culture, biochemical measurements and drafted the manuscript. They contributed equally to the work. WH carried out confocal imaging and the analysis of mitochondria dynamics. LX carried out ATP production assay. YW carried out calpain activity assay. ST participated in the preparation of manuscript. JZ participated in the measurement of LDH and cell viability. QC performed the statistical analysis. AS participated in the design and preparation of manuscript. JC conceived the study and participated in the coordination and

#### REFERENCES


preparation of manuscript. All authors read and approved the final draft.

#### FUNDING

This work was supported by the National Natural Science Foundation of China (31171027, 81371416, 31670778), the Fundamental Research Funds for the Central Universities (HUST: 2016YXMS191, 2014YGYL005) to JC, the National Natural Science Foundation of China (81371418, 31571466), the Specialized Research Fund for the Doctoral Program of Higher Education (20130142110071), the Program for New Century Excellent Talents in University (NCET-13-0234), and the Junior Thousand Talents Program of China to AS.

#### ACKNOWLEDGMENT

We thank Dong Li, Zhenrong Yang, and Xin Fu for the technical assistance.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2017.00049/full#supplementary-material

#### FIGURE S1 | Melatonin ameliorates KA induced mitochondrial

fragmentation in rat primary neurons. (A) The morphology of mitochondria stained by Mito-tracker in KA and/or melatonin treated neurons. (B) Numbers of mitochondria in KA and/or melatonin treated neurons. (C) Average length of mitochondria in KA and/or melatonin treated neurons. <sup>∗</sup>P < 0.05, ∗∗P < 0.01 vs. controls; #P < 0.05, ##P < 0.01 vs. the KA group; significant difference from the respective values determined by one-way analysis of variance test, n = 3.

FIGURE S2 | Melatonin and PBA ameliorates KA induced Ca2<sup>+</sup> elevation.

(A) Fura-2 AM probe was used to determine real-time ratio of F340/F380 in KA and/or melatonin-treated N2a cells. (B) Fura-2 AM probe was used to measure real-time ratio of F340/F380 in KA and/or PBA-treated N2a cells.

pre- and postconditioning. Aging Dis. 5, 430–441. doi: 10.14336/ad.2014. 0500430


infantile encephalopathy. Hum. Mutat. 37, 898–903. doi: 10.1002/humu. 23033


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Xue, Shi, Chen, Hang, Xia, Wu, Tao, Zhou, Shi and Chen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Inhibitory Effect of α/β-Hydrolase Domain-Containing 6 (ABHD6) on the Surface Targeting of GluA2- and GluA3-Containing AMPA Receptors

Mengping Wei1,2† , Moye Jia1,2† , Jian Zhang1,2† , Lulu Yu1,2, Yunzhi Zhao1,2, Yingqi Chen1,2 , Yimeng Ma1,2, Wei Zhang<sup>3</sup> , Yun S. Shi<sup>4</sup> and Chen Zhang1,2 \*

<sup>1</sup> State Key Laboratory of Membrane Biology, School of Life Sciences, Peking University, Beijing, China, <sup>2</sup> PKU-IDG (International Digital Group)/McGovern Institute for Brain Research, Peking University, Beijing, China, <sup>3</sup> Department of Pharmacology, Institute of Chinese Integrative Medicine, Hebei Medical University, Shijiazhuang, China, <sup>4</sup> Ministry of Education (MOE) Key Laboratory of Model Animal for Disease Study, Model Animal Research Center of Nanjing University, Nanjing, China

#### Edited by:

Cong Ma, Huazhong University of Science and Technology, China

#### Reviewed by:

Guiquan Chen, Nanjing University, China Junyu Xu, Zhejiang University, China

\*Correspondence: Chen Zhang ch.zhang@pku.edu.cn †These authors have contributed equally to this work.

Received: 16 December 2016 Accepted: 17 February 2017 Published: 02 March 2017

#### Citation:

Wei M, Jia M, Zhang J, Yu L, Zhao Y, Chen Y, Ma Y, Zhang W, Shi YS and Zhang C (2017) The Inhibitory Effect of α/β-Hydrolase Domain-Containing 6 (ABHD6) on the Surface Targeting of GluA2 and GluA3-Containing AMPA Receptors. Front. Mol. Neurosci. 10:55. doi: 10.3389/fnmol.2017.00055 The α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA)-type glutamate receptors (AMPARs) are major excitatory receptors that mediate fast neurotransmission in the mammalian brain. The surface expression of functional AMPARs is crucial for synaptic transmission and plasticity. AMPAR auxiliary subunits control the biosynthesis, membrane trafficking, and synaptic targeting of AMPARs. Our previous report showed that α/β-hydrolase domain-containing 6 (ABHD6), an auxiliary subunit for AMPARs, suppresses the membrane delivery and function of GluA1-containing receptors in both heterologous cells and neurons. However, it remained unclear whether ABHD6 affects the membrane trafficking of glutamate receptor subunits, GluA2 and GluA3. Here, we examine the effects of ABHD6 overexpression in HEK293T cells expressing GluA1, GluA2, GluA3, and stargazin, either alone or in combination. The results show that ABHD6 suppresses the glutamate-induced currents and the membrane expression of AMPARs when expressing GluA2 or GluA3 in the HEK293T cells. We generated a series of GluA2 and GluA3 C-terminal deletion constructs and confirm that the C-terminus of GluAs is required for ABHD6's inhibitory effects on glutamate-induced currents and surface expression of GluAs. Meanwhile, our pull-down experiments reveal that ABHD6 binds to GluA1–3, and deletion of the C-terminal domain of GluAs abolishes this binding. These findings demonstrate that ABHD6 inhibits the AMPAR-mediated currents and its surface expression, independent of the type of AMPAR subunits, and this inhibitor's effects are mediated through the binding with the GluAs C-terminal regions.

Keywords: AMPA receptor, ABHD6, receptor trafficking, glutamates, protein–protein interactions

### INTRODUCTION

In the mammalian brain, α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA)-type glutamate receptors (AMPARs) are major ionotropic receptors that mediate the majority of fast excitatory synaptic transmissions. The binding of the glutamate released from presynaptic terminals with postsynaptic AMPARs determines the efficiency and plasticity of synaptic

transmission between pairs of neurons. The numbers and biophysical properties of AMPARs remain dynamically modulated during periods of rest and plasticity, and deficits in these processes are strongly linked to psychiatric and neurodegenerative diseases, including Huntington's disease and Alzheimer's disease (Keinänen et al., 1990; Jackson and Nicoll, 2011; Huganir and Nicoll, 2013; Henley and Wilkinson, 2016).

AMPARs are tetramers assembled from four AMPAR subunits: GluA1–4 (Hollmann and Heinemann, 1994). The subunit composition determines the biophysical and molecular properties of AMPARs. The C-terminus of AMPARs interacts with proteins, such as PICK1, NSF, GRIP1, SAP97, and NSF, to traffic the receptors to the plasma membrane and the synapse in both constitute and activity-dependent manners (see reviews Ziff, 2007; Anggono and Huganir, 2012; Cheng et al., 2012; Yang et al., 2014). For example, PICK1 causes a selective decrease in the surface expression levels of GluA2, while the surface expression levels of GluA1 remain unchanged, suggesting that PICK1 regulates the subunit composition and synaptic targeting of surface AMPARs (Terashima et al., 2004; Cao et al., 2007; Xu et al., 2016). GluA2 confers the calcium permeability to the AMPARs; while GluA2-containing receptors are not calcium permeable, GluA1-containing receptors are calcium permeable. All AMPAR subunits are subject to alternative splicing to generate either "flop" or "flip" versions of the receptors, while only GluA2 undergoes Q/R RNA editing (Burnashev et al., 1992).

The gating, pharmacology, trafficking, and localization of AMPARs are regulated by not only the subunit composition of GluA subunits, but also the AMPARs auxiliary subunits proteins (Bredt and Nicholl, 2003; Li et al., 2016). Stargazin was identified as the first member in the family of transmembrane AMPAR regulatory proteins (TARPs; Letts et al., 1998; Chen et al., 1999, 2000; Hashimoto et al., 1999), and ever since, more and more AMPAR-associated proteins are identified, mainly using proteomics approaches. Native AMPARs associate with a variety of regulatory proteins, including TARPs, cornichon-2 and -3, cystine-knot AMPAR modulating protein of 44 kDa (CKAMP44), germ cell-specific gene 1-like protein (GSG1L), α/β-hydrolase domain-containing 6 (ABHD6), porcupine (PORCN), etc. (Vandenberghe et al., 2005; Tomita et al., 2006; Milstein and Nicoll, 2008; Coombs et al., 2012; Schwenk et al., 2012; Gu et al., 2016b; Wei et al., 2016). The functional analysis of these proteins reveals the AMPARs' important and distinct roles in both neurons and heterologous cells. For example, type I TARPs (stargazin, γ-3, γ-4, γ-8) are necessary and sufficient for the delivery of AMPARs to the plasma membrane in cerebellum granule cells. Furthermore, type I TARPs also modulate the properties of the AMPAR channels by reducing desensitization and slowing the deactivation of the AMPARs (Priel et al., 2005; Tomita et al., 2006). GSG1L and ABHD6 suppress AMPAR-mediated synaptic transmission and modulate its kinetics in hippocampal neurons (Gu et al., 2016a; Wei et al., 2016). Cornichon-2 and -3 conditional knock-out mice showed selective reduction of surface GluA1 containing subunits, and impaired strength and kinetics of AMPAR-mediated synaptic transmission. This is likely due to the effect of TARPγ-8, which can mediate the functional interaction between CNIHs and AMPARs, thus promoting the association of CNIHs with the GluA1 subunit and preventing the association of CNIHs with other subunits (Herring et al., 2013).

We previously showed that ABHD6, a monoacylglycerol lipase, can bind to the C-terminus of GluA1. Overexpression of ABHD6 can reduce AMPAR-mediated excitatory neurotransmission in neurons and glutamate-induced currents in HEK293T cells in a 2-arachidonoylglycerol independent manner. Further studies via immunostaining demonstrated that these decreases might have been due to the decreased surface expression levels of GluA1, rather than the overall expression levels (Wei et al., 2016). Thus, ABHD6 seems to functionally interact with GluA1 in both heterologous cells and cultured hippocampal neurons. However, whether the inhibitory effect of ABHD6 on AMPAR function depends on the type of AMPAR subunit remains uncertain. In the present study, we investigated the subunit specificity of ABHD6's inhibition on AMPARs in transfected HEK293T cells.

#### MATERIALS AND METHODS

#### Construction of Expression Vectors

Rat GluA1, GluA2, and GluA3 subunit cDNAs containing internal ribosome entry site linked green fluorescent protein (IRES-GFP) were used in the present study and have been described previously (Shi et al., 2009). ABHD6-2A-GFP was cloned into a pFUGW expression vector using polymerase chain reaction (PCR) methods (Wei et al., 2016). GluA1, GluA2, and GluA3 deletion constructs were generated by PCR. GluA1 deletion 14 (A1D14) ended in SKRMK; GluA2-deletion 1 (A2D1) ended in EGYNV; GluA2-deletion 2 (A2D2) ended in QNSQN; GluA2-deletion 3 (A2D3) ended in SQNSQ; GluA2 deletion 4 (A2D4) ended in SSQNS; GluA2-deletion 5 (A2D5) ended in SSSQN; GluA2-deletion 6 (A2D6) ended in PSSSQ; GluA2-deletion 7 (A2D7) ended in NPSSS; GluA2-deletion 8 (A2D8) ended in KNPQN; GluA2-deletion 9 (A2D9) ended in RMKVA; GluA2-deletion 10 (A2D10) ended in KRMKV; GluA2-deletion 11 (A2D11) ended in AKRMK; GluA2-deletion 12 (A2D12) ended in EAKRM; GluA3-deletion 1 (A3D1) ended in NTQNY; GluA3-deletion 2 (A3D2) ended in KPAPA; GluA3-deletion 3 (A3D3) ended in FKPAP; GluA3-deletion 4 (A3D4) ended in NFKPA; GluA3-deletion 5 (A3D5) ended in KNTQN; GluA3-deletion 6 (A3D6) ended in RMKLT; GluA3-deletion 7 (A3D7) ended in KRMKL; and GluA3 deletion 8 (A3D8) ended in SKRMK. All the C-terminal deletion constructs were tagged with an human influenza hemagglutinin (HA) tag with a linker of GQG (**Figure 3A**). GluA11ATD lacked sequence from ANFPN to DDKFV, GluA21ATD lacked sequence from VSSNS to VDKMV, and GluA31ATD lacked sequence from GFPNT to YERFV. All the N-terminal deletion constructs were tagged with a Flag tag (**Figure 4A**). Myc-ABHD6 was cloned into a pCAG vector using PCR methods. The constructs were verified with Sanger sequencing.

#### FIGURE 1 | Continued

amplitudes and plateaus (right) of 10 mM glutamate-induced currents in HEK293T cells transfected with combinations of GluA3 and stargazin (control: n = 27/3; ABHD6: n = 27/3; peak: p < 0.01; plateau: p < 0.01). (C) Representative traces (left) and summary graphs of the peak amplitudes and plateaus (right) of 10 mM glutamate-induced currents in HEK293T cells transfected with combinations of GluA2, GluA3, and stargazin (control: n = 24/3; ABHD6: n = 24/3; peak: p < 0.001; plateau: p < 0.01). All summary graphs show means ± SEMs; statistical comparisons by Student's t-test yielded: <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

#### HEK293T Cell Culture and Transfection

HEK293T cells (KCB Cat# KCB 200744YJ, RRID:CVCL\_0063) were used for expressing GluAs, stargazin, control, and ABHD6. The HEK293T cells were cultured with 5% CO<sup>2</sup> in a 37◦C incubator (Jiang et al., 2015). The cDNA transfection was performed in 3.5-cm dishes or six-well plates. The total cDNA used for transfection per 3.5-cm dish or per well in six-well plates was 4 µg. When the expression was performed, a 2:3 ratio of GluA to stargazin cDNA was used (we increased the GluA3 to stargazin ratio to 4:1 to get detectable currents). When GluA1 and GluA2 were coexpressed, the ratio of GluA1 to GluA2 was 3:2 (Shi et al., 2009). Transfection was terminated after 3–5 h. All of the HEK293T cell transfections were performed using polyethylenimine (Polysciences, USA). The HEK293T cells were dissociated with 0.05% trypsin and plated on poly-D-lysinepretreated coverslips after counting the cells with an automated cell counter (µScope CellCounter Basic, Zhoushan Chengchuang Electronic Tech. Co., China). Electrophysiological recording was performed 24–48 h after transfection.

### Electrophysiological Recordings

Electrophysiological recordings were performed as previously reported (Zhang et al., 2009, 2010). Whole-cell voltage clamp recordings were performed for HEK293T cells with a MultiClamp 700A amplifier (Molecular Devices). Series resistance was compensated to 60–70%, and recordings with series resistances of >20 M were rejected. The data were analyzed using Clampfit 9.02 (pClamp, RRID:SCR\_011323), Igor 4.0 (WaveMetrics), and Prism 5 (GraphPad Prism, RRID:SCR\_002798). Data were presented as mean ± SEM. Differences in means were tested with Student's t-test and were accepted as significant if p < 0.05. Coverslips with transfected HEK293T cells were maintained during the recordings in an external solution containing (in mM) NaCl 144, KCl 10, CaCl<sup>2</sup> 2, MgCl<sup>2</sup> 1, HEPES 10, and D-glucose 10, with the pH adjusted to 7.4, mOsm/kg 315. Using 3–5 M borosilicate glass pipettes (World Precision Instruments), whole-cell patches were excised from positively transfected cells identified by epifluorescence microscopy. The internal solution contained (in mM) KCl 145, NaCl 5, EGTA 5, MgATP 4, Na2GTP 0.3, and HEPES 10, with the pH adjusted to 7.2, Osm 305. The glutamate-induced currents were recorded by the local puffing of bath solution containing potassium glutamate (10 mM), and the cells were background perfused with bath solution at the speed of 3 mL/min.

#### Immunostaining Analyses

Immunofluorescence analyses were performed as previously described. Experiments were performed under nonpermeabilized conditions to label the surface of the GluAs receptors. The coverslips with transfected HEK293T cells were washed once with phosphate-buffered saline (PBS). The PBS solution contained (g/L) NaH2PO4·H2O 3.1, Na2HPO<sup>4</sup> 10.9 and NaCl 9, with the pH adjusted to 7.4, Osm 310, fixed with 4% formaldehyde in PBS for 12 min at room temperature (RT), washed three times with PBS, and then blocked with PBS containing 3% goat serum and 5% milk for 30 min at RT. The cells were then incubated for 2 h at RT with the primary antibody (Millipore Cat# AB1504 RRID:AB\_2113602; HA 1:1000 Abmart) diluted in a blocking solution. Either a donkey anti-rabbit Alexa Fluor 546-conjugated secondary antibody (Life Technologies) or a goat anti-mouse Alexa Fluor 633-conjugated secondary antibody (Life Technologies) was used at 1:500 according to the source of primary antibody. The cells were incubated for 1 h at RT with the secondary antibody and washed three to five times with PBS. To label the total GluAs, 0.2% triton was used 5 min after fixation. Fluoromount-G (Southern Biotech) was used to mount the cells on microscope slides. Images were acquired with a laser scanning confocal microscope (Olympus), and were further analyzed in a blinded fashion using the National Institutes of Health (NIH) ImageJ program (ImageJ, RRID:SCR\_003070).

#### Affinity Chromatography Experiments and Western Blotting

To examine the binding region of GluAs with ABHD6 in HEK293T cells, 7.5 µg full-length GluAs or GluAs deletion plasmids, together with 2.5 µg ABHD6, were transfected in a 60 mm dish. A pCAG empty vector was used as control. The cells were harvested 48 h after transfection. The HEK293T cells were washed with PBS once, kept at −80◦C overnight, and thawed at 37◦C for 1 min. Then, the cells were collected with PBS and centrifuged at 17,000×g for 1 min at 4◦C to obtain the cell pellets. 200 µl of buffer A (150 mM NaCl, 20 mM HEPES, 2 mM CaCl2, 2 mM MgCl2, 0.1 mM EDTA, 1% Triton, and protease inhibitors) was added to the cell pellets. Proteins were solubilized by gentle rocking at 4◦C for 2 h. Next, the insoluble fractions were removed by centrifugation at 17,000×g for 30 min. A total of 150 µl of supernatant was used for an affinity chromatography assay, and 16 µl were used as inputs. 3 µl anti-myc antibodies (M20002, Abmart) and 24 µl of protein G beads were added to samples and rotated overnight at 4◦C. Then, the beads were washed five times with wash buffer (150 mM NaCl, 20 mM HEPES, 2 mM

FIGURE 3 | The C-terminus of GluAs mediated the inhibitory effect of ABHD6. (A) The amino acid sequences of different GluA1–3 deletion constructs. The arrow points to the mutants after which ABHD6 failed to reduce the amplitude of the glutamate-induced current. (B) Summary graphs of the peak amplitudes of 10 mM glutamate-induced currents in HEK293T cells transfected with different GluA2/GluA3 deletions, stargazin, and a control vector or vectors encoding ABHD6 (A2D1, control: 24/3, ABHD6: 24/3, p < 0.0001; A2D2, control: 26/3, ABHD6: 26/3, p < 0.0001; A2D3, control: 24/3, ABHD6: 24/3, p < 0.0001; A2D4, control: 24/3, ABHD6: 24/3, p < 0.0001; A2D5, control: 25/3, ABHD6: 24/3, p < 0.001; A2D6, control: 27/3, ABHD6: 27/3, p < 0.001; A2D7, control: 32/4, ABHD6: 32/4, p < 0.01; A2D8, control: 29/3, ABHD6: 29/3, p < 0.01; A2D9, control: 34/4, ABHD6: 34/4, p < 0.05; A2D10, control: 26/3, ABHD6: 26/3, p > 0.05; A2D11, control: 27/3, ABHD6: 27/3, p > 0.05; A2D12, control: 26/3, ABHD6: 26/3, p > 0.05; GluA2 full-length, 30/3, ABHD6: 30/3, p < 0.0001; A3D1, control: 24/3, ABHD6: 24/3, p < 0.0001; A3D2, control: 26/3, ABHD6: 26/3, p < 0.001; A3D3, control: 25/3, ABHD6: 25/3, p < 0.01; A3D4, control: 24/3, ABHD6: 24/3, p < 0.05; A3D5, control: 26/3, ABHD6: 26/3, p < 0.05; A3D6, control: 25/3, ABHD6: 25/3, p < 0.05; A3D7, control: 27/3, ABHD6: 27/3, p > 0.05; A3D8, control: 26/3, ABHD6: 26/3, p > 0.05; GluA3 full-length, 27/3, ABHD6: 27/3, p < 0.01). Each graph of the peak amplitude in HEK293T cells transfected with various ABHD6-expressing vectors was normalized to the peak amplitude in HEK293T cells transfected with the control vector. All summary graphs show means ± SEMs; statistical comparisons by Student's t-test yielded: <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

CaCl2, 2 mM MgCl2, 0.1 mM EDTA, 1% Triton, and protease inhibitors), boiled in SDS sample buffer, and subjected to sodium dodecyl sulfate polyacrylamide gel (SDS-PAGE) electrophoresis.

#### Western Blotting

SDS-PAGE was performed using NuPAGE precast gels (10% Bis-Tris gels, Life Technology), then transferred to nitrocellulose (HATF00010, Millipore), and visualized on immunoblots. The bounded secondary antibody (IRDye <sup>R</sup> 680LT Goat anti-Mouse IgG and 800CW Goat anti-Rabbit IgG, Odyssey) was detected by an infrared imaging system (Odyssey). Monoclonal antibodies against HA-tag (M20003; Abmart) and polyclonal antibodies against GluA1 (ab1504; Millipore) were used in this study.

#### RESULTS

#### Overexpression of ABHD6 Decreased GluA2- or GluA3-Mediated Currents in Transfected HEK293T Cells

Previously, we demonstrated that the overexpression of ABHD6 suppressed the glutamate-induced current in HEK293T cells expressing GluA1, GluA1 + stargazin, and GluA1 + GluA2 + stargazin (Wei et al., 2016). Since GluA1–3 are the major AMPAR subunits expressed in the brain (Angulo et al., 1997; Lee et al., 1998; Lu et al., 2009), we focused our analysis on GluA2 and GluA3, either alone or in combination. The overexpression of ABHD6 significantly reduced the peak amplitude and steady-state amplitude of glutamate-induced currents in HEK293T cells expressing GluA2 (**Figure 1A**), GluA3 (**Figure 1B**), or GluA2 + GluA3 (**Figure 1C**), when cotransfected with stargazin. In the absence of stargazin, ABHD6 overexpression also reduced the amplitude of the currents when any two among GluA1–3 were coexpressed in the HEK293T cells (**Figures 2A–C**). However, similar to GluA1, glutamate elicited almost undetectable ligand-gated currents in HEK293T cells expressing either GluA2 or GluA3 alone (**Figures 2D,E**). We want to address if the existence of endogenous ABHD6 affect the conclusion in **Figures 2D–E**, we used immunostaining and western blotting methods to test the ABHD6 expression level in HEK293T cells (**Supplementary Figure S2**). We found that endogenous ABHD6 can hardly be detected compared with ABHD6-transfected cells, which means that the effect of endogenous ABHD6 in HEK293T cells might be neglected. Furthermore, in the absence of GluAs, the expression of stargazin alone in HEK293T cells exhibited zero current in response to the glutamate puffing (**Figure 2F**), demonstrating the absence of endogenously expressed GluAs in native HEK293T cells. These results showed that the overexpression of ABHD6 inhibits glutamate-induced currents mediated by either heterophilic or hemophilic AMPARs in transfected HEK293T cells.

#### FIGURE 5 | Continued

fnmol-10-00055 March 2, 2017 Time: 11:44 # 8

plasmid together with GluA1 and stargazin. The transfected HEK293T cells were stained with or without permeabilization using an anti-GluA1 antibody. The panels show representative images (left) and quantification of the puncta intensity (right) (control: n = 182/3; ABHD6: n = 191/3; surface GluA1: p < 0.0001; control: n = 233/3; ABHD6: n = 218/3; total GluA1: p > 0.05). (B) Measurement of the surface (top) and total (bottom) expression of GluA1 C-terminal deletion A1D14 in HEK293T cells expressing ABHD6 or a control plasmid together with A1D14 and stargazin. The transfected HEK293T cells were stained with or without permeabilization using an anti-GluA1 antibody. The panels show representative images (left) and quantification of the puncta intensity (right) (control: n = 248/3; ABHD6: n = 208/3; surface GluA1: p > 0.05; control: n = 152/3; ABHD6: n = 154/3; total GluA1: p > 0.05). (C) Measurement of the surface (top) and total (bottom) expression of HA-tagged GluA2 in HEK293T cells expressing ABHD6 or a control plasmid together with GluA2 and stargazin. The transfected HEK293T cells were stained with or without permeabilization using an anti-HA antibody. The panels show representative images (left) and quantification of the puncta intensity (right) (control: n = 169/3; ABHD6: n = 187/3; surface GluA2: p < 0.0001; control: n = 147/3; ABHD6: n = 141/3; total GluA2: p < 0.0001). (D) Measurement of the surface (top) and total (bottom) expression of HA-tagged GluA2 C-terminal deletion A2D10 in HEK293T cells expressing ABHD6 or a control plasmid together with A2D10 and stargazin. The transfected HEK293T cells were stained with or without permeabilization using an anti-HA antibody. The panels show representative images (left) and quantification of the puncta intensity (right) (control: n = 97/3; ABHD6: n = 103/3; surface GluA2: p > 0.05; control: n = 115/3; ABHD6: n = 111/3; total GluA2: p > 0.05). (E) Measurement of the surface (top) and total (bottom) expression of HA-tagged GluA3 in HEK293T cells expressing ABHD6 or a control plasmid together with GluA3 and stargazin. The transfected HEK293T cells were stained without or with permeabilization using an anti-HA antibody. The panels show representative images (left) and quantification of the puncta intensity (right) (control: n = 139/3; ABHD6: n = 123/3; surface GluA3: p < 0.0001; control: n = 144/3; ABHD6: n = 111/3; total GluA3: p > 0.05). (F) Measurement of the surface (top) and total (bottom) expression of HA-tagged GluA3 C-terminal deletion A3D7 in HEK293T cells expressing ABHD6 or a control plasmid together with A3D7 and stargazin. The transfected HEK293T cells were stained with or without permeabilization using an anti-HA antibody. The panels show representative images (left) and quantification of the puncta intensity (right) (control: n = 151/3; ABHD6: n = 164/3; surface GluA3: p > 0.05; control: n = 101/3; ABHD6: n = 103/3; total GluA3: p > 0.05). All summary graphs show means ± SEMs; statistical comparisons by Student's t-test yielded: <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

### The C-Terminus of GluAs Mediated the Inhibitory Effect of ABHD6

The C-terminus of GluA1 has been shown to be crucial for ABHD6's inhibitory effect in heterologous cells. ABHD6 failed to reduce the glutamate-induced current in HEK293T cells expressing a GluA1 mutant in which the C-tail was deleted after "SKRMK" (Wei et al., 2016). We then investigated the importance of similar C-terminal regions in GluA2 and GluA3 for ABHD6-induced inhibition. To this end, we cloned 12 GluA2 C-terminal deletion constructs and eight GluA3 C-terminal deletion constructs based on the sequence similarity among GluA1, GluA2, and GluA3 (**Figure 3A**). Using full-length GluA2 or GluA3 as a positive control, ABHD6 failed to reduce the amplitude of the glutamate-induced current in HEK293T cells expressing GluA2 mutants (A2D10, A2D11, A2D12) and GluA3 mutants (A3D7 and A3D8)(**Figure 3B** and **Supplementary Figure S1**). Interestingly, these results point out that the longest ABHD6-resistant GluA mutants, A2D10 (ending with AKRMKV in GluA2) and A3D7 (ending with SKRMKL in GluA3), and the previously identified A1D14 (ending with SKRMK in GluA1), share sequence similarity to some extent (**Figure 3A**). Thus, like GluA1, the C-terminal regions of GluA2 and GluA3 are required for ABHD6's inhibition of AMPAR-mediated currents in transfected HEK293T cells.

In addition, we generated N-terminal partial deletion mutants for GluA1, GluA2, and GluA3, and tested these mutants in transfected HEK293T cells. Because the extracellular ligandbinding domain is essential for the receptors to respond to glutamate, we removed the entire amino-terminal domain (ATD) from GluA1–3 (**Figure 4A**). The results showed that in cells expressing GluA11ATD, GluA21ATD, and GluA31ATD, ABHD6 suppressed the glutamate-induced currents to a similar extent as those of full-length GluA1–3 (**Figure 4B**). These results demonstrated that the ATDs of GluAs are not required for ABHD6's inhibition of AMPAR-mediated currents.

#### Overexpression of ABHD6 Suppressed the Surface Expression of GluA1–3 in the Transfected HEK293T Cells

To investigate whether the observed reduction in AMPARmediated currents in the heterogonous cells is due to a specific loss of surface-localized AMPARs, or an overall reduction in the expression level of total GluAs, we performed quantitative immunostaining of GluA1–3 from both permeabilized and non-permeabilized HEK293T cells expressing various GluAs together with stargazin. Our results revealed that the coexpression of ABHD6 reduced the surface expression of GluA1 (**Figure 5A**), GluA2 (**Figure 5C**), or GluA3 (**Figure 5E**) compared with the control groups. However, the total GluA1 immunostaining signal from permeabilized HEK293T cells showed no difference between HEK293T cells expressing ABHD6 and stargazin together with GluA1 (**Figure 5A**) or GluA3 (**Figure 5E**), compared to control groups. Furthermore, overexpression ABHD6 increased the total expression of fulllength GluA2 (**Figure 5C**). Thus, ABHD6 specifically affected the surface expression of GluA subunits when coexpressed with stargazin.

Because ABHD6 failed to reduce the ligand-induced currents when expressing C-terminal deletion mutants of GluAs (**Figure 3**; GluA2-A2D10, and GluA3-A3D7, and GluA1-A1D14), we next explored whether these mutations are also resistant to the effects of ABHD6 on the surface expression of AMPARs. By quantifying the immunostaining signal from nonpermeabilized and permeabilized HEK293T cells expressing A1D14 (**Figure 5B**), A2D10 (**Figure 5D**), and A3D7 (**Figure 5F**), we found that the ABHD6-induced suppression of surface GluAs was abolished, and the total expression levels of GluAs remained unchanged compared to control groups. These results suggested that the same GluAs C-terminal mutants abolished the inhibitory effect of ABHD6 on glutamate-induced currents and the surface expression of AMPARs.

expressed in transfected HEK293T cells together with myc-tagged ABHD6 by an anti-myc antibody.

## ABHD6 Directly Bound to GluA1–3 Though the C-Terminus of the Receptors

The notion of ABHD6–AMPAR association was first proposed by a high-resolution proteomics study (Schwenk et al., 2012), and later confirmed by our biochemistry study showing that ABHD6 coimmunoprecipitated GluA1 in transfected heterologous cells (Wei et al., 2016). In the present study, we used a similar approach to test whether ABHD6 coimmunoprecipitated GluA2 and GluA3 when transfected in the HEK293T cells. We found that ABHD6 was specifically coimmunoprecipitated with full-length GluA2 (**Figure 6B**) or GluA3 (**Figure 6C**) only when ABHD6 and GluA1 were coexpressed. Next, to test whether the interactions between ABHD6 and GluA2/3 also require the C-tail of the GluA subunits, we used ABHD6 as a bait to coimmunoprecipitate GluA2 and GluA3 C-terminal mutations that abolished ABHD6's inhibitory effect on the ligand-induced currents and surface expression of GluAs in transfected HEK293T cells. We found that the C-terminal deletion mutations, GluA1-A1D14 (**Figure 6A**), GluA2-A2D10 (**Figure 6B**), and GluA3-A3D7 (**Figure 6C**) all failed to coimmunoprecipitate with ABHD6 in the pulldown assay. These results clearly suggested that ABHD6 binds to all three GluA subunits through their C-terminus domains, and implied that this binding might serve as the underlying mechanism for the functional effects observed in **Figures 1–3**.

### DISCUSSION

fnmol-10-00055 March 2, 2017 Time: 11:44 # 10

ABHD6, a monoacylglycerol lipase, was previously found to inhibit the glutamate-induced currents of GluA1-containing AMPARs in both heterologous cells and neurons (Wei et al., 2016). In this study, we extended our observations to GluA2 and GluA3-containing AMPARs. Our results show that the overexpression of ABHD6 significantly reduced the peak amplitude and steady-state amplitude of glutamate-induced currents in HEK293T cells expressing GluA2, GluA3, or GluA2 + GluA3 when co-transfected with stargazin (**Figure 1**). In the absence of stargazin, ABHD6 overexpression also reduced the amplitude of currents when any two among GluA1– 3 were coexpressed in the HEK293T cells (**Figure 2**). Our results also revealed that the suppression effect of ABHD6 on the surface AMPAR levels is independent of the subunit composition of the AMPARs in transfected HEK293T cells (**Figure 5**).

Similar to previous findings in GluA1, the C-terminal domains of GluA2 and GluA3 are required for ABHD6 to inhibit the ligand-gated current in transfected HEK293T cells (**Figure 3**), and to bind with AMPARs in the co-immunoprecipitation (co-IP) experiments. ABHD6 can significantly decrease GluA1–3 surface expression, but without the C-terminus, the GluA1– 3 surface expression level was restored (**Figure 5**). Combining this with the co-IP experiments, we can infer that ABHD6 directly binds to GluA1–3 C-terminal regions, and selectively inhibits the surface delivery of AMPARs. These observations also suggest that ABHD6 functions during receptor trafficking. According to our results, ABHD6 can interact with the GluA1–3 C-terminus, but the specific interaction region or motif still needs further investigation. The binding of ABHD6 with the GluA1–3 C-terminus is consistent with the observation that ABHD6 is a membrane-bound protein, but is different from stargazin, which interacts with the glutamatebinding domain of AMPARs (Tomita et al., 2007). Stargazin increased the ligand-induced current in oocytes expressing fulllength GluA1 or in GluA1 lacking the ATD. However, the introduction of a Lurcher mutation (A636T) in the glutamatebinding domain of GluA1, or a L497Y mutation in the extracellular S1 domain, abolished the stargazin's enhancement of glutamate-induced currents (Tomita et al., 2007). Thus, ABHD6 and stargazin interact with different domains of AMPARs.

Furthermore, our data clearly demonstrated that ABHD6's inhibition on AMPAR trafficking is not dependent on the existence of stargazin. This is in contrast to CNIH-2/3 or PORCN, two other auxiliary proteins recently found in the macrocomplex of AMPARs (Mauric et al., 2013). The overexpression of CNIH-2/3 facilitated the membrane delivery of AMPARs and increased the amplitude of glutamate-elicited currents in heterologous cells and neurons (Schwenk et al., 2009; Kato et al., 2010; Shi et al., 2010; Coombs et al., 2012; Gill et al., 2012; Harmel et al., 2012). CNIH-2/3 colocalized with γ-8 but not found in the AMPAR complex lack stargazin (γ-2) and γ-3 (Schwenk et al., 2009; Kato et al., 2010; Gill et al., 2011). The membrane localization of CNIH-2 is critically dependent on γ-8, reflected as the absence of surface CNIH-2 in the cerebellum where γ-8 was also absent (Gill et al., 2011). This effect is subject to TARP modulation, since coexpression of CNIH-2 with GluAs and γ-2 showed a more profound increase in current intensity in heterologous cells than that in γ-8 (Gill et al., 2012). PORCN is another auxiliary protein found in the AMPAR macro-complex (Schwenk et al., 2012). Similar to ABHD6, PORCN negative regulates the trafficking of AMPARs in both transfected heterologous cells and knockout (KO) neurons. However, the effect of PORCN seems to be mediated with a TAPR-dependent mechanism. Knockdown PORCN dissociates γ-8 from the AMPAR complex and alters the subunit composition of AMPARs in the hippocampal neurons (Erlenhardt et al., 2016). Thus, ABHD6 and stargazin function in a divergent manner in trafficking the AMPARs to the plasma membrane.

#### CONCLUSION

We extend our previous observation on the functional interaction of ABHD6 and GluA1-containing receptors to GluA2- and GluA3-containing receptors. These results revealed a negative mechanism governing the membrane trafficking of AMPARs through ABHD6 that is independent on stargazin. The limitation of these studies is the lack of in vivo studies of this interaction using KO animals, which requires further investigation.

#### AUTHOR CONTRIBUTIONS

MW, MJ, and JZ contributed equally to this work. WZ, YS, and CZ designed research; MW, MJ, and JZ performed research and analyzed data; LY, YZ, YC, and YM analyzed data; WZ, YS, and CZ wrote the paper.

## FUNDING

This work was supported by grants from the National Basic Research Program of China (2014CB942804, 2014BAI03B01, 2015BAI08B02, and 2012YQ0302604), the National Science Foundation of China (31670842 to CZ, 31371061 and 31571060 to YS, and 81573416 to WZ), the Ministry of Education (the Young Thousand Talent Program to WZ), Beijing Institute of Collaborative Innovation (15I-15-BJ), the Seeding Grant for Medicine and Life Sciences of Peking University (2014-MB-11), and Beijing Municipal Science & Technology Commission (Z161100002616021, Z161100000216154).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol.2017. 00055/full#supplementary-material

#### FIGURE S1 | The C-terminus of GluAs mediated the inhibitory effect of

ABHD6. (A) Summary graphs of the peak amplitudes of 10 mM glutamate-induced currents in HEK293T cells transfected with GluA2 deletion A2D10, A2D11, A2D12, and GluA2 full-length. (B) Summary graphs of the peak amplitudes of 10 mM glutamate-induced currents in HEK293T cells transfected with GluA3 deletion A3D7, A3D8, and GluA3 full-length. (A2D10, control: 26/3, ABHD6: 26/3, p > 0.05; A2D11, control: 27/3, ABHD6: 27/3, p > 0.05; A2D12, control: 26/3, ABHD6: 26/3, p > 0.05; GluA2 full-length, 30/3, ABHD6: 30/3, p < 0.0001; A3D7, control: 27/3, ABHD6: 27/3, p > 0.05; A3D8, control: 26/3, ABHD6: 26/3, p > 0.05; GluA3 full-length, 27/3, ABHD6: 27/3, p < 0.01).

#### REFERENCES


FIGURE S2 | The expression level of endogenous ABHD6 in HEK293T cells is very low compared with HEK293T cells transfected with ABHD6. The white lines represent the scale bar (scale bar 10µm). (A) Measurement of the expression level of ABHD6 in HEK293T cells expressing ABHD6 (bottom) or a control plasmid (top). The transfected or untransfected HEK293T cells were stained with permeabilization using an anti-ABHD6 antibody. The panels show representative images. (B) Measurement of the expression level of ABHD6 in HEK293T cells expressing ABHD6 (right) or a control plasmid (left) using Western blotting. ABHD6 was detected using an anti-ABHD6 antibody.


diversity of native AMPA receptor complexes. Neuron 74, 621–633. doi: 10. 1016/j.neuron.2012.03.034


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer GC declared a shared affiliation with one of the authors YS to the handling Editor, who ensured that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Wei, Jia, Zhang, Yu, Zhao, Chen, Ma, Zhang, Shi and Zhang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

fnmol-10-00055 March 2, 2017 Time: 11:44 # 12

# Molecular Mechanisms for the Coupling of Endocytosis to Exocytosis in Neurons

Zhenli Xie1,2,3,4 , Jiangang Long1,2 , Jiankang Liu1,2 , Zuying Chai 3,4\*, Xinjiang Kang3,4,5,6 \* and Changhe Wang1,2,3,4 \*

<sup>1</sup>Center for Mitochondrial Biology and Medicine, The Key Laboratory of Biomedical Information Engineering of Ministry of Education, School of Life Science and Technology, Xi'an Jiaotong University, Xi'an, China, <sup>2</sup>Frontier Institute of Science and Technology, Xi'an Jiaotong University, Xi'an, China, <sup>3</sup>State Key Laboratory of Membrane Biology, Peking University, Beijing, China, <sup>4</sup>Beijing Key Laboratory of Cardiometabolic Molecular Medicine, Institute of Molecular Medicine, Peking University, Beijing, China, <sup>5</sup>College of Life Sciences, Liaocheng University, Liaocheng, China, <sup>6</sup>Key Laboratory of Medical Electrophysiology, Ministry of Education of China, Collaborative Innovation Center for Prevention and Treatment of Cardiovascular Disease, Institute of Cardiovascular Research, Southwest Medical University, Luzhou, China

Neuronal communication and brain function mainly depend on the fundamental biological events of neurotransmission, including the exocytosis of presynaptic vesicles (SVs) for neurotransmitter release and the subsequent endocytosis for SV retrieval. Neurotransmitters are released through the Ca<sup>2</sup><sup>+</sup>- and SNARE-dependent fusion of SVs with the presynaptic plasma membrane. Following exocytosis, endocytosis occurs immediately to retrieve SV membrane and fusion machinery for local recycling and thus maintain the homeostasis of synaptic structure and sustained neurotransmission. Apart from the general endocytic machinery, recent studies have also revealed the involvement of SNARE proteins (synaptobrevin, SNAP25 and syntaxin), synaptophysin, Ca<sup>2</sup><sup>+</sup>/calmodulin, and members of the synaptotagmin protein family (Syt1, Syt4, Syt7 and Syt11) in the balance and tight coupling of exo-endocytosis in neurons. Here, we provide an overview of recent progress in understanding how these neuronspecific adaptors coordinate to ensure precise and efficient endocytosis during neurotransmission.

#### Edited by:

Cong Ma, Huazhong University of Science and Technology, China

#### Reviewed by:

John J. Woodward, Medical University of South Carolina, USA Subhabrata Sanyal, California Life Company (Calico), USA

#### \*Correspondence:

Zuying Chai chaizuying@pku.edu.cn Xinjiang Kang kxj335@163.com Changhe Wang changhewang@xjtu.edu.cn

Received: 31 December 2016 Accepted: 10 February 2017 Published: 13 March 2017

#### Citation:

Xie Z, Long J, Liu J, Chai Z, Kang X and Wang C (2017) Molecular Mechanisms for the Coupling of Endocytosis to Exocytosis in Neurons. Front. Mol. Neurosci. 10:47. doi: 10.3389/fnmol.2017.00047 Keywords: exocytosis, endocytosis, vesicle recycling, calmodulin, synaptotagmin, SNARE

Neurotransmission based on the exocytosis of synaptic vesicles (SVs) and the subsequent SV membrane retrieval through endocytosis are crucial for efficient neuronal communication, the integrity of neuronal circuits, and normal brain function (Chapman, 2002; Sudhof, 2004; Wu L. G. et al., 2014). With the arrival of an action potential, extra-synaptic Ca2<sup>+</sup> flows into the nerve terminals and triggers soluble N-ethylmaleimide-sensitive factor attachment protein receptor (SNARE) protein-dependent vesicle exocytosis (Südhof and Rothman, 2009; Jahn and Fasshauer, 2012; Rizo and Xu, 2015). The released neurotransmitters diffuse across the synaptic cleft and activate or inhibit the postsynaptic compartment. After exocytosis, fused SV components are locally retrieved from the neuronal surface through endocytosis, which is tightly coupled to exocytosis. Precise and efficient endocytosis is critical for the preservation of presynaptic morphology and structural integrity, the replenishment of presynaptic vesicle pools, and sustained neurotransmission during continuous neuronal activity (Saheki and De Camilli, 2012; Wu L. G. et al., 2014; Leitz and Kavalali, 2016).

Several modes of endocytosis operate to ensure a sufficient and precise vesicle-recycling rate during neurotransmission. Clathrin-mediated endocytosis (CME), the best-characterized endocytic pathway, is known to be the predominant route of vesicle retrieval with slow kinetics (time constant: 10–30 s) following exocytosis (Granseth et al., 2006; Jung and Haucke, 2007; McMahon and Boucrot, 2011). The elevated neuronal activity also elicits bulk endocytosis, which internalizes a large area of plasma membrane, forms an endosome-like endocytic structure, and is finally converted into releasable SVs by a mechanism that remains elusive (Clayton et al., 2008; Smith et al., 2008; Saheki and De Camilli, 2012; Wu L. G. et al., 2014). The kiss-and-run mode of exo-endocytosis probably represents the fast component of SV endocytosis, during which SVs release their contents through a transient nanometer-sized fusion pore and are retrieved rapidly without full collapse into the plasma membrane (He and Wu, 2007; Rizzoli and Jahn, 2007; Alabi and Tsien, 2013). In addition, ultrafast endocytosis has been revealed by electron microscopic analysis (Watanabe et al., 2013) and membrane capacitance (Cm) recordings (Wu et al., 2009; Mahapatra et al., 2016), which are not discussed in detail here because of uncertainty about the nature of these endocytic pathways.

#### EXO-ENDOCYTOSIS COUPLING

Although endocytosis is predominantly a constitutive process in most non-neuronal cells, SV endocytosis is primarily an activitydependent form of membrane retrieval that is spatiotemporally coupled to exocytosis. Upon depolarization, docked vesicles diminish while clathrin-coated pits and structures associated with bulk endocytosis increase near the release sites (Gad et al., 1998; Gundelfinger et al., 2003; Hosoi et al., 2009; Wang et al., 2016), representing exocytosis and the tightlycoupled endocytosis. Consistently, Cm recordings have revealed endocytosis as a stimulation-dependent form of membrane retrieval, in which exocytosis is represented as a Cm jump upon depolarization and the subsequent Cm decay indicates the process of compensatory endocytosis (Zhang et al., 2004; Wu and Wu, 2007; Yamashita et al., 2010). Importantly, the Cm traces reliably decay back to baseline within seconds to minutes after exocytosis, indicating that endocytosis retrieves an amount similar to that of exocytosed SV membrane (Lou et al., 2008; Yamashita et al., 2010; Wang et al., 2016). Furthermore, blockade of exocytosis by cleaving SNARE proteins with botulinum neurotoxins also abolishes endocytosis (Wu et al., 2005; Yamashita et al., 2005), implying a critical role of exocytosis in the initiation of endocytosis. Given the limitation of Cm recordings in small conventional synapses, the optical imaging of fluorescent dyes such as FM1–43, or dextran uptake, has permitted studies of vesicle recycling in neuronal terminals (Virmani et al., 2003; Deák et al., 2004; Clayton et al., 2010; Wang et al., 2016). Tagging vesicular proteins with pHluorin, a pH-sensitive green fluorescent protein that allows the direct visualization of exocytosis and the subsequent endocytosis in living nerve terminals, has also confirmed the tight coupling of synaptic endocytosis to exocytosis in terms of both timing and amount (Poskanzer et al., 2003; Ferguson et al., 2007; Hua et al., 2011; Yao et al., 2011).

### Ca2+/CALMODULIN IN EXO-ENDOCYTOSIS COUPLING

Although there is extensive evidence that Ca2<sup>+</sup> influx plays critical role in compensatory endocytosis, whether and how cytosolic Ca2<sup>+</sup> regulates exo-endocytosis coupling is rather controversial. Accumulating evidence has shown that a transient elevation in cytosolic Ca2<sup>+</sup> triggers and accelerates both clathrindependent and clathrin-independent endocytosis in neurons and neuroendocrine cells (Balaji et al., 2008; Hosoi et al., 2009; Sun et al., 2010; Leitz and Kavalali, 2016). However, the Ca2+-dependence of exo-endocytosis is diverse among different preparations (Wu and Wu, 2014; Wu L. G. et al., 2014). Endocytosis can also occur independent of cytosolic Ca2<sup>+</sup> (Ryan et al., 1993; Granseth et al., 2006), and increasing the intracellular Ca2<sup>+</sup> concentration slows exo-endocytosis in many cases (von Gersdorff and Matthews, 1994; Leitz and Kavalali, 2011; Armbruster et al., 2013). Nonetheless, the critical roles of cytosolic Ca2<sup>+</sup> in SV exocytosis make it inconclusive whether Ca2<sup>+</sup> influx directly mediates exo-endocytosis coupling and thus controls the timing and amount of compensatory endocytosis independent of exocytosis, although great efforts have been made to dissect this by manipulating exocytosis (Sun et al., 2002; Wu et al., 2009; Yao et al., 2011). Thus, the exact role of Ca2<sup>+</sup> in the coupling of SV exo-endocytosis remains a pending question and needs more thorough investigations.

Several endocytic Ca2<sup>+</sup> sensors and effectors have been shown to initiate and mediate Ca2+-dependent endocytosis, in which calmodulin is involved in most forms of endocytosis and synaptotagmin is a dual Ca2<sup>+</sup> sensor for both exocytosis and endocytosis. Calcineurin functions as a key mediator of Ca2+/calmodulin in exo-endocytosis by dephosphorylating endocytic proteins known as dephosphins (Cousin and Robinson, 2001; Saheki and De Camilli, 2012). Typically, many proteins involved in different stages of CME (e.g., dynamin, synaptojanin, amphiphysin, epsin and Eps15) are constitutively phosphorylated as an inactive conformation in resting nerve terminals (Liu et al., 1994; Chen et al., 1999; Lee et al., 2004, 2005). During synaptic activity, these dephosphins undergo rapid dephosphorylation by the Ca2+/calmodulinactivated calcineurin to drive endocytosis via their enhanced binding to other endocytic factors or by dephosphorylationdependent activation (Liu et al., 1994; Slepnev et al., 1998; Anggono et al., 2006; Saheki and De Camilli, 2012). The regulation of CME by calcineurin has been confirmed by the inhibition of slow endocytosis with calcineurin blockers, or the knockdown/knockout of calcineurin (Engisch and Nowycky, 1998; Sun et al., 2010; Armbruster et al., 2013; Wu X. S. et al., 2014). In addition, calcineurin also mediates bulk endocytosis by dephosphorylating dynamin 1 during elevated neuronal activity (Clayton et al., 2009, 2010). It has been proposed that the GTPase activity of dynamin is essential for vesicle fission during CME, bulk endocytosis and kiss-and-run, while its phosphorylation-dephosphorylation cycle is also critical for activity-dependent bulk endocytosis (Marks et al., 2001; Yamashita et al., 2005; Anggono et al., 2006; Clayton and Cousin, 2009; Anantharam et al., 2011). However, the dynamindependency of bulk endocytosis remains controversial because it still occurs robustly in the absence of dynamin 1, which might be due to the compensatory effect of other dynamin isoforms (Hayashi et al., 2008; Raimondi et al., 2011; Lou et al., 2012; Fan et al., 2016). Finally, dynamin and the calcineurin-dependent dynamin-syndapin interaction have also been demonstrated to regulate the kiss-and-run mode of exo-endocytosis and the quantal size of neurotransmitter release by limiting the fusion pore dilation under elevated stimulation (Graham et al., 2002; Samasilp et al., 2012).

In addition to calcineurin, myosin light-chain kinase is another co-effector functioning to accelerate both the slow and fast forms of exo-endocytosis through the activitydependent phosphorylation of myosin at the downstream of Ca2+/calmodulin (Yue and Xu, 2014; Li et al., 2016). A recent study has also defined critical roles of calmodulin in regulating the intrinsic membrane-remodeling activity via a Ca2+-dependent interaction with Rvs167 in yeast and several endocytic N-BAR domain proteins such as endophilins and amphiphysins in mammalian cells (Myers et al., 2016).

#### SYNAPTOTAGMIN PROTEINS IN EXO-ENDOCYTOSIS COUPLING

Synaptotagmins (Syts), a family of type I membrane proteins with evolutionarily conserved cytosolic tandem C<sup>2</sup> domains (C2A and C2B), are well-characterized Ca2<sup>+</sup> sensors that initiate SNARE-dependent vesicle fusion during synaptic transmission and hormone secretion (Chapman, 2002; Gustavsson and Han, 2009; Südhof and Rothman, 2009; Pang and Südhof, 2010). At least 17 mammalian Syt isoforms have been identified, the detailed characterizations of which are summarized in recent reviews (Gustavsson and Han, 2009; Pang and Südhof, 2010). All Syt members bind the clathrin-adaptor protein AP-2 with high affinity (K<sup>d</sup> = 0.1–1.0 nM) and some Syts have been shown to function in different endocytic pathways (Zhang et al., 1994; Li et al., 1995; Chapman et al., 1998; Yao et al., 2011). Syt1, the prototypical Syt protein functioning as the primary Ca2<sup>+</sup> sensor for exocytosis, has also been proposed to be a major Ca2+-sensing protein that promotes CME upon exocytosis (Haucke et al., 2000; Jarousse and Kelly, 2001; Poskanzer et al., 2003). Cm recordings, electron microscopy, FM uptake and pHluorin assays have reliably revealed dramatic endocytic defects in Syt1-deficient cells from a variety of organisms (Poskanzer et al., 2003; Nicholson-Tomishima and Ryan, 2004; Yao et al., 2011, 2012). Meanwhile, Syt1 has also been demonstrated to bind the µ2 subunit of the endocytic adaptor protein AP-2 and the µ-homology domain of stonin-2 through its C2B domain (Zhang et al., 1994; Haucke et al., 2000; Jarousse and Kelly, 2001; Walther et al., 2001; Kaempf et al., 2015). However, the direct regulation of Syt1 in CME has been challenged due to that the endocytic defects may be secondary to the impaired exocytosis caused by Syt1 deficiency (Poskanzer et al., 2006; Yao et al., 2011). A recent study has provided direct evidence that Syt1 indeed functions as a Ca2<sup>+</sup> sensor for SV endocytosis by uncoupling the function of Syt1 in exo- and endocytosis in hippocampal neurons (Yao et al., 2011). Then, with cell-attached Cm recordings, another group validated that Syt1 functions to modulate the Ca2+-dependence of CME probably by AP-2 dependently prolonging the duration of fission pore closure (Yao et al., 2012).

Syt7 is ubiquitously expressed at early stage of development but is later restricted to dividing cells, neuroendocrine cells, and presynaptic neuronal structures (Virmani et al., 2003). Syt7 binds Ca2<sup>+</sup> with a high apparent affinity and slow kinetics, and thus mainly functions as a slow Ca2<sup>+</sup> sensor to mediate the slow phase of exocytosis known as asynchronous release, as well as fusion-pore expansion and synaptic facilitation (Maximov et al., 2008; Schonn et al., 2008; Liu et al., 2014; Neuland et al., 2014; Wu et al., 2015). Interestingly, Syt7 is extensively spliced and exhibits a broad variety of alternative splice variants, among which the short Syt7 variant lacking both of the C2 domains inhibits CME but accelerates exo-endocytosis in response to intense stimulation, while the regular full-length Syt7 directs synaptic endocytosis into a slow-recycling CME (von Poser et al., 2000; Virmani et al., 2003). A recent study also defined Syt7 as a Ca2<sup>+</sup> sensor for SV replenishment (Liu et al., 2014), confirming the regulatory role of Syt7 in SV recycling. Furthermore, Syt7 also plays a critical role in the occurrence of kiss-and-run probably by mediating the push-and-pull regulation of fusion pore dilation (Segovia et al., 2010; Neuland et al., 2014). It has been proposed that Ca2<sup>+</sup> binding to the C2A domain of Syt7 is sufficient to trigger fusion-pore opening but the resulting pores are unstable, thus leading to a dramatic increase in kiss-and-run fusion events. In contrast, Ca2<sup>+</sup> binding to the C2B domain facilitates the continuous expansion of fusion pores, making Syt7 a critical regulator of the Ca2+-dependent occurrence of kiss-and-run and full-fusion events (Segovia et al., 2010; Neuland et al., 2014).

Syt4 and Syt11 are classified as non-Ca2+-binding Syts because of an aspartate-to-serine substitution in a Ca2+ coordination site of the C2A domain, and they do not bind Ca2<sup>+</sup> biochemically (von Poser et al., 1997; Dai et al., 2004; Dean et al., 2009). Syt4 has been reported to regulate fusion-pore and fusion modes in both endocrinal cells and neurons, but the effects fail to reach a consensus in these preparations. Syt4 overexpression favors the occurrence of kiss-and-run and increases the duration of fusion pore dilation in PC12 cells (Wang et al., 2001, 2003; Zhang et al., 2010). Cell-attached Cm recording also revealed prolonged lifetime and smaller downward Cm steps of fission pores during endocytosis (Zhang et al., 2010). In contrast, Syt4 deficiency accelerates the rapid component of endocytosis probably through the enhanced kiss-and-run in the peptidergic nerve terminals of posterior pituitary neurons (Zhang et al., 2009). Similarly, Syt4 inhibits BDNF release in both axons and dendrites but with distinct mechanisms, in which presynaptic Syt4 decreases frequency of spontaneous quantal release while postsynaptic Syt4 limits quantal size by favoring kiss-and-run modes of exo-endocytosis (Dean et al., 2009).

Syt11 is a newly-defined endocytic regulator that inhibits CME and bulk endocytosis in neurons probably through distinct mechanisms (Wang et al., 2016). Disruption of this inhibitory role by Syt11-knockdown induces excessive membrane retrieval, accelerates vesicle pool replenishment, and facilitates sustained neurotransmission, indicating a critical role of Syt11 as a clamp protein to ensure the precise coupling and balance of endocytosis to exocytosis during neurotransmission (Wang et al., 2016). Since Syt11 does not bind Ca2<sup>+</sup> biochemically, there may also be a Ca2+-sensitive inhibitor to ensure the precise Ca2+ dependency of exo-endocytosis, especially during sustained neuronal activities.

#### SNARE PROTEINS AND SYNAPTOPHYSIN IN EXO-ENDOCYTOSIS COUPLING

In addition to Ca2<sup>+</sup> influx upon depolarization, exocytosis itself is required for the initiation of compensatory SV endocytosis, which is abolished by the cleavage of SNARE proteins essential for exocytosis with botulinum neurotoxins (Hosoi et al., 2009; Xu et al., 2013). A debated issue is that exocytosis-mediated plasma membrane expansion and surface tension reduction may serve to initiate the local membrane curvature (membrane buds) for internalization (Dai et al., 1997; Anantharam et al., 2010; Diz-Muñoz et al., 2013; Hassinger et al., 2017). Meanwhile, the delivery of PI(4,5)P2-lacking SV membranes to the plasma membrane makes these budding sites competent for the recruitment of endocytic scaffolding proteins and the formation of coated pits (Wenk and De Camilli, 2004; McMahon and Gallop, 2005; Haucke et al., 2011; Saheki and De Camilli, 2012; Puchkov and Haucke, 2013). Furthermore, some classical exocytic proteins, especially Syts, SNARE proteins and synaptophysin, also function to couple exoendocytosis.

SNARE proteins are critical for membrane fusion, while recent studies have also implied a significant contribution of synaptobrevins (also termed VAMPs, vesicle-associated membrane proteins), syntaxin, and SNAP-25 in the coupling of SV exo-endocytosis. Synaptobrevin-2 (VAMP2) deficiency impairs the fast component of compensatory endocytosis and the rapid re-use of SVs in hippocampal neurons (Deák et al., 2004), while the cleavage of VAMP2 and VAMP3 with tetanus toxin blocks both the slow and fast modes of endocytosis in nerve terminals of the calyx of Held (Hosoi et al., 2009; Xu et al., 2013). A recent study has also established an essential role of synaptobrevin in slow endocytosis in hippocampal neurons (Zhang et al., 2013). VAMP4 also plays critical roles in activity-dependent bulk endocytosis in hippocampal neurons (Nicholson-Fish et al., 2015). In addition, an early study also revealed the involvement of t-SNARE proteins (syntaxin and SNAP-25 in targeting membrane) in exo-endocytosis coupling in yeast (Gurunathan et al., 2002).

(Syp) and SNARE proteins, Ca2+-binding (Syt1, Syt7) and non-Ca2+-binding (Syt4, Syt11) Syts, and other proteins such as dynamin (Dyn) coordinate to control the efficient and precise coupling of SV endocytosis to exocytosis. Syt11 is shown in red due to the nature of negative regulator, and Syt4 is shown in gray since that Syt4 shows different effects on different types fusion-fission events. Scaffolding proteins (e.g., N-BAR/F-BAR/BAR domain-containing proteins (such as FCHo, endophilin, amphiphysin, syndapin), intersectin, Rab3, CDC42, N-WASP, SNX9 and Eps15), downstream effectors (e.g., calcineurin, myosin, CDC2 and CDK5), phosphoinositide metabolism (e.g., PI(4,5)P2, cholesterol, synaptojanin, PI 3-kinase and PIPK1γ), and other general endocytic machineries (clathrin, AP-2, AP180, Epsin and stonin) are excluded to simplify the cartoon.

Consistently, SNAP25 knockdown inhibits slow SV endocytosis in hippocampal synapses (Zhang et al., 2013), and the cleavage of SNAP-25 with botulinum neurotoxin E impairs both the fast and slow modes of endocytosis in calyx terminals (Xu et al., 2013). Syntaxin 1 clearance with botulinum neurotoxin C also greatly inhibits SV endocytosis at the calyx (Xu et al., 2013), while syntaxin 1A SUMOylation shows a similar inhibitory effect on SV endocytosis in cortical and hippocampal neurons (Craig et al., 2015).

Synaptophysin is the most abundant SV protein; it is exclusively localized to SVs with uncertain roles in SV exocytosis, endocytosis, synapse formation, and other synaptic functions (Janz et al., 1999; Tarsa and Goda, 2002; Takamori et al., 2006). Synaptophysin interacts with dynamin via its C-terminal cytoplasmic tail region in a Ca2+-dependent manner (Daly et al., 2000; Daly and Ziff, 2002), disruption of which decreases vesicle retrieval and thus neurotransmitter release during intense stimulation, probably due to the impairment of clathrinindependent rapid endocytosis (Daly et al., 2000). A recent study provided direct evidence for the involvement of synaptophysin in exo-endocytosis coupling by using optical imaging of Syt1-pHluorin and SV2-pHluorin. Synaptophysin knockout impairs SV endocytosis during and after sustained neuronal activity, while the C-terminal tail-truncated synaptophysin can only rescue the slow post-stimulus endocytosis (Kwon and Chapman, 2011), indicating the distinct requirement of synaptophysin structural elements in the two phases of exoendocytosis. These findings validate the critical dual roles of synaptophysin and SNARE proteins in both exocytosis and the exo-endocytosis coupling process; however, which specific endocytic pathways are regulated by these fusion machineries and how these proteins are involved in the compensatory SV endocytosis remain largely elusive.

#### CONCLUSION

Recent advances paint an extremely complex picture of the tight exo-endocytosis coupling in neurons. At least three different endocytic pathways, CME, activity-dependent bulk endocytosis, and the kiss-and-run mode of fast endocytosis, cooperate to couple SV endocytosis to exocytosis with different neuronal activities. The Ca2+-calmodulin-calcineurin pathway, synaptophysin and SNARE proteins, Ca2+-binding Syt members, and other positive regulators work together

#### REFERENCES


with endocytic inhibitors such as non-Ca2+-binding Syts to provide a fine-tuning mechanism for the efficient and precise coupling of SV endocytosis to exocytosis (**Figure 1**). Membrane lipid structures and proteins involved in phosphoinositide metabolism also play critical roles in the exo-endocytosis coupling. In addition, scaffolding and effector proteins essential for non-neuronal endocytosis are also necessary for exo-endocytosis coupling in neurons. However, uncertainty about the functions of these endocytic regulators and the co-existence of several other endocytic pathways with distinct kinetics and molecular mediators require a more thorough investigation (Wu et al., 2009; Watanabe et al., 2013; Kononenko and Haucke, 2015). Further studies have been challenged due to the limitation of electrophysiological recordings and live fluorescence imaging assays of single-SV recycling in small nerve terminals. Advances in super-resolution microscopy and correlative light and electron microscopy offer new opportunities in this field. In addition, optogenetic stimulation, two-photon imaging, and acute molecular manipulation in vivo allow a deep functional analysis of the endocytic regulators that associate SV recycling with brain disorders such as Alzheimer disease, Parkinson disease and emotional disorders.

#### AUTHOR CONTRIBUTIONS

ZX drafted the manuscript with help from JLo, JLi, ZC, XK and CW. All authors coordinated, revised and approved the manuscript.

#### FUNDING

This work was supported by the National Natural Science Foundation of China (31400708, 81571235 and 31670843), the Natural Science Foundation of Heilongjiang Province of China (C201453), the Natural Science Foundation of Shandong Province of China (ZR2016CM16) and the Scientific Research Fund of Heilongjiang Provincial Education Department (12531750 and 12531746). XK was supported in part by the start-up funding of Liaocheng University (318051525).

#### ACKNOWLEDGMENTS

We thank Dr. Iain C. Bruce (Peking University) for reading the manuscript.


single presynaptic boutons. Neuron 11, 713–724. doi: 10.1016/0896-6273(93) 90081-2


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Xie, Long, Liu, Chai, Kang and Wang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Hemifusion in Synaptic Vesicle Cycle

Dae-Hyuk Kweon<sup>1</sup> , Byoungjae Kong<sup>1</sup> and Yeon-Kyun Shin<sup>2</sup> \*

<sup>1</sup>Department of Genetic Engineering, College of Biotechnology and Bioengineering, Sungkyunkwan University, Suwon, South Korea, <sup>2</sup>Department of Biochemistry, Biophysics and Molecular Biology, Iowa State University, Ames, IA, USA

In the neuron, early neurotransmitters are released through the fusion pore prior to the complete vesicle fusion. It has been thought that the fusion pore is a gap junction-like structure made of transmembrane domains (TMDs) of soluble N-ethylmaleimidesensitive-factor attachment protein receptor (SNARE) proteins. However, evidence has accumulated that lipid mixing occurs prior to the neurotransmitter release through the fusion pore lined predominantly with lipids. To explain these observations, the hemifusion, a membrane structure in which two bilayers are partially merged, has emerged as a key step preceding the formation of the fusion pore. Furthermore, the hemifusion appears to be the bona fide intermediate step not only for the synaptic vesicle cycle, but for a wide range of membrane remodeling processes, such as viral membrane fusion and endocytotic membrane fission.

Keywords: SNARE, membrane fusion, hemifusion, fusion pore, transmembrane domain

## INTRODUCTION

Neurotransmitter release from the neuron requires fusion of vesicles to the plasma membrane. However, the bilayer structure is highly stable; thus, two bilayers normally do not fuse spontaneously. It is thought that conserved soluble N-ethylmaleimide-sensitive-factor attachment protein receptor (SNARE) proteins mediate synaptic vesicle fusion (Söllner et al., 1993a,b). Three SNARE proteins involved in neuroexocytosis are synaptobrevin-2 (Syb2, also called VAMP2), syntaxin-1 (Stx1), and SNAP-25. Syb2 is the vesicle membrane (v-) SNARE of 116-amino acids with a single transmembrane domain (TMD). Stx1 is the 288-amino acid protein attached to the plasma membrane, likewise with a single TMD. SNAP-25 has lipid anchors in the plasma membrane and forms the target membrane (t)-SNARE complex with Stx1 (**Figure 1A**). Cognate SNARE motifs that protrude respectively from two membranes assemble to form a parallel four-helix bundle (Poirier et al., 1998; Sutton et al., 1998) that drives apposition and subsequent fusion of two membranes (Weber et al., 1998). It is believed that SNARE proteins progressively zipper from the membrane-distal N-terminal region toward the membrane-proximal C-terminal region (Fiebig et al., 1999; Chen et al., 2001; Melia et al., 2002; Matos et al., 2003; Sorensen et al., 2006; Ellena et al., 2009; Gao et al., 2012; Lou and Shin, 2016).

A crucial question towards elucidating the pathway of membrane fusion is what happens to the bilayers when two membranes merge. Obviously, our interests are on the role that SNARE TMDs play in fusion because they are located at or near the epicenter of membrane fusion. There are two proposed fusion pathways that predict two very different roles of the TMDs. The first model (Han et al., 2004), which is primarily based on electrophysiological measurements, argues that the TMDs serve as the principal building block of the fusion pore that is considered a bona fide intermediate for synaptic vesicle fusion. In stark contrast, the second model (Xu et al., 2005), largely based on spectroscopic and structural studies, predicts that lipids are major building blocks that build the ''hemifusion'' and the fusion pore. In the latter, TMDs play a supporting role either as membrane anchors for the soluble SNARE complex or as mechanical levers working at the periphery.

#### Edited by:

Cong Ma, Huazhong University of Science and Technology, China

#### Reviewed by:

Tabrez Jamal Siddiqui, University of Manitoba, Canada Yuji Ishitsuka, University of Illinois at Urbana–Champaign, USA

> \*Correspondence: Yeon-Kyun Shin colishin@iastate.edu

Received: 27 December 2016 Accepted: 27 February 2017 Published: 16 March 2017

#### Citation:

Kweon D-H, Kong B and Shin Y-K (2017) Hemifusion in Synaptic Vesicle Cycle. Front. Mol. Neurosci. 10:65. doi: 10.3389/fnmol.2017.00065

Although the TMD-lined fusion pore model was developed earlier to explain electrophysiological results, much evidence has now accumulated in favor of membrane fusion through hemifusion for SNARE-dependent vesicle fusion. Furthermore, the hemifusion might be a common intermediate shared by many membrane fusion and fission systems including viral-cell membrane fusion (Melikyan et al., 1995; Chernomordik et al., 1998; Chernomordik and Kozlov, 2003) and endocytotic membrane fission (Antonny et al., 2016).

#### TMD-LINED FUSION PORE

There is overwhelming evidence that vesicle fusion transits through the fusion pore (Neher and Marty, 1982; Breckenridge and Almers, 1987; Chow et al., 1992; Alvarez de Toledo et al., 1993; Lollike et al., 1995), a metastable intermediate with a small aqueous opening through two apposed membranes. In this state, the vesicle is expected to remain nearly intact except a pore that continues through the plasma membrane. The size of the fusion pore is estimated to be similar to those of large ion channels. Some small but detectable amounts of neurotransmitters can pass through the fusion pore. Electrophysiological measurements of fusion pore conductance revealed that the fusion-pore diameter remains <3 nm and can persist for a few seconds (Albillos et al., 1997).

In an earlier study, no flow of lipids was observed at the stage of the fusion pore formation (Klyachko and Jackson, 2002). Combining the electrophysiological data, one could envision a gap junction-like fusion pore in which SNARE complex formation brings about docking of two hemi-pores made of vand t-SNARE TMDs, respectively, to produce a longer, complete pore that continues through the two membranes (**Figure 1B**).

Experimental evidence that supports such a protein-lined fusion pore was obtained by amperometric study of Stx1 mutants (Han et al., 2004). Han et al. (2004) found that Trp mutations at the putative pore-lining residues in the TMD interfere with the release of the neurotransmitters, consistent with the model. Based on the conductance measurements, they modeled a helixlined pore consisting of 5–8 TMDs with a 0.5 nm pore at the center (**Figure 1C**).

Although the data qualitatively agree with the model, one caveat is that the impedance of the release due to the Trp mutation is much lower than expected. There is only a 20%–30% decrease in the release with the mutation. In reality, one would expect that 5–8 Trp residues at the same layer in the 0.5-nm-diameter pore would completely fill the space, which would allow little passage of neurotransmitters through the pore. On the other hand, two membranes are still fully separated in this stage, and the model does not explain how structurally and energetically the two bilayers eventually merge to complete the fusion reaction (**Figure 1D**; Chernomordik and Kozlov, 2003).

Recently, Cys-scanning experiments showed that V101 and I105 in the TMD of Syb2 might line the fusion pore (Bao et al., 2016). However, the small nanodisc of ∼6 nm diameter used in the Cys-scanning experiments, contained as few as two copies of Syb2. The reality is that two copies of the TMD would not be able to form a TMD-lined pore. Alternatively, an idea that the fusion pore may be both lipidic and proteinaceous was suggested on the basis of an amperometry study of chromaffin cells expressing C-terminal truncation mutants of SNAP-25 (Fang et al., 2008). The layout of TMDs in this model is however purely imaginary with little experimental support.

The early result that there was no lipid flow at the fusion pore stage is an important basis for developing a gap junction-like fusion model. However, it is not unusual to have poor lipid mixing when the protein density is high, as is shown for influenza virus-cell fusion (Chernomordik et al., 1998). Thus, one could argue that limited lipid mixing might not be a sufficient condition for a gap junction-like fusion pore. After all, the TMD might not be a required component to complete membrane fusion. In fact, TMD-less, lipid-anchored SNAREs are sufficient for healthy neurotransmitter release in the neuron (Zhou et al., 2013), vacuole fusion in yeast (Xu et al., 2011), and proteoliposome fusion in vitro (McNew et al., 2000), raising concerns on the validity of the TMD-lined fusion pore model.

### HEMIFUSION

Let us now consider alternative possibilities to the TMD-lined fusion pore. For membrane fusion between influenza virus and red blood cells, Kemble et al. (1994); Melikyan et al. (1995) made a seminal discovery that uncovered a lipiddominant intermediate in membrane fusion (Kemble et al., 1994; Melikyan et al., 1995). The authors found that a GPI-anchored hemagglutinin mutant arrests membrane fusion at the intermediate state in which lipid mixing is allowed while content mixing is not.

The observation by Kemble et al. (1994); Melikyan et al. (1995) appears to be consistent with a theoretical model for membrane fusion, developed for protein-free fusion of two lipid bilayers, on the basis of the lipid-stalk intermediate (Kozlov et al., 1983). We call the half-merged state in which inner leaflets are intact while the outer leaflets are merged the ''hemifusion''. The predicted hemifusion was later imaged experimentally with x-ray crystallography for the macroscopically aligned, protein-free multi-bilayer (Yang and Huang, 2002).

At hemifusion, lipid mixing may be allowed through outer leaflets (**Figure 1E**). However, lipid mixing through inner leaflets is also possible because the hemifusion could be in equilibrium with small fusion pores that flicker (**Figure 1F**; Chanturiya et al., 1997). In some cases, the hemifusion diaphragm can be formed via expansion of the hemifusion into a large area (Hernandez et al., 2012).

The idea that intracellular membrane fusion might transit through a structure of the curved membrane was percolated through the observation that the exogenously added lysolipids impair exocytosis in cells (Chernomordik et al., 1993). It has been thought that the molecular shape of a lipid correlates with the effective spontaneous curvature. While cylindrical phosphatidylcholine forms the almost flat monolayer, cone-shaped phosphatidylethanolamine and diacylglycerol bulge in the direction of the acyl chains and favor the net negative curvature. In contrast, lysolipids such as lysophosphatidylcholine and polyphosphoinositides are inverted cone-shaped molecules

junction-like fusion pore is formed by TMDs of SNARE proteins. Two hemi-pores, one formed by Stx1 TMDs and the other by Syb2 TMDs, dock to constitute a long pore through which neurotransmitters can be released. (C) Five to eight TMDs may form a pore in the center. (D) The TMD-lined pore model cannot explain how fusion between the two membranes is achieved or how the small fusion pore eventually dilates to complete fusion reaction. (E) In the hemifusion, outer leaflets are merged while inner leaflets remain separate. (F) A lipidic fusion pore is formed by inner leaflet mixing. This small fusion pore is in equilibrium with the hemifusion, which may result in flickering. The small fusion pore would eventually dilate to complete membrane fusion.

with large polar heads and thin acyl chain tails that prefer the positively curved membrane to the negatively curved one. Net negative membrane curvature happens to be a characteristic feature of the hemifusion (Chernomordik and Kozlov, 2008).

#### HEMIFUSION IN SNARE-DEPENDENT MEMBRANE FUSION

The first direct experimental evidence of hemifusion in SNARE-dependent fusion was from in vitro fusion assays employing SNARE-reconstituted proteoliposomes (Lu et al., 2005; Xu et al., 2005). The fluorescence measurements showed that lipids in the outer leaflets mix faster than those in the inner leaflets. Furthermore, the fusion reaction between single proteoliposomes, studied with total internal reflection microscopy, showed distinct steps that reflected the hemifusion (Yoon et al., 2006).

Hemifusion has also been identified in fusion of vacuoles from yeast (Jun and Wickner, 2007) as well as in Ca2+ triggered exocytosis in chromaffin cells (Wong et al., 2007). Additionally, the hemifusion was observed in cell-cell fusion by flipped SNAREs (Giraudo et al., 2006) and in fusion between a SNARE-reconstituted nanodisc and a liposome (Shi et al., 2012), although it is unclear whether the hemifusion observed here is a dead-end product or not. Recently, signals reflecting hemifusion have been detected in the force measurement in SNARE-mediated fusion of proteoliposome to the supported bilayer (Oelkers et al., 2016). Furthermore, a hemifusion structure was visualized in the study of neurons with electron tomography at low resolution (Zampighi et al., 2006).

Although there is overwhelming evidence that the hemifusion does exist in SNARE-dependent membrane fusion, there is still a controversy concerning whether the hemifusion is on- or off-pathway in Ca2+-triggered exocytosis. In fact, Diao et al. (2012) raised the possibility that hemifusion might be an off-pathway intermediate in Ca2+-triggered exocytosis (Diao et al., 2012). In their in vitro experiments, full fusion has occurred within the pool of un-hemifused proteoliposomes, although it is still possible that the hemifusion is too short-lived to be detected under their experimental conditions. Alternatively, in some limited cases, the hemifusion can expand into the hemifusion diaphragm, which may not progress readily to full fusion (Hernandez et al., 2012).

There is still controversy on whether the TMDs are even required in the neuro-exocytosis. While Zhou et al. (2013) found that GPI-anchored Syb2 fully supports the neurotransmitter release (Zhou et al., 2013). Han et al. (2004) showed that various lipid-anchored Syb2 variants provide little support of exocytosis (Chang et al., 2016). Although SNARE TMDs may not be essential, it turns out that they play important, active roles in modulating SNARE-dependent membrane fusion. For example, Shin et al. (2014) have shown that cholesterol could change the conformation of the dimeric Syb2 TMD to be favorable for membrane fusion (Tong et al., 2009). A simulation study found that the SNARE TMDs might play a role in initiating fusion by distorting the lipid packing of the outer leaflets (Risselada et al., 2011). It is also shown that the conformational flexibility of the Syb2 TMD might lower the negative membrane curvature within the outer leaflet of the fusion pore neck to facilitate pore expansion (Dhara et al., 2016).

#### COUPLING OF SNARE ZIPPERING TO THE HEMIFUSION

SNARE motifs assemble into a stable, parallel, four-stranded coiled coil. The SNARE complex is made of 15 (numbered −7 to +8) hydrophobic layers and one ionic zeroth layer at the center. There is evidence that SNARE motifs zipper from the membrane-distal N-terminal region towards the membrane-proximal C-terminal region (Fiebig et al., 1999; Chen et al., 2001; Melia et al., 2002; Sorensen et al., 2006; Su et al., 2008; Ellena et al., 2009). After transitioning through intermediate structures that might serve as structural platforms for interactions with other accessory proteins, the SNARE complex ends up at a cis-conformation representing the post fusion state in which TMDs of Stx1 and Syb2 reside in the same membrane. Prior to cis-complex formation, partial complexes are present, of which the degree of zippering was only recently revealed. It was found, from single-molecule force measurements, that SNARE complex formation may occur in at least two steps, with a pivot at the conserved ''zeroth'' layer in the middle (Li et al., 2007; Gao et al., 2012; Min et al., 2013; Shin et al., 2014; Zorman et al., 2014). Now, given that SNARE zippering drives apposition of two membranes, it must be determined at what stage of SNARE zippering hemifusion occurs. Precise mapping of the degrees of SNARE zippering to specific stages of membrane fusion is prerequisite to the understanding of the mechanism of membrane fusion.

An immediate, related question in the field has been if membranes are hemifused before the Ca2<sup>+</sup> influx. Hemifusion induced by SNARE complex formation before Ca2<sup>+</sup> was reported (**Figure 2B**), for the first time, by Schaub et al in their in vitro investigation of the regulation of SNARE-dependent vesicle fusion by Syt1 and complexin (Schaub et al., 2006). This result was verified with the observations that the hemifusion is a stable intermediate of exocytosis in neuronal cells in vivo (Wong et al., 2007) and that two membranes may be hemifused before Ca2<sup>+</sup> influx (Zampighi et al., 2006). In contrast, it has been proposed that syt1 and Ca2<sup>+</sup> play a role in driving lipid stalk formation (Martens et al., 2007; Hui et al., 2009), indicating that lipid mixing or stalk formation occur after Ca2<sup>+</sup> influx (**Figure 2C**). However, this model has not been verified with in vivo results.

Interestingly, it has been recently shown that SNARE zippering of only the N-terminal half could drive hemifusion (**Figure 2A**; Yang et al., 2010). SNARE complex formation can be arrested at the half-zippered state by a flavonoid myricetin, and it is found that the state arrested by myricetin corresponds to the hemifusion in proteoliposome fusion. Myricetin blocks C-terminal zippering by binding to the middle region, while allowing SNARE zippering at the N-terminal region. Remarkably, all hemifused vesicles arrested by myricetin were completely converted to full fusion when the myricetin clamp is removed by an enzyme laccase and the fusion reaction was triggered by Ca2+. The results imply that the hemifusion is likely to be an on-pathway intermediate in Ca2+-triggered exocytosis (Heo et al., 2016). Further, the study raises the strong possibility that the hemifusion is induced by the N-terminal half-zippered SNARE complex.

One might now wonder how hemifusion is possible despite SNAREs being only half-zippered at the N-terminal region. Under these conditions, the SNARE complex would not be forcing apposition of two membranes due to the flexibility at the C-terminal half. Alternatively, the highly basic juxtamembrane regions might play a role here. In fact, electrostatic stitching of two negatively charged membranes by the polybasic sequence has been previously proposed as a potential mechanism for membrane merging (Williams et al., 2009).

Purely energetically speaking, it was recently shown that only one SNARE complex may be sufficient for hemifusion (van den Bogaart et al., 2010; Shi et al., 2012). Partial SNARE zippering generates about 35 kBT, corresponding closely to the energy needed for hemifusion (Li et al., 2007). Half-zippering releases 26 kBT in the presence of membranes and 35 kBT in the presence of the pre-assembled C-terminal domains (Gao et al., 2012; Zorman et al., 2014). Because there are multiple SNARE complexes in the synaptic vesicles (Lang et al., 2001; Takamori et al., 2006), it is possible that hemifusion is efficiently achieved by multiple half-zippered SNARE complexes.

#### HEMIFUSION IN THE ENDOCYTOTIC PATHWAY

In the endocytotic pathway, vesicles are created by membrane fission. Membrane fission is a topologically opposite process to membrane fusion. But, this reaction might as well transit through the hemifusion. A GTPase dynamin is the best-studied membrane fission protein. The dynamin protein binds to the neck of the spherical membrane sac bulged from the plasma membrane (**Figure 3**). The GTP-driven

hemifission intermediate formed during endocytosis. (C) The membrane sac is detached from the plasma membrane to form an endocytic vesicle.

conformational change of dynamin constricts the neck to detach the vesicle from the plasma membrane (Antonny et al., 2016).

In a clever experiment using a tubular membrane capillary, Bashkirov et al. (2008) have shown that dynamin-induced membrane fission is leak-free, representing the fission through hemifusion (Bashkirov et al., 2008). Here, it is clear that dynamins remain on the periphery and are not integral parts of the intermediate. Thus, the role of the protein is clearly defined as the mechanical energy source that drives the remodeling of the membrane.

Recently, hemifusion and hemifission structures were observed in live cells, which provided strong evidence of the hemifusion model against that of a TMD-lined pore. Zhao et al. (2016) observed membrane fusion directly in live chromaffin cells in real time using super-resolution stimulated emission depletion microscopy. They observed a Ω-shaped hemifusion structure in the live cell, adding further evidence that the hemifusion indeed exists along both the fusion and fission pathways.

#### PERSPECTIVES

Although exocytotic membrane fusion was initially considered to traverse the TMD-lined fusion pore, evidence has accumulated to support an alternative pathway through the lipidic hemifusion

#### REFERENCES


and fusion pore. The fusion pathway through the hemifusion appears now to be shared by many membrane fusion systems including viral-cell and intracellular membrane fusion. Not surprisingly, the common hemifusion intermediate is shared by endocytotic membrane fission as well, where no TMD is directly involved in the process. The core feature of the hemifusion is that it is lipidic in nature, although some regulatory participation of TMDs cannot be ruled out. In this model, the proteins may stay at the periphery, corralling lipids at the fusion center to undergo merging. As cryo EM and other imaging methods are making significant strides in improving resolution, we are sure that the mechanistic models described here will be tested in the very near future.

#### AUTHOR CONTRIBUTIONS

D-HK, BK and Y-KS wrote the article. All authors reviewed the manuscript.

#### FUNDING

This work was supported by the National Institutes of Health GM05290 and 5U54GM087519. Korea Healthcare Technology R&D Project, Ministry of Health and Welfare, Republic of Korea (Grant No.: HN14C01010000).


wedging small hydrophobic molecules into the SNARE zipper. Proc. Natl. Acad. Sci. U S A 107, 22145–22150. doi: 10.1073/pnas.1006899108


fusion during neurotransmitter release. Neuron 80, 470–483. doi: 10.1016/j. neuron.2013.09.010

Zorman, S., Rebane, A. A., Ma, L., Yang, G., Molski, M. A., Coleman, J., et al. (2014). Common intermediates and kinetics, but different energetics, in the assembly of SNARE proteins. Elife 3:e03348. doi: 10.7554/eLife.03348

**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Kweon, Kong and Shin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Planar Supported Membranes with Mobile SNARE Proteins and Quantitative Fluorescence Microscopy Assays to Study Synaptic Vesicle Fusion

Volker Kiessling1,2\*, Binyong Liang1,2 , Alex J. B. Kreutzberger 1,2 and Lukas K. Tamm1,2

<sup>1</sup>Center for Membrane and Cell Physiology, University of Virginia, Charlottesville, VA, USA, <sup>2</sup>Department of Molecular Physiology and Biological Physics, University of Virginia, Charlottesville, VA, USA

Synaptic vesicle membrane fusion, the process by which neurotransmitter gets released at the presynaptic membrane is mediated by a complex interplay between proteins and lipids. The realization that the lipid bilayer is not just a passive environment where other molecular players like SNARE proteins act, but is itself actively involved in the process, makes the development of biochemical and biophysical assays particularly challenging. We summarize in vitro assays that use planar supported membranes and fluorescence microscopy to address some of the open questions regarding the molecular mechanisms of SNARE-mediated membrane fusion. Most of the assays discussed in this mini-review were developed in our lab over the last 15 years. We emphasize the sample requirements that we found are important for the successful application of these methods.

#### Edited by:

Jiajie Diao, University of Cincinnati, USA

#### Reviewed by:

Ruoyi Qiu, Stanford University, USA Erdem Karatekin, Yale University, USA Dixon J. Woodbury, Brigham Young University, USA

\*Correspondence:

Volker Kiessling vgk3c@virginia.edu

Received: 23 January 2017 Accepted: 03 March 2017 Published: 16 March 2017

#### Citation:

Kiessling V, Liang B, Kreutzberger AJB and Tamm LK (2017) Planar Supported Membranes with Mobile SNARE Proteins and Quantitative Fluorescence Microscopy Assays to Study Synaptic Vesicle Fusion. Front. Mol. Neurosci. 10:72. doi: 10.3389/fnmol.2017.00072 Keywords: SNARE, supported membrane, fluorescence microscopy, TIRF, synaptic vesicle fusion

#### INTRODUCTION

Neurotransmitter release at the synapse, accomplished by the fusion of synaptic vesicles with the presynaptic membrane, is a fast and highly regulated Ca2<sup>+</sup> dependent process that is catalyzed by the neuronal SNAREs synaptobrevin-2 (Syb2, VAMP-2), SNAP-25 and syntaxin-1a (Syx1a; Jahn and Scheller, 2006; Jahn and Fasshauer, 2012; Rothman, 2014). Despite tremendous progress in this field, we are still missing a molecular timeline that proceeds from docking of synaptic vesicles to the plasma membrane, through priming of the fusion machinery, and eventually to fusion of the two membranes once an action potential reaches the synaptic terminal (Jahn and Fasshauer, 2012; Südhof, 2013). In addition to SNAREs at the core of the fusion machinery, several protein players that regulate the fusion process have been identified and characterized. Munc18, a 68 kDa soluble protein, is essential for fusion (Hata et al., 1993; Verhage et al., 2000) and has been proposed to arrange the SNAREs within a larger acceptor complex (Ma et al., 2013; Baker et al., 2015). Munc13, a 196 kDa protein, is associated with priming of synaptic vesicles (Ma et al., 2013), while complexin (Cpx), a 15 kDa helical protein, has been identified as inhibitor of spontaneous, i.e., Ca2<sup>+</sup> independent, vesicle fusion (Yang et al., 2010) as well as a facilitator of fast synchronized fusion (Xue et al., 2007; Maximov et al., 2009). Synaptotagmin 1 (Syt1), a 47 kDa membrane protein consisting of two C2 membrane binding domains, has been identified as the Ca2<sup>+</sup> sensor for fast synchronized release in hippocampal neurons Kiessling et al. Supported Membranes and Fluorescence Microscopy

(Perin et al., 1990, 1991; Maximov and Südhof, 2005). The exact role of these proteins and the nature of the events that constitute the intermediates of the fusion machinery during docking, priming and onset of fusion are highly debated. While the importance of the above proteins could be assessed by innovative in vitro and in vivo experiments, the role of lipids and their interactions with proteins has not been explored in comparable detail. Recombinant protein in combination with model membranes has been an important tool to investigate membrane fusion in vitro. Starting with the initial SNARE-mediated liposome fusion assay introduced by Weber et al. (1998) and Rothman (2014) that helped formulate the current version of the ''SNARE hypothesis'', reconstitution experiments that try to measure the fusion reaction itself have received a lot of attention. However, to dissect the molecular details of the docking, priming and fusion steps, the underlying affinities and concentration dependencies between subsets of the relevant proteins and lipids as well as their complex interactions should be mapped and quantitatively controlled. One route that we have taken over the last years is to exploit fluorescence microscopy assays in combination with planar supported membranes (Tamm, 1984; Tamm and McConnell, 1985). The planar geometry of the sample allows the application of sophisticated fluorescence microscopy methods to probe the dynamics, structure and function of lipids and proteins within the lipid bilayer (Tamm and Kalb, 1993). Starting with the necessary sample requirements and preparations, we will describe fluorescence assays that we developed and applied to address some of the puzzling questions that remain around the synaptic vesicle fusion mechanism.

#### SAMPLE PREPARATION

While a thorough discussion of this subject goes beyond the scope of this review, we want to emphasize some characteristics of the sample preparation techniques that we found were most important for meaningful applications.

#### Supported Membranes

There are two fundamentally different techniques for supported lipid bilayer formation: direct vesicle fusion and a combined Langmuir-Blodgett transfer/vesicle fusion (LB/VF) method (Tamm, 1984; Kalb et al., 1992). Perhaps due to its simplicity, most research groups have employed the direct vesicle fusion method. However, integral membrane proteins in this type of bilayers are usually not laterally mobile, and membrane fusion observed in this system often does not reproduce physiological or other in vitro assays (Bowen et al., 2004; Fix et al., 2004; Liu et al., 2005). In the LB/VF method, first, a lipid monolayer is transferred from the air-water interface of a Langmuir trough onto a clean substrate, then liposomes that might contain protein fuse with this supported monolayer to form the second leaflet of the final bilayer (**Figure 1A**). Two properties of the resulting supported membrane are essential for the assays discussed below. First, reconstituted SNARE proteins are highly oriented

FIGURE 1 | Sample preparation and requirements for supported membrane—SNARE applications. (A) Reconstitution of trans-membrane proteins into supported membranes is accomplished by a two-step technique. In step i, a lipid monolayer is transferred from the water-air interface of a Langmuir-Blodgett trough onto an appropriate hydrophilic substrate. In step ii, protein-containing liposomes are fused with the monolayer in a flow-through chamber to assemble the second leaflet of the supported bilayer and to incorporate membrane proteins. (B) Syntaxin-1a (Syx1a), reconstituted by the technique pictured in (A) is oriented with its SNARE motif facing away from the substrate. Incubation of labeled Syx1a (Alexa 546 at residue 192) with Co2<sup>+</sup> results in ∼90% fluorescence quenching while the same application with a symmetrically distributed rhodamine labeled lipid results in ∼50% fluorescence quenching (Liang et al., 2013). (C,D) Formation of 1:1 Syx1a:SNAP-25 complex in DPC as demonstrated by ion-exchange chromatography and SDS-PAGE. Equal molar amounts of Syx1a and SNAP25 are mixed and incubated overnight in DPC before ion-exchange purification. (C) MonoQ elution profile: the blue trace shows UV absorption (left axis) and the red trace shows the eluted buffer conductivity (right axis). The red vertical lines at the bottom denote collected fractions, and fractions run on SDS-PAGE gels are labeled with corresponding capital letters. (D) SDS-PAGE of protein samples purified by MonoQ column chromatography. Since SNAP-25 is about twice the molecular mass of syntaxin (residues 183–288), the SNAP-25 band is twice as strong as the Syx1a band when they are in a molar ratio of 1:1 (Kreutzberger et al., 2016). (A) is reprinted from Kiessling et al. (2015a) with permission from Elsevier. (C,D) are reprinted from Kreutzberger et al. (2016) with permission from Elsevier.

with their cytosolic domain facing away from the substrate (**Figure 1B**). Consequently, they are highly mobile in the plane of the bilayer, even at relatively high protein to lipid ratios (Wagner and Tamm, 2001; Kiessling and Tamm, 2003; Liang et al., 2013). Second, the two step preparation procedures allow the assembly of asymmetric lipid bilayers (Crane et al., 2005; Kiessling et al., 2006). The LB/VF method can also be used to introduce a polymer cushion between substrate and supported membrane to increase mobility of large membrane proteins (Wagner and Tamm, 2000). It is important to point out that for SNAREs with their small C-terminal domain, this polymer is not required (Wagner and Tamm, 2001; Domanska et al., 2009). For a detailed discussion of the supported membrane preparation and its advantages see also two recent reviews (Kiessling et al., 2015a,b).

#### Membrane Proteins

Both Syb2 and Syx1a are integral membrane proteins which are anchored to vesicle or target membranes, respectively, through a single C-terminal transmembrane helix. Both proteins are prone to aggregation in membranes, especially Syx1a. The effect of Syx1a clustering in the membrane has been explored extensively and its biological implication has been a hotly debated topic (van den Bogaart et al., 2013). Although Syx1a is routinely purified with different types of detergents, including octyl-betaglucoside, cholate, or CHAPS, we have recently discovered that dodecylphosphocholine (DPC) can ensure the monomeric form of Syx1a (Liang et al., 2013). Furthermore, by employing DPC during the assembly of a Syx1a:SNAP-25 acceptor complex, we are able to prepare this complex in a strict 1:1 stoichiometry, which results in a highly active acceptor SNARE complex (**Figures 1C,D**; Kreutzberger et al., 2016).

Nascent SNAP-25 is a soluble protein that becomes post-translationally modified and a peripheral membrane protein by palmitoylation of four cysteines in the linker region between the two SNARE motifs. Previously, most in vitro SNARE fusion studies have employed the soluble form of SNAP-25. We have engineered a quadruply dodecylated SNAP-25 through disulfide binding of a dodecyl methanethiosulfonate precursor to the four native cysteines of SNAP-25. This membrane-associated form of SNAP-25 can be employed to form a highly active target acceptor SNARE complex in liposomal or supported membranes that mediates fast and efficient fusion with Syb2-containing proteoliposomes (Kreutzberger et al., 2016).

For some of the fluorescence assays described here, labeling of membrane proteins with a fluorophore is an important aspect of sample preparation. Although membrane fusion can be observed with lipid and/or content labels, the labeling of the proteins themselves offer unique information, such as proteinprotein interaction by FRET or membrane protein orientation by fluorescence interference contrast (FLIC). Protein labeling can be achieved through the genetic addition of a fusion protein, such as GFP, to the N- or C-terminus. Lately, thiol-reactive fluorophores have gained major attraction due to their small sizes, high quantum yield, photo stability, many color options and site-specificity by the relatively easy introduction of single cysteine mutations. The labeling efficiency of this reaction usually approaches unity if samples are handled in an oxygen-free environment. We have routinely labeled membrane proteins before the removal of their purification tags. The labeled samples are then rebound to the affinity medium and subjected to extensive washing to remove unreacted free labels (Liang et al., 2013).

#### MEASURING INTERACTIONS WITHIN MEMBRANES BY FLUORESCENCE RECOVERY AFTER PHOTO-BLEACHING (FRAP) AND SINGLE PARTICLE TRACKING (SPT)

How does the lipid content of membranes interact with their protein content? This is probably the central question for determining what leads to membrane fusion during exocytosis. Systematic measurements of the diffusion behavior of lipids and proteins in membranes is one technique that can elucidate some of these critical interactions.

#### Diffusion of Integral Membrane Proteins

Observing the recovery of fluorescent content in a certain membrane region which was depleted of all or most fluorescence by application of a strong laser pulse is a standard method to probe transport or diffusion mechanisms in biological and model membranes. The planar geometry of supported membranes make them especially suitable for patterned fluorescence recovery after photo-bleaching (FRAP; **Figure 2A**), an easy to apply and analyze version of FRAP that allows the determination of mobile fractions and diffusion coefficients (Smith and McConnell, 1978; Tamm and Kalb, 1993).

In the first supported membrane assays with reconstituted SNARE proteins, Wagner and Tamm (2001) measured the diffusion of Alexa 488 labeled t-SNAREs at different concentrations of the anionic lipids phosphatidylserine (PS) and phosphatidylinositol-4,5-bisphosphate (PIP2). While the protein is highly mobile at relatively high concentration (l/p 400) in pure phosphocholine bilayers, increasing amounts of PS or PIP<sup>2</sup> render large portions of t-SNAREs immobile. The lipid content, although to a lesser degree, also gets partially immobilized. Noteworthy is the much stronger effect that PIP<sup>2</sup> has on lipid and protein mobility compared to PS indicating a strong interaction between PIP<sup>2</sup> and the t-SNARE Syx1a. While the exact reason for the observed PIP<sup>2</sup> and PS dependence of the mobility hasn't been determined, it was later found that Syx1a clustering in model membranes also depends strongly on the lipid environment, especially on the cholesterol, PS and PIP<sup>2</sup> content (Murray and Tamm, 2009; van den Bogaart et al., 2011).

#### Diffusion of Peripheral Membrane Protein Domains

Progress in imaging technology over the last 20 years, especially in electron multiplying charged coupled devices (EMCCD) and the latest complementary metal oxide semiconductor (CMOS) cameras, allows the tracking of individually labeled molecules. Again, the planar geometry of the supported membrane simplifies the measurement and interpretation of molecule trajectories in or at the membrane (**Figure 2B**). We routinely use single particle tracking (SPT) to confirm the mobility of reconstituted membrane proteins (Domanska et al., 2009).

Synaptotagmins are the calcium sensors of exocytotic neurotransmitter release. Their two C2 domains interact with the lipid bilayer in a calcium dependent fashion which successively triggers membrane fusion and pore opening through a still debated mechanism. A detailed characterization of the C2 domain-lipid interactions is therefore of great interest. Vasquez et al. (2014) used SPT to study the diffusion behavior of single and tandem C2 domains of Syt7, the calcium sensor that triggers slow asynchronous release (Schonn et al., 2008). Their results are consistent with C2 domains that interact

FIGURE 2 | Supported membrane—fluorescence microscopy assays. (A) Fluorescence recovery after photobleaching (FRAP) records the recovery of fluorescence due to lateral diffusion in a region of interest in the membrane after application of a strong laser pulse. The total intensity of an area is recorded before and after the bleach pulse has been applied

(Continued)

(Smith and McConnell, 1978). The example graph on the right shows the recovery of Alexa488 labeled t-SNAREs in supported membranes (Wagner and Tamm, 2001). (B) Single particle tracking (SPT). The movement of membrane components labeled with a single fluorophore are tracked within the x/y plane of the lipid bilayer. The statistically analysis of the trajectories can quantify the lateral mobility as well as reveal different modes of diffusion (Schmidt et al., 1995; Kiessling et al., 2006; Vasquez et al., 2014). The example image on the right shows single Alexa647 labeled t-SNAREs in inverse contrast. Movies of the moving protein can be seen on the original publisher's website (Domanska et al., 2009). (C) Binding assay using total internal reflection (TIRF) microscopy. A totally reflected laser beam produces an exponentially decaying electric field at the glass/water interface. Fluorescent molecules or organelles that bind at the membrane surface increase the observable fluorescent intensity I over time (Kalb et al., 1989). Binding isotherms can be determined with various ligand concentrations and acceptor densities. Data on the right shows SNARE-specific binding of Alexa546 labeled Syb2(1-96) to a supported membrane (Domanska et al., 2009). (D) Single vesicle fusion TIRF assay. Fluorescence originating from the membrane (red, IM) or the content (green, IC) of single vesicles can be imaged when they enter the evanescent field of a TIRF microscope (Fix et al., 2004; Domanska et al., 2009). Characteristic intensity traces from docking vesicles can be analyzed to determine docking and fusion efficiencies as well as fusion kinetics (Kiessling et al., 2015b). Data on the right shows fluorescence originating from the membrane and content during a single vesicle fusion event (Kiessling et al., 2015a). (E) Distance measurements by FLIC microscopy. A Si/SiO<sup>2</sup> substrate with different steps is used to probe the interference pattern originating from reflected excitation and emission light (Braun and Fromherz, 1997). Fitting the optical theory to the measured intensities I from different SiO<sup>2</sup> layers allows the determination of the distance of specifically labeled protein residues from the lipid bilayer surface (Lambacher and Fromherz, 2002; Kiessling and Tamm, 2003). Data on the right shows a FLIC curve obtained from Alexa546 labeled Syx obtained under the same conditions as published in Liang et al. (2013). Data in (A) is reprinted from Wagner and Tamm (2001) with permission from Elsevier. Data in (B,C) was originally published in Domanska et al. (2009), © the American Society for Biochemistry and Molecular Biology. Data in (D) is reprinted from Kiessling et al. (2013) with permission from Elsevier.

independently with the lipid bilayer, which is in contrast to Syt1, the calcium sensor for fast synchronous release, for which interdomain cooperativity had been reported (Sun et al., 2007).

### TOTAL INTERNAL REFLECTION FLUORESCENCE (TIRF) MICROSCOPY

The planar geometry of supported membranes make them an excellent candidate for the application of total internal reflection fluorescence (TIRF) microscopy, where a beam of light gets totally reflected at the interface between the (glass-) substrate and water (Axelrod et al., 1983). Fluorescent molecules get excited by the evanescent electric field that decays exponentially, typically with a characteristic penetration depth of ∼100–200 nm.

#### Interactions with the Membrane: Binding

Supported membranes and TIRF microscopy are the ideal platform to record kinetic binding curves or binding isotherms of proteins and other membrane interacting molecules (**Figure 2C**; Kalb et al., 1990). The soluble cytosolic fragment of Syb2 tagged with green fluorescent protein (GFP) was shown to specifically bind to supported membranes containing reconstituted t-SNAREs. The resulting ternary SNARE complex could then be dissociated by the addition of NSF, α-SNAP and ATP mimicking the physiological SNARE assembly and disassembly cycle (Wagner and Tamm, 2001). These experiments represent the first functionally successful reconstitutions of SNARE proteins in supported membranes described in the literature. Binding studies with the soluble C2 domains of Syt and phase separated asymmetric supported bilayers revealed a preference of these domains in binding to liquid-disordered lipid domains that are enriched in anionic lipids over liquid-ordered domains. The same study also showed that Ca2<sup>+</sup> dependent binding of C2 domains can change the lipid partitioning between the two phases (Wan et al., 2011). Reconstituting Syt1 into a planar bilayer allowed monitoring the Ca2<sup>+</sup> and concentration dependent capture of negatively charged liposomes (Lu et al., 2014). Planar bilayers have also been used to investigate the role of the protein calcium activated protein for secretion (CAPS) on docking of Syb2 containing liposomes. CAPS, which has also been implied with docking and priming, promoted increased binding of Syb2 liposomes to planar bilayers containing PIP<sup>2</sup> and Syx1a (James et al., 2009).

#### Membrane-Membrane Interactions: Single Vesicle Fusion

The gold standard for in vitro SNARE experiments is to mimic the physiological vesicle fusion reaction itself. Single vesicle experiments have received increasing attention because they allow the independent observation of the initial docking event from the actual fusion event. While single vesicle assays can be implemented with immobilized proteoliposomes (Yoon et al., 2006; Kyoung et al., 2011; Diao et al., 2013), here, we will only discuss vesicle-supported membrane fusion assays (**Figure 2D**). The different types of commonly used single vesicle fusion assays have been discussed in Kiessling et al. (2015a).

Initial attempts of single vesicle to planar supported bilayers fusion assays resulted in fusion reactions that did not require SNAP-25 (Bowen et al., 2004; Liu et al., 2005), were Ca2<sup>+</sup> dependent in the absence of Syt (Fix et al., 2004) or resulted in vesicle rupture instead of fusion (Wang et al., 2009). In all these experiments the direct vesicle fusion supported membrane preparation method was used and it raised the question if supported membranes were at all suited for SNARE-mediated fusion assays. Utilizing supported membranes prepared by the two-step LB/VF method, we could record fusion events that mimic the physiological SNARE requirement. Here, docking depends on the presence of Syx1a and SNAP-25 in the supported membrane and fusion occurred within tens of milliseconds (Domanska et al., 2009). The fusion reaction was followed by the transfer of fluorescently labeled lipids from Syb2 containing liposomes into the planar membrane. The fast change (∼8 ms) of the fluorophores' dipole orientation relative to the polarization of the evanescent field when the labeled lipids transfer from the spherical vesicle membrane to the planar supported membrane creates a signature signal for the onset of fusion (Kiessling et al., 2010) that can be used to accurately determine the fusion kinetics. Analysis of the fusion kinetics measured in a 1-palmitoyl-2-oleoyl-sn-glycero-3 phosphocholine (POPC)/cholesterol lipid environment revealed that eight parallel reactions had to take place between docking of vesicles and the onset of fusion (Domanska et al., 2009). It is plausible that these reactions are the formation of eight SNARE complexes at the fusion site. Karatekin et al. (2010) later found similar requirements for the number of SNAREs, although the experimental conditions were very different. In that work, a very low concentration of t-SNAREs were reconstituted into a polyethylene-glycol (PEG) containing supported membrane and the fusion delay times were simulated with a diffusion reaction model. We later showed that the fusion kinetics strongly depend on the lipid environment. Systematic introduction of up to 30 mol% phosphoethanolamine (PE) and 5 mol% PS into the vesicle and/or the supported membrane reduced the number of parallel reactions, i.e., the number of necessary SNARE complexes to 3 (Domanska et al., 2010). The variability of the SNARE cooperativity in purely SNARE mediated membrane fusion was further examined by combining bulk and single vesicle fusion assays. While small (∼40 nm) highly curved liposomes were fusogenic with only one Syb2/liposome, large (∼100 nm) liposomes needed 23–30 Syb2/liposome for efficient fusion (Hernandez et al., 2014). Observing the transfer of content dye from liposomes to the small cleft between substrate and supported membrane confirmed productive fusion (Kiessling et al., 2013) and allowed the easy distinction between fulland hemi-fusion events. Increasing amounts of cholesterol in either the supported membrane or the vesicle membrane shift the balance from unproductive hemi-fusion to productive full-fusion (Kreutzberger et al., 2015) and might increase the stability of the fusion pore (Stratton et al., 2016). While cholesterol can influence membrane fusion in many different ways (Yang et al., 2016), we attributed our observation to the stabilization of intermediates by the intrinsic curvature of cholesterol (Kreutzberger et al., 2015). The physiological relevance of this assay was validated when we employed a hybrid system consisting of purified synaptic vesicles from rat brain and a supported membrane containing recombinant plasma membrane SNAREs. Synaptic vesicles and liposomes containing Syb2 and Syt1 showed increased fusion efficiencies in the presence of Ca2<sup>+</sup> and anionic lipids (Kiessling et al., 2013).

#### PROBING PROTEIN CONFORMATIONS IN THE MEMBRANE: FLUORESCENCE INTERFERENCE CONTRAST (FLIC) MICROSCOPY

FLIC microscopy is an interferometric fluorescence microscopy method that measures the distance of a planar fluorescent layer normal to a reflective interface and was originally developed to measure the distance of adhering cell membranes from a substrate (Lambacher and Fromherz, 1996; Braun and Fromherz, 1997, 1998). The interference contrast originates from the excitation and emission lights from fluorophores in front of a mirror at different distances. In practice, the mirror is implemented by the Si/SiO<sup>2</sup> interface of a patterned Si wafer (Braun and Fromherz, 1997, 1998; Kiessling and Tamm, 2003; Liang et al., 2013). Once the sample is prepared on the FLIC substrate a relatively simple epifluorescence microscope with a lamp as excitation source is sufficient to record the images from which the intensities of the different terraces are extracted (**Figure 2E**). The optical theory that takes into account the spectra of excitation and emission of microscope, detector and fluorophore, the numerical aperture of the objective, the dipole orientation and quantum yield of the fluorophore, and the optical layer system that includes the supported membrane is fit to the data in order to get the desired z-distance (Lambacher and Fromherz, 2002). The accuracy of the resulting z-distance is determined by systematic errors caused by assumptions or measurements of the different optical layer thicknesses (oxide, water layer, membrane) and statistical errors caused by the fit and sample inhomogeneity.

Although the first application of FLIC was still limited by the available substrates and the relatively large size of the label, the distance of a N-terminus GFP label in a Syb2 construct within a SNARE complex was consistent with the crystal structure of a cis-SNARE complex published later (Kiessling and Tamm, 2003; Stein et al., 2009). More recently we utilized Alexa labeled proteins to perform site-directed FLIC microscopy. Different label positions allow the determination of an accurate picture of the orientation of the protein with respect to the lipid bilayer. With these bright labels the uncertainty of the absolute distance lies in the order of 1–2 nm. Distance changes that might occur after the protein of interest interacts with ligands that have been added to the sample, can be measured with sub 1 nm accuracy.

When reconstituting labeled integral plasma membrane proteins, we can take advantage of the supported membrane preparation procedure described above to achieve correctly oriented proteins as well as asymmetric lipid compositions between the two leaflets of the bilayer. Recently, by determining the distances of two residues at the N-terminus or the center of the Syx1a SNARE motif from the membrane surface, we showed that the cytoplasmic domain of monomeric Syx1a lies at the bilayer surface. Interestingly, complex formation within a stabilized acceptor SNARE complex consisting of Syx,

#### REFERENCES


SNAP25 and Syb2(49–96) or within a ternary SNARE complex results in a more upright position (Liang et al., 2013).

#### OUTLOOK

To solve the long-standing questions about how the secretory vesicle fusion machinery achieves its precise and reliable function, many different approaches must be utilized. A lot has been learned about the molecular requirements from in vivo and cell experiments and many mechanistic pictures have been drawn based on atomic resolution structures of soluble proteins or protein fragments. However, we need carefully controlled in vitro reconstitution experiments that are performed in the presence of defined lipid bilayers to gain further insights. The above discussed fluorescence assays with supported membranes are, in our view, among the most promising routes forward because they can deliver quantitative biochemical and functional data as well as structural information without neglecting important characteristics of the biological membrane like protein orientation, mobility and lipid asymmetry. Correlating the results of these different assays and considering that the samples with almost identical conditions can be prepared for a range of different structural and functional assays further enhances their value. We are very optimistic that the described approaches will soon deliver significant contributions to solve many of the remaining mysteries about the precise molecular interactions that underlie the mechanism of Ca2+-triggered exocytosis and synaptic vesicle fusion.

#### AUTHOR CONTRIBUTIONS

VK, BL, AJBK and LKT wrote the article.

#### FUNDING

NIH grant P01 GM72694.

#### ACKNOWLEDGMENTS

We thank all former and present members of the Tamm lab that contributed to the assays described in this review article.


the acceptor SNARE complex-containing target membrane. Biophys. J. 99, 2936–2946. doi: 10.1016/j.bpj.2010.09.011


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Kiessling, Liang, Kreutzberger and Tamm. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Neurodegenerative Disease Related Proteins Have Negative Effects on SNARE-Mediated Membrane Fusion in Pathological Confirmation

Chen Hou, Yongyao Wang, Jiankang Liu\*, Changhe Wang and Jiangang Long\*

*Center for Mitochondrial Biology and Medicine and Key Laboratory of Biomedical Information Engineering of the Ministry of Education, School of Life Science and Technology, and Frontier Institute of Science and Technology, Xi'an Jiaotong University, Xi'an, China*

Keywords: neurotransmitter release, membrane fusion, neurodegenerative diseases, single molecule biophysics technology, SNARE

Studies showed that synapses are highly-specialized structures for the communication between preand postsynaptic neurons (Kaeser and Regehr, 2014). Synaptic vesicles carrying neurotransmitters dock at specialized sites of presynaptic membranes termed active zones, which are closely apposed to postsynaptic densities, and then undergo one or more priming reactions to prepare them to a release-ready state. When an action potential invades the nerve terminals, the following membrane depolarization activates voltage-gated calcium channels to influx Ca2+, thus initiates the fusion of synaptic vesicles with presynaptic membrane and transmitters release (Wu and Saggau, 1997).

#### Edited by:

*Cong Ma, Huazhong University of Science and Technology, China*

#### Reviewed by:

*Ruoyi Qiu, Stanford University, USA Jiuwei Lu, University of Cincinnati, USA*

#### \*Correspondence:

*jiankang Liu j.liu@mail.xjtu.edu.cn Jiangang Long jglong@mail.xjtu.edu.cn*

Received: *17 January 2017* Accepted: *27 February 2017* Published: *21 March 2017*

#### Citation:

*Hou C, Wang Y, Liu J, Wang C and Long J (2017) Neurodegenerative Disease Related Proteins Have Negative Effects on SNARE-Mediated Membrane Fusion in Pathological Confirmation. Front. Mol. Neurosci. 10:66. doi: 10.3389/fnmol.2017.00066*

It was reported that synaptic vesicle fusion requires assembly of a conserved proteins family termed soluble N-ethylmaleimide-sensitive factor attachment protein receptors (SNAREs) (Chernomordik and Kozlov, 2008; Wickner and Schekman, 2008). All SNAREs contain an evolutionarily conserved coiled-coil SNARE motif of ∼60–70 amino acids that are arranged in heptad repeats. In synapses, syntaxin and synaptosome-associated protein of 25 kDa (SNAP-25, contains two SNARE motifs) on the plasma membrane (t-SNAREs on target membrane) and synaptobrevin/vesicle-associated membrane protein (VAMP) on synaptic vesicles (v-SNARE) assemble into a tight trans-SNARE complex in a 1:1:1 ratio to bridge synaptic vesicles and the plasma membrane (Brunger, 2005). The trans-SNARE complex promotes membrane fusion by pulling the bilayers together as it zippers up, and the remaining SNARE complexes on the fused membrane are transformed to cis-configuration with lower potential energy, which undergoes disassembly catalyzed by a specialized adenosine triphosphatase (ATPase) N-ethylmaleimidesensitive factor (NSF) and its cofactors soluble NSF attachment proteins (SNAPs) (Jahn et al., 2003). SNAPs bind directly to the SNARE complex, then recruit and activate NSF to completely dissociate SNARE complex and recycle individual SNAREs for a new round of fusion reactions (Sudhof and Rothman, 2009). Thus, it appears that the cycle of SNARE assembly and disassembly is critical for the occurrence, the fidelity and plasticity of synaptic transmission.

Meanwhile, it has been shown that a wide range of neurodegenerative disorders are characterized with neuronal dysfunction and neuron loss, which caused by the aggregation of specific neurotoxic proteins (Caughey and Lansbury, 2003). Typically, α-synuclein (α-syn), constitutes the amyloid fibril form of Lewy bodies (Wang et al., 2016), is a cytosolic neural protein consisting of 140 amino acid residues and is abundantly expressed in presynaptic membrane in monomeric form (Burre et al., 2013). α-syn is closely associated with early-onset of neurodegenerative diseases prominently in familial Parkinson's disease, Alzheimer's disease and Lewy body disease. Aβ, a peptide of 36–43 amino acids, is a key molecule in the pathogenesis of Alzheimer's disease, and Aβ deposition is the necessary prerequisites for synaptic dysfunction and cognitive impairment in Alzheimer's disease (Annaert and De Strooper, 2002). Studies showed that, in neurodegenerative diseases, the neurotransmission was profoundly damaged, and the SNARE protein function and distribution were changed (Garcia-Reitbock et al., 2010; Shen, 2010). Thus, we propose that these proteins in pathological confirmation play roles in SNARE-mediated membrane fusion during neurotransmission by disrupting the SNARE protein assembling and recycling.

Recent research showed that α-syn can be classified into spiral membrane binding form, part of folding state, oligomers, fibrous, and amorphous polymers, etc according to the molecular formation after polymerization (Breydo et al., 2012). It has been verified that α-syn oligomerization contributes to the increased cytotoxicity and thus promotes the dopaminergic neuronal degeneration (Hogen et al., 2012). However, native α-syn shows no damage effect on the efficiency of synaptic vesicle exocytosis, but helps to increase the availability of synthetic vesicles at the synapse (Diao et al., 2013a). In addition, α-syn knockout shows little effect on synaptic transmission (Nemani et al., 2010), while overexpression of α-syn reduces neurotransmitter release by disturbing vesicle docking in exocytosis (Larsen et al., 2006). The possible mechanisms may be due to the reduction of synaptic vesicle recycling pool size, the reduced synaptic vesicle density at the active zone and the defects in re-clustering of synaptic vesicles upon α-syn overexpression. Moreover, exhibiting the supportive role in the folding/refolding of SNARE proteins, study showed that α-syn acts as a non-classical chaperone that facilitates the maintenance of proper SNARE states during SNARE cycle, and promotes SNARE complex assembly by directly binding to synaptobrevin-2/VAMP2 (Burre et al., 2010). For example, monomeric α-syn contributes to neural vesicle aggregation by simultaneously interacting with synaptobrevin-2 (Diao et al., 2013a), however, α-syn oligomers inhibits vesicle docking through interaction with synaptobrevin-2 and negatively charged phospholipids (Choi et al., 2013; Hu et al., 2016). Therefore, it appears that the effects of α-syn on neurotransmission are mainly determined by forms of α-syn polymerization.

Cysteine string protein α (CSPα), a co-chaperone protein, also plays an important role in maintaining SNARE rapid cycling and neuronal activity (Garcia-Junco-Clemente et al., 2010). There were several reports that CSPα expression is greatly decreased in the forebrain from patients with neurodegenerative disorders (Tiwari et al., 2015). It was also shown that CSPα can form a chaperone complex with Hsc70 (Chamberlain and Burgoyne, 1997) and SGT protein (Nemani et al., 2010), and the CSPα–Hsc70–SGT complex binds directly to monomeric SNAP-25 to prevent its polymerization, enabling SNARE complex formation. Consistently, CSPα knockout mice show defects in synaptic function that associated with the abnormal formation of SNAP-25 (Fernandez-Chacon et al., 2004). In contrast, overexpression of CSP suppresses the degradation of SNAP-25 under normal physiological condition (Sharma et al., 2011). Dysfunctional SNAP-25, in the absence of CSPα, is ubiquitinated and degraded by the proteasome in a synaptic activity–dependent manner, leading to the reduction of SNAP-25 (Sharma et al., 2011). In addition, overexpression of α-syn blocks the CSPα deletion-induced neurodegeneration and ameliorates the CSPα deficiency-induced inhibition of SNARE complex assembly, however, the removal of endogenous α-syn deteriorated CSPα deletion-induced symptoms (Chandra et al., 2005). These phenomena imply that α-syn may cooperate with CSPα to maintain SNARE proteins assembly and neurotransmission.

Furthermore, it has been shown that intracellular Aβ oligomers inhibit SNARE-mediated exocytosis by impairing SNARE complex formation through direct interaction with syntaxin 1a (Yang et al., 2015). Aβ42 is reported to regulate neurotransmitter release, probably by disrupting the complex formation of Synaptophysin and VAMP2 through the competitive interaction with Synaptophysin (Russell et al., 2012). Studies also showed that Aβ oligomers contribute to the down-regulation of synapse density and decreased neurotransmission efficiency (Terry et al., 1991; Shankar et al., 2007). In addition, although aggregations of Aβ and α-syn are used as the major pathological markers of AD and PD respectively, these two pathogenic proteins have synergistic effects on neurodegenerative disorders (Choi et al., 2015). Study showed that Aβ promotes the formation of large-size α-syn oligomers, which function to inhibit SNARE-mediated vesicle fusion, accelerate motor or memory deficits or cognitive dysfunction in APP/PS1 transgenic mice (Yang et al., 2015). It suggested that these two proteins may cooperate to suppress membrane fusion, and that the formation of α-syn aggregates requires the participation of Aβ. Meanwhile, some other pathological proteins in Alzheimer's disease such as amyloid precursor protein, presenilin, phosphorylated tau protein, and brain-derived neurotrophic factor are associated with Aβ deposition and contribute to the regulation of neuronal function in Alzheimer's disease (Saura et al., 2004; Schindowski et al., 2008; Peethumnongsin et al., 2010).

The evidences above support the suggestion that these typical neurotoxic proteins have negative effects on the assembly and disassembly cycle of SNARE proteins, and thus on the SNAREmediated membrane fusion during neurotransmission. In can be implied that SNARE-mediated membrane fusion in a functional state is largely depends on molecular chaperone systems, which is exhibited by α-syn, CSPα, Aβ, etc (**Figure 1**). These proteins directly or indirectly interact with one or more components of SNARE complex, chaperoning and maintaining appropriate SNARE protein complex assembly or disassembly under different pathological conditions.

It was also shown that the mechanism of SNARE-mediated membrane fusion during neurotransmission has been analyzed by real time monitoring SNARE protein interactions with singlemolecule FRET (smFRET) imaging (Weninger et al., 2003, 2008; Bowen et al., 2004, 2005). There are two major smFRET assays applied for SNARE-mediated membrane fusion: Fusion proteins monitoring during fusion processes (Brunger et al., 2009) and the lipid molecule mixing between of the fused vesicles (Diao et al., 2009, 2012). In monitoring proteins, SNARE and its accessory proteins that are site-specifically conjugated with fluorescent dyes may be used to analyze the unique protein structural information (Brunger et al., 2009). In monitoring fusion, the FRET efficiency value from each pair of vesicles may be collected to identify different stages of fusion, such

as docking, hemifusion, and full fusion (Diao et al., 2013b). These supports the suggestion that smFRET approach allows the detection down to the conformation of a single biomolecule with two dyes attached, and that smFRET approach is different from the ensemble fusion assay that averaged FRET signal from the entire population is obtained.

Finally, mechanistic studies of neurotransmitter release contribute significantly to clarify the pathogenesis of neurodegenerative diseases in brain, which will help to illuminate the underlying pathogenic mechanisms for neurodegenerative disorders. In recent years, the effects of these neurotoxic proteins at different stages of SNARE-mediated membrane fusion have been extensively investigated by single molecule biophysics technologies, including smFRET imaging (Brunger et al., 2015). SmFRET techniques will certainly benefit further studies about SNARE-mediated membrane fusion, and benefit the cross-talk investigation between neurodegenerative proteins and SNARE cycling during neurodegeneration. We suggest that future research should be integrated to the real-time monitoring of

#### REFERENCES


membrane fusion of neurotransmitter release in vivo (Sakon and Weninger, 2010; Crawford et al., 2013). This helps for the deeper understanding of neural signaling process and the exploration of new treatment for neurological disorders.

#### AUTHOR CONTRIBUTIONS

CH drafted the manuscript; YW drawed the cartoon; JLi drafted the manuscript; CW revised the manuscript; JLo revised the manuscript.

#### FUNDING

This work was supported by the National Basic Research Program of China (973 Program, 2015CB856302 and 2015CB553602), the opening foundation of the State Key Laboratory of Space Medicine Fundamentals and Application, the Chinese Astronaut Research and Training Center (SMFA15K01), and the National Natural Science Foundation of China (31400708 and 31670843).


inhibiting synaptic vesicle reclustering after endocytosis. Neuron 65, 66–79. doi: 10.1016/j.neuron.2009.12.023


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Hou, Wang, Liu, Wang and Long. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Visualization of SNARE-Mediated Hemifusion between Giant Unilamellar Vesicles Arrested by Myricetin

#### Paul Heo<sup>1</sup> , Joon-Bum Park <sup>1</sup> , Yeon-Kyun Shin<sup>2</sup> and Dae-Hyuk Kweon<sup>1</sup> \*

<sup>1</sup>Department of Genetic Engineering, College of Biotechnology and Bioengineering, Sungkyunkwan University, Suwon, South Korea, <sup>2</sup>Department of Biochemistry, Biophysics and Molecular Biology, Iowa State University, Ames, IA, USA

Neurotransmitters are released within a millisecond after Ca<sup>2</sup><sup>+</sup> arrives at an active zone. However, the vesicle fusion pathway underlying this synchronous release is yet to be understood. At the center of controversy is whether hemifusion, in which outer leaflets are merged while inner leaflets are still separated, is an on-pathway or off-pathway product of Ca<sup>2</sup><sup>+</sup>-triggered exocytosis. Using the single vesicle fusion assay, we recently demonstrated that hemifusion is an on-pathway intermediate that immediately proceeds to full fusion upon Ca<sup>2</sup><sup>+</sup> triggering. It has been shown that the flavonoid myricetin arrests soluble N-ethylmaleimide-sensitive factor (NSF) attachment protein receptor (SNARE) mediated vesicle fusion at hemifusion, but that the hemifused vesicles spontaneously convert to full fusion when the myricetin clamp is removed by the enzyme laccase. In the present study, we visualized SNARE-mediated hemifusion between two SNAREreconstituted giant unilamellar vesicles (GUVs) arrested by myricetin. The large size of the GUVs enabled us to directly image the hemifusion between them. When two merging GUVs were labeled with different fluorescent dyes, GUV pairs showed asymmetric fluorescence intensities depending on the position on the GUV pair consistent with what is expected for hemifusion. The flow of lipids from one vesicle to the other was revealed with fluorescence recovery after photobleaching (FRAP), indicating that the two membranes had hemifused. These results support the hypothesis that hemifusion may be the molecular status that primes Ca<sup>2</sup><sup>+</sup>-triggered millisecond exocytosis. This study represents the first imaging of SNARE-driven hemifusion between GUVs.

Keywords: SNARE, membrane fusion, hemifusion, myricetin, calcium, neurotransmitter release

## INTRODUCTION

Membrane fusion constitutes the final step in the secretion and cargo transfer pathways between cellular compartments (Südhof and Rizo, 2011; Scheller, 2013; Rothman, 2014), and it is also essential in many cellular processes, including autophagy (Wang et al., 2016). When two membranes are approaching each other for fusion, free energy is required to overcome electrostatic repulsive forces, steric hindrances and the hydration force between two membranes. The soluble N-ethylmaleimide-sensitive factor (NSF) attachment protein receptor (SNARE) proteins comprise the molecular fusion machine. SNARE proteins provide the free energy required for fusion during

#### Edited by:

Cong Ma, Huazhong University of Science and Technology, China

#### Reviewed by:

Masahiro Kono, Medical University of South Carolina, USA Zhiqi Tian, University of Cincinnati College of Medicine, USA Tae-Young Yoon, Yonsei University, South Korea

#### \*Correspondence:

Dae-Hyuk Kweon dhkweon@skku.edu

Received: 29 December 2016 Accepted: 20 March 2017 Published: 31 March 2017

#### Citation:

Heo P, Park J-B, Shin Y-K and Kweon D-H (2017) Visualization of SNARE-Mediated Hemifusion between Giant Unilamellar Vesicles Arrested by Myricetin. Front. Mol. Neurosci. 10:93. doi: 10.3389/fnmol.2017.00093 the formation of a parallel four-helix bundle called the SNARE complex (Poirier et al., 1998; Sutton et al., 1998). In neurons, there are three SNARE proteins: syntaxin 1a (Stx1) and synaptosome-associated protein 25 (SNAP-25) on the plasma membrane and vesicle-associated membrane protein 2 or Syb2 (VAMP2) on the vesicle membrane. These three SNARE proteins constitute the minimal machinery for fusion between the synaptic vesicle and the presynaptic plasma membrane (Weber et al., 1998). Overwhelming evidence favors the zippering hypothesis, in which SNARE complex formation starts from N-termini and zippers progressively towards the membranes (Melia et al., 2002; Matos et al., 2003; Gao et al., 2012; Lou and Shin, 2016).

Many membrane fusion processes proceed via several sequential intermediates (Kozlov and Markin, 1983; Chernomordik and Kozlov, 2003; Chernomordik et al., 2006; Jahn and Scheller, 2006). When two membranes approach each other, they become locally connected by forming a hemifusion stalk. Proximal leaflets of bilayers are fused, but distal leaflets are separated at this stage. Hemifusion is also shown to be an on-pathway intermediate in SNARE-mediated membrane fusion (Lu et al., 2005; Xu et al., 2005). Subsequently, the stalk expands radially into a hemifusion diaphragm with the distal leaflets remaining separated, though it is possible that hemifusion expansion results in a dead-end product in Ca2+-triggered exocytosis (Diao et al., 2012; Hernandez et al., 2012). Finally, a fusion pore is opened within the hemifusion diaphragm, directly from the hemifusion stalk or from a point of membrane contact. Although hemifusion is considered to be an essential fusion intermediate, its experimental verification and characterization in biological membranes has been very difficult, yielding contradictory results (Zampighi et al., 2006; Wong et al., 2007; Fernández-Busnadiego et al., 2010; Zhao et al., 2016). After identification of the SNARE complex assembly and membrane fusion intermediate, Jahn and Scheller (2006) proposed that straining of lipids at the edge of an extended docking zone initiates fusion (Hernandez et al., 2012). Another cryo-electron microscopy study showed that Ca2+-triggered immediate fusion starts from a point-contact between membranes and proceeds to full fusion without discernible hemifusion intermediates (Diao et al., 2012). In both studies, stable hemifusion diaphragms were kinetically trapped and represented an off-pathway product. However, a study using super-resolution stimulated emission depletion microscopy observed membrane hemifusion directly in live chromaffin cells in real time (Zhao et al., 2016). An -shaped hemifusion structure was observed in the live cells, and it was found that even the 'kiss-and-run' model can be explained by the competition between transitions of hemifusion/hemi-fission to full fusion or to full fission. Recently, we also showed that a stable hemifusion state can proceed to complete fusion and form a fusion pore almost synchronously with Ca2<sup>+</sup> triggering (Heo et al., 2016). When a small molecule flavonoid (myricetin) that halts SNARE zippering in the middle (Yang et al., 2010) was removed from the SNARE complex intermediate with the enzyme laccase, hemifusion proceeded to full fusion. The speed of Ca2+-triggered fusion was comparable to that in neurons, and the pattern of release was reminiscent of the synchronous and asynchronous release of neuroexocytosis depending on the stage of Ca2<sup>+</sup> arrival in the reconstituted systems (Heo et al., 2016). These results provide clear evidence that the hemifusion state is the bona fide intermediate enabling millisecond exocytosis.

Hemifusion mediated by SNARE proteins and myricetin was analyzed again using dynamic light scattering (DLS) spectroscopy (Yang et al., 2015). We verified hemifusion between vesicles in the presence of myricetin by simulating vesicle hydrodynamic radius changes during fusion and by cleaving SNARE proteins with proteinase K. In the present study, we aimed to visualize hemifusion between individual giant unilamellar vesicles (GUVs). GUVs are excellent objects for fluorescence microscopy visualization and analysis because GUV dimensions are larger than light microscopy's intrinsic resolution limit. The mean diameter of GUVs is tens of µm, for which mean and dispersion values are strictly dependent on the method of GUV preparation. GUV size can be made comparable to the plasma membrane of a variety of cells. Membrane fusion processes, including the existence of the protein-free hemifusion diaphragm as a fusion intermediate, can be visualized using GUV and fluorescence microscopy (Lei and MacDonald, 2003; Heuvingh et al., 2004; Nikolaus et al., 2010). GUVs were also elegantly used to visualize the molecular interplay between membranes, accessory proteins and SNARE proteins (Bacia et al., 2004; Tareste et al., 2008; Hui et al., 2009; Malsam et al., 2012). Here, the hemifusion induced by SNARE proteins and myricetin was evaluated with confocal microscopy to directly visualize SNARE-mediated hemifusion.

#### MATERIALS AND METHODS

#### Purification of SNARE Proteins

Neuronal SNAREs from Rattus norvegicus: SNAP-25 (amino acids 1–206), Syb2 (amino acids 1–116) and Stx1 (amino acids 1–288) were expressed in Escherichia coli CodonPlus-RIL (DE3) and purified by a glutathione S-transferase (GST) tag system. In brief, cell pellets were resuspended in PBS (pH 7.4) supplemented with 2 mM 4-(2-aminoethyl)-benzenesulfonyl fluoride, 2 mM ethylenediaminetetraacetic acid (EDTA) and 2 mM dithiothreitol (DTT). After sonication, the supernatant was mixed with GST-agarose beads at 4◦C for 3 h. Excess PBS was used for washing, and each protein of interest was eluted in thrombin cleavage buffer (TCB, 50 mM Tris-HCl and 150 mM NaCl, pH 8.0). For transmembrane proteins, 0.2% Triton X-100 and 0.05% Tween 20 were added to PBS for the lysis and washing steps, and subsequently 1% n-octyl-beta-Dglucopyranoside (OG) was added to TCB instead of Triton X-100 at the elution step. All purified proteins were analyzed by SDS-PAGE and the Bradford assay.

#### Reconstitution into LUV

We used a conventional SNARE reconstitution method to make proteoliposomes. 1-Palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), 1,2-dioleoyl-sn-glycero-3phospho-Lserine (DOPS), 1,2-dipalmitoyl-sn-glycero-3 phosphoethanolamine-N-(7-nitro-2-1,3-benzoxadiazol-4-yl)

(NBD) and 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-(lissaminerhodamine B sulfonyl, Rhod) were purchased from Avanti Polar Lipids Inc. A lipid mixture composed of PC:PS (95:5) was dried with nitrogen gas and further dried under vacuum overnight. For fluorescent GUVs, fluorescent lipids were incorporated into the mixture at the expense of PC. Large unilamellar vesicles (LUV) were formed by extruding the hydrated lipid mixtures through polycarbonate filters with a 100-nm pore size. SNARE proteins were mixed with liposomes at the indicated lipid/SNARE ratio while keeping the 0.8% OG concentration. The proteoliposomes were diluted with dialysis buffer (25 mM HEPES, 150 mM NaCl, pH 7.4) and dialyzed against 2 L dialysis buffer overnight at 4◦C. Residual OG in samples was eliminated by SM2 Bio-Beads at 4◦C for 30 mins. To compare reconstitution efficiency on GUVs depending on the lipid-to-protein (LP) ratio, SNARE proteins labeled with Alexa 647 dye through an amine reaction were used.

#### SNARE-Reconstituted GUV Formation

The GUVs with SNARE proteins were generated by the electroformation method based on previous reports with modifications (Bacia et al., 2004; Tareste et al., 2008; Hui et al., 2009; Malsam et al., 2012). The process began after preparation of SNARE-embedded LUVs. After obtaining LUV pellets by centrifugation at 100,000 g at 4◦C for 2 h, each pellet was resuspended in 10 mL low salt buffer (5 mM HEPES, 5 mM NaCl, pH 7.4). A droplet of the solution was applied to an indium tin oxide (ITO)-coated glass slide. After drying the lipid droplet, two ITO slides were assembled to form a 5 × 5 × 2 mm chamber (**Supplementary Figure S1**). The dried films were rehydrated with 200 mM sucrose while applying a sinusoidal electric field at 0.01 V and 10 Hz for 10 mins. The electric field was increased gradually by 0.1 V per 5 min up to 1.2 V, while maintaining the 10 Hz frequency. Then, 1.2 V (at 10 Hz) was applied for 6 h. Finally, 3 V (at 10 Hz) was applied for 15 min to detach the GUVs from the ITO slide. The SNARE-containing GUV mixture was incubated in 200 mM glucose solution to remove aberrant lipid aggregates. The efficiency of SNARE-containing GUV formation was estimated using fluorescent dye-labeled SNARE proteins and fluorescent lipids. GUVs were used for assays within 1 day of their formation. Note that residual OG strongly inhibited the formation of a dried film on the ITO slides and subsequent GUV formation. GUVs with no membrane proteins were generated following the same protocol as above, except that 2 µl of the lipid mixture in chloroform was directly spread on the ITO slides.

#### Confocal Setup, GUV Fusion Assay and FRAP Assay

For membrane fusion assays, 5 µl of each t- and v-GUV preparation containing the binary acceptor complex (Syx1 and SNAP-25) and Syb2, respectively, were mixed with 5 µl buffer (25 mM HEPES, 400 mM KCl, 1 mM DTT, pH 7.4) and incubated at 37◦C for 40 min. The GUVs deposited on glass slides were observed through a confocal laser-scanning microscope (LSM 700, Zeiss, Germany) equipped with a C-Apochromat 63×/1.2 water immersion objective. After focusing on the focal plane on the bottle-neck of a hemifused GUV pair, the fluorescence intensity distribution was obtained from the points on the line which cross the center of bottle-neck and meets the vesicles at opposite positons. The fluorescence intensities were normalized by the maximum fluorescence intensity. The fluorescence recovery after photobleaching (FRAP) assay was performed following the manufacturer's instructions. After bleaching the regions of interest (ROI) at 100% power, fluorescence recovery was observed at the same power as before bleaching (∼10% of maximum). ROI was ∼10 µm in diameter to select entire single GUV. Further image analysis was performed with ZEN 2010 LSM software and ImageJ software (National Institutes of Health). The fluorescence recovery curve was fitted to an exponential decay function f(t) = A(1-exp(−t/τ)), where, A is the fraction of mobile component, t is the time passed after photobleaching and τ is the time constant. The lateral diffusion coefficient of lipids was calculated with following equation D = r<sup>2</sup> /4τ, where, D is diffusion coefficient, r is the radius of the photobleached GUV, and τ is the characteristic diffusion time.

### RESULTS

#### Reconstitution of SNARE Proteins into GUVs

GUVs containing SNARE proteins were prepared through the electroformation method. To make sure that SNARE proteins were incorporated into GUVs, the SNARE proteins and/or GUVs were labeled with fluorescent dyes. First, Syb2 labeled with Alexa 647 was reconstituted into GUVs containing NBD. We observed many GUVs with the Alexa 647 fluorescent signal and the co-localized NBD signal (**Figure 1A**), indicating that Syb2-containing GUVs were well formed. Next, Stx1 labeled with Alexa 647 was reconstituted into the non-fluorescent GUVs to exclude the possibility that the fluorescence of circles derives from inadequate filtration of lipid fluorescence. We expected to observe the Alexa 647 signal on circles if the t-GUVs were formed as planned. Otherwise, no fluorescence or any signal from aggregates was expected. Though several amorphous dots representing protein aggregates were present, we found many circles representing GUVs containing Stx1 (**Figure 1B**).

We found that GUVs were not well formed when we tried to incorporate SNARE proteins above a certain limit. Because too few SNARE proteins on a GUV may not induce efficient GUV-GUV fusion, we tested several lipid-to-SNARE protein (LP) ratios to find an optimal condition for the efficient formation of SNARE-containing GUVs (**Table 1**). The efficiency of GUV formation was inversely proportional to the concentration of SNARE proteins. When the LP ratio was below 200 for Syb2 and below 500 for Stx1, we did not observe any GUV. In contrast, GUVs did not contain any SNARE proteins when the LP ratio was 4000, though many GUVs were observed. The 1000 LP ratio was optimal for both Stx1 and Syb2. For

TABLE 1 | The effect of lipid-to-protein (LP) ratio on the formation of giant unilamellar vesicles (GUVs) and reconstitution yields.


<sup>1</sup>−, +, ++ and +++ represent the number of GUVs in a single focal plane. −, no GUV detected; +, a few GUVs; ++ 10–30 GUVs; +++ >50 GUVs. <sup>2</sup>% of SNAREcontaining GUVs = (number of GUVs with a signal for Alexa 647-labeled SNARE proteins/number of GUVs detected in the area) × 100.

example, when Syb2 was incorporated into GUVs at an LP ratio of 1000, we observed tens of GUVs in a single focal plane, and approximately 73% of the GUVs contained fluorescently-labeled SNARE proteins (**Table 1**). Thus, we used an LP ratio of 1000 for further experiments.

#### GUV-GUV Fusion by SNARE Proteins

T-GUVs and v-GUVs containing the binary t-SNARE complex (Stx1 and SNAP-25) and Syb2, respectively, were separately prepared following the procedure described above at an LP ratio of 1000. T-GUVs and v-GUVs were labeled with rhodamine and NBD, respectively. After mixing equal amounts of t- and v-GUVs, the mixture was incubated at 37◦C for 40 min. While major populations were unfused, we found that ∼10% of the GUVs exhibited both NBD and rhodamine fluorescence, which is expected to happen when the two vesicles are fully fused (**Figures 2A**, **Supplementary Figure S2**). On the other hand, no vesicle exhibited both fluorescent signals when GUVs containing only fluorescent dyes (but no SNARE proteins) were mixed together (**Figure 2B**). This clearly indicated that GUVs did not spontaneously fuse in the absence of SNARE proteins. Thus, SNARE proteins reconstituted on GUVs mediated GUV-GUV fusion.

### Hemifusion between Two GUVs Arrested by Myricetin

We tested whether the membrane fusion intermediate arrested by myricetin was hemifused or not. It was expected that hemifused vesicles would show strong fluorescence intensities

for both NBD and rhodamine near the stalk. Other regions of t-GUVs were expected to show a high rhodamine intensity with a lower NBD fluorescence intensity than that observed in the v-GUVs, and vice versa (**Figure 3A**, upper panel). If lipids were not mixed but the membranes were merely closely apposed, NBD fluorescence would not be detected in t-GUVs, and vice versa for v-GUVs (**Figure 3A**, lower panel).

The mixture of t- and v-GUVs was incubated at 37◦C for 40 min in the presence of 1 µM myricetin. We observed hourglass-shaped vesicles forming in the presence of myricetin (**Figures 3B**, **Supplementary Figure S3**). One vesicle showed strong NBD fluorescence, while the other showed rhodamine fluorescence; this obviously represented the signal for the v- and t-GUVs, respectively. However, each vesicle also showed the fluorescence signal of the other vesicle, although the intensity was low. The intensity of each fluorescent signal was analyzed at three different positions: one near the stalk (designated b), one on the v-GUV (designated a) and one on the t-GUV (designated c). The asymmetry of the fluorescence intensities of NBD and rhodamine was dependent on the location of the GUV pair, and was consistent with what we expected for the hemifused GUVs (**Figure 3C**). On the other hand, GUV pairs were also observed when the mixture of t- and v-GUVs was incubated at 4◦C (**Figure 3D**), which was a condition in which membrane fusion did not happen while docking of vesicles was allowed. Though the shape was similar to the hemifused vesicle pair, the fluorescent signals of NBD and rhodamine were not detected on opposite vesicles (**Figure 3E**). This result suggested that the hourglass-shaped GUV pairs enriched in the presence of myricetin were hemifused.

#### Fluorescence Recovery after Photobleaching

Hemifusion between the vesicles in the GUV pair arrested by myricetin was confirmed with a FRAP assay. If two GUV outer leaflets are connected continuously, lipid molecules of the outer leaflets will diffuse laterally, leading to recovery of fluorescence after photobleaching. In contrast, diffusion of lipids from one GUV to another is not allowed if the GUV pair is simply docked, but bilayer leaflets are not connected between the two GUVs.

After hemifused GUV pairs were prepared in the presence of myricetin, GUV-GUV pairs that looked like hourglasses were selected. Hemifusion between a pair of GUVs was identified based on the fluorescence asymmetry as described above (**Figure 4A**). After photobleaching, the entire NBD fluorescence in the GUV at the right-hand side was measured as a function of time. The NBD fluorescence gradually recovered over time, suggesting that the NBD of the GUV on the left-hand side moved to the GUV on the right-hand side (**Figure 4B**). This result clearly indicated the GUV pair was hemifused. The lateral diffusion coefficient of the NBD between GUVs was calculated from the kinetics of fluorescence recovery (**Figure 4C**). It was determined to be 0.18 ± 0.03 µm<sup>2</sup> /s from 3 independent GUV pairs. The average decay constant τ was 71 s when the radius of a photobleached GUV was 7.15 µm. When the rhodamine fluorescence in one vesicle of the GUV pair was photobleached instead, fluorescence was recovered within a few minutes, consistent with NBD photobleaching (**Figure 4D**). These results suggested that the GUV pairs treated with myricetin were hemifused to allow lipid diffusion through the continuous outer leaflets of the two GUVs.

### DISCUSSION

#### SNARE-Driven GUV-GUV Fusion

GUVs are sufficiently large to be viewed using optical or fluorescence microscopy, and as such they are excellent samples to directly visualize and analyze the individual membrane fusion process. However, membrane protein incorporation, size control, and molecule encapsulation inside the GUVs are still challenging, although the electroformation-based method is relatively reproducible for protein-free GUV formation (Yamashita et al., 2002; Limozin et al., 2003; Chiantia et al., 2011; Dezi et al., 2013). It is likely that these difficulties have limited direct visualization of SNARE-driven fusion between GUVs, even in studies that have made use of GUVs for the analysis of membrane fusion (Bacia et al., 2004; Tareste et al., 2008; Hui et al., 2009; Nikolaus et al., 2010; Malsam et al., 2012). In the present study, all t- and v-SNARE proteins were successfully reconstituted in GUVs by optimizing the LP ratio and the GUV formation procedure. Reconstituted GUVs containing SNARE proteins enabled us to analyze membrane fusion intermediates occurring during GUV-GUV fusion with a fluorescence microscope.

The copy number of Syb2 in a synaptic vesicle with a 42-nm diameter is ∼70, which corresponds to an LP ratio of ∼176 in reconstituted proteoliposomes (Takamori et al., 2006; Ji et al., 2010). This LP ratio also corresponds to 5 × 10<sup>6</sup> copies of Syb2 in a v-GUV with a diameter of 10 µm. However, we found that such a high protein density did not allow efficient GUV formation. Rather, higher LP ratios enabled more efficient GUV formation, though the probability that the GUVs contained SNARE proteins was lowered. GUV formation efficiency and SNARE incorporation appeared to be somewhat incompatible. We found that an LP ratio of 1000 and 2000 was optimal for both Syb2 and Stx1.

#### Hemifusion Lipid Diffusion Coefficient

Our FRAP assay revealed a lipid diffusion coefficient of 0.18 µm<sup>2</sup> /s at 25◦C. The diffusion constant of POPC (which was also used in our experiments) in multilamellar vesicles at 25◦C is 7–10 µm<sup>2</sup> /s (Gaede and Gawrisch, 2003). This value is much smaller than the values for the cortical granule membrane and the plasma membrane (Wong et al., 2007). This small diffusion constant of lipids indicates the hemifused GUV pair does not share a wide area, and that only a small region is merged. The small shared area is most likely the bottleneck of the lateral lipid movement. It is not likely that the reconstituted SNARE proteins directly hindered the flow of lipid molecules because the protein density is too low to restrict lipid diffusion at such a low LP ratio as 1000. But, it is also possible that SNARE proteins that induced hemifusion do not dissipate from the stalk of hemifusion, restricting the lateral diffusion of lipids from one vesicle to the other (Chernomordik et al., 1998). Regardless of the exact reason for the small diffusion constant, it suggests that hemifusion arrested by myricetin does not expand to a wide area (i.e., the hemifusion diaphragm).

#### Hemifusion in the Pathway to Fusion Pore Opening Observed by Utilizing Myricetin

We previously suggested that N-terminal half zippering might drive hemifusion. SNARE complex formation was arrested at the half-zippered state by a flavonoid (myricetin), and it was found that the membrane fusion intermediate arrested by myricetin corresponded to hemifusion in proteoliposome fusion. Though it is yet unclear how only half-zippering of the SNARE complex induced hemifusion, hemifusion arrested by myricetin could be converted to full fusion when the myricetin was removed from the SNARE complex by the enzyme laccase. The hemifusion was metastable, and Ca2<sup>+</sup> could trigger immediate full fusion and content mixing. FRET-based bulk lipid-mixing assays (Yang et al., 2010), FRET-based single vesicle-vesicle lipidmixing assays (Heo et al., 2016), FRET-based single vesiclevesicle content-mixing assays (Heo et al., 2016), and DLS-based hydrodynamic radius change assays (Yang et al., 2015) were employed to investigate all of the features mentioned above. In the present study, it was shown that GUV-GUV fusion was also arrested in the hemifusion state by myricetin. This result suggested that hemifusion is a bona fide intermediate leading to fusion pore opening, and serves as the primed

#### REFERENCES


state for Ca2+-triggered millisecond exocytosis (Heo et al., 2016).

#### AUTHOR CONTRIBUTIONS

PH and D-HK devised the experiment. PH and J-BP performed the experiments. PH, Y-KS and D-HK wrote the article.

#### ACKNOWLEDGMENTS

This work was supported by a grant from the Korea Healthcare Technology R&D Project, Ministry of Health and Welfare, South Korea (Grant No: HN14C01010000), and by the National Institutes of Health (GM05290 and 5U54GM0 87519). The Advanced Biomass R&D Center (ABC) of Korea Grant (2011- 0031359) funded by the Ministry of Science, ICT and Future Planning.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2017.00093/full#supplementary-material

#### FIGURE S1 | Assembly of ITO slides.

FIGURE S2 | Representative images of fully fused GUVs. V-GUVs and t-GUVs were labeled with 3 mol % NBD and rhodamine, respectively. Fluorescence intensity profile (lower panel) was obtained on the line indicated by an arrow in the merged GUV image.

FIGURE S3 | Hemifused vesicle pairs exhibiting fluorescence intensity asymmetry. (A) Images were taken in the same manner as in Figure 3, but in different experiments. The fluorescently labeled v-GUV (NBD, green) and t-GUV (rhodamine, red) were mixed and incubated at 37◦C for 40 min in the presence 1 µM myricetin. The distribution of fluorescent intensity was measured at various regions of GUV pair. (B) Number of GUVs observed in this study. Major population (80%–90%) of GUVs in a focal plane is unfused free t- or v-GUVs even after optimization of GUV reconstitution scheme because of low probability of GUV-GUV fusion. Tens of glass slides from several independent experiments were used to obtain significant numbers of GUVs. We observed 161 and 117 GUVs in the absence and presence of 1 µM myricetin, respectively. Circles exhibiting both NBD and rhodamine fluorescence were counted as fully fused vesicles. Hourglass-shaped GUV pairs were counted by eye. Hemifusion was confirmed for 7 GUV pairs out of 79 hourglass-shaped GUV pairs by analyzing fluorescence intensity asymmetry. (C) Averaged fluorescence intensities which are asymmetrical depending on the locations on the GUV pair.


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Heo, Park, Shin and Kweon. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Exocytosis, Endocytosis, and Their Coupling in Excitable Cells

Kuo Liang1 †, Lisi Wei 2 † and Liangyi Chen<sup>2</sup> \*

*<sup>1</sup> Department of General Surgery, XuanWu Hospital, Capital Medical University, Beijing, China, <sup>2</sup> State Key Laboratory of Membrane Biology, Beijing Key Laboratory of Cardiometabolic Molecular Medicine, Institute of Molecular Medicine, Peking University, Beijing, China*

Evoked exocytosis in excitable cells is fast and spatially confined and must be followed by coupled endocytosis to enable sustained exocytosis while maintaining the balance of the vesicle pool and the plasma membrane. Various types of exocytosis and endocytosis exist in these excitable cells, as those has been found from different types of experiments conducted in different cell types. Correlating these diversified types of exocytosis and endocytosis is problematic. By providing an outline of different exocytosis and endocytosis processes and possible coupling mechanisms here, we emphasize that the endocytic pathway may be pre-determined at the time the vesicle chooses to fuse with the plasma membrane in one specific mode. Therefore, understanding the early intermediate stages of vesicle exocytosis may be instrumental in exploring the mechanism of tailing endocytosis.

Keywords: exocytosis, endocytosis, kiss and run, kiss and stay, compound fusion, multivesicular exocytosis, clathrin

#### Edited by:

*Cong Ma, Huazhong University of Science and Technology, China*

Reviewed by:

*Qiangjun Zhou, Howard Hughes Medical Institute, USA Xuelin Lou, University of Wisconsin-Madison, USA*

#### \*Correspondence:

*Liangyi Chen lychen@pku.edu.cn*

*†Co-first authors.*

Received: *18 January 2017* Accepted: *31 March 2017* Published: *19 April 2017*

#### Citation:

*Liang K, Wei L and Chen L (2017) Exocytosis, Endocytosis, and Their Coupling in Excitable Cells. Front. Mol. Neurosci. 10:109. doi: 10.3389/fnmol.2017.00109* INTRODUCTION

Vesicle exocytosis is a fundamental cellular process that regulates many biological events, such as the release of neurotransmitters, hormones, and cytokines and delivery of proteins and lipids to the plasma membrane for cell repair, growth, migration, and regulation of cell signaling (Alabi and Tsien, 2013; Wu L. G. et al., 2014). In excitable cells, such as neurons and endocrine cells, regulated exocytosis is triggered within milliseconds after membrane depolarization. Upon strong stimulation, a massive fusion of secretory vesicles could occur at designated release sites within a short period of time. Therefore, compared with constitutive exocytosis in non-excitable cells, regulated exocytosis must be equipped with specialized machinery that enables fast, Ca2+ dependent, spatially defined exocytosis. Tailing endocytosis must match with exocytosis to recycle exocytosed vesicular components and clear release sites on the plasma membrane in a timely fashion. Based on the kinetics, structures, and molecules involved in different cell types, a variety of exocytosis and endocytosis subtypes have been proposed. However, how these mechanisms are coupled in space and time remains mysterious. Here, we have provided an outline of different exocytic and endocytic processes and how they may be coupled by different factors.

## EXOCYTOSIS IN EXCITABLE CELLS

Exocytosis requires a merging of the vesicular membrane into the plasma membrane. Through shielding of the negative charge on the bilayer surface, diminishing electrostatic repulsion force, and overcoming the dehydration barrier, two bilayers can merge into one. Formation of an assembled ternary SNARE complex provides the required energy. Depending on the fate of the vesicular components upon lipid merging, exocytosis can progress by fully inter-mixing the vesicular membrane components with the plasma membrane (full fusion), fusing with the plasma membrane via a transient flickering of the fusion pore (kiss and run; An and Zenisek, 2004; Rizzoli and Jahn, 2007; Alabi and Tsien, 2013), or fusing with a partially retained vesicular membrane structure and components at the exocytic site (kiss and stay; Taraska et al., 2003; An and Zenisek, 2004; Tsuboi et al., 2004; Rizzoli and Jahn, 2007).

#### Different Models of Fusion Pore Formation

Intrinsically, fusion machinery must operate with the formation and the expansion of an omega-shaped pore structure, which minimizes leakage from the vesicle and cytosol during exocytosis. Such a process requires coordinated distortion and controlled disruption of two lipid bilayers to form a water-filling fusion pore, which cannot be observed (van den Bogaart et al., 2010) directly in vivo due to its small size and short lifetime (Lindau and Alvarez de Toledo, 2003). Electrophysiological methods, on the other hand, provide a brief glimpse of some pore intermediates. Based on these indirect estimations, there exists three models that describe the fusion pore, a lipidic (Chanturiya et al., 1997) or a proteinaceous (Han et al., 2004) pore or a hybrid of the lipid and protein composition (Bao et al., 2016; Sharma and Lindau, 2016). For a lipidic pore, fusion starts with protrusion of two bilayers toward each other in a very narrow region, followed by the merge of the two proximal monolayers of each bilayer (stalk), the enlargement of the merged region to form one bilayer (hemifusion), and the final formation of a lipidic fusion pore. Formation of the stalk and hemifusion diagram ensures the expansion of the pore without a leak. Such a fusion pore does not require multiple copies of SNARE complexes. Instead, one pair of SNARE proteins, firmly anchored on the vesicular and plasma membrane with transmembrane (TM) segments, may interact with each other to provide the force to pull the different membranes together (van den Bogaart et al., 2010).

In 1987, Almers and co-workers measured the initial pore conductance during exocytosis of mast cells to be 200–300 pS, equivalent to a pore of diameter of ∼2 nm (Breckenridge and Almers, 1987). This value is similar to the conductance of K channel, inspiring the early hypothesis that the fusion pore is a proteinaceous gap junction channel. In 2004, using tryptophan scanning mutagenesis of the syntaxin TM anchor, Han et al. identified three critical residues that reduced the amplitude of the foot signal of an amperometry recording (Han et al., 2004). These positions are positioned along one face of the alpha-helix, promoting the idea that they might face the inside of a pore. Based on these results, they proposed a provocative hypothesis that the TM domains of 6–8 syntaxin molecules are arranged in a ring to form one half of a gap-junction-like pore, with the other half formed by the TM domains of synaptobrevin (Syb2). After the formation of the protein-lined pore, the pore could close again ("kiss and run"; Albillos et al., 1997; MacDonald et al., 2006) or allow the membrane lipids to enter to facilitate pore expansion and complete membrane merge ("pore dilation" or "full fusion"). This interpretation is supported by the existence of syntaxin clusters on the plasma membrane of endocrine and synapses (Barg et al., 2010; van den Bogaart et al., 2011), as well as three or more copies of SNARE proteins that are required for the fast release of secretory vesicles (Domanska et al., 2009; Mohrmann et al., 2010). This model, however, requires fusogenic proteins to be in perfect alignment to constrict lipid flow during the initial pore opening, which have not been proved experimentally. Changes in the amplitude of foot signals could be due to different extents of vesicular filling (Sombers et al., 2004) and different dissociation kinetics of neurotransmitters from the vesicular matrix (Reigada et al., 2003). The impact of a syntaxin mutation on the release kinetics thus may provide an alternative explanation. Other amperometric investigations also reported controversial results, such as very large fluctuations of foot signals characteristic of variable and lipidic pores.

A combination of these two models yielded a model with a pore that is both lipidic and proteinaceous. Recently, two Syb2 molecules have been shown to be able to be incorporated within a nanodisc with a diameter of 6 nm, which readily fuses with t-SNARE-containing vesicles and can be blocked by mutations of critical residues in the TM domain of Syb2. Given that such a small nanodisc appeared to be too small to accommodate a lipidic pore, and at least three TM domains are required to line up a proteinaceous pore, these results suggest that the pore itself must be a hybrid of both proteins and lipids (Bao et al., 2016). According to the molecular dynamic simulation, the water-filled fusion pore traversing the membrane and the nanodisc constitutes both the lipid head group and the c-termini of the TM domains of Syb2 and syntaxin (Sharma and Lindau, 2016). Whether such a hybrid model works in real cells remains to be determined.

### Pore Opening and Full Fusion

Under electron microscopy (EM), the earliest seen fusion pores of synaptic vesicles were never <20 nm despite the observed 3–4 ms after a single stimulus (Heuser and Reese, 1981) and were mostly ∼150 nm for dense-core vesicle exocytosis in Limulus amebocytes (Ornberg and Reese, 1981). In contrast, using cell-attached membrane capacitance recording, fusion pore conductance of secretory vesicles range from 30 to 1,000 pS (Breckenridge and Almers, 1987; Lindau and Alvarez de Toledo, 2003; He et al., 2006; MacDonald et al., 2006), corresponding to a fusion pore diameter of 1–7 nm. Fusion pores larger than 10 nm, as those observed under EM, will result in pore conductance approaching infinity, rendering estimation of pore size impossible. Additionally, these large fusion intermediates do not restrict the diffusion of neurotransmitters and small neuropeptides such as neuropeptide-Y (NPY) (Tsuboi et al., 2004). Thus, they are undetectable with either amperometry or membrane capacitance recordings. On the other hand, both electrophysiological methods provide an estimation of the pore duration before final dilation to be ∼10–80 ms in endocrine cells on average. Therefore, it is intriguing that a small pore <10 nm was never observed under EM, provided that ultrafast-freezing EM should have sufficient temporal and spatial resolution in principal. Whether the small fusion pore intermediate is a rare event compared to other fusion intermediates or ultrafastfreezing EM lacks the resolution and contrast to resolve such

small pores remains unknown. Live cell fluorescence microscopy, including super-resolution microscopy (Huang et al., 2009; Schermelleh et al., 2010), does not have the sufficient spatial and temporal resolution to observe small fusion pores either.

The large fusion pore observed, on the other hand, may represent an intermediate before full collapse of vesicles. A secretory vesicle contains 60–70 copies of Syb2, of which 1–3 are used during the fusion process. All of these Syb2 molecules, as well as other vesicular proteins such as synaptotagmin, are completely lost on the plasma membrane during the full collapse of the vesicle. Clathrin-mediated endocytosis must be initiated to collect and precisely recycle these vesicular membrane components rapidly. Originally believed to be a slow process, we find that the migration of preformed clathrin-mediate pits (CCPs) on the plasma membrane to the vesicle release sites is key to the clearance of exocytic slots in a timely fashion (**Figure 1**; Yuan et al., 2015), which may explain the fast clathrinmediated endocytosis observed in neurons (time constant of 3–10 s; Granseth et al., 2006; Zhu et al., 2009).

#### Kiss and Run (KR)

Resealing of a small fusion pore leads to a KR event, which is both an exocytic and an endocytic process. KR events are detected as the "stand-alone" foot signals in amperometry recordings (Zhou and Misler, 1996; Albillos et al., 1997) or membrane capacitance flickers in capacitance recordings (Albillos et al., 1997; He et al., 2006; MacDonald et al., 2006). However, electrophysiology methods cannot identify the integrity of the vesicular shape and composition after a vesicle performs a KR event. KR was originally defined by Ceccarelli et al. (1973) under EM as the fusion of vesicles with preservation of vesicle morphology. Therefore, the conservation of the vesicle shape, as observed with EM and live cell fluorescence microscopy, along with a small pore probed with electrophysiology technologies are cornerstone features of KR (Alabi and Tsien, 2013).

Various conditions have been shown to promote KR in a number of secretory cells, including high cytosolic Ca2<sup>+</sup> (Ales et al., 1999) and activation of PKA (MacDonald et al., 2006). Despite this knowledge, the mechanisms of the resealing and flickering of a fusion pore remain elusive. In the lipidic pore model, each intermediate structure is at its free energy minima, and an injection of exogenous energy is needed for the transition between different fusion intermediates. Therefore, fusing vesicles remain connected to the plasma membrane with a narrow pore until the addition of new proteins and lipids to the fusion machinery reverses the process. Indeed, a variety of proteins, such as dynamin (Anantharam et al., 2011; Jackson et al., 2015), myosin II, actin (Aoki et al., 2010), SNARE proteins (Fang et al., 2008; Gucek et al., 2016), synaptotagmin (Wang et al., 2001; Lai et al., 2013), and complexin (Dhara et al., 2014) have been found to affect the fusion pore dynamics in chromaffin and PC12 cells. These studies highlight an active role of these components in impacting the fusion pore. Alternatively, it is hypothesized that the release of energy associated with the formation of one SNARE bundle is insufficient to overcome the restraining force from the intact vesicle-vesicle and vesiclecytoskeleton filamentous web that opposes full vesicle collapse (Alabi and Tsien, 2013). Therefore, upon completion of the trans-SNARE complex formation and diminishing of the countering force against pore constriction, the vesicle pore reseals to be intact again. The difference between these two models is that the force opposing fusion dilation is constitutively present at the release sites in the latter model, therefore bypassing the need for acute and coordinated recruitment of facilitating membrane components.

The physiological significance of KR in endocrine cells has been well-established. In pancreatic β-cells, the KR of large densecore vesicles (LDCV) and small vesicles allows for the selective release of ATP and GABA, respectively. In contrast, insulin crystals within LDCVs are retained within the lumen during the transient flickering of fusion pores (MacDonald et al., 2006). The release of peptides from other endocrine cells is also likely to be limited, since the transient brightening with no diffusion of fluorescent-tagged NPY puncta is regarded as a KR event under total internal reflection fluorescence (TIRF) microscopy (Tsuboi and Rutter, 2003). KR is also identified in the fusion of synaptic vesicles in synapses (He et al., 2006; Zhang et al., 2009), which may add another layer of post-fusional regulation and enables non-quantal synaptic transmission in principal. However, even for the smallest fusion pore opening, the vesicle will be drained of transmitter within tens of milliseconds, long before the fusion pore closes. Therefore, the KR model is unlikely to regulate vesicle release post-fusionally.

Alternatively, KR may confer an ultrafast and efficient recycling process independent of clathrin (He et al., 2006; Zhang et al., 2009). A KR event will lead to a fast and efficient recycling of almost all vesicular components, as well as immediate on-site refilling of neurotransmitters. The same vesicle then can fuse multiple times (Zhang et al., 2009), while the previously used cis-SNARE complex in the previous round of fusion needs to be removed from the vesicle to prevent blockade of the second round of fusion. However, it is unclear how such a cis-SNARE complex passes through the small flickering pore and diffuses into the plasma membrane without affecting the pore, given that the SNARE protein could be part of the pore itself. Nevertheless, by expelling a few Syb2 molecules used for each round of fusion, a KR event is more efficient in maintaining the identity of the vesicle than discharging all Syb2 molecules upon every instance of vesicle exocytosis. By keeping the vesicular V-ATPase, resealed vesicles can gradually re-acidify, which permits pH gradientcoupled refilling of the vesicle with neurotransmitters (Alabi and Tsien, 2013). In theory, this process may promote rapid recovery of neurotransmission during sequential stimulations. However, this hypothetical benefit is in disagreement with the experimental data that KR is prevalent at the beginning of action potential trains but is eventually replaced by full fusion upon sustained firing in hippocampal neurons (Zhang et al., 2009). Therefore, the physiological significance of KR in synapses remains unknown.

#### Kiss and Stay (KS)

In endocrine cells, in addition to probing fusion pores indirectly with electrophysiological methods, imaging technologies such as spinning disc confocal and TIRF microscopy provide the spatiotemporal correlation of fusion events with concurrent

diffusion of vesicular lipids and proteins (Holroyd et al., 2002; Taraska et al., 2003; Tsuboi and Rutter, 2003; Tsuboi et al., 2004). In parallel to the small fusion pores detected from electrophysiological data, imaging reveals retention of vesicle membrane shape and some vesicle compositions after the exocytosis of LDCVs (Holroyd et al., 2002; Taraska et al., 2003; Tsuboi et al., 2004). In contrast to the retention of the majority of composition after a vesicle performs KR, the loss of vesicular lipids and the majority of some vesicular proteins such as Syb2 is obvious (Taraska and Almers, 2004; Tsuboi et al., 2004). Named as KS (or cavicapture), this process is regarded as an allosteric form of KR, sharing similar characteristics such as on-site recycling of vesicular components and its dependence on dynamin (An and Zenisek, 2004). In this sense, it is often regarded as a fast, clathrinindependent endocytosis that occurs at the fusion sites (Holroyd et al., 2002; Taraska et al., 2003; Tsuboi et al., 2004). However, we have shown that clathrin-dependent endocytosis could rapidly synchronized to occur at the fusion sites (Yuan et al., 2015), highlighting a necessity of classifying the identity of endocytosis based on molecules, rather than on kinetics and localization.

Moreover, it is unclear whether KS events identified by different fluorescence probes represent the same or different fusion intermediate stages. For example, discharge of NPY is much faster than that of fluorescent-tagged tissue plasminogen activator (tPA) in adrenal chromaffin cells and pancreatic β-cells (Tsuboi et al., 2004; Weiss et al., 2014). These data were initially interpreted to suggest that large tPAs (∼10 nm in diameter) are prevented from free diffusion by the fusion pore, which do not interfere with diffusion of small lumen contents such as NPY (∼3 nm in diameter). However, overexpressed tPA changes the lumen composition of LDCVs, binds to the exposed luminal surface of fused chromaffin granules and slows down the release kinetics as measured by amperometry recordings (Weiss et al., 2014). A pore as large as 10 nm barely constrains diffusion of small neurotransmitters from vesicle lumen upon exocytosis. Alternatively, KS may represent a fusion intermediate later than the KR, which explains why it shares many mechanistic characteristic with the later. Given the prevalence of KS in endocrine cells, we speculate that the 20–150 nm fusion pores observed under EM may be in the KS state, although future direct proof is still needed.

Clearly, resolving fusion-associated membrane shape retention and dispersion of vesicular membrane during synaptic transmission is difficult due to the small sizes of synaptic vesicles and boutons. However, coupled and recycled vesicular membrane proteins are distinct from those left on the presynaptic membrane after exocytosis (Wienisch and Klingauf, 2006), highlighting a possible KS mechanism operating in synapses as well.

### Sequential Fusion and Multivesicular Exocytosis

After a KS fusion event, the invaginated fusion site before endocytosis can be targeted to harbor the next rounds of exocytosis ("sequential fusion"; Takahashi et al., 2004; Kishimoto et al., 2005), creating deep invaginations on the plasma membrane that may resulted in internalization of one large endocytic vesicle ("bulk endocytosis"; Wu and Wu, 2007; Wen et al., 2012). These invaginations were initially found in non-excitable cells such as pancreatic acinar cells, mast cells, eosinophils and neutrophils (Pickett and Edwardson, 2006) and were later discovered in excitable cells such as pancreatic β-cells and neurons (Kwan and Gaisano, 2005; He et al., 2009). During sequential fusion, cis-SNARE complexes need to be removed from the fusion sites, and new trans-SNAREs on the plasma membrane must diffuse into these invagination structures. By adopting this configuration, vesicles that exist deep within the cytosol readily fuse with the plasma membrane to release their contents. This configuration also creates spatially preferred sites on the plasma membrane, conferring a mechanism for generating exocytosis "hot spots" in non-neuronal cells.

Multivesicular exocytosis is another form of exocytosis where vesicles fuse homotypically before interacting with the plasma membrane. In contrast to compound fusion, a multivesicular exocytosis event leads to a capacitance increase that is several folds higher than that caused by the fusion of a single vesicle (He et al., 2009). However, multivesicular exocytosis is rare, precluding it from being systematically and statistically analyzed. Therefore, whether these large capacitance jumps represent a distribution different from that represented by a fusion of a single vesicle or the long-tail region of one unified distribution remains to be determined. Under EM, multiple vesicles that are connected to each other but not with the plasma membrane are sometime observed, which is also taken as evidence supporting multivesicular exocytosis (Wu L. G. et al., 2014). However, these structures could also be due to the sequential fusion of vesicles to the invaginated site that exhibited pore closure, which has a distinct molecular mechanism. A multivesicular exocytosis event needs homotypical vesicle-vesicle fusion, which presumably uses different sets of SNAREs other than those used for vesicle-plasma membrane exocytosis, similar to what has been proposed for compound fusion (Thorn and Gaisano, 2012). In principal, compared to the fusion of multiple vesicles at one designated site for several rounds, a multivesicular fusion event will be more efficient in emptying vesicular contents within a short period of time given that fusion sites are limited. However, how multivesicular fusion operates in vivo remains elusive.

### COUPLED ENDOCYTOSIS IN EXCITABLE CELLS

Unlike constitutive endocytosis in non-excitable cells, coupled endocytosis following evoked exocytosis must be fast and spatially matched with exocytosis to maintain the balance of surface membrane and the finite size of the readily releasable pool of vesicles. Kinetically, evoked endocytosis often consists of two phases, a fast endocytosis followed by a slow one (Artalejo et al., 2002; He et al., 2008; Lou et al., 2008; Wu et al., 2009). Mechanistically, the fast endocytosis is often regarded as clathrinindependent, while the slow one is often dependent on clathrin (He et al., 2008; Lou et al., 2008). Based on these kinetic and mechanical characteristics, a full collapsing of vesicle fusion is often thought to be associated with the slow, clathrin-dependent endocytosis, while the resealing of a fusion pore during a KR or KS is regarded as the fast mechanism underlying the coupled clathrin-independent endocytosis.

#### Clathrin-Dependent Endocytosis

With immunostaining and confocal microscopy, active zones have been found to be surrounded by a peri-active zone enriched with endocytic proteins such as clathrin and dynamin, which mediate clathrin-mediated endocytosis following synaptic transmission (Cano and Tabares, 2016). Using TIRF microscopy, exocytosis in MIN6 cells was found to be associated with on-site recruitment of the endocytic protein dynamin but not clathrin, epsin, or amphiphysin. These data were interpreted to suggest that only clathrin-independent endocytosis, a form of KS, is spatially coupled to exocytosis in insulin-secreting β-cells (Tsuboi et al., 2004). However, do these spatially confined dynamin recruitments represent bona fide endocytosis? If they indeed represent clathrin-independent endocytosis, do their kinetics match with electrophysiological data? What is their physiological significance? These are questions left unexplored. Recently, we have systematically examined the exocytosis-endocytosis coupling in insulin-secreting cells (Yuan et al., 2015). We have revealed that clathrin can be recruited to the fusion sites in a fast and a slow manner, which were accompanied with the simultaneous recruitment of dynamin (**Figures 1A,B** refer to Figure S2 and **Figure 1C** in Yuan et al., 2015). The slow recruitment represents a de novo formation of clathrin-coated pits (CCPs), while the fast recruitment originates from preformed CCPs stably docked at the fusion sites or rapid movement of CCPs toward fusion sites on the plasma membrane. These spatially confined clathrin recruitments are indeed mediators of the endocytosis of vesicular proteins such as synaptotagmin VII and Syb2 (Yuan et al., 2015). Therefore, clathrin-dependent endocytosis can operate both at a fast and a slow pace, in agreement with similar findings in hippocampus neurons (Granseth et al., 2006; Zhu et al., 2009). We argue that the speed of designated endocytosis depends on the extent of synchronization of individual events, which cannot be used as the sole criteria for distinguishing clathrin-dependent from clathrin-independent endocytosis. As we have shown, the physiological significance of fast, clathrin-dependent endocytosis is also critical for sustained exocytosis during intense stimulation (Yuan et al., 2015), similar to what has been observed in synapses (Hosoi et al., 2009; Kawasaki et al., 2011).

#### Clathrin-Independent Endocytosis

The recruitment of dynamin to fusion sites can be described by three Gaussian functions (**Figure 1A** refers to Figure S2 in Yuan et al., 2015). While the last two time constants match that of clathrin recruitment, the first one represents recruitment to sites ∼2 s after a fusion event and independent of clathrin (**Figure 1A**). Assuming that dynamin 1-dependent endocytosis recycles a vesicle with a size similar to that of a dense-core granule (∼230 nm in diameter) the membrane internalized by the clathrin-independent endocytosis shall be much larger than those internalized by the fast, clathrin-dependent pathway, in consistent of membrane capacitance experiments conducted in rat pancreatic β-cells (He et al., 2008). Such clathrin-independent fast recruitments of dynamin may also profoundly contribute to the fast membrane capacitance decay recorded in synapses and other endocrine cells (Artalejo et al., 2002; Lou et al., 2008; Hosoi et al., 2009). In addition to dynamin, actin also plays an indispensable role in the clathrin-independent endocytosis in pancreatic β-cells (He et al., 2008).

The identity of clathrin-independent, actin-dependent fast endocytosis is unlikely to be KR in β-cells, given that exocytosis with fusion pores smaller than 1 nm lasts <1 s in β-cells (Takahashi et al., 2002). Closure of a large fusion pore formed by KS (cavicapture) is likely to be the corresponding form of the fast clathrin-independent endocytosis. Bulk endocytosis, also found in endocrine cells and synapses (Wen et al., 2012; Watanabe et al., 2013), can be fast and independent of clathrin. Structurally, bulk endocytosis could be the reversal of sequential fusion or multivesicular endocytosis. The main difference between a bulk endocytosis and a cavicapture event is that the quantity of the plasma membrane retrieved by a single endocytic process is larger in the former. To differentiate these possibilities, we must directly visualize the membrane structures of fusion sites on the plasma membrane with imaging techniques. Of course, bulk endocytosis could also be unrelated to sequential exocytosis but related to a continuously invaginated plasma membrane driven by vesicular proteins, lipids, and endocytic machinery.

Finally, a clathrin-independent and dynamin-independent endocytosis is found in calyx neurons (Xu et al., 2008). However, without a definite molecular marker, this endocytic process cannot be studied further. In contrast, an ultrafast, clathrinindependent endocytosis is found in central synapses (Watanabe et al., 2013), which was inhibited by dynasore. However, given that dynasore affects cellular cholesterol, lipid rafts, and actin as well as dynamin (Preta et al., 2015), whether ultrafast endocytosis depends on dynamin remains to be proved. Actin is found to be required for the fast endocytosis in neurons (Delvendahl et al., 2016; Wu et al., 2016; Soykan et al., 2017), similar to what have been demonstrated in endocrine β-cells (He et al., 2008). However, how this fast endocytosis defined by electrophysiological and fluorescence experiments correlates with ultrafast endocytosis defined by the rapidlyfreezing electron microscopy needs to be explored in the future.

#### MOLECULAR MECHANISMS FOR EXO-ENDOCYTOSIS COUPLING

Different factors couple endocytosis with exocytosis, including cytosolic Ca2+, lipids, cytoskeleton and proteins (Wu L. G. et al., 2014). Here, we briefly summarize how they are proposed to function in exo-endocytosis coupling.

## Ca2<sup>+</sup>

Ca2<sup>+</sup> influx through voltage-gated calcium channels triggers exocytosis in excitable cells. Synaptotagmin is the established Ca2<sup>+</sup> sensor for triggering vesicle exocytosis. An increase in [Ca2+]<sup>i</sup> , on the other hand, accelerates but does not affect the amplitude of both clathrin-independent and clathrin-dependent endocytosis in β-cells (He et al., 2008). Not surprisingly, [Ca2+]<sup>i</sup> elevation is found to initiate all forms of endocytosis (fast endocytosis, slow endocytosis, and bulk endocytosis) in calyx neurons (Hosoi et al., 2009; Wu et al., 2009). Because endocytosis is intimately linked to the prior exocytosis, the impact of Ca2<sup>+</sup> influx on endocytosis may be a result of the impact of Ca2<sup>+</sup> on membrane additions due to exocytosis. However, the relationship between the speed of endocytosis and [Ca2+]<sup>i</sup> (He et al., 2008) is different than that between [Ca2+]<sup>i</sup> and exocytosis (Wan et al., 2004). Similarly, deletion or mutation of both the C2A and C2B domains of the calcium-binding domains of synaptotagmin 1 prolongs the time constant of slow endocytosis by 30–50% but does not completely block the endocytosis (Yao et al., 2011). These data suggest that Ca2<sup>+</sup> may affect the endocytic route via a pathway different from that for exocytosis.

The application of various calmodulin inhibitors blocks all types of endocytosis in calyx neurons, suggesting that calmodulin could be one Ca2<sup>+</sup> sensor for endocytosis (Wu et al., 2009). The mechanism by which calmodulin phosphorylation initiates endocytosis remains to be determined. Calcineurin, the phosphatase that dephosphorylates many endocytic proteins (Cousin and Robinson, 2001), could be one main downstream target of calmodulin (Wu X. S. et al., 2014). Calcineurin has been shown to selectively dephosphorylate neuronal specific dynamin 1 and dynamin 3 but not ubiquitous dynamin 2. Such dephosphorylation is associated with the recruitment of F-BAR protein, syndapin I (Anggono et al., 2006), and may be critical for the stimulation of bulk endocytosis in synapses (Clayton et al., 2009). However, a large number of studies using blockers of calcineurin do not reach a consensus (Wu L. G. et al., 2014). Therefore, it is unclear whether such the controversy is due to the different synapses involved or a lack of specificity of pharmacological blockers. Resolving this issue is critical for understanding how calcium influx triggers endocytosis.

#### Lipids

Phosphatidylinositol 4,5-bisphosphate (PIP2) is a minority phospholipid of the inner leaflet of plasma membranes (Suh and Hille, 2008). On the one hand, PIP<sup>2</sup> activates voltage gated Ca2<sup>+</sup> channels and slows channel rundown, which is upstream of vesicle exocytosis. On the other hand, PIP<sup>2</sup> also interacts with a number of proteins essential for the exocytosis machinery, such as syntaxin 1, Munc13, synaptotagmin and Doc2, either via the C2 domain or via an electrostatic interaction with basic amino acids (Koch and Holt, 2012). PIP<sup>2</sup> binds to syntaxin and Munc13, which regulate the readily releasable pool of vesicles, and the PIP2:synaptotagmin interaction seems to be essential for the Ca2+-dependent structure changes that catalyze the SNARE assembly. PIP<sup>2</sup> also serves as a central hub for the organization of different endocytic proteins. Through electrostatic interactions with dynamin, the adaptor protein 2 (AP2), membrane curvature sensing protein FCHo, amphiphysin, and assessor proteins, such as epsin and synaptojanin, PIP<sup>2</sup> facilitates the initiation, assembly, maturation, and final scission of CCPs. Therefore, PIP2, being in the center of recruiting proteins important for exocytosis and endocytosis, could be one crucial coupling factor.

Downstream of both Ca2<sup>+</sup> and PIP2, we have shown that diaglycerol (DAG) could be another lipid that coordinates exocytosis and endocytosis. Ca2<sup>+</sup> influx activates Ca2+ dependent phospholipase C, which breaks down PIP<sup>2</sup> into inositol trisphosphate (IP3) and DAG, which is locally enriched around fusion sites in pancreatic β-cells. In return, DAG binds to Munc13 and activates protein kinase C, both of which are essential to vesicle exocytosis. As a lipid that induces negative membrane curvature, DAG microdomains accumulated at fusion sites reduce the energy of CCP movement on the plasma membrane, thus guiding the movement of preformed CCPs toward fusion sites to mediate fast, clathrin-dependent endocytosis (Yuan et al., 2015).

### Cytoskeleton

Densely packed actin filaments are often seen under the plasma membrane. Actin and related factors, such as Cdc42, N-WASP, and actin binding protein (ABP), interact directly or indirectly with active zone scaffolding proteins such as piccolo, organizing vesicle trafficking to, and fusion at the active zone. Cdc42 and N-WASP also interact with coat proteins of CCPs such as intersectin. These data suggest that actin could act as a bridge between exocytosis and endocytosis (Alabi and Tsien, 2013).

Microtubules, on the other hand, are often believed to bridge between the cell interior to the actin filaments close to the plasma membrane. However, microtubules originated from the Golgi can also touch the plasma membrane by CLASP, a microtubule-associated capping protein (Lansbergen et al., 2006). Through its interaction with LL5β, CLASP interacts with ELKS, another active zone scaffolding protein, and helps to anchor dynamic microtubule filaments at fusion sites. We show that a mutation of CLASP inhibits exocytosis in pancreatic βcells and reduces coupled endocytosis along with a reduction in the simultaneous movement of CCPs toward the fusion sites (Yuan et al., 2015). Therefore, microtubules organized by CLASP and ELKS may be another factor that couples exocytosis with fast clathrin-dependent endocytosis in secretory cells.

### Proteins

SNARE proteins and associated proteins such as synaptotagmin and Munc13 are essential for exocytosis, while also interacting with proteins critical for endocytosis (Wu L. G. et al., 2014). However, different from their active roles in exocytosis, the roles of SNARE and associated proteins in endocytosis may be providing domains for AP2 and other adaptor proteins to recognize and bind. In this sense, their roles in endocytosis are permissive and non-essential. Dynamin is another protein that may participate critically in both exocytosis and endocytosis. As a GTPase, the role of dynamin in mediating fission of endocytic vesicle is well-known. On the exocytosis side, transfecting PC12 cells with a dynamin mutant with elevated GTPase activity shortened the foot duration of amperometry recordings, while the opposite occurred with the overexpression of a dynamin mutant with reduced GTPase activity (Anantharam et al., 2011; Jackson et al., 2015). These experiments place dynamin at the very beginning of exocytosis regulation, where the fusion pore is smaller than 1 nm. How this function of dynamin is correlated with its impact on the endocytic machinery remains elusive. Accordingly, deletion of dynamin-1 impairs both endocytosis and exocytosis at central synapses and produces different synaptic plasticity through distinct mechanisms (Mahapatra et al., 2016; Mahapatra and Lou, 2017); deletion of dynamin-2 in pancreatic β-cells leads to defects in clathrin-mediated endocytosis and biphasic insulin release (Fan et al., 2015).

### SUMMARY AND FUTURE PERSPECTIVES

We have summarized the above-mentioned mechanisms regarding exocytosis, endocytosis and possible coupling factors in **Figure 2**. From a macroscopic view, exocytosis may be matched with endocytosis: full fusion with clathrin-mediated endocytosis, KR and KS with clathrin-independent endocytosis, and sequential fusion and multivesicular exocytosis with bulk endocytosis. In this sense, the fate of the components of the fusing vesicle may be pre-determined at the moment of its choice of fusion modes. Therefore, understanding the early fusion intermediates of a vesicle, such as the hemifusion state, pore opening, dilation, and shape retention, will be instrumental for the understanding of the whole coupled process.

The listed classification of different exocytosis and endocytosis subtypes is not based on molecular mechanism but rather hinges on studies that involve different experiments conducted on different cell types. The terminologies defined by different methods may not be mutually inclusive or exclusive. For example, bulk endocytosis is usually regarded as a subcategory of

clathrin-independent endocytosis. However, the bulk membrane invaginations observed in secretory cells under EM, which are often taken as evidence supporting bulk endocytosis, may support the internalization of small or large chunks of membrane in a clathrin-dependent manner in live cell studies. KR and KS may be one uniform process at different stages but could also be two distinct processes with non-overlapping mechanisms. To differentiate these controversies, it is important to sort out molecules that are exclusively used for some specific processes, in addition to actin for clathrin-independent endocytosis (He et al., 2008; Delvendahl et al., 2016). Alternatively, we shall examine the same process in the same cells using multiple techniques. For example, combining cell-attached membrane capacitance measurements with imaging vesicular lipids in endocrine cells will help clarify whether lipid exchange occurs between the vesicle and the plasma membrane during the flickering of a small fusion pore. Simultaneous imaging of vesicular components and extracellularly applied fluorescent dextran of different sizes will help monitor the dilation of a fusion pore from ∼1 nm to a much larger in diameter (Takahashi et al., 2002). This will differentiate KR and KS and ultimately determine the size of fusion pores accompanying KS exocytosis. Monitoring the shape of the membrane may reveal clues of hemifusion in live cells (Zhao et al., 2016) and will also confirm or disapprove the compound fusion/multivesicular exocytosis theories and their physiological significance. Finally, operating at a nanometer scale with lifetimes of milliseconds,

most of the fusion intermediate structures described here can hardly be directly discerned even with state-of-the-art superresolution microscopy methodologies (Huang et al., 2009; Schermelleh et al., 2010). Despite differences in exocytosis kinetics and the organization of fusion sites between synapses and endocrine cells, we believe that the core exo-endocytosis coupling mechanism is conserved. Therefore, if we can improve the temporal and spatial resolution and duration of current super-resolution imaging technologies, direct visualization of fusion pore intermediates in endocrine cells may invoke new insights that would render much of the discussed theories here obsolete.

#### AUTHOR CONTRIBUTIONS

All authors listed, have made substantial, direct and intellectual contribution to the work, and approved it for publication.

#### ACKNOWLEDGMENTS

The work was supported by grants from the National Natural Science Foundation of China (31327901, 31428004, 31521062, 31570839), the Major State Basic Research Program of China (2013CB531200), the National Science and Technology Major Project Program (2016YFA0500400), the Beijing Natural Science Foundation (7142071), and the Beijing Health System High Level Health Technical Personnel (2014-3-058).

#### REFERENCES


Proc. Natl. Acad. Sci. U.S.A. 105, 17555–17560. doi: 10.1073/pnas.08096 21105


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Liang, Wei and Chen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Systematic RNAi Screen Reveals a Novel Role of a Spindle Assembly Checkpoint Protein BuGZ in Synaptic Transmission in *C. elegans*

Mei Han1, 2, 3 †, Wenjuan Zou1 †, Hao Chang2, 3 †, Yong Yu2 †, Haining Zhang<sup>2</sup> , Shitian Li <sup>1</sup> , Hankui Cheng<sup>1</sup> , Guifeng Wei <sup>2</sup> , Yan Chen<sup>2</sup> , Valerie Reinke<sup>3</sup> , Tao Xu<sup>2</sup> \* and Lijun Kang<sup>1</sup> \*

*<sup>1</sup> Key Laboratory of Medical Neurobiology of the Ministry of Health of China, Department of Neurobiology, Institute of Neuroscience, Zhejiang University School of Medicine, Hangzhou, China, <sup>2</sup> National Laboratory of Biomacromolecules, CAS Center for Excellence in Biomacromolecules, Institute of Biophysics, Chinese Academy of Sciences, Beijing, China, <sup>3</sup> Department of Genetics, Yale University School of Medicine, New Haven, CT, USA*

*Edited by:*

*Jiajie Diao, University of Cincinnati, USA*

#### *Reviewed by:*

*Min Liu, Harvard Medical School, USA Zhitao Hu, University of Queensland, Australia Zhiyong Shao, Fudan University, China*

*\*Correspondence:*

*Tao Xu xutao@ibp.ac.cn Lijun Kang kanglijun@zju.edu.cn*

*† These authors have contributed equally to this work.*

> *Received: 28 February 2017 Accepted: 25 April 2017 Published: 11 May 2017*

#### *Citation:*

*Han M, Zou W, Chang H, Yu Y, Zhang H, Li S, Cheng H, Wei G, Chen Y, Reinke V, Xu T and Kang L (2017) A Systematic RNAi Screen Reveals a Novel Role of a Spindle Assembly Checkpoint Protein BuGZ in Synaptic Transmission in C. elegans. Front. Mol. Neurosci. 10:141. doi: 10.3389/fnmol.2017.00141*

Synaptic vesicles (SV) store various neurotransmitters that are released at the synapse. The molecular mechanisms of biogenesis, exocytosis, and endocytosis for SV, however, remain largely elusive. In this study, using Complex Object Parametric Analysis and Sorter (COPAS) to monitor the fluorescence of synapto-pHluorin (SpH), we performed a whole-genome RNAi screen in *C. elegans* to identify novel genetic modulators in SV cycling. One hundred seventy six genes that up-regulating SpH fluorescence and 96 genes that down-regulating SpH fluorescence were identified after multi-round screen. Among these genes, *B0035.1 (bugz-1)* encodes ortholog of mammalian C2H2 zinc-finger protein BuGZ/ZNF207, which is a spindle assembly checkpoint protein essential for mitosis in human cells. Combining electrophysiology, imaging and behavioral assays, we reveal that depletion of BuGZ-1 results in defects in locomotion. We further demonstrate that BuGZ-1 promotes SV recycling by regulating the expression levels of endocytosis-related genes such as rab11.1. Therefore, we have identified a bunch of potential genetic modulators in SV cycling, and revealed an unexpected role of BuGZ-1 in regulating synaptic transmission.

Keywords: RNAi screen, synaptic transmission, *C. elegans*, C2H2 zinc-finger protein, synaptic vesicles

## INTRODUCTION

Synaptic vesicles (SV) store neurotransmitters, concentrate in the presynaptic nerve terminals, and undergo Ca2+-dependent exocytosis. These steps include biogenesis of SVs, transport to release sites, docking with plasma membrane, priming, and calcium-triggered fusion (Sudhof and Rizo, 2011; Rizo and Xu, 2015). After exocytosis, SVs undergo endocytosis, recycle, and refilling with neurotransmitters for next round of exocytosis (Wu et al., 2014; Rizo and Xu, 2015; Xie et al., 2017). Three modes of exocytosis, including full-collapse fusion, kiss-and-run, and compound exocytosis, are coupled to classical endocytosis, kiss-and-run, and bulk endocytosis, respectively (Wu et al., 2014; Rizo and Xu, 2015; Xie et al., 2017).

The fusion of SVs to the plasma membrane is mechanically driven by the interaction among SNARE complex, a four-helix coiled-coil structure, which consists of a vesicle SNARE (v-SNARE) protein, synaptobrevin (on vesicle membrane), and two target membrane SNARE (t-SNARE) proteins, syntaxin-1, and SNAP-25 (on the plasma membrane) (Sudhof and Rizo, 2011; Südhof, 2013; Rizo and Xu, 2015). Fusion-competent conformations of SNARE proteins are maintained by chaperone complexes including CSPα, Hsc70, and SGT (Sudhof and Rizo, 2011). The synaptic SNARE and SM fusion-machine is controlled by synaptotagmin by Ca2<sup>+</sup> via synaptotagmin and complexin, and is additionally regulated by a presynaptic active zone proteins that includes Munc13 and RIM as central components (Sudhof and Rizo, 2011). Classical endocytosis is clathrin-dependent (Wu et al., 2014). All three SNARE proteins that catalyze exocytosis—synaptobrevin, SNAP25, and syntaxin are also needed for endocytosis initiation (Sudhof and Rizo, 2011; Wu et al., 2014; Xie et al., 2017). Some molecules such as amphiphysin, endophilin, AP180, auxilin, and dynamin have been implicated to be involved in endocytosis (Wu et al., 2014). Additionally, endocytosis is prolonged by depletion of clathrin, AP2, stonin 2, endophilin, and auxilin (Wu et al., 2014). Although the principle steps and some molecules have been identified, the exact mechanisms of biogenesis, exocytosis and endocytosis of SVs remain largely elusive.

Synaptobrevin is the key molecule on SVs, so synaptopHluorin (SpH), a pH-sensitive variant of GFP (pHluorin) fused to the luminal domain of synaptobrevin, is widely used to quantitatively measure the exocytosis and endocytosis of SVs (Miesenböck et al., 1998; Sankaranarayanan et al., 2000; Dittman and Kaplan, 2006; Afuwape and Kavalali, 2016). At rest, SpH fluorescence is quenched by the luminal acidic pH of the vesicle. After stimulation, vesicles fuse with the plasma membrane exposing the lumen to the neutral pH of the extracellular medium and causing an increase in SpH fluorescence. The fluorescence is then quenched once again after endocytosis and reacidification (Sankaranarayanan et al., 2000; Afuwape and Kavalali, 2016).

C. elegans is an excellent model system for studying SV cycling and performing systematic RNAi screen (Richmond and Broadie, 2002; Jadiya et al., 2016). With the availability of the C. elegans whole-genome RNAi feeding library, the expression of endogenous genes can be specifically knocked down by feeding bacteria expressing double-stranded RNA (dsRNA) of corresponding genes (Kamath et al., 2003). In order to get a clearer picture of the SV cycling, we aimed to identify novel genes required for synaptic vesicle cycling via C. elegans whole-genome RNAi screen, using SpH as the probe. Previous whole genome RNAi screens in C. elegans mostly detected and scored phenotypes by eyes, with limited quantifiable results, quality control, and systematic analysis. In this study, we detected the SpH fluorescence intensity of individual worms via Complex Object Parametric Analysis and Sorter (COPAS, Union Biometrica), which provides a method for high-throughput, reproducible quantitative analysis (Pulak, 2006; Dupuy et al., 2007). We screened two RNAi feeding libraries, including both Ahringer Library and Vidal Library (Kamath et al., 2003; Rual et al., 2004; Kim et al., 2005), together covering 94% of predicted C. elegans genes. We identified 176 genes up-regulating SpH fluorescence and 96 genes downregulating SpH fluorescence after multi-round screen. Among these candidate genetic modulators of SV cycling, the C2H2 zinc-finger protein BUGZ-1, an ortholog of mammalian spindle assembly checkpoint protein BuGZ/ZNF207, is critically required for SV cycling, suggesting an unidentified role of spindle assembly checkpoint proteins in synaptic transmission.

### RESULTS

#### A Whole-Genome RNAi Screen Identifies Novel Genes Required for Synaptic Vesicle Cycling

To identify novel genes related to synaptic vesicle (SV) cycling, we performed an automatic whole-genome RNAi screen in C. elegans. We used a pan-neuronal expressed snb-1 promoter to drive SNB-1::pHluorin (SpH) expression in the nerve system for optical measurements of presynaptic activity. SpH is a PH-sensitive variant of GFP fused to the luminal domain of synaptobrevin. Previous studies have confirmed that the fluorescence of SpH is quenched in the acidic environment of the SV lumen but increased dramatically when the SVs fused to the plasma membrane (Miesenböck et al., 1998; Sankaranarayanan et al., 2000; Dittman and Kaplan, 2006; Afuwape and Kavalali, 2016). The transgenic strain Is[snb-1::pHluorin] was crossed with the RNAi hypersensitive strain KP3948 eri-1(mg366); lin-15b(n744) (Sieburth et al., 2005) to generate worm for wholegenome RNAi screen. We performed screen with COPAS, a machine which is able to effectively detect the changes of fluorescence intensity in C. elegans by line scanning, and profile the fluorescent intensity for each worm (Figure S1, Dupuy et al., 2007; Han et al., 2013). For high-throughput screen using COPAS, worms were incubated in liquid culture in standard flat-bottomed 96-well plates according to previous reports with some modification (Lehner et al., 2006). We developed series of programs to batch process the raw data of COPAS. Relative florescent signals (RFS) were used to represent the SpH fluorescent signal, and robust Z-score were calculated to normalize SpH fluorescent intensity from different experimental 96-well plates (see Materials and Methods).

We chose unc-11/AP180, which has been reported to play an important role in recycling synaptobrevin from plasma membrane (Nonet et al., 1999; Dittman and Kaplan, 2006; Sudhof and Rizo, 2011), as an up-regulated control (up regulate SpH fluorescent intensity). Knocking down unc-11 AP180 6-fold increased the fluorescent intensity of SpH in the probe strain (Figure S1). We used the RNAi bacteria expressing dsRNA of gfp, whose sequence is similar to pHluorin, as a down-regulated control (down regulate SpH fluorescent intensity). Knocking down pHluorin by feeding worms with gfp RNAi bacteria decreased the fluorescent intensity of SpH to <0.5-fold in worms. The fluorescent intensities of worms cultured in 96-well plates in liquid which detected by COPAS were consistent with the fluorescent intensities of worms cultured on RNAi plates which detected by the confocal microscope. These results indicated that positive genes can be knocked down by feeding worms with corresponding RNAi bacteria in liquid culture, and the changes of fluorescent intensity can be detected efficiently by the COPAS.

In order to monitor the quality of RNAi treatment for each 96 well plate, we added up-regulated control of unc-11 RNAi clones, down-regulated control of gfp RNAi clones, and empty vector control of L4440 clones into the empty wells of each 96-well plate. We rearranged a few clones in 96-well plates to insure each experimental plate contains two empty vector controls, one upregulated control and one down-regulated control. Synchronized L1 worms were fed with each RNAi clone in each well of 96 well plates, and SpH fluorescence was detected in their progeny via the COPAS. For those RNAi clones that caused embryonic lethal or sterile, SpH fluorescence was detected in young adults of the same generation (**Figure 1A**). All bacteria RNAi clones were duplicated in the whole-genome screen. The plates with low repeatability (fold change of the florescent signals between the two repeats larger than 1.5) were redone. We examined the repeatability of two repeats for each clone, the overall correlation of the paired repeats is 0.81, which means that most paired repeats have very similar values (**Figure 1B**). Based on the genomic distribution of fluorescent intensity change after RNAi treatment, genes fell in the two tails (5%) were chosen as candidates. The candidates were rearranged into new 96 well plates with controls for further validation. In the secondary validation, worms were retested under the same condition. Each RNAi clones had four repeats. Genes that caused stable and consistent SpH signal changes were chosen as positive hits.

The fluorescent intensity changes of SpH could also due to unspecific reasons, especially the changing of protein expression level, so it is necessary to exclude unrelated genes for SV cycle. We checked the expression level of GFP in neurons, using the RNAi hypersensitive worm strain nre-1(hd20);lin-15b(hd126);rhIs13[unc-119p::GFP + dpy-20(+)] (Schmitz et al., 2007). Genes that lead to significant fluorescent intensity changes (p < 0.05) were excluded. Finally, 176 genes with up-regulated fluorescent intensity of SpH and 96 genes with down-regulated fluorescent intensity of SpH were identified after multi-round screen (Table S2). Seventy-six percent of these genes are evolutionarily conserved. The functional classes of candidate genes indicated that diverse groups of genes were taken part in SV cycling (**Figure 1C** and Table S3). Although some genes were already identified to be required for SV cycling, there still a big portion of genes with less known, which maybe new important regulators for SV cycling.

#### Mutants of Candidate Genes Display Acetylcholine Release Defects

In C. elegans, steady-state acetylcholine (ACh) secretion can be indirectly detected by measuring their resistance to the acetylcholine esterase inhibitor aldicarb (Mahoney et al., 2006b). Accumulation of ACh at synapsis caused by aldicarb leads to acute paralysis of worms and finally death. Blocking of SV cycle could lead to less release of ACh, which can be detected by aldicarb resistance analysis (Lackner et al.,

Frontiers in Molecular Neuroscience | www.frontiersin.org

percentage distribution chart of candidate genes within indicated functional classes. The total number of genes in each group and its percent to the total number of candidate genes are indicated in the parenthesis. 272 candidate genes were found for synaptic vesicle cycling after multi-round RNAi screen (Tables S2, S3).

1999; Mahoney et al., 2006b). To validate the participation of candidates in neurotransmitter release, we examined the resistance to aldicarb in 14 mutations which lead to dramatic SpH fluorescent intensity changes when knocking down by RNAi treatment. We used unc-32(e189) as a positive control in aldicarb resistance assay (Wiese et al., 2012). unc-32 encodes an ortholog of subunit of the membrane-bound domain of vacuolar proton-translocating ATPase. unc-32(e189) is a specific mutant that is important for locomotion and SV morphology in motoneurons (Wiese et al., 2012). Nine mutants showed significant aldicarb-resistance phenotypes, including B0035.1(tm578), Y71G12B.11(ok1648), Y25C1A.7(tm2889), F41D9.3(ok695), F45E4.3(ok2285), Y76A2A.2(gk107), T23F11.1(tm3480), F59E12.11(tm3828), C18B12.2(tm1690) (**Figure 2A**). The other five mutants showed no resistance (data not shown). All these candidate genes we identified are evolutionarily conserved, of which orthologs can be found in human, mouse, and fly, indicating their critical roles in diverse species.

In order to identify whether candidate genes function in presynaptic or postsynaptic terminal, we examined their resistance to the drug levamisole. Levamisole is a cholinergic receptor agonist that directly activates postsynaptic ACh receptors (Richmond and Jorgensen, 1999; Culetto et al., 2004). Wild type worms and mutants defected in presynaptic terminal display similar paralyzed phenotypes. All these 9 mutants exhibited similar paralysis rates with wild type (Figure S2), indicating a presynaptic function of candidate genes. Besides, SpH fluorescent intensities were significantly changed when knocking down these genes with RNAi treatment in the primary and the secondary screen, while GFP signals unaltered in nerve system of VH624 worms after RNAi treatment. These results also indicated that these genes probably function in presynaptic nerve terminal.

FIGURE 2 | Deletion mutants of candidate genes show acetylcholine secretion defects. Acetylcholine release was detected by determining the proportion of paralyzed animals exposed to 1 mM aldicarb. (A–I) Proportion paralyzed over time from indicated deletion mutant worms. Worm strains of *B0035.1(tm578)*, *Y71G12B.11(ok1648), Y25C1A.7(tm2889)*, *F41D9.3(ok695)*, *F45E4.3(ok2285), Y76A2A.2(gk107), T23F11.1(tm3480)*, *F59E12.11(tm3828)*, and *C18B12.2(tm1690)* exhibited aldicarb-resistance. *UNC-32(e189)* as aldicarb resistant positive control. 25–35 young adult animals were tested for each experiment, at least three independent experiments were performed. Error bars represent SEM.

#### BuGZ-1 Is a C2H2 Zinc-Finger Protein Expressed in the Nuclei of Neurons and Muscles

Among these 9 candidate genes, B0035.1 (bugz-1) encodes an ortholog of mammalian C2H2 zinc-finger protein BuGZ/ZNF207. Very interestingly, recent studies have been implicated that mammalian BuGZ binds to and stabilizes spindle check point protein Bub3 during interphase and facilitates mitosis (Jiang et al., 2014; Toledo et al., 2014). Depletion of BuGZ in cancer cells causes chromosome misalignment and mitotic arrest followed by massive cell death (Jiang et al., 2014; Toledo et al., 2014). Our observation implicates that BUGZ may also have an unidentified role in synaptic transmission, thus we focused our following study on BUGZ-1. The C. elegans gene bugz-1 encodes two isoforms of BuGZ-1, BuGZ-1S (B0035.1a) and BuGZ-1L (B0035.1b), both containing a nuclear localization sequence (NLS) and two C2H2 zinc finger domains at the N-terminal of the protein (**Figures 3A,B**). BUGZ-1 is highly conserved from worm to human and exhibits 68% identity between amino acids 1 and 107 of mouse and human homologs (**Figure 3C**). We generated a bugz-1p::gfp construct by driven GFP under the promoter of the bugz-1 gene. GFP signals were observed throughout nervous system and muscles, including

exons; blue lines indicate introns. 8X backcross of *bugz-1(tm578)* to wild type was performed before doing experiments. (C) Multiple sequence alignment of the highly conserved N-terminal of BuGZ-1 from different species, including humans (*H. sapiens*), mouse (*M. musculus*), and worm (*C. elegans*). Perfect sequence conservation is indicated in black. (D) Transcriptional expression pattern of *bugz-1p::gfp. bugz-1* is expressed in the nervous system, intestine and muscles. Images were taken from head, valve, and tail in ventral view, head to the left. (E) Translational expression pattern of *bugz-1::gfp.* Neurons are marked with *rab-3p::mCherry*. Images were taken from head and tail in lateral view, head to the left. BuGZ-1 is mainly localized to the nuclei of neurons. Scale bars indicate 20 µm.

the nerve ring and ventral nerve cord (**Figure 3D**). In order to confirm the nervous system expression of BuGZ-1, we co-injected the constructs of bugz-1p::gfp and rab-3p::mCherry. BUGZ-1 is co-localized well with RAB-3 in most neurons in the nerve ring and the tail. Notably, BUGZ-1 is localized to the nuclei of these cells, consistent with the subcellular location of mammalian BuGZ (**Figure 3E**, Jiang et al., 2014; Toledo et al., 2014).

#### Depletion of BuGZ-1 Results in Locomotion Defects

To further investigate the function of BuGZ-1 in synaptic vesicle cycling, we used a mutant strain bugz-1(tm578) generated by the National Bioresource Project (Tokyo, Japan). Five out of six exons are deleted in bugz-1(tm578) mutant worms. The deletion starts from the promoter region of bugz-1, which makes bugz-1(tm578) a null allele (**Figure 3B**). While wild type animals exhibited a smooth and continuous sinusoidal movement, bugz-1(tm578) animals displayed more hesitating and uncoordinating movement with decreased trajectories and locomotion speed (**Figures 4A,B**). Moreover, mutants prefer to move backward (ratio of total time = 25.3 ± 10.4%) or stay immobile (3.85 ± 4.01%) when compared with wild type worms (5.8 ± 7.3% and 0.76 ± 1.66%, respectively) (**Figures 4C,D**).

### BuGZ-1 Regulates Neurotransmitter Release at NMJs

C. elegans locomotion is controlled by nervous system that innervate specific muscles. Defects of either neuronal or muscle activities lead to locomotion abnormal (Richmond and Jorgensen, 1999; Dittman and Kaplan, 2006). We observed that bugz-1 mutants showed a dramatic resistance to aldicarb, an inhibitor for acetylcholine esterase (Dittman and Kaplan, 2006; Mahoney et al., 2006b, **Figure 4E**). The aldicarb-resistance phenotype and locomotion defects in bugz-1 mutants can be rescued by transgenic expression of BUGZ-1 cDNA driven by its own promoter or by the pan-neuronal promoter rab-3 (**Figures 4B–E**, Mahoney et al., 2006a), suggesting that BuGZ-1 predominately functions in nervous system to regulate synaptic transmission.

We then directly measured neurotransmitters release by recording the endogenous excitatory postsynaptic currents (EPSC) and evoked EPSCs at NMJs in zxIs6 animals, in which a light-gated cation channel, channelrhodopsin-2 (ChR2), is expressed specifically in cholinergic motor neurons (Liewald et al., 2008; Yang et al., 2015). The evoked EPSCs were recorded in voltage-clamped muscles with 100 ms blue light stimulation. Endogenous EPSC of bugz-1 mutants decreased 50% in the frequency whereas the amplitudes of release events didn't altered, indicating a shrink in spontaneous release events (**Figures 5A–C**). The evoked response in bugz-1 mutants also displayed a 55% decrease, suggesting a significant defect in regulated release (**Figures 5D,E**). The defects in both endogenous and evoked EPSCs were rescued by neuronal-expression of BuGZ-1 (**Figures 5A–E**). Furthermore, 1 mM acetylcholineinduced currents in body muscles didn't show any differences between wild type and bugz-1 mutant animals (**Figure 5F**). These results suggest that the decrease of EPSCs at NMJs is due to decreased neurotransmitter release from motor neurons rather than altered response to neurotransmitters of body muscles.

### BuGZ-1 Promotes Synaptic Vesicle Endocytosis

The SpH fluorescence was largely increased when BuGZ-1 was knocked down by RNAi treatment (1.43-fold change compared with L4440 empty vector). Significant increases of the SpH fluorescence were also observed in the nerve ring, the ventral nerve cord, as well as the axonal SpH puncta and inter-puncta of the dorsal nerve cord in bugz-1(tm578) mutations (**Figure 6**), suggesting an increase in SV exocytosis or a decrease in endocytosis in bugz-1(tm578) background. Given the neurotransmitter release was decreased at NMJs in bugz-1 mutant animals (**Figure 5**), we propose BuGZ-1 regulates synaptic transmission by promoting SV recycling, particularly endocytosis.

#### BuGZ-1 Regulates the Expression Levels of Endocytic Genes

The nuclei expression pattern of BuGZ-1 suggested that this protein may regulate neurotransmitter release by altering the expression level of some crucial genes related to synaptic vesicle recycling. Based on this speculation, we performed a high-throughput sequencing of C. elegans cDNA generated by isolating total RNA (RNA-seq) from wild type and bugz-1(tm578) worms (Table S4). We examined expression levels of coding sequence on a genome-wide and quantified the differences of expression levels for each gene between wild type and mutants. The expression levels of known genes essential for clathrinmediated endocytosis including unc-11/AP180, chc-1/clathrin, and snb-1/synaptobrevin (Sudhof, 2004; Wu et al., 2014) were significantly reduced in bugz-1(tm578) worms (**Figure 7A**). Rab11.1, a small GTPases Rab required for endocytic recycling in many eukaryotic species (Sato et al., 2008), also showed a reduced expression level in bugz-1(tm578) worms (**Figure 7A**). We asked whether BuGZ-1 and endocytic genes function sequentially in the recycling pathway or in parallel pathways. If they function in the same pathway, simultaneous knockout or knockdown would be expected to give a phenotype similar to single mutants of either gene. However, in bugz-1(tm578);unc-11(e47) double mutant worms, more significant increases of the SpH fluorescence were observed both in the nerve ring and the ventral nerve cord compared to either bugz-1(tm578) or unc-11(e47) single mutant worms. These strongly enhanced phenotypes in double mutants suggest that BuGZ-1 and UNC-11 mainly function in parallel pathways, though loss of BuGZ-1 may result in down-regulation of UNC-11 (**Figures 7B,C**). Notably, whereas rab-11.1(RNAi) shows similar phenotypes with increased SpH fluorescence with bugz-1(RNAi) worms, no synthetic effect was observed in bugz-1(RNAi) combined with rab-11.1(RNAi) depletion, suggesting

genotypes compared. Error bars represent SEM.

that BuGZ-1 and RAB-11.1 may function sequentially in the recycling pathway (**Figures 7D–F**). Considering BuGZ-1 is a predicted transcription factor and has nuclei localization, we performed chromatin immunoprecipitation followed by deep sequencing (ChIP-seq) of single copy transgenic worm BuGZ-1::gfp at young adult stage to determine binding sites of BuGZ-1 in the genome. We found that BuGZ-1 directly binds to rab-11.1 (Figure S3 and Table S5), indicates that BuGZ-1 may function in endocytosis by regulating RAB-11.1, which consistent with our conclusion that BuGZ-1 and RAB-11.1 may function sequentially in SV recycling pathway.

### DISCUSSION

In this study, we performed a high-throughput and quantitative whole-genome RNAi screen in C. elegans using COPAS. We identified 176 genes with up-regulated SpH fluorescence and

96 genes with down-regulated SpH fluorescence, which may be essential modulators for SV cycling. The functional classes of candidate genes indicated that diverse groups of genes, such as genes related to transcription, signaling, and kinase, transporter and channels, and membrane trafficking, are involved in SV cycling, supporting the vision that SV cycling is a complex and dynamic process involving a variety of mechanisms carried out by diverse groups of molecules.

Mammalian BuGZ/ZNF207 is a spindle assembly checkpoint protein associating with spindle microtubules and regulates chromosome alignment (Jiang et al., 2014; Toledo et al., 2014). Inhibition of BuGZ results in loss of both Bub3 and its binding partner Bub1 from kinetochores, and lethal chromosome congression defects in cancer cells (Jiang et al., 2014; Toledo et al., 2014). Nevertheless, a role of spindle assembly checkpoint proteins such as BuGZ in regulation of SV cycling have not been reported. By combining molecular genetics, optogenetics, electrophysiological recording and behavioral analysis, here we demonstrate that the C. elegans homolog of BuGZ is also a key modulator of synaptic vesicle recycling. Whether, BuGZ and other mitosis-associated proteins in other species have a role in synaptic transmission has yet to be identified. We envision that the functional characterization of C. elegans BuGZ may provide very useful insights into the functions of BuGZ and other mitosis-associated proteins in general.

The genetic interactions between bugz-1 and endocytic genes such as unc-11 and rab-11.1 strongly suggest that bugz-1 functions through, or in parallel to, clathrin-mediated endocytosis and vesicle recycling pathway during synaptic transmission. BuGZs are C2H2 zinc finger proteins, which represent the second largest gene family in humans after the odorant receptor family (Tadepally et al., 2008). Most of the

characterized C2H2 zinc finger genes code for transcription factors which bind DNA through their zinc finger region; others may bind RNA and protein motifs (Wang et al., 2006; Tadepally et al., 2008). However, their exact function is as yet unknown (Tadepally et al., 2008). Given that BuGZ-1 localizes to nuclear and down-regulates the expression level of some key endocytic proteins, it is likely that BuGZ-1 functions as a transcription factor or a posttranscriptional modulator for proteins essential for endocytosis. This transcriptional or posttranscriptional modulation may provide a fine tune mechanism for synaptic transmission.

Taken together, our study provides an example that performing a high-throughput and quantitative whole-genome RNAi screen in C. elegans, and identified diverse groups of molecules may be involved in SV cycling. These molecules may have a broader impact on protein and vesicle trafficking. Furthermore, we have directly correlated the role of a C2H2 zinc finger spindle assembly checkpoint protein with synaptic transmission in an in vivo setting.

#### MATERIALS AND METHODS

#### *C. elegans* Strains

All C. elegans strains were maintained on NGM plates at 20◦C using standard methods unless otherwise statement (Brenner, 1974). The plasmid SNB-1::pHluorin (SpH, gift from Joshua Kaplan) was injected into wild-type N2 worms and integrated into chromosome by UV/TMP method and backcrossed 6 times with N2 worms to remove background mutations. The integrated strain was crossed with a RNAi hypersensitive strain KP3948 eri-1(mg366);lin-15b(n744) (Sieburth et al., 2005) to make the transgenic strain eri-1(mg366);lin-15b(n744);kanIs8[snb-1::pHluorin] for RNAi screening.

The following strains were used in this study:

bugz-1(tm578), Y25C1A.7(tm2889), T23F11.1(tm3480), F59E12.11(tm3828), C18B12.2(tm1690), RB1445 Y71G12B.11(ok1648), RB861 F41D9.3(ok695), RB1777 F45E4.3(ok2285), VC194 Y76A2A.2(gk107), CB189 unc-32(e189), VH624 rhIs13[Punc-119::GFP + dpy-20(+)]; nre-1(hd20); lin-15b(hd126); zxIs6[Punc-17p::chop-2(H134R)::yfp, lin-15(+)] (a gift from Dr. Mei Zhen). Single copy worm strain xtl1186 kanIs54[bugz-1p::BUGZ-1::gfp, unc-119(+)] II; unc-119(ed9) III kanEx378[bugz-1p::BUGZ-1::gfp+ rab-3p::mCherry]

kanEx379[bugz-1p::gfp+ rab-3p::mCherry]

bugz-1(tm578); kanEx380[lin-44p::gfp+ rab-3p:: BUGZ-1:: mCherry]

bugz-1(tm578); kanEx392[lin-44p::gfp+ myo-3p:: BUGZ-1:: mCherry]

bugz-1(tm578);kanEx380[lin-44p::gfp+ rab-3p:: BUGZ-1:: mCherry]; zxIs6[unc-17p::chop-2(H134R)::yfp, lin-15(+)] bugz-1(tm578);kanEx392[lin-44p::gfp+ myo-3p:: BUGZ-1:: mCherry]; zxIs6[unc-17p::chop-2(H134R)::yfp, lin-15(+)].

#### RNAi Screen

Bacterial glycerol stocks of RNAi library was replicated onto LB-agar square plates containing 25 µg/ml carbenicillin and 15 µg/ml tetracycline using a 96-pin replicator. After overnight culture at 37◦C, bacterial clones were replicated from square plates into 1.2 mL 96-well plates containing 400 µl LB liquid medium with 25 µg/ml carbenicillin in each well. The RNAi bacterial clones in 1.2 mL 96-well plates were cultured overnight at 37◦C with shaking. Added IPTG to bacterial cultures to a final 1 mM concentration to induce transcription of doublestranded RNA and incubated at 37◦C for 1 h with shaking. Bacterial clones were spun down and resuspended with 200 µl S-Basal buffer containing 50 µg/ml carbenicillin and 1 mM IPTG. Forty microliters concentrated bacterial suspension was added into each well of standard 96-well plates. RNAi bacteria

0.001 by two-tails Student's *t*-test). N.S. indicates no significant difference between genotypes compared. Error bars represent SEM.

from original 96-well RNAi library were rearranged to make each experimental 96-well plate contained two L4440 controls, one gfp RNAi down-regulated control and one unc-11 RNAi up-regulated control.

Adult hermaphrodite worms were bleached using standard method. Synchronized L1s were washed and resuspended in S-Basal containing 1 mM IPTG, 50 µg/ml carbenicillin, and 0.01% Tween-20. 5–8 L1s/10 µl S-Basal for F1 generation screen or 120 L1s/10 µl for P0 generation screen. Ten microliters worm solution was added into each well of 96-well plates which pre-added 40 µl concentrated RNAi bacterial suspension. Experimental 96-well plates containing bacterial clones and L1s were incubated in humid chambers at 20◦C. To increase RNAi efficiency, we used two-generation RNAi treatment for wholegenome screen: synchronized L1 worms were cultured for 6 days, and fluorescent intensity of their progeny were detected by COPAS (Han et al., 2013). The RNAi clones which knocking down reduced the progeny number to less than 20% of average progeny number of the wells in the same plate were defined as sterile or lethal. We rearranged them into 25 new 96-well plates with controls. For those clones, worms were cultured for 2.5 days and L4-young adult stage of the same generation were detected by COPAS.

#### Statistical Analysis the Screen Data of COPAS

We use the relative fluorescent signal (RFS) to represent synaptopHluorin (SpH) signal of each worm. The green fluorescent signals (represent all the SpH fluorescent signals of the worm) divided by the EXT signals (extinction integral value, represented the worm size) would result in the normalized SpH signals. Log2transformation was used to transform the fluorescent signals linearly. The median of RFS in each well was used to represent the signal of each well.

$$\text{RFS} = \log\_2 \frac{Green \text{ fluorescent signals}}{EXT \text{ signals}}$$

Robust Z-score (rZ) was used to normalize screen data in all the plates in the whole-genome screen. Robust Z-score was similar to Z-score except that the sample median and sample median absolute deviation (MAD) were used instead of sample mean and sample standard deviation, and thus was not sensitive to outliers (Birmingham et al., 2009).

$$\text{rZ} = \frac{X - median\,(X)}{MAD\,(X)}$$

For variable X, MAD is defined as the median of the absolute deviations from the median of X .

After robust Z-score normalization, the rZ score of two repeats were averaged to represent the SpH fluorescent intensity of the corresponding genes upon RNAi treatment.

In the secondary and cytoplasmic GFP screen, student's ttest was used to confirm the positive hits. For each plate, after removing the low repeatability genes (fold change between the two repeats was >2), one tailed t-test was used to identify the experimental RNAi bacteria that were significantly different from L4440 empty control.

Those RNAi bacteria with p < 0.05 in secondary screen and >0.05 in the cytoplasmic GFP screen were considered as positive hits.

#### Aldicarb-Sensitivity Assay

Aldicarb-sensitivity experiments were performed on NGM plates containing 1 mM aldicarb as previously described (Wiese et al., 2012). Prepare aldicarb plates 24 h before the assay and put them at room temperature to let the plates dry. Pick 25–35 L4 worms to a fresh seeded NGM plate 16 h before the assay. Place a small spot of OP50 in the middle of aldicarb plates and let it dry thoroughly. Young adult animals were transferred to aldicarb plates and tested for paralysis every 10 min for 2 h with a harsh touch on the head. Worms failed to response to the touch were identified as paralyzed.

### Constructs and Transgenes

All expression plasmids were based on the pPD95.75 vector unless otherwise statement. The putative bugz-1 promoter region including 5 kb upstream of the gene bugz-1. bugz-1p::gfp fusion construct was generated by PCR to amplify the bugz-1 promoter region. The PCR product of bugz-1 promoter was digested with XmaI and AgeI–HF restriction enzymes and ligated into the pPD95.75 vector. Based on the highly consistency of functional domains between the two isoforms of bugz-1, we cloned the long isoform of bugz-1 for all the rescue studies. BuGZ-1 cDNA was cloned from a cDNA library using primers targeted to the start and stop codons of the long isoform, bugz-1b. To generate bugz-1p:: gfp, we cloned bugz-1 cDNA into bugzp-1::GFP using AgeI single restriction enzyme. A 1.3 kb rab-3 promoter was used to drive bugz-1 expression in nervous system and a 1.2 kb myo-3 promoter was used to drive bugz-1 expression in body muscles. bugz-1 cDNA was digested by AgeI single restriction enzyme and then inserted into rab-3p::mCherry or myo-3p::mCherry to generate rab-3p::BUGZ-1::mCherry or myo-3p::BUGZ-1::mCherry. GFP or mCherry was fused to the Cterminus of bugz-1 cDNA as a reporter. Transgenic worm strains were obtained by microinjection of corresponding plasmids with Plin-44::gfp or Prab-3::mCherry as a marker. For expression analysis, constructs were injected at 80 ng/µl. For rescue assays, constructs were injected at 10 ng/µl. All the markers used were injected at 30∼50 ng/µl.

### Single Copy Insertion

Transgenic worms were generated by injection constructs into EG4322 [ttTi5605; unc-119(ed9)] animals (Frøkjaer-Jensen et al., 2008). The standard injection mix consisted of 50 ng/µl bugz-1::gfp, unc-119(+) repair template, 50 ng/µl Mos1 transposase pJL43.1(Pglh-2::transposase), 5 ng/µl myo-3p::mCherry which as a negative marker. Injected animals were transferred to NGM plates, one worm per plate. Individual injected worms were allowed to exhaust the food source of each plate. Once starved, L1 progeny were screened for insertion events with GFP fluorescence and wild-type movement but lack of co-injection marker expression.

#### Image Analysis

All the images were obtained using a FV 1000 laser scanning confocal microscope (Olympus). Confocal images were captured using a 60X objective with NA 1.4 at 1x or 2x digital zoom. Worms were immobilized with 30 mM NaN<sup>3</sup> (Sigma) on agarose pads. For quantitative analysis, images were acquired in young adult worms and maximum fluorescent intensity of Z-series was stacked. Background fluorescence was subtracted before analysis. P-values were calculated by student t-tests. Images were quantified and analyzed using FV10-ASW Viewer and Image J. For fluorescent analysis of SpH signal in the nerve ring and ventral nerve cord, average of stacked maximum fluorescent intensity were calculated to represent florescent intensity. For fluorescent analysis in the dorsal nerve cord, florescent intensity of puncta, and inter-puncta were measured using Igor Pro software.

#### Locomotion Analyses

Locomotion experiments were performed by capturing movies of animals free moving on NGM plates with fresh seeded E. Coli OP50 8–12 h before. Fifteen young adult animals (12–14 h after L4 stage) were transferred to each 60 mm NGM plate. Ten min after transfer, a 1 min movie of animal moving was recorded using a digital camera installed on a Zeiss dissecting microscope. Movies of locomotion behaviors were analyzed by the "Imaging the Behavior of Nematodes (iBeN)" system developed in our lab. Software of iBeN system is developed based on the computer vision library, OpenCV (Open Source Computer Vision). Locomotion rates and movement status were achieved automatically by iBeN system.

#### Electrophysiology

Electrophysiology assays were performed at the neuromuscular junctions of dissected C. elegans as previously described (Richmond and Jorgensen, 1999; Kang et al., 2010; Yang et al., 2015). Day 2 adult worms were glued on the surface of Sylgardcoated coverslips using cyanoacrylate-based glue (Zou et al., 2017), and a dorsolateral incision was made using a sharp glass pipette to expose the body wall muscles for recording. Wholecell recordings of ventral body wall muscles were carried out by a HEKA EPC10 amplifier using the Patchmaster software. Recording pipettes were pulled from borosilicate glass capillaries (Sutter Instruments) to a resistance of 3–4 M on a P-97 micropipette puller (Sutter Instruments). The bath solution contained 145 mM NaCl, 2.5 mM KCl, 5 mM CaCl2, 1 mM MgCl2, 20 mM glucose and 10 mM HEPES (325–335 mOsm, pH 7.3). The pipette solution contained145 mM KCl, 2.5 mM KCl, 5 mM MgCl2, 0.25 mM CaCl2, 10 mM HEPES, 10 mM glucose, and 5 mM EGTA, 5 mM ATP, 0.5 mM GTP (325∼335 mOsm, pH 7.2). Membrane potential was clamped at −60 mV. For acetylcholine-activated experiments, 500 µM acetylcholine was perfused to the bath solution. The zxIs6 strain, in which light-gated cation channel channelrhodopsin-2 (ChR2)-YFP was expressed in cholinergic neurons, was used for recording evoked EPSCs (Liewald et al., 2008; Yang et al., 2015). All-trans retinal was added to the NGM plates at a final concentration of 2.5 µM to mediate light stimulation of ChR2. L4 animals were transferred to NGM plates containing all-trans retinal and the next generation of 2-day-old hermaphrodite adults were used for electrophysiological studies. Blue light stimulation were performed by LAMBDA XL (Sutter Instruments) with a GFP filter controlled by the Patchmaster software.

#### RNA Isolation and qPCR Analysis

RNA was extracted from young adult animals using Trizol reagent (Invitrogen). Total RNA was reversed to cDNA using Reverse Transcription System A3500 (Promega). For quantitative RT-PCR, Bio-Rad CFX96 real-time PCR Amplifier was used to run the cycles. qPCR was performed in triplicate for three independent biological experiments. Relative gene expression levels were calculated by 1Ct method.

#### ChIP-Sequencing

ChIP assays were performed as previously described (Kudron et al., 2013; Kasper et al., 2014). Young adult worms were collected after 50 h post synchronized L1. Worms were crosslinked in 2% formaldehyde for 30 min and then quenched with 1 M Tris 7.5. Worm sample was sonicated to obtain 200– 800 bp DNA fragment. 4.4 mg cell extract from the sonicated worm sample was immunoprecipitated with 7.5 µg of GoatVαGFP antibody (gift from Kevin White). Deep sequencing was performed on the Illumina Hiseq 2500 platform for the immunoprecipitated DNA fragments and genomic DNA input control. Sequencing consortium version of C. elegans WS235 was used to align reads. Significant binding peaks were called with SPP and IDR algorithms (Landt et al., 2012). The closest coding gene to the peak maximum of a binding site was considered as a target of the transcription factor. The ChIP-seq raw data have been uploaded to the ENCODE wedsite for public viewing and downloading (https://www.encodeproject.org/experiments/ ENCSR450GPA/).

#### Statistical Analysis

Data analysis was performed using Excel or Igor 5. All data were presented in mean ± SEM. Unpaired two-tailed t-test was used for data comparison and P < 0.05 were considered to be statistically significant.

### AUTHOR CONTRIBUTIONS

MH, YY, LK, and TX conceived and designed the experiments. MH, WZ, HC, YY, HZ, SL, HKC, GW, and YC performed molecular genetics, optogenetics, behavioral, and electrophysiological experiments. MH, VR, LK, and TX analyzed and interpreted results. MH, YY, and LK wrote the manuscript and modified by all the other authors.

#### ACKNOWLEDGMENTS

We thank Nan Qiao and Jing-Dong J. Han for help with COPAS data analysis. We thank the modERN consortium, especially Michelle Kudron and Swapna Samanta, for help with ChIPseqencing, LaDeana Hillier for target calling analysis. We thank Shohei Mitani for providing tm deletion worm strains. We thank the CGC for providing worm strains (P40 OD010440). This work was supported by grants from the Major National Scientific Research Projects of the Ministry of Science and Technology of China (2013CB945603, 2015AA020512), the National Foundation of Natural Science of China (31271180, 31471023),

#### REFERENCES


and Zhejiang Natural Science Funds for Distinguished Young Scholars of China (LR14C090001).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2017.00141/full#supplementary-material


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Han, Zou, Chang, Yu, Zhang, Li, Cheng, Wei, Chen, Reinke, Xu and Kang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Unique Structural Features of Membrane-Bound C-Terminal Domain Motifs Modulate Complexin Inhibitory Function

David Snead<sup>1</sup> , Alex L. Lai<sup>2</sup> , Rachel T. Wragg<sup>1</sup> , Daniel A. Parisotto<sup>1</sup> , Trudy F. Ramlall<sup>1</sup> , Jeremy S. Dittman<sup>1</sup> , Jack H. Freed<sup>2</sup> and David Eliezer<sup>1</sup> \*

<sup>1</sup> Department of Biochemistry, Weill Cornell Medicine, New York, NY, United States, <sup>2</sup> Department of Chemistry and Chemical Biology, Cornell University, Ithaca, NY, United States

Complexin is a small soluble presynaptic protein that interacts with neuronal SNARE proteins in order to regulate synaptic vesicle exocytosis. While the SNARE-binding central helix of complexin is required for both the inhibition of spontaneous fusion and the facilitation of synchronous fusion, the disordered C-terminal domain (CTD) of complexin is specifically required for its inhibitory function. The CTD of worm complexin binds to membranes via two distinct motifs, one of which undergoes a membrane curvature dependent structural transition that is required for efficient inhibition of neurotransmitter release, but the conformations of the membrane-bound motifs remain poorly characterized. Visualizing these conformations is required to clarify the mechanisms by which complexin membrane interactions regulate its function. Here, we employ optical and magnetic resonance spectroscopy to precisely define the boundaries of the two CTD membrane-binding motifs and to characterize their conformations. We show that the curvature dependent amphipathic helical motif features an irregular element of helical structure, likely a pi-bulge, and that this feature is important for complexin inhibitory function in vivo.

## Keywords: complexin, amphipathic helix, membrane curvature, ESR, NMR, micelles, synaptic transmission,

### INTRODUCTION

pi-bulge

Synaptic function requires the precise regulation of synaptic vesicle exocytosis by a number of accessory proteins that modulate SNARE complex assembly and synaptic vesicle fusion with the plasma membrane (Rizo and Rosenmund, 2008; Sudhof and Rizo, 2011; Rizo and Xu, 2015; Schneggenburger and Rosenmund, 2015). Complexins are a family of highly conserved cytoplasmic proteins that facilitate calcium-evoked synchronous exocytosis but that can also inhibit spontaneous synaptic vesicle fusion (Brose, 2008; Mohrmann et al., 2015; Trimbuch and Rosenmund, 2016). By clamping synaptic vesicles in a fusion competent but inhibited state until evoked fusion is required, complexins help to prevent premature depletion of the synaptic vesicle pool and keep spontaneous synaptic noise to a minimum (Huntwork and Littleton, 2007; Maximov et al., 2009; Cho et al., 2010; Hobson et al., 2011; Martin et al., 2011; Kaeser-Woo et al., 2012; Wragg et al., 2013). The precise mechanisms by which complexins fulfill two discrete functions remain incompletely delineated. It has been clearly established, however, that complexins contain four

#### Edited by:

Jiajie Diao, University of Cincinnati, United States

#### Reviewed by:

Mark Bowen, Stony Brook University, United States Rosemary Beth Cornell, Simon Fraser University, Canada

\*Correspondence:

David Eliezer dae2005@med.cornell.edu

Received: 20 March 2017 Accepted: 08 May 2017 Published: 24 May 2017

#### Citation:

Snead D, Lai AL, Wragg RT, Parisotto DA, Ramlall TF, Dittman JS, Freed JH and Eliezer D (2017) Unique Structural Features of Membrane-Bound C-Terminal Domain Motifs Modulate Complexin Inhibitory Function. Front. Mol. Neurosci. 10:154. doi: 10.3389/fnmol.2017.00154 fusion.

fnmol-10-00154 May 22, 2017 Time: 16:51 # 2

domains (**Figure 1**): a highly conserved SNARE-binding central helix necessary for all complexin functions (Pabst et al., 2000; Bracher et al., 2002; Chen et al., 2002; Xue et al., 2007), a facilitatory N-terminal domain (Xue et al., 2007, 2010; Lai et al., 2016), an inhibitory accessory helical domain (Xue et al., 2007; Bin et al., 2010; Krishnakumar et al., 2011; Kümmel et al., 2011; Li et al., 2011; Bykhovskaia et al., 2013; Cho et al., 2014; Radoff et al., 2014; Trimbuch et al., 2014), and an inhibitory C-terminal domain (CTD) (the focus of this work) (Xue et al., 2009; Martin et al., 2011; Wragg et al., 2013; Snead et al., 2014).

helical AH motif and the subsequent short unstructured CT motif, localize and position complexin to engage the SNARE complex and inhibit synaptic vesicle

The complexin CTD is necessary for the inhibition of spontaneous exocytosis in both worms (Martin et al., 2011; Wragg et al., 2013) and flies (Xue et al., 2009; Cho et al., 2010; Iyer et al., 2013), insofar as the frequency of spontaneous fusion events increases upon deletion or mutation of this domain in these organisms. The function of the mammalian CTD is somewhat more enigmatic, and it has been suggested to both facilitate and inhibit vesicle fusion in different experimental contexts. Mammals contain multiple complexin isoforms with differing CTDs that may contribute in different ways to exocytosis, exerting either a facilitatory or inhibitory effect depending on the specific isoform (Reim et al., 2005; Schaub et al., 2006; Malsam et al., 2009; Seiler et al., 2009; Kaeser-Woo et al., 2012; Yang et al., 2013; Lai et al., 2014). Interestingly, the CTDs of both worm complexin-1 (CPX-1) (Wragg et al., 2013; Snead et al., 2014) and mouse complexin-1 (mCpx1) (Seiler et al., 2009) bind to phospholipid vesicles through a putative amphipathic helical region and in the case of the worm, this membrane interaction is necessary for inhibition of spontaneous synaptic vesicle exocytosis (Wragg et al., 2013; Snead et al., 2014). Fly complexin, too, binds to liposomes in vitro (Cho et al., 2015) and contains a putative amphipathic region (Wragg et al., 2013), though whether this region contributes to membrane-binding and protein function in flies remains unknown.

The CPX-1 CTD contains tandem lipid binding motifs – termed the C-terminal (CT) and amphipathic helix (AH) motifs – that together sense membrane curvature to preferentially bind highly curved membranes (Snead et al., 2014). Membrane curvature sensing by these two regions likely directs complexin to the highly curved synaptic vesicle membrane (Snead et al., 2014), its likely membranous binding target in vivo. Notably, worm complexin appears to function from the synaptic vesicle, but not the plasma membrane (Wragg et al., 2013). Both motifs are critical for the ability of complexin to inhibit synaptic vesicle exocytosis in worms, as perturbation of either motif impairs inhibitory activity in vivo. The CT motif contains a mix of bulky hydrophobic and positively charged residues that likely mediate its interaction with lipid membranes; a CPX-1 variant with this motif deleted cannot bind membranes in vitro and cannot inhibit exocytosis in vivo. The adjacent amphipathic region is intrinsically disordered in the free state and adopts helical structure only upon binding to highly curved membrane surfaces. Mutations that impair membrane-binding of the amphipathic motif also perturb inhibitory function in vivo; interestingly, mutations that selectively decrease helix formation, but not lipid binding, similarly fail to inhibit vesicle exocytosis, suggesting that helix formation by the amphipathic motif is critical for CPX-1 inhibitory function (Snead et al., 2014). Putative amphipathic helical regions have been identified for not only worm but also fly, and mammalian complexins, suggesting that it may be a conserved structural feature of the complexin CTD. In fact, phospholipid membrane-binding by a putative amphipathic helix of the complexin CTD was first established for mammalian complexin-1 (Seiler et al., 2009). Interestingly, many complexins contain a C-terminal CAAX box motif, including mammalian complexin-3 and -4. Farnesylation of this CAAX box motif mediates membrane association for these variants, further arguing that complexin CTD-membrane interactions are of critical functional significance (Brose, 2008; Xue et al., 2009; Cho et al., 2010; Buhl et al., 2013; Iyer et al., 2013).

Why is amphipathic helix formation required for CPX-1 function? To answer this question requires a clearer picture of the structural properties of the membrane-bound CPX-1 CTD. Prior structural characterization of the vesicle-associated CPX-1 CTD using solution-state NMR was limited by the fact that this slowly tumbling high molecular-weight protein–lipid complex is invisible to standard experiments. Optical methods such as circular dichroism (CD) spectroscopy, which are not limited by size, provided some information on helix formation, but only at low resolution (Snead et al., 2014). In this paper, we use two established and complementary techniques to directly observe and so better delineate the structure of the worm complexin-1 CTD in the membrane bound state at a residue-specific level. We first characterize the structure of the CTD in the presence of dodecylphosphocholine (DPC) micelles, a membrane mimetic amenable to study by

standard solution-state NMR experiments. We then apply electron spin resonance (ESR) spectroscopy with site-directed spin labeling to directly characterize the AH and CT motifs of phospholipid vesicle bound CPX-1 CTD. Together, these experiments provide a clearer picture of the structure and boundaries of the membrane-binding motifs of the CPX-1 CTD, which is confirmed in an accompanying study (Wragg et al., 2017) using functional assays in living worms. The results provide evidence for an interruption of regular helical structure in the AH motif at residue Gly 116. Deletion of Gly 116 restores regular helical structure to the AH motif and leaves membrane-binding unimpaired, but impairs CPX-1 inhibitory function in vivo. We conclude that specific structural features of membrane-bound complexin beyond simple amphipathic helix formation are required for proper inhibitory function and speculate that these act to template functionally important protein–protein interactions. Together with an accompanying study (Wragg et al., 2017), our work suggests that these interactions have undergone divergent evolution, potentially explaining reported differences in the inhibitory activity or nematode and mammalian complexin-1.

#### MATERIALS AND METHODS

#### Protein Purification

Worm CPX-1 wild-type full-length (lacking the N-terminal methionine), and the CTD fragment (residues 91–143) were cloned into the pET SUMO vector. Single cysteine mutants were generated via site-directed mutagenesis of the full-length wild-type construct. 1116 was similarly generated via site-directed mutagenesis of the isolated CTD fragment construct. Proteins were expressed either as previously described (Wragg et al., 2013; Snead et al., 2014) or with direct inoculation of isotopically labeled minimal media. Briefly, BL21(DE3) Escherichia coli cells were transformed with the relevant plasmid and grown at 37◦C for 5 h in a starter culture of Luria Broth with kanamycin. 2 mL starter culture was then diluted into 100 mL overnight culture – either LB/kanamycin or isotopically labeled minimal media/kanamycin – which was then grown at 37◦C for 16–20 h. The overnight culture was diluted to 1 L and grown to an optical density ∼0.6. For unlabeled growths (in LB) and for directly inoculated isotopically labeled growths (in minimal media in either H2O or D2O containing the appropriate isotopic reagents) protein expression was induced for 3–4 h by addition of IPTG. Alternately, to produce labeled protein as previously described, LB cultures at an optical density of 0.6 were pelleted, washed, and then transferred to isotopically labeled minimal media prior to induction with IPTG. Depending on the specific labeling scheme required, the minimal media used contained the appropriate mixture of 15N-labeled ammonium chloride, <sup>13</sup>C-labeled glucose, <sup>13</sup>C <sup>2</sup>H labeled glucose, or and/or <sup>12</sup>C 2H labeled glucose in either H2O or D2O.

Proteins were purified as previously described as well. Briefly, cells were lysed by sonication on ice and clarified by centrifugation at 40,000 rpm for 45 min. SUMO-tagged fusion protein was purified on a Ni-NTA column and then dialyzed into 20 mM Tris pH 8, 150 mM NaCl, 1 mM dithiothreitol. After cleavage by SUMO protease, purified complexin was isolated in the flow-through of a second run over the Ni-NTA column. Protein was dialyzed into distilled water and then frozen and lyophilized.

#### Protein Spin Labeling

CPX-1 mutants with single cysteine mutations spanning residues 110–136 were generated as noted above. These proteins were dissolved in PBS (pH 6.1) at 50 µM and mixed with S-(2,2,5,5-tetramethyl-2,5-dihydro-1H-pyrrol-3-yl)methyl methanesulfonothioate (MTSL, from Santa Cruz Biotechnology, Dallas, TX, United States) in a 1:10 ratio overnight at 4◦C and dialyzed against PBS buffer (pH 6.1) to remove excess MTSL. The spin-labeled proteins were mixed with DOPC:DOPE:DOPS (60:25:15) small unilamellar vesicles (SUVs) at about a 1:200 protein:lipid ratio for 30 min at room temperature in PBS buffer (pH 6.1) before measurements.

#### Liposome and Micelle Preparation

Lipids were purchased from Avanti Polar Lipids and were stored at −20◦C. Lipids – including DPC – were mixed at the desired ratios and dried under nitrogen gas; residual solvent was then removed by centrifugation under vacuum for 1 h or, for the ESR experiments, in a vacuum drier overnight. The resulting lipid film was then resuspended to the desired stock concentration. Preparation of DPC micelles was complete at this point, and this stock of DPC micelles was mixed with protein for sample preparation. For SUVs, the resuspended lipid film was bath sonicated until the solution became transparent. The SUV solution was then further clarified by ultracentrifugation at 60,000 rpm for 2 h (or, for the ESR experiments, at 13,000 g for 10 min). For large unilamellar vesicles (LUVs), the resuspended lipid film was frozen in liquid nitrogen and subsequently thawed in warm water for 10 freeze/thaw cycles. LUVs were then prepared using a 1 mL Avanti Mini-Extruder from Avanti Polar Lipids by extruding 21 times each through 400 nm and then 100 nm pore size polycarbonate films. SUVs and LUVs prepared in this manner typically exhibit a size distribution centered at ∼30 and ∼120 nm diameter, respectively (Snead et al., 2014). SUVs and LUVs were stored at 4◦C, while DPC micelles were stored at room temperature; all samples were used within 1 week. Concentration were estimated based on the amount of starting lipid used.

#### NMR Spectroscopy

For HSQC experiments, typical spectral widths (ppm) and number of complex data points were 14/1024 and 25/256 for proton and nitrogen, respectively, with some variation from experiment to experiment. Perdeuterated CPX-1 CTD and perdeuterated DPC were used for triple resonance experiments used to assign the CTD in the presence of DPC (15N, <sup>13</sup>C, <sup>2</sup>H labeled protein) and for the HSQC-NOESY-HSQC experiment for the CTD with DPC micelles (15N, <sup>2</sup>H labeled protein). Complete assignments of resonances observed in the presence of DPC micelles were obtained using the following experiments and parameters: (1) TROSY-based HNCACB – Bruker AVANCE

900 MHz spectrometer equipped with a cryoprobe (New York Structural Biology Center) with spectral widths (ppm) and number of complex data points at 14/1024, 25/64, and 69.9/128 in proton, nitrogen, and carbon dimensions, respectively. (2) TROSY-based HN(CO)CACB – Bruker AVANCE 900 MHz spectrometer equipped with a cryoprobe (New York Structural Biology Center) with spectral widths (ppm) and number of complex data points at 14/1024, 25/64, and 69.9/128 in proton, nitrogen, and carbon dimensions, respectively. (3) TROSY-based HNCACO – Bruker AVANCE 900 MHz spectrometer equipped with a cryoprobe (New York Structural Biology Center) with spectral widths (ppm) and number of complex data points at 14/1024, 25/64, and 15/72 in proton, nitrogen, and carbon dimensions, respectively. (4) TROSY based HNCO – Varian Unity Inova spectrometer equipped with a cryoprobe with spectral widths (ppm) and number of complex data points at 18/1536, 26/44, and 16/128 in proton, nitrogen, and carbon dimensions, respectively. Sequential NOEs for DPC-bound were obtained from an HSQC-NOESY-HSQC experiment with the following parameters – Bruker AVANCE 900 MHz spectrometer equipped with a cryoprobe (New York Structural Biology Center) with spectral widths (ppm) and number of complex data points at 14/1024, 25/72, and 25/72 in proton, nitrogen, and nitrogen dimensions, respectively.

To transfer wild-type CTD/DPC assignments to the 1116 variant, the following triple resonance experiments were acquired for <sup>15</sup>N, <sup>13</sup>C labeled 1116 CTD in the presence of perdeuterated DPC: (1) HNCA – Bruker AVANCE 900 MHz spectrometer equipped with a cryoprobe (New York Structural Biology Center) with spectral widths (ppm) and number of complex data points at 12/1024, 26/64, and 32/128 in proton, nitrogen, and carbon dimensions, respectively; (2) CBCACONH – Bruker AVANCE 900 MHz spectrometer equipped with a cryoprobe (New York Structural Biology Center) with spectral widths (ppm) and number of complex data points at 12/1024, 26/66, and 70/130 in proton, nitrogen, and carbon dimensions, respectively; (3) HNCO – Bruker AVANCE 700 MHz spectrometer equipped with a cryoprobe (New York Structural Biology Center) with spectral widths (ppm) and number of complex data points at 12/1024, 22/64, and 15/1028 in proton, nitrogen, and carbon dimensions, respectively; (4) HN(CA)CO – Bruker AVANCE 700 MHz spectrometer equipped with a cryoprobe (New York Structural Biology Center) with spectral widths (ppm) and number of complex data points at 12/1024, 22/64, and 15/1028 in proton, nitrogen, and carbon dimensions, respectively.

Resonance assignments for free 1116 CPX-1 could be reliably transferred for all resonances save those residues in the AH motif proximal to the modification. However, intensity ratio analyses could still be performed for the AH motif as a whole because all shifted but unassigned residues necessarily resided within this motif.

All NMR data were indirectly referenced to 4,4-dimethyl-4 silapentane-1-sulfonic acid and ammonia based on the position of the water resonance. Cα–Cβ secondary shifts were calculated as the difference between the observed carbon chemical shifts and random coil values tabulated from linear hexapeptides in 1 M urea at pH 5.0 and 25◦C (Wishart et al., 1995). Amide chemical shift perturbations were calculated as

$$
\Delta \delta\_{\text{avg}} = \sqrt{(1/2)(\Delta \delta\_{\text{HN}}^2 + (\Delta \delta\_{\text{N}}/5)^2)}
$$

where 1δ 2 HN and 1δ<sup>N</sup> are the amide proton and amide nitrogen chemical shift differences, respectively. Nitrogen shifts are scaled by a factor of five because of intrinsically larger chemical shift values of amide nitrogens.

#### Power Saturation ESR

The cw-ESR measurement spectra were collected on an ELEXSYS ESR spectrometer (Bruker Instruments, Billerica, MA, United States) at X-band (9.5 GHz) at RT. The power saturation experiments were performed in air, argon, and 20 mM Ni(II)-diammine-2,2<sup>0</sup> -(ethane-1,2-diyldiimino) diacetic acid (NiEDDA) with argon conditions. The latter two conditions were achieved by repeatedly degassing and saturating the sample with argon. In each condition, the spectra were recorded as a function of microwave power varied from 0.1 to 200 mW in 30 steps. The number of scans depended on the quality of the signal. The half-saturation parameter (P1/2) is obtained by fitting the equation

$$A = I^\* \sqrt{P^\*} [1 + (2^{1/e} - 1)^\* P / P\_{1/2}]^{-e},$$

where P is the microwave power applied, A is the peak-to-peak value of the central line of the spectra, ε is the line-homogeneity parameter that was obtained from the fitting (we usually obtained ε = 1.5 for the best fit). The accessibility parameters 5(O2) and 5(Ni) are calculated by the equation

$$\Pi(\mathcal{O}\_2) = \left[P\_{1/2}(\mathcal{O}\_2)/\Delta \, H(\mathcal{O}\_2) - P\_{1/2}(\text{Ar})/\Delta \, H(\text{Ar})\right]/$$

$$\left[P\_{1/2}(\text{ref})/\Delta \, H(\text{ref})\right], and \Pi(\text{Ni}) = \left[P\_{1/2}(\text{Ni})/\Delta \, H(\text{Ni})\right]$$

$$-P\_{1/2}(\text{Ar})/\Delta \, H(\text{Ar})]/\left[P\_{1/2}(\text{Ref})/\Delta \, H(\text{Ref})\right],$$

where 1H is the line width of the central line measured at 2 mW. We used data for Q119C in argon as a reference. The insertion depth parameter 8, which is independent of the reference, was calculated by the equation

$$\Phi = \ln[\Pi(\mathcal{O}\_2)/(\Pi(\text{NiEDPA}))](\text{refl}, 2, 3).$$

Experiments were repeated two or three times to ensure reproducibility (Altenbach et al., 1994; Frazier et al., 2002; Georgieva et al., 2014).

The plot of 8 vs. residue number was fit using the equation

$$\Phi = A^\* \sin(2\pi^\* R n / N + B) + C,$$

where Rn is the residue number, N is the periodicity, A is the scaling factor, B is the periodicity correction factor, and C is the offset. To compare the periodicity of the CTD with a standard alpha helix, we fit the plot in two ways: in the first fitting, N, A, B, and C were allowed to float; and in the second fitting N was fixed at the canonical alpha-helix periodicity of 3.6 residues per turn, and A, B, and C were allowed to float. The two approaches produced nearly identical fits. Fitting was carried out using the program Origin (Northampton, MA, United States).

#### CD Spectroscopy

fnmol-10-00154 May 22, 2017 Time: 16:51 # 5

Far-ultraviolet CD spectroscopy experiments were performed on an AVIV Biomedical Model 410 CD Spectrometer and were performed and analyzed as previously described (Snead et al., 2014). Briefly, spectra were obtained from ∼200 to 250 nm at 25◦C with a wavelength step of 1 nm, an averaging time of 1.7 s, 3–4 averaged scans per sample, a cell path length of 0.02 cm, and protein concentrations typically from 50 to 100 uM. CD data are normalized so that the number of helical residues in the absence of lipids matches the predicted free state helicity as estimated by NMR data, as previously described (Snead et al., 2014). Briefly, Cα–Cβ secondary chemical shifts have been tabulated for each amino acid in the presence of fully formed helical structure. We can thus integrate the positive Cα–Cβ secondary chemical shifts for free CPX-1 and normalize by the expected average secondary shift for a fully formed helix to estimate the overall fraction of helicity and the number of helical residues in the free state. By this analysis, the wild-type CTD construct should have roughly 12 residues helical in the absence of liposomes. Once the CD data are properly normalized in the absence of liposomes, the number of helical residues in the presence of lipids is determined using the equation

$$f\_{\mathbf{h}} = \frac{([\theta]\_{222} - 3000)}{-39000},$$

where f<sup>h</sup> is the fractional helicity, and [θ]<sup>222</sup> is the mean residue ellipticity at 222 nm.

#### Lipid Binding Data Analysis

Liposome binding of CPX-1 and CPX-1 mutants and/or truncations was assessed by solution-state NMR spectroscopy, as previously described (Snead et al., 2014). Briefly, HSQC spectra were obtained in the presence and absence of liposomes, and per-residue intensity ratios were calculated as cross-peak intensity with liposomes divided by cross-peak intensity in the free state. Intensity ratios directly reflect the unbound fraction for each residue (Bodner et al., 2009). To generate binding curves for the AH and CT motifs, intensity ratios for residues 111–136 (AH motif) and 137–143 (CT motif) were averaged, and fraction bound was calculated as 1 minus the intensity ratio. As noted above, full assignments were not obtained for free 1116 CPX-1, though all shifted and unassigned peaks for this variant were necessarily located within the AH motif and so could be incorporated into an average intensity ratio for this motif.

#### Strains

Caenorhabditis elegans were maintained on agar nematode growth media (NGM) at 20◦C and seeded with OP50 bacteria as previously described (Brenner, 1974). Strains employed in this study are N2, cpx-1 (ok1552), tauIs90 (Psnb-1::CPX-1::GFP); cpx-1, and tauEx454 [Psnb-1::CPX-1(1G116)::GFP]; cpx-1. Robust synaptic expression of all arrays was verified by measuring synaptic fluorescence to check expression levels against those that can fully rescue complexin mutants as previously described (Martin et al., 2011; Wragg et al., 2013).

#### Pharmacological Assays

To measure aldicarb sensitivity, 20–25 young adult animals were placed on agar plates containing 1 mM aldicarb (Chem Services) and scored for paralysis at 10 min intervals for 2 h. Each genotype was tested 8–10 times and paralysis curves were generated by averaging paralysis time courses for each plate as described previously (Dittman and Kaplan, 2008; Martin et al., 2011; Wragg et al., 2013).

### RESULTS

#### Use of DPC-Micelles to Mimic the Vesicle-Bound State of CPX-1 CTD

Detergent and phospholipid micelles provide a membrane-like environment yet are small enough for high-resolution structural analysis by NMR, and are therefore often utilized as membrane mimetics. Micelles have proven particularly useful for the structural characterization of peripheral membrane-interactions by proteins such as alpha-synuclein and tau (Eliezer et al., 2001; Bussell and Eliezer, 2003; Barré and Eliezer, 2006; Borbat et al., 2006; Georgieva et al., 2008, 2010), among others. To facilitate solution-state NMR studies of the membrane-binding AH and CT motifs of CPX-1, we produced a recombinant CTD fragment lacking the N-terminal domain, accessory domain, or central helix (**Figure 2A**) in order to reduce spectral complexity and overlap. Proton-nitrogen correlation spectra (HSQC) of the truncated CTD construct free in solution overlay well with the subset of signals in the corresponding spectrum of the full-length protein (**Figure 2B**). Specifically, each peak in the spectrum of the truncated CTD (red signals in **Figure 2B**) is situated at a position that can be unambiguously identified as belonging to an assigned resonance in the corresponding spectrum of the full-length protein (black signals in **Figure 2B**), allowing us to transfer the previously published backbone assignments from the full-length protein to the truncated CTD.

To assess whether the isolated CTD binds to lipids in a manner similar to that observed in the context of the full-length protein, we obtained HSQC spectra in the presence of lipid vesicles. Resonances corresponding to residues that engage in interactions with the vesicle surface are undetectable in such spectra, which therefore provide, for each residue, a measure of the fraction of protein for which that residue is bound to the vesicle surface. In other words, a decrease in the intensity ratio (±lipid) indicates an interaction with the vesicle. The resulting data shows that the AH and CT motifs in the isolated CTD bind to phospholipid vesicles in a manner similar to that previously observed in the context of full-length CPX-1 (**Figure 2C**). Specifically, the top panel shows strong binding to highly curved SUVs for residues 110–143, containing the AH and CT motifs, as previously reported (Snead et al., 2014), which is comparable for the full-length protein and the truncated CTD. Weaker binding is observed for residues 94–107, containing the AH2

motif, as previously observed, but in this region binding of the truncated CTD is decreased compared to the full-length protein. Importantly this effect is well-separated from the AH and CT motifs. The bottom panel shows that a greater concentration of less highly curved LUV vs. more highly curved SUV membranes is required to achieve comparable binding at the very C-terminus for both constructs, in keeping with the established curvature dependence of the CTD interaction. As previously noted (Snead et al., 2014), binding to LUVs decreases moving away from the CT motif to the AH and AH2 motifs. LUV binding by the truncated CTD is very similar to that observed for the full-length protein.

To identify a suitable micelle system for characterizing CTD binding, HSQC spectra were acquired in the presence of micelles composed of SDS, DPC, and the lysophospholipids LPPG and LOPC. HSQC spectra of the free CTD exhibit a poor dispersion of amide proton resonances, with many signals clustered together and in many cases overlapping (**Figure 2B**), consistent with a predominantly disordered polypeptide in which individual residues are highly dynamic and experience a similar

average local environment. Interestingly, the presence of SDS, LPPG, LOPC, or DPC micelles results in chemical shift changes and a somewhat increased dispersion, with the amide proton signals spread over a wider spectral region and better resolved (**Figures 3A,B** and **Supplementary Figure S1**), suggesting some ordering of the polypeptide chain upon micelle binding. Samples prepared using DPC gave the best spectral quality, and this lipid was therefore selected for subsequent structural studies.

To assess the resemblance of the micelle- and vesicle-bound states, we used CD spectroscopy to monitor secondary structure formation. Upon addition of increasing amounts of DPC, CD spectra of the CPX-1 CTD exhibited a clear dose dependent increase in CD spectral signatures indicative of helical secondary structure formation, including minima at 208 and 222 nm (**Figure 3C**). Importantly, the overall gain in helical structure, corresponding to approximately 13 residues, closely matched that observed previously upon SUV binding (Snead et al., 2014), suggesting that the micelle- bound state may accurately represent the more physiologically relevant SUV-bound state.

### Structural Features of DPC-Micelle Bound CPX-1 CTD

Based on liposome binding profiles such as those in **Figure 1C**, we postulated previously (Snead et al., 2014) the existence of

three structurally distinct motifs within the CPX-1 CTD: (1) the CT motif spanning residues 128–143 containing multiple bulky hydrophobic phenylalanine residues; (2) the AH motif spanning residues 110–124 that is unstructured in isolation and when bound to flat membranes, but that adopts helical structure upon binding to highly curved membranes; and (3) an AH2 (amphipathic helix 2) motif from residues 96–103 that forms a stable helix even in the free state, that may interact with membranes, and that, like the C-terminal and amphipathic motifs, is necessary for CPX-1 inhibitory function (perturbation of this motif impairs CPX-1 function in vivo).

Unlike the liposome-bound state, which cannot be visualized directly using solution-state NMR, the DPC micelle-bound state allowed us to directly establish the structure and boundaries of the CTD in a bound state. A standard set of triple resonance spectra acquired using perdeuterated CTD in the presence of perdeuterated DPC micelles enabled complete NMR backbone and Cβ resonance to be obtained (**Figure 3B**). Comparing the amide group chemical shifts to those of the free CTD in solution (**Figure 4A**) reveals large differences for signals originating from residues 110–143, likely indicating direct protein/micelle interactions in this region, and consistent with the previously proposed location of the AH and CT motifs. Considerably smaller differences are observed for signals corresponding to the N-terminal portion of this construct (residues ∼90–110), suggesting that this region may not interact with micelles. These smaller changes are likely caused by differences in the conditions under which the free and bound states are prepared.

Carbon chemical shifts are sensitive to local secondary structure, which leads to deviations (secondary shifts) from tabulated shifts observed in model random coil peptides (Wishart et al., 1995). The Cα–Cβ secondary shifts observed for the micelle-bound CTD (**Figure 4B**) feature two distinct regions, each exhibiting a nearly contiguous stretch of positive values, while the remaining residues mostly feature smaller values near zero with only isolated exceptions. Residues 96–104 exhibit positive secondary shifts indicative of marginally stable helical structure (average value of 1.8 ± 0.3) and subsume the location of the previously noted AH2 motif. CO secondary shifts (**Figure 4C**) are also positive for residues 96–103 with values consistent with partly populated helical structure. Notably, similar Cα–Cβ secondary shifts were observed in this region for the free protein (Snead et al., 2014). When combined with the lack of micelle-induced amide group chemical shift changes in this region, this suggests that the AH2 motif is marginally stable independently of the presence of DPC micelles, that it does not interact directly with the micelles and that its secondary structure is not influenced by micelle binding.

Larger positive Cα–C<sup>β</sup> secondary shifts (average 4.2 ± 0.3) are observed for residues 111–114 and 118–136, indicating the presence of stable helical structure in these regions. Although the free state of the protein exhibits positive Cα–Cβ secondary shifts for residues 109–126 (Snead et al., 2014), they are of significantly smaller amplitude (average of 0.9 ± 0.4), indicating that while this region transiently samples helical structure in the free state, stable helix formation depends on binding to the micelle. Positive CO secondary shifts are also observed for residues 110–114 and 118–136 consistent with helix formation in the micelle-bound state. Interestingly, residues 115–117, which connect two regions with large positive secondary shifts, exhibit small or negative Cα–Cβ and oscillating CO secondary shifts, suggesting that they may populate non-helical phi-psi peptide bond angles.

Residues 137–143 exhibit mostly small Cα–Cβ secondary shifts that oscillate around zero, but CO secondary shifts are somewhat negative in this region, perhaps indicating a weak preference for extended conformations. Although the significance of this observation is unclear, molecular dynamics simulations in an accompanying paper (Wragg et al., 2017) support a somewhat extended conformation for this region of the protein when bound to lipid membranes.

To further assess the conformation of these different CTD regions when bound to micelles, we measured NH-NH NOEs (Nuclear Overhauser Effects) between neighboring residues. These signals, which reflect inter-nuclear distances between the amide protons of successive residues, are strong in helical regions and weaker in regions of random coil or extended structure. NOEs were detected almost exclusively in regions exhibiting chemical shifts indicative of helical structure (**Figure 4D**), including a few scattered NOEs in the region corresponding to the AH2 motif, and strong NOEs from residues 110–136. Interestingly, residues 115–117 exhibit strong NOEs, indicating short NH-NH distances, despite having secondary shifts that are not typical of helical structure. As discussed below, this region of the protein likely adopts a pi-bulge conformation.

#### Secondary Structure of the AH Motif Bound to Lipid Vesicles

Although our CD data suggest that the micelle-bound CTD resembles the vesicle-bound protein, it remains possible that micelles engender a conformation of the protein that is not physiologically relevant. To validate the NMR analysis of the micelle-bound protein, we employed ESR spectroscopy combined with site-directed spin-labeling to probe the environment of each residue in a region encompassing that for which strong NOEs were observed (residues 110–136). Continuous wave ESR spectra were obtained for full-length CPX-1 labeled at each position with MTSL [S-(1-oxyl-2,2,5,5 tetramethyl-2,5-dihydro-1H-pyrrol-3-yl) methyl methanesulfonothioate] in the presence of SUVs and either O2, which selectively partitions into the membrane, or NiEDDA, which partitions into solution. For each residue, MTSL accessibilities to O<sup>2</sup> (membrane probe) and NiEDDA (solution probe) were determined and used to calculate the membrane insertion depth parameter Phi. Secondary structure results in a regular periodicity of Phi as a function of residue number (Hubbell et al., 1998; Georgieva et al., 2014). SUVs composed of 60:25:15 DOPC:DOPE:DOPS were employed in order to approximate the lipid composition and size (∼30 nm diameter) of synaptic vesicles, as previously described (Snead et al., 2014).

For residues 117–136, the ESR data (**Figure 5A**) show a clear periodicity that matches that 3.6 residue per turn periodicity expected for an ideal alpha helix reasonably well. However,

this periodicity is not maintained through residues 115–116, where interruptions in the secondary chemical shift data were observed, confirming that regular helical structure is interrupted in this region. Residues 110–114 exhibit periodicity that is again consistent with helical structure. Thus, the observed periodicity deviates from that expected for a canonical alpha helix in a short region centered on residue 116, while the two sequences to either side of this residue appear to adopt regular alpha-helical structure in the bound state. Importantly, the fact that both the ESR (**Figure 5A**) and NMR (**Figure 4**) data indicate an interruption in regular helical structure around position 116 alleviates any concern that the introduced spin label may somehow cause the anomalous behavior.

An explanation for the interruption of regular helical structure around residue 116 can be obtained by plotting the sequence of CPX-1 CTD starting at residue 110 as an helical wheel projection (**Figure 5B**). The resulting helix has hydrophobic residues positioned all around the helix axis. Such a helix does not possess the amphipathic character that is typically seen in helices that bind to membrane surfaces. Closer examination of the helical wheel reveals that residues 110–116 maintain an amphipathic character by clustering hydrophobic residues Leu111, Ile112, and Leu115 opposite from polar/small residues Ser110, Gly113, Gln114, and Gly 116. The next residue in the sequence, however, Leu 117, falls in the middle of the apolar cluster of residues, disrupting the amphipathic moment of the helix.

was ∼55 mM and Protein concentrations varied from ∼100 uM for HSQC experiments to ∼500 uM for NOESY experiments. Lipid-to-protein ratios were all above

the ∼100:1 saturation point indicated in the CD data. Gly116, deletion of which is discussed later in the text, is shown in bold and marked with a 1.

wild-type CPX-1 singly labeled with MTSL at residues 110–136. Shown is the insertion depth parameter 8, related to the ratio of accessibility to O2 and to NiEDDA as 8 = ln(5O2/5NiEDDA). Increasing 8 values indicate deeper immersion into the membrane. The blue line indicates the best fit to a cosine function with a fixed periodicity of 3.6 amino acids per turn (i.e., that expected for an ideal alpha helix). (B) Helical wheel diagram for the proposed amphipathic region of wild-type CPX-1. Residues are color coded according to their hydrophobicity as hydrophobic (black), neutral (white), polar (yellow), negatively charged (red), and positively charged (blue). The wild-type sequence does not appear amphipathic when modeled as a canonical alpha-helix. (C) Same data as shown in (A) but with residue 116 omitted. The blue line again indicates the best fit to a cosine function with a fixed periodicity of 3.6 amino acids per turn. (D) Helical wheel diagram for the proposed amphipathic region of the CTD with glycine 116 omitted. Removal of this residue results in a clear amphipathic character. Helical wheel plots were generated using the HELNET program suite (Jones et al., 1992). Gly116 is shown in bold and marked with a 1, which also marks its location when omitted.

The same phenomenon is observed by examining a helical wheel projection of the CTD sequence starting at position 116. This results in an amphipathic helix with well-defined polar and apolar faces (**Supplementary Figure S2A**), as shown previously (Wragg et al., 2013) and consistent with the NMR chemical shift data. Attempting to extend this helix in the N-terminal direction, however, results in Leu115 falling in the middle of the polar face of the helix (**Supplementary Figure S2B**), suggesting that a structural rearrangement is required to generate a continuous amphipathic character.

#### Deletion of Gly116 Results in a Single Continuous Amphipathic Helix without Altering Membrane-Binding or Curvature Sensing by the AH Motif

Based on the observation that Gly116 effectively leads to a discontinuity in the amphipathic moment of the helical structure of CPX-1 CTD, we hypothesized that removal of this residue might result in a single uninterrupted amphipathic helix extending from residue 110–136. This hypothesis was

FIGURE 6 | Deletion of glycine 116 alters the helical structure of the amphipathic motif, though overall helicity is maintained. (A) Amide proton/nitrogen chemical shift differences for wild-type versus 1116 CTD. Amide shift perturbations are localized to the area of the deletion. Overlay of Cα–C<sup>β</sup> secondary shifts (B) and CO secondary shifts (C) for wild-type (black) versus 1116 (red) CTD in the presence of DPC micelles. 1116 complexin shows moderately positive Cα–C<sup>β</sup> secondary shifts at residues 115 and 117, while near zero values are observed for wild-type complexin for residues 115–117, suggesting that deletion of residue 116 allows helical structure to propagate through this region. CO secondary shifts also exhibit a clear change toward values consistent with helical structure for residues 115 and 117. Note that for ease of comparison with the 1116 variant, data for residue 116 of wild-type complexin is not shown. (D) Helix formation by wild-type (black) versus 1116 (red) CTD as a function of increasing DPC concentration monitored by CD spectroscopy. The number of helical residues was estimated based on [θ]<sup>222</sup> as described in methods. Inset: CD spectra for the isolated CTD of 1116 complexin in the presence of increasing amounts of DPC (spectra shown from black to increasingly lighter shades of gray correspond, respectively, to: 0, 0.75, 1.25, 1.875, 2.5, 5, 10, and 67.5 mM). Gly116 is shown in bold and marked with a 1, which also marks its location when deleted.

supported by the observation that excision of Gly116 from the ESR data leads to a continuous helical periodicity throughout this region (**Figure 5C**) and that a helical wheel projection of the sequence lacking Gly116 is amphipathic (**Figure 5D**). To directly test the hypothesis, we generated a mutant CTD with residue Gly116 deleted, termed 1116, and obtained backbone and Cβ NMR resonance assignments in the presence of DPC. Substantial amide chemical shift changes for DPC-bound 1116 versus wild-type CTD were observed for ∼residues 111–121, centered on the deletion site, indicating that the mutation only affected local structure, with no impact further toward the N- or C-termini (**Figure 6A**). Carbon secondary shifts (**Figures 6B,C**) indicate that DPC-bound 1116 does not contain the helical break observed for the wild-type CTD, with Cα–Cβ carbon secondary shifts of ∼2 for 1116 residues 115 and 117 (compared with values of −0.5 to 1 for wt residues 115–117). Carbonyl secondary shifts, too, are positive for residues 115 and 117 in 1116, in contrast with the wild-type where they are negative. Thus NMR secondary chemical shifts suggest that whereas micelle-bound wild-type CTD contains a helical break at residues 115–117, 1116 likely adopts a continuous helical structure extending from residue 110 to 136. CD spectra confirm that 1116 forms a similar degree of helical structure in the micelle-bound state, as wild-type CTD (**Figure 6D**).

To determine whether deletion of glycine 116 alters phospholipid vesicle binding or membrane curvature sensing by CPX-1, we titrated wild-type and 1116 CTD with SUVs or LUVs composed of either 85/15 POPC/POPS or 60/25/15

DOPC/DOPE/DOPS and monitored binding using NMR HSQC spectra. Binding curves for the AH and CT motifs were then generated by averaging the data over all residues within each motif (**Figure 7**). For the wild-type protein, as previously reported (Snead et al., 2014), binding of the both the CT and AH motifs is enhanced for SUVs vs. LUVs of similar composition (black vs. red circles, all panels), reflecting the membrane curvature sensitivity of both motifs. In addition, binding of the AH motif is enhanced for liposome composed of lipids with increased acyl chain unsaturation (DO liposomes, upper panels, vs. PO liposomes, lower panels), likely reflecting the role of membrane packing defects in modulating AH motif binding. Interestingly, deletion of Gly116 does not significantly perturb membrane-binding or curvature sensing by the CTD, as binding curves for the AH and CT motifs of 1116 are similar to those of the wild-type (circles vs. squares, all panels). Both variants display a similar curvature sensitivity in membrane-binding, with preferential binding to small versus large vesicles (black vs. red symbols, all panels). For both wild-type and 1116,

1116 (squares) CTD. Error bars represent standard error of the mean.

too, increasing acyl chain desaturation (DO versus PO lipids) increases binding to lower curvature membranes (upper panels vs. lower panels). Thus, deletion of residue 116 does not significantly alter binding properties of the CPX-1 CTD.

To characterize whether deletion of glycine 116 alters helical folding of the AH motif upon liposome binding, we monitored titrations using CD spectroscopy to assess secondary structure (**Figure 8A**). Deletion of glycine 116 does not appear to significantly alter or perturb helical folding by the AH upon binding to liposomes with any of the compositions and sizes tested. The increase upon addition of liposomes in the number of helical residues estimated based on the CD signal at 222 nm is comparable for both wild-type and 1116 CTD. Together with the above results, these observations suggest that the 1116 CTD binds to vesicles with a slightly altered bound state structure, but with binding properties and overall helicity otherwise comparable to wild-type CPX-1. Where wild-type CTD contains a helical break in the bound state, 1116 CTD instead adopts a single continuous helix.

#### Irregular Structure at Gly116 Is Necessary for Inhibition of Spontaneous Exocytosis in C. elegans

To determine whether altered helical periodicity at residue 116 is important for CPX-1 function, we examined the inhibitory function of wild-type versus 1116 CPX-1 using a well-established functional assay in C. elegans (Miller et al., 1996; Mahoney et al., 2006; Martin et al., 2011). Briefly, upon treatment with the acetylcholinesterase inhibitor aldicarb, wild-type worms paralyze with a characteristic time course. Mutations that increase acetylcholine release at the neuromuscular junction cause worms to paralyze more rapidly. Knockout of cpx-1 abolishes its inhibition of synaptic vesicle exocytosis, and thus cpx-1 knockout worms are hypersensitive to aldicarb treatment. This phenotype can then be fully rescued by re-expression of wild-type CPX-1. Interestingly, expression of 1116 CPX-1 in the cpx-1 knockout background only partially rescues the aldicarb phenotype (**Figure 8B**), suggesting that the inhibitory function of 1116 CPX-1 is impaired in vivo. Given that 1116 CPX-1 lacks the helix interruption found in the wild-type, this implies that altered helical periodicity at residue 116 somehow contributes to inhibition of synaptic vesicle exocytosis by CPX-1. Interestingly, the 1116 mutant is unique in that membrane-binding and overall helicity are not perturbed, though protein function is still impaired. Thus, some property of the CPX-1 CTD beyond membrane-binding and helix formation is likely required for proper inhibitory function.

#### DISCUSSION

Membrane-binding by worm CPX-1 via its CTD is required to localize complexin to synaptic vesicles at presynaptic nerve terminals in living worms (Wragg et al., 2013). Membrane-binding requires residues at the very C-terminus of the CTD (CT motif). Furthermore, helix formation by an adjacent amphipathic region (AH motif) upon binding to highly curved membranes is required for efficient inhibition of neurotransmitter release by CPX-1 (Wragg et al., 2013; Snead et al., 2014). A third amphipathic motif (AH2 motif) upstream of the AH motif is also partly helical and important for proper CPX-1 function. The three CTD motifs were defined previously based on NMR measurements of residue-resolved binding of the CTD to lipid vesicles. Here, solution-state NMR and ESR spectroscopy applied to the CPX-1 CTD in the presence of DPC micelles allow us to directly characterize the structural properties of the CTD in a membrane or membrane-like environment while also refining the previously proposed structure and boundaries of these motifs. We find that the AH2 motif, which is marginally stable even in the absence of lipids, is unaffected by the presence of DPC. The CT motif, which is absolutely required for CTD membrane-binding, is found to consist of only the final seven residues, 137–143, of the CPX-1 sequence, and does not adopt regular secondary structure upon binding to micelles. The AH motif is now observed to extend from residue 110 to residue 136, and adopts an amphipathic helix that is interrupted in the region surrounding residue Gly116. Removal of Gly116 results in a continuous helical structure throughout the AH motif, without perturbing membrane-binding or curvature-sensitive helix formation. Despite this, deletion of Gly116 significantly reduces the ability of CPX-1 to inhibit neurotransmitter release in living worms.

#### AH2 Motif

The AH2 motif was defined to include residues 96–104 based on the observation that this region possesses amphipathic character, binds to SUVs, albeit more weakly than the CT or AH motifs, and possesses significant helical character even in the free state of CPX-1 (Snead et al., 2014). Surprisingly, we observe no indications of interactions between the AH2 motif and DPC micelles, as neither amide nor carbon chemical shifts are altered

in the presence of DPC. This likely results from the choice of our CTD construct, which begins at CPX-1 residue 90, only a few residues away from the beginning of the AH2 motif. We expected that this construct would properly capture the behavior of this motif, but unfortunately it appears that the close proximity of the N-terminus of this fragment may interfere with proper lipid interactions in this region, as observed in our SUV binding data (**Figure 2C**). Obtaining a clearer picture of the structure of the AH2 motif in its lipid-bound state will require a model system that faithfully recapitulates its observed interaction with vesicles.

### CT Motif

We documented the presence of a membrane-binding C-terminal motif in CPX-1 based on the observation that residues at the very C-terminus of CPX-1 bound most tightly to lipid vesicles in vitro, combined with the observation that deletion of the final 12 residues of CPX-1 completely abolished membrane-binding by the CTD and significantly impaired inhibitory function in vivo (Wragg et al., 2013; Snead et al., 2014). Data presented here suggests, however, that the CT motif likely consists of only the final seven residues of CPX-1, as residues upstream of this region form part of a continuous helical structure that belongs to the adjacent AH motif. Based on small Cα–Cβ secondary chemical shifts that oscillate around zero and on the paucity of sequential amide proton NOEs, we conclude that the CT motif likely binds to micelles in a disordered conformation devoid of secondary structure. This is consistent with previous observations using CD spectroscopy showing no spectral changes upon binding of CPX-1 to LUV liposomes under conditions where the CT motif binds but the AH motif does not become helical (Snead et al., 2014). Membrane-binding by the CT motif is likely mediated by insertion of its three bulky, hydrophobic phenylalanine residues. Based on these new structural data we show in an accompanying paper (Wragg et al., 2017) that mutants in which the final six residues of CPX-1 are deleted, or in which all three CT motif phenylalanine residues are replaced by alanine, bind very weakly to liposomes in vitro and are unable to inhibit synaptic vesicle exocytosis in worms. This provides strong support for this new structure-based demarcation of the CT motif. Notably, we previously established that mCpx1 also binds preferentially to highly curved membranes, possibly through tandem lipid binding motifs (Snead et al., 2014) and showed that the putative mCpx1 CT motif consists of seven residues, indicating that the length of the CT motif is likely similar between worms and mice. Despite this similarity, we show in an accompanying paper (Wragg et al., 2017) that the mouse CT motif cannot substitute for the worm motif in vivo, suggesting some degree of divergence in the precise function(s) of this motif.

## AH Motif

The AH motif was initially identified by inspection of the CPX-1 sequence (Wragg et al., 2013), though a similar sequence was noted in mammalian complexins by others (Takahashi et al., 1995; Seiler et al., 2009). We confirmed membrane-association of the AH motif using NMR binding assays, and demonstrated that the AH motif becomes helical upon binding to highly curved liposomes using a combination of CD spectroscopy and mutational analysis (Seiler et al., 2009; Wragg et al., 2013; Snead et al., 2014). Here, we directly observe the helical structure of the AH motif when bound to micelles. The motif is found to extend from residues 110 to 136, encompassing a longer stretch of the CPX-1 CTD than was initially proposed based on either sequence analysis (Wragg et al., 2013) or NMR binding assays (Snead et al., 2014). Our secondary chemical shift data indicate, however, that

the helical structure of the AH motif deviates from that of a typical alpha helix in the vicinity of residue 116. This deviation can be explained by the fact that the amphipathic moment of the N-terminal part of the AH motif, comprised of residues 110–116, is mismatched with that of the remaining sequence of the motif (residues 117–136), so that formation of a continuous helix throughout the sequence would result in the loss of the overall amphipathic moment. Examination of the sequence reveals that this mismatch effectively results from a single-residue insertion, at position 116, into what would otherwise be a continuous amphipathic helix. Indeed, we show that deletion of Gly116 removes the discontinuity in the AH motif in results in a single continuous helix.

### Pi-bulge in the CPX-1 CTD

The insertion of an extra residue into a helix-forming sequence often results in a rearrangement of the helical turn including that residue into a non-canonical helical structure (Heinz et al., 1993; Keefe et al., 1993). This type of irregularity often takes the form of a pi-helix or a pi-bulge, though such motifs have been referred to also as alpha-bulges or alpha aneurysms, among other names (Keefe et al., 1993; Riek and Graham, 2011). Pi-bulge formation at residue Gly116 is suggested by small, negative and/or oscillating values of Cα–Cβ and carbonyl secondary shifts for residues 115–117, combined with the presence of strong sequential amide proton NOEs through this region. Together these observations indicate a non-canonical helical structure that is nevertheless well-organized and compact, not flexible and disordered. Although our current NMR data are not sufficient to unambiguously establish the presence of a Pi-bulge or to calculate a robust atomic resolution structure, ESR data further corroborated the presence of an irregular helical structure at residue 116; sequences to either side of this residue appear to adopt a canonical alpha helical periodicity that is not propagated through position 116. Interestingly, analysis of the fly complexin sequence, which also contains an AH motif (Wragg et al., 2013; Cho et al., 2015), suggests that the presence of a pi-bulge in the AH motif may be conserved between worms and flies. Similar to worm complexin, residues 103–131 of fly complexin isoform 7B, which, like CPX-1, lacks a C-terminal CAAX box prenylation motif, when modeled using a helical wheel, do not exhibit a clear amphipathic nature (**Figure 9A**). The C-terminal part of this region, starting at residue Glu 112, does form a clearly amphipathic helix (**Supplementary Figure S3A**), but when propagated in the N-terminal direction, residue Gln111 and Glu 110 fall on the hydrophobic face of the helix (**Supplementary Figure S3B**). As in the worm protein, removal of a single residue, Glu112, from the sequence results in a single continuous amphipathic alpha helix (**Figure 9B**). Ongoing efforts to determine structures of both worm and fly complexins in their micelle-bound states should establish the precise nature of their helical structures.

Pi-bulges are often enriched at functional sites within proteins (Cooley et al., 2010) and associated with specific protein functions, for example by acting as a scaffold to align key residues for, e.g., metal ion coordination (Riek and Graham, 2011) or ligand binding (Cartailler and Luecke, 2004). Of greater potential relevance for the case of complexin is the observation that pi-bulges can cause helix bending and helical kinks (Cartailler and Luecke, 2004), structural features which could impact membrane-binding and membrane curvature sensing. To assess whether the presence of a pi-bulge at Gly116 in CPX-1 might contribute to membrane-binding, curvature sensing and inhibitory function, we generated and characterized 1116 CPX-1. NMR secondary chemical shifts show that deletion of Gly116 removes the pi-bulge, as 1116 formed a continuous amphipathic helix in the micelle-bound state. Surprisingly, however, membrane-binding, membrane curvature sensitivity, and overall helicity are unperturbed for 1116. Despite the lack of change in membrane-binding, curvature sensing and helix formation, 1116 was less able to inhibit neurotransmitter release in C. elegans, arguing that the pi-bulge of wild-type complexin contributes to its inhibitory function.

If membrane-binding, curvature sensing and helicity are not impaired, why does 1116 CPX-1 display decreased inhibitory activity in vivo? Though this remains unclear at present, one possibility is that pi-bulge formation contributes to some functionally requisite protein/protein interaction, and removal of Gly116 results in an altered conformation that is unable to properly form this interaction. In an accompanying paper (Wragg et al., 2017) we show that the AH motif of mCpx1 is unable to substitute for the worm AH in CPX-1 despite the fact that the mouse AH motif also binds membranes in a curvature-sensitive fashion (Snead et al., 2014; Gong et al., 2016). We speculate that while membrane-binding and curvature sensing are a requisite conserved features of both AH motifs, additional interactions with unknown binding partners are also required to effect proper complexin function, and that the interactions with these partners, as well as the identity of these partners, may have diverged between worm and mouse, as well as among other phyla. It will thus be critically important to identify any such putative yet currently unknown AH motif protein binding partners.

### AUTHOR CONTRIBUTIONS

Conceptualization: DS, JD, JF, and DE; methodology: DS, AL, RW, DP, TR, JD, JF, and DE; investigation: DS, AL, RW, DP, TR, and DE; writing: DS, AL, JD, JF, and DE; funding acquisition: DS, DP, JD, JF, and DE.

### FUNDING

Funding provided by NIH grants R37AG019391 and R01GM117518 (DE), F30MH101982 and MSTPGM07739 (DS), R01GM095974 (JD), and R01EB003150 and P41GM103521 (JF) and DFG Fellowship PA2679/2-1 (DP). DE is a member of the New York Structural Biology Center. Data collected at NYSBC were made possible by a NYSTAR and ORIP/NIH facility improvement grant CO6RR015495. The 900 MHz spectrometers were funded by NIH grant P41GM066354, the Keck Foundation, New York State Assembly, and the US Department of Defense.

#### ACKNOWLEDGMENT

fnmol-10-00154 May 22, 2017 Time: 16:51 # 16

We thank Clay Bracken (WCMC) and Shibani Bhattacharya (NYSBDG) for assistance with NMR spectrometers and specifically for help setting up and running NMR experiments.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2017.00154/full#supplementary-material

FIGURE S1 | Dodecylphosphocholine (DPC) micelles lead to increased spectral dispersion in 2D spectra of CPX-1 CTD. Overlay of HSQC spectra for the CPX-1 CTD in the free state (black) and in the presence of DPC micelles (cyan). Cyan and black horizontal and vertical arrows highlight the increased proton and nitrogen spectral widths spanned by the DPC micelle versus free state spectrum. Inset: expansion of the glycine region similarly shows increased dispersion of these residues.

#### REFERENCES


FIGURE S2 | Propagating the CPX-1 AH motif helix in the N-terminal direction past residue Gly116 results in a loss of amphipathic character. (A) Helical wheel projection of CPX-1 residues 116–132 shows clearly defined polar and apolar faces. (B) Helical wheel projection of CPX-1 residues 115–132 shows that residue Leu115 falls in the middle of the polar face of the otherwise amphipathic helix, disrupting its amphipathic character. Residues are color coded according to their hydrophobicity as hydrophobic (black), neutral (white), polar (yellow), negatively charged (red) and positively charged (blue). Helical wheel plots were generated using the HELNET program suite (Jones et al., 1992).

#### FIGURE S3 | Propagating the fly complexin isoform 7B AH motif helix in the N-terminal direction past residue Glu112 results in a loss of

amphipathic character. (A) Helical wheel projection of fly 7B complexin residues 112–131 shows clearly defined polar and apolar faces. (B) Helical wheel projection of fly 7B complexin residues 110–131 shows that residues Gln111 and Glu110 fall on the apolar face of the otherwise amphipathic helix, disrupting its amphipathic character. Residues are color coded according to their hydrophobicity as hydrophobic (black), neutral (white), polar (yellow), negatively charged (red), and positively charged (blue). Helical wheel plots were generated using the HELNET program suite (Jones et al., 1992).



**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Snead, Lai, Wragg, Parisotto, Ramlall, Dittman, Freed and Eliezer. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Evolutionary Divergence of the C-terminal Domain of Complexin Accounts for Functional Disparities between Vertebrate and Invertebrate Complexins

Rachel T. Wragg<sup>1</sup> , Daniel A. Parisotto<sup>1</sup> , Zhenlong Li<sup>2</sup> , Mayu S. Terakawa<sup>1</sup> , David Snead<sup>1</sup> , Ishani Basu<sup>1</sup> , Harel Weinstein2,3, David Eliezer<sup>1</sup> and Jeremy S. Dittman<sup>1</sup> \*

<sup>1</sup> Department of Biochemistry, Weill Cornell Medical College, New York, NY, United States, <sup>2</sup> Department of Physiology and Biophysics, Weill Cornell Medical College, New York, NY, United States, <sup>3</sup> Institute for Computational Biomedicine, Weill Cornell Medical College, New York, NY, United States

Edited by:

Jiajie Diao, University of Cincinnati, United States

#### Reviewed by:

Changhe Wang, Xi'an Jiaotong University, China Xiaochu Lou, University of Wisconsin–Madison, United States Jeremy Leitz, Stanford University, United States

> \*Correspondence: Jeremy S. Dittman jed2019@med.cornell.edu

Received: 03 March 2017 Accepted: 30 April 2017 Published: 26 May 2017

#### Citation:

Wragg RT, Parisotto DA, Li Z, Terakawa MS, Snead D, Basu I, Weinstein H, Eliezer D and Dittman JS (2017) Evolutionary Divergence of the C-terminal Domain of Complexin Accounts for Functional Disparities between Vertebrate and Invertebrate Complexins. Front. Mol. Neurosci. 10:146. doi: 10.3389/fnmol.2017.00146 Complexin is a critical presynaptic protein that regulates both spontaneous and calciumtriggered neurotransmitter release in all synapses. Although the SNARE-binding central helix of complexin is highly conserved and required for all known complexin functions, the remainder of the protein has profoundly diverged across the animal kingdom. Striking disparities in complexin inhibitory activity are observed between vertebrate and invertebrate complexins but little is known about the source of these differences or their relevance to the underlying mechanism of complexin regulation. We found that mouse complexin 1 (mCpx1) failed to inhibit neurotransmitter secretion in Caenorhabditis elegans neuromuscular junctions lacking the worm complexin 1 (CPX-1). This lack of inhibition stemmed from differences in the C-terminal domain (CTD) of mCpx1. Previous studies revealed that the CTD selectively binds to highly curved membranes and directs complexin to synaptic vesicles. Although mouse and worm complexin have similar lipid binding affinity, their last few amino acids differ in both hydrophobicity and in lipid binding conformation, and these differences strongly impacted CPX-1 inhibitory function. Moreover, function was not maintained if a critical amphipathic helix in the worm CPX-1 CTD was replaced with the corresponding mCpx1 amphipathic helix. Invertebrate complexins generally shared more C-terminal similarity with vertebrate complexin 3 and 4 isoforms, and the amphipathic region of mouse complexin 3 significantly restored inhibitory function to worm CPX-1. We hypothesize that the CTD of complexin is essential in conferring an inhibitory function to complexin, and that this inhibitory activity has been attenuated in the vertebrate complexin 1 and 2 isoforms. Thus, evolutionary changes in the complexin CTD differentially shape its synaptic role across phylogeny.

Keywords: complexin, membrane binding, synaptic transmission, synaptic vesicles, molecular dynamics, evolutionary conservation and diversification, SNAREs, C. elegans

### INTRODUCTION

fnmol-10-00146 May 23, 2017 Time: 16:8 # 2

Precise synaptic transmission is critical for proper nervous system function, and over the past 25 years, most of the proteins required for this process have been identified and characterized. A mechanistic picture has emerged based on the assembly of SNARE proteins residing on the synaptic vesicle (SV) and plasma membrane (Sollner and Rothman, 1994). This assembly is tightly orchestrated by a set of conserved proteins including Munc13, Munc18, and synaptotagmin (Sudhof, 2013; Rizo and Xu, 2015). Although the complete molecular picture of the events controlling SV fusion is far from fully developed, there is general agreement on the impact of perturbing the SNAREs, Munc13, Munc18, and synaptotagmin in many highly divergent experimental preparations such as the squid giant synapse, worm and fly neuromuscular junctions, as well as rodent cultured neurons and acute rodent brain slices (Augustine et al., 1996; Weimer and Jorgensen, 2003; Sudhof and Rothman, 2009; Kochubey et al., 2011; Sudhof and Rizo, 2011). Consensus of both sequence and function across such a broad range of synapses implies a deep mechanistic conservation of calciumregulated secretion in all animals, consistent with the assertion that there is a single overarching molecular pathway for SV fusion in neurons shared across phylogeny. However, another key SNARE-binding protein has proven more difficult to fit into this picture. Complexin is a small (130–150 residue) cytoplasmic protein that binds directly to the assembled SNARE complex via a highly conserved alpha helical domain termed the central helix (CH) (McMahon et al., 1995; Melia, 2007; Brose, 2008; Trimbuch and Rosenmund, 2016). Human complexin mutations (CPLX1 gene) are associated with severe epilepsy, cortical atrophy, and intellectual disability (Karaca et al., 2015; Redler et al., 2017). Loss-of-function studies in different model synapses revealed similarities as well as prominent differences in complexin function. For instance, while almost all studies agree that loss of complexin leads to a decrease in calciumtriggered exocytosis, the regulation of spontaneous fusion by complexin appears to have diverged between vertebrates and invertebrates (Trimbuch and Rosenmund, 2016). SV fusion in the absence of calcium influx (spontaneous fusion) is either decreased or slightly increased in several mammalian synapses lacking complexin depending on the preparation and the details of complexin removal (Xue et al., 2007; Maximov et al., 2009; Strenzke et al., 2009; Lin et al., 2013; Yang et al., 2013). In contrast, spontaneous SV fusion is highly elevated (between 10- and 20-fold) in worm and fly synapses lacking complexin (Huntwork and Littleton, 2007; Hobson et al., 2011; Martin et al., 2011). Interestingly, vertebrate complexin 3/4 isoforms have been proposed to inhibit spontaneous release in retinal bipolar cell synapses (Vaithianathan et al., 2013, 2015), suggesting a functional divergence between complexin isoforms within the vertebrate subphylum. Another recent study found that the rate of SV fusion in the calyx of Held is transiently elevated by a factor of more than 10-fold in the absence of mCpx1, but only for a brief time lasting a few 100 ms after SVs initially dock and prime following a previous SV fusion event (Chang et al., 2015). These observations hint at a transient role for mammalian complexin 1 in preventing premature fusion during the process of docking and priming, whereas invertebrate complexins are constitutively required to inhibit spontaneous fusion.

Relative to the other core SV fusion machinery, complexin is a poorly conserved protein. The 25 residues defining the CH constitute the only extensive region of complexin exhibiting strong conservation between phyla. This CH domain mediates a direct SNARE interaction and is required for all known complexin function in both vertebrate and invertebrate synapses (Giraudo et al., 2006; Xue et al., 2007; Maximov et al., 2009; Cho et al., 2010; Martin et al., 2011; Yang et al., 2015). Three additional regions of complexin have been defined both structurally and functionally: the N-terminal domain (NTD), the accessory helix domain (AH), and the C-terminal domain (CTD) comprising the latter half of complexin (Trimbuch and Rosenmund, 2016). The NTD serves a positive function in regulating fusion whereas the AH and CTD contribute to an inhibitory activity of complexin (Xue et al., 2007, 2010; Kummel et al., 2011; Martin et al., 2011; Kaeser-Woo et al., 2012; Buhl et al., 2013; Iyer et al., 2013; Wragg et al., 2013; Cho et al., 2014; Lai et al., 2014; Radoff et al., 2014). How does a protein domain with little or no primary sequence homology share a conserved function? The complexin CTD lacks meaningful sequence identity between phyla but a common motif predicted in all known complexin CTD sequences is an amphipathic helix region near the end of the protein (Seiler et al., 2009; Wragg et al., 2013; Snead et al., 2014; Gong et al., 2016). Several recent studies have proposed that the amphipathic region of the CTD mediates a curvature-sensitive membrane binding interaction that directs both mammalian and nematode complexin to SVs (Wragg et al., 2013; Snead et al., 2014; Gong et al., 2016). Without the CTD, the inhibitory function of complexin is impaired (Xue et al., 2009; Kaeser-Woo et al., 2012; Wragg et al., 2013), as is complexin localization at the synapse (Buhl et al., 2013; Iyer et al., 2013; Wragg et al., 2015). In addition to the amphipathic region, the CTD of all complexins terminates with either a second hydrophobic lipid-binding motif or a lipidated CAAX box motif, further emphasizing a potential membrane-interacting role for this region of complexin (Reim et al., 2005; Cho et al., 2010; Buhl et al., 2013; Iyer et al., 2013). Despite these conserved membrane-interacting features, several studies have described a range of imperfect functional rescue between species when exchanging mouse and fly complexins (Xue et al., 2009; Cho et al., 2010). Is the CTD functionally conserved despite the wide variety of primary sequences across phyla? Do the differences in CTD sequences account for the functional differences between vertebrate and invertebrate complexins? We systematically investigated nematode and mammalian complexin 1 orthologs using a combination of in vitro, in vivo, and computational approaches and found that differences in the CTD account for divergence of complexin inhibitory function. Moreover, these differences are not simply due to large variations in membrane binding. We propose that other divergent protein interactions within the CTD account for functional differences in complexin across phylogeny.

#### MATERIALS AND METHODS

fnmol-10-00146 May 23, 2017 Time: 16:8 # 3

#### Animals

Caenorhabditis elegans were maintained on agar nematode growth media (NGM) at 20◦C and seeded with OP50 bacteria as previously described (Brenner, 1974). Strains employed in this study are listed in Supplementary Table S3. Robust synaptic expression of all arrays was verified by measuring synaptic fluorescence to check expression levels against those that can fully rescue complexin mutants as previously described (Martin et al., 2011; Wragg et al., 2013).

### Pharmacological Assays

To measure aldicarb sensitivity, 20–25 young adult animals were placed on agar plates containing 1 mM aldicarb (Chem Services) and scored for paralysis at 10 min intervals for 2 h. Each genotype was tested 8–10 times and paralysis curves were generated by averaging paralysis time courses for each plate as described previously (Dittman and Kaplan, 2008; Martin et al., 2011; Wragg et al., 2013). Percent rescue based on t0.<sup>5</sup> was calculated by first interpolating the time at which 50% of the worms paralyzed for each trial, averaging the single-trial t0.<sup>5</sup> values together, and normalizing to wild type (100%) t0.<sup>5</sup> and cpx-1 (0%) t0.<sup>5</sup> values according to **Equation 1**.

$$\text{\% Rescue} \left[ \text{Strain} \right] = 100 \cdot \left( t\_{0.5} \left[ \text{Strain} \right] - t\_{0.5} \left[ \text{cpx} \right] \right) / t\_{0.5} \left[ \text{WT} \right] \tag{1}$$

#### Steady-State Fluorescence Imaging and Quantification

To measure protein expression levels, animals were immobilized using 2,3-butanedione monoxime (Alfa Aesar) (30 mg/mL) mounted on 2% agarose pads. An inverted Olympus microscope (IX81), using a laser scanning confocal imaging system (Olympus Fluoview FV1000 with dual confocal scan heads) and an Olympus PlanApo 60X 1.42 NA objective was used. Rescuing complexin constructs were C-terminally tagged with GFP separated by a 12 residue linker (GGSGGSGGSAAA). Synaptic protein levels were estimated by measuring background-subtracted fluorescence within dorsal cord varicosities. A fluorescent slide was imaged to monitor laser stability over time and the dorsal cord axonal fluorescence was normalized to the slide value for all measurements. For the data plotted in **Figure 3F**, the normalized axonal fluorescence for all three strains was normalized to the worm CPX-1:: GFP strain for comparison. Data were analyzed with custom software in IGOR Pro (WaveMetrics, Lake Oswego, OR, United States) (Burbea et al., 2002; Dittman and Kaplan, 2006). As previously reported, we did not observe a correlation between expression levels and rescue efficiency (Wragg et al., 2013; Radoff et al., 2014). The single-copy CPX-1:: GFP transgene fully rescued in all behavioral assays even though it was expressed near the lower limit of our imaging sensitivity. All transgenic strains used in this study displayed a higher expression level of CPX relative to this single-copy strain.

### Protein Purification

CPX-1+W (JP767), CPX-1 (1CT)+W (JP773), mCpx1+W (JP790), mCpx1 (1CT)+W (JP791), CPX-1(FFF/AAA)+W (JP793), CPX-1 (FFF/III)+W (JP794), CPX-1 (LV/EE)+W (JP915), CPX-1(16)+W (JP916), CPX-1 (16mouse7)+W (JP917) constructs were cloned into the pET28a vector using standard techniques. These constructs contain a His<sup>6</sup> tag, a T7 tag and a thrombin cleavage site to facilitate purification. BL21-DE3 Escherichia coli were transformed and grown in Luria-Bertani media (LB) with kanamycin (50 µg/mL) to an optical density of 0.6. Cells were induced with isopropyl thiogalactopyranoside (IPTG) (400 µg/ml), grown for 3 h at 37◦C, pelleted, resuspended in buffer (350 mM NaCl, 20 mM imidazole, 20 mM Tris–HCl pH 8, 1.5 mM BME, 2 mM DTT), lysed by sonication, and pelleted at 40,000 r.p.m. for 40 min. The supernatant was purified on a Ni-NTA column (Qiagen, Hilden, Germany). Protein was eluted in elution buffer (350 mM NaCl, 250 mM Imidazole, 20 mM Tris–HCl, 1.5 mM BME, 2 mM DTT) then dialyzed into buffer (150 mM NaCl, 50 mM Tris–HCl pH 8). Protein was then concentrated and FPLC was performed. Sephadex G-25 Fine beads (Sigma) were then used for buffer exchange (150 mM NaCl, 50 mM Tris–HCl pH 8, 5 mM EGTA). Protein concentrations were estimated by absorbance at 280 nm using the extinction coefficient. For the mouse mCpx1 NMR studies, the CTD construct (residues 71–134) was cloned into a SUMO fusion vector and expressed and purified as previously described for worm complexin using nickel-affinity chromatography. Briefly, BL21(DE3) E. coli cells were transformed with the relevant plasmid, grown in LB media for 4 h as a small culture and transferred to 100 mL minimal media containing <sup>15</sup>N-labeled ammonium chloride and <sup>13</sup>C-labeled glucose and grown overnight. Cells were grown in 1 L minimal media to an optical density of 0.6 before induction with IPTG for 3–4 h. Cells were lysed by sonication on ice, supernatants were clarified by centrifugation at 40,000 r.p.m. for 45 min. The SUMO-tagged fusion protein was purified from supernatants on a Ni-NTA column and dialyzed into 20 mM Tris pH 8, 150 mM NaCl, 1 mM dithiothreitol, followed by cleavage of the SUMO tag using the SUMO protease Ulp1. A second Ni-NTA affinity purification was used to remove the SUMO tag. Proteins were then dialyzed into distilled water, frozen and lyophilized. For NMR, lyophilized proteins were dissolved in 50 mM phosphate pH 6.1, 1 mM dithiothreitol, 0.5 mM EDTA, with 60 mM NaCl. Protein concentrations were estimated by absorbance at 280 nm using the coefficients of the individual amino acids in the protein sequence.

## Small Unilamellar Vesicle (SUV) Preparation

Lipids were obtained from Avanti Polar Lipids and stored at −20◦C. A lipid mixture composed of 85% 1-palmitoyl-2-oleoyl-phosphatidylcholine (POPC), and 15% 1-palmitoyl-2 oleoyl-phosphatidylserine (POPS) was dried under a stream of N<sup>2</sup> gas then residual solvent was removed under vacuum for 2 h. The lipid film was then rehydrated in assay buffer

(150 mM NaCl, 50 mM Tris–HCl pH8, 5 mM EGTA) to obtain a lipid concentration of 4 mM. The resulting SUVs underwent bath sonication and pelleted at 60,000 r.p.m. for 2 h (Sorvall RC M120 EX Ultracentrifuge, S120AT2 rotor). Vesicle size and purity were verified by dynamic light scattering using a Zetasizer Nano-S (Malvern Instruments). Lipid concentration was determined based on the amounts of starting lipid and using a phosphate quantification assay. Perchloric acid was added to lipid samples and heated to 150◦C for 1 h. Ammonium molybdate and ascorbic acid were added to samples and heated for 10 min at 100◦C. Absorbance was measured at 797 nm and lipid concentrations were obtained through comparison to phosphate standards. Vesicles were stored at 4◦C and used within 1 week.

#### Fluorescence Titration Measurements

Tryptophan fluorescence was measured at 22◦C with either a spectrofluorometer (Photon Technology International) or a SpectraMax M5 microplate reader (Tecan). For the spectrofluorometer, emission spectra were recorded between 300 and 450 nm (1 nm step) with an excitation wavelength of 280 nm, at slit widths of 4 nm. For the plate reader, emission at 350 nm was recorded in a 96-well plate using an excitation wavelength of 280 nm with 6 flashes per read. Protein–lipid binding was determined from the increase in tryptophan emission fluorescence intensity upon addition of SUVs corrected for fluorescence in SUVs alone. The data were analyzed using custom software in IGOR Pro (WaveMetrics, Lake Oswego, OR, United States).

#### NMR Spectroscopy

Perdeuterated CTD and perdeuterated DPC (Avanti Polar Lipids) were used for triple resonance experiments used to assign the CTD in the presence of DPC (i.e., <sup>15</sup>N, <sup>13</sup>C, <sup>2</sup>H labeled protein) and for the HSQC-NOESY-HSQC experiment for the CTD with DPC micelles (i.e., <sup>15</sup>N, <sup>2</sup>H labeled protein). DPC micelles were prepared by resuspending a dried lipid film at the desired stock concentration (Snead et al., 2017). Experiments included TROSY versions of HNCACB, HN(CO)CACB, HNCACO, HNCO, HNCA, and HNCOCANH. Data were collected on 600 MHz (Weill Cornell) and 900 MHz (New York Structural Biology Center) cryoprobe-equipped spectrometers and indirectly referenced to 4,4-dimethyl-4-silapentane-1-sulfonic acid and ammonia based on the position of the water resonance. Cα-Cβ secondary shifts were calculated as the difference between the observed carbon chemical shifts and random coil values tabulated from linear hexapeptides in 1M urea at pH 5.0 and 25◦C.

#### Calculation of Amphipathic Moments and Helicity

The amphipathic moment vector was defined by **Equation 2** where −→µ<sup>H</sup> is the net moment vector of an N-residue helix (in complex notation), r<sup>k</sup> = hydrophobicity of the k th residue using the Moon-Fleming scale (multiplied by −1) and δ = 100◦ is the angle between successive residue side chains moving counterclockwise (Eisenberg et al., 1982; Moon and Fleming, 2011).

$$\overrightarrow{\mu\_{\rm H}} = \sum\_{k=1}^{N} r\_k \left\{ \cos \left( \left( k - 1 \right) \cdot \delta \right) - i \cdot \sin \left( \left( k - 1 \right) \cdot \delta \right) \right\} \tag{2}$$

The distribution of amphipathic moment magnitudes for random 12-mer and 18-mer peptides was estimated by generating 10<sup>6</sup> random peptide sequences (excluding proline from all but the first two and last two residues) and computing the amphipathic moment for each peptide. The proline-free constraint was implemented to allow for stable alpha helix packing. The cumulative distributions of these ensembles are shown in **Figure 9G**. Because the aspartate and glutamate hydrophobicity were assigned at low pH in the Moon-Fleming scale, we substituted those values with the octanol hydrophobicity values (−3.64 and −3.63 kCal/mol, respectively). Percent helicity was computed using Agadir as described previously (Radoff et al., 2014), and average values were normalized to the average nematode helicity for comparison.

#### Molecular Dynamics Simulations

The peptide-membrane binding free energy profiles (potentials of mean force, PMF) were computed along the normal of a model lipid bilayer, using Molecular Dynamics (MD) simulations with the CHARMM36 all-atom force field (MacKerell et al., 1998). The lipid bilayer was modeled by a compositionally symmetric mixture of 100 DOPC:DOPE:DOPS lipids (mole ratio of 60:25:15) pre-assembled using the CHARMM-GUI server (Jo et al., 2008). The bilayer surfaces were aligned parallel to the XY plane and solvated in a cubic water box (70 Å × 70 Å × 110 Å) with periodic boundary conditions (PBCs). Two complexin peptides were positioned near the bilayer (one on each side) to exploit available symmetry. Both peptides were modeled as initially disordered (Snead et al., 2014), and both ends of each peptide were capped with neutral end groups (acetylated N-terminus and amidated C-terminus). In addition, each system was brought to electrical neutrality and adjusted to a NaCl concentration of 0.15 M by randomly replacing water molecules with ions. The equilibration phase of the simulations was conducted with the NAMD software (version 2.10) under isothermal-isobaric (NPT) ensemble (P = 1 atmosphere, T = 310 K) (Phillips et al., 2005). A 2000-step energy minimization and 2 nanoseconds (ns) MD simulation with harmonic restraints (force constant k = 5 kcal/mol) were conducted on the positions of both the lipid heavy atoms and peptide backbone atoms using an integration time interval of 1 femtosecond (fs). The system was further equilibrated for another 10 ns with an integration time interval of 2 fs after removal of restraints on the lipid heavy atoms. In all simulations, the particle mesh Ewald (PME) algorithm was used for long-range electrostatic interactions, while a 14-Å cutoff distance was used for van der Waals interactions (Darden et al., 1993).

### Free Energy Calculation

fnmol-10-00146 May 23, 2017 Time: 16:8 # 5

After equilibration, the bilayer-binding PMF of each peptide was explored using united free energy dynamics (UFED) (Cuendet and Tuckerman, 2014), an enhanced sampling approach that combines the advantages of driven adiabatic free energy dynamics (dAFED) and metadynamics methods (Laio and Parrinello, 2002; Zhu et al., 2002). Within the framework of dAFED, the vertical peptide-bilayer separation distance, 1z was chosen as a collective variable (CV) for each peptide and computed during the simulations. The 1z is defined as the distance between the center of mass (COM) of sidechain heavy atoms of the four residues in the middle (residues 138–141 for worm complexin C-terminal peptide) and the COM of all lipid phosphate atoms along the bilayer normal. To ensure the system quickly crosses large free energy barriers, 1z was harmonically (k = 50 kcal/mol/Å) coupled to a fictitious particle. By choosing a much larger virtual mass for the fictitious particle (2 × 10<sup>11</sup> kJ/mol/Å<sup>2</sup> ) than the total mass of the physical system, the dynamics of the fictitious particle was adiabatically decoupled from the physical system. This allowed assignment of a high temperature (T = 2500 K) to the dynamics of the fictitious particle that is able to cross high free energy barriers and drive the physical system to evolve faster along 1z at room temperature. An external biasing Gaussian potential of fixed height (0.1 kcal/mol) and width (0.1 Å) was added to the Hamiltonian of the system every 2500 time steps as a history-dependent function of 1z. All UFED simulations were conducted using the ACEMD molecular dynamics software (Harvey et al., 2009) with the pluMED software as a plugin that supplies the UFED function (Abrams and Tuckerman, 2008; Bonomi et al., 2009). With the same atomic coordinates from the final configuration of the equilibration run, the atomic velocities were randomly regenerated at 310 K to start 10 independent replicas of UFED simulations that sample the one-dimensional free energy space in a parallel manner. The upper limit of 1z was set to 30.0 Å to avoid interactions between the two peptides. By adopting the hydrogen mass repartitioning scheme, we conducted all UFED simulations with a time step of 4 fs under canonical (NVT) ensemble (T = 310 K) (Bonomi et al., 2009). For each system, the total sampling time exceeded 3 µS, and the time series of 1z and its corresponding virtual counterpart (i.e., the trajectory of the fictitious particle) were used to reconstruct the PMF along 1z based on the thermodynamic forces (Cuendet and Tuckerman, 2014).

#### Statistics and Protein Sequence Analysis

For single comparisons, statistical significance was defined as p < 0.01 by Student's t-test. In cases where multiple comparisons were made using the same data sets, ANOVA followed by the post hoc Tukey–Kramer method was used to compute significance as defined by p < 0.01. Multiple protein sequence alignment was performed using Clustal Omega. Helical wheels were generated using custom software implemented in Igor Pro. Helical propensity was computed with Agadir (Munoz and Serrano, 1997).

### RESULTS

### Conserved Features of Complexin–SNARE Interactions

We first examined highly homologous regions of complexin shared between mouse and worm. The major defining feature of complexin is its CH, the 25 residue alpha-helical region of complexin that directly binds to the assembled SNARE complex (**Figure 1A**). This is by far the most conserved domain of complexin and shows comparable conservation to the SNARE domains of the neuronal SNARE proteins – especially VAMP2 and Syntaxin 1, its two binding partners (**Figure 1B**). The SNARE domains of VAMP2 and Syntaxin 1 share 87 and 85% sequence identity, respectively, between C. elegans and mouse. Likewise, the CH of C. elegans CPX-1 is 76% identical to that of mouse complexin 1 (hereon referred to as mCpx1). The other protein domains of complexin are far less conserved as shown in **Figure 1C**. Several crystal structures of complexin bound to the ternary SNARE complex as well as biochemical and in vivo studies have identified specific residues required for the tight association of the CH and the ternary SNARE bundle, and these residues are almost perfectly conserved throughout phylogeny (Nonet et al., 1998; Saifee et al., 1998; Bracher et al., 2002; Chen et al., 2002; Giraudo et al., 2006; Xue et al., 2007; Maximov et al., 2009) (**Figure 1D**). To disrupt CPX-1 binding, several of these conserved SNARE residues were mutated to alanines (**Figure 1E**). We assessed the impact of perturbing CH binding at cholinergic synapses in C. elegans employing acute sensitivity to the cholinesterase inhibitor aldicarb by quantifying the rate of paralysis upon exposure to 1 mM aldicarb (**Figure 2A**). Numerous studies have established that impairment of ACh release decreases sensitivity to aldicarb and slows the rate of paralysis, whereas hypersecretory mutations accelerate paralysis (Rand and Russell, 1985; Miller et al., 1996; Nurrish et al., 1999; Mahoney et al., 2006; Martin et al., 2011). As described previously, worms rapidly paralyzed in the absence of complexin due to a high rate of spontaneous SV fusion (Hobson et al., 2011; Martin et al., 2011; Wragg et al., 2013). This hypersecretion was fully suppressed to wild-type levels by expressing a full-length CPX-1 in all neurons (**Figure 2B**). Both deletion of the central helix (1CH) and substitution of two key residues in the central helix (KY/AA) completely eliminated CPX-1 inhibitory function by this assay (**Figures 2B,C**). Note that all CPX-1 variants used in this study were tagged with a C-terminal GFP and that the fulllength CPX-1:: GFP fusion protein fully rescued cpx-1 mutants either as a multi-copy array or single-copy integrant (**Figure 3B**) (Martin et al., 2011; Wragg et al., 2013). Synaptic expression levels were assessed by imaging fluorescence in the dorsal nerve cord for all strains, and representative measurements of expression for several strains are shown in **Supplementary Table S1**. All strains analyzed in this study expressed complexin at higher levels than the single-copy integrant.

In principle, the hypersecretion observed in cpx-1 mutants could emerge from an independent secretion pathway unrelated to canonical SV fusion at the synapse. Perhaps an unanticipated change in trafficking in the absence of CPX-1 could account for

residues used to define the domains are listed for each species below. (D) 33 resides of the SNARE domains for VAMP2 (Top) and Syntaxin 1 (Bottom) are aligned for mouse, worm, and fly with 9 central SNARE layer residues highlighted in green. The residues that directly interact with complexin are depicted in blue and orange (arrowheads). Note that all 8 of these residues are conserved between the three species. (E) The C. elegans orthologs of the neuronal SNAREs. Key residues that contribute to the binding reaction are mutated to alanines in the '3A' snb-1 VAMP2 mutant and '5A' unc-64 Syntaxin 1 mutant.

the hypersensitivity to cholinesterase inhibitors. However, the additional secretion events observed in cpx-1 mutants relied on the same exocytosis machinery as in wild-type animals since SNARE hypomorphic mutants in snb-1 synaptobrevin 1 and unc-64 syntaxin 1 strongly suppressed the hypersecretion phenotype of cpx-1 (**Figure 2D**). Furthermore, a weak hypomorphic mutant in the critical SV fusion protein unc-13 Munc13 also suppressed cpx-1 (**Figure 2D**). These observations

FIGURE 2 | Function impact of disrupting complexin–SNARE interactions. (A) Cartoon of the worm neuromuscular junction (NMJ) depicting acetylcholine (ACh, red) release as well as synaptic cleft cholinesterase (AChE, green), and the cholinesterase inhibitor aldicarb. (B) Average paralysis time course on 1 mM aldicarb for wild type (black), cpx-1(ok1552) null mutant (red), full-length rescue CPX-1 (blue), and rescue with CPX-1 either lacking its central helix (1CH, green), or with two alanine substitutions in two conserved central helix residues (KY/AA, orange) to disrupt SNARE binding. All rescue strains are in cpx-1(ok1552) null mutant background. (C) Quantification of aldicarb rescue based on the time to 50% paralysis (t0.5) for full-length CPX-1 (blue), central helix deletion (1CH, green), and the central helix double point mutant (KY/AA, orange). (D) Aldicarb paralysis time course for wild type (black), cpx-1 (red), or cpx-1 double mutant together with a hypomorphic mutant of either unc-64(e246) Syntaxin 1 (blue), snb-1(md247) Synaptobrevin 1 (green), or unc-13(e1091) Munc13 (pink). (E) Average aldicarb time course for wild type (black), cpx-1 null mutant (red), and two SNARE null mutants rescued with mutated SNARE domains. The snb-1 synaptobrevin null mutant (js104) was rescued with mutations in three complexin-binding residues (3A, blue) as indicated in (E). The unc-64 syntaxin 1 null mutant (js115) was rescued with mutations in five complexin-binding residues (5A, green). Note that SNARE mutants lacking the ability to bind complexin phenocopy the cpx-1 mutant. (F) Voltage-clamp recordings of spontaneous neuromuscular junction cholinergic fusion events in the absence of external calcium for wild type, cpx-1 mutants, and rescue of cpx-1 with a variant lacking its last 12 residues (112) as indicated. Blue dots indicate individual vesicle fusion events. (G) Plot of percent rescue of spontaneous fusion rates versus percent rescue of aldicarb sensitivity (based on t0.5) for several distinct cpx-1 mutants. This data was reanalyzed from Martin et al. (2011) (red), Wragg et al. (2013) (green), and Radoff et al. (2014) (blue) as indicated by symbol color. Data are mean ± SEM. ∗∗p < 0.01 by Tukey–Kramer test for multiple comparisons. n.s. = not significant.

indicate that neurotransmitter secretion remained highly sensitive to the neuronal SNAREs and essential SNARE-binding proteins in the absence of CPX-1.

The deep conservation of the SNARE residues that interact with the CH domain suggests that complexin inhibition relies on this interaction. To test for this possibility, we rescued neuronal SNARE mutants with mutated SNARE proteins designed to eliminate complexin binding (Maximov et al., 2009). Three complexin-binding residues of the VAMP2 ortholog SNB-1 were mutated to alanine (DLV/AAA = '3A'), and this construct was expressed in snb-1 null mutants to replace endogenous vSNAREs. The SNB-1(3A) constructs fully phenocopied cpx-1 null mutants in the presence of endogenous complexin (**Figures 1E**, **2E**). Similar results were reported for mCpx1 in a previous study

FIGURE 3 | Mouse complexin 1 fails to inhibit secretion in worm synapses. (A) Aldicarb paralysis time course for wild type (black), cpx-1 mutant (red), and cpx-1 mutant rescued with either worm CPX-1 (blue) or mouse mCpx1 (gray). (B) Aldicarb paralysis comparison of wild type (black), cpx-1 null mutant (red), and cpx-1 rescue using either a multi-copy array integrant (CPX-1, blue) or a single-copy integrant (s.c. CPX-1, green). (C) Percent rescue based on 50% paralysis time point for a multi-copy array of worm CPX-1 (blue), multi-copy array of mouse Cpx1 (gray), and a single-copy array of worm CPX-1 (green). (D) Representative confocal images of dorsal cord axonal worm CPX-1:: GFP (Top) co-expressed with mCherry::RAB-3 (Middle), and a merged display of both images (Bottom). Scale bar is 5 µm. (E) Representative confocal images of axonal mouse Cpx1:: GFP (Top) co-expressed with mCherry::RAB-3 (Middle), and a merged display of both images (Bottom). (F) Quantification of axonal protein abundance for the multi-copy arrays of worm CPX-1 (CPX-1, blue) and mouse Cpx1 (Cpx1, gray) as well as the single-copy array of worm CPX-1 (s.c. CPX-1, green), normalized to an internal fluorescent standard (see Materials and Methods). (G) Aldicarb paralysis time course of slo-1 K(Ca) mutants expressing the HisCl channel under a cholinergic promoter in the presence (blue) and absence (black) of histamine. (H) Summary of aldicarb paralysis in the absence and presence of histamine for three genetic backgrounds all expressing the HisCl channel: wild type (Left), slo-1 K(Ca) mutant (Middle), and cpx-1 expressing mCpx1 (Right). The particular time point in the aldicarb assay is indicated above the bars for each genotype. (I) Aldicarb paralysis time course for wild type (black) or mCpx1 over-expressed in wild type animals (pink). Data are mean ± SEM. ∗∗p < 0.01 by Tukey–Kramer test for multiple comparisons in (C,F) or Student's t-test for (H). n.s. = not significant.

(Maximov et al., 2009). Likewise, a 5-residue mutant of the Syntaxin 1 ortholog UNC-64 (LMDMD/AAAAA = '5A') expressed in the unc-64 null mutant background identically phenocopied cpx-1. Both of these SNARE variants were functional since the transgenic animals expressing them were living, highly mobile, and displayed excessive ACh secretion, whereas null mutants in either snb-1 or unc-64 die at an early larval stage (Nonet et al., 1998; Saifee et al., 1998). Furthermore, hypomorphic alleles of these SNAREs are known to be severely uncoordinated and display strong resistance to aldicarb (Miller et al., 1996; Nonet et al., 1998; Saifee et al., 1998). Thus, the aldicarb hypersensitivity phenotype of cpx-1 arose specifically from the loss of a complexin–SNARE interaction rather than through some unidentified complexin function. Prior electrophysiological studies in cpx-1 mutants demonstrated that spontaneous fusion in the absence of external calcium is highly elevated when complexin function is impaired (**Figure 2F**) (Hobson et al., 2011; Martin et al., 2011; Wragg et al., 2013; Radoff et al., 2014). Replotting data from several of these studies against the percent rescue of spontaneous fusion rate versus the percent rescue of aldicarb sensitivity for a variety of CPX-1 structural mutations revealed a strong correlation (**Figure 2G**). The acute aldicarb sensitivity assay therefore provides a reasonable quantitative assessment of CPX-1 inhibitory function in the context of an intact behaving animal.

### Mouse Complexin 1 Fails to Inhibit Secretion in Worm Synapses

To examine the functional conservation of complexin across distantly related species, mouse mCpx1 was expressed as a multi-copy array in the nervous system of C. elegans cpx-1 mutants lacking endogenous CPX-1. These transgenic animals exhibited only a small degree of functional rescue based on aldicarb sensitivity compared to either single-copy CPX-1:: GFP (s.c. CPX-1) or over-expressed CPX-1:: GFP (**Figures 3A–C**). Failure to rescue could have resulted from poor protein expression or non-synaptic localization of mouse complexin. However, mCpx1:: GFP synaptic localization was similar to CPX-1:: GFP when co-expressed with a SV marker (**Figures 3D,E**). Furthermore, mCpx1:: GFP synaptic abundance was quantitatively similar to CPX-1:: GFP (**Figure 3F**). The GFP fusion itself did not impair mCpx1 function because untagged mCpx1 also failed to rescue (data not shown). Thus, mCpx1 failed to restore proper function in cpx-1 mutants, and neither expression levels nor mislocalization could account for this failure.

An important difference between mammalian and invertebrate complexins is the relative impact on promoting calcium-triggered release versus inhibiting spontaneous fusion. In mouse, loss of mCpx1/2 causes a significant decrease in calcium-triggered fusion while spontaneous fusion is either increased or decreased over a broad range depending on the neuronal subtype and perhaps on the methodologies employed (Xue et al., 2007; Maximov et al., 2009; Strenzke et al., 2009; Lin et al., 2013; Yang et al., 2013). Expression of mCpx1 in fly synapses significantly boosts calcium-triggered neurotransmitter release (Cho et al., 2010). However, in both worm and fly, the most conspicuous effect of losing complexin is a profound increase in the rate of spontaneous fusion (Huntwork and Littleton, 2007; Cho et al., 2010; Hobson et al., 2011; Martin et al., 2011; Wragg et al., 2013). Accordingly, the hypersecretion observed in worm cpx-1 mutants expressing mCpx1 could arise from an upregulation of calcium-triggered release rather than a failure to suppress spontaneous release. To examine this possibility in vivo, we expressed a fly histamine-gated chloride channel (HisCl) in worm cholinergic neurons and performed aldicarb sensitivity assays in the presence and absence of histamine (Pokala et al., 2014). Partial silencing of cholinergic neurons by activation of hyperpolarizing HisCl channels would be expected to decrease calcium-triggered ACh release and to delay paralysis on aldicarb relative to control animals. To demonstrate this effect, HisCl was expressed in slo-1 K(Ca) channel mutants (**Figure 3G**). These mutants are hypersensitive to aldicarb due to elevated calcium-triggered secretion in the absence of a repolarizing K(Ca) current (Wang et al., 2001; Martin et al., 2011). As anticipated, the addition of histamine significantly decreased ACh secretion in slo-1 mutants as well as wild-type animals (**Figure 3H**). However, the same treatment had no effect on either cpx-1 mutants or cpx-1 mutants expressing mCpx1, suggesting that the enhanced secretion observed in these transgenic animals derives from enhanced spontaneous fusion rather than increased calcium-triggered fusion (**Figure 3H**). Finally, no enhancement of secretion was observed when mCpx1 was over-expressed in wild-type animals to determine if this variant could drive additional secretion via its facilitatory function (**Figure 3I**). Taken together, these experiments indicate that a conventional spontaneous SV fusion pathway is strongly elevated in cpx-1 null mutants and that mCpx1 suppresses this fusion pathway to only a small extent despite being highly expressed and properly localized to worm synapses.

#### The C-terminal Domain of Mouse Cpx1 Accounts for Its Failure to Inhibit in Worm

Having established that mCpx1 fails to inhibit SV fusion at worm synapses, we next explored individual complexin domains within mCpx1 to identify which domains failed to substitute for their homologous worm complexin domains (**Figure 4A**). Each domain of worm CPX-1 was substituted with the corresponding region of mCpx1 and expressed in cpx-1 null mutants to assess the degree of functional rescue by aldicarb sensitivity. As shown in **Figure 4B**, some substitutions such as the AH domain fully restored wild-type complexin function (Radoff et al., 2014). In fact, of the four protein domains within complexin, only introduction of the mouse CTD recapitulated a failure to restore function to the same degree as full-length mCpx1 (**Figure 4C**). These findings indicate that the lack of functional rescue originates in the CTD of mCpx1. To further explore this region, several chimeras with varying lengths of the mCpx1 C-terminus were expressed in cpx-1 null mutants. We found that even replacing only the last six residues of worm CPX-1 with the corresponding mouse mCpx1 residues strongly impaired the

FIGURE 4 | The C-terminal domain of mouse Cpx1 accounts for its failure to inhibit in worm. (A) Protein sequence alignment for worm CPX-1 and mouse Cpx1 showing the N-terminal domain (NTD), accessory helix domain (AH), central helix domain (CH), and C-terminal domain (CTD). Identical residues are indicated with a blue diamond. (B) Aldicarb time course for wild type (black, open circles), cpx-1 (red), and cpx-1 rescued with either worm CPX-1 (+CPX-1, black, filled circles), mouse Cpx1 (mCpx1, gray), or a chimeric worm CPX-1 variant containing the mouse accessory helix [+CPX-1(mAH), blue]. (C) Summary of aldicarb rescue of cpx-1 mutants (by 50% paralysis time) for full-length worm CPX-1 (black), mouse Cpx1 (gray), and four chimeric worm CPX-1 variants containing mouse Cpx1 domains: mouse NTD (orange), mouse AH (blue), mouse CH (red), and mouse CTD (purple). (D) Summary of aldicarb rescue of cpx-1 mutants by 50% paralysis time for full-length worm CPX-1 (light gray), mouse Cpx1 (dark gray), and three CTD chimeras containing various lengths of mouse Cpx1 substituted for the corresponding worm CPX-1 sequence as indicated (light/dark gray). (E) cpx-1 mutant rescue with worm CPX-1 harboring various deletions in the CTD lacking the last 50 residues (150), last 12 residues (112), or last 6 residues (16). Data are mean ± SEM. ∗∗p < 0.01 by Tukey–Kramer test for multiple comparisons and #not significantly different from rescue with full=length mCpx1.

inhibitory function of CPX-1 (**Figure 4D**). If the majority of the CTD sequence was deleted rather than replaced, complexin function was impaired to a similar extent, suggesting that adding back mouse complexin residues failed to restore functionality lost with the deleted worm residues (**Figure 4E**). Thus, the highly divergent CTD of complexin in mouse and worm accounts for the lack of functional rescue.

#### Monitoring Membrane Interactions with a C-terminal Tryptophan

Why does the CTD of mCpx1 fail to restore function in worm synapses? Previous studies in worm demonstrated that a major role of the CPX-1 CTD is to properly localize complexin to SVs via a membrane-binding region comprising the last ∼34 residues of CPX-1 (Wragg et al., 2013, 2015; Snead et al., 2014). mCpx1 also contains a membrane-binding region (Seiler et al., 2009; Snead et al., 2014; Gong et al., 2016), and this region is required for proper inhibition of spontaneous fusion in cultured mouse hippocampal neurons (Kaeser-Woo et al., 2012). The diameter of a typical SV is ∼30 nm in C. elegans and ∼40 nm in mammalian neurons, making it one of the most highly curved membranes within a cell (Rostaing et al., 2004; Takamori et al., 2006). To examine the membranebinding properties of complexin on highly curved membranes in vitro, recombinant CPX-1 terminating with an added tryptophan (CPX-W) was incubated with small unilamellar vesicles (SUVs) and the fluorescence spectrum of tryptophan excited at 280 nm was monitored (**Figure 5A**). Typical SUV preparations comprised a relatively uniform population of vesicles with an average diameter of 35–45 nm as determined by dynamic light scattering (**Figure 5B**). Note that neither worm nor mouse complexin 1 contains endogenous tryptophan residues. CPX-1 preferentially binds to highly curved membranes irrespective of lipid head-group composition (Snead et al., 2014). The functionality of CPX-1 containing a C-terminal tryptophan was confirmed in vivo by rescue of cpx-1 mutants with a CPX-W construct (**Figure 5C**). Thus, including a terminal tryptophan did not impair CPX-1 inhibition of ACh secretion. Full-length recombinant CPX-W incubated with increasing concentrations of SUVs exhibited a corresponding increase in peak emission fluorescence (**Figure 5D**). Furthermore, the location of the emission peak shifted toward shorter wavelengths at high lipid concentrations, consistent with tryptophan partitioning into the low dielectric environment of the SUV lipid bilayer (Ladokhin et al., 2000; Ladokhin and White, 2001). This membrane partitioning depended on the C-terminal region of CPX-1 as deletion of the last 34 residues (but retaining the C-terminal tryptophan) eliminated detectable increase in emission and peak blue-shift (**Figures 5E,F**), indicating that partitioning is dependent on and reflects membrane-binding by complexin. Corresponding in vitro experiments with mouse mCpx1-W confirmed that mammalian complexin also bound SUVs with a similar affinity in a manner strictly depending on the presence of the CTD (**Figure 5G**). Thus, the worm and mouse complexin 1 bound to SUVs with comparable affinities in vitro despite significant differences in their primary sequence.

## Impact of C-terminal Hydrophobic Residues

Inspection of the primary amino acid sequence of the CTDs of worm and mouse complexin revealed little similarity beyond the previously described amphipathic region (L117 – K136 in worm CPX-1 and E114 – P125 in mouse mCpx1) (Wragg et al., 2013; Snead et al., 2014). To better understand the contribution of the last few residues to CPX-1 function and lipid binding, we explored several mutations, focusing on the three C-terminal phenylalanines characteristic of nematode complexins (**Figure 6A**). Structural studies (Snead et al., 2017) highlighted a potential role in membrane binding for this C-terminal motif. Several substitutions of one or more phenylalanines were made to alter the overall hydrophobicity of these last six residues, and the effective hydrophobicity was estimated using an empirical scale created by Moon and Fleming (2011). This scale was generated from measurements of the free energy change for moving an amino acid side chain from the cytoplasm to the middle of a lipid bilayer in the context of a folded transmembrane protein. All mutations in this C-terminal region of CPX-1 produced significant impairments in complexin inhibitory function as measured by aldicarb sensitivity (**Figures 6B,C**). Moreover, a strong correlation was observed between the hydrophobicity of the six residue C-terminal motif and the ability of CPX-1 to inhibit ACh secretion (**Figure 6D**). These same mutations were introduced into recombinant CPX-W to quantify the degree to which membrane binding was impaired. To monitor changes in binding affinity, we calculated both the initial slope of normalized fluorescence increase at 350 nm versus lipid concentration and the relative increase in fluorescence in the presence of 0.9 mM lipid versus lipid-free medium (**Figure 6E**). These approaches minimized inaccuracies encountered at high lipid concentrations due to light scattering (Ladokhin et al., 2000; Ladokhin and White, 2001). By either measure of binding affinity, the two perturbations that most strongly disrupted lipid binding were the deletion of the last six residues (16) and the substitution of all three phenylalanines for alanines (3 × F/A) (**Figure 6F**). We noted that mCpx1 is considerably less hydrophobic than worm CPX-1 over the last six residues (**Figure 6A**), but surprisingly, replacing the last six CPX-1 residues with the last seven mCpx1 residues (16m7) had only a modest impact on membrane binding. When the relative functionality of the C-terminal CPX-1 variants was plotted versus their relative membrane binding, three variants failed to show a correlation between membrane binding and CPX-1 inhibition (**Figure 6G** pink region). Full-length mCpx1, CPX-1 with the mCpx1 last seven residues, and a double point mutation in the amphipathic region (L117E V121E) all exhibited reasonably strong lipid binding but failed to rescue in vivo. Taken together with the other C-terminal mutations, these results demonstrate that membrane binding by the CTD is necessary but not sufficient for CPX-1 inhibitory function. Finally, since almost all known non-prenylated complexins terminate with a lysine, we substituted this lysine with either arginine (K/R) or alanine (K/A) and rescued cpx-1 mutant animals with these variants (**Figure 6H**). Both substitutions significantly impaired

FIGURE 5 | Monitoring membrane interactions with a C-terminal tryptophan. (A) Cartoon of the CPX-1 C-terminal region containing a tryptophan (W) binding a small unilamellar vesicle (SUV) either with (full-length) or without the last 50 residues of CPX-1 (1CTD). Note that tryptophan fluorescence at 350 nm increases when it penetrates into the lipid bilayer (White ref). (B) Representative dynamic light scattering (DLS) data for the SUVs used in this study. SUVs containing 85% POPC and 15% POPS were prepared by sonication (see Materials and Methods for details). The mean SUV diameter was estimated to be 35 ± 4 nanometers for this sample. Histogram was fit to a log-normal distribution (red). (C) Aldicarb time course for wild type (black), cpx-1 mutants (red), and cpx-1 mutants rescued with a CPX-1 harboring a C-terminal tryptophan (CPX-W, blue). (D,E) Emission spectrum for recombinant full-length worm CPX-W or CPX-W lacking the last 50 residues (1CT) excited at 280 nm in the presence of varying concentrations of lipids as indicated. Data is normalized to the fluorescence measured at 300 nm after background subtraction. Emission peaks are indicated with pink arrowheads. (F) Location of the emission peak is plotted as a function of lipid concentration for full-length CPX-W (black) and CPX-W lacking its C-terminal 50 residues (1CT, green). (G) Normalized emission fluorescence at 350 nm is plotted as a function of lipid concentration for worm CPX-1 (black) and mouse Cpx1 (red) for full-length constructs (solid circles) and variants lacking their C-terminal domains (open circles). Data are mean ± SEM.

CPX-1 inhibition, indicating that the conserved terminal lysine was required for CPX-1 inhibitory function. The in vitro and in vivo experiments described here demonstrate that although membrane binding by the CTD of complexin is critically important, other features of this region beyond membrane binding appear to play a role, and these features are poorly conserved between species.

### Molecular Dynamics of Membrane Interactions with the Complexin C-terminal Motif

Perhaps the simplest hypothesis for the failure of the mouse C-terminal motif to function when swapped into worm complexin is a large difference in hydrophobicity of these residues. However, in vitro membrane-binding experiments (**Figures 5**, **6**) indicated that both the mCpx1 CTD and the 16m7 variant bound to membranes with a similar affinity to CPX-1 despite being significantly less hydrophobic. To further investigate the nature of complexin membrane interactions in a structural context, we performed all-atom MD simulations of peptides comprising the last eight residues of either worm CPX-1 or mouse mCpx1 in proximity to a lipid bilayer containing phosphatidylcholine (POPC), phosphatidylethanolamine (POPE), and phosphatidylserine (POPS) in a ratio of 60:25:15 (see Materials and Methods for details of the MD simulations). The simulations revealed that both worm and mouse Cpx peptides partitioned into the lipid bilayer and exhibited energetics that favored membrane binding to a similar degree. In fact, the mouse peptide displayed a somewhat deeper free energy trough than the worm peptide (−3.3 kcal/mol vs. −5.4 kcal/mol for worm and mouse, respectively) (**Figure 7A**). The energy minimum occurred at a penetration depth of 10 Å from the bilayer center for CPX-1 and 15 Å for mCpx1. The C-terminal

FIGURE 7 | Molecular dynamic simulations of complexin C-terminal peptides binding to the bilayer. (A) Bilayer-binding PMFs of four peptides: CPX-1 = KGFPFFGK (black), 3×F/A = KGAPAAGK (red), 3×F/I = KGIPIIGK (purple), mCpx1 = PLQDMFKK (blue). The positions of free energy minima for CPX-1, 3×F/I, and mCpx1 are indicated with orange arrowheads and average phosphorous (P) atom position is indicated with a green vertical line. (B) Density profiles of lipid atoms in the head-group and linker regions and water molecules: N, nitrogen, P, phosphate, O, oxygen atoms of the glycerol group, C2 is the start of the two acyl chains, tail, all atoms of the two lipid tails. X-axis represents the vertical distance (z) away from the bilayer mid-plane along the bilayer normal. The vertical green line (same as in A) indicates the average z position of the P atoms. (C,E) Backbone conformation of bilayer-bound peptides. Atom color scheme: C (cyan), N (blue), O (red), S (yellow). Alignment of peptide backbone ensemble onto one representative conformation. (D,F) Average residue insertion depth for the backbone was measured as the average vertical distance of the Cα atom to the bilayer center, and the insertion depth for the sidechain was measured as the average vertical distance of the sidechain heavy atoms to the bilayer center. All snapshots were rendered using VMD (Humphrey et al., 1996). Ramachandran plot heat maps are shown for the worm CPX-1 peptide (G) and mCpx1 peptide (H) using the same color scale. The torsion angle regions outlined in magenta correspond to either helical or extended conformations as indicated. (I) The fractional occupancy of helical (black) or extended (blue) states was computed based on the proportion of torsion angles located in a 20◦ × 20◦ region centered on (–62◦ ,–43◦ ) for a right-handed helix or (–55◦ , +150◦ ) for a beta strand.

motifs of both complexins entered the hydrophobic core of the lipid bilayer in accordance with the density profiles of the lipid head group atoms and water molecules as shown in **Figure 7B**. The structural organization of the peptides embedded in the lipid bilayer are indicated by the ensembles of peptide backbone conformations observed in the simulations for CPX-1 (**Figure 7C**) and mCpx1 (**Figure 7E**). To clarify the position of the side chains, the conformational ensemble cluster for each peptide is superimposed on a cluster representative for each complexin in the corresponding figure (**Figures 7C,E**). The

typical configurations adopted by the two complexin peptides were strikingly different. Worm CPX-1 dipped uniformly into the hydrophobic core via its phenylalanines anchored by lysines snorkeling back to the head group layer on either side (**Figure 7D**). In contrast, mCpx1 adopted a helical bend with L128 and M131 directed into the bilayer while Q129 and D130 were directed toward the aqueous phase (**Figure 7F**). This difference in configuration was also quantified by computing the distribution of peptide backbone dihedral angles for CPX-1 and mCpx1 as shown in **Figures 7G–I**. Notably, a point mutation in L128 (L128M) of the human mCpx1 ortholog has been identified in a patient with significant intellectual disability, severe seizures, myotonia, and conductive hearing loss (Redler et al., 2017). In summary, MD simulations suggest that the final eight residues of mammalian complexin adopt a more structured configuration that promotes membrane binding despite the relatively low hydrophobicity of this region compared to nematode complexin.

The dissimilar bound states of the worm and mouse C-terminal motifs can stabilize distinct orientations and positions relative to the upstream peptide as well as alter the availability of the C-terminal side chains for other protein interactions. Moreover, there may be functionally significant kinetic differences in the binding and unbinding of worm and mouse complexin (see Discussion). Such a kinetic difference is made more likely given the critical role played by the three phenylalanines of worm CPX-1. While the free energy of binding is clearly dependent on their special mode of insertion (substitution of the phenylalanines with isoleucines diminished the estimated free energy of binding by twofold, whereas alanine substitutions eliminated the free energy trough altogether – **Figure 7A**), it is possible that their coordinated withdrawal from the lipid membrane could slow the unbinding process.

#### NMR Spectroscopic Analysis of the C-terminal Domain

The structural differences between the C-terminal motifs of worm and mouse complexin in a membrane-like environment were further investigated with NMR spectroscopy. Worm CPX-1 and mCpx1 were incubated with dodecylphosphocholine (DPC) micelles (**Figure 8A**), a membrane mimetic that is amenable to solution-state NMR spectroscopy, and sequence-specific NMR backbone resonance assignments for both micelle-bound C-terminal motifs were obtained (Snead et al., 2017). Carbon chemical shifts for each C-terminal motif were then used to assess the degree of secondary structure in the micelle-bound state by calculating their deviation (secondary shift) from tabulated shifts characteristic of random coil behavior. In particular, positive carbon secondary shifts indicate alpha-helical structure (Wishart and Sykes, 1994). The six residues in the C-terminal motif of worm CPX-1 bound to DPC micelles in a configuration lacking any regular secondary structure, exhibiting small secondary shifts with no contiguous secondary shift patterns that might have suggested a transient helical structure (**Figure 8B**). In contrast, the C-terminal residues of mCpx1 exhibited a contiguous stretch of five positive carbon secondary shifts beginning with Proline 127 (**Figure 8C**), suggesting a significant population of helical

structure in the mouse C-terminal motif. Thus, the NMR data corroborate the MD simulations, supporting the conclusion that a segment of the mCpx1 C-terminal motif interacted with lipids in the form of an alpha helical turn. The MD and NMR spectroscopy results, together with the results from mutagenesis studies (**Figures 4**–**6**), reveal significant differences between the C-termini of two Cpx1 orthologs when bound to membrane, and these differences profoundly affect complexin function. However, as the C-terminal motif is not the only region of the CTD exhibiting high variability across phylogeny, other regions may also account for functional differences between complexin isoforms, as discussed below.

chemical shift consistent with alpha helix formation as illustrated (inset). Hydrophobic residues are indicated in blue while polar/charged residues are depicted in red. The CTD peptide consisted of the last 53 residues for CPX-1

and 64 residues for mCpx1.

and vertebrate Cpx4 homologs (vert 4, purple) occur in the 85–99th percentile, indicating relatively strong hydrophobic moments. (H) The amphipathic moment is plotted versus the overall hydrophobicity for each phylogenetic group as indicated. The mean amphipathic moment and hydrophobicity of random 18-mer peptides are shown in gray. (I) The average angle of the hydrophobic moment vector relative to the first residue in the amphipathic region is shown for each of the four groups. Error bars in (F–I) are standard deviation. See Supplementary Table S2 for a list of all species used in this analysis.

### Conserved and Divergent Features of the Complexin C-terminal Amphipathic Region

In addition to the C-terminal motifs explored above, upstream amphipathic regions of both worm CPX-1 and mCpx1 are known to adopt a helical conformation upon membrane binding and to confer selective binding to highly curved membranes such as SV membranes (Snead et al., 2014, 2017, co-submitted; Gong et al., 2016). In this region of the CTD, there is little or no primary sequence similarity between these two complexin orthologs. A systematic assessment of sequence and

secondary structure conservation suggested that the CTD is in fact conserved within a particular phylum, but is highly divergent between phyla. For example, nematode CPX-1 and vertebrate mCpx1 homologs reveal a strong degree of intraphylum conservation both at the primary sequence level and in the nature of the amphipathic sequence pattern (**Figures 9A,B**). To generate quantitative comparisons of both the strength and orientation of the amphipathic region, the amphipathic moment µ<sup>H</sup> of each sequence, modeled as an alpha helix, was computed by vector addition from the moments of individual residues (each rotated 100◦ relative to the preceding residue) weighted by the hydrophobicity of the side chain (using the Moon & Fleming metric for hydrophobicity – see Materials and Methods), and normalized to the number of residues (Eisenberg et al., 1982). By definition, the amphipathic moment points toward the hydrophobic interaction surface (e.g., lipid bilayer) and its magnitude provides a measure of the degree of asymmetry between the hydrophilic and hydrophobic sides of the helix. The CAAX-box containing complexins harbored a conserved amphipathic region with a similar µ<sup>H</sup> to other complexins (**Figures 9C–E**). Overall, the predicted helical tendency of the amphipathic region did not differ extensively between the representative phyla (**Figure 9F**). Compared to random peptides of identical length, the complexin amphipathic moment resided in the 85th99th percentile for all species analyzed (**Figure 9G**). Interestingly, the invertebrate complexins and the CAAX-box isoforms of the vertebrate complexins generally shared the same pattern of amphipathic residues and no net hydrophobicity (**Figure 9H**). In contrast, vertebrate Cpx1 (and Cpx2) exhibited both a higher amphipathic moment and a greater hydrophobicity distributed over a shorter stretch of residues (∼3 helical turns for vertebrate Cpx1/2 versus more than 5 turns for the other Cpx homologs). Moreover, the spatial orientation of the amphipathic moment of complexin 1/2 was markedly rotated when compared to the other homologs (**Figure 9I**). By these metrics, the invertebrate complexins and vertebrate Cpx3/4 isoforms were more similar to each other than to Cpx1/2. The broad conservation of the CTD amphipathic motif together with its functional importance in worm synapses suggests that this region generally plays an important role in proper complexin function. Is this function mechanistically conserved across phylogeny despite the differences described above?

#### Substitutions and Rotations in the Amphipathic Region

We explored the functional importance of the amphipathic region in worm CPX-1 first by introducing small perturbations to its structure and orientation as shown in **Figure 10A**. Either two or four consecutive alanines (+AA and +AAAA, respectively) were inserted just after the initial leucine of the amphipathic region (L117). The +AA insertion significantly rotated and diminished the µ<sup>H</sup> whereas the +AAAA insertion produced a more subtle change in µH. Despite their different effects on the amphipathic moment, both CPX-1 insertions equally failed to restore inhibitory function in the cpx-1 null mutant (**Figure 10B**). Thus, the amphipathic region was highly sensitive to alterations in the primary amino acid sequence. Furthermore, chimeric substitutions between worm and mCpx1 failed to restore complexin function despite attempts to match the CPX-1 amphipathic moment (**Figures 10C,D**). Interestingly, the mouse mCpx3 amphipathic region restored about 50% of CPX-1 functionality. This rescue was highly sensitive to the precise mCpx3 sequence as a two-residue shift in the substituted region destroyed both the amphipathic moment and the functionality of this chimeric variant even though the amino acid content and most of the sequence were identical in these two chimeras (**Figures 10C,D**). Thus, even relatively minor alterations in the amphipathic region abolished CPX-1 inhibition whereas similar amphipathic regions from highly divergent complexin isoforms were able to restore the CPX-1 inhibitory function, but only when precisely substituted in the correct orientation.

### DISCUSSION

It has proven difficult to deduce the structure/function relationship for complexin given the observed mixture of almost perfectly conserved regions and highly divergent domains. In the experiments described here, several features of complexin conservation were explored in the context of in vivo synaptic function using a simple behavioral assay as well as in vitro membrane binding assays. Five conclusions arise from these experiments. First, mouse complexin 1 fails to restore inhibitory function in worm synapses, and this is largely due to differences in the CTDs of mouse and worm complexin. Second, membrane-binding by several C-terminal residues is necessary but not sufficient for proper worm CPX-1 function. Third, the C-terminal motifs of worm and mouse complexins adopt distinct configurations, indicated by both MD simulations and NMR chemical shifts, perhaps explaining some of the failure of mCpx1 to function in worm CPX-1. Fourth, a deeply conserved amphipathic region is shared across both prenylated and non-prenylated complexins, with the vertebrate Cpx1/2 isoforms deviating from an otherwise characteristic pattern. Fifth, relatively subtle alterations in the amphipathic region can profoundly impact worm CPX-1 function, supporting the notion that the amphipathic region confers more than a generic membrane-binding capacity to complexin function. These results suggest that, in addition to membrane binding, there is another aspect of the complexin C-terminal region that has diverged across species.

### Conserved versus Divergent Regions of Complexin

Despite its small size, complexin displays a prominent heterogeneity in protein sequence conservation. The 25 residues comprising the CH are almost perfectly preserved both within phyla and between phyla. In effect, these residues define the complexin genes since the rest of the protein sequence has markedly diverged between phyla. To date, the only known binding partner of the CH is at the interface formed by the assembled synaptobrevin and syntaxin 1 SNARE helices in the

FIGURE 10 | Substitutions and rotations in the amphipathic region. (A) Helical wheel depictions of the 20-mer amphipathic regions in wild-type worm CPX-1 as well as variants with either two alanines (+AA, green) or four alanines (+AAAA, orange) inserted at the start of the region just after the initial Leu 117 (arrowhead). The additional alanines are in dark blue. The amphipathic moment magnitude (| µ| ) and angle (θ) are given for a 20-residue window beginning with L117 in the three variants. Note that two alanines invert the orientation of the amphipathic moment whereas four alanines approximately restore it. (B) Aldicarb time course for wild type (black open circles), cpx-1 mutant (red), and cpx-1 mutant rescued with full-length worm CPX-1 (black filled circles), or CPX-1 containing either two additional alanines (+AA, green) or four additional alanines (+AAAA, orange) following L117. (C) Chimeric amphipathic region substitutions were made in worm CPX-1 (black) using 14 residues from either mCpx1 (m1AR, orange), mCpx3 (m3AR, blue) or mCpx3 shifted two residues toward its N-terminus (m3AR –2, green) to perturb its amphipathic moment. For each sequence, the amphipathic moment was computed between L117 and the last lysine as indicated. The angle is measured relative to L117 as above. (D) Aldicarb time course for wild type (black open circles), cpx-1 mutant (red), and cpx-1 mutant rescued with CPX-1 containing either the mCpx1 amphipathic region (m1AR, orange), mCpx3 amphipathic region (m3AR, blue), or mCpx3 shifted by two residues (m3AR –2, green) as described in (C). Data are mean ± SEM.

ternary SNARE complex (Pabst et al., 2000; Bracher et al., 2002; Chen et al., 2002). Thus, the CH sequence is highly constrained by the equally conserved SNARE sequence (Kloepper et al., 2007). Examining secondary structure rather than primary sequence, conservation is more evident throughout the complexin protein. A stable alpha helix characterizes the AH domain across

phylogeny (Radoff et al., 2014), while the lipid-binding CTD displays highly charged stretches, amphipathic helices, and hydrophobic regions in stereotyped locations (Snead et al., 2014, 2017, co-submitted). The AH conservation is both structural and functional since the mouse domain was fully operational in worm synapses when substituted into worm CPX-1 (Radoff et al., 2014). For the CTD, functional conservation is less clear. A previous study tested the impact of swapping mouse Cpx1 and fly Cpx (DmCpx) domains including the CTD (Xue et al., 2009). Introducing the fly C-terminus onto mCpx1 enhanced mCpx1 suppression of spontaneous neurotransmitter release in hippocampal autaptic cultures, consistent with the notion that invertebrate CTDs endow Cpx with a potent inhibitory function (Xue et al., 2009). Of note, five cases of homozygous mutations in the human Cpx1 ortholog CPLX1 have been reported, and all involve either truncations that effective delete the CTD or a point mutation near the end of the CTD (Karaca et al., 2015; Redler et al., 2017). These mutations are associated with severe epilepsy as well as intellectual disability.

Across the animal kingdom, the Cpx superfamily can be divided into prenylated versus non-prenylated Cpx isoforms (Reim et al., 2005, 2009; Brose, 2008; Yang et al., 2015). While the Drosophila genome harbors only a single complexin gene, alternative splicing produces both prenylated and non-prenylated DmCpx variants (Buhl et al., 2013; Cho et al., 2015). The sequence analysis and structural comparisons presented here suggest that both prenylated and non-prenylated complexins share attributes in the CTD across phylogeny with the most divergence arising from vertebrate Cpx1/2 isoforms. Of particular interest, mammalian and fish Cpx3/4 isoforms, which are prenylated and limited in expression to specialized nervous tissue such as the retina, show functional similarity to the invertebrate complexins in their suppression of neurotransmitter release (Vaithianathan et al., 2013, 2015; Mortensen et al., 2016). Interestingly, a recent study utilized a distantly related cnidarian Cpx isoform from Nematostella vectensis (NvCpx1) in mammalian synapses and concluded that this non-bilaterian lineage Cpx lacked inhibitory activity while still facilitating calcium-triggered fusion (Yang et al., 2015). The Cpx CTDs of Nematostella and other basal animals (such as Trichoplax) lack an obvious amphipathic helix motif, suggesting that this aspect of CTD function might not be conserved outside of bilateria. The experiments presented here are consistent with these observations and indicate a correlation between Cpx inhibitory function and amphipathic properties of the CTD. We propose that the ancestral bilaterian Cpx homolog mediated an inhibitory function at the synapse via its CTD and that this function was attenuated in chordates for Cpx1/2 homologs. As these complexins subserve the central and peripheral nervous systems in vertebrates, loss of a constitutive inhibitory function in Cpx may have coincided with other specializations of the vertebrate nervous system relevant to the physiology of vertebrate chemical synapses. This raises the question of whether the inhibitory role of Cpx has been eliminated entirely in vertebrate Cpx1/2 isoforms. The inhibitory role of Cpx has not been entirely lost in the vertebrate central nervous system since inhibitory function of mammalian Cpx1/2 isoforms has been observed in hippocampal neurons and at the calyx of Held (Maximov et al., 2009; Kaeser-Woo et al., 2012; Chang et al., 2015). In addition, expression of mCpx1 in fly synapses lacking endogenous DmCpx rescued Cpx inhibitory activity to a large degree, indicating that mCpx1 retains some inhibitory activity that can operate in a distantly related synapse (Cho et al., 2010). Regardless of the evolutionary pressures that reshaped Cpx1/2 function in vertebrates, it is clear that alterations in C-terminal structure and function have a profound impact on the regulation of synaptic transmission, and a better understanding of this mechanism will shed light on the molecular control of SV fusion at all synapses.

### Membrane-Binding Properties of Complexins

Previous studies have established that both nematode and mammalian Cpx1 isoforms bind to either flat or curved membranes, but with a preference for relatively high curvature. This curvature-sensitive binding is accompanied by a transition of the C-terminal amphipathic region from an unstructured configuration to an alpha helix, with some similarity to a protein motif known as an ALPS domain (Antonny, 2011; Snead et al., 2014). Moreover, this conformational switch from disordered to helical is required for efficient CPX-1 inhibitory function in worm synapses (Snead et al., 2014). The last few residues of nematode complexin contain a hydrophobic stretch that binds to membranes irrespective of curvature, and this interaction is also required for CPX-1 inhibition. In contrast, the corresponding residues in mCpx1 lack the same degree of hydrophobicity but as shown here, appear to adopt an amphipathic helical turn that mediates comparable membrane binding. Interestingly, substituting the mCpx1 motif into worm CPX-1 lacking its final hydrophobic stretch of residues failed to restore CPX-1 inhibitory function in vivo even though this chimeric protein bound to SUVs with a similar affinity to CPX-1 as measured by tryptophan fluorescence. Based on our findings described here, there are several possible explanations for this failure to function.

First, although equilibrium binding does not appreciably differ between these two complexin isoforms, the binding kinetics may differ significantly. Little is currently known about the details of complexin binding kinetics (Gong et al., 2016), and yet the rate at which complexin binds or unbinds vesicle membrane may have functional consequences at the synapse. Indeed, a major inhibitory effect of mammalian complexin appears to be limited to a small time window during the priming process (Chang et al., 2015).

Second, the tryptophan fluorescence measurements only probe membrane interactions in pure lipid membranes whereas complexin interactions in vivo involve SVs packed with both integral membrane proteins and membrane-associated proteins (Takamori et al., 2006). It is possible that divergence of other protein interactions required for in vivo function accounts for the failure of mammalian Cpx C-terminal motif to replace the nematode motif in vivo. For instance, the mCpx1 C-terminus

central helix of complexin binds the assembling SNARE bundle to prevent full assembly and spontaneous fusion (3). (B) Complexins from divergent species possess similar lipid- and SNARE-binding properties but other interactions depending on the C-terminal domain may not be well conserved. Potential differences in C-terminal binding partners (a) or kinetics of C-terminal membrane binding (b) may account for functional differences across phylogeny. SV, synaptic vesicle; PM, plasma membrane.

presented an electrostatic interaction surface mediated by Q129 and D130 directed out of the lipid bilayer in MD simulations that was not observed with the worm CPX-1 C-terminus, possibly capable of binding other synaptic proteins.

Third, the C-terminal tryptophan only reports the local membrane-binding behavior of the peptide. Other aspects of the membrane interaction within the amphipathic region or even the NTD may differ in this chimera without altering the C-terminal tryptophan fluorescence (Lai et al., 2016). The MD simulations suggested that the relative position of the C-terminal motif compared to the amphipathic region may differ significantly between worm and mouse complexin. In particular, side chains located six and seven residues upstream of the C-terminus (K136 G137) resided close to the membrane surface, whereas the corresponding residues of the mouse CT motif (P127 L128) were buried nearly 10 angstroms into the membrane. This difference in membrane penetration may alter the position of the neighboring amphipathic helix. However, since even when the entire CTD region was swapped between mouse and worm, the mouse version failed to rescue synaptic inhibition, a disruption of the coupling between the amphipathic region and CT motif cannot fully explain the species differences.

Finally, C-terminal prenylation is a common feature of many complexin isoforms, and its biological role is currently not well understood beyond a synaptic targeting function (Reim et al., 2005; Buhl et al., 2013; Iyer et al., 2013). Disruption of prenylation impairs Cpx inhibitory function, but it is still unclear what membrane-binding characteristics are endowed by prenylation in any complexin isoform (Reim et al., 2005, 2009; Cho et al., 2010; Buhl et al., 2013; Iyer et al., 2013).

## Model of Complexin Inhibitory Action at the Synapse

A number of previous studies have contributed to the growing knowledge of complexin membrane interactions and their role in complexin function (Seiler et al., 2009; Kaeser-Woo et al., 2012; Wragg et al., 2013, 2015; Snead et al., 2014; Gong et al., 2016).

We previously showed that two membrane-binding modules within the CTD of complexin direct complexin to SVs. Moreover, upon binding to highly curved membranes, the amphipathic region adopts a functionally important alphahelical structure (Snead et al., 2014). We propose that this helical conformation interacts with other proteins that help complexin to engage the assembling SNARE complex as vesicles dock and prime (**Figure 11A**). Based on observations described here and in other studies, we conclude that there are structural differences between the amphipathic and C-terminal motifs of worm CPX-1 and mouse mCpx1 (Snead et al., 2014, 2017). We further speculate that these differences are associated with non-conserved protein–protein interactions, so that the differences in conformational and sequence requirements for these interactions explain the failure of mCpx1 to restore proper inhibitory function in worm synapses (**Figure 11Ba**). Another possibility is that differences in the kinetics of membrane binding and unbinding by the CTD that result from the conformational and sequence differences may lead to the observed functional consequences (**Figure 11Bb**). A recent study revealed a potent but transient inhibitory function for mCpx1 during vesicle priming (Chang et al., 2015). It is possible that, in vertebrate

synapses, complexin inhibition is required only during the priming process, whereas a longer term association of complexin and the fusion machinery is utilized in invertebrate synapses to minimize spontaneous fusion. The CTD provides a means of controlling the inhibitory strength of complexin, and the phylogenetic diversity of C-terminal sequences may reflect evolutionary divergence of synaptic regulation. It is noteworthy that in addition to its inhibitory activity, complexin has a separate positive role in calcium-triggered fusion. This function appears to be universally shared among all complexin isoforms, although a detailed understanding of this facilitatory mechanism is currently lacking (Xue et al., 2009; Cho et al., 2010; Martin et al., 2011; Yang et al., 2015; Trimbuch and Rosenmund, 2016).

#### Controversies Regarding Complexin Function

Beginning with the first reports of complexin inhibitory activity in invertebrates, a stark contrast has persisted between mammalian and invertebrate complexin function at the synapse. Studies in numerous model synapses over the past decade have produced some consensus on the positive role of complexin in calcium-triggered neurotransmitter release as well as on the absolute dependence of this function on the SNARE-binding CH domain (Xue et al., 2007; Cho et al., 2010; Martin et al., 2011; Trimbuch and Rosenmund, 2016). However, the inhibitory function of complexin varies across synapses and across species, perhaps reflecting diverse demands on synaptic function in these different contexts. The work presented here emphasizes the contribution of two membrane-binding modules within the CTD of complexin as major drivers of this functional diversity. Nevertheless, the mechanistic basis of both the facilitatory and inhibitory roles of complexin remains poorly understood. Future studies will be required to place the diverse functions of the CTD into a mechanistic picture of complexin action, and comparisons across synapses of different species will aid and enrich our understanding of this fascinating regulatory protein.

#### AUTHOR CONTRIBUTIONS

Conceptualization: RW, DE, and JD. Methodology: RW, IB, DP, DS, MT, DE, HW, and JD. Investigation: RW, DP, MT, DS, ZL, IB, and JD. Writing: RW, HW, DE, and JD. Funding acquisition: DP, HW, DE, and JD.

### REFERENCES

Abrams, J. B., and Tuckerman, M. E. (2008). Efficient and direct generation of multidimensional free energy surfaces via adiabatic dynamics without coordinate transformations. J. Phys. Chem. B 112, 15742–15757. doi: 10.1021/ jp805039u

#### FUNDING

This work was supported by the National Institutes of Health grants R01-GM095674 (JD), R37-AG019391 (DE), and R01- GM117518 (DE), P01-DA012408 (HW), the Membrane Protein Consortium U54-GM087510, NIH/NIMH F30MH101982 (DS), MSTPGM07739 (DS), the Rita Allen Foundation Award 188423 (JD), and a DFG Fellowship PA2679/2-1 (DP). Computational results utilized in this work were carried out at resources of the Oak Ridge Leadership Computing Facility (ALCC allocation BIP109) at the Oak Ridge National Laboratory, which is supported by the Office of Science of the U.S. Department of Energy under Contract No. DE-AC05-00OR22725; an XSEDE allocation at the Texas Advanced Computing Center at the University of Texas at Austin (Stampede supercomputer, projects TG-MCB090132, and TG-MCB120008), and the computational resources of the David A. Cofrin Center for Biomedical Information in the Institute for Computational Biomedicine at Weill Cornell Medical College. DE is a member of the New York Structural Biology Center. Data collected at NYSBC were made possible by a NYSTAR and ORIP/NIH facility improvement grant CO6RR015495. The 900 MHz spectrometers were funded by NIH grant P41GM066354, the Keck Foundation, New York State Assembly, and the US Department of Defense.

#### ACKNOWLEDGMENTS

We thank Tim Ryan, Cindy Liang, and Murugesh Padmanarayan for help, advice, and for critically reading the manuscript. We also thank Cori Bargmann and Navin Pokala for the HisCl plasmid. We are grateful to Michel A. Cuendet for significant help with the united free energy dynamics (UFED) calculations.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol. 2017.00146/full#supplementary-material

TABLE S1 | Summary of axonal protein expression data for various

transgenic animals. Quantification of axonal CPX-1::GFP expression for several transgenic animals used in this study. Details of the imaging and quantification are described in the Methods. All average expression values were normalized to the full-length rescuing CPX-1::GFP transgenic (tauIs90), which is a multi-copy integrated array. In addition to the protein abundance for each transgenic, its qualitative ability to restore CPX-1 inhibitory function is also indicated (rescue quality). Importantly, the ability of CPX-1 variants to rescue the cpx-1 null mutant generally did not correlate with their protein expression levels.




Zhu, Z., Tuckerman, M. E., Samuelson, S. O., and Martyna, G. J. (2002). Using novel variable transformations to enhance conformational sampling in molecular dynamics. Phys. Rev. Lett. 88:100201. doi: 10.1103/PhysRevLett.88. 100201

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Wragg, Parisotto, Li, Terakawa, Snead, Basu, Weinstein, Eliezer and Dittman. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Membrane Fusion Involved in Neurotransmission: Glimpse from Electron Microscope and Molecular Simulation

Zhiwei Yang1,2,3 , Lu Gou<sup>1</sup> , Shuyu Chen<sup>1</sup> , Na Li <sup>1</sup> , Shengli Zhang<sup>1</sup> \* and Lei Zhang<sup>1</sup> \*

<sup>1</sup>Department of Applied Physics, Xi'an Jiaotong University, Xi'an, China, <sup>2</sup>Department of Applied Chemistry, Xi'an Jiaotong University, Xi'an, China, <sup>3</sup>School of Life Science and Technology, Xi'an Jiaotong University, Xi'an, China

Membrane fusion is one of the most fundamental physiological processes in eukaryotes for triggering the fusion of lipid and content, as well as the neurotransmission. However, the architecture features of neurotransmitter release machinery and interdependent mechanism of synaptic membrane fusion have not been extensively studied. This review article expounds the neuronal membrane fusion processes, discusses the fundamental steps in all fusion reactions (membrane aggregation, membrane association, lipid rearrangement and lipid and content mixing) and the probable mechanism coupling to the delivery of neurotransmitters. Subsequently, this work summarizes the research on the fusion process in synaptic transmission, using electron microscopy (EM) and molecular simulation approaches. Finally, we propose the future outlook for more exciting applications of membrane fusion involved in synaptic transmission, with the aid of stochastic optical reconstruction microscopy (STORM), cryo-EM (cryo-EM), and molecular simulations.

Edited by:

Jiajie Diao, University of Cincinnati, United States

#### Reviewed by:

Dechang Li, Beijing Institute of Technology, China Rui Su, University of Cincinnati, United States Xiaochu Lou, University of Wisconsin-Madison, United States

#### \*Correspondence:

Shengli Zhang zhangsl@mail.xjtu.edu.cn Lei Zhang zhangleio@mail.xjtu.edu.cn

Received: 21 March 2017 Accepted: 15 May 2017 Published: 07 June 2017

#### Citation:

Yang Z, Gou L, Chen S, Li N, Zhang S and Zhang L (2017) Membrane Fusion Involved in Neurotransmission: Glimpse from Electron Microscope and Molecular Simulation. Front. Mol. Neurosci. 10:168. doi: 10.3389/fnmol.2017.00168 Keywords: membrane fusion, neurotransmission, neurotransmitter release machinery, electron microscope, molecular simulation

### INTRODUCTION

Neurotransmission is composed of the delivery of neurotransmitters from presynaptic neuron to another neuron, and the feedback of postsynaptic neuron (Jahn and Scheller, 2006; Burnstock, 2007). It is a chemical event which is involved in the transmission of the impulse (Xu et al., 2017), and relies on: the availability of the neurotransmitter; the neurotransmitter release (exocytosis); the binding of the neurotransmitter to the postsynaptic receptor, the excitatoryinhibitory interaction in the postsynaptic cell (Bonifacino and Glick, 2004); and the subsequent removing or deactivating of the neurotransmitter (Iversen, 1971; Heuser and Reese, 1973). Hence, this process requires the controlled release of neurotransmitter from synaptic vesicles by membrane fusion with the presynaptic plasma membrane (Martens and Mcmahon, 2008). Soluble N-ethylmaleimidesensitive factor attachment protein receptors (SNAREs) are the core constituents of the protein machinery which is responsible for synaptic membrane fusion (Jahn and Scheller, 2006). In general, the SNAREs-mediated fusion event is thought to involve a hemifusion diaphragm between the fusion talk and the fusion pore (hemifusion intermediate, **Figure 1**), where the outer lipid bilayers have been fused, whereas not the inner ones (Zimmerberg, 1987; Brunger et al., 2015). Beyond that, there exists a direct pathway where pre-fusion contact translates into fusion pore without the hemifusion state (Gerst, 1999; Xu et al., 2005; Brunger et al., 2015).

An integral part in the SNAREs-mediated fusion is the assembly of synaptic vesicle transmembrane protein (synaptobrevin) with the target plasma membrane proteins synaptosome-associated protein with relative molecular mass 25 K (SnAP25) and syntaxin (Bommert et al., 1993), which is thought to provide the driving force for the fusion (Haucke et al., 2011). During this process, the extended α-helices of these proteins trend to assemble together, with the formation of four-helix bundles (Mehta et al., 1996). The formed trans-SNARE complex then facilitates the close proximity of vesicular and plasma membranes (∼3–4 nm), and induces the membrane fusion (Martens and Mcmahon, 2008).

The spatial regulation of membrane shape, curvature and fluidity are strongly influenced by the lipid composition and topology, during the processes of membrane fusion and fission. Effective neurotransmission requires the precise spatial regulation of lipid-protein interactions for synaptic vesicle targeting, docking, priming and fusion at the active zone (Rohrbough and Broadie, 2005). For an in-depth understanding of the neuronal membrane fusion, various points and states will be summarized in this article, including the process and mechanism of membrane fusion, electron microscope (EM) approaches and molecular simulation results.

#### PROCESS AND MECHANISM OF MEMBRANE FUSION

Membrane fusion is identified as a process where two separate phospholipid bilayers merge into an interconnected structure. It is a fundamental physiological and pathological process at the level of cell, organelle and vesicle, resulting in the mixing of the two bilayers of lipids and proteins, as well as the mixing of the contents (Jahn et al., 2003).

### Process of Membrane Fusion

Despite derived by diverse proteins, all fusion reactions processes four fundamental steps (Jahn and Südhof, 1999): membrane aggregation (approaching each other), membrane association (coming into a very close apposition), lipid rearrangement (highly-localized lipid rearrangements of adjacent two bilayers) and lipid and content mixing (complete fusion; **Figure 1**; Wilschut and Hoekstra, 1986; Plattner et al., 1992; Blijleven et al., 2016).

Hemifusion of lipid bilayers is an important intermediate state in the membrane fusion, which might be boosted by negative spontaneous curvature of monolayer (monolayer trends to bulge toward the hydrophobic tails) and deformation of monolayer induced by the distortion of lipid monolayer (inclusion of amphiphilic peptides; **Figure 1**; Chernomordik et al., 2006). In addition, bilayers hemifuse when brought to distances of the polar heads of lipids much smaller than the one in the equilibrium state, by adding polyethylene glycol to draw water from the contact zone or by a direct dehydration of multileveled lipid sample. To complete the fusion process, the hemifusion state should proceed to a full fusion pore (Geisow and Fisher, 1986). The pore might open directly from a fusion stalk (stalk-pore pathway) or from a hemifusion state with discernible hemifusion intermediates (hemifusionfusion pathway; Chernomordik et al., 2006). The formation and closure of the fusion pore are usually regulated by the conformational change with a high activation energy and phase separated lipids, respectively (Oberhauser et al., 1992). The hemifusion diaphragm is a possible intermediate between the stalk and the final fusion pore (Chernomordik et al., 2006). However, Diao et al. (2012) observed more fast fusion (on the ms scale) upon Ca2<sup>+</sup> addition starting from a hemifusion-free state. This discovery revealed that the neurotransmitter release (especially the fastest event) is more dependent on the immediate pathway, and then stimulated a substantially revised membrane fusion paradigm for the membrane fusion (Wickner and Rizo, 2017).

#### Fusion of Protein-Free Lipid Bilayers

The major constituent of the most bilayer lipid, especially the membranes of mammalian cells, is phosphatidylcholine (PC) which has no spontaneous fusion for hours or days (Wang, 2010). Only the tension (or dehydration contact zones) of these bilayers could fuse by the interposition of polyethylene glycol (Sharma and Lindau, 2016). A monolayer protruding into the layer of polar heads seems to consist of molecules with a reasonable inverted cone (Epand, 1998) and positive spontaneous curvature (Chernomordik et al., 2006). A lipid monolayer that orients toward the hydrocarbon tails usually possesses a negative curvature, composed by cone shaped lipid molecules (Chizmadzhev, 2004). A possible explain is that the action of proteins in fusion process do not directly facilitate the formation of fusion intermediates, also generate fusogenic lipids (sphingomyelinase and phospholipase; Epand, 2000; Chernomordik and Kozlov, 2008). It is promoted by the defects created of the bilayers referring membrane perturbation, including the vicinity of lipid phase transition, the separation of lateral phase or the generation of domain, the high local curvature of membrane, osmotic or electric stress in or on the membrane; the present of amphipaths or macromolecules within the membrane, etc. (Cevc and Richardsen, 1999). High concentrations of lysolipids, which increase the intrinsic curvature of the monolayer, inhibit several biological membrane fusion processes (Shangguan et al., 1996; Söllner, 2004).

### Protein-Mediated Membrane Fusion

The context of membrane fusion in vivo is more complicated since biological fusion is always mediated by protein. However, the specific mechanisms of these processes still remain elusive, especially on proteins which promote the development of hemifusion and fusion pore. Viral membrane fusion and synaptic membrane fusion are widely studied among current

protein-mediated membrane fusions. The former promotes the combine between the viral membrane and the host cell membrane, then inducing the release of viral genome into the cytoplasm, as well as the replication cycle of virus. There are two mechanisms of activating viral fusion proteins: exposure to low pH and pH-independent (Earp et al., 2005).

The latter carries the neurotransmitter across the synapses (neurotransmitter release) and plays an important role in the signals traveling in the central nervous system. Neurotransmitter release during the synaptic membrane fusion process requires a protein family that have termed SNAREs which can be divided into four categories (Diao et al., 2012; Hughson, 2013): (1) vesicle-anchored (v) and target-membrane–anchored (t) SNAREs; (2) N-ethylmaleimide–sensitive factor (NSF) and NSF attachment proteins (SNAPs); (3) Rab GTPases and multicomponent vesicle tethering complexes; and (4) Sec1/Munc18 (SM) proteins. So far, the mechanism of this family still has a few basic doubts, such as function, conformational changes, etc. The SNARE-mediated membrane fusion was conducted the zippering mechanism which pulls two membranes together (Diao et al., 2013b). In general, synaptic membrane fusion requires a consecutive two-step pathway. First, the N-terminal domain of the vesicle (v-) SNARE, synaptobrevin-2, docks to the target membrane (t-) SNARE, thereby results in a conformational rearrangement of a half-zippered SNARE complex. Then, the assembled SNARE complex locks the C-terminal portion of the t-SNARE into the same way as the four-helix bundle, which is formed with syntaxin and SNAP-25. Besides, this fusion is greatly accelerated by synaptotagmin-Ca2<sup>+</sup> (Lai et al., 2013). Reconstitutions of synaptic vesicle fusion indicated that the interactions between the Ca2+-binding loops of the synaptotagmin-1 and phospholipids are critical to release of neurotransmission, when little content mixing occurs in the absence of Ca2<sup>+</sup> (Diao et al., 2013b; Wickner and Rizo, 2017).

#### ELECTRON MICROSCOPE ON MEMBRANE FUSION

In general, neuronal communication is mediated by neurotransmitters release induced by Ca2+-induced synaptic vesicle exocytosis. It is a long-sought goal that understanding the mechanism of synaptic vesicle fusion, associated with the development of in vivo synthetic system. With the advent of modern electron microscopic techniques, we could particularly investigate the neurotransmission and the consequent membrane alterations.

During neurotransmitter release, several SNARE protein complexes involve with synaptobrevin (vesicle (v-) SNAREs) and syntaxin and SNAP-25 (target membrane (t-) SNAREs) to mediate the fusion of two membranes. Meanwhile, this vesicle membrane fusion is acutely triggered in a Ca2+-dependent manner (**Figure 2**). Munc18 (also called neuronal Sec1) forms a tight complex with syntaxin, with a closed conformation that is unable to bind other SNAREs. Regarding as Munc13, a synaptic vesicle ''priming'' protein, catalyzes the transition from syntaxin-Munc18 complex to fully assembled v/t-SNARE complex (syntaxin—SNAP-25—synaptobrevin), via bridging the vesicle and plasma membranes and controlling vesicle tethering (Hughson, 2013; Ma et al., 2013; Wickner and Rizo, 2017). Chen et al. (2002) explored the atomic structure of the complexin/SNARE complex, using the X-ray and TROSY-based NMR methods. The results revealed that complexin presents an antiparallel helical conformation, stabilizes the interface between two helices of synaptobrevin and syntaxin, thus enables the Ca2+ evoked neurotransmitter release with the exquisitely high speed.

Over the past 10 years, cryo-electron microscopy (cryo-EM) has been developing rapidly, which combines the potential of three-dimensional (3D) imaging at molecular resolution with a close-to-life preservation of biological samples. Rapid freezing followed by the investigation of the frozen-hydrated samples avoids the artifacts caused by chemical fixation and dehydration procedures. Furthermore, the biological material is observed directly, without heavy metal staining, avoiding artifacts caused by the unpredictable accumulation of staining material (Luci´c et al., 2005). The vesicle clusters induced by Ca2+-bound C<sup>2</sup> domains of synaptptagmin-1 was visualized by the Cryo-EM method, and the tomographic 3D reconstruction of a vesicle cluster revealed that this process might be induced by Ca2+ dependent phospholipid binding of the C2AB fragment, where the C2B domain cooperates with the SNAREs bring the membranes together, as well as the multiple interactions between the C2B domain and phospholipids (diacyglycerol and phosphatidylinositol 4,5-bisphosphate (PIP2); Araç et al., 2006).

In fact, there is still an active debate regarding whether SNAREs are linked to pre-fusion contact to a fusion pore or participate later in the fusion process by facilitating hemifusion, through the formation of tight SNARE complexes and gathering of the vesicle and plasma membranes. With the aid of single-vesicle fluorescence fusion assay and EPR, the direct observation of two-faceted functions of complexin revealed the formation of a complex substrate (SNARE complexes, complexins and phospholipids) for Ca2<sup>+</sup> and Ca2+-sensing fusion effectors in the release process of neurotransmitter (Yoon et al., 2008). Neuron firing gives rising of the intracellular Ca2<sup>+</sup> concentration, with the triggering of synaptic vesicles fusion to carry neurotransmitter molecules. Diao et al. (2012) used recently developed Cyro-EM method to monitor the temporal sequence of both content and lipid exchange upon Ca2+-triggering between single pairs of donor and acceptor vesicles on a 100-ms time scale (Diao et al., 2012). Their system performed a quantitative analysis of all observed cryo-EM images (before and after Ca2+-injection) and achieved a Ca2<sup>+</sup> sensitivity in the 250–500 µM range (**Figure 2**). During their experiments, hemifusion diaphragms were observed, as well as points where liposomes contacted each other without the shape change of membrane. Extended tight contacts of membrane were not observed, without the present of Ca2+. With the addition of Ca2+, there merely exists the process from point-contacts to fast fusion (Diao et al., 2012). It was found that alone neuronal SNAREs cannot efficiently induce the complete fusion. The combination of SNAREs with selected components (small-head group lipids, Munc18-1, Munc 13 and synaptotagmin-Ca2+) could lower the activation barriers during the fusion process, because of enhancing the kinetic control by complexin (Kyoung et al., 2011; Wickner and Rizo, 2017). Bharat et al. (2014) performed reconstitutes synaptic fusion and applied large-scale, automated cryo-electron tomography to observe this in vitro system. Afterwards docking and priming of vesicles with the fast Ca2+-triggered fusion, a local protrusion in the plasma membrane will be induced by the SNARE proteins, with the direction towards the primed vesicle and allowing synchronous and instantaneous fusion upon the complexin clamp release (Zhang et al., 2015b).

#### MOLECULAR SIMULATION ON MEMBRANE FUSION

Many efforts have been devoted to modeling the membrane fusion process involved in synaptic transmission via molecular simulations, such as Coarse-grained (CG) molecular dynamics (MD) simulations. The results of these studies revealed the mechanics of membrane fusion involved in synaptic transmission and some key physical properties of lipid monolayers and related proteins.

The SNARE complex between opposing membranes promotes membrane fusion of synaptic transmission (Mayer, 1999; Pfeffer, 1999). In vivo, the formation of complex connects the opposing membranes and pulls two membranes together using their α-helical transmembrane domains (TMD; Ossig et al., 2000). In addition, SNARE complexes are also deemed to overcome the fusion barriers and to accelerate the fusion process (Chen and Scheller, 2001; Hong, 2005; Risselada and Grubmüeller, 2012).

The SNARE complex is represented by a twisted bundle of four α-helices which generally consists of SNAP-25, syntaxin-1, and synaptobrevin-2, confirmed by coarse-grain MD simulations (Durrieu et al., 2009; Tekpinar and Zheng, 2014) There is mechanistically link between the conformational flexibility of SNARE TMD helices and their ability to induce lipid mixing (Nagy et al., 2005; Stelzer et al., 2008). The basic residues (positive charged) at the C terminal of SNAP-25 is benefit for the tight zippering of SNARE complex and the binding with negatively charged lipid head groups, improving the high frequency and clipping neurotransmitter release (Fang et al., 2015). Besides, the transmembrane domain of synaptobrevin II (sybII TMD) influences both the natural helicity and flexibility of SNARE (Gao et al., 2012; Zheng, 2014; Han et al., 2016). The assemblage of SNARE complex is regulated by complexin, a cytoplasmic neuronal protein and the MD results suggest that the α-accessory helix of complexin (Cpx AH) make partially unzipped state of the SNARE bundle being stabilized by its functions in relate in the clamping of synaptic vesicle fusion (Ghahremanpour et al., 2010; Bykhovskaia et al., 2013; Lai et al., 2014, 2016; Gong et al., 2016).

The composition of several parts of the SNARE completes its function of modulate membrane fusion (**Figure 3**). MD simulations have also been used in the study of other functions in relation with membrane fusion. During the synaptic transmission, vesicles filled with neurotransmitter molecules are required to be docked to the membrane (Knecht and Grubmüller, 2003; Bock et al., 2010; Lai et al., 2015). Therefore, the SNARE complex use attractive forces to counterbalance the long-range repulsion between the vesicle and membrane (Diao et al., 2013a; Fortoul et al., 2015). More recently, the formation of a transient pore by using the MD method was first reported. The close contact of two membranes gives rise to a high local transmembrane voltage. The decrease of the distance of the

opposed bilayers brings out the increase of the transmembrane voltage. When the distance is under a critical value, the local transmembrane voltage is enough high to induce the transient of membrane pores (**Figure 4**; Ribrault et al., 2011; Bu et al., 2016). Finally, some findings have offered new structural and dynamic details of SNARE disassembly mechanism based on CG modeling (Zheng, 2016).

#### OUTLOOK

The controlled release of neurotransmitter by membrane fusion, from synaptic vesicles to presynaptic cell, is an important step in the synaptic transmission (Trimbuch and Rosenmund, 2016). This universal fusion can be accelerated by synaptotagmin-Ca2+, with the assembly of specialized proteins (such as SNAREs) within the opposing membrane bilayers. Electron density map, 3-D topography and simulation studies of the SNARE ring complex, advance that membrane-associated SNAREs overcome repulsive forces to process the two membranes being close to each other (just 2.8 Å apart; Chen and Scheller, 2001). However, people are actually working on an assumption that all proteins are at the right position for inducing membrane fusion. Since all proteins are unlabeled, one is not able to tell the real story on the protein side, which is also an important part for the study of membrane fusion in synaptic transmission. Thus, in the future, it requires the high spatial resolution techniques in order to monitor these interactions simultaneously during the fusion processes, such as stochastic optical reconstruction microscopy (STORM), is required (Diao et al., 2011). In STORM, single biomolecules containing photo-switchable fluorophores are turned on and off repeatedly, to find their positions precisely

molecular dynamics (MD) simulations (A–C): the process of fusion pore formation. (D–F): electric potential alteration during the process (Bu et al., 2016).

with ∼20 nm resolution by determining the center position of the point-spread function from reconstructed images for each biomolecule (Rust et al., 2006).

Cryo-electron microscopy, abbreviated as ''cryo-EM'', is a form of transmission EM (TEM) technique which observes the sample (generally biological sample) at cryogenic temperatures in order to void the ultrastructural changes (Doerr, 2016). It has been rapidly developed in the decade, with increasing popularity in structural biology (Callaway, 2015; Nogales et al., 2016). As cryo-EM became matured, it has been adopted by an ever-increasing range of disciplines to offer tools for providing a cell-like yet simplified environment for investigating the membrane fusion, especially the dynamic structural change of important proteins and the dynamic mechanism of fusion process. In particular, optimized negative-staining (OpNS) EM images have revealed several important physical attributes of CETP and substantial molecular basis for the CETP-mediated lipid exchange (Zhang et al., 2012, 2015a).

Active zones of synaptic plasma membranes are known to concentrate the components that drive membrane fusion, such as the SNAREs, Munc 18-1, Munc 13-1 and small-head group lipids (e.g., diacyglycerol and PIP2), while the participation and characteristic of these macromolecular complexes are still not fully understood (Rizo and Xu, 2015; Ryu et al., 2016; Wickner and Rizo, 2017). Strikingly, current experimental techniques do not achieve a resolution better than ms/µs in time, and neuronal membrane fusion normally occurs at ms timescale (Diao et al., 2012). CG MD simulation is one of effective solutions to overcome the time-scale gap between computational and experimental methods (Bhushan, 2016). With the aid of residue-based and shape-based CG approaches, the regulatory mechanisms of SNARE proteins have been briefly outlined, focused on the intricate molecular mechanisms between proteins and membranes (Bu et al., 2016; Zhang et al., 2016; Han et al., 2017). However, the application of CG models sacrifice degrees of freedom and accurate molecular interactions to get the requirement of less resources (Bhushan, 2016). Though the prohibitive computational cost usually limits the simulation times and system sizes of all-atom models

#### REFERENCES


less than 1000 ns and 10 nm, it will provide the description of SNARE-mediated membrane fusions with all-atom detail, such as the specific lipid properties for stimulating fusion, the tethering/SM protein complex, the lipid-protein (such as Munc18-1, Munc13-1, and complexin) interactions, and membrane architecture. Nevertheless, molecular simulation selectivity leads to the factitious results of synaptic membrane fusion, therefore, the computational methods and initial models should be amending continuously by the sufficient basic parameters derived from the non-invasive experimental observations (such as STORM and cryo-EM; Diao et al., 2013b; Wickner and Rizo, 2017).

As STORM, cryo-EM, and all-atoms molecular simulations continue to develop through advances in technological innovation, the combination of the three techniques will be a powerful tool for the in-depth investigation on the regulatory mechanisms of synaptic membrane fusion at atomistic resolution, uncovering the recruitment process of Sec1-Munc 18 family proteins to catalyze SNARE assembly, specific lipid properties which be crucial for fusion, and the intricate balance of protein-lipid interactions. We expect to see more exciting applications of synaptic membrane fusion with continued advances in these methods.

#### AUTHOR CONTRIBUTIONS

The manuscript was initially drafted by ZY, SZ and LZ and then further edited after discussion with LG, SC and NL.

#### ACKNOWLEDGMENTS

We thank the anonymous reviewers for the helpful suggestions. This research was supported by the National Natural Science Foundation of China (No. 11504287, 11374237), Fundamental Research Funds for the Central Universities, China Postdoctoral Science Foundation (2017M613147) and Shaanxi Province Postdoctoral Science Foundation (2015). LZ acknowledges Young Talent Support Plan of Xi'an Jiaotong University.

dynamics simulations. Biophys. J. 99, 1221–1230. doi: 10.1016/j.bpj.2010. 06.019


complex: molecular-dynamics model of the fusion clamp. Biophys. J. 105, 679–690. doi: 10.1016/j.bpj.2013.06.018


different temperature dependencies. kinetic analysis of single fusion events in patch-clamped mouse mast cells. Biophys. J. 61, 800–809. doi: 10.1016/S0006- 3495(92)81884-2


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer RS and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review.

Copyright © 2017 Yang, Gou, Chen, Li, Zhang and Zhang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Stimulation Function of Synaptotagmin-1 in Ternary SNARE Complex Formation Dependent on Munc18 and Munc13

#### Yun Li † , Shen Wang† , Tianzhi Li, Le Zhu, Yuanyuan Xu and Cong Ma\*

Key Laboratory of Molecular Biophysics of the Ministry of Education, College of Life Science and Technology, Huazhong University of Science and Technology, Wuhan, China

The Ca<sup>2</sup><sup>+</sup> sensor synaptotagmin-1 (Syt1) plays an essential function in synaptic exocytosis. Recently, Syt1 has been implicated in synaptic vesicle priming, a maturation step prior to Ca<sup>2</sup><sup>+</sup>-triggered membrane fusion that is believed to involve formation of the ternary SNARE complex and require priming proteins Munc18-1 and Munc13-1. However, the mechanisms of Syt1 in synaptic vesicle priming are still unclear. In this study, we found that Syt1 stimulates the transition from the Munc18-1/syntaxin-1 complex to the ternary SNARE complex catalyzed by Munc13-1. This stimulation can be further enhanced in a membrane-containing environment. Further, we showed that Syt1, together with Munc18-1 and Munc13-1, stimulates trans ternary SNARE complex formation on membranes in a manner resistant to disassembly factors NSF and α-SNAP. Disruption of a proposed Syt1/SNARE binding interface strongly abrogated the stimulation function of Syt1. Our results suggest that binding of Syt1 to an intermediate SNARE assembly with Munc18-1 and Munc13-1 is critical for the stimulation function of Syt1 in ternary SNARE complex formation, and this stimulation may underlie the priming function of Syt1 in synaptic exocytosis.

#### Edited by:

Christian Henneberger, University of Bonn, Germany

#### Reviewed by:

Xiaochu Lou, University of Wisconsin-Madison, United States Qiangjun Zhou, Howard Hughes Medical Institute, United States

#### \*Correspondence:

Cong Ma cong.ma@hust.edu.cn

†These authors have contributed equally to this work.

> Received: 17 June 2017 Accepted: 28 July 2017 Published: 15 August 2017

#### Citation:

Li Y, Wang S, Li T, Zhu L, Xu Y and Ma C (2017) A Stimulation Function of Synaptotagmin-1 in Ternary SNARE Complex Formation Dependent on Munc18 and Munc13. Front. Mol. Neurosci. 10:256. doi: 10.3389/fnmol.2017.00256 Keywords: synaptotagmin-1, SNARE complex, synaptic exocytosis, membrane traffic, synaptic vesicle priming

### INTRODUCTION

Neurotransmitter release by synaptic exocytosis is accomplished by the fusion of synaptic vesicles to the plasma membrane upon Ca2<sup>+</sup> influx into the nerve terminal (Südhof, 2004; Rizo and Rosenmund, 2008). In contrast to most intracellular membrane fusion processes, synaptic vesicle fusion occurs in a sub-millisecond timescale in response to Ca2<sup>+</sup> (Augustine et al., 1987; Südhof, 2013). To achieve this goal, most of synaptic vesicles undergo a series of maturation steps before Ca2+-triggered fast fusion, which include: (i) ''tethering'': recruitment of synaptic vesicles to specialized sites at the presynaptic membrane called active zones (Pfeffer, 1999); (ii) ''docking'': close attachment of synaptic vesicles to the fusion sites (Schimmöller et al., 1998); and (iii) ''priming'' that renders the docked vesicles in a semi-stable state ready for fast membrane fusion (Klenchin and Martin, 2000).

As the fusion machinery for synaptic exocytosis, SNARE proteins syntaxin-1, SNAP-25 on the presynaptic membrane and synaptobrevin-2 on synaptic vesicles are involved in vesicle docking, priming and fusion steps (Chen and Scheller, 2001; Brunger, 2005; Jahn and Scheller, 2006; Südhof and Rothman, 2009; Rizo and Xu, 2015). Syntaxin-1 initially interacts with Munc18-1, a Sec1/Munc18-like (SM) protein essential for exocytosis, to form a ''closed'' heterodimeric complex (Misura et al., 2000; Burkhardt et al., 2008), playing a function in vesicle docking (Voets et al., 2001; de Wit et al., 2006). Afterwards, syntaxin-1 assembles with SNAP-25 and synaptobrevin-2 into the ternary SNARE complex composed of a four-helical bundle that forces membranes into close proximity (Jahn and Scheller, 2006; Südhof and Rothman, 2009; Jahn and Fasshauer, 2012). The transition from the Munc18-1/syntaxin-1 complex to the ternary SNARE complex appears to underlie the vesicle priming reaction, where this transition is catalyzed by active zone priming factors, such as Munc13s (Augustin et al., 1999; Ma et al., 2011; Yang et al., 2015). Finally, full assembly of the ternary SNARE complex towards the C-terminal membrane anchors coincides with Ca2+-triggered membrane fusion and this step is highly regulated by neuronal specific proteins such as synaptotagmin-1 (Syt1) and complexins (Südhof and Rothman, 2009; Südhof, 2013; Rizo and Xu, 2015).

Syt1 is one of the major Ca2<sup>+</sup> sensors that mediate synchronous neurotransmitter release (Xu et al., 2007; Bacaj et al., 2013). Syt1 anchors in the synaptic vesicle membrane via a transmembrane region, and comprises two Ca2+ binding C2 domains, known as C2A and C2B domains (Perin et al., 1991), which bind to acidic phospholipids in both a Ca2+-independent and a Ca2+-dependent manner (Fernandez et al., 2001; Fernández-Chacón et al., 2002). Ca2+-dependent binding of the C2 domains to phosphatidylserines (PS) is essential for Ca2+-triggered synchronous neurotransmitter release (Fernandez et al., 2001; Fernández-Chacón et al., 2002; Chapman, 2008; Bacaj et al., 2013). Moreover, recent studies positioned Syt1 as a candidate factor that regulates the maturation steps prior to Ca2+-triggered fusion. For instance, extensive biochemical and biophysical evidence have shown that the Ca2+-independent binding of the C2B domain to phosphatidylinositol 4,5-bisphosphate (PI(4,5)P2) and to the SNAREs (i.e., SNAP-25 and/or syntaxin-1) promotes docking of synaptic vesicles to the fusion sites (de Wit et al., 2009; van den Bogaart et al., 2011; Park et al., 2015).

In addition to the roles in synaptic vesicle docking and Ca2+-evoked fusion, Syt1 has been suggested to be critical for synaptic vesicle priming in more recent studies (Yoshihara and Littleton, 2002; Okamoto et al., 2005; Liu et al., 2009; Wang et al., 2011; Mohrmann et al., 2013; Bacaj et al., 2015). Unlike vesicle tethering and docking steps that can be morphologically assessed by measuring distances (within ∼ nm ranges) between synaptic vesicles and the plasma membrane using electron microscopy (EM; Siksou et al., 2009), priming is characterized by the size of the readily releasable pool (RRP) of vesicles, which is generally measured by monitoring neurotransmitter release induced by hypertonic sucrose solution in a Ca2+-independent manner (Rosenmund and Stevens, 1996). In most previous observations, individual deletion of Syt1 and its isoforms (e.g., Syt2 and Syt7) exhibits no effect on the size of the RRP (Sun et al., 2007; Luo et al., 2015). However, a recent study found that simultaneous loss-of-function of both Syt1 and Syt7 dramatically decreases the RRP size in mass cultures of hippocampal neurons (Bacaj et al., 2015). Consistently, another study reported that the RRP size, the recycling pool size and release probability are all reduced when deletion of Syt1 in cultures containing multiple interconnected neurons (Liu et al., 2009). In line with these evidence, some earlier studies found that deletion of Syt1, especially its C2B domain, in Drosophila neuromuscular junctions (NMJs) leads to a severe reduction of the RRP size (Yoshihara and Littleton, 2002; Okamoto et al., 2005). Altogether, the RRP-promoting effect of Syt1 strongly suggests a function of Syt1 in synaptic vesicle priming. However, the underlying mechanism remains unclear.

Priming is generally believed to involve formation of the ternary SNARE complex (Sørensen et al., 2006). To explore the priming mechanism of Syt1 in synaptic exocytosis, in this study, we investigated the functional importance of Syt1 with the priming factors (i.e., Munc18-1 and Munc13-1) and the SNARE disassembly factors (i.e., NSF and α-SNAP, which are also required for priming; Xu et al., 1999; Burgalossi et al., 2010; Wickner, 2010; Ma et al., 2013) for ternary SNARE complex formation. We observed a strong Ca2+-independent stimulation effect of Syt1 on both cis and trans ternary SNARE complex formation in the presence of Munc18-1 and Munc13-1. In addition, this stimulation of Syt1 is resistant to the SNARE disassembly activity of NSF and α-SNAP, in a manner that critically requires Munc18-1 and Munc13-1. Our results correlate with recent in vivo observations that Syt1 promotes the RRP, and suggest a priming mechanism of Syt1 in synaptic exocytosis.

### MATERIALS AND METHODS

### Plasmids

Rat Syt1 cytoplasmic domain (residues 140–421, known as C2AB), C2A domain (residues 140–266), C2B domain (residues 270–421) and its mutants C2B2RQ (residues 270–421, R398Q and R399Q), C2b (residues 270–421, D363N and D365N), full-length rat Munc18-1, full-length synaptobrevin-2 and its cytoplasmic domain (residues 29–93), rat syntaxin-1a cytoplasmic domain (residues 2–253) and its SNARE domain (H3, residues 191–253), rat Munc13-1 MUN domain (also known as MUN<sup>933</sup> , residues 933–1407, EF, 1453–1531), full-length Cricetulus griseus NSF and full-length Bos taurus α-SNAP were cloned into pGEX-KG vector. Full-length human SNAP-25a (with its four native cysteines mutated to serines), its C-terminal truncation SNAP-25a ∆9 (residues 1–197, with its four native cysteines mutated to serines) and full-length rat syntaxin-1a were cloned into pET28a vector (Novagen). The co-expressed full-length rat Munc18-1 and full-length rat syntaxin-1a, full-length rat Munc18-1 and rat syntaxin-1a cytoplasmic domain (residues 1–261) were cloned to pETDuet-1 vector (Novagen). Rat Munc13-1 C1-C2B-MUN fragment (residues 529–1407, EF, 1453–1531) was cloned into pFastBacTMHtB vector (Invitrogen). All of the mutants used in this study were generated by using QuikChange Site-Directed Mutagenesis Kit (Stratagene).

### Recombinant Protein Expression and Purification

Rat Munc13-1 C1-C2B-MUN fragment was expressed in Sf9 insect cells as previous described (Ma et al., 2013). All other proteins were expressed in E. coli BL21 DE3 by culturing the cells to 0.6–0.8 O.D.<sup>600</sup> at 37◦C, then induced with 0.4 mM isopropyl-β-D-thiogalactoside (IPTG, Amresco) for 16–20 h at 20◦C. Cultured cells were harvested by centrifuging at 4200 rpm in a J6-MI centrifuge equipped with JS-4.2 rotor (Beckman Coulter) at 4◦C.

For GST-tagged proteins, cell pellets from 1 L of culture were resuspended with 50 mM Na2HPO4-NaH2PO<sup>4</sup> pH 7.6, 300 mM NaCl, 10% (v/v) glycerol, 0.5% Triton X-100 (Sigma; lysis buffer A) supplied with 1 mM phenylmethanesulfonyl fluoride (PMSF, Amresco) and 5 mM 2-mercaptoethanol (2-ME, Amresco). Cells were broken using an AH-1500 Nano Homogenize Machine (ATS Engineering Inc.) at 1200 bar for three times at 4◦C. Cell lysates were centrifuged at 16,000 rpm in a JA-25.50 rotor (Beckman Coulter) at 4◦C. The supernatants were collected and mixed with 1 ml glutathione Sepharose 4B (GE Healthcare) affinity media. After 2 h rotation at 4◦C, the mixture were washed twice with lysis buffer A supplied with 2 mM 2-ME. For purification of GST-syntaxin-1a H3 domain, GST-Munc18-1 and GST-MUN, proteins were eluted by a buffer containing 50 mM Tris-Cl, 300 mM NaCl, 10% (v/v) glycerol and 20 mM L-glutathione (GSH, Amresco), pH 8.0. Eluted proteins were desalted by PD-10 desalting column (GE Healthcare) using a buffer containing 25 mM HEPES-KOH, 150 mM KCl and 10% (v/v) glycerol, pH 7.6 (buffer H) supplied with 0.2 mM tris(2-carboxyethyl)phosphine (TCEP, Sigma). For purification tag-free proteins, GST-tag were removed by mixing 10 U/ml of thrombin (from bovine pancreatic) into the media in buffer H supplied with 0.2 mM TCEP at 4◦C overnight. Eluted proteins were further purified by ion exchange chromatography (Source 15S/15Q, GE Healthcare) followed by size-exclusion chromatography (Superdex 75 pg/200 pg, GE Healthcare). For hexa-histidine tagged proteins, cell pellets from 1 L of culture were resuspended with 50 mM Tris-Cl pH 8.0, 300 mM NaCl, 10% (v/v) glycerol, 0.5% Triton X-100 (Sigma; lysis buffer B) supplied with 1 mM PMSF and 2 mM 2-ME. Cells were broken using an AH-1500 Nano Homogenize Machine (ATS Engineering Inc.) at 1200 bar for three times at 4◦C. Cell lysates were centrifuged at 16,000 rpm in a JA-25.50 rotor (Beckman Coulter) at 4◦C. The supernatants were collected and mixed with 1 ml Nickel-NTA agarose (Qiagen) affinity media. After 2 h rotation at 4◦C, the mixture were washed twice with lysis buffer B supplied with 1 mM 2-ME and 30 mM imidazole followed by an additional wash step with Triton X-100-free lysis buffer B supplied with 1 mM 2-ME and 30 mM imidazole. Proteins were eluted with a buffer containing 20 mM Tris-Cl pH 8.0, 150 mM NaCl, 10% (v/v) glycerol (buffer T) supplied with 300 mM imidazole and 0.2 mM TCEP. For full-length syntaxin-1a and co-expressed full-length Munc18-1/syntaxin-1, proteins were eluted with buffer T supplied with 300 mM imidazole, 0.2 mM TCEP and 1% (w/v) sodium cholate (Aladdin, Shanghai, China). Eluted proteins were desalted by using PD-10 desalting column (GE Healthcare) and further purified by ion exchange chromatography (Source 15Q, GE Healthcare) followed by size-exclusion chromatography (Superdex 75 pg/200 pg, GE Healthcare).

Purification buffers for proteins with transmembrane domain were supplied with 1% (w/v) n-octyl-β-D-glucoside (β-OG, Amresco; for full-length synaptobrevin-2) or 1% (w/v) sodium cholate (Aladdin, Shanghai, China; for full-length syntaxin-1a and co-expressed full-length Munc18-1/syntaxin-1). Protein purities were checked using SDS-PAGE (>95%) and final concentrations were determined by A<sup>280</sup> on a P330 NanoPhotometer (Implen).

#### Fluorescent Labeling of Purified Proteins

BODIPY FL [N-(2-aminoethyl)maleimide] (BDPY, Molecular Probes) and tetramethylrhodamine-5-maleimide, single isomer (TMR, Molecular Probes) were used as fluorescence resonance energy transfer (FRET) donor and acceptor in this study. Dye-protein conjugation was achieved by maleimide-cysteine conjugation (all cysteine mutants for dye conjugation were described in the figure). Powdered dyes were dissolved by using dimethylsulfoxide (DMSO, Sigma; for BDPY) or dimethyl formamide (DMF, Aladdin, Shanghai, China; for TMR) to a concentration of 40 mM and stored at −20◦C. Purified proteins were dialyzed with buffer H containing 0.2 mM TCEP and 0.2 mM EDTA and diluted to a concentration of 20–40 µM. Dyes were gently added to the protein solutions with a protein-to-dye ratio of 1:5. Dye-protein mixtures were gently rotated overnight at 4◦C in the dark room. Excess dyes were bleached by adding 10 mM dithiothreitol (DTT, Amresco) and removed by using PD-10 desalting column (GE Healthcare). Labeling/conjugation efficiency was calculated by:

$$\mathcal{E} = \frac{A\_{\rm D} \ast \varepsilon\_{\rm P}}{\varepsilon\_{\rm D} \ast (A\_{280} - A\_{\rm D} \ast C\_{280})} \ast 100\% \tag{1}$$

where A<sup>D</sup> is the peak absorbance of the dye; A<sup>280</sup> is the absorbance of the dye-protein compound at 280 nm; ε<sup>p</sup> is the extinction coefficient of the protein at 280 nm; ε<sup>D</sup> is the extinction coefficient of the dye at the absorption maximum; C<sup>280</sup> is the so called dye correction factor since some dyes display an absorbance at 280 nm. The C<sup>280</sup> is given as:

$$\mathcal{C}\_{280} = \ \varepsilon\_{\rm D280}/\varepsilon\_{\rm D} \tag{2}$$

where εD280 is the extinction coefficient of the dye at 280 nm. About 90% of the labeling efficiency was reached for each samples.

#### GST Pull-Down Assay

Purified GST-syntaxin-1a H3 domain (residues 191–253) was incubated with purified SNAP-25a and the cytoplasmic domain of synaptobrevin-2 (residues 29–93) at 4◦C overnight. The sample was analyzed by SDS-PAGE to confirm the SNARE complex formation. For GST pull-down assay, 20 µl 50% (v/v) glutathione Sepharose 4B affinity media (GE Healthcare) was mixed with 5 µM GST-SNARE complex, 5 µM GST-Munc18-1, 5 µM GST-Munc13-1 MUN domain (residues 933–1407, EF, 1453–1531), 5 µM GST, respectively and 10 µM Syt1 C2B domain to a final volume of 50 µl. The samples were gently rotated for 2 h at 4◦C followed by washing with buffer H three times. Finally, the samples were analyzed by SDS-PAGE followed by coomassie brilliant blue staining and immunoblotting (anti-Syt1, rabbit polyclonal, Proteintech, Wuhan, China). All of the experiments were performed in the absence of Ca2+.

#### Liposome Preparation

1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE), 1,2-dioleoyl-sn-glycero-3-phospho-L-serine (sodium salt; DOPS), L-α-phosphatidylinositol-4,5-bisphosphate

(Brain, Porcine; ammonium salt; PI[4,5]P2), cholesterol (ovine wool; Chol) and 1-2-dioleoyl-sn-glycerol (DAG) were all obtained from Avanti Polar Lipids. T-liposome reconstituted with Munc18-1/syntaxin-1 and t'-liposome containing syntaxin-1/SNAP-25∆9 comprises 37% POPC, 20% POPE, 20% DOPS, 2% DAG, 1% PI(4,5)P2 and 20% Chol and v-liposome reconstituted with BDPY-labeled synaptobrevin-2 comprises 45% POPC, 20% POPE, 15% DOPS and 20% Chol, which are demonstrated in the schematic diagrams in each figures. Lipid mixtures in glass tubes were dried under nitrogen gas flow in water bath at 40◦C followed by vacuuming in a vacuum desiccator for 3 h. Lipid films were dissolved in buffer H containing 0.2 mM TCEP and 1% (w/v) sodium cholate. After vortexing for 5 min, purified proteins (i.e., full-length syntaxin-1a, full-length Munc18-1/syntaxin-1 or BDPY-labeled full-length synaptobrevin-2) were added into the mixture to a protein-to-lipid ratio of 1:200. An additional round of vortex were performed to mix the proteins and lipid mixture. The proteoliposomes were formed by desalting the protein-lipid mixtures with PD-10 desalting column (GE Healthcare) in buffer H containing 0.2 mM TCEP. To remove any residual detergents and contaminants, a routine dialysis step was performed with buffer H containing 0.2 mM TCEP, 0.2 mM EDTA and 0.5 g/L Bio-beads SM2 (Bio-Rad).

#### Fluorescence Experiments

All of the FRET assays in this study were carried out on a PTI QM-40 spectrophotometer with excitation and emission wavelength of 485 nm and 513 nm, respectively at 37◦C.

For Syt1 stimulating ternary SNARE complex formation with and without Munc18-1 and Munc13-1 in membrane-free system (**Figure 1** and Supplementary Figure S1), 5µM syntaxin-1a cytoplasmic domain (residues 2–253) were mixed with 5 µM TMR-labeled SNAP-25a (S187C), 1 µM BDPY-labeled synaptobrevin-2 cytoplasmic domain (residues 29–93, S61C) and 2 µM Syt1 fragment. Alternatively, pre-assembled syntaxin-1a cytoplasmic domain (residues 2–253) and TMR-labeled SNAP-25a (S187C) were added to 5 µM as well. Five micromolar co-expressed transmembrane-free Munc18-1/syntaxin-1 (residues 1–261) were mixed with 5 µM TMR-labeled SNAP-25a (S187C), 1 µM BDPY-labeled synaptobrevin-2 cytoplasmic domain (residues 29–93, S61C), 30 µM Munc13-1 MUN domain (residues 933–1407, EF, 1453–1531) and 2 µM Syt1 fragment. CaCl<sup>2</sup> were added, if indicated, at a final concentration of 1 mM.

For Syt1 stimulating ternary SNARE complex formation with Munc18-1 and Munc13-1 in the presence of membranes (**Figures 2**, **3**), general procedures are the same to those in membrane-free system. Particularly, Munc13-1 MUN domain was replaced by Munc13-1 C1-C2B-MUN fragment (0.5 µM) and the amount of Syt1 fragments were reduced to 0.5 µM due to the recruiting effect of PI(4,5)P2 on t-liposomes. For trans SNARE complex formation, TMR labeled SNAP-25a ∆9 (residues 1–197, S187C) was applied. 0.4 µM full-length NSF, 1 µM α-SNAP, 2 mM ATP-2Na<sup>+</sup> (Sangon, Shanghai, China) and 2 mM MgCl<sup>2</sup> were incorporated as indicated. 0.2 mM EDTA was present all the time to chelate residual Ca2<sup>+</sup> in the reaction buffer.

#### Graphing and Mathematical Methods

Prism 6.01 (GraphPad) was used to graph and perform non-linear curve fits.

### RESULTS

#### Syt1 Stimulates Ternary SNARE Complex Formation with Munc18-1 and Munc13-1

As introduced above, synaptic vesicle priming is generally believed to involve ternary SNARE complex formation. However, the action of Syt1 in ternary SNARE complex formation has been unclear. To this aim, we first examined whether the cytoplasmic fragment of Syt1 (referred to as C2AB) stimulates formation of the ternary SNARE complex in solution. Fluorescence donor and acceptor were separately labeled to synaptobrevin-2 (residues 29–93, S61C) and SNAP-25 (S187C), respectively, and ternary SNARE complex formation was measured by a decrease of donor fluorescence based on FRET (**Figure 1A**). Addition of C2AB did not influence formation of the ternary SNARE complex in the absence and presence of Ca2<sup>+</sup> (**Figures 1A,C**). Similarly, C2AB did not affect formation of synaptobrevin-2 with preassembled syntaxin-1/SNAP-25 heterodimers in the absence and presence of Ca2<sup>+</sup> (**Figures 1B,C**). These results are consistent with previous observations (Stein et al., 2007; Chicka et al., 2008), suggesting that Syt1 has no obvious effect on ternary SNARE complex formation when beginning merely with the three SNAREs.

In a physiological context, SNARE complex formation is highly regulated by Munc18-1 and Munc13-1, the two key factors required for synaptic vesicle priming (Augustin et al., 1999; Südhof and Rothman, 2009). We therefore explored whether Syt1 stimulates ternary SNARE complex formation in the presence of Munc18-1 and Munc13-1 using the FRET assay. As observed in our previous studies (Ma et al., 2011, 2013; Yang et al., 2015), the MUN domain, which contains the minimal priming activity of Munc13-1 (Basu et al., 2005), catalyzed the transition from the Munc18-1/syntaxin-1 complex to the ternary SNARE complex (**Figures 1D,F**). Intriguingly, further addition of C2AB strongly enhanced this transition in a Ca2+-independent manner (**Figures 1D,F** and Supplementary Figure S1), whereas this enhancement was not observed when the MUN domain was absent (**Figures 1D,F**). These results indicate that Syt1 stimulates ternary SNARE complex formation in a Munc18/Munc13-dependent manner.

We further identified the minimal requirement of Syt1 for its stimulation activity. We found that the C2B domain was sufficient for the stimulation, whereas the C2A domain was not (**Figures 1D,F**). These results suggest that the C2B

FIGURE 1 | Synaptotagmin-1 (Syt1) stimulates ternary SNARE complex formation with Munc18-1 and Munc13-1. (A,B) Syt1 actions on ternary SNARE complex formation started with isolated syntaxin-1 and SNAP-25 (A) and preassembled syntaxin-1/SNAP-25 heterodimers (B) in the absence (0.2 mM EDTA) and presence of 1 mM Ca2+. Inset displays a fluorescent-labeling scheme of the SNARE complex (PDB entry 1N7S) and Syt1 C2B domain (PDB entry 1TJX) with indicated poly-basic stretch (K326, K327), bottom face (R398, R399) and Ca2+-binding sites (D363, D365). (C) Quantification of the results in (A,B). (D) Syt1 stimulates ternary SNARE complex formation with Munc18-1 and the MUN domain through its C2B domain in the absence of Ca2<sup>+</sup> (0.2 mM EDTA). (E) The bottom face (i.e., R398, R399) of Syt1 is essential for stimulating ternary SNARE complex formation with Munc18-1 and the MUN domain in the absence of Ca2<sup>+</sup> (0.2 mM EDTA). (F) Quantification of the results in (D,E). Schematic diagrams were displayed on the top of each charts. Representative traces from one of three independent experiments are shown. Data in each bar chart were presented as means ± SD, n = 3, technical replicates. neg.ctrl., excess unlabeled cytoplasmic domain of synaptobrevin-2 was incorporated. Syb, synaptobrevin-2; SN25, SNAP-25.

presence of SNAP-25 ∆9, the Munc13-1 C1-C2B-MUN fragment, 0.2 mM EDTA. (D) Quantification of the results in (C). Schematic diagrams and lipid compositions were displayed on the top of each charts. Representative traces from one of three independent experiments are shown. Data in the bar charts were presented as

domain mediates the stimulation activity of Syt1 in ternary SNARE complex formation dependent on Munc18-1 and Munc13-1.

Structurally, the C2B domain comprises several conserved basic residues at the ''bottom'' (the R398 R399 region) and at the ''side'' (a poly-basic stretch, referred to as the K326 K327 region; **Figure 1A**). Abundant positive charges contributed by these two basic regions endow the C2B domain with the ability to bind acidic SNARE complexes and acidic phospholipids, which helps to couple the functions of Syt1 and the SNAREs in synaptic exocytosis (Mohrmann et al., 2013; Brewer et al., 2015; Zhou et al., 2015; Wang et al., 2016). The K326 K327 region was implicated in recruitment of Syt1 onto the PI(4,5)P2-enriched plasma membrane (van den Bogaart et al., 2011; Park et al., 2015), whereas the R398 R399 region has been recently found to interact with the SNAREs and the SNARE complex (Zhou et al., 2015; Wang et al., 2016).

Therefore, we asked whether disruption of the R398 R399 region that is supposed to impair the SNARE binding affects the stimulation activity of the C2B domain. Indeed, mutation of the R398 R399 region of the C2B domain (R398Q/R399Q, referred to as C2B2RQ) impeded the stimulation activity of the C2B domain in MUN-catalyzed transition from the Munc18-1/syntaxin-1 complex to the ternary SNARE complex (**Figures 1E,F**). In contrast, mutation of the Ca2+ binding sites of the C2B domain (D363N/D365N, referred to as C2b) had no such effect (**Figures 1E,F**). These results suggest that the C2B/SNARE interaction is important for the stimulation function of Syt1 in ternary SNARE complex formation dependent on Munc18-1 and Munc13-1.

means ± SD, n = 3, technical replicates.

On the other hand, we also tested the interactions of the C2B domain with the MUN domain and/or Munc18-1 for its stimulation activity. However, no detectable binding between the C2B domain and the MUN domain or Munc18-1 can be observed using GST pull-down assay in solution (Supplementary Figure S2). Despite that, considering that the stimulation effect of the C2B domain on ternary SNARE complex formation strictly requires both Munc18-1 and Munc13-1 (**Figure 1D**), there might have weak and cooperative interactions among the C2B domain, Munc18-1 and Munc13-1, which orchestrate the C2B/SNARE interaction. Therefore, we expected that, in the presence of membranes, the interactions among the above proteins would be increased due to the protein-recruitment function of several lipid molecules (e.g., PI(4,5)P2 and PS; Martin, 2015; Pinheiro et al., 2016), resulting in enhanced stimulation activity of Syt1 in SNARE complex formation.

### The Stimulation Effect of Syt1 Is Enhanced in the Presence of Membranes

We next explored the stimulation activity of Syt1 in the transition from the Munc18-1/syntaxin-1 complex to the ternary SNARE complex in a membrane-containing environment. To this goal, we developed a membrane-based FRET assay in which the Munc18-1/syntaxin-1 complex was reconstituted into liposomes (referred to as t-liposomes), whereas synaptobrevin-2 was present either in solution (using the cytoplasmic domain of synaptobrevin-2 (residues 29–93, S61C), see **Figure 2A**) or reconstituted onto liposomes (using full-length synaptobrevin-2 (residues 1–116, S61C), referred to v-liposomes, see **Figure 2C**). Note that, cis ternary SNARE complex formation was monitored by using SNAP-25 (S187C) and the cytoplasmic domain of synaptobrevin-2 in the presence of t-liposomes (**Figure 2A**); in contrast, full-length synaptobrevin-2 and a C-terminal truncated SNAP-25 fragment (residues 1–197, S187C, referred to as SNAP-25 ∆9) were applied to monitor trans ternary SNARE complex formation between t- and v-liposomes (**Figure 2C**), because SNAP-25 ∆9 can assemble into the trans ternary SNARE complex without inducing membrane fusion (Binz et al., 1994; Lu, 2015). Fluorescence donor and acceptor were separately labeled to synaptobrevin-2 and SNAP-25 as shown in **Figures 2A,C**, respectively. In this assay, a Munc13-1 C1-C2B-MUN fragment (which includes the C<sup>1</sup> and C2B domains that bind to DAG and PI(4,5)P2 on t-liposomes, respectively), rather than the MUN domain, was applied, due to its high ability for membrane binding (Ma et al., 2013; Liu et al., 2016). Consistent with the results observed in solution (**Figures 1D–F**), we observed that both C2AB and C2B of Syt1 were able to stimulate the transition of the Munc18-1/syntaxin-1 complex to the ternary SNARE complex on membranes with the addition of the C1- C2B-MUN fragment and SNAP-25 (or SNAP-25 ∆9), no matter


TABLE 1 | Data summary of the non-linear curve fits of synaptotagmin-1-stimulated SNARE complex formation in membrane-free, one-membrane and two-membrane systems.

†Rate constant for fast phase; ‡Time constant for fast phase. Data plots were fitted to two-phase exponential decay function except for "membrane-free", which was fitted to one-phase exponential decay function. The data were presented as means ± SE.

synaptobrevin-2 was present in a soluble state (**Figures 2A,B**) or embedded in the membrane (**Figures 2C,D**). As control, this stimulation was abrogated in the absence of the C1-C2B-MUN fragment (**Figures 2A–D**). In addition, C2B2RQ lost the stimulation activity (**Figures 2A–D**). These data suggest that Syt1 stimulates both cis and trans ternary SNARE complex formation on membranes in the presence of Munc18-1 and Munc13-1.

We next analyzed the kinetics of the Syt1-stimulated transition reactions in the absence and presence of membranes (**Table 1** and Supplementary Figure S3). Compared to membrane-free system (Supplementary Figure S3A), one-membrane system (i.e., t-liposomes and the cytoplasmic domain of synaptobrevin-2, see Supplementary Figure S3B) exhibited comparable reaction rate in the absence of C2AB (1.299 ± 0.100 × 10−<sup>3</sup> s −1 vs. 1.352 ± 0.039 × 10−<sup>3</sup> s −1 ), but displayed faster rate in the presence of C2AB (3.540 ± 0.090 × 10−<sup>3</sup> s −1 vs. 15.20 ± 0.69 × 10−<sup>3</sup> s −1 ; **Table 1**). This acceleration arises likely from robust association of C2AB with membranes via binding to PI(4,5)P2, thus increasing C2AB binding probability to the SNAREs, Munc18-1 and Munc13-1 on the membrane. In contrast to the membrane-free and one-membrane systems, two-membrane system (i.e., t-liposomes and v-liposomes, see Supplementary Figure S3C) showed even higher reaction rate both in the absence and presence of C2AB (12.29 ± 1.76 × 10−<sup>3</sup> s −1 and 24.93 ± 1.02 × 10−<sup>3</sup> s −1 , respectively; **Table 1**), which arises likely because the C1-C2B-MUN fragment efficiently associates t- and v-liposomes into close proximity and thus further increases collision probabilities among these proteins (Liu et al., 2016). These results suggest that the membrane behaves as a perfect platform, enabling Syt1 cooperation with Munc18-1 and Munc13-1 to stimulate trans ternary SNARE complex formation more efficiently.

#### Syt1 Stimulates Trans SNARE Complex Formation in the Presence of Munc18-1, Munc13-1, NSF and α-SNAP

It was previously shown that, started with liposomes containing syntaxin-1/SNAP-25 heterodimers, the efficient lipid mixing with synaptobrevin-2 liposomes in the presence of Ca2<sup>+</sup> and Syt1 (C2AB) was abolished by incorporation of NSF and α-SNAP because they disassemble syntaxin-1/SNAP-25 heterodimers (Ma et al., 2013; Liu et al., 2016). In the presence of NSF and α-SNAP, efficient membrane fusion requires both Munc18-1 and Munc13-1 (the C1-C2B-MUN fragment) to orchestrate SNARE- mediated membrane fusion in an NSF/α-SNAP-resistant manner (Ma et al., 2013; Liu et al., 2016). These results suggest that assembly and disassembly factors (i.e., Munc18-1/Munc13-1 and NSF/α-SNAP, respectively) function in alternation to catalyze cycles of membrane fusion (Hughson, 2013; Ma et al., 2013; Rizo and Xu, 2015). In this regard, we investigated whether Syt1 stimulates trans ternary SNARE complex formation starting with syntaxin-1/SNAP-25 heterodimers in the presence of Munc18-1, Munc13-1, NSF and α-SNAP in a membranecontaining environment.

To monitor the formation of trans ternary SNARE complexes on membranes, we routinely used synaptobrevin-2 (fulllength, S61C) and SNAP-25 ∆9 (S187C) as FRET donor and acceptor, respectively (**Figure 3A**). Mixing syntaxin-1/SNAP-25 ∆9 liposomes (t'-liposomes) and synaptobrevin-2 liposomes (v-liposomes) produced trans ternary SNARE complex formation in the presence of Munc18-1, Munc13-1 (the C1-C2B-MUN fragment), NSF, α-SNAP, ATP and Mg2<sup>+</sup> (**Figures 3B,C**). This result is in agreement with the notion that Munc18-1 captures the syntaxin-1 from the membraneembedded syntaxin-1/SNAP-25 heterodimer assisted by NSF and α-SNAP, leading to the transition of the Munc18-1/syntaxin-1 complex to the ternary SNARE complex catalyzed by Munc13-1 through an NSF/α-SNAP-resistant pathway (Ma et al., 2013). Intriguingly, we observed that both C2AB and C2B, instead of C2A, were able to further stimulate trans ternary SNARE complex formation in the presence of Munc18-1, Munc13-1, NSF and α-SNAP (**Figures 3B,C**). In contrast, C2B2RQ, but not C2b, abolished this stimulation (**Figures 3B,C**), reinforcing the importance of the C2B/SNARE interaction for the stimulation function of Syt1. However, the stimulation of Syt1 was abrogated when Munc18-1 and Munc13-1 were absent (**Figures 3B,C**), as NSF and α-SNAP disassemble the syntaxin-1/SNAP-25 complex to prevent SNARE complex formation (Supplementary Figure S4). These results imply that Syt1 cooperates with Munc18-1 and Munc13-1 to stimulate trans ternary SNARE complex on membranes in an NSF/α-SNAPresistant manner.

#### Syt1 Might Stabilize a Ternary SNARE Assembly with Munc18-1 and Munc13-1

We recently found that binding of Munc13-1 to the Munc18- 1/syntaxin-1 complex induces a conformational rearrangement in the syntaxin-1 linker region that cause higher reactivity towards ternary SNARE complex formation (Wang et al., 2017;

FIGURE 4 | Energy landscape of the transition from the Munc18-1/syntaxin-1 complex to the ternary SNARE complex in the presence of Munc13-1 and Syt1. (A) Binding of Munc13-1 to the Munc18-1/syntaxin-1 complex induces a conformational rearrangement in the syntaxin-1 linker region (low Gibbs free energy) thus (B) providing a potential nucleation site for N-terminal assembly among syntaxin-1, SNAP-25 and synaptobrevin-2 (metastable state with rather high Gibbs free energy, which is below the energy barrier (black dashed line)); (C) propagation of the four helical SNARE bundle toward the C-terminal end would eventually dissociate the H3 domain of syntaxin-1 from Munc18-1 to reach a stable ternary SNARE complex (extremely low Gibbs free energy). (D) Binding of Syt1 C2B domain to the partially-assembled ternary SNARE complex (metastable state) would help to lower the Gibbs free energy of the total system (gray solid line to gray dashed line), thus stimulating the reaction. Munc18-1, SNAP-25, synaptobrevin-2, Munc13-1, Syt1 C2B domain are colored in purple, green, blue, pink and orange, respectively. Syntaxin-1 is colored in yellow/black/red (yellow: the Habc domain; black: the linker region; red: the H3 domain). The bottom face of Syt1 C2B domain (i.e., R398, R399) is displayed as blue hexagon.

**Figure 4A**). With Munc13-1, transit of syntaxin-1 from the Munc18-1/syntaxin-1 complex to the ternary SNARE complex strictly requires the simultaneous existence of SNAP-25 and synaptobrevin-2 (Wang et al., 2017). It is thus conceivable that, the conformational change in the syntaxin-1 linker region induced by Munc13-1 would expose the N-terminal end of the H3 domain of syntaxin-1, providing a nucleation site for SNAP-25 and synaptobrevin-2 binding (**Figure 4B**). Subsequently, propagation of the four helical SNARE bundle toward the C-terminal end would eventually dissociate the H3 domain from the closed Munc18-1/syntaxin-1 complex (**Figure 4C**). This model challenges the current dogma on the role of the steadily preassembled syntaxin-1/SNAP-25 heterodimer (''t-SNARE'') as an ''on-pathway'' product in neuronal SNARE complex formation (Bhalla et al., 2006; Wiederhold and Fasshauer, 2009; Hui et al., 2011), suggesting a simultaneous ''N-to-C'' assembly of the three SNAREs chaperoned by Munc18-1 and Munc13-1 (**Figures 4A–C**). Based on this model and our present results, we hypothesize that, binding of Syt1 to a partially installed ternary SNARE assembly with Munc18-1 and Munc13-1 (**Figure 4D**), promotes SNARE zippering forward to the C-terminal end of the ternary SNARE complex. Indeed, this hypothesis is supported by our observation that disruption of the C2B/SNARE interaction by introducing the R398Q/R399Q mutations abolishes the stimulation function of Syt1 in ternary SNARE complex formation dependent on Munc18-1 and Munc13-1 (**Figures 1**–**3**).

### DISCUSSION

Recent in vivo studies suggested a function of Syt1 in synaptic vesicle priming (Yoshihara and Littleton, 2002; Okamoto et al., 2005; Liu et al., 2009; Wang et al., 2011; Mohrmann et al., 2013; Bacaj et al., 2015), but the underlying mechanism of Syt1 is still a mystery. In this study, we observed a strong stimulation activity of Syt1 in ternary SNARE complex formation in the presence of Munc18-1 and Munc13-1. We suggest that this stimulation function of Syt1 may underlie the action of Syt1 in synaptic vesicle priming.

Previous in vitro reconstitution studies found that Syt1 strongly promotes SNARE-dependent membrane fusion in response to Ca2<sup>+</sup> (Tucker et al., 2004; Bhalla et al., 2006; Xue et al., 2008; van den Bogaart et al., 2011; Wang et al., 2011). One aspect of this promotion was suggested to involve an ability of Syt1 in accelerating assembly of syntaxin-1/SNAP-25 heterodimers or the SNARE complex on membranes in a Ca2+-dependent manner (Bhalla et al., 2006; Lai et al., 2014). However, this result is not consistent with in vivo observations that Syt1 promotes the RRP in a Ca2+-independent manner (Bacaj et al., 2015). In addition, although syntaxin-1/SNAP-25 heterodimers appears to be highly reactive when pairing with synaptobrevin-2 for membrane fusion (in the presence of Syt1 and Ca2+), recent in vitro studies indicated that fusion is totally abolished when the disassembly factors NSF and α-SNAP are included (Ma et al., 2013; Liu et al., 2016). Therefore, the actual neuronal SNARE complex formation pathway need to be reconsidered, with increasing evidence indicative of the vital functions of Munc18-1 and Munc13-1 in protecting ternary SNARE complex formation (Ma et al., 2013; Liu et al., 2016).

Our finding that Syt1 efficiently stimulates Munc13 catalyzed transition from the Munc18-1/syntaxin-1 complex to the ternary SNARE complex in the absence of Ca2<sup>+</sup> (**Figures 1D–F**) correlates well with a recently proposed SNARE complex formation pathway (Ma et al., 2013). In this pathway, the Munc18-1/syntaxin-1 complex, rather than the syntaxin-1/SNAP-25 heterodimer, is suggested to represent the physiologically relevant starting point for ternary SNARE complex formation and membrane fusion. In our present study, we showed that the Ca2+-independent stimulation effect of Syt1 in trans ternary SNARE complex formation on membranes is totally blocked in the presence NSF and α-SNAP, but can be restored with the addition of Munc18-1 and Munc13-1 (**Figures 3B,C**). These data reinforce the chaperon function of Munc18-1 and Munc13-1 in protecting SNAREmediated membrane fusion (Hughson, 2013; Ma et al., 2013), and suggest that in this Munc18-1/Munc13-1-chaperoned SNARE complex assembly pathway, the vital role of Syt1 in promoting ternary SNARE complex assembly can be effectively reconstituted.

In our proposed model (**Figure 4**), we suggest that Syt1 acts cooperatively with Munc18-1 and Munc13-1 for its function in promoting ternary SNARE complex formation with the following reasons: (i) the stimulation function of Syt1 in ternary SNARE complex formation can only be observed in the presence of Munc18-1 and Munc13-1 (**Figure 1**); (ii) this stimulation can be further enhanced in the presence of the membranes (**Figure 2**), arising likely because Munc13-promoted association of the membranes, together with PI(4,5)P2-mediated recruitment of multiple priming factors, induces high efficiency of protein-protein assembly; and (iii) multiple transient and low affinity interactions can provide high avidity and specificity, while maintaining the reversibility necessary to orchestrate dynamic assemblies (**Figure 4**).

A recent crystallographic study revealed three distinct Syt1/SNARE binding interfaces; among which the largest one (''the primary interface'' that involves the conserved residues R398 and R399 at the bottom of the C2B domain) interacts with several negatively charged residues on SNAP-25 and syntaxin-1 (residues D51, E52, E55 on SNAP-25 and D231, E234, E238 on syntaxin-1, respectively) and plays a key function in Ca2+-triggered release (Xue et al., 2008; Zhou et al., 2015). Our present work extends this evidence, suggesting an important role of R398 and R399 in synaptic vesicle priming, as mutation of R398 and R399 abolishes the stimulation function of Syt1 in ternary SNARE complex formation dependent on Munc18-1 and Munc13-1 (with and without NSF and α-SNAP; **Figures 1**–**3**). Moreover, we suspect that Munc18-1 and Munc13-1 likely help to render the C2B domain of Syt1 in a proper configuration, leaving the bottom face of the C2B domain (R398 R399) in close contact with the SNARE four-helical bundle (**Figure 4D**, see also the region II of the priming interface in the Syt1/SNARE structure; Zhou et al., 2015).

In addition to the functional interplay of Syt1 with Munc18-1 and Munc13-1 in ternary SNARE complex formation we described above, Syt1 was found to stabilize

#### REFERENCES


and promote ternary SNARE complex formation with complexins (Bacaj et al., 2015) reported in a recent study using a reduced system (i.e., in HEK293T). Moreover, it was suggested that Syt1 induces multimeric assemblies of the ternary SNARE complex in a Ca2+-independent manner (Zhou et al., 2015). All of these activities of Syt1 may contribute to promoting synaptic vesicle priming.

In conclusion, our present study suggests a molecular mechanism of Syt1 in SNARE complex formation that couples its actions in synaptic vesicle priming. Future in vivo research will be required to investigate whether R398 and R399 are involved in promoting the RRP size that underlies synaptic vesicle priming, and decipher the membrane-reconstituted Munc13-1/Munc18-1/Syt1/SNARE assembly that underlies the priming state.

#### AUTHOR CONTRIBUTIONS

YL, SW and CM designed the study and analyzed the data; wrote the article. YL, SW, TL, LZ and YX purified the proteins. YL and SW performed the experiments. CM supervised the study.

#### FUNDING

This work was supported by the National Science Foundation of China (31670846 and 31370819 to CM), the National Key Basic Research Program of China (2015CB910800 and 2014CB910203 to CM) and Program for Changjiang Scholars and Innovative Research Team in University (PCSIRT: IRT13016).

#### ACKNOWLEDGMENTS

We thank Josep Rizo (University of Texas Southwestern Medical Center at Dallas) for providing constructs, Josep Rizo, Tao Xu and Mingjie Zhang for insightful discussions and comments on the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fnmol.2017.002 56/full#supplementary-material

of the readily-releasable pool of synaptic vesicles. PLoS Biol. 13:e1002267. doi: 10.1371/journal.pbio.1002267


vesicle priming in retinal ribbon synapses. J. Neurosci. 35, 11024–11033. doi: 10.1523/jneurosci.0759-15.2015


acting upstream of SNARE nucleation. Nat. Struct. Mol. Biol. 18, 805–812. doi: 10.1038/nsmb.2061


properties in subsets of neurons. Neuron 54, 567–581. doi: 10.1016/j.neuron. 2007.05.004


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Li, Wang, Li, Zhu, Xu and Ma. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Regulation of Exocytotic Fusion Pores by SNARE Protein Transmembrane Domains

Zhenyong Wu1,2 , Sathish Thiyagarajan<sup>3</sup> , Ben O'Shaughnessy <sup>4</sup> and Erdem Karatekin1,2,5,6 \*

<sup>1</sup>Department of Cellular and Molecular Physiology, School of Medicine, Yale University, New Haven, CT, United States, <sup>2</sup>Nanobiology Institute, Yale University, West Haven, CT, United States, <sup>3</sup>Department of Physics, Columbia University, New York, NY, United States, <sup>4</sup>Department of Chemical Engineering, Columbia University, New York, NY, United States, <sup>5</sup>Department of Molecular Biophysics and Biochemistry, Yale University, New Haven, CT, United States, <sup>6</sup>Laboratoire de Neurophotonique, Université Paris Descartes, Faculté des Sciences Fondamentales et Biomédicales, Centre National de la Recherche Scientifique (CNRS), Paris, France

Calcium-triggered exocytotic release of neurotransmitters and hormones from neurons and neuroendocrine cells underlies neuronal communication, motor activity and endocrine functions. The core of the neuronal exocytotic machinery is composed of soluble N-ethyl maleimide sensitive factor attachment protein receptors (SNAREs). Formation of complexes between vesicle-attached v- and plasma-membrane anchored t-SNAREs in a highly regulated fashion brings the membranes into close apposition. Small, soluble proteins called Complexins (Cpx) and calcium-sensing Synaptotagmins cooperate to block fusion at low resting calcium concentrations, but trigger release upon calcium increase. A growing body of evidence suggests that the transmembrane domains (TMDs) of SNARE proteins play important roles in regulating the processes of fusion and release, but the mechanisms involved are only starting to be uncovered. Here we review recent evidence that SNARE TMDs exert influence by regulating the dynamics of the fusion pore, the initial aqueous connection between the vesicular lumen and the extracellular space. Even after the fusion pore is established, hormone release by neuroendocrine cells is tightly controlled, and the same may be true of neurotransmitter release by neurons. The dynamics of the fusion pore can regulate the kinetics of cargo release and the net amount released, and can determine the mode of vesicle recycling. Manipulations of SNARE TMDs were found to affect fusion pore properties profoundly, both during exocytosis and in biochemical reconstitutions. To explain these effects, TMD flexibility, and interactions among TMDs or between TMDs and lipids have been invoked. Exocytosis has provided the best setting in which to unravel the underlying mechanisms, being unique among membrane fusion reactions in that single fusion pores can be probed using high-resolution methods. An important role will likely be played by methods that can probe single fusion pores in a biochemically defined setting which have recently become available. Finally, computer simulations are valuable mechanistic tools because they have the power to access small length scales and very short times that are experimentally inaccessible.

#### Edited by:

Jiajie Diao, University of Cincinnati, United States

#### Reviewed by:

Yingjie Sun, Harvard University, United States Qiangjun Zhou, Stanford University, United States Ying Chuan Li, University of Texas Southwestern Medical Center, United States

> \*Correspondence: Erdem Karatekin erdem.karatekin@yale.edu

Received: 06 June 2017 Accepted: 19 September 2017 Published: 10 October 2017

#### Citation:

Wu Z, Thiyagarajan S, O'Shaughnessy B and Karatekin E (2017) Regulation of Exocytotic Fusion Pores by SNARE Protein Transmembrane Domains. Front. Mol. Neurosci. 10:315. doi: 10.3389/fnmol.2017.00315

Keywords: exocytosis, SNARE, transmembrane domain, fusion pore, membrane fusion

## INTRODUCTION

Coordinated neuronal communication and motor activity rely on tightly controlled release of neurotransmitters. Secretion of hormones is likewise finely tuned, since these compounds control the physiological activities of organs and cells. Both neurotransmitters and hormones are packaged into intracellular secretory vesicles (synaptic vesicles or secretory granules, respectively) and are secreted via calcium-triggered exocytosis. Exocytosis is a multi-step process, involving translocation of secretory vesicles to release sites at the plasma membrane, maturation (called ''priming'') to a state of fusion-readiness, and opening of a fusion pore in response to an increase in the local calcium concentration (Sudhof and Rothman, 2009; Jahn and Fasshauer, 2012; Rizo and Xu, 2015).

The late stages of exocytosis (from maturation at the fusion site to pore dilation) involve about a dozen proteins, many of which are essential. Munc13 is a large priming factor that cooperates with Munc18 to direct SNARE assembly (Rizo and Xu, 2015; Baker and Hughson, 2016). Synaptotagmin-1 (Syt1) and Complexin (Cpx) cooperate to inhibit fusion at resting (low) calcium and to induce rapid fusion upon a rise in calcium (Chapman, 2008; Diao et al., 2012; Rizo and Xu, 2015; Lai et al., 2016). The fusion step itself requires formation of trans complexes between vesicular v- and plasma (target) membrane t-SNAREs that bridge the two membranes (Sudhof and Rothman, 2009). Syt and Cpx may contribute to pore creation (Martens et al., 2007; Hui et al., 2009; Kyoung et al., 2011; Brunger et al., 2015), as Syt couples calcium binding to fusion (Rizo and Xu, 2015) and Cpx somehow increases the efficiency of this process (Lai et al., 2016).

The neuronal/exocytotic soluble N-ethyl maleimide sensitive factor attachment protein receptors (SNAREs) consist of the v-SNARE Synaptobrevin/VAMP2 (Syb2) and the t-SNAREs Syntaxin-1 (Stx) and SNAP25 (SN25; Sollner et al., 1993). The α-helical SNARE domains of these proteins (highly conserved 60–70 residue cytoplasmic regions) assemble in a parallel coiled coil (with all the N-termini at the membrane-distal end) that brings the membranes to be fused into close proximity (**Figure 1**; Sutton et al., 1998). It is less clear what happens as the SNARE complex assembly proceeds toward the membrane-proximal ends. The juxtamembrane regions (JMRs) have a propensity to zipper (Gao et al., 2012), with possible functional implications (Stein et al., 2009; Hernandez et al., 2012). These domains are rich in positively charged residues (Neumann and Langosch, 2011) that bind and recruit acidic phospholipids, including PI(4,5)P<sup>2</sup> (van den Bogaart et al., 2011; Honigmann et al., 2013) and PI(3,4,5)P<sup>3</sup> (Khuong et al., 2013) to vesicle docking and fusion sites (Barg et al., 2010; Gandasi and Barg, 2014).

Fusion formally occurs once the vesicular lumen is connected to the extracellular space via a fusion pore. Nevertheless, the dynamics of the pore can further control the release process (Breckenridge and Almers, 1987; Zimmerberg et al., 1987; Monck and Fernandez, 1996; Lindau and Alvarez de Toledo, 2003). The pore is initially small (a few nm in diameter) and can flicker open and closed repeatedly before resealing or dilating irreversibly. In transient ''kiss and run'' fusion, the pore reseals before complete emptying of the vesicle. Alternatively, the pore may dilate irreversibly as the fused vesicle's membrane collapses into the plasma membrane and the entire cargo is released (full fusion). Thus, beyond affecting the kinetics and the amount of cargo released, pore dynamics also control the mode of vesicle recycling. Transient fusion is a well-established mode of hormone release by neuroendocrine cells (Breckenridge and Almers, 1987; Zimmerberg et al., 1987; Monck and Fernandez, 1996; Lindau and Alvarez de Toledo, 2003; Fulop et al., 2005). It is also documented for synaptic vesicle exocytosis (Staal et al., 2004; He et al., 2006; He and Wu, 2007; Alabi and Tsien, 2013), but its prevalence and significance are debated, in part due to technical challenges in probing fusion pores during neurotransmitter release.

Molecular mechanisms that regulate pore dynamics are not well understood (Lindau and Alvarez de Toledo, 2003; Harata et al., 2006; He and Wu, 2007; Lindau, 2012). Fusion pore properties are affected by calcium (Hartmann and Lindau, 1995; Chiang et al., 2014), dynamin (Anantharam et al., 2011; Chiang et al., 2014), the actin cytoskeleton and/or membrane tension (Bretou et al., 2014; Wen et al., 2016), phosphorylation (Staal et al., 2004), molecular crowding (Wu et al., 2017) and mutations in many of the components of the fusion machinery. In this review article, we emphasize the role of SNARE TMDs in regulating fusion pore dynamics. We choose this focus because SNAREs and Syt are the only TMD proteins known to be involved in the late stages of pore opening and dilation, and few systematic studies of the role of Syt1 TMD in exocytosis are available (Lee and Littleton, 2015). However, we stress that in addition to affecting pore dynamics (Han et al., 2004; Borisovska et al., 2005; Chang et al., 2015, 2016; Dhara et al., 2016), SNARE TMDs also regulate pre-fusion stages (Chang et al., 2016; Dhara et al., 2016). In addition, for nearly all of the proteins mentioned above there are mutations with fusion pore phenotypes (Wang et al., 2001, 2003; Jorgacevski et al., 2011; Dhara et al., 2014). In particular, interfering with synaptotagmin's calcium binding and/or membrane penetration (Paddock et al., 2011; Lai et al., 2015) influence fusion pore opening and pore properties (Chapman, 2008). Since Synaptotagmin also binds SNAREs (Lai et al., 2014; Zhou et al., 2015), its effects on fusion pores may be difficult to disentangle from those of SNARE protein TMDs. Finally, in addition to possible direct influences, TMDs of other proteins may indirectly influence the fusion pore via their interactions with SNARE TMDs (e.g., the synaptophysin TMD interacts with the Syb2 TMD, Adams et al., 2015).

SNARE TMDs seem to possess some special features. First, compared to other tail-anchored proteins, SNARE TMDs are enriched in beta-branched Ile and Val residues (Neumann and Langosch, 2011). Beta-branched residues (Ile, Val, or Thr) contain two non-hydrogen substituents attached to their C-β carbon, compared to other amino acids that contain only one (Popot and Engelman, 2000). The increased bulkiness near the protein backbone makes it harder for β-branched amino acids to adopt α helical conformations in solution. Nevertheless, such residues are frequently found in α-helical TMDs where they are thought to increase the conformational flexibility of the TMD (Popot and Engelman, 2000). Second, the tiny, helix-perturbing

post-fusion cis SNARE complex (Stein et al., 2009). (A) A synaptic vesicle is docked at the plasma membrane by trans-SNARE complexes. The exocytotic/neuronal v-SNARE Syb2 (blue) and the t-SNARE (Stx1 and SNAP25, depicted together in red) are anchored to the synaptic vesicle and the plasma membrane, respectively via their transmembrane domains (TMDs). Further zippering, coupled with the action of the calcium sensor Synaptotagmin-1 (not shown) and possibly other factors leads to the opening of a fusion pore (C). Possible intermediate structures on the pathway to opening of the fusion pore include a hemifusion state (B) wherein the proximal leaflets, but not the distal ones, are fused. An alternative intermediate is a channel-like structure formed by oligomerization of the TMDs of soluble N-ethyl maleimide sensitive factor attachment protein receptors (SNAREs) in both membranes (B<sup>0</sup> ). Hetero-oligomerization of SNAREs with other proteins may also contribute to the channel structure. Expansion of the proteinaceous pore would lead to invasion of the pore's walls by lipids. In (C), the SNAREs are shown fully zippered at the waist of the pore, but the actual structure is unknown. The fusion pore can fluctuate in size, and flicker open and shut multiple times before expanding further, leading to full fusion (D), or resealing, concluding a transient or kiss-and-run fusion event (E). (F) Structure of the cis-SNARE complex (adapted from Stein et al., 2009), PDB file 3HD7, rendered in PyMol). The t-SNAREs Syntaxin 1A and SNAP25 are shown in red and salmon, respectively; the v-SNARE Syb2 is shown in blue; TMDs are shown inserted into a membrane. Beta-branched residues are indicated in orange (Stx1) or cyan (Syb2). Note that the ultimate residue in Syb 2 and the last two C-terminal residues in Stx1 were not resolved in the structure and are absent from the image. The plasma membrane thickness is slightly larger than the TMD lengths (Sharpe et al., 2010). (G) Alignment of TMD sequences of Syb2 and Stx1 according to the crytal structure shown in (F). Contacts between residues observed in the crystal structure are indicated as black lines. The dashed lines indicate residues that face one another, but are further apart, as the two helices veer apart toward the very C-termini. Beta-branched residues are indicated in orange (Stx1) or cyan (Syb2), as in (F) modified from Stein et al. (2009). (H) Alignment of TMDs of Syb2 from several species. Uniprot identifiers (http://www.uniprot.org) are indicated in parentheses. The arrows mark the Gly in position 100 (using the rat sequence as reference) and the tiny residue in position 103. See Hastoy et al. (2017) for a more comprehensive alignment.

Gly is enriched in the N-terminal portion of the TMD (Neumann and Langosch, 2011) which may allow a kink in the TMD helix (Han et al., 2016b). In comparison, viral fusion protein TMDs are also enriched in Gly (Cleverley and Lenard, 1998), but at a more central position (Neumann and Langosch, 2011). Finally, the TMDs of exocytotic SNAREs are exceptions to the observation that for most proteins the TMD length matches the thickness of membrane wherein the protein resides (Sharpe et al., 2010). Neuronal SNARE TMDs are shorter than the average plasma membrane thickness.

Exocytosis is unique among all biological fusion reactions in that pore dynamics can be observed with sub-millisecond temporal resolution under native conditions using high resolution electrophysiological and electrochemical methods (Travis and Wightman, 1998; Lindau, 2012). In addition, high temporal resolution of single-pore measurements was recently achieved in biochemically defined systems, promising to illuminate many mechanistic questions (Nikolaus and Karatekin, 2016; Stratton et al., 2016; Wu et al., 2016, 2017). Finally, atomistic (Blanchard et al., 2014; Han et al., 2016b) and coarse-grained (CG) simulations (Risselada et al., 2011; Han et al., 2015; Mostafavi et al., 2017) of fusogen protein-membrane systems have provided important insights into the role of TMDs in fusion pore regulation.

### MEMBRANE FUSION PATHWAYS

Fusion between purely lipidic membranes is non-specific and relatively slow (Chanturiya et al., 1997; Warner and O'Shaughnessy, 2012a). In consequence, biological membrane fusion requires specific proteins to perform the recognition and fusion steps. Despite great diversity in biological fusion reactions, from enveloped virus infection, cell-cell fusion, intracellular trafficking and wound repair to exocytosis, the fusogens involved share some general evolutionary principles and drive fusion through a limited number of pathways (**Figure 1**). First, at some stage the fusogens must be anchored to both of the bilayers. Second, a conformational change in the fusogens brings the hydrated phospholipid head groups into close contact. Fusion requires that substantial hydration forces be overcome (Rand and Parsegian, 1989) and that intermediate high energy states be transiently assumed, in which lipid arrangements are far from that of the equilibrium bilayer. How the associated barrier to fusion is overcome by fusion proteins has been much debated. It was proposed that fusion is triggered when SNARE proteins cooperatively generate entropic forces that clear the contact zone between apposing membranes and push them into close proximity, causing rapid fusion due to thermally driven collisions (Mostafavi et al., 2017). Fusion may be promoted by local destabilization of the bilayer structure, for example by bulging (Martens et al., 2007; Hui et al., 2009). In a radically different possible scenario, the initial fusion pore is a channel-like structure formed by assembly of two hemi-channels in the two fusing membranes (Breckenridge and Almers, 1987; Jackson and Chapman, 2008; Chang et al., 2017; **Figure 1B**<sup>0</sup> ). In this view the channel somehow subsequently dilates, allowing lipids to invade the pore (Chang et al., 2017).

The fusion pathway may be different in different systems and remains controversial, but in some cases has been shown to pass through or terminate in a hemifused state in which only the proximal leaflets of the apposing membranes are fused while the distal leaflets engage in an extended bilayer region called a hemifusion diaphragm (**Figure 1B**). These include calciummediated fusion of protein-free giant unilamellar vesicles (GUVs; Nikolaus et al., 2010; Warner and O'Shaughnessy, 2012b), fusion of yeast vacuoles (Reese et al., 2005; Jun and Wickner, 2007), and fusion between liposomes mediated by SNAREs alone (Lu et al., 2005; Hernandez et al., 2012) or together with Syt and/or Cpx (Schaub et al., 2006; Diao et al., 2012). Hemifusion was recently observed during exocytosis in chromaffin cells (Zhao et al., 2016).

Truncating SNARE TMDs or replacing them with lipids spanning a single membrane leaflet usually impairs fusion and results in hemifusion (McNew et al., 2000; Xu et al., 2005; Fdez et al., 2010; Chang et al., 2016). Thus TMDs can affect the pathway, either by helping to bypass dead-end hemifusion, or by converting hemifusion to fusion.

#### REGULATION OF EXOCYTOTIC FUSION PORES BY SNARE TMDs

Once the two membranes have fused, the SNARE complex is now in cis, i.e., both the Syb2 and Stx TMDs are embedded in the same membrane. There are several features of the TMDs which may influence pore opening and dynamics at this stage: (1) flexibility of the TMDs; (2) specific interactions between TMDs; and (3) TMD-lipid interactions. These effects are difficult to disentangle. In addition, membrane properties (curvature and tension) and the soluble portions of the fusogens and their interactions with one another and with other proteins will constrain the configurations available to the TMDs.

#### TMD Flexibility

Increased flexibility may promote membrane fusion by allowing TMDs to sample conformations compatible with membrane shape changes that accompany fusion (Langosch et al., 2007; Neumann and Langosch, 2011). Consistent with this view, in reconstituted bulk fusion assays fusion activity correlated with β-branched residue content in the TMD sequence (Hofmann et al., 2004; Langosch et al., 2007), although some of these experiments used small sonication-generated liposomes that can be prone to fusion even without fusogens, and leakiness during fusion was not always tested.

TMD flexibility was also identified as a key factor in secretion kinetics in recent studies of exocytosis from mouse chromaffin cells. Dhara et al. (2016) studied cells lacking Syb2 and Cellubrevin (Syb3), that expressed exogenous Syb2 with the entire TMD replaced with a sequence containing various combinations of only Ile, Leu or Val. The results ranged from severe impairment to normal exocytosis. The degree of restoration of secretion correlated well with the fraction of β-branched residues that the Syb2 TMD sequence contained in its N-terminal half (that portion embedded in the cytoplasmic leaflet of the vesicular/plasma membrane). Replacing the wild-type TMD with a polyL stretch reduced the amplitudes of the rapid phases of exocytosis ∼5 fold and the slow phase <2 fold as measured by whole-cell capacitance. Interestingly, however, the kinetics were unaffected. In amperometric measurements of single-vesicle release events, replacing the native TMD by polyV or polyI led to shorter, faster rising spikes (∝ amount of catecholamine flux reaching the detector), and shorter, higher amplitude pre-spike features (related to release through the initial pore). Further, pre-spike fluctuations increased in frequency and amplitude. Replacing the TMD by an α-helix-stabilizing polyL sequence produced the opposite effects. These findings suggest that β-branched residues destabilize the pore, facilitating its nucleation (increasing event frequency), and accelerating pore dilation (shorter pre-spike duration and sharper spikes).

In the VAMP2 TMD, a highly conserved Gly100 is followed by another tiny residue three positions later (Cys, Gly, or Ala, **Figure 1H**), which may be important in allowing a kink in the TMD toward the middle of the bilayer (Han et al., 2016b). Hastoy et al. (2017) studied the role of VAMP2 G100 and C103 by substituting them with Val. These substitutions should impair the ability of the TMD to kink. Using both short peptides encompassing the TMD and full-length purified VAMP2 reconstituted in artificial lipid membranes, the authors found that substitution of C103 and especially of G100 with Val rendered the membranes less fluid and impaired a transition from alpha-helical to beta-sheet conformation of the TMD as the protein concentration increased. These effects correlated with reduced exocytosis in PC12 and INS-1 cells when the knocked-down endogenous VAMP2 expression was rescued with the mutants. Fusion pores expanded faster for the Val mutants, but they also resealed faster after discharging less cargo—a result compatible with transient, kiss and run fusion.

These effects of TMD flexibility on fusion pores are further discussed in the ''Computer Simulations of SNARE TMDs and Their Influence on the Fusion Pore'' section.

#### TMD-TMD Interactions

Interactions among v- and t-SNARE TMDs can take three forms: homotypic (vTMD-vTMD or tTMD-tTMD), heterotypic (vTMD-tTMD) or interactions of SNARE TMDs with TMDs of other proteins, e.g., synaptophysin (Adams et al., 2015).

Homodimerization of v-SNARE TMDs has long been known (Washbourne et al., 1995; Roy et al., 2004; Langosch et al., 2007), but it has been argued that these weak interactions are of little consequence for exocytosis (Bowen et al., 2002; Fdez et al., 2010; Dhara et al., 2016). Neuronal t-SNAREs form clusters in artificial membranes (Bacia et al., 2004; Murray and Tamm, 2009), neuroendocrine cells (Lang et al., 2001; Barg et al., 2010; van den Bogaart et al., 2011; Honigmann et al., 2013, Honigmann-NSMB13; Gandasi and Barg, 2014) and at the neuromuscular junction (Khuong et al., 2013). However, these clusters seem to arise from electrostatic interactions between phospholipids and the JMR of Stx, or from recruitment by vesicle docking (Gandasi and Barg, 2014) rather than specific TMD-TMD interactions.

It has also been proposed that both t- and v-SNARE TMDs may homo-oligomerize into channel-like structures (Han et al., 2004; Jackson and Chapman, 2008; Chang et al., 2015). Remarkably, systematic mutagenesis showed that residues that affected fusion pore currents all fall on one side of the t-SNARE TMD helix, possibly facing the pore's lumen (Han et al., 2004). In order to release cargo from the vesicular lumen to the extracellular space, a pore lined with t-SNARE TMDs would require a complementary pore formed on the vesicular side by oligomerization of v-SNARE TMDs. Although some evidence supports this idea (Chang et al., 2015; Bao et al., 2016), it is less compelling than that for t-SNARE TMDs. It is also possible that the vesicular hemi-channel includes TMDs from another vesicular protein such as synaptophysin (Chang et al., 2017). A channel-like pore might constitute only the initial structure, yielding a lipid-lined pore once the pore expands (Chang et al., 2017). Since the initial pore lasts only a few milliseconds, it is difficult to confirm or refute such a highly transient, channel-like structure. The notion of a channel-like structure would gain credibility if the contacts between channel-forming units could be identified and manipulated, e.g., to stabilize the pores.

Interactions between v- and t-SNARE TMDs were reported nearly two decades ago (Poirier et al., 1998; Margittai et al., 1999). More recently, Stein et al. (2009) solved the crystal structure of the neuronal SNARE complex including the Syb2 and Stx TMDs in the presence of detergent. The α-helices of Syb2 and Stx1 continued beyond the SNARE domain to the C-termini, spanning the linker region and the TMDs. Contacts between certain Syb2 and Stx1 residues were identified in both linker and TMD domains (**Figures 1F,G**).

This raises an interesting question: do the TMD-TMD contacts represent specific interactions promoting SNARE complex zippering through the bilayer and affecting the fusion process, or are they artifacts of crystal packing constraints? Wu et al. (2016) sought to answer this question using a novel approach in which single fusion pores can be probed in a biochemically defined system. Three isoleucines in the Syb2 TMD that contact Stx1 TMD residues in the crystal structure were mutated to alanines, and the fusion rate and individual pore properties were monitored. The manipulation reduced the fusion rate moderately, but increased pore lifetimes 10-fold, from ∼6 s to ∼60 s. Replacing the entire TMD with that of a non-exocytotic v-SNARE or a lipid anchor spanning the entire bilayer resulted in qualitatively similar outcomes. These results suggest that specific interactions between Syb2 and Stx1 TMDs are not essential, but may help fine-tune the fusion reaction.

In a later study, Wu et al. (2017) found that pore dilation does not rely on putative v- and t-SNARE TMD interactions, but rather their results support a dilation mechanism from entropic forces generated by crowding of SNARE complexes at the fusion pore.

#### TMD-Lipid Interactions

Clusters formed by neuronal t-SNAREs are cholesteroldependent in artificial membranes (Bacia et al., 2004; Murray and Tamm, 2009; Milovanovic et al., 2015) and in live neuroendocrine cells (Lang et al., 2001). Unlike raft-associated proteins, t-SNAREs were found to be enriched in cholesterol-poor membrane regions. Milovanovic et al. (2015) showed that the TMD of Stx1 alone is sufficient for cholesteroldependent clustering and argued that this effect originated in hydrophobic mismatch. Cholesterol-rich membrane regions tend to form thicker liquid-ordered L<sup>o</sup> domains, whereas cholesterol-poor regions tend to form thinner, liquid-disordered (Ld) domains. Remarkably, the TMD length of most membrane proteins matches the thickness of the membrane in which they normally reside (Mitra et al., 2004; Sharpe et al., 2010), while neuronal/exocytotic SNAREs appear to be exceptions in that their TMDs are considerably shorter than the thickness of the plasma membrane (Sharpe et al., 2010). This length mismatch may introduce lipid-packing defects that can be minimized if the offending TMDs are clustered (Milovanovic et al., 2015).

Addition of even a single charged residue to the lumenal C-terminus of Syb2 inhibits fusion (Ngatchou et al., 2010). In contrast, Syb2 fused to a pH-sensitive GFP via a flexible linker sequence composed of S and G supports exocytosis (Miesenböck et al., 1998). Ngatchou et al. (2010) interpreted this as evidence that a few of the lumenal Syb2 residues adjacent to the TMD may need to move toward the hydrophobic core of the membrane. This would destabilize the vesicular membrane and help open a fusion pore. D'Agostino et al. (2016) argued that at least during yeast vacuolar fusion this ''penetration model'' is unlikely to hold.

#### COMPUTER SIMULATIONS OF SNARE TMDs AND THEIR INFLUENCE ON THE FUSION PORE

Computational studies are of particular value to the fields of membrane fusion and exocytosis because experimental characterization of the small, short-lived fusion pore is challenging. These small scale and short-lived features are accessible to computer simulations. The most detailed approaches are atomistic, currently accessing up to microsecond timescales (Han et al., 2016b), while CG methods can probe considerably larger times (Cooke et al., 2005; Marrink et al., 2007; Monticelli et al., 2008; Mostafavi et al., 2017).

#### TMD Flexibility

Conformational flexibility of the TMDs has been proposed to play a role in fusion (Langosch et al., 2007; Stelzer et al., 2008). Atomistic simulations of a v-SNARE C-terminal fragment in a membrane identified three types of flexibility possessed by the α-helical linker-TMD regions: (i) tilt relative to the membrane normal; (ii) a kink feature at the Gly100 residue; and (iii) conformational flexibility of the entire backbone (Blanchard et al., 2014; Han et al., 2016b). The tilt and kink angles were uncorrelated, as expected for a flexible TMD, yet confined to a narrow range ∼10<sup>0</sup> (Blanchard et al., 2014). Similar kinked conformations were seen in simulations with CG representations of TMDs and lipids (Durrieu et al., 2009; Lindau et al., 2012), and in an atomistic study of a t-SNARE C-terminal fragment (Knecht and Grubmuller, 2003).

Taken together, simulations and experiments suggest that TMD conformational flexibility (in particular kinking and/or backbone flexibility—types (ii) and (iii) above) promotes exocytosis. In chromaffin cell experiments with the Syb2 native TMD replaced by sequences containing only Val, Ile, or Leu, β-branched residue content correlated with restoration of secretion (Dhara et al., 2016). Simulations appear to identify flexibility as the relevant property, because polyI and polyV substitutions increased simulated backbone TMD flexibility, while polyL substitution decreased the flexibility (Han et al., 2016b). By contrast, in these simulations all three substitutions reduced the tilt and straightened the TMD compared to wildtype, while fluctuations in the associated angles were reduced. These results suggest tilt and kink flexibilities are of minor importance to pore dilation. However, mutations of the Syb2 residues Gly100 and Cys103 to Val that would be expected to reduce the N-terminal kinking resulted in a decrease in exocytosis and pores that expanded and closed faster following partial release (Hastoy et al., 2017). Therefore, both types of flexibility (kinking and backbone fluctuations) appear to be relevant.

Thus, increased TMD flexibility may favor pore nucleation and expansion. What is the underlying mechanism? Greater TMD flexibility in the N-terminal portion of the TMD might splay lipids and so relieve the high negative curvature in the outer phospholipid leaflet in a small fusion pore, thereby promoting pore nucleation (Dhara et al., 2016). Dhara et al. (2016) also proposed that this mechanism would promote pore expansion, but this is less clear since pore geometry is complex. Consistent with the ability of TMDs to moderate lipid ordering, TMDs disturbed lipid ordering in their vicinity in atomistic and Martini simulations (Risselada et al., 2011; Han et al., 2016b). TMD flexibility may also promote fusion more directly, by helping membranes to assume configurations required to navigate pathways to fusion. Further, several known or potential TMD-TMD interactions may be affected by Syb2 TMD mutations. One might expect that any specific interactions would be disrupted by replacing the TMD with a simple I, L, or V repeat sequence (Dhara et al., 2016). However, specificity in TMD-TMD interactions may be encoded in just a single TMD residue in a cellular context, likely due to packing interactions (Heim et al., 2015).

Many questions remain unanswered regarding the role of TMD flexibility in fusion. For example, if indeed greater flexibility translates to more frequent and faster exocytosis, one might anticipate that lipid-anchored Syb2 would provide the most efficient fusion. However, no such enhancement is seen when Syb2 is anchored by palmitoylated cysteine string protein (CSP) in chromaffin cells (Chang et al., 2016; Dhara et al., 2016), neurons (Chang et al., 2016), or when a lipid anchor that spans the entire bilayer is used (McNew et al., 2000; Wu et al., 2016, 2017).

A mathematical model suggested that another possible mechanism whereby TMD flexibility could promote fusion is by enhancing entropic forces tending to expand the pore. In the model, increased TMD flexibility would increase orientational fluctuations of and mutual steric interactions among the cis-SNARE complexes, increasing the entropic pore expansion force (Wu et al., 2017).

#### TMD-TMD Interactions

The mathematical model of Wu et al. (2017) identified a critical role for specific and non-specific TMD-TMD interactions in fusion. These interactions drive zippering of cis-SNARE complexes to the fusion pore waist, forcing the SNAREs to interact sterically and thus generating entropic forces that drive pore expansion.

TMD interactions may be important at other stages of fusion. Simulations show v-SNARE TMDs interact and can homodimerize or form higher order oligomers (Fleming and Engelman, 2001; Han et al., 2015, 2016a), with an interaction energy of ∼10 kBT between v-SNARE C-terminal fragments measured in a hybrid atomistic-MARTINI approach (Han et al., 2015). A MARTINI study suggested that the fusion pathway passes through a hemifused state with a HD, and that homodimerization of SNARE TMDs restricts the HD to remain small and therefore to transit more readily to a fusion pore (Risselada et al., 2011).

#### TMD-Lipid Interactions

Interactions between lipids and SNARE TMDs or JMRs may assist fusion. In Martini simulations, post fusion SNARE complexes surrounding the fusion pore were constrained to retain their Y shape by the energy penalty associated with moving the C-terminal polar residues through the hydrophobic membrane core (Risselada et al., 2011). Thus, the bending energy stored in the C-terminal portion of the complexes could be released only by pore expansion. Other MARTINI and hybrid atomistic/CG studies have shown that PI(4,5)P<sup>2</sup> concentrates at t-SNARE JMRs due to interactions with the charged Lys and Arg residues (Khelashvili et al., 2012; Sharma et al., 2015). These effects are thought to help cluster neuronal t-SNAREs (van den Bogaart et al., 2011).

#### AUTHOR CONTRIBUTIONS

ZW, ST, BO'S and EK contributed to the review of the literature, and to the writing and editing of the manuscript. ZW and EK produced the initial draft. EK coordinated the work.

#### REFERENCES


#### FUNDING

This work was supported by National Institute of General Medical Sciences, NIH grant R01GM108954 and a Kavli Neuroscience Scholar Award to EK.

#### ACKNOWLEDGMENTS

We thank all members of the Karatekin and O'Shaughnessy labs and Donald Engelman (Molecular Biophysics and Biochemistry, Yale University) for thoughtful discussions.


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Wu, Thiyagarajan, O'Shaughnessy and Karatekin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Productive and Non-productive Pathways for Synaptotagmin 1 to Support Ca2+-Triggered Fast Exocytosis

#### Jaewook Kim and Yeon-Kyun Shin\*

Roy J. Carver Department of Biochemistry, Biophysics and Molecular Biology, Iowa State University, Ames, IA, United States

Ca2+-triggered SNARE-mediated membrane fusion is essential for neuronal communication. The speed of this process is of particular importance because it sets a time limit to cognitive and physical activities. In this work, we expand the proteoliposome-to-supported bilayer (SBL) fusion assay by successfully incorporating synaptotagmin 1 (Syt1), a major Ca2<sup>+</sup> sensor. We report that Syt1 and Ca2<sup>+</sup> together can elicit more than a 50-fold increase in the number of membrane fusion events when compared with membrane fusion mediated by SNAREs only. What is remarkable is that ∼55% of all vesicle fusion events occurs within 20 ms upon vesicle docking. Furthermore, pre-binding of Syt1 to SNAREs prior to Ca2<sup>+</sup> inhibits spontaneous fusion, but intriguingly, this leads to a complete loss of the Ca2<sup>+</sup> responsiveness. Thus, our results suggest that there is a productive and a non-productive pathway for Syt1, depending on whether there is a premature interaction between Syt1 and SNAREs. Our results show that Ca2<sup>+</sup> binding to Syt1 prior to Syt1's binding to SNAREs may be a prerequisite for the productive pathway. The successful reconstitution of Syt1 activities in the physiological time scale provides new opportunities to test the current mechanistic models for Ca2+-triggered exocytosis.

Keywords: SNARE proteins, synaptotagmin 1, membrane fusion, supported bilayer, single molecule biophysics, exocytosis

#### INTRODUCTION

One of the truly remarkable features of the neuron is its ability to release neurotransmitters in <1 ms, in response to the Ca2<sup>+</sup> influx (Schneggenburger and Neher, 2000; Burgalossi et al., 2012). Neurotransmitter release is achieved by means of synaptic vesicle fusion onto the plasma membrane. The protein components that mediate fast membrane fusion have largely been identified and most of them are well-characterized biochemically and structurally. The highly conserved soluble N-ethylmaleimide-sensitive factor attachment protein receptor (SNARE) proteins are the central fusogen (Südhof and Rothman, 2009). Cognate vesicle (v-) SNARE and target plasma membrane (t-) SNAREs associate to form a four-stranded coiled coil which drives membrane fusion (Poirier et al., 1998; Sutton et al., 1998; Weber et al., 1998). Additionally, synaptotagmin 1 (Syt1) is known to be a major Ca2+-sensor (Brose et al., 1992; Fernández-Chacón et al., 2001; Chapman, 2008) while complexins (Cpx) (Giraudo et al., 2006; Tang et al., 2006; Xue et al., 2007) and sec1/Munc18-like (SM) (Südhof and Rothman, 2009; Ma et al., 2013) proteins

#### Edited by:

Jiajie Diao, University of Cincinnati, United States

#### Reviewed by:

Dechang Li, Beijing Institute of Technology, China Chiayu Chiu, Interdisciplinary Center of Neuroscience of Valparaíso, University of Valparaíso, Chile Ying Chuan Li, University of Texas Southwestern Medical Center, United States

#### \*Correspondence:

Yeon-Kyun Shin colishin@iastate.edu

Received: 20 July 2017 Accepted: 01 November 2017 Published: 15 November 2017

#### Citation:

Kim J and Shin YK (2017) Productive and Non-productive Pathways for Synaptotagmin 1 to Support Ca2+-Triggered Fast Exocytosis. Front. Mol. Neurosci. 10:380. doi: 10.3389/fnmol.2017.00380 are both believed to play intimate roles in tightly regulating membrane fusion. However, the mechanisms whereby these protein components orchestrate synchronized vesicle fusion in such a short time scale is still unclear at the molecular level.

An effective and powerful approach to delineate the mechanism may be the in vitro reconstitution of the membrane fusion reaction. For example, fusion of v-SNARE-reconstituted proteoliposomes to t-SNARE-containing supported bilayer (SBL), when analyzed with single molecule spectroscopy, revealed that SNAREs alone are capable of mediating membrane fusion in <25 ms (Liu et al., 2005; Domanska et al., 2009; Kiessling et al., 2013). However, attempts to functionally incorporate Ca2+-sensor Syt1 using this platform have not been successful. Thus, it was impossible to test the proposed mechanistic models of Ca2+-triggered exocytosis with this method.

Meanwhile, an alternative experimental platform which monitors fusion between two proteoliposomes has proven effective in dissecting the functions of individual protein components. This approach has been used to demonstrate that SNARE proteins are the core fusion machinery (Weber et al., 1998; Yoon et al., 2006), that Syt1 is the Ca2+-sensor (Lee et al., 2010; Kyoung et al., 2011), that Cpxs are the clamping agent for spontaneous fusion (Schaub et al., 2006) and that Munc18 is part of the core fusion machinery (Shen et al., 2007). It has also been shown that Munc13 is a critical component for quality-controlling t-SNAREs to be ready for productive membrane fusion (Liu et al., 2016). However, when fusion between two single proteoliposomes was analyzed with single molecule technique, it turned out that the speed of membrane fusion was in the time scale of several seconds (Lai et al., 2013), as much as four orders of magnitude slower than the physiological rate. Such slow speed raises concerns of some critical, missing components which are not incorporated in the assay or whether the proteoliposome fusion assay does not faithfully reproduce vesicle fusion in vivo. Some suspect that the discrepancy might be due to the tight membrane curvature of proteoliposomes which may not closely mimic the relaxed curvature of the plasma membrane. Others wonder if our long-standing dogma that SNAREs are the core membrane fusion machinery is valid (Wickner and Rizo, 2017).

In this work, we expand the proteoliposome-to-SBL fusion assay by successfully incorporating Syt1. We observe a more than 50-fold increase in the number of membrane fusion events in the presence of Ca2<sup>+</sup> when compared with those without Syt1. Most importantly, more than ∼55% of all vesicle fusion occurs within our instrumental time limit, 20 ms, after docking to the bilayer surface. Further analysis reveals that Syt1 binding to t-SNAREs prior to Ca2<sup>+</sup> clamps spontaneous fusion. However, this pre-binding leads to a failure to respond to Ca2<sup>+</sup> in promoting synchronized membrane fusion. Thus, our results show that there may be productive and non-productive pathways for Syt1 in supporting fast membrane fusion. We then suggest possible mechanisms whereby Syt1 might be steered to the productive pathway. Importantly, the improved membrane fusion assay provides new opportunities to test the mechanistic models for Ca2+-triggered exocytosis in a time scale ever closer to the physiological one.

### RESULTS

### SNAREs Are Capable of Driving Fast Membrane Fusion

Previously, fast fusion between v-SNARE-reconstituted vesicles (v-vesicles) and t-SNARE-reconstituted supported bilayers (t-SBL) was successfully demonstrated (Liu et al., 2005; Domanska et al., 2009; Karatekin et al., 2010; Kiessling et al., 2013; Stratton et al., 2016). This environment displayed the fusion kinetics that better mimicked what was observed in vivo. The results suggested that SNAREs alone, without the help of any auxiliary proteins, are capable of driving sub 25 ms membrane fusion.

To investigate SNARE-mediated single vesicle-to-SBL fusion in our hands, we prepare the t-SBL with PEGylated liposomes following the method previously reported by Karatekin et al. (2010). It has been shown that the PEGylated lipids provide a hydrated cushion of ∼4 nm in thickness between the SBL and the imaging surface which allows for critical dynamic movement of transmembrane proteins (Karatekin and Rothman, 2012). Once the SBL is properly formed we place the slide under the total internal reflection fluorescence (TIRF) microscope and start video acquisition as we introduce the v-vesicles (**Figure 1A**). The v-vesicles contain lipophilic cationic fluorescent dye 1,1′ dioctadecyl-3,3,3′ ,3′ -tetramethylindocarbocyanine perchlorate (DiI) (excitation at 549 nm, emission at 565 nm). Thus, we are able to monitor individual vesicle docking events by counting the immobilized fluorescent spots in the TIRF video (**Figure 1B**). The fusion of the docked v-vesicles onto the SBL is identified by the characteristic two-dimensional diffusion of the fluorescent lipid dyes (**Figure 1C**). With this setup, we are able to monitor the docking rate, the docking-to-fusion delay, and the fusion efficiency in real-time (**Figures 1D–G**).

We first examine the rate of v-vesicles docking onto the t-SBL. We find, under our experimental conditions, that ∼20 vesicles dock in our viewing area (∼55 × ∼110µm) during a 1 min video recording (**Figure 1E**). It is well-established that SNARE complex formation requires VAMP2, syntaxin 1A, and SNAP-25, and even a single missing component fails to elicit docking and fusion. Thus, as a control, we examine the vvesicles docking onto t-SBLs prepared without SNAP-25 or without both SNAP-25 and syntaxin 1A. In either case, the vesicles fail to associate with the SBL. Furthermore, to test if vesicle docking is exclusively SNARE-mediated we pre-incubate t-SBL with soluble VAMP2 without the transmembrane domain (VpS). VpS has been frequently used as a competitive inhibitor for SNARE-mediated vesicle docking. We observe no docking in the presence of VpS as well, indicating that the v-vesicle docking to the SBL is SNARE-mediated. Because Ca2+-sensor Syt1 is not included at this time, we expect that Ca2<sup>+</sup> should not affect either docking or membrane fusion. Indeed, when the v-vesicles premixed with 500µM Ca2<sup>+</sup> are injected into

FIGURE 1 | SNAREs are capable of mediating fast membrane fusion. (A) Schematics for the single vesicle-to-SBL fusion assay. (B) Screen shot (55 × 55µm) of the video recording. The red box indicates a docked vesicle. (C) Time lapse of a single membrane fusion event; each image is 20 ms apart. The red arrow indicates the time delay between docking and the start of fusion. The time delay here is 40 ms. (D) Time traces of slow (upper) and fast (below) membrane fusion events are shown. (E) Docking events with SNAREs only (red) and SNAREs with 500µM Ca2<sup>+</sup> (gray). The controls with t-SBL deactivated by VpS (magenta), with t-SBLs constructed without SNAP-25 (blue), and without both SNAP-25 and syntaxin 1A (green) are also shown. Representative traces of cumulative counts of docking against time (upper) and the number of docked vesicles during the 1 min observation period. We observed a total of 62, 63, 5, 4, and 1 docking events for SNAREs, SNAREs/Ca2+, SNAREs/VpS, w/o SNAP-25, and w/o t-SNAREs, respectively. (F) A representative histogram of docking-to-fusion delay distribution with only SNAREs in the absence (upper) and presence of Ca2<sup>+</sup> (below). We observed a total of 23 and 20 number of fusion events in the absence and presence of Ca2<sup>+</sup> from three independent video recordings, respectively. Each frame is 20 ms apart. (G) The number of fusion events occurring within the 1st frame (fast fusion) and the sum of those occurring thereafter (delayed fusion) within a 1 min recording. For (E) and (G) the data are shown as means ± SD. Statistical significance was assessed by Student's t-test (\*\*\*p < 0.005; NS, no significant difference; n = 3 independent experiments).

the flow chamber we do not observe any apparent changes in docking kinetics when compared with the results without Ca2<sup>+</sup> (**Figure 1E**).

Next, we take a closer look into the individual docked vvesicles to find out if membrane fusion happens. We analyze three independent recordings and count a total of 23 fusion events out of 62 docked v-vesicles (∼35% fusion efficiency) and 20 fusion events out of 63 docked v-vesicles (∼32% fusion efficiency) in the absence and presence of Ca2+, respectively (**Figure 1F**). Furthermore, approximately half of the fusion events occur within the first 20 ms after docking for both cases (**Figure 1F and G**). Thus, our results show that SNAREs alone are capable of mediating fast vesicle fusion within 20 ms.

### Syt1 Increases the Probability of Fast Membrane Fusion in the Presence of Ca2<sup>+</sup>

Syt1 harbors tandem Ca2+-binding C2 domains that are tethered to the vesicle membrane via a transmembrane helix (Perin et al., 1991; Chapman, 2008). Previously, in their seminal study, Sudhof and coworkers used the gain of function and the loss of function mutants of Syt1 to demonstrate that the main function of Syt1 as a Ca2+-sensor is to increase the probability of fast membrane fusion in response to the Ca2<sup>+</sup> signal (Fernández-Chacón et al., 2001).

Having observed that SNAREs alone can mediate fast membrane fusion with the SBL platform, we now ask if Syt1 and Ca2<sup>+</sup> together enhance the membrane fusion probability as Sudhof and coworkers observed in the neuron. If so, how much enhancement does it have?

To answer these questions, we reconstitute both VAMP2 and Syt1 into the vesicles in a 1:1 ratio. In order to follow through from docking to membrane fusion in a continuous time frame, the vesicles are premixed with Ca2<sup>+</sup> and the mixture is injected into the flow cell prepared with t-SBL. The vesicle concentration here is set to be equal to that in the previous section, in the absence of Syt1, for the proper comparison (**Figure 2A**).

We first examine the effect of Syt1 and Ca2<sup>+</sup> on docking. It is established that Syt1 assists docking either by its t-SNARE interaction (Rickman et al., 2004; Loewen et al., 2006) or by direct binding to negatively charged lipids (Schiavo et al., 1996; Kim et al., 2012). Remarkably, we observe more than ∼540 docking events during our 1 min video recordings with 500µM Ca2+, a ∼25-fold enhancement over those in the absence of Syt1 (**Figures 2B,C**, **Supplementary Video 1**). As for membrane fusion, we find that ∼70% of the docked v-vesicles exhibit membrane fusion, which is a 2-fold enhancement when compared with the percentage with SNARE only. Among those, ∼55% of the fusion population display docking-to-fusion delay shorter than 20 ms, similar to the case with SNAREs only. In direct comparison with the fusion population observed with v-vesicles without Syt1, we find that Syt1 and 500µM Ca2<sup>+</sup> together elicit over a ∼50-fold increase in both fast and total fusion population (**Figure 2D**). Intriguingly, however, most of the enhancement of fast membrane fusion by Syt1 and Ca2<sup>+</sup> stems from the ∼25-fold enhancement of docking while there is only a ∼2-fold increase in the fusion-to-docking ratio.

It is often recognized that lipid mixing does not necessarily report the formation of a fusion pore. To make sure that the fusion pore is indeed formed, we prepare v-vesicles that are loaded with both lipid-reporter DiD (excitation at 644 nm, emission at 665 nm) and the content-reporter sulforhodamine B (SRB) (excitation 565 nm, emission at 586 nm) (Kiessling et al., 2013; Stratton et al., 2016) (**Supplementary Figure 1A**). We are able to monitor both lipid mixing and content release via simultaneous excitation with red (640 nm) and green (532 nm) lasers. We find that v-vesicles showing the lipid diffusion, characteristic to lipid mixing lose the SRB signal instantaneously at the onset of lipid mixing (**Supplementary Figure 1B** left), while the v-vesicles showing no lipid diffusion maintain their SRB signal (**Supplementary Figure 1B** right). This implies that lipid mixing and fusion pore opening occur simultaneously within our experimental time resolution of 20 ms. Furthermore, it appears that using DiD instead of DiI or the incorporation of SRB do not significantly affect the overall docking and fusion characteristics (**Supplementary Figures 1C,D**).

#### Dissection of Membrane Fusion Step

In the experiments presented in the previous section, we premixed Ca2<sup>+</sup> with the v-vesicles. This experimental preparation mimics the pre-binding of Ca2<sup>+</sup> to Syt1 prior to Syt1's binding to SNAREs and the target membrane. However, in reality, it is possible that Syt1 pre-binds to SNAREs in the absence of Ca2<sup>+</sup> in preparation for the Ca2<sup>+</sup> influx. In fact, it is shown that Syt1 has the capacity to bind the binary t-SNARE complex in the absence of Ca2<sup>+</sup> (Rickman et al., 2004; Loewen et al., 2006). To explore the outcomes of this alternative situation, we examine docking and fusion in the absence of Ca2<sup>+</sup> (**Figure 3A**) and investigate whether Ca2<sup>+</sup> can trigger membrane fusion of a priorily docked vesicles (**Figure 3B**).

When we examine vesicle docking in the absence of Ca2<sup>+</sup> we observe a ∼6-fold increase in docking compared to the results from the vesicles with only VAMP2 (**Figure 3C**). This enhancement factor falls significantly short of those observed when both Ca2<sup>+</sup> and Syt1 are present, indicating that Ca2<sup>+</sup> plays a role in vesicle docking.

To our surprise, no apparent accumulation of fast membrane fusion is observed within the first 20 ms time period. This is in sharp contrast to that of which was observed for the experiments with only SNAREs. Membrane fusion events among docked vesicles are all scattered randomly over the period of our experimental time (1 min) (**Figures 3D,E**). Nevertheless, the results appear to be somewhat consistent with the proposal that Syt1 might function as a fusion clamp of spontaneous fusion (Chicka et al., 2008).

Even more surprising is that the strong enhancement of membrane fusion by Ca2+, observed when Ca2<sup>+</sup> was pre-bound to Syt1, is completely lost. When we inject Ca2<sup>+</sup> (**Figure 3B**), after any unbound vesicles are washed out with sufficient amount of buffer, we do not detect any synchronization of membrane fusion events in the first 20 ms window. Mere ∼7% of docked vesicles fuses with SBL with 500µM Ca2<sup>+</sup> and fusion events are sporadically scattered in the time frame of our observation (1 min) (**Figures 3F,G**). Thus, the results show that a priori binding of Syt1 to SNARE complexes in the absence of Ca2<sup>+</sup> leads to an irreversible off-pathway that does not bring about Ca2+-triggered synchronized membrane fusion.

As controls, we evaluate docking and fusion of vesicles reconstituted with both VAMP2 and Syt1, with only VAMP2, with only Syt1, onto SBLs with t-SNAREs, with only syntaxin

1,161 fusion events were observed for SNAREs (red, also shown in Figure 1F) and SNAREs/Syt1/Ca2<sup>+</sup> (gray), respectively. For (C) and (D) the data are shown as means ± SD. Statistical significance was assessed by Student's t-test (\*\*\*p < 0.005; NS, no significant difference; n = 3 independent experiments).

1A, with t-SNAREs disabled by VpS, and without any proteins (**Figure 4A**). Experiments with all possible combinations are separately carried out either in the absence or presence of 500µM Ca2<sup>+</sup> in **Figures 4B,C**, respectively. Even in the absence of VAMP2, we observe docking with v-vesicles reconstituted with only Syt1 (**Figure 4B**). This is due to Syt1 binding to either the binary t-SNARE complex, the ternary SNARE complex or phosphatidylinositol-4,5-bisphosphate (PIP2) (Rickman et al., 2004; Loewen et al., 2006). Furthermore, it is established that Ca2<sup>+</sup> increases binding between Syt1 and PIP2 which is shown in **Figure 4C** (Schiavo et al., 1996; Kim et al., 2012). However, we note that these docking events did not lead to membrane fusion.

#### DISCUSSION

SNAREs alone may have the capability to drive fast membrane fusion. SNARE-driven fast fusion in the millisecond time scale was observed in previous studies but the results had caveats: the requirement of SNAP-25 was not met (Liu et al., 2005) and fast membrane fusion was only observed with the addition of a small soluble fragment of VAMP2 (Domanska et al., 2009). More recently, significant fusion enhancement has been observed with Ca2<sup>+</sup> even in the absence of Syt1, adding further confusion to the issue (Kiessling et al., 2013). However, our experiments display fast membrane fusion that meets the requirement of SNAP-25 and we observe strong competitive inhibition by VpS, confirming that fast membrane fusion is strictly SNARE-dependent. Thus, our results confirm that SNAREs are indeed capable of driving fast millisecond-time scale membrane fusion.

When both Syt1 and Ca2<sup>+</sup> are incorporated into our system, membrane fusion is dramatically enhanced by as much as ∼50 fold. Previously, using gain of function and loss of function mutants, Sudhof and coworkers reported the increase of the release probability by Syt1 with Ca2<sup>+</sup> without a significant change in the time scale of the release (Fernández-Chacón et al., 2001). Our results are largely consistent with this in vivo observation and show that, while SNAREs are responsible for the fusion kinetics, Syt1 and Ca2<sup>+</sup> plays a role in dramatically

FIGURE 3 | Dissection of membrane fusion steps. (A) Schematics for the single vesicle-to-SBL fusion assay with Syt1 and without Ca2+. (B) Schematics for Ca2+-triggered single vesicle-to-SBL fusion assay of a priori docked vesicles. (C) Representative cumulative count of docking events against time from one individual recording. The gray trace indicates v-vesicles reconstituted with Syt1 and VAMP2 and the red trace indicates vesicles with just Syt1. Controls without SNAP-25 and with VpS are also shown in blue and green, respectively. (D) A representative histogram of fusion events plotted against docking-to-fusion frame delay (20 ms) from one individual recording. (E) The number of fast and delayed fusion events. The red bars indicate v-vesicles reconstituted with just VAMP2 (23 fusion events, also shown in Figure 1G) and gray bars indicate v-vesicles reconstituted with VAMP2 and Syt1 (90 fusion events). (F) Ca2+-triggered fusion events plotted against time. A representative histogram of frame delay (20 ms) between Ca2<sup>+</sup> injection and fusion within the first 100 s (left) and the cumulative plot during the entire 1 min recording is shown (right). (G) The fusion probability of Ca2+-triggered events within a 1 min recording. A total of 74 fusion events among 1,106 docked vesicles were observed from three independent recordings. For (E,G) the data are shown as means ± SD. Statistical significance was assessed by Student's t-test (\*\*\*p < 0.005; NS, no significant difference; n = 3 independent experiments).

increasing the fusion probability. However, caution is warranted due to our instrumental limitation (20 ms acquisition time). Although we have observed accumulation of fusion events within the first 20 ms with Ca2+, it is still possible that these Ca2+-triggered fusion events happen faster than 20 ms. Further investigations with a faster instrumentation would resolve this issue.

It is however quite intriguing that the enhancement of the membrane fusion events by Syt1 and Ca2<sup>+</sup> are mainly due to the stimulation of vesicle docking although we observed

by Student's t-test (\*\*\*p < 0.005; NS, no significant difference; n = 3 independent experiments).

a factor of two increase of fusion probability for the docked vesicles. Here, one must be careful in interpreting the in vitro data because vesicle docking in in vitro setup closely correlates with an initial step in SNARE complex formation. It is believed that initial preassembly of the N-terminal half of the SNARE motifs occurs concurrently with vesicle docking. Given that, we speculate that Ca2+-activated Syt1 might play a role in promoting the SNARE preassembly at the N-terminal region (Zhou et al., 2015)

Surprisingly, when v-vesicles are allowed to bind to the t-SBL, a priorily in the absence of Ca2+, the results are quite different from the aforementioned results. First, the fast membrane fusion that we observed with only SNAREs, without Syt1, disappears. The results show that Syt1 might act as an inhibitor of fast SNARE-mediated membrane fusion. Such "clamping" effects were previously observed by Chapman and coworkers (Chicka et al., 2008). Moreover, the system did not respond to the subsequent addition of Ca2<sup>+</sup> at all. Thus, it appears that Ca2<sup>+</sup> binding to Syt1 prior to Syt1's interaction with either t-SNAREs or PIP2, residing on the plasma membrane, may be an essential requirement for the productive Ca2+-triggered fast membrane fusion.

Taking these observations into consideration we propose two pathways for Syt1: the productive pathway and nonproductive pathway. In the productive pathway, shown in black arrows in **Figure 5**, Syt1 is able to bind Ca2<sup>+</sup> prior to ternary SNARE complex formation and vesicle docking. This significantly increases the probability of proper SNARE assembly and membrane fusion. However, in the non-productive pathway, shown in red arrows in **Figure 5**, SNAREs and Syt1 form a premature complex in the absence of Ca2+, which results in the failure to respond productively to the Ca2<sup>+</sup> influx.

How would the proposed non-productive pathway be avoided in vivo by Syt1? It is most likely that Syt1 interacts with the binary t-SNARE complex of syntaxin 1A and SNAP-25. In fact, the Ca2+-independent interaction between these two has been documented many times (Rickman et al., 2004; Loewen et al., 2006). One strategy to avoid such an inadvertent interaction would be to keep syntaxin 1A and SNAP-25 separated until Ca2<sup>+</sup> influx. Rizo and coworkers proposed a fusion model where Munc13 and Munc18 choreograph the assembly of the tbinary complex and subsequent SNARE complex formation in a Ca2<sup>+</sup> dependent manner (Ma et al., 2013; Liu et al., 2016). Alternatively, the Syt1 interaction with the binary t-SNARE complex could be sterically avoided. It is also possible that Munc13 bridges the synaptic vesicle and plasma membrane and enable the formation of a primed state (Liu et al., 2016). If this gap is large enough the non-productive pathway could be avoided. Very recently, Brunger and coworkers proposed that a tripartite SNARE-Cpx-Syt1 complex locks membrane fusion until the arrival of Ca2<sup>+</sup> (Zhou et al., 2017). However, it remains unclear if the tripartite complex represents the preor the post-fusion complex, assuming that the SNARE complex is not fully zippered in the pre-fusion state (Lou and Shin, 2016).

One might ask why fast time scales were not observed in the vesicle-to-vesicle fusion assay. This assay has been widely used to gain insights into SNARE-mediated membrane fusion and functions of the accessory proteins (Lee et al., 2010; Kyoung et al., 2011; Ma et al., 2013; Lai et al., 2014). There may be a bilayer curvature issue in this configuration. Specifically, t-vesicles do not represent the planar presynaptic plasma membrane well. It is possible that fusion between two curved lipid membranes is energetically unfavorable in comparison to fusion between a curved vesicle and the planar membrane, warranting further investigation. More importantly, previous attempts focused on triggering fusion of pre-docked vesicle pairs with the Ca2<sup>+</sup> injection. This was due in part to the

assumption that fusion machinery be pre-assembled prior to the Ca2<sup>+</sup> influx. However, our experiments show that such preassembly of the fusion machinery may not be required and the fast millisecond-time scale membrane fusion can happen even though there are no pre-assembly of the fusion machinery and no Cpx.

## MATERIALS AND METHODS

### Plasmid Construct and Site-Directed Mutagenesis

We prepared DNA sequences encoding rat syntaxin 1A (amino acids 1–288 with three native cysteines replaced by alanines), VAMP2 (amino acids 1–116 with C103 replaced by alanines), VpS (amino acids 1-94), SNAP-25 (amino acids 1–206 with four native cysteines replaced by alanines) and inserted them into pGEX-KG vector as N-terminal GST fusion proteins. We also prepared rat Syt1 (amino acids 50-421 with four native cysteines C74, C75, C77 and C79 replaced by alanines and another C82 replaced by serine) and inserted it into pET-28b vector as Cterminal His-tagged proteins. DNA sequences were confirmed by the Iowa State University DNA Sequencing Facility.

## Protein Expression and Purification

Escherichia coli BL21 Rosetta (DE3) pLysS (Novagen) was used to express the recombinant GST fusion proteins VAMP2, SNAP-25, syntaxin 1A and VpS. The cells were grown in LB medium at 37◦C with ampicillin (100µg/mL) until the ∼0.6–0.8 absorbance at 600 nm. The cells were further grown for another 12 h at 16◦C after induction with the addition of Isopropyl β-D-1 thiogalactopyranoside (0.3 mM final concentration). The cell pellets were then harvested via centrifugation at 6,000 × g for 10 min and resuspended in PBS at pH 7.4 containing 2 mM 4- (2-aminoethyl)-benzenesulfonyl fluoride (AEBSF), 2 mM EDTA, and 2 mM dithiothreitol. For the transmembrane proteins, in addition to the resuspension buffer we added 0.5% Triton X-100. The cells were lysed with sonication immersed in an ice bath. The lysed cells were centrifuged at 15,000 × g for 20 min and the supernatant was added onto columns with glutathione-agarose beads for 4◦C for 2 h. The unbound proteins were thoroughly washed off and thrombin (Sigma-Aldrich) was added to cleave off the GST fusion proteins at 4◦C for 16 h. We note that the thrombin cleavage buffer was 50 mM Tris-HCl, 150 mM NaCl, and pH 8.0 and additional 1% n-octyl glucoside was added for transmembrane proteins.

The C-terminal His-tagged Syt1 (amino acids 51–421) was also expressed in E. coli BL21 Rosetta (DE3) pLysS and was purified identically to the aforementioned protocol with the exception of using Ni-NTA beads (Qiagen) instead of GST beads. Also, the elution buffer was prepared with 25 mM HEPES, 400 mM KCl, 500 mM imidazole, and 0.8% OG. All purified proteins were examined with 15% SDS-PAGE and the purity was at least 90%.

### V-vesicle Reconstitution

We used the following lipid molecules to make v-vesicles: 1, 2-dioleoyl-sn-glycero-3-phospho-L-serine (DOPS), 1-palmito yl-2-oleoyl-snglycero-3-phosphocholine (POPC), 1,1′ -diocta decyl-3,3,3′ ,3′ -tetramethylindocarbocyanine perchlorate (DiI, Invitrogen), 1,1′ -dioctadecyl-3,3,3′ ,3′ -tetramethylindodicarbocy anine perchlorate (DiD, Invitrogen), and cholesterol. All lipids in this study were obtained from Avanti Polar Lipids if not otherwise specified.

For v-vesicles with just DiI, we first mixed lipids with molar ratios of 5:54:40:1 (DOPS: POPC: Cholesterol: DiI) and dried it into lipid film using gentle nitrogen gas in a glass tube. The lipid film was stored overnight in a desiccator under house vacuum. The lipid film was resuspended with buffer containing 25 mM HEPES 100 mM KCL pH 7.4. We made liposomes with 10 flash freeze-thaw cycles and large unilamellar vesicles (∼100 nm in diameter) were prepared by extrusion through the polycarbonate filter (Avanti Polar lipids). VAMP2 and Syt1 was mixed with the vesicles such that the lipid to protein ratio was 200:1. The liposome/protein mixture was diluted by adding three times the volume of the protein lipid mixture and then dialyzed in 2 L dialysis buffer at 4◦C overnight.

We also prepared v-vesicles with both lipid-reporter DiD and the content-reporter sulforhodamine B (SRB). These vesicles were prepared identical to the aforementioned v-vesicles with the exception of substituting DiD for DiI. Also, the SRB concentration was kept constant (20 mM) after resuspension of the dried lipid film. Free SRB was removed with the PD-10 desalting column (GE healthcare) after overnight dialysis.

#### T-SBL Preparation

The lipid film for the t-SBL was prepared identically to the case of v-vesicles with the exception of lipid composition. We used the following lipids: DOPS, POPC, phosphatidylinositol-4,5-bisphosphate (PIP2, from porcine brain), and 1,2-dipalmitoyl-sn-glycero-3-phosphoethanolamine-N-[methoxy(polyethylene glycol)-2000] (PEG2000). We mixed the lipids using the following molar ratio of 15:78:2:5 (DOPS: POPC: PIP2: PEG2000). The lipid film was dried and stored in a desiccator overnight. The lipid film was resuspended with buffer containing 25 mM HEPES 100 mM KCL 1% OG pH 7.4. We mixed the resuspended lipid with t-binary complex which was prepared by mixing syntaxin 1A and SNAP-25 (1:1.5 molar ratio) for 30 min at room temperature. The lipid to t-binary complex ratio was 2000:1. Then the liposome/protein mixture was diluted by adding three times the volume of the protein lipid mixture using 25 mM HEPES 100 mM KCL pH 7.4 and then dialyzed overnight at 4◦C in a 2 L beaker.

We subjected the imaging quartz slide to piranha cleaning, boiling mixture of sulfuric and hydrogen peroxide, for 20 min. The slides were rinsed with de-ionized water and placed in a sonicator for 20 min to rid of any residual acid. We quickly assembled the slides and gently injected the t-proteoliposomes into the flow chamber. We let the t-SBL form for 2 h at room temperature, washed out excess liposomes and let the samples settle for 2 h.

#### Single Vesicle-to-SBL Fusion Assay

Once the t-SBL was prepared, we mounted the imaging quartz slide on the total internal reflection (TIR) fluorescence microscope. The TIR angle of the laser (532 nm) was properly adjusted and then we initiated the real-time movie acquisition with an imaging area of ∼55 × ∼110µm and 20 ms time resolution for 90 s. We then gently injected the v-vesicles (prepared according to the experiment) into the flow chamber until the buffer within the flow chamber was exchanged with the v-vesicle solution. The injection pump was promptly stopped in order to prevent any unwanted fusion events due to the flow effect. Once the 90 s movie was obtained, we analyzed a 60 s segment of the movie starting after the injection pump was stopped with our custom built program.

With our custom-built analysis program we monitored the fluorescent DiI signals (DiD and SRB signals for the experiments in **Figures 2E,F**) from the v-vesicles in order to determine docking and fusion. First, we scrolled through the entire 60 s movie clip and handpicked every immobilized spot which represents docked v-vesicles. By selecting the immobilized spot with our program we were able to record the x- and y-coordinates as well as the time at which the vesicle docked on to the t-SBL. Using the x- and y- coordinates, the selected spots were visually marked in order to avoid any recounting or omission.

We then re-examined the movie clip and looked for the characteristic two dimensional diffusion of the fluorescent lipids. The docking-to-fusion delay time was determined by counting the number of frames between immobilization and two dimensional expansion of the fluorescence signal. Thus, the fast fusion events (sub 20 ms fusion delay time) displayed expansion of the fluorescence signal in the absence of a distinct immobilization step. From the analysis process we were able to monitor individual v-vesicles and obtained the docking rate, the fusion percentage, and the fusion delay time between docking and fusion within a movie clip. This analysis process was executed on each movie recording. Three independent recordings were analyzed to obtain the statistics and the statistical significance for each data set.

## AUTHOR CONTRIBUTIONS

JK and YKS designed the experiments. JK performed the experiments. JK and YKS wrote the paper.

### FUNDING

This work was supported by the US National Institutes of Health grant (R01 GM051290) to YKS.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2017.00380/full#supplementary-material

Supplementary Figure 1 | Accompanying Figure 2, single vesicle-to-SBL fusion assay with v-vesicles loaded with both fluorescent lipid dye DiD and content dye SRB. (A) v-vesicles loaded with the fluorescent lipid dye DiD as well as the content dye SRB for simultaneous monitoring of lipid mixing and content release. (B) Representative time traces of DiD lipid dye (red) and SRB content dye (green) of membrane fusion (left) and just docking (right) are shown. (C) Number of docking

events with v-vesicles reconstituted with VAMP2/Syt1 in the presence of 500µM Ca2<sup>+</sup> (left). Due to the rise in background signal from the released content dye, we analyzed the first 30 s of the 1 min recording and reduced the v-vesicle concentration 2-folds compared to the experiments performed in Figure 2. The docking events for v-vesicles reconstituted with just VAMP2 are shown in the middle. We increased the vesicle concentration 20-folds (indicated by the red arrow) in order to observe a comparable number of docking events. A bar graph of the normalized docking events taking the vesicle concentration into

consideration is also shown (right). (D) The percentage of fast and delayed fusion events observed among total docked vesicles (left). A bar graph of the normalized

#### REFERENCES


fusion events taking the vesicle concentration into consideration is also shown (right). A total of 66 and 249 fusion events were observed for SNAREs (red) and SNAREs/Syt1/Ca2<sup>+</sup> (gray), respectively. For (C) and (D) the data are shown as means ± SD. Statistical significance was assessed by Student's t-test (∗∗∗p < 0.005; NS, no significant difference; n = 3 independent experiments).

Supplementary Video 1 | Accompanying Figure 2, a representative movie acquisition of docking and fusion of v-vesicles with VAMP2, Syt1, and Ca2<sup>+</sup> on t-SLB. This movie is a 10 s excerpt from a 1 min recording filmed at 50 frames per second (fps) and played at 50 fps.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2017 Kim and Shin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Accessory and Central α-helices of Complexin Selectively Activate Ca2+ Triggering of Synaptic Exocytosis

Yi Yu1† , Su Chen1† , Xiaoqiang Mo<sup>2</sup> , Jihong Gong<sup>3</sup> , Chenhong Li <sup>1</sup> and Xiaofei Yang<sup>1</sup> \*

<sup>1</sup>Key Laboratory of Cognitive Science, Hubei Key Laboratory of Medical Information Analysis and Tumor Diagnosis & Treatment, Laboratory of Membrane Ion Channels and Medicine, College of Biomedical Engineering, South-Central University for Nationalities, Wuhan, China, <sup>2</sup>Youjiang Medical University for Nationalities, Baise, China, <sup>3</sup>Key Laboratory of Molecular Biophysics of Ministry of Education, College of Life Science and Technology, Huazhong University of Science and Technology, Wuhan, China

Complexins, binding to assembling soluble NSF-attachment protein receptor (SNARE) complexes, activate Ca<sup>2</sup><sup>+</sup> triggered exocytosis and clamp spontaneous release in the presynaptic terminal. Functions of complexin are structural dependent and mechanistically distinct. To further understand the functional/structural dependence of complexin, here we show that the accessory and central α-helices of complexin are sufficient in activation of Ca<sup>2</sup><sup>+</sup> triggered vesicle fusion but not in clamping spontaneous release. Targeting the two α-helices to synaptic vesicle suppresses spontaneous release, thus further emphasizing the importance of curvature membrane localization in clamping function.

#### Edited by:

Jiajie Diao, University of Cincinnati, United States

#### Reviewed by:

Xiaochu Lou, University of Texas Southwestern Medical Center, United States Zhitao Hu, The University of Queensland, Australia Wei Dong, Southwest Medical University, China

#### \*Correspondence:

Xiaofei Yang sunlittlefly@hotmail.com

†These authors have contributed equally to this work.

Received: 05 December 2017 Accepted: 13 February 2018 Published: 26 February 2018

#### Citation:

Yu Y, Chen S, Mo X, Gong J, Li C and Yang X (2018) Accessory and Central α-helices of Complexin Selectively Activate Ca2+ Triggering of Synaptic Exocytosis. Front. Mol. Neurosci. 11:61. doi: 10.3389/fnmol.2018.00061 Keywords: complexin, Ca2+ triggered exocytosis, spontaneous release, SNARE protein, synaptic vesicle

### INTRODUCTION

Neurotransmitter release is mediated by Ca2<sup>+</sup> triggered synaptic vesicle fusion. Like most intracellular membrane fusion, synaptic vesicle fusion mediated by soluble NSF-attachment protein receptor (SNARE) and SM (for ''Sec1-Munc18 like'') proteins (Rizo and Rosenmund, 2008; Südhof and Rothman, 2009). In presynaptic terminals, syntaxin-1, SNAP-25 and synaptobrevin/VAMP2 (vesicle-associated membrane protein) form a tight complex that forces membranes into close proximity, and Munc18-1 binds to the SNARE complex to catalyze fusion (Südhof and Rothman, 2009). During this process, many other proteins have been involved to promote fusion, including Munc13, synaptotagmin and complexin (Cpx; Geppert et al., 1994; McMahon et al., 1995; Wang et al., 2017). Munc13 is critical in maintaining the readily releasable pool size (Richmond et al., 1999; Varoqueaux et al., 2002) and interacts with Munc18-syntaxin complex to explore the syntaxin linker region in initiation of vesicle fusion (Wang et al., 2017). Synaptotagmin, known as Ca2<sup>+</sup> sensors, triggers fusion pore opening via the binding of its C2 domains to phospholipids and SNARE complexes (Fernández-Chacón et al., 2001; Wang et al., 2016). And Cpx functionally cooperates with synaptotagmin in regulating synaptic exocytosis (McMahon et al., 1995; Reim et al., 2001; Zhou et al., 2017).

Cpxs are small (∼130 residues) and evolutionarily conserved SNARE-binding proteins (Yang et al., 2015). The functions of Cpx in activating Ca2+-triggered vesicle release and in clamping spontaneous exocytosis are confirmed not only in C. elegans and Drosophila (Hobson et al., 2011; Martin et al., 2011; Buhl et al., 2013), but also in mice (Maximov et al., 2009; Yang et al., 2010). Among the four mammalian Cpxs, Cpx1 and Cpx2 are the major expressed isoforms in neurons (McMahon et al., 1995). Knockout both Cpx1 and Cpx2 leads the mice lethal (Reim et al., 2001). Here, we focus on the Cpx1 isoform to briefly summarize most relevant results. Cpx can be divided into four domains: flexible N- and C-terminal domains, an accessory and a central α-helices (Chen et al., 2002). A large number of studies of Cpx function with different approaches has discovered different functions for Cpx in synaptic fusion with distinct sequence requirements. The N-terminal domain (residues 1–26) of Cpx plays a role for fast synchronous Ca2<sup>+</sup> triggering of exocytosis (Xue et al., 2007; Maximov et al., 2009). The accessory domain (residues 27–47) clamps spontaneous fusion in neurons (Xue et al., 2007; Yang et al., 2010) and it suppresses Ca2+-independent fusion in vitro systems (Giraudo et al., 2006; Lai et al., 2014; Krishnakumar et al., 2015). The C-terminal domain (residues 71–134) is not only important for both clamping and priming roles (Kaeser-Woo et al., 2012; Dhara et al., 2014; Wragg et al., 2017), but also sensitive to membrane curvature, and it thus localizes Cpx to the synaptic membrane (Wragg et al., 2013; Snead et al., 2014, 2017; Gong et al., 2016). Binding to SNARE complex via the central α-helical domain (residues 48–70; Bracher et al., 2002; Chen et al., 2002) is essential for all functions of Cpx (Maximov et al., 2009; Yang et al., 2013). The differential functional/structural dependence of Cpx strongly argues that these functions are mechanistically distinct.

Although the functions of Cpx and the roles of each domain in the protein have been extensively investigated, the question that what is the minimal functional sequence of Cpx is still unsolved. To address this question, here we investigate all the possible domain combinations of Cpx in Cpx1/2 knockdown (KD) neurons. We found that the central α-helix alone didn't reverse any phenotype in Cpx deficient mouse neurons, while the accessory and central α-helices together rescued the inactivation of evoked neurotransmitter release, but did not clamp the spontaneous mini release. Moreover, we demonstrated that the vesicular localization helped the two α-helices in suppressing spontaneous fusion. Our results thus suggested that the accessory and central α-helices were the minimal functional structure to activate exocytosis.

#### MATERIALS AND METHODS

#### Plasmid Construction

The complexin KD and wild-type complexin (CpxWT) rescue constructs were described previously (Maximov et al., 2009). The mutants contained central α-helix alone (Cpx48–70), accessory and central α-helices together (Cpx27–70), flanking N-terminal domain to central α-helix (Cpx1–27+48–70) or accessory and central α-helices fused with cysteine-string protein-α (CSPα; Cpx27–70−CSP<sup>α</sup> ) were generated by gene synthesis and were cloned downstream of the human ubiquitin promoter in the L309 lentiviral vector.

#### HEK293T Cell Culture

HEK293T cells (CRL-11268, ATCC) were grown in a humidified atmosphere incubator (Thermo) with 5% CO<sup>2</sup> at 37◦C. The culture medium contained Dulbecco's modified Eagle's medium (Gibco), 10% fetal bovine serum, and penicillin-streptomycin (50 µg/ml and 50 µg/ml).

#### Neuronal Culture

The dissociated cortical neurons were dissected from postnatal day 0 (P0) of WT Kunming mice, dissociated by 0.25% trypsin-EDTA digestion for 12 min at 37◦C, plated at 12 mm diameter circular glass coverslips coated with poly-L-lysine (Sigma), and cultured in MEM (GIBCO) supplemented with 2 v/v% B27 (GIBCO), 0.5 w/v% glucose, 100 mg/l transferrin, 5 v/v% fetal bovine serum (GIBCO) and 2 mM Ara-C (Sigma). Wild type mice were fed by mouse facility of South-Central University for Nationalities. No live animals were directly used in this study. All animal procedures were performed in accordance with South-Central University for Nationalities animal use rules and the requisite approvals of animal use committees.

#### Lentiviruses Preparation

Lentiviral expression vectors and three helper plasmids (pRSV-REV, pMDLg/pRRE and pVSVG) were co-transfected into HEK293 cells. The transfections were carried out using the polyethylenimine (PEI, 1 mg/ml in ddH2O) method with the ratio at PEI:pFUGW:pVSVg:RRE:REV = 24:3:1:2:2. The viruscontaining medium was harvested 48 h after transfection and subsequently cleaned with a 3000 g centrifuge and a 0.45 µm filtration (Millipore). The virus was then concentrated by a sucrose-containing buffer (50 mM Tris-HCl, pH 7.4, 100 mM NaCl, 0.5 mM ethylene diaminetetraacetic acid [EDTA]) at a 4:1 v/v ratio and centrifuged at 4◦C. For re-suspension the virus, Phosphate Buffered Saline (PBS) was added to the tube at the fridge with a cover for recovery overnight. All steps were performed under level II biosafety conditions. Neurons were infected with lentiviruses at days in vitro (DIV) 5–6 and analyzed at DIV 13–14.

#### Immunocytochemistry

Various lentiviruses infected mouse cortical neurons were fixed in 4% paraformaldehyde and permeabilized with 0.2% Triton X-100, incubated with anti-complexin (polyclonal; L669) and anti-vGlut1 (monoclonal; N28/9 (Neuromab)) primary antibodies in PBS with 5% BSA, washed, and stained with polyclonal anti-complexin and monocolonal anti-vGlut1 and visualized using Alexa Fluor 488 goat anti-rabbit and Alexa Fluor 546 goat anti-mouse secondary antibodies (Molecular Probes). Images were acquired by using a Nikon C2 confocal microscope equipped with a 60× oil-immersion objective. We measured the average pixel intensities by manually tracing each dendrite, with a >2-fold background signal. Identical settings were applied to all samples in each experiment.

#### Electrophysiological Recordings

Electrophysiological recordings were performed in whole-cell patch-clamp mode at room temperature using concentric extracellular stimulation electrodes. Patch pipettes were pulled from borosilicate glass capillary tubes (World Precision

(B,C) Sample traces (B) and summary graphs of the amplitude (C) of action-potential evoked AMPAR-mediated EPSCs monitored in cultured cortical neurons that were infected with a control lentivirus (control) or lentiviruses expressing complexin shRNAs (GFP) without or with coexpression of CpxWT or Cpx48–70, respectively. (D–F) Sample trace (D) and summary graphs of the frequency (E) and amplitude (F) of miniature excitatory postsynaptic currents (mEPSCs) recorded in neurons as described for (B). Data shown in summary graphs are means ± SEM; numbers of cells/independent cultures analyzed are listed in the bars. Statistical assessments were performed by the Student's t-test comparing each condition to the indicated control experiment (∗∗p < 0.01, ∗∗∗p < 0.001).

Instruments, Inc.) by using a P-97 pipette puller. The resistance of pipettes filled with intracellular solution varied between 3 MOhm and 5 MOhm. After formation of the whole-cell configuration and equilibration of the intracellular pipette solution, the series resistance was kept less than 20 MOhm and then compensated to 8–10 MOhm. The whole-cell pipette solution contained 120 mM CsCl, 10 mM HEPES, 10 mM EGTA, 0.3 mM Na-GTP, 3 mM Mg-ATP and 5 mM QX-314 (pH 7.2, adjusted with CsOH). The bath solution contained 140 mM NaCl, 5 mM KCl, 2 mM MgCl2, 2 mM CaCl2, 10 mM HEPES-NaOH and 10 mM glucose (pH 7.4). In all the recordings, neurons were voltage clamped at −70 mV. Evoked synaptic responses were recorded with a bipolar electrode placed 100–150 mm from the soma of neurons. Synaptic currents were monitored with an EPC10 amplifier (HEKA). Single extracellular stimulus pulses (90 µA, 1 ms) were controlled with a Model 2100 Isolated Pulse Stimulator (A-M Systems, Inc.) for all evoked EPSCs measurements. EPSCs were pharmacologically isolated by adding the GABAA-receptor blockers picrotoxin (100 µM) to the extracellular solution. Spontaneous miniature excitatory postsynaptic currents (mEPSCs) were monitored in the presence of tetrodotoxin (TTX; 1 µM) to block action potentials. The data were digitized at 10 kHz with a 2-kHz low-pass filter. Miniature events were analyzed in Clampfit 10 (Molecular Devices) using the template matching search and a minimal threshold of 5 pA and each event was visually inspected for inclusion or rejection by an experimenter blind to the recording condition.

## Statistical Analysis

Prism 6.01 (Graphpad) was used for statistical tests, all of which are described in figure legends.

## RESULTS

#### Cpx Central α-Helix Itself Is Not Sufficient for Either Clamping Spontaneous Exocytosis or Activation of Ca2+ Triggered Exocytosis

To identify the functions of various Cpx structures, we generated Cpx deficient neurons using shRNA dependent KD of Cpx1 and Cpx2 and performed rescue experiments by expressing a series of Cpx1 mutants that were introduced into the same lentivirus used for the KD. Since all the functions of Cpx were relied on the binding of central α-helix to SNARE complex (Maximov et al., 2009), we first examined whether the central α-helix alone could regulate vesicle release. In the action-potential-evoked exocytosis measurement, consistent with previous results, Cpx deficit caused a significant decrease in the amplitude of evoked release. The wild type (CpxWT) but not the central α-helix mutant (Cpx48–70) of Cpx1 (**Figure 1A**) rescued the decrease induced by Cpx1/2 KD (**Figures 1B,C**), indicating that only the central α-helix couldn't perform the activated role. To test whether the central α-helix of Cpx1 could clamp spontaneous release, we then measured mEPSCs. As expected, Cpx KD increased the frequency of

mEPSCs and the CpxWT reversed the increase as previous reported. But no obvious suppressing effect in the frequency of mEPSCs was observed in Cpx48–70 group (**Figures 1D,E**), suggesting the central α-helix of Cpx1 was not involved in the clamping function as well. The amplitude of mEPSCs was unchanged in all conditions, suggesting that the effects on the mEPSCs frequency are because of presynaptic processes (**Figure 1F**). Taking together, our data demonstrated that the central α-helix of Cpx1 alone couldn't perform any function of Cpx, thus domain combinations were required for Cpx.

#### The Accessory and Central α-helices Work Together to Support the Facilitation of Ca2+ Triggered Neurotransmitter Release

Besides the central α-helix, Cpx1 has an N-terminal domain, a C-terminal domain and an accessory α-helix. We next want to address whether any of the other three domain could rescue the function of Cpx together with the central α-helix domain. Previous research has clarified that the truncated mutant only containing the central α-helix and C-terminal domain of Cpx1 was not observed any function (Maximov et al., 2009). On the other hand, the N-terminal domain of Cpx1 was reported to trigger vesicle fusion in reconstituted systems (Lai et al., 2016). Therefore, we flanked the N-terminal sequence to central α-helix domain (Cpx1–27+48–70 , **Figure 2A**) to investigate whether these two domains could rescue any function of Cpx. Surprisingly, different from in vitro results, Cpx1–27+48–70 could not reverse the decrease in excitatory postsynaptic currents (EPSCs) amplitude measurement induced by Cpx KD (**Figures 2B,C**), indicating only the N-terminal and central α-helix domains were not sufficient to activate exocytosis in neurons. To further analyze the clamp effect of these two domains, the mEPSCs were measured. Our data revealed Cpx1–27+48–70 had no effect in suppressing the increased frequency of spontaneous release caused by Cpx KD either (**Figures 2D–F**), thus further suggesting the N-terminal domain was unable to perform the roles of Cpx with the exist of central α-helix domain. The difference between in vitro and in vivo systems was possible due to more spatial barrier in neurons.

Since neither the N-terminal nor the C-terminal domain could perform any function of Cpx together with the central αhelix domain, we then wondered whether the accessory α-helix could help the central α-helix domain. For this purpose, the mutant expressed the two α-helices was introduced in Cpx deficient neurons. Interestingly, we found that the accessory and central α-helices (Cpx27–70 , **Figure 3A**) expressed together rescued the amplitude of EPSCs (**Figures 3B,C**),

reflecting the activation ability was restored. Moreover, we observed no significant effect in the response time of EPSCs measurements between control and Cpx27–70 mutant (Supplementary Figure S1), indicating the synchronization of EPSCs is not altered by Cpx27–70. On the contrary, the frequency of mEPSCs was not rescued by expressing the Cpx27–70 mutant (**Figures 3D–F**). Therefore, our results identified that the two α-helices of Cpx1 promoted action-potential triggered neurotransmitter release without the help of N- and C-terminal of Cpx. Consistent with previous results, the clamping role of Cpx was unable to be restored by the accessory and central α-helices also indicated Cpx functions are mechanistically distinct.

#### The Vesicular Localization Is Important for Clamping Spontaneous Release

We further wanted to uncover the possible reason for the lack of a clamping activity of Cpx27–70 in our experiments. Previous studies has demonstrated that lacking the C-terminal domain may cause mislocation of Cpx1 and result the increase of mEPSCs frequency (Gong et al., 2016). Targeting Cpx lacking C-terminal sequence mutant to vesicle membrane by fusing the C-terminal palmitoylated sequence of CSPα (Zhou et al., 2013) suppressed spontaneous mini release. To test whether synaptic vesicle localization could rescue the clamping activity of Cpx27–70, we designed a chimera that fused CSPα at the end of Cpx27–70 (Cpx27–70-CSP<sup>α</sup> , **Figure 4A**). To test whether the C-terminal sequence of CSPα could really drive the two α-helices of Cpx1 to vesicle membrane, immunocytochemistry experiments were performed in lentivirus-infected neurons. Consistent with previous results, comparing to the Cpx deficit neurons, the synaptic located Cpx27–70-CSP<sup>α</sup> but not Cpx27–70 significant increased (**Figures 4B–D**), confirming the vesicle located ability of CSPα. However, the synaptic signal of Cpx27–70-CSP<sup>α</sup> was lower than control neurons, suggesting Cpx27–70-CSP<sup>α</sup> only partially rescued the vesicle localization of Cpx1. We then found that Cpx27–70-CSP<sup>α</sup> rescued the decrease of EPSCs amplitude induced by Cpx deficit, however the efficiency is slight but not significant lower than that of CpxWT (**Figures 4E,F**), confirming the two α-helices are able to activate Ca2<sup>+</sup> triggered fast synchronous exocytosis. Again, Cpx27–70-CSP<sup>α</sup> doesn't affect the synchronization of EPSCs by reflecting as an unaltered response time (Supplementary Figure S2). Moreover, comparing to the neurons lacking Cpxs, the increase of mEPSCs frequency was partially but significantly reversed by Cpx27–70-CSP<sup>α</sup> expression (**Figures 4G–I**), consistent with the partially rescued localization caused by Cpx27–70-CSP<sup>α</sup> , confirming synaptic vesicle localization had positive role in clamping spontaneous vesicle fusion for Cpx1. However, different from the mutant only lacking C-terminal domain but fused with CSPα (Gong et al., 2016), the mini frequency in Cpx27–70-CSP<sup>α</sup> expressed neurons was still higher than control neurons (**Figures 4G–I**), thus arguing the N-terminal sequence might have a role in clamping effect of Cpx1 as well.

FIGURE 4 | Synaptic vesicle location is important in suppressing spontaneous release. (A) Schematic structures of wild type (CpxWT) and the C-terminal sequence of CSPα linked accessory and central α-helices of Cpx (Cpx27–70-CSP<sup>α</sup> ). (B–D) Representative images (B) and summary graphs of vGlut1 intensities (C) and ∆F (Fsynapse-Fnon−synapse) of Cpx fluorescence intensities (D) of cultured mouse cortical neurons infected with a control lentivirus (control) or lentiviruses expressing complexin shRNAs (GFP) without or with coexpression of Cpx27–70 or Cpx27–70-CSP<sup>α</sup> , respectively. (E,F) Representative traces (E) and summary graphs of the amplitude (F) of action-potential evoked AMPAR-mediated EPSCs recorded in cultured cortical neurons that were infected with a control lentivirus (control) or lentiviruses expressing complexin shRNAs (GFP) without or with coexpression of CpxWT or Cpx27–70-CSP<sup>α</sup> , respectively. (G–I) Sample trace (G) and summary graphs of the frequency (H) and amplitude (I) of mEPSCs monitored in neurons as described for (E). Data shown in summary graphs are means ± SEM; numbers of cells/independent cultures analyzed are listed in the bars. Statistical assessments were performed by the Student's t-test comparing each condition to the indicated control experiment (∗∗∗p < 0.001) or comparing Cpx27–70-CSP<sup>α</sup> to Cpx knockdown (KD) alone (GFP) experiment (#p < 0.05, ###p < 0.001).

#### DISCUSSION

Through binding to SNARE complex, Cpxs perform at least two physiological functions: activating fast synchronous release and clamping spontaneous vesicle fusion (Yang et al., 2010; Gong et al., 2016; Trimbuch and Rosenmund, 2016). These functions of Cpx selectively depend on distinct modular sequences (Xue et al., 2007; Yang et al., 2010, 2015). The conserved activation function of Cpx is found across all species and different types of Ca2<sup>+</sup> triggered synaptic neurotransmitter release (Reim et al., 2001; Maximov et al., 2009; Kaeser-Woo et al., 2012; Cao et al., 2013; Yang et al., 2013), thus considering as the key role of Cpx. While clamping spontaneous release by Cpx is less conserved among species and varies depending on experimental conditions. Knockout Cpx in mice autaptic neurons claims no obvious clamping effect (Reim et al., 2001; Xue et al., 2007). In contrast, KD Cpx in high density mouse neuronal cultures, as well as knockout Cpx in Drosophila and C. elegans, increases spontaneous release (Huntwork and Littleton, 2007; Yang et al., 2010; Martin et al., 2011; Jorquera et al., 2012; Kaeser-Woo et al., 2012), supporting the fusion clamp model.

In the present study, we made the following observations. First, expressing the central α-helices of Cpx alone is not sufficient to support its function. Second, the accessory α-helix but not the N-terminal domain of Cpx together with central α-helix plays a role in activation of Ca2<sup>+</sup> triggered synchronous release but not in clamping spontaneous release. Third, targeting the two α-helices of Cpx to synaptic vesicle suppresses the spontaneous vesicle fusion, but the role of N-terminal sequence cannot be ignored in clamping.

Our results extend some of the previous studies, for example, confirming the role of C-terminal domain of Cpx in clamping via its synaptic vesicle localization ability and reemphasizing the distinct structural dependent functions of Cpx. But our data are inconsistent with some researches as well. The main difference is how to understand the role of N-terminal domain. Lacking N-terminal domain of Cpx in culture neurons was reported an inactivation in Ca2<sup>+</sup> triggered release (Maximov et al., 2009), and expressing N-terminal domain of Cpx independently activate Ca2<sup>+</sup> triggering fusion in reconstituted single-vesicle fusion assay (Lai et al., 2016). These observations strongly argued an activation function for N-terminal sequence. Previous studies also identified that the N- and C-terminal domain interacted with plasma membrane (Lai et al., 2016) and vesicle membrane (Gong et al., 2016), respectively. Therefore a hypothesis is led that the N- and C-terminal domain keep a balance position for Cpx via their trans-membrane interaction. Lacking N-terminal domain kept Cpx away from plasma membrane, thus abolished the activation ability. While Lacking C-terminal domain pushed Cpx more close to plasma membrane to impair the clamping effect. The absence of both N- and C-terminal domain lost all membrane interaction, thus free the two α-helices to activate vesicle fusion. The activation was slightly decreased when CSPα sequence drove the two α-helices more close to vesicle membrane also supported

#### REFERENCES


this hypothesis. Moreover, flanked expressing the N-terminal domain and central α-helix may not simultaneously bind to SNARE complex and plasma membrane in culture neurons, thus not activate release.

In conclusion, we here reveal the general activation role of Cpx via its accessory and central α-helices, and argue the synergistic effect of N-terminal sequence and synaptic vesicular localization by C-terminal domain in clamping spontaneous synaptic vesicle exocytosis.

#### AUTHOR CONTRIBUTIONS

YY and SC carried out the experiments. YY, SC, XM, JG and CL analyzed the data. XY contributed to the planning of the work and wrote the article.

#### FUNDING

This work was supported by the National Natural Science Foundation of China (31670850 to XY and 81403186 to SC).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnmol. 2018.00061/full#supplementary-material


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Yu, Chen, Mo, Gong, Li and Yang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Endophilin A1 Promotes Actin Polymerization in Dendritic Spines Required for Synaptic Potentiation

Yanrui Yang1,2 \* † , Jiang Chen<sup>3</sup>† , Zhenzhen Guo1,2,4, Shikun Deng1,2,4, Xiangyang Du1,4 , Shaoxia Zhu1,2, Chang Ye<sup>3</sup> , Yun S. Shi<sup>3</sup> and Jia-Jia Liu1,2,5 \*

<sup>1</sup> State Key Laboratory of Molecular Developmental Biology, Institute of Genetics and Developmental Biology, Chinese Academy of Sciences, Beijing, China, <sup>2</sup> CAS Center for Excellence in Brain Science and Intelligence Technology, Chinese Academy of Sciences, Shanghai, China, <sup>3</sup> State Key Laboratory of Pharmaceutical Biotechnology and MOE Key Laboratory of Model Animal for Disease Study, Model Animal Research Center, Nanjing University, Nanjing, China, <sup>4</sup> Graduate School, University of Chinese Academy of Sciences, Beijing, China, <sup>5</sup> College of Life Sciences, University of Chinese Academy of Sciences, Beijing, China

Edited by: Jiajie Diao, University of Cincinnati, United States

#### Reviewed by:

Xiaochu Lou, University of Texas Southwestern Medical Center, United States Wei Dong, Southwest Medical University, China Xiaofei Yang, South-Central University for Nationalities, China

#### \*Correspondence:

Yanrui Yang yryang@genetics.ac.cn Jia-Jia Liu jjliu@genetics.ac.cn

†These authors have contributed equally to this work.

> Received: 03 April 2018 Accepted: 09 May 2018 Published: 28 May 2018

#### Citation:

Yang Y, Chen J, Guo Z, Deng S, Du X, Zhu S, Ye C, Shi YS and Liu J-J (2018) Endophilin A1 Promotes Actin Polymerization in Dendritic Spines Required for Synaptic Potentiation. Front. Mol. Neurosci. 11:177. doi: 10.3389/fnmol.2018.00177 Endophilin A1 is a member of the N-BAR domain-containing endophilin A protein family that is involved in membrane dynamics and trafficking. At the presynaptic terminal, endophilin As participate in synaptic vesicle recycling and autophagosome formation. By gene knockout studies, here we report that postsynaptic endophilin A1 functions in synaptic plasticity. Ablation of endophilin A1 in the hippocampal CA1 region of mature mouse brain impairs long-term spatial and contextual fear memory. Its loss in CA1 neurons postsynaptic of the Schaffer collateral pathway causes impairment in their AMPA-type glutamate receptor-mediated synaptic transmission and long-term potentiation. In KO neurons, defects in the structural and functional plasticity of dendritic spines can be rescued by overexpression of endophilin A1 but not A2 or A3. Further, endophilin A1 promotes actin polymerization in dendritic spines during synaptic potentiation. These findings reveal a physiological role of endophilin A1 distinct from that of other endophilin As at the postsynaptic site.

Keywords: endophilin A1, learning and memory, synaptic transmission, synaptic potentiation, actin polymerization, structural plasticity, dendritic spine, AMPAR

### INTRODUCTION

Endophilin A1 (or endophilin 1, EEN1) is a member of the evolutionarily conserved endophilin A family that is expressed almost exclusively in brain (de Heuvel et al., 1997; Ringstad et al., 1997, 2001), featuring an amino-terminal amphipathic helix-Bin/amphiphysin/Rvs (N-BAR) domain with membrane bending and curvature sensing capacities (Farsad et al., 2001; Gallop et al., 2006; Frost et al., 2009; Bai et al., 2010), and a carboxylterminal Src Homology 3 (SH3) domain that binds to a number of protein partners (**Figure 1A**; Gad et al., 2000; Vinatier et al., 2006; Nakano-Kobayashi et al., 2009; Fu et al., 2011; Pechstein et al., 2015; Yang et al., 2015). Previous studies have established roles

for endophilin As in recycling of synaptic vesicles through its interaction with the endocytic proteins synaptojanin, dynamin, and intersectin (Verstreken et al., 2002, 2003; Schuske et al., 2003; Milosevic et al., 2011; Pechstein et al., 2015), and regulation of neurotransmitter exocytosis through its binding to the glutamate transporter VGLUT1 (Weston et al., 2011). In mammalian cells, they also mark and control a clathrin-independent fast endocytic pathway of transmembrane receptors (Boucrot et al., 2015; Renard et al., 2015). Most recently endophilin As were found to be involved in autophagosome formation and protein homeostasis at presynaptic terminals of neuromuscular junctions (NMJ) in Drosophila and mammalian neurons (Murdoch et al., 2016; Soukup et al., 2016).

In hippocampal neurons, all three members of endophilin A family localize to both pre- and post-synaptic sites (Chowdhury et al., 2006; Yang et al., 2015). Although knockout (KO) of individual endophilin A genes in mice does not affect life span and fertility, double knockout (DKO) of endophilin A1 and A2 (or endophilin 2, EEN2) genes causes progressive ataxia and neurodegeneration, and triple knockout (TKO) causes perinatal lethality (Milosevic et al., 2011; Murdoch et al., 2016), suggesting functional redundancy among them in neurons. At the presynaptic terminal, DKO or TKO causes accumulation of clathrin-coated vesicles and impairment in synaptic transmission (Milosevic et al., 2011). Intriguingly, although cell biological studies and electron microscopy analysis of DKO and TKO synapses reveal a role of endophilin As in clathrin uncoating after scission of endocytosed synaptic vesicles at the presynaptic site, a decrease in the amplitude of spontaneous miniature excitatory postsynaptic currents (mEPSC) was detected in TKO neurons (Milosevic et al., 2011), implying changes in the number of the AMPA-type glutamate receptors (AMPARs) in postsynaptic plasma membrane that cannot be explained by their known functions.

At the postsynaptic site, endophilin A2 and A3 (or endophilin 3, EEN3) interact with the immediate early protein Arc/Arg3.1 to enhance endocytic trafficking of the AMPARs that likely contributes to synaptic plasticity and memory consolidation (Chowdhury et al., 2006; Rial Verde et al., 2006). In dendritic spines, membrane protrusions from dendrites that are major postsynaptic sites for excitatory inputs, endophilin A1 interacts with the cytoskeleton regulator p140Cap and regulates spine morphogenesis and synapse formation during early neurodevelopment (Yang et al., 2015). Whether or not postsynaptic endophilin A1 also functions in synaptic plasticity is unclear. Moreover, the physiological function(s) of individual endophilin As in the mammalian central nervous system (CNS) remain elusive.

In this study, we have investigated the postsynaptic function(s) of endophilin A1 using single gene KO mice and mature mouse hippocampal neurons. Endophilin A1 KO mice exhibit significantly impaired contextual fear memory and spatial learning and memory. Hippocampal CA1-specific KO of endophilin A1 in adult animals causes similar memory phenotypes to those of whole brain KO, indicating that its function in the hippocampus CA1 is required for long-term memory. Moreover, ablation of endophilin A1 in CA1 neurons impairs AMPAR-mediated synaptic transmission and long-term potentiation (LTP). We further show that endophilin A1 promotes actin polymerization required for the morphological and functional changes in dendritic spines of cultured hippocampal neurons during chemically induced LTP. These findings uncover a postsynaptic role of endophilin A1 in synaptic plasticity and long-term memory.

### MATERIALS AND METHODS

### Ethics Statement

All animal experiments were approved (approval code AP2013003 and AP2015002) by the Animal Care and Use Committee of the Institute of Genetics and Developmental Biology, Chinese Academy of Sciences. The Nestin-Cre-Tg C57BL/6J mice were obtained from Nanjing Biomedical Institution of Nanjing University (Tronche et al., 1999). The Thy1-EGFP-Tg C57BL/6J mice were obtained from the Jackson Laboratory (Feng et al., 2000). All animals were housed in standard mouse cages at 22–24◦C on a 12 h light/dark cycle with access to food and water freely.

#### Generation of Endophilin A1 Knockout Mice

The targeting vector for EEN1 was obtained from European Mouse Mutant Cell Repository (EuMMCR, PRPGS00060\_A\_A02). The endophilin A1 KO first and EEN1fl/fl C57BL/6J mice were generated at Nanjing Biomedical Institution of Nanjing University. EEN1 CNS-specific KOs were generated by crossing EEN1fl/fl mice with Nestin-Cre-Tg mice. Mice with a limited subset of green fluorescent protein (GFP)-labeled neurons for analysis of spine morphology were generated by crossing Nestin-Cre+/−; EEN1fl/fl mice to Thy1-EGFP+/−; EEN1fl/fl mice. Genotyping of mouse lines was performed by genomic PCR. PCR genotyping of tail prep DNA from offspring was performed with the following primer pairs:

loxPF/loxPR: 5<sup>0</sup> -CAAGGACTCCCAGAGACCTAGCATC-3<sup>0</sup> and 5<sup>0</sup> -GAGATGGCGCAACGCAATTAAT-3<sup>0</sup> [PCR primer locations are shown in **Figure 1B** resulting in a PCR product of 375 base pairs in EEN1 KO first mice but none in wild-type (WT) mice].

zptF/zptR: 5<sup>0</sup> -GTAAGCGGCTCTAGCGCATGTTCT-3<sup>0</sup> and 5 0 -GCAGGGGCATGTAGGTGGCTCAAC-3<sup>0</sup> (PCR primer locations are shown in **Figure 1B**. Genomic PCR results in a PCR product of 466 base pairs in WT mice, none in EEN1 KO first mice, and of 627 base pairs in EEN1fl/fl mice).

The Nestin-Cre transgene was detected using the following primer pairs:

5 0 -TGCCACGACCAAGTGACAGCAATG-3<sup>0</sup> and 5<sup>0</sup> -ACCAG AGAGACGGAAATCCATCGCTC-3<sup>0</sup> .

FIGURE 1 | EEN1 KO mice are impaired in long-term spatial and contextual fear memory. (A) Domain structure of EEN1. (B) Schematic representation of the EEN1 gene locus, the KO-first, floxed and mutant alleles after homologous recombination. zptF/R and loxPF/R: primer pairs used for genotyping. neo, the neomycin resistance cassette. (C) Immunoblots of tissue lysates from mouse littermates, probed with antibodies to EEN1 and EEN2. β-Actin serves as loading control. 1, hippocampus; 2, cortex; 3, cerebellum; 4, liver. (D) No differences in the body weight of EEN1+/+, EEN1+/−, and EEN1−/<sup>−</sup> mice were detected during development (9 EEN1+/+, 11 EEN1+/−, and 14 EEN1−/−). (E–G) No effects of EEN1 KO on the performance in assays of rotarod (E and F) and Y maze (G). Data represent mean ± SEM for each group (18 EEN1+/+, 26 EEN1+/−, and 22 EEN1−/−). (H–K) No effects of EEN1 KO on the social affiliation and sociability (H and I) (Continued)

#### FIGURE 1 | Continued

fnmol-11-00177 May 24, 2018 Time: 15:50 # 4

or social memory and novelty (J and K). Data represent mean ± SEM (9 EEN1+/+, 10 EEN1+/−, and 8 EEN1−/−). (L–S) The Morris water maze test. Shown are escape latency or traveled distance before escaping to the platform among groups in the visible-platform training (L and M), escape latency, and traveled distance before escaping to the platform in the invisible-platform training (N and O), number of crossing with the 1.5× platform area over 35 days after training and the swim trace 7 days after training in the probe test (P and Q), the swim trace and recall ability following training once again on day 35 (R and S). Red circle indicates position of the platform. Data represent mean ± SEM (9 EEN1+/+, 10 EEN1+/−, and 8 EEN1−/−), <sup>∗</sup>p < 0.05, ∗∗p < 0.01. (T and U) Contextual fear conditioning. Shown are levels of freezing behavior after 24, 48, and 72 h from contextual fear training, and levels of freezing when animals were exposed to a novel context. Data represent mean ± SEM (9 EEN1+/+, 11 EEN1+/−, and 10 EEN1−/−), ∗∗p < 0.01.

The Thy1-EGFP transgene was detected using the following primer pairs:

5 0 -TCTGAGTGGCAAAGGACCTTAGG-3<sup>0</sup> and 5<sup>0</sup> -CGCTGA ACTTGTGGCCGTTTACG-3<sup>0</sup> .

#### Constructs, Viruses, and Stereotaxic Injection

The pAOV-CaMKIIα-EGFP-2A-EEN1 construct was generated by cloning EEN1 cDNA amplified from pCMV-Tag2B-EEN1 into pAOV-CaMKIIα-EGFP-2A. All other constructs used in this study (EEN1-LentiGFP, pCMV-Tag2B-EEN1, pCMV-Tag2B-p140Cap, pCMV-Tag2B-EEN1 Y343A, and LifeActmCherry) were described previously (Yang et al., 2015). Viral particles of adeno-associated virus (AAV) carrying pAOV-CaMKIIα-EGFP-2A-Cre, pAOV-CaMKIIα-EGFP-2A-EEN1, or the control construct pAOV-CaMKIIα-EGFP-2A-3FLAG were purchased from Obio Technology (Shanghai) Corp. Ltd. (Shanghai, China).

For viral injection, 8-week-old mice were anesthetized with isoflurane (1–2% mixed with oxygen) and placed in a stereotaxic apparatus. After being sterilized with iodophors and 75% (vol/vol) alcohol, the scalp was incised along the midline between the ears. Holes were drilled in the bilateral skull. The coordinates of viral injection relative to bregma were as follows: 2.0 mm posterior, 1.8 mm lateral, and 1.4 mm ventral. Using a microinjection system (World Precision Instruments), viral particles carrying pAOV-CaMKIIα-EGFP-2A-Cre, pAOV-CaMKIIα-EGFP-2A-EEN1, or vector (1 µl, 2.0 × 10<sup>12</sup> viral genomes/ml) were injected in the hippocampal CA1 region at a rate of 0.125 µl/min, the needle was kept in place for 5 min before withdrawal, the skin was sutured, and the mice were placed beside a heater for recovery (Barbash et al., 2013).

#### Antibodies

The following antibodies were obtained from commercial sources: goat anti-endophilin A1 (S-20) and endophilin A2 (E-15), mouse anti-synaptophysin (SYP) (D-4) and mouse anticortactin (sc-55588) from Santa Cruz Biotechnology (Santa Cruz, CA, United States); rabbit anti-endophilin A1 (Synaptic Systems GmbH, Germany); rabbit and mouse anti-GFP (MBL598, D153- 3), rabbit and mouse anti-RFP (PM005 and M155-3) which recognize DsRed and mCherry from Medical & Biological Laboratories (Naka-ku, Nagoya, Japan); mouse anti-MAP2 (MAB3418, Chemicon, CA, United States); mouse anti-PSD95 (75-028) for immunofluorescence staining (NeuroMab, Davis, CA, United States); mouse anti-PSD95 for western blotting (BD Biosciences, San Diego, CA, United States); mouse anti-GluA1 (MAB2263, Millipore, Billerica, MA, United States); rabbit anti-FLAG M2 (F7425), mouse anti-α-tubulin (T9026), and mouse anti-β-actin (A5441) (Sigma–Aldrich, St. Louis, MO, United States). Rabbit anti-p140Cap was described previously (Yang et al., 2015). Secondary antibodies for immunofluorescence staining were from Molecular Probes (Invitrogen, Carlsbad, CA, United States).

#### Histology

Adult mice were anesthetized with 1% sodium pentobarbital and transcardially perfused with normal saline followed by 4% paraformaldehyde (PFA) in 0.01 M phosphate-buffered saline (PBS). Mouse brain was dissected out and post-fixed with 4% PFA/PBS for 4 h at 4◦C. Fixed brain was incubated with PBS + 20% sucrose overnight and then PBS + 30% sucrose overnight. The brain was stored at −80◦C until usage. Thirtymicron cyrosections were cut using cryostat and mounted on the slide-glass for immunostaining.

For LacZ staining, slide-glass was incubated with 1 mg/ml X-gal in the staining buffer supplemented with 5 mM potassium ferricyanide and 5 mM potassium ferrocyanide overnight at 37◦C. Stained samples were washed with PBS three times then dehydrated in ethanol of ascending purity (50, 75, 90, and 100%, 2 min each). Slides were mounted on Permount and stored at room temperature (RT) (Kokubu and Lim, 2014).

For immunostaining of brain sections, floating 30-µm-thick slices were rinsed with PBS and permeabilized in 0.4% Triton X-100 in 0.01 M PBS. Cyrosections were blocked with 1% BSA in PBS containing 0.4% Triton X-100 for 1 h at RT, then incubated with primary antibodies overnight at 4◦C. Appropriate secondary antibodies conjugated with Alexa Fluor 488, Alexa Fluor 555, or Alexa Fluor 647 were used for detection. Sections were then incubated with DAPI for nuclear staining for 1 h at RT. Following rinsing, cyrosections were mounted on gelatin-coated slides and covered with coverslip with mounting medium. Confocal images were collected using the Spectral Imaging Confocal Microscope Digital Eclipse C1Si (Nikon, Tokyo, Japan) with a 10× Plan Apochromat DIC N1 0.45 objective or 40× Plan Apochromat VC NA 1.40 oil objective (Yang et al., 2015).

#### Behavioral Analyses

Ten-week-old male animals were used for behavioral analyses.

#### Rotarod

Motor coordination and balance were assayed with an accelerating rotarod (Ugo Basile, Italy). Mice were placed

on a slowly rotating drum for 1 min at 4 rpm for three times to habituate, then the rod accelerated gradually from 4 to 40 rpm over a period of 5 min. The latency and the velocity to fall off the rod were recorded.

#### Y-Maze Spontaneous Alternation

The Y-maze apparatus is made of three identical arms at 120◦ angle with respect to each other. Mouse was put in the center of the maze and allowed to freely explore its three arms for 6 min. Alternations were defined as successive entries into each of the three arms as on overlapping triplet sets (i.e., ABC, BCA, . . .). Percentage of spontaneous alternations was defined as the ratio of actual (= total alternations) to possible (= total arm entries−2) number of alternations × 100.

#### Social Interaction

The social interaction test was based on the method described by Kaidanovich-Beilin et al. (2011). Briefly, mouse was placed into the middle chamber and habituated for 5 min. Then the walls between chambers were removed to allow the mouse to freely explore the three chambers with two empty wire containment cups placed in the middle of both side chambers for the first 10 min. Then "stranger 1" mouse was placed inside cup located in one of the side chambers for a second 10 min. For a third 10 min, "stranger 2" mouse was placed inside cup located in the opposite side chamber. Direct contact between the subject mouse and the containment cup, or stretching of the body of subject mouse in an area 3–5 cm around the cup was counted as an active contact. Duration and number of direct (active) contacts between the subject mouse and the containment cup housing or not housing the mouse for each chamber individually were monitored by a centrally placed video camera and analyzed with an automated video tracking software (the Anilab System, AniLab Software and Instruments Co., Ltd.).

### Morris Water Maze

The water maze procedure was similar to previously established protocols (Bromley-Brits et al., 2011) with minor modifications. The water tank is a 120 cm diameter circular pool. Cues with different shapes are pasted on the wall of the tank above water surface in four different directions. A circular black curtain around the tank eliminates competing environmental cues. Nontoxic white tempura paint was used to opacify the water, which was maintained at 19–23◦C. For the visible trial, a flag was placed on the platform to increase its visibility, then the flag was removed and additional water was added to the pool to submerge the platform which was kept in fixed position to 1 cm below the water surface. Acquisition training was then performed for 8 or 11 days and four trials per day with different waterentering site (at north, south, east, and west positions adjacent to the pool wall). During each trial, mouse must learn to use cues to navigate a path to the hidden platform within 90 s. If they failed to locate the platform within time, they were gently guided to it, and kept on it for 10 s. The escape latency (the average value of time duration from entering water to finding the platform of four trials per day) and traveled distance were calculated for each mouse. After acquisition training, the hidden platform was removed and probe testing was performed with one trial each day for 5 days or one trial with 2 or 7 days interval for 35 days at the distal water-entering site away from the platform. A 1.5× platform circle area where the platform was placed was monitored. The number of crossing the 1.5× platform circle area of each mouse within 60 s was analyzed. For recall training, mouse was placed in the same pool without platform to examine memory extinguishment at least 1 month after training. Similarly, the numbers of crossing 1.5× area of each mouse within 60 s were analyzed. Afterward, the platform was placed back to pool and recall training was performed for 1 day with one trial at the farthest water-entering site away from the platform. The second day, the platform was removed again and mouse was placed in pool at the farthest water-entering site away from the platform. The number of crossing 1.5× area of each mouse was analyzed. The mouse trajectory in the pool was monitored and analyzed with an automated system (Smart 3.0, Panlab SMART video tracking system).

### Contextual Fear Conditioning

Mice were trained in a standard fear conditioning apparatus (Harvard Apparatus Ltd., Holliston, MA, United States). They were allowed to explore freely for 2 min. A 2 s, 0.9 mA foot shock (unconditioned stimulus) was delivered and mice stayed in the chamber for 30 s. Mice were re-exposed to the same chamber for 2 min on the second, the third, and the fourth day. After 3 h on the fourth day, mice were exposed to a novel chamber. Freezing was scored and analyzed automatically using FREEZING software (Harvard Apparatus Ltd., Holliston, MA, United States), with thresholds set to give agreement with blinded human observation.

### Electrophysiology in Slice Cultures

Hippocampi of postnatal day 6–8 (P6-8) EEN1fl/fl mice were isolated in the ice cold dissection solution [MEM (Gibco, 12360-038) with 25 mM HEPES (Gibco, 12360-038), penicillin– streptomycin (Gibco, 15140-122), and 10 mM Tris, pH 7.2]. The isolated hippocampi were sliced to 400 µm sections with tungsten filament slicer (Siskiyou, MX-TS). Sections were cultured with medium containing 50% MEM, 25% HBSS (24020-117), 25% heat-inactivated horse serum (Gibco, 16050-122), 1 mM L-glutamine (Gibco, 35050-061), 1% penicillin–streptomycin (Gibco, 15140-122), 12 µg/ml ascorbic acid, and 1 µg/ml insulin, and supported by sterile 30-mm diameter, porous (0.4 µm), transparent, and low protein-binding membrane (Millicell-CM, Millipore, Billerica, MA, United States). The slices were infected with AAV-GFP-2A-Cre for 24 h in culture. Experiments were done 2–3 weeks after AAV infection. Slices were maintained in artificial cerebrospinal fluid (ACSF, in mM, NaCl 119, KCl 2.5, NaH2PO<sup>4</sup> 1, NaHCO<sup>3</sup> 26, CaCl<sup>2</sup> 2.5, MgCl<sup>2</sup> 1.3, glucose 11) supplemented with 10 µM 2-chloroadenosine to dampen epileptiform activity, and GABA receptors were blocked with picrotoxin (PTX, 0.1 mM) and bicuculline (Bic, 0.01 mM), in a solution saturated with 95% O2/5% CO2. CA1 pyramidal

cells were visualized by infrared differential interference contrast microscopy. The internal solution contained (in millimolar) CsMeSO<sup>4</sup> 115, CsCl 20, HEPES 10, Na3-GTP 0.4, Na2-ATP 4, EGTA 0.6, QX-314 5, and spermine 0.1. Cells were recorded with 4- to 6-M borosilicate glass pipettes, following stimulation of Schaffer collaterals (SC) with concentric biopolar electrode (FHC, CBBRC75) placed in stratum radiatum at the CA1 region. All paired recordings involved simultaneous whole-cell recordings from one GFP-positive neuron and one neighboring GFP-negative neuron. GFP-positive neurons were identified by epifluorescence microscopy. Series resistance was monitored and not compensated, and cells in which series resistance was above 30 M or varied by 25% during recording session were discarded. Synaptic responses were collected with the Multiclamp 700B amplifier and Digidata 1550 data acquisition system (Axon Instruments), filtered at 2 kHz, digitized at 10 Hz. The stimulus was adjusted to evoke a measurable, monosynaptic EPSC in the control cell. AMPAR-mediated responses were isolated by voltage-clamping the cell at −70 mV, whereas NMDAR responses were recorded at +40 mV and amplitudes measured at 150 ms after stimulation to avoid contamination by AMPAR current.

#### Electrophysiology in Acute Slices

EEN1fl/fl mice within 24 h after birth were injected with hightiter AAV stock carrying pAOV-CAMKIIα-GFP-2A-Cre (AAV-GFP-2A-Cre) (about 1 ∼ 5 × 10<sup>13</sup> IU/ml). Newborns were anesthetized on ice for 2–3 min and then mounted in a custom ceramic mold before being injected with about 10 nl of viral solution at seven sites targeting the hippocampus at each cerebral hemisphere with microsyringe (Sutter Instrument) and a beveled glass injection pipette. Injected pups were returned to home cage and used for recording 2–3 weeks afterward. Transverse 350 µm hippocampal slices were cut from viral injected EEN1fl/fl mice on a Leica vibratome (VT1000 S) in high sucrose cutting solution containing (in mM): KCl 2.6, NaH2PO<sup>4</sup> 1.25, NaHCO<sup>3</sup> 26, CaCl<sup>2</sup> 0.75, MgCl<sup>2</sup> 7, sucrose 211, glucose 10. Freshly cut slices were placed in an incubating chamber containing ACSF, and recovered at 32◦C for about 90 min before recording. The slices were perfused with ACSF containing PTX/Bic and saturated with 95% O2/5% CO<sup>2</sup> in whole-cell LTP experiments. CA1 pyramidal cells were voltage-clamped at −70 mV and AMPAR EPSCs were evoked by stimulation at SC. LTP was induced by stimulating SC axons at 2 Hz for 90 s while clamping the cell at 0 mV, after recording a stable 3- to 5-min baseline, but no more than 6 min after breaking into the cell (Granger et al., 2013; Diaz-Alonso et al., 2017). To minimize run-up of baseline responses during LTP, cells were held cell-attached for about 1–2 min before breaking into the cell.

### Primary Neuronal Culture and Transfection

Primary neuronal cultures from hippocampi were prepared as described previously (Banker and Goslin, 1988). Briefly, Hippocampi were dissected from P0 C57BL/6J mice, dissociated with 0.125% trypsin in Hank's balanced salt solution without Ca2<sup>+</sup> and Mg2<sup>+</sup> at 37◦C for 20 min, triturated in DMEM, 10% F12, and 10% fetal bovine serum. Hippocampal neurons were plated on poly-D-lysine-coated coverslips in 24-well plates at a density of 2 × 10<sup>4</sup> cells/well. The medium was replaced with the serum-free Neurobasal (NB) media supplemented with 2% B27 supplement and GlutaMAX (Gibco, Invitrogen, Carlsbad, CA, United States) 4 h after plating. Half of the media were changed every 3 days until use.

For neuronal morphology and immunofluorescence staining, neuronal transfections were performed using Lipofectamine LTX according to the manufacturer's instructions (Invitrogen, Carlsbad, CA, United States) on 12–14 days in vitro (DIV) after plating. Briefly, DNA (1.0 µg/well) was mixed with 1 µl PLUS reagent in 50 µl NB medium, then mixed with 2.0 µl Lipofectamine LTX in 50 µl NB medium, incubated for 20 min, and then added to the neurons in NB at 37◦C in 5% CO<sup>2</sup> for 1 h. Neurons were then rinsed with NB and incubated in the original medium at 37◦C in 5% CO<sup>2</sup> for 4–5 days. For co-transfection, neurons were transfected with 1.0 µg of DNA consisting of two plasmids (0.50 µg each).

### Chemical LTP Stimuli

Neurons were treated with glycine (200 µM) in Mg2+-free extracellular solution (mM: 125 NaCl, 2.5 KCl, 2 CaCl2, 5 HEPES, 33 glucose, 0.2 glycine, 0.02 bicuculline, and 0.003 strychnine, pH 7.4) for 10 min. Neurons were then kept in extracellular solution without glycine for 30 min (Park et al., 2006; Fortin et al., 2010).

#### Immunofluorescence Staining, Image Acquisition, and Analysis

For surface GluA1 labeling, neurons were fixed for 7 min at RT in PBS containing 4% PFA/4% sucrose, rinsed with PBS, blocked with 10% normal goat serum in PBS for 30 min, and incubated with mouse anti-GluA1 (anti-N terminus) antibodies in PBS with 1% normal goat serum overnight at 4◦C, followed with appropriate fluorescence-conjugated secondary antibodies. Neurons were then permeabilized with 0.4% Triton X-100 for 30 min at RT followed by labeling with other primary antibodies. For all other labeling, neurons were fixed in 4% PFA/4% sucrose in PBS at RT for 15 min. After blocking with 1% BSA in PBS containing 0.4% Triton X-100 for 1 h at RT, neurons were incubated with primary antibodies for 1 h at RT or overnight at 4◦C, and appropriate secondary antibodies conjugated with Alexa Fluor 488, Alexa Fluor 555, or Alexa Fluor 647 were used for detection.

Confocal images were collected using the Spectral Imaging Confocal Microscope Digital Eclipse C1Si (Nikon, Tokyo, Japan) with a 100× Plan Apochromat VC NA 1.40 oil objective. Images were z projections of images taken at 0.15–0.2 µm step intervals. The number of planes, typically 5–7, was chosen to encompass the entire dendrite from top to bottom.

The procedure for morphometric analysis of dendritic protrusions was described previously (Yang et al., 2015). GFP

or DsRed was used as a cell-fill. The GFP- or DsRed-labeled dendrites or spines were outlined manually. Maximum image projections used in measurements of spine density, spine head area, or fluorescent signal intensity were rendered with the NIS-Elements AR software (Nikon, Tokyo, Japan) from confocal z-series images. Dendritic segments 40–120 µm from the neuronal cell body were selected for analysis. To quantify enrichment of F-actin in spines, we measured the mean intensity of LifeAct–mCherry fluorescence within the center of spines and normalized each measurement with the fluorescent signal along the adjacent dendritic shaft. The GFP- or mCherry-labeled dendrites or spines were outlined manually. All quantitative analyses were done with the NIS-Elements AR software.

To examine spine number and morphology in vivo, spines located at the apical dendrites or basal dendrites of dorsal hippocampal CA1 and CA3 regions were imaged in 100 µm-thick coronal sections from Thy1-EGFP-Tg mice. Z-stack Images (0.25 µm step intervals) were captured at 100× magnification with 4× optical zoom and reconstructed by maximum projections with the NIS-Elements AR software (Nikon, Tokyo, Japan). Spines were examined over 1000 µm dendritic segments from more than 15 dendrites for each mouse from two experimental groups.

To measure changes in the density and morphology of dendritic spines, and surface levels of GluA1 in spines upon chemical LTP, the spine number, spine head area, or fluorescence intensity of GluA1 in spines of glycine-treated EEN1+/<sup>+</sup> or EEN1−/<sup>−</sup> neurons was subtracted by the average of those without glycine application.

#### PSD Fractionation

Cytosol, synaptosome, synaptosomal membrane, and PSD fractions from mouse brain were prepared using a small-scale modification of the procedure previously described (Carlin et al., 1980; Cho et al., 1992; Jaworski et al., 2009). In brief, hippocampi were homogenized on ice using 20 strokes of a Teflon-glass homogenizer in 1 ml of HEPES-buffered sucrose (0.32 M sucrose, 4 mM HEPES, pH 7.4) containing freshly added protease inhibitors, then homogenized with a syringe (20– 30 strokes), followed by centrifugation at 800–1000 × g for 10 min at 4◦C to remove the pelleted nuclear fraction (P1). The supernatant (S1, a.k.a Homogenates or Total) was centrifuged at 10,000 × g for 15 min to yield the crude synaptosomal pellet (P2). P2 was washed once in 1 ml HEPES-buffered sucrose and lysed by hypoosmotic shock in 900 µl ice-cold 4 mM HEPES, pH 7.4 plus protease inhibitors, and homogenized by pipetting and rotating for 30 min at 4◦C. The lysate was centrifuged at 25,000 × g for 20 min to yield supernatant (S3, crude synaptic vesicle fraction) and pellet (P3, lysed synaptosomal membrane fraction). To prepare the PSD fraction, P3 was resuspended in 900 µl of ice-cold 50 mM HEPES, pH 7.4, 2 mM EDTA, protease inhibitors, and 0.5% Triton X-100, rotated for 15 min at 4◦C and centrifuged at 32,000 × g for 20 min to obtain the PSD pellet. PSD pellets were resuspended in 60 µl icecold 50 mM HEPES pH 7.4, 2 mM EDTA plus protease inhibitors.

#### Western Blotting

For expression analysis, tissues were dissected from C57BL/6J mice and rinsed once in ice-cold PBS, pH 7.4. Frozen samples were homogenized in lysis buffer (50 mM Tris–Cl pH 7.4, 150 mM NaCl, 5 mM EDTA, 0.5% Triton X-100) supplemented with protease inhibitors. Twenty micrograms of protein was loaded in each lane for subsequent Western blot analysis. Immunoblots were imaged with an Epichemi3 Darkroom system (UVP BioImaging Systems, Upland, CA, United States). For densitometric analysis, immunoreactive bands were quantified using ImageJ (National Institutes of Health, Bethesda, MD, United States).

#### Statistical Analysis

All data are presented as the mean ± SEM. GraphPad Prism 5 (GraphPad Software, LaJolla, CA, United States) was used for statistical analysis. For two-sample comparisons vs. controls, Student's t-test was used. One-way analysis of variance with a Dunnett's multiple-comparison or Newman–Keuls multiple comparison hoc test was used to evaluate statistical significance of three or more groups of samples. A p-value of <0.05 was considered statistically significant.

#### Results

#### Pan-Neural Knockout of Endophilin A1 Causes Impairment in Spatial and Contextual Fear Memory

Endophilin A1 KO mice generated by removing the first exon have normal life span and show no obvious phenotypes such as neurodegeneration (Milosevic et al., 2011), suggesting functional redundancy or compensatory effects among the endophilin A family members. To investigate physiological functions of endophilin A1 that might be distinct from that of A2 and A3, we generated a reporter KO of endophilin A1 (KOfirst or KO), in which the endophilin A1 gene (EEN1) was inactivated by insertion of a lacZ-Neomycin cassette before exon 3 (**Figure 1B**). Immunoblotting analysis showed that, compared with WT littermates (EEN1+/+), endophilin A1 expression was dramatically reduced in the brain of KO mice (EEN1−/−), whereas no decrease in endophilin A2 levels was detected (**Figure 1C**). Consistent with the previous study (Milosevic et al., 2011), endophilin A1 KO mice were viable and had normal body weight with no obvious phenotypic defects (**Figure 1D**).

To investigate role(s) of endophilin A1 in brain function, we analyzed the motor coordination, working memory, social interaction, and hippocampal-dependent memory of the EEN1+/+, EEN1+/<sup>−</sup> (heterozygous, or HET), and EEN1−/<sup>−</sup> mice. The performance of KO mice was indistinguishable from that of the WT and HET littermates in the rotarod test of motor coordination (**Figures 1E,F**). KO mice also exhibited normal working memory in the Y-maze test (**Figure 1G**) and normal social interaction in the three-chamber test (**Figures 1H–K**). In the Morris water maze test, KO mice spent a similar latency

to escape and traveled similar swimming distances before escaping onto the visible platform (**Figures 1L,M**). In the training phase, all three genotypes improved their performance with repetitive training, with WT and HET mice showing a steeper learning curve and reaching a better performance level in fewer training days (**Figures 1N,O**). After prolonged training, KO mice could catch up to the performance level of WT and HET mice (**Figures 1N,O**). In the probe trial, although all three genotypes showed similar platform crossings 24 h after training, while WT and HET mice still remembered the platform location 1 month later, KO mice forgot the platform location within 7 days (**Figures 1P,Q**). Moreover, KO mice were unable to recall the platform location after training once again (**Figures 1P,R,S**). These data indicate that endophilin A1 deficiency impairs spatial learning and long-term retention of spatial memory after training has been finished.

Next we examined the effect of endophilin A1 KO on contextual fear memory. Mice were trained to associate a particular environment with a mild foot shock and tested for fear memory. Compared with WT and HET, although KO mice displayed similar levels of freezing when they were tested 24 h after training, their freezing behavior decreased significantly at 72 h (**Figures 1T,U**). Notably, all three genotypes exhibited similar levels of freezing when exposed to a novel context (**Figures 1T,U**), indicating that short-term memory was intact in KO mice. Together these data indicate that endophilin A1 deficiency impairs long-term retention of contextual fear memory.

#### Expression of Endophilin A1 in Hippocampal CA1 Is Required for Spatial and Contextual Fear Memory

To investigate mechanisms underlying memory deficits caused by ablation of endophilin A1 in the CNS, first we examined its expression pattern in the brain with the endophilin A1 promoter-driven LacZ reporter (**Figure 1B**). β-Galactosidase staining of sagittal brain sections revealed enrichment of signals in CA1 and CA3 pyramidal cells in the hippocampus (**Figure 2A**). Consistently, immunofluorescence staining showed that endophilin A1 was highly expressed in hippocampal CA1 and CA3 but not in CA2 or dentate gyrus (DG) in WT mice, and its expression was dramatically downregulated in KO mice (**Figure 2B**). As both spatial memory and contextual fear memory are hippocampal-dependent functions that involve the SC-CA1 synapses (Tsien et al., 1996; Nakazawa et al., 2002; Daumas et al., 2005; Zelikowsky et al., 2012), the high expression of endophilin A1 in the hippocampus prompted us to investigate whether endophilin A1 in the CA1 region is required for spatial and contextual fear memory. To determine whether ablation of endophilin A1 in CA1 recapitulates the behavioral phenotypes of pan-neural KO mice, and to avoid disruption of endophilin A1 function during early neurodevelopment, first we generated floxed EEN1 alleles (EEN1fl/fl ) by mating the KO first mice to FLPeR (flipper) mice (**Figure 1B**). We then applied bilateral stereotaxic injections of adeno-associated viral vectors encoding both enhanced GFP and the Cre recombinase (AAV-GFP-2A-Cre) or only EGFP (AAV-GFP) under the CaMKIIα promoter into hippocampal CA1 regions of 8-week-old EEN1fl/fl mice (**Figure 2C**). Immunostaining of brain sections from mice 21 days after injection indicated that endophilin A1 expression was dramatically reduced in AAV-GFP-2A-Creinfected pyramidal neurons in CA1 (**Figure 2D**).

Next we examined hippocampus-associated memory of mice 21 days after viral injection. Based on the results obtained from the Morris water maze test on WT, HET, and KO mice, we shortened the training phase to 8 days and tested the memory retention of animals with probe trial for 5 days and memory recall 1 month after training. Compared with AAV-GFP-injected mice, AAV-GFP-2A-Creinjected mice were slightly retarded in learning the position of the hidden platform (**Figures 2E–H**). They also exhibited rapid forgetting of platform position in the probe trial (**Figures 2I,J**). Moreover, they exhibited defect to recall the platform location after training once again (**Figures 2I,K,L**). Further, ablation of endophilin A1 in the CA1 region also caused impairment of contextual fear memory. Compared with AAV-GFP-injected mice, AAV-GFP-2A-Cre-injected mice displayed lower levels of freezing behavior 72 h after training (**Figures 2M,N**). Collectively these data indicate that endophilin A1 in the hippocampal CA1 region is required for the retention of spatial and contextual fear memory in mature animals.

#### Expression of Endophilin A1 in CA1 of Mature KO Mouse Brain Is Sufficient for Restoration of Spatial and Contextual Fear Memory

To determine whether the memory impairment in EEN1−/<sup>−</sup> mice is irreversible or can be reversed by expression of endophilin A1 in adult brain, we injected AAV vectors coexpressing endophilin A1 and GFP (AAV-GFP-2A-EEN1) or GFP only into bilateral hippocampal CA1 regions of 8-week-old endophilin A1 KO mice (**Figure 3A**). Immunostaining of brain sections verified endophilin A1 expression in neurons infected with AAV-GFP-2A-EEN1 (**Figure 3B**).

The hippocampus-associated memory of mice was assayed 21 days after viral injection. All injected mice spent a similar time and traveled similar swimming distances before escaping onto the visible platform in the water maze test (**Figures 3C,D**). The performance of KO mice injected with AAV-GFP-2A-EEN1 was similar to that of WT mice in both the training phase and the probe trial (**Figures 3E–H**). Moreover, memory recall of the platform position exhibited by AAV-GFP-2A-EEN1-injected KO mice was indistinguishable from that by WT mice (**Figures 3G,I,J**). Consistently, compared with WT and AAV-GFP-injected KO mice, the AAV-GFP-2A-EEN1-injected KO mice did not exhibit any deficits in contextual fear memory (**Figures 3K,L**). Together, these data indicate that expression of endophilin

FIGURE 2 | EEN1 expression in the hippocampal CA1 region of adult mice is required for long-term spatial and contextual fear memory. (A) LacZ staining in the sagittal brain section of 10-week-old EEN1 KO-first (EEN1−/−) mice. Right panel is magnification of the hippocampus. Scale bars, 1 mm in the left panel and 100 µm in the right panel. (B) Immunofluorescence staining of EEN1 in hippocampal CA1, CA2, CA3, and DG regions of 10-week-old EEN1+/<sup>+</sup> and EEN1−/<sup>−</sup> mouse brains. Scale bar, 20 µm. (C) AAV virus was stereotaxically injected into the CA1 regions of EEN1fl/fl mice to express GFP alone or Cre and GFP. Shown are GFP signal and DAPI labeling of nuclei 21 days after viral injection. Scale bar, 1 mm. (D) Immunofluorescence staining of EEN1 in brain slices 21 days after injection of AAV virus into the CA1 region of EEN1fl/fl mice. Lower panels are magnification of the boxed areas. Scale bar, 100 µm. (E–L) The Morris water maze test. Shown are escape latency or traveled distance before escaping to the platform in the visible-platform training (E and F), escape latency, and traveled distance before escaping to the platform in the invisible-platform training (G and H), number of crossing with the 1.5× platform area and the swim trace 5 days after training in probe test (I and J), the swim trace and recall ability following training once again 1 month after training (K and L). Data represent mean ± SEM (11 GFP, 13 Cre), <sup>∗</sup>p < 0.05, ∗∗p < 0.01. (M and N) Decrease in freezing behavior 72 h after contextual fear training in the Cre virus-injected group. Data represent mean ± SEM (9 GFP, 12 Cre), ∗∗p < 0.01.

FIGURE 3 | EEN1 overexpression in hippocampal CA1 of adult KO mice restores long-term spatial and contextual fear memory. (A) AAV virus was stereotaxically injected into the CA1 regions of EEN1−/<sup>−</sup> mice to express GFP alone or EEN1 and GFP. Shown are confocal images of GFP signal and DAPI labeling of nuclei 21 days after viral injection. Scale bar, 1 mm. (B) Immunofluorescence staining of EEN1 in CA1 neurons of brain slices 21 days after injection of AAV virus into the CA1 region of EEN1−/<sup>−</sup> mice. Right panels are magnification of the boxed areas. Scale bar, 100 µm. (C–J) The Morris water maze test. Shown are escape latency or traveled distance before escaping to the platform in the visible-platform training (C and D), escape latency and traveled distance before escaping to the platform in the invisible-platform training (E and F), number of crossing with the 1.5× platform area and the swim trace 5 days after training in probe test (G and H), the swim trace and recall ability following training once again 1 month after training (I and J). Data represent mean ± SEM (11 EEN1+/<sup>+</sup> + GFP, 10 EEN1−/<sup>−</sup> + GFP, 14 EEN1−/<sup>−</sup> + EEN1), <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. (K and L) Restoration of freezing behavior in EEN1-overexpressed EEN1−/<sup>−</sup> mice, compared with control mice. Data represent mean ± SEM (11 EEN1+/<sup>+</sup> + GFP, 10 EEN1−/<sup>−</sup> + GFP, 14 EEN1−/<sup>−</sup> + EEN1), <sup>∗</sup>p < 0.05, ∗∗∗p < 0.001.

FIGURE 4 | Morphological and functional alterations of EEN1-deficient hippocampal neurons. (A) Confocal micrographs showing spines on GFP-positive apical or basal dendrites of pyramidal cells in hippocampal CA1 or CA3 regions of 10-week-old Thy1-GFP;EEN1fl/fl and Thy1-GFP;nestin-Cre;EEN1fl/fl mice. Scale bar, 5 µm. (B–E) Quantification of spine density or spine head area in A (CA1 apical/basal: 42/34 cells, 2691/3240 spines, total length of dendrites >1500 µm and CA3 apical/basal: 32/31 cells, 2742/2258 spines, total length of dendrites >1000 µm for Thy1-GFP;EEN1fl/fl . CA1 apical/basal: 44/37 cells, 3666/2282 spines, total length of dendrites >1200 µm, and CA3 apical/basal: 39/32 cells, 3998/3002 spines, total length of dendrites >1500 µm for Thy1-GFP;Nestin-Cre;EEN1fl/fl ). Data represent mean ± SEM, <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. (F–G) Shown are the average of average spine head area of each CA1 and CA3 pyramidal cells, respectively. Data represent mean ± SEM, n = 31–44, <sup>∗</sup>p < 0.05, ∗∗∗p < 0.001. (H) Immunoblotting of indicated proteins in homogenates (total) and PSD fractions from hippocampi of EEN1+/+, EEN1+/−, and EEN1−/<sup>−</sup> mice. (I–M) Quantification of protein levels in H, normalized to levels of EEN1+/<sup>+</sup> mice. Data represent (Continued)

#### FIGURE 4 | Continued

fnmol-11-00177 May 24, 2018 Time: 15:50 # 12

mean ± SEM, N = 4, <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. (N) Dual recording analysis of AMPAR-mediated synaptic responses. Scatter plots show amplitudes of AMPAR-eEPSCs for single pairs (open circles) and mean ± SEM (filled circle). The current amplitudes of infected neurons (Cre) were plotted on the ordinate and those of the control neurons (Ctrl) were plotted on the abscissa. Inset shows sample current traces from a pair of infected (green) and control (black) neurons. Scale bar, 100 pA and 20 ms. Bar graph shows mean ± SEM of AMPAR amplitudes represented in the scatter plots. Control, 132.1 ± 16.3 pA; Cre, 105.1 ± 16.8 pA, n = 20, <sup>∗</sup>p = 0.030, paired t-test. (O) NMDAR-mediated eEPSC. Currents were recorded at +40 mV. Data were collected at 150 ms after electric stimulation (arrow), when the AMPAR-mediated EPSC had completely decayed. Scale bar, 50 pA and 50 ms. The NMDA eEPSCs were 35.7 ± 4.6 pA for control and 37.1 ± 5.7 pA for Cre-expressing neurons. n = 16, p = 0.74, paired t-test. (P) Paired-pulse recording of AMPAR eEPSCs. Two identical stimulus pulses were delivered in an interval of 50 ms and AMPAR eEPSCs were recorded at −70 mV. Left were sample traces of eEPSCs from a pair of infected and control neurons. Scale bar, 100 pA and 25 ms. The paired-pulse ratio (PPR) was the enhancement of the second eEPSC relative to the first eEPSC. Bar graph shows mean ± SEM of PPRs. Control, 1.55 ± 0.05; Cre, 1.53 ± 0.10, n = 10, p = 0.81, paired t-test. (Q) LTP was severely reduced in EEN1-deficient neurons. Relative amplitudes of AMPAR-eEPSCs (mean ± SEM) in control and Cre-expressing neurons before and after a whole-cell LTP-pairing protocol (arrow), Vm = 0 mV, 2 Hz SC stimulation for 90 s, normalized to average eEPSC amplitude prior to LTP induction. n = 10 decreased to 6 cells for control and n = 9 decreased to 6 cells for Cre-expressing neurons, respectively. Right shows sample traces of control and Cre before and 40 min after pairing. Sale bar: 100 pA and 20 ms. The potentiation ratio is significantly decreased in EEN1-deficient neurons 40 min after LTP induction, p = 0.020, t-test.

A1 in the hippocampal CA1 region of mature brain is sufficient to rescue the memory deficits exhibited in EEN1−/<sup>−</sup> mice.

#### Loss of Endophilin A1 Impairs Postsynaptic Function and Long-Term Potentiation of Hippocampal CA1 Pyramidal Cells

To investigate changes in synaptic functions caused by ablation of endophilin A1 at the cellular level, first we examined neuronal morphology in the hippocampal CA1 and CA3 regions by crossing Nestin-Cre+/−; EEN1fl/fl mice with Thy1-EGFP+/−; EEN1fl/fl mice and imaging sparsely labeled neurons in brain sections by confocal microscopy (**Figure 4A**). Quantitative analysis indicated that there was a decrease in the size of spines of both the apical and basal dendrites of CA1, and apical dendrites of CA3 pyramidal cells in EEN1−/<sup>−</sup> mice (**Figures 4B–D,F,G**). There was also a slight increase in spine density of CA1 apical dendrites (**Figure 4B**), possibly an in vivo compensation for the reduction in spine size. No statistically significant change in the number and size of spines was detected in CA3 basal dendrites (**Figures 4E,G**).

We also determined whether ablation of endophilin A1 causes changes in the levels of postsynaptic proteins in mouse hippocampi. Consistent with previous findings that endophilin A1 recruits p140Cap and cortactin, its downstream effectors, to dendritic spines (Yang et al., 2015), immunoblotting of the PSD fraction detected significant decreases in their amount in KO mice (**Figures 4H–K**). In agreement with the mild phenotype in neuronal morphology of CA1 and CA3 pyramidal cells, no significant changes in the postsynaptic levels of PSD95 in the whole hippocampus were detected (**Figures 4H,L**). Notably, the amount of endophilin A2, another member of the endophilin A family, in the PSD fraction was similar in EEN1+/+, EEN1+/−, and EEN1−/<sup>−</sup> mouse hippocampi (**Figures 4H,M**), indicating that loss of endophilin A1 did not cause its upregulation at the postsynaptic site. Unfortunately, we were unable to test expression of endophilin A3 because of lack of reliable antibodies for immunoblotting.

Spine size and synaptic strength are significantly correlated. As morphological changes in dendritic spines were detected in apical dendrites of CA1 pyramidal cells, which receive excitatory inputs from CA3, next we sought to determine whether synaptic transmission is impaired in EEN1−/<sup>−</sup> CA1 neurons by electrophysiological analysis. To investigate exclusively effect(s) of endophilin A1 ablation in postsynaptic neurons, we eliminated the EEN1 gene in a small subset of CA1 neurons by injection of organotypic hippocampal slice culture from EEN1fl/fl mice with AAV-GFP-2A-Cre (Niu et al., 2017). By simultaneous recording of evoked excitatory postsynaptic currents (eEPSCs) on virus-infected and adjacent uninfected cells, we detected a decrease in AMPAR-mediated synaptic transmission in endophilin A1-deficient neurons (**Figure 4N**), whereas the NMDA-type glutamate receptor (NMDAR)-mediated eEPSCs and the paired-pulse ratio of AMPAR eEPSCs were unaffected (**Figures 4O,P**), indicating that the impairment of synaptic function was not due to reduction of presynaptic glutamate release. Further, we asked whether synaptic potentiation at SC-CA1 pathways was altered in the absence of endophilin A1. To this end, we injected the hippocampal CA1 region of EEN1fl/fl mice with AAV-GFP-2A-Cre at P0 and induced LTP in SC synapses by whole-cell recording of CA1 pyramidal cells in acute slices from virus-injected animals at P14-21 (Granger et al., 2013; Diaz-Alonso et al., 2017). In WT neurons, LTP induction caused a robust increase in EPSC that persisted throughout the 40-min recording period, whereas in AAV-GFP-2A-Cre-infected neurons the magnitude of LTP was significantly lower (**Figure 4Q**). Together these data indicate that both AMPAR-mediated basal transmission and LTP were impaired with removal of endophilin A1 in postsynaptic neurons.

#### Endophilin A1, Not Endophilin A2 or A3, Is Required for the Structural and Functional Plasticity of Dendritic Spines Undergoing Synaptic Potentiation

Long-term potentiation is a form of long-term synaptic plasticity, the cellular correlate of learning and memory. At SC-CA1 synapses, LTP occurs when Ca2<sup>+</sup> influx through the activated NMDARs in the postsynaptic membrane initiates

FIGURE 5 | EEN1 is required for the structural and functional plasticity of dendritic spines. (A) Cultured EEN1+/<sup>+</sup> and EEN1−/<sup>−</sup> hippocampal neurons were transfected with pLL3.7.1 on DIV12-13 to express DsRed as volume marker, fixed and immunostained for EEN1 and DsRed on DIV19. Shown are representative confocal images of dendrites. (B) Quantification of EEN1 fluorescent signals in spines in A, normalized to levels of EEN1+/<sup>+</sup> neurons. Data represent mean ± SEM, n > 10 neurons, >600 spines per group, ∗∗∗p < 0.001. (C) Cultured EEN1+/<sup>+</sup> and EEN1−/<sup>−</sup> neurons co-transfected with DsRed expression construct and Flag vector, and EEN1−/<sup>−</sup> neurons co-transfected with constructs expressing DsRed- and Flag-tagged EEN1, EEN2, or EEN3 on DIV12-13 were treated with glycine to induce chemLTP with or without MK801 pretreatment on DIV18, and immunostained for surface GluA1, Flag, and DsRed. Shown are representative confocal images of dendrites. (D) Quantification of spine density in C. (E) Changes of spine density in C. (F) Quantification of spine head area in C. (G) Changes of spine head area in C. (H) Quantification of surface GluA1 levels in spines in C. (I) Changes of surface GluA1 levels in spines in C. Data represent mean ± SEM in D–I, n > 15 neurons per group, >850 spines per group, <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. Scale bars, 2 µm.

induce LTP with or without MK801 pretreatment on DIV18, followed by immunostaining for surface GluA1, Flag, and DsRed. Shown are representative confocal images of dendrites. (E) Quantification of spine density in D. (F) Changes of spine density in D. (G) Quantification of spine head area in D. (H) Changes of spine head area in D. (I) Quantification of surface GluA1 levels in spines in D. (J) Changes of surface GluA1 levels in spines in D. Data represent mean ± SEM in E–J, n > 15 neurons, >850 spines per group, <sup>∗</sup>p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. (K) EEN1+/<sup>+</sup> neurons co-transfected with LifeAct–mCherry and GFP constructs, and (Continued)

#### FIGURE 6 | Continued

fnmol-11-00177 May 24, 2018 Time: 15:50 # 15

EEN1−/<sup>−</sup> neurons co-transfected with LifeAct–mCherry and GFP or EEN1-GFP constructs on DIV12-13 were treated with glycine with or without MK801 pretreatment on DIV18, followed by immunostaining with antibodies against GFP and mCherry. Shown are representative confocal images of dendrites. (L) Quantification of actin enrichment in dendritic spines in K. Data represent mean ± SEM, n > 12 neurons, >700 spines per group, ∗∗p < 0.01,∗∗∗p < 0.001. (M) EEN1+/<sup>+</sup> and EEN1−/<sup>−</sup> neurons co-transfected with LifeAct–mCherry construct and Flag vector, and EEN1−/<sup>−</sup> neurons co-transfected with LifeAct–mCherry and Flag-tagged p140Cap or EEN1 Y343A constructs on DIV12-13 were treated with glycine on DIV18, followed by immunostaining with antibodies against mCherry and Flag. Shown are representative confocal images of dendrites. (N) Quantification of actin enrichment in dendritic spines in M. (O) Quantification of spine head area in M. Data represent mean ± SEM in N and O, n > 10 neurons, >500 spines per group, ∗∗p < 0.01, ∗∗∗p < 0.001. Scale bars, 5 µm in A and 2 µm in D, K, and M.

downstream signaling cascades, leading to the structural and molecular remodeling of dendritic spines that eventually result in potentiation of AMPAR-mediated synaptic transmission (Herring and Nicoll, 2016). The finding that postsynaptic ablation of endophilin A1 in CA1 neurons impaired wholecell LTP prompted us to investigate whether it functions in the structural and functional plasticity of dendritic spines. To this end, we used a well-characterized pharmacological approach (Park et al., 2006; Fortin et al., 2010) to chemically induce LTP (chemLTP) in mature hippocampal neurons in dissociated culture (**Figures 5A,B**). LTP induction by application of glycine led to a rapid increase in both the number and size of spines as well as surface levels of the AMPAR subunit GluA1 in EEN1+/<sup>+</sup> neurons, which was fully inhibited by the addition of MK801, an NMDAR antagonist (**Figures 5C,D,F,H**). In contrast, not only the activity-dependent increase in spine density was abolished in EEN1−/<sup>−</sup> cells, changes in spine morphology and GluA1 surface expression were also significantly inhibited (**Figures 5C–I**). The impairment in structural and functional plasticity was rescued by overexpression of endophilin A1 but not A2 or A3 (**Figures 5C–I**). Moreover, endophilin A1 overexpression failed to restore plasticity in MK801 treated EEN1−/<sup>−</sup> neurons (**Figures 5C,D,F,H**), indicating that endophilin A1 functions in NMDAR-mediated spine growth and synaptic potentiation. Notably, the absolute activity dependent increase in not only spine number but also spine size in endophilin A1-overexpressing EEN1−/<sup>−</sup> neurons was similar to those in EEN1+/<sup>+</sup> neurons (**Figures 5E,G**). Since overexpression of endophilin A1 caused enlargement of spines in steady state EEN1−/<sup>−</sup> neurons (**Figure 5F**), this result is consistent with previous findings that chemLTP induction causes similar modifications in small and large spines (Kopec et al., 2006). Taken together, these data indicate that endophilin A1, not endophilin A2 or A3, is specifically required for NMDAR-mediated synaptic potentiation of dendritic spines.

#### Endophilin A1 Promotes Actin Polymerization in Dendritic Spines of Hippocampal Neurons Undergoing Synaptic Potentiation

During early stages of synaptic development, endophilin A1 contributes to dendritic spine morphogenesis and stabilization by recruiting p140Cap to spines to promote actin polymerization (Yang et al., 2015). Recent studies show that p140Cap regulates synaptic plasticity through Src-mediated and Citron-N-mediated actin reorganization (Repetto et al., 2014). To explore the mechanistic role of endophilin A1 in synaptic plasticity of postsynaptic neurons, first we asked whether recruitment of p140Cap to dendritic spines by endophilin A1 is regulated by neural activity. Indeed, an increase in p140Cap signal intensity in dendritic spines was detected in glycine-treated EEN1+/<sup>+</sup> neurons, which was attenuated in EEN1−/<sup>−</sup> neurons (**Figures 6A–C**).

Next we asked whether the endophilin A1-p140Cap pathway contributes to the increases in spine number and size and postsynaptic surface expression of GluA1 during chemLTP (**Figure 6D**). Intriguingly, although overexpression of endophilin A1 restored both structural and functional plasticity in EEN1−/<sup>−</sup> neurons, overexpression of p140Cap failed to ameliorate the defects in morphological changes of spines and upregulation of the postsynaptic GluA1 levels (**Figures 6D–J**). Since p140Cap is a downstream effector of endophilin A1, these data suggest that mechanism(s) other than the endophilin A1 p140Cap interaction are required for endophilin A1-mediated synaptic plasticity of spines. Alternatively, the interaction might be spatiotemporally regulated by activity-dependent signals upstream of endophilin A1. Nevertheless, since actin reorganization is crucial for spine plasticity (Hotulainen and Hoogenraad, 2010), next we asked whether endophilin A1 promotes actin polymerization in spines during chemLTP. To this end, we transfected neurons with constructs expressing EGFP and the F-actin probe LifeAct–mCherry. Quantification of the spine:shaft ratio of red fluorescence mean intensity revealed that indeed, the increase in F-actin content in spines was inhibited in glycine-treated EEN1−/<sup>−</sup> neurons, which was restored by overexpression of endophilin A1 (**Figures 6K,L**). Moreover, endophilin A1 overexpression failed to rescue activity-dependent F-actin accumulation in spines of MK801 treated EEN1−/<sup>−</sup> neurons (**Figures 6K,L**), indicating that endophilin A1 promotes actin polymerization via the NMDARmediated signaling pathway. Further, although overexpression of p140Cap fully rescued levels of F-actin in spines of steady state EEN1−/<sup>−</sup> neurons, overexpression of a p140Cap-binding deficient mutant of endophilin A1 (Y343A) (Yang et al., 2015) did not (**Figures 6M,N**). Intriguingly, neither p140Cap nor the endophilin A1 Y343A mutant could restore the increase in F-actin content or the size of spine head in glycinetreated EEN1−/<sup>−</sup> spines (**Figures 6M–O**), suggesting that the endophilin A1-p140Cap interaction is required not only for actin polymerization during spine morphogenesis and maturation, but also for spatiotemporal regulation of actin dynamics crucial for the activity-dependent morphological changes of spines. Collectively, these data indicate that endophilin A1 promotes actin polymerization in dendritic spines during synaptic potentiation.

## DISCUSSION

fnmol-11-00177 May 24, 2018 Time: 15:50 # 16

In this study, we uncovered a postsynaptic role of endophilin A1 in synaptic plasticity distinct from that of endophilin A2 and A3. Specifically, endophilin A1 is required for the physical enlargement and upregulation of AMPAR expression in the postsynaptic membrane of dendritic spines during synaptic potentiation, whereas previous studies have indicated that endophilin A2 and A3 cooperate with the immediate early protein Arc/Arg3.1 to downregulate surface AMPARs by accelerating their endocytosis (Chowdhury et al., 2006). In agreement with its function in synaptic plasticity, KO of endophilin A1 in mouse brain causes deficits in spatial and contextual fear memory, which can be rescued by its overexpression in the hippocampal CA1 region. Whether or not endophilin A1 is also involved in higher brain function(s) that requires brain areas other than the hippocampus remains to be determined.

Endophilin A1 is highly expressed in the CA1 and CA3 regions of the hippocampus. Notably, ablation of endophilin A1 in the hippocampal CA1 region of mature brain is sufficient to cause phenotypes in spatial and contextual fear memory similar to those of pan-neural KO mice. Conversely, expression of endophilin A1 in CA1 of mature brain fully rescues the memory deficits of KO mice. Given that the CA2 subfield is essential for social memory, but is not critical for spatial and contextual memory (Hitti and Siegelbaum, 2014; Mankin et al., 2015), the region-specific expression of endophilin A1 might explain the learning and memory deficits of the KO mice, as both spatial and contextual fear memories involve the CA3–CA1 pathway.

Notably, there is only one endophilin A in Drosophila melanogaster and Caenorhabditis elegans. A role for endophilin A in synaptic vesicle recycling has been established across species (Ringstad et al., 1999; Gad et al., 2000; Verstreken et al., 2002; Schuske et al., 2003). Most recent studies reveal that endophilin A also functions in neuronal activity and stress-induced macroautophagy at presynaptic terminals of NMJ in Drosophila, which mediates protein turnover and is crucial for neuronal homeostasis and survival (Soukup et al., 2016). Endophilin A1 phosphorylated at the Ser75 residue by the kinase LRRK2 (Matta et al., 2012) promotes autophagosome formation by creating highly curved membrane zones in preautophagosomes that serves as docking sites for autophagic factors (Soukup et al., 2016). In mammalian cells, all three endophilin As interact with the E3 ubiquitin ligase FBXO32 that is involved in protein homeostasis, and both endophilin A1 and A2 are needed for autophagosome formation in mouse neurons (Murdoch et al., 2016). Moreover, recent studies also indicate that endophilin As regulate endosomal sorting and trafficking of the BDNF–TrkB neurotrophic signal complex to mediate dendrite development and survival of hippocampal neurons (Fu et al., 2011; Burk et al., 2017). The endophilin A DKO and TKO mice but not single KO mice show ataxia and motor impairment caused by neurodegeneration in the brain (Murdoch et al.,

2016), suggesting functional redundancy of endophilin A family members in autophagosome formation and endosomal sorting.

In higher eukaryotes, however, the biological functions of endophilin A family members at the postsynaptic site are diverse. Removal of endophilin A in Drosophila causes a reduction in the frequency but not amplitude of miniature excitatory junctional potentials (mEJPs) at the NMJ (Verstreken et al., 2002). In contrast, both the frequency and amplitude of mEPSCs of dissociated cultured hippocampal neurons from the TKO mice are lower than the WT animals (Milosevic et al., 2011). As a decrease in the amplitude of mEPSCs indicates impaired synaptic response to neurotransmitter release from a single vesicle, together these findings suggest that the postsynaptic function(s) of endophilin A is required to maintain synaptic function in mammalian neurons. Consistent with previous findings that transient knockdown of endophilin A1 expression in cultured hippocampal neurons causes a decrease in the frequency of mEPSCs (Yang et al., 2015), AMPARmediated basal transmission is impaired in endophilin A1 deficient CA1 neurons. Moreover, ablation of endophilin A1 in CA1 neurons also causes impairment in LTP. Notably, the impaired structural and functional plasticity of dendritic spines of endophilin A1-deficient neurons cannot be rescued by other endophilin A family members, indicating that its role in synaptic plasticity is distinct from those of A2 and A3.

What is the mechanistic role(s) of endophilin A1 in synaptic plasticity? Given that it facilitates actin polymerization by recruiting p140Cap and cortactin to dendritic spines during spine morphogenesis and maturation (Yang et al., 2015), it is conceivable that endophilin A1 also promotes actin polymerization required for formation of new spines and increase in the size of existing spines during synaptic potentiation. Intriguingly, overexpression of p140Cap, its downstream effector, does not rescue the phenotypes of endophilin A1 KO neurons during chemLTP. As the turnover rates and locations of distinct F-actin pools in single dendritic spines are dynamically regulated during synaptic plasticity (Honkura et al., 2008; Frost et al., 2010), these findings suggest that actin polymerization promoted by the p140Cap pathway is not sufficient for restoration of actin cytoskeleton remodeling in spines, and that other regulatory factor(s) acts via endophilin A1 to achieve the spatiotemporal control of molecular events required for the growth and synaptic potentiation of dendritic spines. A recent study reports that calmodulin binds to the N-BAR domains of endophilin A1 and A2 in vitro and promotes the membrane tubulation activity of endophilin A2 in COS7 cells (Myers et al., 2016). Ca2+/calmodulin-dependent activation of calmodulin-dependent protein kinase (CaMKII) plays a central role for the induction of LTP (Herring and Nicoll, 2016). Given that the Ca2+-calmodulin–CaMKII pathway controls signaling cascades that regulate both branched actin polymerization and receptor trafficking at synapses, and that endophilin A1 also functions in promoting actin polymerization in spines, it will be of great interest to explore whether and how endophilin A1 functions downstream of these two master regulators of intracellular signaling during synaptic plasticity.

#### AUTHOR CONTRIBUTIONS

fnmol-11-00177 May 24, 2018 Time: 15:50 # 17

YY, YSS, and J-JL designed the experiments. YY, JC, ZG, SD, XD, SZ, and CY performed the experiments. YY, JC, ZG, and J-JL analyzed the data. YY, YSS, and J-JL wrote the paper.

#### FUNDING

This work was supported by funding from the National Natural Science Foundation of China (31530039 and 31325017 to J-JL, 31571056 to YY, and 31571060 to YSS), National Key

#### REFERENCES


R&D Program of China (2016YFA0500100 to J-JL), Ministry of Science and Technology of the People's Republic of China (2014CB942802 to J-JL, 2014CB942804 and 2015BAI08B00 to YSS), the Fundamental Research Funds for the Central Universities (090314380021 to YSS), and Natural Science Foundation of Jiangsu Province Grant (BK20140018 to YSS).

#### ACKNOWLEDGMENTS

We thank Professor Youming Lu (Huazhong University of Science and Technology) for advice on the Morris water maze test and Dr. Qingfeng Wu (Institute of Genetics and Developmental Biology, Chinese Academy of Sciences) for critical comments on the manuscript.

spectral variants of GFP. Neuron 28, 41–51. doi: 10.1016/S0896-6273(00)00084- 2



stages in clathrin-mediated synaptic vesicle endocytosis. Neuron 24, 143–154. doi: 10.1016/S0896-6273(00)80828-4


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 Yang, Chen, Guo, Deng, Du, Zhu, Ye, Shi and Liu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# New Insights Into Interactions of Presynaptic Calcium Channel Subtypes and SNARE Proteins in Neurotransmitter Release

Rongfang He1,2† , Juan Zhang1† , Yiyan Yu1† , Laluo Jizi <sup>3</sup> , Weizhong Wang<sup>4</sup> \* and Miaoling Li <sup>1</sup> \*

<sup>1</sup>Key Laboratory of Medical Electrophysiology, Ministry of Education, Collaborative Innovation Center for Prevention and Treatment of Cardiovascular Disease, Institute of Cardiovascular Research, Southwest Medical University, Luzhou, China, 2 Infectious Disease Department, The Affiliated Hospital of Southwest Medical University, Southwest Medical University, Luzhou, China, <sup>3</sup>Department of Neurology, Liangshan Hospital of Integrated Traditional and Western Medicine, Xichang, China, <sup>4</sup>Department of Physiology and Center of Polar Medical Research, Second Military Medical University, Shanghai, China

Action potential (AP) induces presynaptic membrane depolarization and subsequent opening of Ca<sup>2</sup><sup>+</sup> channels, and then triggers neurotransmitter release at the active zone of presynaptic terminal. Presynaptic Ca<sup>2</sup><sup>+</sup> channels and SNARE proteins (SNAREs) interactions form a large signal transfer complex, which are core components for exocytosis. Ca<sup>2</sup><sup>+</sup> channels serve to regulate the activity of Ca<sup>2</sup><sup>+</sup> channels through direct binding and indirect activation of active zone proteins and SNAREs. The activation of Ca<sup>2</sup><sup>+</sup> channels promotes synaptic vesicle recruitment, docking, priming, fusion and neurotransmission release. Intracellular calcium increase is a key step for the initiation of vesicle fusion. Various voltage-gated calcium channel (VGCC) subtypes exert different physiological functions. Until now, it has not been clear how different subtypes of calcium channels integrally regulate the release of neurotransmitters within 200 µs of the AP arriving at the active zone of synaptic terminal. In this mini review, we provide a brief overview of the structure and physiological function of Ca<sup>2</sup><sup>+</sup> channel subtypes, interactions of Ca<sup>2</sup><sup>+</sup> channels and SNAREs in neurotransmitter release, and dynamic fine-tune Ca<sup>2</sup><sup>+</sup> channel activities by G proteins (Gβγ), multiple protein kinases and Ca<sup>2</sup><sup>+</sup> sensor (CaS) proteins.

#### Edited by:

Jiajie Diao, University of Cincinnati, United States

#### Reviewed by:

Guo-Xing Zhang, Soochow University, China De-Pei Li, University of Texas MD Anderson Cancer Center, United States

#### \*Correspondence:

Weizhong Wang wangwz68@hotmail.com Miaoling Li limiaolingcc@swmu.edu.cn

†These authors have contributed equally to this work.

> Received: 12 April 2018 Accepted: 30 May 2018 Published: 16 July 2018

#### Citation:

He R, Zhang J, Yu Y, Jizi L, Wang W and Li M (2018) New Insights Into Interactions of Presynaptic Calcium Channel Subtypes and SNARE Proteins in Neurotransmitter Release. Front. Mol. Neurosci. 11:213. doi: 10.3389/fnmol.2018.00213 Keywords: Ca2+ channel subtypes, SNAREs, Ca2+ sensor, active zone, membrane fusion, neurotransmitter release

#### INTRODUCTION

Influx of Ca2<sup>+</sup> through presynaptic calcium channels into presynaptic terminals at active zone is a crucial step in synaptic vesicle exocytosis and rapid neurotransmitter release (Catterall, 2011). Interactions of Ca2<sup>+</sup> channel and soluble N-ethyl-maleimide-sensitive factor attachment protein receptor (SNAREs) complex contribute to reduce the distance between vesicles and the presynaptic membrane (Catterall and Few, 2008). The close distance provides a spatial structure that can ensure triggering of the fast neurotransmitter release within milliseconds of the action potential (AP) arriving at the synaptic terminal (Südhof, 2013; Mochida, 2017). Changes in the kinetic properties of Ca2<sup>+</sup> channels (such as channels open, close, inactivate and so on) directly or indirectly induce modulation of the exocytosis of the synaptic vesicle, and subsequently modulate the release of neurotransmitters in a negative or positive way (Atlas, 2013). Multiple subtypes of Ca2<sup>+</sup> channels are present in the nervous system with diverse physiological functions (Mochida, 2018). Furthermore, a single neuron also contains different types of Ca2<sup>+</sup> channel isoforms. Thus, the channel isoforms play a key role in integral regulation of the synaptic vesicle exocytosis. Until now, it has not been clear how the Ca2<sup>+</sup> channel isoforms coordinate well and accurately regulate fast neurotransmitter release at synaptic terminals. In this mini review, we focus on the molecular structures and regulatory mechanisms of multiple Ca2<sup>+</sup> channel isoforms, and the interactions of Ca2<sup>+</sup> channels and SNAREs involving vesicles fusion and neurotransmitter releases.

#### DIVERSITY OF Ca2<sup>+</sup> CHANNELS IN NEUROUS SYSTEM

The diverse subtypes of Ca2<sup>+</sup> channels display different biological structures and distribution in the nervous system. The diversity of channels corroborates its different physiological functions.

According to the unique electrophysiological and pharmacological properties, voltage-gated calcium channel (VGCC) have been classified into N-, P/Q-, R-, L- and T-type (Ertel et al., 2000). N-, P/Q-, R- and L-type is termed as high-voltage activated Ca2<sup>+</sup> channel, while T-type is low-voltage activated Ca2<sup>+</sup> channel (<−40 mV). High-voltage activated Ca2<sup>+</sup> channels are composed of the pore-forming Cavα1 and four auxiliary subunits (Cavα2, Cavβ, Cavγ and Cavδ; Catterall, 2000), while T-type contains only the Cavα1 subunit (Cavα1G, Cavα1H and Cavα1I; **Figures 1A,B**). The neuronal Cavα1 subunit (190–250 kDa) is the largest and main subunit, which is composed of about 2000 amino-acid residues. The molecular weights of α2, β, γ and δ subunits are 143 kDa, 53–70 kDa, 30 kDa and 24–27 kDa, respectively. The Cavα1 contains four homologous domains (I–IV; **Figure 1C**), and each domain of Cavα1 is comprised of six transmembrane α helices (S1–S6). The transmembrane S5–S6 segments form a p loop, and the S1–S4 segments serve as the voltage sensor (Yu et al., 2005). Diversity of Cavα1 isoforms determine the channel subtypes. Ten different types of Ca2<sup>+</sup> channels have been identified (Yu and Catterall, 2004). The Cavα1 subunit genes are classified as Cav1.1–1.4 (L-type), Cav2.1–2.3 (P/Q-, N-, and R-type) and Cav3.1–3.3 (T-type; **Figure 1B**), each of them belongs to CACNA1x gene families. N-type and P/Q-type Ca2<sup>+</sup> channels are the main Ca2<sup>+</sup> channels in nerve terminals and play an important role in fine-tuning of rapid neurotransmitter release at synaptic terminals (Ariel et al., 2012). The R-type (Cav2.3) Ca2<sup>+</sup> channels are present in the peripheral nervous system (PNS) and central nervous system (CNS). Though R-type Ca2<sup>+</sup> channels are not the main Ca2<sup>+</sup> channels, they are also involved in presynaptic plasticity and neurotransmitter release (Breustedt et al., 2003; Dietrich et al., 2003; Naidoo et al., 2010). T-type Ca2<sup>+</sup> channel present in peripheral, central synapses and neuroendocrine cells, play a key role on tuning of basal neurosecretion near resting potential with a mild stimulation (Lambert et al., 2014).

The auxiliary subunits of Ca2<sup>+</sup> channels include α2, β, γ and δ subunits (**Figure 1A**). The α2 and δ subunits are encoded by the same gene that bind together with disulfide linkage to form α2-δ subunit complex. α2-δ subunit exerts a

**229**

role of increased calcium current and upregulation of gene expression. The interactions of α2-δ subunit with extracellular Cavα1 modulate the binding of divalent cations (Cantí et al., 2005). α2δ-1 isoform is encoded by the gene Cacna2d1. The interaction of α2δ-1 and NMDA receptors significantly increased in neuropathic pain (Dolphin, 2012; Patel et al., 2013). Chen et al. (2018) demonstrate that gabapentin reduces neuropathic pain by inhibiting of the interaction between the C terminus of α2δ-1 and NMDA receptors. The whole β subunit is located in cytoplasm. The functional role of β subunit is to ensure the α1 subunit binding to the plasma membrane and prevent it trafficking to the endoplasmic reticulum. β subunit regulates membrane protein expression and gating of Ca2<sup>+</sup> channels (Arikkath and Campbell, 2003). The γ subunit comprises of four transmembrane α helices, that can slightly reduce Ca2<sup>+</sup> current density and change kinetic properties by interacting with the Cavα1 subunit (Osten and Stern-Bach, 2006). The specific polypeptide toxins from snail and spider venoms, block multiple Ca2<sup>+</sup> channel subtypes. ω-conotoxin GVIA (ω-Cgtx GVIA) blocks the N-type channels irreversibly in central nervous system (CNS) and Peripheral nervous system (PNS; Prashanth et al., 2014). ω-agatoxin IVA (ω-AgaIVA) blocks the Cav2.1 (P-type) and ω-Aga IVA blocks the Cav2.1 (Q-type) with a lower affinity in CNS (Arranz-Tagarro et al., 2014; Ricoy and Frerking, 2014). ω-conotox in MVIIC is a toxin from the venom of marine conus snail, which targets Cav2.1 with high affinity and targets Cav2.2 with low affinity (Catterall et al., 2005). α-conotox in Vc1.1 does not affect Cav2.1 but strongly inhibits Cav2.3 Ca2<sup>+</sup> channels through GABA<sup>B</sup> receptor (Berecki et al., 2014). Cav2.3 Ca2<sup>+</sup> channels were potently blocked by Zn2<sup>+</sup> (IC<sup>50</sup> = 0.78 ± 0.07 µmol/L; Traboulsie et al., 2007). The tetraline derivative of mibefradil and the peptide blocker of scorpion toxin kurtoxin have been evaluated as potential Cav3 Ca2<sup>+</sup> channel inhibitors in CNS and PNS (Chuang et al., 1998).

The diversity of Cavα1 and auxiliary subunits confirms distinct molecular structures, synaptic properties and distributions that are involved in the regulation of various physiology functions in neurotransmitter release. P-/Qtype Ca2<sup>+</sup> channels mediated GABA release in the most of GABA releasing inhibitory neurons (Lonchamp et al., 2009). Glutamate-release is often mediated by integrated interactions of P-/Q- and N-type Ca2<sup>+</sup> channels in the vast majority of glutamatergic cortical and cerebellar synapses (Ladera et al., 2009). Furthermore, P-/Q-type Ca2<sup>+</sup> channels decrease fusion pore stability and trigger vesicle fusion, N-type and L-type Ca2<sup>+</sup> channels slow down fusion pore expansion (Ardiles et al., 2007). In the axon terminal, P-/Q- type Ca2<sup>+</sup> channels are close to the release zone than other Ca2<sup>+</sup> channels in various synapses. As a result, P-/Q- type Ca2<sup>+</sup> channels (Cav2.1) may lead to higher local presynaptic Ca2<sup>+</sup> concentrations and frequently co-localized with synaptotagmin-containing vesicle clusters, whereas the N-type channel (Cav2.2) and R-type channel (Cav2.3) are only partially involved in vesicle clusters (Wu et al., 1999). The significant role of N-type Ca2<sup>+</sup> channels is involved in neurotransmitter release in cortical and hippocampal synapses. L- and T-type Ca2<sup>+</sup> channels are involved in neurotransmission release in various retinal neurons. T-type Ca2<sup>+</sup> channels play a crucial role in neurotransmitter release and its regulation in special reciprocal synapses. Functionally, P-/Q-type Ca2<sup>+</sup> channels may be mainly related to fast, synchronous exocytosis, and N-type Ca2<sup>+</sup> channels may contribute to exocytosis in neurons processing information, P-/Q-type Ca2<sup>+</sup> channels have been shown to be more efficient in neurotransmitter release than N-type Ca2<sup>+</sup> channels in most investigated synapses, as in entorhinal stellate neurons, different inhibitory interneurons, cerebrocortical synapses or cerebellar parallel fiber terminals (Ladera et al., 2009).

#### INTERACTIONS OF Ca2<sup>+</sup> MEDIATED MEMBRANE FUSION BY SNARE PROTEINS AND ACTIVE ZONE PROTEINS

Ca2<sup>+</sup> entry through presynaptic Ca2<sup>+</sup> channels can trigger vesicle fusion by assembly of the SNARE proteins complex [t-SNARE proteins syntaxin-1 and SNAP-25, v-SNARE protein synaptobrevin (VAMP)] (Südhof, 2004; Bao et al., 2018; **Figure 2A**). SNARE function is widely reported to be associated with the processing of physiology and pathophysiology. It is reported that modifying SNARE function through regulating exocytosis can provoke metabolic diseases such as obesity (Valladolid-Acebes et al., 2015), which is improved by many therapies such as exercise training (Ramos-Miguel et al., 2015; Roh and So, 2017; Roh et al., 2017). The release of neurotransmitter requires localization of both calcium channels and synaptic vesicle proteins to the presynaptic active zone (Südhof, 2012). Rab3 interacting molecules (RIM) localizes in active zone (**Figure 2B**), which contain an N-terminal zinc finger domain, a central PDZ domain, C-terminal C2A and C2B domain and a conserved sequence between the two C-terminal domains (Wang and Südhof, 2003). RIM plays an essential role for synaptic vesicle docking and priming (Deng et al., 2011; Han et al., 2011). Munc13-1 is a large multidomain protein in active zone that plays a central role in synaptic vesicle priming (Brose et al., 1995; Augustin et al., 1999; Fukuda, 2003). The interaction of SNARE and SM (sec1/Munc18) proteins control the millisecond timescale presynaptic fusion after AP. Before priming, the Munc13 C2A-domain forms a constitutive homodimer (inactive state; **Figure 2C**). When Munc13 transforms from inactive state to an active state, Munc13-1 switches from a homodimer to a heterodimer (Munc13-1-RIM), may regulate synaptic vesicle priming (Lu et al., 2006). RIM-binding proteins (RIM-BPs) are also large multidomain proteins (∼200 kDa) in active zone, that tightly bind to RIM. PDZ-domain of both RIM and RIM-BPs bind to Ca2+-channels for tethering Ca2<sup>+</sup> channels to an active zone (Han et al., 2011; Kaeser et al., 2011). Deletion of RIM or RIM-BP (Liu et al., 2011; Kaeser et al., 2012) causes loss of Ca2<sup>+</sup> channels from active zone and decreases Ca2<sup>+</sup> entry. The central PDZ-domain of RIM can bind directly with N-type and P/Q-type, without binding with L-type Ca2<sup>+</sup> channels (Kaeser et al., 2011). RIM that lacks the PDZ-domain exhibits loss binding abilities

induced Munc13 from inactive homodimer to active heterodimer, which promoted Sec1/Munc18-1 (SM) protein dissociated with syntaxin-1. Syntaxin-1 changes from closed formation to open formation. Syntaxin-1 and SNAP-25 interacted with synaptobrevin to form SNAREs. Ca2<sup>+</sup> entry through Ca2<sup>+</sup> channel induced interaction with synaptotagmin, which trigger vesicle fusion.

with Ca<sup>v</sup> channels. Ca<sup>v</sup> channels are recruited to active zone for synaptic vesicle fusion by a tripartite complex formation (RIM, RIM-BP and the C-terminal tail of Ca2<sup>+</sup> channels) that needs assistance by Munc13-1. Munc13-1- RIM heterodimer formation is a key component for fusion. Furthermore, the C2B domain of RIM can modulate Ca2<sup>+</sup> channel activation (Kaeser et al., 2012). Recently, it was reported that Munc13, independent with Munc18, promotes the syntaxin-1-synaptobrevin complex formation during the assembly of the triplet SNARE complex. Interaction with Munc18 and Munc13 contributes to syntaxin/SNAP-25 complex formation (Lai et al., 2017).

Before forming the SNARE complex (**Figure 2A**), syntaxin-1 is presented in a closed conformation by interaction with SM proteins which cannot promote SNARE complex formation. When the zinc-finger of RIM binds to the C2A domain of Munc13, Munc13 is activated by homodimer dissociation. Subsequently, the activation of Munc13 drags RIM closer to the presynaptic membrane. Ma et al. (2011) have demonstrated that Munc13 can accelerate the transfer from the closed syntaxin-1-Munc18-1 heterodimer to an open syntaxin-1 for promoting SNARE complex formation. SM proteins are fundamental for synaptic vesicle trafficking. However, another study reported that the SM proteins exert no effect on spontaneous fusion and Ca2+-triggered fusion with SNAREs, complexin-1 and syt-1 (Zhang et al., 2015). Stable SNAREs complex provide energy for membrane fusion (Weber et al., 1998).

Complexin is a small soluble protein that controls (activates or suppresses) the trigger-release and spontaneous release (Fernández-Chacón et al., 2001; Pang et al., 2006; Mohrmann et al., 2015; Yu et al., 2018). The central helix of complexin binds to the interface of the v- and t-SNAREs close to the membrane (Fernández-Chacón et al., 2001; Chen et al., 2002; Tang et al., 2006). Complexin displays an activated effect in fast synchronous release and an inhibited effect in spontaneous release (Maximov et al., 2009; Kaeser-Woo et al., 2012). The synchronous function of complexin-1 is promoted by interactions with the SNARE complex at the N-terminal, whereas the suppressive action of spontaneous fusion is involved in binding with the C-terminal domain of complexin-1, but not the N-terminal domain (Lai et al., 2014). Lai et al. (2016) have demonstrated the mechanism that the N-terminal domain of the complexin can independently modulate the interaction of presynaptic membrane and the SNAREs. Furthermore, Gong et al. (2016) have revealed that the C-terminal domain is pivotal for regulation of spontaneous release and suppression of Ca2+-independent fusion in a curvature-dependent phase. Misplacement of complexin to the plasma membrane increases the variableness and the mean decay time constant of synchronization with NMDA-type glutamate receptor initiated postsynaptic currents.

Synaptotagmin (syt) is a Ca2<sup>+</sup> sensor that can evoke fast and synchronous neurotransmitter release (Xu et al., 2007). Syt-1 contains two homologous Ca2<sup>+</sup> sensor modules: C2 domains (C2A and C2B) and transmembrane domain. Syt-1, Syt-2 and Syt-9 bind to Ca2<sup>+</sup> to promote synchronous transmitter release, while Syt-7 evokes a asynchronous transmitter release (Bacaj et al., 2013; Brewer et al., 2015; Zhou et al., 2015; Pérez-Lara et al., 2016).

Ca2<sup>+</sup> binding to syt abolishes the complexin clamp and triggers synaptic vesicle fusion. Recent study has revealed that complexin may regulate fusion in cooperation with Syt. Syt1-SNARE and complexin-SNARE cooperate to activate synchronous release and regulate synchronous release after the AP has arrived at the synaptic terminal (Jorquera et al., 2012; Dhara et al., 2014). Recent study has demonstrated that the tripartite SNARE complexin-syt-1 complex at a synaptic vesicle docking site exerts an open state for trigger fusion. Interaction of interfaces are fundamental for Ca2+ triggered neurotransmitter release. Disruption of tripartite interface cannot trigger neurotransmitter release, although the primary interface is intact. It implied that both the primary and tripartite interfaces are required for Ca2+-triggered synaptic vesicle fusion (Akyuz et al., 2013; Gipson et al., 2017). Before the Ca2<sup>+</sup> trigger, syt interacts with SNARE proteins in the targeted membrane to prevent SNARE complex assembly (Chicka et al., 2008). Ca2<sup>+</sup> entry through Ca2<sup>+</sup> channels increases the affinity of syt-1 with syntaxin-1 for approximately two orders of magnitude (Chapman et al., 1995; Bhalla et al., 2006). Munc13, notably, enhances the transforms from the Munc18-1 syntaxin-1 complex to the SNARE complex (Ma et al., 2011) that can open the closed form of the SNARE protein (Lu et al., 2006; Kaeser et al., 2011). NSF (Sec18) and α-SNAP (Sec17) form a molecular chaperone for dynamic modulation of the disassembly of cis-SNAREs. Recently, Song et al. (2017) demonstrated that Sec17 residue K159 contributes to enhance the synaptic vesicle fusion. Furthermore, Sec18 can augment the interactions of Sec17 and trans-SNARE (Schwartz et al., 2017; Song et al., 2017). Lai et al. (2013) also have found that syt1 and Ca2<sup>+</sup> are required for pore formation and expansion. Furthermore, SNAREs alone are sufficient in promoting membrane hemifusion.

### INTERACTION OF Ca2<sup>+</sup> CHANNEL SUBTYPES AND SNARE PROTEINS COMPLEX

Presynaptic VGCCs have been classified into three super families (Cav1, Cav2 and Cav3). Cav2 (P/Q-, N- and R-type) are the dominant channel subtypes for fast presynaptic transmitter. Cav2.2 (N-type) interacts with active zone proteins (RIM, RIM-BP) and SNAREs to regulate the vesicle fusion. RIM-C2A and RIM-C2B bind the pore-forming subunit of N-type Ca<sup>v</sup> channels in a Ca2+-independent manner that weakly interacts with the Cav1.2 (L-type, α1c), but do not interact with the Cav1.3 (L-type, α1D). Furthermore, RIM (C2 domain) enhances the interaction with syt-1 when intracellular Ca2<sup>+</sup> concentration is increased. Removal of RIM domain heavily reduces the channel current and number of docking vesicles resulting in decreasing Ca2<sup>+</sup> channel coupling with vesicle. The central PDZ-domain of RIM interacts with the C-terminal of presynaptic N- and P/Q-type Ca2+-channels, with no act on L-type Ca2<sup>+</sup> channels. Deletion of RIM inhibits most neurotransmitter release due to impairing the synaptic priming and decreasing the Ca2<sup>+</sup> channels localization in presynaptic membrane (Kaeser et al., 2011; Han et al., 2015). It is well-known that vesicle priming and Ca2<sup>+</sup> influx do not require RIM C2B domains. Recently, studies have found that C2 domains of RIM do not bind to Ca2+, but bind to PIP2. PIP2 binding to RIM C2B domains exerts a critical role for vesicle priming and Ca2<sup>+</sup> channel tethering to PIP2 containing targeted membranes (de Jong et al., 2018).

Active zone scaffold protein Bassoon directly binds to RIM-BP to modulate synaptic vesicle docking via an indirect contact with Cav2.1. Genetic deletion of Bassoon or an acute interference with Bassoon RIM-BP interaction reduces synaptic amount of CaV2.1, which gently regulates P/Q-type Ca2<sup>+</sup> current to trigger synaptic transmission (Davydova et al., 2014). Both genetic ablation of Bassoon or interference of the link between Bassoon and RIM-BP reduced the numbers of Cav2.1 in active zone, decelerated AP-triggered neurotransmitter release and impaired the synaptic transmission. Cav2.2 current was increased for compensation for Cav2.1-induced decreases (Acuna et al., 2015). RIMs-mediated vesicle priming is not produced by coupling with Munc13, whereas it is directly activated by Munc13. Zn2<sup>+</sup> finger domain of RIM binds to Munc13 to promote vesicle priming, thereby dissociating Munc13 from heterodimer to homodimer and promotes priming in Munc13-deficient synapses. Hence, homodimer of Munc13 inhibits priming, and RIM activates priming by disrupting Munc13 homodimer (Deng et al., 2011). At rod photoreceptor ribbon synapses, RIM causes a dramatic loss of Ca2<sup>+</sup> entry through Cav1.4 channels and reduces trigger release. RIM induces Ca2<sup>+</sup> entry, which in turn promotes release by modulating Cav1.4 channel opening (Grabner et al., 2015). Alternative splicing (exons of 44 and 47) of Cav2.1α1 (P/Q-type) induces gene variants of the C-terminal region (CTD) of Cav2.1. The two exons interact with RIM (1α and 2α), impair the binding of CTD with RIM and implied suppressive effect of RIM on voltage-dependent inactivation (Hirano et al., 2017). Syntaxin, SNAP-25 and syt-1 possess specific ''synprint'' binding site interaction with CaV2.1 and CaV2.2 at the intracellular loop linking domains II and III (LII-III; **Figure 1C**). Diversity of VGCC types display distinct tissue specificity, subcellular localizations, kinetics performance and amount of Ca2<sup>+</sup> influx. CaV2.1 is the most abundant expression in neurons (Cabañes, 2008; Catterall and Few, 2008; Davies et al., 2011; Jahn and Fasshauer, 2012; Davydova et al., 2014; Wang and Augustine, 2014; Chai et al., 2017; Silva et al., 2017). GSK-3β displays inhibitory effects in presynaptic vesicle exocytosis by phosphorylating CaV2.1 and disturbing SNARE complex formation. A mutation in the first intracellular loop of CaV2.1 prevents interaction with SNARE proteins and impair SNAREs complex formation. SNAREs proper interact with synprint site to help vesicles docking near the Ca2<sup>+</sup> entry pathway, and modulate steadystate inactivation of Cav2.1 (Serra et al., 2018). R-type (Cav2.3) channels are localized at the presynaptic terminal and trigger neurotransmitter release by enhancing presynaptic Ca2<sup>+</sup> levels. Wu et al. (1999) reported that R-type (Cav2.3) Ca2<sup>+</sup> channels contributed to about 26% of the total Ca2<sup>+</sup> current during a medial nucleus of the trapezoid body presynaptic AP, but display a lower efficacy than other types of Ca2<sup>+</sup> channels. R-type Ca2<sup>+</sup> channels are also involved in fast synaptic excitation (Naidoo et al., 2010). Recently, researchers revealed that R-type Ca2<sup>+</sup> channels linked with NOS to induce NO release by controlling gastrointestinal smooth muscle relaxation in the guinea pig ileum via a purine transmitter (Rodriguez-Tapia et al., 2017).

Aplysia pleural sensory neurons are involved in the forms of presynaptic plasticity. The Aplysia CaV2α1 subunit EF-hand tyrosine Y1501 are targets for modulation by GPCRs through Src kinase. The heterosynaptic depression of the CaV2 channel current is inhibited when channel is combined with a Y-F mutation at the conserved Src phosphorylation. It implies that the inhibition of the Cav2 calcium current is partially, at least, responsible for the inhibition of neurotransmitter release with heterosynaptic depression (Dunn et al., 2018). Ca2<sup>+</sup> channels are also involved in nerve injury. Lu et al. (2018) first demonstrated that lycopene depress glutamate release through inhibition of voltage-dependent Ca2<sup>+</sup> entry (N-type and P/Q-type channels) and protein kinase C in rat cerebrocortical nerve terminals and not by intracellular Ca2<sup>+</sup> release.

The CaV3 family CaV3.1(α1G), CaV3.2(α1H), and CaV3.3(α1I) mediate T-type Ca2<sup>+</sup> currents. T-type channels have been revealed to regulate neurotransmitter release in central, peripheral synapses and neuroendocrine cells that modulate basal neurosecretion close to resting potential with mild stimulations. Although T-type channels have no directly binding peptide (no synprint binding site), Cav3.2 channels interact with syntaxin 1A and SNAP-25. The interactions form nanodomains that can be regulated transiently and low voltages controlling neural activity and neuroendocrine. Interaction of T-type channels, secretory vesicles, and SNAREs form a nanodomains complex. T-type Ca2<sup>+</sup> channels can directly interact with SNAREs (syntaxin 1A-Cav3.2-SNAP25) to control exocytosis. It is clear that T-type channels contribute to synaptic transmission in neurons and neuroendocrine cells under conditions of rest and mild stimulation. T-type Ca2<sup>+</sup> channels are also involved in the development of a neuropathic pain. T-type Ca2<sup>+</sup> channel subunit CaVα2δ interaction with the extracellular matrix protein thrombospondin-4 (TSP4) contributes to initiate, but not for the maintenance of excitatory synaptogenesis. Treatment with gabapentin blocks the early pain state but does not reverse the delayed state. It implies that early intervention with gabapentin may prevent the development of injury-induced chronic pain, one of the reasons is that CaVα2δ1/TSP4 initiates abnormal synapse formation (Yu et al., 2018).

Interestingly, Diao et al. (2013) have found that native presynaptic protein α-Synuclein (α-Syn) has little effect on Ca2+-triggered synaptic fusion efficiency or kinetics in neurotransmitter releases. On the contrary, α-Syn plays a key role in clustering of v-vesicles. Parkinson's disease induces α-Syn mutant at A30P. Pathogenic α-Syn reduces the clustering ability that resulted in affecting neurotransmission (Diao et al., 2013). Furthermore, N-terminal acetylation can significantly decrease α-Syn oligomerization that can preserve its native conformation against pathological aggregation (Bu et al., 2017).

### Ca2<sup>+</sup> CHANNELS REGULATION AND SYNAPTIC TRANSMISSION

The activity of presynaptic calcium channels is also modulated by βγ-subunits of G proteins (Gβγ), protein kinases (PKC, CaMKII) and Ca2<sup>+</sup> sensor (CaS) proteins. Gβγ negatively regulates the neurotransmitter release by inhibition of CaV2 (P/Q- and N-type) Ca2<sup>+</sup> channels in synaptic terminals. Gβγ directly binds to CaV2.2α1 at the N-terminal45–55 (Cantí et al., 1999), the intracellular loop domains between I and II (LI-II) at 377393 (Zamponi et al., 1997) and the C-terminus at 22572336 (Li et al., 2004). Only the N-terminal can suppress CaV2 channels activity. The site at the N terminus and intracellular loop (LI-II) produces a more potent effect (Stephens and Mochida, 2005; **Figure 1C**). Furthermore, it has also been demonstrated that the CaV2.2 alternative splicing isoform, e37a, exerts an increase in the expression of N-type Ca2<sup>+</sup> channels and also increases the channel opening compared to Cav2.2 channels that contain e37b (Castiglioni et al., 2006). Injection of N-terminal or a I-II loop interaction domain peptide into sympathetic superior cervical ganglion (SCG) neurons attenuates noradrenaline-initiated G protein regulation, and reduces synaptic transmission, and decreases Ca2<sup>+</sup> current density. Furthermore, mutation at N-terminal abolishes the inhibitory effects of the N-terminal peptide (Bucci et al., 2011). Gβγ binding to N-terminal and loop I–II of CaV2.2 contributes to regulate the function of CaV2.2. Interestingly, the SNARE protein syntaxin 1A co-localizes with Ca2<sup>+</sup> channels and Gβγ. Co-expression of syntaxin 1A with N-type channels induces tonic inhibition mediated by Gβγ (Jarvis et al., 2000). Nevertheless, syntaxin 1B does not display such effect (Lü et al., 2001). It is suggested that the spatial localization of the G proteinsynaprint-CaV2.2 complex is critical for neurotransmitter release (Yoon et al., 2008). The synaptic protein cysteine string protein promotes interaction between G proteins and the synprint site on CaV2.2 channel for enhancing neurotransmitter release (**Figure 2C**).

Protein kinases (such as PKC and CaMKII) are localized in presynaptic terminals that can phosphorylate both Ca2<sup>+</sup> channels and SNAREs (**Figure 2C**). PKC and CaMKIIphosphorylation of Ca2<sup>+</sup> channels at the synprint site induce forceful inhibition of its binding to syntaxin-1A and SNAP-25 (Yokoyama et al., 1997). Phosphorylation of Ca2<sup>+</sup> channels at the synprint by PKC is located at serines 774 and 898 which resulted in modulating the interaction with syntaxin-1A and SNAP-25. However, PKC phosphorylation failed to dissociate CaV2.2/syntaxin 1A complexes. Auxiliary subunits of Cav2.2 also participate in regulation of the function of Cav2.2 channel and then modulate the transmitter release. The acetyl-βmethylcholine (MCh) or PKC isozymes (βII or ε) are unable to potentiate Cav2.2 current in the presence of CaVβ subunits. Cavβ subunits complete suppression of the interactions between PKC and Ser/Thr sites of Cav2.2α1 subunits (Thr-422, Ser-425, Ser-1757, Ser-2108 and ser-2132; Rajagopal et al., 2014). The mutation of PKC sites (Thr-422, Ser-1757 and Ser2132) can abolish MCh potentiation on Cav2.2α1 currents. The stimulatory sites at Thr-422, Ser-2108 or Ser-2132 and inhibitory sites at Ser-425 of Cav2.2α1 are identified by binding to PKCs βII and ε subunits. Whereas, the stimulatory sites at Thr-365, Ser-1995 and Ser-2011 and the inhibitory sites at Ser-369of Cav2.3α1 subunits are homologous with Cav2.2α1. The stimulatory effects of PKC at the site of Thr-365 or Ser-1995 were fully offset by inhibitory site at Ser-369. PKC cannot inhibit the effects via the coexistence with Thr-365 and Ser-1995 (Rajagopal et al., 2017).

The phosphorylation of Core-conserved residues inside the SNARE domain can suppress vesicle fusion. Studies revealed that secretory protein VAMP8 phosphorylation by PKC at multiple residues in the SNARE domain mediated vesicle fusion, where protein kinase activation decreases and phosphatase activation increases the capacity of VAMP8 (Malmersjö et al., 2016).

CaMKII potently inhibits the interactions between syntaxin-1A and SNAP25 by phosphorylation at Ser 784 and 896 (Yokoyama et al., 2005). Each site of phosphorylation modulates syntaxin-1 and SNAP-25 binding to the synprint site. PKC or CaMKII phosphorylates Cavα1 at the synprint sites that manipulates a biochemical switch for controlling the interaction of synprint and SNAREs. It implied that switch role provides a potential functional link between neurotransmitter release and protein phosphorylation for tethering and docking synaptic vesicle in an optimal position to respond to the Ca2<sup>+</sup> signal from presynaptic Ca2<sup>+</sup> channels (Catterall and Few, 2008).

In neurons, multiple Ca2<sup>+</sup> sensor (CaS) proteins are involved in neuronal Ca2<sup>+</sup> signaling transmitter. The distance between voltage-gate Ca2<sup>+</sup> channels and CaS for exocytosis determines the timing and probability of neurotransmitter release (Nakamura et al., 2018). Calmodulin (CaM) is one of the members of a subfamilies of CaS proteins. Vesicle protein synaptotagmin is also a CaS protein for fast neurotransmission. Interactions of Ca2+/CaM binding to the CaM-binding domain (CBD) and IQ-like motif (IM) of CaV2.1 contribute to facilitate and inactivate Cav2.1 channels. Mutation of the motifs of CBD and IM prevents synaptic facilitation. Nanou et al. (2018) demonstrate a direct link between regulation of CaV2.1 channels and short-term synaptic plasticity in native hippocampal excitatory and inhibitory synapses. CaBP1 and VILIP-2 are neurospecific CaM-like CaS proteins that potently modulate CaV2.1 channels function. Ca2+-binding protein (CaBP1), Visinin-like protein 2 (VILIP-2) and neuronal calcium sensor-1 (NCS-1) are the key CaS proteins for synaptic transmission. CaBP1 is highly expressed in the brain and retina, and colocalized in the CBD of Cav2.1α1 (Lee et al., 2002). CaBP1 binds to CBD in a Ca2<sup>+</sup> independent profile. Leal et al. (2012) demonstrated that CaBP1 performed a blockade effect on Ca2+ dependent facilitation of Cav2.1, and reduced facilitation of synaptic transmission in superior cervical ganglion neurons. Nanou et al. (2018) also demonstrated CaBP1/caldendrin as the CaS protein interacting with CaV2.1 channels to mediate rapid synaptic depression in the inhibitory hippocampal synapses. On the contrary, VILIP-2 blocked Ca2+-dependent inactivation of CaV2.1 current, and notably reduced synaptic depression and showed increasing facilitation. VILIP-2 is highly expressed in neocortex and hippocampus, and plays a complementary effect on CaBP1. These studies reveal that CaBP1 and VILIP-2 bind to the same site with opposite effects on Cav2.1. The integrated effect contributes to modulating short-term synaptic plasticity (Leal et al., 2012; Catterall et al., 2013). The N-terminal myristoylation site and EF-hand motifs of CaBP1 and VILIP-2 determine their differential regulated role on CaV2.1 channels. CaS proteins serve as bidirectional switch that fine-tune the relationships of Ca<sup>V</sup> and synaptic transmission. Thereby, the balance between facilitation and depression is a key role on neurotransmitter release (Leal et al., 2012).

Neuronal calcium sensor-1 (NCS-1) has been also shown to enhance synaptic facilitation. NCS-1 directly interacts with IQ-like motif and CBD site at the C-terminal domain of CaV2.1. NCS-1 reduces Ca2+-dependent inactivation of Cav2.1 through interaction with the IQ-like motif and CBD. NCS-1 modulates Ca2<sup>+</sup> current amplitude or kinetics activity. These studies indicate that NCS-1 directly binds to CaV2.1 to serve short-term synaptic facilitation and confirm that CaS proteins are crucial in fine-tuning short-term synaptic plasticity (Yan et al., 2014).

### AUTHOR CONTRIBUTIONS

RH, JZ, YY and LZ contributed to the review of the literature, and editing of the manuscript. WW and ML wrote the draft manuscript. All authors read and approved the submission.

#### ACKNOWLEDGMENTS

We thank the undergraduate students Zenan Fan and Xiaoli Gao for drawing figures of this manuscript.

#### REFERENCES


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2018 He, Zhang, Yu, Jizi, Wang and Li. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Molecular Mechanisms of Synaptic Dysregulation in Fragile X Syndrome and Autism Spectrum Disorders

#### Michael Telias\*

Department of Molecular and Cell Biology, University of California, Berkeley, Berkeley, CA, United States

Fragile X syndrome (FXS) is the most common form of monogenic hereditary cognitive impairment. FXS patient exhibit a high comorbidity rate with autism spectrum disorders (ASDs). This makes FXS a model disease for understanding how synaptic dysregulation alters neuronal excitability, learning and memory, social behavior, and more. Since 1991, with the discovery of fragile X mental retardation 1 (FMR1) as the sole gene that is mutated in FXS, thousands of studies into the function of the gene and its encoded protein FMR1 protein (FMRP), have been conducted, yielding important information regarding the pathophysiology of the disease, as well as insight into basic synaptic mechanisms that control neuronal networking and circuitry. Among the most important, are molecular mechanisms directly involved in plasticity, including glutamate and γaminobutyric acid (GABA) receptors, which can control synaptic transmission and signal transduction, including short- and long-term plasticity. More recently, several novel mechanisms involving growth factors, enzymatic cascades and transcription factors (TFs), have been proposed to have the potential of explaining some of the synaptic dysregulation in FXS. In this review article, I summarize the main mechanisms proposed to underlie synaptic disruption in FXS and ASDs. I focus on studies conducted on the Fmr1 knock-out (KO) mouse model and on FXS-human pluripotent stem cells (hPSCs), emphasizing the differences and even contradictions between mouse and human, whenever possible. As FXS and ASDs are both neurodevelopmental disorders that follow a specific time-course of disease progression, I highlight those studies focusing on the differential developmental regulation of synaptic abnormalities in these diseases.

Keywords: synaptic plasticity, fragile X syndrome, autism spectrum disorders, mouse models, human pluripotent stem cells, human embryonic stem cells, human induced pluripotent stem cells

**Abbreviations:** AMPA, α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid; aNSCs, adult neural stem cells; ASDs, autism-spectrum disorders; BDNF, brain-derived neurotrophic factor; CNS, central nervous system; cAMP, cyclic adenosine monophosphate; GABA, gamma-(γ)-aminobutyric acid; GABAA, ionotropic GABA receptors; GABAB, metabotropic GABA receptors; GluRs, glutamate receptors; iGluRs, ionotropic glutamate receptors; mGluRs, metabotropic glutamate receptors; GSK3β, glycogen synthase kinase 3 beta; FMR1, human fragile X mental retardation 1 (gene); Fmr1, mouse fragile X mental retardation 1 (gene); FMRP, FMR1/Fmr1 protein; hPSCs, human pluripotent stem cells; hESCs, human embryonic stem cells; hiPSCs, human induced-pluripotent stem cells; KO, knock-out; LTD, long-term depression; LTP, long-term potentiation; NMDA, N-methyl-D-aspartate; TFs, transcription factors; UTR, untranslated region; WT, wild-type.

#### Edited by:

Jiajie Diao, University of Cincinnati, United States

#### Reviewed by:

Franziska Scharkowski, Max-Planck-Institut für Experimentelle Medizin, Germany Maija Liisa Castrén, University of Helsinki, Finland

> \*Correspondence: Michael Telias mtelias@berkeley.edu

Received: 28 February 2018 Accepted: 12 February 2019 Published: 07 March 2019

#### Citation:

Telias M (2019) Molecular Mechanisms of Synaptic Dysregulation in Fragile X Syndrome and Autism Spectrum Disorders. Front. Mol. Neurosci. 12:51. doi: 10.3389/fnmol.2019.00051

### INTRODUCTION

Fragile X Syndrome (FXS) is the most prevalent form of inherited intellectual disability (Penagarikano et al., 2007). It is caused by a CGG triplet repeat expansion, in the 5<sup>0</sup> UTR region of the fragile X mental retardation 1 (FMR1) human gene, located in the X chromosome (Verkerk et al., 1991). If this genomic region expands to more than 200 CGG-repeats, the promoter of FMR1 becomes hyper-methylated, resulting in the inactivation of the gene and the absence of its encoded protein: FMR1 protein (FMRP; O'Donnell and Warren, 2002; Mor-Shaked and Eiges, 2018). FMRP is an RNA-binding protein that plays important roles in regulation of translation as well as in other processes in the central nervous system (CNS; Bagni and Oostra, 2013; Fernández et al., 2013). It is estimated that FMRP has hundreds of mRNA and microRNA targets, making the study of FMRP's role a very challenging field (Ascano et al., 2012; Pasciuto and Bagni, 2014). Accordingly, FXS pathology is complex too. Patients with FXS suffer from mild to severe cognitive impairment, epilepsy, auditory hypersensitivity, repetitive behavior, social withdrawal and other neurological symptoms, as well as other disorders outside of the CNS, like cartilage malformations and macroorchidism (Penagarikano et al., 2007; Kidd et al., 2014). Perhaps the most remarkable fact about FXS for the purposes of this review is a high comorbidity with autism spectrum disorders (ASDs): 50% of male FXS patients and 20% of female FXS patients are diagnosed with ASDs (Kaufmann et al., 2017). Therefore, studying FMRP's role in the CNS can shed light not only on the etiology of FXS, but also on basic mechanisms shared with other neurodevelopmental disorders, while expanding current scientific understanding of synaptic plasticity and brain physiology (Telias and Ben-Yosef, 2014).

A critical feature of FXS pathophysiology is related to the timing of FMR1 expression and inactivation during human embryonic development (Colak et al., 2014). It is safe to assume that the full CGG-repeat expansion exists already in the zygote, since it can be detected as early as in human blastomeres (Malcov et al., 2007). Yet, despite the CGG-repeat expansion present, the FMRP protein is expressed and detected in FXS human embryonic tissue, at least up to the end of the first trimester of pregnancy (Willemsen et al., 2002; Mor-Shaked and Eiges, 2018). After this time-point, the FMR1 promoter becomes gradually hyper-methylated and the gene increasingly silenced. This means that, in human FXS embryos, early neurodevelopmental events such as the formation of the neural tube take place in the presence of FMRP; while FMRP absence probably affects late developmental stages, such as rapid neurogenesis from progenitors, migration and synaptogenesis, during the late phases of cortical and neo-cortical development (i.e., second and third trimesters). This is a critical feature to consider when assessing the mechanisms behind FXS pathology, but also when choosing research tools, because the CGG-repeat expansion at the FMR1 locus, and its ensuing developmentallyregulated disappearance of FMRP, is unique to humans and is not recapitulated in Fmr1 knock-out (KO) mice (Eiges et al., 2007; Telias and Ben-Yosef, 2014).

In this review, I will summarize the main hypotheses and mechanistic models proposed to explain synaptic dysregulation in FXS and ASDs (see **Table 1**). All these hypotheses ultimately reflect the current state of knowledge regarding the role of FMRP in CNS neurons, during embryonic development and postnatal life. I will include studies conducted on the Fmr1 KO mouse model, and emphasize how they compare to more recent research carried-out on human pluripotent stem cells (hPSCs), including human embryonic stem cells (hESCs) obtained from donated in vitro fertilization human blastocysts, and human induced pluripotent stem cells (hiPSCs) derived from somatic cells obtained from patients' biopsies.

#### CHEMICAL SYNAPTIC TRANSMISSION

#### Glutamate-Dependent Synaptic Transmission

Glutamate is the most prevalent excitatory neurotransmitter in the brain. Glutamate receptors (GluRs) are divided into two major families: ionotropic and metabotropic. Both types of GluRs are intrinsically involved in the activation of long-term potentiation (LTP) and long-term depression (LTD). Given that intellectual disability constitutes perhaps the most important aspect of FXS pathology and that synaptic plasticity is considered to be at the base of learning and memory, researchers raised the hypothesis that FMRP could be involved in the regulation of luRs. This approach could also uncover pharmacological targets for a possible therapy. Surprisingly, one of the first studies that assessed synaptic transmission in Fmr1 KO mice showed no conclusive abnormalities (Godfraind et al., 1996). The affected mice showed normal acquisition of new behavior as compared to healthy counterparts, but difficulties during extinction of the learned behavior and the acquisition of a new one, suggesting impaired LTP. However, electrophysiological recordings showed no significant differences in LTP recordings carried out on hippocampal CA1 neurons in wild-type (WT) vs. Fmr1 KO mice. The same study also showed that Fmr1 expression is not affected by the induction of LTP in WT neurons, but it did not address the question whether LTP-responsive genes, including GluRs, are differentially expressed in WT as compared to Fmr1 KO.

Breakthrough research by Huber et al. (2002) showed an increase in the expression of postsynaptic metabotropic GluR type-I (mGluRI) in Fmr1 KO hippocampal neurons. mGluRs are G-protein coupled receptors that mediate slow response to glutamate. There are eight different mGluRs divided into three groups: mGluRI(1,5) , mGluRII(2,3) , and mGluRIII(4,6,7,8) (Maj et al., 2016; Ribeiro et al., 2017). According to this hypothesis, mGluRI expression is negatively regulated by FMRP, and therefore, loss of FMRP results in an abnormal increase of mGluRI in Fmr1 KO neurons, enhancing mGluRdependent LTD. An increase in LTD, seemingly at the expense of LTP, would be consistent with intellectual disability and cognitive impairment, since these mechanisms have been shown to directly affect learning and memory. This fundamental result, the increase in mGluRI-dependent LTD in correlation with FMRP loss in mice, was later confirmed by many independent studies (Todd et al., 2003; Antar et al., 2004;


TABLE 1 | Summary of mechanisms involved in Fragile X Syndrome (FXS) pathology.

Table 1 summarizes the mechanisms discussed in the review, found to be involved in the brain pathophysiology of FXS, according to the target gene or protein proposed. The table indicates the developmental stage and organism model used, as well as a representative article for each mechanism.

Aschrafi et al., 2005; Desai et al., 2006; Huang et al., 2015) giving rise to the formulation of the ''mGluR theory of FXS'' (Bear et al., 2004; Bear, 2005), which will eventually rise to almost dominate the field of FXS research. Enhanced LTD mediated by mGluRs not only provides a possible biological explanation for the intellectual disability associated with FXS, but also provide highly-specific drug targets for a potential pharmacological treatment, or cure, of FXS (Sourial et al., 2013; Berry-Kravis, 2014; Gandhi et al., 2014).

Yet, the mGluR-based explanation of synaptic dysregulation in FXS has some weak points that need to be addressed. First, the molecular mechanism and the cascade of cellular events that lead from FMRP loss to mGluRI functional upregulation remains unresolved. Second, none of the molecular and physiological hallmarks of the ''mGluR theory'' have ever been conclusively confirmed in any human model for FXS or ASDs. Third, from a more neurodevelopmental perspective, the question of the timing of mGluRI hyperactivation remains open. If mGluRI hyperactivation is caused by FMRP downregulation, it is important to remember that, as mentioned before, the Fmr1 mutation in KO mice does not recapitulate the much later timing of developmentally-regulated transcriptional inactivation, as observed in human embryos and hESCs (Telias et al., 2013; Telias and Ben-Yosef, 2014). One study tackled these questions directly, comparing developing neural progenitors obtained from WT and Fmr1 KO mice and differentiated in vitro from healthy and FXS-hiPSCs (Achuta et al., 2017). By measuring Ca2<sup>+</sup> responses to dihydroxyphenylglycine (DHPG, an agonist of mGluRI(1,5)), with or without the presence of 2-methyl-6-(phenylethynyl)pyridine (MPEP, a selective mGluR<sup>5</sup> antagonist), they could measure the contribution of mGluR5 to mGluRI-hyperactivation. In murine Fmr1 KO cells, the presence of MPEP did not significantly change Ca2<sup>+</sup> responses to DHPG, while in human FXS-hiPSCs derived neural progenitors, blocking mGluR<sup>5</sup> tripled the number of cells positively responding to DHPG. These results, even though not yet confirmed by other studies, seem to indicate that the alterations in mGluR-dependent signaling found in mice do not necessarily correlate to the pathophysiology found in human models. A clinical trial by Novartis with mavoglurant (AFQ056), an mGluR5 antagonist, was discontinued after it failed to show improvement over placebo in FXS patients (NCT01482143). Another study, analyzing the data from two separate phase-IIb trials of mavoglurant administration to fully-methylated FXS patients, reached the conclusion that neither of the studies achieved the primary efficacy end-point of improvement on behavioral symptoms (Berry-Kravis et al., 2016).

Finally, it is worth mentioning that mGluRI activation is found to be abnormal in other neurodevelopmental disorders, in which FMR1 and FMRP are not mutated in any way, and are normally expressed in the CNS. For example, in one study conducted in the cerebellar inner granular layer of Neuroligin-3 KO mice (with normal expression of Fmr1), an increase in the expression of mGluR1α was found, together with an increase in the phosphorylation of the GluA2 subunit of ionotropic GluRs (iGluRs) and the occlusion of mGluR-LTD upon treatment with DHPG (Baudouin et al., 2012). Another model for ASDs, known as the BTBR mouse (Meyza and Blanchard, 2017), was used to show that behavioral deficits associated with ASDs in these mice, are reversed by treatment with MPEP (Seese et al., 2014). Mice displaying haploinsufficiency of synaptic GTPaseactivating protein (Syngap+/−) are another monogenic model for ASDs (Jeyabalan and Clement, 2016). One study directly compared hippocampal physiology in Fmr1 KO and Syngap+/<sup>−</sup> mice, finding that they both show the same elevated mGluRIdependent LTD (Barnes et al., 2015). In summary, increased mGluRI signaling in Fmr1 KO mice: (1) is caused by a molecular mechanism that has not been yet fully elucidated; (2) it has never been shown to be true in human neuronal tissue; in vivo or in vitro; (3) it has failed to provide an effective drug target to ameliorate FXS; and (4) it has been shown to exist in many other mouse models of intellectual disability and ASDs, regardless of the Fmr1 KO mutation. Taken together these findings suggest that increased mGluRI-dependent LTD is a common consequence of intellectual disability and not the cause of it.

iGluRs include α-amino-3-hydroxy-5-methyl-4 isoxazolepropionic acid (AMPA), N-methyl-D-aspartate (NMDA) and kainate receptors. Research into the possible involvement of iGluRs in FXS and ASDs synaptic pathophysiology, has been less prevalent as compared to the study of mGluRs (Uzunova et al., 2014). As mentioned before, in perhaps the first of such studies, Godfraind et al. (1996) did not find evidence for altered LTP in the hippocampi of adult Fmr1 KO mice. However, since then, new research has emerged challenging this concept, but also providing somewhat conflicting results. One study showed reduced LTP and decreased levels of GluR1-containing AMPA receptors in the cortex of Fmr1 KO mice, but normal LTP in the hippocampus and normal levels of GluR1 in both hippocampus and cerebellum (Li et al., 2002). In contrast, another study showed reduced LTP in hippocampal CA1 neurons of Fmr1 KO mice, associated with reduced delivery of GluR1-containing AMPA receptors to the active synapse, but without a change in GluR1 expression (Hu et al., 2008). The same study also showed that enhancement of the Ras-PI3K signaling pathway rescues LTP in these mice, but the exact mechanism linking loss of FMRP to this pathway remains unknown. In this context, one study explored the question whether changes in LTP associated with FXS are developmentally regulated (Pilpel et al., 2009). They found that CA1 hippocampal neurons of 2-weeks old Fmr1 KO mice show down-regulation of AMPA receptors and up-regulation of NMDA, significantly changing the AMPA to NMDA ratio, resulting in enhanced NMDA-dependent LTP. Most interestingly, they found that by age 6–7 weeks, these abnormalities in synaptic plasticity disappeared, suggesting the existence of a critical developmental period in which loss of FMRP could account for impaired plasticity. More recent studies further show the complexity of FXS-associated synaptic deficiencies and their time-dependency. For example, one study showed increased mGluR-dependent LTD in Fmr1 KO hippocampal neurons upon NMDA receptor blockade, as expected (Toft et al., 2016). However, while this was the case at P30, the same was not found at P60, indicating that enhanced mGluR-LTD in Fmr1 KO mice is NMDA-dependent and developmentally regulated.

All of the aforementioned studies analyzing the possible role of iGluRs in synaptic dysregulation in FXS were conducted using the mouse model for FXS, and examining mostly hippocampal CA1 neurons. Far less studies have been conducted in other brain regions, and even less in other FXS models. One recent study in Fmr1 KO mice focused on the Mossy fiber pathway which innervates CA3 hippocampal neurons, showing increased excitatory postsynaptic potentials coupled with enhanced AMPA receptor activation (Scharkowski et al., 2018). A study conducted on 8-weeks old Fmr1 KO rats, showed deficient AMPA receptors-mediated responses as compared to WT in hippocampal CA3-CA1 synapses (Tian et al., 2017). Interestingly, this study also shows a decrease in both LTP and LTD in Fmr1 KO rats as compared to WT, and an increase in DHPG-induced LTD in Fmr1 KO, which was independent of protein synthesis. Finally, a recent study conducted on human neurons derived from FXS-hiPSCs shows, for the first time, functional changes in AMPA receptors in a human model (Achuta et al., 2018). This study analyzed the differential expression and activation of iGluRs during the process of in vitro differentiation, which can be correlated, to some extent, to human embryonic development. The results show decreased GluR2 expression, and increased expression of Ca2+ permeable AMPA receptors, in human FX neurons as compared to non-mutated controls. The number of cells co-expressing AMPA and NMDA was higher in FX neurons, too. However, in striking opposition to most studies conducted on Fmr1 KO mice, there was no significant difference in the fraction of DHPG-responsive cells in FX vs. control. These findings exemplify how disappearance of FMRP has different effects on human neurons as compared to rodent counterparts, and can also hint at a critical (and maybe overlooked) difference between the role of FMRP during embryonic development and early life on one hand, and during adulthood on the other.

#### Gamma-Aminobutyric Acid-Dependent Synaptic Transmission

The major inhibitor neurotransmitter in the brain is an enzymatic product of glutamate break-down, gamma-(γ) aminobutyric acid (GABA). Two major families of GABA receptors exist: GABA<sup>A</sup> (ionotropic) and GABA<sup>B</sup> (metabotropic; Fritschy and Panzanelli, 2014; Mele et al., 2016). GABA<sup>A</sup> receptors are expressed in the whole CNS, and their activation is coupled with a fast increase in chloride conductance and hyperpolarization of the postsynaptic neuron, inhibiting neuronal activity. GABA<sup>B</sup> receptors have a similar inhibitory function, but through slower G-protein mediated activation of K <sup>+</sup> channels. GABA<sup>A</sup> receptors are pentameric, and composed of combinations of several different subunits. Some of these subunits also include several different isoforms, which makes the study of GABA<sup>A</sup> receptors structure and composition a specially challenging field. GABA<sup>B</sup> receptors are similar in structure to mGluRs, and are divided into two subtypes that assemble as heterodimers. Two key symptoms, present in both FXS and autism, are hyperexcitability and hypersensitivity, which could be caused by reduced GABA-mediated inhibition. If this hypothesis is true, then FXS neurons should exhibit decreased GABA receptors expression, reduced GABA secretion, or both, reducing inhibition and therefore increasing uncontrolled excitation.

And indeed, both GABA<sup>A</sup> and GABA<sup>B</sup> receptors have been found to be involved in FXS and ASDs pathology, during embryonic development and in adulthood. Several studies have shown a reduction in the mRNA expression of several GABA<sup>A</sup> receptor subunits in correlation with the loss of FMRP (D'Hulst and Kooy, 2007; Paluszkiewicz et al., 2011; Braat and Kooy, 2015; Braat et al., 2015), but the mechanism of this effect remains unclear, especially since FMRP is known as a negative regulator of translation. Seminal work by D'Hulst and Kooy (2007) showed that, in the cortex of Fmr1 KO mice but not in their cerebellum, mRNA levels of eight different GABA<sup>A</sup> subunits displayed a down-regulation of 35%–50%, including most prominently the δ subunit, as well as α1, α3, α4, β1, β2, γ1 and γ2 (D'Hulst et al., 2006). However, the authors did not provide any insight on whether these changes in GABA<sup>A</sup> receptor subunit mRNA expression are developmentally regulated. This is important, since GABA<sup>A</sup> receptor subunit expression is itself developmentally-regulated (Luján et al., 2005). Many follow-up studies have confirmed the fundamental results obtained by D'Hulst et al. (2006) consistently showing a reduction in GABA<sup>A</sup> subunits expression in different parts of the brain, including the hippocampus and the amygdala, concomitant with the expected alterations in GABAergic synaptic transmission, such as reduced inhibitory postsynaptic potentials and currents (Olmos-Serrano et al., 2010; Sabanov et al., 2017; Zhang et al., 2017). However, while the literature is rich in reports confirming the ''GABAergic theory of FXS'' in esoteric models of the disease, such as zebrafish and Drosophila, the same literature to-date includes only one single report attempting to test this hypothesis in FXS human patients (D'Hulst et al., 2015). This study, using positron emission tomography (PET) to map GABA<sup>A</sup> receptor availability in 10 FXS patients, found an average reduction throughout the brain of only 10%, far from the almost 50% reduction expected from mouse studies, with the thalamus being the brain region showing the most significant reduction (17% as compared to control subjects). Since the thalamus is not a brain area typically associated with learning, memory and complex social behavior, the results shown in this study actually raise more questions than they solve. In spite the lack of confirmation in humans and contradicting results, clinical trials with different drugs aimed at enhancing GABA<sup>A</sup> signaling were approved and conducted (Erickson et al., 2011; Braat and Kooy, 2015), failing to provide the expected clinical improvement.

As for GABA<sup>B</sup> receptors, it has also been shown that their expression is reduced in the forebrain of adult Fmr1 KO mice (D'Hulst et al., 2009). This study proposes that hyperexcitability in FXS is caused by decreased GABAB-mediated attenuation of glutamate secretion at presynaptic terminals. It was subsequently reported that treating adult (8–12 weeks old) autistic mice with R-Baclofen, a GABA<sup>B</sup> specific agonist, effectively reversed social deficits and reduced repetitive behavior (Silverman et al., 2015). More recent data seems to support the idea that of a specific FXS-associated deficit in GABA<sup>B</sup> receptors subunit expression in presynaptic terminals, which could lead to excess secretion of glutamate in the hippocampi of 5-week-old Fmr1 KO mice (Kang et al., 2017). Here too, the researchers made the effort to confirm, in post-mortem human brain tissue, the observations collected from the mouse model. The results, although not significant, showed a trend toward human validation of the mouse results: a decrease in the protein levels of a few GABA<sup>B</sup> subunits. However, and most importantly, this study also shows that treatment with R-Baclofen does not rescue abnormalities in synaptic activity characterizing Fmr1 KO neurons. Currently, one human clinical trial (NCT01013480), testing the efficacy of R-Baclofen as a candidate drug for FXS treatment, has reported unsuccessful results.

The only other published study so far, aimed at testing the ''GABAergic theory of FXS'' in human neurons is our own (Telias et al., 2016). In it, FXS human neurons were differentiated in vitro from three different lines of FXS-hESCs, all affected with the naturally occurring >200 CGG expansion mutation. By puffing GABA during whole-cell patch-clamp recordings, we showed that developing human neurons display either a mature response (bicuculine-sensitivity, no current desensitization), and an immature response (bicuculine-insensitive, fast and lasting desensitization). While 60% of the WT neurons tested were classified as immature and 40% mature, ∼90% of the developing FX human neurons showed immature responses. Furthermore, the transcriptional levels of the GABA<sup>A</sup> β2-subunit were dramatically reduced in FX human neurons, in accordance to some of the findings in Fmr1 KO mice. However, the expression of δ was similar in FX and WT, and the expression of α<sup>2</sup> was increased in FX neurons, two results that directly contradict the evidence obtained using the Fmr1 KO mouse model, which shows significant reduction in δ and no effect on α2 (D'Hulst et al., 2006). None of the cells analyzed in our study, FX or WT, mature or immature, responded to baclofen, demonstrating a lack of functional GABA<sup>B</sup> expression during this developmental stage, regardless of FMR1/FMRP expression.

As mentioned before, GABAergic synaptic transmission and chloride gradient regulation, play a crucial role in brain development, and might also explain the developmental aspects of FXS and ASDs. During early development of the CNS, GABA<sup>A</sup> receptors are key players in an excitatory-to-inhibitory developmental switch (Ben-Ari, 2014). Impairment of this developmental switch has been proposed as a pathological molecular mechanism shared by several neurodevelopmental disorders (Ben-Ari, 2017). And indeed, in Fmr1 KO mice, it has been found that this developmental switch is delayed, in correlation with a significant increase in the expression of the neuronal chloride transporter NKCC1 (He et al., 2014). Using tissue sections containing somatosensory cortex from P5-15 Fmr1 KO and WT mice, this study demonstrated delayed maturation of Cl<sup>−</sup> currents in the affected mice, in correlation to protein expression levels measured by Western blot. According to this model, during human embryonic development, FXS-neurons remain depolarized for longer, presumably delaying their maturation and affecting synaptogenesis, during a critical period in neurogenesis. In the follow up to this study, the researchers show that inhibition of NKCC1 in Fmr1 KO during this critical period, corrects the Cl<sup>−</sup> imbalance and rescues the phenotype in vivo (He et al., 2018). This is a powerful model, that can both, explain many of the symptoms associated with FXS, as well as providing a pathway for a possible treatment.

### NEURONAL EXCITABILITY

Beyond aberrant expression and function of neurotransmitter receptors, other cellular and molecular neuronal mechanisms could be disrupted in FXS and ASDs, including abnormal excitability, neurotransmitter release and synaptogenesis. Studying the effect of FMRP loss on basic neuronal electrical properties, such as action potential (AP) firing, membrane resistance, ion channel expression and current conductance; as well as release probability and dynamics, and vesicle composition; could prove important in understanding the pathophysiology of FXS and ASDs. One of the first and more interesting studies to tackle this question, made used of an unorthodox FXS mouse model: a mouse displaying mosaic expression of Fmr1 (as it is in FXS females), including a reporter gene (GFP), to allow discrimination between cells based on whether they express FMRP, or not (Hanson and Madison, 2007). Electrophysiological recordings from coupled cells (in four possible combinations), from CA3 pyramidal neurons, showed a reduction in the proportion of active synaptic connections from 70% when the presynaptic cell expressed Fmr1, to 40% when the presynaptic neuron was Fmr1 KO, while the average amplitude of excitatory postsynaptic currents did not change in correlation to FMRP expression. This seems to indicate that FMRP absence results in reduced synaptogenesis, but within those connections that successfully developed, the postsynaptic response seems to be unaffected. Lack of altered postsynaptic activity could explain why the phenotype of the mosaic FXS female is much milder than that found in the majority of FXS males. Importantly, these recordings took place during the critical period at age P5-6, but were not compared to recordings carried out after it. Another study seems to independently support this idea, by analyzing the proteasome expression profile of isolated synaptic membranes from P14 WT and Fmr1 KO hippocampi (Klemmer et al., 2011). This screening showed that lack of FMRP affects several presynaptic proteins, including a reduction of ∼40% in the expression of β-Catenin [Ctnnb1, see glycogen synthase kinase 3 beta (''GSK3β'') below], and an increase of ∼25%–40% in the expression of Synapsin (Syn1), and Synaptophysin (Syp), which are involved in regulation of synaptic vesicle release and the formation of new synaptic connections. The study also shows that FMRP loss is associated with a reduced density of vesicles per cluster and a higher proportion of docked vesicles, indicating reduced synaptic activity, in line with the work of Hanson and Madison (2007).

Two groundbreaking studies were published by the Klyachko lab (Deng et al., 2011, 2013), in which they reported important electrophysiological abnormalities directly affecting short-term plasticity in hippocampal pyramidal neurons of P15-25 Fmr1 KO mice. They found that absence of FMRP in presynaptic neurons is correlated with enhanced responses to high-frequency stimuli and reduced short-term plasticity; as well as an increase in Ca2<sup>+</sup> influx, synaptic vesicle recycling, and vesicle pool size. They also found that these FXS-enhanced excitatory responses to high-frequency stimuli were independent of GABAergic transmission. FMRP was found to regulate neurotransmitter release by modulating AP duration. This work demonstrated a critical role for FMRP through direct interaction with the regulatory β4 subunit of big potassium channels (BK). This protein-protein interaction was found to be translationindependent, expanding the spectrum of FMRP functions beyond negative regulation of translation through mRNA sequestration, and providing a new model to explain FXS and ASDs' molecular pathophysiology.

Other important ion channels, regulating spiking and neurotransmitter release, might be affected in FXS. For example, Ferron et al. (2014) showed an increase in the function, density, and expression of N-type voltage-gated Ca2+-channels, in cultured E18 rat dorsal root ganglion cells, following shRNAmediated knock-down of Fmr1 expression. This study further demonstrated a direct protein-protein interaction between FMRP and Cav2.2, somewhat similar to that found between FMRP and BK channels. Our own work was the first to report presynaptic abnormalities in human FX neurons differentiated in vitro from FXS-hESCs (Telias et al., 2013, 2015a). First, we found FXS-associated impairments in both neurogenesis and synaptogenesis, including the inability of human FX neurons to fire trains of consecutive APs. We also found specific abnormalities in inward and outward ionic currents and synaptic vesicle release dynamics, associated with FMRP loss. We demonstrated that impairments in early synaptogenesis associated with FXS display a presynaptic component, by co-culturing early human neurons with adult, fully-developed, rat neurons: in regular cultures human FX neurons showed poor spontaneous synaptic activity, but when these cells were differentially labeled and co-cultured with normal adult rat neurons, normal postsynaptic activity was restored. The study of FMRP's effect on the expression of ion channels through protein-protein interactions is a fascinating new prospect in the research of FXS. However, it is still unclear how these new discoveries can be implemented in a clinical set-up to device therapeutic strategies.

#### CELL SIGNALING, GROWTH FACTORS AND GENE EXPRESSION

Synaptic transmission affects downstream cell signaling, and it is itself affected by upstream gene regulation events. Therefore, molecular interactions occurring ''far away'' from the synaptic domain, can have a direct impact on synaptic communication. In order to explore the involvement of growth factors, enzymes and transcription factors (TFs) in FXS and ASDs, we need first to assess whether these mechanisms are important in embryonic neurodevelopment, in adult neurogenesis, or in both. Molecular players affecting embryonic neurogenesis and synaptogenesis can be helpful in explaining how the impairments observed in FXS and ASDs patients came to be. Mechanisms affecting adult neurogenesis can help answer the question whether these impairments are reversible or not. Finally, mechanisms affecting both, embryonic and adult neurogenesis and synaptogenesis, could prove essential in designing therapeutic approaches to ameliorate or even cure neurodevelopmental disorders. Hypothetically speaking, it could be possible 1 day to treat FXS and autism patient through gene therapy as soon as they are born or even in utero, with the hopes of correcting any synaptic abnormalities before the brain is fully developed. Another hypothetical treatment for FXS and ASDs might 1 day be the implant of unmutated adult neural stem cells (aNSCs), that can re-populate the hippocampus and other areas with properly functioning neurons (Telias and Ben-Yosef, 2015), in the same way as today mutated hematopoietic stem cells are replaced with healthy ones. However, no strategy currently exists to re-populate the brain with healthy neurons, that can be directed to recapitulate proper neuronal wiring. In any case, for any therapeutic approach to work, elucidation of mechanistic abnormalities must be achieved first. Next, I summarize some of the most interesting studied mechanisms known to affect embryonic and adult neurogenesis, that have also been shown to be impaired in FXS and ASDs.

#### BDNF

Brain-derived neurotrophic factor (BDNF) is a pivotal neuronal growth factor, involved in embryonic and adult neurogenesis (Park and Poo, 2013; Vilar and Mira, 2016), it is secreted from neurons and other cells in an activity-dependent manner, coupled to firing of APs, exerting its effect through both paracrine and autocrine routes, and binding of two different receptors: TrkB and p75. BDNF expression and secretion regulates synaptic plasticity in the hippocampus (Leal et al., 2015), is linked to AMPA and NMDA-mediated increase in Ca2<sup>+</sup> influx, and can trigger PKC-mediated inhibition of GABAergic postsynaptic signaling (Henneberger et al., 2002; Slack et al., 2004). Critical roles have been found for BDNF in neurogenesis, dendritogenesis and synaptogenesis, while abnormalities in BDNF activity and specific polymorphisms in its sequence are associated with several diseases, including neurological and psychiatric disorders such as schizophrenia, epilepsy and drug addiction (Nagahara and Tuszynski, 2011). Impairments in BDNF-mediated signaling have also been associated with autism (Connolly et al., 2006; Correia et al., 2010).

The possible role of BDNF dysregulation in FXS and autism has been pioneered by the Castrén lab (Castrén and Castrén, 2014). First, they found that Fmr1 mRNA levels are reduced in vitro in cultured mouse hippocampal neurons, when these cells are incubated with BDNF (Castrén et al., 2002). This effect was BDNF-specific and could not be mimicked by another neurotrophin, NT-3. Moreover, they showed in vivo that transgenic mice, overexpressing TrkB, display reduced hippocampal Fmr1 and FMRP levels. However, they found no change in BDNF or TrkB expression in Fmr1 KO mice. Later work uncovered a more complex picture regarding the involvement of BDNF in FXS (Louhivuori et al., 2011). BDNF protein levels were found to be increased in the hippocampus, and decreased in the cortex, of Fmr1 KO mice, indicating that the possible role of BDNF in FXS is region-specific. However, both cortical and hippocampal neurons from Fmr1 KO mice showed increased dendritic localization of BDNF-mRNA, in basal conditions and upon pharmacological induction of seizures. These observations are consistent with the role of FMRP as a local-dendritic negative regulator of mRNA translation. Furthermore, this work also suggested that abnormalities in BDNF-TrkB signaling might explain abnormal differentiation and migration of Fmr1 KO neural precursor cells. Subsequently, Bdnf KO mice were cross-bred with Fmr1 KO mice, creating double-mutant FXS animals with reduced BDNF expression (Uutela et al., 2012). In these double mutant mice, hippocampusdependent learning and memory (i.e., Morris water maze) was shown to be negatively affected, as compared to WT, Bdnf KO, and Fmr1 KO mice. However, other behaviors were positively impacted in the double-mutant mice, including locomotor activity and startle responses to loud noises, showing again region-specificity for BDNF role in FXS. They also explored the role of fluoxetine (Prozac), a selective serotonin reuptake inhibitor, in Fmr1 KO mice (Uutela et al., 2014). Chronic administration of fluoxetine was previously shown to increase the expression of BDNF-induced LTP-associated genes in a brain region-specific pattern in mice (Alme et al., 2007), and to increase BDNF serum levels in humans (Liu et al., 2014). Fmr1 KO mice treated with fluoxetine showed mixed results. Anxiety-like behavior was reduced in Fmr1 KO mice, but also in WT counterparts. Locomotor hyperactivity was corrected in Fmr1 KO mice by fluoxetine, but exploratory activity was abnormally high following treatment. At the cellular level, they found that fluoxetine significantly increased hippocampal cell proliferation in WT, but not in Fmr1 KO mice, consistent with the idea that FMRP loss dysregulates BDNF signaling, and indicating that this effect might not be reversible. Finally, a more recent study showed the potential clinical significance of BDNF as therapeutic target in FXS (Nomura et al., 2017). In this study, the physiological maturation of fast-spiking interneurons in the sensory cortex of neonates was found to be delayed in Fmr1 KO mice, and rescued by chronic delivery of a TrkB agonist. The study shows how a temporary decrease in TrkB activation during a critical period of synaptogenesis and circuit formation could be responsible for many of the deficits observed in FXS, and how restoration of TrkB activation could reverse these effects. However, the study did not show any data indicating that adult Fmr1 KO mice show signs of behavioral rescue, if they are treated with a TrkB agonist as neonates. Yet, it strengthens the idea that BDNF role in FXS could be pivotal to the development of a successful treatment or cure.

#### cAMP

The cyclic adenosine monophosphate (cAMP) signaling pathway is virtually ubiquitous in mammalian tissues. In neurons, cAMP signaling through the TF cAMP-responsive element bindingprotein (CREB), has been involved in regulation of synaptic transmission, plasticity, and neurogenesis (Nicol and Gaspar, 2014; Ortega-Martínez, 2015). G-protein coupled receptors activate the membrane-bound enzyme adenylyl cyclase, which synthesizes cAMP from ATP, activating CREB. cAMP itself also serves as a ligand to many enzymes, most importantly PKA and cyclic nucleotide-gated channels such as HCN, which are crucial for membrane depolarization and neuronal spiking (Waltereit and Weller, 2003; Seino and Shibasaki, 2005; Biel and Michalakis, 2009; Baudry et al., 2015).

Several studies point toward possible abnormalities in the cAMP pathway in FXS and ASDs. Early studies showed a reduced production of cAMP in human FXS platelets, and that overexpression of FMR1 in human neural cells in vitro results in increased cAMP levels (Berry-Kravis and Sklena, 1993; Berry-Kravis et al., 1995; Berry-Kravis and Ciurlionis, 1998). These results were later corroborated in the same human in vitro cellular systems, as well as in Fmr1 KO mice (Kelley et al., 2007), suggesting the existence of a conserved role for an ''FMRPcAMP pathway.'' Importantly, altered cAMP signaling has also been correlated with autism (Kelley et al., 2008). Based on this body of evidence, cAMP became the target of a new therapeutic strategy, in which the goal is to increase cAMP levels in FXS patients. One study used Rolipram, an inhibitor of PDE-4, the cAMP-degrading phosphodiesterase, to increase the levels of cAMP in ex vivo acute hippocampal slices of Fmr1 KO mice (Choi et al., 2015). Acute treatment with Rolipram resulted in a decrease in mGluR-dependent LTD, rescuing the Fmr1 KO phenotype. A subsequent study raised the hypothesis that other drug candidates for FXS and ASDs treatment, including antagonists against mGluRs and GSK3β, also work by increasing cAMP levels (Choi et al., 2016). The method used to quantify the effect of these drugs was semi-quantitative western blot ratios between target proteins in their phosphorylated state vs. non-phosphorylated, and the ratio of the target protein to Tubulin, in hippocampal lysates of WT and Fmr1 KO mice. Also, ERK activation was measured in lymphocytes. However, and most importantly, no electrophysiological or behavioral data from mice was provided, and it was not shown whether these treatments indeed increase cAMP concentration in vivo in mice. Moreover, many important questions remain open, it is yet not clear how does FMRP increase cAMP levels, and how exactly reduced levels of cAMP cause enhanced mGluR-dependent LTD.

#### GSK3β

The Wnt/GSK3β pathway is essential for both embryonic and adult neurogenesis (Toledo et al., 2008; Kuwabara et al., 2009; Gage and Temple, 2013; Bengoa-Vergniory and Kypta, 2015). Wnt is the collective name of a family of secreted protein ligands that bind to membrane-bound receptors in the target cell, activating a signaling cascade. This cascade hits a main ''crossroads'' when it activates cytoplasmic GSK3β, phosphorylating it. Phospho-GSK3β activates and inhibits several different pathways, including MAP-kinase, Cyclin and Akt, as well as β-Catenin as part of the canonical Wnt signaling pathway, resulting in complex gene regulation events that are yet not fully understood (Hur and Zhou, 2010; Seira and Del Río, 2014).

In FXS, it has been shown that GSK3β is elevated in the hippocampi of Fmr1 KO mice, leading to abnormal adult hippocampal neurogenesis (Portis et al., 2012). Ablation of FMRP in aNSCs caused an increase in GSK3β protein levels, causing a phosphorylation-dependent decrease in β-Catenin activation and an increase in β-Catenin degradation, impairing neuronal differentiation (Luo et al., 2010), while pharmacological inhibition of GSK3β led to rescue of impaired neurogenesis in vivo (Guo et al., 2012). Moreover, several studies showed that lithium, an FDA-approved drug that inhibits GSK3β activity, alone or in conjunction with mGluR5 antagonists, mitigated many of the symptoms of FXS in Fmr1 KO mice, from the cellular level to behavior associated with FXS and ASDs (Min et al., 2009; Yuskaitis et al., 2010; Mines and Jope, 2011).

However, many different lines of evidence argue against the idea of GSK3β as a legitimate target for FXS treatment. First, a study in which mice were fed with control or lithiumcontaining chow, reported that a lithium-rich diet failed to significantly alter the phosphorylated-to-non-phosphorylated GSK3β ratio (p-GSK3β/GSK3β), in the hippocampi of WT or Fmr1 KO mice (Choi et al., 2016). They also demonstrated that even under a normal control diet, WT and Fmr1 KO mice display non-significant difference in their p-GSK3β/GSK3β ratio. Second, it is possible that hyperactivation of GSK3β in Fmr1 KO mice is rodent-specific. When FXS-hESCs were differentiated into human neural progenitor cells, no significant change in the transcription or translation of GSK3β or β-Catenin was found (Telias et al., 2015b). Overexpression of FMR1 in FXS-cells and siRNA-mediated knock-down of FMR1 expression in WT counterparts, failed to alter the levels of GSK3β or β-Catenin. Pharmacological inhibition of either GSK3β or β-Catenin affected neuronal differentiation of WT and FXS-human neural precursor cells similarly, regardless of FMRP levels. Finally, and most importantly, in the first clinical trial for lithium treatment of FXS patients, in which a placebo control was not included, the results sadly fell short of the stipulated clinical goals, while improvements were observed only in minor aspects of FXS pathology (Berry-Kravis et al., 2008; Liu and Smith, 2014). Currently, 10 years later, there are no active clinical trials involving lithium treatment for FXS or autism.

#### Transcription Factors

During embryonic and adult neurogenesis, all gene expression must be triggered by the timely and correct activation of TFs. Different groups of TFs dictate the embryonic and adult neurogenesis of neural progenitor cells. One such group of TFs is known as the SOX superfamily, which in humans includes 20 different genes arranged in nine different sub-groups (Wegner and Stolt, 2005; Kiefer, 2007). For example, activation of SOX2 is critical for the development of neural progenitor cells, but also for maintaining their specific identity, meaning that for the progenitor to progress into a fully differentiated neuron, SOX2 expression should be de-activated. Our own study has shown a causative link between disappearance of FMRP and a significant increase in SOX2 levels in human FXS-neural precursor cells derived from FXS-hESCs (Telias et al., 2015b), coupled with poor neurogenesis in FX cells as compared to WT counterparts. Another member of the same family, SOX9, has been shown to induce neural crest development, gliogenesis and chondrogenesis (Marshall and Harley, 2000; Lee and Saint-Jeannet, 2011; Lefebvre and Dvir-Ginzberg, 2017). Several studies have pointed toward abnormal gliogenesis as a contributing factor in FXS and autism pathophysiology (Cheng et al., 2012), and loose, incompletely formed cartilage is a hallmark of non-neuronal symptoms of FXS, as well as craniofacial abnormalities (Penagarikano et al., 2007). We found that loss of FMRP results in a significant decrease in SOX9 expression, which is reversed by overexpressing FMR1 in human FXS neural progenitor cells and mimicked in WT cells by inhibiting FMR1 expression.

Other studies have also shown abnormal expression of TFs in human FX cells. We reported on abnormal expression of PAX6 and NOTCH1 during neural differentiation of FXS-hESCs (Telias et al., 2013). Halevy et al. (2015) studied the expression of repressor element-1 silencing TF (REST) in undifferentiated FXS-hiPSCs and their derived neurons. They found that undifferentiated FXS-hiPSCs have reduced expression of REST as compared to WT counterparts, but differentiated FXS-neurons display increased REST expression. REST is a negative regulator of neuronal development, it is expressed in early neural lineages but inactivated in mature neurons, similar to SOX2. Therefore, data from independent studies carried out on different in vitro human models of FXS (Halevy et al., 2015; Telias et al., 2015b), seem to indicate that FMRP loss results in a failure to inactivate the expression of negative regulators of neurogenesis, an idea that is consistent with the role of FMRP as a negative regulator of translation, even though direct protein-mRNA interaction was not demonstrated. In addition, this line of research does not provide a clear and identifiable pharmacological target with therapeutic potential.

#### SUMMARY AND CONCLUSIONS

Since the discovery of the FMR1 mutation as the cause of FXS, many hypotheses have been proposed on how lack of FMRP results in dysfunctional synaptic activity. In this review, I tried to summarize most of these hypotheses, focusing predominantly on evidence obtained from the Fmr1 KO mouse and, more recently, from human-based models, especially hPSCs.

Given the symptoms of FXS and ASDs, rationally-built hypotheses were put to the test, resulting in outstanding scientific discoveries. The ''mGluR theory of FXS'' proposes that lack of FMRP leads to an increase in mGluRI-dependent LTD, resulting in reduced LTP, therefore explaining the cognitive impairment and intellectual disability. The ''GABAergic theory of FXS'' proposes that lack of FMRP leads to reduced GABA-mediated inhibition, explaining neuronal hyperexcitation, behavioral hyperactivity and epilepsy in FXS and ASDs. Newer studies suggest that loss of FMRP affects the intrinsic properties of the neuron itself, resulting in abnormal ion channel activity and firing pattern, decreased neurotransmitter release and overall reduced synaptogenesis, which can explain many of the symptoms characterizing FXS, especially from a developmental perspective. Many of these studies have also provided the clinical world with accessible targets for pharmacological treatment: from neurotransmitter receptors, through ion channels, to cytosolic enzymes. However, so far, none of

#### REFERENCES


these hypotheses has been shown as definitive, or fully elucidated from a mechanistic point of view; nor as the source of a successful treatment or cure for FXS and ASDs in clinical trials.

The criticism of current FXS hypotheses in this review does not call into question the quality and the value of the research done and the data obtained, but the relevance of the model employed, in each specific case. Extrapolating from mouse to human, or from cell cultures to whole organisms, is a complex issue, for which definitive standards do not necessarily apply. For many years, it was hard to compare the results from mice to anything else. Today, hPSCs, and other human-based models, can be used to answer basic scientific questions, as well as to find molecular correlates with mouse data, increasing the relevance of the research to accelerate the finding of a suitable treatment for FXS and ASDs.

#### AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and has approved it for publication.

#### FUNDING

This work was supported by University of California, Berkeley, Berkeley, CA, USA.

rodent models of autism. Science 338, 128–132. doi: 10.1126/science.12 24159


**Conflict of Interest Statement**: The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Telias. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

digital media

of impactful research

article's readership