# UBIQUITIN AND THE BRAIN: ROLES OF PROTEOLYSIS IN THE NORMAL AND ABNORMAL NERVOUS SYSTEM

EDITED BY: Ashok N. Hegde and Fred W. van Leeuwen PUBLISHED IN: Frontiers in Molecular Neuroscience

### *Frontiers Copyright Statement*

*© Copyright 2007-2017 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.*

*The compilation of articles constituting this e-book, wherever published, as well as the compilation of all other content on this site, is the exclusive property of Frontiers. For the conditions for downloading and copying of e-books from Frontiers' website, please see the Terms for Website Use. If purchasing Frontiers e-books from other websites or sources, the conditions of the website concerned apply.*

*Images and graphics not forming part of user-contributed materials may not be downloaded or copied without permission.*

*Individual articles may be downloaded and reproduced in accordance with the principles of the CC-BY licence subject to any copyright or other notices. They may not be re-sold as an e-book.*

*As author or other contributor you grant a CC-BY licence to others to reproduce your articles, including any graphics and third-party materials supplied by you, in accordance with the Conditions for Website Use and subject to any copyright notices which you include in connection with your articles and materials.*

> *All copyright, and all rights therein, are protected by national and international copyright laws.*

*The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use.*

ISSN 1664-8714 ISBN 978-2-88945-285-9 DOI 10.3389/978-2-88945-285-9

# 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

# **UBIQUITIN AND THE BRAIN: ROLES OF PROTEOLYSIS IN THE NORMAL AND ABNORMAL NERVOUS SYSTEM**

Topic Editors:

**Ashok N. Hegde,** Georgia College and State University, United States **Fred W. van Leeuwen**, Maastricht University, Netherlands

Foreground: Dorsal view of the human brain with left hemisphere showing normal brain and the right hemisphere showing the brain shrunken as a result of Alzheimer's disease (AD). Overlaid on the brain are the schematic figures of the proteasome (center), protein substrate tagged with polyubiquitin (top right) and fragments of the polyubiquitinated protein (bottom left) depicting degradation by the proteasome.

Background: Immunohistochemistry of a 50 µm section of the hippocampus taken from the postmortem brain of a 92 year old female AD patient who had suffered from the sporadic form of the disease. The staining shows UBB+1, a frameshifted mutant form of ubiquitin with 19 extra amino acids at the C-terminus, in the pyramidal cells in the CA1 region.

Copyright ©2017 by Ashok N. Hegde and Fred W. van Leeuwen

Proteolysis by the ubiquitin-proteasome pathway (UPP) in the nervous system has been extensively studied both in the context of normal physiological function as well as abnormal pathological conditions. Although ubiquitin was used as a marker of brain pathology, the normal functions of the UPP were not studied much in the nervous system until the 1990s. The early investigations focused on synaptic plasticity which was followed by studies on the roles of protein degradation in the development of the nervous system. Research on the role of abnormal roles of the UPP follows a parallel trajectory. Since the 2000s, the field has grown to encompass many subareas of research and several model systems. Despite the progress made, many unanswered questions still remain. For example, there are many unknowns about the precise spatial and temporal control of protein degradation in the normal nervous system. With respect to the roles of proteolysis in brain pathology a major challenge is to elucidate the connection between impaired protein degradation and disease progression. In addition, in-depth studies of the roles of ubiquitin-proteasome-mediated proteolysis in neurodegenerative diseases are promising in identifying therapeutic targets. This ebook contains original research papers and insightful reviews that cover several aspects of proteolysis by the UPP and its physiological as well as pathological functions in the nervous system.

**Citation:** Hegde, A. N., van Leeuwen, F. W., eds. (2017). Ubiquitin and the Brain: Roles of Proteolysis in the Normal and Abnormal Nervous System. Lausanne: Frontiers Media. doi: 10.3389/978-2-88945-285-9

# Table of Contents


*124 The Ubiquitin-Proteasome System: Potential Therapeutic Targets for Alzheimer's Disease and Spinal Cord Injury*

Bing Gong, Miroslav Radulovic, Maria E. Figueiredo-Pereira and Christopher Cardozo

*140 The ubiquitin-proteasome system in neurodegenerative diseases: precipitating factor, yet part of the solution*

Nico P. Dantuma and Laura C. Bott

*158 The ubiquitin proteasome system in glia and its role in neurodegenerative diseases*

Anne H. P. Jansen, Eric A. J. Reits and Elly M. Hol

*172 Subcellular Clearance and Accumulation of Huntington Disease Protein: A Mini-Review*

Ting Zhao, Yan Hong, Xiao-Jiang Li and Shi-Hua Li


# Editorial: Ubiquitin and the Brain: Roles of Proteolysis in the Normal and Abnormal Nervous System

Ashok N. Hegde<sup>1</sup> \* and Fred W. van Leeuwen<sup>2</sup>

*<sup>1</sup> Department of Biological and Environmental Sciences, Georgia College and State University, Milledgeville, GA, United States, <sup>2</sup> Department of Neuroscience, Faculty of Health, Medicine and Life Sciences, Maastricht University, Maastricht, Netherlands*

Keywords: proteasome, synaptic plasticity (LTP/LTD), neurodegenerative diseases, Alzheimer disease, Parkinson's disease, Huntington's disease, glutamate receptors, local protein degradation

**Editorial on the Research Topic**

## **Ubiquitin and the Brain: Roles of Proteolysis in the Normal and Abnormal Nervous System**

Proteolysis by the ubiquitin-proteasome pathway (UPP) is now widely recognized as a major molecular mechanism playing a role in numerous normal functions of the nervous system as well as in malfunctions of the brain in several neurodegenerative diseases. In the UPP, attachment of a small protein, ubiquitin, tags the substrates for degradation by a multi-subunit complex called the proteasome. Linkage of ubiquitin to protein substrates is highly specific and occurs through a series of well-orchestrated enzymatic steps. Protein degradation has key functions in the nervous system including fine-tuning of synaptic connections during development and synaptic plasticity in the adult organism (Hegde, 2017). In neurons, several physiological processes are regulated by proteolysis. From gene transcription to posttranslational modification of proteins, several quality checks are essential before the protein is ready for biological action, locally in dendrites or distantly via axonal transport. Beyond regulation at the RNA level, proteolysis by the UPP provides a quick and efficient way to regulate the amount of protein in neurons. In addition, accurate folding and control over levels of many proteins must be tightly regulated both spatially and temporally. To achieve this function, the cell possesses a network of different protein quality control systems (PQC) for protein folding via molecular chaperones as the first line of defense against protein misfolding and aggregation (Ciechanover and Kwon, 2017). Subsequently accurate protein degradation by the UPP and the autophagosomal-lysosomal system is the second line of defense. In addition, recent data suggest an additional PQC pathway in which misfolded proteins are excreted actively after encapsulation at the endoplasmic reticulum, a process dubbed MAPS (Misfolded Associated Protein Secretion) (Lee et al., 2016). It is even speculated that a dysfunctional PQC contributes to the process of proteopathic seeding in Alzheimer's disease (AD) (Gentier and van Leeuwen, 2015). In addition, dysfunction of the UPP is linked to Parkinson's, Huntington's, and other neurodegenerative diseases (Hegde, 2017). Perturbation in the UPP is also believed to play a causative role in mental disorders such as Angelman syndrome (Jiang and Beaudet, 2004).

Many questions pertaining to the UPP in the nervous system remain unanswered. How is the UPP-mediated degradation regulated spatially and temporally in neurons? What is the role of local protein degradation? How does the interplay between proteolysis and protein synthesis affect synaptic plasticity and memory? Do perturbations in the UPP have a role in the pathology of neurodegenerative diseases?

### Edited and reviewed by:

*Nicola Maggio, The Chaim Sheba Medical Center, Israel*

> \*Correspondence: *Ashok N. Hegde ashok.hegde@gcsu.edu*

Received: *08 June 2017* Accepted: *26 June 2017* Published: *18 July 2017*

### Citation:

*Hegde AN and van Leeuwen FW (2017) Editorial: Ubiquitin and the Brain: Roles of Proteolysis in the Normal and Abnormal Nervous System. Front. Mol. Neurosci. 10:220. doi: 10.3389/fnmol.2017.00220*

The papers in this Frontiers in Molecular Neuroscience Research Topic-Special Issue on Ubiquitin and the Brain: Roles of Proteolysis in the Normal and Abnormal Nervous System cover a wide range of topics from development of methods and primary research studies to reviews. These articles from researchers working on yeast, neuronal and glial cell culture and mouse models, often validated in postmortem human brain tissue, cover a wide array of topics such as receptor endocytosis and synaptic plasticity in the normal nervous system to abnormalities of the nervous system such as AD and Huntington's disease.

The paper by Pinto et al. reports the results of studies on visualizing a specific type of ubiquitin linkage to substrate proteins in which ubiquitin molecules are linked to each other through a covalent linkage to Lysine-48 (K48) in the ubiquitin sequence. The K-48 linkage targets substrate proteins for degradation. Pinto et al. study adapted a technique to monitor K-48-liked ubiquitin molecules in cultured rat hippocampal neurons. In this technique, a yellow fluorescent protein called Venus is split into non-fluorescent N- and C- termini and fused to sequences containing ubiquitin-interacting-motifs (UIMs). These parts come together when they bind to closely spaced ubiquitin molecules in a polyubiquitin chain assembled through K-48 linkage. Using this technique, the authors show that proteins tagged with K-48-linked ubiquitin chains accumulate in presynaptic terminals when synapses newly form.

The studies by Scott Wilson and colleagues (Vaden et al.) builds on their previous work on the role of a deubiquitinating (DUB) enzyme called USP14 in the mammalian neuromuscular junction (NMJ). Their studies suggest that the role of USP14 in maintaining the free ubiquitin levels is critical for NMJ structure but the function of NMJ (such as muscle coordination) is likely to be regulated by USP14 through a separate pathway. Apart of USP14, many other DUBs play a role in the nervous system. Ristic et al. argue in their review that looking at the UPP from the point of view of DUBs can provide novel insights into the exquisite regulation of proteolysis in the nervous system and might be helpful in devising therapeutic approaches.

Goo et al. take a look at the role of ubiquitin in regulating trafficking of AMPA-type glutamate receptors. After internalization AMPA receptors can either be recycled back to the plasma membrane or degraded by the lysosome or the proteasome. For example, Nedd4-1, a ubiquitin ligase, is responsible for targeting AMPA receptors to proteasomemediated degradation. Two deubiquitinating enzymes called Usp8 and Usp46 remove the ubiquitin attachment on AMPA receptors and thus appears to be important for rescuing the receptors from degradation and recycling them back to plasma membrane.

Rodriguez et al. examine the question of increased longevity in female mice with chronic rapamycin treatment and the effect of the drug on proteasome activity and expression of chaperone proteins. They show that rapamycin treatment has different effects on various tissues of the body and increases the amount of proteasome-interacting chaperone proteins in female but not in male mice.

Stojkovic et al. review the key role of ubiquitination in modifying the clock proteins which determine the circadian rhythm. They propose that ubiquitination is a key part of the posttranslational modification "code" that determines the fate and function of a particular clock protein.

Hegde et al. review the accumulated evidence on local roles of protein degradation in induction and maintenance of longterm synaptic plasticity. Because neurons are highly polarized cells, understanding local proteolysis is likely to be important both from a basic science point of view and translational research. Local neuronal function is also emphasized in the mini-review by Zhao et al. with respect to clearance of mutant huntingtin (mHtt) protein. Tagging of mHtt by ubiquitin plays a role in clearance by the proteasome as well as through autophagy. The authors present strategies to study local clearance and accumulation in neuronal sub-compartments.

Jarome and Helmstetter review the evidence for the roles of protein degradation in long-term memory (LTM) storage. They discuss the interplay between protein degradation and protein synthesis in the hippocampus and amygdala and other cortical areas during formation and consolidation of LTM.

The coverage of the role of proteolysis by the UPP in abnormalities of the nervous system ranges from drug abuse to neuroinflammation and neurodegenerative diseases. Massaly et al. review the evidence for role of the UPP in mediating the effects of drug abuse on the nervous system. They details the role of the UPP in regulating molecules that mediate drug-abuse-related neuroplasticity. In addition, they present evidence from the literature regarding how drugs of abuse regulate amounts and function of the various components of the UPP. The review by Figueiredo et al. discusses the role in neuroinflammation of a specific class of prostaglandins called J2 which are the toxic products of cyclooxygenases. Among the pleiotropic effects of J2 prostaglandins is perturbation of UPP. The authors suggest that the targeting neuroinflammatory pathways stimulated by J2 prostaglandins might be beneficial in treating several neurodegenerative diseases such as AD, Parkinson's and amyotrophic lateral sclerosis. Gong et al. discuss the role of components of the UPP such as the deubiquitinating enzyme Uch-L1, F-Box protein Fbx2, and the proteasome in development of AD and spinal cord injury. They suggest pharmacological manipulations of these components could be developed as a therapeutic strategy. In addition, the role of UPP is highlighted in neurodegenerative diseases such as AD, Parkinson's disease and Huntington's disease (papers by Ortega and Lucas; Atkin and Paulson; Jueneman et al.).

Until recently, the investigation with regard to the role of proteolysis in neurodegenerative diseases largely focused on neurons and role of the UPP in glia was largely ignored. This has been true also of other diseases with protein conformational abnormalities such as Amyotrophic Lateral Sclerosis and Parkinson's disease. The review by Jansen et al. tackles this issue and discusses among other things the differences between neurons and glia in how the two cell types react to impaired proteolysis by the UPP.

The studies on the role of the UPP in neurodegenerative diseases can be greatly benefited by work on simple model systems such as yeast. This model system provides an easy readout of perturbations in the UPP and allows quick analysis of action of numerous chemicals on the UPP (see review by Brau), which can be subsequently validated using mammalian model systems. The group of Dantuma has developed many tools to detect proteasomal activity (e.g., constructs with GFP and YFP) in vitro as well asin vivo. The authors (Dantuma and Bott) discuss "the paradigm shift that has repositioned the UPS from being a prime suspect in the pathophysiology of neurodegeneration to an attractive therapeutic target that can be harnessed to accelerate the clearance of disease-linked proteins" (review by Dantuma and Bott). For example, downregulation misframed ubiquitin (UBB+<sup>1</sup> ) that interacts in AD with γ-secretase (Gentier and van Leeuwen, 2015) could be a potential therapeutic target to restore neuronal function perturbed by abnormalities in the UPP. Based on accumulating evidence, it is clear that the neuronalglial network of the brain functions effectively when protein degradation is normal and impairments in the UPP contribute to the early cellular abnormalities seen in AD.

The research articles and reviews in this research topic now collected as an e-book highlight the complex role of ubiquitinproteasome-mediated proteolysis in the neuronal and glial physiology and pathology. In spite of the considerable progress made over the last two decades, many challenges remain. For example, with respect to normal roles of the UPP in the nervous system, the roles of the UPP in memory formation especially in relation to the role of protein synthesis need to be better understood. The exact contributions and the interplay of the UPP with other protein-clearance systems such as autophagy in various neurodegenerative diseases also need to be delineated. As many new technological developments such as CRISPR/Cas9, allow precise testing of hypotheses, we may expect to see many additional interesting discoveries on both these fronts.

# AUTHOR CONTRIBUTIONS

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

# ACKNOWLEDGMENTS

The research in the ANH's laboratory was supported by a grant from National Institute of Neurological Disease and Stroke (NINDS) (NS098405) and startup funds from Georgia College and State University.

# REFERENCES


proteasome dysfunction in mammalian cells. Nat. Cell Biol. 18, 765–776. doi: 10.1038/ncb3372

**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 Hegde and van Leeuwen. 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.

# Visualizing K48 Ubiquitination during Presynaptic Formation By Ubiquitination-Induced Fluorescence Complementation (UiFC)

Maria J. Pinto1,2 , Joana R. Pedro<sup>1</sup> , Rui O. Costa<sup>1</sup> and Ramiro D. Almeida1,3,4 \*

<sup>1</sup> Center for Neuroscience and Cell Biology (CNC), University of Coimbra, Coimbra, Portugal, <sup>2</sup> PhD Programme in Experimental Biology and Biomedicine (PDBEB), Center for Neuroscience and Cell Biology, University of Coimbra, Coimbra, Portugal, <sup>3</sup> School of Allied Health Technologies, Polytechnic Institute of Porto (ESTSP-IPP), Vila Nova de Gaia, Portugal, 4 Institute for Interdisciplinary Research, University of Coimbra, Coimbra, Portugal

In recent years, signaling through ubiquitin has been shown to be of great importance for normal brain development. Indeed, fluctuations in ubiquitin levels and spontaneous mutations in (de)ubiquitination enzymes greatly perturb synapse formation and neuronal transmission. In the brain, expression of lysine (K) 48-linked ubiquitin chains is higher at a developmental stage coincident with synaptogenesis. Nevertheless, no studies have so far delved into the involvement of this type of polyubiquitin chains in synapse formation. We have recently proposed a role for polyubiquitinated conjugates as triggering signals for presynaptic assembly. Herein, we aimed at characterizing the axonal distribution of K48 polyubiquitin and its dynamics throughout the course of presynaptic formation. To accomplish so, we used an ubiquitination-induced fluorescence complementation (UiFC) strategy for the visualization of K48 polyubiquitin in live hippocampal neurons. We first validated its use in neurons by analyzing changing levels of polyubiquitin. UiFC signal is diffusely distributed with distinct aggregates in somas, dendrites and axons, which perfectly colocalize with staining for a K48-specific antibody. Axonal UiFC aggregates are relatively stable and new aggregates are formed as an axon grows. Approximately 65% of UiFC aggregates colocalize with synaptic vesicle clusters and they preferentially appear in the axonal domains of axo-somatodendritic synapses when compared to isolated axons. We then evaluated axonal accumulation of K48 ubiquitinated signals in bead-induced synapses. We observed rapid accumulation of UiFC signal and endogenous K48 ubiquitin at the sites of newly formed presynapses. Lastly, we show by means of a microfluidic platform, for the isolation of axons, that presynaptic clustering on beads is dependent on E1-mediated ubiquitination at the axonal level. Altogether, these results indicate that enrichment of K48 polyubiquitin at the site of nascent presynaptic terminals is an important axon-intrinsic event for presynaptic differentiation.

### Keywords: ubiquitination, presynaptic terminal, presynaptic differentiation, axon development, lysine 48 polyubiquitin

**Abbreviations:** K, lysine; PDL, poly-D-lysine; UiFC, ubiquitination-induced fluorescence complementation; UPS, ubiquitin-proteasome system; VGluT1, vesicular glutamate transporter 1.

### Edited by:

Ashok Hegde, Georgia College and State University, USA

### Reviewed by:

Ralf J. Braun, Universität Bayreuth, Germany Shasta L. Sabo, Case Western Reserve University, USA

> \*Correspondence: Ramiro D. Almeida ramirodalmeida@gmail.com

Received: 29 February 2016 Accepted: 24 May 2016 Published: 10 June 2016

### Citation:

Pinto MJ, Pedro JR, Costa RO and Almeida RD (2016) Visualizing K48 Ubiquitination during Presynaptic Formation By Ubiquitination-Induced Fluorescence Complementation (UiFC). Front. Mol. Neurosci. 9:43. doi: 10.3389/fnmol.2016.00043

# INTRODUCTION

Neurons are highly complex and polarized cells with a remarkable network of functionally active processes that extend outwards the cell body. Within the brain, each neuron's axon establishes thousands of synaptic contacts with neighboring or fairly distantly located neurons. Differentiation of presynaptic terminals occurs early in development (Steward and Falk, 1991; Bury and Sabo, 2010) and it comprises recruitment and coordinated clustering of presynaptic material that can be found along the axon in the form of cell body-derived mobile units (Jin and Garner, 2008; Pinto and Almeida, 2016). Specificity to this phenomenon is conferred by cues derived from the postsynaptic partner that by activating axonal receptors trigger presynaptic assembly (Johnson-Venkatesh and Umemori, 2010; Siddiqui and Craig, 2011). Despite the huge number of proteins known to be implicated in presynaptic differentiation, this is a highly rapid event that occurs in a time-scale of minutes to few hours (Friedman et al., 2000; Bresler et al., 2004). Furthermore, axons are extremely long and presynaptic terminals are to be formed in remote sites. On top of this, each presynaptic site is an individual micro-domain, meaning that changes in one do not necessarily affect adjacent segments. In light of these circumstances, axons are believed to rely on intra-axonal mechanisms to support and sustain their prompt response to cues, subsequently leading to a site-specific clustering of presynaptic components. Indeed, there are predefined sites along the axon shaft in which en passant presynaptic terminals selectively form (Krueger et al., 2003; Sabo et al., 2006), thus establishing the importance of intrinsic axonal mechanisms.

Ubiquitin is a highly conserved small protein that is covalently attached to other proteins in the form of a single monomer, monoubiquitination, or as a chain of ubiquitins, polyubiquitination (Komander and Rape, 2012). All seven internal lysines in ubiquitin can serve as attachment sites for other ubiquitins, and so, different chain types can be formed, which differently alter properties of the target protein and are involved in a multitude of cellular processes (Komander and Rape, 2012; Sadowski et al., 2012). Of particular relevance is its role as a tag for proteasomemediated degradation mainly through lysine 48 and 11-linked polyubiquitin chains, in the so-called ubiquitin-proteasome system (UPS; Kulathu and Komander, 2012; Kleiger and Mayor, 2014).

Although very much less explored, signaling through ubiquitin is also likely to play a role in presynapse development. The ataxia mice ax<sup>J</sup> , with a loss-of-function mutation in the proteasome-associated deubiquitinating enzyme Usp14 and concomitant decreased synaptic levels of monomeric and conjugated ubiquitin, display severe malformation of the neuromuscular junction and impaired presynaptic function (Wilson et al., 2002; Chen et al., 2009). These defects are rescued by restoration of ubiquitin levels (Chen et al., 2011). Contrariwise, transgenic mice overexpressing ubiquitin also display impaired formation of presynapses (Hallengren et al., 2013), thus reinforcing that tightly balanced ubiquitin levels are crucial for proper synaptic development. Furthermore, similar presynaptic defects are also observed in mice carrying mutations in the E3 ubiquitin ligases HERC1 (Bachiller et al., 2015) and PHR (Burgess et al., 2004; Saiga et al., 2009). Interestingly, the Drosophila and C. elegans homologs of PHR have been shown to function locally in modulating the triggering cascades that guide presynaptic differentiation (Liao et al., 2004; Nakata et al., 2005; Collins et al., 2006). Altogether, these observations point to a fundamental role for ubiquitination in the events launching presynaptic assembly.

Notwithstanding, the mechanistic role of ubiquitin in vertebrate presynaptic formation is still unclear. We have made significant advances in the field by demonstrating that proteasome-related polyubiquitin signals trigger presynaptic assembly (Pinto et al., 2016), which is in line with the higher expression of lysine 48 ubiquitin chains at the peak of synapse formation (Chen et al., 2011) and the high number of embryonic ubiquitinated proteins involved in synaptogenesis (Franco et al., 2011). In the study reported here, we exploited the ubiquitination-induced fluorescence complementation (UiFC) assay (Chen et al., 2013) to look closely to K48 ubiquitination along axons and its relation to sites of presynaptic clustering. In contrast to previous ubiquitinbased fluorescence complementation approaches, which allow for detection of substrate-specific ubiquitination (Fang and Kerppola, 2004; Kerppola, 2006), UiFC detects endogenous conjugation of K48 ubiquitin chains due to favored binding of UiFC's ubiquitin-interacting motifs and reconstitution of fused Venus fragments (Chen et al., 2013). Using UiFC in neuronal cultures, we show that axonal aggregates of enhanced K48 polyubiquitination are mostly stable and preferably located at axo-somatodendritic regions. Moreover, the majority colocalizes with presynaptic clusters. We further observed that axonal localized enrichment of K48 polyubiquitin occurs rapidly upon contact with a synaptic partner. Lastly, we demonstrate that clustering of presynaptic material requires ubiquitination at the axon level. Overall, we propose that site-specific enrichment of K48 polyubiquitinated conjugates supports presynaptic formation and this effect is dependent on axonal E1-mediated ubiquitination.

# MATERIALS AND METHODS

# Primary Neuronal Cultures

Animals were maintained at the animal house of the Center for Neuroscience and Cell Biology (University of Coimbra, Portugal) approved by the Portuguese National Authority for Animal Health (DGAV). Primary cultures of rat hippocampal neurons were prepared from E17 Wistar rat embryos, as previously described (Baptista et al., 2010; Baeza et al., 2012) with minor changes. Hippocampi were dissected and dissociated in 0.045% trypsin/0.01% v/v deoxyribonuclease in HBSS for 15 min at 37◦C. Then washed once in plating media (supplemented with 0.026 MEM NaHCO3, 0.025 M glucose, 1 mM sodium pyruvate and 10% FBS), mechanically dissociated and cell density determined. Cells were diluted in plating medium and plated in poly-D-lysine (PDL)-coated surfaces. Plating medium was replaced by culture medium (neurobasal medium supplemented with 2% B27, 25 µM glutamate, 0.5 mM glutamine and 1:400 penicillin-streptomycin) 2–4 h after plating. Cells were maintained in a humidified incubator with 5% CO2/95% air at 37◦C. The mitotic inhibitor 5-FDU was added at days in vitro (DIV) 3/4. Experiments were performed at DIV 7/8.

# Microfluidic Devices

Microfluidic devices were prepared by assembling a molded poly-dimethylsiloxane (PDMS) chamber onto a glass coverslip (Taylor et al., 2005; Pinto et al., 2016). The molds for the PDMS devices used in this study were fabricated by Noo Li Jeon (School of Mechanical and Aerospace Engineering, Seoul National University, Seoul, Korea). PDMS was prepared from the Sylgard 184 Silicone elastomer kit (Dow Corning), poured onto the microfluidic molds and cured for 4–6 h. PDMS devices were assembled on top of glass coverslips (Marienfeld) coated with PDL and laminin.

# Neuron Transfection

F(syn)WRBN-VGluT1mCherry, a vector for expression of a fusion version of VGluT1 to mCherry controlled by the synapsin promoter, was kindly offered by Prof. Etienne Herzog (Interdisciplinary Institute for Neuroscience, Bordeaux, France; Herzog et al., 2011). The constructs for ubiquitinationinduced fluorescence complementation (UiFC), pcDNA3-UiFC-C (UiFC-C) and pcDNA3-UiFC-N (UiFC-N), were kindly offered by Prof. Shengyun Fang (Department of Biochemistry and Molecular Biology, University of Maryland, Baltimore, Maryland, MD, USA; Chen et al., 2013). Also, pcDNA3.1- GFP and pcDNA3.1-mCherry were used for the expression of GFP and mCherry, respectively. Deoxyribonucleic acid (DNA) was recombinantly expressed in primary hippocampal neurons using calcium phosphate transfection at DIV 6/7 (Almeida et al., 2005). Expression was allowed to occur for 16–20 h.

# PDL-Coated Beads and Drugs

PDL-coated beads were prepared at the experiment day. 4.5 µmdiameter aliphatic amine latex beads (Life Technologies) were incubated with PDL for 30 min at 37◦C, washed twice in sterile mQH2O and diluted in culture medium or HEPESbuffered solution imaging medium (119 mM NaCl, 5 mM KCl, 2 mM CaCl2, 2 mM MgCl2, 30 mM glucose, 10 mM HEPES, pH 7.4) for experiments requiring fixation or live imaging, respectively. Beads were added to cultures for the indicated periods of time and incubated at 37◦C. Drug treatment [IU1 (75 µM, Tocris Bioscience), MG132 (1µM, Calbiochem), PR619 or Ziram (1µM, Sigma Aldrich)] was performed in conditioned medium (either culture or imaging medium) by diluting the drug from a 1000×-concentrated stock in DMSO. Equal amounts of DMSO were added to the control condition. Ziram was always prepared fresh before experiment.

# Immunocytochemistry and Image Acquisition

Immunostaining of cultured neurons was performed similarly to previously described (Baptista et al., 2013). Fixation in 4% paraformaldehyde (in PBS with 4% sucrose) for 10 min was followed by washes in PBS. Cells were then permeabilized in PBS with 0.25% Triton X-100 for 5 min, washed once and blocking performed with 3% BSA for at least 30 min. These steps were performed at room temperature. Incubation with primary antibodies in 3% BSA was performed either overnight at 4◦C or for 2 h at 37◦C followed by three washes in PBS. Secondary antibodies were incubated for 1 h at room temperature in 3% BSA and washed (twice with PBS with 0.1% Triton X-100 and once with PBS). The following primary antibodies were used: Bassoon (1:400; #ADI-VAM-PS003; Enzo Life Sciences), GFP (1:1000, #598, MBL), K48 polyubiquitin (Apu2; 1:500; #05-1307, Millipore), MAP2 (1:5000; #AB5543; Chemicon), tau (1:1000; #AB75714; Abcam) and ubiquitin (1:200; #Z0458; Dako, Denmark). Alexa-conjugated secondary antibodies (405, 488, 568 and 647) were used (1:1000; Life Technologies).

Imaging was performed either in a Zeiss Observer Z.1 microscope equipped with a Plan-Apochromat 20× air objective (0.8 NA) or a Plan-NeoFluar 63× oil objective (1.4 NA), an AxioCam HRm camera and Zen Blue 2011 software or in a spinning disk confocal imaging system (CSU-X1M, Yokogawa) configured to a Zeiss Axio Observer Z1 microscope with a LCI Plan-NeoFluar 63× water or glycerol objective (1.3 NA) coupled to an EM-CCD Evolve Delta camera and Zen Black 2012 Software. All conditions within an experiment were processed simultaneously and imaging settings (exposure time and laser power) were conserved. The spinning disk system was used for fixed preparations with beads and for obtaining XY reconstructions of microfluidic devices (with both somal and axonal compartments). For the experiment in which we were interested in determining the density of axonal UiFC and VGluT1mCherry puncta at sites with or without somatodendritic contact, XY reconstructed images were obtained with the 63× objective in the spinning disk.

# Live-Imaging

Experiments involving live cells were all performed in a spinning disk confocal imaging system (CSU-X1M, Yokogawa) configured to a Zeiss Axio Observer Z1 microscope with a LCI Plan-NeoFluar 63× water or glycerol objective (1.3 NA) coupled to an EM-CCD Evolve Delta camera and Zen Black 2012 Software. All experiments were done in imaging medium at 37◦C in a humidified atmosphere to avoid medium evaporation. Correction for focal drift was accomplished by the definite focus feature of Zen Software. z-stacks encompassing the sample of interest were acquired at each time frame. Images were obtained every 1, 5 or 10 min depending on the experiment. Drug treatment or addition of beads was preceded by acquisition of three frames.

Frontiers in Molecular Neuroscience | www.frontiersin.org June 2016 | Volume 9 | Article 43 |

# Quantitative Analysis

Except for analysis of kymographs, quantification was performed using ImageJ Software. All images were converted to 8-bit for quantification purposes. To quantify the number of UiFC aggregates, K48 aggregates and VGluT1mCherry puncta, dendritic and/or axonal markers were used to select branches to quantify and their length determined. Particle analysis was applied to UiFC, K48 ubiquitin and VGluT1mCherry images after appropriately thresholded to quantify number of puncta in each neuronal segment of known length. Due to differences in the expression level of the reporters, threshold to identify puncta was differently adjusted in each image so that distinct puncta would be considered whilst discarding diffuse signal along the axon shaft. In order to determine the density of axonal UiFC and VGluT1mCherry puncta, at sites with or without contact, somas were identified in the brightfield image and their ROIs used to determine whether an axonal domain was in contact with somas (axosomatic) or isolated. Identification of axo-somatodendritic and axo-dendritic domains was carried out by staining for MAP2 and creating ROIs encompassing MAP2-positive structures. The length of ''axo-somatodendritic'' and ''isolated'' axons was determined and used to quantify the density of puncta within each domain. To quantify the number of UiFC-VGluT1mCherry clusters, the presence or absence of UiFC aggregates within VGluT1mCherry puncta ROIs was determined. For quantifying signal intensity on beads, raw intensity values of the signal of interest (UiFC, VGluT1mCherry and stained Bassoon, K48 ubiquitin and tau) within ROIs encompassing beads was quantified from z-projections (sum of all slices) of original z-stacks. Signal at equal-sized offbead ROIs adjacent to each bead was also quantified. When comparing changes in Bassoon, K48 ubiquitin and tau on beads at different time-points and upon ziram treatment, background signal was subtracted and the ratio of signal intensity between on-bead and the correspondent off-bead calculated.

For quantification of changes in time-lapse series, z-stacks were sum projected and aligned with the StackReg plugin. The brightfield image was used to locate somas or beads in contact with axons and ROIs created. Fluorescence intensity within ROIs was measured in each frame of the time-lapse series and normalized to the frame preceding drug treatment or addition of beads (0 min). Kymograph analysis was performed in ICY Software. Kymographs were extracted from axonal segments from a 1 h time-lapse (frames every 1 min). The kymograph tracking tool was used to trace the path of UiFC aggregates on kymographs. For each aggregate, the net run length and mean instant speed were quantified. Puncta with mean instant speed values greater than 0.05 µm/min and net displacements greater than twice their width (limit set to 0.08 µm because UiFC aggregates' average diameter is 0.04 µm) were considered as mobile. Per each axonal segment, the number of mobile and stable puncta was calculated.

All images were processed and prepared for presentation using Photoshop and Illustrator (Adobe).

# Statistical Analysis

Unless otherwise indicated, results are presented as averaged values ± SEM. Graphs and statistical analysis were performed in Graph Pad Prism 5 Software. Statistical differences were examined by non-parametric tests (the test used for each experiment is indicated in the figure legends). Values of p < 0.05 were considered statistically significant (∗p < 0.05; ∗∗p < 0.01 and ∗∗∗p < 0.001).

# RESULTS

# An Ubiquitin-Induced Fluorescence Complementation (UiFC) Approach to Monitor K48 Ubiquitination in Axons

Bimolecular fluorescence complementation has been extensively used in neurons (Feinberg et al., 2008; Unoki et al., 2012; Ramaker et al., 2013; Das et al., 2015; Macpherson et al., 2015) to achieve spatiotemporal resolution in the unmasking of cellular and synaptic aspects. Herein, we pioneered the use of an ubiquitin-based fluorescence complementation approach in neuronal cells to look deep into ubiquitination phenomena in the developing axon. The recently developed UiFC assay allows for live monitoring of K48 polyubiquitination and consists of two constructs, UiFC-C and UiFC-N, each bearing ubiquitin interacting motifs fused to either the N- or C-terminal non-fluorescent fragments of Venus (Chen et al., 2013). Upon polyubiquitination, preferably on lysine 48, interaction of ubiquitin interacting motifs with growing chains reconstitutes Venus fluorescence, whose appearance can be detected within 10 min (Chen et al., 2013).

We first asked if ubiquitination-induced Venus reconstitution was observed in primary hippocampal neurons upon expression of UiFC plasmids. To accomplish so, cells were singly or doubly transfected with UiFC-C and/or UiFC-N. To control for transfection efficacy, co-transfection with mCherry was performed. Transfection of each UiFC plasmid alone did not yield any Venus fluorescent signal despite the presence of mCherry-expressing cells (**Figures 1A,B**). However, when UiFC-C and UiFC-N were co-transfected approximately 10% of cells displayed fluorescence for UiFC (**Figures 1A,B**), thus demonstrating that the signal specifically reveals sites of Venus reconstitution. Importantly, approximately 100% of mCherry-expressing cells exhibit Venus signal when both UiFC constructs were expressed (**Figure 1C**), clearly showing that lack of Venus signal upon single UiFC-C or UiFC-N expression is not due to absence of transfected cells.

We then assessed UiFC ability to faithfully recapitulate changes in cellular ubiquitination levels. To accomplish so, we performed time-lapse imaging in cultured neurons expressing UiFC upon treatment with specific UPS drugs that alter the lifetime of ubiquitinated conjugates (**Figure 1D**). We used the proteasome inhibitor MG132 (1 µM) and the proteasome activator IU1 (75 µM). The latter is a selective inhibitor of the proteasome-associated deubiquitinase Usp14, which limits

degradation by trimming ubiquitin chains from substrates (Lee et al., 2010). Its inhibition by IU1 thus accelerates proteasome degradation of ubiquitin-tagged proteins. Throughout time, MG132 led to increased UiFC intensity, whilst IU1 decreased it in comparison to the DMSO-treated condition (**Figures 1D,E**), which is in accordance with their opposing effects on proteasome activity and concomitant effect on the levels of ubiquitinated proteins. Moreover, we used an inhibitor of deubiquitinases with broad specificity, PR619 (at 1 µM; Altun et al., 2011), which inhibits removal of Ub chains, thus leading to overall enhanced cellular polyubiquitination (Pinto et al., 2016). Indeed, a gradual increase in UiFC signal was observed when neurons were exposed to PR619 (**Figures 1D,E**). Therefore, we demonstrated that UiFC is sensitive to changing levels of endogenous ubiquitination in neurons.

We next assessed the distribution pattern of UiFC in neurons. To accomplish so, UiFC-expressing samples were immunostained for ubiquitin or for K48-linked ubiquitin (Newton et al., 2008) along with the somatodendritic marker MAP2 (**Figures 2A,B**). UiFC showed a diffused staining in somas with occasional bright aggregates (**Figures 2A,B**), which had also been previously observed decorating HeLa cells

(Chen et al., 2013). Close attention to dendrites (**Figure 2C**) and axons (**Figure 2D**) revealed similar pattern of UiFC distribution. Remarkably, both diffuse and aggregated UiFC signal colocalized with staining for K48-specific ubiquitin chains (**Figures 2B–D**), but not with ubiquitin staining (**Figure 2A**). In axons, the great majority of UiFC aggregates (87%) were also positive for aggregates of K48 ubiquitin (**Figure 2E**) and vice-versa (77%), thus suggesting that UiFC aggregates faithfully represent endogenous sites of enrichment of K48 ubiquitinated conjugates along axons. This is in agreement with previous biochemical and imaging analysis showing that UiFC preferentially detects K48 ubiquitin chains over K11 or K63, most likely due to favorable chain conformation allowing for Venus reconstitution (Chen et al., 2013).

Binding of UiFC's ubiquitin interacting motifs to growing ubiquitinated chains on substrates may interfere with the normal stability of the protein or its dynamic ubiquitination. An in vitro ubiquitination reaction with increasing doses of UiFC fragments, showed that UiFC does not interfere with normal polyubiquitination (Chen et al., 2013). To assess whether neuronal expression of UiFC alters levels of its targets in axons, neurons expressing UiFC or a control GFP-expressing plasmid were stained for K48 polyubiquitin (**Figure 2F**). No differences in the total intensity of K48 ubiquitin signal were observed between axons expressing GFP, UiFC or untransfected neighbors (**Figures 2F,G**), thereby demonstrating that under basal conditions UiFC does not affect stability of K48 ubiquitin-tagged conjugates in axons. We then assessed whether UiFC alters the distribution pattern of K48 ubiquitin in axons by quantifying the density and intensity of K48 ubiquitin aggregates. Although no differences were observed in the intensity of K48 ubiquitin aggregates (**Figures 3A,C**), UiFC-expressing axons have a higher number of aggregates in comparison to untransfected neighbors and GFP-expressing ones (**Figures 3A,B**). These results reveal that UiFC alters the pattern of K48 ubiquitination along axons with more K48 ubiquitin aggregates populating the axon. This effect is likely due to enhanced aggregation of polyubiquitinated conjugates through interaction with UiFC's ubiquitin interacting motifs, generating highly stable macromolecular complexes.

We then asked whether such enhanced aggregation of K48 ubiquitin by UiFC has a functional role in the axon or merely represents a functionless artifact. We have previously proposed that localized accumulation of proteasome-related polyubiquitinated conjugates triggers presynaptic assembly (Pinto et al., 2016). Considering that UiFC enhances the number of aggregates of K48 ubiquitin along axons (**Figures 3A,B**), we asked whether this would be positively reflected on the density of presynaptic clusters being formed onto a postsynaptic partner. To investigate this possibility, we coexpressed UiFC or control GFP with the excitatory presynaptic reporter VGluT1mCherry, which consists of a fusion version of the vesicular glutamate transporter 1 (VGluT1) to the fluorescent protein mCherry. Cultures were then stained for MAP2 to locate dendrites. In order to compensate for UiFC and VGluT1mCherry fluorescence loss following the staining

procedure, cultures were also stained for GFP (the antibody used also recognizes Venus) and mCherry to enhance the signal of both reporters. Attention was given to axons expressing both VGluT1mCherry and UiFC or GFP at sites of contact with MAP2<sup>+</sup> structures. Higher density of VGluT1mCherry clusters at axo-somatodendritic regions was observed in axons expressing UiFC (**Figures 3D,E**). The fact that UiFC enhances K48 ubiquitin aggregation on axons and also increases the number of presynaptic terminals argues in favor of the role of K48 ubiquitination in presynaptic assembly proposed in our recent study (Pinto et al., 2016). Moreover, it clearly indicates that UiFC enhancing effect on the aggregation pattern of axonal K48 ubiquitin can exert a biological signaling role.

In this set of results, we demonstrate that UiFC signal can be used in neurons for tracking and monitoring K48 polyubiquitination. We further show that aggregates of K48 ubiquitinated conjugates can be found in axons and reliably detected with UiFC signal. It should be noted, however, that UiFC expression produces an enhancing effect on K48 aggregation along axons.

# Stable Axonal Aggregates of K48 Polyubiquitin

We have recently shown that enhanced presynaptic concentration of proteins in an ubiquitinated state functions as a trigger for presynaptic assembly. In particular, K11 and K48 polyubiquitin chains (Pinto et al., 2016), which normally drive proteins to proteasome clearance (Sadowski et al., 2012). Hence, deep characterization of the distribution and dynamics of accumulated K48 ubiquitination along axons is of immediate relevance. We therefore directed our attention to the UiFC aggregates dispersedly observed in neurons (**Figure 2**). Quantification of the density of UiFC aggregates showed that on average approximately three and six aggregates can be found per 100 µm of dendritic and axonal length, respectively (**Figures 4A,B**). The difference in the density of UiFC aggregates between dendrites and axons (**Figures 4A,B**) likely reflects different subcellular needs and involvement in distinct events.

We then sought to examine the properties of axonal UiFC aggregates. Considering that the axon is densely populated with mobile packets of material actively transported along

FIGURE 4 | Dynamics of UiFC aggregates along axons. (A) Distribution of UiFC aggregates in dendrites and axons. Immunostaining for MAP2 (blue) and tau (red) was performed to identify dendrites (top) and axons (bottom) of neurons expressing UiFC (green). Arrowheads indicate UiFC aggregates. Scale bars represent 5 µm. (B) Quantification of number of UiFC aggregates per 100 µm of dendritic and axonal segment shows a higher density of UiFC aggregates in axons than in dendrites. Results are represented as Mean ± SEM. Statistical significance assessed by Mann Whitney test (∗p < 0.05). A total of 35 dendritic and 25 axonal segments were analyzed from three independent experiments. (C) Mobility of UiFC aggregates along axons. Time-lapse imaging, every 1 min for 1 h, was performed in UiFC-expressing axons to evaluate aggregates' mobility. Top, Representative segment of a UiFC-expressing axon at the beginning of the time lapse. Arrowheads and arrow correspond to stable and mobile UiFC aggregates, respectively. Middle, representative kymograph obtained from a continuous movie at 1 min/frame (1 h total time) showing trajectories of UiFC aggregates. Bottom, schematic representation of UiFC trajectories. Gray lines and dashed green line represent tracks of UiFC aggregates at arrowheads and arrow, respectively. UiFC aggregates along axons are mostly stable. (D) Quantification of the number of (Continued)

### FIGURE 4 | Continued

mobile [mean instant speed greater than 0.05 µm/min and net run length greater than twice aggregates' width (0.08 µm)] and stationary UiFC aggregates per 100 µm of axon length. Results are shown as Mean ± SEM. Statistical significance assessed by Wilcoxon paired t-test (∗∗∗p < 0.001). A total of 93 axonal segments were analyzed from three independent experiments. (E,F) Fraction distribution of individual UiFC aggregates' (E) mean instant speed (µm/min) and (F) net run length (µm). Dashed gray and green lines indicate averaged mean for stable and mobile UiFC aggregates, respectively. Data from 334 individual UiFC aggregates (285 stable and 49 mobile) from three independent experiments.

the axon (Maday et al., 2014), we first questioned about the mobility rate of UiFC aggregates. We were mostly interested in understanding whether UiFC aggregates represented stable platforms of polyubiquitinated conjugates or mobile ubiquitintagged material. To accomplish so, we imaged UiFC-positive axons every 1 min for a total time of 1 h and converted signal along axons into kymographs for analysis of UiFC aggregates' movement (**Figure 4C**). Tracks for each UiFC aggregate were traced (**Figure 4C**, bottom) and their moving behavior analyzed. Most of the aggregates were relatively stable in approximately the same position, showing only minor dislocations along the axon length throughout the imaging period (**Figure 4C**, white arrowheads and correspondent gray lines). In order to have a quantitative idea of the general moving behavior of axonal UiFC aggregates, we divided them into two groups: stable and mobile aggregates. For a UiFC aggregate to be considered as stable two requirements had to be met: display speed values lower than 0.05 µm/min [below the rate of the slower component of slow axonal transport (Maday et al., 2014)]; and dislocate less than the length of twice their width. Most of UiFC aggregates found along the axon were stable (approximately 5/100 µm of axon; **Figures 4C,D**). In contrast, less than one UiFC aggregate per 100 µm displayed considerable movement within a 1 h interval (**Figures 4C,D**, green arrow and correspondent dashed green line). Indeed, for the majority of UiFC aggregates, their mean speed and net displacement were close to zero (**Figures 4E,F**), thus further reinforcing that the majority were stably maintained in the same axonal site.

In few occasions, kymographs revealed new aggregates being formed and others lost from view along the axonal shaft. Interestingly, as an axon grows and extends through the culture, dynamic formation of new UiFC aggregates was observed (**Figure 5**). Whilst some persisted only for a relatively short period of time (**Figure 5**, arrowhead), others steadily remained in the same spot of its first appearance (**Figure 5**, arrow). Therefore, aggregates of K48 polyubiquitin with limited mobility populate axons and are formed as an axon navigates.

# Enrichment of K48 Polyubiquitination at Presynaptic Sites

In order to better understand the biological significance of K48 ubiquitin accumulations, we studied the distribution of UiFC aggregates along axons. We asked whether their

density would be influenced upon synapse formation. In other words, are UiFC aggregates preferentially located at the presynaptic domain of axo-somatic synapses? To answer this question, we fixed cells following UiFC expression and compared the density of UiFC aggregates along axonal segments with or without contact with neuronal somas (**Figures 6A,B**). As it can be observed, along the length of the same axon, UiFC aggregates are found at higher density at axonal domains that represent sites of axosomatic synapses rather than ''bare'' segments (**Figures 6A–C**, green bars).

The fact that UiFC aggregates are preferentially placed at soma-contacting axonal domains raises the question of whether they are at presynaptic sites or in their vicinity. To investigate this possibility, we co-expressed UiFC with the presynaptic reporter VGluT1mCherry. VGluT1mCherry puncta were mainly found at soma-contacting axonal domains (**Figures 6A–C**, red bars), thus demonstrating that somas induced in axons formation of presynaptic clusters. Extra-somatic VGluT1mCherry puncta present in isolated axons (**Figures 6B,D**, red bars) are likely to represent mobile packets of synaptic vesicles that later give rise to terminals (Kraszewski et al., 1995; Ahmari et al., 2000). Colocalization analysis between UiFC aggregates and VGluT1mCherry puncta demonstrated that the majority of axonal UiFC aggregates colocalized perfectly (white arrows) with VGluT1mCherry puncta (**Figures 6J,K**, green bars). Moreover, clusters in which a VGluT1mCherry puncta and a UiFC aggregate colocalize are preferentially found at axonal domains in contact with somas, axo-somatic regions, as opposed to isolated extra-somatic regions (**Figures 6B,C**, yellow bars). These results show that enrichment of K48 polyubiquitin along axons is more likely at soma-triggered sites of presynaptic clustering.

We then asked whether such preferential distribution of K48 ubiquitin enriched aggregates is also observed at axonal sites in contact with dendrites. As previously described (**Figure 3D**), cultures expressing UiFC and VGluT1mCherry were stained for MAP2, GFP and mCherry (**Figure 6D**). In accordance to the concentration of UiFC aggregates at axo-somatic regions (**Figures 6A–C**), higher density of axonal UiFC aggregates is observed at regions juxtaposing MAP2<sup>+</sup> dendrites (axodendritic regions) in comparison to isolated axonal segments (**Figures 6D–G**, green bars). In a way of reinforcing the synaptic nature of these domains, a higher density of VGluT1mCherry puncta was also observed (**Figures 6D–G**, red bars). We then determined the density of clusters containing both UiFC and VGluT1mCherry and again observed their higher concentration on axo-dendritic domains (**Figures 6D–G**, yellow bars). It should be noted that density of axonal K48 ubiquitinated aggregates and presynaptic puncta at MAP2<sup>+</sup> somatodendritic regions (**Figures 6D–G**) is similar to the one obtained at somatic domains detected in brightfield (**Figures 6A–C**). Moreover, no changes were observed between axo-somatodendritic and axo-dendritic domains (**Figures 6D–G**), thus revealing that distribution of UiFC aggregates and VGluT1mCherry puncta have no clear preference for dendrites over somas.

To guarantee that increased detection of UiFC aggregates at somatodendritic domains is due to a real increase in the number of aggregates rather than global enhanced intensity of the reporter, total UiFC signal along the axon was quantified (**Figure 6H**). UiFC intensity is higher at domains juxtaposed to somatodendritic structures in comparison to isolated segments (**Figure 6H**). We then asked whether such difference arises from a higher intensity of UiFC aggregates at these regions. Indeed, axonal aggregates of UiFC are more intense when in contact with somas and dendrites (**Figure 6I**). Therefore, somatodendritic contact elicits in the contacting axon appearance of UiFC aggregates of higher intensity.

In order to clarify the extent to which UiFC aggregates and VGluT1mCherry puncta overlap, we quantified the fraction of VGluT1 puncta at UiFC aggregates, and vice-versa, along an axon at sites of contact and outside (**Figures 6J,K**). Approximately 45% of VGluT1 puncta colocalized with UiFC aggregates regardless of contact with somas and dendrites (**Figures 6J,K**). On the contrary, UiFC aggregates are more prone to overlap with presynaptic puncta at sites of somatodendritic contact (approximately 65% against 44% along an isolated axon; **Figures 6J,K**). These results suggest that the localization of an axonal UiFC aggregate at somatodendritic contact sites increases the likelihood of a presynaptic bouton being formed. Although at a much lower density (**Figures 6C,G**) and fraction of isolated UiFC aggregates (**Figures 6J,K**), UiFC-VGluT1mCherry clusters can be found in an isolated axonal segment. This may be indicative of periods of co-transport or formation of orphan terminals. On the other hand, the fact that a fraction of axo-somatodendritic VGluT1mCherry puncta do not colocalize with UiFC aggregates (**Figures 6J,K**) suggests one of two possibilities: (1) their presence does not coincide with sites of polyubiquitin enrichment or (2) concentration of polyubiquitin occurred at an earlier time-point.

Altogether, our data indicate that the majority of aggregates enriched in K48 polyubiquitin are present at presynaptic sites of axo-somatodendritic synapses.

(green) and the presynaptic reporter VGluT1mCherry (red) was performed in hippocampal neurons. Along axons, UiFC aggregates and VGluT1mCherry puncta were preferentially located at soma contacting axonal regions. Dashed area encircles axon of interest expressing UiFC and VGluT1mCherry. Scale bar represents 50 µm. (B) Enlarged images of boxes in (A) showing axonal segments contacting somas (a, axo-somatic) and without contact (b, isolated axon). Dashed lines outline somatic regions detected in the brightfield. Higher density of both UiFC aggregates and VGluT1mCherry puncta were observed at somas. Arrowheads indicate UiFC-VGluT1mCherry clusters. Scale bars represent 10 µm. (C) Averaged density of UiFC aggregates (green), VGluT1mCherry puncta (red) and

(Continued)

### FIGURE 6 | Continued

UiFC-VGluT1mCherry clusters (yellow) per axonal length at segments contacting somas and isolated ones. Results are normalized to density along the total length of each axon. UiFC-VGluT1mCherry clusters correspond to colocalization events between UiFC aggregates and VGluT1mCherry puncta. Twenty axonal segments (length between 50 and 300 µm) were analyzed from three independent experiments. (D) Distribution of axonal UiFC aggregates and presynaptic clusters at somatodendritic structures and isolated axons. Following dual expression of UiFC (green) and VGluT1mCherry (red), staining for MAP2 (blue), GFP and mCherry was performed. Higher density of axonal UiFC aggregates and VGluT1mCherry puncta at MAP2<sup>+</sup> structures. Dashed area encircles axon of interest. Scale bar represents 20 µm. (E) Straightened images of dashed area in (D). Dashed lines highlight sites of somatodendritic contact. (F) Enlarged images of boxes in (D) showing axonal segments contacting somas (a), dendrites (b) and isolated axons (c). Arrowheads indicate UiFC-VGluT1mCherry clusters. Scale bars represent 5 µm. (G) Averaged density of UiFC aggregates (green), VGluT1mCherry puncta (red) and UiFC-VGluT1mCherry clusters (yellow) per axonal length at axonal segments contacting somatodendritic and dendritic elements and isolated ones. Results are normalized to density along the total length of each axon. (H,I) Intensity of (H) UiFC and (I) UiFC aggregates along the axon at sites with and without contact with MAP2<sup>+</sup> structures. For each axonal segment, results were normalized to intensity values in the total axon. (J) UiFC aggregates co-localize with presynaptic clusters. Colocalization between UiFC (green) and VGluT1mCherry (red) signal along axons showed that the majority of UiFC aggregates overlap with presynaptic puncta. Arrowheads and arrows point to clusters of UiFC-VGluT1mCherry and UiFC aggregates alone, respectively. Scale bar represents 5 µm. (K) Fraction of axonal UiFC aggregates colocalizing with VGluT1mCherry puncta (green bars) and vice-versa (red bars) at sites with and without somato and/or dendritic contact. (C,G–I,K) Results are shown as Mean ± SEM. Statistical significance by Kruskal-Wallis test followed by the Dunn's multiple comparison test (∗p < 0.05, ∗∗p < 0.01 and ∗∗∗p < 0.001 when compared to total axon and ##p < 0.01 and ###p < 0.001 between indicated conditions). (G–I,K) 45 axonal segments (length between 50 and 400 µm) were analyzed from three independent experiments.

# Enhanced K48 Ubiquitination at Sites of Presynaptic Formation

We next asked whether K48 ubiquitination concentrates at the site of nascent presynaptic clusters. Beads with a cationiccoating have been shown to constitute a spatiotemporally controlled way of triggering formation of functional presynaptic sites (Burry, 1980, 1982; Burry et al., 1986; Lucido et al., 2009). Upon their contact with neuronal cells, appearance of presynaptic-like elements occurs as rapidly as after 1–2 h of contact (Burry, 1982; Lucido et al., 2009; Suarez et al., 2013). Beads induce aggregation of presynaptic material in axonal contact sites as well as cytoskeleton rearrangements in the form of enhanced localized actin filaments (Lucido et al., 2009), which have a prominent role in the initial formation of the presynapse (Zhang and Benson, 2001; Lucido et al., 2009; Nelson et al., 2013). Likewise, considering that enhanced UPS-related ubiquitination propels presynaptic differentiation (Pinto et al., 2016), we hypothesized that K48 polyubiquitin signals become enriched at bead-contacting axonal sites. To address this hypothesis, expression of UiFC and VGluT1mCherry in neurons was followed by addition of PDLcoated beads and incubation for 8 h (**Figures 7A,B**). Intense clustering of VGluT1mCherry was observed at bead sites in comparison to an off-bead adjacent domain (**Figures 7A–C**), in accordance to previous studies (Lucido et al., 2009). More importantly, UiFC intensity was increased in these axonalbead contact sites (**Figures 7A–C**), thus indicating a strong accumulation of K48 ubiquitination at sites of induction of presynaptic formation. In order to determine the fraction of responsive beads, we calculated the percentage of beads inducing enhanced accumulation of VGluT1mCherry, UiFC and both in contacting axons (**Figure 7D**). The great majority of beads triggered recruitment of the presynaptic reporter (83%), the K48 polyubiquitination reporter (67%) and both (60%; **Figure 7D**). Importantly, from the pool of beads accumulating UiFC the majority also led to VGluT1mCherry clustering (89%), thus suggesting that localized accumulation of K48 polyubiquitination strongly relates to presynaptic clustering.

In order to look specifically to the response of developing axons not exposed to influences from somatodendritic elements or glia, we used a microfluidic system for the compartmentalization of axons (Taylor et al., 2005; Pinto et al., 2016). In microfluidic devices, two facing compartments are connected by a set of microgrooves, which are sufficiently long and narrow to prevent crossing of somas and dendrites while allowing axons to reach a physically and fluidically isolated chamber (**Figure 8A**; Taylor et al., 2005). This system has been fulfilling the rising demand for technical approaches capable of overcoming the neuronal polarity issue, and in fact it has been used as a tool for the study of axon-intrinsic mechanisms (Hengst et al., 2009; Taylor et al., 2009; Magnifico et al., 2013; Cristovão et al., 2014; Neto et al., 2014; Kim and Jaffrey, 2016; Pinto et al., 2016).

We used time-lapse imaging to study the dynamics of axonal K48 ubiquitin accumulation on nascent presynaptic terminals. Expression of UiFC at the compartment in which neuronal cells have been plated results in UiFC-expressing axons crossing the microgrooves into the axonal side (**Figures 8A,B**). We therefore used this system to monitor axon-autonomous effects on K48 polyubiquitin accumulation on beads following their addition to the axonal chamber and contact with resident axons (**Figures 8A,B**). Bead contact induced in isolated axons a rapid (10 min) local increase in the intensity of the polyubiquitination reporter, UiFC, as opposed to adjacent sites (**Figures 8C,D**). The sudden increase in UiFC signal remains equally high until the end of the time-lapse (90 min) with no further considerable increases (**Figures 8C,D**). Importantly, this effect is not the result of increased axonal volume on bead-contacting domains (Pinto et al., 2016). To examine the relevance of initial accumulation of K48 ubiquitin to presynaptic clustering on beads, we stained axons for the presynaptic active zone marker Bassoon 4–5 h post-imaging (**Figure 8E**). Beads were divided into two groups according to their positive or negative effect on UiFC signal at 10 min post-contact (approximately 70% and 30% of beads, respectively; **Figure 8F**). In comparison to an adjacent offbead site, clustering of Bassoon was enhanced on beads that were capable of rapidly increasing UiFC intensity upon contact (**Figures 8E,F**). These results indicate that accumulation of K48 polyubiquitinated conjugates at the initial stages

of presynaptic formation is related to subsequent formation of the active zone.

Assembly of presynaptic terminals has been proposed to respect a temporal order of material recruitment (Friedman et al., 2000; Suarez et al., 2013). In order to rigorously examine the timing of K48 ubiquitin enrichment in relation to recruitment of presynaptic material to the nascent terminal, axons in microfluidic devices were stained for K48 ubiquitin and Bassoon at different times post-bead contact (**Figures 8G–I**). Bassoon is trafficked in Piccolo-Bassoon transport vesicles (Zhai et al., 2001; Shapira et al., 2003), mobile packets of active zone material which are believe to be first-comers to the presynapse (Friedman et al., 2000; Suarez et al., 2013), thereby making it an excellent temporal marker of presynaptic assembly. Initial clustering of Bassoon on beads occurred surprisingly fast, with detectable increases only after 15 s of contact (ratio between on- and off-bead of approximately 1.4 in 67% of beads), followed by progressive clustering until 1 h (2.8 on/off ratio in 85% of beads; **Figures 8G–I**). Enrichment of K48 ubiquitin on beads was also detected at the shortest time-interval tested (1.9 on/off ratio in 78% of beads; **Figures 8G–I**). However, contrary to Bassoon, with K48 an initial period of intense enrichment was rapidly observed (approximately 2.3 on/off ratio between 30 s to 2 min) followed by a slight decrease at around 5 min (**Figures 8G–I**), which is comparable to the level of UiFC

FIGURE 8 | Rapid increase in UiFC intensity on beads correlates with presynaptic formation. (A) Schematic representation and (B) representative image of a microfluidic platform to monitor changes in axonal polyubiquitination upon bead contact. (A,B) Expression of UiFC at the somal side of microfluidic devices resulted in UiFC<sup>+</sup> axons reaching the axonal compartment, to which beads were subsequently added and their effect on the contacting axon monitored by live-cell imaging. Thicker branches on soma side represent dendrites that do not cross into the axonal compartment. (B) UiFC signal was differently adjusted between right and left images to prevent over-saturation in somas. Enlarged image on right upper corner corresponds to dashed box. Scale bars represent 20 µm and 100 µm. (C) Profile of axonal UiFC intensity upon bead contact at on- and off-bead sites. Individual frames of a time-lapse series showing the initial contact of a bead (dashed circle) with a UiFC-expressing axon (firelut, green). Images were taken every 5 min. Bead contact resulted in a rapid and strong increase in local UiFC intensity, as opposed to an off-bead adjacent site (solid circle). (D) Left, Quantification of UiFC intensity at beads (on-bead, green) and adjacent sites (off-bead, black) throughout time. Statistical significance was assessed by 2-way ANOVA (∗∗∗p < 0.001 and ∗∗p < 0.01 between on- and off-bead at each time-point). Right, Comparison of UiFC intensity values at 0 min and 10 min post-bead contact for both on- and off-bead sites. Statistical analysis by Wilcoxon paired t-test (∗∗∗p < 0.001 when compared to the correspondent 0 min time-point). Results are normalized to 0 min, which corresponds to the frame preceding addition of beads, and shown as Mean ± SEM. A total of 193 beads and equivalent off-bead sites analyzed from two independent experiments. (E) Correlation between enhanced UiFC signal on beads and posterior presynaptic clustering. Retrospective labeling for the active zone marker Bassoon (red) was performed 4–5 h post-UiFC (green) time-lapse on beads. Clustering of presynaptic material was greater on beads that displayed increased UiFC intensity shortly after bead contact (yellow vs. white dashed circle).

(Continued)

### FIGURE 8 | Continued

Solid circle indicates off-bead site. (F) Averaged raw Bassoon intensity values at off- and on-bead sites. Beads were divided into two pools according to their effect on axonal UiFC intensity at 10 min contact: beads not changing UiFC signal (∼30%, white bar) and beads eliciting increases in UiFC signal above off-bead levels (∼70%, yellow bar). Statistical significance by Kruskal-Wallis test followed by the Dunn's multiple comparison test ( ∗∗∗p < 0.001 compared to off-bead). Data from 191 beads (57 and 134 beads without and with UiFC increase at 10 min bead contact) and equivalent off-bead sites from two independent experiments. (G) Time-course of K48 ubiquitin and Bassoon accumulation on beads. Axons in microfluidic devices were exposed to beads for the indicated periods of time and stained for Bassoon (green) and K48 ubiquitin (red). Both K48 ubiquitin and Bassoon rapidly build up on beads. Initial strong accumulation of K48 ubiquitin is followed by Bassoon clustering. The scale bar is 5 µm. (H) Percentage of beads with accumulation of K48 ubiquitin and Bassoon. (I) Ratio of K48 ubiquitin and Bassoon intensity between bead and corresponding adjacent site. (H,I) For individual K48 ubiquitin (red) and Bassoon (green) timelines, statistical significance by Kruskal-Wallis test followed by the Dunn's multiple comparison test (###p < 0.001, ##p < 0.01 and #p < 0.05 compared to 15'). Comparison between K48 ubiquitin and Bassoon at each time-point by 2-way ANOVA (∗p < 0.05). Data from 112 (15"), 76 (30"), 89 (1'), 134 (2'), 154 (5'), 129 (20') and 114 (1 h) beads from 3–4 independent experiments. (J) Correlation between the amount of K48 ubiquitin and Bassoon at beads (2 h contact). Values at bottom right are the ratios of intensities at on- and off-site for each bead shown. The scale bar is 5 µm. (K) Positive correlation between K48 ubiquitin and Bassoon at beads (r, Spearman's rank correlation coefficient; ∗∗∗p < 0.001). A total of 297 beads analyzed from two independent experiments.

change on beads at these later time-points (**Figures 8C,D**). So, K48 ubiquitin and Bassoon are hastily accumulated on beads in a synchronized manner. An early intense accumulation of K48 ubiquitin is followed by sustained Bassoon clustering. We and others (Lucido et al., 2009; Suarez et al., 2013; Pinto et al., 2016) have previously proposed that clustering of synaptic vesicle markers on beads occurs at a later time than the initial recruitment time for Bassoon herein observed (**Figures 8G–I**), which is in agreement with the proposed timeline for presynaptic assembly onto an axo-dendritic contact (Bassoon—active zone formation—accumulation of synaptic vesicles; Friedman et al., 2000). Therefore, local enrichment of K48 ubiquitin occurs coincidently to the initial recruitment of active zone material (**Figures 8G–I**), and precedes clustering of synaptic vesicles. Likewise, enhanced accumulation of a proteasome reporter on beads was observed prior to clustering of a synaptic vesicle marker (Pinto et al., 2016).

Finally, we tested for a correlation between the amount of accumulated K48 ubiquitin on beads and Bassoon clustering. Do beads with higher levels of K48 ubiquitin tend to cluster more presynaptic material and vice-versa? A positive correlation was found between K48 ubiquitin and Bassoon on beads at 2 h contact (**Figures 8J,K**). We can therefore conclude that beads' efficiency to cluster Bassoon is positively correlated to the amount of accumulated K48 polyubiquitin signals. Altogether, this set of results reveals that K48 ubiquitinated conjugates accumulate locally at the initial stages of presynaptic assembly and is intimately associated to efficient clustering of active zone material.

# Presynaptic Clustering Requires Axonal E1 Ubiquitin-Activating Enzyme Activity

To confirm the biological relevance of ubiquitination as a functional means to presynaptic formation, we investigated dependence of presynaptic clustering on E1-mediated ubiquitination. The enzymatic cascade catalyzing ubiquitination of substrates comprises sequential activity of the E1 ubiquitinactivating enzyme, E2 ubiquitin-conjugating enzyme and E3 ubiquitin-ligase (Neutzner and Neutzner, 2012; Heride et al., 2014). In order to prevent ubiquitination we used ziram, an inhibitor of the E1 ubiquitin-activating enzyme. This inhibitor reduces E1 activity by preventing formation of E1-ubiquitin conjugates (Chou et al., 2008; Rinetti and Schweizer, 2010), thus compromising subsequent transfer of ubiquitin moieties to the E2 active site and consequently ubiquitination. The axonal chamber of microfluidic devices was treated with increasing doses of ziram (0, 1, 2 and 5 µM) in the presence of beads. A pre-incubation of 10 min with ziram was performed to guarantee that E1 was inhibited prior initial axon contact with beads. Importantly, because ziram loses efficiency in the reduction of E1-ubiquitin conjugation in longer incubation periods (Rinetti and Schweizer, 2010), we decided to shorten the experiment time to 2 h, after which devices were fixed and stained for the axonal marker tau and Bassoon (**Figure 9A**). In control conditions (vehicle), beads elicited intense clustering of Bassoon in the juxtaposed axonal segment in relation to adjacent regions, with a high ratio between Bassoon intensity at on- and off-bead (**Figures 9A,B**). Increasing doses of ziram progressively decreased the amount of Bassoon clustered on beads, whilst having no effect on tau (**Figures 9A,B**), an axonal microtubule-associated protein that diffusely distributes along healthy axons (Kosik and Finch, 1987; Black et al., 1996). The lack of change on tau distribution between on- and off-bead sites indicates that ziram effect is specific towards clustering of presynaptic components. The fact that no substantial changes were observed in the percentage of beads inducing Bassoon clustering above off-bead levels (**Figure 9C**), shows us that ziram, rather than diminishing the number of beads capable of triggering clustering, affects the amount of recruited presynaptic material per bead. Hence, clustering of presynaptic material on beads requires activity of the E1-ubiquitin activating enzyme at the axon level, thus emphasizing the crucial role of local ubiquitination in the mechanisms governing presynaptic differentiation.

# DISCUSSION

In this study, we evaluated the dynamics of K48 ubiquitination along axons and its correlation to sites of presynaptic formation. We observed that the axon contains aggregates of K48 polyubiquitinated conjugates whose majority is relatively stable (**Figures 2–5**). Secondly, these aggregates are mainly found at sites of clusters of presynaptic material and in the axonal counterpart of axo-somatodendritic synapses (**Figure 6**). We then questioned about enrichment of K48 polyubiquitination at the site of nascent presynaptic terminals from a temporal

perspective and found that it occurs simultaneously to the initial recruitment of Bassoon (**Figures 7**, **8**). An early temporal window of intense K48 ubiquitin enrichment is followed by progressive

active zone clustering (**Figure 8**). Finally, we attributed a fundamental role for axonal ubiquitination, with a particular focus on the local activity of E1 ubiquitin-activating enzyme, on presynaptic assembly (**Figure 9**). The data reported here together with our previous findings (Pinto et al., 2016), suggests that site-specific enrichment of K48 polyubiquitinated conjugates requires axonal ubiquitination and acts as an initial local signal for presynaptic differentiation (**Figure 10**). However, the precise molecular mechanism underlying this phenomenon has yet to be determined.

Throughout this study we made use of the recently developed UiFC approach to monitor K48 ubiquitination along axons. UiFC is capable of sensing changing levels of ubiquitination (**Figure 1**), does not alter stability of K48 ubiquitinated targets and shows high degree of colocalization with endogenous K48 ubiquitin in neurons (**Figure 2**). Our results also show that UiFC mildly enhanced the density of axonal K48 ubiquitin aggregates (**Figure 3**). Nevertheless, UiFC robustly identifies sites of enhanced accumulation of K48 ubiquitin along axons, and so, can be used to monitor dynamics of K48 ubiquitination. Such conclusion is based on the following observations: (1) ∼87% of axonal UiFC aggregates colocalize with endogenous aggregates of K48 ubiquitin, thus showing that UiFC aggregates are bona fide sites of K48 ubiquitin enrichment in axons; (2) K48 aggregates can be found in untransfected axons and not only on UiFC-expressing ones, thus indicating that under basal conditions axonal sites of accumulated K48 ubiquitination exist (**Figure 2D**). UiFC further enhances/stabilizes their presence in axons (**Figures 3A,B**); (3) UiFC expression leads to higher density of presynaptic clusters onto somatodendritic structures (**Figures 3D,E**), which suggests that UiFC-enhanced aggregation of K48 ubiquitin in axons has a functional role in the axon; and (4) distribution of UiFC aggregates in axons is not random but biased towards contact sites with somatodendritic elements and presynaptic clusters (**Figure 6**).

Axonal sites of enriched K48 polyubiquitination are in close association to presynaptic sites. Our analysis of UiFC aggregates along axons of hippocampal neurons shows that approximately 85% of K48 polyubiquitinated aggregates are stationary, barely moving from the same spot, whilst 15% are mobile, although with a low movement speed (**Figure 4**). Of important note is the fact that the average instant speed by which these mobile aggregates move (0.2 µm/min) does not conform to that of fast axonal transport or slow axonal transport of cytosolic proteins (14–280 µm/min and 0.7–7 µm/min, respectively; Maday et al., 2014; Roy, 2014), thus discarding the possibility that K48 polyubiquitinated aggregates are actively transported by microtubule-mediated axonal transport. Instead, it is tempting to speculate that they constitute freely-diffusible aggregates, similar to a previously described diffusion-like motion of the proteasome along axons (Otero et al., 2014). As for the stationary aggregates, their confined location suggests that they represent stable hotspots enriched with K48 polyubiquitinated conjugates in axons. The fact that new aggregates are stably formed as an axon extends through its environment (**Figure 5**); that they are mainly present in axonal domains in contact with somas and dendrites and that the majority colocalize with clusters of presynaptic vesicles (**Figure 6**), gives support to the idea that differentiation of presynaptic sites in a developing axon is assisted by enriched sites of K48 polyubiquitin-tagged conjugates. It is important, however, to stress the possibility that UiFC aggregates correspond to axonal protein aggregates where K48-tagged proteins await to be degraded. Indeed, one may surmise a scenario in which clearance of ubiquitinated aggregates either by proteasome degradation or intense deubiquitination follows their signaling role in the developing axon.

Localized accumulation of K48 polyubiquitinated conjugates and dependence on axonal ubiquitination underlie presynaptic clustering. Reconstitution of UiFC fluorescence and abundance of endogenous K48 ubiquitin were greatly and rapidly potentiated upon axon contact with synapse formation-inducing beads (**Figure 8**). Importantly, this local phenomenon occurred simultaneously to the initial clustering of active zone material (Bassoon) on beads. An early period of intense enrichment of K48 ubiquitin is followed by sustained recruitment of active zone material (**Figure 8**), thus suggesting a sequential ordering of events leading to presynaptic assembly that comprises enhanced on-site ubiquitination as an initial step. The temporal dynamics of accumulation of K48 polyubiquitin conjugates on beads, assessed by UIFC, correlates remarkably with the local decrease in proteasome activity that we previously observed at bead-contacting axons (from 10 min of bead contact on; Pinto et al., 2016). Likewise, the same proportion of beads (approximately 70%) triggers in overlapping axons a decreased rate of degradation of a proteasome reporter (Pinto et al., 2016) as well as higher UiFC levels (**Figure 8**). Together, these data open the likely possibility that accumulation of K48 ubiquitin signals on beads is the direct outcome of a localized halt in UPS degradation. It should be noted, however, that staining for the endogenous pool of K48 ubiquitin at shorter time intervals (in the seconds range) reveals faster enrichment (**Figures 8G–I**), which would not probably be possible to observe with UiFC due to the time required for reconstitution of Venus fluorescence (Chen et al., 2013). In addition, clustering of presynaptic material on beads (**Figure 9**), as well as axonal proteasome inhibition-induced presynaptic assembly (Pinto et al., 2016), requires axonal E1-catalyzed ubiquitin activation. These results are in agreement with previous work in C. elegans (DiAntonio et al., 2001) and mouse lines (Burgess et al., 2004; Chen et al., 2011; Bachiller et al., 2015) that assign paramount importance to protein ubiquitination for presynaptic formation, however not further explored.

Noticeably, the effect of beads on the local levels of K48 ubiquitination and the dependence of presynaptic clustering on ubiquitin conjugation have also been observed when considering actin polymerization (Lucido et al., 2009). Indeed, presynaptic clustering on beads depends on localized reorganization of actin filaments (Lucido et al., 2009). In addition, nascent presynapses along axons are associated with enhanced levels of filamentous actin (Zhang and Benson, 2002) and increased K48 polyubiquitin signal (Pinto et al., 2016). In terms of actin cytoskeleton, it is currently believed that formation of a filamentous actin network recruits synaptic vesicles to discrete sites along the axon by acting as a scaffold for nascent presynapses (Nelson et al., 2013). It is therefore possible that actin and ubiquitin function together to erect a local platform for the recruitment of presynaptic material. Thus, it is essential to unveil how multiple intra-axonal events may interact synergistically and cooperatively to yield presynaptic differentiation efficiently.

Appendage of ubiquitin chains is a versatile way of functionally altering proteins and thereby controlling diverse cellular processes. An interesting finding tell us that only 5% of total ubiquitin in the brain can be found as polyubiquitin chains on substrates (Kaiser et al., 2011). This raises the possibility that cells are equipped to respond rapidly to new intracellular polyubiquitination signals, rather than requiring proteins to remain accumulated in their polyubiquitinated state for long periods. In the same line of thought, together with our recent work (Pinto et al., 2016), we herein propose that quick and site-specific enrichment of K48 polyubiquitinated conjugates occurs in the early steps of presynaptic assembly to assist in clustering of material. This model suggests that K48-linked ubiquitin chains are able to exert a different biological function, other than signaling proteasomal removal. It further suggests that the same type of ubiquitin chain attached to a protein may engage it into diverse roles depending on signal duration, localization and available downstream ubiquitin interpreters and effectors. For instance, β-catenin, an intervenient of the Wnt signaling pathway that among other functions induces presynaptic clustering (Hall et al., 2000), can undergo enhanced stability (Hay-Koren et al., 2011) or be targeted for degradation (Dimitrova et al., 2010) both by attachment of K11 polyubiquitin chains. Nonetheless, the means by which locally enriched K48 polyubiquitin signals trigger presynaptic formation remains unclear. One possibility is that an ubiquitinated enriched platform, composed of either axonal cytoskeleton proteins or presynaptic scaffolding elements, created at axonal domains undergoing synaptogenesis recruits presynaptic material. Indeed, several proteins involved in synaptogenesis and structural elements of the presynaptic terminal can be found in an ubiquitinated state (Franco et al., 2011; Na et al., 2012). Therefore, identification of the ubiquitinated conjugates featuring presynaptogenic properties would be a key issue to disclose, so that the mechanism could be better understood. Ubiquitin proteomics to material clustered on beads at different time-points after initial contact would help to characterize the presynaptic ubiquitome on demand for presynaptic formation.

Neurodevelopmental diseases may arise from abnormal function of the UPS. It is suggested that genetic mutations in enzymes of the ubiquitination cascade or abnormal expression of ubiquitin signaling machinery may underlie neurodevelopmental defects (Hegde and Upadhya, 2011). For instance, mutations in the E1 ubiquitin-activating enzyme or its reduced levels with subsequent disruption of ubiquitin homeostasis contribute to the development of spinal muscular atrophy (Ramser et al., 2008; Wishart et al., 2014). Furthermore, the E3 ubiquitin-ligase UBE3A is implicated in both Angelman syndrome (Kishino et al., 1997; Matsuura et al., 1997) and autism spectrum disorders (Baron et al., 2006; Glessner et al., 2009). In schizophrenia, ubiquitin signaling is hindered due to reduced expression of several involved proteins, with concomitant decreases in monomeric and conjugated ubiquitin (Middleton et al., 2002; Altar et al., 2005; Rubio et al., 2013).

# REFERENCES


This substantial prevalence as a causative factor for the onset of neurodevelopmental diseases emphasizes the need to fully understand and characterize the physiological role of ubiquitin and the proteasome. Particularly, given the data presented here, it is of great importance to pursue the role of localized ubiquitination for presynaptic differentiation.

# AUTHOR CONTRIBUTIONS

Study conception and design by MJP and RDA. MJP: conducted the experiments. RDA and MJP: wrote the article. JRP and ROC: performed experiments for revision.

# FUNDING

This work was supported by the individual grants SFRH/BD/51196/2010 (MJP), SFRH/BD/77789/2011 (JRP), SFRH/BPD/84593/2012 (ROC) by Fundação para a Ciência e Tecnologia (FCT); by FEDER through Programa Operacional Factores de Competividade—COMPETE; by national funds through FCT: PTDC/SAU-NEU/104100/2008, EXPL/NEU-NMC/0541/2012 and UID/NEU/04539/2013; and by Marie Curie Actions—International reintegration Grant, 7th Framework programme, EU.

# ACKNOWLEDGMENTS

We thank Prof. Shengyun Fang for kindly offering the constructs for UiFC, pcDNA3-UiFC-C and pcDNA3-UiFC-N and also Prof. Etienne Herzog for kindly providing F(syn)WRBN- VGluT1mCherry. We thank Prof. Ana Luísa Carvalho, Prof. Carlos Duarte, Dr. João Peça and Prof. Anne Taylor, and all their laboratory members, for helpful discussion and feedback. We also thank Dr. Luísa Cortes for technical microscope support.


**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 Pinto, Pedro, Costa and Almeida. 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.

# An optimal ubiquitin-proteasome pathway in the nervous system: the role of deubiquitinating enzymes

# *Gorica Ristic 1, Wei-Ling Tsou1,2 and Sokol V. Todi 1,2\**

*<sup>1</sup> Department of Pharmacology, Wayne State University School of Medicine, Detroit, MI, USA*

*<sup>2</sup> Department of Neurology, Wayne State University School of Medicine, Detroit, MI, USA*

### *Edited by:*

*Ashok Hegde, Wake Forest School of Medicine, USA*

### *Reviewed by:*

*Jose J. Lucas, CBMSO-UAM, Spain Scott Michael Wilson, University of Alabama at Birmingham, USA*

*\*Correspondence: Sokol V. Todi, Wayne State University School of Medicine, 540 E Canfield, Scott Hall Rm. 3108, Detroit, MI 48201, USA e-mail: stodi@med.wayne.edu*

The Ubiquitin-Proteasome Pathway (UPP), which is critical for normal function in the nervous system and is implicated in various neurological diseases, requires the small modifier protein ubiquitin to accomplish its duty of selectively degrading short-lived, abnormal or misfolded proteins. Over the past decade, a large class of proteases collectively known as deubiquitinating enzymes (DUBs) has increasingly gained attention in all manners related to ubiquitin. By cleaving ubiquitin from another protein, DUBs ensure that the UPP functions properly. DUBs accomplish this task by processing newly translated ubiquitin so that it can be used for conjugation to substrate proteins, by regulating the "where, when, and why" of UPP substrate ubiquitination and subsequent degradation, and by recycling ubiquitin for re-use by the UPP. Because of the reliance of the UPP on DUB activities, it is not surprising that these proteases play important roles in the normal activities of the nervous system and in neurodegenerative diseases. In this review, we summarize recent advances in understanding the functions of DUBs in the nervous system. We focus on their role in the UPP, and make the argument that understanding the UPP from the perspective of DUBs can yield new insight into diseases that result from anomalous intra-cellular processes or inter-cellular networks. Lastly, we discuss the relevance of DUBs as therapeutic options for disorders of the nervous system.

**Keywords: brain, glia, neuron, neurodegeneration, protease, proteasome, synapse, ubiquitin**

# **INTRODUCTION**

The Ubiquitin Proteasome Pathway (UPP) is responsible for degrading the majority of proteins in eukaryotic cells. The UPP is important to all steps and processes of the nervous system, including cell fate specification, differentiation, migration, networking, and maturation, and is critical in maintaining neuronal homeostasis during ageing. As post-mitotic cells, neurons cannot disperse toxic or misfolded proteins through cell division, but must instead continuously rid themselves of cellular components whose accumulation could be detrimental. The importance of the UPP to the ageing nervous system is exemplified by neurodegenerative disorders such as Alzheimer's, Parkinson's and other diseases, where pathological hallmarks are the accumulation and aggregation of proteins with toxic properties.

Degradation of proteins by the UPP is a highly selective process. In order for proteins to be degraded by the proteasome, they require a "tag" to identify them as substrates. This "tagging" task is accomplished through the post-translational modification of substrates with the small modifier protein ubiquitin (Ub). Ubiquitination involves the covalent attachment of Ub via an isopeptide bond to a lysine residue of target proteins through the coordinated action of a Ub activating enzyme (E1; two such enzymes are known in mammals), a Ub conjugating enzyme (E2; ∼50–75 in mammals), and a Ub ligase (E3; *>*500 in mammals) (**Figure 1**).

Ub can be added onto a protein as a monomer or as a poly-Ub chain. Poly-Ub can take on different topographies due to the presence of seven lysine residues on Ub itself, creating chains of different linkage types: K6, K11, K27, K29, K33, K48, and K63. Different types of chains signal different outcomes. In the case of UPP, K48-linked chains consisting of at least four Ub most commonly identify a protein as a proteasome substrate (Thrower et al., 2000; Komander and Rape, 2012; Heride et al., 2014). K11 chains have also been implicated in protein degradation during mitosis (Jin et al., 2008), although it is not clear how widely these chains are utilized by the UPP in other cellular processes.

The proteasome is able to bind both K63- and K48 linked chains in reconstituted systems *in vitro* (Kim et al., 2007; Peth et al., 2010). However, in mammalian cells there appears to be selectivity for K48-linked poly-Ub by the proteasome (Nathan et al., 2013). K63-linked poly-Ub chains are bound by proteins involved in the Endosomal Sorting Complex Required for Transport (ESCRT) pathway, which do not function directly with the proteasome. Binding of K63-linked poly-Ub by ESCRT proteins seemingly precludes the proteasomal degradation of proteins modified with this type of chain (Nathan et al., 2013). Additional specificity in poly-Ub recognition is provided by a family of proteasomeassociated proteins, including the Rad23 orthologs hHR23A and hHR23B that selectively bind K48-linked chains (Nathan et al., 2013).

The 26S proteasome is a macromolecular structure composed of a catalytic 20S subunit and one or two 19S regulatory subunits. Ubiquitinated substrates are recognized by and bind to the 19S particle. This process is accomplished in part by the integral 19S receptors S5a and Adrm1. Ubiquitinated proteins that are bound by the 19S proteasome are deubiquitinated and unfolded. The unfolded proteins can then pass through the hollow, cylindrical core of the 20S particle, where they are enzymatically degraded.

Erroneous ubiquitination of a protein could send it to the proteasome prematurely, or could target it for the wrong pathway (e.g., autophagy rather than the UPP), leading to unintended consequences for cells. Specificity for which proteins are ubiquitinated and the type of Ub linkage formed rests in large part with the E2/E3 pair that performs the ubiquitination process (Komander and Rape, 2012; Heride et al., 2014). An additional level of control is provided by enzymes known as deubiquitinases (DUBs), which reverse the isopeptide bond and thus help to control the status of protein ubiquitination. An increasing number of reports is being published on the role of DUBs in nearly all cellular pathways, tissues and organs, in normal homeostasis and in various diseases, including disorders of the nervous system (Todi and Paulson, 2011; Clague et al., 2013).

The nearly 95 DUBs that are encoded by the human genome are subdivided into five categories based on homology at the catalytic domain. The Ubiquitin C-terminal Hydrolases (UCH), Ubiquitin Specific Proteases (USP), Machado-Joseph Disease Proteases (MJD), and Otubain (OTU) Proteases are cysteine proteases, while JAB1/MPN/Mov34 Metallo-enzyme (JAMM) proteases are zinc-dependent metallo-proteases (**Figure 2**). DUBs maintain the cellular pool of mono-Ub available for conjugation by processing Ub precursors; they replenish mono-Ub by cleaving poly-Ub chains and recycling Ub; they can fully deubiquitinate substrates and reverse their outcome; or they can edit poly-Ub chains on substrates to help direct them toward a specific pathway (**Figure 3**). Although at the most basic level catalytically active DUBs perform a similar function—disassembly of Ub-protein bonds—*in vitro* and *in vivo* studies have collected evidence that these proteases have various non-redundant roles (Clague et al., 2013). Distinct roles stem in part from differences in the structure of the catalytic domains of DUBs and in part from interaction domains and subcellular localization signals encoded in their amino acid sequences.

Three DUBs associate directly with the proteasome: PSMD14, USP14, and UCHL5. PSMD14 is a stoichiometric component of the 19S regulatory subunit, whereas USP14 and UCHL5 associate transiently with it during protein degradation. Several other DUBs function in conjunction with the UPP at steps that precede the proteasome (e.g., during substrate ubiquitination) or following substrate binding to the 19S (e.g., during ubiquitin chain disassembly and ubiquitin recycling; **Figure 4**). The following sections provide details on UPP-related DUBs in the nervous system, with examples from each sub-class of this family. The exception are OTU proteases, for which there are few mechanistic data relating them to the UPP in the nervous system. Information that has been reported so far on OTUs is included in **Table 1**, which is a comprehensive list of DUBs implicated in the nervous system, and also includes proteases that do not necessarily function through the UPP.

# **UPP-RELATED DUBs IN THE NERVOUS SYSTEM PROTEASOME-ASSOCIATED DUBs** *PSMD14*

This metallo-protease belongs to the JAMM sub-family of DUBs. It is also known as POH1. PSMD14 is a constitutive subunit of the proteasome, a member of the 19S regulatory complex, where it neighbors the ubiquitin receptor Adrm1 (Beck et al., 2012). Its reported role is to deubiquitinate proteins at the proteasome before they are degraded. By removing Ub chains *en bloc* from substrates, PSMD14 plays an important role in facilitating protein degradation, as well as in Ub recycling. Once poly-Ub is removed from substrates by PSMD14 it is disassembled by other DUBs, for example USP5 (Clague et al., 2013; Liu and Jacobson, 2013).


PSMD14 and its role at the proteasome were uncovered through studies in yeast that examined how deubiquitination and degradation of a substrate are coupled. In yeast, PSMD14 is known as Rpn11. Through the use of an unfolded, ubiquitinated model substrate, it was shown that deubiquitination occurs after a protein is bound by the proteasome. Mutating Ub so that it can no longer be cleaved by DUBs leads to a significant reduction of substrate degradation. Other studies showed that substrate

### **Table 1 | DUBs in the nervous system.**



### **Table 1 | Continued**

deubiquitination at the proteasome is insensitive to molecules that block cysteine proteases, whereas a zinc chelator blocks deubiquitination, suggesting a metallo-enzyme. Mutations in Rpn11 stabilize UPP substrates and are lethal in yeast. These and other studies pointed to Rpn11 as the DUB functioning at the proteasome, bridging substrate deubiquitination and degradation (Verma et al., 2002; Yao and Cohen, 2002).

Work conducted in mammalian cell culture implicated PSMD14 in maintaining the post-mitotic status of neurons. RNAi-mediated knockdown of PSMD14 in mouse midbrain and cortical cultures leads to *de novo* synthesis of DNA and an increase in apoptosis (Staropoli and Abeliovich, 2005). The mechanism behind these observations is unclear, but these data suggest a critical role for UPP in preventing neurons from re-entering mitosis. Perhaps disturbance of the UPP (from the perspective of PSMD14) triggers a general cellular stress response that ultimately leads to neuronal death. Alternatively, the UPP could be involved in the continuous turnover of specific factors that induce mitotic re-entry. Highlighting the importance of PSMD14 to the nervous system, RNAi-mediated knockdown of its ortholog in neurons in *Drosophila* causes larval lethality (Tsou et al., 2012).

PSMD14 is one of very few DUBs with a specific function, in this case, the removal of poly-Ub chains in a single swoop from substrates at the proteasome. As it will become clear through the remainder of this review, it is difficult to categorize most DUBs as operating in only one step, or acting on only one substrate, pathway, tissue, or organ. Perhaps PSMD14 is an exception because of its continuous presence at the 19S subunit, where it is precluded from interacting with other potential partners. Most other DUBs do not appear to reside in a single constitutive molecular complex and can thus conduct different functions depending on their interactions.

# *UCHL5*

This DUB is a member of the UCH sub-family and is also known as UCH37. Unlike PSMD14, UCHL5 is not a constitutive subunit of the proteasome but associates with it by binding to the Ub receptor Adrm1 on the 19S subunit (Lee et al., 2010a,b). The proposed role for this DUB at the proteasome is one of poly-Ub chain trimming by hydrolyzing Ub chains at their distal end and releasing mono-Ub for re-utilization (Yao et al., 2006). UCHL5 deubiquitinating activity has been shown to increase once it is bound to the proteasome (Qiu et al., 2006; Yao et al., 2006). Based on more recent work, UCHL5 stimulates the activity of the proteasome by regulating ATP hydrolysis and 20S gate opening (Peth et al., 2009, 2013a,b).

Studies in a *Uchl5*-null mouse line showed that this protease is important for brain development. *Uchl5* knockout embryos have malformations in various brain areas, including the telencephalon, mesencephalon, and metencephalon (Al-Shami et al., 2010). The reasons for these specific anomalies are not known. Other proteasome-associated DUBs, such as USP14, which is discussed next and which also processes chains to yield mono-Ub, appear unable to compensate for the loss of UCHL5, indicating a non-redundant role for this DUB in the brain.

## *USP14*

USP14 belongs to the USP sub-family of DUBs. It reversibly associates with the 19S proteasome, where it is responsible for trimming Ub chains on substrates that are destined for degradation and thus recycling mono-Ub. The catalytic activity of USP14 is enhanced by its binding to the proteasome several hundred-fold (Lee et al., 2010a,b), suggesting that this DUB needs the proteasome in order to conduct functions that require its protease activity.

Depletion of USP14 alone enhances proteasome activity, whereas co-depletion with UCHL5 inhibits the proteasome (Koulich et al., 2008), indicating some cross-dependence for these DUBs during protein degradation. Critical information on the function of USP14 has come from yeast studies, where depletion of Ub leads to an upregulation of its ortholog, Ubp6, which restores Ub balance (Hanna et al., 2006). Not all functions of USP14 at the proteasome depend on its catalytic activity: binding of USP14 to poly-Ub chains and unfolded peptides results in opening of the gates of the 20S proteasome and stimulation of the ATPase activity of the proteasome (Peth et al., 2009, 2013a,b). Collectively, these findings implicated USP14 with keeping proteasomal activity under check.

A role for USP14 in the nervous system was first demonstrated by the identification of a recessive mutation in the *Usp14* gene, which causes ataxia in the *ax<sup>J</sup>* mouse line. Homozygous *axJ* mice suffer from progressive motor impairment including ataxia and tremor, reduced body and brain mass, paralysis and death by 2 months of age. The *axJ* mutation results from the insertion of an intracisternal-A particle into intron 5 of *Usp14*, leading to ∼95% lower levels of USP14 protein in the brain of homozygous mice (Wilson et al., 2002; Anderson et al., 2005). In contrast to many mouse models of neurodegenerative diseases, *axJ* mice do not show neuronal loss or abnormal protein accumulation (Wilson et al., 2002). Instead, *ax<sup>J</sup>* mice have reduced mono-Ub levels, particularly in synaptosomal fractions (Anderson et al., 2005; Chen et al., 2009, 2011). The introduction of Ub in neurons in an *axJ* background largely suppresses the phenotype (Chen et al., 2011), indicating that a primary role of USP14 in the central nervous system is Ub homeostasis.

USP14 is critically important at the neuromuscular junction (NMJ). Recordings in *axJ* mice reveal defective release of the neurotransmitter acetylcholine (ACh) at the NMJ, and loss of USP14 leads to developmental anomalies at the pre- and postsynaptic terminals of the NMJ (Chen et al., 2009). The NMJs of *axJ* mice are swollen and poorly arborized, with aberrant nerve terminals and an immature morphology of ACh receptor clusters. These changes correlate with loss of mono-Ub, and are corrected by exogenous expression of Ub in neurons (Chen et al., 2009, 2011).

The ataxia phenotype in the *axJ* mice suggests that the cerebellum is negatively affected by nearly absent USP14. A potential role for this DUB in regulating the activity of cerebellar Purkinje cells has been reported. In *ax<sup>J</sup>* mice, Purkinje cells have increased cell surface expression of GABAA Receptors (GABAAR) in extrasynaptic regions, with a concomitant increase in inhibitory GABAergic currents that effectively reduces cerebellar output (Lappe-Siefke et al., 2009). GABAAR is ubiquitinated in cells and USP14, which interacts directly with this receptor, may deubiquitinate it (Lappe-Siefke et al., 2009). Generally speaking, mono-ubiquitination controls recycling of various receptors to and from the cell membrane, including GABAAR (Saliba et al., 2007). Consequently, USP14 deficiency may cause ataxia in the *axJ* mice in part by perturbing the turnover and/or cell surface distribution of GABAAR. Still, since it is not clear that USP14 directly deubiquitinates GABAAR, it is also worth considering that this DUB may regulate the receptor indirectly. After all, USP14 is a sluggish protease outside of the proteasomal context, leading one to wonder whether it can deubiquitinate GABAAR in the absence of factors that enhance its catalytic activity.

Because of the early lethality phenotype in the *ax<sup>J</sup>* mice, it has been difficult to investigate whether USP14 is important in adults. Recently, another mutation in *Usp14* was identified (Marshall et al., 2013). This mutation, *nmf375*, leads to ∼95% reduction in USP14 protein levels when homozygous, similar to the reduction observed in *ax<sup>J</sup>* mice. However, unlike *axJ* homozygous mice, ones carrying two copies of *nmf375* do not present with phenotypes early in life. Deterioration of motor performance is observed around 12 months of age (*ax<sup>J</sup>* mice die by 2 months) and is associated with mono-Ub depletion (Marshall et al., 2013). These data indicate that USP14 plays an important role not only for NMJ development, but also for its maintenance. They also highlight the importance of genetic modifiers: mutations that lead to similar reduction in USP14 protein levels in different genetic backgrounds have markedly different phenotypic onset and progression. Placing the *nmf375* into the same genetic background as the *axJ* leads to a dramatically earlier phenotype, even more severe than *axJ* (Marshall et al., 2013). Perhaps future genetic analyses will identify factors that modulate the phenotype caused by this UPP-related DUB. Could these modifiers be other DUBs related to the UPP, or are they E3 ligases that counteract USP14 function?

Studies from yeast, mammalian cell culture and mice together indicate that USP14 plays an important role in recycling mono-Ub by functioning at the proteasome. Other studies have presented the possibility that this DUB has specific substrates, potentially outside of the UPP. As described later, there is even evidence from cell culture that USP14 prevents the degradation of some ubiquitinated substrates at the proteasome, adding further complexity to the functions of this protease (see the Section on "The use of DUBs for therapeutic purposes"). Future work may find distinct binding partners of USP14 that depend on cell type, the state of neuronal activity, and on sub-cellular localization. In turn, these partners may dictate the precise function of USP14 as a DUB, or even as a scaffolding protein.

# **DUBs THAT FUNCTION IN CONJUNCTION WITH THE UPP** *UCHL1*

The first reported DUB with a neuronal function, UCHL1 is a member of the UCH sub-family of DUBs. The catalytic area of UCHL1 has a loop positioned over the active site, which limits the size of Ub adducts that can be processed by it to small peptides (Johnston et al., 1999; Das et al., 2006). Based on *in vitro* biochemical reactions, structural data and observations from animal studies, UCHL1 is proposed to function largely by maintaining a stable pool of mono-Ub for use in ubiquitination reactions (Clague et al., 2013). Newly translated Ub contains amino acids following the terminal glycine residue that is used for isopeptide bond formation. UCHL1 can cleave off these additional amino acids in order to expose the final glycine of Ub for conjugation. UCHL1 can also help maintain mono-Ub by reversing accidental modifications that can form during Ub activation (Larsen et al., 1998).

UCHL1 was first described in Aplysia, where its ortholog is known as Ap-UCH (Hegde et al., 1997). This protease was found to have a role in synaptic plasticity because it was one of the genes that was markedly upregulated in sensory neurons following LTF (long-term facilitation) and LTD (long-term depression). Inhibition of Ap-UCH induction or blockage of its function inhibits synaptic plasticity (Chain et al., 1995; Hegde et al., 1997; Fioravante et al., 2008).

UCHL1 is among the most abundant proteins in the brain, by some estimates reaching 1–2% (Jackson and Thompson, 1981; Doran et al., 1983; Wilkinson et al., 1992). Similar to Ap-UCH in Aplysia, mammalian UCHL1 is linked to synaptic function. Studies of mouse knockouts of the *Uchl1* gene indicate that this DUB is important for the structure and function of the NMJ. *Uchl1* knockout mice develop normally, but die prematurely after a period of spasticity and paralysis. At the level of the NMJ, *Uchl1* knockouts show a significant decrease in the release of ACh from the synaptic terminal (Chen et al., 2010), which could be due to perturbed Ub-dependent pathways as a result of decreased Ub recycling. This reduction in content release is accompanied by hindered synaptic plasticity, nerve terminal retraction and axonal degeneration (Chen et al., 2010). Supportive evidence for a role for UCHL1 at the synapse also comes from the *gad* (gracile axonal dystrophy) mouse line, which has an intragenic *Uchl1* deletion (Saigoh et al., 1999). Similar to the *Uchl1* knockout mice, ones homozygous for *gad* present with a dying-back type of axonal degeneration. Lastly, studies in rat hippocampal slices also collected evidence that UCHL1 is important for synaptic plasticity by maintaining mono-Ub. Increased UCHL1 activity leads to higher levels of mono-Ub, whereas pharmacological inhibition of this DUB has the opposite effect and is associated with anomalous synaptic spine structure (Cartier et al., 2009). Importantly, anomalies at the synaptic level during UCHL1 inhibition are rescued by the introduction of mono-Ub (Cartier et al., 2009). Together, these data indicate that a primary role for UCHL1 in the nervous system is to maintain mono-Ub available for utilization during inter-cellular communication. This pool of mono-Ub could be utilized by the UPP as well as other types of cellular pathways and processes, such as gene transcription, receptor internalization, autophagy, etc.

UCHL1 is important to the ageing nervous system, as highlighted by the connection of this DUB to age-related diseases, including Alzheimer's (AD) and Parkinson's (PD). Based on proteomic studies, UCHL1 is a major target of oxidative damage in AD and PD post-mortem human brains (Choi et al., 2004). UCHL1 protein levels are reduced in the hippocampus of a transgenic mouse model of AD that has learning deficits and impaired LTF (Gong et al., 2006). Similarly, soluble UCHL1 levels are decreased in post-mortem AD brains, potentially due to its sequestration in neurofibrillary tangles (Setsuie and Wada, 2007). Introduction of UCHL1 in transgenic AD mice and in cultured cells alleviates cognitive defects and restores synaptic plasticity in a manner dependent on its catalytic activity (Gong et al., 2006). These and other findings support previously mentioned data that UCHL1 is important at synapses, and suggest that increased UCHL1 activity could counteract certain symptoms in AD.

Other work proposes that UCHL1 has specific substrates and may increase UPP-dependent degradation of the β-site amyloid precursor protein (APP) cleaving enzyme 1 (BACE1). BACE1 sequentially cleaves APP into amyloid ß (Aß) protein, which is a major component of neuritic plaques, a hallmark of AD. Pharmacological inhibition of UCHL1 in cultured cells leads to an increase in BACE1 protein levels, whereas an increase in UCHL1 levels has the opposite effect and is associated with reduced levels of Aß protein. Studies in cultured cells are supported by investigations in *gad* mice. Measuring hippocampal levels of BACE1 by western blotting shows an increase in its protein levels in *gad* mice compared to wild type counterparts (Zhang et al., 2012). Whether UCHL1, with its physically constrained catalytic site, can deubiquitinate BACE1 and how this would accelerate its degradation is presently unclear. Another possibility is that reduced mono-Ub levels as a result of UCHL1 perturbation inhibits the UPP more generally, which in turn impacts BACE1 stability.

In relation to PD, a missense mutation in UCHL1 (I93M) was described in 1998 as the cause of dominantly-inherited PD in a family (Leroy et al., 1998). In *in vitro* reconstituted assays, UCHL1I93M was found to have reduced DUB activity (Nishikawa et al., 2003), thus it was initially hypothesized that PD could be due to partial loss of UCHL1 activity. However, mice lacking UCHL1 do not develop neurodegenerative hallmarks of PD, such as loss of dopaminergic neurons, indicating that PD in the I93M family might not be due to reduced UCHL1 activity, but could result from a gain-of-function. To examine this latter possibility, transgenic mice were generated expressing *UCHL1*I93M. These mice show loss of nigral dopaminergic neurons, as is characteristically seen in PD, and develop Ub- and UCHL1-positive inclusions, although not Lewy Bodies, which are the histopathological hallmark of PD (Setsuie and Wada, 2007; Setsuie et al., 2007; Tan and Skipper, 2007). According to these results, UCHL1I93M could cause neurodegeneration through a gain-of-function mechanism. The molecular process through which this mutation would impact UCHL1 cellular function, localization or other properties is unclear.

Another report that linked UCHL1 to PD presented evidence that the DUB activity of a farnesylated, membrane-bound form of this protease rescues the PD protein, α-synuclein, from lysosomal degradation (Liu et al., 2009). α-synuclein is implicated in the buildup of aggregated structures in patient brains, and this buildup is believed to be one cause of death of dopaminergic neurons in PD (Kruger et al., 1998; Athanassiadou et al., 1999; Tan and Skipper, 2007). If it is confirmed *in vivo* that farnesylated UCHL1 has a significant role in the turnover of α-synuclein, it would suggest this DUB as a potential therapeutic target for PD. Severing UCHL1 from the membrane might increase the degradation of α-synuclein, effectively reducing levels of this aggregation-prone protein and alleviating neuronal stress. As it will be discussed later, α-synuclein can be degraded by the proteasome as well as through autophagy. Each type of degradation rests on Ub-dependent processes that are controlled by DUBs such as UCHL1 and USP9X (see below). Consequently, targeting the stability of this disease-linked protein for therapy will require the consideration of various regulatory processes.

Lastly, in 2013 three siblings from a Turkish family were discovered with a recessive, missense mutation in the Ub-binding domain of UCHL1 (mutation E7A), which leads to markedly reduced catalytic activity of this DUB *in vitro*. Patients from this family show an early-onset progressive neurodegenerative syndrome that includes cerebellar ataxia, spasticity, blindness, and nystagmus (Bilguvar et al., 2013). These symptoms are different from those of PD patients, including the ones who carry the I93M mutation. Such phenotypic variability could be due to different effects of the E7A and I93M mutations on the activity and interactions of UCHL1, and might be compounded by other genetic differences between the families.

As we near the end of this section, it is important to note that mutations in two different DUBs, USP14 and UCHL1, which function at least by maintaining mono-Ub, cause problems at the NMJ and neurological defects in mice. Does this mean that they share duties at the molecular, cellular, and tissue level? These DUBs are linked through mono-Ub: *Uchl1* transcription is upregulated in *ax<sup>J</sup>* mice and, vice versa, *Usp14* transcription is increased in *Uchl1*-deficient mice (Walters et al., 2008), suggesting a molecular circuitry that controls the expression of these genes in response to depleted mono-Ub. However, this upregulation at the transcription level does not lead to suppression of the phenotype, or restitution of mono-Ub levels. Neurological symptoms differ among *Uchl1*- and *Usp14*-deficient mice, which could be due to differences in genetic background. As discussed above, genetic background can have a profound effect on symptoms and progression of ataxia in *Usp14*-mutant mice (Marshall et al., 2013). Another reason why mutations in USP14 and UCHL1 lead to different phenotypes could stem from partially nonredundant functions. Perhaps there are specific cellular processes (or substrates), populations of neurons (or glia), or stages of development for which one DUB is more critical than the other. While USP14 and UCHL1 have similar effects on the levels of mono-Ub available in neurons, they likely also play divergent roles in other aspects of neuronal homeostasis that remain to be elucidated.

# *USP7*

A member of the USP sub-family of DUBs, USP7 is also known as HAUSP. It has gained attention because it is viewed as a potential therapeutic target for malignancies. Based on numerous studies, USP7 functions closely with the UPP, where it opposes the proteasomal degradation of various substrates, including the E3 Ub ligase murine double mutant 2 (mdm2) and its substrate, p53, a tumor suppressor that causes cell cycle arrest and apoptosis (Clague et al., 2013). USP7 deubiquitinates both mdm2 (which enhances p53 degradation) and p53 (which inhibits p53 degradation) (Li et al., 2002, 2004). It is the interplay of p53 and USP7 that may be critical for the nervous system. Through a brain-restricted knockout strategy for *Usp7* in mice, it was shown that this DUB is essential: mice lacking USP7 in the brain die soon after birth and have anomalous brain development, attributed in part to p53 protein stabilization as a result of increased mdm2 turnover in the absence of USP7 (Kon et al., 2011). p53-independent mechanisms may also be involved in neonatal lethality, because inactivation of p53 through a knockout strategy fails to fully rescue lethality in the absence of USP7 (Kon et al., 2011).

Another potential role for USP7 in the nervous system is the regulation of the repressor element 1-silencing transcription factor (REST), which inhibits neural cell differentiation. Knockdown of USP7 in neuronal progenitor cells leads to a decrease in the levels of REST protein. USP7 and REST co-immunoprecipitate, and USP7 knockdown results in higher levels of poly-Ub REST, suggesting that USP7 controls REST by deubiquitinating it (Huang et al., 2011). During differentiation, REST is targeted for proteasomal degradation by multiple E3 ligases. The levels of one such ligase, ß-TrCP, increase during neuronal differentiation, when levels of USP7 and REST are lower. When both USP7 and ß-TrCP are knocked down, an intermediate level of ubiquitinated REST is observed compared to its levels in cells where only one protein is targeted (Huang et al., 2011). According to these results, ß-TrCP and USP7 counterbalance each other in regulating the stability of REST. This interaction between USP7 and ß-TrCP is reminiscent of the USP7/mdm2 exchange with respect to p53. It is presently unclear whether p53-independent anomalies in brain development as a result of *Usp7* knockout are due to perturbation in REST signaling. Collectively, these studies suggest at least two major pathways that can be regulated by USP7 in the nervous system: neuronal differentiation and cell viability.

USP7 may also regulate other processes in the brain, including in some neurodegenerative diseases. USP7 interacts with the gene transcription protein ataxin-1 (Hong et al., 2002), mutations in which cause the age-related neurodegenerative disease Spinocerebellar Ataxia Type 1 (Williams and Paulson, 2008). The physiological implications of this interaction are uncertain, but suggest the possibility of USP7 roles in stages following nervous system development. Lastly, recent studies indicated that the above-mentioned transcription factor, REST, is positively involved in normal ageing, and its loss may be implicated in AD and mild cognitive impairment (Lu et al., 2014). Considering the role of USP7 in REST turnover during development, one can extrapolate that this DUB may be neuroprotective in the ageing brain by suppressing REST degradation.

# *USP9X*

Another member of the USP sub-family of DUBs implicated in the nervous system is USP9X. The *Drosophila* ortholog of USP9X, Faf, has been linked to the development of fly eyes and the NMJ (Fischer-Vize et al., 1992; Diantonio et al., 2001). Faf overexpression in *Drosophila* results in NMJ overgrowth, including an increase in synaptic span and number of synaptic boutons. Faf genetically interacts with the E3 ligase Highwire (Hiw), because loss of function in Hiw results in a phenotype similar to Faf overexpression (Diantonio et al., 2001). These findings implicate yet another E3 ligase/DUB pair balancing each other's functions, although the molecular events downstream of Faf and Hiw remain to be uncovered.

In fly eyes, Faf prevents over-neuralization during development. Fruit flies lacking this DUB or expressing a version that is catalytically inactive have supernumerary photoreceptors due to aberrant differentiation of cells that normally acquire nonneural fates (Fischer-Vize et al., 1992; Huang et al., 1995). The fate of these cells is specified through Notch-Delta signaling. Faf genetically interacts with Liquid facets (Lqf), an ortholog of mammalian epsins involved in endocytosis (Cadavid et al., 2000; Chen et al., 2002). Lqf is important for Delta internalization during Notch signaling and thus is a critical component of cell fate specification. In *Drosophila*, decreasing Lqf levels enhances the supernumerary photoreceptor phenotype of Faf mutants, while increasing Lqf levels renders Faf unnecessary (Cadavid et al., 2000; Chen et al., 2002). Ubiquitination of Lqf is stabilized in Faf-less eyes, Faf and Lqf co-immunoprecipitate, and Lqf protein levels are lower in Faf-null mutants (Chen et al., 2002). These findings lead to the conclusion that Faf regulates the ubiquitination status and stability of Lqf: in the absence of Faf, ubiquitinated Lqf could be targeted for proteasomal degradation. It is possible that Lqf ubiquitination also modulates its activity. The ability of human epsins to interact with partners can be regulated by mono-ubiquitination (Nijman et al., 2005; Yi and Ehlers, 2007). Consequently, Faf-dependent deubiquitination of Lqf may control both its function and stability. The human version of Faf, USP9X, also interacts with the Lqf ortholog, epsin-1, and co-localizes with it at the synapse (Chen et al., 2003). Since RNAi against USP9X stabilizes ubiquitinated epsin-1 in cultured cells (Chen et al., 2003), we can infer an evolutionarily conserved role for USP9X in regulating epsin-1 through deubiquitination.

USP9X has been implicated in two neurodegenerative disorders: PD and Diffuse Lewy Body Disease (DLBD). USP9X expression is altered in a mouse model of PD (Zhang et al., 2010), and a portion of USP9X localizes to Lewy Bodies in PD and DLBD (Rott et al., 2011). Supporting the possibility of a role for USP9X in disease, studies have linked this DUB to the stability of α-synuclein, which has been implicated in the etiology of both PD and DLBD. USP9X is reported to regulate the cellular turnover of α-synuclein by deubiquitinating it. RNAi targeting USP9X leads to higher levels of monoubiquitinated α-synuclein in cultured cells and these two proteins co-immunoprecipitate from cultured cells and rat brain lysates. Higher levels of mono-ubiquitinated α-synuclein are not well tolerated by cells (Rott et al., 2011), indicating that USP9X activity can protect against toxicity from this protein. By deubiquitinating α-synuclein, USP9X seems to specify the degradative pathway through which this disease protein is disposed: UPP or autophagy. Unlike other proteins, α-synuclein reportedly only requires mono-ubiquitination to be degraded by the proteasome. However, upon deubiquitination by USP9X, α-synuclein is removed through autophagy (Rott et al., 2011). Perhaps this information on USP9X and α-synuclein can be used for therapeutics for PD and DLBD. Activators of USP9X are expected to reduce levels of mono-ubiquitinated α-synuclein—which is toxic to cells—and to increase its degradation through autophagy, resulting in a neuroprotective effect.

As mentioned above, membrane-bound, farnesylated UCHL1 can also regulate α-synuclein in cell culture by protecting it from lysosomal degradation through a mechanism that remains to be elucidated (Liu et al., 2009). It is tempting to speculate that UCHL1 prevents the lysosomal degradation of α-synuclein by directly deubiquitinating it, but this model would not necessarily fit with the data summarized above on USP9X and the degradation of this disease protein. Potentially, UCHL1 regulates α-synuclein indirectly by controlling other proteins that dictate its turnover, including perhaps USP9X. Ultimately, therapeutic approaches based on the role of USP9X or UCHL1 on α-synuclein need to consider how these DUBs might affect each other's regulatory effect on this aggregation-prone protein.

### *Ataxin-3*

Ataxin-3 is a member of the MJD sub-family of DUBs. It first received attention because it is the disease protein in the neurodegenerative disorder Spinocerebellar Ataxia Type 3 (SCA3), also known as Machado-Joseph Disease. SCA3 is an age-related disease that belongs to the family of triplet repeat disorders, more specifically the polyglutamine repeat-related diseases that include Huntington's, Spinobulbar Muscular Atrophy, Dentatorubral-Pallidoluysian Atrophy and five more SCAs (SCA1, 2, 6, 7, and 17) (Todi et al., 2007). SCA3, which is believed to be the most common dominantly inherited ataxia in the world, is a progressive ataxia accompanied by dystonia, dysarthria, spasticity, rigidity, ophthalmoparesis, dysphagia, and neuropathy. Pathology includes degeneration of cerebellar pathways and nuclei, pontine and dentate nuclei, substantia nigra, globus pallidus, cranial motor nerve nuclei, and anterior horn cells. SCA3 is caused by a CAG repeat expansion in the gene *ATXN3*, which encodes the DUB ataxin-3 (Costa Mdo and Paulson, 2012).

Based on *in vitro* biochemistry, cell-based studies and *in vivo* work in mice, *C. elegans* and *Drosophila*, ataxin-3 seems to function in the UPP (Matos et al., 2011; Costa Mdo and Paulson, 2012). Under some circumstances, ataxin-3 may enhance the degradation of some proteasome substrates (e.g., in Endoplasmic Reticulum (ER)-Associated Degradation, Wang et al., 2006, and in relation with the Ub ligase CHIP, Scaglione et al., 2011), while under other conditions it may decelerate proteasomal degradation of other proteins (Zhong and Pittman, 2006). Since ataxin-3 interacts with several E3 Ub ligases (CHIP, E4B, Hrd1, Parkin) and with the proteasome-associated proteins hHR23A, hHR23B, and VCP/p97 (Costa Mdo and Paulson, 2012), this DUB may regulate the fate of numerous UPP substrates. The precise outcome of ataxin-3 function—increased or decreased stability of proteins—most likely depends on the protein partners with which it interacts.

Through a series of biochemical assays, it was shown that the E3 Ub ligase CHIP and its E2 partner Ubch5C attach long poly-Ub chains onto model substrates. Ataxin-3 cooperates with CHIP/Ubch5C to restrict or edit the length of these poly-Ub species. This collaboration serves to enhance, rather than prevent, proteasomal degradation of CHIP substrates such as iNOS (Scaglione et al., 2011), probably because very long poly-Ub can hinder proteasomal activity (Kim et al., 2007, 2009). Thus, in contrast to what was described above for USP7 and mdm2, which can oppose each other's activities, ataxin-3 and CHIP appear to work together to enhance the turnover of at least some proteins. Such collaborative interactions between ligases and DUBs could be common, because a proteomic study of DUBs identified numerous E3 Ub ligases co-precipitating with these proteases (Sowa et al., 2009).

There is also evidence that ataxin-3 can suppress the degradation of some UPP substrates in cells. Work conducted on ER-Associated Degradation indicates that this DUB deubiquitinates some misfolded proteins synthesized in the ER, thus preventing their proteasomal degradation. Ataxin-3 appears to perform this function in relation with the proteasome-associated protein VCP/p97 (Zhong and Pittman, 2006), although it has not been ruled out that ataxin-3 may also directly oppose the function of ER ligases such as Hrd1 or AMFR.

Three different knockout mouse lines for *atxn3* are viable and appear to live normal lives, indicating that *atxn3* is a non-essential gene (Schmitt et al., 2007; Reina et al., 2010; Switonski et al., 2011). However, this DUB might be required under certain physiological conditions. For example, mouse embryonic fibroblasts that lack ataxin-3 fair poorly during heat shock (Reina et al., 2010, 2012). Also, exogenous ataxin-3 suppresses toxicity from polyglutamine proteins in *Drosophila* (Warrick et al., 2005; Tsou et al., 2013). Whether ataxin-3 has a protective role *in vivo* in mice is less clear. In one study, evidence was presented that wild type ataxin-3 suppresses pathology from its disease-causing version in mouse models of SCA3 (Cemal et al., 2002). But, in a recent publication lack of ataxin-3 did not seem to enhance pathology in a Huntington's Disease mouse model (Zeng et al., 2013). As this latter study did not test whether higher levels of ataxin-3 had a protective effect, or whether catalytically inactive ataxin-3 worsened the HD phenotype, it remains to be clarified whether this DUB has a neuroprotective effect in mammals.

The molecular mechanisms that lead to polyglutaminedependent neurodegeneration in SCA3 are uncertain. Based on *in vitro* studies, K48- and K63-linked poly-Ub binding and cleaving capabilities of ataxin-3 are not affected by expansions in the polyglutamine repeat (Burnett et al., 2003; Winborn et al., 2008; Todi et al., 2009), although its ability to cleave K27- and K29-linked poly-Ub is enhanced in the disease-causing version (Durcan et al., 2011). In cells, polyglutamine-expanded ataxin-3 is less efficient at reducing levels of ubiquitinated species, suggesting that some of its cellular roles are affected by expansion (Winborn et al., 2008). Supporting the notion that polyglutamine expansion alters ataxin-3 function, a disease-causing version of this protease targets the E3 Ub ligase parkin for degradation through autophagy, unlike wild type ataxin-3 (Durcan et al., 2011). Based on studies from other polyglutamine proteins such as ataxin-1 (Lam et al., 2006), polyglutamine repeat expansion may lead to both a partial loss-of- and a toxic gain-of-function for ataxin-3, although this remains to be elucidated. Since ataxin-3 appears to be non-essential in mice, perhaps one effective therapeutic route for SCA3 is to get rid of the protein, potentially without the need to discriminate between normal and pathogenic forms.

## **THE USE OF DUBs FOR THERAPEUTIC PURPOSES**

The UPP has attracted particular attention from a therapeutic point of view because of its critical role in protein quality control and, consequently, in regulating numerous cellular pathways. Different components of the UPP have been targeted for therapy, most commonly the proteasome itself, although not directly for neurological diseases. Therapies focusing on the proteasome have been developed and have shown promise in the clinic. Bortezomib (PS-341/Velcade®), which inhibits the activity of the 20S component, is a treatment option for multiple myeloma and mantle cell lymphoma (Clague et al., 2013). Another inhibitor, carfilzomib (PR-171/Kyprolis®), was generated to treat patients desensitized to bortezomib. Although both of these inhibitors show preference for killing cancer cells, inhibiting the proteasome itself is largely nonspecific and leads to various undesirable side effects (Colland, 2010; Mujtaba and Dou, 2011). Focusing on more specific steps and components upstream of the proteasome could provide desired therapeutic effects without perturbing general proteostasis. This is where DUBs might be particularly useful. Structural work has revealed significant differences in the catalytic areas of DUBs (Clague et al., 2013) that could be utilized to design DUB-specific inhibitors. High-throughput screens have identified small molecule inhibitors for USP7 and UCHL1 (Colland, 2010; Todi and Das, 2012), although the efficacy and overall effect of these inhibitors in intact animals, particularly in the nervous system, is not clear.

One DUB that has been targeted therapeutically for the nervous system is USP14. As discussed earlier, this DUB associates reversibly with the proteasome and is proposed to be important in maintaining a pool of mono-Ub for utilization. USP14 may also function by rescuing certain proteins destined for degradation by deubiquitinating them at the proteasome. Cells that do not express USP14 have enhanced clearance of several diseaserelated proteins, including tau (related to AD), TDP-43 (AD and Amyotrophic Lateral Sclerosis), and ataxin-3 (SCA3, discussed above) (Lee et al., 2010a). This finding suggests that inhibiting the catalytic activity of USP14 may have therapeutic benefits. It is not clear what cellular conditions or substrate particularities dictate the ability of USP14 to rescue some substrates from degradation.

A screen was conducted to identify inhibitors of USP14, and one particularly promising compound was isolated, 1- [1-(4-fluorophenyl)-2,5-dimethylpropyl-3-yl]-2-pyrrolidin-1 ylethanone (abbreviated as IU1), with remarkable specificity against the catalytic activity of USP14 (Lee et al., 2010a). Treatment with IU1 enhances the clearance of disease-linked proteins in cultured cells in a manner that is proteasome dependent (Lee et al., 2010a). A series of *in vitro* and cell-based experiments indicated that trimming of poly-Ub chains by USP14 has an antagonistic effect on protein degradation by the proteasome: proteasomes incubated with USP14 show decreased degradation of a model ubiquitinated substrate compared to inactive USP14, which does not inhibit proteasomal degradation (Lee et al., 2010a). By inhibiting the deubiquitinating activity of USP14, IU1 would presumably prevent the rescue of some ubiquitinated neurotoxic proteins at the proteasome (such as tau and ataxin-3), leading to their degradation. It remains to be determined if IU1 can prevent neurodegeneration *in vivo,* particularly in light of more recent work, which found that overall levels of ataxin-3 and tau were not different in USP14-deficient mice (Jin et al., 2012). These potential discrepancies in findings from cultured cells and intact animals could result from different handling of ataxin-3 and tau *in vitro* vs. *in vivo*. Acute inactivation of USP14 could increase the degradation of some disease proteins by the proteasome without perturbing Ub homeostasis, whereas chronic inactivation could cause an overall disturbance in mono-Ub availability that leads to compensatory mechanisms of protein turnover (such as lysosomal-dependent degradation). Additonally, inactivation or depletion of USP14 may not have

beneficial effects in all neurodegenerative diseases. USP14 can reduce aggregates formed by mutant huntingtin (which causes Huntington's Disease) and can suppress cellular degeneration caused by this protein (Hyrskyluoto et al., 2014). The molecular mechanism through which USP14 has this neuroprotective effect in cells is not known, but it might involve: (1) mono-Ub availability for proper UPP function, (2) the deubiquitination of huntingtin aggregates, which are heavily ubiquitinated, for better recognition, access and degradation by the proteasome, as well as (3) changes in ER-Associated Degradation (Hyrskyluoto et al., 2014). Altogether, these findings stress the importance of a deep understanding of the UPP before targeting specific components for blanket therapeutics in neurodegenerative diseases.

Other studies have hinted at potential molecules that could be used to regulate DUB activities in the nervous system. A recent report that investigated the mechanism through which metal complexes inhibit proteasome function for cancer therapy found that copper pyrithione (CuPT) acts at two distinct steps: (1) by inhibiting the activity of the 20S component, and (2) by inhibiting USP14 and UCHL5 (Liu et al., 2014). CuPT might be a promising cancer therapeutic by inhibiting UPP and causing cellular death in malignant cells, although its utility in neurodegenerative diseases is doubtful. Inhibiting neuronal UPP would be detrimental to the nervous system and would exacerbate pathology. Still, CuPT might be an important experimental reagent to understand the function of USP14 and UCHL5 in the nervous system, especially if this complex can be modified so that its inhibitory roles on the 20S and USP14 and

**FIGURE 4 | Summary of DUBs involved in various steps of the UPP.** Summary of the various steps of substrate ubiquitination and degradation during which specific DUBs involved in the UPP are reported to function. UCHL1 maintains a pool of mono-Ub for conjugation by removing additional amino acids present in newly translated Ub, or by processing accidental thiol or amine modifications formed during Ub reactions. Once a substrate has been ubiquitinated, DUBs such as USP7, USP9X and under some

circumstances, ataxin-3 can deubiquitinate and rescue it from proteasomal degradation. Ataxin-3 has also been reported to enhance the degradation of a few substrates by editing poly-Ub chains for better recognition by and access to the proteasome. Once at the proteasome, premature deubiquitination of substrates by USP14 can prevent degradation. If a proteasome-bound substrate has been committed to degradation and is being unfolded, deubiquitination by PSMD14, UCHL5, and USP14 recycles Ub.

UCHL5 are dissociated. A molecule with such properties in fact exists: b-AP15, which inhibits the DUB activity of USP14 and UCHL5 (D'Arcy et al., 2011). To the best of our knowledge, neither compound has reportedly been utilized to investigate the nervous system.

# **CONCLUSIONS**

Understanding the functions of enzymes important for the removal of Ub from substrates targeted for UPP-dependent degradation provides a deeper appreciation of a basic cellular process essential to all eukaryotic cells. We hope to have made a convincing case that DUBs that function in conjunction with the UPP play critical roles in nervous system development, function, and disease.

The job of the UPP is to identify proteins that need to be degraded and selectively send them to be destroyed by the proteasome through the utilization of an identifier, in this case, ubiquitination. DUBs are among the various checkpoints that ensure the ubiquitination of correct substrates. Some such proteases, like ataxin-3, can function by editing the type of chain attached to a substrate in order to enhance the proteasomal degradation of a specific protein. Others, exemplified by USP7 and USP9X, can oppose the function of specific E3 Ub ligases, leading to the stabilization of some proteasomal substrates. DUBs such as UCHL1, UCHL5, and PSMD14 seem critical for mono-Ub homeostasis; others, like USP14, control the UPP through multiple, seemingly contradictory, activities: by recycling Ub for reuse, by increasing proteasome function, and sometimes by rescuing specific substrates from degradation (**Figure 4**).

As there are nearly 95 genes encoding DUBs in humans, one would think that there would be redundancy built into the functions of this family of proteases. However, studies in various animal models indicate that redundancy is not common among DUBs (Clague et al., 2013). Mutations or knockouts of different DUBs lead to distinct developmental anomalies or neurological disorders, supporting the idea that while some DUBs share duties, they also have roles in pathways, cells, tissues, organs, and stages of development that cannot be fully compensated by other members of their family.

While we have learned a tremendous amount about the UPP and the role of DUBs in it, much remains to be uncovered, particularly in relation to the nervous system. Each time a door is opened by new research, a spiraling corridor is revealed with more possibilities to consider and opportunities to explore. Details on UPP-related DUBs in specific neuronal circuits are limited. We lack an expression and localization atlas of DUBs during development and in adults, during increased periods of activity and in resting times, in different areas of the brain and spinal cord, in different types of neuronal and glial cells and in distinct sub-cellular compartments. Such information is critical if we are to understand why the perturbation of DUBs that are supposed to have somewhat related functions has different outcomes in intact organisms. Rewards from the continued study of DUBs and UPP in the nervous system are expected to be great, both in terms of comprehending basic mechanisms of neuronal physiology, and in understanding and treating diseases of the nervous system.

# **AUTHOR CONTRIBUTIONS**

Wrote Manuscript: Gorica Ristic, Wei-Ling Tsou, Sokol V. Todi. Prepared Figures: Gorica Ristic, Wei-Ling Tsou. Prepared Table: Gorica Ristic, Sokol V. Todi.

# **NOTE**

While this manuscript was under review and in production, further work was published on DUBs important to the nervous system, including reports on UCHL1 and its importance in the enteric nervous system (Coulombe et al., 2014) and in tau phosphorylation and AD pathogenesis (Zhao et al., 2014). We regret the inability to discuss this work as well as other reports in this review.

# **FUNDING**

This work was supported by R01 NS086778 from NINDS to Sokol V. Todi.

# **REFERENCES**


human ubiquitin carboxyl-terminal hydrolase L1 variants. *Biochem. Biophys. Res. Commun.* 304, 176–173. doi: 10.1016/S0006-291X(03)00555-2


**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.

*Received: 30 April 2014; accepted: 10 July 2014; published online: 19 August 2014. Citation: Ristic G, Tsou W-L and Todi SV (2014) An optimal ubiquitin-proteasome pathway in the nervous system: the role of deubiquitinating enzymes. Front. Mol. Neurosci. 7:72. doi: 10.3389/fnmol.2014.00072*

*This article was submitted to the journal Frontiers in Molecular Neuroscience. Copyright © 2014 Ristic, Tsou and Todi. 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.*

# Distinct effects of ubiquitin overexpression on NMJ structure and motor performance in mice expressing catalytically inactive USP14

*Jada H. Vaden1, Jennifer A. Watson1, Alan D. Howard1, Ping-Chung Chen2, Julie A. Wilson1 and Scott M. Wilson1\**

*<sup>1</sup> Evelyn F. McKnight Brain Institute, Department of Neurobiology and Civitan International Research Center, University of Alabama at Birmingham, Birmingham, AL, USA, <sup>2</sup> Department of Structural Biology, St. Jude Children's Research Hospital, Memphis, TN, USA*

### *Edited by:*

*Ashok Hegde, Wake Forest School of Medicine, USA*

### *Reviewed by:*

*Izhak Michaelevski, Tel Aviv University, Israel Sokol V. Todi, Wayne State University School of Medicine, USA*

### *\*Correspondence:*

*Scott M. Wilson, Evelyn F. McKnight Brain Institute, Department of Neurobiology and Civitan International Research Center, University of Alabama at Birmingham, 1825 University Boulevard, Shelby 914, Birmingham, AL 35294, USA livvy01@uab.edu*

> *Received: 17 February 2015 Accepted: 06 April 2015 Published: 23 April 2015*

### *Citation:*

*Vaden JH, Watson JA, Howard AD, Chen P-C, Wilson JA and Wilson SM (2015) Distinct effects of ubiquitin overexpression on NMJ structure and motor performance in mice expressing catalytically inactive USP14. Front. Mol. Neurosci. 8:11. doi: 10.3389/fnmol.2015.00011* Ubiquitin-specific protease 14 (USP14) is a major deubiquitinating enzyme and a key determinant of neuromuscular junction (NMJ) structure and function. We have previously reported dramatic ubiquitin depletion in the nervous systems of the USP14-deficient *ataxia* (*axJ)* mice and demonstrated that transgenic ubiquitin overexpression partially rescues the *ax<sup>J</sup>* neuromuscular phenotype. However, later work has shown that ubiquitin overexpression does not correct the *ax<sup>J</sup>* deficits in hippocampal short term plasticity, and that transgenic expression of a catalytically inactive form of USP14 in the nervous system mimics the neuromuscular phenotype observed in the *ax<sup>J</sup>* mice, but causes a only a modest reduction of free ubiquitin. Instead, increased ubiquitin conjugates and aberrant activation of pJNK are observed in the nervous systems of the USP14 catalytic mutant mice. In this report, we demonstrate that restoring free ubiquitin levels in the USP14 catalytic mutant mice improved NMJ structure and reduced pJNK accumulation in motor neuron terminals, but had a negative impact on measures of NMJ function, such as motor performance and muscle development. Transgenic expression of ubiquitin had a dose-dependent effect on NMJ function in wild type mice: moderate levels of overexpression improved NMJ function while more robust ubiquitin overexpression reduced muscle development and motor coordination. Combined, these results suggest that maintenance of free ubiquitin levels by USP14 contributes to NMJ structure, but that USP14 regulates NMJ function through a separate pathway.

Keywords: neuromuscular junction, ubiquitin, USP14, pJNK, proteasomes, motor neuron and deubiquitinating enzyme

**Abbreviations:** Tg*CA* (transgene expressing catalytically inactive USP14, generated by mutating the active site cysteine to an alanine residue, expressed under the neuronal *Thy1.2* promoter; also refers to mice expressing this transgene); Tg*Ub* (transgene expressing ubiquitin under the *Thy1.2* promoter; also refers to mice expressing this transgene); Tg*Ub-H* (a transgenic founder line with more robust ubiquitin overexpression than Tg*Ub*); Tg*CA*,Tg*Ub* (mice expressing both Tg*CA* and Tg*Ub*).

# Introduction

Ubiquitination can regulate a diverse array of cellular pathways depending on the number of ubiquitin monomers conjugated onto a protein and the internal lysine residue used to form the ubiquitin chain. These pathways regulate protein stability (Hershko and Ciechanover, 1998), activity (Schmukle and Walczak, 2012; Humphrey et al., 2013; Zhou et al., 2013), and localization (Tanno and Komada, 2013). The accumulation of ubiquitin-positive aggregates is a hallmark of neurodegenerative diseases (Perry et al., 1987; Lowe et al., 1988; Jara et al., 2013), suggesting that sequestration of ubiquitin, and the consequent reduction of ubiquitin availability, could contribute to neuronal dysfunction. This suggestion is supported by the hypothalamic dysfunction and neurodegeneration observed in mice lacking the ubiquitin gene *Ubb* (Ryu et al., 2008).

Although ubiquitin is encoded by four genes in the mammalian genome (Lund et al., 1985; Wiborg et al., 1985; Baker and Board, 1987, 1991; Finley et al., 1989; Redman and Rechsteiner, 1989), the equilibrium between ubiquitin conjugated onto proteins and free ubiquitin is largely controlled by the opposing actions of ubiquitin ligases and deubiquitinating enzymes (DUBs; Hallengren et al., 2013). We have previously shown that loss of the proteasome-associated DUB ubiquitin-specific protease 14 (USP14) in the *ataxia* (*axJ* ) mice results in perinatal lethality, reduced muscle development, and structural and functional defects at the neuromuscular junction (NMJ). Loss of USP14 also severely alters ubiquitin homeostasis by causing a dramatic depletion of free ubiquitin in the brain and spinal cord (Anderson et al., 2005; Chen et al., 2009). Ubiquitin levels were most severely affected at the synapse, suggesting that the NMJ deficits in the *axJ* mice are due to ubiquitin depletion (Chen et al., 2009). In fact, neuronal-specific transgenic expression of ubiquitin corrects the reduced muscle development, altered NMJ structure, reduced synaptic transmission, and perinatal lethality observed in the *axJ* mice (Chen et al., 2011). Although these findings suggest a role for USP14's in the maintenance of free ubiquitin pools in the nervous system, later work showed ubiquitin complementation does not rescue the deficits in hippocampal short-term plasticity observed in the *axJ* mice (Walters et al., 2014).

Furthermore, we recently reported that transgenic expression of a dominant-negative, catalytically inactive USP14 species in the murine nervous system recreates many of the essential phenotypes observed in the *axJ* mice, including deficits in NMJ structure, poor motor performance, and reduced muscle development (Vaden et al., 2015). In contrast to the *axJ* mice, however, the USP14 catalytic mutant mice have a normal lifespan and only a modest decrease in free ubiquitin. Instead, the spinal cords of the USP14 catalytic mutant mice show an increase in levels of ubiquitin conjugates linked through lysine 63 (K63), which are known to promote kinase activation (Yang et al., 2010; Humphrey et al., 2013). Consistent with this finding, the USP14 catalytic mutant mice have increased activation of the ubiquitin-dependent mixed lineage kinase 3 (MLK3) pathway, which signals through c-Jun N-terminal kinase (JNK; Vaden et al., 2015). As expected from the well-documented role of pJNK in rearranging synaptic architecture in invertebrates (Etter et al., 2005; Collins et al., 2006; Drerup and Nechiporuk, 2013), *in vivo* inhibition of JNK significantly improved NMJ structure and muscle development in the USP14 catalytic mutant mice (Vaden et al., 2015).

These findings led us to hypothesize that, in addition to maintaining ubiquitin homeostasis, USP14's catalytic activity is required for the termination of ubiquitin signaling cascades and, consequently, that the increase in ubiquitin conjugates observed in the spinal cords of the USP14 catalytic mutant mice can alter neuronal function independent of the reduction of free ubiquitin. To directly test this hypothesis, we generated mice expressing both catalytically inactive USP14 (Tg*CA*) and ubiquitin (Tg*Ub*) under the neuronal *Thy1.2* promoter. The spinal cords of the resulting Tg*CA*, Tg*Ub* mice had control levels of free ubiquitin and a nearly twofold increase in ubiquitin conjugates. The motor performance and muscle development of the double transgenic Tg*CA*,Tg*Ub* mice were reduced compared to both Tg*CA* mice and controls, underlining the need for USP14-dependent deubiquitination events in the nervous system. However, despite the detrimental effects of ubiquitin complementation on the overall phenotype of the Tg*CA* mice, the structure of the NMJ was improved, and the amount of pJNKpositive pathology was decreased, in the Tg*CA*,Tg*Ub* mice compared to Tg*CA* mice. Additionally, we found that overexpression of ubiquitin had a bidirectional, dose-dependent effect on muscle mass, motor function, and measures of synaptic transmission even in wild type mice where USP14's activity was intact. These data suggest that ubiquitin complementation in *axJ* mice indirectly corrects the functional deficits caused by loss of USP14 and demonstrate that increased protein ubiquitination alters motor function.

# Materials and Methods

# Animals

Wild type C57BL/6J, *Thy1-YFP* mice (16JRS, Jackson Laboratories, Bar Harbor, ME, USA), transgenic mice expressing USP14 (Tg*Usp14)* catalytically inactive USP14 (Tg*CA;* previously referred to as Tg*Usp14CA*), ubiquitin (Tg*Ub*), or both catalytically inactive USP14 and ubiquitin (Tg*CA*,Tg*Ub*) have been maintained in our breeding colony at the University of Alabama at Birmingham, which is fully accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International. We have previously described the generation of the Tg*Usp14* (Crimmins et al., 2006), and Tg*Ub* (Chen et al., 2011), and Tg*CA* mice (Vaden et al., 2015), and all transgenes are expressed from the neuronal *Thy1.2* promoter. Generation of Tg*Ub* mice resulted in two different founder lines with differing ubiquitin expression, Tg*Ub-H* (high expresser) and Tg*Ub* (low expresser). Double transgenic (Tg*CA*,Tg*Ub*) mice were generated by breeding male Tg*CA* mice to female Tg*Ub* mice. All mouse lines were maintained on a C57BL/6J background and transgenic mice were heterozygous for the transgene(s) of interest. Research was conducted without bias toward the sex of animals used for each study, and equal numbers of male and female mice were used. All research complied with the United States Animal Welfare Act and other federal statutes and regulations relating to animals and experiments involving animals, and adhered to principles stated in the Guide for the Care and Use of Laboratory Animals, United States National Research Council. In addition, all experiments were carried out with the approval of the University of Alabama at Birmingham's Institutional Animal Care and Use Committee.

## Isolation of Proteins

Mice were deeply anesthetized with isoflurane prior to rapid decapitation. Spinal cords were removed and homogenized in modified RIPA buffer containing 50 mM Tris, pH 7.5, 150 mM NaCl, 5 mM MgCl2, 0.5 mM EGTA, 1 mM EDTA, 0.5% SDS, 1% Triton X-100, and 1% sodium deoxycholate. Complete protease inhibitor (Roche, Indianapolis, IN, USA), phosphatase inhibitor cocktail III (Sigma Aldrich, St. Louis, MO, USA), and 50 μM PR-619 (inhibitor of DUBs, Life Sensors, Malvern, PA, USA) were added to the homogenization buffer per manufacturer instructions. Following homogenization, samples were sonicated and centrifuged at 17,000 × *g* for 10 min at 4◦C. The supernatants were removed and stored immediately at −80◦C. Protein concentrations were determined using the bicinchoninic acid (BCA) protein assay kit from Pierce (Rockford, IL, USA).

### Immunoblotting and Quantitation

Proteins were resolved on 10% Tris-glycine gels and transferred onto nitrocellulose membranes. BSA (2%) in PBS containing 0.1% NP-40 was used to block the membranes and dilute the primary and secondary antibodies. HRP-conjugated secondary antibodies (Southern Biotechnology Associates, Birmingham, AL, USA) and SuperSignal West Pico Chemiluminescent Substrate (Thermo Scientific, Rockford, IL, USA) were used for detection. For detection of ubiquitin, proteins were resolved on 4–12% NuPage Tris-bis gels (Life Technologies, Grand Island, NY, USA) and transferred onto PVDF membranes. Membranes were treated with 0.1% glutaraldehyde in PBS for 20 min prior to blocking in PBS containing 2% BSA and 0.1% NP-40. The secondary antibody was diluted into PBS containing 1% nonfat dry milk and 0.1% NP-40. For quantitation, all blots were scanned using a Hewlett-Packard Scanjet 3970, and band density was quantified with UN-SCAN-IT gel digitizing software (Silk Scientific, Inc., Orem, UT, USA).

### Antibodies

The following antibodies were used: USP14 (Anderson et al., 2005), β-tubulin (Developmental Studies Hybridoma Bank, Iowa City, IA, USA); Ubiquitin (UAB Hybridoma Facility, Birmingham, AL, USA); pJNK, and JNK (Cell Signaling Technology, Danvers, MA, USA).

### Open Field

Animals were handled 1 day prior to open field testing. Locomotor activity was measured in an open field chamber (43.2 cm × 43.2 cm × 30.5 cm) for 15 min by an automated video tracking system (Med Associated, St. Albans, VT, USA).

The first 5 min were not analyzed to account for habituation to the chamber.

## Rotarod

Motor coordination was tested by placing mice on a rotating rod (ENV-575, Med Associates), which accelerated from 3.5 to 35 rpm over a 5-min period. Latency to fall was recorded over three trials, each separated by 1 h, and the individual trials for each animal were averaged.

### Quantitative PCR

Total RNA was isolated from gastrocnemius muscles or spinal cords using RNA-STAT60 (Tel-Test, Friendswood, TX, USA) and reverse transcribed using the Superscript VILO cDNA synthesis kit (Life Technologies) per manufacturer instructions. Individual gene assays were purchased from Applied Biosystems for each of the RNAs analyzed: *AChR*-α (Mm00431629\_m1), *AChR-*ε (Mm00437411\_m1), and *AChR-*γ (Mm00437419\_m1). --Ct values were generated using *Gapdh* (Mm99999915\_g1) as an internal standard. All values are reported as mean ± SEM of at least three different animals per genotype, run in triplicate.

### NMJ immunostaining and confocal imaging

Whole mount immunostaining of the tibialis anterior (TA) muscle was performed as described (Vaden et al., 2015). Briefly, the TA muscle was immersed in ice-cold PBS containing 2% PFA for 1 h following dissection and immediately teased into thin bundles. Muscle bundles were then transferred to PBS containing 1% PFA and 1% Triton (PBS-T) and incubated overnight at 4◦C with constant rocking. To improve visualization of axons and ultra-terminal sprouting, all mice used for NMJ immunostaining carried the *Thy1-YFP* transgene in addition to the transgene(s) of interest. Endplates were labeled by a 1 h room temperature incubation with rhodamine-conjugated α-bungarotoxin (α-BTX) and prepared for antibody application by a 1 h incubation in a blocking buffer containing 2% bovine serum albumin and 4% normal goat serum in PBS-T. For pJNK immunostaining, muscle bundles were incubated with primary antibody (pJNK, ##81E11, Cell Signaling Technology) for 5 days at 4◦C with constant rocking. All images were captured using a Zeiss LSM 510 Meta confocal microscope (Carl Zeiss, Oberkochen, Germany). Endplate size was determined by tracing the circumference of the α-BTX-positive post-synaptic AChR cluster and computing area using ImageJ software (NIH, Bethesda, MD, USA).

### Statistical Analysis

All analyses were carried out using GraphPad Prism version for 6.0 for Mac OS X (GraphPad Software, La Jolla, CA, USA). For scaled variables, significant differences among genotypes were determined with a one-way ANOVA after assumptions for normality were verified with a D'Agostino and Pearson omnibus normality test with alpha set to 0.05. *Post hoc* comparisons were made with independent samples *t*-tests with Bonferronicorrected alpha. In samples in which one or more genotype did not meet the assumption of normality, significant differences were determined with a Kruskal–Wallis test and *post hoc* comparisons were made with Mann–Whitney tests with Bonferronicorrected alpha.

# Results

# Loss of USP14's DUB Activity Increases Ubiquitin Conjugates

To assess the interaction of USP14 with cellular ubiquitin pools, we performed immunoblot analysis on spinal cord extracts from wild type, Tg*CA*, and *axJ* mice (**Figure 1A**). Tg*Usp14* mice, which overexpress wild type USP14 in the nervous system (Crimmins et al., 2006), were included as a control. As previously reported (Anderson et al., 2005; Chen et al., 2009), loss of USP14 in the *axJ* mice resulted in significant decreases in free ubiquitin levels, whereas neuronal expression of catalytically inactive USP14 in the Tg*CA* mice led to only a modest reduction in free ubiquitin levels (**Figures 1A,C**). Instead, loss of USP14's catalytic activity led to the retention of more proteins in a ubiquitinated state, as indicated by the increase in ubiquitin conjugates observed in the spinal cords of Tg*CA* mice compared to controls (**Figures 1A,B**). We also observed both an increase in free ubiquitin and a reduction in ubiquitin conjugates in the spinal cords of the mice overexpressing wild type USP14 (Tg*Usp14*), suggesting that USP14's DUB activity makes a significant contribution to protein deubiquitination.

# Neuronal-Specific Over-Expression of Ubiquitin is Detrimental to Tg*CA* Mice

To test the hypothesis that increased ubiquitin conjugates, and not depletion of free ubiquitin, cause the motor neuron dysfunction in Tg*CA* mice, we restored free ubiquitin levels in the Tg*CA* mice by generating double transgenic mice expressing both Tg*Ub* and Tg*CA* under the neuronal *Thy1.2* promoter. The resulting Tg*CA*,Tg*Ub* mice expressed wild type levels of free ubiquitin as well as an increase in ubiquitin conjugates above what was observed in the Tg*CA* mice (**Figures 2A–C**). Consistent with our previous report (Vaden et al., 2015), there was no difference in body mass between 4-weeks-old Tg*CA* mice and controls (**Figure 2D**), but there was a significant reduction in the mass of the gastrocnemius muscles in the Tg*CA* mice compared to controls (**Figure 2E**). While transgenic expression of ubiquitin increased body and muscle mass in the *axJ* mice (Chen et al., 2011), both parameters were significantly reduced in the Tg*CA*,Tg*Ub* mice compared to both control and Tg*CA* mice (**Figures 2D,E**). Furthermore, whereas rotarod performance in *axJ* Tg*Ub* mice is improved over *axJ* mice (Chen et al., 2011), the Tg*CA*,Tg*Ub* mice did not perform better than Tg*CA* mice in this assay (**Figure 2F**). In fact, when we measured motor function in the less demanding open field assay, the Tg*CA*,Tg*Ub* mice showed reduced ambulatory distance (**Figure 2G**) and velocity (**Figure 2H**) compared to both wild type and Tg*CA* mice. In contrast, ubiquitin overexpression alone, in Tg*Ub* mice, caused increased body and gastrocnemius mass (**Figures 2D,E**) and ambulatory distance in the open field assay (**Figure 2G**) compared to controls.

We have previously shown that deficits in motor function and NMJ synaptic transmission are inversely related to muscle *AChR* transcript levels, and, in particular, that expression of the fetal *AChR-*γ subunit is dramatically increased in adult animals with motor and synaptic deficits (Chen et al., 2009, 2011; Vaden et al., 2015). As predicted by their poor open field performance, *AChR-*α, -ε, and *-*γ transcripts were significantly increased in Tg*CA*,Tg*Ub* mice over all other genotypes, including Tg*CA* (**Figures 3A–C**). Finally, there was a significant decrease in *AChR-*γ transcripts in Tg*Ub* mice compared to controls (**Figure 3C**), and a trend

muscle mass, or motor deficits in Tg*CA* mice. (A) Representative immunoblots of ubiquitin and USP14 from spinal cords of 4-weeks-old wild type; Tg*CA*, Tg*CA*, Tg*Ub*, and Tg*Ub* mice. β-tubulin was used as a loading control. (B) Quantitation of the levels of ubiquitin conjugates, normalized to wild type levels; [*F*(3,28) = 17.36, *p* < 0.0001, one-way ANOVA]. (C) Quantitation of the levels of free ubiquitin, normalized to wild type levels; [*F*(3,12) = 11.70, *p* < 0.001, one-way ANOVA]. (D) Body mass [*F*(3,56) = 9.03, *p* < 0.0001, one-way ANOVA] and (E) gastrocnemius muscle mass [*F*(3,70) = 94.72, *p* < 0.0001, one-way ANOVA] of

4-weeks-old mice. *n* = at least 12 animals per genotype. (F) Latency to fall from beam during a rotarod assay [*F*(3,26) = 38.36; *p* < 0.0001, one-way ANOVA]. (G) Total ambulatory distance [*F*(3,29) = 9.26; *p* < 0.001, one-way ANOVA] and (H) velocity [*F*(3,26) = 6.56; *p* < 0.01, one-way ANOVA] during 10 min open field assay. For (F–H), *n* = at least five animals per genotype. All data are shown as mean ± SEM. Symbols represent unpaired *t*-tests compared against wild type and corrected for multiple comparisons with a Bonferroni adjustment, where ∗*p* < 0.05, ∗∗*p* < 0.01, ∗∗∗*p* < 0.001. In (B,C), an additional unpaired *t*-test with Bonferroni adjustment was used to compare Tg*CA* and Tg*CA*,Tg*Ub* mice.

toward decreased *AChR-*<sup>α</sup> and *-*<sup>ε</sup> transcripts (**Figures 3A,B**). Together with our previous report, these data demonstrate that increased ubiquitin expression in the presence of catalytically inactive USP14 (in the Tg*CA*,Tg*Ub* mice) has drastically different effects than when USP14 is absent (*axJ* Tg*Ub* mice).

# Ubiquitin has a Dose-Dependent Effect on Motor Function in Wild Type Mice

Loss of USP14's catalytic activity leads to a greater increase in ubiquitin conjugates in Tg*CA*,Tg*Ub* spinal cords than ubiquitin overexpression alone causes in the spinal cords of Tg*Ub* mice (**Figure 2**). To determine whether increased protein ubiquitination in the nervous system could alter neuromuscular function independently of reduced USP14 activity, we compared muscle mass, *AChR* expression, and rotarod performance in the Tg*Ub* mice used in previous experiments with another transgenic founder line (Tg*Ub-*H) that has higher levels of ubiquitin expression (**Figures 4A–C**). The gastrocnemius muscles of 4- to 6-weeks-old female Tg*Ub* mice were significantly larger than those of controls (**Figure 4D**). In contrast, the increased ubiquitin expression in the Tg*Ub-H* mice resulted in significantly decreased gastrocnemius mass compared to controls (**Figure 4D**). The same was true for male wild type, Tg*Ub*, and Tg*Ub-*H mice (data not shown). Further, we found that Tg*Ub-*H mice had a significant increase in the abundance of fetal *AChR-*γ transcripts (**Figure 4E**), a marker of synaptic deficits and motor dysfunction in adult *axJ* animals (Chen et al., 2009, 2011). Finally, whereas overexpression of ubiquitin in the Tg*Ub* mice had no effect on rotarod performance, increased levels of ubiquitin in the Tg*Ub-*H mice caused them to fall from the rotating beam more quickly than controls (**Figure 4F**), and display an abnormal gait while performing this task (data not shown). Together, these data show that ubiquitin has dose-dependent effects on motor function and muscle development.

# Ubiquitin Overexpression Corrects Presynaptic Structural Deficits at the Tg*CA* NMJ

We have previously reported that the NMJs of 4- to 6 weeks-old USP14-deficient *axJ* mice have poor motor endplate arborization, swollen presynaptic terminals, and ultraterminal sprouting, and that these deficits are corrected in *axJ* Tg*Ub* mice (Chen et al., 2011). Because Tg*CA* mice recapitulate these *axJ* NMJ deficits (Vaden et al., 2015), we investigated whether NMJ structure was improved in Tg*CA*,Tg*Ub* mice (**Figure 5A**). Whereas 56% of Tg*CA* NMJs had presynaptic swellings, ubiquitin overexpression improved NMJ structure and only 11% of Tg*CA*,Tg*Ub* terminals were swollen (**Figure 5B**). Similarly, we observed ultra-terminal or ultraaxonal sprouting in 50% of the terminals of the Tg*CA* mice compared to only 17% in the Tg*CA*,Tg*Ub* mice (**Figure 5C**). Terminal swelling and sprouting were seen only rarely in wild type and Tg*Ub* mice (**Figures 5B,C**). We found a significant effect of genotype on endplate area, with smaller, more plaque-like endplates in the Tg*CA* and Tg*CA*,Tg*Ub* mice than wild type and Tg*Ub* mice (**Figure 5D**), consistent with the increase in *AChR* mRNA abundance observed in these mice (**Figure 3**).

# Ubiquitin Complementation Reduces pJNK at the NMJs, but Not in the Spinal Cords, of the Tg*CA* Mice

We recently reported that pJNK is a prominent component of the terminal swellings and sproutings in Tg*CA* mice, and that the extent of this pathology is reduced by treatment with the JNK inhibitor SP600125 (Vaden et al., 2015). Because ubiquitin complementation improved the structure of the Tg*CA* endplates (**Figure 5**), we investigated whether

this reduction in pathology was associated with a reduction of pJNK at the NMJ. As previously reported, we observed strong pJNK immunostaining in the presynaptic terminals of Tg*CA* mice (**Figure 6A**): 55.8% of terminals contained pJNKpositive swellings (**Figure 6B**) and 43.5% of terminals contained pJNK-positive ultra-terminal sprouts (**Figure 6C**). In contrast, ubiquitin overexpression in the Tg*CA*,Tg*Ub* mice reduced the pathology, and only 8.3% of terminals contained pJNK-positive swelling or sprouting (**Figures 6B,C**). pJNK-positive pathology was negligible in wild type mice and absent in Tg*Ub* mice. However, when we immunoblotted spinal cord lysates from wild type, Tg*CA*, Tg*CA*,Tg*Ub*, and Tg*Ub* mice with a pJNK-specific antibody, we found that the levels of pJNK in Tg*CA*,Tg*Ub* mice were elevated over the levels observed in Tg*CA* mice, and that spinal cord lysates of both genotypes had significantly more pJNK than spinal cord lysates from wild type mice (**Figures 7A,B**). There was no change in total JNK abundance in the Tg*CA* or Tg*CA*,Tg*Ub* mice compared to controls, indicating that USP14 and ubiquitin regulate JNK activation and not JNK stability. We also observed a significant decrease in the abundance of pJNK, but not total JNK, in Tg*Ub* spinal cord lysates compared to controls (**Figures 7A,B**). A schematic summarizing the levels of pJNK abundance observed in the spinal cords, distal motor neuron axons, and NMJs of wild type, Tg*CA*, and Tg*CA*,Tg*Ub* mice is shown in **Figure 7C**.

# Discussion

Three main conclusions can be drawn from these studies. First, overexpression of ubiquitin increased the abundance of ubiquitin conjugates in the spinal cord (**Figure 4**), demonstrating that free ubiquitin is a limiting factor in some ubiquitindependent processes in the nervous system. Second, USP14's ubiquitin hydrolase activity plays a significant role in regulating the distribution of free and conjugated ubiquitin pools in neurons (**Figure 1**). The increased abundance of ubiquitin conjugates, and not depletion of the free ubiquitin pool, directly correlated with impaired motor function when USP14's ubiquitin hydrolase activity was inhibited (**Figures 2** and **3**). Third, both ubiquitin levels and the catalytic activity of USP14 impact the abundance of pJNK at the NMJ and in the spinal cord (**Figures 6** and **7**) and, in turn, the levels of pJNK at the NMJ impact the structure of the presynaptic terminal.

Neuronal overexpression of ubiquitin led to increased ubiquitin conjugates in a dose-dependent manner in two transgenic founder lines, Tg*Ub* and Tg*Ub-H*, with low and high levels of ubiquitin overexpression, respectively (**Figure 4A**). These ubiquitin conjugates had dose-dependent effects on motor function and muscle development in wild type mice (**Figure 4**). Mild ubiquitin overexpression led to increased muscle development and reduced expression of the fetal γ subunit of the muscle *AChR*, which is correlated with enhanced NMJ synaptic transmission (Chen et al., 2009, 2011; Vaden et al., 2015), in Tg*Ub* mice compared to controls (**Figures 4B–D**). In contrast, more dramatic ubiquitin overexpression reduced muscle development, increased *AChR-*γ abundance, and hindered motor performance in the Tg*Ub-*H mice compared to controls

(**Figures 4B–D**). Moreover, we have previously reported that the Tg*Ub*-H mice develop adult-onset motor endplate disease (Hallengren et al., 2013). Together, these data highlight the importance of maintaining ubiquitin levels within a relatively narrow window.

Ubiquitin-specific protease 14 is an important regulator of ubiquitin homeostasis. We found that increasing the level of wild type USP14 in the nervous system resulted in an increase in the free ubiquitin pool and decreased levels of ubiquitin conjugates as compared to controls. Because both the *axJ*

mice and Tg*CA* mice exhibit severe motor endplate disease, we were surprised to find differences in ubiquitin homeostasis between these two mouse lines. Loss of USP14 in the *axJ* mice significantly reduced the abundance of both free and conjugated ubiquitin. In contrast, expression of catalytically inactive USP14CA in the Tg*CA* mice led to a 40% increase in ubiquitin conjugates and a 25% reduction of free ubiquitin. Further studies will be required to determine if these effects of USP14 on ubiquitin pools are due to changes in ubiquitin-dependent degradation by the proteasome. These differences in ubiquitin homeostasis in the spinal cords of the *axJ* and Tg*CA* mice, along with the dose-dependent impact of increased ubiquitin conjugates on motor function and muscle development, may provide a framework for understanding the opposite effects of ubiquitin overexpression in the *axJ*

and Tg*CA* mice. One possibility for these differences is that USP14CA may have an increased affinity for its substrates compared to USP14. Binding of USP14CA may slow their eventual deubiquitination by another DUB and result in prolonged ubiquitin-dependent signaling. Alternatively, increased affinity of USP14CA to ubiquitin conjugates may block their proteasomal degradation and lead to protein aggregation that could affect synaptic function.

In the *axJ* , Tg*Ub* mice, the levels of ubiquitin conjugates and free ubiquitin are slightly elevated over what is normally observed in wild type mice (Chen et al., 2011). We have now shown that a modest increase in ubiquitin conjugates and free ubiquitin in the Tg*Ub* mice correlated with increased muscle development and motor function, even when wild type USP14 was present (**Figure 4**). In contrast, the levels of ubiquitin conjugates in the spinal cords of the Tg*CA*, Tg*Ub* mice were nearly twofold what is observed in wild type mice (**Figures 2A,B**), and reduced muscle development and mobility were observed in the Tg*CA*,Tg*Ub* mice compared to both wild type and Tg*CA* mice. Our studies of the Tg*Ub-*H mice indicated that a twofold increase in ubiquitin conjugates resulted in deficits in motor function and muscle development even when USP14's catalytic activity was intact (**Figure 4**). Together, these data open the possibility that the restoration of motor neuron function in *axJ* Tg*Ub* mice is indirect and can be attributed to increased ubiquitin conjugates, while suggesting that ubiquitin overexpression aggravates the accumulation of ubiquitin conjugates caused by loss of USP14's DUB activity in Tg*CA* mice to exaggerate the existing phenotype (**Figures 2** and **3**). Since both increases and decreases in ubiquitin pools are associated with deleterious effects in cells and animals models (Parag et al., 1987; Ferguson et al., 1990; Osaka et al., 2003; Anderson et al., 2005; Hirsch et al., 2006) the ubiquitin depletion observed in the absence of functional USP14 may contribute to other neurological deficits observed in the Tg*CA* and *axJ* mice.

However, when considered together with our previous work demonstrating a role for pJNK in the NMJ pathology observed in the Tg*CA* mice (Vaden et al., 2015), this study suggests that ubiquitin complementation corrects these structural deficits directly, by reducing pJNK-positive pathology (**Figures 5** and **6**). While we have not examined pJNK abundance at the NMJs of *axJ* mice, the increased JNK activation in *axJ* spinal cords suggests that the mechanism underlying NMJ pathology caused by loss of USP14 is the same as that caused by loss of USP14's DUB activity (Vaden et al., 2015). Given the reduction of pJNK at Tg*CA,*Tg*Ub* NMJs compared to Tg*CA* NMJs (**Figure 6**), we were surprised by the dramatic increase in pJNK levels in the spinal cords of the Tg*CA*,Tg*Ub* mice (**Figures 7A,B**). However, we have recently reported that loss of USP14's DUB activity leads to enhanced K63-linked ubiquitination of MLK3 (Vaden et al., 2015), which drives it to dimerize, autophosphorylate, and activate its immediate downstream targets, MKK4/7, which, in turn, activate JNK (Humphrey et al., 2013). It is therefore possible that the increased JNK activation that we observed in Tg*CA*,Tg*Ub* spinal cords compared to Tg*CA* spinal cords (**Figures 7A,B**) results from increased activation of MLK3.

The means by which ubiquitin complementation reduced pJNK-positive pathology at the NMJ remains unclear. One possibility is that the increased ubiquitin in Tg*CA*,Tg*Ub* mice stimulates the degradation of pJNK or its upstream kinases at the

# References


NMJ or, alternatively, that ubiquitin contributes to the activation of phosphatases that act on pJNK. Both of these explanations are consistent with the reduction of pJNK abundance in Tg*Ub* spinal cords compared to wild type, but not with the increase in pJNK levels over both wild type and Tg*CA* levels in the spinal cords of the Tg*CA*,Tg*Ub* mice (**Figures 7A,B**). In contrast, the well-documented retrograde axonal transport of activated JNK (Cavalli et al., 2005; Lindwall and Kanje, 2005; Shin et al., 2012; Drerup and Nechiporuk, 2013) and the pJNK-positive swellings at Tg*CA* NMJs (**Figure 6**) are all consistent with a deficit in the retrograde transport of pJNK out of axon terminals in the Tg*CA* mice. The lack of pJNK-positive pathology in the nerve terminals (**Figure 6**) of the Tg*CA*,Tg*Ub* mice, combined with the robust elevation of pJNK abundance in Tg*CA*,Tg*Ub* spinal cords (**Figures 7A,B**), the site of motor neuron cell bodies, suggest that ubiquitin may stimulate the retrograde axonal transport of pJNK (**Figure 7C**). Although a role for ubiquitin in retrograde axonal transport has not been demonstrated directly, it was recently reported that altered ubiquitin homeostasis contributes to the deficits in TrkB retrograde transport caused by the exposure of cultured neurons to Aβ oligomers (Poon et al., 2013).

Finally, our findings also suggest that the functional deficits observed in the Tg*CA* and *axJ* mice do not arise because of altered NMJ structure. This is consistent with our previous report that acute inhibition of USP14 at the NMJs of wild type mice causes deficits in synaptic transmission that are similar to what is observed in the Tg*CA* and *axJ* mice, while intramuscular injections of the same inhibitor given over the course of a week do not cause NMJ pathology (Vaden et al., 2015). Together, these findings may indicate that USP14 regulates synapse structure and function through distinct pathways. The same is true of the ubiquitin ligase highwire, which regulates the structure and function of the *Drosophila* NMJ through separate pathways (Collins et al., 2006).

# Acknowledgments

This work was supported by grants to UAB from the Howard Hughes Medical Institute through the Med into Grad Initiative (JV), the Civitan Research Center, the Evelyn F. McKnight Brain Institute, and the National Institutes of Health (R01 NS047533 and R21 NS074456 to SW).


**Conflict of Interest Statement:** Scott Wilson is a paid consultant for Progenera Inc. The collection, analysis, and interpretation of the data presented in this manuscript were not influenced by my relationship with Progenra Inc.

*Copyright © 2015 Vaden, Watson, Howard, Chen, Wilson and Wilson. 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.*

# Ubiquitin-dependent trafficking and turnover of ionotropic glutamate receptors

## Marisa S. Goo, Samantha L. Scudder and Gentry N. Patrick \*

*Section of Neurobiology, Division of Biological Sciences, University of California, San Diego, La Jolla, CA, USA*

Changes in synaptic strength underlie the basis of learning and memory and are controlled, in part, by the insertion or removal of AMPA-type glutamate receptors at the postsynaptic membrane of excitatory synapses. Once internalized, these receptors may be recycled back to the plasma membrane by subunit-specific interactions with other proteins or by post-translational modifications such as phosphorylation. Alternatively, these receptors may be targeted for destruction by multiple degradation pathways in the cell. Ubiquitination, another post-translational modification, has recently emerged as a key signal that regulates the recycling and trafficking of glutamate receptors. In this review, we will discuss recent findings on the role of ubiquitination in the trafficking and turnover of ionotropic glutamate receptors and plasticity of excitatory synapses.

### Edited by:

*Ashok Hegde, Georgia College and State University, USA*

### Reviewed by:

*Izhak Michaelevski, Tel Aviv University, Israel Nien-Pei Tsai, University of Illinois at Urbana-Champaign, USA*

### \*Correspondence:

*Gentry N. Patrick, Section of Neurobiology, Division of Biological Sciences, University of California, San Diego, 9500 Gilman Drive, La Jolla, CA 92093-0347, USA gpatrick@ucsd.edu*

> Received: *30 July 2015* Accepted: *22 September 2015* Published: *16 October 2015*

### Citation:

*Goo MS, Scudder SL and Patrick GN (2015) Ubiquitin-dependent trafficking and turnover of ionotropic glutamate receptors. Front. Mol. Neurosci. 8:60. doi: 10.3389/fnmol.2015.00060* Keywords: glutamate receptor, ubiquitin, E3 ligase, deubiquitinating enzyme (DUB), proteasome, lysosome, postsynaptic density, synaptic plasticity

# Introduction

Glutamatergic synapses mediate the majority of excitatory synaptic transmission in the mammalian central nervous system (CNS). Arguably, AMPA receptor (AMPAR) trafficking to and from the postsynaptic membrane plays a significant role in many forms of synaptic plasticity (Shepherd and Huganir, 2007). AMPARs are tetrameric receptors comprised of four different subunits (GluA1- A4) and these subunits can combine in different stoichiometries to form ion channels with distinct functional properties (Hollmann and Heinemann, 1994; Rosenmund et al., 1998). A large body of evidence suggests that AMPARs are not statically localized at the synapse, but rather are dynamically trafficked in and out of the postsynaptic membrane under specific signaling cues.

Phosphorylation is one well-studied post-translational modification that regulates AMPAR trafficking. Protein kinases can phosphorylate AMPARs, which signals them to move to and from the synapse, potentially leading to either long-term potentiation (LTP) or long-term depression (LTD) (Lu and Roche, 2012). Recently, ubiquitination, a distinct post-translational modification, has emerged as an important regulator of AMPAR trafficking and function. Ubiquitin, a 76 amino acid protein, is covalently linked to lysine residues on a protein substrate via an isopeptide bond (Pickart, 2004). The addition of the ubiquitin moiety occurs through a series of enzymatic reactions involving an activating enzyme (E1), a conjugating enzyme (E2), and a ligase (E3) (Mabb and Ehlers, 2010; Berndsen and Wolberger, 2014). Alternatively, removal of the ubiquitin moiety is facilitated by deubiquitinating enzymes (DUBs). Depending on the chain length and topology, the ubiquitin moiety can then send the target protein to various fates in the cell including proteasomal or lysosomal degradation (Pickart and Eddins, 2004). Degradation via the proteasome typically involves a ubiquitin chain length of four or more. On the other hand, ubiquitination in the form of single (mono) or short-chain ubiquitin modifications can result in the endocytosis of integral membrane proteins (Clague and Urbé, 2010). Ubiquitinated proteins are then sorted by the endosome sorting complexes required for transport (ESCRTs) into multivesicular bodies (MVBs) and eventually the lysosome (Hicke and Dunn, 2003; Piper and Luzio, 2007). Conversely, if a DUB acts on the protein in the early endosome, the protein can be recycled back to the plasma membrane.

Since the first investigations of glutamate receptor ubiquitination in the nematode Caenorhabditis elegans (C. elegans) (Burbea et al., 2002) and later in mammals (Schwarz et al., 2010; Fu et al., 2011; Lin et al., 2011; Lussier et al., 2011), more recent studies have further defined the role of ubiquitination on glutamate receptor trafficking and function (See **Figure 1**). In this review we will highlight recent findings on the ubiquitin-dependent trafficking and turnover of glutamate receptors in neurons and the distinct regulatory signals involved.

# Signals that Induce AMPAR Ubiquitination

Surface AMPARs are internalized in a constitutive manner, but their trafficking can also be controlled through synaptic activity (Huganir and Nicoll, 2013). Activation of glutamate receptors with the agonists AMPA and NMDA can both induce receptor internalization through independent pathways that result in different receptor fates (Shepherd and Huganir, 2007). Studies in recent years have shown that certain types of stimulation can induce AMPAR internalization through a pathway that

FIGURE 1 | Schematic of known E3 ligases and DUBs that target ionotropic glutamate receptors. AMPARs are ubiquitinated by ligases Nedd4-1, RNF167 and APCCdh1. Short-term treatment with bicuculline (on the order of min) leads to AMPAR ubiquitination by RNF167, while long-term treatment with bicuculline (on the order of hours to days) leads to AMPAR ubiquitination by either Nedd4-1 or APCCdh1. Once internalized, ubiquitinated AMPARs can either be deubiquitinated by DUBs (USP8 or USP46), which can promote their recycling, or they can be targeted to the lysosome or the 26S proteasome for degradation. Ubiquitination of NMDARs occurs by the E3 ligase Mind bomb-2 (Mib2) and is, in part, regulated by receptor phosphorylation. In heterologous cells involving transfection strategies, Nedd4-1 has been shown to target NMDARs for ubiquitination, though data in neurons has not been verified. Furthermore, retrotranslocated NMDARs can be ubiquitinated by F-box Protein 2 E3 ligase, Fbx2, which recognizes high-mannose glycans found on the extracellular region of GluN subunits. Finally, KARs have been found to be ubiquitinated by Cul3 (through interactions with actinfilin) or parkin, which has been implicated in Parkinson's disease. Alternatively, covalent modification of KARs by the SUMO conjugating enzyme PIAS3, has been shown to regulate kainate-receptor mediated synaptic transmission.

involves receptor ubiquitination. In this section, we discuss what is currently known about the synaptic cues that induce AMPAR ubiquitination.

The first study of mammalian AMPAR ubiquitination determined that direct activation of receptors with the agonist AMPA causes robust ubiquitination of the GluA1 subunit (Schwarz et al., 2010), a finding which has been confirmed in recent studies (Scudder et al., 2014; Widagdo et al., 2015). Interestingly, another group found that AMPA promotes ubiquitination of the GluA2 subunit rather than GluA1 (Lussier et al., 2011). However, a recent report indicates that AMPA induces the ubiquitination of all four AMPAR subunits (Widagdo et al., 2015). Regardless of this discrepancy, all available data support the conclusion that direct activation of AMPARs with agonists promotes their ubiquitination and internalization. This ligand-induced effect requires calcium entry, provided mainly through voltage-gated calcium channels while NMDA receptor (NMDAR) signaling appears unnecessary (Schwarz et al., 2010; Lussier et al., 2011; Widagdo et al., 2015).

In addition to bath application of receptor agonist, many groups have used alternate methods to examine activity-induced AMPAR ubiquitination. The GABA<sup>A</sup> receptor antagonist bicuculline is commonly used to globally raise activity in cultured neurons, and when applied to neurons for a prolonged amount of time can induce a negative feedback process termed synaptic scaling (Turrigiano et al., 1998; Siddoway et al., 2014). Bicuculline treatments have been shown to promote AMPAR ubiquitination after short-term and long-term treatments (Lussier et al., 2011; Scudder et al., 2014; Widagdo et al., 2015). Furthermore, long-term treatment promotes the recruitment of the E3 ligase Nedd4-1 to synapses (Scudder et al., 2014). However, unlike the AMPA-induced scenario, this form of receptor modification appears to require NMDAR signaling (Lussier et al., 2011; Widagdo et al., 2015), which suggests there may be slight differences in the pathways that involve ubiquitin conjugation. Application of an agonist activates both synaptic and extrasynaptic receptors while bicuculline should only activate synaptic AMPARs, and this difference may activate different cellular pathways and perhaps even lead to different receptor fates. Alternatively, these two scenarios may simply differ in the source of calcium; NMDARs could provide the calcium influx during bicuculline treatments while AMPA treatments instead rely on calcium influx through voltage-gated calcium channels and calcium-permeable AMPARs.

The specific ligase responsible for bicuculline-induced ubiquitination is debated; short treatments induce AMPAR ubiquitination that requires the ligase RNF167 (Lussier et al., 2012) while longer treatments (>20 h) recruit Nedd4-1 to synapses and increase the overall protein levels of Nedd4- 1 (Scudder et al., 2014). The E3 ligase complex APCCdh1 also appears to become engaged upon long-term bicuculline treatment, as loss of this protein prevents bicuculline's effects on synaptic strength, though it is unclear whether this is due to direct targeting of AMPARs (Fu et al., 2011). RNF167 may handle short-term regulation of surface AMPARs while Nedd4-1 and APCCdh1 act on a longer time scale to homeostatically control synaptic strength, supported by the fact that bicuculline-induced synaptic scaling is blocked by the loss of either of these ligases (Fu et al., 2011; Scudder et al., 2014) (**Figure 2**).

In addition to pharmacological manipulations, AMPAR ubiquination has been studied using various other techniques. Using light-gated glutamate receptors to activate a subset of cultured neurons, Hou et al. demonstrated that synapses receiving 30 min of prolonged activity reduced their total and surface GluA1 and also experienced a site-dependent increase in polyubiquitin conjugates and the ligase Nedd4-1 (Hou et al., 2011). These data suggest that homeostatic scaling via ubiquitindependent pathways can occur on a single synapse level. To date, only a few groups have examined the role of receptor ubiquitination in vivo (Yuen et al., 2012; Atkin et al., 2015). Yuen et al. found that repeatedly exposing rats to stress leads to ubiquitination of GluA1 and the NMDAR subunit NR1 in the prefrontal cortex by Nedd4-1 and Fbx2, respectively, and that this results in reduced levels of these receptors and reduced glutamatergic transmission, which may underlie the stress-induced cognitive deficits observed.

# Function of Ubiquitination in AMPAR Trafficking

The field has converged on the idea that ubiquitination plays a critical role in regulating the abundance and localization of AMPARs in neurons. However, the exact role that ubiquitin conjugation plays remains debated. Direct conjugation to AMPAR subunits at the cell surface may function as a signal for internalization by triggering the assembly of endocytic machinery. Alternatively, the internalization process may occur prior to conjugation, with ubiquitination instead serving to direct endocytosed receptors toward a fate of degradation and prevent them from recycling to the surface.

In C. elegans it was first observed that the abundance of GLR-1, the C. elegans non-NMDA type glutamate receptor, is regulated by ubiquitin (Burbea et al., 2002; Juo and Kaplan, 2004). GLR-1 was found to be ubiquitinated in vivo, and mutations in GLR-1 which block ubiquitination increase the abundance of the receptor at synapses and alter locomotion behavior in a manner consistent with increased synaptic strength. In this system, overexpression of ubiquitin caused a decrease in GLR-1 abundance, and mutations in unc-11, which encodes the clathrin adaptin protein AP180, blocked the effect. Additionally, ubiquitin-conjugated GLR-1 accumulated in neurons lacking functional AP180. While the authors acknowledged that it is possible that ubiquitination is occurring in endosomes to control degradation, their data strongly supports a model where ubiquitination of GLR-1 occurs at the surface, prior to internalization through clathrin-mediated endocytosis (Burbea et al., 2002).

Three of the first papers to study mammalian AMPAR ubiquitination argued in favor of ubiquitination at the postsynaptic membrane (Schwarz et al., 2010; Fu et al., 2011; Lin et al., 2011). In these these studies, blocking ubiquitination by mutating the relevant GluA1 lysines (GluA1-4KR) or knocking down the E3 ligase responsible (Nedd4-1 or Cdh1, activator of APC) prevented the detection of internalized GluA1 during a stimulation-induced internalization assay. Thus, the authors concluded that ubiquitination of AMPAR subunits is a necessary

step in the internalization of stimulated receptors. However, an alternate explanation could be that surface receptors are indeed internalized in these conditions, but upon failure of ubiquitination in a nascent endosomal vesicle, receptors are recycled back to the membrane during the time-frame of the internalization assay. In that case, the lack of a sorting signal by ubiquitin would cause the same observed effect as the lack of internalization. However, in agreement with the aforementioned studies, a recent paper identified a role for the endocytic adaptor protein Eps15, which is known to be critical in supporting the internalization of epidermal growth factor receptor (Goh and Sorkin, 2013), in ubiquitin-dependent receptor trafficking (Lin and Man, 2014). The authors found that levels of Eps15 affected surface expression of GluA1 through ubiquitin-dependent interactions with this subunit and also demonstrated that clathrin-mediated endocytosis is necessary for the ubiquitininduced enhancement in receptor internalization (Lin and Man, 2014). Since Eps15 is involved in the recruitment of endocytic machinery at the surface, the authors conclude that this ubiquitin-mediated interaction is occurring prior to AMPAR internalization.

There have been studies of AMPAR ubiquitination which concluded that ubiquitination is not necessary for the internalization step of this pathway (Lussier et al., 2011; Widagdo et al., 2015). In investigating activity-induced GluA2 ubiquitination, Lussier et al. utilized dynasore to block dynaminmediated endocytosis and sucrose to prevent formation of clathrin-coated pits and observed that these manipulations prevent the detection of bicuculline-induced ubiquitination of GluA2 (Lussier et al., 2011). This supports a model where ubiquitin conjugation occurs after internalization to control receptor sorting. Similarly, a recent paper observed that inhibition of dynamin-mediated endocytosis with dynasore abolishes bicuculline- or AMPA-induced ubiquitination of all four AMPAR subunits (Widagdo et al., 2015). Curiously, this study also utilized GluA1 mutants that cannot be ubiquitinated but found that these lysine mutations did not prevent agonistinduced internalization, in contrast to the previously described papers (Schwarz et al., 2010; Lin et al., 2011). Instead, the lysine mutations reduced the amount of internalized GluA1 that co-localized with LAMP-1 positive late endosome/lysosomes and allowed more GluA1 to return to the surface. As a result, the authors conclude that ubiquitination occurs after receptors have been internalized, likely in early endosomes.

While the role of ubiquitination in regulating AMPARs has only been explored fairly recently, extensive work has been done to identify the role of ubiquitin in controlling surface proteins in non-neuronal cells. Epidermal Growth Factor Receptor (EGFR) has been the subject of numerous studies, as its ligand-induced removal from the cell surface is regulated by the E3 ubiquitin ligase c-Cbl (Goh and Sorkin, 2013). Though considerable debate continues to exist surrounding the various internalization pathways, ample evidence has shown that direct ubiquitination of EGFR by c-Cbl can serve as a signal for the assembly of clathrin-mediated endocytic machinery (including Eps15) and subsequent endocytosis (Stang et al., 2000; de Melker et al., 2001). However, EGFR can also be internalized through a nonubiquitin-dependent pathway, and ubiquitination can instead occur while EGFR is located in endosomes, where it serves as a signal for degradation (Levkowitz et al., 1998; Huang et al., 2007; Goh and Sorkin, 2013). Thus, it is reasonable that AMPAR internalization could occur through multiple pathways that involve ubiquitin, and that ubiquitination can occur on the surface to signal internalization or at early or late endosomes to control receptor fate. The exact conditions (i.e., type and intensity of neuronal stimulation) could determine which pathway surface AMPARs engage in. Since there appear to be a few distinct E3 ligases that can target AMPAR subunits, these ligases may engage the receptors at different points in the internalization process. Additionally, since AMPARs can be composed of four different subunits, the exact composition of surface receptors may also be a factor.

# Fate of Internalized AMPARs

Though there is still uncertainty about the exact function of AMPAR ubiquitination, the ultimate fate of ubiquitinated AMPARs can be determined by the ubiquitin chain length and topology as well as the duration of the modification. Mono- and short-chain ubiquitination of membrane proteins often leads to their internalization and degradation by the lysosome, while ubiquitin chain lengths of four or more (polyubiquitination) typically targets substrates for proteasomedependent degradation (Clague and Urbé, 2010). The duration and dynamics of the ubiquitinated state are, in many cases, regulated by other post-translational modifications and the activity of DUBs which counteract the ubiquitin conjugation forward reaction (Pickart, 2004). In the following sections we discuss evidence for both lysosomal and proteasomal degradation of AMPARs.

# Lysosomal Degradation of AMPARs

The lysosome is a membrane-bound organelle which contains hydrolytic enzymes that break down cellular components and allow them to be recycled. It maintains a low pH (pH < 5) through proton pumps nested inside the lysosomal membrane, which provides an ideal environment for the hydrolytic enzymes to function. Two main degradative pathways converge at the lysosome: the ESCRT pathway and the autophagy pathway. In the ESCRT pathway, membrane proteins are endocytosed and routed to the MVB and then to the lysosome (Hurley, 2008; Henne et al., 2011). In the autophagy pathway, cytoplasmic components are engulfed in an autophagosome, which fuses with the lysosome to form an autolysosome (Shintani and Klionsky, 2004; Levine and Kroemer, 2008).

Trafficking of AMPARs to the lysosome was first characterized in work by Ehlers MD, where he showed that bath application of AMPA targets AMPARs to the early endosome and subsequently to late endosome/lysosome compartments. Furthermore, AMPAinduced degradation was blocked by lysosomal inhibition (Ehlers, 2000). On the other hand, other studies have shown AMPAR subunit composition controls its trafficking. For instance, ectopic expression of tagged GluA subunits in cultured hippocampal neurons, which favors homomeric assembly, revealed that specific synaptic cues govern internalization and endocytic sorting to recycling or degradation pathways (Lee et al., 2004), while subunit-specific interactions with stargazin and PKC may control endocytic sorting to lysosomes (Kessels et al., 2009). The distinct trafficking of AMPARs based on subunit composition highlights the cell's ability to fine-tune surface proteins in order to tightly control changes in synaptic plasticity.

The large majority of AMPAR trafficking studies have revealed that the phospho-status of carboxy terminal residues tightly controls the stability of AMPARs at the synaptic membrane (Shepherd and Huganir, 2007). Our group, however, was the first to demonstrate that activity-dependent AMPAR ubiquitination by the E3 ligase Nedd4-1 targets AMPARs to the lysosome for degradation (Schwarz et al., 2010). Inhibition of the lysosome not only prevented AMPA-induced degradation of ubiquitinated GluA1-containing AMPARs but it also increased colocalization of AMPARs with lysosomes when Nedd4-1 was overexpressed (Schwarz et al., 2010).

One form of synaptic plasticity in which AMPARs are internalized and potentially degraded by the lysosome is longterm depression (LTD). Interestingly, however, one group found that inhibition of the lysosome did not affect LTD induction. Rather, expression of a dominant negative Rab7, which regulates trafficking from the late endosome to the lysosome, significantly reduced LTD expression compared to controls (Fernández-Monreal et al., 2012). The authors suggest that the sorting decision of internalized AMPARs between Rab7- or Rab11 dependent trafficking (which route to lysosome or back to synaptic membrane, respectively) is a key determinant for LTD induction. The authors also show that dephosphorylation of S845 on GluA1 is correlated with AMPAR degradation by the lysosome (Fernández-Monreal et al., 2012). Indeed, in GluA1 S845A phosphomutant mice, LTD is altered and AMPARs are constitutively degraded by the lysosome (He et al., 2009). This study supports the idea that LTD involves AMPAR internalization and degradation by the lysosome. However, it is still to be determined if AMPAR ubiquitination is required for LTD.

Recently, however, autophagy-dependent degradation of AMPARs has been shown to occur during LTD. Shehata et al. discovered that chemical LTD (chemLTD) induces autophagydependent degradation of AMPARs via inhibition of the PI3K-Akt-mTOR pathway (Shehata et al., 2012). Furthermore, they indicate that autophagosomes can enter dendritic spines in an activity-dependent manner, suggesting autophagy can degrade AMPARs (Shehata et al., 2012). One interesting avenue of research would be to explore lysosomal trafficking during synaptic plasticity and examine how lysosomal trafficking may change during these activity manipulations.

# Proteasomal Degradation of AMPARs

The ubiquitin proteasome system (UPS) is one of the most widely studied pathways for protein degradation in eukaryotic cells. Polyubiquitinated proteins are recognized by the 26S proteasome where they can be degraded into small peptides and amino acids. The 26S proteasome is a large energy-dependent protease formed by the co-assembly of a 20S proteasome (the catalytic component) and 19S cap (regulatory particle which binds ubiquitinated proteins) (Hershko and Ciechanover, 1998). It was first demonstrated by Zhang et al. that AMPAR turnover was proteasome-dependent. In this study, the authors showed that Na,K-ATPase (NKA) inhibition led to rapid degradation of AMPAR subunits which was blocked by proteasome inhibitors (Zhang et al., 2009). Additionally, they showed that AMPAR degradation is sodium-dependent during NKA inhibition. They reasoned that since Nedd4-1, an E3 ligase demonstrated by our lab to target AMPAR for ubiquitination (Schwarz et al., 2010), is also regulated by sodium, it is likely the ligase that ubiquitinates AMPARs and targets them for proteasomedependent in response to NKA inhibition (Zhang et al., 2009). A follow-up study from this group found Nedd4-1 does indeed ubiquitinate AMPARs and that under basal conditions, inhibition of the proteasome leads to a build-up of ubiquitinated AMPARs (Lin et al., 2011). In addition, Hou et al. used light-controlled activity stimulation of synapses and found that AMPARs are degraded after repeated stimulation and inhibiting the proteasome prevented this loss, while lysosomal inhibition has no effect (Hou et al., 2011). While the observation of proteasomedependent turnover of AMPARs differs from findings by our group, which showed that AMPAR activation leads to Nedd4- 1-dependent ubiquitination and degradation of AMPARs by the lysosome (Schwarz et al., 2010; Scudder et al., 2014), it suggests there are multiple signaling pathways that can control turnover of AMPARs. Regardless, given how AMPAR trafficking and degradation must be tightly regulated it is not surprising that AMPARs can be degraded by both machineries.

# Deubiquitination of AMPARs and Recycling

While the ubiquitin signal can have a profound cellular effect, in some cases ubiquitinated proteins are spared from degradation. The ubiquitination process can be counteracted by DUBs, which remove the ubiquitin moiety. For membrane proteins such as AMPARs, deubiquitination can facilitate their recycling to the surface of the cell. There are 5 major classes of DUBs and they can function to cleave ubiquitin-linked molecules to (1) maintain the ubiquitin pool, (2) rescue proteins targeted for degradation, or (3) prevent UPS-dependent protein degradation. Two DUBs have been implicated in AMPAR deubiquitination: USP8 and USP46. USP8, which is found in the somatic, dendritic and synaptic compartments of neurons, becomes dephosphorylated and activated upon calcium influx (Scudder et al., 2014). Our group found that NMDAR activation negatively regulates AMPAR ubiquitination, suggesting that the influx of calcium through NMDAR channels activates USP8. This causes the deubiquitination of AMPARs, resulting in their ability to escape degradation and recycle back to the membrane. The functional importance of USP8 was further demonstrated when overexpression of USP8 prevented bicuculline-induced downscaling (Scudder et al., 2014). Since USP8 counteracts Nedd4-1's ability to ubiquitinate and target AMPARs for degradation, this study provides the first mechanistic evidence for opposing activity-dependent control of a ubiquitin ligase and DUB in the regulation of homeostatic plasticity.

USP46 has also been implicated in AMPAR deubiquitination. In the ventral nerve cord of C. elegans it was found that USP46 binds to GLR-1, and negatively regulates the levels of its ubiquitination. Conversely, mutant USP46 increases ubiquitinated GLR-1 (Kowalski et al., 2011). Mechanistically, USP46 can bind with WD40-repeat (WDR) proteins WDR-20 and WDR-48 to stimulate USP46 catalytic activity and increase GLR-1 levels (Dahlberg and Juo, 2014). Recently, in dissociated rat neuronal cultures, USP46 was found to deubiquitinate AMPARs (Huo et al., 2015). It appears that both USP8 or USP46 knockdown lead to elevated AMPAR ubiquitination and reduced miniature excitatory postsynaptic currents (mEPSC) amplitude while overexpression of either DUB leads to a reduction in AMPAR ubiquitination and an increase in surface AMPAR abundance (Scudder et al., 2014; Huo et al., 2015). Given that multiple E3 ubiquitin ligases and DUBs have been shown to target AMPARs, it will be of great interest to understand how the dynamics of AMPAR ubiquitination and deubiquitination are regulated or influenced by other post-translational modifications such as phosphorylation.

# Ubiquitination and SUMOylation of Non-AMPA Glutamate Receptors

While the trafficking of AMPARs to and from the synapse is thought to underlie most changes in synaptic strength at excitatory synapses, control of other glutamate receptors is also critical in regulating transmission and the capacity for plasticity. The number of NMDA and kainate receptors (KARs) at the postsynaptic membrane can be controlled by multiple mechanisms, including direct ubiquitination. In this section we review what is currently known about ubiquitination and ubiquitin-like modification of NMDARs and KARs.

Like AMPARs, kainate receptors are internalized through separate pathways in response to NMDA treatment or direct activation by an agonist (kainate). Activation of NMDARs promotes the targeting of internalized KARs to recycling endosomes, allowing them to return to the surface, while direct activation of KARs causes the majority of internalized receptors to be degraded via lysosomes, thus reducing surface and total levels (Martin and Henley, 2004). Kainate-evoked endocytosis requires phosphorylation of the GluK2 subunit by protein kinase C (PKC) and the conjugation of the small ubiquitin-like modifier SUMO-1 by the SUMO conjugating enzyme PIAS3 (Martin et al., 2007). Treatment with kainate causes phosphorylation of Cterminal sites of GluK2, which then induces the SUMOylation of this region and causes the subsequent internalization and degradation of these receptors (Konopacki et al., 2011). SUMOylation is thought to occur at surface receptors and serve as a signal for endocytosis, as non-SUMOylatable GluK2 does not undergo agonist-induced endocytosis and kainate-induced SUMOylation of surface GluK2 is detected when internalization is blocked by sucrose (Martin et al., 2007). These studies indicate that SUMOylation serves as a critical signal to control surface expression of KARs and KAR-mediated synaptic transmission. The authors theorize that this mechanism may exist to protect neurons from excitotoxic damage. Additionally, one recent report indicates that SUMOylation of GluK2 is necessary for the long-term depression of KAR-mediated synaptic transmission evoked by low-frequency stimulation at mossy fiber-CA3 synapses, demonstrating a role for SUMO conjugation in activitydependent synaptic plasticity (Chamberlain et al., 2012).

Kainate receptors can also be ubiquitinated by the E3 ligase parkin and the Cul3-containing E3 ligase complex (Salinas et al., 2006; Helton et al., 2008). GluK2 interacts with the postsynaptically-located protein actinfilin, which serves as a scaffold to bring the receptor subunit in contact with Cul3. Reduction of Cul3 or actinfilin leads to increased surface GluK2 and reduced ubiquitination of this subunit (Salinas et al., 2006). However, it is not yet known whether this occurs at surface KARs or whether GluK2 is ubiquitinated and degraded via an endoplasmic reticulum-associated degradation (ERAD) pathway. Since the neuronal activity-dependence of this phenomenon was not explored, this mechanism may constitutively control KAR levels. A recent study has also reported GluK2 ubiquitination, in this case by the E3 ligase parkin, a protein known to be mutated in many cases of Parkinson's disease (Maraschi et al., 2014). Parkin mutations in mice and human patients cause large increases in total levels of GluK2. Parkin appears to directly ubiquitinate this subunit and control its surface expression in neurons, and the interaction between these proteins increases after treatment with glutamate. As reported in other studies, loss of parkin increases the susceptibility of neurons to excitotoxic damage and death after treatment with kainate (Staropoli et al., 2003; Helton et al., 2008). Thus, the authors conclude that this ubiquitination pathway serves to protect neurons from excitotoxic damage and loss of this pathway through parkin mutations may contribute to the pathology of Parkinson's disease. Collectively, these studies indicate that a combination of phosphorylation, ubiquitination, and SUMOylation work to control KAR surface abundance and allow for synaptic plasticity and protection from excitotoxic stress.

Control of surface NMDARs is critical in regulating synaptic transmission and synaptic plasticity, and also in limiting excitotoxicity. In response to prolonged increases in activity caused by bicuculline in vitro, the subunit GluN1 was found to be ubiquitinated by the E3 ligase Fbx2 and the synaptic levels of this subunit are reduced, suggesting that ubiquitination may serve as a mechanism to reduce receptor levels during synaptic scaling (Kato et al., 2005). However, ubiquitin conjugation occurs at an extracellular domain of GluN1, through a mechanism involving retrotranslocation of NMDARs. Ubiquitination of GluN1 by Fbx2 was also reported to occur in vivo in the prefrontal cortex as a result of repeated stress. This mechanism appears to partially underlie stress-induced cognitive impairments, in conjunction with GluA1 ubiquitination by Nedd4-1 (Yuen et al., 2012). Recent studies from Fbx2 knockout mice showed increases in GluN1 and GluN2A but no changes to GluN2B levels and that the increased GluN1 subunits are mostly found at the cell surface. The build-up of unused NMDAR subunits results in an accumulation at non-synaptic sites leading to the formation of shaft synapses (Atkin et al., 2015). It was found that highmannose glycans reside on the extracellular region of GluN subunits and that Fbx2 can bind to high-mannose glycans (Atkin et al., 2015). This suggests that internalization must precede Fbx2-directed ubiquitination of GluN subunits.

Ubiquitination of the subunit GluN2B by the ubiquitin ligase Mind bomb-2 (Mib2) has also been reported (Jurd et al., 2008). In this pathway, phosphorylation of this subunit causes direct ubiquitination and downregulation of surface NMDARs, potentially to prevent the pathological effects of excessive NMDAR activation. Nedd4-1 has also recently been reported to ubiquitinate the GluN2D subunit and decrease NMDAR signaling, though this has not yet been verified in neurons (Gautam et al., 2013) (See **Figure 1**, Nedd4-1<sup>∗</sup> ). Taken together, these studies indicate that the surface expression of NMDARs is tightly regulated by many mechanisms, several of which involve direct ubiquitin conjugation to receptor subunits. These pathways likely work to both homeostatically control surface expression and protect neurons from excitotoxic stress.

# Degradation of Glutamate Receptor Interacting and Postsynaptic Scaffold Proteins

The trafficking of glutamate receptors to and from the postsynaptic membrane in part relies on direct and indirect interactions with other proteins and these interacting proteins can regulate many forms of plasticity at excitatory synapses. Several of these proteins act as scaffolds or regulatory proteins to ensure the proper postsynaptic insertion, removal, or stabilization of glutamate receptors. As such, the ubiquitindependent degradation of these proteins could therefore have profound effects on glutamate receptor trafficking and function as well as synaptic plasticity.

One of the first studies that examined protein turnover at synapses revealed that the ubiquitination and degradation of several PSD proteins was regulated by synaptic activity. Interestingly, these effects were controlled by chronic activity modulation and found to be bi-directional. The turnover of key ionotropic and metabotropic glutamate receptor scaffolding molecules including Shank, AKAP79/150 (AKAP), and GKAP was found to be mediated by UPS-dependent degradation (Ehlers, 2003). Subsequently, the E3 ubiquitin ligase TRIM3 was identified to target GKAP (also known as SAPAP) for ubiquitin-dependent degradation (Hung et al., 2010). Furthermore, activity- and phosphorylation-dependent ubiquitination and degradation of GKAP was shown to be important for global remodeling of synapses. Altering GKAP levels at synapses by overexpression or knockdown alters the remodeling of PSD-95 and Shank and blocks bidirectional synaptic scaling (Shin et al., 2012). This indicates that half-life control of specific PSD scaffolds and regulatory proteins is important for the overall activity-dependent remodeling of synapses.

The ubiquitination of PSD-95, a major PSD scaffold that links both NMDA- and AMPA-type glutamate receptors to signaling complexes and to the actin cytoskeleton (Kim and Sheng, 2004), has been reported by several groups. Colledge et al. found PSD-95 to be ubiquitinated by the E3 ligase Mdm2 in response to NMDA receptor activation. Furthermore, they showed that preventing PSD-95 ubiquitination and degradation blocked NMDA-induced AMPAR internalization and synaptically-induced LTD (Colledge et al., 2003). In contrast, Bianchetta et al. found that increased cyclin-dependent kinase 5 (Cdk5) activity promotes PSD-95 ubiquitination by increasing Mdm2 association with PSD-95. In this case, however, they found that PSD-95 levels were unchanged (Bianchetta et al., 2011). The authors therefore proposed a nonproteolytic role for PSD-95 ubiquitination involving increased interaction with the clathrin adaptor protein complex protein AP-2 to promote NMDAR-induced internalization of AMPARs (Bianchetta et al., 2011). More recently, the ubiquitination and degradation of PSD-95 has been linked to autism spectrum disorders (ASDs). Tsai et al. found that the myocyte enhancer factor 2 (MEF2) and fragile X mental retardation protein (FMRP)-regulated ASD-linked gene, protocadherin 10 (Pcdh10), links ubiquitinated PSD-95 to proteasomal turnover. In contrast, blocking Pcdh10 interaction with proteasomes prevented PSD-95 degradation and synapse elimination (Tsai et al., 2012).

Interestingly, negative regulators of synaptic AMPARs are also degraded by the UPS. Arc is an important synaptic protein that has been shown to promote the internalization of AMPARs (Chowdhury et al., 2006; Shepherd et al., 2006). While investigating the function of Ube3A, the gene mutated in the neurological disorder Angelman syndrome, Greer et al. found that loss of Ube3A prevents Arc ubiquitination and degradation with a concomitant decrease in AMPARs (Greer et al., 2010). It has been shown, however, that Ube3A may regulate Arc protein levels independent of direct ubiquitination (Kühnle et al., 2013). More recently, Mabb et al. found that the RING domain ubiquitin ligase Triad3A/RNF216 targets Arc for ubiquitination and degradation. In the absence of Triad3A, Arc levels are increased, leading to a loss of AMPARs and disruption of Arc-dependent forms of synaptic plasticity (Mabb et al., 2014).

PICK1 and GRIP1, two other AMPAR interacting and scaffold proteins, have also been shown to be regulated by ubiquitindependent protein degradation. The E3 ligase parkin, encoded by a gene involved in Parkinson's disease, was found to target PICK1 (Joch et al., 2007). In this case, however, PICK1 was found to be mono-ubiquitinated by parkin, which negatively regulates acid-sensing ion channels (ASIC). Therefore, it is speculated that enhanced ASIC activity could promote neurodegeneration in Parkinson's disease (Joch et al., 2007). While yet to be determined, it is plausible that parkin-mediated ubiquitination of PICK1 could regulate its interaction with AMPARs. GRIP1, which is primarily complexed with GluA2-containing AMPA receptors stabilized at the postsynaptic membrane (Kim and Sheng, 2004), was found to be rapidly degraded in an activity and calcium-dependent manner. Proteasome inhibition blocked these effects, indicating GRIP1 turnover to be proteasome dependent. Advancements in ubiquitin proteomics, where diglycine affinity strategies are now being used to enrich substrates and identify sites of ubiquitination (Na et al., 2012), will inevitably uncover other synaptic proteins that regulate glutamate receptor trafficking, function, and synaptic plasticity.

# Conclusion

In recent years, glutamate receptor ubiquitination has emerged as a key post-translational modification that can control glutamate receptor trafficking and degradation. The discovery that glutamate receptors can be tagged by ubiquitin in an activitydependent manner highlights its importance in modulating synaptic plasticity. Interestingly, while several studies have revealed that internalized receptors can be recycled back to the synaptic membrane, the ultimate degradative fate of the receptors has been far less studied. In this review we discussed ubiquitination as a signal for glutamate receptor degradation by the lysosome or the proteasome. Available data suggests that glutamate receptors can be degraded by these cellular components but detailed mechanisms for their trafficking have not been fully elucidated and would be a particularly interesting area of research. Additionally, other post-translational modifications such as phosphorylation have also been shown to play a role in glutamate receptor trafficking. Since the phosphorylation status of AMPAR subunits is a key determinant of their synaptic abundance, it remains to be determined how phosphorylation and ubiquitination of glutamate receptors are coordinated. It may be that degradation of glutamate receptors is regulated by the dynamic interplay between receptor phosphorylation and ubiquitination. Pursuing these questions would ultimately provide insight into how neurons regulate receptor trafficking and turnover with high specificity in response to signaling cues.

# Acknowledgments

We thank members of the Patrick laboratory for helpful discussion. This work was supported by NSF Graduate Research Fellowships (MG, SS), Ford Graduate Research Fellowship (MG), NIH Grant NS060847 (GP), and Grant P50-GMO85764 from the Center for Systems Biology.

# 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 © 2015 Goo, Scudder and Patrick. 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.

# Local ubiquitin-proteasome-mediated proteolysis and long-term synaptic plasticity

# *Ashok N. Hegde\*, Kathryn A. Haynes†, Svitlana V. Bach† and Brenna C. Beckelman†*

Department of Neurobiology and Anatomy, Wake Forest University Health Sciences, Winston-Salem, NC, USA

### *Edited by:*

Nicola Maggio, The Chaim Sheba Medical Center, Israel

### *Reviewed by:*

Nicola Maggio, The Chaim Sheba Medical Center, Israel Diasynou Fioravante, University of California, Davis, USA

### *\*Correspondence:*

Ashok N. Hegde, Department of Neurobiology and Anatomy, Wake Forest University Health Sciences, Medical Center Boulevard, Winston-Salem, NC 27157, USA e-mail: ahegde@wakehealth.edu

†Kathryn A. Haynes, Svitlana V. Bach and Brenna C. Beckelman have contributed equally to this work.

The ubiquitin-proteasome pathway (UPP) of protein degradation has many roles in synaptic plasticity that underlies memory.Work on both invertebrate and vertebrate model systems has shown that the UPP regulates numerous substrates critical for synaptic plasticity. Initial research took a global view of ubiquitin-protein degradation in neurons. Subsequently, the idea of local protein degradation was proposed a decade ago. In this review, we focus on the functions of the UPP in long-term synaptic plasticity and discuss the accumulated evidence in support of the idea that the components of the UPP often have disparate local roles in different neuronal compartments rather than a single cell-wide function.

**Keywords: ubiquitin, proteasome, learning and memory, protein degradation, ubiquitin conjugation**

Many years of research on synaptic plasticity using various model systems, from slugs and flies to mammals, has yielded a wealth of information on mechanisms that underlie both short- and longterm synaptic plasticity (Kandel and Schwartz, 1982; Belvin and Yin, 1997; Mayford, 2007; Walters and Moroz, 2009; Abrams, 2012). It is now generally accepted that short-term synaptic plasticity requires modification, usually by phosphorylation, of pre-existing proteins. Long-term plasticity, however, requires gene transcription and translation of newly transcribed mRNAs (Hernandez and Abel, 2008; Sossin, 2008; Katche et al., 2013). Research over the last couple of decades has revealed another major mechanism with roles in both short- and long-term synaptic plasticity: protein degradation by the ubiquitin-proteasome pathway (UPP; Hegde, 2010; Fioravante and Byrne, 2011; Jarome and Helmstetter, 2013).

Previously one of us proposed a role for local proteolysis by the UPP in synaptic plasticity (Hegde, 2004). The gist of this theory is that proteolysis by the UPP performs disparate functions in different parts of the neuron. Since then, investigations on different model systems have obtained evidence to support the function of local proteolysis. The idea has gained acceptance by other researchers as well (Segref and Hoppe, 2009). In the nervous system, to achieve synapse-specific effects, proteolysis needs to be spatially restricted. Therefore, local protein degradation is likely to be critical during development of synaptic connections as well as synaptic plasticity in adult organisms.

How might local protein degradation be achieved in neurons? A simple way would be to restrict the protein substrate or the enzymes of the UPP to a subcellular location. For example, proteins whose expression is largely restricted to the synapses could be locally degraded because all the requisite UPP components

are present at the synapse. In addition, substrates can be made susceptible (or resistant) to ubiquitination by locally controlled phosphorylation in neurons. Likewise ubiquitin ligases can be activated or inactivated locally by phosphorylation or other posttranslational modifications (e.g., conjugation of ubiquitin-like protein Nedd8 to Cul1 that activates SCF ligases; Osaka et al., 2000; Lyapina et al., 2001) can be locally controlled as well (Hegde and Upadhya, 2007; Hegde, 2010). Moreover, specific E3 ligases can also be sequestered to specific cellular compartments (Tsai et al., 2012; Ichimura et al., 2013). Experimental evidence has been obtained for some of these possibilities. Evidence gathered over recent years indicates that proteasome activity is differentially regulated in different neuronal compartments as well. A few examples of local roles of ubiquitination and those of the proteasome in neuronal compartments are discussed below.

# **ROLES FOR LOCAL UBIQUITINATION AND DEUBIQUITINATION**

As described previously, the specificity of ubiquitination is largely controlled at the level of E3 ubiquitin ligases (Glickman and Ciechanover, 2002). The specificity of ubiquitination could also be regulated at the level of E2s because E2s are diverse and many unique E2–E3 combinations can be generated (Glickman and Ciechanover, 2002; Hegde, 2010). Several studies show evidence for local roles of E2s and E3s as well as for deubiquitinating enzymes (DUBs) during development of synaptic connections.

### **E2s**

In *Drosophila*, an E2 called ubcD1 controls dendritic pruning where local degradation appears to be critical. In this insect, most of the larval neurons die during metamorphosis, except for a cluster of peripheral sensory neurons named class IV dendritic arborization (C4da) neurons which survive to adulthood (Kuo et al., 2005). These neurons extensively remodel their dendrites by completely degrading the old arborization and by growing a new elaborate set of dendrites. During the remodeling of dendrites, axons are kept intact. Hence the molecular processes have to be spatially restricted. It was found that perturbations of the UPP by overexpression of an exogenous DUB called UBP2 from yeast, or mutations in E1 or a 19S proteasome subunit disrupted dendritic pruning. Subsequent studies identified the essential role of ubcD1 in this process (Kuo et al., 2006). Mutations in ubcD1 led to a blockade of dendritic pruning and retention of larval dendrites in C4da neurons. Based on additional experiments it was inferred that ubcD1 targets *Drosophila* inhibitor of apoptosis 1 (DIAP1) an E3 ubiquitin ligase. DIAP1 is required for degradation of a caspase called Dronc. Therefore, degradation of DIAP1 enables activation of the Dronc caspase locally in dendrites. Because the Dronc caspase is critical for severing dendrites of C4da neurons, restricted dendritic activation of this caspase allows preservation of C4da neurons while removing their dendrites (Kuo et al., 2005, 2006).

### **E3s**

A genetic screen for isolating mutants that enhance synaptic growth at the *Drosophila* neuromuscular junction revealed that a loss-of-function mutation in a gene called *highwire* (*hiw*) causes a substantial increase in number of synapses. The protein product of the *hiw* gene contains a RING finger domain which is a key part of some ubiquitin ligases (Wan et al., 2000).

Subsequent work carried out on the *Caenorhabditis elegans* homolog of the *hiw* gene called RPM-1 showed that the ligase functions to regulate presynaptic differentiation. RPM-1 protein is localized to the periactive zone, a presynaptic region excluded from the active zone and synaptic vesicles. RPM-1 combines with an F-box protein called FSN-1 and *C. elegans* homologs of SKP1 and Cullin toform an SCF-like ubiquitin ligase complex. The localizedfunction of this ubiquitin ligase in the periactive zone is critical for presynaptic differentiation in *C. elegans* (Liao et al., 2004). The downstream target of RPM-1 in *C. elegans* is a MAP kinase kinase kinase (MAPKKK) called delta-like homolog 1 (DLK-1) which is also localized to the periactive zone like RPM-1. Inactivation of the DLK-1 cascade suppresses RPM-1 loss-of- function phenotypes whereas DLK-1 overexpression causes synaptic aberrations similar to the ones seen with RPM-1 mutations (Nakata et al., 2005). In *Drosophila*, the downstream target of hiw is a MAPKKK encoded by a gene called *wallenda* (Collins et al., 2006). Although the downstream effectors of DLK-1 and wallenda proteins are different, attenuation of the signaling mediated by these proteins inhibits synaptic growth in similar ways (Fulga and Van, 2008).

In vertebrates, a homolog of *hiw* called *Phr1* regulates development of neuronal connections. Studies using mice with a mutation in the *Phr1* gene (a mutation called *Magellan*), which lacks the *C*-terminal ligase domain, revealed that the Phr1 protein is localized to axon shaft and is largely excluded from growth cones and distal processes. The substrate of Phr1 is most likely DLK in mice as well. Distribution of DLK is non-overlapping with that of Phr1; DLK is present in growth cones with only low levels in axon shaft (Lewcock et al., 2007).

Local regulation of other E3 ligases has also been reported. For example, in hermaphrodite-specific motor neurons of *C. elegans*, an SCF ligase containing the protein SKR-1 and an F-box protein called SEL-10 mediates developmental elimination of synapses. A synaptic adhesion molecule called SYG-1 binds to SKR-1 and blocks the assemblage of the SCF complex which protects the nearby synapses (Ding et al., 2007). In vertebrates, the ligase anaphase promoting complex containing the substrate-binding protein Cdh1 curtails axonal growth when it is nuclear, but when it is cytosolic it promotes dendrite growth without affecting axons (Kim et al., 2009). Another ligase KLHL20-Cullin 3 promotes neurotrophin-induced neurite outgrowth by targeting RhoGEF for proteolysis (Lin et al., 2011).

### **DEUBIQUITINATING ENZYMES**

Ubiquitination can be reversed by removal of the attached ubiquitin molecules by DUBs. Thus DUBs provide important negative regulation of protein degradation. Like ligases, DUBs can act locally to reverse ubiquitination. A search for molecules that regulate the size and strength of synapses in *Drosophila* found that a DUB encoded by the *fat facets (faf)* gene functions in synapse formation. During development of the *Drosophila* nervous system, *faf* overexpression leads to overgrowth of synapses and disruption of synaptic transmission. A similar phenotype is observed when a yeast DUB is expressed in the fruit fly CNS (DiAntonio et al., 2001).

Deubiquitinating enzymes also have a local synaptic role in mammals. A DUB called Usp14 is essential for synaptic development and function in mouse neuromuscular junctions. The role for Usp14 in the nervous system was originally discovered through studies on mice with the *ataxia* (*axj* ) mutation, a recessive mutation characterized by severe tremors, hind limb paralysis, and postnatal lethality (Wilson et al., 2002). The *ax<sup>j</sup>* gene encodes Usp14 the protein product of which associates with the proteasome and is believed to help disassemble polyubiquitin chains and recycle ubiquitin thus maintaining ubiquitin levels in the cell. Accordingly, loss of Usp14 results in reduced ubiquitin levels in many tissues of the *ax<sup>j</sup>* mice including the brain (Anderson et al., 2005). The motor defects of the *ax<sup>j</sup>* mice were rescued and viability was restored with transgenic Usp14 suggesting that Usp14 deficiency is the cause of neurological defects in these mice (Crimmins et al., 2006). Subsequent studies demonstrated that in Usp14 deficient *axj* mice ubiquitin loss occurred in the spinal cord and sciatic nerve. Biochemically, the majority of the loss was found to occur in synaptosomal fractions indicating that Usp14 at synaptic sites was critical. Loss of Usp14 caused presynaptic defects such as poor arborization of motor nerve terminals, and transgenic expression of Usp14 rescued these defects. Thus it appears that local Usp14 function is critical for maintaining ubiquitin levels and hence protein degradation at the synapse (Chen et al., 2009).

# **LOCAL ROLES OF THE PROTEASOME IN SYNAPTIC PLASTICITY**

Recently it has become clear that the proteasome is not the same in terms of its activity and function throughout the neuron. This realization came from attempts to resolve conflicting results obtained with proteasome inhibitors on long-term facilitation (LTF) in *Aplysia*. Initially, proteasome inhibitors were found to block induction of LTF (Chain et al., 1999). Later studies on LTF, however, showed that when the active form of lactacystin, *clasto*lactacystin β-lactone (hence forth referred to as β-lactone), was applied in the culture medium to sensory-motor neuron synapses, this resulted in enhanced LTF and an increase in neurite outgrowth in isolated sensory neuron (Zhao et al., 2003). The results on enhanced neurite elongation are consistent with those obtained in PC12 and Neuro2A cells in which lactacystin induces neurite outgrowth (Fenteany et al., 1994). The discrepancies between these results can be resolved by hypothesizing that the proteasome has different roles in different cellular compartments (Hegde, 2004). In a given neuron, the proteasome is likely to carry out dissimilar tasks in various subcellular compartments resulting in distinct physiological outcomes at separate neuronal locales. Thus, blocking discrete roles of the proteasome during induction of memory would lead to distinctive and even opposite effects on synaptic strength. For example, the proteasome degrades transcription repressors. Proteolytic removal of transcription repressors should enable transcription activators to induce gene expression and consequent development of LTF. If the proteasome is inhibited only in the nucleus before the repressors are degraded, gene expression and hence induction of LTF should be blocked. Degradation of the CREB repressor CREB1b by the UPP in response to LTF-inducing protocols (Upadhya et al., 2004) supports this idea. In contrast, if the proteasome is inhibited at the synapse causing accumulation of LTF-inducing proteins, LTF should be augmented. As postulated earlier, transcription is necessary during induction of LTF (or other forms of plasticity underlying long-term memory) for supplying mRNAs for synthesis of proteins that turn over rapidly in the early stages of memory formation (Hegde, 2004). Blocking the degradation of such proteins should cause long-term memory to form without transcription. Consistent with this notion, it was found that proteasome inhibitor-induced synaptic strengthening depends on translation but not transcription (Zhao et al., 2003).

Results from direct measurement of proteasome activity also support differential function of the proteasome in distinct neuronal compartments. Proteasome activity in the synaptic terminals is much higher compared to that in the nucleus in *Aplysia* nervous system as well as the mouse brain. Furthermore, the synaptic and nuclear proteasome activities are differentially regulated by protein kinases with a key role in synaptic plasticity such as PKA, PKC, and MAP kinase (Upadhya et al., 2006). Recently others have found that CaMKII can stimulate proteasome activity in cultured hippocampal neurons (Djakovic et al., 2009).

As discussed above, differential proteasome activity in the invertebrate *Aplysia* might explain conflicting results obtained in different studies. Does the idea of proteasomal activity affecting synaptic plasticity differentially hold true for vertebrates? It has been observed that the proteasome has differential roles during the induction and maintenance parts of late-phase long-term potentiation (L-LTP) in the murine hippocampus (Dong et al., 2008), which is discussed in detail in the next section.

The proteasome has been shown to be dynamically and locally regulated at the dendrites in cultured rat hippocampal neurons. In response to NMDA receptor activation proteasome was found to be redistributed from dendritic shafts to synaptic spines. What is the mechanism of redistribution of the proteasome? Neuronal activity increased the entry of the proteasome into dendritic shafts only to a small extent but drastically reduced their exit. In addition, proteasome was found to be sequestered persistently in the spines through association with cytoskeleton (Bingol and Schuman, 2006). Later investigations showed that a protein called NAC1, which is induced by psychostimulants, modulates the recruitment of the proteasome into the dendritic spines (Shen et al., 2007). Much of the evidence from these studies, however, was on the catalytic 20S core of the proteasome. Thus, it is not clear whether recruitment of the full 26S proteasome complex that degrades polyubiquitinated proteins is also regulated by NAC1. A recent study has suggested that the CaMKIIα subunit acts as a scaffold for the proteasome (Bingol et al., 2010). It is not clear how or if the functions of NAC1 and CaMKIIα relate to each other in sequestering the proteasome. Local proteolysis by the proteasome also has been shown to be critical for regulating spine outgrowth (Hamilton et al., 2012).

It is also likely that the proteasome operates locally to regulate other molecular processes required for synaptic plasticity such as translation of mRNA. For instance, proteasome is known to regulatefragile X mental retardation protein (FMRP),which modulates translation of a subset of mRNAs in dendrites. Moreover, proteasomal regulation of FMRP is required for metabotropic glutamate receptor-dependent LTD (Hou et al., 2006).

## **DISPARATE LOCAL ROLES OF THE PROTEASOME IN DENDRITES AND THE NUCLEUS: OPPOSITE CONSEQUENCES FOR INDUCTION AND MAINTENANCE OF L-LTP**

Investigations on hippocampal L-LTP showed that the proteasome inhibitor application to hippocampal slices prior to induction of L-LTP caused an increase in the magnitude of the early, induction phase but an inhibition of the late, maintenance phase (Dong et al., 2008). What is the basis of these differential effects of the proteasome on phases of L-LTP? The enhancement of the early part of L-LTP (referred to as Ep-L-LTP for convenience) by the proteasome inhibitor β-lactone is blocked by prior application of the translation inhibitor anisomycin but not by a transcription inhibitor actinomycin D. The increase in Ep-L-LTP caused by β-lactone is also prevented by prior application of rapamycin which blocks signaling that controls translation of a subset of mRNAs (Gingras et al., 2001). Moreover, Ep-L-LTP is augmented by β-lactone in dendrites isolated from the cell body by means of a surgical cut. These lines of evidence suggest that proteasome inhibition enhances Ep-L-LTP by stabilizing proteins locally translated from pre-existing mRNAs (Dong et al., 2008; **Figure 1A**).

How does proteasome inhibition block maintenance of L-LTP? The proteasome inhibitor β-lactone blocks maintenance of L-LTP only if applied prior to induction of L-LTP but not if applied 2 h after induction of L-LTP. Previous studies by others have established that the critical time window for transcription required for maintenance of L-LTP is2h(Nguyen et al., 1994). These results

**FIGURE 1 | Differential local roles of the proteasome in dendrites and in the nucleus during L-LTP. (A)** Proteasome Active: the proteasome in dendrites is highly active, translational activators such as eIF4E are degraded (broken green spheres) and protein substrates that positively regulate L-LTP are degraded (broken spheres). Therefore extent of L-LTP is limited and only normal L-LTP ensues. A retrograde signal is likely transmitted to the nucleus. Proteasome aids transcription of genes by degrading the CREB repressor ATF4 (broken squares in the nucleus) thus allowing for normal L-LTP maintenance. Transcribed mRNAs (triangles) travel to activated synapses. **(B)** Proteasome Inactive: when the proteasome is inhibited (indicated by X marks on the proteasome), translational activators are stabilized (intact green spheres) leading to increased protein synthesis

in dendrites. Also the newly synthesized proteins in dendrites are stabilized (intact spheres) and L-LTP-inducing stimulation protocols dramatically increase (upward arrow) the early part of L-LTP (Ep-L-LTP). Proteasome inhibition obstructs CREB-mediated transcription by preventing the degradation of transcription repressor ATF4 (intact squares in the nucleus). Proteasome inhibition could also inhibit the generation of the retrograde signal. Therefore, L-LTP is not maintained but decays (downward arrow). Proteasome inhibition also causes failure of sustained translation because of stabilization of translation repressors such as 4E-BP (intact red spheres) which accumulate after induction of L-LTP thus contributing to blockade of L-LTP maintenance. [Modified from Hegde (2010) and reprinted with permission from Cold Spring Harbor Laboratory Press].

suggest that proteasome inhibition blocks maintenance of L-LTP by inhibiting transcription. Additional molecular evidence supports this notion. Application of β-lactone to hippocampal slices significantly reduced induction of *brain-derived neurotrophic factor* (*BDNF*) mRNA by chemically induced LTP (cLTP) or L-LTP induced by a theta-burst protocol (Dong et al., 2008). *BDNF* is a CREB-inducible gene linked to maintenance of L-LTP (Barco et al., 2005).

How might proteasome inhibition block transcription? One possibility is that normally the UPP aids the degradation of transcription repressors. Hence proteasome inhibition would result in accumulation of these repressors thus blocking transcription. Consistent with this concept, it was found that a CREB repressor ATF4 is degraded by the UPP during cLTP and β-lactone application to hippocampal slices prevents degradation of ATF4. Furthermore, ATF4-ubiquitin conjugates accumulate during cLTP when the proteasome is inhibited (Dong et al., 2008; **Figure 1B**).

These studies have also revealed the changing role of the proteasome even in dendrites through progression of L-LTP. Application of β-lactone to isolated dendrites also blocks maintenance of the dendritic L-LTP (Dong et al., 2008). Under these conditions, there is no supply of newly transcribed mRNA from the cell body. Thus blockade of transcription by proteasome inhibition does not explain this phenomenon. The most likely possibility is that proteasome inhibition leads to a slow accumulation of translation repressors in dendrites. Buildup of translation repressors would also occur in the cell body which would hinder translation of newly transcribed mRNAs. Thus late stages of translation in both

dendrites and the cell body would be blocked by stabilization of translation repressors by proteasome inhibition. In support of this idea, confocal microscopy experiments at various time points after L-LTP induction showed that proteasome inhibition causes accumulation of translational activators eukaryotic initiation factors 4E (eIF4E) and eukaryotic elongation factor 1A (eEF1A) early during L-LTP (Dong et al., 2014). Translational repressors such as polyadenylate-binding protein interacting protein 2 (Paip2) and eukaryotic initiation factor 4E-binding protein 2 (4E-BP2) buildup at later stages of L-LTP in response to proteasome inhibition (Dong et al., 2014). Other negative regulators of translational repressors such as Mov10 might be stabilized by proteasome inhibition as well. For example, in cultured hippocampal neurons Mov10, which inhibits translation of key plasticity-related mRNAs such as that of *CaMKII*α*,* is degraded by the proteasome in an NMDA- and activity-dependent manner (Banerjee et al., 2009).

Other studies have investigated the effect of proteasome inhibition on LTP. These studies failed to discern differential roles of the proteasome in LTP because one investigation used MG-132 (Karpova et al., 2006) which is not a highly specific proteasome inhibitor (Chain et al., 1999; Tang and Leppla, 1999) and the other utilized proteasome inhibitors lactacystin and epoxomicin at nanomolar concentration (Fonseca et al., 2006) which is substantially lower than the effective concentration (micromolar) essential to block proteasome activity.

## **FUTURE DIRECTIONS**

The evidence accumulated over the past decade has further supported the importance of local protein degradation in synaptic plasticity during brain development as well as in the adult brain. What has been lacking is research on the possible mechanisms by which local proteolysis is regulated in neurons. Looking ahead, we can expect to see exciting new discoveries on the local roles of protein degradation in the normal nervous system as well as in many neurodegenerative diseases.

# **ACKNOWLEDGMENT**

The research in the laboratory of Ashok N. Hegde is supported by grants from National Institutes of Health (NS066583; AG040975).

# **REFERENCES**


autoubiquitylation and cytoplasmic body formation. *J. Cell Sci.* 126, 2014–2026. doi: 10.1242/jcs.122069


**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.

*Received: 25 September 2014; accepted: 14 November 2014; published online: 01 December 2014.*

*Citation: Hegde AN, Haynes KA, Bach SV and Beckelman BC (2014) Local ubiquitinproteasome-mediated proteolysis and long-term synaptic plasticity. Front. Mol. Neurosci. 7:96. doi: 10.3389/fnmol.2014.00096*

*This article was submitted to the journal Frontiers in Molecular Neuroscience.*

*Copyright © 2014 Hegde, Haynes, Bach and Beckelman. 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 andthatthe original publication inthis journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Protein degradation and protein synthesis in long-term memory formation

### *Timothy J. Jarome1,2 and Fred J. Helmstetter <sup>2</sup> \**

*<sup>1</sup> Department of Neurobiology, University of Alabama at Birmingham, Birmingham, AL, USA*

*<sup>2</sup> Department of Psychology, University of Wisconsin-Milwaukee, Milwaukee, WI, USA*

### *Edited by:*

*Ashok Hegde, Wake Forest School of Medicine, USA*

### *Reviewed by:*

*Jorge Medina, Universidad de Buenos Aires, Argentina Diasynou Fioravante, University of California, Davis, USA*

### *\*Correspondence:*

*Fred J. Helmstetter, Department of Psychology, University of Wisconsin-Milwaukee, Garland Hall, PO Box 413, Milwaukee, WI 53201, USA e-mail: fjh@uwm.edu*

Long-term memory (LTM) formation requires transient changes in the activity of intracellular signaling cascades that are thought to regulate new gene transcription and *de novo* protein synthesis in the brain. Consistent with this, protein synthesis inhibitors impair LTM for a variety of behavioral tasks when infused into the brain around the time of training or following memory retrieval, suggesting that protein synthesis is a critical step in LTM storage in the brain. However, evidence suggests that protein degradation mediated by the ubiquitin-proteasome system (UPS) may also be a critical regulator of LTM formation and stability following retrieval. This requirement for increased protein degradation has been shown in the same brain regions in which protein synthesis is required for LTM storage. Additionally, increases in the phosphorylation of proteins involved in translational control parallel increases in protein polyubiquitination and the increased demand for protein degradation is regulated by intracellular signaling molecules thought to regulate protein synthesis during LTM formation. In some cases inhibiting proteasome activity can rescue memory impairments that result from pharmacological blockade of protein synthesis, suggesting that protein degradation may control the requirement for protein synthesis during the memory storage process. Results such as these suggest that protein degradation and synthesis are both critical for LTM formation and may interact to properly "consolidate" and store memories in the brain. Here, we review the evidence implicating protein synthesis and degradation in LTM storage and highlight the areas of overlap between these two opposing processes. We also discuss evidence suggesting these two processes may interact to properly form and store memories. LTM storage likely requires a coordinated regulation between protein degradation and synthesis at multiple sites in the mammalian brain.

### **Keywords: ubiquitin, proteasome, fear conditioning, protein degradation, protein synthesis, amygdala, hippocampus**

# **INTRODUCTION**

The ubiquitin-proteasome system (UPS) is a complex network of different ubiquitin ligases and interconnected protein structures involved in the regulation of protein degradation in neurons. This system has been reviewed extensively by others (Hegde, 2010; Mabb and Ehlers, 2010; Bingol and Sheng, 2011), but in general proteins become targeted for degradation through a series of steps in which the small protein modifier ubiquitin is covalently bound to a target substrate. The substrate can acquire anywhere from 1 to 7 ubiquitin modifiers, which link together at specific lysine residues forming polyubiquitin chains. In general, longer ubiquitin chains and lysine-48 linkage provide the maximal signal for degradation (Fioravante and Byrne, 2011) while lysine-63 and M1 linkage (linear ubiquitination) often target substrates for other non-proteolytic functions (Rieser et al., 2013). The ubiquitination process determines what proteins will be targeted for degradation by the proteasome, but the proteasome ultimately controls which of these polyubiquitinated substrates will be degraded.

The catalytic structure in the UPS is the 26S proteasome, which consists of a core (20S) and two regulatory particles (19S) (Bedford et al., 2010). The 20S proteasome is the catalytic core of the proteasome structure and it possesses three types of proteolytic activity controlled by the β-subunits. The activity of the 20S core is regulated by the 19S regulatory particles, which contain the only six ATP-sensitive subunits of the proteasome known as the Rpt subunits. The Rpt6 subunit has received the most attention as it has been shown that increases in its phosphorylation regulates increases in proteasome activity *in vitro* (Bingol et al., 2010; Djakovic et al., 2012) and correlates with increased proteasome activity *in vivo* (Jarome et al., 2013), suggesting that phosphorylation of Rpt6 (at Serine-120) may be the primary regulator of activity-dependent changes in proteasome activity in the brain. Additionally, the 19S proteasome contains numerous deubiquitinating enzymes which generally facilitate the degradation process by removing ubiquitin moieties as the substrate enters the proteasome, thus maintaining the ubiquitin pool (Kowalski and Juo, 2012). However, some deubiquitinating enzymes, such as the ubiquitin-specific protease 14 (USP14), actually seem to inhibit the degradation of certain substrates (Lee et al., 2010; Jin et al., 2012). This suggests that not only does the proteasome degrade polyubiquitinated substrates, but it can actually determine which of these substrates will ultimately be degraded.

In recent years numerous studies have suggested a role for the proteolytic activity of the UPS in activity-dependent synaptic plasticity. For example, bidirectional activity-dependent homeostatic scaling requires UPS-mediated protein degradation (Ehlers, 2003). Interestingly, this proteasome-dependent homeostatic scaling is largely regulated by phosphorylation of the Rpt6 subunit at Serine-120 (Rpt6-S120) (Djakovic et al., 2012) which enhances proteasome activity (Djakovic et al., 2009), suggesting that Rpt6-mediated increases in proteasome activity are critical for activity-dependent synaptic plasticity. Consistent with this, protein degradation is involved in new dendritic spine growth that is regulated by phosphorylation of Rpt6-S120 (Hamilton et al., 2012; Hamilton and Zito, 2013). Additionally, proteasome inhibitors alter long-term potentiation (LTP) in the hippocampus (Fonseca et al., 2006; Dong et al., 2008) and long-term facilitation (LTF) in *Aplysia* (Chain et al., 1999; Lee et al., 2012), suggesting that protein degradation is critical for various forms of synaptic plasticity.

Recently, attention has turned to the potential role of protein degradation in learning-dependent synaptic plasticity. Indeed, there is now convincing evidence that UPS-mediated protein degradation is likely involved in various different stages of memory storage. However, while some studies have suggested potential roles for protein degradation in long-term memory (LTM) formation and storage (Kaang and Choi, 2012), one intriguing question is whether protein degradation is linked to the wellknown transcriptional and translational alterations thought to be critical for memory storage in the brain (Johansen et al., 2011). Here, we discuss evidence demonstrating a role for protein degradation and synthesis in the long-term storage of memories in the mammalian brain, highlighting instances in which a requirement for protein degradation correlates with a requirement for protein synthesis. Additionally, we discuss evidence suggesting that both protein degradation and synthesis may be regulated by CaMKII signaling during LTM formation. Collectively, we propose that LTM storage requires coordinated changes in protein degradation and synthesis in the brain, which may be primarily controlled through a CaMKII-dependent mechanism.

# **MEMORY PARADIGMS**

A variety of different rodent behavioral paradigms have been used to study the molecular neurobiology of LTM formation, and these behavioral tasks often result in alterations in synaptic plasticity in many different brain regions. This review will focus mostly on fear-related conditioning paradigms, but will also discuss results from the Morris water maze (MWM) and the object recognition paradigm. One of the most common rodent behavioral procedures is Pavlovian fear conditioning, in which a neutral conditional stimulus (CS) becomes associated with a noxious or aversive unconditional stimulus (UCS). As a result of this association, the CS can elicit an emotional response based on the memory of the UCS. A simple form of Pavlovian fear

conditioning is contextual fear conditioning, in which rodents learn to fear the specific context or training environment in which the UCS occurred. Memories in this paradigm require an intact and active amygdala and hippocampus for their formation and long-term storage (Kim and Fanselow, 1992; Phillips and Ledoux, 1992). Auditory delay fear conditioning involves a discrete auditory cue (CS) that coterminates with the UCS. Memories formed in this paradigm require an intact and active amygdala, but unlike contextual fear conditioning, do not require the hippocampus (Helmstetter, 1992a,b; Phillips and Ledoux, 1992; Ledoux, 2000). A more complex form of fear conditioning is auditory trace fear conditioning in which the auditory CS predicts the UCS but the two stimuli are separated in time. The introduction of this "trace interval" recruits the prefrontal cortex (Gilmartin and McEchron, 2005b; Gilmartin et al., 2013b). Trace fear conditioning also requires the hippocampus and amygdala (Gilmartin and McEchron, 2005a; Kwapis et al., 2011; Gilmartin et al., 2012). Fear conditioning can be measured in several ways including behavioral observation of freezing (e.g., Helmstetter, 1992a,b) or CS modulation of reflex responses (Hitchcock and Davis, 1991; Rosen et al., 1991). Inhibitory avoidance is another popular aversive learning procedure. In this task, rodents are placed into the lit (white) compartment of a black-white shuttle box. Once the partition is opened, the animal will go to the dark (black) side of the box where it will receive a shock. Animals will then learn to avoid the dark compartment to prevent receiving the shock again. Memories formed using this paradigm require the amygdala and hippocampus for their long-term storage (Taubenfeld et al., 2001; Milekic et al., 2007). Finally, another form of aversive classical conditioning is conditioned taste aversion, in which rodents will acquire an aversion to a specific food due to experiencing illness associated with it. Memories formed using this paradigm require the amygdala and insular cortex for their longterm storage (Rodriguez-Ortiz et al., 2011). In general, memories formed through aversive conditioning require the amygdala but the contribution of the hippocampus, prefrontal cortex and insular cortex depend on the specific behavioral paradigm used.

Two of the most common non-shock based spatial paradigms used to study LTM formation are the MWM and object recognition procedures. In the MWM, a spatial paradigm that can also be stressful and aversive (D'Hooge and De Deyn, 2001), rodents are typically placed into a pool where a hidden platform is positioned in one of four quadrants, each of which has specific spatial cues surrounding it. The animal is given several trials to learn where the platform is, and its latency to swim to the platform will decrease as it learns the task. On the probe day, the platform is removed and the amount of time the animal spends searching each quadrant is measured. If the animal learned the task, it will spend a majority of its time searching the target quadrant where the platform had been during training. Due to the spatial nature of the task, memories acquired using this paradigm require the hippocampus (Artinian et al., 2008). In the objection recognition paradigm, a non-aversive form of spatial learning, rodents are allowed to explore two objects for a period of time. At later test one of the objects is replaced with a new object and the amount of time the animal spends exploring the objects is recorded Object memory is indicated by more time spent exploring the novel object during the test phase. As in the MWM, memories formed in this paradigm require the hippocampus for their formation and long-term storage (Rossato et al., 2007).

# **THE ROLE OF PROTEIN SYNTHESIS IN MEMORY STORAGE**

For several decades there has been a general consensus that *de novo* protein synthesis is critical for the formation and stability of LTM (Davis and Squire, 1984; Helmstetter et al., 2008; Johansen et al., 2011). Consistent with this, numerous studies using pharmacological, molecular, and genetic approaches have implicated a role for increased transcriptional and translational regulation in various brain regions during LTM formation for a variety of different behavioral tasks. **Table 1** summarizes some the findings for inhibitors of gene transcription and protein synthesis on LTM formation and stability following retrieval. Some of the first evidence suggesting that new protein synthesis may be necessary for LTM after fear conditioning came from a study showing that infusions of a broad spectrum mRNA synthesis inhibitor in the amygdala impaired memory for auditory and contextual fear conditioning (Bailey et al., 1999). Consistent with this, infusions of broad spectrum inhibitors of gene transcription or protein synthesis into the amygdala can impair LTM for auditory delay fear conditioning, auditory trace fear conditioning, contextual fear conditioning, fear potentiated startle, inhibitory avoidance, and conditioned taste aversion memories (Schafe and Ledoux, 2000; Bahar et al., 2003; Lin et al., 2003; Yeh et al., 2006; Milekic et al., 2007; Jarome et al., 2011; Kwapis et al., 2011), suggesting that increased protein synthesis is critical for the storage of fear memories in the amygdala (Hoeffer et al., 2011). Inhibiting protein synthesis in the hippocampus impairs LTM for contextual fear conditioning, MWM spatial memories, inhibitory avoidance memories, and object recognition memories (Bourtchouladze et al., 1998; Taubenfeld et al., 2001; Rossato et al., 2007; Artinian et al., 2008), while inhibiting protein synthesis impairs LTM for conditioned taste aversion memories in the insular cortex and trace fear memories in the medial prefrontal cortex (Blum et al., 2006; Moguel-Gonzalez et al., 2008; Reis et al., 2013). Additionally, infusions of more selective inhibitors of protein synthesis which block mTOR-mediated translation impair fear memory and object recognition memory formation in the amygdala and hippocampus, supporting that *de novo* translation is critical for LTM formation in these regions (Parsons et al.,

### **Table 1 | The role of protein synthesis in memory consolidation and reconsolidation.**


?*,Denotes that the role of protein synthesis has not been tested for this type of memory during the indicated stage of memory storage.*

2006b; Gafford et al., 2011; Jobim et al., 2012a,b). Collectively, these results suggest new protein synthesis is required for the processes of memory consolidation.

In addition to a role for new protein synthesis in LTM consolidation, numerous studies have shown that the use or retrieval of an established memory results in a second phase of increased protein synthesis, a process referred to as memory reconsolidation. Protein synthesis inhibitors infused into the amygdala can impair the reconsolidation of auditory delay fear memories, contextual fear memories, inhibitory avoidance memories, and conditioned taste aversion memories (Nader et al., 2000; Parsons et al., 2006a; Milekic et al., 2007; Jarome et al., 2011, 2012; Rodriguez-Ortiz et al., 2012). Additionally, inhibiting protein synthesis in the hippocampus impairs the reconsolidation of contextual fear memories (Debiec et al., 2002; Lee et al., 2008; Gafford et al., 2011), MWM spatial memories (Artinian et al., 2008), and object recognition memories (Rossato et al., 2007), though it has no effect on inhibitory avoidance memories in the hippocampus (Taubenfeld et al., 2001) or conditioned taste aversion memories in the insular cortex (Garcia-DeLaTorre et al., 2009). These results indicate that new protein synthesis is a necessary step in the transfer of a retrieved memory back to long-term storage, suggesting that both the consolidation and reconsolidation of memories requires *de novo* protein translation in several brain regions.

Consistent with the evidence from the broad spectrum inhibitors, several studies have implicated a role for specific intracellular signaling molecules thought to be "upstream" of protein synthesis in LTM formation and storage (Johansen et al., 2011). For example, inhibition of NMDA receptor (NMDAR) function impairs the consolidation of auditory delay fear and contextual fear memories, fear potentiated startle, and conditioned taste aversion memories in the amygdala (Walker and Davis, 2000; Yasoshima et al., 2000; Rodrigues et al., 2001), auditory trace and contextual fear memories in the prefrontal cortex (Gilmartin and Helmstetter, 2010; Gilmartin et al., 2013a), contextual fear memories, MWM and objection recognition spatial memories in the hippocampus (Liang et al., 1994; Izquierdo et al., 1999; Czerniawski et al., 2012; Da Silva et al., 2013; Warburton et al., 2013), and conditioned taste aversion memories in the insular cortex (Escobar et al., 1998). Inhibition of signaling molecules thought to be downstream of NMDAR activity but upstream of protein synthesis such as Protein Kinase A, ERK-MAPK, and CaMKII impairs memory consolidation for auditory fear memories, contextual fear memories, inhibitory avoidance memories, MWM spatial memories, and conditioned taste aversion memories (Schafe and Ledoux, 2000; Schafe et al., 2000; Sacchetti et al., 2001; Koh et al., 2002; Quevedo et al., 2004; Rodrigues et al., 2004; Leon et al., 2010; Ota et al., 2010; Chen et al., 2012; Halt et al., 2012). Many of these signaling molecules are thought to regulate the transient changes in gene expression hypothesized to be critical for LTM formation in the brain. Consistent with this, epigenetic modifications such as acetylation, phosphorylation, and methylation of histones and DNA methylation are critical for LTM formation in the amygdala and hippocampus (Levenson et al., 2004; Lubin et al., 2008; Gupta et al., 2010; Maddox and Schafe, 2011; Jarome and Lubin, 2013), suggesting that LTM formation requires dynamic changes in gene transcription and protein translation in multiple regions of the brain.

# **THE ROLE OF PROTEIN DEGRADATION IN MEMORY STORAGE**

The idea that protein degradation could contribute to activitydependent synaptic plasticity was first identified more than a decade ago by a study demonstrating that the induction of LTF in *Aplysia* resulted in increased expression of *Ap-uch*, which encodes a ubiquitin C-terminal hydrolase, and that a loss of this gene impaired LTF (Hegde et al., 1997). Additionally, application of the proteasome inhibitor lactacystin impairs LTF (Chain et al., 1999), suggesting that functional proteasome activity is critical for synaptic plasticity. Consistent with this identified role of ubiquitin-proteasome mediated protein degradation in activity-dependent synaptic plasticity, homeostatic changes in synaptic strength that result from chronic stimulation or inhibition of cultured hippocampal neurons requires activity of the UPS (Ehlers, 2003). This activity-dependent homeostatic scaling requires phosphorylation of the proteasome subunit Rpt6 at Serine-120, a CaMKII target site (Djakovic et al., 2009; Bingol et al., 2010). Remarkably, enhancements in Rpt6 phosphorylation is sufficient to drive long-term changes in synaptic strength (Djakovic et al., 2012) and new dendritic spine growth *in vitro* (Hamilton et al., 2012), suggesting that protein degradation is a critical regulator of synaptic plasticity. Furthermore, numerous studies have indicated a role for protein degradation in LTP, the proposed cellular analog of memory (for review, see Hegde, 2010). For example, inhibiting protein degradation or protein synthesis individually impairs late-LTP (Fonseca et al., 2006), suggesting that protein degradation is critical for maintaining LTP following its induction. Interestingly, proteasome inhibitors can also enhance the induction of LTP (Dong et al., 2008), an effect that is largely due to the proteasomes targeting of translational activators during LTP induction (Dong et al., 2014). Thus, it is well-established that protein degradation is a critical regulator of synaptic plasticity *in vitro*.

While not considered in traditional memory models, the theory that protein degradation accompanies changes in protein synthesis during LTM formation has become increasingly popular. Indeed, numerous studies now clearly demonstrate a role for UPS-mediated protein degradation in LTM formation and storage in neurons (Felsenberg et al., 2012; Kaang and Choi, 2012; Jarome and Helmstetter, 2013). Interestingly, strong evidence suggests that changes in protein degradation are correlated with changes in protein synthesis in specific brain regions during both memory consolidation and reconsolidation. **Table 2** summarizes the findings for genetic and pharmacological manipulations of ubiquitin-proteasome activity on LTM formation and stability following retrieval. Proteasome inhibitors infused in the amygdala alter LTM for contextual and auditory fear conditioning (Jarome et al., 2011), fear potentiated startle (Yeh et al., 2006) but not conditioned taste aversion memories (Rodriguez-Ortiz et al., 2011). Additionally, proteasome inhibitors impair MWM and inhibitory avoidance memories in the hippocampus (Lopez-Salon et al., 2001; Artinian et al., 2008), trace fear memories in the prefrontal cortex (Reis et al., 2013), and conditioned taste


?*,Denotes that the role of protein degradation has not been tested for this type of memory during the indicated stage of memory storage.*

aversion memories in the insular cortex (Rodriguez-Ortiz et al., 2011). These results suggest a strong overlap between the protein degradation and synthesis processes during LTM formation. Consistent with this, fear conditioning simultaneously increases protein polyubiquitination and mTOR phosphorylation in the amygdala 1 h after behavioral training (Jarome et al., 2011), a time when protein synthesis is increased in the amygdala (Hoeffer et al., 2011), suggesting a potential overlap between protein degradation and synthesis processes. However, inhibiting proteasome activity in the hippocampus does not alter LTM for contextual fear conditioning. Consistent with this finding, a genetic loss of a ubiquitin E3 ligase alters memory for amygdala but not hippocampus dependent fear memories (Pick et al., 2013a,b), suggesting that there is not a perfect overlap between the protein degradation and synthesis processes during LTM formation. In general, these results do suggest though that protein degradation and protein synthesis are both critical for LTM formation in the same brain regions and they likely overlap in time following behavioral training.

Protein degradation is also critical for the reconsolidation of fear memories, though much less is known about the role of the UPS in this stage of memory storage. For example, proteasome inhibitors alter the reconsolidation of auditory and contextual fear memories in the amygdala (Jarome et al., 2011), contextual fear memories and MWM memories in the hippocampus (Artinian et al., 2008; Lee, 2008, 2010; Lee et al., 2008) and conditioned taste aversion memories in the insular cortex (Rodriguez-Ortiz et al., 2011), though it is currently unknown if protein degradation regulates the reconsolidation of trace fear memories, inhibitory avoidance memories, and object recognition memories in the brain. Despite this, there does appear to be a clear overlap between the protein degradation and protein synthesis processes during memory reconsolidation as manipulation of protein degradation in the amygdala and hippocampus prevent the effectiveness of protein synthesis inhibitors at disrupting LTM for fear conditioning tasks. Collectively, these results suggest a strong correlation between protein degradation and protein synthesis during the reconsolidation process.

In addition to the protein degradation function, nonproteolytic functions of the UPS are also critical for LTM and correlate with the demand for increased protein synthesis. For example, non-proteolytic monoubiquitination of the cytoplasmic polyadenylation element binding protein 3 (CPEB3) by the E3 ligase Neuralized1 is critical for the consolidation of hippocampus dependent memories (Pavlopoulos et al., 2011). Interestingly, monoubiquitination of CPEB3 by Neuralized1 was critical for activity dependent increases in GluR1 and GluR2, suggesting that non-proteolytic ubiquitination regulates protein synthesis during LTM formation. Additionally, recent evidence suggests that the USP14 is a critical regulator of LTM formation in the amygdala (Jarome et al., 2014). USP14 acts as a negative regulator of protein turnover (Lee et al., 2010) and regulates presynaptic plasticity *in vitro* (Wilson et al., 2002; Walters et al., 2008; Bhattacharyya et al., 2012), suggesting that USP14 regulates memory formation through a mechanism independent of protein degradation. These studies highlight that both proteolytic and non-proteolytic functions of the UPS regulate LTM formation in the amygdala and hippocampus, suggesting that overall changes in ubiquitinproteasome activity correlates with the increased demand for *de novo* protein synthesis during memory consolidation.

# **THE LINKING OF PROTEIN DEGRADATION AND SYNTHESIS THROUGH NMDA-CaMKII SIGNALING**

In addition to the potential overlap between protein degradation and synthesis in the brain during LTM formation, both of these mechanisms are likely regulated by a similar signaling pathway: NMDA receptor (NMDAR) dependent changes in CaMKII activity. As stated above, similar to protein degradation and synthesis, NMDAR activity at the time of training has been shown to be critical for the formation of different types of memories in various brain regions. NMDAR activity is thought to regulate the changes in protein synthesis necessary for LTM formation by altering the activity of a number of downstream signaling pathways (Johansen et al., 2011; Jarome and Helmstetter, 2013). One such signaling molecule is CaMKII which is thought to be involved in changes in gene transcription and protein synthesis through its regulation of CREB (Wayman et al., 2008), a critical regulator of LTM formation and stability in the neurons (Josselyn et al., 2001; Han et al., 2007, 2008, 2009). Indeed, numerous studies have suggested that CaMKII is critical for LTM formation and that this requirement for CaMKII signaling overlaps with the requirement for protein degradation and synthesis during memory consolidation (Barros et al., 1999; Rodrigues et al., 2004; Von Hertzen and Giese, 2005; Halt et al., 2012; Da Silva et al., 2013). Collectively, these results suggest that changes in NMDAR and CaMKII activity correlate with changes in both protein degradation and protein synthesis during LTM formation in neurons.

In addition to overlapping with an increased need for protein degradation, recent evidence suggests that NMDAR and CAMKII activity can regulate UPS-mediated protein degradation during LTM in neurons (Jarome et al., 2011, 2013). Inhibiting the activity of NR2B-containing NMDARs, a manipulation which impairs LTM (Rodrigues et al., 2001), prevents learning-dependent increases in degradation-specific polyubiquitination in the amygdala (Jarome et al., 2011). This supports the previously identified *in vitro* relationship between NMDAR and ubiquitin-proteasome activity (Bingol and Schuman, 2006; Bingol et al., 2010) and suggests that NMDAR activity regulates changes in protein degradation in neurons during memory consolidation. Additionally, inhibiting CaMKII signaling in the amygdala during LTM formation prevents learning-induced increases in proteasome activity (Jarome et al., 2013), suggesting that CaMKII signaling is critical for changes in protein degradation in neurons during memory consolidation. Interestingly, pharmacological manipulation of CaMKII also prevented learning-induced increases in the phosphorylation of the proteasome regulatory subunit Rpt6, which is critical for changes in proteasome activity, synaptic plasticity and dendritic spine growth *in vitro* (Djakovic et al., 2009, 2012; Hamilton et al., 2012), suggesting that CaMKII may regulate protein degradation through its actions on the proteasome ATPase subunits. Importantly, manipulation of protein kinase A (PKA), another NMDAR-dependent signaling molecule that is critical for LTM formation in neurons (Schafe and Ledoux, 2000; Tronson et al., 2006), did not alter the changes in proteasome phosphorylation and activity during memory consolidation. This demonstrates that PKA, which can regulate protein degradation *in vitro* (Upadhya et al., 2006; Zhang et al., 2007), does not regulate protein degradation during memory formation and suggests that not all NMDAR-dependent signaling pathways regulate ubiquitin-proteasome activity during LTM formation. Collectively, these results suggest that NMDA-CaMKII-dependent changes in ubiquitin-proteasome mediated protein degradation are critical for LTM formation in neurons, suggesting that protein degradation and synthesis could be linked during memory formation by CaMKII signaling, however, it is still unknown if CaMKII actually does regulate protein synthesis during memory formation.

# **WHAT COMES FIRST, DEGRADATION OR SYNTHESIS?**

A majority of the studies discussed here reveal a strong correlation between protein degradation and synthesis during LTM formation. This leads to one important question: Which comes first? While the exact relationship between protein degradation and synthesis during memory formation currently remains equivocal, the available evidence suggests that protein degradation likely regulates protein synthesis. For example, fear conditioning leads to an increase in polyubiquitinated proteins being targeted for degradation by the proteasome (Jarome et al., 2011). While a majority of the proteins being targeted by the proteasome for degradation remain unknown, the RNAi-induced Silencing Complex (RISC) factor MOV10 has been identified as a target of the proteasome during increases in activity-dependent protein degradation *in vitro* (Banerjee et al., 2009) and following behavioral training and retrieval *in vivo* (Jarome et al., 2011). Increases in the degradation of MOV10 are associated with increased protein synthesis *in vitro*, suggesting that the proteasome could regulate protein synthesis during LTM formation through the removal of translational repressor proteins such as various RISC factors. However, it is currently unknown if the selective degradation of MOV10, or any RISC factor, is critical for memory formation in neurons. Nonetheless, studies such as these provide indirect evidence that protein degradation by the UPS could regulate protein synthesis during memory formation in the brain.

Some of the best evidence that protein degradation may be upstream of protein synthesis during memory storage comes from studies examining memory reconsolidation following retrieval. For example, inhibiting proteasome activity can prevent the memory impairments that normally result from post-retrieval blockade of protein synthesis in the hippocampus, amygdala, and nucleus accumbens (Lee et al., 2008; Jarome et al., 2011; Ren et al., 2013) as well as during LTF in aplysia (Lee et al., 2012), suggesting that protein degradation is upstream of protein synthesis during memory reconsolidation. This remains some of the best evidence directly linking protein degradation to protein synthesis during memory storage, but it is possible that the rescue of memory impairments in the face of protein synthesis inhibition may occur as an indirect consequence of blocking protein degradation rather than a direct effector. Additionally, some studies find that proteasome inhibitors impair memory reconsolidation when administered on their own (Artinian et al., 2008; Rodriguez-Ortiz et al., 2011), suggesting that this relationship between protein degradation and synthesis may not exist for all types of memories. However, in cell cultures protein degradation has been shown to regulate mTOR activity (Ghosh et al., 2008), a translational control pathway critical for memory formation and storage (Parsons et al., 2006b; Gafford et al., 2011), though it is unknown if this relationship exists *in vivo* during memory formation. Collectively, the current evidence suggests that protein degradation may determine the requirement for protein synthesis during memory storage, though this relationship has never been directly proven.

While it is currently not known if protein degradation regulates protein synthesis during memory formation, one recent study found that protein degradation regulates protein synthesis during LTP, a proposed cellular analog of memory (Dong et al., 2014). Inhibiting proteasome activity enhances the induction but impairs the maintenance of LTP (Dong et al., 2008). Surprisingly, the enhanced induction of LTP following proteasome inhibition can be blocked by inhibiting the activity of several proteins involved in mTOR-mediated protein synthesis, suggesting that the proteasome targets translational activators during LTP induction. However, proteasome inhibitors cause an increase in translational repressors during L-LTP, suggesting that the proteasome acts to enhance protein synthesis during LTP maintenance. Collectively, these results demonstrate for the first time that protein degradation can directly regulate protein synthesis during activity-dependent synaptic plasticity. While it is currently unknown if the proteasome targets similar proteins during memory formation, this study provides some of the best evidence to date that protein degradation may regulate protein synthesis during learning-dependent synaptic plasticity.

# **REGULATION OF MEMORY STORAGE BY PROTEIN DEGRADATION AND SYNTHESIS**

While the functional significance of the observed overlap between protein degradation and synthesis during LTM formation is unknown, it is possible that a coordinated balance between protein degradation and synthesis may be necessary for memory formation and storage in neurons (Jarome and Helmstetter, 2013). This theory has been described extensively elsewhere, but in this model CaMKII-dependent increases in proteasome activity leads to reductions in a number of different proteins that normally repress transcription and translation or prevent alterations to the postsynaptic structure. Consistent with this, the proteasome targets the RISC factor MOV10 and synaptic scaffold Shank following behavioral training and memory retrieval (Lee et al., 2008; Jarome et al., 2011), and inhibiting proteasome activity can prevent learning-induced changes in GluR2 expression at synapses (Ren et al., 2013). Thus, the changes in protein degradation could be necessary to remove repressor proteins that prevent the dynamic changes to the postsynaptic structure that characterize memory storage in neurons (Ostroff et al., 2010) while protein synthesis could produce the proteins that are necessary for these changes. In addition, an attractive new addition to this model is that protein degradation and synthesis could regulate dynamic changes in the atypical protein kinase C isoform PKMζ during memory consolidation. A majority of studies have shown that inhibiting PKMζ impairs LTM even after the consolidation process has completed (Pastalkova et al., 2006; Shema et al., 2007; Kwapis et al., 2009, 2012), though a few exceptions have been noted (Parsons and Davis, 2011; Volk et al., 2013), suggesting that PKMζ regulates the maintenance of memories in neurons (Sacktor, 2012; Kwapis and Helmstetter, 2013). Interestingly, one recent study (Vogt-Eisele et al., 2014) found that protein degradation can regulate PKMζ levels during memory storage through targeting of KIBRA (Kidney/BRAin protein). Since behavioral training is thought to increase PKMζ levels in the brain (Sacktor, 2012), this suggests that a balance between protein degradation and synthesis may control LTM formation and storage through regulation of PKMζ levels. Whether a balance between protein degradation and synthesis is necessary for learningdependent changes in PKMζ levels will be of interest in future studies.

# **PROTEIN DEGRADATION AND SYNTHESIS AS INDEPENDENT PROCESSES DURING MEMORY STORAGE**

While the evidence discussed here suggests that protein degradation and protein synthesis may be interconnected processes necessary for memory formation and storage, an alternative theory is that these are actually opposing processes and that memory formation requires an appropriate balance between them. For example, both proteasome and protein synthesis inhibitors impair LTP when applied individually, but actually rescue these deficits when applied simultaneously (Fonseca et al., 2006). This would suggest that memory impairments that result from blocking protein degradation or synthesis individually are caused by an inappropriate balance between the two processes. While an intriguing theory, few studies have directly tested this *in vivo* during memory formation as most times proteasome inhibitors are given alone and not in combination with protein synthesis inhibitors. Despite this, some studies suggest that an altered balance between protein degradation and synthesis cannot account for memory impairments following disruption of the UPS. For example, proteasome inhibitors infused into the amygdala result in similar memory impairments for a fear conditioning task as those produced by the broad spectrum protein synthesis inhibitor anisomycin and simultaneous infusion of both inhibitors does not rescue these impairments (Jarome et al., 2011). This suggests that an inappropriate balance between protein degradation and synthesis likely cannot account for memory deficits observed following infusions of either inhibitor individually. However, simultaneous blockade of protein degradation and synthesis can rescue memory impairments that results from blocking protein synthesis individually following memory retrieval (Lee et al., 2008; Jarome et al., 2011), though in these cases proteasome inhibitors have no effect on memory on their own. This would lend more toward the theory that protein degradation is upstream of protein synthesis during memory storage following retrieval. Thus, while the idea that protein degradation and protein synthesis are independent processes and a balance between them is the primary factor that underlies memory formation is intriguing, very few studies have directly tested this to date.

## **SUMMARY AND FUTURE DIRECTIONS**

The studies reviewed here suggest a clear correlation between protein degradation and protein synthesis in LTM formation in the brain. However, we are currently far from understanding the exact relationship between these two opposing processes during in memory. As discussed above, the current evidence suggests that protein degradation and synthesis are both important regulators of LTM for auditory delay and trace fear conditioning, contextual fear conditioning, inhibitory avoidance, conditioned taste aversion, MWM, and object recognition memories and simultaneously occur in multiple brain regions including the amygdala, hippocampus, prefrontal cortex, and insular cortex. However, it is unknown if this relationship holds for all brain regions and behavioral tasks. For example, contextual and trace fear memories also require the retrosplinal cortex for their long-term storage (Kwapis et al., 2014), though it is unknown if protein degradation and protein synthesis are required. Future research should determine which behavioral tasks and brain regions require protein degradation and protein synthesis to better determine the extent of overlap in these processes. More importantly, future research should focus on determining if and how protein degradation regulates activity driven protein synthesis during memory formation in the already identified instances in which these two processes locally correlate. Such information is critical to determine not only if protein degradation and synthesis are dissociable during memory formation, but also what the functional role of the changes in protein degradation is. Furthermore, identifying the targets of the proteasome during increased protein degradation levels will also help determine the functional role of protein degradation during memory formation. Understanding the exact relationship between protein degradation and synthesis is a critical step in understanding the molecular neurobiology of LTM formation and storage in neurons.

Additionally, future studies should focus on how CaMKII signaling regulates increases in proteasome activity during memory formation and whether this occurs through Rpt6 phosphorylation. Considering that CaMKII is the only intracellular signaling molecule known to regulate protein degradation during LTM formation (Jarome et al., 2013) and is thought to regulate protein synthesis, it is critical to understand how changes in CaMKII activity contribute to the appropriate regulation of protein degradation and synthesis during LTM formation in neurons. Thus, CaMKII might serve as an attractive target to better understand the role of protein degradation in LTM formation and its relationship to *de novo* protein synthesis. Finally, future studies should focus on identifying the non-proteolytic functions of the UPS during memory formation and storage. Currently, only two non-proteolytic roles for ubiquitin-proteasome activity exist (Pavlopoulos et al., 2011), though it is likely that many more remain to be discovered. Considering that one of these nonproteolytic functions has been suggested to regulate some protein synthesis, the degradation-independent functions of the UPS may prove to be among the most important regulators of translational regulation by the UPS.

# **CONCLUSIONS**

Here, we reviewed a number of studies suggesting that changes in protein degradation correlate with changes in protein synthesis during memory formation and storage in neurons. Additionally, we discussed evidence demonstrating that protein degradation is regulated by NMDAR activity and the intracellular signaling molecule CaMKII, two well-known regulators of memory formation that are thought to regulate protein synthesis, suggesting that protein degradation and synthesis may be linked through CaMKII signaling. Finally, we reviewed evidence demonstrating that protein degradation may be upstream of protein synthesis during LTM formation, suggesting that protein degradation may supersede protein synthesis during the memory consolidation process. Collectively, the studies outlined in this review suggest that protein degradation and protein synthesis are not likely two independent regulators of LTM formation but rather directly interact to modify active synapses and store memory for a variety of different behavioral tasks.

### **ACKNOWLEDGMENT**

This work was supported by NIMH grants R01-06558 (Fred J. Helmstetter).

# **REFERENCES**


Yeh, S. H., Mao, S. C., Lin, H. C., and Gean, P. W. (2006). Synaptic expression of glutamate receptor after encoding of fear memory in the rat amygdala. *Mol. Pharmacol.* 69, 299–308. doi: 10.1124/mol.105. 017194

Zhang, F., Hu, Y., Huang, P., Toleman, C. A., Paterson, A. J., and Kudlow, J. E. (2007). Proteasome function is regulated by cyclic AMP-dependent protein kinase through phosphorylation of Rpt6. *J. Biol. Chem.* 282, 22460–22471. doi: 10.1074/jbc.M702439200

**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.

*Received: 30 April 2014; paper pending published: 16 May 2014; accepted: 09 June 2014; published online: 26 June 2014.*

*Citation: Jarome TJ and Helmstetter FJ (2014) Protein degradation and protein synthesis in long-term memory formation. Front. Mol. Neurosci. 7:61. doi: 10.3389/fnmol. 2014.00061*

*This article was submitted to the journal Frontiers in Molecular Neuroscience. Copyright © 2014 Jarome and Helmstetter. 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 central role for ubiquitination within a circadian clock protein modification code

# *Katarina Stojkovic1, Simon S. Wing2 and Nicolas Cermakian1\**

<sup>1</sup> Douglas Mental Health University Institute, McGill University, Montréal, QC, Canada

<sup>2</sup> Polypeptide Laboratory, Department of Medicine–McGill University Health Centre Research Institute, McGill University, Montréal, QC, Canada

### *Edited by:*

Ashok Hegde, Wake Forest School of Medicine, USA

### *Reviewed by:*

Jason DeBruyne, Morehouse School of Medicine, USA Kazuhiro Yagita, Kyoto Prefectural University of Medicine, Japan

### *\*Correspondence:*

Nicolas Cermakian, Douglas Mental Health University Institute, McGill University, 6875 LaSalle Boulevard, Montréal, QC H4H 1R3, Canada e-mail: nicolas.cermakian@mcgill.ca Circadian rhythms, endogenous cycles of about 24 h in physiology, are generated by a master clock located in the suprachiasmatic nucleus of the hypothalamus and other clocks located in the brain and peripheral tissues. Circadian disruption is known to increase the incidence of various illnesses, such as mental disorders, metabolic syndrome, and cancer. At the molecular level, periodicity is established by a set of clock genes via autoregulatory translation–transcription feedback loops. This clock mechanism is regulated by post-translational modifications such as phosphorylation and ubiquitination, which set the pace of the clock. Ubiquitination in particular has been found to regulate the stability of core clock components but also other clock protein functions. Mutation of genes encoding ubiquitin ligases can cause either elongation or shortening of the endogenous circadian period. Recent research has also started to uncover roles for deubiquitination in the molecular clockwork. Here, we review the role of the ubiquitin pathway in regulating the circadian clock and we propose that ubiquitination is a key element in a clock protein modification code that orchestrates clock mechanisms and circadian behavior over the daily cycle.

**Keywords: circadian clock, clock gene, ubiquitin, ubiquitin ligase, deubiquitinase, stability**

# **INTRODUCTION: THE MOLECULAR CIRCADIAN CLOCK**

Circadian rhythms are endogenous ∼24 h cycles in physiology and behavior generated by a master clock in the suprachiasmatic nucleus of the hypothalamus, and clocks located in most other tissues. Circadian clocks enable organisms to anticipate predictable daily occurrences, such as changes in light, temperature, or food availability (Dibner et al., 2010). The importance of circadian clocks is illustrated by the impacts of circadian disruption in humans. For example, shift work increases the risk of developing various illnesses, such as mental disorders, metabolic syndrome, and cancer (Evans and Davidson, 2013).

At the molecular level, the circadian clock relies on selfsustained transcription-translation feedback loops involving "clock genes" (**Figure 1**; Duguay and Cermakian, 2009). In mammals, CLOCK and BMAL1 dimerize and activate *Period (Per) 1* and *2* and *Cryptochrome* (*Cry*) *1* and *2* genes. The PER1/2 and CRY1/2 proteins then enter the nucleus and inhibit the activity of CLOCK/BMAL1, thereby repressing their own transcription. However, the mechanism is more complex, with additional interlocking feedback loops, including one that involves the induction of the *Rev-erb* and *Ror* genes, whose protein products regulate *Bmal1* gene transcription. One consequence of these feedback loops is that the mRNAs and proteins of many clock genes present circadian rhythms in their abundance. Moreover, hundreds of clock-controlled genes, which do not participate in the clock mechanism, but whose transcription is under the control of the clock molecular machinery, also present rhythms at the RNA and protein levels (Storch et al., 2002; Yan et al., 2008), thus linking the circadian clock with cellular physiology.

The timing of these feedback loops is dictated by posttranslational modifications (PTMs; Gallego and Virshup, 2007; Duguay and Cermakian, 2009). Indeed, clock proteins are subject to phosphorylation, ubiquitination, acetylation, SUMOylation, and other PTMs (**Figure 1**). Ubiquitination is of particular interest due to the diversity of signals that it can generate. In particular, its direct role in determining protein half-life is crucial for proteins with a daily rhythm in abundance. In this article, we review the current state of knowledge on ubiquitination of clock proteins and their ubiquitin-modifying enzymes in animal models, with a special focus on the mammalian clock (**Table 1**).

### **UBIQUITINATION IN THE CIRCADIAN CLOCK**

### **UBIQUITINATION OF CRYPTOCHROMES BY FBXL UBIQUITIN LIGASES**

*N*-Ethyl-*N*-nitrosourea screens led to the discovery of mice exhibiting free-running periods of locomotor activity rhythms ∼2–3 h longer than normal (Godinho et al., 2007; Siepka et al., 2007). These mice had loss-of-function mutations in the gene encoding the F-box protein FBXL3. Loss of FBXL3 activity leads to CRY protein stabilization due to decreased ubiquitination (Siepka et al., 2007). Further work on FBXL3 revealed that ubiquitination of CRY1/2 by the FBXL3-containing SCF E3 ubiquitin ligase complex was necessary for the timely degradation of the CRY proteins and the reactivation of BMAL1/CLOCK (Busino et al., 2007). A prolonged inhibition of BMAL1/CLOCK-mediated transcription in the mutant mice leads to reduced peak levels and delayed rhythms of the *Per* and *Cry* mRNAs in mutant mouse SCN, cerebellum, and liver (Godinho et al., 2007; Siepka et al., 2007).

Interestingly, FBXL3 cannot undergo SCF complex formation in the absence of its CRY substrates (Yumimoto et al., 2013). X-ray crystallography revealed that FBXL3 binds to the FAD-binding pocket of mammalian CRY, which may also be bound by FAD or PER proteins (Czarna et al., 2013; Xing et al., 2013), which suggests a mechanism for the protection of CRYs from degradation in the presence of PER (Yagita et al., 2002).

expression of Bmal1 (and also Cry1 and Clock). For each protein or pair of proteins, the colored boxes list the post-translational

An FBXL3 paralog, FBXL21, was identified in sheep, where it was also found to bind to CRY1, thereby affecting transcriptional activation by CLOCK/BMAL1 (Dardente et al., 2008). Despite the high similarity between FBXL3 and FBXL21, they appear to have non-redundant roles within the clock. Indeed, while *Fbxl3* gene mutant or knock-out (KO) mice display a longfree-running period of locomotor activity rhythms, *Fbxl21*-mutant or KO mice present either a short (Yoo et al., 2013) or a normal (Hirano et al., 2013) period. Moreover, when the mutant lines are crossed, the *Fbxl21* mutation attenuates the long-period phenotype of *Fbxl3*-mutant mice.

The distinct roles of the FBXL proteins may be based on the timing of their expression and that of their substrates. While *Fbxl3* is expressed at constant levels over the day, *Fbxl21* expression has a pronounced circadian rhythm in the mouse SCN, with a peak by the end of the subjective day (Dardente et al., 2008), thus restricting its action to only part of the cycle. Interestingly, while FBXL3 protein levels do not vary over time, its action on CRYs is conditional on their phosphorylation by AMPK, whose expression and nuclear abundance vary over the day (Lamia et al., 2009). Ligase intracellular localization also plays a role: while FBXL3 protein is restricted to the nucleus, FBXL21 is located both in the nucleus and cytoplasm (Hirano et al., 2013; Yoo et al., 2013). The work of both laboratories supports a two-step mode of action of FBXL21. First, FBXL21 allows CRYs to accumulate in the cytoplasm. This occurs when CRY levels rise around the end of the day or beginning of the night. Shortly thereafter, after CRYs have entered the nucleus, FBXL21 might counteract FBXL3: FBXL21 binds CRYs more stably and with a higher affinity than FBXL3,

ADP-ribosylation; PER, Period; CRY, Cryptochrome; ROR, Retinoic acid receptor-related orphan receptor; DUB, deubiquitinating enzyme.


### **Table 1 | Ubiquitin-modifying enzymes involved in the regulation of mammalian clock proteins.**

suggesting that FBXL21 may in part stabilize CRYs by preventing FBXL3 binding (Yoo et al., 2013). Then, when FBXL21 levels have decreased, FBXL3 can finally act on CRYs and target them to degradation. It is interesting to note that if this model is further confirmed, the roles of FBXL21 in the clock will turn out to be partly non-degradative (regulation of nuclear entry, protection from the action of another F-box protein, FBXL3), in contrast to other ubiquitin ligases involved in the clock, which target clock proteins to proteasomal degradation. Finally, CRY ubiquitination mechanisms might be even more complex, as it was suggested that another ubiquitin ligase might be involved in regulating CRY accumulation (Kurabayashi et al., 2010; Hirano et al., 2013).

### **UBIQUITINATION OF PERIOD PROTEINS BY β-TRCP UBIQUITIN LIGASES**

In *Drosophila*, the F-box component of an SCF ligase, SLIMB, was shown to be critical for ubiquitination and degradation of PER protein over the course of the circadian cycle (Grima et al., 2002; Ko et al., 2002). SLIMB binds to PER after its phosphorylation by Doubletime (DBT), in particular on serine 47 within the SLIMB recognition site (Chiu et al., 2008).

SLIMB has two homologs in mammals, β-TRCP1/FBW1A and β-TRCP2/FBW1B. Similarly to the action of DBT and SLIMB on PER in the fly, β-TRCP1/2 are recruited to PER2 following phosphorylation of this protein by the kinases CK1δ and CK1ε (mammalian homologs of DBT), which leads to polyubiquitination and subsequent degradation of PER2 (Eide et al., 2005). Indeed, expression of a dominant negative form of β-TRCP leads to the inhibition of PER2 ubiquitination and degradation. β-TRCP1/2 also interact with PER1 in a CK1ε-dependent manner and a knockdown of both β-TRCPs was found to stabilize PER1, and reduce levels of transcriptional activation by CLOCK/BMAL1 (Shirogane et al., 2005). Accordingly, preventing the action of β-TRCP1/2 on PER proteins leads to long-period or dampened circadian rhythms in cultured fibroblasts (Reischl et al., 2007; Ohsaki et al., 2008). Surprisingly though, mice lacking β-TRCP1 neither show alteration in circadian locomotor behavior nor differences in SCN PER2 levels when compared to WT controls, suggesting either that the SCN clock behaves differently from clocks in fibroblasts or that there is redundancy at the level of the ubiquitin ligases (Ohsaki et al., 2008). Finally, similarly to PER proteins protecting CRYs (see Ubiquitination of Cryptochromes by FBXL Ubiquitin Ligases), PER proteins are protected from ubiquitination and degradation upon association with CRYs (Yagita et al., 2002).

### **UBIQUITINATION OF REV-ERBα**

The stability of REV-ERBα is also regulated by a sequence of phosphorylation, ubiquitination, and proteasomal degradation. Indeed, REV-ERBα is stabilized following phosphorylation by GSK3β (Yin et al., 2006). Treating cells with lithium, a GSK3β inhibitor, leads to the quick degradation of REV-ERBα and therefore, to increased expression of *Bmal1* (**Figure 1**). Subsequent work identified HUWE1/ARF-BP1 and PAM/MYCBP2 as E3 ligases involved in this lithium-induced REV-ERBα degradation (Yin et al., 2010). Their depletion in cells stabilized REV-ERBα, decreased *Bmal1* gene expression, and disrupted oscillations of other clock genes. HUWE1 and PAM may not be the only ubiquitin ligases acting on REV-ERBα. In *Fbxl3*-mutant mice, REV-ERBα levels are higher and consequently its repression of *Bmal1* and *Cry1* genes is enhanced. Creation of double-mutant *Fbxl3*/*Rev-erb*α−/<sup>−</sup> mice rescues the *Fbxl3*-mutant phenotype (Shi et al., 2013) indicating that FBXL3, in addition to its role on CLOCK/BMAL1-mediated transcription via destabilization of CRYs, also has an effect on REV-ERBα-mediated repression of target genes. Although this effect may be indirect, it does indicate a role for this F-box protein as a coordinator of different clock transcription factors.

### **UBIQUITINATION OF BMAL1**

Many studies have indicated a tight regulation of BMAL1 stability. Indeed, BMAL1 undergoes different phosphorylation events that either target it for ubiquitination and degradation (e.g., GSK3β, Sahar et al., 2010) or on the contrary for deubiquitination and stabilization (e.g., PKCγ, Zhang et al., 2012). Importantly, BMAL1 ubiquitination and proteasome-mediated proteolysis appear to coincide with the time of highest transcriptional activity (Kwon et al., 2006; Lee et al., 2008; Stratmann et al., 2012), whereas in conditions where CLOCK/BMAL1 activity is repressed (e.g., presence of CRYs), BMAL1 is stabilized (Kondratov et al., 2006; Dardente et al., 2007). However, no BMAL1-specific ubiquitin ligase had been uncovered until a recent report, which described UBE3A as an E3 ligase that binds and destabilizes BMAL1 (Gossan et al., 2014). Knockdown of this ligase in mammalian cells and in *Drosophila*

clock neurons leads to a strong dampening of circadian oscillations or even arrhythmicity.

# **SUMOylation IN THE CIRCADIAN CLOCK**

The small ubiquitin-related modifier (SUMO) proteins also play a role in the clock. Like other PTMs, SUMOylation is reversible and the conjugation/deconjugation mechanisms are reminiscent of the ubiquitin pathway (Muller et al., 2001). In contrast to ubiquitination though, SUMOylation does not directly target proteins for degradation but rather regulates other functions such as nuclear localization, protein–protein interactions, transcriptional activity and, interestingly, ubiquitination itself (Desterro et al., 1998; Buschmann et al., 2000).

SUMOylation was first implicated in the clock following the discovery of a SUMOylation consensus motif in BMAL1 (Cardone et al., 2005). Co-expression of BMAL1 and SUMO showed that BMAL1 could indeed be SUMOylated. In the liver, this occurs in a rhythmic manner, with peak SUMOylation in the second half of the light phase. This timing coincides with peak BMAL1 phosphorylation and activity, suggesting an interplay between these PTMs. In further support of this, a functional CLOCK protein is required for both BMAL1 SUMOylation and phosphorylation (Kondratov et al., 2003; Cardone et al., 2005; Dardente et al., 2007). SUMOylated BMAL1 is most abundant when the CLOCK/BMAL1 targets *Dbp* and *Rev-erb*α show their highest mRNA levels, again supporting that SUMOylation of BMAL1 is involved in its transcriptional activity (Lee et al., 2008). Indeed, BMAL1 binding to the *Dbp* promoter was reduced when the lysine required for SUMOylation was mutated (Lee et al., 2008). Interestingly, SUMOylation of BMAL1 is a prerequisite for its subsequent ubiquitination, again highlighting the interplay of different PTMs in the circadian clock.

### **DEUBIQUITINATION IN THE CIRCADIAN CLOCK**

Given the importance of ubiquitination within the clock, it appears reasonable to assume that deubiquitination plays a role as well. Interestingly, the mRNA levels of a deubiquitinating enzyme (DUB), ubiquitin-specific protease 2 (USP2), show rhythmicity in most tissues examined (Kita et al., 2002; Storch et al., 2002; Yan et al., 2008). This is notable, because among the hundreds of clock-controlled transcripts, only a small minority cycles in multiple locations. The circadian rhythm of *Usp2* is blunted in *Clock* mutant and *Bmal1* KO mice (Oishi et al., 2003; Molusky et al., 2012b), and the *Usp2* promoter is activated by CLOCK/BMAL1 (Molusky et al., 2012b), indicating that *Usp2* is a direct target of these transcription factors. In addition to its circadian regulation, *Usp2* expression is also induced by starvation and it was therefore proposed that USP2 integrates nutritional and circadian timing cues (Molusky et al., 2012b). In turn, liver USP2 appears to be involved in the generation of a diurnal rhythm in glucose metabolism (Molusky et al., 2012a).

However, the circadian role of USP2 is not limited to mediating the rhythmic control of cellular processes by the molecular clock. Since the short list of genes rhythmic in multiple tissues is enriched for clock components, USP2 was hypothesized to exert a role within the clock mechanism. To address this, *Usp2* KO mice were generated by two laboratories. In one case, they revealed no alteration of the free-running period of locomotor rhythms (Scoma et al., 2011). In contrast, our *Usp2* KOs display a period longer than WT littermates (Yang et al., 2012), implying a role within the clockwork. In line with this, the absence of USP2 affects the mRNA levels of several clock genes (Scoma et al., 2011; Yang et al., 2012, 2014), and USP2 interacts with clock proteins. In our hands, whereas it forms a complex with several clock proteins, USP2 directly binds only to PER1 (Yang et al., 2012). Accordingly, PER1 is deubiquitinated in the presence of USP2, but notably, this does not lead to PER1 stabilization. Instead, USP2 appears to regulate PER1 intracellular localization (Yang et al., 2014). Interestingly, the only other DUB that to our knowledge has been implicated in clock mechanisms, *Drosophila* USP8, also seems to act in a non-degradative manner: it deubiquitinates CLOCK, thereby inhibiting transcriptional activity of CLOCK/CYCLE (CYCLE is the *Drosophila* homolog of BMAL1; Luo et al., 2012).

In contrast, the work of other groups showed a stabilization of other clock proteins due to deubiquitination by USP2. BMAL1 levels are lower in the SCN of *Usp2* KO mice (Scoma et al., 2011), whereas in cultured cells, USP2 stabilized BMAL1 (Scoma et al., 2011) and reduced its ubiquitination (Lee et al., 2008). Interestingly, a report suggested the involvement of PKCγ-triggered deubiquitination of BMAL1 in the resetting of peripheral clocks by feeding schedules, but the DUB involved in this pathway remains unknown (Zhang et al., 2012). In addition to PER1 and BMAL1, USP2 deubiquitinates CRY1 in cultured cells in response to a serum shock, and in the mouse liver, *Usp2* knockdown increases CRY ubiquitination and decreases CRY1 protein levels (Tong et al., 2012).

Data also support a role for USP2 in the response of the clock to external cues. We found that *Usp2* KO mice exhibit larger phase delays than WT mice after light treatment in the first part of the night, and reduced phase advances, upon light treatment later in the night (Yang et al., 2012). Thus, USP2 appears to be involved in the response of the SCN clock to light, which is also supported by data of Scoma et al. (2011), which show increased phase-shifting in response to low irradiance light in the early night. USP2 may also mediate the response of the clock to inflammation, as the expression of the gene is increased in response to TNFα treatment, and CRY1 protein induction in response to this cytokine is abrogated when *Usp2* expression is knocked down (Tong et al., 2012).

Together, these studies ascribe a pivotal role to USP2, and deubiquitination in general, not only in the circadian clock mechanism, but also as an integrator of environmental and physiological signals, and in output pathways linking the molecular clockwork to cellular and physiological functions.

## **A CLOCK PROTEIN MODIFICATION CODE?**

Overall, the work described above underscores the importance of PTMs within the circadian timing mechanism. Given that different modifications often converge on the same clock protein, we propose the existence of a *clock protein modification code* whereby the fate/function of a given protein is determined by the precise combination and/or the consecutive occurrence of different PTMs. This clock protein modification code is proposed to exist at different levels:


also occurs at this time and leads to increased recruitment of CRY (Hirayama et al., 2007). CRYs themselves are good examples of substrates for sequential PTMs over the 24 h day and across the progression of the clock feedback loop (see Ubiquitination of Cryptochromes by FBXL Ubiquitin Ligases). Therefore, each clock protein undergoes a daily wave of PTMs, in a sequential and often conditional manner, which determines the expression, localization, and activity of the protein and its partners.

# **CONCLUSION**

In conclusion, ubiquitination and deubiquitination are involved in the regulation of key core clock components. On one hand, ubiquitin ligases are selectively acting on one or a few clock proteins. A given clock protein can even be the target of two or three different E3 ligases, depending on the time of day and cellular compartment. On the other hand, DUBs seem less specific, and only one was identified as a mammalian clock component so far: USP2. This DUB regulates the stability and function of PER1, CRY1, BMAL1 and perhaps other clock proteins, as well as components of the input and output pathways of the clock. Moreover, there is a complex interplay of ubiquitination with other PTMs. It will be crucial in future years to precisely define ubiquitin chain configurations and conjugation sites on clock proteins, to unravel the precise regulation of their addition and removal and identify all the actors involved. Furthermore, ubiquitination becomes an attractive drug target. Indeed, recent chemical screens of compounds binding CRY proteins have identified molecules modulating their ubiquitin-induced degradation (Hirota et al., 2012), suggesting the possibility of therapeutic resetting of the circadian clock by drug-mediated ubiquitin modulation of clock components.

### **ACKNOWLEDGMENTS**

The authors thank Dr. Kai-Florian Storch for critical review of the manuscript and all members of Nicolas Cermakian's laboratory for discussions. This work was supported by grants from the Natural Sciences and Engineering Research Council (RGPIN 249731-12) to Nicolas Cermakian and the Canadian Institutes of Health Research (MOP 115106) to Simon S. Wing. Katarina Stojkovic was supported by a fellowship from McGill Faculty of Medicine and Nicolas Cermakian by a salary award from the Fonds de Recherche du Québec—Santé.

## **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.

*Received: 14 May 2014; accepted: 08 July 2014; published online: 07 August 2014. Citation: Stojkovic K, Wing SS and Cermakian N (2014) A central role for ubiquitination within a circadian clock protein modification code. Front. Mol. Neurosci. 7:69. doi: 10.3389/fnmol.2014.00069*

*This article was submitted to the journal Frontiers in Molecular Neuroscience.*

*Copyright © 2014 Stojkovic, Wing and Cermakian. 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 andthatthe original publication inthis journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Divergent tissue and sex effects of rapamycin on the proteasome-chaperone network of old mice

# *Karl A. Rodriguez 1,2, Sherry G. Dodds 1,3, Randy Strong1,4, Veronica Galvan1,2, Z. D. Sharp1,3 and Rochelle Buffenstein1,2\**

*<sup>1</sup> Sam and Ann Barshop Institute for Aging and Longevity Studies, University of Texas Health Science Center at San Antonio, San Antonio, TX, USA*

*<sup>2</sup> Department of Physiology, University of Texas Health Science Center at San Antonio, San Antonio, TX, USA*

*<sup>3</sup> Department of Molecular Medicine, University of Texas Health Science Center at San Antonio, San Antonio, TX, USA*

*<sup>4</sup> Department of Pharmacology, University of Texas Health Science Center at San Antonio, San Antonio, TX, USA*

### *Edited by:*

*Ashok Hegde, Wake Forest School of Medicine, USA*

### *Reviewed by:*

*Dudley Lamming, University of Wisconsin-Madison, USA Sokol V. Todi, Wayne State University School of Medicine, USA Michal Masternak, University of Central Florida, USA*

### *\*Correspondence:*

*Rochelle Buffenstein, Sam and Ann Barshop Institute for Aging and Longevity Studies, University of Texas Health Science Center at San Antonio, 15355 Lamda Drive, San Antonio, TX 78245, USA e-mail: buffenstein@uthscsa.edu*

Rapamycin, an allosteric inhibitor of the mTOR kinase, increases longevity in mice in a sex-specific manner. In contrast to the widely accepted theory that a loss of proteasome activity is detrimental to both life- and healthspan, biochemical studies *in vitro* reveal that rapamycin inhibits 20S proteasome peptidase activity. We tested if this unexpected finding is also evident after chronic rapamycin treatment *in vivo* by measuring peptidase activities for both the 26S and 20S proteasome in liver, fat, and brain tissues of old, male and female mice fed encapsulated chow containing 2.24 mg/kg (14 ppm) rapamycin for 6 months. Further we assessed if rapamycin altered expression of the chaperone proteins known to interact with the proteasome-mediated degradation system (PMDS), heat shock factor 1 (HSF1), and the levels of key mTOR pathway proteins. Rapamycin had little effect on liver proteasome activity in either gender, but increased proteasome activity in female brain lysates and lowered its activity in female fat tissue. Rapamycin-induced changes in molecular chaperone levels were also more substantial in tissues from female animals. Furthermore, mTOR pathway proteins showed more significant changes in female tissues compared to those from males. These data show collectively that there are divergent tissue and sex effects of rapamycin on the proteasome-chaperone network and that these may be linked to the disparate effects of rapamycin on males and females. Further our findings suggest that rapamycin induces indirect regulation of the PMDS/heat-shock response through its modulation of the mTOR pathway rather than via direct interactions between rapamycin and the proteasome.

**Keywords: rapamycin, mTOR, proteasome, heat shock proteins, sexual dimorphic effects, longevity**

# **INTRODUCTION**

Rapamycin, an allosteric inhibitor of mechanistic (mammalian) target of rapamycin (mTOR) reportedly increases longevity in mice even when given at an advanced age as an encapsulated formulation in chow. This effect is significantly greater in females (Miller et al., 2014). The biological mechanism and relevance of this sex divergence in response to rapamycin is unclear. Surprisingly, rapamycin has also recently been shown to be an allosteric inhibitor of proteasome activity *in vitro* (Osmulski and Gaczynska, 2013). Further, 20S proteasome activity is reduced in liver tissues of elderly (25 month old) mice treated with rapamycin for 6 months when compared to control animals (Zhang et al., 2014a). However, gene expression studies undertaken in tissues harvested from the same animals showed equivocal effects on proteolytic pathways with differentially expressed genes linked to these pathways both up-regulated and also downregulated (Fok et al., 2014). Down-regulated proteolytic pathways are at variance with the widely accepted theory that the decline in functionality during aging is linked to an accrual of damaged or misfolded proteins (Ross and Poirier, 2004; David et al., 2010). This breakdown in proteostasis during aging is attributed at least in part to an age-associated decline in the efficiency of the proteolytic machinery (Rodriguez et al., 2010; Grimm et al., 2012). This leads to the concomitant accrual of aggregation-prone cytotoxic proteins that underlie many age-associated pathologies (Bucciantini et al., 2002; Ross and Poirier, 2004).

Age-related changes contributing to compromised proteasome and autophagy function in short-lived species include decreased transcription of some catalytic subunits, altered proteasome subcellular distribution, disruption of lysosomal control, and reduced degradative capacity of both proteolytic machineries (Ferrington et al., 2005; Massey et al., 2006; Rodriguez et al., 2010). In contrast to these declines in proteolytic degradation processes, the proteasome-mediated protein degradation system (PMDS), that includes both ubiquitin-dependent and independent proteasome machinery and associated molecular chaperones, is more robust in naturally long-lived species (Rodriguez et al., 2012; Edrey et al., 2014), long-lived mutants (Kruegel et al., 2011), calorically-restricted animals (Bonelli et al., 2008), and centenarians (Chondrogianni et al., 2000). As such it appears hard to reconcile a decline in proteasome function with rapamycininduced extended longevity.

The 20S core when doubly capped by 19S regulatory particles is called the 26S proteasome and is primarily responsible for ubiquitinylated protein degradation and the bulk of PMDS related proteolytic activity (Demartino and Gillette, 2007). While 20S proteasomes can exist un-capped *in vivo*, they are mostly inactive and contribute only 20–30% of the total proteasome content in a cell (Babbitt et al., 2005). The 19S regulatory caps through a combination of ATP-dependent (RPT) and ATP-independent (RPN) subunits mediate substrate uptake, by unfolding, deubiquitinylating, and moving proteins through to the 20S core (Demartino and Gillette, 2007). Upstream trafficking of substrates to the proteasome is partially controlled by the heat-shock protein (HSP) family of molecular chaperones. HSP70/72 together with its cochaperones may have a key regulatory role in PMDS and prolonged efficient function in long-lived species (Grune et al., 2011; Rodriguez et al., 2014). Further, the smaller molecular chaperone, HSP25, has been implicated in the mitigation of protein aggregation in vertebrates (Goldbaum et al., 2009). While chaperone interactions in the PMDS may be critical for maintained protein homeostasis, recently the drug rapamycin has been shown to suppress the heat-shock response in cell culture (Chou et al., 2012), yet nevertheless facilitates prolonged healthspan and extended longevity *in vivo* (Harrison et al., 2009; Miller et al., 2011, 2014; Zhang et al., 2014a).

We question whether this observation of reduced proteasome activity (Osmulski and Gaczynska, 2013; Zhang et al., 2014a) and suppression of the heat-shock response seen *in vitro* (Chou et al., 2012), is observed *in vivo* especially in light of rapamycin-mediated transcriptional regulation of certain proteasome-related genes (Fok et al., 2014). If this is the case, it may reveal that the mechanisms facilitating the increase in healthspan and longevity seen by improved protein homeostasis are independent of rapamycin-induced longevity and healthspan. We test the hypothesis that rapamycin counters the deleterious effects of aging in C57BL/6 mice through differentially suppressing the PMDS. We also ask if there are sex and/or tissue differences in in the various proteasome activities and molecular chaperone responses to rapamycin and whether changes in the mTOR pathway elucidate a mechanism for this rapamycin induced modulation of proteolytic function and lifespan.

# **METHODS**

# **ANIMALS**

Care of animals followed UT Health Science Center Institutional Animal Care and Use Committees approved procedures. Specific pathogen-free C57BL/6 mice were purchased from the National Institutes of Health colony reared in the Charles River Laboratories at 19 months of age. Mice were maintained under barrier conditions by the UTHSCSA Nathan Shock Center Aging Animal and Longevity Assessment Core and started on the

visceral fat ChTL (0.7X) and TL (0.7X) activity in females but not in males. PGPH activity did not show any significant differences between treated or untreated animals in any of the tissues examined. The y-axis indicates net proteasome activity in pmol/min/μg lysate for each of the respective activities measured. Statistically significant differences (Two-Way ANOVA, *p <* 0*.*05) comparing treatment vs. vehicle (∗) and/or male vs. female untreated (#) or male vs. female treated (\$) are indicated (*n* = 5 brain, fat, *n* = 6 liver).


**Table 1 | Rapamycin-influenced changes in markers of proteasome, chaperone, and mTOR pathway in old male and female brain, liver, and visceral fat lysates.**

*"UP" indicates a rapamycin-dependent significant increase. "DN" represents a rapamycin-dependent significant decrease.*

rapamycin/eudragit control diet at ∼19 months of age. Six months later at ∼25 months of age, the *ad libitum* fed animals were anesthetized with isofluorane, euthanized by cardiac exsanguination and the tissues immediately excised and flash frozen in liquid nitrogen. All the tissues from these animals that were not fasted prior to sacrifice were stored at −80◦C until used in analyses.

# **DIET PREPARATION**

Mice were fed a diet containing either encapsulated rapamycin or empty capsules (eudragit control). Microencapsulated rapamycin or empty microcapsules were incorporated into Purina 5LG6 diet. The rapamycin diet was prepared at 14 ppm using methods described by Harrison et al. (2009) and blood levels checked to confirm appropriate dosing. Data pertaining to the rapamycin blood levels, lifespan, and healthspan effects of these animals have been previously published (Fok et al., 2014; Zhang et al., 2014a).

# **LYSATES**

Mouse tissue (∼50–100 mg fat, ∼15–30 mg liver, 100–200 mg brain) was cryofractured under liquid nitrogen with a mortar and pestle. Powdered fat was lysed by homogenization in 2× dry weight in Protein Homogenization Buffer (50 mM HEPES, pH 7.6, 150 mM sodium chloride, 20 mM sodium pyrophosphate, 20 mM ß-glycerophosphate, 10 mM sodium fluoride, 2 mM sodium orthovanadate, 2 mM ethylenediaminetetraacetic acid, 1% IGEPAL, 10% glycerol, 1 mM magnesium chloride, 1 mM calcium chloride, 1 mM phenylmethylsulfonyl fluoride, and one tablet/10 ml Complete Mini (EDTA free) protease inhibitor tablets from Roche) for Western blots and in Re-suspension Buffer (RSB) (10 mM HEPES, pH 6.2, 10 mM NaCl, 1.4 mM MgCl2) without protease inhibitors at a weight per volume ratio of 1 g of tissue to 2 mL of buffer for peptidolytic assays. The RSB buffer was supplemented with the addition of 1 mM ATP, 0.5 mM DTT, 5 mM MgCl2 to help maintain intact 26S subassemblies (Liu et al., 2006). Powdered liver was lysed by homogenization in 5× dry weight in modified RIPA Buffer (150 mM sodium chloride, 50 mM Tris-HCl, pH 7.4, 1 mM ethylenediaminetetraacetic acid, 1 mM phenylmethylsulfonyl flouride, 1% Triton X-100, 1% sodium deoxycholic acid, 0.1% sodium dodecyl sulfate, 1μM Bortezomib proteasome inhibitor, and one tablet/10 ml protease and phosphatase inhibitor mini tablets (Thermo Fisher Scientific, Waltham, MA, USA) for Western blots. For peptidolytic assays, liver powder was lysed in RSB without protease inhibitors and with ATP supplemented as described above at a weight per volume ration of 1 g of tissue to 5 mL of buffer. Brain tissue was separated and lysed in RSB containing the protease/phosphatase cocktail tablet described above (Thermo Fisher Scientific) for Western blot analyses or RSB without protease inhibitors containing the ATP supplement for peptidolytic assays. Debris was cleared by centrifugation (3000 g for 12 min). The Bio-Rad Protein Assay (Bio-Rad Life Sciences, Hercules, CA) or BCA Assay (Thermo Fisher Scientific) was used to determine protein concentrations.

# **PEPTIDOLYTIC ACTIVITY**

In each assay 20μg of whole tissue lysates were incubated with 100μM of substrate specific for the type of proteasome activity. A saturating concentration of proteasome inhibitor N-(benzyloxycarbonyl) leucinyl-leucinyl-leucinal (MG132) was added to parallel samples. The difference of the fluorescence released with and without inhibitor was used as a measure of the net peptidolytic activity of proteasome as previously described using model peptide substrates to represent cleavage after hydrophobic (Chymotrypsin-like; ChTL) residues, basic residues (Trypsin-like; TL) and acidic residues (Post-glutamyl, peptide-hydrolizing; PGPH) (Rodriguez et al., 2010).

# **WESTERN BLOTS**

Tissue lysates were separated using a 4–12% SDS-PAGE (Biorad Life Sciences) and transferred to nitrocellulose membranes (Biorad Life Sciences). The membranes were probed with antibodies against the following proteasome and chaperone proteins: HSP90 (mouse mAb, SPA831, 1:20K), HSF1 (rabbit pAb, SPA901, 1:5K), HSP70/72 (mouse mAb, SPA810, 1:10K), HSP40 (HDJ1) (rabbit pAb, SPA400, 1:2.5K), HSP25/HSPB1 (rabbit pAb, SPA801, 1:10K), α7 (mouse mAb, PW8110, 1:5K,), RPT5 (mouse mAb, PW8310, 1:5K) (Enzo Life Sciences, Plymouth Meeting, PA, USA) and Carboxyl-terminus of HSP70 Interacting Protein (CHIP) [rabbit mAb(C3B6), #2080, 1:5K] (Cell Signaling Technology, Inc., Danvers, MA, USA). Antibodies against GAPDH (mouse mAb, G8795, 1:30K) (Sigma-Aldrich, St. Louis, MO, USA) were used as a loading control. For mTOR pathway proteins, the following antibodies were used

all at 1:1K dilutions: mTOR (rabbit pAb, #2972), phosphomTOR (Ser2448) (rabbit pAb, #2971), AKT (rabbit pAb, #9272), phospho-AKT (Ser473) (rabbit, pAb, #9271), S6 ribosomal protein (mouse mAb (54D2), #2317), phospho-S6 ribosomal protein (Ser240/244) (rabbit pAb, #2215), 4EBP1 (rabbit pAb, #9452) and phospho-4EBP1 (Thr37/46) (236B4) (rabbit mAb, #2855) (Cell Signaling Technology). Either the GAPDH antibody used above or pan-Actin (mouse mAb, MS-1295-P0, 1:20K) (Thermo-Fisher Scientific) was used as a loading control.

Primary antibodies were detected using anti-mouse IRDye 680LT, or anti-rabbit IR Dye800 CW (Li-Cor, Lincoln, NB, USA) conjugated antibodies. Secondary antibodies were incubated at 1:10K (anti-rabbit) or 1:20K (anti-mouse) for 2 h at room temperature and images were captured and subsequently quantified using the Odyssey Imaging System (Li-Cor) by quantifying fluorescent signals as Integrated Intensities (I.I. K Counts) using the Odyssey Infrared Imaging System, Application Version 3.0 software. We used a local background subtraction method to subtract independent background values from each box, more specifically, the "median" background function with a 3 pixel width border above and below each box was subtracted from individual counts. We calculated ratios for each antibody against the pan-actin or GAPDH loading control using I.I. K Counts. The respective antibody to pan-actin or GAPDH ratio was then used to calculate phosphorylated protein to total protein ratio were applicable.

### **NATIVE GEL ELECTROPHORESIS**

In this current study, we use an ATP-reconstituting system to maintain the integrity of the 26S proteasome (Liu et al., 2006), however it cannot be ruled out that the fluorogenic proteasome assay measures total proteasome activity for all the proteasome subassemblies, nor does it directly report ubiquitin-dependent activity of the whole 26S. With that caveat, however, we supplement this reliable indirect measure with in-gel assays on native gels.

Proteasome ChTL function was also measured using Native Gel Electrophoresis. This technique enabled us to discriminate if 26S or 20S proteasome activity predominated, the relative quantities of both the double-capped and uncapped proteasome and proteasome specific activities. Fifty micrograms of fractionated lysate from each of the sample groups prepared as described in Subcellular Fractionation above (q.v) were run on a 3–12% non-denaturing, gradient polyacrylamide gel (Life Technologies, Carlsbad, CA). The gel was run at 30 V for 30 min in a 4◦C cold

cabinet, thereafter the voltage was increased to 35 V for 1 h, 50 V for 1 h and further increased to 75 V for three more hours (Elasser et al., 2005; Tai et al., 2010).

Peptidolytic activity of proteasomes was detected after incubating the gels in a Suc-LLVY-MCA substrate dissolved in 50 mM Tris pH 8.0, 5 mM MgCl2, 1 mM DTT, 2 mM ATP, and 0.02% SDS for 15, 30, and 60 min at 37◦C. Proteasome bands were identified by the release of highly fluorescent, free AMC (Elasser et al., 2005; Vernace et al., 2007; Rodriguez et al., 2012). Fluorescence was quantitated using ImageJ software (http://imagej*.*nih*.*gov/ ij/). Following the in-gel assay, the protein from the gel was transferred to nitrocellulose using the i-blot transfer apparatus (Life Technologies) and subjected to Western blotting analyses using the α7 antibody described above to determine where the 26S and 20S proteasome complexes lay. The IRDye conjugating antibodies and the Odyssey Imaging System (Li-Cor) were used to quantitate the amounts of α7 signal (See Western Blots above).

### **STATISTICAL ANALYSES**

Prism 5 (GraphPad Software, Inc.) was used to analyze and graph the mTOR pathway Western blot data. An unpaired two-tailed *t*-test or Mann Whitney test was used to obtain *p*-values. *P*-values below 0.05 were considered significant. Proteasome and chaperone data was analyzed with Sigma Plot (v.11) using Two-Way ANOVA. Statistical significance was set at the *p <* 0*.*05 level with Tukey and Holm-Sidak corrections to counteract the probability of false positives. Cluster Analysis was performed using Cluster 3.0 and Treeview v 1.16r2 to generate output (open source code http://bonsai*.*hgc*.*jp/~mdehoon/software/cluster/; University of Tokyo, 2002). Data were log transformed to normalize the activity and protein expression data, and the cluster was generated using the group medians and hierarchical clustering with an uncentered Pearson Correlation to generate a complete linkage to create the most unbiased set of clusters (Yeung and Ruzzo, 2001; De Hoon, 2002).

# **RESULTS**

# **RAPAMYCIN-TREATED FEMALES SHOWED INCREASES IN BOTH CHYMOTRYPSIN-LIKE AND TRYPSIN-LIKE 26S PROTEASOME ACTIVITY IN BRAIN, BUT NOT IN LIVER OR VISCERAL FAT**

Peptidolytic activity of the proteasome was measured in old (25 month) male and female, rapamycin-treated and untreated, brain, liver, and visceral fat lysates using model peptide substrates specific for each of the three types of catalytic sites. The core particle contains three cleavage sites that degrade polypeptides or unfolded proteins severing peptide bonds on the carboxyl side of hydrophobic ("chymotrypsin-like"; ChTL), basic ("trypsinlike"; TL), or acidic ("peptidylglutamyl peptide hydrolyzing" or "caspase-like"; PGPH) residues (Demartino and Gillette, 2007).

To determine net proteasome activity, assays were run in parallel with and without the proteasome inhibitor MG132. Divergent responses were evident with regards to treatment, sex, and peptidolytic activities (**Figure 1**; summarized in **Table 1**). In untreated animals, males showed significantly higher ChTL activity in brain (*p* = 0*.*01), liver (*p* = 0*.*002), and fat (*p* = 0*.*03) (**Figure 1**, top panels) than observed in females. TL activity, while still higher for males in liver (*p* = 0*.*009), was the same in brain and higher in females visceral fat (*p* = 0*.*02). PGPH activity was higher in lysates from female animals compared to males in both brain (*p* = 0*.*03) and fat (*p* = 0*.*01), but in liver, PGPH activity was higher in untreated males compared to females (*p* = 0*.*003) (**Figure 1**, bottom panels).

Rapamycin effects were also tissue-specific but in general proteasome activity in females was more responsive to rapamycin treatment. In whole brain tissue lysates males showed no change in proteasome activity, whereas in females, the change in proteasome activity was significant, being nearly 3-fold higher for ChTL (*p* = 0*.*0009) and 2-fold higher for TL (*p* = 0*.*0005) substrates in treated compared to untreated samples (**Figure 1**, top left). In the liver lysates only male TL activity changed significantly, with a 20% decline in activity in the rapamycin treated (*p* = 0*.*02) (**Figure 1**, middle). Finally, proteasome activity from visceral fat lysates showed significant declines in ChTL (*p* = 0*.*05) and TL (*p* = 0*.*01) activities of 30% in samples from treated females compared to controls. Activity in male fat lysates did not change with treatment (**Figure 1**, right panels). Rapamycin treatment had no effect on PGPH activity in either sex or in any of the tissues tested (**Figure 1**, left panels). Interestingly, ChTL activity showed the greatest sexual dimorphism with sex-differences evident in the activities of samples from both rapamycin-treated and untreated controls. Activity was higher in male control samples, whereas females showed the higher activity in rapamcyin treated samples (**Figure 1**, top left).

Native gel electrophoresis has been used to determine if the proteasome remains intact in a higher molecular weight form (i.e., 26S) or exists disassembled (20S) (Elasser et al., 2005; Rodriguez et al., 2012). We used this technique to determine if either the 20S or 26S proteasome assembly, was affected by rapamycin treatment or showed sex-specific differences (**Figures 2**–**4**). In tissue lysates from female mouse brains, the in-gel assay measuring ChTL activity (**Figures 2A,B**) showed a significant increase in 26S proteasome activity in a rapamycindependent manner (*p* = 0*.*02), but not in 20S activity. In contrast ChTL activity in male mouse brains did not change (*p >* 0*.*05). The immunblot analyses of α7 (**Figures 2C,D**) revealed that the 26S proteasome content of this protein significantly increased in brain lysates of rapamycin treated female mice (*p* = 0*.*003). Interestingly, the proteasome α7 decreased significantly in rapamycin-treated male mice brain samples compared to those of the male control group (**Figures 2C,D**; *p* = 0*.*02), though this did

not reflect on activity. 20S proteasome content did not change for either sex.

Zymograms for 26S proteasome activity showed a significant decline in both liver (**Figures 3A,B**) and fat (**Figures 4A,B**) tissue lysates of rapamycin-treated females (*p* = 0*.*008 brain, *p* = 0*.*009 fat) with no change in proteasome activity for either sex at the 20S site. A significant decline in α7 protein content occurred in 26S proteasomes measured in liver samples from rapamycin-treated females (*p* = 0*.*004). Further, 20S α7 protein content declined in both male (*p* = 0*.*02) and female liver samples (*p* = 0*.*001) (**Figures 3C,D**). Proteasome content in fat tissue lysates (α7) also decreased significantly for both the 26S (*p* = 0*.*003) and 20S (*p* = 0*.*04) sites of treated female sample whereas α7 content did not change in the fat samples from males.

# **LEVELS OF PROTEASOME-RELATED CHAPERONES CHANGED MORE IN TREATED FEMALES THAN IN TREATED MALES**

Western blot analyses using a panel of antibodies representing major chaperone families and key proteasome subunits was undertaken in the tissues harvested from rapamycin-fed and Eudragit-(vehicle) fed controls. Tissues from three male and three female 25 month old mice were analyzed. In lysates from male samples, rapamycin had no effect on chaperone levels (**Figure 5A**). In female brain lysates, the levels of HSP70, CHIP, and HSP25 proteins increased in treated samples compared to controls (**Figure 5A**). Further, the 19S cap protein RPT5 also showed a higher protein content in these brain lysates from rapamycin-treated females (**Figure 5A**). Proteins were normalized to GAPDH whose mRNA is eIF4E insensitive and would not change with rapamycin treatment (Livingstone

et al., 2010), These data are indicative of an increased 26S population and RPT5 levels correlated with an increase in ChTL and TL proteasome activity in brain lysates (**Figure 1**). In liver, with the exception of HSP25 (which did not change), there was a decline in heat shock proteins in both sexes with treatment (**Figure 5B**). While α7 levels also decreased in males and females, this did not correlate with any loss in proteasome activity (**Figures 1**, **5B**). In visceral fat, HSF1 as well as HSP90, HSP70, HSP40, and HSP25 decreased significantly in samples from treated female animals (**Figure 5C**). Both RPT5 and α7 also showed protein declines in fat samples from rapamycin-treated females that correlated with a decline in ChTL and TL proteasome activity. Interestingly, samples from treated males showed declines in CHIP, HSP, and HSP25 (**Figure 5C**). Lastly, HSP90, HSP70, and RPT5 showed significant differences between males and females sample in their immuno-detection (**Figure 5C**).

these markers of the mTOR pathway measured in brain lysates from 25 month-old male and female mice (rapamycin-fed vs. control) and

# **mTOR PATHWAY MARKERS RESPONDED TO RAPAMYCIN TREATMENT IN FEMALES BUT THESE CHANGES WERE ALSO TISSUE DEPENDENT**

total protein. Significant *p*-values as derived by Two-Way ANOVA (*n* = 3) using Prism GraphPad Software is indicated for each set of proteins.

To examine for a correlation between the changes in proteasome activity and chaperone profile with the mTOR pathway, we immunoassayed a subset of mTOR pathway proteins and their phosphorylated forms (p) in brain, liver, and visceral fat lysates from old male and female, rapamycin-fed animals and compared to similarly aged controls relative to Actin (also insensitive to rapamycin) or GAPDH (Livingstone et al., 2010) (**Figure 6B**). Female rapamycin-treated brains had higher levels of p-mTOR (Ser2448), and both total and phosphorylated AKT (Ser473) and rpS6 (Ser240/244). We also observed an increase in total 4EBP1 and unchanged p-4EBP1 (Thr36/47) levels which led to significant decline in the ratio of p-4EBP1/4EBP1 (**Figure 6C**). In male lysates there was no significant difference in the phosphorylated or total protein expression of these four mTOR pathway proteins, with rapamycin-treatment (**Figures 6B,C**).

Rapamycin had little effect on protein expression of mTOR related proteins in liver tissue (**Figures 7**, **8**). Only p-4EBP1 (Thr36/47) in liver lysates of females was significantly higher with rapamycin (**Figures 7A,B**) whereas in liver lysates from males, rapamycin induced increased expression of both p-4EBP1 (Thr36/47) and total 4EBP1 (**Figures 8A,B**) as well as total AKT. The latter change led to a lower ratio of p-AKT (Ser473) to total AKT (**Figure 8B**).

Visceral fat lysates from rapamycin-treated females showed a trend toward a decrease in p-AKT (Ser473), total mTOR and downstream effectors p-rp-S6 (Ser240/244) and total 4EBP1 (**Figures 9A,B**). These rapamycin mediated declines in protein expression led to altered ratios of the phosphorylated form to the total proteins (**Figure 7B**). The effects of rapamycin on the expression of these proteins were attenuated in males. Only phosphorylated AKT (Ser473) and the ratios of p-mTOR (Ser2448) to total mTOR significantly increased (**Figures 10A,B**).

# **CLUSTER ANALYSES REVEALED THAT BRAIN PROTEASOME ACTIVITY IS MOST INFLUENCED BY RAPAMYCIN**

**Table 1** summarizes the observed rapamycin-mediated changes in proteasome activity, chaperones, and mTOR pathway proteins tested in this study. To determine whether the combined proteasome activities, chaperone, and mTOR pathway data assembled into common patterns of rapamycin-associated changes, cluster analyses were performed (**Figure 11**). First, proteasome activity and proteasome-related chaperones were clustered (**Figure 11A**). Cluster analysis revealed that groups separated by tissue type, weighted toward the higher levels of proteasome activity in liver and brain. These groups were further divided in the analysis by sex, but treatment was indistinguishable (**Figure 11A**). In this comparison, two distinct clusters formed. The first major cluster consisted of proteasome activity, the 19S ATPase RPT5 and HSF1in one sub-cluster with α7 and HSP25 in another. The second major cluster showed that the large chaperones (HSP90 and HSP70) and the HSP70 co-chaperones HSP40 and CHIP sub-divided into sub-clusters (**Figure 11A**).

The mTOR pathway proteins were added into the next cluster analysis (**Figure 11B**). This revealed that rapamycin-treated female brain and treated male liver formed two distinct groups. Three clusters formed in this analysis, with the mTOR pathway proteins 4EBP1 and p-AKT joining the "proteasome activity cluster" with the same proteins associated in the first Chaperone-Proteasome analysis from panel A (RPT5, α7, HSP25, and HSF1). A second cluster containing HSP90 and the co-chaperones HSP40 and CHIP also contained p-4EBP1, AKT, and rpS6 (**Figure 11B**). The last cluster in this analysis was characterized by low p-mTOR and low p-rpS6 (brain), and associated with HSP70 and total mTOR (**Figure 11B**).

# **DISCUSSION**

In this study we evaluated if rapamycin-mediated changes in mTOR signaling and the ubiquitin proteasome system (i.e., proteasome activity and content and associated protein levels various chaperones) correlated with the sexually dimorphic effects of rapamycin on lifespan. Rapamycin-induced changes in proteasome activity and expression of the various molecular chaperones and mTOR pathway proteins were both sex- and tissue-specific (**Figure 11**). Rapamycin effects are readily apparent from both individual variables and cluster analyses (**Figure 11**). The latter were generated by applying multiple peptidolytic assays and associated chaperone proteins as well as a sampling of mTOR pathway proteins across the three tissues harvested from the same individuals. Cluster analyses reveal there are interactions between the proteasome-chaperone network and the mTOR pathway. Rapamycin treatment led to elevated proteasome activities in the brain of females, potentially promoting the removal of oxidatively damaged and misfolded proteins and creating improved proteostasis in female mice that are likely to contribute to the better protection of their brains against environmentally mediated (e.g., oxidatively damaged/ glucose crosslinking) protein aggregation. Better protection of the female brain may be an evolutionary life-history tradeoff and these traits may possibly vary with reproductive status (non-reproductive pregnant, lactation). Improved proteostasis and concomitant neuroprotection in the brains of rapamycin treated females supports previous studies using genetically engineered mouse models of Alzheimer's

**pathway markers changed significantly in visceral fat lysates from rapamycin-treated females.** Measuring protein content of several mTOR pathway markers analyzed from visceral fat lysates taken from 25 month rapamycin-fed and control female mice via immunoblot

disease in which rapamycin firstly improved cognition and when subjected to a chronic high sugar diet prevented the accrual of protein aggregates, plaques and neurofibrillary tangles (Orr et al., 2014).

### **RAPAMYCIN EFFECTS ARE TISSUE-SPECIFIC**

Contrary to a previous data that examined only the 20S catalytic core ChTL activity (Zhang et al., 2014a), our extensive study revealed that rapamycin treatment did not globally suppress the PMDS or its associated chaperones. Indeed in our current study, declines in proteasome were observed predominantly in visceral fat (**Figures 1**, **4**) in contrast the previous study using mice in the same cohort showed decreases in both heart and liver 20S ChTL proteasome activity in both sexes of rapamycin-treated mice (Zhang et al., 2014a). The decline in proteasome activity in peripheral tissues both in this study and that of Zhang, is in keeping with *in vitro* findings that rapamycin can allosterically decrease in p-rp-S6 which also lowered the ratio (phosphorylation state) of p-rpS6 to total rp-S6. Total 4EBP1 also declined, increasing the phosphorylation state. *P*-values, derived using Prism GraphPad Software, are indicated for each dot plot (*n* = 5).

inhibit proteasome activity (Osmulski and Gaczynska, 2013) However, this cannot explain the tissue specific increases in ChTL and TL proteasome activity in the brains of rapamycin treated animals.

While ChTL is considered the pivotal peptidolytic activity of oxidatively damaged proteins, other peptidase activities such as TL and PGPH activity may be more sensitive to rapamycin. Female mice treated with rapamycin showed the most divergent responses compared to untreated females for both the ChTL and TL catalytic sites in fat (decrease) and brain (increase) (**Figure 1**). PGPH proteasome activity did not change suggesting that rapamycin-mediated effects target specific types of proteins and peptides for cleavage driving changes in proteasome function rather than a restructuring of the proteasome although there is some evidence that mTOR signaling can induce proteasome subunits through both complex 1 or complex 2 (Lamming et al., 2014; Zhang et al., 2014b).

Given that rapamycin is likely to first reach the liver via the hepatic portal vein and be at its most concentrated there, it is surprising that of the three tissues examined, the liver was least sensitive to rapamycin treatment as shown by the peptide assays (**Figure 1**). It is possible that rapamycin is metabolized in such a way in the liver that it has minimal effects on liver proteasome function. Alternately, as the liver has several pathways involving nutrient signaling, a slight perturbation in the mTOR pathway is likely to be compensated for by other signaling pathways. Unfortunately, published RNA-sequence analyses data revealing differentially expressed genes in the transcriptome with rapamycin were equivocal (Fok et al., 2014) and did not reveal an obvious explanation for a lack of a change in 26S proteasome activity or a decline only in 26S TL activity as was seen in males with treatment (**Figure 1**). However, the in-gel assay on native gels did show a decline ChTL activity in lysates from rapamycintreated samples from females (**Figure 3**) and a decline in α7 proteasome content suggesting alterations in liver proteasome expression and quantity by rapamycin(Zhang et al., 2014a). Given the low blood titer of C57BL6 compared to the UM-HET3 mice (∼3–4 ng/L compared to 13.4 ng/L, respectively, after 6 months of treatment) (Harrison et al., 2009; Zhang et al., 2014a), we may have only observed effects in tissues that are particularly sensitive to rapamycin. Also, there definitely seem to be straindependent differences in at least rapamycin uptake or metabolism which could dictate the effect of the drug on measured parameters.

### **RAPAMYCIN DECREASES PROTEASOME ACTIVITY IN VISCERAL FAT**

We chose to include fat tissue in our analyses because we hypothesized it may be more sensitive to alterations in mTOR signaling, being a nutrient storage tissue, and was previously shown to be responsive male and female mice (Harrison et al., 2009). In C57BL6 mice, proteasome activity declines with age in adipose tissue (Dasuri et al., 2011). Our study shows that in old animals, there is a further reduction in proteolytic activity in visceral fat harvested from female mice after rapamycin-treatment (**Figure 1**) which was supported by a similar observation on native gel

selected mTOR pathway proteins **(B)**. Groups representing the various lysates collected from animals treated with rapamycin and corresponding controls are organized in columns represent cases. Four letter codes define the groups with the first two in the code representing gender

zymograms (**Figure 4**). This was accompanied by a treatmentrelated reduction in the key proteasome subunits α7 and RPT5 (**Figures 4**, **5**). Thus, rapamycin appears to affect upstream control of critical genes in the PMDS in visceral fat though at this time it is unknown whether this effect is transcriptional or translational. Optimal function of the 26S is dependent on a tightly regulated ratio of ATPases such as RPT5 in the 19S regulatory cap (Smith et al., 2011). Alternatively, or in combination with an upstream regulation of RPT5 and/ or other proteasome ATPases, rapamycyin has been shown to directly block proteasome activity *in vitro* by interfering with the attachment of the 19S cap (Osmulski and Gaczynska, 2013). A similar mechanism may be in place in fat tissue whereby the inhibition of mTOR signaling in adipose tissue may block the binding of the regulatory cap to the catalytic core and impair proteasome function. Reduction of the proteasome activity has been shown to induce cytotoxicity, upregulate cell stress responses, and lead to various pathologies in other tissues (Goldbaum et al., 2006; Grimm et al., 2012). Taken together these data suggest that rapamycin-induced declines in visceral fat PMDS function could be detrimental and contribute to some of the observed peripheral tissue pathologies associated with age (Dasuri et al., 2011). For example, the development of obesity and insulin signaling in type 2 diabetes is influenced maps were prepared with TreeView v.1.16r2. Color scheme corresponds to the normalized values of variables where bright green represent the lowest (approaching −3.0) and bright red the highest (close to +3.0) values. The lengths of tree branches are proportional to a relative similarity between variables and between cases.

by the PMDS as insulin receptor substrate 1 is inactivated by degradation through this system (Sun et al., 1999; Chang et al., 2009). Thus, effects that mimic aging in PMDS function could substantially contribute to age-related insulin resistance in adipose tissue and other deleterious features associated with both fat and liver metabolism (Umemura et al., 2014). While we did not measure changes in insulin signaling in this cohort of animals, others have shown consistently that rapamycin affects the insulin pathway in much the same manner in multiple strains of mice including C57BL6, namely creating glucose intolerance, but causing insulin sensitivity (Lamming et al., 2013; Orr et al., 2014; Yu et al., 2014). PMDS could be a potential pathway to explain this phenomenon.

# **RAPAMYCIN INCREASES PROTEASOME ACTIVITY IN THE BRAIN**

The most pronounced effects to proteasome activity were evident in the brain suggesting that it is not the rapamycin directly inducing these changes but rather the down-stream signaling it induces in rapamycin-responsive tissues. In contrast to the potential detrimental effects of rapamycin seen in visceral fat proteolytic function, the increase of PMDS in the brain could be beneficial and could suggest that organisms selectively protect certain more vulnerable tissues. Here, we observe that rapamycin

induced an increase in proteasome activity, a phenotype commonly observed in long-lived species in the periphery as well as in the brain (Chondrogianni et al., 2000; Rodriguez et al., 2012; Edrey et al., 2014). Brain lysates from treated females showed both an increase in ChTL and TL activity (**Figures 1**, **2**) as well as significant increases in 26S proteasome content on non-denaturing gels (**Figure 2**). Proteasome-related chaperones HSP70, C-terminal HSP-interacting protein (CHIP), RPT5, and HSP25 also showed significant increases in rapamycin-treated samples compared to control (**Figure 5**). Further, α7, RPT5,HSF1, and HSP25correlated strongly with higher levels of proteasome activity (**Figure 11**) in the cluster analysis. Other studies examining brains from rapamycin-treated mice have seen a similar enhancement of protein homeostasis and/or the HSP system (Spilman et al., 2010; Pierce et al., 2013; Orr et al., 2014). For example, rapamycin-fed mice showed the enhanced expression of the small chaperone gene, alpha-crystallin B chain (*CRYAB*) in the brains of a mouse Alzheimer's model (Pierce et al., 2013). Further, crossing this model with an HSF1-transgenic mouse (Pierce et al., 2010), which showed increased HSF1, HSP90, and CryAB, reduced toxic brain amyloid-β levels and improved cognitive function (Pierce et al., 2013). Interestingly, we did not see an increase in HSF1 in brain tissue with rapamycin treatment though cluster analyses suggested a correlation (**Figures 5**, **11**). However, several chaperones showed increases in rapamycin-treated female brain samples (i.e., HSP70, CHIP, HSP25) suggesting both HSF1 and non-canonical upregulation of HSPs. One possibility is that if rapamycin indeed inhibits proteasome activity *in vivo* as suggested by *in vitro* studies (Osmulski and Gaczynska, 2013) an alternate heat shock factor such as HSF2 may be triggered inducing the same set of chaperones as HSF1 (Mathew et al., 1998).

The differences seen between the effects of rapamycin on the brain and peripheral tissues also suggests a tissue-specific decoupling of rapamycin function. In a study in which mice genetically mutated to serve as transgenic mouse models for Alzheimer's disease a tissue-specific decoupling of rapamycin effects were observed when these mice were fed a high sucrose diet and developed insulin resistance. While the exacerbation of plaques were ameliorated by the simultaneous treatment with rapamycin, rapamycin had no effect on peripheral insulin resistance and liver protein levels (Orr et al., 2014).

In this study, we did not examine markers of autophagy. Counter-intuitively, a previous investigation reports that, autophagy in brain tissue lysates was not increased in rapamycintreated non-transgenic mice but only was manifest in animals genetically engineered to express high levels of amyloid-β or tau (Spilman et al., 2010; Orr et al., 2014). So while autophagy may have a key role in reducing aggregates from disease pathologies, non-aggregated protein degradation may instead be enhanced by the PMDS. The increase in chaperone-E3 ligase CHIP in the brain and a significant decline in liver tissue (**Figure 5**) may hold a clue. CHIP has been shown to be essential in modulating oxidative load and degradation of oxidized proteins through the PMDS as well as act as the E3 ligase for the degradation of tau, parkin, and polyglutamine expansions in brain tissue (Dickey et al., 2007; Sisoula and Gonos, 2011). As such, in the brain chronic rapamycin treatment could trigger an E3 ligase like CHIP to protect against proteotoxic stress, increasing the translation of chaperone proteins and the proteasome-mediated, protein degradation machinery to maintain protein equilibrium.

# **CHANGES IN mTOR SIGNALING CORRELATE WITH SEX AND TISSUE DIFFERENCES**

In both female brain and fat tissues the mTOR pathway proteins, as expected, showed the most changes with rapamycin-treatment (**Figures 3**, **7**; **Table 1**). However, while the changes in female fat samples suggest rapamycin blocks mTOR signaling, rapamycinmediated changes in the brain did not (**Figures 6**, **9**; **Table 1**) (reviewed in Wullschleger et al., 2006). Rather, these data suggest that mTOR remains active in the brain, and that the inhibitory effects of rapamycin are suppressed in brain tissue. This uncoupling of brain and peripheral effects on mTOR was unexpected but could also explain the lack of increased levels of autophagy in the brain of rapamycin-treated control animals seen in previous studies (Spilman et al., 2010; Orr et al., 2014). Further, the increase in p-rpS6 in the female brain (**Figure 6**) was very different to what was observed in female fat (**Figure 9**) or in the intestine of rapamycin-treated familial adenomatous polyposis mice both which showed a decrease in the phosphorylation state (ratio of phosphorylated to non-phosphorylated protein) of rp-S6 and other mTOR signaling molecules (Hasty et al., 2014). This paradox may be linked to the increase in both total and phosphorylated AKT (**Figure 6**, **Table 1**). A similar induced activation of AKT leading to a resistance of rapamycin treatment has been observed in human lung cancer cells and human rhabdomyosarcoma cell lines and in rodent cells overexpressing insulin-like growth factor (IGF) II (Sun et al., 2005; Wan et al., 2007). Unlike these cell systems, whereby the phosphorylation of both downstream mTOR targets S6K1 and 4EBP1 were suppressed, this was not observed in brain lysates from rapamycin-treated old females in this study (**Figure 6**). Instead we observed an increase in phosphorylated ribosomal protein S6, the target of S6K1 (**Figure 6**), responsible for 5 terminal oligopyrimidine (TOP)-dependent translation (Wullschleger et al., 2006). Further, there was a decline in the phosphorylation state of 4EBP1 (**Figure 6**) which is responsible for control of cap-dependent translation (Wullschleger et al., 2006) and inhibition of TOPdependent translation (Thoreen et al., 2012). This may be also be reflected by the cluster analysis correlation of low 4EBP1 with proteasome activity (**Figure 11**). These changes in translational control suggest a focus on TOP-dependent translation, and have further repercussions on the PMDS, as several proteasomerelated HSP mRNAs can be preferentially translated through TOP-dependent translation during stress (Cuesta et al., 2000; Pierce et al., 2013). Conversely, in fat we observed a decline in phosphorylated rp-S6 and an increase in the phosphorylation state of 4EBP1 suggesting a decrease in translation (**Figure 9**) (Thoreen et al., 2012) that could in turn influence the observed reduction of proteasome activity (**Figure 1**).

Phosphorylation of AKT also protects against the apoptotic effects of proteasome inhibition further linking PMDS to the AKT cell survival program (Yu et al., 2006; Zanotto-Filho et al., 2012). As total AKT is also induced by 17-β estradiol (Haynes et al., 2000), this may explain why rapamycin-treated female animals have more robust effects when treated by the drug. Taken together, the sex-specific increase seen in brains of rapamycin treated mice in our comprehensive measures of proteasome activity could be linked to a sex-dependent change in AKT signaling driven by mTOR regulation of the heat-shock pathway through activated rp-s6 (as we observed in treated female brains; **Figure 3**). Interestingly, mTOR signaling through rp-S6 has also been shown to increase proteasome subunits with a dependence on nuclear factor erythroid-derived 2-related factor 1 (NRF1) (Zhang et al., 2014b). NRF1 also mediates the recovery of proteasome activity after stress or proteotoxic insults (Radhakrishnan et al., 2010; Balasubramanian et al., 2012).

# **CONCLUSIONS**

Our data indicate that the sexually dimorphic effects of lifespan extension induced by rapamycin (Harrison et al., 2009; Fok et al., 2014; Miller et al., 2014; Zhang et al., 2014a) may be linked to sex differences in tissue-specific responsiveness of the regulators and components of mTOR, HSPs and PMDS pathways. Proteolytic activity is augmented in the brain facilitating more efficient removal of damaged or unfolded proteins, and as a consequence thereof, improved brain structural and functional integrity. This protective response to rapamycin treatment in brain tissue is uncoupled from that of the response in peripheral fat tissue. The latter, appears to be left more vulnerable to the potentially detrimental peripheral proteotoxic effects of rapamycin (Wilkinson et al., 2012; Ponticelli, 2014; Zhang et al., 2014a). A proposed summary of what happens to the PMDS in female brain vs. the periphery (specifically visceral fat) with rapamycin treatment is shown in **Figure 12**. Here we show that rapamycin appears to influence up- stream regulators of the proteasome-chaperone network through stimulation of the AKT pathway (**Figure 12**). Defining how the AKT pathway is stimulated or repressed in various tissues in a sex-dependent manner can give us insight on how to better understand rapamycin's effects in a therapeutic context. Upstream regulators of the PMDS including E3 ligases such as CHIP such as HSF1, 2 or NRF1 could be activated or suppressed by the AKT stress response in a sex-dependent manner thereby differentially affecting proteostasis in peripheral and brain tissues. Maintenance of protein homeostasis in the female brain may play a pivotal role in the extended longevity observed only in the rapamycin-treated female C57BL6 mice.

# **AUTHOR CONTRIBUTIONS**

Karl A. Rodriguez and Sherry G. Dodds conducted the experiments. Karl A. Rodriguez and Rochelle Buffenstein wrote the initial draft of the paper. Randy Strong, Veronica Galvan, Z. D. Sharp, and Rochelle Buffenstein contributed materials for the study. All authors helped with editing and revising the manuscript.

# **ACKNOWLEDGMENTS**

The authors would like to thank the Aging Animal and Assessment core supervised by Vivian Diaz at the Sam and Ann Barshop Center for Longevity and Aging Studies for care and handling of the animals. This work was supported by an NIA Training Grant (T32 AG021890-08) (Karl A. Rodriguez), an NIA R21 (1R21AG043912) (Rochelle Buffenstein), an RC2AG036613 (Z. D. Sharp and Randy Strong), a P30-AG013319 (Rochelle Buffenstein), an Interventions Testing Program, U01 AG022307 (Randy Strong), and this research was conducted while Karl A. Rodriguez was an Ellison Medical Foundation/AFAR Postdoctoral Fellow.

# **REFERENCES**


effect of chronic inhibition of TOR by rapamycin and is sufficient to ameliorate Alzheimer's-like deficits in mice modeling the disease. *J. Neurochem.* 124, 880–893. doi: 10.1111/jnc.12080


**Conflict of Interest Statement:** None of the contributing authors received payment or services from a third party for any aspect of this submission. Z. D. Sharp, Randy Strong, and Veronica Galvan are uncompensated scientific advisors for Rapamycin Holdings, Inc. Z. D. Sharp, Randy Strong, and Veronica Galvan are part holders of a patent (#13/128/800 pending) for the use of encapsulated rapamycin in treating or preventing an age-related disease, condition, or disorder. There are no other relationships that could have influenced or give the appearance of potentially influencing this work. 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.

*Received: 29 July 2014; accepted: 11 October 2014; published online: 04 November 2014.*

*Citation: Rodriguez KA, Dodds SG, Strong R, Galvan V, Sharp ZD and Buffenstein R (2014) Divergent tissue and sex effects of rapamycin on the proteasome-chaperone network of old mice. Front. Mol. Neurosci. 7:83. doi: 10.3389/fnmol.2014.00083 This article was submitted to the journal Frontiers in Molecular Neuroscience.*

*Copyright © 2014 Rodriguez, Dodds, Strong, Galvan, Sharp and Buffenstein. 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.*

# Ubiquitin pathways in neurodegenerative disease

# *Graham Atkin\* and Henry Paulson*

*Department of Neurology, University of Michigan, Ann Arbor, MI, USA*

### *Edited by:*

*Fred Van Leeuwen, Maastricht University, Netherlands*

### *Reviewed by:*

*Hansen Wang, University of Toronto, Canada Baojin Ding, University of Massachusetts Medical School, USA*

### *\*Correspondence:*

*Graham Atkin, Department of Neurology, University of Michigan, Rm. 4160, Biomedical Science Research Building, 109 Zina Pitcher Place, Ann Arbor, MI 48104, USA e-mail: atking@umich.edu*

Control of proper protein synthesis, function, and turnover is essential for the health of all cells. In neurons these demands take on the additional importance of supporting and regulating the highly dynamic connections between neurons that are necessary for cognitive function, learning, and memory. Regulating multiple unique synaptic protein environments within a single neuron while maintaining cell health requires the highly regulated processes of ubiquitination and degradation of ubiquitinated proteins through the proteasome. In this review, we examine the effects of dysregulated ubiquitination and protein clearance on the handling of disease-associated proteins and neuronal health in the most common neurodegenerative diseases.

**Keywords: ubiquitin, neurodegenerative diseases, proteasome, protein quality control, Alzheimer's disease, Parkinson disease, Amyotrophic Lateral Sclerosis, Huntington's disease**

# **INTRODUCTION**

The unique demands placed on neurons by their exquisitely complicated and dynamic architecture have been appreciated by investigators for over 100 years (Golgi, 1886; Ramon y Cajal, 1909; Herculano-Houzel, 2011). A recent study places the number of neurons in the human brain at approximately 85 billion (Azevedo et al., 2009). With each neuron making upwards of 10 thousand synaptic connections, the estimated total number of synapses reaches toward 8.5 hundred trillion (8*.*<sup>5</sup> <sup>×</sup> <sup>10</sup>14). Astonishingly, plasticity can occur selectively at particular subsets of synapses within a neuron, even down to the level of a single specified synapse (Lee et al., 2010). Failure to maintain these synaptic connections and their proper plasticity are hallmarks of a host of neurodegenerative diseases, and loss of synaptic connections correlates with diminished cognitive function even before neurons degenerate (Masliah et al., 1991a,b). With 2788 unique proteins already identified as integral to the composition of each synapse (Pielot et al., 2012), the management of unique protein environments requires sufficiently complex and modifiable systems of protein quality control. The Ubiquitin Proteasome System (UPS), a set of interacting enzymes and associated proteins, is able to address these diverse proteostatic needs through the orchestrated activity of over 500 components working in versatile combinations to regulate protein-protein interactions and eliminate unwanted proteins. As newly synthesized proteins form new structures and connections, the UPS works to ensure that old proteins are degraded to make way and that the proper complement of building materials is available. To achieve these functions, components of the UPS are recruited to dendritic spines in response to synaptic activity (Bingol and Schuman, 2006; Bingol et al., 2010). Evidence continues to mount for the necessity of UPS involvement in the dynamic remodeling of synaptic structures following synaptic activity (Hegde et al., 1997; Ehlers, 2003; Pak and Sheng, 2003; Patrick et al., 2003; Bingol and Schuman, 2004; Hung et al., 2010; Fu et al., 2011). Many synaptic components have been identified as activity-dependent substrates for agents of the UPS. These include postsynaptic receptors (Kato et al., 2005; Lussier et al., 2011) and the scaffolding proteins which hold them in place (Colledge et al., 2003; Shin et al., 2012). Transsynaptic proteins such as Beta-Catenin and Beta-Integrin align and connect pre- and post-synaptic elements and are degraded through the UPS (Yoshida et al., 2002; Dreier et al., 2005). The shape, size, and placement of dendritic spines are also determined in part by proteins that regulate cytoskeletal organization and are themselves subject to ubiquitin-mediated degradation (Bingol and Schuman, 2005; Marland et al., 2011). These contributions to synapse maintenance and synaptic plasticity require that the UPS functions properly. For example, pharmacologic inhibition of the UPS leads to a frank reduction in activity-dependent synaptic plasticity (Ehlers, 2003) and a dose-dependent loss of synaptic connections (Bajic et al., 2012). Robust loss of synaptic connections is evident in all of the major neurodegenerative disorders. The full role of ubiquitination pertaining to synaptic structure and function remains incompletely understood, but has been the focus of significant investigation (Hegde and Upadhya, 2007; Bingol and Sheng, 2011; Hanus and Schuman, 2013).

As important as the contributions of the UPS to maintaining the plasticity of synapses are its diverse roles in ensuring general cell health, including the elimination of misfolded or damaged proteins, mediation of receptor signaling pathways, response to DNA damage and oxidative stress, and progression of the cell cycle, among other roles (Bernassola et al., 2010; Shang and Taylor, 2011). UPS function is essential to cell health and survival in all cell types. Improper clearance of proteins is believed to be a causative or contributing factor in many neurodegenerative diseases, which are often characterized by the accumulation of aggregated proteins (Alves-Rodrigues et al., 1998; Huang and Figueiredo-Pereira, 2010). Whether aggregation itself is the cause of toxicity or merely represents a strategy by which neurons sequester toxic proteins remains contested (Ross and Poirier, 2005), but the failure of quality control pathways to eliminate these unwanted proteins is evident. UPS dysfunction has been reported in the most common neurodegenerative diseases, including Alzheimer's disease (AD), Parkinson's disease (PD), Amyotrophic Lateral Sclerosis (ALS), and Huntington's disease (HD), as well as less common disorders and various animal models of protein aggregation (Keller et al., 2000; Bence et al., 2001; McNaught et al., 2003; Seo et al., 2004; Lonskaya et al., 2013). How the UPS becomes impaired in disease states is not always clear, as too little is known about how disease-related proteins are handled under normal conditions. Research into these areas has begun to reveal just how intricate and extensive ubiquitinmediated processes within the cell are, and accordingly, just how significant their dysregulation in a pathogenic context must be. Recent studies describing the plurality of ubiquitin-related abnormalities seen in disease states suggest the intriguing possibility that the disruption of one such process leads into that of another. Hopefully, further investigations will uncover potential therapeutic interventions.

# **UBIQUITINATION AND THE UBIQUITIN PROTEASOME SYSTEM**

Ubiquitin is a 76 amino acid (∼8 kDa) protein expressed in all eukaryotic cells. It is highly conserved throughout evolution; the amino acid sequence of human ubiquitin is identical to that of Aplysia (Hegde et al., 2000) and nearly identical to that of yeast (Finley and Chau, 1991). Through the coordinated activity of multiple enzymes, ubiquitin is covalently added to substrate proteins through the formation of an iso-peptide bond between the C-terminal diglycine motif of ubiquitin and lysine residues on the target (**Figure 1**) (Hershko and Ciechanover, 1998). This cascade begins with the ATP-dependent attachment of ubiquitin's C-terminal glycine through a thio-ester bond to an active-site cysteine on an Ubiquitin Activating Enzyme (E1). This ubiquitin is then transferred to a Ubiquitin Conjugating Enzyme (E2) through a thio-ester bond. From there, the E2 enzyme will cooperate with a Ubiquitin Ligase (E3) to transfer the ubiquitin molecule to the target lysine on a substrate protein. This last step occurs differently depending on the type of E3 involved, but it is at this step that substrate specificity is believed to occur, with E3 ligases selecting substrates for ubiquitination. However, this view has recently come under some scrutiny, as evidence emerges for a role for E2s in the process of substrate selection (Scaglione et al., 2013).

E3s are typically grouped into three classes, with each defined by the specific protein domains it possesses. These domain-based classes include: (1) Really Interesting New Gene (RING) fingercontaining E3s, (2) Homologous to E6-AP (HECT) domaincontaining E3s, and (3) E3s composed of multiple subunits. RING finger-containing E3s bring the E2 and the substrate protein into sufficiently close proximity for the transfer of the ubiquitin to its target lysine (Lorick et al., 1999). HECT-domain containing proteins possess an active-site cysteine to which the ubiquitin is first transferred from the E2 before being passed to the substrate protein (Huibregtse et al., 1995). In contrast to these single-unit E3 ligases, multi-subunit ligases are composed of multiple adaptor proteins, cofactors, and scaffolding proteins that confer substrate specificity and facilitate ubiquitination (Cardozo and Pagano, 2004). The Skp1/Cul1/F-box (SCF) protein complex and the Anaphase-promoting Complex (APC) are among the best studied of these multi-subunit E3s. SCF complexes can include various combinations of scaffolding proteins called cullins, F-box proteins, and substrate adaptors (Hao et al., 2005). APC is less variable, containing Apc2, Apc11, and either Cdh1 or Cdc20 for substrate recognition (Biggs et al., 2006). Each E3 can recognize multiple substrates; Fbxw1/β-TRCP1, for example, has upwards of 40 documented substrates itself (Skaar et al., 2009). Further contributing to the complexity of these multi-subunit E3 ligases is the developmental and spatial restriction of their expression within cells (Pines, 2006). There are estimated to be at least 500 different E3 ligases, with more continuing to be discovered (Ardley and Robinson, 2005).

This highly regulated process adds ubiquitin to target lysines on the substrate, but it can also add other ubiquitin molecules onto lysines of a ubiquitin already conjugated to a substrate. By this process, ubiquitin chains of varying length and composition can be formed. The elongation of ubiquitin chains can occur at any of ubiquitin's own seven lysines, resulting in the formation of different linkage types (Peng et al., 2003). Although all possible linkage types are present in cells, their precise functions remain only partially understood (Xu et al., 2009). Chains formed through the addition of ubiquitin exclusively at lysine 48 (K48) have been recognized to signal protein degradation (Glickman and Ciechanover, 2002), whereas K63-linked ubiquitin chains seem to subserve diverse functions beyond protein degradation (Jacobson et al., 2009). For example, K63-linked chains regulate NF-κB signaling not by promoting protein degradation but by influencing ubiquitin-dependent protein-protein interactions (Hadian et al., 2011). Elsewhere, however, they have been implicated in promoting the lysosomal degradation of the low density lipoprotein receptor (LDLR) (Zhang et al., 2013) and the epidermal growth factor receptor (EGFR) (Huang et al., 2013). Both K48- and K63-linked chains have been observed to modify kinase activity in response to cellular stress, as have K11-linked chains (Ben-Neriah, 2002; Bertrand et al., 2011). K11-linked chains are critical for cell-cycle regulation and cell division (Matsumoto et al., 2010). Other linkages, including atypical, mixed-type linkages are less well studied, but have been implicated in similar processes within the cell (Ikeda and Dikic, 2008; Husnjak and Dikic, 2012).

The functions thus far attributed to specific chain linkages represent only a fraction of the diverse roles ubiquitin is known to play in cellular processes (Husnjak and Dikic, 2012), many of which do not depend on proteasome function (Mukhopadhyay and Riezman, 2007). Ubiquitination regulates DNA repair (Jackson and Durocher, 2013), protein localization and endocytosis (Haglund et al., 2003; Hicke and Dunn, 2003; Schnell and Hicke, 2003), and protein-protein interactions (Hoege et al., 2002; Moldovan et al., 2007). Free, unanchored chains of ubiquitin molecules are also present in cells and can regulate numerous functions including kinase activation (Xia et al., 2009). The UPS has been also been implicated in the turnover of mRNA, although whether this regulation is direct or indirect remains unclear (Cano et al., 2010). Intriguingly, E3 ligases can

**FIGURE 1 | The process of ubiquitin conjugation.** 1. Ubiquitin (U) is bound via a thioester bond to the active-site cysteine of an E1 Ubiquitin-Activating enzyme, through a process requiring ATP. 2. The ubiquitin molecule is then passed to an E2 Ubiquitin-Conjugating Enzyme through trans(thio)esterification. 3. An E3 ubiquitin ligase brings the E2 into sufficiently close proximity and correct alignment with a substrate protein to facilitate the transfer of ubiquitin to a target residue. In the case of HECT-type E3s, the ubiquitin is first transferred to an active site cysteine on the E3 before being conjugated to the substrate. E3s can also exist as multi-subunit

themselves be targeted for ubiquitination, offering an additional level of control for this important pathway (Buschmann et al., 2000; Xiong et al., 2009). The ubiquitination of E3 ligases can result either in their degradation or in the modification of their activity (Scaglione et al., 2011).

Once a substrate is tagged for elimination through the addition of a ubiquitin chain, it must be targeted to the cell's degradation machinery by chaperone proteins that recognize and bind to poly-ubiquitin chains. One such chaperone is Valosin-containing protein (VCP), which has been shown to physically interact with and shuttle poly-ubiquitinated substrates to the proteasome to facilitate their degradation (**Figure 1**) (Dai and Li, 2001). The proteasome contains a catalytic protein complex referred to as the 20S core (Lowe et al., 1995), which is capped at each end by a regulatory protein complex (19S) (Finley, 2009; Zhang et al., 2009). It is responsible for breaking down substrate proteins into small peptides (Hough et al., 1987; Hadari et al., 1992).

The ubiquitination of substrate proteins is a reversible process. The removal of ubiquitin is carried out by De-ubiquitinating Enzymes (DUBs). DUBs play two important roles, allowing for the editing of existing chains (Wilkinson, 1997) and the removal of ubiquitin chains altogether from a substrate. As the presence of ubiquitin chains prevents substrates from entering the proteasome due to spatial restrictions, DUBs play an essential role in determining the rate of protein clearance in cells (Hershko and Ciechanover, 1998; Wilkinson, 2009), and certain DUBs including USP14 are known to associate directly with the19S regulatory complex of the proteasome (Borodovsky et al., 2001).

complexes including scaffolding and adaptor proteins that confer substrate specificity to the process of ubiquitin transfer. 4. Additional ubiquitin molecules can be added onto the first to create polyubiquitin chains on substrate proteins. 5. The 26S proteasome is composed of a cylindrical, proteolytic 20S core which is capped at both ends by a 19S regulatory cap. Polyubiquitinated substrate proteins, typically bearing K48-linked chains, are targeted to the proteasome by trafficking proteins. The proteasome digests these substrates into smaller peptides and free ubiquitin molecules, which then can be used to modify further substrates.

It is important to consider that while numerous studies report proteasome dysfunction in disease states, proteasomal degradation is only one of several potential outcomes of ubiquitination that may contribute to pathogenesis when impaired. Indeed, there are numerous examples of dysregulated protein handling due to failures in ubiquitination that do not necessarily implicate the proteasome. Whether the effects of proteasomal failure reach upstream to impact the activity and efficacy of E1s, E2s, or E3s is not clear. Given that K48-linked chains are among the most common found in cells (Ziv et al., 2011), it is conceivable that impaired proteolysis could deplete cellular pools of free ubiquitin and thereby induce further dysfunction. Accordingly, an imbalance in ubiquitination/deubiquitination activities may result in improper chain formation, preventing the proteasome from recognizing and handling targeted substrates. To examine these two separate yet related phenomena, it will be helpful to distinguish whether dysfunction occurs in UPS *per se* or, more broadly in the Ubiquitin Signaling System (USS) which includes the labeling of proteins with ubiquitin for any number of intended outcomes beyond degradation. While impaired clearance characterizes UPS failure, it remains to be seen whether and how the other, non-proteasomal components of the USS contribute to disease processes. One example of this distinction is evident in the ubiquitin-mediated fate of Caspase proteins.

Though the precise methods by which neurons degenerate in disease remain unclear, substantial evidence supports a role for proteases of the Caspase family. For example, activated caspases are elevated in AD patient tissue (Bredesen, 2009). Upon activation, caspases can either activate other proteases (initiator caspases) or damage essential components of the cell and promote apoptosis (effector caspases). Sublethal amounts of caspase activation have been linked to synaptic dysfunction in an animal model of AD (D'Amelio et al., 2011). Caspase-mediated effects can be inhibited both by the inactivation of caspase enzyme activity and by their targeted degradation through the proteasome. The X-linked Inhibitor of Apoptosis Protein (XIAP) and its family member cIAP1 are RING-type E3 ligases that directly bind to and target caspases for degradation; XIAP is also able to inhibit the proteolytic activity of caspases (Suzuki et al., 2001; Choi et al., 2009). The regulation of caspases by XIAP is limited, however, in oxidative stress conditions. Both acute and chronic inflammation, which are often associated with disease, elevate the levels of nitric oxide in neurons. This elevation can lead to the aberrant addition of nitric oxide to proteins (nitrosylation). Dysregulated nitrosylation of proteins is evident in several neurodegenerative diseases, including AD (Nakamura et al., 2013). The addition of nitric oxide (nitrosylation) to the active cysteine of XIAP's RING domain inactivates its E3 ligase activity. Nitrosylated XIAP is unable to ubiquitinate caspases and thus unable to inhibit apoptosis (Nakamura et al., 2010). The relative amount of nitrosylated XIAP is increased in AD patient tissue, suggesting a role in the accelerated apoptosis observed in disease (Nakamura et al., 2010). Here, then, is one example in which the state of the proteasome in disease is secondary to the decreased ability of E3 ligases to target caspases for degradation. Therefore, it is essential to consider how disease-related proteins are modified by agents of the USS while also examining whether affected neurons can execute the intended consequences of that modification.

# **UBIQUITINATION IN NEUROPATHOLOGY**

To further illustrate the complexity and fundamental significance of ubiquitin-mediated pathway involvement, we review key findings about the handling of disease-related proteins by the USS and UPS in several of the most common neurodegenerative diseases.

### **ALZHEIMER'S DISEASE**

Alzheimer's disease is the most common form of dementia and the most common neurodegenerative disorder. Its symptoms include a progressive decline in memory and other cognitive functions. Histologically, AD involves extensive neurodegeneration and loss of synaptic connections, resulting in progressive atrophy of the temporal, frontal and parietal lobes of the cerebral cortex. It is characterized by the hallmark deposition of intracellular, filamentous aggregates mainly consisting of hyper-phosphorylated Tau (neurofibrillary tangles, NFTs) and extracellular plaques rich in Amyloid-Beta (amyloid plaques) (Glenner et al., 1984; Goedert et al., 1988; Wischik et al., 1988). Ubiquitinated forms of Tau and Amyloid-Beta, as well as other ubiquitinated proteins, are major components of these aggregates (Perry et al., 1987). Amyloid-Beta, produced by the cleavage of the Amyloid Precursor Protein (APP), is thought to have numerous deleterious effects on neuronal health and connectivity (Thinakaran and Koo, 2008). Mutations in the genes encoding APP or the presenilin protease enzymes that cleave APP to generate Amyloid-Beta cause earlyonset AD, and support the notion that Amyloid-Beta metabolism is a central component of AD pathogenesis (Bertram et al., 2010). The generation of hyper-phosphorylated Tau is less clearly understood. It has been suggested that Amyloid-Beta is able to stimulate the kinase GSK3-B, resulting in the aberrant phosphorylation of Tau (Hernandez et al., 2010). The mechanisms governing the synthesis, processing, and degradation of these proteins remain incompletely understood. Inhibition of the proteasome causes an increase in Amyloid-Beta (Kumar et al., 2007), and numerous studies have begun to describe an extensive role for both ubiquitination and the proteasome in the production and handling of APP, Amyloid-Beta, and Tau (**Figure 2**).

APP is produced in the endoplasmic reticulum (ER). HRD1, an E3 ligase associated with the clearance of newly synthesized proteins through ER-associated degradation (ERAD), has been shown to interact with APP. Decreasing HRD1 expression evokes ER stress and apoptosis, accompanied by the accumulation of APP and Amyloid Beta. The disease relevance of the HRD1/APP relationship is supported by the reduced levels of HRD1 in AD brain tissue (Kaneko et al., 2010). Similarly, the levels of Fbxo2, a brain-enriched E3 ligase substrate adaptor protein also implicated in ERAD, have been reported to be decreased in AD patient tissues (Gong et al., 2010). Fbxo2 was also found to facilitate the degradation of APP, and a knockout mouse model for Fbxo2 revealed elevated levels of APP and Amyloid Beta in a brain region-specific manner (Atkin et al., 2014).

In the Golgi apparatus, APP is ubiquitinated by unknown E3 ligases stimulated by ubiquilin-1, a protein with chaperone-like properties. This ubiquitination is K63-linked and does not cause the degradation of APP. Instead, it causes the retention of APP in the early secretory pathway, impairing its maturation and delaying its subsequent processing by secretases into Amyloid Beta (El Ayadi et al., 2012). The process by which this ubiquitination of APP is normally removed to allow processing remains unclear. Ubiquilin-1 levels are significantly decreased in AD patient brain tissues, further suggesting a role in AD pathogenesis (El Ayadi et al., 2012). Single-nucleotide polymorphisms (SNPs) in the UBQLN1 gene have recently been linked to late-onset AD (Stieren et al., 2011).

After being produced and trafficked to the surface, APP is internalized via endocytosis into the endosome and trans-Golgi network and sequentially cleaved to produce Amyloid-Beta (Thinakaran and Koo, 2008). The C-terminus of HSP70 Interacting Protein (CHIP) is an E3 ligase previously associated with polyglutamine neurodegenerative disorders (Miller et al., 2005; Williams et al., 2009). In AD models, CHIP has been shown to interact with Amyloid-Beta in the Golgi, in a manner that increases upon inhibition of the proteasome. Over-expression of CHIP results in a decrease in Amyloid-Beta levels and may stabilize levels of APP (Kumar et al., 2007).

Beyond targeting APP or Amyloid-Beta directly, the USS affects numerous other components of the Amyloid pathway, including the secretase enzymes responsible for the production of Amyloid Beta. FBXW7 (SEL-10) facilitates the ubiquitination of the gamma-secretase component Presenilin 1 (PS1), although this ubiquitination unexpectedly increases Amyloid Beta production through mechanisms that remain unclear (Li et al., 2002). Fbxo2, described above, has also been implicated in the turnover

**FIGURE 2 | Regulation of protein trafficking, receptor signaling, and protein clearance by ubiquitination in Alzheimer's disease.** Improper processing of APP and Tau contribute to the pathology of AD. Excessive or improperly folded APP is cleared from the ER by HDR1 and Fbxo2 through the addition of K-48 linked chains. Ubiquitin-mediated processes are indicated by dashed lines. Maturation of APP is arrested in the early secretory pathway by non-degradative, K-63 linked ubiquitination that is stimulated by Ubiquilin 1. Surface APP is endocytosed to the late Golgi, where it is cleaved by secretases including BACE-1 and PS-1, whose levels are regulated by the E3 ligases Fbxo2 and FBXW-7, respectively. The cleavage of APP results in the production of Amyloid-Beta, which can be targeted for degradation by CHIP. Uncleared Amyloid-Beta is exocytosed to the extracellular space, where it aggregates to form plaques. Amyloid-Beta can influence NMDA receptor signaling. NMDA

of the Beta-secretase protein BACE1 (Gong et al., 2010). Similarly, overexpression of the E3 ligase Parkin, described below for its role in PD, has been shown to reduce APP expression, Amyloid-Beta burden, and inflammation while also restoring the formation of long-term hippocampal synaptic potentiation in an AD mouse model (Hong et al., 2014).

The effects of Amyloid-Beta on neurons are numerous, and can also influenced by the complex actions of the USS in more indirect methods than just the degradation of the Amyloid-Beta peptide or its precursor. For example, exposure to excess Amyloid-Beta causes diminished Brain-Derived Neurotrophic Factor (BDNF) signaling in AD patient tissue and mouse models (Peng et al., 2005); however, this diminution can be reversed by the overexpression of the de-ubiquitinating enzyme UCHL-1, which, through an unknown mechanism, restores the trafficking of BDNF receptors (Poon et al., 2013). Intriguingly, UCHL-1 levels are decreased in AD patient brains (Poon et al., 2013).

As another example, the intracellular response to Amyloid-Beta is thought to be mediated through numerous proteins including GSK3B, a kinase that is reportedly increased in AD and has Tau as one of its substrates. GSK3B has been investigated as a receptor activation stimulates the kinase Cdk5, which results in the downstream inhibition of the E3 ligase APC and blocks the degradation of cyclin B1. NMDA receptor signaling also increases the activation of GSK3B, which phosphorylates Tau. GSK3B is targeted for degradation by NUB1, which also blocks the interaction between GSK3B and Tau. GSK3B activity is increased by complexing with p53, whose levels are regulated by the E3 ligase MDM2. Under conditions of cellular stress, MDM2 auto-ubiquitinates and targets itself for degradation. The E3 ligase CHIP targets Tau, but can have divergent effects on its handling. When working in combination with Hsp70, CHIP targets Tau for degradation. However, when Hsp90 is involved, CHIP facilitates an alternative ubiquitination of unknown linkage type, resulting in the accumulation of phosphorylated tau. Through an unknown mechanism, this accumulated Tau then forms insoluble protein aggregates.

possible link between the two characteristic protein pathologies of AD (Hernandez et al., 2010). The kinase activity of GSK3B is enhanced by forming a complex with the tumor suppressor protein p53. Under normal conditions, this interaction is limited by the degradation of p53 by the E3 ligase Mouse double minute 2 homolog (MDM2). Under conditions of cellular stress, however, MDM2 levels are decreased, leading to the increased association of GSK3B and p53. This enhancement is thought to contribute to the hyperphosphorylation of Tau seen in AD (Proctor and Gray, 2010). The observed decrease in MDM2 under conditions of stress actually stems from its own governance by the UPS. Under normal conditions, MDM2 is modified by the small, ubiquitin-like modifier protein SUMO. Sumoylation of MDM2 prevents its auto-ubiquitination. Under conditions of cellular stress, sumoylation of MDM2 is diminished, causing it to become auto-ubiquitinated which then triggers its own degradation by the proteasome (Buschmann et al., 2000). The level of GSK3B in neurons is also mediated by yet another E3-ligase, Nedd8 ultimate buster 1 (NUB1). NUB1 directly binds to GSK3B and promotes its degradation by the proteasome, while also inhibiting the interaction between GSK3B and Tau. NUB1 activity thereby diminishes the levels of hyper-phosphorylated Tau and Tau aggregates (Richet et al., 2012).

NUB1 was originally identified for its role in regulating Nedd8 (Kamitani et al., 2001; Kito et al., 2001). Nedd8 is a small signaling protein similar to ubiquitin both in structure (approximately 60% identity and 80% homology to human ubiquitin) and in function (Kumar et al., 1993; Kamitani et al., 2001). The conjugation of Nedd 8, "neddylation," to cullin proteins promotes the ubiquitination activity of SCF complexes (Duda et al., 2008). Nedd8 also modifies other proteins, including transmembrane proteins such as APP, and may influence their degradation (Chen et al., 2012). Dysregulated clearance of Neddylated proteins in disease states is evinced by the accumulation of Nedd8 in ubiquitin-positive tau filamentous inclusions in some cases of AD and in Lewy bodies in PD (Dil Kuazi et al., 2003).

CHIP, described above for its role in the turnover of Amyloid-Beta, has also been shown to play a role in regulating phosphorylated Tau. Through its interaction with two heat-shock-induced proteins, Hsp70 and Hsp90, CHIP is able to ubiquitinate phosphorylated Tau. Under normal conditions, this ubiquitination leads to an accumulation of ubiquitinated tau, which then becomes aggregated into high molecular-weight, detergent-insoluble aggregates. CHIP immunoreactivity decorates NFTs in several tauopathies, including AD (Petrucelli et al., 2004). However, over-expression of Hsp70 shifts the handling of ubiquitinated Tau toward a pathway of clearance through the UPS, rather than aggregation, and reduces Tau levels *in vitro* (Petrucelli et al., 2004). Intriguingly, *in vitro* the ubiquitination of Tau also is carried out in an E3-ligase independent manner by the E2 Ube2w (Scaglione et al., 2013). Ube2w, whose levels are increased under cellular stress, also ubiquitinates CHIP and regulates its activity (Scaglione et al., 2011). Additionally, a role for CHIP in the ubiquitin-mediated turnover of caspases has been described. Mice that lack CHIP show increased levels of cleaved and uncleaved caspase-3 (Dickey et al., 2006).

Although not considered causative themselves, additional proteins have been shown to exacerbate disease pathogenesis, and are similarly subject to regulation by ubiquitination. NMDA receptors are ionotropic glutamate receptors whose regulated conductance of calcium into postsynaptic sites has been linked to learning and memory. Aberrant activation and signaling of these receptors has been implicated in numerous disease processes, including AD (Hardingham and Bading, 2010). It has been suggested that the Amyloid-Beta-induced loss of synapses requires the activation of extra-synaptic NMDA receptors containing the GluN2B subunit (Shankar et al., 2007; Talantova et al., 2013), underscoring the importance of regulating NMDA receptor levels and localization in AD pathology. GluN2B is ubiquitinated in response to synaptic activity by the E3 ligase Mind Bomb-2 (Jurd et al., 2008). Fbxo2, described above for its regulation of APP, also regulates the levels of NMDA receptors *in vitro* by facilitating the activity-dependent ubiquitination and elimination of the NMDA receptor subunit GluN1 (Kato et al., 2005).

The downstream effects of NMDA receptor activation on cell health include the regulation of ubiquitin-dependent pathways. In one example, following NMDA receptor activation, the cyclindependent kinase 5 (Cdk5) phosphorylates Cdh1, a key regulator of the E3 ligase complex APC; in doing so, the turnover of its cyclin B1 by APC is inhibited, promoting neurotoxicity following NMDA receptor activation (Maestre et al., 2008).

### **PARKINSON'S DISEASE**

Parkinson's disease is the second most common neurodegenerative disease and the most common neurodegenerative movement disorder. It is characterized by progressive abnormalities in gait and posture, as well as difficulty initiating and completing voluntary and involuntary movements. Histopathologically, PD involves the loss of dopaminergic neurons of the substantia nigra and locus ceruleus accompanied by astrocytosis and increased numbers of microglia. Throughout the brainstem of PD patients, proteinaceous intracellular aggregates called Lewy bodies can be found, comprised primarily of alpha-synuclein. With slightly altered appearance, these synuclein-containing aggregates can be seen in other brain regions, and the question of their contribution to disease pathogenesis remains an area of intense study (Syme et al., 2002). Although the majority of PD cases are of sporadic origin, the small fraction of inherited cases has provided valuable insight into the role of the USS and UPS in PD pathology (**Figure 3**).

Mutations in alpha-synuclein have been described in cases of familial PD (Kruger et al., 1998; Riess et al., 1998), and the proteasome is at least partially responsible for the turnover of this protein (Shimura et al., 2001). Proteasome dysfunction is reported in the substantia nigra in PD, and proteasomal inhibition itself can cause the formation of protein inclusions and eventual degeneration of neurons in the substantia nigra in rats, although with some inconsistency (McNaught and Jenner, 2001; McNaught et al., 2002; McNaught and Olanow, 2006). The expression of mutant alpha-synuclein induces the formation of filaments which interact directly with the 20S core of the proteasome and decrease its proteolytic activity (Lindersson et al., 2004). This deficit in proteasome function is ameliorated by the concomitant expression of the E3 ligase Parkin (Petrucelli et al., 2002).

Of patients with inherited PD, nearly 40 percent of those with an early onset of symptoms have mutations in the gene encoding Parkin (Lucking et al., 2000). First described in Japan, multiple mutations in Parkin have now been described in one or both alleles (Khan et al., 2003). While mutations in alpha-synuclein cause autosomal dominant PD, mutations in Parkin cause autosomal recessive PD. These mutations result in the loss of Parkin function, one aspect of which is to regulate the ubiquitination of alpha-synuclein. Because the accumulation of ubiquitinated alpha-synuclein is evident in both familial and sporadic PD, Parkin dysfunction may also play a role in idiopathic PD and the formation of Lewy bodies (Shimura et al., 2000; Hardy, 2003). Consistent with this view, targeted expression of mutant Parkin in disease-related brain regions of animal models yields similar pathologies to those observed in PD (Lu et al., 2009).

Parkin is also involved in the intricate regulation of prosurvival signaling through the Akt pathway by Epidermal Growth Factor (EGF) (Fallon et al., 2006). EGF-Akt signaling occurs through EGFRs which are decreased in disease-related brain regions of patients with PD (Iwakura et al., 2005). Accordingly, increased activation of EGFR in an animal model of PD prevents

**FIGURE 3 | Ubiquitination in pro-survival pathways, mitochondria stability, and protein clearance in Parkinson's disease.** Dysregulation of Parkin and UCHL1, both of which are associated with PD, has widespread effects on neuronal health. The Epidermal Growth Factor (EGF) binds to EGFR and initiates pro-survival signaling through the Akt and mTOR pathways. Ligand-binding to EGFR stimulates its ubiquitination, which allows for its UIM-dependent recognition by Eps15. Eps15 internalizes EGFR, allowing it to be trafficked to the proteasome for degradation. The EGFR-Eps15 interaction is blocked by the interaction of the UBL of the E3 ligase Parkin with the UIM of Eps15; this interaction increases Parkin's ligase activity, causing Eps15 to become ubiquitinated and dissociated from its

the loss of dopaminergic neurons (Iwakura et al., 2005). EGF signaling is regulated by ubiquitin in several ways. Ligand binding of EGFR stimulates Akt signaling, but triggers the endocytosis and degradation of EGFR, thereby limiting the extent of EGF signaling. This process is mediated by the activity-dependent ubiquitination of EGFR by an unknown E3 ligase, which makes it eligible for recognition by the ubiquitin-interacting motif (UIM) of the EGFR Protein tyrosine kinase Substrate #15, Eps15. Interaction with Eps15 is necessary for the internalization of EGFR, as Eps15 links ubiquitinated EGFRs to the machinery which traffics them to the proteasome for degradation (van Bergen En Henegouwen, 2009). Parkin, which contains an amino-terminal ubiquitin-like (UBL) domain, blocks the Eps15-EGFR interaction by binding to the UIM of Eps15 through this UBL domain, and then facilitating the ubiquitination of Eps15. In this manner, EGFR internalization is prevented, and EGF signaling and activation of pro-survival pathways are enhanced. But interaction between Eps-15 and Parkin suggests yet a further level of ubiquitindependent regulation. Eps-15 promotes its own ubiquitination by stimulating Parkin's E3 ligase activity upon interaction. This ubiquitination, which is believed to be K63-linked, prevents the interaction of ubiquitinated Eps15 with other ubiquitinated proteins such as EGFR and Parkin itself. Notably, Parkin deficiency UIM-dependent interactors. Downstream of EGFR signaling, mTOR's participation in the protein complex mTORC1 requires the ubiquitination of Raptor by DDB1 and Cul4. This ubiquitination is undone by the de-ubiquitinating enzyme UCHL1. Parkin also plays a role in the degradation of mitochondrial outer membrane proteins including Tom20, and through its involvement with PINK1 and Dj-1, facilitates the degradation of Synphilin-1 and itself. Synphilin interacts with alpha-synuclein, which can be degraded by Parkin in association with CHIP. Dimerized UCHL1 may also modify alpha-synuclein, though the result is non-degradative, K-63 linked ubiquitination which promotes its accumulation. Accumulated alpha-synuclein binds to the 20S core of the proteasome and inhibits proteasome function.

causes a reduction in EGF signaling (Fallon et al., 2006). Here, again, dysfunctional ubiquitin signaling, independent of proteasomal cleavage, is able to regulate the health of neurons in a disease context.

Parkin has been shown to function as part of a macromolecular E3 ligase complex which includes the pten-induced kinase (PINK1) and the peptidase Dj-1 (Xiong et al., 2009). Mutations in both PINK1 and Dj-1 cause hereditary PD (Abou-Sleiman et al., 2003; Valente et al., 2004). Together, these proteins facilitate the degradation of Parkin substrates. Synphillin-1, which interacts with alpha-synuclein and is also present in Lewy bodies, is one such substrate, as is Parkin itself (Xiong et al., 2009). Additionally, the interaction between PINK1 and Parkin facilitates the degradation and turnover of mitochondria (Greene et al., 2003; Vincow et al., 2013). In order to do so, Parkin targets proteins in the outer mitochondrial membrane for degradation, including Tom20 (Yoshii et al., 2011). These findings suggest that Parkin dysfunction may contribute to global cellular health problems, even beyond the accumulation of alpha-synuclein and synphillin-1.

Rare mutations associated with familial, early-onset PD are also found in the de-ubiquitinating enzyme UCH-L1, described above for its role in AD (Das et al., 2006). These disease-associated mutations reduce the DUB activity of UCHL1. UCH-L1's role is significant for the health and function of neurons through its regulation of substrates involved in governing mRNA transcription, protein translation, synaptic plasticity, and pro-survival signaling. These processes are potently influenced by the activity of a serine/threonine protein kinase originally identified for its susceptibility to inhibition by Rapamycin, the Mammalian Target of Rapamycin, mTOR (Sabatini et al., 1994; Hay and Sonenberg, 2004; Ghosh et al., 2008; Hoeffer and Klann, 2010). Intriguingly, mTOR function requires both the correct ubiquitination of interacting proteins and proteasome activity (Ghosh et al., 2008; Quy et al., 2013). mTOR asserts its effects through the formation of large protein complexes. mTOR can function in combination with Raptor (mTOR Complex 1 or mTORC1) or Rictor (mTORC2). mTORC1 regulates transcription and is important for local protein synthesis, a necessary component of several types of synaptic plasticity. mTORC2, on the other hand, influences Akt signaling, cytoskeletal dynamics, and actin polymerization (Costa-Mattioli and Monteggia, 2013). The balance between mTORC1 and mTORC2 formation depends on the interaction of Raptor with a multi-subunit E3 ligase composed of DNA Damage-Binding Protein (DDB1), Cullin 4A (CUL4), and the RING-type E3 ligase RING box 1 (RBX1) (Ghosh et al., 2008). Although the exact mechanism is still unclear, the DDB1-CUL4 complex appears to ubiquitinate Raptor in a manner that facilitates its incorporation into the mTORC1 complex including DDB1, CUL4, Raptor, and mTOR (Ghosh et al., 2008). UCHL1 deubiquitinates Raptor, thereby destabilizing mTORC1 complex formation and shifting the balance toward mTORC2-dependent signaling (Hussain et al., 2013). This shift may have a significant impact on the ability of neurons to modify their synaptic content to promote learning and memory, and may also represent a neuroprotective response mechanism.

The loss of UCHL1's DUB activity results in an accumulation of alpha-synuclein at presynaptic terminals rather than increased clearance (Cartier et al., 2012). Surprisingly, UCH-L1 had been proposed to act as an E3 ligase when it self-dimerizes; this new activity is independent of the state of UCHL1's DUB activity (Liu et al., 2002). Acting as an E3 ligase, UCHL1 causes the K63-linked ubiquitination of alpha-synuclein, which, in turn, worsens PD pathology (Liu et al., 2002). Accordingly, a mutant form of UCH-L1 with decreased E3 ligase activity upon dimerization, but normal DUB activity, decreases PD pathogenesis (Liu et al., 2002). The consequences of UCHL1's ubiquitination of alpha-synuclein remain unclear.

There have been several ubiquitin-linked proteins whose involvement in neurotoxicity overlaps between AD and PD. Of those described above, CHIP, MDM2, and HRD1 support the importance of the UPS in pathological mechanisms in multiple disease states. CHIP also interacts with Parkin and stimulates its ligase activity (Imai et al., 2002). Even without additional insult, the down-regulation of MDM2 is toxic to dopaminergic neurons (Nair et al., 2006). And HRD1 levels are increased in dopaminergic neurons under conditions of neurotoxicity, suggesting a role in the response to cellular stress (Mei and Niu, 2010).

## **AMYOTROPHIC LATERAL SCLEROSIS**

Amyotrophic Lateral Sclerosis (ALS) is the most common fatal neurodegenerative disease affecting motor neurons. The loss of motor neurons in the brain and spinal cord of ALS patients typically results in progressive paralysis and death within a few years of onset. Non-motor pathologies, particularly cognitive deficits, can also accompany motor deficits. Histopathologically, ubiquitin-rich cytoplasmic inclusions appear in the remaining motor neurons of ALS patients, accompanied by marked gliosis. Ubiquitin-positive inclusions can also be observed in cortical brain regions, consistent with the observed cognitive decline in many ALS patients (Lowe, 1994; Walling, 1999). Sporadic ALS (SALS) represents greater than 90 percent of all ALS cases, with familial ALS (FALS) accounting for the remaining *<*10 percent. The etiology of ALS remains unclear, and what is known to date has been determined largely through the study of genes originally identified as mutated in FALS. Importantly, the disease genes that cause FALS are also found mutated in some SALS cases, and the histopathologies of familial and sporadic cases are essentially indistinguishable (Andersen and Al-Chalabi, 2011), suggesting that studies of FALS will offer useful insight into all forms of ALS. Among the disease genes associated with FALS, mutations in the gene encoding Superoxide Dismutase 1 (SOD1) were described first (Rosen et al., 1993), and mutations in the Transactivation Response DNA-binding protein 43 (TDP-43) are among the most common (Gitcho et al., 2008). Recently, genome-wide association studies revealed a large hexanucleotide expansion—upwards of 1600 repeats—within the non-coding region of the C9ORF72 gene to be causative in a Finnish cohort (Laaksovirta et al., 2010). Numerous other studies have now observed this expansion in 20–50 percent of FALS cases, making it the most common cause (Boeve et al., 2012; Chio et al., 2012; Simon-Sanchez et al., 2012). The protein products of these genes accumulate in ubiquitinpositive inclusions in disease-related neuronal populations, and the mechanisms of their turnover and clearance provide insight into the importance of the UPS in ALS (**Figure 4**).

SOD1 plays an important role in the elimination of free superoxide radicals. Its expression is regulated by several agents of the UPS. The canonical HECT-type E3 ligase E6-AP directly binds to and ubiquitinates SOD1 (Mishra et al., 2013). Levels of E6-AP are decreased prior to the onset of neurodegeneration in a mouse model of ALS expressing mutant SOD1, and overexpression of E6-AP reduces the aggregation of mutant SOD1 *in vitro* (Mishra et al., 2013). NEDL1, a homolog of E6-AP, is reported to selectively bind to mutant SOD1 but not wildtype SOD1, facilitating its ubiquitination and clearance (Miyazaki et al., 2004). NEDL1 is thought to function in concert with the endoplasmic reticulum translocon-associated protein TRAP-δ, suggesting a role for NEDL1in the ERAD of mutant SOD1 (Kunst et al., 1997). The interaction between NEDL1 and mutant SOD1 increases with disease severity, and NEDL1 immunoreactivity is observed in SOD1-positive inclusions in human spinal cord tissue from FALS patients (Miyazaki et al., 2004). Also reportedly present in SOD1-positive inclusions is the RING-finger type E3 ligase Dorfin, which shares NEDL1's selectivity in ubiquitinating mutant, but not wild-type, SOD1 (Niwa et al., 2002). Over-Expression of Dorfin in a mouse model of FALS reduces

expression and aggregation of SOD1 while also diminishing motor neuron degeneration (Sone et al., 2010).

with HDAC6. This modification can include both K48 and K63-linked chains

The protein TDP-43, extensively aggregated in many in ALS cases, normally has a predominantly nuclear localization. It has two RNA binding domains and has been implicated in RNA splicing and trafficking (Buratti and Baralle, 2001, 2010). Ubiquitinated TDP-43 is a major component of cytoplasmic inclusions in ALS (Neumann et al., 2006). Mutations associated with ALS result in the redistribution of TDP-43 from the nucleus to the cytoplasm (Igaz et al., 2011), although the precise mechanisms by which TDP-43 contributes to ALS pathogenesis remain the subject of intensive study (Janssens and Van Broeckhoven, 2013). TDP-43 expression inhibits proteasome function and increases levels of Parkin mRNA and protein (Hebron et al., 2013) by binding to Parkin mRNA (Polymenidou et al., 2011). Parkin, in turn, is able to mediate both K48 and K63-linked ubiquitination of TDP-43, though this modification does not cause a reduction in the levels of TDP-43 (Hebron et al., 2013). TDP-43 is unlike other substrates of Parkin-dependent K63-linked ubiquitination, which are targeted for degradation through the autophagic pathway (Olzmann and Chin, 2008). The interaction between TDP-43 and Parkin requires the histone deacetylase protein HDAC6 and promotes the translocation of TDP-43 from the nucleus to the cytoplasm (Hebron et al., 2013).

In C9ORF72-related ALS cases, TDP-43 is similarly present in ubiquitin-positive, cytoplasmic aggregates (Mackenzie et al., 2014). However, a second ubiquitin-positive pathology is also observed: ubiquitin- and p62-positive inclusions that are TDP-43 negative, but stain positive for dipeptide products of abnormal transcripts from the hexanucleotide repeats in C9ORF72 (Mackenzie et al., 2014). These dipeptide-repeat (DPR) products are made by unconventional, repeat-associated, non-ATG initiated (RAN) translation of transcripts containing the hexanucleotide repeat expansion (Ash et al., 2013). The pathogenic contribution of these dipeptide products is as yet unclear, but their extensive colocalization with ubiquitin and p62 may suggest a failed attempt by the cells to clear them and the toxic response they may evoke (van Blitterswijk et al., 2012).

# **HUNTINGTON'S DISEASE**

colocalize with TPD-43.

Huntington's disease is the most common of the polyglutamine (polyQ) disorders. While polyglutamine repeats are common motifs that may facilitate protein-protein interactions, expansion of these repeats is associated with at least nine different neurodegenerative disorders (Ross, 1995; Orr, 2012). The threshold length at which this expansion begins to elicit adverse effects varies among these diseases, determined in part by the specific protein context neighboring the repeat in the expanded protein (Robertson et al., 2011). Expansion of the CAG repeat in exon 1 of the gene coding for the Huntingtin protein increases the length of the polyglutamine domain (Myers et al., 1993; Snell et al., 1993). The repeat length threshold for disease in HD is 38 or more repeats (Chong et al., 1997), with longer repeats causing earlier onset and more severe disease. Symptoms of HD include progressive cognitive impairment, mood disorder and other psychiatric symptoms, and highly disordered movement and motor control, including abnormal movements such as chorea. These progressive symptoms lead to mortality within 20 years of onset (Walker, 2007). Histopathologically, HD is characterized by the degeneration of striatal and cortical neurons and the presence of intraneuronal, intranuclear aggregates, and dystrophic neurites. Both of these latter features are marked by the presence of ubiquitinated Huntingtin. These aggregates can be found in cortical and striatal neurons, with some variance in cortical localization related to the age of onset (DiFiglia et al., 1997).

The normal function of Huntingtin remains unknown. Structural analyses and numerous other studies have led scientists to propose several functions including intracellular protein trafficking, modulation of gene transcription, and regulation of scaffolding proteins and NMDA receptors at the synapse (Cattaneo et al., 2005; Parsons et al., 2013). NMDA receptors are improperly localized to extra-synaptic sites in HD, and their signaling at those sites contributes to the onset of symptoms in HD animal models (Milnerwood et al., 2010; Gladding et al., 2012). This mislocalization decreases synaptic NMDA receptor activation, which has been shown to promote Huntingtin aggregation and neuronal survival (Okamoto et al., 2009). Whether HD arises exclusively from a toxic gain of function for mutant Huntingtin or from an additional partial loss of normal Huntingtin function remains unclear. In either case, ubiquitin plays a significant role in the handling of normal and mutant Huntingtin protein (**Figure 5**).

The presence of ubiquitinated Huntingtin suggests a failure of the UPS. Huntingtin is ubiquitinated (Kalchman et al., 1996), but it has been proposed that the proteasome may be unable to process expanded polyglutamine stretches, resulting instead in the accumulation of peptide fragments containing polyglutamine (Venkatraman et al., 2004). The fate of these peptide fragments is unclear, but they may remain associated with the proteasome and inhibit its function (Holmberg et al., 2004; Raspe et al., 2009). Intriguingly, components of the UPS, including proteasomes, have been identified within Huntingtin aggregates (Suhr et al., 2001). This may represent deleterious proteasomal sequestration in aggregates, even though aggregation of mutant Huntingtin may itself be neuroprotective (Holmberg et al., 2004; Okamoto et al., 2009). The idea that Huntingtin-induced pathology includes a deficiency in UPS-mediated clearance is supported by the beneficial effects observed following efforts to increase proteasomal activity in models of HD (Seo et al., 2007). The upregulation of UPS agents through the action of histone deacetylase inhibitors may also improve the aggregation phenotype of HD model mice (Jia et al., 2012).

The precise mechanisms by which Huntingtin is ubiquitinated are not well understood. It is unclear whether, like TDP-43, ubiquitination of Huntingtin includes both K48 and K63 linkages. Most studies investigating this question have used models overexpressing a fragment of Huntingtin with a polyQ expansion. While such over-expression models induce aggregation and toxicity, their physiological relevance to the human disease state remains an area of some debate. That said, in these model systems several components of the UPS contribute to the clearance of mutant Huntingtin. HRD1, described above for its role in several diseases, is implicated in the clearance of mutant Huntingtin. The activity of HRD1 increases with Huntingtin polyQ expansion length, suggesting that HRD1 regulation may not typically handle normal, unexpanded Huntingtin (Yang et al., 2007). The Tumor Necrosis Receptor Associated Factor 6 (TRAF6) is an E3 ligase which binds to both unexpanded and mutant Huntingtin and facilitates non-canonical ubiquitination through K6, K27, and K29-linked chains (Zucchelli et al., 2011). The physiological role of this modification is unclear, but it promotes the aggregation of mutant Huntingtin without changing the localization of the wild-type protein. Additionally, NUB1, described above for its role in AD, works in conjunction with Cullin 3 to facilitate the ubiquitination and clearance of mutant Huntingtin (Lu et al., 2013).

Sumoylation also can regulate HD pathogenesis. Rhes, a striatal protein which acts as a SUMO E3 ligase, binds to mutant Huntingtin selectively and facilitates its sumoylation (Subramaniam et al., 2009). The sumoylation of mutant Huntingtin decreases its aggregation; however, disaggregated, sumoylated, mutant Huntingtin inhibits transcription, increases cytotoxicity through Caspase-3, and may impair the induction of autophagy (Choo et al., 2004; Steffan et al., 2004; Mealer et al., 2013).

Huntingtin is also reported to interact with the E2 enzyme hE2-25k (Ube2K) (Kalchman et al., 1996). The presence of Ube2k immunoreactivity in HD patient brains has been observed, and the Huntingtin-Ube2k interaction promotes aggregation and cytotoxicity in a manner that requires E2 catalytic function (de Pril et al., 2007). Ube2k has been shown to interact with numerous RING-finger E3 ligases (Lee et al., 2001). It is possible that Ube25k cleaves the polyubiquitin chains attached to Huntingtin to an extent that degradation is no longer signaled.

# **CONCLUSIONS: TOO BIG NOT TO FAIL**

Befitting its extensive contributions to a wide range of cellular processes, the systems of protein ubiquitination and ubiquitindependent clearance are implicated at many levels of the most common neurodegenerative disorders. Is this implied involvement in disease incidental, or does dysregulated ubiquitination and clearance represent an inevitable and common pathway for the worsening of all neurodegenerative diseases? Because alterations in ubiquitin-dependent gene transcription, translation, control of protein quality and maturation, trafficking, mitochondrial turnover, and the handling of protein-protein interactions are found in combination in neurodegenerative disease, it seems likely that dysregulated ubiquitination will not remain limited to a single ubiquitin-dependent process in a given disease. Instead, the UPS and USS will be widely and progressively involved. To what degree this involvement might begin the pathogenic cascade is currently unclear. But based on the studies reviewed here, it seems more likely that through a complex web of dysfunction in the UPS and USS, involvement of these systems causes the pathogenicity of aberrant proteins to diversify and flourish, affecting additional systems and promoting the loss of synapses and cell death. As questions remain about the mechanism by which the deposition of abnormal protein progresses through regions of a diseased brain, mounting evidence strongly supports the view that this

**FIGURE 5 | Ubiquitin-mediated handling of the pathologic Huntingtin protein in Huntington's disease.** The Huntingtin protein is improperly cleared from neurons in HD. Mutant Huntingtin can be targeted for degradation in the ER by HRD1, and in the cytosol by NUB1. However, this ubiquitination can be edited by UCHL1, inhibiting the degradation of Huntingtin and promoting its aggregation in the cytosol. Traf6 facilitates the non-canonical ubiquitination of Huntingtin

through the formation of K6, K27, and K29-linked polyubiquitin chains, and these modifications selectively promote the aggregation of mutant Huntingtin. Aggregates of mutant Huntingtin include components of the UPS including proteasomes. Rhes promotes the sumoylation of Huntingtin, which reduces its aggregation, but causes sumoylated, mutant Huntingtin to interfere with other cellular processes including gene transcription in the nucleus.

spread occurs within the context of weakening protein quality control and ubiquitin-mediated pathways.

The UPS is not alone in its handling of unwanted or toxic proteins. The autophagic system is the other major pathway by which protein clearance is achieved. Autophagy especially regulates the turnover of organelles and aggregated protein species, and accordingly its role in neurodegenerative disease has been extensively pursued. Upon failure of the proteasome, autophagic mechanisms can be induced as a compensatory response to UPS inhibition in numerous neurodegenerative diseases (Garcia-Arencibia et al., 2010; Metcalf et al., 2012; Nixon, 2013). Further activation of autophagic mechanisms through pharmacologic or genetic manipulation has proven useful as a therapeutic intervention in model systems (Hochfeld et al., 2013). It seems evident, however, from the continued accumulation of ubiquitin-positive proteins in neurodegeneration, that autophagic induction is not sufficient to overcome this failure. Moreover, it is unclear what effect sustained, heightened autophagy might have on already weakened neurons.

The overlapping involvement of certain E3 ligases like HRD1, NUB1, and CHIP not only speaks to their significance as key regulators of proteostasis but also nominates them as important targets for research. While knocking out CHIP has been shown to exacerbate polyglutamine pathology (Williams et al., 2009), CHIP over-expression can reduce proteotoxicity and the effects of cellular stress in numerous models (Miller et al., 2005; Adachi et al., 2007; Lee et al., 2013). But CHIP induction without concurrent upregulation of Hsp70 might worsen tau pathology, suggesting that combinatorial approaches to therapy will be required.

In considering the specific goals and appropriate timing for interventions intended to prevent or delay disease onset, it is important to acknowledge that the loss of synapses can precede neurodegeneration and is thought to underlie many of the earliest cognitive impairments in numerous diseases. As such, it may represent a key step in the disease cascade and a critically important target for therapies. Alternatively, synaptic connections may simply be too costly for unhealthy neurons to properly maintain, and the loss of synaptic connections is, in effect, a response intended to limit inappropriate and potentially dangerous synaptic signaling in disease states. In the latter case, this dauer-like state is ultimately ineffective in staving off neurodegeneration, but may slow the process. With respect to human disease, perhaps restoring synapses could improve quality of life—regardless of whether such a change ultimately lengthens or shortens the life-span of affected neurons. For this reason, elucidation of the processes by which ubiquitin governs synapse formation, maintenance, and removal under normal conditions may prove invaluable.

### **REFERENCES**


mouse model. *J. Neurosci.* 27, 5115–5126. doi: 10.1523/JNEUROSCI.1242- 07.2007


in response to proteasome inhibition. *J. Biol. Chem.* 281, 39550–39560. doi: 10.1074/jbc.M603950200


Proctor, C. J., and Gray, D. A. (2010). GSK3 and p53 - is there a link in Alzheimer's disease? *Mol. Neurodegener.* 5, 7. doi: 10.1186/1750-1326-5-7

Quy, P. N., Kuma, A., Pierre, P., and Mizushima, N. (2013). Proteasome-dependent activation of mammalian target of rapamycin complex 1 (mTORC1) is essential for autophagy suppression and muscle remodeling following denervation. *J. Biol. Chem.* 288, 1125–1134. doi: 10.1074/jbc.M112.399949

Ramon y Cajal, S. (1909). *Histologie du Systeme Nerveux de L'homme & des Vertebres,* Vol. 2. *Paris: A. Maloine*.


Rosen, D. R., Siddique, T., Patterson, D., Figlewicz, D. A., Sapp, P., Hentati, A., et al. (1993). Mutations in Cu/Zn superoxide dismutase gene are associated with familial amyotrophic lateral sclerosis. *Nature* 362, 59–62. doi: 10.1038/362059a0

Ross, C. A. (1995). When more is less: pathogenesis of glutamine repeat neurodegenerative diseases. *Neuron* 15, 493–496. doi: 10.1016/0896-6273(95)90138-8


C9ORF72 hexanucleotide repeat expansions. *Brain* 135(Pt 3), 723–735. doi: 10.1093/brain/awr353


Hsp70-interacting protein (CHIP) supports an aggregation model of pathogenesis. *Neurobiol. Dis.* 33, 342–353. doi: 10.1016/j.nbd.2008.10.016


*Methanocaldococcus jannaschii*. *Mol. Cell* 34, 473–484. doi: 10.1016/j.molcel. 2009.04.021


**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.

*Received: 29 April 2014; accepted: 19 June 2014; published online: 08 July 2014. Citation: Atkin G and Paulson H (2014) Ubiquitin pathways in neurodegenerative disease. Front. Mol. Neurosci. 7:63. doi: 10.3389/fnmol.2014.00063*

*This article was submitted to the journal Frontiers in Molecular Neuroscience. Copyright © 2014 Atkin and Paulson. 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 Ubiquitin-Proteasome System: Potential Therapeutic Targets for Alzheimer's Disease and Spinal Cord Injury

### Bing Gong1,2 , Miroslav Radulovic1,2,3 , Maria E. Figueiredo-Pereira<sup>4</sup> and Christopher Cardozo1,2,3 \*

<sup>1</sup> Department of Medicine, Mount Sinai School of Medicine, New York, NY, USA, <sup>2</sup> Medicine, James J. Peters Veteran Affairs Medical Center, Bronx, NY, USA, <sup>3</sup> National Center of Excellence for the Medical Consequences of Spinal Cord Injury (SCI), Bronx, NY, USA, <sup>4</sup> Department of Biological Sciences, Hunter College, and the Graduate School and University Center, The City University of New York, New York, NY, USA

The ubiquitin-proteasome system (UPS) is a crucial protein degradation system in eukaryotes. Herein, we will review advances in the understanding of the role of several proteins of the UPS in Alzheimer's disease (AD) and functional recovery after spinal cord injury (SCI). The UPS consists of many factors that include E3 ubiquitin ligases, ubiquitin hydrolases, ubiquitin and ubiquitin-like molecules, and the proteasome itself. An extensive body of work links UPS dysfunction with AD pathogenesis and progression. More recently, the UPS has been shown to have vital roles in recovery of function after SCI. The ubiquitin hydrolase (Uch-L1) has been proposed to increase cellular levels of mono-ubiquitin and hence to increase rates of protein turnover by the UPS. A low Uch-L1 level has been linked with Aβ accumulation in AD and reduced neuroregeneration after SCI. One likely mechanism for these beneficial effects of Uch-L1 is reduced turnover of the PKA regulatory subunit and consequently, reduced signaling via CREB. The neuron-specific F-box protein Fbx2 ubiquitinates β-secretase thus targeting it for proteasomal degradation and reducing generation of Aβ. Both Uch-L1 and Fbx2 improve synaptic plasticity and cognitive function in mouse AD models. The role of Fbx2 after SCI has not been examined, but abolishing ß-secretase reduces neuronal recovery after SCI, associated with reduced myelination. UBB+1, which arises through a frame-shift mutation in the ubiquitin gene that adds 19 amino acids to the C-terminus of ubiquitin, inhibits proteasomal function and is associated with increased neurofibrillary tangles in patients with AD, Pick's disease and Down's syndrome. These advances in understanding of the roles of the UPS in AD and SCI raise new questions but, also, identify attractive and exciting targets for potential, future therapeutic interventions.

Keywords: ubiquitin-proteasome system, Alzheimer's disease, spinal cord injuries, beta-amyloid clearance, neuroregeneration

### Edited by:

Ashok Hegde, Georgia College & State University, USA

### Reviewed by:

Fred J. Helmstetter, University of Wisconsin—Milwaukee, USA Sokol V. Todi, Wayne State University School of Medicine, USA

\*Correspondence:

Christopher Cardozo christopher.cardozo@va.gov

Received: 07 September 2015 Accepted: 07 January 2016 Published: 26 January 2016

### Citation:

Gong B, Radulovic M, Figueiredo-Pereira ME and Cardozo C (2016) The Ubiquitin-Proteasome System: Potential Therapeutic Targets for Alzheimer's Disease and Spinal Cord Injury. Front. Mol. Neurosci. 9:4. doi: 10.3389/fnmol.2016.00004

# INTRODUCTION

The ubiquitin-proteasome system (UPS) is a major intracellular protein degradation system (Schwartz and Ciechanover, 2009). Initially, ubiquitin is activated through the ATP-dependent formation of a high energy thioester bond between the active site cysteine of the ubiquitin activating enzyme (E1), and the carboxyl terminus of ubiquitin. Ubiquitin is then transferred to ubiquitin conjugases (E2), and is finally conjugated to lysine residues within the substrate, or to the N-terminal amino group, by an ubiquitin (E3) ligase, which recognizes specific motifs in the targeted protein. Additional ubiquitin molecules are conjugated onto the first, forming a polyubiquitinated adduct that is recognized by the 26S proteasome leading to substrate degradation (**Figure 1**; Bence et al., 2001; Adams, 2003).

Dysfunction of the UPS is associated with many neurological diseases including Alzheimer's disease (AD), Amyotrophic Lateral Sclerosis (ALS), neuronal degeneration, remodeling and regeneration after spinal cord injury (SCI), Parkinson's disease (PD), Transmissible Spongiform Encephalopathies (TSE), and Huntington's disease (HD; Leigh et al., 1991; Neumann et al., 2006). The impact of the UPS in these disorders may be related to deficits in the clearance of misfolded proteins leading to intracellular protein aggregation, cytotoxicity, and cell death (Demuro et al., 2005; Schwartz and Ciechanover, 2009). Ubiquitinated proteins are detected in oligomeric beta-amyloid (Aβ) plaques and neurofibrillary tangles, and a mutation in the UPS associated gene UBB+1 causes neuronal degeneration, and has been linked to AD (van Leeuwen et al., 1998; Tan et al., 2007) and to spatial memory impairment (van Tijn et al., 2011).

There are interesting similarities in the role of the UPS in SCI and AD although these neurological conditions differ in etiology, age at onset, and region of the central nervous system (CNS) affected. AD is the most common cause of dementia and is a degenerative disorder of the characterized by memory loss, and neuropathologically by accumulation of Aβ in neurofibrillary tangles. SCI is trauma whereas AD is a degenerative disorder of older individuals. SCI results from traumatic injury to the spinal cord that is most commonly due to motor vehicle accidents, sports or falls, and which results in partial or complete loss of sensation and voluntary movement below the site of the injury. SCI is most common in younger men although a bimodal distribution is now recognized reflecting the growing number of new cases of SCI in the elderly. SCI affects approximately 275,000 individuals in the united states with approximately 12,000 new cases each year. UPS dysfunction is closely associated with the pathological hallmarks of AD including the accumulation of Aβ, formation of Aβ plaques, and hyper-phosphorylation of Tau, which is the major component of intracellular neurofibrillary tangles. The UPS is also involved in neuronal signaling pathways that regulate neurotransmitter release, synaptic membrane receptor turnover and synaptic plasticity (Zhao et al., 2003; Patrick, 2006). Compelling evidence indicates that the UPS also plays a role in the recovery from SCI (Coleman and Perry, 2002). While the accumulation/aggregation of ubiquitinated protein is a biomarker in the early response to SCI due to the impairment of proteasome function (Li and Farooque, 1996; Myeku et al., 2012), an increase in ubiquitin levels is essential for neuronal regeneration after SCI (Schwartz and Ciechanover, 2009; Noor et al., 2011). Furthermore, the UPS also has a role in neuronal development, differentiation, and programming of neuro-stem and neuroprogenitor cells. Regulation of Aβ production by the UPS may have a role in both AD pathogenesis and recovery from SCI. There are several excellent reviews that have discussed the mechanisms and relationships of UPS and tau phosphorylation in AD development (Ciechanover and Kwon, 2015; Gentier and van Leeuwen, 2015; Kumar et al., 2015), and also many reviews that highlight deubiquitination enzymes and the E3 ligases in the role of substrate recognition and ubiquitination. In this review, we have focused on several recent areas of new research: the role of ubiquitin C-terminal hydrolase L1 (Uch-L1), Ubiquitin-B+1 (UBB+1), F-box protein 2 (Fbxo2), and aggregation-promoting chaperones (CRAM-1 and MOAG-4) in Amyloid precursor protein (APP) and Aβ related metabolism as well as the related signaling pathways in AD and SCI, and discuss pressing questions in these fields of research. We also discuss the possibility that one or more of these molecules are potential targets for pharmaceuticals.

# TARGETING Uch-L1 AND Fbx2 IN AD

# Uch-L1 Regulates the CREB Signaling Pathway in AD

UPS dysfunction slows or stops the degradation of many proteins, including misfolded proteins, and results in protein aggregation in the cytoplasm, nucleus and extracellular space of neurons. Formation of inclusion bodies disrupts cell homeostasis, axonal transport, synaptic receptor activity, and neurotransmitter release (Ehlers, 2003; Jarome et al., 2013), and also results in mitochondrial swelling and damage (Maharjan et al., 2014), impaired cell differentiation, degeneration, and neuronal death via apoptosis or necrosis (Schwartz and Ciechanover, 2009; Kaang and Choi, 2012). In neurons, the cyclic adenosine 3<sup>0</sup> ,50 -monophosphate (cAMP)-cAMPdependent protein kinase (PKA)-cAMP response elementbinding protein (CREB) signaling pathway is implicated in synaptic plasticity and cognitive function, and is regulated by the UPS through degradation of the regulatory subunit of PKA (Vitolo et al., 2002). The activation of PKA and the increase of CREB phosphorylation are essential for the formation of stable long term memory (Chain et al., 1999; Fioravante and Byrne, 2011). This signaling pathway has been shown to be compromised by Aβ both in vitro, using oligomeric Aβ treated cultured neurons or brain slices, and in vivo as evidenced by Aβ overproducing transgenic AD mouse models (Vitolo et al., 2002; Gong et al., 2004; Smith et al., 2009). Consequently, a decrease in PKA-pCREB levels may contribute to impairment of synaptic plasticity and learning function in AD (Gong et al., 2006; Atkin and Paulson, 2014). It was postulated that compounds that enhance levels of pCREB in brain, such as cAMP phosphodiesterases (PDE) 4, 5 inhibitors, have beneficial

effects on cognitive function (Gong et al., 2004; Navakkode et al., 2004).

In the search for UPS regulators to modulate PKApCREB levels in the brain, it was found that in the snail Aplysia, the ubiquitin C-terminal hydrolase (Ap-Uch, a homolog of vertebrate Uch-L1) enhances the degradation of the polyubiquitinated R subunit of PKA (Hegde et al., 1997). Uch-L1 is an abundant deubiquitinase (1–5% of soluble protein in brain) that is expressed in neurons, testis and ovaries (Wilkinson et al., 1989; Osaka et al., 2003). Uch-L1 exists as both a soluble form in the cytosol, which accounts for 70–80% of Uch-L1 protein, and a membrane-bound form (Bishop et al., 2014). While the reduction of cytosolic Uch-L1 may be linked to the pathogenesis of AD by contributing to abnormal tau metabolism and Aβ plaque, membrane bound Uch-L1 may be associated with α-synuclein dysfunction in PD (Liu et al., 2009; Chen et al., 2013). Uch-L1 is a highly conserved protein with a catalytic Cys at residue 100 and active site His at residues 107 and 181. The major proposed functions of Uch-L1 are: (1) cleavage of α- and ε-amino acid linked ubiquitin molecules from polyubiquitin chains or polyubiquitinated proteins, respectively (Wilkinson et al., 1989; Larsen et al., 1998) to provide a pool of monoubiquitin and regulate ubiquitin homeostasis; and (2) directly binding to mono-ubiquitin to mask mono-ubiquitin activating sites such as Glu64, Phe4, Leu8 and Leu43 (Nakatsu et al., 2000; Shih et al., 2000) to avoid the mono-ubiquitination of substrates that would otherwise shunt them to the endosomallysosome pathway to be degraded (Osaka et al., 2003). Uch-L1 C-terminal hydrolase activity can be measured in vitro by an enzyme assay using the fluorogenic substrate Ub-AMC, and its activity can be effectively inhibited by the specific inhibitor LDN-57444 (Wilkinson et al., 1989; Gong et al., 2006). Interestingly, Uch-L1 also acts as an ubiquitin ligase and through such activity may have links with PD (Liu et al., 2002, 2006). Studies in AD brains have shown that Uch-L1 deficits are linked to the accumulation of Aβ in the ascending gracile tract as demonstrated by a mouse model of gracile axonal dystrophy (GAD; Osaka et al., 2003), and there is an association between Uch-L1 gene S18Y polymorphisms and sporadic AD (Xue and Jia, 2006; Zetterberg et al., 2010). Uch-L1 levels decrease in postmortem brains of AD patients and in AD transgenic mouse models, coinciding with the accumulation of ubiquitinated protein in Aβ plaques and neurofibrillary tangles (Gong et al., 2006; Zetterberg et al., 2010). Similar to Ap-Uch, human Uch-L1 increases the ubiquitination of the R (regulatory) subunit of PKA in the neuronal cytoplasm by providing mono-ubiquitin to promote proteasomal degradation of the R subunit, thus freeing the PKA catalytic subunit to phosphorylate CREB in the brain (Vitolo et al., 2002; Poon et al., 2013). Uch-L1 also modulates the turnover of the glutamate receptors NMDAR and AMPAR, and has effects on neurotransmitter release (Cartier et al., 2009). Enhancing Uch-L1 expression in the brain could have beneficial effects on synaptic function and improve cognition in AD. This premise is supported by studies showing that overexpressing Uch-L1 or pharmacologically enhancing Uch-L1 activity, improves long term potentiation (LTP) and cognition in AD transgenic mouse models (Gong et al., 2006), whereas knocking out Uch-L1 or pharmacologically inhibiting Uch-L1 causes a decline in cognition (Kurihara et al., 2001; Gong et al., 2006; Chen et al., 2010; **Figure 2**). This suggests that factors regulating Uch-L1 activity may be potential targets for AD therapeutics. However, more studies are needed to uncover (1) what other signaling pathways that are regulated by Uch-L1 and (2) the effect of such regulatory influences on neuronal function. For example, it remains to be determined how Uch-L1 is involved in the activation of the transcription factor of nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) or in tumor necrosis factor (TNF)-induced necroptosis.

# Uch-L1/UBB+ are involved in the Regulation of Aβ Production

Earlier studies failed to demonstrate a significant effect of Uch-L1 manipulations on APP processing and Aβ production in the brain (Gong et al., 2006). However, recent publications show that Uch-L1 is indeed involved in the degradation of APP and beta-site APP cleaving enzyme 1 (BACE1), one of the two essential enzymes cleaving APP and producing Aβ in brain. Overexpressing Uch-L1 reduces BACE1 activity and Aβ levels, whereas Uch-L1 null mice exhibit increased number of Aβ plaques and neurofibrillary tangles, thus displaying an inverse relationship with Uch-L1 levels (Choi et al., 2004; Qing et al., 2004). The discrepancy in the results of various

complex toward the inactive tetramer. As a consequence, the transcription factor CREB cannot be phosphorylated and initiates transcription. Uch-L1 re-establishes normal proteasomal activity leading to normal levels of RIIa subunit which in turn phosphorylates CREB, thus rescuing synaptic function.

Gong et al. UPS in AD and SCI

trials regarding the effect of Uch-L1 on Aβ may be due to different dosages and durations of the Uch-L1 treatment in the experimental mice. The mechanism by which Uch-L1 reduces Aβ accumulation and toxicity is unclear. It is postulated that it could be linked to increased APP degradation in Golgi or ER, thus reducing APP levels available to be cleaved by BACE1. Uch-L1 may also affect APP axonal translocation to synaptic terminals, thus reducing Aβ induced neuronal toxicity at the synapse. The latter is consistent with Uch-L1 improving synaptic plasticity in AD mouse models. Thus, there may be multiple pathological mechanisms contribute to the cognitive dysfunction and development of AD. Moreover, drugs may improve cognitive function without altering Aβ levels, which is in agreement with reports suggesting that the clinical phenotype is not always closely related with the Aβ plaque numbers or load in postmortem AD brains (Aizenstein et al., 2008).

A frameshift mutation in the ubB (ubiquitin) gene (UBB+1) selectively expressed in brain was associated with increases in neurofibrillary tangles and plaques in AD, Pick's disease and Down's syndrome patients, but was not found in the healthy young population, which suggests a link between the UPS and AD pathogenesis (van Leeuwen et al., 1998, 2006). The polyubiquitin chain, which is essential for degradation of most substrates by the 26S proteasome, is formed through isopeptide bonds between the carboxyl terminus of Gly76 of one ubiquitin and the amino group of the lysine residues at positions 48 (K48) of another ubiquitin molecule (Chau et al., 1989). UBB+1 is a result of a dinucleotide deletion during transcription causing frameshift that replaces the glycine at the C-terminus of wild-type ubiquitin with a 19 residue insert (UBB+). The precise mechanism by which UBB+1 influences protein degradation is not clear. While K48 is preserved, UBB+1 cannot form polyubiquitin chains because of the absence of a C-terminal Gly; overexpression of UBB+1 results aggregation of ubiquitinated protein in UBB+1 overexpressing cells (van Tijn et al., 2007) and mice (Irmler et al., 2012). Moreover, UBB+1 conjugated proteins are toxic and directly inhibit 26 proteasome activity, causing aggregation of ubiquitinated protein (Ciechanover and Kwon, 2015). This aberrant UBB+1 has been linked with the accumulation of Aβ and tau in the brain in AD in both post-mortem brains from patients and transgenic mice (Lindsten et al., 2002; Shabek et al., 2009; Dennissen et al., 2010). Studies with AD mouse models showed that knockdown of UBB+1 decreases the steady-state levels of APP, as well as the production of sAPPα, APP C-terminal fragments (CTF), C83 and APP intracellular domain (AICD; Gentier and van Leeuwen, 2015). Conversely, overexpression of UBB+1 increased the levels of APP and Aβ in cultured neuron cells, possibly by reducing the degradation of APP by proteasomes, thus shifting APP into an Aβ production pathway (Massey et al., 2004; Hiltunen et al., 2006; Viswanathan et al., 2011). Therefore, UBB+1 is involved in both secretase-dependent and independent pathways for Aβ generation and this could be a reasonable explanation as to why UBB+ conjugated protein aggregation has been found to be associated with Aβ plaques in transgenic UBB+1 AD brain (van Leeuwen et al., 2006; Zhang et al., 2007; El Ayadi et al., 2012). Since Uch-L1 and UBB+1 exert opposite effects on Aβ processing, metabolism and polyubiquitin chain formation, pharmacological modulation of UBB+1 or Uch-L1 levels in the brain may be a potential therapeutic strategy for AD.

Because Uch-L1 has been found play a role in synaptic function and BACE1 degradation (see below), it will be interesting to know whether UBB+1 has the opposite effect on synaptic plasticity in the AD brain, what signaling pathways are involved, and if UBB+1 also linked BACE1 degradation. Further studies in animal models produced by crossing the UBB+1/Uch-L1 with AD mouse models such as APP or APP/S1 are needed. These studies will provide the basis for screening compounds affecting Uch-L1 and UBB+1 levels that could improve synaptic and cognitive function in AD patients.

# Role of the E3 Ligase Fbox2 in Aβ Processing and Synaptic Function

The UPS is postulated to be involved in the degradation of APP and the attenuation of Aβ accumulation in the brain (Kang et al., 2010; Zhang et al., 2012). Several E3 ligases have been reported to be linked with APP and Aβ processing such as HRDI (Kaneko et al., 2010), the C-terminus of HSP70 interacting protein (CHIP; Kumar et al., 2007) and FBXW7 (Li et al., 2002) which is possibly regulated through the ER associated protein degradation pathway (ERAD). Reductions of E3 ligase function may cause APP accumulation and Aβ generation, whereas, overexpression of these E3 ligases may result in decreased level of Aβ in in vitro models. This field has been recently reviewed in depth (Atkin and Paulson, 2014). Here, we will focus our discussion on a newly identified, neuron-specific F-box protein, Fbx2. Fbx2 was found to contribute to the rapid elimination of neuron-specific membrane proteins, including neurotransmitter receptors and postsynaptic scaffolding proteins, by promoting their ubiquitination (Yoshida et al., 2003; Lin and Man, 2013; Atkin et al., 2014). Fbx2 is a member of the SCF-Fbx2-E3 ligase complex. Through its FBA (F-box associated) domain, Fbx2 attaches polyubiquitin chains to high-mannose N-linked glycosylated substrates, thus targeting them for proteasomal degradation via the ERAD pathway (Nelson et al., 2006). Recently, we found that Fbx2 specifically recognizes and binds BACE1 through its FBA domain, thus increasing BACE1 ubiquitination (Gong et al., 2010). Normally, Fbx2 is expressed in various intracellular organelles in brain neurons, and BACE1 is co-localized with Fbx2 mainly in the early endosome and ER (Ko and Puglielli, 2009). Moreover, APP, which is cleaved by BACE as a critical step in the production of Aβ in neurons, is located in the endosome and Golgi complex, and is co-localized with BACE1. During metabolic stresses such as increased oxidative phosphorylation or changes in cytoplasmic homeostasis that induce ER stress, the UPS will be compromised, which would most likely result in a reduction in BACE1 degradation by the UPS and lysosomes. The increased levels of BACE1 therefore are associated with an increase in APP cleavage and overproduction of Aβ (O'Connor et al., 2008; Gong et al., 2010; Singh and Pati, 2015).

We established that Fbx2 overexpression in primary neurons derived from Tg2576 transgenic mice, a well characterized AD mouse model (Hsiao et al., 1995), significantly promoted BACE1 degradation and reduced Aβ production. Not surprisingly, Fbx2 overexpression regulated LTP in the hippocampus. In our in vivo studies with Tg2576 mice, we found that stereotactic delivery of adenoviral-Fbx2 to the brain significantly increased Fbx2 expression and decreased synaptic deficits coinciding with a reduction of BACE1, which is essential for Aβ formation (Gong et al., 2010, 2013b). Moreover, we and others demonstrated that in addition to BACE1, the SCF-Fbx2-E3 ligase is involved in binding and ubiquitination of the NMDA and AMPA receptors at the synaptic membrane, affecting receptor turnover during neuronal activation (Kato et al., 2005; Gong et al., 2010). Recently, it was reported that APP is also one of the substrates for Fbx2 ligase (Atkin et al., 2014). APP levels are decreased in the Fbx2 overexpressing cells and increased in hippocampal neurons of Fbx2 knockout mice. The increase in total APP and decrease in surface membranebounded APP coincides with higher levels of Aβ in the Fbx2 knockout mice due to subsequent cleavage by β- and γsecretases.

In summary, there may be two major mechanisms by which Fbx2 regulates synaptic plasticity in AD mouse models: one is through a direct regulation of synaptic neurotransmitter release and receptor turnover; the second may be due to stimulation of the degradation of BACE1/APP to reduce Aβ levels in the brain. Therefore, the Fbx2 E3 ligase could be another potential target for AD treatment. For example, we previously reported that nicotinamide riboside, a potent inducer of Fbx2 expression in neurons, significantly increased the binding of BACE1 to ubiquitin, enhanced BACE1 degradation by proteasome, and thus reduced Aβ toxicity (Gong et al., 2013a). Thus, it would be meaningful to screen compounds that have similar function as nicotinamide that able to boost Fbx2 expression/function in neuronal cells, and then apply these compounds in animal models to test their in vivo functions and effects.

# Functional Links between the Ubiquitin-Proteasome System and Mitochondrial Impairment in AD

It is clear that the function of the UPS is perturbed in AD (Ding et al., 2006; Oddo, 2008; Riederer et al., 2011). Defective proteasome activity is detected in the early phase of AD along with synaptic dysfunction, and in late AD stages is associated with the accumulation and aggregation of ubiquitinated (Ub) proteins which precedes tangle formation (García Gil et al., 2001; Upadhya and Hegde, 2007; Oddo, 2008; Bedford et al., 2009). Mitochondrial impairment is also involved in AD (Eckert et al., 2011; Stranahan and Mattson, 2012; Selfridge et al., 2013). In particular, reduced brain glucose metabolism is detected early in AD patients and is implicated in the initiation and progression of AD pathology (Mosconi et al., 2008). In neurons, even a modest restriction of ATP production by mitochondria seems to far outweigh the effects of reactive oxygen species (ROS; Nicholls, 2008). We will discuss the functional link between UPS and mitochondrial dysfunction, and its impact on AD.

The UPS plays a critical role in the pathogenesis of AD (Riederer et al., 2011). Proteasome impairment could be an upstream event leading to the AD pathogenic cascade. Supporting this notion, we showed that in rat cerebral cortical cultures, short term (4–8 h) proteasome inhibition results in the accumulation of soluble Ub-proteins without significant loss in neuronal viability (Metcalfe et al., 2012). However, longer proteasome inhibition (16–24 h) induces a rise in Ub-protein aggregates, coinciding with caspase-activation, caspase-cleavage of the microtubule associated protein Tau to an aggregation prone form, and neuronal death (Metcalfe et al., 2012). The identity of the factors that induce the transition from soluble to aggregated Ub-proteins is poorly defined. Recent studies with the nematode C. elegans, revealed a novel protein known as ''cytotoxicity-related aggregation meditaor-1'' or CRAM-1 that acts as an aggregation modulator (Ayyadevara et al., 2015). Glutamine/asparagine-rich proteins, which favor hydrophobic interactions with random-coiled domains, seem to be the most susceptible to aggregation (Ayyadevara et al., 2015). CRAM-1 is a highly disordered protein lacking defined protein motifs, and has the ability to sequester unstructured or unfolded proteins. CRAM-1 binds to the polyubiquitin tags of these proteins and blocks their proteasomal degradation and/or targeting to autophagy. This crude sequestration mechanism ultimately leads to an unsustainable aggregate burden (Ayyadevara et al., 2015). Another C. elegans protein that plays a similar sequestration role, but functions independent of ubiquitin binding, is MOAG-4 (modifier of aggregation 4) which is homologous to human SERF1 (van Ham et al., 2010). MOAG-4/SERF1 is a specialized and autonomous amyloid factor that promotes aggregation of a broad range of amyloidogenic proteins but is inactive against non-amyloid aggregation (Falsone et al., 2012). Both of these primitive chaperones, CRAM-1 and MOAG/SERF1, impair survival and could act as non-competitive inhibitors of the proteasome in an age-dependent manner relevant to AD (Ayyadevara et al., 2015).

The high levels of oxidized proteins detected in the brains of AD patients are an indirect indication of proteasome impairment, since this proteolytic complex degrades the majority of oxidatively modified proteins (Grune et al., 2004). Impaired clearance of oxidatively modified proteins can cause their aggregation and directly promote the progression of the neurodegenerative process (Grune et al., 2004). Moreover, the accumulation and aggregation of Ub-proteins detected in most neurodegenerative disorders including AD, is also a sign of UPS dysfunction, since this pathway degrades Ub-proteins (Alves-Rodrigues et al., 1998). Impairment in proteolysis mediated by the UPS has profound ramifications for neuronal viability. It leads to the accumulation of modified, potentially toxic proteins in neurons and can cause neuronal injury or premature neuronal death by apoptosis or necrosis (Dasuri et al., 2013). While it is accepted that proteasome function decreases in AD, the question remains why is there a decline in regulated protein degradation by the UPS under these conditions.

One of the emerging factors associated with UPS impairment is mitochondrial dysfunction (Livnat-Levanon and Glickman, 2011). A mechanism by which mitochondrial dysfunction contributes to aging and neurodegeneration is via a net increase in the production of ROS (Zhu et al., 2013). Oxidative stress induced by ROS alters the structure of cellular proteins (Stadtman and Berlett, 1998), which, if not repaired, must be removed by proteolysis to prevent their accumulation and aggregation. Although one of the major roles of the proteasome is to degrade oxidatively modified proteins, whether ubiquitination is required remains elusive (Shang and Taylor, 2011). This is an interesting issue since there is ample evidence that neural tissues are especially vulnerable to oxidative stress, which is linked to AD pathogenesis (Eckert et al., 2011). Neurons exhibit a higher sensitivity to proteasome inhibition than astrocytes, mostly because they exhibit increased levels of oxidized proteins (Dasuri et al., 2010). Interestingly, cysteine deubiquitinases are proposed to be highly sensitive redox biosensors that become inactive and promote a quick response independent of protein synthesis or protein modification (Cotto-Rios et al., 2012; Eletr and Wilkinson, 2014). Furthermore, K63-linked polyubiquitination was proposed to be a new modulator of the oxidative stress response induced by peroxides (Silva et al., 2015). In S. cerevisiae, three different components of the UPS, the Ub-conjugase Rad6, the Ub-ligase Bre1, and the deubiquitinase Ubp2, were shown to work in concert to regulate K63-linked polyubiquitination of various ribosomal proteins (Silva et al., 2015). Disrupting K63 linked polyubiquitination under peroxide-induced oxidative stress destabilizes polysomes, impairs protein synthesis, and increases the vulnerability of cells to stress (Silva et al., 2015). Thus K63-linked polyubiquitination seems to play a new role as a redox-regulator of protein translation by ribosomes, while K48-linked polyubiquitination regulates protein degradation by proteasomes.

Another deleterious mechanism associated with mitochondrial dysfunction is the limitation in ATP production that can cause an energy crisis in neurons (Nicholls, 2008). Degradation of proteins by the 26S proteasome is highly dependent on ATP binding and hydrolysis (Liu et al., 2006). We investigated the effects of mitochondrial impairment in rat cerebral cortical neurons treated with oligomycin, antimycin, and rotenone, all of which inhibit different elements of the mitochondrial electron transport chain and thus cause a significant decline in ATP (Huang et al., 2013). Our study revealed that a decline in ATP leads to multiple adverse effects: (1) A shutdown of the ubiquitination cascade occurred and was caused by inhibition of ubiquitin adenylation carried-out in an ATP-dependent manner by the E1 ubiquitin activating enzyme; (2) Disassembly of the 26S proteasome was observed and was triggered in part by the selective cleavage of Rpn10 by calpain, which is activated under conditions of ATP deficit. No other proteasome subunit tested was cleaved by calpain. Rpn10 could act as a 26S proteasome gatekeeper sensitive to ATP deficits; (3) There was, accordingly, an increase in the activity of the 20S proteasome, which in concert with immunoproteasomes is postulated to degrade oxidatively modified proteins (Grune et al., 2004); and (4) Cleavage of Tau in a calpain-dependent manner was found, leading to a ∼17 kDa fragment which is presumed to be toxic (Garg et al., 2011). These data are highly significant to AD because: (1) mitochondrial dysfunction (Selfridge et al., 2013) and calpain-activation are linked to AD (Saito et al., 1993; Ferreira and Bigio, 2011); and (2) calpain-dependent Tau fragments are detected in the brains of AD patients (Ferreira and Bigio, 2011; Garg et al., 2011). In conclusion, both deleterious consequences of mitochondrial impairment, i.e., restricted ATP generating capacity and ROS production, impair UPSdependent proteolysis in neurons and are highly relevant to the pathogenesis of AD.

Not only are mitochondria associated with UPS function in regards to ATP production and the generation of ROS, the UPS is also linked to mitochondrial function, as recent studies revealed that the UPS degrades mitochondrial proteins and thus contributes to mitochondrial quality control (Taylor and Rutter, 2011). Mitochondrial proteins that are UPS substrates include: (1) damaged and/or misfolded nuclear-encoded proteins that are destined for import into the mitochondria; (2) defective proteins at the outer mitochondrial membrane (OMM) extracted by p97 and delivered to the proteasome; and (3) non-OMM proteins, although the mechanistic details of how these proteins in the inner compartments retrotranslocate to the OMM remain poorly defined (Farhoud et al., 2012). UPS impairment impacts several aspects of mitochondrial function that include: (1) mitochondrial dynamics (Carlucci et al., 2008) via the scaffold protein AKAP121; (2) mitochondrial biogenesis (Farhoud et al., 2012) via Peroxisome proliferator-activated receptor-gamma coactivator (PGC-1α); and (3) mitochondrial fission via Drp1 (Kanamaru et al., 2012). These factors are turned over by the UPS, and are perturbed in AD by either reduced expression, mis-localization, or an interaction with Tau (Carlucci et al., 2008; Trausch-Azar et al., 2010; Merrill et al., 2011; Manczak and Reddy, 2012; Sheng and Cai, 2012; Sheng et al., 2012).

In conclusion, it is clear that there is a mutual dependance between the UPS and mitochondria (Livnat-Levanon and Glickman, 2011). New discoveries regarding the functional links between UPS and mitochondrial impairment and their roles in overall neuronal homeostasis could lead to novel and successful therapeutic approaches for AD (Huang and Figueiredo-Pereira, 2010). Thus, future studies are needed to better understand how PGC-1α promotes the mitochondrial biogenesis and how the proteasome affects the inner compartments of mitochondrial retrotranslocation to the OMM, and more practically, to test whether stabilization of K63-linked polyubiquitination is able to resist the oxidative stress induced impairment of mitochondrial function in animal models.

# THE UPS IN SPINAL CORD INJURY

# Ubiquitin and Uch-L1 Participate in Neuronal Regeneration After SCI

SCI most commonly results from traumatic injuries to the spinal cord. The initial mechanical disruption of white matter tracts is rapidly exacerbated by multiple secondary sequelae that include ischemia, inward calcium flux, infiltration with inflammatory cells and liberation of high levels of ROS which evolve over time to produce a chronic state of inflammation. The acute and chronic production of ROS in the injured spinal cord may be expected to have multiple deleterious effects. ROS are reported to cause oxidation of the catalytic cysteine residue at the active site of E1 and E2 enzymes and promote the formation of E1/E2 containing disulfide complexes, which prevents these enzymes from catalyzing ubiquitination [note: this would decrease levels of ubiquitin conjugates] (Hunter, 2007). Severe oxidative stress also impairs proteasome activity thereby increasing ubiquitin conjugates (Shang and Taylor, 2011). Consistent with the notion that proteasome function is impaired after SCI, ubiquitinated proteins have been demonstrated to accumulate in human spinal cord axons in after SCI (Ahlgren et al., 1996). Ubiquitinated protein aggregates have also been found in neurons of dorsal root ganglia and spinal cord gray matter of rats after spinal cord compression (Li and Farooque, 1996). Selective motor neuron death, induced by transient spinal cord ischemia associated with trauma or surgery, was found to correlate with redistribution of Aβ, ubiquitin, Uch-L1 and the formation of aggregates of proteins in neuronal cytoplasm and nuclei (Yamauchi et al., 2007; Myeku et al., 2012). The increase of ubiquitin and Uch-L1 was observed at the early stage of re-perfusion, after the 15 min spinal cord ischemia, and had resolved by 6 h after re-perfusion in experimental animal models. These findings suggest a strong relationship between vulnerability of motor neurons and the ubiquitin-mediated stress response (Yamauchi et al., 2008).

A critical area of research is understanding mechanisms of neuronal death after SCI and identification of strategies to increase numbers of neurons that survive after the trauma, or so called neuroprotection. Even if damage to the descending and ascending axons in white matter can not be prevented, interneurons present in gray matter may provide a means to rewire spinal cord circuits to restore some connectivity between the cortex and periphery (Courtine et al., 2008). Several lines of evidence link ubiquitin, Uch-L1 and E3 ubiquitin ligases to survival of such interneurons. Specifically, spinal cord interneurons with increased ubiquitin and Uch-L1 expression have been reported to have greater cell survival after SCI (Noor et al., 2013). Similarly, ubiquitin and Uch-L1 up-regulation has also been observed in complete spinal cord transection (Fry et al., 2003; Lane et al., 2007). In addition, vulnerability of motor neurons of the spinal cord can be partially attributed to differences in ubiquitin-mediated stress responses after transient ischemia (Sakurai et al., 1998; Yamauchi et al., 2008). Therefore, a failure to upregulate ubiquitin in motor neurons could represent a great disadvantage for motor neuron recovery (Mengesdorf et al., 2002). Moreover, the ubiquitin E3 ligase ZNRF1 (zinc and ring finger 1) has been demonstrated to be associated with Wallerian degeneration after SCI (Araki and Milbrandt, 2003). Thus the significant changes in ubiquitin expression and cellular distribution in response to SCI, and its effects on the neuronal regeneration and myelination, suggest that the UPS could have a significant role in recovery from SCI caused by ischemia.

Once the acute inflammatory response to SCI has resolved, spinal cord neurons engage in a process of rewiring (neuroplasticity) and in efforts to extend new axons across the injured region (neuroregeneration). Studies in Monodelphis domestica and mice suggest a linkage between neuroregeneration and expression levels of components of the UPS, although the underlying mechanisms are poorly understood. For example, following spinal cord transection in Monodelphis domestica, axonal regeneration and regrowth of neuronal tissue was found at the injury site in animals injured at 7 days after birth (P7), which was associated with upregulation of ubiquitin mRNA and protein expression. However if injury occurred at postnatal 28 day (p28), little or no upregulation in ubiquitin expression occurred and no neuronal regeneration was observed (Ji et al., 2008; Yan et al., 2010; Noor et al., 2013). The levels of mono-ubiquitin were decreased at sites both caudal and rostral to the SCI injury site in aged animals, which also have reduced neuroregenerative capabilities; the magnitude of the decrease was found to be age-related. The reduction in ubiquitin levels may also be associated with involvement of ubiquitin in multiple functions and substrates in an organism outside of the proteasome system such as receptor internalization and modifications to chromatin structure (Noor et al., 2011). Moreover, the distribution of ubiquitin in response to the SCI also plays a role in the recovery of the spinal cord. For example, in motor neurons or interneurons of the ventral horn, ubiquitin is significantly increased in the cytosol after injury of postnatal 7 day (p7) mice which demonstrate significant neuroregenerative capabilities. By contrast, in mature mice, where the capacity of neuron regeneration is lost, there is minimal change in ubiquitin levels in the cytosol of ventral horn neurons, in which ubiquitin was primarily localized to the nuclei (Noor et al., 2011, 2013). These findings suggest that the growth inhibitory or stimulatory factors that explain the difference in neuroregenerative capacity between young and old mice regulate ubiquitin expression and localization and support the view that certain levels of ubiquitin in injured neurons are essential for neuron regeneration. An open question is whether certain deubiquitinating enzymes such as Uch-L1 could be beneficial for the SCI. It is notable therefore that Uch-L1 has been reported to promote neurogenesis and regulate differentiation of neuronal progenitors and has been proposed as a potential stem-cell based therapeutic target in the SCI (Sakurai et al., 2006; Naujokat, 2009).

Several neuroprotective drugs have been shown to reduce accumulation of aggregated, ubiquitinated proteins. Elevation of cAMP in rat spinal cord neurons by rolipram, a PDE4 inhibitor, or dibutyryl-cAMP, is neuroprotective and is associated with the increase in proteasome function by enhancing 19S regulatory particle ATPase 6 (Rpt6), and β5 subunit of the central beta-rings of 20S proteasome activity and other components of the UPS including UBB and the E3 ligase CHIP, resulting in a decrease in aggregation of ubiquitinated protein in SCI (Myeku et al., 2012). Since Uch-L1 enhances the down-stream signaling pathways of cAMP by acting on PKA, this provides another rationale for the targeting of Uch-L1 in SCI treatment.

In summary, the UPS is impaired after SCI, and ubiquitin levels are positively associated with neuroregeneration. While it is logical to assume that elevation of Uch-L1 levels is beneficial after SCI, this possibility has not been tested. Further animal studies are needed in SCI models to determine if the overexpression of ubiquitin or Uch-L1 can lead to resistance to neural injury or promote neuroregeneration or neuroplasticity with benefits to functional recovery.

# The Role of Proteasome After SCI

Dysfunction of the proteasome may explain or contribute to the accumulation of ubiquitinated proteins after SCI (Myeku et al., 2012). However, the role of the proteasome in neuronal regeneration is controversial and likely complex. It has been found that treatment with proteasome inhibitors such as MG132 or lactacystin induces secondary axotomy of stretch-injured axons 72 h after the injury (Du et al., 2008), and arrests neurite outgrowth in primary cultured neurons (Laser et al., 2003). The axonal stretch model replicates the mechanical forces acting on brain or spinal cord following traumatic brain injury (TBI) or SCI (Staal et al., 2007). Of relevance to traumatically induced SCI, motor neurons are also more vulnerable to proteasome inhibitor induced neurotoxicity, which includes apoptotic nuclear changes and mitochondrial dysfunction (Kikuchi et al., 2002). In a rat SCI model, cAMP induced PKA activation reduced the accumulation of ubiquitinated protein that was caused by proteasome inhibition in the spinal cord, thus demonstrating a potential therapeutic approach in the prevention of neuronal damage after SCI (Moore and Kennedy, 2006; Myeku et al., 2012), and raising the possibility that Uch-L1 and PDE4 inhibitors may be beneficial after an SCI.

On the other hand, the drug bortezomib, a 26S proteasome inhibitor currently used in treatment of multiple myeloma, was found to be effective in reducing brain ischemia size in clinical studies (Shah and Di Napoli, 2007) by increasing ubiquitin expression in the infarct. One might posit that treatment of bortezomib could also be beneficial after SCI to preserve motor neurons and to improve neuronal function (Henninger et al., 2006). Furthermore, following nerve transection, MG132, a peptide aldehyde inhibitor of the proteasome as well as multiple cysteine and serine proteases, delays the onset of Wallerian degeneration (Zhai et al., 2003). Similar to the immunohistological studies, microarray analysis showed that inhibition of the proteasome may have both neuroprotective and pro-apoptotic response in primary neuron cells through upregulation of neuroprotective heat-shock proteins (HSPs; Yew et al., 2005), of which the main subtype expressed in motor neurons of spinal cord are Hsc70 and Hsp27 (Chen and Brown, 2007). Of interest, the induction of Hsc70 by proteasome inhibitors prevents neuregulin-induced motor nerve demyelination and improves nerve conductance; these changes may be related to the clearance of c-Jun (Li et al., 2012). However, proteasome inhibition activates antioxidant and HSP signaling pathways, which may result in neuroprotection early after SCI and an apoptotic response at later time points (Yew et al., 2005; Yang et al., 2006). The effects of proteasome inhibitors are thus complex, and likely depend on specificity of the inhibitor used and cell type, but also may be influenced by duration and timing relative to the initial spinal cord trauma. Additionally, proteasomes participate in myriad cellular processes in addition to housekeeping functions such as disposal of damaged proteins. Differences in results of studies with proteasome inhibitors may reflect the relative activation state of one or more signaling pathways regulated by proteasomes, or of expression levels of E3 ligases or other components of ubiquitin machinery. Such differences could explain, in part the divergent and at times apparently contradictory findings with proteasome inhibitors. Thus, further studies in animal models of SCI are needed to explore the role of proteasomes in neuroprotection and neuroregeneration and to define the timing of administration of proteasome inhibitors in the protection of spinal neuron recovery.

# The Role of Ubiquitin Ligases, APP and Aβ in Recovery After SCI

As noted above, the UPS plays critical roles in modulating levels of APP and Aβ. Understanding of the role of Fbox2 or other E3 ubiquitin ligases in neuroregeneration or injury to spinal cord neurons after SCI is limited to extrapolations from our understanding of the role of these factors in Aβ metabolism and AD pathogenesis outlined above. There is some understanding of relationships between APP, Aβ and outcomes after SCI which allows greater understanding of how E3 ubiquitin ligases might be expected to influence neuroprotection or neuroregeneration after SCI. APP is carried along the axon by fast axonal transport. It exists in both pre- and postsynaptic sites and has a role in synaptic function and axonal protein transport. Generation, processing and resultant levels of APP have been identified to be very sensitive to CNS insults and have been extensively used as a biomarker of traumatic axonal injury. Levels of APP immunostaining provide a goldstandard for clinical identification of diffuse axonal injury both in brain and SCI (Blumbergs et al., 1995; Nashmi and Fehlings, 2001). Experimental sciatic nerve axotomy showed an axonal accumulation of Aβ and promotion of the pathological formation of Aβ plaques in the motor neurons in gray matter and the glial cells present in white and gray matter of the spinal cord, which may be a reactive response (Ahlgren et al., 1996; Rose and Odlozinski, 1998). However, inhibition of Aβ accumulation by inhibiting APP-γ-secretase and BACE1 was found to reduce the recovery after SCI indicating that Aβ may also have neuroprotective role after SCI (Ahlgren et al., 1996; Rose and Odlozinski, 1998; Pajoohesh-Ganji et al., 2014). At 1 day after SCI, accumulation of APP and Aβ, the loss of the cytoskeletal proteins such as microtubule associated protein-2 (MAP-2), and axonal demyelination have been found both within and away from injury epicenter (Li et al., 1995; Huang et al., 2007). Such findings suggest that both the initial injury and subsequent secondary injury caused by the activation of cascades which spread the injury away from the trauma site are involved in the above pathological changes. The accumulation of APP in traumatized axons at the epicenter of the injury, and also rostral and caudal to it, correlates with loss of motor neuron function, further demonstrating the crucial link between APP and the SCI. These findings implicate impairment of both fast anterograde and retrograde axonal transport (Medana and Esiri, 2003; Ward et al., 2014), which causes an accumulation of APP, other axonal transport proteins, intracellular components such as lysosomes, neurotrophin, mitochondria and synaptophysincarrying vesicles. Transport of these components to synapses is essential for normal neuronal function and survival (Lazarov et al., 2007) and could be Aβ independent (Stokin et al., 2008). On the other hand, the build-up of APP in axons and at synapses could cause rapid local increases in Aβ levels following the increased cleavage of APP by γ- and βsecretase. The increased Aβ further impairs axonal transport causing axonal swelling due to damage to axonal microtubules. Similar to the reported toxicity of Aβ in brain, accumulation of Aβ in spinal cord neurons disrupts intracellular calcium homeostasis, impairs mitochondrial function, and results in abnormal regulation of synaptic receptors and ion channels (Pike et al., 1993). It also induces the release of cytokines causing neuronal inflammation and defective endo-lysosomal trafficking (Pimplikar et al., 2010), altered intracellular signaling cascades, or impaired neurotransmitter release thereby contributing to synaptic dysfunction (Tang-Schomer et al., 2010). Thus, drugs regulating ubiquitin expression after spinal injury could have the potential to spare neurons from injury by minimizing Aβ accumulation (Yokobori et al., 2015).

In summary, Aβ accumulation is an early response after SCI, and its accumulation affects neuronal recovery via similar mechanisms to those involved in AD, such as impairing axonal transport, activating cytokine induced inflammation and impairing neurotransmitter release in spinal cord neurons. Further studies are needed to define which signaling pathway is involved in Aβ induced neuronal dysfunction, if the sensory neurons and motor neurons are shared same mechanisms, and the role of the UPS in the regulation of Aβ accumulation in SCI.

# Role of BACE1 After SCI

Increased accumulation of APP and Aβ have been reported in axons in white matter about 1 mm from the lesion site

after SCI in mice (Kobayashi et al., 2010). There is evidence that in APP overexpressing mice (TgCRND8 mice), there were increased levels of Aβ at the synaptic terminal and extracellular accumulation of Aβ, which were predominately located in the gray matter of the dorsal horn and the central part of dorsal column of the spinal cord. The enhanced Aβ accumulation coincides with loss motor neurons (Xie et al., 2000). In ALS patients, it has been found that Aβ co-localized with oxidative damage markers such as heme oxygenase-1, and nitro-tyrosine in abnormal neurons and an increased cleaved caspase-3 immunoreactivity in motor neurons (Calingasan et al., 2005).

BACE1 is an aspartyl protease of the pepsin family that is mainly expressed in neuronal late Golgi and trans-Golgi (TGN) compartments as well as the cytoplasmic membrane (Vassar et al., 1999; Yan et al., 2001). Since BACE1 is an essential enzyme for Aβ generation as described above, in the past decade, many compounds have been developed to block BACE1 activity. Immunostaining showed that BACE1, APP, and Aβ are closely co-localized with large motor neurons in the ventral horn of the spinal cord in AD mouse models and that Aβ plaques are largely derived from the dystrophic presynaptic axonal terminals originating from glutamatergic neurons arising in the brain and projecting to spinal cord neurons (Li et al., 2013). Moderate inhibition of BACE1 activity reduces Aβ production, enhances clearance of myelin debris and improves axonal regeneration after SCI (Farah et al., 2011). However, completely abolishing BACE1 by knocking out the BACE1 gene has adverse effects on recovery after SCI such as delayed myelination and decreased myelin thickness (hypomyelination) due to a dysregulation of neuregulin-1 type III, an EGF-like growth factor (Hu et al., 2006; Willem et al., 2006). Moreover, animal studies have shown that complete inhibition of BACE1 activity results in an elimination of Aβ in neurons, but Aβ is required for certain physiological neuronal functions such as regulation of neuronglia signaling by enhancing spontaneous astrocyte calcium transient signaling via α7-nAChRs (Lee et al., 2014), activation of M2 microglia cells after SCI promoting plasticity, axonal outgrowth and inhibition of inflammation (Kigerl et al., 2009; Pajoohesh-Ganji and Byrnes, 2011). Since both accumulation and complete elimination of BACE1/Aβ adversely affect recovery after SCI, biological regulation of Aβ or BACE1 levels after SCI by modulation of components of the UPS such as Fbx2-Uch-L1 via peroxisome proliferator-activated receptor-gamma co-activator (PGC)-1alpha or gene expression (Gong et al., 2010, 2013b)

# REFERENCES


may be a feasible strategy for the SCI treatment and AD (**Figure 3**).

In summary, BACE1 plays multiple, critical functions after SCI. Reducing BACE1 expression prevents the accumulation of Aβ in neurons after SCI, while completely blocking BACE1 activity may result demyelination of axons in white matter tracts. Thus future in vivo studies are needed to test when biologically expression of ubiquitin/Uch-L1 as well as other components of the UPS could beneficial to the recovery of SCI.

# SUMMARY

Multiple lines of evidence indicate that dysregulation of any of several components of the UPS, or mutations in genes that encode them, exacerbate AD progression and are deleterious to the recovery of function after SCI. It is notable that many UPS components such as ubiquitin, Uch-L1, Fbox2 and the proteasome itself are critical determinants of disease severity in both SCI and AD. There are, however, many questions to be answered regarding the mechanisms by which these UPS components contribute to AD progression and facilitate or impede recovery of function after SCI. Finally, ubiquitin, Uch-L1, and the proteasome can be manipulated pharmacologically providing convenient methods for research applications and suggesting the possibility that further efforts in this field should include investigation of the potential utility of drugs that target these molecules. The major obstacle in developing such therapies will be the ubiquitous presence of the UPS in eukaryotes which may require nanomedicine approaches or other tissue-targeting strategies to minimize adverse effects of such therapies.

# AUTHOR CONTRIBUTIONS

BG, MR, MEF-P and CC wrote this article. BG and CC did the revision.

# ACKNOWLEDGMENTS

The studies described here were supported in part by grant from Alzheimer's Association (IIRG-12-242345) to BG and by the Rehabilitation Research and Development Service Center for the Medical Consequences of Spinal Cord Injury (B9212C). We thank Dr. Anand and Mr. Jin for critical reading of the manuscript and preparation of the figures.


neurodegenerative disease models by forming proteasome-blocking complexes. Aging Cell 14, 35–48. doi: 10.1111/acel.12296


**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 Gong, Radulovic, Figueiredo-Pereira and Cardozo. 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 ubiquitin-proteasome system in neurodegenerative diseases: precipitating factor, yet part of the solution

### *Nico P. Dantuma1 \* and Laura C. Bott 1,2*

*<sup>1</sup> Department of Cell and Molecular Biology, Karolinska Institutet, Stockholm, Sweden*

*<sup>2</sup> Neurogenetics Branch, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD, USA*

### *Edited by:*

*Fred Van Leeuwen, Maastricht University, Netherlands*

### *Reviewed by:*

*A. Kimberley McAllister, University of California, USA Elly M. Hol, University Medical Center Utrecht, Netherlands*

### *\*Correspondence:*

*Nico P. Dantuma, Department of Cell and Molecular Biology, Karolinska Institutet, von Eulers väg 3, S-17177 Stockholm, Sweden e-mail: nico.dantuma@ki.se*

The ubiquitin-proteasome system (UPS) has been implicated in neurodegenerative diseases based on the presence of deposits consisting of ubiquitylated proteins in affected neurons. It has been postulated that aggregation-prone proteins associated with these disorders, such as α-synuclein, β-amyloid peptide, and polyglutamine proteins, compromise UPS function, and delay the degradation of other proteasome substrates. Many of these substrates play important regulatory roles in signaling, cell cycle progression, or apoptosis, and their inadvertent stabilization due to an overloaded and improperly functioning UPS may thus be responsible for cellular demise in neurodegeneration. Over the past decade, numerous studies have addressed the UPS dysfunction hypothesis using various model systems and techniques that differ in their readout and sensitivity. While an inhibitory effect of some disease proteins on the UPS has been demonstrated, increasing evidence attests that the UPS remains operative in many disease models, which opens new possibilities for treatment. In this review, we will discuss the paradigm shift that repositioned the UPS from being a prime suspect in the pathophysiology of neurodegeneration to an attractive therapeutic target that can be harnessed to accelerate the clearance of disease-linked proteins.

**Keywords: neurodegeneration, ubiquitin, proteasome, proteolysis, protein quality control**

# **INTRODUCTION**

Proteinopathies form a large group of pathologies that are characterized by the presence of abnormally folded proteins in affected cells (Carrell and Lomas, 1997). Among those diseases are hereditary neurodegenerative disorders, such as Huntington's disease (HD), spinal and bulbar muscular atrophy (SBMA), and several forms of autosomal dominant spinocerebellar ataxias (SCAs; types 1-3, 6, 7, 17), caused by polyglutamine (polyQ) repeat expansions in unrelated proteins (Orr and Zoghbi, 2007; La Spada and Taylor, 2010). The observation that the expanded polyQ repeat renders proteins prone to aggregation has raised the idea that members of this disease family may share a common pathogenic mechanism (Scherzinger et al., 1997). It has further been suggested that related pathogenic events may be an underlying cause in other neurodegenerative diseases characterized by the presence of protein aggregates, such as Alzheimer's disease (AD), Parkinson's disease (PD) and amyotrophic lateral sclerosis (ALS) (Sherman and Goldberg, 2001). The technical advantage of working with monogenic polyQ diseases has brought together scientists from a broad variety of research disciplines who have studied the effects of the pathogenic proteins on a wide spectrum of cellular processes in cultured cells and animal models.

Since the discovery of repeat expansions as a genetic basis of hereditary neurodegenerative diseases more than two decades ago (La Spada et al., 1991), we have learned that polyQ proteins have an impact on diverse cellular processes such as transcription, transport, neuronal function, and viability. The long list of cellular functions affected by these proteins suggests that the disease-linked proteins disturb one or more systems central to these processes, causing many downstream pathways to collapse during the course of pathology. The ubiquitin-proteasome system (UPS) has received particular attention in the study of neurodegenerative disorders due to its role as a critical regulator of protein homeostasis in eukaryotic cells. It keeps the cellular environment free of misfolded, defective, and aggregation-prone proteins, which have been found to accumulate in neurodegenerative diseases (Ciechanover and Brundin, 2003). The protein quality control function of the UPS is, however, only one of the essential processes that engages this multitasking proteolytic system which governs also cell cycle progression and induction of apoptosis (Hershko and Ciechanover, 1998). It has been proposed that the vast amounts of aggregation-prone polyQ proteins may overwhelm the UPS and compromise other essential functions of the machinery required for maintaining cellular homeostasis (Mayer et al., 1989). According to this model, blockade of the UPS by the disease protein would result in global accumulation of proteasome substrates and thus provides an explanation for the broad and diverse effects on cellular homeostasis in affected cells.

Many laboratories have addressed the UPS dysfunction hypothesis over the last decade and it has become evident that UPS activity is preserved in the majority of neurodegenerative disorders. In the course of these studies, adaptive cellular responses have been identified that help to alleviate the burden of aggregation-prone proteins to keep ubiquitin-dependent proteolysis operative. Here we will discuss the transition of our view on the UPS from a dysfunctional system that catalyzes cytotoxicity in neurodegenerative diseases to a powerful proteolytic system that may be exploitable in therapeutic strategies aimed at clearing aggregation-prone proteins from cells.

# **THE UBIQUITIN-PROTEASOME SYSTEM AS PRIME SUSPECT**

The UPS, which is the principal pathway for the clearance of short-lived, damaged, and misfolded proteins in the nucleus and cytoplasm, consists of two separate, consecutive steps: ubiquitylation and proteasomal degradation (Hershko and Ciechanover, 1998; Kleiger and Mayor, 2014). An enzymatic cascade composed of ubiquitin activator, conjugase, and ligase catalyzes the covalent attachment of ubiquitin to a substrate protein. Ubiquitin is conjugated via its carboxy-terminal glycine to an internal lysine (Lys) residue or, less commonly, to the free amino (N) terminus of the substrate (Pickart, 2001). Multiple rounds of ubiquitylation lead to the formation of a polyubiquitin chain, which can function as a signal for degradation by the proteasome, a multi-protein complex consisting of a 20S core particle and 19S regulatory particles, at one or both ends. The active sites responsible for the chymotrypsin-like, trypsin-like, and caspase-like activities of the proteasome are situated in the interior surface of the 20S core particle, thereby shielding their proteolytic activities from the rest of the cellular proteome (Bedford et al., 2010). The proteasome unfolds substrates and threads the polypeptide chains through the inner channel, where they are cleaved into short peptides (Bhattacharyya et al., 2014). Following their release from the barrel, peptides are rapidly processed into amino acids by cellular aminopeptidases and recycled (Reits et al., 2003).

Ubiquitylation has many other roles in cells besides proteasomal degradation (**Figure 1**). The destiny of a given substrate protein is determined by the type of ubiquitin assembly to which it is connected. This is possible because ubiquitin contains seven Lys residues in its amino acid sequence at positions 6, 11, 27, 29, 33, 48, and 63, which can serve as acceptors for additional ubiquitin monomers in the construction of polyubiquitin chains (Komander and Rape, 2012). As a result, many different chain topologies can be formed, which are recognized by specific ubiquitin-binding adaptors in the relevant cellular pathways. Chain topology and substrate specificity are determined by a large spectrum of ubiquitin-ligating and -modifying enzymes, whose expression levels and activities are tightly regulated in a tissue-, cell-, and compartment-specific manner. Several different types of polyubiquitin linkages target substrates to the proteasome, such as Lys11 and Lys29 in addition to the canonical Lys48-linked chains, as well as conjugation of a single ubiquitin molecule (Kravtsova-Ivantsiv and Ciechanover, 2012). Monoand polyubiquitin chains also regulate non-proteolytic functions in cells, such as protein activity and localization (Seet et al., 2006). While most ubiquitin chains can target proteins for proteasomal degradation, it has been shown that Lys63-linked ubiquitin chains are the only modification that do not behave as a proteasome targeting signal *in vivo* (Nathan et al., 2013). Instead these chains play pivotal roles in signaling, endocytosis, and DNA repair. More recently, this chain topology has been implicated in macroautophagy (Kraft et al., 2010), a pathway that targets

**FIGURE 1 | Structure and function of common ubiquitin modifications.** Ubiquitin may be conjugated to protein substrates as either a monomer or a polymeric chain, in which one of seven internal lysine (Lys) residues of ubiquitin, or the N-terminal methionine, serves as an acceptor for additional ubiquitin moieties. The type of polyubiquitin linkage dictates the topology of the resulting chain. Ubiquitin modifications can regulate protein function or act as a signal in many cellular processes. Examples for functions of monoubiquitylation, and homogenous Lys11-, Lys48-, and Lys63-linked polyubiquitin chains are shown.

cytoplasmic proteins and organelles for degradation in lysosomes (Nakatogawa et al., 2009). Therefore, the UPS is highly interconnected with other proteolytic and non-proteolytic cellular processes at multiple levels, whereby it controls many diverse functions in cells.

Given the vast amount of proteins that are involved in the UPS, it is perhaps not surprising that some of them have been genetically linked to neurodegenerative disorders (Ciechanover and Brundin, 2003). On first sight, the finding that mutations in genes encoding components of the UPS can cause or predispose for neurodegeneration supports the notion of inefficient ubiquitin-dependent proteolysis as a shared pathogenic mechanism. However, it should not be overlooked that the UPS and, in particular, the ubiquitin-targeting step are connected to many different processes besides proteasomal degradation. Several cases of neurodegeneration-linked mutations in UPS components are now known to affect ubiquitin-dependent processes that do not target proteins to the proteasome, while the UPS remains largely operative.

One of the best-studied examples of a UPS component that has been linked to neurodegeneration is the ubiquitin ligase Parkin, mutations in which cause an autosomal recessive juvenileonset PD (Kitada et al., 1998). Early studies revealed that Parkin can target endoplasmic reticulum (ER)-derived proteins (Yang et al., 2003) and polyQ proteins for proteasomal degradation (Tsai et al., 2003) suggesting that a defect in the UPS-mediated protein quality control may be responsible for this pathology. However, during recent years it has become apparent that Parkin is also involved in the autophagy pathway that results in the degradation of dysfunctional mitochondria in lysosomes, a process known as mitophagy (Ashrafi and Schwarz, 2013). Parkin cooperates with the mitochondrial kinase PINK1, which has also been linked to PD, thus strengthening the genetic association between mitophagy and neurodegeneration (Clark et al., 2006; Park et al., 2006). Another theory suggests that Lys63-linked polyubiquitin chains generated by Parkin may be important for targeting aggregation-prone proteins to inclusions bodies (IBs) (Olzmann et al., 2007). Although the possibility remains that UPS dysfunction contributes to PD, recent studies point to autophagyrelated processes as central to the pathology caused by Parkin mutations.

Another example is ubiquilin, a ubiquitin-binding shuttle factor that is involved in escorting polyubiquitylated proteins to the proteasome for degradation (Elsasser and Finley, 2005). Overexpression of ubiquilin has a neuroprotective effect in mice expressing a fragment of the polyQ protein causative for HD, huntingtin (El Ayadi et al., 2012). A single nucleotide polymorphism (SNP) that causes alternative splicing of the ubiquilin transcript predisposes for late-onset AD (Bertram et al., 2005). Given the function of ubiquilin as a shuttling substrate receptor in the UPS, it is tempting to speculate that the SNP in ubiquilin alters the ability of cells to recognize and destroy misfolded proteins. However, the AD-linked ubiquilin variants do not cause a general block of the UPS, but instead, were found to selectively cause accumulation of presenilin-1, which is involved in amyloid precursor protein (APP) processing (Viswanathan et al., 2011). Moreover, ubiquilin-1 functions as a molecular chaperone that regulates trafficking and processing of APP, which has been linked to its ability to stimulate polyubiquitylation of APP with non-proteolytic Lys63-linked polyubiquitin chains (El Ayadi et al., 2012). Together this suggests that while ubiquilin-1 may regulate the production of β-amyloid peptide at multiple levels in a ubiquitin-dependent fashion, the variants linked to AD do not seem to trigger a general failure of the UPS.

The ubiquitin-selective chaperone valosin-containing protein (VCP) has also attracted attention from researchers who study the link between the UPS and neurodegenerative disorders. Mutations in VCP cause multisystem proteinopathy (MSP), with among its spectrum of symptoms frontotemporal dementia (Watts et al., 2004), as well as the motor neuron disease amyotrophic lateral sclerosis (ALS) (Johnson et al., 2010). VCP is critical for proteasomal degradation of certain proteins and it is believed that it does so by means of its ATP-dependent chaperone activity that can unfold or segregate proteasome substrates from their environment (Stolz et al., 2011). However, ubiquitin-dependent functions of VCP are not limited to proteasomal degradation (Dantuma and Hoppe, 2012; Meyer et al., 2012) and include ubiquitin-selective autophagy (Ju et al., 2009; Tresse et al., 2010), the clearance of stress granules (Buchan et al., 2013), mitochondrial integrity (Bartolome et al., 2013), Parkindependent mitophagy (Kim et al., 2013), and the DNA damage response (Acs et al., 2011; Meerang et al., 2011). Overexpression of VCP mutants linked to MSP and ALS did not inhibit the UPS (Tresse et al., 2010), whereas some of the VCP-mediated events mentioned above were altered (Ju et al., 2009; Tresse et al., 2010; Bartolome et al., 2013; Fujita et al., 2013; Kim et al., 2013), suggesting that the pathology caused by mutant VCP likely involves essential ubiquitin-dependent processes that are different from proteasomal degradation.

A unique example of a protein responsible for a polyQ neurodegenerative disease that is involved in ubiquitin-dependent proteasomal degradation is ataxin-3. It functions as a deubiquitylation enzyme that reverses the ubiquitin mark on proteins by disassembling Lys48- and Lys63-linked polyubiquitin chains with a preference for the latter (Winborn et al., 2008). Expansion of the polyQ repeat that resides in ataxin-3 are the underlying cause for the most common form of autosomal dominant SCA, known as SCA-3 or Machado-Joseph disease. Notably, ataxin-3 physically interacts with VCP and regulates proteasomal degradation of ER-derived substrates (Wang et al., 2006; Zhong and Pittman, 2006). Overexpression of either wild-type or polyQ-expanded ataxin-3 compromises the functionality of the UPS resulting in increased levels of proteasome substrates (Burnett et al., 2003). Interestingly, non-expanded ataxin-3 is also a known suppressor of polyQ toxicity in models of SCA-3 and other polyQ diseases (Warrick et al., 2005). Although the reason for this protective effect is poorly understood, it may be attributed to the ability of ataxin-3 to stimulate sequestration of misfolded proteins (Burnett and Pittman, 2005). Finally, ataxin-3 regulates the ubiquitylation status of Parkin and stimulates its degradation by autophagy, linking also this protein to other ubiquitin-dependent systems (Durcan et al., 2011). This study, like the previous examples, suggests a complex relationship between the UPS and neurodegeneration and shows that the effects of disease-associated proteins on ubiquitin-dependent proteolysis are difficult to predict.

Two important lessons can be drawn from the aforementioned examples. First, each one of these examples underscores the prevalent notion that the UPS is tightly connected to neurodegenerative disorders. The fact that unrelated diseases share the presence of causative mutations in UPS components leaves little doubt that the UPS plays a central role in these pathologies. Second, they illustrate that care should be taken when extrapolating genetic data to disease mechanisms. Thus, the general view that a compromised UPS is responsible for the typical accumulation of misfolded proteins in neurodegenerative diseases is problematic due the complex and multilayered connection between the UPS and a variety of cellular functions. The fact that a large number of essential cellular functions other than protein quality control require ubiquitin conjugation or proteasomal degradation leaves the possibility open that the role of the UPS in these pathologies may be largely unrelated to ubiquitindependent protein quality control. A definite answer on whether, and if so, to what extent, impairment of the UPS contributes to the development of neurodegenerative diseases can only be obtained by empirically assessing the functional status of the UPS in each of these diseases.

# **UBIQUITIN-POSITIVE INCLUSIONS**

A hallmark of neurodegenerative diseases is the accumulation of abnormal proteins in insoluble deposists, or IBs. The idea that impaired clearance of misfolded proteins may be central to neurodegenerative diseases originates in part from the observation that IBs contain ubiquitin, proteasome subunits, and other UPS components (Cummings et al., 1998). Mori et al. who first described the presence of ubiquitin in IBs, already speculated that either the protease responsible for degrading ubiquitylated proteins (which was not known at that time to be the proteasome) was dysfunctional, or that ubiquitylated proteins residing in the inclusions resisted degradation (Mori et al., 1987).

For a long time, the nature of IBs has been under debate (Sisodia, 1998). While some researchers argued that they were innocent bystanders, others favored the view that IBs were directly implicated in the cellular pathology caused by aggregation-prone proteins (Sherman and Goldberg, 2001). One hypothesis that has appeared in different forms with various key players suggests that physical entrapment of proteins in IBs removes functional proteins from critical cellular processes. With respect to the UPS and chaperones, one can picture how sequestration may compromise their household function and elicit cellular pathology. According to yet another model, IBs are part of the cellular defense mechanism that protects cells against toxicity by functioning as a sink for misfolded proteins. A wealth of data in support for the latter option comes from cell biologists who found that ectopic overexpression of abnormal proteins triggers a sequence of reactions that results in the deposition of the misfolded proteins into specific perinuclear structures, which were coined aggresomes (Wojcik et al., 1996; Johnston et al., 1998; Garcia-Mata et al., 1999). These findings argue against the view that IBs are formed by passive aggregation, but instead require active, ATPdependent processes in cells (Kopito, 2000). Moreover, related protein deposits have been documented in budding yeast, which suggests that this cellular response is conserved throughout evolution and serves cytoprotective functions in both unicellular and multicellular eukaryotic organisms (Kaganovich et al., 2008). In dividing cells, IBs are distributed asymmetrically to daughter cells during mitosis ensuring the generation of healthy progeny devoid of proteineous deposits (Rujano et al., 2006).

The most compelling evidence supporting a role of IBs as an adaptive response comes from live cell recordings, which show that IB formation correlates with increased survival in primary neurons expressing a mutant huntingtin fragment (Arrasate et al., 2004). This is consistent with the finding that reducing the load of IBs, achieved by genetically inhibiting ubiquitylation, enhances polyQ-mediated neurodegeneration in a SCA-1 mouse model (Cummings et al., 1999). Inclusions are reversible and highly dynamic structures, as switching off expression of a mutant huntingtin fragment (Yamamoto et al., 2000) or mutant ataxin-1 (Zu et al., 2004) in conditional mouse models results in the clearance of ubiquitin-positive IBs. When it comes to the UPS components and chaperones associated with IBs, it has been demonstrated that these factors do not appear to be physically trapped in IBs (Kim et al., 2002; Stenoien et al., 2002; Holmberg et al., 2004). A recent study even suggests that proteasomes play an active role in maintaining or dissolving these structures (Schipper-Krom et al., 2014). Furthermore, POH1/Rpn11, a proteasome subunit with deubiquitylation activity, stimulates the formation of IBs by generating free ubiquitin chains in proximity of the aggregated proteins and results in the recruitment of HDAC6 (Hao et al., 2013), which orchestrates ubiquitin-dependent transport of proteins to the aggresome (Kawaguchi et al., 2003). Altogether, these findings support a protective function of IBs in cells facing large amounts of misfolded proteins. Increasing evidence suggests that IBs may act as a hub for misfolded and aggregated proteins and redirect them to alternative destruction mechanisms, which will be discussed later.

# **PROTEASOME ACTIVITY IN NEURODEGENERATIVE DISEASES**

One possibility to get a better insight into the functionality of the UPS is by assessing the individual enzymatic activities involved in ubiquitin-dependent proteasomal degradation (Lindsten and Dantuma, 2003). Ubiquitylation, on one hand, is the net result of a large family of enzymes that are involved in proteolytic and non-proteolytic processes and are therefore not straight-forward to address or to interpret. The proteasome, on the other hand, is the final destination of all ubiquitylated substrates to be degraded and creates a bottleneck in the UPS pathway. Its function is readily traceable to the individual proteolytic subunits whose activities can be measured by using specific fluorogenic substrates (Kisselev and Goldberg, 2005) or activity probes (Verdoes et al., 2006). It is therefore not surprising that the proteasome has received a lot of attention in studies that assess the functionality of the UPS in neurodegenerative diseases. However, correlation of proteasome activity measurements and UPS impairment in neurodegenerative diseases is complicated by the fact that it is presently unknown to what extent altered proteasome activity affects the overall flux of degradation of ubiquitylated substrates. Fibroblasts derived from mice with a heterozygous deletion of PSMC1/Rpt2, one of the ATPase subunits of the 19S regulatory particle, develop without obvious defects despite reduced proteasome function (Rezvani et al., 2012). Moreover, proteasome activity can be regulated through expression of individual subunits. For example, increasing the amount of active proteasomes, either through overexpression of the 20S core component PSMB5/β5 responsible for the chymotrypsin-like activity or the proteasome assembly chaperone hUMP1/POMP, improves resistance to oxidative stress insults in human fibroblasts (Chondrogianni et al., 2005; Chondrogianni and Gonos, 2007). Likewise, levels of the PSMD11/Rpn6 proteasome subunit, which stabilizes the interaction between the regulatory and core particle and facilitates ubiquitin-dependent degradation (Pathare et al., 2012), can regulate proteasome activity (Vilchez et al., 2012a,b).

Translating how certain levels of proteasome inhibition will jeopardize the functional status of ubiquitin-dependent proteasomal degradation poses another unresolved problem in assessing the status of the UPS through proteasome activity measurements. Experiments with yeast strains expressing mutant proteasome subunits suggest that the proteolytic sites are non-redundant and vary in their contribution to overall protein degradation (Rubin et al., 1998). The chymotrypsin-like activity of the proteasome needs to be reduced by more than 80 percent in human cells before the clearance of ubiquitylated proteins becomes noticeably delayed (Dantuma et al., 2000; Bence et al., 2001). Even when this threshold is not reached, a lower degree of inhibition of the chymotrypsin-like activity does, however, impair the cells' ability to deal with an acute increase in the flux of ubiquitylated proteins during stress conditions (Dantuma et al., 2000). In mammalian cells, individual proteasome activities also differ in their contribution to overall protein degradation since the caspase-like activity can be chemically ablated without affecting the ability of cells to clear ubiquitylated proteins (Myung et al., 2001). Therefore, the differential contribution of individual proteolytic activities, the high level of redundant proteasome activity, and indirect effects of the stress status, which determine the load of ubiquitylated substrates, complicate the interpretation of proteasome activity measurements when it comes to the functional status of the UPS.

Initial studies have measured proteasome activity in postmortem human tissues and reported decreased activity in neurodegenerative disorders (**Table 1**). However, follow-up experiments with purified proteasomes, cell lines, and animal models have not always been consistent with those early reports. While in some studies a decrease in proteasome activity has been reported, others have found that the activity is unchanged or even increased (for references see **Table 1**). Increased proteasome activity could be an adaptive response to the augmented load of misfolded proteins in these diseases but there are also alternative explanations. In this respect, it is important to point out there is a set of alternative proteolytic subunits that can replace the constitutive proteases in the proteasome in response to interferon γ (Kniepert and Groettrup, 2014). The resulting immunoproteasome plays an important role in generating peptide fragments for antigen display by major histocompatibility complex class I proteins on the cell surface. Thus, changes in the activity of the proteasome may also be attributed to induction of immunoproteasomes in inflammatory response, which are commonly observed in neurodegenerative diseases (Czirr and Wyss-Coray, 2012). Indeed induction of immunoproteasome subunits have been reported in mouse models of polyQ disorders (Diaz-Hernandez et al., 2003) and other neurodegenerative diseases (Mishto et al., 2006; Cheroni et al., 2009; Orre et al., 2013).

An exceptional case, in which the presence of an aggregationprotein has been mechanistically linked to inhibition of the proteasome, is the prion protein, which causes when misfolded the fatal, transmissible neurodegenerative disorder Creutzfeld-Jacob disease. Wild-type prion proteins residing in the cytosolic compartment of cells are efficiently degraded in a ubiquitindependent fashion (Yedidia et al., 2001) whereas cytosolic mutant prion proteins form aggresomes, which is accompanied by signs of apoptotic cell death (Kristiansen et al., 2005). It has been shown that oligomers of mutant prion protein effectively inhibit the activity of the proteasome *in vitro* and *in vivo* (Kristiansen et al., 2007) through direct binding of oligomers to the 20S proteasome core particle, which results in stabilization of a closed confirmation of the proteasome (Deriziotis et al., 2011). As a consequence, prion protein inhibits proteolysis by preventing substrate access to the proteasome core.

The main constituent of the Lewy bodies in PD, α-synuclein, has also been shown to bind to proteasomes (Snyder et al., 2003) and inhibit UPS function *in vitro* and *in vivo* (Stefanis et al., 2001; Snyder et al., 2003; Chen et al., 2006). Cytoplasmic amyloid β peptide can also interact with the proteasome (Gregori et al., 1995), however, its effect on proteasome activity is debated (for references see **Table 1**). Moreover, it is not clear if proteasome inhibition by α-synuclein or β-amyloid is mechanistically related to that of prion proteins.

For polyQ proteins, it has been shown that ubiquitylated filamentous aggregates of mutant huntingtin isolated from mouse or human brain samples can selectively inhibit the proteolytic activity of the proteasome (Diaz-Hernandez et al., 2006). We will discuss below that several studies suggest that this inhibitory activity of huntingtin aggregates does not compromise the UPS *in vivo* (Bett et al., 2009; Maynard et al., 2009; Ortega et al., 2010). Moreover, mutant huntingtin fragments have not been found to inhibit degradation of ubiquitylated substrates by the 26S proteasome *in vitro* (Hipp et al., 2012). Whether or not polyQ proteins can be degraded by the mammalian proteasome has been the subject of a considerable debate (Venkatraman et al., 2004; Pratt and Rechsteiner, 2008; Juenemann et al., 2013). Puromycin-sensitive aminopeptidase has been identified as the main cytosolic protease to efficiently clear expanded polyQ peptides generated by the proteasome *in vitro* (Bhutani et al., 2007). It has been hypothesized that proteasome-derived polyQ fragments may enhance aggregate formation (Venkatraman et al., 2004; Raspe et al., 2009), but it is unclear whether these species are actually generated *in vivo*. In cultured cells, aggregation of mutant huntingtin fragment is exacerbated following treatment with the proteasome inhibitor lactacystin (Waelter et al., 2001). Moreover, enhancing UPS-mediated clearance of polyQ proteins reduces their levels and toxicity, suggesting that the net outcome of accelerated proteasomal degradation is beneficial in polyQ diseases (Verhoef et al., 2002; Michalik and Van Broeckhoven, 2004).

# **FUNCTIONAL STATUS OF THE UPS IN NEURODEGENERATIVE DISEASES**

UPS functionality, which describes the relative rate at which cells ubiquitylate and degrade proteins at a given time, can be addressed in cells and tissues in several ways. These approaches are based on the assumption that cells that cannot maintain a constant flux through the UPS will gradually build up ubiquitylated proteasome substrates irrespective of their nature. As explained earlier, the relationship between UPS activity and levels of substrates is often complex. Reduced proteasome activity does not necessarily lead to functional impairment of the UPS as long as the activity is sufficient to process ubiquitylated proteins targeted for destruction. *Vice versa*, ubiquitin-dependent proteolysis in cells can be severely impaired despite the presence of a normal proteasome activity profile. Furthermore, cells may mobilize compensatory mechanisms to assist a suboptimal but functioning UPS.

A straight-forward approach to assess UPS activity in cells or tissues measures the abundance or quality of ubiquitylated proteins. However, ubiquitylation is involved in many cellular processes other than proteasomal degradation (Komander and Rape, 2012) and it is therefore difficult to extrapolate whether a possible build up in ubiquitylated proteins is a consequence of UPS impairment or reflects other changes in the ubiquitin homeostasis (Groothuis et al., 2006). Substrate flux through the UPS


### **Table 1 | Overview of measurements of the proteolytic activity of the proteasome in neurodegenerative disease models.**

*aFilamentous htt but not inclusions isolated from HD94 mice was shown to inhibit the 26S proteasome in vitro without affecting 20S proteasome function.*

*bDecreased proteasome activity observed only at early time points; no difference compared with wild type at late disease stages.*

*cDecreased chymotryptic activity, but no decreased tryptic- and caspase-like activity.*

can be assessed by determining the levels or half-lives of cellular proteins. An important caveat with this approach is that the degradation of endogenous proteasome substrates is often tightly regulated and likely to change depending on internal or external cues. Particularly in the context of neurodegenerative diseases, altered degradation rates of endogenous substrates likely reflect functional differences in protein stabilization rather than changes in overall UPS activity. This problem can be circumvented by the use of reporter substrates, which have been developed for monitoring protein degradation by the UPS. These reporters are typically based on fluorescent proteins fused to degradation signals that target the fusion protein for constitutive turnover via the UPS (Neefjes and Dantuma, 2004). They include the ubiquitindependent reporters UbG76V-green fluorescent protein (GFP) and GFPu that are targeted via an N-terminal ubiquitin or a short CL1 peptide motif, respectively (Dantuma et al., 2000; Bence et al., 2001). Another commonly used reporter, based on the degron derived from mouse ornithine decarboxylase (ODC), is degraded in a ubiquitin-independent fashion (Murakami et al., 1992). UPS reporter substrates are in principle devoid of intrinsic regulatory elements to eliminate the impact of variables such as posttranslational modifications, which can affect the half-life of endogenous protein substrates. Although all these reporter proteins are subject to degradation by the 26S proteasome, they are recognized by distinct targeting pathways and therefore differ in sensitivity and signal-to-background ratio. It has been shown that levels of the reporters inversely correlate with UPS activity (Dantuma et al., 2000; Bence et al., 2001). Accumulation of reporter protein relative to baseline levels is typically interpreted as reduced UPS function, however, reporter protein levels may be affected by changes in expression (Bowman et al., 2005; Tokui et al., 2009) or protein synthesis, to which they are very sensitive as a consequence of their short half-life (Li et al., 1998). These reporter substrates are important tools for studying UPS functionality in cell and mouse models of neurodegenerative diseases (**Table 2**).

In cell culture experiments, an N-terminal fragment of polyQexpanded huntingtin has been shown to impair the degradation of the endogenous substrate p53 (Jana et al., 2001) and GFPu (Bence et al., 2001). UPS inhibition appears to be a consequence of protein aggregation and not a unique feature of polyQ proteins since the unrelated aggregation-prone protein mutant cystic fibrosis membrane conductance regulator (CFTR) has a similar effect on the UPS (Bence et al., 2001). GFPu reporter accumulation correlates with induction of apoptosis, supporting a model in which UPS impairment induced by protein aggregation proteins is responsible for cell death although it cannot exclude that other adverse effects are responsible for the cytotoxicity (Bence et al., 2001). UPS impairment as a result of polyQ proteins affects both the nuclear and cytosolic UPS and does not seem to be confined to the compartment that accumulates the aggregation-prone proteins, arguing for general interference with UPS function (Bennett et al., 2005). A study utilizing a version of the GFPu reporter targeted to the synapse of neurons suggested that mutant huntingtin primarily affects UPS function in the synaptic compartment (Wang et al., 2008). It is noteworthy that later studies found the artificial CL1 degradation signal in GFPu to render proteins aggregation-prone (Menendez-Benito

et al., 2005; Link et al., 2006), which is not surprising if one considers that this artificial degradation signal mimics targeting signals present in ER-derived proteasome substrates (Gilon et al., 2000). Subsequent studies have shown that the polyQ huntingtin fragment also affects the degradation of other UPS reporters, such as UbG76V-GFP, signifying that this effect is not limited to aggregation-prone proteins (Maynard et al., 2009; Mitra et al., 2009; Hipp et al., 2012). Analysis in a cellular model demonstrates that aggregates of N-terminal huntingtin do not directly impair the proteasome, instead, the increase in ubiquitylated proteins likely reflects a general disturbance of the cellular proteostasis network (Hipp et al., 2012).

Two transgenic mouse models expressing GFP-based reporter substrates for the UPS, UbG76V-GFP (Lindsten et al., 2003) and GFPu (Bove et al., 2006), have been instrumental in addressing UPS functionality in neurodegeneration *in vivo*. Using these mice, it has been demonstrated that UPS function is preserved in animals expressing polyQ-expanded proteins such as the N-terminal huntingtin fragment, R6/2 (Bett et al., 2009; Maynard et al., 2009; Ortega et al., 2010), androgen receptor (AR) responsible for SBMA (Tokui et al., 2009), and ataxin-7 that causes SCA-7 (Bowman et al., 2005). The lack of UPS impairment in the R6/2 mouse model has been particularly puzzling since the evidence for a globally dysfunctional UPS in cultured cells has largely been based on ectopic expression of the very same huntingtin fragment that is expressed in these mice (Bence et al., 2001; Jana et al., 2001; Bennett et al., 2007; Wang et al., 2008).

The development of quantitative mass spectrometry combined with efficient purification of ubiquitylated proteins or signature peptides from biological samples has been invaluable in undertaking detailed analysis of the ubiquitylated proteome in human diseases (Kessler, 2013). This method not only allows the determination of total ubiquitin levels, but also enables discrimination of different types of ubiquitin chains. Quantitative mass spectrometry has been successfully applied to investigate the composition of ubiquitin conjugates in R6/2 mice and HD patient brain and revealed accumulation of Lys11-, Lys48-, and Lys63-linked ubiquitin chains (Bennett et al., 2007). In theory, the increase in Lys11- and Lys48-linked ubiquitin chains would be consistent with a functional blockade of the UPS as both chain topologies can target proteins to the proteasome (Komander and Rape, 2012). However, Lys63-linked polyubiquitin chains also accumulate in HD and are not normally associated with proteasomal degradation (Nathan et al., 2013), suggesting that the effect of polyQ-expanded huntingtin on ubiquitin homeostasis is more complex than simple blockade of the UPS. In support of this view, two independent studies have not detected stabilization of UbG76V-GFP (Maynard et al., 2009) or GFPu (Bett et al., 2006) in R6/2 mice. Ubiquitin conjugates observed in R6/2 brains have also been shown to differ qualitatively from those observed upon proteasome inhibition (Maynard et al., 2009), indicating that the observed increase in ubiquitylation in HD may be due to changes in ubiquitin-dependent processes other than proteasomal degradation.

While data obtained with mouse models for polyQ diseases have unequivocally supported the presence of a preserved UPS, UPS impairment has been detected in models of


**Table 2 | Overview of measurements of the functionality of the ubiquitin/proteasome system in neurodegenerative disease models.**

*aTransient UPS impairment.*

*bUPS impairment by mutant AR is ligand-dependent.*

*cUPS impairment by mutant AR occurs only in absence of ligand; ligand treatment restored UPS functionality.*

*dReduced UPS activity observed only at late stages in cells which display advanced ALS pathology (e.g., enlarged vacuoles).*

other neurodegenerative diseases. For example studies with the UbG76V-GFP reporter mice have shown that UPS impairment occurs during progression in an ALS mouse model expressing mutant SOD1 (Cheroni et al., 2009). Likewise, UPS impairment has been demonstrated in a mouse model of prion pathology, which is consistent with a role for direct UPS impairment by the prion protein (Kristiansen et al., 2007). Opposing findings have been reported in the well-characterized APPswe/PS1dE9 mouse model for AD. A recent study has demonstrated accumulation of GFPu reporter as well as the endogenous proteasome substrate p53 in this model (Liu et al., 2014a), whereas the same mouse model did not accumulate UbG76V-GFP according to another report (Orre et al., 2013). Although documented for some disorders, global inhibition of ubiquitin-dependent proteolysis does not appear to be a universal feature of neurodegenerative diseases.

# **COMPENSATORY MECHANISMS**

The absence of functional impairment of the UPS in mouse models of polyQ diseases does not imply that the UPS is unaffected by the presence of aggregation-prone proteins. In cellular models, polyQ proteins have been shown to elicit UPS inhibition, indicating that they can have a general negative impact on intracellular proteolysis. Transgene expression levels are an important difference between the cell lines and mouse models. In cells, proteins are typically transiently overexpressed in an acute manner whereas proteins are expressed chronically at modest levels in transgenic mice, more similar to the situation in human patients. The effect of acute induction of the N-terminal mutant huntingtin fragment has been addressed *in vivo* (Ortega et al., 2010) using a mouse model in which huntingtin expression can be regulated by administration of doxycyclin (Yamamoto et al., 2000). Interestingly, acute overexpression of the mutant huntingtin fragment in these mice is accompanied by increased levels of the UbG76V-GFP reporter substrate in neuronal and non-neuronal cells. Importantly, accumulation of the UPS reporter is of a transient nature, suggesting that cells activate adaptive responses to restore UPS activity (Ortega et al., 2010). Restoration of UPS function coincides with IB formation in neurons (Mitra et al., 2009; Ortega et al., 2010), which is consistent with a protective role of inclusions through sequestration of toxic oligomeric species. In support of this model, the UbG76V-GFP reporter has been shown to accumulate also in R6/2 mice following treatment with chemical inhibitors of protein aggregation, which block the formation of IBs (Heiser et al., 2002; Ortega et al., 2010).

The autophagy pathway is tightly connected with the UPS and can play a compensatory role in maintaining intracellular protein degradation under conditions of reduced UPS activity. Autophagy counteracts the toxicity of mutant huntingtin, possibly through promoting the removal of aggregated, oligomeric species (Ravikumar et al., 2004). Moreover, autophagy induction reversed UPS impairment in a *Drosophila* model of SBMA (Pandey et al., 2007). Chemicals that inhibit lysosomal degradation also compromise the UPS, suggesting that autophagy is required for proper functioning of ubiquitin-dependent proteasomal degradation (Korolchuk et al., 2009). Autophagy likely assists the UPS in the removal of a pool of problematic proteasome substrates, such as aggregated or damaged proteins, which would otherwise impede UPS activity. It has been suggested that, analogous to UPS dysfunction in neurodegeneration, autophagy may be compromised in polyQ diseases (Ravikumar et al., 2004). Interestingly, experiments in mice have shown that autophagy dysfunction results in a neurodegeneration phenotype associated with ubiquitin-positive inclusions, indicating that an impaired autophagolysosomal system can recapitulate neurodegeneration (Hara et al., 2006; Komatsu et al., 2006).

Ubiquitin, which serves a wide range of cellular functions besides proteasomal degradation, is closely involved in these compensatory mechanisms (**Figure 2**). It has been shown to target proteins to aggresomes by a mechanism that involves HDAC6, a cytosolic deacteylation enzyme that binds unanchored ubiquitin chains and facilitates the sequestration of aggregated proteins in IBs (Kawaguchi et al., 2003). The ability of specific adaptor proteins, such as p62 and NBR1, to simultaneously bind polyubiquitylated cargo and the autophagosome marker LC3 also supports a central role for ubiquitin in selective autophagy (Kraft et al., 2010). However, the role of ubiquitin as a decisive signal in autophagy has been questioned and instead it has been argued that protein oligomerization and not ubiquitylation is the primary signal (Riley et al., 2010). Both the formation of aggresomes (Olzmann et al., 2007) and ubiquitin-selective autophagy (Tan et al., 2008) are typically associated with Lys63-linked ubiquitin chains, although the involvement of other chain topologies is less clear. Notably, monoubiquitin can suffice as a signal to direct proteins to the autophagosome (Kim et al., 2008). More experiments are needed to dissect the functions of ubiquitin and the relative contribution of individual chain topologies in cellular processes.

The fact that ubiquitin is shared between targeting mechanisms that direct substrates to the proteolytic machinery and adaptive responses, which counteract toxic proteins, suggests a functional significance in directing the joint efforts of these pathways in eliminating harmful proteins (Groothuis et al., 2006).

**FIGURE 2 | Cellular pathways that counteract protein aggregation are ubiquitin-dependent processes.** Proteins linked to neurodegenerative diseases, such as α-synuclein, β-amyloid peptide and polyQ proteins, are prone to misfolding and aggregation in the cellular environment. The proteasome, autophagy, and inclusion bodies form a network of quality control systems which reduces levels of misfolded proteins and counteracts aggregation. All three pathways are regulated by ubiquitylation. It is possible that the accumulation of polyubiquitin conjugates in polyQ diseases may be due to upregulation of ubiquitindependent adaptive responses, such as the formation of inclusions or activation of ubiquitin-selective autophagy, which through their actions may preserve UPS activity. Thus, rather than being a consequence of a functional impaired UPS, the accumulation of ubiquitin conjugates may be part of the reason why the UPS remains operative even in the challenging intracellular environment of polyQ disease.

# **REDUCING LEVELS OF NEURODEGENERATION-ASSOCIATED PROTEINS**

In monogenic, dominant neurodegenerative disorders in which the identity of the disease-causing mutant proteins is known, as in the case of the polyQ disorders, an obvious but technically less trivial therapeutic approach would be to reduce the levels of the mutant protein. However, some of the gene products fulfill important functions in cells and may even be essential for viability, as is the case for huntingtin (Dragatsis et al., 2000) and ataxin-7 (Helmlinger et al., 2004). This problem may be circumvented by designing strategies to selectively target the mutated protein without affecting the wild-type form. Additionally, experiments in inducible mouse models for HD (Yamamoto et al., 2000) and SCA-1 (Zu et al., 2004) have shown that switching off expression of polyQ proteins reverses disease symptoms and neuronal pathology. A recent study has demonstrated that decreasing expression of polyQ-expanded ataxin-7 by 50 percent results in a full phenotypic rescue in a conditional SCA-7 mouse model (Furrer et al., 2013), which suggests that modest reduction of polyQ proteins may suffice for therapeutic intervention in humans. Targeting expression of mutant transcripts using RNA interference or antisense oligonucleotide technology is a promising approach to achieve this goal (Bonini and La Spada, 2005).

An alternative strategy for reducing levels of disease proteins is to accelerate their turnover by the UPS, whose function is preserved even at late stages of pathology. However, the question remains whether or not the proteasome can efficiently degrade disease-linked proteins. Some studies have suggested that the proteasome cannot degrade the polyQ proteins (Dyer and McMurray, 2001; Jana et al., 2001; Holmberg et al., 2004; Venkatraman et al., 2004) whereas others other reports show that they can be efficiently degraded by the proteasome as long as they remain in a soluble state (Verhoef et al., 2002; Kaytor et al., 2004; Michalik and Van Broeckhoven, 2004; Juenemann et al., 2013; Tsvetkov et al., 2013). Also, α-synuclein, expression levels of which can predispose individuals to PD (Singleton et al., 2003), can be degraded by the proteasome (Bennett et al., 1999), suggesting that other aggregation-prone proteins can be targeted by the UPS.

Several studies show that stimulating ubiquitylation and degradation of disease proteins through the UPS can rescue pathology in cell and animal models. For example, increasing the pool of free ubiquitin through genetic overexpression suppresses α-synuclein-induced neurodegeneration in *Drosophila* (Lee et al., 2009). The phenotypic rescue depends on the ability of the transgenic ubiquitin to form Lys48-linked polyubiquitin chains that typically target proteins for proteasomal degradation. Overexpression of certain ubiquitin ligases, which determine substrate specificity in the UPS pathway, also can confer protection in neurodegeneration models. Increasing levels of the ubiquitin ligase CHIP (C terminus of Hsc-70 interacting protein) delays the disease phenotype in SCA-1 (Al-Ramahi et al., 2006) and SBMA animal models (Adachi et al., 2007) through enhanced ubiquitylation and subsequent clearance of polyQ-expanded proteins. Moreover, Parkin has been shown to ubiquitylate ataxin-3 and reduces polyQ toxicity in cells (Tsai et al., 2003; Morishima et al., 2008). These findings show that the UPS is a plastic and versatile system that can be harnessed to accelerate the clearance of disease-linked proteins.

# **LEVERAGING PROTEASOMAL DEGRADATION**

Augmenting UPS activity or targeting its activity toward diseaseassociated proteins may be achieved through small molecules and, though challenging, opens up the possibility of counteracting protein accumulation in neurodegeneration (**Figure 3**). One such strategy takes advantage of the fact that aggregation-prone proteins are better substrates for the UPS in their monomeric, soluble state compared to oligomeric assemblies. Compounds, which stimulate expression or activity of heat shock proteins, effectively counteract aggregation and increase the clearance of misfolded proteins through the UPS. The compound arimoclomol induces chaperone expression and has been shown to ameliorate disease in SBMA (Malik et al., 2013) and ALS mice (Kalmar et al., 2012). YM-1, a small molecule that increases the ability of the molecular chaperone Hsp70 to bind unfolded substrates, increases degradation of polyQ-expanded AR in cell culture and rescues toxicity in SBMA flies (Wang et al., 2013).

Stimulation of UPS activity is another means by which degradation of disease protein can be achieved. An important question is whether such a complex system involving a large number of proteins can be effectively activated by small molecules in a therapeutic setting. Recent studies have revealed that the capacity of the UPS can be regulated via two related transcription factors, Nrf1 and Nrf2. Nrf1 is synthesized as an ER-anchored protein and only becomes transcriptionally active in the nucleus when increased UPS function is required (Radhakrishnan et al., 2010; Grimberg et al., 2011). Nrf2 regulates the antioxidant response in cells and can stimulate expression of proteasome subunits, which likely increases degradation of oxidized proteins during stress (Pickering et al., 2012). Nrf2 can be induced by the small molecule sulforaphane, which increases proteasome levels and activity (Kwak et al., 2003, 2007), and enhances UPS function *in vivo* (Liu et al., 2014b). Sulforaphane has been shown to reduce mutant huntingtin protein and mitigate polyQ toxicity in neuronal cells (Liu et al., 2014b).

Ubiquitin-dependent proteasomal degradation of aggregation-prone proteins can also be stimulated through direct modulation of proteasome function (Lee et al., 2010). The regulatory particle of the proteasome harbors two deubiquitylation enzymes, namely UCH-L5 and USP14/Ubp6, that counteract proteasomal degradation by trimming ubiquitin chains of recruited substrates (Finley, 2009). It is has been proposed that chain trimming by proteasome-associated

deubiquitylation enzymes may function as a molecular timer that determines the time window during which the proteasome can initiate degradation, thereby rescuing poorly ubiquitylated proteins from degradation (Lam et al., 1997). IU1, a selective small molecule inhibitor of USP14, has been shown to accelerate proteasomal degradation of aggregation-prone proteins, including proteins associated with neurodegenerative diseases such as tau, TDP43 and the polyQ protein ataxin-3, as well as oxidized proteins (Lee et al., 2010). However, it remains to be seen whether IU1 is effective in mitigating disease manifestations in animal models of neurodegenerative disorders.

# **SELECTIVE PROTEASOMAL TARGETING OF DISEASE PROTEINS**

Increasing overall degradation by stimulating UPS activity may be accompanied by unwanted side effects due to their general nature. Selectively targeting the disease-causing proteins for proteasomal is therefore expected to have many advantages over general stimulation of UPS activity. The specificity of the UPS action towards disease-linked proteins can be increased by taking advantage of the natural regulatory systems that dictate the degradation rates of these proteins. While this method has the potential of reducing side-effects of increasing overall UPS activity, it requires detailed insights in the biological function and regulation of the disease-causing protein.

Among the proteins implicated in polyQ diseases, AR is probably the one whose functions are best understood. AR is a nuclear hormone receptor that is linked to various types of cancers, most notoriously prostate cancer (Matsumoto et al., 2013). The idea of using small molecule inhibitors of heat shock protein 90 (Hsp90) as therapeutic tools in SBMA originates from cancer studies which revealed that AR is an Hsp90 client and requires heat shock proteins for proper functioning (Prescott and Coetzee, 2006). It has been shown that geldanamycin analogs 17-AAG and 17-DMAG accelerate degradation of AR and other Hsp90 client proteins and reduce polyQ-expanded AR toxicity in cell and mouse models (Waza et al., 2005; Tokui et al., 2009). Interestingly, geldanamycin is also effective in preventing aggregation of mutant huntingtin, which has been attributed to the ability of Hsp90 inhibitors to induce a general heat shock response (Sittler et al., 2001). Activation of the heat shock response may also contribute to lowering of AR half-life by geldanamycin analogs in SBMA models, as overexpression of heat shock proteins has been shown to also promote AR degradation (Bailey et al., 2002). Small molecules that promote the deposition of aggregation-prone proteins into IBs may also be beneficial in neurodegenerative diseases. For example, the compound B2 has been shown to increase inclusion formation of the mutant huntingtin fragment, polyQexpanded AR, and α-synuclein, and thereby suppresses toxicity in cell and fly models (Bodner et al., 2006; Palazzolo et al., 2010).

Protein stability can be influenced by posttranslational modifications. One example for this is again AR, whose levels and subcellular localization is controlled by the PI3K-Akt signaling pathway. It has been shown that phosphorylation of mutant AR by the Akt kinase, which can be induced by insulin-like growth factor-1 (IGF-1), accelerates clearance of the receptor by the proteasome (Palazzolo et al., 2007). Importantly, IGF-1 reduces mutant AR aggregation and toxicity in SBMA models *in vitro* and *in vivo* (Palazzolo et al., 2009; Rinaldi et al., 2012). More recently, it has been shown that the turnover of ataxin-1 is regulated by phosphorylation by MSK1 and, accordingly, the levels of polyQ-expanded ataxin-1 could be reduced by depletion of MSK1 (Park et al., 2013). The phosphorylation event is controlled by components in the MAPK-Ras pathway, and pharmacological curtailing of the MAPK pathway with small compound inhibitors of MEK1/2 or Raf1 similarly reduced levels and toxicity of mutant ataxin-1 (Park et al., 2013).

In recent years, scientists have begun to explore the possibility to mobilize branches of the UPS that are not normally involved in handling aggregation-prone proteins. Earlier studies have shown that engineered chimeric ubiquitin ligases can be used to redirect the ubiquitylation machinery to proteins of interest (Zhou et al., 2000). Engineered ubiquitin ligases typically contain two domains: a recognition domain specific to the protein of interest, and a ubiquitylation domain. Examples for the latter are RING, HECT, and U-box domains, which directly catalyze the addition of ubiquitin to the substrate. Ubiquitylation domains may also guide the substrate protein to the enzymatic activity of a ubiquitin ligase, such as the SCF complex via the F box domain (Zhang and Zhou, 2005). An example of engineered ubiquitin ligases in the context of neurodegenerative diseases is a chimeric Dorfin-CHIP fusion that combines the substrate binding domain of the ubiquitin ligase Dorfin with the U-box of CHIP. The resulting fusion protein was found to ubiquitylate mutant SOD1 more efficiently than the wild-type Dorfin or CHIP ubiquitin ligases (Ishigaki et al., 2007). Off-target effects due to residual activity toward their native substrates, or unwanted regulation in the cellular microenvironment are a major caveat with current chimeras based on domains derived from endogenous proteins. However, an important advantage of the chimeras is that they can be designed with high specificity to certain sub-populations of the protein-of-interest. For instance, a recent study reported the generation of an SCF-βTrCP ubiquitin ligase that selectively targets only activated ErbB receptor tyrosine kinases for degradation (Kong et al., 2014).

Nanobody-based fusions are variations on protein chimeras based on single chain antibodies, which, when expressed in the intracellular environment, bind to target proteins with high specificity (Muyldermans, 2013). A fusion consisting of an F-box domain and a nanobody directed to GFP has been successfully used to deplete cytoplasmic, nuclear, and transmembrane GFPfusion proteins in cultured cells and *in vivo* via the SCF complex (Caussinus et al., 2012). In theory, this approach may also be used to engineer intrabody-based ubiquitin ligases specific to diseaseassociated proteins, or even to pathogenic protein conformations. Although engineered ubiquitin ligases can be of great value in experimental settings and provide us with a proof-of-principle that ubiquitylation can be tailored to accelerated degradation of desired proteins, there are currently many practical limitations that limit their adaptation to therapeutic settings, such as safety and delivery to target tissues.

A number of studies have explored whether small molecules can be used to redirect endogenous ubiquitylation enzymes to defined target substrates. Most interesting in this respect is the work on molecules known as Protacs (Proteolysis-targeting chimeric molecules) and SNIPERs (Specific and non-genetic IAP-dependent Protein Erasers) (Buckley and Crews, 2014). Both types of molecules combine a ubiquitin ligase interaction peptide with a small molecule ligand specific for the target protein (Sakamoto et al., 2001). Protacs have mainly been developed to target cancer-relevant gene products but may be applicable as well to misfolded proteins in neurological disorders. One example is a chimeric molecule that unites a peptide motif from HIF1α with the AR ligand dihydrotestosterone (DHT) and can be used to accelerate degradation of AR in cultured cells (Rodriguez-Gonzalez et al., 2008). Although it would be interesting to test whether this molecule can also accelerate the degradation of mutant AR in SBMA or whether derivatives can be generated for other polyQ proteins, a limiting factor in the development of Protacs is poor permeability across cell membranes and potential antigenicity due to their peptidic nature. However, in the cases where the ubiquitin ligase can be recruited by small synthetic molecules instead of polypeptides, it may be possible to obtain compounds with potential therapeutic value. A functional cellpermeable Protac combines two active molecules, the AR ligand DHT and the Mdm2-interacting compound Nutlin, to increase ubiquitylation of AR by Mdm2 (Schneekloth et al., 2008). In SNIPERs, an IAP1-interacting bestatin ester replaces the peptide domain found in Protacs that links the chimeric molecule to ubiquitin ligase activity. SNIPERs have been used successfully in the context of steroid receptors, including AR, and efficiently reduce receptor protein levels at lower molar concentrations than Protacs (Okuhira et al., 2013). Small molecules for targeted degradation of disease-associated proteins by the UPS have so far mainly been explored in the cancer field, but the successful application in malignant cells may serve as a proof-of-principle to warrant their investigation in neurodegenerative disorders.

# **CONCLUDING REMARKS**

Increasing evidence suggests that ubiquitin-dependent proteolysis is largely operative in many neurodegenerative diseases, and hence a causal relationship does not exist between aggregationprone proteins and global UPS impairment. While perturbations in UPS function due to disease proteins cannot be excluded, adaptive responses such as IB formation and autophagy likely contribute to the restoration of cellular protein homeostasis. This knowledge justifies further exploration of the protein degradation machinery for treating or preventing debilitating neurodegenerative disorders. However, in order to fully appreciate the potential of the UPS as a therapeutic target, we need to continue to decipher how the UPS and compensatory pathways coordinate the detoxification and clearance of misfolded proteins.

While the current data do not support a scenario in which global impairment of the protein quality control function of the UPS lies at the basis of proteinopathies, our present understanding strongly supports an involvement of ubiquitin-dependent processes in the development or progression of neurodegenerative diseases. Dissecting the molecular mechanisms and key players responsible for changes in ubiquitylation may help to identify suitable therapeutic targets. Because it appears to be unlikely that this inhibitory activity is a shared intrinsic feature of aggregation-prone proteins, it will be important to study the interplay between the UPS and the individual disease-associated proteins in the context of their native functions.

This change of perspective from the UPS as a potential cause for neurodegeneration to it being a preserved proteolytic pathway that may be exploited in therapeutic approaches has been an important contribution of the large number of studies that have probed into the role of the UPS in polyQ disorders. Inspiring examples can be found in the identification of compounds which stimulate UPS activity and the development of small molecules that target desired proteins for proteasomal degradation. Although these studies provide us with proofs-of-principle for redirecting proteins to endogenous ubiquitin ligases, there is a clear need for compounds with more favorable drug-like properties in order to accomplish this aim in a physiological setting. If successful, such small molecules could be used on their own or in combination with other polyQ protein-reducing therapies, such as RNA interference, to reduce the levels of the toxic proteins in patients. Despite the fact that much needs to be done before we can start to evaluate the applicability of such approaches, the recent insight that at least the UPS is still on our side in fighting the toxic effects of the these rogue proteins opens other opportunities in the pursuit for therapeutics for neurodegenerative diseases.

## **ACKNOWLEDGMENTS**

We thank Florian Salomons and Emily Foran for critical reading of the manuscript. The Dantuma lab is supported by the Swedish Research Council, the Swedish Cancer Society and the Karolinska Institute.

# **REFERENCES**


repeat androgen receptor in a cellular model of spinal and bulbar muscular atrophy. *Hum. Mol. Genet.* 11, 515–523. doi: 10.1093/hmg/11.5.515


mutant fALS-typical superoxide dismutases. *J. Neurochem.* 83, 1019–1029. doi: 10.1046/j.1471-4159.2002.01232.x


clearance without affecting proteasome catalytic activity. *Cell Death Dis.* 2:e196. doi: 10.1038/cddis.2011.81


model of spinal and bulbar muscular atrophy. *Neuron* 63, 316–328. doi: 10.1016/j.neuron.2009.07.019


autophagic clearance of protein inclusions associated with neurodegenerative diseases. *Hum. Mol. Genet.* 17, 431–439. doi: 10.1093/hmg/ddm320


**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.

*Received: 11 May 2014; accepted: 09 July 2014; published online: 31 July 2014. Citation: Dantuma NP and Bott LC (2014) The ubiquitin-proteasome system in neurodegenerative diseases: precipitating factor, yet part of the solution. Front. Mol. Neurosci. 7:70. doi: 10.3389/fnmol.2014.00070*

*This article was submitted to the journal Frontiers in Molecular Neuroscience.*

*Copyright © 2014 Dantuma and Bott. 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 ubiquitin proteasome system in glia and its role in neurodegenerative diseases

# *Anne H. P. Jansen1, Eric A. J. Reits 1\* and Elly M. Hol 2,3,4 \**

<sup>1</sup> Department of Cell Biology and Histology, Academic Medical Center, Amsterdam, Netherlands

<sup>2</sup> Department of Translational Neuroscience, Brain Center Rudolf Magnus, University Medical Center Utrecht, Utrecht, Netherlands

<sup>3</sup> Netherlands Institute for Neuroscience, Institute of the Royal Netherlands Academy of Arts and Sciences, Amsterdam, Netherlands

<sup>4</sup> Swammerdam Institute for Life Sciences, Center for Neuroscience, University of Amsterdam, Netherlands

### *Edited by:*

Fred van Leeuwen, Maastricht University, Netherlands

### *Reviewed by:*

Eldi Schonfeld-Dado, Stanford University, USA Inbal Goshen, The Hebrew University, Israel

### *\*Correspondence:*

Eric A. J. Reits, Department of Cell Biology and Histology, Academic Medical Center, Meibergdreef 9, 1105 AZ Amsterdam, Netherlands e-mail: e.a.reits@amc.uva.nl; Elly M. Hol, Department of Translational Neuroscience, Brain Center Rudolf Magnus, University Medical Center Utrecht, Universiteitsweg 100, Utrecht, Netherlands e-mail: e.m.hol-2@umcutrecht.nl

The ubiquitin proteasome system (UPS) is crucial for intracellular protein homeostasis and for degradation of aberrant and damaged proteins. The accumulation of ubiquitinated proteins is a hallmark of many neurodegenerative diseases, including amyotrophic lateral sclerosis, Alzheimer's, Parkinson's, and Huntington's disease, leading to the hypothesis that proteasomal impairment is contributing to these diseases. So far, most research related to the UPS in neurodegenerative diseases has been focused on neurons, while glial cells have been largely disregarded in this respect. However, glial cells are essential for proper neuronal function and adopt a reactive phenotype in neurodegenerative diseases, thereby contributing to an inflammatory response. This process is called reactive gliosis, which in turn affects UPS function in glial cells. In many neurodegenerative diseases, mostly neurons show accumulation and aggregation of ubiquitinated proteins, suggesting that glial cells may be better equipped to maintain proper protein homeostasis. During an inflammatory reaction, the immunoproteasome is induced in glia, which may contribute to a more efficient degradation of disease-related proteins. Here we review the role of the UPS in glial cells in various neurodegenerative diseases, and we discuss how studying glial cell function might provide essential information in unraveling mechanisms of neurodegenerative diseases.

**Keywords: astrocytes, microglia, oligodendrocytes, gliosis, ubiquitin proteasome system, neurodegenerative diseases**

# **THE UBIQUITIN PROTEASOME SYSTEM**

Protein homeostasis is essential for proper function of a cell; therefore both protein synthesis and degradation are tightly regulated. Although there is a complex interplay between the two systems, in general, there are two main machineries involved in protein degradation: autophagy and the ubiquitin proteasome system (UPS). Autophagy involves lysosomal degradation by the formation of intracellular vesicles (autophagosomes) and is subdivided in chaperone-mediated autophagy (CMA), microautophagy, and macroautophagy. In CMA, chaperone proteins bind specifically to a KFERQ domain in a cytosolic protein, afterward it is internalized and degraded. Microautophagy is the direct lysosomal digestion of cytoplasmic content, which is trapped by random invagination of the lysosomal membrane. Macroautophagy functions as bulk degradation of mainly longlived proteins, protein aggregates and organelles, and involves autophagosome formation to isolate cytoplasmic proteins. The UPS selectively targets individual proteins, including short-lived, damaged or defectively folded proteins, which accounts for about 80–90% of all intracellular proteins (Rock et al., 1994; Lilienbaum, 2013).

The UPS consists of two key components: the ubiquitination system, which selects and targets proteins towards degradation by ubiquitinating them, and the proteasome, a multimeric protein complex that actually performs the degradation. Protein degradation is a tightly regulated process: before a protein is cleaved by the proteasome, an elaborate process of selection and targeting has taken place exerted by the ubiquitin (Ub) system. This system mediates the conjugation of Ub, a small 76-aminoacid-long modifier. Binding of Ub to the target protein takes place in a three-step reaction. First, Ub is linked to an Ub-activating enzyme (E1) in an ATP-dependent manner. Subsequently, the activated Ub is transferred to an E2 conjugating enzyme, followed by attachment of E2 to a specific E3 ubiquitin ligase enzyme that binds the target protein. Lastly, Ub is transferred by the E2 enzyme to the target protein (**Figure 1**). Ub itself can be ubiquitinated at one of its seven lysine residues, resulting in various polyubiquitin chain types that each has its own specific signal function. Regulation of protein degradation is also mediated by deubiquitinating enzymes (DUBs) that can reverse ubiquitination by removing Ub residues of mono- or polyubiquitinated proteins (Glickman and Ciechanover, 2002; Lilienbaum, 2013).

Polyubiquitinated proteins are generally degraded by the 26S proteasome, which consists of a barrel-shaped 20S core that is the actual protease and a 19S regulatory complex that recognizes and unfolds the ubiquitinated substrate (**Figure 1**). The 20S core consists of two rings of seven α subunits, flanking two rings of seven β subunits. Three of these seven β subunits are catalytically active, with a caspase-like (β1), trypsin-like

(β2), or chymotrypsin-like (β5) activity that cleave after negatively charged, positively charged, and hydrophobic amino acids, respectively (Lilienbaum, 2013). The 19S regulatory complex recognizes and unfolds the ubiquitinated substrate and guides it into the 20S core. The 19S complex consists of 19 individual proteins, of which six are ATPases of the AAA family (Baumeister et al., 1998). Binding of ATP is necessary for assembly of the 20S core with the 19S cap, subsequently ATP hydrolysis is required for protein unfolding, while for the next steps, gate opening, translocation and degradation, only ATP binding is required (Smith et al., 2005).

Next to the 19S complex, also other regulatory proteasome activators (PA) exist, including 11S (PA28), PI31, and PA200, which bind the 20S core in an ATP-independent way (Lilienbaum, 2013). PA28αβ expression is induced upon secretion of interferon gamma (IFNγ) and plays a role in processing peptides for MHC class I antigen presentation (Realini et al., 1994; Sijts and Kloetzel, 2011). When PA28αβ binds to 20S core, the activities of all β-subunits are increased, which is probably due to an increased accessibility rather than alterations within the active sites themselves. In contrast to PA28αβ, the exact mechanism that PA28γ, which is only expressed in the nucleus, uses to exert its function is not known (Rechsteiner and Hill, 2005).

IFNγ not only induces different PA caps but also induces the expression and incorporation of the proteasomal immunosubunits β1i (LMP2), β2i (Mecl-1), and β5i (LMP7) that substitute β1, β2, and β5, respectively. With this replacement, the chymotrypsin-like activity increases and the immunoproteasome shows an altered cleavage pattern, resulting in different peptides being generated that are subsequently presented by MHC class-I molecules to the cells of the immune system (Sijts and Kloetzel, 2011; Basler et al., 2013).

# **GLIAL CELLS**

Neuroglia were first described by the German pathologist Rudolf Virchow already more than 150 years ago. For a long time these cells were considered as nerve glue, a kind of connective tissue holding the brain together. In contrast to neurons, glia are not electrically excitable. This made it difficult to study these cells, and as a consequence, most neuroscientific research was focused on neuronal function. When sophisticated molecular tools became available which made it possible to study the physiology of glial cells, this has led to a change in the neurocentric view of neuroscientists. During the last decades, it has become evident that glial cells are essential for proper neuronal function, are actively involved in neuronal communication, and form the immune system of the brain. As diverse as different glial cells are in morphology and origin, as diverse they are likely in their function. They take care of the general homeostasis in the brain, insulate neurons, and protect against pathogenic invaders (Kettenmann and Verkhratsky, 2008). The ratio between neurons and glia in the human central nervous system (CNS) is about 1:1, with oligodendrocytes being the most abundant type of glial cells (75.6%), followed by astrocytes (17.3%) and microglia (6.5%) in human male brains (Pelvig et al., 2008). Below, the most important glial cell types and their functions are described (**Figure 2**).

# **ASTROCYTES**

Astrocytes perform a variety of crucial tasks in the CNS. Similar to neurons, astrocytes are of ectodermal origin. During the first stages of embryogenesis, the first astrocyte precursor cells arise from neural stem cells. These radial glia function as neural progenitors and form the scaffold that is used by immature neurons to migrate toward their final location. In a later stage, these cells give rise to the progenitors of oligodendrocytes as well as to the different kinds of astrocytes in the brain. The astrocyte population is broad and heterogeneous and the classification of subtypes is in its initial stage. Astrocytes contact multiple blood vessels with their endfeet, and they connect to neighboring astrocytes via gap junctions. Besides, astrocytes envelope multiple synapses and fibrous astrocytes in the white matter contact several nodes of Ranvier, which are the small gaps in myelinated axons (Sofroniew and Vinters, 2010; Molofsky et al., 2012). Subventricular astrocyte-like cells are the stem cells in the brain. The progeny of these cells migrate through the rostral migratory stream into the olfactory bulb, and differentiate into interneurons. When isolated and brought in culture, they can differentiate in both neurons and glia thus exhibiting multipotent properties (Mamber et al., 2013). These cells are even still present in aged brains, suggesting regenerative potential until late in life (Van Den Berge et al., 2010).

In the mature brain, astrocytes influence synaptic transmission directly by the release of gliotransmitters including glutamate, ATP, GABA, and D-serine, although it is not fully understood how the cells release these transmitters. Astrocytes also respond to neurotransmitters, which leads to a surge in

calcium. Although astrocytes do have voltage-gated channels (Pappalardo et al., 2014), they are not able to respond with an increase in calcium in a millisecond time scale. Hence, astrocytes play a modulatory role: they are likely to influence synaptic communication of the many neurons in their domain. Astrocytes can also regulate synaptic transmission by mediating homeostasis of ions, pH and (neuro)transmitters. A well-known function of astrocytes is uptake of glutamate from the synaptic cleft via glutamate transporter Glt-1. Moreover, astrocytes contain transporters for the neurotransmitters GABA and glycine, for K+ ions, bicarbonate, water and monocarboxylic acid (Sofroniew and Vinters, 2010). Recently, it has been shown that astrocytes are able to influence synapses also by synaptic pruning both in the developing and in the adult mouse brain. This phagocytosis of synapses is dependent on neuronal activity and is mediated via the Mertk and Megf10 pathways. Thus, astrocytes actively contribute to activity-dependent synapse elimination and CNS remodeling (Chung et al., 2013). Next to the phagocytosis of synapses, activated astrocytes are able to phagocytize amyloid β deposits *in vitro* and *in situ* (Wyss-Coray et al., 2003).

Furthermore, astrocytes form a key compartment of the blood brain barrier (BBB); they are not only involved in induction and development of the BBB, they also regulate BBB permeability (Abbott, 2002). Astrocytes are connecting the blood vessels with many neuronal perikarya, axons and synapses. Therefore, astrocytes are the ideal cells to make an important contribution to the energy supply of the brain; for example by taking up glucose from the blood via their glucose transporters.

In conclusion, astrocytes are a heterogeneous group of glial cells that exert numerous tasks essential for proper neuronal function. They are indispensable in neurodevelopment, synaptic communication and brain homeostasis. As a consequence, loss or change in astrocyte function can contribute to pathogenesis of neurodegenerative diseases. A clear example is the leukodystrophy Alexander's disease, which is caused by a mutation in the astrocyte specific gene GFAP that leads to astrocyte pathology and dysfunction (Brenner et al., 2001; Sosunov et al., 2013).

### **OLIGODENDROCYTES**

Oligodendrocytes are the myelinating cells of the brain, and form an insulating myelin sheath around the axonal segments of neurons. Like astrocytes, oligodendrocytes are of ectodermal origin; they arise from the neural stem cells in the neuroepithelium during early embryogenesis. The onset of myelination, as well as the selection of axons that require myelination, are both tightly regulated. Neuronal activity and degree of differentiation, together with several surface receptors (e.g., LINGO-1), influence the brain area and specific axons that are myelinated (Baumann and Pham-Dinh, 2001; Bradl and Lassmann, 2010). As oligodendrocytes have only a short time window wherein they are capable of myelination, this provides temporal control during early differentiation as well (Baumann and Pham-Dinh, 2001). A single oligodendrocyte can enwrap multiple axons, while a single axon can have adjacent myelin sheets belonging to different oligodendrocytes. Myelination of axons is important for a highspeed, reliable conduction of electrical signals between neurons. The myelinated fibers contain small gaps named nodes of Ranvier that are important for the fast signal conduction, so the action potential can jump from node to node. During the peak of myelination, about three times the weight of the oligodendrocyte is produced in myelin each day. Small changes in protein homeostasis will therefore have huge consequences for protein production and quality control, which can result in misfolding and accumulation of proteins and cause dysfunctional oligodendrocytes and eventually demyelination. Illustratively, transgenic

rat overexpressing the oligodendrocyte proteolipid protein showed oligodendrocyte apoptosis and dysmyelination (Baumann and Pham-Dinh, 2001; Bradl and Lassmann, 2010). Axons can be remyelinated by new oligodendrocytes arising from the NG2+ cells that are stimulated to divide and differentiate by activated microglia and astrocytes. However, the resulting myelin sheath is thinner, possibly due to the lack of stimulating growth factors that were initially secreted by the growing axon. Oligodendrocyte dysfunction is directly detrimental for neuronal function and leads to neurodegeneration in the α-synucleinopathy multiple system atrophy (MSA), but also in multiple sclerosis and other leukodystrophies (Baumann and Pham-Dinh, 2001; Kuzdas et al., 2013).

## **MICROGLIA**

Microglia primarily function as the immune cells of the CNS. In contrast to astrocytes and oligodendrocytes, microglia are from mesodermal origin (Ginhoux et al., 2010) and are derived from hematopoietic stem cells in the yolk sac early in embryonic development (Alliot et al., 1999). Although microglia are mostly known for their function in the immune response under disease conditions, they also play an important role in CNS homeostasis. During neuronal development, microglia regulate brain plasticity, by controlling synapses (Kettenmann et al., 2013). Next to phagocytizing apoptotic cells and pruning synapses during development, microglia are actively involved in promoting apoptosis, which is also an important aspect of neuronal development (Kettenmann et al., 2013). In the adult brain, microglia also phagocytize dead neurons and express several receptors for neurotransmitters, -peptides, and -modulators, thereby sensing (changes in) neuronal activity. Under healthy conditions, microglia have a small nucleus and long, thin, very motile processes, and although called "resting microglia," these cells are constantly monitoring the environment and when dying neurons are detected, microglia become activated. Subsequent phagocytosis is a tightly regulated process involving multiple receptors like CD36, lectin, and integrin receptors (Kettenmann et al., 2013).

Next to neuronal damage, invading pathogens are detected and cleaned up by microglia (Aloisi, 2001). To recognize and eliminate foreign entities, a variety of pattern recognition receptors are expressed including complement receptor 3 and toll-like receptors (TLR) 2, 4, and 9 that promote phagocytosis and recognize bacterial lipopolysaccharide (LPS; TLR2, 4) or DNA (TLR9). These cytokines are secreted by CNS associated macrophages, astrocytes and by other microglia as a paracrine activation mechanism. TNFα, IL-1, and IFNγ are pro-inflammatory cytokines that activate the immune functions of microglia including phagocytosis, cytokine production and antigen presentation. The immune response is regulated by anti-inflammatory cytokines such as IL-10, TGF-β, which downregulate the expression of proinflammatory cytokines, reactive oxygen species (ROS), chemokines and other molecules associated with phagocytosis in microglia (Aloisi, 2001; Tambuyzer et al., 2009).

# **GLIA IN PATHOLOGY**

Early pathological studies showed astrogliosis and microglial proliferation in damaged brain tissues in several neurodegenerative diseases including Huntington's disease (HD; Faideau et al., 2010), Alzheimer's disease (AD; Sheng et al., 1997; Mrak, 2009), but also in inflammatory diseases, brain trauma, ischemia and infection (Gray et al., 1996; Tambuyzer et al., 2009; Amor et al., 2013; Pekny et al., 2014). Since microglia sense neuronal damage, they might be the initiators of the glial reaction to neuronal pathology, which is called neuroinflammation (reactive gliosis). Activated microglia have been shown to secrete cytokines that induce astrocyte reactivity. This is a delayed reaction; microglia are sensitive for the smallest pathological changes in the CNS, but only when the damage is severe enough this will lead to astrogliosis. Astrogliosis is characterized by increased expression of the astrocyte-specific intermediate filament GFAP (Pekny et al., 2014) and vimentin (Parpura et al., 2012). Astrocytes express cytokine receptors for, e.g., IL-1, IL-6, and TNF-α, and microglia might disturb normal astrocyte function by activating astrocytes via these receptors. For example, TNF-α affects glutamate transmission directly by inhibiting expression of astrocytic glutamate transporters (Cumiskey et al., 2005). As a result of activation, astrocytes release cytokines that can in turn influence microglial function (Amor et al., 2013). Hereby a feedback loop is created, in which factors from both astrocytes and microglia regulate each other (Zhang et al., 2010).

It is unclear whether glial activation under pathological circumstances is beneficial or detrimental. Activation of astrocytes is initially meant to protect neurons by forming a protective border around the damaged site (Pekny et al., 2014), but loss-of-function of astrocytes such as reduced glutamate uptake may also result in neuronal dysfunction (Cumiskey et al., 2005). LPS-induced microglial activation is associated with subsequent astrocyte and oligondendrocyte impairment and demyelination. LPS induction of microglia caused a decrease in expression in several astrocytic proteins that are important for their normal function (e.g., Aqp4, a water transport channel). Astrocytes connect to each other and to oligodendrocytes via gap junctions, and activated astrocytes show reduced expression of connexins (30 and 43) that mediate these junctions. As ions, water and osmolites are exchanged via these connexins, loss of this connection leads to decreased oligodendrocyte function and subsequent demyelination and neuronal impairment (Sharma et al., 2010). In astrocytes, Dicer ablation causes changes in the transcriptome resulting in the upregulation of many immature/reactive astrocyte genes, while astrocytic genes related to mature astrocyte functions are downregulated (e.g., GLT-1). The alterations in astrocyte phenotype occur already before neuronal deficits were perceived, underlining the importance of proper astrocyte function in the brain (Tao et al., 2011).

# **A DIVERGENCE IN UPS BETWEEN NEURONS AND GLIA**

Since glia are essential for proper neuronal function, disturbances in glial function can lead to excitotoxicity and neurodegeneration. However, in neurodegenerative diseases with protein inclusions (proteinopathies), neurons appear to be most vulnerable, as protein aggregation and cell degeneration is mainly observed in neurons. Possibly these differences are due to dissimilarities in their protein homeostasis. Indeed, distinct cell types show divergent protein synthesis and degradation (Vilchez et al., 2012). In addition, differences may exist in UPS levels and activity between neurons and the various glial cells. The UPS is the main protein complex involved in the degradation of oxidized proteins, and oxidation of nucleic acids, lipids and proteins is associated with aging and neurodegenerative diseases (Mariani et al., 2005). Proteasome inhibition causes an increase in nucleic acid oxidation in both primary neurons and astrocytes; however, this increase is much larger in neurons. This suggests that neurons are more sensitive to proteasome inhibition, or are more prone to oxidative stress, which is associated with neurodegenerative diseases like AD and Parkinson's disease (PD; Ding et al., 2004). The UPS appears to be less active in neurons in comparison to white matter glia (Tydlacka et al., 2008). Intriguingly, in response to cytokines such as TNF-α, IL-2, G-CSF, and IFNγ, neurons upregulate the Ub-like protein ubiquitin D (or Fat10; Lisak et al., 2011), which is able to target long-lived proteins to the proteasome. Conversely, both astrocytes and microglia failed to upregulate Fat10 upon cytokine stimulation (Hipp et al., 2005), also indicating that the UPS is regulated in neurons in another way. In contrast to most other neurodegenerative diseases, in intranuclear inclusion body disease (INIBD) the largest proportion of aggregates are found in glial cells (about 5% in glial cells, compared to 1% in neurons). Here, most aggregates are present in astrocytes, and a smaller proportion in oligodendrocytes. All aggregates are positive for Ub, however, glial cells seem to have a significantly smaller proportion of aggregates positive for the Ub-like proteins NEDD8, NUB1, and SUMO1 than neurons. This slightly dissimilar composition of the aggregates can be explained by either variations in protein expression or protein recruitment toward the aggregates in the cell types (Mori et al., 2012). In conclusion, neurons and glia are differently reacting to proteasome inhibition and cytokines. Besides, presence of UPS components in inclusion bodies seem to diverge between neurons and glia, which could be indications why neurons are more vulnerable in neurodegenerative diseases.

# **UPS IN GLIA IN RELATION TO NEURODEGENERATIVE DISEASES**

The tight balance between protein synthesis and degradation is essential for cellular function. In the brain, disturbances in either of the two cause the accumulation and aggregation of short-lived and misfolded proteins. In aging neurons, UPS activity has been shown to be decreased (Tydlacka et al., 2008), accordingly, age-related alterations in proteasomal activity are implicated in various neurodegenerative diseases (Tydlacka et al., 2008; Tashiro et al., 2012; Lin et al., 2013; Orre et al., 2013). It remains to be examined in more detail whether and how proteasome levels and activity differ between neurons and glia, which would obviously affect the capacity of cells to maintain proper protein homeostasis. In many diseases protein aggregates are hallmarks that are clearly visible in neurons, and are indicative for neuronal dysfunction. However, aggregates are also

described in glial cells, for example in HD (Shin et al., 2005; Tong et al., 2014) and amyotrophic lateral sclerosis (ALS; Bruijn et al., 1997), although they are usually smaller or less abundant. The discrepancy between the occurrence of neuronal and glial inclusions implies that the UPS system is more efficient to remove aggregation-prone proteins in glia. Yet, while glia might be better able to handle these proteins, this does not mean that glia are not affected. Mainly astrocytes and microglia react to the increase of aggregated proteins in the brain and show altered function, thereby contributing to neuronal dysfunction. Many studies have shown that neuronal UPS dysfunction plays an important role in several neurodegenerative diseases. The first indications were found already more than 25 years ago, when components of the UPS were found in inclusions of several neurological diseases including PD, Pick's disease but also in Rosenthal fibers in astrocytomas (Lowe et al., 1988). Depletion of 26S proteasome in mice neurons led to significant neurodegeneration and inclusion body formation (Bedford et al., 2008). Although the depletion was restricted to neurons, changes in glia were observed as astrocytes showed an increase in GFAP and vimentin protein expression, indicative of reactive gliosis (Elkharaz et al., 2013). Interestingly, astrocytes themselves seem to be less sensitive for proteasome inhibition. It has recently been suggested that this is due to the high expression of the heat shock protein HSP25 (Goldbaum et al., 2009). Although inhibition of proteasomes by the reversible proteasome inhibitor MG-132 caused aggresome formation and cytoskeletal disturbances in cultured primary astrocytes, cell viability was only diminished by 20%. Besides, cytoskeletal disturbances appeared to be reversible in astrocytes, whereas in oligodendrocytes proteasome inhibition massively induced apoptosis. This effect might be caused by the induction of the HSP25, which is expressed at a much higher level in astrocytes and interacts with all three types of cytoskeletal filaments in astrocytes. As downregulation of this protein indeed resulted in fragmentation of actin networks, these results suggest a protective role of HSP25 in the astrocyte cytoskeleton during proteasome inhibition (Goldbaum et al., 2009). Proteasome inhibition does not only change cell viability, it has also been shown to decrease intermediate filament transcription in astrocyte cell lines. Treatment with proteasome inhibitors epoxomicin and MG-132 decreased the mRNA levels of the astrocyte-specific intermediate filament GFAP in different astrocyte cell lines. Similarly, vimentin and nestin expression were significantly decreased upon proteasome inhibition in astrocytes, but not in neuronal cells. Moreover, rat brains treated with proteasome inhibitors through a cannula showed less astrogliosis around the cannula. It appeared that the proteasome can affect GFAP promotor activity by the degradation of its transcription factors, thereby linking UPS activity directly to astrogliosis (Middeldorp et al., 2009).

Cultured primary oligodendrocytes seem to be much more sensitive to proteasome inhibition when compared to astrocytes. Treatment of these cells with MG-132 caused oxidative stress, mitochondrial dysfunction and apoptosis after 18 h (Goldbaum et al., 2006). While inhibition of the 26S regulatory subunit 7 (RPT1, part of 19S) specifically reduced pro-inflammatory cytokine secretion in LPS-stimulated cultured microglia (Bi et al., 2012), in general also microglia showed a decreased survival and an increase in pro-inflammatory response following proteasome inhibition, including upregulation of nitric oxide and TNF-α secretion (Kwon et al., 2008). LPS induction also triggers the secretion of these cytokines by microglia, while intracellularly the pro-inflammatory NF-κB pathway is upregulated. This process is controlled by UPS components, with E3 ligase RING finger protein 11 (RNF11) as one of the key negative regulators of the NF-κB pathway in microglia (Dalal et al., 2012). In neurons and microglia, IFNγ caused induction of the immunoproteasome by replacing the constitutive subunits β1, β2, and β5 with the inducible immuno subunits β1i, β2i, and β5i and the association with the regulatory complex PA28αβ. Synergistically, LPS-induced neuroinflammation can aggravate neurodegeneration that is triggered by proteasome inhibition (Stohwasser et al., 2000; Pintado et al., 2012).

Since both proteasome inhibition and neuroinflammation are associated with neurodegenerative diseases, studying the interplay between those two will provide new insights in neurodegenerative disease mechanisms. In the following section, proteasome dysfunction and its effect on glia in several age-related neurodegenerative diseases are described and in **Table 1** an overview of the literature on a wider range of neurodegenerative diseases is given.

### **ALZHEIMER'S DISEASE**

Alzheimer's disease is the most prevalent form of dementia and is characterized by irreversible memory loss and impaired cognitive functions. Magnetic resonance imaging (MRI) scans of AD patient brains reveal atrophy and an enlargement of the ventricles caused by neuronal shrinkage or cell loss, and loss of (synaptic) connections is thought to be the main contributor to the cognitive and memory impairments (Bozzali et al., 2011). Neuropathological hallmarks of AD are senile plaques and neurofibrillary tangles. Neurofibrillary tangles are formed by the hyperphosphorylated microtubule-associated protein tau and are located mainly inside neurons, while extracellular the senile plaques consist of aggregates formed by the peptide amyloid β (Aβ). Over the last two decades, the concept of a neuroinflammatory involvement in AD has been established. This started in 1986 with the discovery of plaque-associated reactive microglia, and additional studies described the association of both activated microglia and astrocytes with Aβ plaques expressing pro-inflammatory markers (Rozemuller et al., 1986; Griffin et al., 1989; Mrak, 2009). Astrocytes change their phenotype when contacting extracellular Aβ, and their response includes upregulation of glial fibrillary acidic protein (GFAP), vimentin and S100β (Griffin et al., 1989; Kamphuis et al., 2012). In AD, both astrocytes and microglia play an important role in clearance of Aβ plaques (Wyss-Coray et al., 2003; Mandrekar et al., 2009; Mulder et al., 2014). Besides, attenuation of astrocyte


**Table 1 | Overview of published data on the role of glia and the UPS in several neurodegenerative diseases.**

activation leads to a higher plaque load and increased microglial activation, probably as a compensatory mechanism (Kraft et al., 2013). A loss in metabolic function of activated astrocytes worsens the disease, and upregulation of GFAP has been shown to correlate with downregulation of the astrocyte-specific glutamate transporter EAAT2 (Simpson et al., 2010; Yan et al., 2013).

Changes in the UPS pathway have been associated with AD in several studies. Ub is accumulating in both plaques and tangles (Chu et al., 2000). Next to the regular Ub protein in the brains of AD and Down syndrome patients, both having the plaque pathology, a unique form of ubiquitin B can be found. This protein originates from a frameshift mutation in the mRNA and leads to a protein with an aberrant C-terminus, called UBB+1. It can be detected in all AD and Down syndrome patients, but in none of the non-demented controls (van Leeuwen et al., 1998). Because of the absence of the carboxyterminal Gly-76, UBB+1 cannot ubiquitinate other proteins, but is itself efficiently ubiquitinated. However, the proteasome seems unable to degrade large amounts UBB+1, leading to proteasome inhibition (Lam et al., 2000; van Tijn et al., 2007). In addition, UBB+1 expression also changes the immune response in astrocytes upon TNF-α and IFNγ treatment as secretion of the chemokines CCL2 and CXCL8 is increased dramatically. This is probably due to an upregulation of several proteins in the NF-κB and JNK pathways, suggesting that a disruption in the UPS directly influences pro-inflammatory signaling in astrocytes and *vice versa* (Choi et al., 2013).

Importantly, alterations in the UPS can influence the degradation of Aβ in both neurons and astrocytes. Although mainly the neuronal viability is affected in response to increasing Aβ levels, also astrocytes showed a similar upregulation in UPS related proteins and a decrease in proteasome activities upon Aβ treatment as neurons (Lopez Salon et al., 2003). Treatment with Aβ oligomers decreased proteasome activity *in vitro* and in mouse brain lysates, and a lower proteasome activity was also observed in lysates of several brain areas of AD patients (Gregori et al., 1995; Keller et al., 2000; Tseng et al., 2008; Zhao and Yang, 2010). However, these measurements were mainly performed in whole brain homogenates or in (neuronal) cell lines, and may therefore fail to see alterations in immunoproteasome levels and activity in glial cells since that was not investigated in these studies. More recently, it was shown that Aβ treatment led to an increase in proteasome activity in both cultured neurons, astrocytes and microglia. In addition, it was discovered that both mRNA and protein levels of immunoproteasome subunits β5i and β1i were upregulated in reactive astrocytes and microglia around plaques in AD mice and human AD patient material, whereas in AD mice the levels of β2i were similar to controls. Consistently, the activity of all immunoproteasome subunits was increased with increasing plaque load in both AD mice and in post-mortem material of human AD patients (Nijholt et al., 2011; Orre et al., 2013). These results are in line with earlier data showing that β1i expression levels in neurons and astrocytes were increased with age, and in the brain areas most affected in AD (Mishto et al., 2006). While it becomes clear that especially immunoproteasome activity is increased in glia during AD, it remains to be examined whether these changes are beneficial or not.

## **PARKINSON'S DISEASE**

Parkinson's disease is after AD the most common neurodegenerative disease with an incidence of 0.3% in the total population. The incidence increases with age leading to 4% of people above the age of 80 suffering from this mostly idiopathic movement disorder (Imai et al., 2000; de Lau and Breteler, 2006). The etiology of the disease is not known, although there are some familial cases caused by mutations in the genes of, e.g., alpha-synuclein (α-syn) and E3 ligase parkin. Also mutations in UCH-L1, a Ub c-terminal hydrolase, have been associated with PD but this finding is controversial (Hardy et al., 2009). In PD, the dopaminergic neurons in the substantia nigra pars compacta degenerate, which is accompanied with eosinophillic intracytoplasmatic inclusions known as Lewy bodies (LBs) and gliosis. While microgliosis is dominantly present in PD, reactive astrogliosis is virtually absent (Mirza et al., 1999; Van Den Berge et al., 2010; Halliday and Stevens, 2011). Although they do not change their phenotype, astrocytes are affected by α-syn accumulation. Astrocytes are able to take up α-syn released from neurons, and subsequently secrete proinflammatory cytokines (TNF-α and CXCL1), thereby activating microglia (Lee et al., 2010). These results were confirmed in a study using a mouse model expressing α-syn in astrocytes only, which caused similar disease symptoms, astrocyte functional loss and activated microglia. As the pro-inflammatory response of these cells caused neurodegeneration (Gu et al., 2010), this suggests that glial dysfunction plays a significant role in the etiology of PD.

A role for the UPS in PD was first described in familial PD when it was discovered that mutations in the protein parkin were associated with familial PD. Parkin functions as a Ub ligase in association with proteasomal degradation (Imai et al., 2000). A subset of the mutations in parkin, but also post-translational modifications of the protein, have been shown to cause a loss of function of the E3 ligase, and have been associated with UPS impairment and abrogation of the neuroprotective effects of parkin (Tsai et al., 2003; Yamamoto et al., 2005; Ali et al., 2011). In addition, Lewy bodies are mainly composed of α-syn and cytoskeletal proteins, but next to these proteins Lewy bodies contain the UPS related proteins parkin and Ub (Shults, 2006). Induced mutant α-syn expression was reported to decrease proteasome activity in cultured cells and a decreased proteasome activity was detected in the substantia nigra of PD patients (Tanaka et al., 2001; McNaught et al., 2003). These findings are in contrast with a study that showed that overall proteasome activity was not decreased in brain regions with Lewy body pathology (Tofaris et al., 2003). The discrepancy between these results can be explained by the technique that was used to measure proteasome activity in these studies. The AMC-peptides are rather non-specific since also other proteases can cleave these peptides, thereby influencing the results. Therefore, the exact role of UPS in PD is still under debate. Cultured parkin knock-out neurons appeared to be more resistant to mild proteasome inhibition than wild type neurons due to upregulation of anti-oxidant scavenger protein glutathione and autophagy-related proteins. In contrast, parkin knock-out glia (mixed population, mainly astrocytes) were more susceptible to

epoxomicin-induced cell death than wild type glia, suggesting that parkin dysfunction in PD mainly affects glia. In contrast to neurons, this effect on parkin knock-out glia was associated with lower levels of glutathione, a decreased HSP70 response, and an increase in poly-ubiquitinated proteins, all contributing to glial dysfunction. Besides, conditioned medium of parkin knock-out glia was less neuroprotective. So proteasome inhibition causes glial dysfunction and is therefore possibly the reason for the pathology in Parkin knock-out mice (Solano et al., 2008; Casarejos et al., 2009). Correspondingly, proteasome inhibitors had deleterious effects in mice expressing α-syn only in oligodendrocytes. The mice showed motor symptoms and severe degeneration in the nigrostriatal pathway caused by myelin disruption and demyelination in oligodendrocytes that had accumulated α-syn fibrils in their cytoplasm (Stefanova et al., 2012). Overall in PD, inhibition of the UPS system seems to contribute to glial dysfunction, thereby also affecting neuronal function. However, similar to AD the exact mechanism causing proteasome changes and neurodegeneration still needs to be unraveled.

### **HUNTINGTON'S DISEASE**

HD is a severe familial neurodegenerative disease typified by chorea (abnormal involuntary limb movements), incoordination, cognitive decline, and behavioral difficulties (Walker, 2007). The prevalence of this autosomal dominant heritable disease is five to seven affected individuals per 100,000 people in the Western world. HD is characterized by neuronal and glial aggregates, neuronal dysfunction and neurodegeneration, starting in the striatum and the cerebral cortex (corticostriatal pathway). An individual is affected when the polyglutamine (polyQ) repeat present in the disease-related huntingtin (HTT) protein exceed 36–41 glutamines (Walker, 2007). The encoding CAG repeats are not unique for HD as at least eight other neurodegenerative disorders are caused by a similar polyQ expansion in different proteins. Although the cause is already known for decades, the exact disease mechanism is still unclear. Interestingly, even though every cell in the body expresses the polyQ-expanded proteins, mainly neurons are affected and degenerate (Han et al., 2010). Fragments of the polyQ-expanded HTT are thought to initiate aggregation, and the resulting protein aggregates or inclusion bodies (IB) are an important hallmark of HD. More recently it became clear that these IBs are not necessarily causing neurodegeneration, as the formation of IBs did not correlate with cell death in cultured striatal neurons (Saudou et al., 1998). The formation of IBs may even be a protection mechanism of the cell to sequester toxic monomers and oligomers, thereby promoting cell survival (Arrasate et al., 2004; Truant et al., 2008).

Intriguingly, glial cells hardly show aggregates in HD patient material, while the expanded HTT is also expressed in these cells (Shin et al., 2005). Still, neuropathological studies show glial involvement in HD by phagocytosis of dead neurons and activated astrocytes (Reddy et al., 1998; Li et al., 2003), and the severity of both astrogliosis and microgliosis correlates with disease progression (Sapp et al., 2001; Faideau et al., 2010). Increased GFAP expression and activated microglia are already observed in presymptomatic stages in HD patients.With increased aggregation

in neurons also the number of activated microglia and astrocytes increases (Sapp et al., 2001; Faideau et al., 2010). In addition, both microglial cell lines expressing polyQ-expanded HTT and primary microglia from early postnatal HD mice were strongly impaired in their migration toward chemotactic stimuli, and showed a retarded response to laser-induced brain injury *in vivo* (Kwan et al., 2012). This suggests that polyQ-expanded HTT expression influences the appearance and function of reactive microglia. Similarly, astrocytes showed a loss of their regulatory functions: in the R6/2 HD mouse model decreased levels of glutamate transporters and glutamate uptake were observed, eventually leading to excitotoxicity (Shin et al., 2005). Intriguingly, excitotoxicity also occurred when polyQ-expanded HTT is only expressed in striatal astrocytes but not in neurons (Faideau et al., 2010;Wang et al., 2012). In addition, astrocytes isolated from R6/2 mice suppressed the secretion of the chemokine CCL5/RANTES and of BDNF, which is a growth factor important for neuronal survival, thereby inhibiting the trophic functions of astrocytes (Chou et al., 2008). So despite the low abundance of IBs, exclusive expression of polyQ-expanded HTT in astrocytes caused a decrease in glutamate uptake followed by excitotoxicity, neurodegeneration and an age-dependent HD-like phenotype in mice (Bradford et al., 2009), suggesting a crucial role for astrocytes in HD that is directly affected by polyQ-expanded HTT. As astrocytes and microglia hardly show IB formation in HD and when they appear, they are much smaller in size (Shin et al., 2005), differences in HTT ubiquitination or proteasome activity might explain these dissimilarities.

An important role for the UPS in HD was suggested in the late nineties, when Ub and proteasomes were found to be colocalized with HTT aggregates as shown by immunohistochemistry in HD mice and post-mortem patient material (Davies et al., 1997; DiFiglia et al., 1997). This led to the hypothesis that components of the UPS were irreversibly sequestered into HTT aggregates, as was also suggested by FRAP studies that showed no fluorescence recovery after photobleaching in fluorescently tagged proteasomes that were present in aggregates (Holmberg et al., 2004). However, we recently showed with fluorescent pulse-chase experiments that proteasomes were exchanged in HTT aggregates but with slower kinetics than it would be detectable by FRAP analysis (Schipper-Krom et al., 2014). This indicates that proteasomes are dynamically recruited into HTT aggregates. Since soluble HTT can be ubiquitinated, HTT can be targeted for proteasomal degradation (Thompson et al., 2009; Juenemann et al., 2013). However, various studies have reported an impairment of the proteasome system in both HD cell models and in the brain of HD patients (Seo et al., 2004; Hunter et al., 2007). Moreover, impairment of the proteasome causes accumulation of the potentially toxic aggregation-prone N-terminal HTT fragments (Li et al., 2010). Yet, proteasomes are able to degrade polyQ-expanded HTT fragments efficiently and entirely, but only when HTT is efficiently targeted for degradation by ubiquitination (Juenemann et al., 2013). Together these results suggest a crucial role for the UPS in HD pathogenesis.

Cellular stress due to polyQ-expanded HTT in neurons can lead to proteasome subunit changes, as increased levels of the IFNγ-inducible immunoproteasome subunits LMP2 and LMP7 were observed in cortex and striatum of both HD mice and post-mortem HD patient material. Interestingly, this increase was mainly attributed to degenerating neurons (Díaz-Hernández et al., 2003), and may be the result of glial cytokine secretion. It is unknown whether glial cells upregulate immunoproteasomes in HD, and whether the induction of the immunoproteasome is beneficial in HD. While these proteasomes are better capable to deal with protein aggregates under oxidative stress (Seifert et al., 2010), it is unknown whether immunoproteasomes are also better capable to degrade polyQ-expanded HTT fragments. Still, the apparent absence of HTT aggregates in glial cells remains intriguing, and may be explained by observed differences in proteasome activity in wild type mouse brains where glial cells showed increased proteasome activity when compared to neurons. It remains, however, to be examined whether this explains the differences in aggregate size and abundance. Remarkably, when polyQ-expanded HTT was present in R6/2 mice, proteasome activity levels were not altered (Tydlacka et al., 2008), although this does not reflect any possible changes in proteasome composition. The latter finding is underscored by a recent study, where it was shown by using proteasome activity-based probes that proteasomes recruited into aggregates were accessible for substrates and remained active. Besides, no differences were observed in overlay proteasome activity between HD and wild type mice (Schipper-Krom et al., 2014). Together, this indicates that the overall activity of the proteasome does not change in HD, although the composition of proteolytic subunits undergoes some alterations. Therefore, studying the role of the UPS in glial cells in HD is of high importance.

### **AMYOTROPHIC LATERAL SCLEROSIS**

Amyotrophic lateral sclerosis is a late onset, rapidly progressive and ultimately fatal neurological disorder, characterized by muscle weakness and atrophy leading to the inability to control voluntary movements. These symptoms are caused by the loss of motor neurons in the brain and spinal cord. Prevalence is increasing with age from overall 4–6 cases per 100,000 to up to 33 cases per 100,000 in 60–75-year-olds. About 10% of these patients have the familial form of ALS, often caused by mutations in the gene coding for copper-zinc superoxide dismutase (SOD1; Majoor-Krakauer et al., 2003). Motor neuron degeneration is often preceded by the formation of nuclear inclusions containing ALS associated proteins SOD1, TDP-43. or FUS and are usually also Ub positive (Leigh et al., 1991; Bendotti et al., 2012). Glial pathology is strongly present in ALS as observed in cell models, mice and post-mortem patient material (Alexianu et al., 2001; Petrik et al., 2007; Evans et al., 2013). Although SOD1 expression in microglia or astrocytes alone is not sufficient to cause ALS phenotype in mice, transplantation of healthy microglia in SOD1 mutated mice slows down disease progression (Gong et al., 2000; Clement et al., 2003; Beers et al., 2006). In addition, SOD1 expression alone in motor neurons does not cause ALS pathology, suggesting that glial function is important in ALS development (Pramatarova et al., 2001). Normally, astrocytes are secreting numerous trophic factors, like VEGF and BNDF to promote neuronal survival, and they take up glutamate from the synaptic cleft. However, activated astrocytes in ALS showed disturbances in these functions, eventually

leading to excitotoxicity (Trotti et al., 1999; Evans et al., 2013). Recently, it was shown that a mutation in C9orf72 could lead to aggregates in neurons, astrocytes, microglia and oligodendrocytes (Mizielinska et al., 2013). Besides, affected astrocytes caused neuronal cell death when co-cultured, suggesting an important role of astrocytes in this disease (Meyer et al., 2014). Microglia have been shown to secrete IL-1β as a response to purified mutant SOD1 stimulation, which is thus an important pro-inflammatory trigger (Meissner et al., 2010). Also secretion of several other pro-inflammatory cytokines such as TFN-α, IFNγ, and IL-6 was reported, resulting in (more) activation of astrocytes and microglia (Evans et al., 2013).

Similar to the neurodegenerative diseases described above also impairment of several components of the UPS is observed in ALS, which is likely a cause or consequence of the protein aggregates that are widely present (Kabashi et al., 2008; Bendotti et al., 2012). Impairment of the UPS but not of the autophagic pathway led to ALS-like symptoms in mice. Conditional knock-out of the 19S proteasome subunit RPT3 specifically in motor neurons caused accumulation of ALS related proteins (FUS and TDP-43) in these cells, and led to reactive gliosis and motor dysfunction. A knock-out of autophagy associated gene Atg5 only led to accumulations of Ub and p62 in the cytoplasm but not of ALS-related proteins (Tashiro et al., 2012). While impairment of the UPS in motor neurons induced an ALS-like phenotype, immunohistochemical studies showed significantly increased levels of Ub and proteasome subunits in both motor neurons and astrocytes in ALS (Aquilano et al., 2003; Mendonça et al., 2006). This contradictory result could be explained by findings in an ALS mouse model where a decrease in constitutive proteasome and an increase in immunoproteasome levels were observed, which correlated with the glia-mediated inflammatory response. Also the levels of the reactive astrocyte marker GFAP, the microglia marker CD68, and secretion of the cytokine TNF-α were upregulated in the ALS mice in a presymptomatic stage (Cheroni et al., 2009). Motor neurons in ALS also showed an upregulation of PA28αβ which could lead to a decrease in the turnover of ubiquitinated proteins as this requires the 19S activator (Bendotti et al., 2012). However, when mutant SOD1 mice were crossed with β1i/LMP2 knockout mice, the absence of immunoproteasomes *in vivo* could not significantly prevent mutant SOD1-induced disease; as no changes in disease symptoms were observed (Puttaparthi et al., 2007). So while activated astrocytes and microglia secrete cytokines that result in the induction of immunoproteasomes in both glia and neurons, these changes seem to be not beneficial in the clearance of accumulating SOD1 proteins.

# **CONCLUDING REMARKS AND FUTURE PERSPECTIVES**

Studies on the role of the UPS in neurodegenerative disease are mainly focusing on neurons, as these diseases are often considered to be cell autonomous due to the fact that neurons appear to be the most severe affected. Recently, however, the role of glial cells in neurodegenerative diseases is emerging. The fact that glial activation can already be observed in early presymptomatic stages advocates for an important if not crucial role for glial cells in the initiation of these diseases (Tai et al., 2007; Faideau et al., 2010). Glial cells express most of the disease-related proteins, and glial aggregates are observed in some of these neurodegenerative diseases albeit with lower frequencies (Bruijn et al., 1997; Shin et al., 2005; Tong et al., 2014). In addition, the disease-causing proteins can alter glial function by the induction of secretion of various cytokines. This makes glial cells important players in the various diseases, as the interplay between disease-causing proteins and the altered UPS not only affects glial cell homeostasis, but indirectly also neuronal function. A common hallmark of the here discussed various neurodegenerative diseases is the general downregulation of the constitutive proteasome accompanied by an upregulated immunoproteasome. This is in line with the observed phenotypical alterations that microglia and astrocytes undergo during disease progression, and the resulting pro-inflammatory environment correlates perfectly with the induction of the immunoproteasome. The induced immunoproteasomes may be better able to degrade intracellular protein aggregates, as suggested for protein aggregates that are induced following IFNγ-induced stress (Seifert et al., 2010). Since IFNγ is secreted by microglia (Kawanokuchi et al., 2006) and is able to activate both microglia and astrocytes in several neurodegenerative diseases (Choi et al., 2013; Evans et al., 2013), most likely the observed immunoproteasome induction is mediated by microglia. Nevertheless, while immunoproteasomes are directly induced by IFNγ, this cytokine also induces the accumulation of oxidized proteins that may accelerate aggregation and affect protein homeostasis in time. Still, the induced immunoproteasomes appear better capable to degrade polyubiquitinated proteins under these circumstances, and proteolytic activity of immunoproteasomes is higher compared to standard proteasomes (Seifert et al., 2010). These data suggest that activation of glia and induction of the immunoproteasome may be beneficial for clearing disease-causing proteins.

Yet, similar to chronic activation of glia, it is unknown whether induction of the immunoproteasome is on the long term beneficial or detrimental in neurodegenerative diseases. After all, the increased immunoproteasome induction in these diseases and in aging could also indicate undesirable protein accumulation and chronic inflammation (Díaz-Hernández et al., 2003; Mishto et al., 2006; Cheroni et al., 2009; Nijholt et al., 2011; Orre et al., 2013; Bellavista et al., 2014). When the induction of the immunoproteasome is not beneficial in targeting the accumulating proteins, specific inhibitors targeting immunoproteasome subunits could be an efficient therapeutic approach. Although several of these inhibitors are already developed (Miller et al., 2013), more insight is required in the exact function of the immunoproteasome in these disorders.

Next to focusing on proteasome activity, for future approaches it is also important to examine which other alterations in the UPS observed in glia can be used as possible therapeutic targets. Although proteasome compositional changes appear in glial cells during disease progression, in general proteasomes stay active and accessible even in cells showing large protein aggregates (Tydlacka et al., 2008; Basler et al., 2013; Orre et al., 2013; Schipper-Krom et al., 2014). Yet, it is unclear whether the diseased proteins are correctly targeted for degradation, as inefficient ubiquitination of mutant and misfolded proteins could explain the observed protein accumulations and neuronal vulnerability. Improved targeting of the disease-causing proteins toward the proteasome by modifying

ubiquitination (or with other Ub-like proteins) would increase their degradation and delay onset of disease. It is unknown whether the ubiquitination patterns of the various disease-related proteins in glia and neurons is dissimilar and whether they change during disease progression. Therefore, determining the ubiquitination patterns of these proteins in different stages of disease in both neurons and glia, along with determining the involved specific ubiquitin ligases and DUBs in these cells should lead to the identification of new and more specific therapeutic targets.

# **ACKNOWLEDGMENTS**

This work was supported by an AMC PhD scholarship (Anne H. P. Jansen), the Internationale Stichting Alzheimer Onderzoek [ISAO 08504 and 12509 to Elly M. Hol], and the Netherlands Organization for Scientific Research [NWO; VICI grant 865.09.003 to Elly M. Hol].

# **REFERENCES**


immune response. *Prog. Neurobiol.* 97, 101–126. doi: 10.1016/j.pneurobio.2011. 10.001


neuronal sense and antisense RNA foci. *Acta Neuropathol.* 126, 845–857. doi: 10.1007/s00401-013-1200-z


**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.

*Received: 28 May 2014; accepted: 10 July 2014; published online: 08 August 2014.*

*Citation: Jansen AHP, Reits EAJ and Hol EM (2014) The ubiquitin proteasome system in glia and its role in neurodegenerative diseases. Front. Mol. Neurosci. 7:73. doi: 10.3389/fnmol.2014.00073*

*This article was submitted to the journal Frontiers in Molecular Neuroscience.*

*Copyright © 2014 Jansen, Reits and Hol. 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.*

# Subcellular Clearance and Accumulation of Huntington Disease Protein: A Mini-Review

Ting Zhao<sup>1</sup> , Yan Hong<sup>1</sup> , Xiao-Jiang Li 1,2\* and Shi-Hua Li <sup>1</sup> \*

<sup>1</sup> Department of Human Genetics, Emory University School of Medicine, Atlanta, GA, USA, <sup>2</sup> State Key Laboratory of Molecular Developmental Biology, Institute of Genetics and Developmental Biology, Chinese Academy of Sciences, Beijing, China

Huntington's disease (HD) is an autosomal dominant, progressive neurodegenerative disease caused by an expanded polyglutamine (polyQ) tract in the N-terminal region of mutant huntingtin (mHtt). As a result, mHtt forms aggregates that are abundant in the nuclei and processes of neuronal cells. Although the roles of mHtt aggregates are still debated, the formation of aggregates points to deficient clearance of mHtt in brain cells. Since the accumulation of mHtt is a prerequisite for its neurotoxicity, exploring the mechanisms for mHtt accumulation and clearance would advance our understanding of HD pathogenesis and help us develop treatments for HD. We know that the ubiquitinproteasome system (UPS) and autophagy play important roles in clearing mHtt; however, how mHtt preferentially accumulates in neuronal nuclei and processes remains unclear. Studying the clearance of mHtt in neuronal cells is a challenge because neurons are morphologically and functionally polarized, which means the turnover of mHtt may be distinct in different cellular compartments. In this review, we discuss our current knowledge about the clearance and accumulation of mHtt and strategies examining mHtt clearance and accumulation in different subcellular regions.

### Edited by:

Ashok Hegde, Georgia College and State University, USA

### Reviewed by:

Jose J. Lucas, Centro de Biología Molecular Severo Ochoa (CBMSO)-Universidad Autónoma de Madrid (UAM), Spain Nihar Ranjan Jana, National Brain Research Centre, India

### \*Correspondence:

Xiao-Jiang Li xli2@emory.edu; Shi-Hua Li sli@emory.edu

Received: 24 February 2016 Accepted: 01 April 2016 Published: 21 April 2016

### Citation:

Zhao T, Hong Y, Li X-J and Li S-H (2016) Subcellular Clearance and Accumulation of Huntington Disease Protein: A Mini-Review. Front. Mol. Neurosci. 9:27. doi: 10.3389/fnmol.2016.00027 Keywords: huntingtin, neurodegeneration, proteasome, autophagy, aggregation

# HUNTINGTON'S DISEASE AND SELECTIVE NEUROPATHOLOGY

Huntington's disease (HD) is an autosomal dominant neurodegenerative disease that is characterized by motor abnormalities, cognitive decline and psychiatric problems (Munoz-Sanjuan and Bates, 2011). The disease is caused by the expansion of a trinucleotide CAG repeat in exon 1 of the HD gene, which encodes an expanded polyglutamine (polyQ) tract in the N-terminal region of mutant huntingtin (mHtt). While most HD patients carry CAG repeats in the range of 38–55 and develop neurological symptoms in mid-life, larger repeats (>60Q) can cause juvenile onset HD (Ross et al., 2014).

Despite the ubiquitous expression of mHtt in the brain and peripheral tissues, the major pathological feature of HD is selective neurodegeneration (Vonsattel and DiFiglia, 1998; Munoz-Sanjuan and Bates, 2011). Similarly, selective neurodegeneration is also seen in many other neurodegenerative diseases, such as Alzheimer's and Parkinson's diseases, suggesting that multiple factors may contribute to the selective neurodegeneration in these diseases.

**Abbreviations:** AAV, Adenoviral-associated vectors; HD, Huntington's disease; HDAC6, Histone deacetylase 6; LGP, Lateral globus pallidus; mHtt, Mutant huntingtin; polyQ, Polyglutamine; SN, Substantia nigra; UPS, Ubiquitin-proteasome system.

Given the known genetic mutation in HD and its wellcharacterized neuropathology, HD makes an ideal model for investigating how selective neuropathology occurs with aging. In HD, neuronal degeneration is characterized by the preferential loss of neuronal cells in the striatum in the early disease stage and extensive neurodegeneration in a variety of brain regions in later disease stages (Ross et al., 2014). This progressive neurodegeneration is consistent with the late-onset neurological symptoms of HD.

# PREFERENTIAL ACCUMULATION OF mHtt IN NEURONAL NUCLEI AND PROCESSES

The age-dependent neurodegeneration and neurological symptoms in HD correlate with the accumulation of misfolded forms of mHtt in neuronal cells. The expansion of the polyQ repeat in mHtt causes the misfolding of mHtt and formation of mHtt aggregates in neuronal nuclei and neuropils in HD patient brains (DiFiglia et al., 1997; Gutekunst et al., 1999). These aggregates are recognized by antibodies to the Nterminal region of mHtt, suggesting that the aggregates are formed by mHtt N-terminal fragments that are produced by the proteolysis of full-length mHtt. Indeed, transgenic mice expressing N-terminal mHtt show abundant mHtt aggregates in their neuronal nuclei and processes (Davies et al., 1997; Schilling et al., 1999). Studies of these transgenic HD mice also demonstrate that N-terminal mHtt fragments preferentially accumulate in neuronal cells to form nuclear aggregates and neuropil aggregates. HD knock-in mice express full-length mHtt at the endogenous level under the control of the mouse Htt gene, and therefore mimic HD patients genetically. In HD knock-in mice, mHtt also forms aggregates first in the neuronal nuclei of the striatum (Wheeler et al., 2000; Li et al., 2001; Lin et al., 2001). Although this preferential formation of neuronal nuclei aggregates mirrors the vulnerability of striatal neurons in HD patients, mHtt appears to form more neuropil aggregates than nuclear aggregates in HD patients in early disease stages (DiFiglia et al., 1997; Gutekunst et al., 1999). Despite this species-dependent difference, in mice expressing full-length mHtt and modeling early stages of HD, neuropil aggregates form preferentially in the lateral globus pallidus (LGP) and substantia nigra (SN; Li et al., 2001). The majority of striatal neurons extend their axons to the LGP and SN, two brain regions degenerated more significantly during the early stage of HD (Reiner et al., 1988; Richfield et al., 1995). Thus, the preferential accumulation of mHtt in both the neuropils and nuclei of striatal neurons may account for the selective striatal neurodegeneration in HD.

Many studies report that mHtt aggregates can be either toxic or protective. It is possible that mHtt aggregates are both harmful and beneficial depending on the disease stage, subcellular localization and their association with other partners or organelles. Nevertheless, because mHtt aggregates are formed by N-terminal mHtt fragments that are misfolded, the mHtt aggregates reflect the accumulation of misfolded mHtt. Growing evidence indicates that the misfolded mHtt exerts its neurotoxicity by disturbing a wide range of cellular functions (Ross et al., 2014). The wide range of cellular toxicity from mHtt is due perhaps to its ability to interact with a variety of proteins and to interrupt the function of these interactors (Li and Li, 2004; Shirasaki et al., 2012) via a gain-of-function mechanism. Thus, the accumulation of misfolded mHtt is a prerequisite to its neuronal toxicity, and the clearance of mHtt is key to the treatment of HD.

# INTRACELLULAR CLEARANCE OF mHtt

Two proteolytic machineries are critical for clearing misfolded proteins. One is the ubiquitin-proteasome system (UPS), which mainly clears soluble and short-lived proteins in eukaryotic cells. The other is autophagy, which removes long-lived proteins, aggregated proteins and damaged organelles.

Because the formation of mHtt aggregates is age-dependent, the initial notion was that mHtt impairs proteasomal function (Bence et al., 2001; Venkatraman et al., 2004). However, targeting expanded polyQ for proteasomal degradation did not compromise proteasome activity (Michalik and Van Broeckhoven, 2004). Also, mHtt-exon1, the shortest N-terminal fragment of mHtt, was completely digested by the proteasome (Juenemann et al., 2013). In addition, in vivo studies show that proteasome activity in HD mouse brains is not perturbed by mHtt expression (Wang et al., 2008a; Bett et al., 2009). Although no global impairment of proteasomal activity is seen, proteasomal dysfunction probably occurs in certain subcellular regions, such as axonal terminals, which may lead to the accumulation of mHtt in these places (Wang et al., 2008a). It is established that mHtt compromises the axonal transport of mitochondria (Orr et al., 2008; Reddy and Shirendeb, 2012). Since the UPS is a highly ATP-dependent system (Schrader et al., 2009), defective mitochondria transport may lead to ATP deficiency in neurites and nerve terminals, which can impair the local degradation of mHtt by the proteasome and causes the aggregation of mHtt in these subcellular regions.

Using various HD cell and animal models, previous studies have shown that upregulation of autophagy leads to a reduction of mHtt aggregates, indicating that autophagy plays a role in the clearance of mHtt aggregates (Qin et al., 2003; Sasazawa et al., 2015). K63-linked polyubiquitination on mHtt is proposed to confer the selectivity on the degradation of mHtt aggregates by autophagy since K63-ubiquitinated substrates are recognized by autophagy receptors, such as p62, which has been shown to bind Htt (Tan et al., 2008). However, Bhat et al. (2014) found that ubiquitination of K48 linkage on mHtt switches to K63 linkage with aging, which promotes aggregation. On the other hand, mHtt compromises autophagy by perturbing cargo recognition and autophagosome motility (Martinez-Vicente et al., 2010; Wong and Holzbaur, 2014). Since the ubiquitination of both K48 and K63 is involved in the clearance of misfolded proteins by the UPS and autophagy, how the UPS and autophagy work together to remove mHtt remains to be clarified. Histone deacetylase 6 (HDAC6) probably links autophagy to UPS for clearance of mHtt as autophagy is activated in an HDAC6-dependent manner when proteasomal function is impaired (Pandey et al., 2007).

# STRATEGIES TO EXAMINE SUBCELLULAR mHtt CLEARANCE AND ACCUMULATION

Given the fact that mHtt preferentially accumulates in neuronal nuclei and processes, it is important to determine mHtt clearance at the subcellular level. Nuclei and synaptosomes can be isolated via biochemical fractionation, and the fractions can be analyzed by Western blotting. The advantage of this assay is that aggregated and monomer proteins can be separated in SDS gel and assessed for their relative levels on the same blot. Quantifying the levels of soluble and aggregated mHtt can yield valuable information about the levels of mHtt in these subcellular regions; however, the combination of subcellular fractionation and western blotting cannot monitor the degradation of mHtt in real time. Moreover, it is difficult to separate the fraction that is enriched in neuronal processes or neuropil. In addition, fractionation requires the use of brain homogenates that cannot distinguish cell types in which mHtt may differentially accumulate or be cleared.

It is not feasible to study the degradation of mHtt in living neurons until phototransformable fluorescent proteins are invented. Phototransformable fluorescent proteins can be classified into three types-photoactivating, photoconverting and photoswitching-based on their responses to light. Currently, these proteins are used widely to study the dynamics of molecules and cells in a spatiotemporal manner (Adam et al., 2014). Tsvetkov et al. (2013) recently used Dendra2, one of the photoconvertible fluorescent proteins, to study the degradation of mHtt-exon1 in cultured striatal neurons. Dendra2 is a green-to-red photoconvertible fluorescent protein featuring fast maturation and bright fluorescence (Chudakov et al., 2007). Light irradiation at a 405-nm wavelength can efficiently activate Dendra2 and consequently switches its fluorescent color irreversibly from green to red. The linkage of Dendra2 to mHtt does not perturb its neurotoxicity or the property of aggregation in neurons. After transient transfection of Htt-exon1-Dendra2 into striatal neurons, Tsvetkov et al. (2013) converted green Htt-exon1-Dendra2 to red fluorescence by activating Dendra2 with 405-nm laser light. Therefore, after irradiation, the reduction of red Httexon1 over time reflects its degradation in live cells. Using this strategy, called the ''optical pulse chase'' assay, Tsvetkov et al. (2013) found that mHtt-exon1 was cleared faster than its wild-type counterpart in the body of the cultured striatal neurons.

Although optical pulse chase has been used mainly with in vitro cultured cells, it should also be useful for determining the turnover of mHtt in live cells at subcellular levels, such as neuronal processes and terminals. Examining the change of red fluorescence avoids the potential influence from newly

# REFERENCES

Adam, V., Berardozzi, R., Byrdin, M., and Bourgeois, D. (2014). Phototransformable fluorescent proteins: future challenges. Curr. Opin. Chem. Biol. 20, 92–102. doi: 10.1016/j.cbpa.2014.05.016

synthesized proteins as they are labeled by inactive green Dendra2, so that the degradation of mHtt can be quantified. The Dentra-2 fusion proteins can be expressed via adenoviralassociated vectors (AAV), and stereotaxic injection would allow the delivery of the AAV vectors to specific brain regions, such as the striatum. The injected brain region can then be isolated and sectioned into brain slices, which can be observed in medium under a fluorescent microscope. Photoconversion of Dendra-2 will then be achieved to measure the clearance of mHtt in neuronal processes or nerve terminals in brain slices. Moreover, using different promoters would confer the expression of Dendra-2-mHtt in specific types of cells, making it possible to study the cell type-dependent degradation of mHtt in the brain. Further, tagging the Dendra-2 fusion protein with organelle targeting sequences would allow us to study mHtt's degradation in specific organelles. Given that suppression of neuropil aggregates ameliorated the neurological symptoms of HD mice (Wang et al., 2008b), promoting clearance of mHtt by the UPS in neuropils should be therapeutically beneficial for HD. In support of this idea, overexpressing ube3a, an ubiquitin E3 ligase, can activate the UPS and decrease mHtt aggregates in the brains of HD knock-in mice (Bhat et al., 2014). In addition, up-regulating autophagy has also been found to eliminate mHtt aggregates. (Ravikumar et al., 2004; Sasazawa et al., 2015).

# CONCLUSION

Although mHtt is known to be cleared by the UPS and autophagy, how mHtt is cleared at the subcellular level remains unknown. Understanding this interesting issue will shed light on the pathogenesis of HD and also help us find ways of accelerating the clearance of this toxic protein. Biochemical fractionation would allow one to simultaneously examine the relative levels of aggregated and monomer mHtt on the same Western blot. The optical pulse chase assay would enable the study of the degradation of mHtt in different subcellular regions in living neurons. Such studies will answer the question of whether mHtt degradation is subcellular region dependent and could also be extended to studies of other types of misfolded proteins in different neurodegenerative diseases.

# AUTHOR CONTRIBUTIONS

TZ, YH, X-JL and S-HL wrote this review article.

# ACKNOWLEDGMENTS

This work was supported by NIH Grants (AG19206 and NS041449 to X-JL, NS095279 and NS095181 to S-HL).

Bence, N. F., Sampat, R. M., and Kopito, R. R. (2001). Impairment of the ubiquitinproteasome system by protein aggregation. Science 292, 1552–1555. doi: 10. 1126/science.292.5521.1552

Bett, J. S., Cook, C., Petrucelli, L., and Bates, G. P. (2009). The ubiquitinproteasome reporter GFPu does not accumulate in neurons of the R6/2 transgenic mouse model of Huntington's disease. PLoS One 4:e5128. doi: 10. 1371/journal.pone.0005128


disease. Proc. Natl. Acad. Sci. U S A 85, 5733–5737. doi: 10.1073/pnas.85.15. 5733


**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 Zhao, Hong, Li and Li. 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.

# Ubiquitin–proteasome system involvement in Huntington's disease

# *Zaira Ortega\* and Jose J. Lucas*

Department of Molecular Biology, Centro de Biología Molecular "Severo Ochoa," Consejo Superior de Investigaciones Científicas (CSIC), Universidad Autónoma de Madrid (UAM), Centro Investigación Biomédica en Red Enfermedades Neurodegenerativa (CIBERNED), Madrid, Spain

### *Edited by:*

Fred Van Leeuwen, Maastricht University, Netherlands

### *Reviewed by:*

Stefano Sensi, University of California, Irvine, USA Eric Reits, University of Amsterdam, Netherlands

### *\*Correspondence:*

Zaira Ortega, Laboratory 209, Department of Molecular Biology, Centro de Biología Molecular "Severo Ochoa," Consejo Superior de Investigaciones Científicas (CSIC), Universidad Autónoma de Madrid (UAM), Centro Investigación Biomédica en Red Enfermedades Neurodegenerativa (CIBERNED), 28049 Madrid, Spain e-mail: Zairaortegallorente@ gmail.com

Huntington's disease (HD) is a genetic autosomal dominant neurodegenerative disease caused by the expansion of a CAG repeat in the huntingtin (htt) gene. This triplet expansion encodes a polyglutamine stretch (polyQ) in the N-terminus of the high molecular weight (348-kDa) and ubiquitously expressed protein htt. Normal individuals have between 6 and 35 CAG triplets, while expansions longer than 40 repeats lead to HD. The onset and severity of the disease depend on the length of the polyQ tract: the longer the polyglutamine stretch (polyQ) is, the earlier the disease begins and the more severe the symptoms are. One of the main histopathological hallmarks of HD is the presence of intraneuronal proteinaceous inclusion bodies, whose prominent and invariant feature is the presence of ubiquitin (Ub); therefore, they can be detected with anti-ubiquitin and antiproteasome antibodies. This, together with the observation that mutations in components of the ubiquitin–proteasome system (UPS) give rise to some neurodegenerative diseases, suggests that UPS impairment may be causative of HD. Even though the link between disrupted Ub homeostasis and protein aggregation to HD is undisputed, the functional significance of these correlations and their mechanistic implications remains unresolved. Moreover, there is no consistent evidence documenting an accompanying decrease in levels of free Ub or disruption of Ub pool dynamics in neurodegenerative disease or models thus suggesting that the Ub-conjugate accumulation may be benign and just underlie lesion in 26S function. In this chapter we will elaborate on the different studies that have been performed using different experimental approaches, in order to shed light to this matter.

**Keywords: ubiquitin–proteasome system, Huntington's disease, inclusion body, degron-fluorescent proteins, animal models**

# **INTRODUCTION**

Huntington's disease (HD) is a genetic autosomal dominant neurodegenerative disease (Wexler et al., 1987) that affects approximately 1 out of 10.000 individuals in most of the populations with European background (Harper, 1992). It shows symptoms in midlife and patients often die 15–20 years after the onset of the symptoms (Ambrose et al., 1994). Currently, there is no effective treatment to prevent or delay disease progression (Vonsattel and DiFiglia, 1998). HD patients suffer from motor dysfunction (chorea, rigidity, dystonia, and oculomotor dysfunction among others), cognitive decline also known as dementia (subcortical dementia, including affective and personality changes, and problems acquiring new knowledge), and psychopathological dysfunction (depression, suicide, and mania are the most frequent ones). Emotional and cognitive changes often precede motor dysfunction by several years (about 3 years). These symptoms are the result of the selective neurodegeneration that occurs preferentially in the striatum of the patients (Graveland et al., 1985; Gusella et al., 1993; Vonsattel and DiFiglia, 1998).

Huntington's disease is included in a group of neurodegenerative diseases called proteinopathies [which include pathologies such as Alzheimer's disease (AD) or Parkinson's disease (PD)] due to the fact that aggregate-prone proteins cause all of them. The main histopathological hallmark of these diseases is the presence

of aggregates constituted by the mutant or modified proteins: these inclusion bodies (IBs) can be predominantly cytosolic (such as in PD, and HD), intranuclear [for example, spinocerebellar ataxia type 1 (SCA1)], aggregated in the endoplasmic reticulum (as seen with neuroserpin mutations that cause familial encephalopathy with neuroserpin IBs) or extracellularly secreted (for example amyloid- β in AD). In HD, the mutated protein is the ubiquitously expressed protein huntingtin (htt) and the mutation consists of an expansion of a CAG repeat located in the 5 terminus of the htt gene (HDCRG, 1993) which translates into a polyQ in the N' terminus of the protein (Gusella et al., 1993; Locke et al., 1993). Normal individuals have between 6 and 35 CAG triplets, while expansions longer than 40 repeats lead to HD (Andrew et al., 1993; HDCRG, 1993). The onset and severity of the disease depend on the length of the polyQ tract, the longer the polyQ is, the earlier the disease begins and the more severe the symptoms are (Andrew et al., 1993; Snell et al., 1993). Apart from HD, there are eight additional hereditary diseases caused by CAG/polyQ expansion (Zoghbi and Orr, 2000; Ross, 2002), all of them are neurological diseases, despite the different nature of the proteins involved and their ubiquitous expression, suggesting a selective vulnerability of the neurons for polyQ (**Figure 1**). In all these diseases the triplet expansions are within the coding sequence of the gene, and they are always translated in the reading frame that produces a polyQ

sequence. Moreover, the threshold for the expansion to become pathogenic is around 40 repeats in most of these diseases (Zoghbi and Orr, 2000) both *in culture* and *in vivo*. Interestingly, the threshold length for *in vitro* aggregation correlates with the pathogenic repeat length threshold (Scherzinger et al., 1997), thus suggesting that PolyQ aggregation is a key element in the pathogenesis.

Although the hemizygous loss of function of normal proteins in polyQ diseases, and particularly in HD, may contribute to some aspects of the pathologies, a toxic gain of function of the expanded polyQ is the most likely determinant of the disease. It can cause disease by conferring additional properties on the mutant gene product that may include hyperactivity of normal function and/or new toxic properties unrelated to normal function. Mice lacking one htt allele are essentially normal, although complete loss of htt causes embryonic lethality (Duyao et al., 1995; Nasir et al., 1995; Zeitlin et al., 1995). Humans with Wolf–Hirschhorn syndrome have hemizygous loss of the tip of chromosome 4p, which includes the HD gene, interestingly, these individuals do not show features of HD (Harper, 1996). Following these findings, a number of transgenic animal models have been made that express a mutant htt∗ transgene, comprising either the whole coding region, an amino-terminal fragment or simply isolated expanded polyQ repeats. Despite also having two normal htt orthologs, these animals recapitulate many of the features of the human disease (for a review, see Rubinsztein, 2002). Furthermore, expression of hypoxanthine phosphoribosyl transferase, a non-disease related protein that doesnot express polyQ, with expanded polyQ caused the mice to develop a progressive neurological disorder with clinical and pathological features reminiscent of HD, which implies that transferring the polyQ tract itself is sufficient to induce aggregation and disease (Ordway et al., 1997).

### **PROTEIN DEGRADATION PATHWAYS**

In cells, the efficient folding of new polypeptides and the efficient elimination of misfolded or damaged proteins is critical to the maintenance of protein homeostasis and cellular health. The presence of IBs in neurons in proteinopathies suggests a failure in the degradation pathways. Eukaryotic cells have two main routes for clearing misfolded or toxic proteins, the ubiquitin–proteasome system (UPS) and the autophagy-lysosome pathways. The UPS works both in the nucleus and in the cytoplasm and is responsible for the recycling and degradation of most of the short-lived and misfolded soluble proteins (Hershko and Ciechanover, 1998). On the contrary, the autophagy-lysosome pathway mainly degrades long-lived proteins and degenerated organelles, and requires the formation of double-membrane-bounded autophagosomes (Klionsky et al., 2003, 2008; Suzuki and Ohsumi, 2007) and it is thus restricted to the cytoplasm. Both pathways have been suggested to play a role in HD (Rubinsztein, 2006; Thompson et al., 2009), although recent studies suggest that the UPS is more important than autophagy for removing toxic- N-terminal htt∗ fragments (Li et al., 2010). If an impairment of the degradation pathway is the triggering step or a secondary effect in HD is still unclear. There are previous reviews on this matter (Valera et al., 2005; Ortega et al., 2007; Rubinsztein, 2007) and in this review we also discuss more recent reports on the status of the UPS in HD.

### **THE UBIQUITIN–PROTEASOME SYSTEM**

As one of the main routes of protein degradation, the UPS is involved in many cellular mechanisms in the nervous system such as neuronal plasticity, memory, and regulation of neurotransmission at pre- and post-synaptic sites, thus it plays a critical role in neuronal signaling (Krug et al., 1984; Fonseca et al., 2006; Karpova et al., 2006). It represents a major defense against misfolded proteins, particularly in post-mitotic neurons that are unable to divide to reduce their burden of damaged proteins. Despite being highly conserved across species, structurally and functionally distinct subunit compositions of the proteasome have been identified in different tissues (Glickman and Raveh, 2005; Drews et al., 2007; Tai et al., 2010). These variants have

been attributed to alterations in ubiquitin (Ub) ligase activity, proteasome subunit composition, and tissue-specific proteasomeinteracting proteins (Glickman and Raveh, 2005; Drews et al., 2007; Tai et al., 2010). In UPS degradation pathway there are two differentiated steps: (1) targeting of the protein for degradation and (2) substrate proteolysis in the proteasome. There are many molecules involved in these steps. In protein targeting for proteasomal degradation substrates must be covalently modified with Ub, which is conjugated through its carboxy terminus to form chains of four or more Ub molecules linked by lysines at residue 48 (Hershko and Ciechanover, 1998; Thrower et al., 2000; Pickart, 2001; Kuhlbrodt et al., 2005). This conjugation typically involves three types of enzyme: E1 (ubiquitin-activating enzyme) hydrolyses ATP and forms a thioester-linked conjugate between itself and Ub; E2 (ubiquitin-conjugating enzyme) receives Ub from E1 and forms a similar thioester intermediate with Ub; and E3 (ubiquitin-ligase) binds both E2 and the substrate, and transfers the Ub to the substrate (**Figure 2**; Hershko and Ciechanover, 1998; Pickart, 2001). Polyubiquitinated proteins are recognized and subsequently degraded by the 26S proteasome. This ATP-dependent proteolytic complex consists of a 20S core particle and one or two 19S regulatory particle(s). The barrel-shaped 20S complex is

composed of four heptagonal rings where the proteolytic activities reside (Groll et al., 1997; DeMartino and Slaughter, 1999). The 19S regulatory particles are important for the recognition, unfolding, and translocation of ubiquitinated substrates into the 20S core subunit for degradation (Voges et al., 1999; Hartmann-Petersen et al., 2003). The polyUb chains are not degraded by the proteasome, deubiquitilating enzymes (DUBs) remove the chain from the substrates once they have been recognized by the 19S subunit of the proteasome and separate it into monomers ready to be reused (Kawakami et al., 1999).

# **THE UPS IN HD**

There is cellular and genetic evidence supporting the hypothesis of UPS impairment in neurodegenerative diseases. Regarding genetic evidence several neurodegenerative diseases have been described to be caused by mutations in different components at different levels of the labeling and degradation (ubiquitilation, deubiquitilation, and substrate delivery) of substrates by the UPS (**Table 1**). These pathologies suggest that primary genetic deficiencies of components of the UPS are sufficient to cause neurodegeneration. Regarding cellular evidence, the presence of IBs constituted by the mutant proteins has already been described



**Table 1 |**

**Continued** both in human brain (DiFiglia et al., 1997) and in animal models of HD (Davies et al., 1997). In addition, complementary pharmacological data show that both in cellular (Waelter et al., 2001) and animal models (McNaught et al., 2004), the inhibition of the proteasome, by using specific inhibitors, produces an increase in aggregation of htt∗ that is critically dependent on the proteasomal activity and can cause parkinsonian features, including Lewy body-like aggregates (McNaught et al., 2004). Similarly, the reversal of aggregates that takes place in primary neurons from the HD inducible mouse model upon shutdown of htt∗ expression no longer takes place in the presence of proteasome inhibitors (Martin-Aparicio et al., 2001). These IBs are labeled with antibodies that recognize Ub and different proteasome subunits (in the 20S core and 19S caps; DiFiglia et al., 1997; Cummings et al., 1998; Goedert et al., 1998; Sherman and Goldberg, 2001; Díaz-Hernández et al., 2003; Schmitt, 2006), suggesting a direct (htt∗ protein) or indirect (proteins associated to htt∗) sequestration of the proteasome into the IBs. Besides, it was reported the spatial restriction of proteasomes within aggregates in fluorescence recovery after photobleaching (FRAP) experiments (Holmberg et al., 2004). However, recent results support that proteasomes are dynamically and reversibly recruited into IBs (Schipper-Krom et al., 2014) and that they remain catalytically active and accessible to substrates. This challenges the concept of proteasome sequestration and impairment in HD, and supports the absence of proteasome impairment in mouse models of HD. In line with this, an increase in proteasome activity in the insoluble cellularfractions of mtHtt-Q150 expressing neuronal cells has also been described (Jana et al., 2001).Taking all these data as a starting point, many experiments testing the capability of the IBs of directly inhibiting the proteasome have been conducted. It has been tested whether 26S activity is inhibited by IBs in HD by incubating 26S purified proteasomes with *in vitro* generated polyQ aggregates (Bennett et al., 2005). No impairment in any of the proteasome catalytic activities were detected what argues against a decreased proteasome activity. However, if ubiquitilation of the aggregates were important for its potential inhibitory interaction with 26S proteasomes, the previous experiments would not detect it. To overcome this limitation, similar experiments were performed with aggregates purifiedfrom mouse models and post-mortem human brains (Díaz-Hernández et al., 2006) instead of *in vitro* generated polyQ aggregates this approach in fact detected 26S activity impairment upon incubation with isolated microaggregates such as htt filaments although not when incubated with isolated IBs. To test the hypothesis in a more physiological environment, similar experiments were performed with HD mouse model brain extracts (Díaz-Hernández et al., 2003; Bowman et al., 2005; Bett et al., 2006). By this approach, not only there was no decrease in the catalytic activities, but also there was a selective increase in the chymotrypsin- and trypsin-like activities that were believed to be a result of a qualitative change in the subunit composition of the proteasomes (Díaz-Hernández et al., 2003, 2004a). These sets of experiment report that the UPS remained active in HD and argue against the postulated inhibition of proteasomes. Thus, opening the possibility of reinterpreting the meaning of the marked accumulation of PolyUb chains observed in the brains of these mouse models and in human HD patients (Bennett et al., 2007). On the other hand, when these experiments were performed on human post-mortem HD brains tissue, a decreased activity of the proteasome was reported (Seo et al., 2004) suggesting differences in the UPS but, due to inherent limitations associated to analysis of enzymatic activities in post-mortem tissue, it is difficult to conclude whether proteasome activity was really altered or not. In summary, it is not possible to draw a definite conclusion from all these studies due to the limitations associated to each of the employed techniques to monitor status or function of UPS components (degradation of small fluorogenic peptides, degradation of ubiquitylated proteins, detection of Ubconjugates, etc.) and the differences in the analyzed systems or samples (*in vitro* incubation of proteasomes with PolyQ species at different degrees of aggregation and ubiquitylation vs. tissue homogenates with the latter being eitherfreshfrozen animal model tissue or human frozen tissue with varying extents of post-mortem intervals).

In Konstantinova et al. (2008) the proteasome was described as dynamic structure in terms of its composition as 26S proteasome is constantly assembling and disassembling, and its 19S and 20S subunits are targets for a great number of post-translational modifications including phosphorylation and acetylation. 26S proteasomes would get assembled just to degrade the substrates and immediately disassembled again. These data led to hypothesize that IBs could interact with the disassembled 19S and 20S subunits preventing them from assembling again to degrade the substrates and retaining them in the IBs. Experiments testing the catalytic activities of the 26S proteasome and the 20S subunit alone in the presence of IBs purified from mouse models and post-mortem human brains (Díaz-Hernández et al., 2006) were performed. Experiments involving the 20S subunit were performed in the presence of sodium dodecyl sulfate (SDS) to facilitate the entrance of the substrate into the catalytic chamber however, as in the case of the 26S proteasome; they failed to detect enzymatic activity inhibition. These results confirm the previously obtained results supporting the absence of influence of IBs upon proteasome activity.

The above mentioned results were obtained from experimental approaches that assume that the proteasome degrades htt, however, the three endoproteolytic activities of the proteasome (trypsin-like, chymotrypsin-like and PGPH) cut the peptide bonds after basic, hydrophobic, or acid residues respectively (DeMartino and Slaughter, 1999) and glutamine does not really fit in any of these categories. This fact brought new hypothesis for a possible UPS impairment such as the possibility of htt∗ getting clogged in the channel of the 20S core subunit blocking access to other ubiquitinated substrates and therefore impairing the proteostasis of the cell. Several experiments have been performed in order to answer whether proteasome can degrade htt or not and results supporting both hypothesis can be found. Experiments performed with peptides containing polyQ tracts and incubated with purified eukaryotic proteasomes showed no digestion of the polyQ tract by the proteasome (Holmberg et al., 2004;Venkatraman et al., 2004), and more importantly, when degrading expanded polyQcontaining proteins, proteasomes might be generating the most toxic and aggregation-prone fragments. Such polyQ sequences (38–300Qs) exceed the lengths of normal proteasome products

(2–25 residues) and a failure of theses fragments to exit the proteasome may interfere with proteasome function and it is known that expression of pure polyQ peptides is sufficient for aggregation (IB formation) and toxicity in cells (Yang et al., 2002; Raspe et al., 2009). However, more recent studies showed through quantitative flow cytometry and live-cell time-lapse imaging that N-htt—whether aggregated or not—does not choke or clog 26S and proposes that 26S activity is compromised only indirectly as a result of disrupted protein folding homeostasis (Hipp et al., 2012). As a matter of fact, UPS has been proved to be able to degrade htt∗-exon1 completely, including the expanded polyQ sequence (Michalik and Van Broeckhoven, 2004; Juenemann et al., 2013), although it has also been shown that the degradation signal that accompanies the polyQ tract is determinant to obtain this result (Juenemann et al., 2013).

There has always been a controversy regarding the pathogenicity of the IBs. Studies using transfected cells further suggested that toxicity might be induced by aggregates (Waelter et al., 2001). On the other hand, more recent observations support they hypothesis that aggregates are not pathogenic, or even that they might be protective (Saudou et al., 1998; Kopito, 2000; Arrasate et al., 2004; Bennett et al., 2005; Bowman et al., 2005) and that the pathogenic species could be the intermediate species that generate during IB formation. Htt∗ aggregation in mouse brain is not only an early event, but occurs rapidly (Gong et al., 2012). It has been found that IBs are present in the cortex of HD brains before any sign of degeneration can be detected, and many MSSNs in the striatum lack IBs despite the presence of significant neuronal loss (Gutekunst et al., 1999). Furthermore, there are also some transgenic mouse models of HD in which IBs appear only after symptoms onset (Menalled et al., 2003), and in transfected primary cultured neurons, their ability to build IBs protects them from the toxicity elicited by htt∗ (Arrasate et al., 2004). In Poirier et al. (2002) described in detail the process of IB formation. They are dynamic structures that require constant production of htt∗ to maintain them. If the influx of the mutated protein is inhibited, IBs disappear and the neurologic phenotype of the disease improves (Yamamoto et al., 2000; Díaz-Hernández et al., 2004b, 2005). IBs are not amorphous associations of N-terminal htt∗ fragments but highly organized structures. During the formation of an IB, several intermediate species are constituted and their organized interactions give rise to the IB. The most simple species, and therefore the one that appears earlier in the disease, is the N-terminal htt∗ fragments, also called monomers. These monomers carry the expanded polyQ that renders them highly prone to aggregate. Monomers associate to form globular assemblies with an average size of 4–5 nm called oligomers. These oligomers serve as nucleation seeds to form more complicated aggregation structures; they linearly associate to form protofibrils that have an undefined length. Finally, protofibrils assemble through polar zippers to obtain β-laminas called fibers. The unorganized assembly of these fibers gives rise to IBs. In order to confirm or discard the hypothesis that intermediate species as the pathogenic structures in HD, Bennett et al. tested the effect of *in vitro*-generated soluble htt∗ fragments (Bowman et al., 2005), highly aggregated fibrillar species or soluble oligomeric aggregates (Chen and Wetzel, 2001; Poirier et al., 2002) on the

degradation of ubiquitin-dependent and ubiquitin-independent substrates by purified 26S proteasomes. No differences in proteasome activity were observed in any of the analyzed species, which argues against the notion that a direct interaction between 26S proteasomes and monomers or aggregates of expanded polyQ could result in decreased proteasome activity. However, as in those experiments performed with purified IBs the ubiquitilation process is not considered, if it were impaired it would not be detected. To overcome this limitation, Díaz-Hernández et al. (2006) tested the potential inhibition of the 26S and 20S proteasome by polyQ-containing filaments isolated from the brain of the Tet/HD94 inducible mouse model or from post-mortem HD human brain tissue. Filaments isolated from brain inhibited the endoproteolytic activities of the 26S proteasome. However, as above mentioned, when the same experiments were performed with the 20S subunit (in the presence of SDS to facilitate the entrance of the substrate into the catalytic chamber) no inhibition was detected. The selective inhibition of 26S proteasome but not of 20S subunit activities suggested a direct interaction of the ubiquitinated filaments and the 19S ubiquitin-interacting regulatory caps of the 26S proteasome. Interestingly, this interaction was confirmed by immunoelectron microscopy (Díaz-Hernández et al., 2006). These results advocate that fibrillar, and possibly also oligomeric, ubiquitinated polyQ aggregates have the potential to interfere with 26S proteasome through interacting with its 19S subunit but only when theses aggregates are not recruited into IBs. These results therefore strengthen the notion that IB formation may be protective, in this case, by neutralizing the inhibitory action of dispersed ubiquitynated polyQ smaller aggregates.

# **SHORT-LIVED FLUORESCENT UPS REPORTER PROTEINS**

All the above mentioned experimental approaches focus their attention only on the proteasome activity without taking into account the complexity of the UPS. As shown by numerous genetic evidence, neurodegenerative diseases can be caused by alterations not only at the proteasome level but also at any other step that affects a substrate to be targeted to UPS degradation such as E1, E2, or E3 ubiquitin-ligating processes. In the human genome there are only 16 subtypes of E1 (Ardley and Robinson, 2005), which reflects the low specificity of this step as one E1 can recognize 100s of substrates. There are only 53 E2 coded in the human genome (Ardley and Robinson, 2005), this shows a higher specificity than E1 enzymes but still a single E2 can recognize several substrates. However, more than 527 E3 enzymes have been described (Ardley and Robinson, 2005), hence indicating that they are highly specialized to certain families of substrates. Taking into account the specificity of the pathogenic hallmarks of each neurodegenerative disease, if we are considering the possibility of finding UPS impairment in the ubiquitilating process, the best candidates would be the E3 ligases due to their substrate specificity. A new tool has been recently developed to test the implication of the ubiquitilating process in a very physiological environment. This tool consists of the use of degron-reporter proteins. These reporter proteins result from fusing a UPS degradation signal to a fluorescent protein that converts it into a reporter of UPS activity. These modified proteins have an extremely

short half-life and will accumulate only in those cells where the UPS is not working efficiently thus offering cellular resolution (Dantuma et al., 2000; Stack et al., 2000; Neefjes and Dantuma, 2004). Many degradation signals can be attached to the protein and each one will undergo ubiquitilation through different combinations of E1-E2-E3 enzymes, as a consequence, if the combination is not the one affected in the disease, the impairment could be undetected.

The most frequently used degradation signals in HD models are CL1 degron and ubiquitin fusion degradation (UFD) signal. CL1 degron is a 16 amino acid sequence that easily destabilizes proteins by labeling them for ubiquitilation. CL1 degron has been used mainly in cellular models (Bence et al., 2001; Bennett et al., 2005) in which global impairment of the UPS, that results from an intrinsic property of N-terminal htt∗ and not from its sequestration into IBs, was detected. It should be noted though that CL1 is aggregation-prone so would also co-aggregate and increase in levels when the proteostasis system slows down, independent of proteasome function. Moreover, Wang et al used this reporter protein in a mouse model of HD, the R6/2 mouse model, and reported a synapse-specific loss of proteasome activity in R6/2 mice by measuring peptidase activity in isolated synaptosomes.

Ubiquitin fusion degradation signal is an N-terminal-linked Ub molecule that, on one hand has a G76V substitution that prevents removal of the Ub by DUBs and, on the other hand, serves as acceptor for polyUb chains. This degradation signal has been used both in cellular and animal HD models. To generate a mouse model to explore UPS dysfunction, the UFD signal was fused to the GFP protein and the transgene is under the control of the cytomegalovirus immediate-early enhancer and the chicken β-actin promoter so the mice show ubiquitous expression of the reporter (Dantuma et al., 2000). Experiments performed in double transgenic R6/2 ubiquitin-reporter mouse models reported that the UPS remained functionally active in HD and that an agedependent decline in UPS activity was found to correlate with the age-related accumulation and aggregation of htt∗ in HD mouse brains (Maynard et al., 2009). These results appear to contradict the accumulation of polyUb chains in the brain of R6/2 mice and human HD patients; however, experiments with the inducible HD94 mouse model (Ortega et al., 2010) reconcile the data from cell models supporting polyQ-induced UPS impairment with the contradictory findings of no impairment in constitutive mouse models by showing that the expression of htt∗ does have the potential to induce UPS impairment*in vivo* in mouse models, thus fitting with previous observations in cell models expressing fluorescent reporter proteins. However, htt∗-induced UPS impairment *in vivo* is transient and, in good agreement with previous reports combining polyQ mouse models with the same or similar reporter mice, it is not detected with constitute htt∗ expression in adult mice. That the aggregate formation correlates with UPS recovery had also been reported in a cell model (Mitra et al., 2009), and Ortega et al. (2010) was able to demonstrate causality with the use of anti-aggregation compounds in a cell model and also *in vivo* in mouse models supporting the notion that formation of IBs has a beneficial effect by sequestering the smaller and more toxic species of htt∗.

# **UPS IMPAIRMENT AS SECONDARY EFFECT AND ITS THERAPEUTIC IMPLICATIONS**

Most of the experiments conducted to elucidate the implication of the UPS in HD have been pursued considering htt∗ directly involved with the UPS. However, UPS impairment could be a consequence of the impairment of other metabolic pathways in which htt participates. N line with this, it has been suggested that UPS impairment might originate at the level of mitochondrial function/dysfunction. As an ATP-dependent process, the efficiency of ubiquitinated substrate degradation by the proteasome is linked to mitochondrial respiration. Htt∗ has been shown to interfere with mitochondria, leading to reduction in mitochondrial trafficking (Orr et al., 2008), and reduced ATP content has been detected in synaptosomes fractions prepared from the brains of HD knock-in mice (Orr et al., 2008; Wang et al., 2008) These findings suggest that mitochondrial dysfunction may contribute to UPS impairment in HD by depleting criticalATP levels. More recently, a theory of global proteostasis network dysfunction has been proposed in which a rising concentration of htt∗ causes delayed maturation of other cellular chaperon clients, promoting their ubiquitilation and proteasomal degradation (Hipp et al., 2012). This would trigger a competition between increasing numbers of ubiquitinated substrates that may result in UPS dysfunction, independent of any impairment in proteasome activity.

Regarding the therapeutic implications of the current knowledge of the UPS in HD, it seems reasonable to think that any agent that directly or indirectly increases proteolytic processing might be benefitial. While pharmacological inhibitors of the proteasome such as bortezomib are available and have reached the clinic for treatment of various types of cancer (Papandreou and Logothetis, 2004; Nawrocki et al., 2005; Richardson et al., 2006), no pharmacological activators of the proteasome are available. Interestingly, 36% of cancer patients treated with the proteasome inhibitor bortezomib develop peripheral neuropathy (Richardson et al., 2006), thus confirming the neurotoxicity decreased proteasome activity and strengthen the notion of a potential neuroprotective action of agents able to boost proteasome activity. Another possibility would be the use of pharmacological chaperones that bind to and stabilize the folded, functional form of a mutant protein or help to direct it to degradation or refolding pathways (Balch et al., 2008). In the meantime in the absence of proteasome activating drugs, any other agent able to diminish the load of unfolded proteins could be beneficial and, so far, related clinical trials for HD are based on compounds that might indirectly alleviate burden on proteasome by decreasing protein aggregation or by increasing degradation through other pathways like autophagy as is the case of trehalose and rapamycin (Sarkar and Rubinsztein, 2008).

# **ACKNOWLEDGMENTS**

This work was supported by the Spanish Ministry of Science MINECO (SAF2012-34177 grant to Jose J. Lucas), by CIBERNED and by an institutional grant from Fundación Ramón Areces.

### **REFERENCES**

Ambrose, C. M., Duyao, M. P., Barnes, G., Bates, G. P., Lin, C. S., Srinidhi, J., et al. (1994). Structure and expression of the Huntington's disease gene: evidence

against simple inactivation due to an expanded CAG repeat. *Somat. Cell Mol. Genet.* 20, 27–38. doi: 10.1007/BF02257483


**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.

*Received: 30 April 2014; accepted: 10 September 2014; published online: 29 September 2014.*

*Citation: Ortega Z and Lucas JJ (2014) Ubiquitin–proteasome system involvement in Huntington's disease. Front. Mol. Neurosci. 7:77. doi: 10.3389/fnmol.2014.00077*

*This article was submitted to the journal Frontiers in Molecular Neuroscience. Copyright © 2014 Ortega and Lucas. 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.*

# Detection of ubiquitinated huntingtin species in intracellular aggregates

# **Katrin Juenemann\*, Anne Wiemhoefer and Eric A. Reits**

Department of Cell Biology and Histology, Academic Medical Center, University of Amsterdam, Amsterdam, Netherlands

### **Edited by:**

Fred Van Leeuwen, Maastricht University, Netherlands

### **Reviewed by:**

Takanobu Nakazawa, Osaka University, Japan Philipp Koch, University of Bonn, Germany

### **\*Correspondence:**

Katrin Juenemann, Department of Cell Biology and Histology, Academic Medical Center, University of Amsterdam, Meibergdreef 15, 1105 AZ Amsterdam, Netherlands e-mail: K.Junemann@amc.uva.nl

Protein conformation diseases, including polyglutamine (polyQ) diseases, result from the accumulation and aggregation of misfolded proteins. Huntington's disease (HD) is one of nine diseases caused by an expanded polyQ repeat within the affected protein and is hallmarked by intracellular inclusion bodies composed of aggregated N-terminal huntingtin (Htt) fragments and other sequestered proteins. Fluorescence microscopy and filter trap assay are conventional methods to study protein aggregates, but cannot be used to analyze the presence and levels of post-translational modifications of aggregated Htt such as ubiquitination. Ubiquitination of proteins can be a signal for degradation and intracellular localization, but also affects protein activity and proteinprotein interactions. The function of ubiquitination relies on its mono- and polymeric isoforms attached to protein substrates. Studying the ubiquitination pattern of aggregated Htt fragments offers an important possibility to understand Htt degradation and aggregation processes within the cell. For the identification of aggregated Htt and its ubiquitinated species, solubilization of the cellular aggregates is mandatory. Here we describe methods to identify post-translational modifications such as ubiquitination of aggregated mutant Htt. This approach is specifically described for use with mammalian cell culture and is suitable to study other disease-related proteins prone to aggregate.

**Keywords: Huntington's disease, huntingtin, aggregation, formic acid, ubiquitination**

# **INTRODUCTION**

Various age-dependent neurodegenerative disorders, such as Huntington's disease (HD), Parkinson's disease (PD) or Alzheimer's disease (AD), are hallmarked by aggregation of misfolded proteins in neurons with age (Ross and Tabrizi, 2011; Benilova et al., 2012; Irwin et al., 2013). HD is one of nine polyglutamine (polyQ) diseases caused by an expansion of a CAG trinucleotide repeat that encodes a polyQ tract in the huntingtin (Htt) protein that triggers its aggregation (Yamada et al., 2008). A neuropathological hallmark of HD is the presence of intracellular inclusion bodies composed of mutant Htt N-terminal fragments found in human postmortem brain, animal models, and cell culture models (DiFiglia et al., 1997; Gutekunst et al., 1999; Lunkes et al., 2002; Schilling et al., 2007; Juenemann et al., 2011). The idea that intracellular inclusions contain N-terminal fragments of Htt has been supported by immunological studies, where inclusions of HD patients could be stained by Htt antibodies against the N-terminal part of Htt but not antibodies against the C-terminal part (Lunkes et al., 2002). Moreover, pathological changes are accelerated in HD mouse models overexpressing N-terminal fragments of mutant Htt compared to those with full-length mutant Htt (Mangiarini et al., 1996; Hodgson et al., 1999; Schilling et al., 1999; Wheeler et al., 2000). Therefore, proteolytic processing of mutant Htt is assumed to play a key role in the pathogenesis of HD.

The age-related decline in protein homeostasis challenges the capacity of neurons to counteract the accumulation of misfolded proteins, such as Htt fragments, and this may explain in part the late onset of HD and other protein conformation diseases (Balch et al., 2008). However, the role of inclusions in HD has been challenged and inclusions have even been thought to be protective (Arrasate et al., 2004; Arrasate and Finkbeiner, 2012). Interestingly, neuronal Htt aggregates found in human postmortem brain material are positive for ubiquitin, indicating a role of the ubiquitin-proteasome related protein quality control system (DiFiglia et al., 1997; Gutekunst et al., 1999; Sieradzan et al., 1999). Previous studies showed that wildtype and mutant Htt are cleared by the proteasomal and autophagosomal pathways (Wyttenbach et al., 2000; Waelter et al., 2001; Qin et al., 2003; Thompson et al., 2009; Li et al., 2010; Juenemann et al., 2013). Furthermore, Htt can be ubiquitinated at its N-terminal region, suggesting specific ubiquitin-mediated degradation by cellular clearance mechanism (Steffan et al., 2004; Juenemann et al., 2013; Lu et al., 2013; Bhat et al., 2014).

Covalent attachment of ubiquitin to a lysine side chain of a target protein is a multistep process. Different enzymatic components are required for protein ubiquitination, including ubiquitin-activating enzyme (E1), ubiquitin-conjugating enzyme (E2), and ubiquitin-ligating enzyme (E3) (Hershko and Ciechanover, 1998). Ubiquitin itself contains seven lysine residues that can be modified by successive rounds of ubiquitination resulting in isopeptide-linked ubiquitin chains of distinct topologies determining the cellular purpose of the ubiquitinated substrate. Protein ubiquitination is important for diverse intracellular processes such as degradation and localization (Komander and Rape, 2012). Previous studies have shown that Htt can be ubiquitinated via the action of specific E3 ubiquitin ligases. Transient overexpression of the co-chaperone and E3 ligase C-terminus of HSC70-interacting protein (CHIP) increases the ubiquitination and clearance of polyQ-expanded Htt and ataxin-3 in a cell culture model (Jana et al., 2005). Overexpression of Ube3a in an HD knock-in mouse model enhances mutant Htt degradation via the proteasomal pathway resulting in reduced Htt aggregation (Bhat et al., 2014). Furthermore, other E3 ubiquitin ligases, such as ERAD-associated E3 ubiquitin-protein ligase (HRD1), TNF receptor-associated factor 6 (TRAF6), and ubiquitin-like with PHD and ring finger domains (UHRF-2), are supposed to ubiquitinate Htt independent of the polyQ-length (Yang et al., 2007; Iwata et al., 2009; Zucchelli et al., 2011).

Studying the intracellular ubiquitination pattern of soluble and insoluble mutant N-terminal Htt fragments offers a valuable insight into Htt aggregation, and reveals whether Htt is properly ubiquitinated for proteasomal or autophagosomal degradation.

To study the composition of Htt species in aggregates solubilization thereof is mandatory. Hazeki et al. showed that Htt aggregates occuring in transfected COS cells overexpressing mutant Htt-exon1 can be solubilized by pure formic acid into monomeric forms (Hazeki et al., 2000). High molecular weight complexes of mutant Htt dissociate in concentrated formic acid and HCl, whereas 8M Urea, 6M Guanidine/HCl, 1N NaOH and pure acedic acid showed low potential to interfere with Htt complex stability. In addition, Iuchi et al. showed that inclusions in PC12 cells formed by polyQstretches of the mutant androgen receptor are dissolved to monomers by the use of concentrated formic acid (Iuchi et al., 2003). Dissociation of inclusion bodies in T-hd122 cells expressing truncated mutant Htt revealed ubiquitination of the Htt proteolytic cleavage fragment cp-A (Lunkes et al., 2002).

Here we describe protocols for solubilization of Htt aggregates, based on sodium dodecyl sulfate (SDS) and formic acid treatment, which allow to identify post-translational modifications such as ubiquitination. The described methods for solubilization of intracellular inclusion bodies composed of mutant Htt N-terminal fragments can be potentially transferred in order to study further neurodegenerative diseases.

# **MATERIAL AND METHODS**

### **CONSTRUCTS**

The Htt-exon1-97Q construct was generated by replacing the C-terminal green fluorescent protein (GFP) sequence of Htt-exon1-97Q-GFP (kindly provided by R. Kopito, Stanford University, USA) for a stop codon and the Httexon1-97Q-H4 construct was generated by cloning the

Htt-exon1-97Q sequence with a 5' *XhoI* and 3' *BamHI* site into a vector encoding a C-terminal H4-tag (His-HA-HA-His, kindly provided by J. Steffan, University of California, USA). The reporter construct Htt-exon1-25Q-GFP was kindly provided by R. Kopito and the ubiquitin constructs HA-Ub-wt and HA-Ub-K0 were a kind gift from N. Zelcer (Academic Medical Center, University of Amsterdam, The Netherlands).

# **CELL CULTURE AND TRANSFECTION**

Neuro-2a cells were maintained in Dulbecco's modified Eagle's medium (DMEM) (Invitrogen) supplemented with 10% fetal calf serum, 1 mM glutamine, 100 U/mL penicillin, and 100 µg/mL streptomycin in a humidified incubator with 5% atmospheric CO2. Neuro-2a cells were seeded in 6-well plates and transfected with polyethylenimine according to the manufacturer's instructions (Polysciences Europe).

### **WESTERN BLOT ANALYSIS OF TRITON X-100 SOLUBLE PROTEINS**

Neuro-2a cells were harvested in lysis buffer (50 mM Tris/HCl pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, 20 mM NEM, supplemented with complete mini protease inhibitor cocktail (Roche)) 24 h after transfection. Total cell lysate was boiled for 10 min at 99◦C with 1x laemmli sample loading buffer (350 mM Tris/HCl pH 6.8, 10% SDS, 30% glycerol, 6% β-mercaptoethanol, bromphenol blue), fractionated on a 12% SDS-gel by SDS-PAGE gel electrophoresis and transferred to a polyvinylidene fluoride (PVDF) membrane (0.45 µm pore size, Schleicher and Schuell). Western blot membranes were blocked with 5% milk, incubated with primary antibodies anti-Htt 1C2 (1:1000, Millipore, MAB1574), anti-Htt N18 (1:1000, Enzo, BML-PW0595-0100), anti-HA (1:1000, Sigma- Aldrich, H3663), anti-β-actin (1:1000, Santa Cruz, SC-130656), and anti-ubiquitin (1:100, Sigma-Aldrich, U5379), and subsequently incubated with secondary antibodies IRDye 680 or IRDye 800 (1:10,000; LI-COR Biosciences). Infrared signal was detected using the Odyssey imaging system (Licor).

## **FILTER TRAP ASSAY**

For analysis of Triton X-100-insoluble aggregates, filter trap assay was performed with the pellet obtained after centrifugation of the Triton X-100 cell lysate (15 min at 14,000 rpm at 4◦C). Pellet with aggregates was resuspended in benzonase buffer (1 mM MgCl2, 50 mM Tris/HCl pH 8.0) and incubated for 1 h at 37◦C with 125 U Benzonase (Merck). Reaction was stopped with 2x termination buffer (40 mM EDTA, 4% SDS, 100 mM DTT). Samples with a protein concentration of 50 µg were diluted in 2% SDS buffer and filtered through a 0.2 µm pore size cellulose acetate membrane (Schleicher and Schuell), pre-equilibrated in 2% SDS wash buffer (2% SDS, 150 mM NaCl, 10 mM Tris/HCl pH 8.0) and spotted on the membrane in doublets. Filters were washed twice with 0.1% SDS buffer (0.1% SDS, 150 mM NaCl, 10 mM Tris pH 8.0) and blocked with 5% milk for further treatment similar to western blot membranes.

# **FLUORESCENCE MICROSCOPY**

Neuro-2a cells were seeded on coverslips in a 6-well plate and transfected with the indicated DNA constructs 24 h prior to imaging. Transfected cells were fixed with 4% paraformaldehyde (PFA) and images were obtained using a confocal microscope equipped with an Ar/Kr laser and a 63x objective (Leica TCS SP8).

# **SDS-SOLUBLE/INSOLUBLE FRACTIONATION FOR ANALYSIS OF POST-TRANSLATIONAL MODIFICATIONS**

### **Procedure**


**!Caution**: The SDS-insoluble pellet is invisible and detaches very easily from the tube wall.


**!Caution**: If the formic acid is not fully evaporated after step 12 the sample will be still acidic and the bromphenol blue indicator will switch color from blue to green/yellow. In this case, the sample can be neutralized by adding 5 µl 1 M Tris/HCl pH 8.0.


**!Caution**: To compare several SDS-insoluble fractions, equivalent volumes have to be loaded.

16. Not immediately used sample fractions were stored at 4◦C.

# **Buffers**


# **RESULTS AND DISCUSSION**

Htt contains three lysine residues (K6, K9, K15) at the N17 region for putative ubiquitination (**Figure 1A**). Previous work from Steffan et al. has shown that mutation of these three lysines to arginines obviates ubiquitination and SUMOylation of soluble Htt-exon1 in cell culture (Steffan et al., 2004). An AQUA-MS approach revealed that less than 1% of soluble N-terminal Htt expressed in HEK 293 cells is ubiquitinated without considering inhibition of cellular degradation pathways (Hipp et al., 2012). Immunoprecipitation and mass spectrometry analysis are appropriate methods to identify specific soluble protein-ubiquitin conjugates. However, once misfolded proteins start to aggregate and interacting partners are co-sequestered, this represents a challenge in terms of isolation and detection of specific protein-ubiquitin conjugates.

In the last two decades, various methods have been established to detect and analyze aggregates of misfolded proteins related to neurodegeneration. Htt with an expanded polyQ-stretch is prone to misfold, triggering its aggregation, a property tightly associated with its neurotoxicity (Takahashi et al., 2010; Miller et al., 2011). Htt can be found *in vitro* and *in vivo* in a monomeric or aggregated state, and they are both detectable with different methods, including microscopic techniques, fluorescence-activated cell sorting (FACS) analysis or biochemical approaches. Until now, these methods are not applicable for analyzing post-translational modified Htt species trapped in aggregates, especially with respect to distinguishing between Htt-ubiquitin conjugates and Htt that is non-covalently associated with monoubiquitin, polyubiquitin conjugates or other co-sequestered polyubiquitinated proteins.

To investigate the connection between ubiquitination and intracellular aggregation of Htt, cells overexpressing Httexon1-97Q and as a control a mutant form lacking three N-terminal lysine residues (3xR mutant) necessary for Htt ubiquitination were analyzed by fluorescence microscopy, western blot analysis and filter trap assay. To study whether aggregation of Htt depends on its ubiquitination Neuro-2a cells were co-transfected with either Htt-exon1-97Q or its 3xR mutant variant and the aggregation reporter Htt-exon1-25Q-GFP,

respectively, and analyzed using confocal fluorescence microscopy (**Figure 1B**). The aggregation reporter Htt-exon1-25Q-GFP shows a diffuse cellular staining unless sequestered into aggregates (Juenemann et al., 2013). Both wildtype and 3xR mutant Htt-exon1-97Q form inclusion bodies within 24 h after transfection as detected with the GFP-tagged Httexon1-25Q reporter protein that becomes sequestered into the mutant Htt-exon1-Q97 aggregates, indicating that Htt aggregation is independent of its ubiquitination. To evaluate whether Htt aggregates are detectable by western blot Triton X-100 soluble fractions of cells transfected with Htt-exon1- 97Q-H4 or 3xR mutant Htt-exon1-97Q-H4 were separated by SDS-PAGE. Both Triton X-100 insoluble Htt aggregates in the stacking gel (asterisk) and the soluble Htt-exon1 monomer were recognized by using a mouse monoclonal HA antibody (**Figure 1C**). The Triton X-100 insoluble aggregates of the same transfected Neuro-2a cells were applied to a filter trap assay. The filter trap membrane was stained against Htt-exon1-97Q-H4 and ubiquitin using an HA and a ubiquitin antibody (**Figure 1D**). Both wildtype and 3xR mutant

Htt-exon1-97Q aggregates showed a positive ubiquitin staining, which could either mean that the three Htt N-terminal lysine residues are not essential for ubiquitination, there is aggregate association with ubiquitin or other ubiquitinated proteins are co-sequestered.

Together, this indicates that with microscopy, immunoblot and filter trap assay there is no discrimination between ubiquitin covalently bound to aggregated Htt-exon1 and noncovalently associated monoubiquitin, polyubiquitin conjugates or sequestered polyubiquitinated proteins. To identify ubiquitinated Htt species besides monomeric Htt within aggregates an adequate solubilization is required that is capable of destroying the strong non-covalent interactions between the aggregated proteins.

Here we provide a protocol suitable for subsequent biochemical protein analysis of aggregates, consisting of mutant Htt, by solubilizing them with 100% formic acid (**Figure 2**). Formic acid disrupts protein interactions such as hydrogen bonds. Previous work indicated that aggregation of Htt results from the formation of hydrogen bonds between polyQ stretches forming β-sheet structures (Perutz et al., 1994). Hazeki et al. showed that only formic acid and HCl but not acedic acid, Urea, Guanidine/HCl or NaOH dissociate higher molecular weight complexes of mutant Htt (Hazeki et al., 2000).

In order to solubilize SDS-insoluble protein pellets containing aggregated Htt from Neuro-2a cells transfected with the constructs encoding Htt-exon1-97Q and its 3xR Htt mutant variant, pellets were treated with 100% formic acid followed by SDS-PAGE and immunoblot analysis. The formic acid treated SDS-insoluble fraction showed the presence of monomeric Htt-exon1-97Q at around 45 kDa (arrow) and at higher molecular weight post-translational modified species of Httexon1-97Q in form of distinct bands (**Figure 3A**). Htt was stained by the Htt-specific antibodies 1C2 against the polyQ tract and N18 against the N-terminal region. With the 3xR Htt mutant variant the higher molecular Htt bands disappeared, indicating that the post-translational modification of Htt is dependent on the N-terminal lysine residues. The size difference between the distinct bands suggests a ubiquitination of Htt. The observation that the aggregated 3xR Htt mutant does not show this high molecular weight bands indicates that the Htt N-terminal lysine residues are the ubiquitination sides and not, as described as more rare forms of ubiquitination, serine, threonine and cysteine residues or the N-terminal α-amino group ubiquitination in a linear fashion (McDowell and Philpott, 2013).

To confirm that the higher molecular weight protein ladder of formic acid-dissolved HA-tagged Htt-exon1-97Q indeed

represents polyubiquitinated Htt, membranes were stained in parallel with an HA antibody and a ubiquitin antibody. Both

Neuro-2a cells transiently transfected with Htt-exon1-97Q-H4. In addition to

antibodies detect the same protein bands indicating that the specific bands are ubiquitinated species of Htt-exon1 (asterisks)

ubiquitin N-terminal HA-tag (asterisks).

(**Figure 3B**). Worth mentioning is the difference between the used antibodies regarding the recognition of formic acid-nondissolved aggregates in the SDS stacking gel. The Htt specific N-terminal polyclonal N18 antibody and the HA antibody against the Cterminal HA-tag of the Htt protein compared to the Htt-specific monoclonal 1C2 antibody and the ubiquitin antibody are capable of detecting higher molecular Htt aggregates in the SDS stacking gel. The solubilization of Htt aggregates by pure formic acid is incomplete, since aggregates are still detectable by certain antibodies in the stacking gel. This might be a consequence of high Htt accumulation due to overexpression and part of the aggregates are not solved in time, or of the potentially formic acid-resistant aggregated Htt species. Previous studies suggested that aggregated Htt is crosslinked by covalent bonds such as those formed by transglutaminase (Kahlem et al., 1996; Cooper et al., 2002; Iuchi et al., 2003; Zainelli et al., 2003, 2004).

To validate Htt ubiquitination in cellular aggregates, SDSinsoluble Htt from Neuro-2a cells co-transfected with Htt-exon1- 97Q and either HA-tagged Ub-wt or Ub-K0, where all the lysine residues are exchanged to an arginine to prevent ubiquitination, was analyzed. Incorporation of HA-Ub into the Htt polyubiquitin chain is indicated by shifts of each Htt-Ub bands upwards in the size of one (about 1 kDa, asterisks) or more HA-tags creating indistinct protein bands as detected by the 1C2 antibody (**Figure 3C**). Co-staining with the HA antibody confirms covalent binding of the overexpressed HA-Ub-wt to the Htt-exon1 protein and its ubiquitinated species (asterisks) showing that Htt gets post-translational modified by ubiquitin. However, co-expression of HA-Ub-K0 reveals no incorporation of ubiquitin lacking lysine residues.

Here we provide a protocol for the solubilization of aggregated Htt proteins using formic acid to detect ubiquitinated species of overexpressed mutant Htt-exon1. Whether this method is sensitive enough to detect ubiquitination of endogenous expressed Htt depends on the quantity of aggregated Htt and the rate of Htt ubiquitination/deubiquitination. The presented methods might be transferrable to study other neurodegenerative diseases besides HD that are hallmarked by misfolded proteins prone to aggregate. In addition, the protocol may be applied to analyze other post-translational modifications of misfolded proteins in SDS-insoluble aggregates, such as SUMOylation, phosphorylation or acetylation.

While studying the role of ubiquitination of a specific misfolded protein within the cell it is important to distinguish between the analyzed misfolded protein covalently binding ubiquitin and ubiquitin only associated with the aggregate such as co-sequestered polyubiquitinated proteins.

Further research and the identification of the ubiquitin linkage pattern of soluble and insoluble Htt dependent on the length of the polyQ stretch or cellular localization is important to understand the fate of this protein within the cell. Unraveling the disease-related alterations of Htt post-translational modifications will certainly fill a crucial gap between the current knowledge of Htt fragment generation and aggregation on the one side, and regulation of potent degradation machineries on the other side that are capable of obviating toxicity of Htt species within the cell.

# **ACKNOWLEDGMENTS**

The authors thank J. Steffan (University of California, USA) and R. Kopito (Stanford University, USA) for generously sharing plasmids. This research project was supported by the Dutch Organization for Scientific Research with a VIDI grant (NWO-Zon-MW, 91796315), by the Prinses Beatrix Fonds (W.OR10-25) and the Hersenstichting (KS2010(1)-06).

# **REFERENCES**


fragments are entirely degraded by mammalian proteasomes. *J. Biol. Chem.* 288, 27068–27084. doi: 10.1074/jbc.M113.486076


**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.

*Received: 04 August 2014; accepted: 05 January 2015; published online: 28 January 2015*.

*Citation: Juenemann K, Wiemhoefer A and Reits EA (2015) Detection of ubiquitinated huntingtin species in intracellular aggregates. Front. Mol. Neurosci. 8:1. doi: 10.3389/fnmol.2015.00001*

*This article was submitted to the journal Frontiers in Molecular Neuroscience*.

*Copyright © 2015 Juenemann, Wiemhoefer and Reits. 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*.

### *Maria E. Figueiredo-Pereira1 \*, Patricia Rockwell 1, Thomas Schmidt-Glenewinkel <sup>1</sup> and Peter Serrano2*

*<sup>1</sup> Department of Biological Sciences, Hunter College, The Graduate School and University Center, City University of New York, New York, NY, USA <sup>2</sup> Department of Psychology, Hunter College, The Graduate School and University Center, City University of New York, New York, NY, USA*

### *Edited by:*

*Ashok Hegde, Wake Forest School of Medicine, USA*

### *Reviewed by:*

*Björn Spittau, Albert-Ludwigs-University Freiburg, Germany Nicolaos Robakis, Mount Sinai School of Medicine, USA Yuling Chi, Albert Einstein College of Medicine, USA*

### *\*Correspondence:*

*Maria E. Figueiredo-Pereira, Department of Biological Sciences, Hunter College, The Graduate School and University Center, City University of New York, 695 Park Avenue, Room 827N, New York, NY 10065, USA e-mail: pereira@ genectr.hunter.cuny.edu*

The immune response of the CNS is a defense mechanism activated upon injury to initiate repair mechanisms while chronic over-activation of the CNS immune system (termed neuroinflammation) may exacerbate injury. The latter is implicated in a variety of neurological and neurodegenerative disorders such as Alzheimer and Parkinson diseases, amyotrophic lateral sclerosis, multiple sclerosis, traumatic brain injury, HIV dementia, and prion diseases. Cyclooxygenases (COX-1 and COX-2), which are key enzymes in the conversion of arachidonic acid into bioactive prostanoids, play a central role in the inflammatory cascade. J2 prostaglandins are endogenous toxic products of cyclooxygenases, and because their levels are significantly increased upon brain injury, they are actively involved in neuronal dysfunction induced by pro-inflammatory stimuli. In this review, we highlight the mechanisms by which J2 prostaglandins (1) exert their actions, (2) potentially contribute to the transition from acute to chronic inflammation and to the spreading of neuropathology, (3) disturb the ubiquitin-proteasome pathway and mitochondrial function, and (4) contribute to neurodegenerative disorders such as Alzheimer and Parkinson diseases, and amyotrophic lateral sclerosis, as well as stroke, traumatic brain injury (TBI), and demyelination in Krabbe disease. We conclude by discussing the therapeutic potential of targeting the J2 prostaglandin pathway to prevent/delay neurodegeneration associated with neuroinflammation. In this context, we suggest a shift from the traditional view that cyclooxygenases are the most appropriate targets to treat neuroinflammation, to the notion that J2 prostaglandin pathways and other neurotoxic prostaglandins downstream from cyclooxygenases, would offer significant benefits as more effective therapeutic targets to treat chronic neurodegenerative diseases, while minimizing adverse side effects.

**Keywords: J2 prostaglandins, neuroinflammation, UPP, mitochondria, neurodegeneration**

# **INTRODUCTION**

Chronic neuroinflammation is recognized as a primary mechanism involved in the pathogenesis of a variety of neurodegenerative disorders including Alzheimer, Parkinson, and Huntington diseases as well as amyotrophic lateral sclerosis (Wyss-Coray and Mucke, 2002; Liu and Hong, 2003; Glass et al., 2010; Herrup, 2010). Neuroinflammation is an active process detectable in the earliest stages of these diseases (Zagol-Ikapitte et al., 2005; Liang et al., 2007; Yoshiyama et al., 2007). The neurotoxicity associated with inflammation makes it a potential risk factor in their pathogenesis. Characterizing the self-perpetuating cycle of inflammatory processes involving microglia and astrocytes in the brain that drives the slow progression of neurodegeneration could be critical for preventing/arresting these devastating disorders (Schwab and McGeer, 2008; Herrup, 2010).

Major players in inflammation are the cyclooxygenases COX-1 and COX-2 (**Figure 1**), which function as homodimers and are key enzymes in the biosynthesis of prostaglandins (Smyth et al., 2009). Although the brain expresses COX-1 and COX-2 under normal physiological conditions, it is clear that cyclooxygenases are implicated in neurodegeneration (Liang et al., 2007; Bartels and Leenders, 2010). COX-1 is generally viewed as being the homeostatic isoform, but studies suggest that it is actively involved in some forms of brain injury (Choi et al., 2009; Aid and Bosetti, 2011). The expression and activity of COX-2 are largely responsive to adverse stimuli, such as inflammation and physiologic imbalances (Yamagata et al., 1993). COX-2 activity is markedly induced in a range of neurodegenerative disorders subsequently leading to neuronal injury (Feng et al., 2003; Klivenyi et al., 2003; Teismann et al., 2003). COX-2 up-regulation following CNS injury is not restricted to neurons since COX-2 induction is also apparent in glia (Consilvio et al., 2004). Although many studies support the notion that COX-2 is involved in neurodegeneration, its contribution to the neurodegenerative process remains poorly defined. Inhibiting cyclooxygenases with nonsteroidal anti-inflammatory drugs (NSAIDs) is being explored

by specific prostaglandin synthases that differ in their cell type distribution. Of these products, PGD2 is highly unstable (estimated brain half-life of 1.1 min) resulting in the non-enzymatic formation of J2 prostaglandins.

converted to prostanoid products (PGE2, PGF2α, PGD2, PGI2, and TXA2)

as a therapeutic strategy to mitigate chronic inflammation and prevent the onset/progression of neuropathology (Klegeris et al., 2007; Vlad et al., 2008). However, the effectiveness of NSAIDs could be counterproductive by blocking the generation of all prostaglandin products of cyclooxygenases (**Figure 1**).

Current animal and cell models of neurodegenerative diseases fail to address how prostaglandins redirect cellular events to promote neurodegeneration. This is a crucial gap since some prostaglandins are neuroprotective and others neurotoxic (Lucin and Wyss-Coray, 2009; Iadecola and Gorelick, 2005). Since prostaglandins act as potent local regulators of physiologic and pathologic pathways linked to CNS inflammation, elucidating the prostaglandin-dependent pathologic pathways will have a major impact on blocking neurotoxicity linked to chronic neuroinflammation with fewer undesirable side effects, and could lead to preventing/delaying neurodegeneration.

# **FORMATION OF J2 PROSTAGLANDINS**

Prostaglandins (PGs) are a family of 20-carbon unsaturated fatty acids produced via the cyclooxygenase pathway in response to numerous extrinsic and intrinsic stimuli (**Figure 2**). The initial step in prostaglandin synthesis involves the hydrolysis of membrane sn-2 glycerophospholipids (phosphatidylcholine, phosphatidylethanolamine, and phosphatidylinositol) by phospholipase A2 (PLA2 group IVA) to release arachidonic acid (Tassoni et al., 2008; Smyth et al., 2009; Astudillo et al., 2012). PLA2 is activated by increased calcium levels and phosphorylation. This event leads to the translocation of PLA2 from the cytoplasm to intracellular membranes including the endoplasmic reticulum and nuclear envelope, to allow its access to arachidonic acid-containing phospholipid substrates (Shimizu et al., 2008).

Cyclooxygenases, which are bifunctional enzymes inserted into the ER and nuclear membranes, will then catalyze the cyclooxygenation of arachidonic acid to PGG2 followed by hydroperoxidation of PGG2 to PGH2 (Kulkarni et al., 2000; Smith et al., 2000; Simmons et al., 2004). PGH2 diffuses from the ER lumen through its membrane to the cytoplasm to be converted to more polar prostanoids via synthases localized on the cytoplasmic face of the ER (Schuster, 2002). The coupling of PGH2 synthesis with the respective downstream synthase enzymes that produce the different types of prostaglandins is intricately orchestrated in a tissue and/or cell specific manner (Funk, 2001).

J2 prostaglandins (**Figure 3**) are derived from PGD2, which is the most abundant prostaglandin in the brain (Ogorochi et al., 1984; Hertting and Seregi, 1989; Uchida and Shibata, 2008; Ricciotti and FitzGerald, 2011), and the one that changes the most under pathological conditions (Liang et al., 2005). PGD2 is produced by two distinct prostaglandin D2 synthases (PGDS), which carry out the isomerization of PGH2 to PGD2: (i) the hematopoietic enzyme (H-PGDS) and (ii) the lipocalin enzyme (L-PGDS) (Urade and Hayaishi, 2000; Urade and Eguchi, 2002). H-PGDS is a cytosolic protein found abundantly in mast cells, antigen presenting cells, and T helper type 2 (Th2) cells (Kanaoka and Urade, 2003). L-PGDS is localized in the CNS, heart and male genital organs (Urade and Eguchi, 2002). L-PGDS is one of the most abundant CSF proteins produced in the brain (Kanekiyo et al., 2007), representing 3% of total CSF protein (Xu and Venge, 2000). Secreted L-PGDS in the CSF has a dual function: it increases CSF-PGD2 levels (Scher and Pillinger, 2005) and also acts as a lipophilic-ligand carrier (Urade and Hayaishi, 2000), being a major endogenous Aβ-chaperone in the brain (Kanekiyo et al., 2007).

PGD2 is unstable and readily undergoes *in vivo* and *in vitro* non-enzymatic dehydration to generate the biologically active cyclopentenone J2 prostaglandins (**Figure 3**), which include PGJ2, -12-PGJ2, and 15-deoxy--12,14-PGJ2 (15d-PGJ2) (Shibata et al., 2002; Uchida and Shibata, 2008; Gilroy,

2010). The half-life of PGD2 in the brain was estimated to be 1.1 min and in the blood 0.9 min (Suzuki et al., 1986).

PGJ2 and its metabolites are not stored in tissues or cells and their production increases with diverse stimuli. Prostaglandins are largely produced in the brain by activated microglia, reactive astrocytes and neurons. During CNS inflammation, these cells make large quantities of prostaglandins such as PGE2 and PGD2 (Liu et al., 2003) as well as J2 prostaglandins (Bernardo et al., 2003). For example, LPS-activated microglia in culture, produced ∼3 ng/ml media of 15d-PGJ2 upon 72 h, and ∼2 ng/ml of PGD2 upon 24 h (Bernardo et al., 2003). J2 prostaglandins have been detected *in vivo* in human body fluids (Hirata et al., 1988), human atherosclerotic plaques (Shibata et al., 2002) and tissues of patients with sporadic ALS (Kondo et al., 2002; Zhang et al., 2010). In addition, a range of studies showed that J2 prostaglandins are generated *in vivo* upon various conditions related to brain injury (see below).

**FIGURE 3 | Generation of J2 prostaglandins.** PGJ2 is generated by non-enzymatic dehydration of PGD2. The J2 metabolites -12-PGJ2 and 15d-PGJ2 are formed from PGJ2 either by reactions catalyzed by human serum albumin (HSA) or by dehydration (–H2O), respectively. Asterisks indicate α, β-unsaturated carbonyl groups.

# *IN VIVO* **LEVELS OF J2 PROSTAGLANDINS IN THE CNS**

Prostaglandins are present in body fluids in the pico to nanomolar range reaching low micromolar levels at local sites of acute inflammation (Offenbacher et al., 1986; Hertting and Seregi, 1989). For example, in human airways PGD2 rose in 9 min to an average of 150-fold in five patients in response to an allergen (Murray et al., 1986). Moreover, exosomes, which are extracellular bioactive vesicles released from multivesicular bodies that mediate intercellular signaling (Subra et al., 2010), were found to contain a large panel of free fatty acids, including arachidonic acid and its derivatives, such as PGE2 and PGJ2 (Subra et al., 2010). In fact, the levels of these prostaglandins within exosomes was determined to be in the micromolar range, thus at concentrations capable of triggering prostaglandin-dependent biological effects (Subra et al., 2010).

J2 prostaglandins (**Table 1**) are bioactive cyclopentenone prostaglandins produced *in vivo* during inflammation (Rajakariar et al., 2007). Like their precursor, J2 prostaglandins can be considered some of the most abundant prostaglandins in the brain (Katura et al., 2010). For example, plasma levels of 15d-PGJ2 increased 12-fold and 23-fold in patients following acute stroke or with vascular risk factors and atherothrombotic infarcts, respectively (Blanco et al., 2005). In rodents, stroke (cerebral ischemia) and traumatic brain injury (TBI) elevate PGJ2 levels in the brain to concentrations similar to those shown to be neurotoxic *in vitro*


(Kunz et al., 2002; Hickey et al., 2007; Liu et al., 2013a,b,c; Shaik et al., 2014). Accordingly, the *in vivo* concentration of free PGJ2 in the brain upon stroke and TBI, increases from almost undetectable to the 100 nM range (Liu et al., 2011, 2013a). These levels represent average brain concentrations, but it is predicted that local cellular and intracellular concentrations of J2 prostaglandins are much higher (Liu et al., 2013b). It is also clear that this is an underestimation of the overall J2 prostaglandin levels *in vivo*, as they bind covalently to proteins (see below), and therefore reported levels of free J2 prostaglandins do not represent their total amounts.

# **MODES OF ACTION OF J2 PROSTAGLANDINS**

Despite their lipid nature, prostaglandins are charged anions thus have low intrinsic permeability across the plasma membrane, but they cross it twice (**Figure 4**): once following their synthesis as they are released from the cytoplasm into the extracellular environment (efflux), and then again when they undergo reuptake into the cytoplasm (influx), a process that mimics neurotransmitter reuptake (Schuster, 1998; Chan et al., 2002; Chi et al., 2014). Prostaglandin efflux is mediated (a) by diffusion driven by pH and the membrane potential, and (b) by the action of transporters such as multidrug resistance-associated proteins (MRPs) and prostaglandin transporters (PGTs) (Schuster, 2002; Ohkura et al., 2014). Prostaglandin influx is mediated by PGTs as well (Schuster, 2002; Chi et al., 2006). Overall, prostaglandin transport requires further investigation. A well-characterized PGT belongs to the family of the 12-transmembrane organic anion transporting polypeptides (OATPs), and mediates the influx of prostaglandins only (Chi et al., 2006; Chi and Schuster, 2010). PGT mediated influx is a required step for prostaglandin metabolism (Nomura et al., 2004), which occurs intracellularly. PGD2 is a substrate for PGT (Itoh et al., 1996) and its metabolism to PGJ2 highly likely involves PGT mediated uptake. Interestingly, this PGT is a lactate/prostaglandin exchanger in which prostaglandin influx varies with lactate levels, so that cells engaged in glycolysis thus producing high levels of lactate, are energetically poised to uptake prostaglandins via this PGT (Chan et al., 2002; Banu et al., 2003). This PGT is present in many tissues including the brain (Kanai et al., 1995; Lu et al., 1996).

Prostaglandins also can be transferred from cell to cell via exosomes, which are extracellular bioactive vesicles released from multivesicular bodies that mediate intercellular signaling (Subra et al., 2010). As such, exosomes are considered intercellular "signalosomes" as they carry arachidonic acid, phospholipases (A2 and D2), COX-1 and 2, and a whole set of prostaglandins, including PGD2, E2, and J2 (Subra et al., 2010). In fact, the concentrations of these prostaglandins within exosomes is in the

membrane. Trafficking of PGJ2 and its metabolites in and out of cells can also occur via exosomes.

micromolar range, thus at concentrations capable of triggering prostaglandin-dependent biological effects (Subra et al., 2010). Moreover, exosome internalization by neighboring cells is considered a mechanism for prostaglandins to reach their intracellular targets (Subra et al., 2010). Exosomes are released from and taken up by neurons in a synaptic activity-dependent manner that is also regulated by calcium (Lachenal et al., 2011; Perez-Gonzalez et al., 2012; Chivet et al., 2013; Morel et al., 2013). Furthermore, exosomes have recently been considered to be propagation vehicles for spreading of toxic proteins (Bellingham et al., 2012) as well as prostaglandins, such as neurotoxic PGJ2, and could thus play a significant role in the spread of pathology in a variety of neurodegenerative disorders (Schneider and Simons, 2013).

As they are unstable, prostaglandins exert their effects near their sites of synthesis thus acting as autocrine or paracrine ligands (Scher and Pillinger, 2005). The efflux of newly synthesized prostaglandins mediates their biological actions through their cell surface receptors, and prostaglandin influx from the extracellular milieu mediates their action through specific nuclear receptors or their inactivation (Banu et al., 2003). Prostaglandin inactivation in the cytoplasm is carried out by the enzyme NAD(+)-linked 15 hydroxyprostaglandin dehydrogenase (15-PGDH); its expression and that of COX-2 are reciprocally regulated in cancer, thus both enzymes control the cellular levels of prostaglandins by opposing means (Tai et al., 2006; Tai, 2011).

The effects of J2 prostaglandins are mediated by at least three different means.

### **G PROTEIN-COUPLED RECEPTORS (GPCR, FIGURE 4)**

In general, the prostaglandin GPCRs are present not only at the plasma membrane (Ricciotti and FitzGerald, 2011) but also at the nuclear membrane, thus providing for intracrine (intracellular) signaling (Zhu et al., 2006). PGD2, the precursor of PGJ2, binds to the DP1 and DP2 receptors (Urade and Eguchi, 2002). While the activation of the DP1 receptor is coupled to the G protein G*s*, resulting in increased cAMP levels, activation of the DP2 receptor and its coupling to G*i* decreases cAMP levels, and increases intracellular calcium (Hata and Breyer, 2004). J2 prostaglandins bind to DP1 and DP2, however they have a higher affinity for DP2 (as much as 100-fold) and bind to it with an affinity similar to PGD2, i.e., in the nanomolar range (Monneret et al., 2002; Pettipher et al., 2007). DP2 activation was shown to potentiate neuronal injury in hippocampal neuronal cultures and organotypic slices, while DP1 activation is neuroprotective (Liang et al., 2005, 2007).

### **NUCLEAR RECEPTORS (FIGURE 4)**

15d-PGJ2 and -12-PGJ2 are endogenous ligands for the nuclear peroxisomal proliferator activator receptor (PPARγ), to which they bind with high affinity (Gilroy, 2010; Paulitschke et al., 2012), although this remains controversial (Ricciotti and FitzGerald, 2011). PPARγ plays a major role in the regulation of adipogenesis, glucose homeostasis, cellular differentiation, apoptosis and inflammation (Qi et al., 2010). PPARγ agonists promote neuroprotection in models of stroke, AD, HD, PD, MS, and spinal cord injury, via anti-inflammatory or antioxidant-dependent mechanisms (Kapadia et al., 2008; Kiaei, 2008).

J2 prostaglandins also act through PPARγ-independent mechanisms including activation of the MAPK and JNK pathways (Wilmer et al., 2001; Li et al., 2004a), stabilization of the transcription factor Nrf2 via its interaction with Keap1 (Itoh et al., 2004; Kaspar et al., 2009; Haskew-Layton et al., 2013), and inhibition of the NFκB pathway (Rossi et al., 2000; Straus et al., 2000). This may account for the different effects of 15d-PGJ2 and other PPAR ligands (Scher and Pillinger, 2005).

# **DIRECT INTERACTION WITH INTRACELLULAR PROTEINS CAUSING A SPECIFIC POST-TRANSLATIONAL MODIFICATION (FIGURE 5)**

J2 prostaglandins are unique among the prostaglandin family in that they have α, β-unsaturated carbonyl groups (*asterisks*, **Figure 5**), promoting Michael addition reactions with free sulfhydryl groups of cysteines in glutathione and cellular proteins (Straus and Glass, 2001). These cyclopentenone PGs covalently modify several proteins, including the p50 subunit of NFκB, which may explain its anti-inflammatory effects (Cernuda-Morollon et al., 2001). They also modify thioredoxin reductase, an enzyme that protects against oxidative damage (Moos et al., 2003) and activate Ras, a small GTPase oncogene known to activate Erk signaling pathways (Oliva et al., 2003).

Electrophile binding to key protein cysteine(s) by endogenous compounds such as PGJ2 is regarded as playing an important role in determining whether neurons will live or die (Satoh and Lipton, 2007). In fact, with neuronal cultures, we established that PGJ2 was by far the most neurotoxic of four prostaglandins tested, including PGA1, D2, E2, and J2, with PGE2 being the least neurotoxic of the four under the conditions tested (Li et al., 2004b). The response to J2 prostaglandins is different from that of agonists that do not form covalent adducts with proteins. A slow steady stream-like release of J2 prostaglandins as a result of chronic neuroinflammation could be cumulative, leading overtime to accumulation of covalent PGJ2-protein adducts until they reach a toxic threshold. Thus, covalent protein modification in the brain by highly reactive electrophiles such as J2 prostaglandins, represents a novel pathologic post-translational change (Higdon et al., 2012b) and could play a critical role in progressive neurodegeneration.

# **J2 PROSTAGLANDIN: POTENTIAL TRANSITION TO CHRONIC INFLAMMATION AND PATHOLOGY SPREADING**

The mechanisms underlying the transition from acute to chronic inflammation are poorly understood. One hypothesis supported by animal studies is that in addition to their role in mediating acute inflammation, prostaglandins also function in the transition and maintenance of chronic inflammation, culminating in long-lasting effects (Aoki and Narumiya, 2012). Prostaglandins accomplish this by amplifying cytokine signaling, up-regulating COX-2 (**Figure 6**), inducing chemokines, and recruiting inflammatory cells such as macrophages (Aoki and Narumiya, 2012). We and others demonstrated that J2 prostaglandins induce COX-2 up-regulation in cancer cells (Kim et al., 2008; Kitz et al., 2011) and neuronal cells (Li et al., 2004a) and this event can be driven by MAPK activation (Li et al., 2004a; Kitz et al., 2011) or Akt/AP-1 activation (Kim et al., 2008).

It is also possible that J2 prostaglandins mediate the spread of neurodamage within the brain via exosomes (**Figure 6**), as discussed above under "Modes of action of J2 prostaglandins." This is important because exosomes were recently considered propagation vehicles for spreading neuropathology in a range of neurodegenerative diseases including AD, PD, and HD (Schneider and Simons, 2013). *In vitro* studies demonstrate that exosomes are released from neurons in a synaptic activity-dependent manner regulated by calcium (Lachenal et al., 2011). This later aspect of the regulation of the exosome-release by calcium could be particularly relevant to J2 prostaglandins. They are potent agonists (EC50 of ∼10 nM) of the DP2 receptor (Monneret et al., 2002), which signals through elevation of intracellular calcium and reduction in intracellular cAMP (Pettipher et al., 2007), thus J2 prostaglandins could induce exosome-release.

These findings support the notion that targeting prostaglandin signaling for therapeutics represents a highly innovative strategy to prevent/block chronic neuroinflammation and disease progression.

# **J2 PROSTAGLANDIN TARGETS: UPP AND MITOCHONDRIA**

COX-2 neurotoxicity seems to be mediated by PGD2 but not by PGE2 (Liang et al., 2005). PGD2 is the most abundant prostaglandin in the brain (Abdel-Halim et al., 1977; Narumiya et al., 1982; Hertting and Seregi, 1989). For example, in young rats (16–18 post-natal) subjected to a 12-min asphyxial cardiac arrest, the brain levels of PGE2 assessed by UPLC–MS/MS were ∼35.5 pmol/g of tissue, while those of PGD2 were at least 26 fold higher, reaching ∼937 pmol/g of tissue (Shaik et al., 2014). PGD2 elicits its cytotoxicity via its bioactive metabolites J2 prostaglandins (Liu et al., 2013b). In contrast to other reviews (Musiek et al., 2005; Uchida and Shibata, 2008; Scher and Pillinger, 2009; Surh et al., 2011; Oeste and Perez-Sala, 2014), our review addresses in detail the effects of J2 prostaglandins on two targets that play key roles in the neurodegenerative process, namely, the ubiquitin-proteasome pathway and mitochondrial function (**Figure 7**).

# **UBIQUITIN-PROTEASOME PATHWAY (UPP, FIGURE 7)**

It is well-established that in neuronal cells J2 prostaglandins trigger the accumulation/aggregation of ubiquitinated proteins (see for example Li et al., 2004b; Ogburn and Figueiredo-Pereira, 2006; Liu et al., 2011, 2013a). This protein accumulation/aggregation can be mediated by at least two mechanisms shown to be affected by J2 prostaglandins.

### *Inhibition of the 26S proteasome*

We and others demonstrated that J2 prostaglandins impair the 26S proteasome (Shibata et al., 2003; Ishii and Uchida, 2004;

**FIGURE 7 | J2 prostaglandins target the ubiquitin proteasome pathway (UPP) and mitochondria.** J2 prostaglandins affect the UPP by: (1) impairing the 26S proteasome by inducing oxidation of proteasome subunits, or promoting its disassembly, (2) inhibiting de-ubiquitinating enzymes (DUBs), and (3) covalently modifying specific active site cysteines on UPP components such as E1 activating enzymes, E2 conjugating enzymes, and some E3 ligases. J2 prostaglandins can also inhibit mitochondrial function by: (1) inhibiting NADH-ubiquinone reductase in complex I, (2) reducing membrane potential, (3) blocking fission, and (4) inducing the generation of reactive oxygen species (ROS) and apoptosis.

Wang et al., 2006; Koharudin et al., 2010). In neuronal cells, these prostaglandins induce the oxidation of at least one proteasome subunit, i.e., S6 ATPase (Rpt5), which seems to be one of the proteasome subunits most vulnerable to protein carbonylation (Ishii et al., 2005). PGJ2 also promotes dissociation of the 20S core particle from the 19S regulatory particle (Wang et al., 2006), resembling the effects of agents that induce oxidative stress (Aiken et al., 2011). The effect of J2 prostaglandins on the proteasome are attributed to their electrophilic properties, since a 15d-PGJ2 analog that lacks the double bond in the cyclopentenone ring failed to inhibit the proteasome (Shibata et al., 2003). Furthermore, 15d-PGJ2/proteasome conjugates were detected in neuronal cells treated with biotinylated 15d-PGJ2 (Shibata et al., 2003).

These data indicate that one of the effects of inflammation mediated by J2 prostaglandin is proteasome inhibition. Conversely, proteasome inhibition suppresses inflammation (Bi et al., 2012). It was recently shown that the mechanisms by which the antibiotic rifampicin suppresses microglia activation upon LPS-treatment, is by downregulating the Rpt1 (MSS1) proteasome subunit (Bi et al., 2012). Likewise, downregulation of the proteasome subunit Rpn9 (PSMD13) by siRNA suppresses microglial activation and diminishes the production of inflammation associated factors such as nitric oxide synthase, nitric oxide, cyclooxygenase-2, and prostaglandin E2 (Bi et al., 2014).

A close relation between proteasome and inflammation is further supported by the finding that cyclooxygenases 1 and 2 are turned over by the 26S proteasome (Rockwell et al., 2000; Yazaki et al., 2012) via the endoplasmic reticulum-associated degradation (ERAD) pathway (Mbonye et al., 2006, 2008). Proteasomalmediated turnover of COX-2 is regulated by the G proteincoupled receptor prostaglandin E1 (EP1 for PGE2) independently of receptor activation (Haddad et al., 2012). EP1 is present both at the plasma membrane and at the inner and outer membrane of the nuclear envelop (Bhattacharya et al., 1998). EP1 may act as a scaffold for an E3 ligase that ubiquitinates COX-2 (Haddad et al., 2012). In addition to the cyclooxygenases, prostaglandin synthases are also turned over by the proteasome. As such, increases in intracellular calcium result in the rapid proteasomal-mediated degradation of the PGD2 prostaglandin synthase H-PGDS in human megakaryocytic cells (Yazaki et al., 2012). Degradation of other prostaglandin synthases by the proteasome remains to be investigated.

## *Inhibition of de-ubiquitinating enzymes (DUBs)*

The UPP and autophagy play a critical role in protein quality control and thus have attracted special attention for drug development (Edelmann et al., 2011). In regard to the UPP, which is the focus of this review, the initial effort was directed toward proteasome inhibitors, although not particularly for neurological conditions (Ristic et al., 2014). However, proteasome inhibition can be quite unspecific and thus may lead to a range of undesirable side effects. De-ubiquitinating enzymes known as DUBs are relevant UPP targets upstream of the proteasome and are the focus of recent attention for drug development. Several DUBs have been targeted for cancer treatment (Crosas, 2014), and one for neurological conditions (USP14) (Lee et al., 2010). At least one hundred (if not more) DUBs from five different gene families are present in humans, and four of these families are thiol proteases and one is a metalloprotease (Eletr and Wilkinson, 2014). These DUBs carryout different functions, some of which involve processing newly translated ubiquitin (Ub) to provide monomers for conjugation and chain formation, or trimming mono-Ub from the distal end of a poly-Ub chain, or disassembling poly-Ub chains, or removing poly-Ub chains from their substrates (Clague et al., 2013). All together these functions provide a means for DUBs to regulate "where, when, and why" ubiquitinated substrates are degraded by the 26S proteasome (Ristic et al., 2014).

We and others showed that J2 prostaglandins inhibit some of the thiol DUBs including UCH-L1 and UCH-L3 (Mullally et al., 2001; Li et al., 2004b; Liu et al., 2011). In particular, UCH-L1 is predicted to be one of the most abundant proteins in the brain reaching 1–2% of the total protein content (Wilkinson et al., 1992), and its activity is highly diminished by J2 prostaglandins (Li et al., 2004b; Koharudin et al., 2010). These prostaglandins trigger UCH-L1 unfolding and aggregation by forming a covalent adduct with a single thiol group on Cys 152 (Koharudin et al., 2010). UCH-L1 plays an important role in aging and neurodegenerative disorders such as AD and PD, thus its alterations by J2 prostaglandins are highly relevant to the neurodegenerative process as discussed below under "neurodegenerative disorders" and reviewed in (Ristic et al., 2014).

# *Alteration of other UPP components that have cysteines at the active site*

Other components of the UPP, such as E1 activating enzymes, E2 conjugating enzymes, and some E3 ligases also have active site cysteines (Metzger et al., 2014). Whether J2 prostaglandins affect these and/or other thiol DUBs remains to be investigated. Protein modification by J2 prostaglandins seems to be highly selective where only a specific set of cysteine residues are vulnerable to covalent modification and this event occurs independently of protein thiol content (Higdon et al., 2012a; Vasil'ev et al., 2014). Of the several hundred cellular proteins with potentially reactive thiols, only ∼10% form covalent adducts with J2 prostaglandins, and are considered members of the "electrophile-responsive proteome" (Ceaser et al., 2004). Although J2 prostaglandins interact with selective targets, they can covalently modify a wide variety of intracellular proteins. Proteomic approaches established that J2 prostaglandins covalently bind to specific sites within the plasma membrane, nuclear and cytosolic subcellular fractions (Yamamoto et al., 2011). At least eleven plasma membrane proteins were identified as binding biotinylated 15d-PGJ2 and they were distributed into three functional groups: glycolytic enzymes, molecular chaperones, and cytoskeletal proteins (Yamamoto et al., 2011). Furthermore, J2 prostaglandin binding alters protein catalysis, binding, structural function, and transport (Marcone and Fitzgerald, 2013; Oeste and Perez-Sala, 2014).

# **MITOCHONDRIA (FIGURE 7)**

J2 prostaglandins inhibit mitochondrial function leading to oxidative stress and apoptosis (Kondo et al., 2001; Lee et al., 2009; Paulitschke et al., 2012). These prostaglandins inhibit the activity of the enzyme NADH-ubiquinone reductase from mitochondrial respiratory complex I, most likely by adduct formation (Martinez et al., 2005). J2 prostaglandins also induce the generation of reactive oxygen species (ROS), trigger a drop in membrane potential (Pignatelli et al., 2005; Gutierrez et al., 2006), interact with the cytoskeleton (Stamatakis et al., 2006) and block mitochondrial division through covalent modification of dynamin-related protein 1 (Drp1) which regulates mitochondrial fission (Mishra et al., 2010). A mitochondrial targeted analog of 15d-PGJ2 (mito-15d-PGJ2) was more efficient than 15d-PGJ2 at inducing apoptosis, was less potent at up-regulating Keap1-dependent antioxidant expression of HO-1 and GSH, and caused profound defects in mitochondrial bioenergetics and mitochondrial membrane depolarization (Diers et al., 2010). This interesting approach that involved specifically targeting 15d-PGJ2 to mitochondria by conjugation to a lipophilic cation, demonstrates the feasibility of manipulating its biological effects (Diers et al., 2010).

### **UPP/MITOCHONDRIA INTERACTION**

UPP and mitochondrial function are inherently connected, and since J2 prostaglandins affect both processes we will discuss this interaction. On the one hand the UPP degrades various mitochondrial proteins contributing to mitochondrial quality control (Taylor and Rutter, 2011). Mitochondrial proteins that are UPP substrates include (a) damaged and/or misfolded nuclear encoded proteins that are destined for import into mitochondria, (b) defective proteins at the outer mitochondrial membrane (OMM) extracted by p97 and delivered to the proteasome, and (c) non-OMM proteins, although the mechanistic details of how these proteins in the inner compartments retrotranslocate to the OMM remains poorly defined (Shanbhag et al., 2012). In addition, both the ubiquitin ligase parkin (Narendra and Youle, 2011) and the de-ubiquitinating enzyme USP30 (Bingol et al., 2014) seem to regulate mitochondrial turnover via mitophagy. The PINK1-Parkin pathway proposed to promote mitophagy remains controversial in neurons (Exner et al., 2012).

On the other hand, mitochondria provide ATP for protein ubiquitination and for 26S proteasome function (Livnat-Levanon and Glickman, 2011). E1 activity, the first step of the ubiquitination cascade, requires ATP for formation of a thiol ester adduct with ubiquitin (Haas et al., 1982; Schulman and Harper, 2009). If E1 activity is impaired all protein ubiquitination should be diminished. Moreover, the degradation of proteins by the 26S proteasome is highly dependent on ATP binding and hydrolysis (Liu et al., 2006).

It is postulated that in neurons even a modest restriction of ATP production by mitochondria far outweighs the negligible effects of ROS, although the underlying mechanisms are not clearly understood (Nicholls, 2008). In a recent study with neurons (Huang et al., 2013), we demonstrated that low ATP levels caused by mitochondrial dysfunction, correlated with impairment of the UPP: there was a decline in E1 and 26S proteasome activities with a concomitant rise in 20S proteasomes. This decline in UPP function occurred upon acute and long-term mitochondrial impairment. Notably, upon energy depletion, calpain activation led to the selective cleavage of Rpn10, a 26S proteasome subunit, without affecting other proteasome subunits tested. Rpn10 cleavage combined with ATP depletion, contributed to the decline in 26S proteasome function. We postulated that upon mitochondrial dysfunction, ATP-depletion and calpain activation contribute to the demise of protein turnover by the UPP in favor of unregulated and energy-independent protein degradation by 20S proteasomes. This adaptive response to energy deficiency may be suitable for short-term periods to promote degradation of randomly unfolded oxidized proteins. However, if chronic it can lead to neurodegeneration, as regulated protein degradation by the UPP is essential for neuronal survival (Huang et al., 2013).

# **COMPLEX EFFECTS OF J2 PROSTAGLANDINS**

Overall, the role of J2 prostaglandins in inflammation is complex (Harris et al., 2002; Wall et al., 2012). On the one hand, they have emerged as key anti-inflammatory agents as they inhibit the production of pro-inflammatory mediators such as iNOS, TNFα, and IL1β, suppress microglia and astrocyte activation and induce apoptosis (Eucker et al., 2004; Giri et al., 2004; Mrak and Landreth, 2004). On the other hand, J2 prostaglandins are pro-inflammatory agents. They stimulate the production of proinflammatory mediators such as IL8 and activate MAPK (Meade et al., 1999; Zhang et al., 2001). Furthermore, 15d-PGJ2 seems to play a role in the regulation of human autoimmune diseases and to inhibit inflammation in models of arthritis, ischemiareperfusion injury, inflammatory bowel disease, lupus nephritis, and AD (Scher and Pillinger, 2009).

J2 prostaglandins also display both protective and destructive effects. Their biological activities include antiviral and antitumoral effects, modulation of the heat shock response, induction of oxidative stress, and apoptosis (Uchida and Shibata, 2008), as well as up-regulation of the death receptor 5 (DR5), sensitization to TRAIL-induced cytotoxicity, and caspase 8 activation (Kondo et al., 2002; Nakata et al., 2006; Su et al., 2008; Metcalfe et al., 2012). Although their anti-proliferative and pro-apoptotic effects are most frequently described, J2 prostaglandins also induce the proliferation of different forms of cancer cells when used at nanomolar to low micromolar concentrations (Oliva et al., 2003).

The conflicting effects of J2 prostaglandins most likely depend on their intracellular targets and downstream pathways which can be dose- and cell-type dependent (Servidei et al., 2004). J2 prostaglandins may exert some of their anti- or pro-inflammatory as well as anti- or pro-survival effects through PPARγ-dependent mechanisms. However, these prostaglandins also exert their actions through PPARγ-independent pathways as discussed above under "modes of action of J2 prostaglandins." As such, it is known that specific PPARγ ligands, such as pioglitazone, do not replicate all of the effects attributed to J2 prostaglandins. Identifying the mechanism by which J2 prostaglandins exert their neurotoxic effects could lead to new strategies to prevent and/or delay neurodegeneration linked to inflammation.

# **J2 POSTAGLANDINS AND NEURODEGENERATIVE DISORDERS (TABLE 2)**

Chronic neuroinflammation is recognized as a primary mechanism involved in neurodegenerative diseases such as AD, PD, and ALS (Liu and Hong, 2003; Glass et al., 2010; Cudaback et al., 2014; Mosher and Wyss-Coray, 2014). Moreover, UPP and mitochondrial dysfunction as interdependent cellular events (Livnat-Levanon and Glickman, 2011; Taylor and Rutter, 2011) are both impaired in many neurodegenerative disorders (Lin and Beal, 2006; Paul, 2008). However, the underlying mechanisms that bring about and/or maintain these malfunctions are unknown. A self-perpetuating cycle of inflammatory processes involving brain immune cells (microglia and astrocytes) could drive the slow progression of the neurodegenerative process, leading to dysfunction in protein degradation by the UPP and ATP production by mitochondria. Preventing/arresting this self-perpetuating inflammatory cycle is a very promising neuroprotective strategy for these disorders (Lima et al., 2012). We will focus on a review of studies showing that J2 prostaglandins are associated with AD, PD, and ALS, as well as stroke, TBI and Krabbe disease (**Table 2**).

### **ALZHEIMER DISEASE**

PGD2 levels were found to be significantly increased in the frontal cortex of AD patients compared to age matched controls (Iwamoto et al., 1989; Yagami, 2006). In AD patients and in Tg2576 mice, a well-established AD model (Hsiao et al., 1996), the levels of the PGD2 synthase H-PGDS and the PGD2 receptor DP1 were found to be selectively up-regulated in microglia and astrocytes within senile plaques (Mohri et al., 2007). These results support the notion that PGD2 acts as a mediator of plaque associated inflammation in the AD brain and they could also explain the pharmacologic mechanisms underlying the favorable response of patients with AD to non-steroidal anti-inflammatory drugs (Mohri et al., 2007).

The other PGD2 synthase L-PGDS, which is one of the most abundant CSF proteins produced in the brain, was localized in amyloid plaques in both AD patients and Tg2576 mice (Kanekiyo et al., 2007). Secreted L-PGDS in the CSF has a dual function: it increases CSF-PGD2 levels (Scher and Pillinger, 2005) and also acts as a lipophilic-ligand carrier (Urade and Hayaishi, 2000). L-PGDS was found to bind Aβ monomers and prevent Aβ aggregation, suggesting that L-PGDS is a major Aβ chaperone and disruption of this function could be related to the onset and progression of AD (Kanekiyo et al., 2007).

L-PGDS and PGD2 also promote migration and morphological changes of microglia and astrocytes that resemble those exhibited under reactive gliosis (Lee et al., 2012). This L-PGDS function is mediated by its interaction with myristoylated alanine-rich protein kinase C substrate (MARCKS), which in turn activates the AKT/Rho/JNK pathway (Lee et al., 2012). MARCKS is a plasma membrane resident abundant in the nervous system, and that depending on its phosphorylation by PKC acts as an actin cross-linker to regulate cellular adhesion and spreading, migration, proliferation, and fusion through its interaction with the cytoskeleton (Arbuzova et al., 2002). MARCKS also plays a role in the maintenance of dendritic spines and contributes to PKCdependent morphological plasticity in hippocampal neurons (Calabrese and Halpain, 2005). Together these results support an essential role for L-PGDS in the regulation of glial cell migration and morphology, and perhaps neuronal plasticity, within the CNS (Lee et al., 2012).

The transport of prostaglandins across cell and organelle membranes involves, among others, a prostaglandin transporter (PGT) (Kanai et al., 1995). Immunohistochemical and immunofluorescent analyses of this PGT in human brains showed its localization in neurons, microglia, and astrocytes in all brain tissues assessed (Choi et al., 2008). In addition, PGT levels were lower in AD than in age-matched control brain homogenates, suggesting that prostaglandins might not be cleared at the normal rate in AD brains (Choi et al., 2008).

In relation to J2 prostaglandins, the finding that they impair the UPP is highly relevant to AD (Selkoe, 2004; Shaw et al., 2007). Defective proteasome activity is linked to the early phase of AD characterized by synaptic dysfunction, as well as to late AD stages linked to accumulation and aggregation of ubiquitinated (Ub)-proteins in both senile plaques and neurofibrillary tangles (Upadhya and Hegde, 2007; Oddo, 2008). Moreover, J2 prostaglandins inhibit the de-ubiquitinating enzyme UCH-L1 (Li et al., 2004b; Liu et al., 2011), which is down-regulated in AD brains; UCH-L1 down-regulation is inversely proportional to the number of neurofibrillary tangles (Choi et al., 2004).

Besides the effects on the UPP, we showed in rat primary cerebral cortical cultures that PGJ2 induced accumulation of Ub-proteins, caspase-activation, TAU cleavage at Asp421, and neuritic dystrophy (Arnaud et al., 2009; Metcalfe et al., 2012).

### **Table 2 | J2 postaglandins and neurodegenerative disorders.**


TAU cleavage at Asp421 was identified as an early event in AD tangle pathology (Gamblin et al., 2003; Rissman et al., 2004; de Calignon et al., 2010). In summary, J2 prostaglandins mimic many pathological processes observed in AD.

### **PARKINSON DISEASE (PD)**

Up-regulation of PGD2 or J2 prostaglandins in PD patients has not been addressed yet, but there is ample evidence linking PGD2, the precursor of PGJ2, to PD. Changes in PGD2 levels occurring in PD brains will lead to parallel changes in the highly reactive cyclopentenone J2 prostaglandins because PGD2 is unstable and is spontaneously metabolized to J2 prostaglandins. It was calculated that the half-life of PGD2 in the brain is 1.1 min (Suzuki et al., 1986).

Significant changes in L-PGDS isoforms were detected in the CSF of at least 20 idiopathic PD patients compared to 100 controls (Harrington et al., 2006). These alterations reflected up/down regulation of L-PGDS isoforms likely (a) to represent pathology at the cellular level expected to impact prostaglandin production, and (b) to correlate with disease symptoms (Harrington et al., 2006). It was speculated that these altered isoforms could be candidate diagnostic PD biomarkers and may have predictive value (Harrington et al., 2006).

A number of studies suggest that α-synuclein plays a role in brain fatty acid metabolism including arachidonic acid, through modulation of ER-localized acyl-CoA synthetase activity (Castagnet et al., 2005; Golovko et al., 2006). Acyl-CoA synthetase is an enzyme that converts fatty acids to acyl-coA for subsequent beta oxidation. In α-synuclein KO mice, exogenous addition of wild-type mouse or human α-synuclein restored acyl-CoA synthetase activity, while mutant (A30P, E46K, and A53T) forms of α-synuclein did not (Golovko et al., 2006). In addition, the levels of several prostaglandins in brains following a 30 s global ischemia were compared in wild type vs. α-synuclein KO mice (Golovko and Murphy, 2008). Among all prostaglandins assayed (E2, D2, F2α, TxB2, and 6-ketoF1α) PGD2 showed the greatest increase (two-fold) in the α-synuclein KO mice relative to the wild type. The levels of PGD2 in brains of α-synuclein KO mice reached ∼35 ng/g following the 30 s global ischemia. Under normal physiological conditions, α-synuclein ablation had no effect. Together these studies suggest that α-synuclein could play a role in brain inflammatory responses through modulation of arachidonic acid metabolism and downstream PGD2/PGJ2 production.

As far as we know, we established the first model of inflammation in which the endogenous highly reactive product of inflammation PGJ2, induces PD pathology convincingly (Pierre et al., 2009). Microinfusion of PGJ2 into the brainstem and striatum of adult FVB male mice led to a dose-dependent reduction in the number of dopaminergic (TH+) neurons in the *substantia nigra pars compacta*, with little damage to local GABAergic interneurons, and to the appearance of Lewy-like bodies and activated glial cells. PGJ2-treatment resulted in a PD-like phenotype exhibiting gait disturbance and impaired balance. More recently we showed that this PGJ2-induced mouse model that mimics in part chronic inflammation, exhibits slow-onset PD-like pathology (Shivers et al., 2014). In this mouse model, microglia activation was evaluated *in vivo* by PET with [11C](R)PK11195 (Banati, 2002) to provide a regional estimation of brain inflammation. We also demonstrated that PACAP27, a peptide that increases intracellular cAMP levels (Moody et al., 2011), reduced dopaminergic neuronal loss and motor deficits induced by PGJ2, without preventing microglia activation. The latter could be problematic in that persistent microglia activation can exert long-term deleterious effects on neurons and behavior. In conclusion, this PGJ2-induced mouse model is optimal for testing diagnostic tools such as PET, which is a powerful technique to quantitatively assess neuroinflammation *in vivo* (Stoessl, 2014), as well as therapies designed to target the integrated signaling across neurons and microglia, to fully benefit patients with PD.

## **AMYOTROPHIC LATERAL SCLEROSIS (ALS)**

There is evidence supporting the involvement of J2 prostaglandins in ALS. As such, 15d-PGJ2 was shown to accumulate in spinal motor neurons of patients with amyotrophic lateral sclerosis (ALS) (Kondo et al., 2002; Zhang et al., 2010). Moreover, astrocytes from mice carrying the *SOD1*G93A mutation were shown to be toxic to stem cell*-*derived human motor neurons but not to interneurons (Di Giorgio et al., 2008). The astrocyte induced neurotoxicity was mediated by up-regulation of PGD2 signaling, and was prevented by MK05524, an antagonist for the PGD2 receptor DP1 (Di Giorgio et al., 2008). In a more recent study, an *in vivo* genetic approach validated the importance of this DP1-mediated mechanism for neuronal degeneration. As such, genetic ablation of DP1 in *SOD1*G93A mice extended their life span, decreased microglial activation, and reduced motor neuron loss (de Boer et al., 2014). These results suggest that blocking DP1 may be a therapeutic strategy in ALS (Di Giorgio et al., 2008; de Boer et al., 2014).

### **STROKE**

Stroke and silent brain infarcts are high risk factors for dementia and neurodegenerative diseases such as AD (Vermeer et al., 2003) and PD (Becker et al., 2010; Rodriguez-Grande et al., 2013). The cyclooxygenase pathway was considered to be a valuable therapeutic target for stroke, however while COX-2 inhibitors are able to diminish injury in stroke models, they also produce an unbalance in prostanoid synthesis that promotes damaging vascular effects (Iadecola and Gorelick, 2005). For this reason, new therapeutic strategies targeting the factors that mediate the damage downstream from COX-2 may offer stroke patients powerful new tools to ameliorate brain damage and improve their functional outcome (Iadecola and Gorelick, 2005). Some of these factors could be J2 prostaglandins, as their levels in the brain are highly elevated in rodent models of cardiac arrest and stroke (Liu et al., 2011, 2013a,c; Shaik et al., 2014). Due to their inhibitory effects on the UPP, J2 prostaglandins could play an important role in brain ischemia brought about by cardiac arrest or stroke, as UPP impairment is highly relevant to these conditions (Caldeira et al., 2014). Thus, therapeutic strategies targeting the deleterious effects of J2 prostaglandins could offer great promise.

### **TRAUMATIC BRAIN INJURY (TBI)**

TBI is another neurological condition associated with J2 prostaglandins. In rodents, TBI elevates J2 prostaglandin levels in the brain to concentrations similar to those shown to be neurotoxic *in vitro* (Kunz et al., 2002; Hickey et al., 2007). TBI initiates an inflammatory cascade that leads to acute pathologic processes as well as long-term neuronal damage (Ziebell and Morganti-Kossmann, 2010). The nuclear receptor PPARγ, for which 15d-PGJ2 is an endogenous ligand, is considered a major anti-inflammatory and neuroprotective target for treating patients with TBI. However, PPARγ activation can also trigger apoptosis (Qi et al., 2010). These opposing effects seem to be related to the level of PPARγ agonists produced (Clay et al., 1999; Na and Surh, 2003). Low PPARγ agonist levels exert neuroprotective and anti-inflammatory effects that include down-regulation of inflammatory responses, reduction of oxidative stress, inhibition of apoptosis, and promotion of neurogenesis, while high levels induced apoptosis (Qi et al., 2010). The regulatory mechanisms and signaling cascades underlying the opposing PPARγ effects require further elucidation (Qi et al., 2010).

### **KRABBE DISEASE**

This disease is associated with demyelination, for which the *twitcher* mouse is an authentic animal model (Duchen et al., 1980; Kobayashi et al., 1980). In this mouse model, myelination proceeds normally up to post-natal day 30, when demyelination initiates due to oligodendrocyte apoptosis accompanied by microglia activation and astroglyosis (Mohri et al., 2006). Remarkably, a blockade of the PGD2 signaling cascade in the *twitcher* mouse via knock-out of the PGD2 synthase H-PGDS or the PGD2 receptor DP1, or treating these mice with the H-PGDS inhibitor HQL-79 [4-benzhydryloxy-1-[3-(1*H*-tetrazol-5-yl)-propyl]piperidine], suppressed astroglyosis, demyelination, twitching and spasticity. These results support the notion that PGD2 and perhaps its metabolites, are key to the pathological demyelination occurring in the *twitcher* mouse, and the neuroprotective potential of manipulating the PGD2-signaling to overcome demyelination in Krabbe disease (Mohri et al., 2006; Bosetti, 2007).

# **POTENTIAL J2 PROSTAGLANDIN THERAPEUTIC TARGETS**

While inflammation can be beneficial, failure to adequately control its abatement when the injurious agent is neutralized, is increasingly believed to be one of the major causes of chronic inflammation (Gilroy, 2010). The therapeutic potential of the prostanoid pathway is supported by the clinical efficacy of NSAIDs, although they present risky side effects as current NSAIDs non-selectively inhibit the overall synthesis of all prostaglandins (**Table 3**).

A more recent approach of diminishing prostaglandininduced neuroinflammation is to block the activity of the enzyme monoacylglycerol lipase (MAGL) (**Table 3**). MAGL, which hydrolyzes endocannabinoids, can also regulate arachidonic acid release in the brain but not in the gastrointestinal tract, that culminates in the generation of neuroinflammatory prostaglandins (Nomura et al., 2011; Piro et al., 2012). A novel MAGL selective and irreversible inhibitor (JZL184, which is 4-nitrophenyl-4-[bis(1,3-benzodioxol-5-yl)(hydroxy)methyl] piperidine-1-carboxylate) has shown promising results for blocking the neuroinflammation in the brain associated with PD and other neurodegenerative disorders (Legg, 2011; Nomura et al., 2011).

Additional specific targets in the prostaglandin pathways, such as prostaglandin synthases, prostaglandin transporters, and prostaglandin receptors have also emerged as drug targets (**Table 3**). For example, recently a highly selective competitive inhibitor of the prostaglandin transporter PGT was found to prolong PGE2 half-life, *in vitro* and *in vivo* (Chi et al., 2011). It is expected that further delineation of the prostaglandin pathway will yield novel beneficial therapeutics in the years to come (Smyth et al., 2009).

As discussed throughout this review, further knowledge on the neurotoxic mechanisms mediated by J2 prostaglandins and their contribution to the progression and longevity/resolution of the inflammatory response are needed to develop novel and more effective neuroprotective therapeutic strategies to attenuate inflammation. J2 prostaglandins represent attractive therapeutic targets because of their important roles in the development and resolution of inflammation. In particular, J2 prostaglandindependent therapeutics should target mechanisms of action that include receptor activation and Michael addition.

### **RECEPTOR MODULATION (TABLE 3)**

It is critical to fully understand the dual roles of these prostaglandins as pro- and anti-inflammatory agents. The opposing effects of PGD2/J2 prostaglandins on inflammation are reflected by the responses induced by activation of the DP1 or DP2 receptors, for which a range of antagonists are undergoing clinical evaluation (Pettipher et al., 2007; Sandig et al., 2007; Ricciotti and FitzGerald, 2011). Some of these potential drugs can reach the cerebrospinal fluid upon oral administration (Pettipher et al., 2007). While DP1 agonists seem to alleviate brain damage upon stroke (Ahmad et al., 2010), DP1 antagonists are being proposed to treat disorders such as ALS (de Boer et al., 2014) and Krabbe disease (Mohri et al., 2006). DP2 antagonists are highly relevant to treating allergies and asthma (Norman, 2014). The effective anti-inflammatory properties of these drugs may be relevant to blocking inflammation


**Table 3 | Potential J2 prostaglandin therapeutic targets.**

associated with neurodegeneration and pain (Jones et al., 2009).

PPARγ agonists seem to protect neurons not only following acute CNS injury including stroke, spinal cord injury and TBI after which massive inflammation plays a detrimental role, but also in neurodegenerative conditions including multiple sclerosis (Diab et al., 2002; Chaudhuri, 2013), AD (Combs et al., 2000) and PD (Kapadia et al., 2008). While 15d-PGJ2 is thought to be the endogenous ligand for PPARγ, the thiazolidinediones (TZDs) are used as potent exogenous agonists exerting their neuroprotective effects via prevention of microglial activation, inflammatory cytokine and chemokine expression, and promoting the antioxidant mechanisms in the injured CNS (Kapadia et al., 2008). More recent evidence suggests that the nuclear receptor (NR) superfamily of transcription factors including Nuclear receptorrelated factor1 (Nurr1), PPARs, retinoic acid and glucocorticoid receptors show promise as therapeutic targets for PD. Since they are known to regulate an array of inflammatory mediators, it is postulated that modulating Nurr1 expression or NR receptor activation, including PPARs, via agonists would protect against dopaminergic neuronal death induced by inflammation (Nolan et al., 2013).

Overall, the effectiveness of J2 prostaglandin receptor agonists/antagonists on the treatment of neurodegenerative conditions needs to be carefully investigated as these receptors may act sequentially to initiate and sustain disease states, and/or play complementary roles. Potentially, a combination of these receptor agonists/antagonists could prove to be a very promising therapeutic approach (Jones et al., 2009). Whether these drugs can be applied to preventing chronic long-term neuroinflammation remains to be explored.

### **MICHAEL ADDITION (TABLE 3)**

The best characterized mechanism of action of J2 prostaglandins is the covalent modification of proteins at cysteine residues through Michael addition, which is attributed to their electrophilic nature (Oeste and Perez-Sala, 2014). In this regard, proteomic approaches used in the past few years have provided much needed knowledge about the pharmacological actions and signaling mechanisms of J2 prostaglandins (Oeste and Perez-Sala, 2014). Detailed investigation of the protein targets directly affected by these lipid mediators of inflammation, is critical to the development of more specific and effective therapeutic approached against the deleterious effects of neuroinflammation. Identification of these protein targets will provide important clues on the pathways modulated by J2 prostaglandins and the mechanisms underlying their beneficial or deleterious effects.

As we discussed above, two of these pathways that are affected by J2 prostaglandins and that are highly relevant to neurodegeneration are the UPP and mitochondrial function. We recently investigated *in vitro* (Metcalfe et al., 2012) and *in vivo* (Shivers et al., 2014) a potential therapeutic approach to overcome J2 prostaglandin neurotoxicity, based on elevating intracellular cAMP with PACAP27 (pituitary adenylate cyclaseactivating polypeptide). PACAP27 is a potent neuroprotective lipophilic peptide in different models of neuronal injury such as stroke, PD, HD, TBI, retinal degeneration, and others, where it exhibits anti-apoptotic, anti-inflammatory and anti-oxidant effects (Reglodi et al., 2004; Atlasz et al., 2010; Ohtaki et al., 2010; Dejda et al., 2011; Mao et al., 2012). While we confirmed the anti-apoptotic effects of PACAP27 in our studies, it was clear that PACAP27 as tested and by itself was not sufficient to overcome all of the neurodamaging effects of PGJ2 (Metcalfe et al., 2012; Shivers et al., 2014). Ideal therapeutic interventions against J2 prostaglandins may require a combinatorial approach to effectively prevent the pleiotropic effects of these highly reactive endogenous mediators of inflammation.

In contrast, negatively charged cyclopentenone prostaglandinbased liposomes (LipoCardium) were developed to specifically deliver prostaglandins, in this case PGA2, to injured arterial wall cells of atherosclerotic mice (Homem de Bittencourt et al., 2007). Anti-inflammatory, anti-proliferative, anti-cholesterogenic and cytoprotective effects were obtained with LipoCardium. This strategy opens up new avenues for specific prostaglandin delivery to humans, including a chronic slow delivery.

In conclusion, much remains to be discovered about the biology of the J2 prostaglandins, to prevent their neurotoxic effects and possibly "trick" ongoing inflammation into resolution. Further investigations may provide important clues "to bring us into a new era of inflammation research, which, if approached with creativity and persistence, might provide numerous benefits for those suffering from inflammation-mediated diseases" (Gilroy, 2010).

# **ACKNOWLEDGMENTS**

Please note that this review is not intended to be comprehensive and we apologize to the authors whose work is not mentioned. We would like to thank Mr. Omer Goodovich for the illustrations. This work was supported by National Institutes of Health (NIH) [CTSC GRANT UL1-RR024996 (pilot award to Maria E. Figueiredo-Pereira, co-investigator and another pilot award to Thomas Schmidt-Glenewinkel), NIGMS 1SC3GM086323 to Thomas Schmidt-Glenewinkel, 5R24DA012136-13 (pilot award to Peter Serrano)]; PSC-CUNY Research Award to Patricia Rockwell and Maria E. Figueiredo-Pereira.

### **REFERENCES**


beta-chaperone in human cerebrospinal fluid. *Proc. Natl. Acad. Sci. U.S.A.* 104, 6412–6417. doi: 10.1073/pnas.0701585104


prostaglandin D2-independent effects. *J. Biol. Chem.* 287, 9414–9428. doi: 10.1074/jbc.M111.330662


potentiates TRAIL-induced apoptosis. *Mol. Cancer Ther.* 5, 1827–1835. doi: 10.1158/1535-7163.MCT-06-0023


fragments from the cell into the brain extracellular space. *J. Biol. Chem.* 287, 43108–43115. doi: 10.1074/jbc.M112.404467


**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.

*Received: 03 November 2014; accepted: 17 December 2014; published online: 13 January 2015.*

*Citation: Figueiredo-Pereira ME, Rockwell P, Schmidt-Glenewinkel T and Serrano P (2015) Neuroinflammation and J2 prostaglandins: linking impairment of the ubiquitin-proteasome pathway and mitochondria to neurodegeneration. Front. Mol. Neurosci. 7:104. doi: 10.3389/fnmol.2014.00104*

*This article was submitted to the journal Frontiers in Molecular Neuroscience.*

*Copyright © 2015 Figueiredo-Pereira, Rockwell, Schmidt-Glenewinkel and Serrano. 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.*

# Roles of the ubiquitin proteasome system in the effects of drugs of abuse

# *Nicolas Massaly1,2,3†, Bernard Francès 1,3 and Lionel Moulédous 2,3\**

*<sup>1</sup> Centre de Recherches sur la Cognition Animale, Centre National de la Recherche Scientifique UMR 5169, Toulouse, France*

*<sup>2</sup> Institut de Pharmacologie et de Biologie Structurale, Centre National de la Recherche Scientifique UMR 5089, Toulouse, France*

*<sup>3</sup> Université Paul Sabatier Toulouse III, Toulouse, France*

### *Edited by:*

*Ashok Hegde, Wake Forest School of Medicine, USA*

### *Reviewed by:*

*Kasia Radwanska, Nencki Institute, Poland Marina Wolf, Rosalind Franklin University of Medicine and Science, USA*

### *\*Correspondence:*

*Lionel Moulédous, Institut de Pharmacologie et de Biologie Structurale, Centre National de la Recherche Scientifique UMR 5089, 205 route de Narbonne, 31077 Toulouse 4, France e-mail: lionel.mouledous@ipbs.fr*

### *†Present address:*

*Nicolas Massaly, Department of Anesthesiology, Columbia University Medical Center, New York, NY, USA*

**THE UBIQUITIN PROTEASOME SYSTEM (UPS)**

The role of protein turnover mediated by the ubiquitin proteasome system (UPS) in neuronal plasticity and memory has been studied for about two decades. Here we will only briefly summarize the basic functioning of this system that has been described in more detail in several reviews (Ciechanover, 2005; Patrick, 2006; Hegde, 2010; Mabb and Ehlers, 2010; Bingol and Sheng, 2011). The UPS controls the degradation of misfolded newly synthesized proteins as well as the turnover of specific target proteins. Its function can be described as a two-step process: the tagging of target proteins and their degradation. Ubiquitin molecules can be attached one to another and form a poly-ubiquitin chain which acts as a specific tag to direct proteins to proteasomedependent degradation (**Figure 1A**). This enzymatic linkage is dependent on the activity of three types of enzymes: Ubiquitinactivating enzymes (E1), Ubiquitin-conjugating enzymes (E2), and Ubiquitin ligases (E3). E1 enzymes form a thioester bond with a ubiquitin molecule to activate it. The combined action of E2 and E3 enzymes then permits its linkage to a specific target protein. E3 enzymes mark the proteins that have to be degraded with a poly-ubiquitin chain (linked through Lysine 48 residues) but can also mediate mono- or other types of polyubiquinitation to affect different processes such as protein trafficking and kinase activation (see Bingol and Sheng, 2011 for a more detailed description). Another important class of enzymes is also involved in the regulation of poly-ubiquitination and UPS activity: the desubiquitinating enzymes (DUBs). They oppose the action of E3 ligases by removing ubiquitin. Thus, E1, E2, E3, and

Because of its ability to regulate the abundance of selected proteins the ubiquitin proteasome system (UPS) plays an important role in neuronal and synaptic plasticity. As a result various stages of learning and memory depend on UPS activity. Drug addiction, another phenomenon that relies on neuroplasticity, shares molecular substrates with memory processes. However, the necessity of proteasome-dependent protein degradation for the development of addiction has been poorly studied. Here we first review evidences from the literature that drugs of abuse regulate the expression and activity of the UPS system in the brain. We then provide a list of proteins which have been shown to be targeted to the proteasome following drug treatment and could thus be involved in neuronal adaptations underlying behaviors associated with drug use and abuse. Finally we describe the few studies that addressed the need for UPS-dependent protein degradation in animal models of addiction-related behaviors.

**Keywords: addiction, drug abuse, nicotine, opioid, plasticity, proteasome, stimulants, ubiquitin**

DUB enzymes tightly regulate the addressing of proteins to the proteasome. The second step of UPS function relies on the proteolytic activity of the 26S proteasome. This complex of proteins can be sub-divided into two components: the 20S proteasome which is the catalytic core where degradation takes place and the 19S proteasome which acts as a regulatory complex. The 20S proteasome is made of two external and two internal rings of proteins. External rings are composed of seven alpha type proteins (numbered from α1 to α7). They are involved in the regulation of the access of tagged proteins to the inner core (internal rings) of the 20S proteasome. The internal rings are composed of seven beta type proteins (numbered from β1 to β7) which are responsible for the catalytic activity of the proteasome. Three subunits are directly involved in degradation processes: the β1, β2, and β5 subunits which are responsible for caspase-like, trypsin-like and chymotrypsin-like activity respectively. The other types of β subunits have been proposed to play a structural role in the complex and to be involved in the binding of targeted proteins during their degradation by β1, β2, and β5 subunits. Different complexes can be associated with the 20S proteasome, the 19S proteasome being the most frequent. The 26S proteasome possesses two 19S proteasome regulatory complexes located at each extremity of the 20S core. They can also be divided in two distinct subparts: the lid and the base. The lid is composed by 9 regulatory particle non-ATPase (Rpn) proteins and possesses two main roles: the recognition of poly-ubiquitinated proteins and the removal of ubiquitin from the targeted proteins. The base is composed of 10 proteins with or without ATPase activity, Regulatory

particle ATPases (Rpt) and Rpn proteins respectively. It is physically connected to the proteasome 20S and is involved in the unfolding of proteins and the regulation of their entry into the catalytic core. Thanks to the combined actions of E1, E2, E3, and DUB enzymes and the 19S proteasome complex, the UPS can finely control the identity of the proteins to be targeted and degraded by the catalytic core located in the inner part of the 20S proteasome.

# **THE UPS IN NEURONAL PLASTICITY AND MEMORY**

Changes in neuronal activity can result in the regulation of many proteins by the UPS. A descriptive study showed that increases or decreases in the activity of cultured hippocampal neurons produce UPS-dependent changes in the amount of several proteins in post-synaptic densities (PSD), including proteins involved in PSD morphology, cytoskeleton organization and scaffolding of signaling complexes (Ehlers, 2003). This result suggests a close relationship between synaptic plasticity and protein degradation. Indeed, it has been reported that protein degradation by the UPS contributes to the formation and maintenance of longterm potentiation (LTP) and long-term depression (LTD). The first study reporting this involvement was conducted in Aplysia during induction of long-term facilitation (LTF) (Hegde et al., 1993). It demonstrated that regulatory subunits of the Protein kinase A (PKA) are targeted to the proteasome for degradation allowing prolonged action of PKA and Aplysia behavioral sensitization (Hegde et al., 1993). Later on, studies in rodents have shown that blocking the UPS in the hippocampus can alter N-methyl-D-aspartate (NMDA)- and/or metabotropic glutamate receptor (mGluR)-dependent LTD or LTP (Colledge et al., 2003; Citri et al., 2009). This deleterious effect of UPS blockade on long term changes in neurons has been suggested to be due to an alteration in the balance between protein synthesis and degradation (Fonseca et al., 2006). Indeed the authors showed that the deleterious effects produced by inhibiting either protein synthesis or degradation on LTP can be reversed by inhibition of the two processes at the same time. In addition to synaptic proteins the UPS is also involved in the regulation of the activity of transcription factors, thus revealing a close relationship between protein synthesis and proteasome action. For example IκB and CREM (cAMP-responsive element modulator), repressors of the transcription factors NF-κB and CREB (cAMP response element binding) respectively, can be ubiquitinated and degraded by the UPS (Woo et al., 2010; Liu and Chen, 2011). In that sense the UPS clearly plays a major role in the regulation of protein turnover implicated in neuronal plasticity acting directly through the degradation of some proteins and indirectly through the modulation of transcriptional activity and protein synthesis.

Unsurprisingly considering its role in neuronal plasticity, a strong involvement of UPS function has also been observed during learning and memory processes. These results are reported in detail in a recent review (Jarome and Helmstetter, 2014). More than 10 years ago a first study demonstrated the role of the proteasome in the dorsal hippocampus during the acquisition phase of inhibitory avoidance memory (Lopez-Salon et al., 2001). In this work the authors reported an increase in the rate of protein poly-ubiquitination in the hippocampus during training. They also found that the repressor of NF-κB, IκB, was present in this poly-ubiquitinated protein pool showing an involvement of the UPS in transcription factor activation. More recent studies confirmed the necessity of protein degradation in the hippocampus during consolidation and reconsolidation processes in rodents in spatial memory and fear conditioning (Artinian et al., 2008; Lee et al., 2008). The hippocampus is not the only region were proteasome activity is required for the creation and maintenance of memory. The involvement of protein degradation in both the prefrontal cortex and the amygdala during fear learning has also been demonstrated (Jarome et al., 2011; Reis et al., 2013). Proteasome action also appears to be necessary in both the insular cortex and the amygdala for aversive taste learning (Rodriguez-Ortiz et al., 2011).

The precise mechanisms underlying the involvement of the proteasome in memory are just beginning to be discovered but it is now clearly established that, in addition to protein synthesis, neuronal protein degradation by the UPS is a mandatory process to create, store and maintain memories and in that sense participates to adaptive behaviors of mammals. Since drug addiction shares common mechanisms with memory processes (Hyman et al., 2006; Milton and Everitt, 2012) it is important to question the role of the UPS in the long term effects of drugs of abuse such as opioids, stimulants, ethanol, nicotine and cannabinoids.

# **DRUGS OF ABUSE REGULATE THE UPS**

In recent years, many transcriptomic and proteomic studies have described the global effects of treatments with drugs of abuse on the brain, or on neuronal or glial cell lines. Proteasome subunits or proteins involved in the ubiquitination process are often found to be regulated in these studies (**Table 1**, **Figure 1B**). In the case of opioids, it was shown in a cellular model that a prolonged 72 h morphine treatment modifies the abundance of two proteasome subunits (α3 and β6) (Neasta et al., 2006). *In vivo*, intra-cerebro-ventricular (icv) infusion of morphine for 72 h results in an increase in the tyrosine-phosphorylated form of the β4 subunit in the rat frontal cerebral cortex (Kim et al., 2005). A longer intermittent treatment (2 weeks) produces a decrease in the amount of the DUB Ubiquitin C-terminal hydrolase L-1 in the nucleus accumbens (Nacc) (Li et al., 2006). 4 days after morphine withdrawal, the quantity of this enzyme, as well as that of the α3 subunit of the proteasome, increases in rat dorsal root ganglia (Li et al., 2009). Similarly, chronic treatment (90 days) and drug withdrawal have been shown to have opposite effects on the amount of α5 subunit in the Nacc of rhesus monkeys (Bu et al., 2012). The levels of Ubiquitin-conjugating enzyme E2 and of Ubiquitin C-terminal hydrolase L-3 are also modulated in this model. Finally, in a morphine-induced conditioned place preference (CPP) paradigm which tests the rewarding properties of the drug, both development, extinction and re-instatement are accompanied by a down-regulation of several DUBs and α and β subunits (Lin et al., 2011).

Changes in expression of proteins of the UPS are not specific to opioid treatment. The amount of Ubiquitin C-terminal hydrolase L-1 is increased and that of RN3 (β7 catalytic subunit) is decreased in the striatum of rats acutely treated with methamphetamine while repeated injections induce an increase in Ubiquitin C-terminal hydrolase L-1 and a decrease in several proteasome subunits in the frontal cortex (Iwazaki et al., 2006; Faure et al., 2009; Kobeissy et al., 2009). The development of cocaine CPP comes with an increase in the expression of the Ubiquitinconjugating enzyme E2N, of the catalytic α2 subunit and of the 26S proteasome regulatory subunit p45/SUG (Guan and Guan, 2013). Moreover, mouse cortical neurons grown in the presence of ethanol for 5 days show decreased amounts of mRNA coding for several ubiquitin-conjugating enzymes, as well as catalytic and regulatory subunits of the proteasome (Gutala et al., 2004) while the quantities of Ubiquitin C-terminal hydrolase L-1 decrease and


**Table 1 | UPS-related molecular and cellular consequences of the treatment with drugs of abuse.**

### **Table 1 | Continued**


that of ubiquitin and Ubiquitin-conjugating enzyme 7 increase in the white matter of the brain of alcoholic patients (Alexander-Kaufman et al., 2006; Kashem et al., 2007). Again at the mRNA level, chronic treatment of rats with nicotine produces an elevated expression of ubiquitin-conjugating enzymes, proteasome regulatory and catalytic subunits and DUBs in the prefrontal cortex whereas their level is decreased in the medial basal hypothalamus (Kane et al., 2004). Variations can be of opposite direction within a single cell type with for example an up-regulation of the E2 ubiquitin-conjugating enzyme E2J2 associated with the down-regulation of a proteasome regulatory subunit and two DUBs in mouse dopaminergic neurons chronically treated with nicotine (Henley et al., 2013). Finally an up-regulation of the DUB Ubiquitin specific protease 3 was observed in human astrocytes exposed for 48 h to -9-THC (tetra-hydro-cannabinol) (Bindukumar et al., 2008).

All drugs of abuse can thus affect the expression and abundance of key UPS proteins. However, the data reported above are only descriptive. Moreover, UPS components are affected differently depending on the drug type, its method of administration, the duration of the treatment and the cell type or brain region considered (**Table 1**). Complementary studies have also found that drugs of abuse modify the activity of the UPS in parallel with changes in the expression of its various components. Indeed morphine was demonstrated to inhibit the activity of the 20S proteasome in human neuroblastoma cells, with neuroprotective consequences (Rambhia et al., 2005). On the contrary, PKC-dependent inhibition of the UPS was linked to the autophagy-mediated toxicity of methamphetamine in dopaminergic neurons (Lin et al., 2012). In addition it has been proposed that the higher toxicity of methamphetamine compared to cocaine was due to its long inhibitory effect on proteasome activity (Dietrich et al., 2005). Finally, a recent study demonstrated that chronic ethanol induces toxicity in mice through a Toll-like receptor 4-dependent impairment of the UPS (Pla et al., 2014). This deleterious effect could depend on a shift in proteasome composition from classical to immunoproteasome subunits, a phenomenon known to play a role in the neurotoxicity observed in neurodegenerative diseases, and on an increase in chymotrypsin-like and trypsin-like activities (Pla et al., 2014).

So far we have only described global changes in the composition and/or activity of the UPS associated with beneficial or deleterious effects on the functioning or survival of neurons. However, more subtle and finely regulated mechanisms need to be considered to explain the plasticity phenomena underlying the development of addiction-related behaviors. These mechanisms do not necessarily imply a global modification of UPS activity but rather the degradation of specific targets in precise cellular locations. Unfortunately fewer studies have focused more specifically on synaptic and/or signaling proteins degraded by the UPS in relation with the administration of drugs of abuse.

# **UPS TARGETS INVOLVED IN DRUG-INDUCED PLASTICITY**

Drugs of abuse target receptors, channels and transporters located in the plasma membrane. However, membrane proteins are not typical proteasome substrates but are rather degraded in lysosomes. Proteasome-mediated degradation only occurs for misfolded membrane proteins through the ERAD (Endoplasmicreticulum-associated protein degradation) pathway before their export to the plasma membrane (Christianson and Ye, 2014) but ubiquitination can also modulate the degradation of membrane proteins after endocytosis by influencing their sorting to lysosomes through the ESCRT (endosomal sorting complexes required for transport) system (Macgurn et al., 2012). This phenomenon involves HECT (Homologous to the E6-AP Carboxyl Terminus) E3 ligases and will not be discussed in detail here since it is proteasome-independent. However, it is worth mentioning that, since proteasome inhibitors cause the accumulation of ubiquitinated proteins and thus reduce the available pool of free ubiquitin, they can affect indirectly ubiquitin-dependent proteasome-independent processes such as sorting to lysosome (Mimnaugh et al., 1997).

Mu opioid (MOP) receptors play a role in the rewarding and reinforcing properties of opioids but also of most non-opioid abused drugs (Le Merrer et al., 2009). They are ubiquitinated following activation. Proteasome inhibitors increase their basal abundance and decrease agonist-induced down-regulation in recombinant cells (Chaturvedi et al., 2001). The increase in basal receptor expression following proteasome inhibition could be due to the blocking of the ERAD pathway whereas the reduction in agonist-induced down-regulation could result from the indirect effect of proteasome inhibitors on ubiquitin-dependent sorting to lysosomes. Indeed it was recently shown that the ubiquitination of the first intracellular loop of the MOP receptor facilitates its lysosomial degradation by promoting its transfer to intralumenal vesicles downstream of the ESCRT system (Hislop et al., 2011). It was also proposed that different translational forms of the receptor showed different sensitivities to the ERAD pathway because of additional ubiquitination sites (Song et al., 2009). Besides the MOP receptor, the nicotinic receptor is another example of drug target which has been shown to be regulated by the UPS. Here again the ERAD pathway seems to be involved and the subunit composition of pentameric nicotinic receptors has an influence on their sensitivity to this pathway (Govind et al., 2012; Mazzo et al., 2013).

UPS-dependent changes have also been identified downstream of receptor activation. In SH-SY5Y human neuroblastoma cells, long-term morphine treatment induces proteasome-dependent degradation of the Gβ subunit of heterotrimeric G proteins (Mouledous et al., 2005). This degradation could reduce G protein-coupled receptor signaling and restore the activity of effectors normally inhibited by Gβ subunits such as adenylyl cyclase. In the same cells, opioids have also been shown to induce the ubiquitination and degradation of regulator of G protein signaling 4 (RGS4), a protein that controls the duration of G protein signaling by acting as a GTPase accelerating protein (GAP) (Wang and Traynor, 2011). RGS4 is an unstable protein known to be subjected to the N-end rule pathway, a particular type of regulation based on the removal of the N-terminal methionine and the arginylation of the resulting N-terminal cysteine to promote ubiquitination and proteasome degradation. Its down-regulation affects the signaling of other G proteincoupled receptors present in the same cell. Overall, by regulating the abundance of several signaling molecules sensitive to opioid treatment, the UPS participates in the homeostatic processes involved in the development of opioid tolerance and dependence (Bailey and Connor, 2005; Christie, 2008). In mice, chronic morphine treatment induces a decrease in the total amount of ubiquitinated proteins in the striatum. In parallel, the heat-shock protein HSP70 was shown to be overexpressed, probably because of a lower ubiquitination rate (Yang et al., 2014). The higher expression of this protein could participate in the behavioral sensitization induced by morphine (Qin et al., 2013). However, the HSP70 cellular effect mediating this process is currently unknown. Besides changes in signaling, long-term drug treatment is known to affect neuronal structural plasticity (Robinson and Kolb, 2004). Small G proteins can influence cellular architecture and it is thus significant to note that, in neuro-2A cells, cannabinoids induce neurite outgrowth by activating the small G protein Rap1 through the proteasome-dependent degradation of one of its GAP, Rap1GAPII (Jordan et al., 2005).

The neuronal adaptations described so far are homeostatic non-associative phenomena. They result from the direct activation of the drug target and its downstream signaling and are not sufficient to explain the associative processes involved in addiction. Similarly to classical forms of memory, drug addiction involves activity-dependent plasticity at excitatory synapses within neuronal circuits, notably those controlling motivated behaviors (Kauer and Malenka, 2007; Russo et al., 2010). Following drug administration, the UPS system regulates the abundance of several proteins at the glutamatergic synapse but very few studies have identified these proteasome targets. In the case of opioids, an increase in ubiquitinated proteins in the synaptosomal fraction of the mouse Nacc was observed following morphine conditioning (Massaly et al., 2013) but the identity of the UPS targeted proteins was not reported. So far, the only glutamate-related proteins shown to be degraded by the proteasome following chronic morphine treatment are glutamate transporters EAAC1, GLSAT, and GLT-1 but these changes were observed in the rat spinal cord and were related to analgesic tolerance rather than addiction (Yang et al., 2008a,b). Concerning nicotine, one study addressed the effect of its intraperitoneal injection in mice on the expression of synaptic proteins (Rezvani et al., 2007). It suggested that the observed increase in the amount of GluR1 α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor subunits, NR2A NMDA receptor subunits, metabotropic receptor mGluR1α, and PSD95 (a scaffolding protein of the PSD), but also the decrease in the quantity of Shank (another scaffolding protein), were due to an inhibition of proteasome activity. The exposure to stimulants was also reported to have UPS-dependent synaptic effects. NAC1 (nucleus accumbens-associated protein 1), the product of an immediate early gene up-regulated by psychostimulants, takes part in the recruitment of the proteasome to the PSD by interacting with Cullin-based E3 ubiquitin ligases and the 19S ATPase subunit Mov34 (Shen et al., 2007). The UPS could also contribute to the phenomenon of synaptic scaling that is observed in the Nacc following cocaine withdrawal and results from the addition of AMPA receptors to the synapse (Sun and Wolf, 2009). UPS-dependent synaptic changes in the striatum have been shown to contribute to the behavioral sensitization induced by repeated amphetamine injections in rats. Contrarily to acute nicotine injection, chronic amphetamine treatment produced a decrease in NMDA receptor subunits and anchoring proteins in the PSD. Only Shank and GKAP (guanylate-kinaseassociated protein) were ubiquitinated and it was proposed that the degradation of these important anchoring proteins indirectly leads to a loss of PSD95 and NR1 and NR2B subunits of the NMDA receptor at the synapse (Mao et al., 2009). Finally, the retrieval of cocaine place preference in rats has been shown to result in an increase in protein poly-ubiquitination in the core of the Nacc, and in particular in the degradation of NSF (Nethylmaleimide-sensitive fusion), a protein of the PSD involved in synaptic plasticity (Ren et al., 2013). In conclusion, the UPS is involved in the synaptic plasticity that underlies some of the behavioral adaptations to drug exposure. However, the molecular details are still poorly known and will probably depend on the drug type, the location of the synapse in the neuronal circuit and the phase of the addiction process under study. It is thus critical to implement studies to establish direct causal relationship between the degradation of neuronal proteins by the UPS in a particular brain region and a given addiction-related behavior.

# **UPS AND ADDICTION-RELATED BEHAVIORS**

Few studies have assessed the role of protein degradation by the proteasome in drug-related behaviors. Recently we found that UPS function in the Nacc is crucial in several types of opioid-induced behaviors (Massaly et al., 2013). Our goal was to assess the role of protein degradation by the proteasome in the development of drug seeking behaviors and the motivation to obtain opioids. By using proteasome inhibitors our study demonstrated a clear role of the UPS in the Nacc during acquisition of non-operant tasks, namely CPP and context-dependent locomotor sensitization in mice. Intra-Nacc proteasome inhibitors also prevented the acquisition of operant tasks in mice (intra-VTA self-administration) and rats (intra-venous self-administration). However, these behavioral paradigms do not enable us to clearly discriminate between an effect of proteasome inhibitors on drug-induced memory and on non-associative drug effects. The behavioral sensitization procedure can be implemented in a context-dependent or -independent way (Valjent et al., 2006). **Figures 2A–C** shows the comparison between the effects of proteasome inhibition in a context-dependent (Massaly et al., 2013) and a context-independent paradigm. In both experiments mice were submitted to 5 daily morphine injections followed by a 2 day withdrawal period. On day 1 basal horizontal activity was measured during 1 h directly after i.p. morphine injection (**Figures 2B,C**, empty bars). On days 2, 3, 4, and 5 mice received intra-Nacc injection of DMSO or the proteasome inhibitor lactacystin 1 h prior to i.p. morphine treatment and were then directly placed in their home cage to prevent association between drug and cues present in activity boxes (context-independent, **Figure 2C**) or in activity boxes (context-dependent, **Figure 2B**). Three days after the last opioid treatment, animals were challenged with an i.p. morphine injection and locomotor activity was measured during 1 h to assess behavioral sensitization. Control groups showed a significant locomotor sensitization on day 8 (**Figures 1C**, **2B**, DMSO group). Lactacystin injections prevented behavioral sensitization only in the context-dependent procedure (**Figure 2B**). It thus appears that the UPS in the Nacc is not necessary for the development of behavioral sensitization when this adaptation is context-independent (**Figures 2A–C**, refer to (Massaly et al., 2013) for details). This result, together with the fact that intra-Nacc injection of proteasome inhibitors prevents the consolidation of morphine place preference, strongly suggests that UPS activity in the Nacc is involved in drug-context association rather than non-associative motivational effects of opioids. However, this distinction might not be true for each type of drug of abuse and in particular for stimulants. Indeed, proteasome inhibitors have been shown to inhibit the development of behavioral sensitization to amphetamine after intra-Nacc injection in rats in a context-independent paradigm (Mao et al., 2009).

### **FIGURE 2 | Continued**

after morphine injection before (day 1; empty bars) and after conditioning (day 8; black bars). Two-Way ANOVA followed by Bonferroni *post-hoc* tests: n.s., non-significant; ∗∗∗*p* < 0.001. **(D)** Schematic representation of the protocol used for assessing the role of the UPS in reconsolidation. **(E)** Intra-Nacc bilateral injection of lactacystin 1 h before a drug-context re-exposure abolishes drug-induced place preference when tested 24 h

Consolidation is the process by which stable long-term memories are formed following a learning session. Reconsolidation refers to the fact that the memory trace can return to an active labile state after recall (Alberini and Ledoux, 2013). Interfering with this process could thus offer a way to erase drug-context association with important therapeutic consequences. If lactacystin is injected in the Nacc before a new drug-compartment association performed 1 week after the last morphine place preference conditioning (Massaly et al., 2013, **Figure 2D**), mice do not display any place preference for the morphine compartment on the following day, contrarily to the DMSO control group (**Figure 2E**). The UPS in the Nacc thus seems to be involved in reconsolidation of place preference induced by morphine. Our conclusions concerning the involvement of the UPS in consolidation and reconsolidation of drug-associated memories are only partly consistent with those of the only other study examining this question (Ren et al., 2013). Ren et al. found that inhibiting UPS activity in the Nacc interfered with drug-reward memories using cocaine CPP in rats. However, in their model, proteasome inhibitors blocked CPP extinction when injected following each extinction session but were not efficient on memory consolidation during the learning phase. Moreover, they did not interfere directly with memory reconsolidation following a reactivation session but counteracted the inhibitory effect of protein synthesis inhibition on this process. These apparent discrepancies are likely due to many differences in experimental conditions between the two studies: animal model (rats vs. mice), drug type (cocaine vs. morphine), conditioning procedure (4 drug/saline injections over 8 days vs. 3 injections over 3 days), timing of injection of proteasome inhibitors (before vs. after memory reactivation), method of induction of memory reconsolidation (reactivation in the absence vs. in the presence of drug). Taken together, the two studies confirm that the UPS in the Nacc plays a key role in drug-reward memories although further work is needed to fully understand under which exact circumstances it is recruited.

In conclusion, even if some discrepancies can be observed between studies depending on the model under investigation, it appears that the UPS plays a role in drug-related behaviors and the adaptation to the exposure to drugs of abuse. Future work will certainly bring us new evidence to complete the picture of the involvement of proteasome-dependent protein degradation in the brain during addiction.

### **FUTURE DIRECTIONS**

It is clear from the studies reviewed here that the UPS plays an essential role in neuronal plasticity associated with longterm exposure to drugs of abuse. However, the UPS is involved in so many cellular processes that we are still a long way from understanding its specific contribution to each aspect of after this new association (*n* = 6) whereas DMSO treated-animals still express place preference (*n* = 6). Data are expressed as percentage of time spent in the drug-associated compartment ± SEM during pre-conditioning tests (empty bars) and post-conditioning tests (filled bars). Two-Way ANOVA followed by Bonferroni *post-hoc* tests: n.s., non-significant; ∗*p* < 0.05. See Massaly et al. (2013) for details about behavioral procedures.

drug use and abuse. Several outstanding questions need to be addressed to achieve this goal. What are the most significant cellular targets of the UPS during neuronal plasticity associated with drug addiction? So far, studies have focused on synaptic and signaling proteins but other types of proteins, such as for example transcription factors, are mandatory for enduring neuronal plasticity and could be regulated directly or indirectly by UPS-dependent processes (Carle et al., 2007; Dong et al., 2008). Where does the regulation take place? All the behavioral studies reported here focused on the Nacc but different UPS-dependent changes will probably occur depending on the brain region. Also alterations in protein content will vary according to the type of plasticity occurring in each individual neuron or synapse. When does the protein need to be degraded? Different UPS targets will be concerned depending on the phase of the addiction process. For example changes appearing along the course of drug administration will probably differ from those resulting from withdrawal. Why is a UPSdependent regulation occurring? In particular it will be important to distinguish homeostatic regulations involved in nonassociative tolerance or sensitization from more integrated plasticity phenomena responsible for associative context-dependent aspects of addiction. How are the proteins targeted to the proteasome? Identifying the mechanism by which each important UPS substrate is targeted to the proteasome (post-translational modification preceding ubiquitination, type of E3 ligase. . . ) will offer opportunities to control addiction-related processes more specifically than by blocking the catalytic activity of the proteasome. Many technical limitations may prevent us from answering fully to these questions but the link between the control of protein expression through UPS-dependent degradation and plastic changes involved in addiction clearly deserves further investigation.

# **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.

*Received: 29 October 2014; accepted: 08 December 2014; published online: 06 January 2015.*

*Citation: Massaly N, Francès B and Moulédous L (2015) Roles of the ubiquitin proteasome system in the effects of drugs of abuse. Front. Mol. Neurosci. 7:99. doi: 10.3389/ fnmol.2014.00099*

*This article was submitted to the journal Frontiers in Molecular Neuroscience.*

*Copyright © 2015 Massaly, Francès and Moulédous. 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.*

# Ubiquitin-dependent proteolysis in yeast cells expressing neurotoxic proteins

Ralf J. Braun<sup>∗</sup>

Institut für Zellbiologie, Universität Bayreuth, Bayreuth, Germany

Critically impaired protein degradation is discussed to contribute to neurodegenerative disorders, including Parkinson's, Huntington's, Alzheimer's, and motor neuron diseases. Misfolded, aggregated, or surplus proteins are efficiently degraded via distinct protein degradation pathways, including the ubiquitin-proteasome system, autophagy, and vesicular trafficking. These pathways are regulated by covalent modification of target proteins with the small protein ubiquitin and are evolutionary highly conserved from humans to yeast. The yeast Saccharomyces cerevisiae is an established model for deciphering mechanisms of protein degradation, and for the elucidation of pathways underlying programmed cell death. The expression of human neurotoxic proteins triggers cell death in yeast, with neurotoxic protein-specific differences. Therefore, yeast cell death models are suitable for analyzing the role of protein degradation pathways in modulating cell death upon expression of disease-causing proteins. This review summarizes which protein degradation pathways are affected in these yeast models, and how they are involved in the execution of cell death. I will discuss to which extent this mimics the situation in other neurotoxic models, and how this may contribute to a better understanding of human disorders.

Edited by:

Fred Van Leeuwen, Maastricht University, Netherlands

### Reviewed by:

Nico P. Dantuma, Karolinska Institute, Sweden Zhiqun Tan, University of California, Irvine, USA

### \*Correspondence:

Ralf J. Braun, Institut für Zellbiologie, Universität Bayreuth, 95440 Bayreuth, Germany ralf.braun@uni-bayreuth.de

Received: 20 October 2014 Accepted: 24 February 2015 Published: 12 March 2015

### Citation:

Braun RJ (2015) Ubiquitin-dependent proteolysis in yeast cells expressing neurotoxic proteins. Front. Mol. Neurosci. 8:8. doi: 10.3389/fnmol.2015.00008 Keywords: ubiquitylation, ubiquitin-proteasome system, autophagy, ubiquitin-dependent vesicular trafficking, neurodegeneration, cell death, Saccharomyces cerevisiae

# Introduction

Ubiquitin is a highly conserved protein with 76 amino acids (Weissman et al., 2011; Finley et al., 2012). It is covalently linked to lysine side chains of substrate proteins by the sequential action of the ubiquitin-activating enzyme (E1), ubiquitin-conjugating enzymes (E2), and substrate-specific ubiquitin ligases (E3) (Weissman et al., 2011; Finley et al., 2012). Deubiquitylating enzymes (DUBs) recycle ubiquitin, replenishing the cellular pool of free ubiquitin. The high variety of ubiquitin modifications, including mono- and polyubiquitylations, destines the degradation, localization, and/or function of substrate proteins. Consequently, numerous cellular processes are regulated by ubiquitylation, including protein degradation, cell death control,

**Abbreviations:** AD, Alzheimer's disease; ALS, Amyotrophic lateral sclerosis; DUB, Deubiquitylating enzyme; ERAD, ER-associated degradation; HD, Huntington's disease; IPOD, Insoluble protein deposit/ perivacuolar aggregates; MAD, Mitochondrion-associated degradation; MVB, Multivesicular bodies; PD, Parkinson's disease; polyQ, Proteins with abnormal glutamine expansions; ROS, Reactive oxygen species; UFD, Ubiquitin-fusion degradation pathway; UPR, Unfolded protein response; UPS, Ubiquitin-proteasome system.

and vesicular trafficking (Weissman et al., 2011; Finley et al., 2012). These ubiquitin-dependent processes are highly conserved from yeast to humans (Finley et al., 2012).

Polyubiquitylated proteins are degraded within the proteasome, a cylindrical multiprotein complex with chymotrypsin-, trypsin-, and caspase-like proteolytic activities. This ubiquitindependent degradation of proteins via the proteasome is called the ubiquitin-proteasome system (UPS) (Weissman et al., 2011; Finley et al., 2012). The ubiquitylation of many plasma membrane proteins promotes their targeting into endosomes and multivesicular bodies (MVB) leading to their degradation by multiple proteases in the lysosomes (or vacuoles in yeast) (Finley et al., 2012; MacGurn et al., 2012). This MVB pathway of protein degradation (also called endosomal-lysosomal pathway) is a ubiquitin-controlled vesicle-based protein degradation pathway, which is independent from proteasomes. Ubiquitylation can also be involved in the degradation of substrate proteins via autophagy (Kuang et al., 2013; Lu et al., 2014). Autophagy is a cellular process where proteins, protein aggregates, or organelles are enclosed by a double membrane forming autophagosomes, which eventually fuse with lysosomes (vacuoles) for degradation. One mechanism to ensure specificity during autophagy relies on the ubiquitylation of target proteins or organelles and the consequent use of specific adaptors that connect the ubiquitin system with the autophagy pathway (Kuang et al., 2013; Lu et al., 2014). The UPS, the MVB pathway, and autophagy share some components, such as the AAA-ATPase p97/VCP (or Cdc48 in yeast) (Bug and Meyer, 2012; Dargemont and Ossareh-Nazari, 2012), and the E3 ligase Nedd4 (or Rsp5 in yeast) (MacGurn et al., 2012; Fang et al., 2014; Lu et al., 2014), making these proteins to potential key players that decide by which pathway a protein is to be degraded.

Accumulation of aggregated proteins is a common hallmark of many neurodegenerative disorders, and believed to contribute to neuronal dysfunction (Lansbury and Lashuel, 2006). In Parkinson's disease (PD) cytoplasmic Lewy bodies are protein aggregates mainly comprised by the protein α-synuclein (Uversky, 2007), and nuclear protein aggregates of the polyglutamine protein hungtingtin are typical for Huntington's disease (HD) (Ross and Tabrizi, 2011). In Alzheimer's disease (AD) the hydrophobic peptide β-amyloid is produced in cells but accumulates in extracellular plaques (Laferla et al., 2007). The microtubule-associated protein tau (MAPT), and UBB+<sup>1</sup> , the frameshift variant of human ubiquitin B, is enriched in intracellular inclusions during AD (van Leeuwen et al., 1998; Mandelkow and Mandelkow, 2012). Accumulation of cytoplasmic aggregates of disease-causing proteins, such as TDP-43 or FUS/TLS, occurs during the motor neuron disease amyotrophic lateral sclerosis (ALS) (Andersen and Al-Chalabi, 2011). Most of these disease-associated proteins or aggregates are ubiquitylated. Therefore, the observed accumulation of protein aggregates has been explained by dysfunctional protein degradation pathways, including the UPS and autophagy (Dennissen et al., 2012; Dantuma and Bott, 2014). However, the precise role of ubiquitin-dependent proteolysis and its importance for the progression of the human disorders remains poorly understood.

Yeast is an established model for measuring cytotoxicity and programmed cell death, and for dissecting conserved mechanisms of apoptosis and necrosis (Carmona-Gutierrez et al., 2010). Diverse roles of ubiquitin-dependent protein degradation have been described in distinct yeast cell death scenarios. The ubiquitin-dependent and proteasome-independent routing of misfolded proteins to the MVB pathway for their degradation protects cells from cytotoxicity (Wang et al., 2011). Elevated proteasome capacity extends the replicative life span and fitness of yeast cells, which are more resistant against proteotoxic stress (Kruegel et al., 2011). Decreased proteasome capacity by proteasome inhibition leads to disturbances in the amino acid homeostasis, thereby executing cell death (Suraweera et al., 2012). Impairment of distinct branches of the UPS pathway, including the ER- and the mitochondrion-associated degradation (ERAD/MAD) are sufficient to trigger cell death in yeast, emphasizing their cytoprotective role in the homeostasis of the ER and mitochondria (Braun et al., 2006; Zischka et al., 2006; Heo et al., 2010). However, UPS impairment (due to proteasome inhibition) can also prevent from cell death, e.g., when cell death is triggered by acetic acid and the chemotherapeutic drug cisplatin, respectively (Valenti et al., 2008; Cunha et al., 2013). Thus, ubiquitindependent proteolysis is involved in both the execution of and the prevention from yeast cell death.

In recent years, many yeast models have been established to analyze the influence of human neurotoxic protein expression on yeast cell survival, including models for PD, HD, AD, and ALS (Gitler, 2008; Miller-Fleming et al., 2008; Winderickx et al., 2008; Bharadwaj et al., 2010; Braun et al., 2010; Khurana and Lindquist, 2010; Bastow et al., 2011; Mason and Giorgini, 2011). Here, I summarize how ubiquitin-dependent protein degradation is impaired in yeast cell death models expressing neurotoxic proteins, and which role proteolysis plays in the execution of cell death. Further, I will discuss some similarities between the yeast models expressing neurotoxic proteins, and the animal and cell culture disease models.

# Yeast Models Expressing Neurotoxic Proteins

# Parkinson's Disease (PD)

PD is the most prevalent age-related movement disorder characterized by a progressive loss of dopaminergic neurons in the substantia nigra, leading to the impairment of normal motor function culminating in resting tremor, bradykinesia and rigidity (Lees et al., 2009). In most familiar and sporadic cases, PD is associated with Lewy bodies, i.e., intracellular cytoplasmic aggregates composed of the protein α-synuclein (Uversky, 2007). Missense mutations in the SNCA gene, resulting in the expression of α-synuclein variants (A18T, A29S, A30P, A53T, E46K, H50Q, G51D), as well as duplication and triplication of SNCA, leading to elevated α-synuclein levels, are causative for PD in some familiar forms of the disorder (Fujioka et al., 2014). Numerous α-synuclein disease models expressing wild-type and diseaseassociated variants have been established, including several yeast models.

# In Yeast, α-Synuclein is a Membrane-Associated Protein, which is Degraded Via the UPS, Autophagy, and Potentially the MVB Pathway

When expressed in yeast, α-synuclein binds to vesicles of the secretory pathway, leading to its localization to the plasma membrane (Outeiro and Lindquist, 2003; Dixon et al., 2005; Sharma et al., 2006; Zabrocki et al., 2008). This depends on α-synuclein phosphorylation (Basso et al., 2013; Tenreiro et al., 2014), and can be interrupted by genetic manipulation (e.g., α-synuclein-A30P) (Outeiro and Lindquist, 2003; Dixon et al., 2005; Sharma et al., 2006). Upon high expression levels α-synuclein forms cellular aggregates in a nucleation-dependent manner, which starts at the plasma membrane and eventually leads to cytoplasmic inclusions (Outeiro and Lindquist, 2003). These inclusions co-localize with markers of different vesicles, including Ypt1 (ER-to-Golgi), Ypt31 (late Golgi), Sec4 (secretory vesiclesto-plasma membrane), Ypt6 (endosome-to-Golgi), Vps21 and Ypt52 (early-to-late endosome) and Ypt7 (late endosome-tovacuole) (Gitler et al., 2008). Thus, α-synuclein is an aggregationprone membrane-associated protein.

Although α-synuclein is ubiquitylated in yeast (Outeiro and Lindquist, 2003), to which extent the UPS contributes to its degradation in human cells (Xilouri et al., 2013) or more specifically in yeast (see below), remains an open debate. Treatment of yeast cells expressing α-synuclein with the proteasome inhibitor lactacystin resulted in increased α-synuclein aggregation (Zabrocki et al., 2005; Lee et al., 2008). Consistently, expression of α-synuclein in the yeast strain sen3-1, which harbors a mutation in the gene encoding the regulatory proteasome subunit Rpn2, led to increased steady-state levels of α-synuclein (using untagged α-synuclein) and to increased formation of aggregates (using GFP-tagged α-synuclein) (Sharma et al., 2006). These data suggest that α-synuclein is a UPS substrate in yeast.

In other studies with yeast cells expressing GFP-tagged αsynuclein, α-synuclein aggregate clearance was neither affected by treatment with the proteasome inhibitor MG132 (Petroi et al., 2012; Tenreiro et al., 2014), nor by mutation in the gene encoding the regulatory proteasome subunit Rpt6 (cim3-1) (Petroi et al., 2012). Further, the steady-state levels of α-synuclein were not affected by MG132 treatment (Tenreiro et al., 2014). These data argue against the contribution of the UPS in α-synuclein degradation in yeast. Here, α-synuclein aggregate clearance was dependent on autophagy and vacuolar protease activity. Treatment of yeast cells expressing α-synuclein-GFP with the protease inhibitor phenylmethylsulfonyl fluoride (PMSF), an inhibitor of vacuolar proteases, resulted in a significant reduction of this clearance (Petroi et al., 2012). Similarly, genetic interruption of autophagy (1atg1 or 1atg7) delayed aggregate clearance (Petroi et al., 2012; Tenreiro et al., 2014), and increased the steady-state levels of α-synuclein (1atg7) (Tenreiro et al., 2014). Consistently, inducing autophagy by rapamycin promoted aggregate removal (Zabrocki et al., 2005), confirming that α-synuclein aggregates are degraded via autophagy.

Aggregate clearance was not limited to autophagy, because aggregate clearance still took place in the absence of autophagy and upon very low UPS activity (cim3-1 1atg1 strain), suggesting for an additional cellular clearing mechanism independent from autophagy and the UPS (Petroi et al., 2012). Indeed, the yeast E3 ligase Rsp5, and its homolog Nedd4 in mammalian cells, play critical roles in α-synuclein degradation (Tofaris et al., 2011). In yeast, α-synuclein was identified as an Rsp5 target for ubiquitylation, and upon RSP5 mutation (rsp5-1 strain), both the steadystate level of α-synuclein, as well as the number of cells showing α-synuclein aggregation were increased as compared to wild-type strain (Tofaris et al., 2011). Although Rsp5/Nedd4 is critically involved in many cellular processes, including the MVB pathway, ubiquitin-dependent autophagy and the proteasome-dependent degradation of misfolded proteins (MacGurn et al., 2012; Fang et al., 2014; Lu et al., 2014), there are some line of evidence suggesting that the MVB pathway contributes to α-synuclein degradation (Tofaris et al., 2011). Mammalian Nedd4 promotes the degradation of endogenous α-synuclein by lysosomes, and the targeting of α-synuclein to the lysosomes depends on the endosomal sorting complex (ESCRT) (Tofaris et al., 2011). Due to the high conservation of protein degradation pathways between mammalian cells and yeast, the critical contribution of Rsp5 in αsynuclein degradation in yeast might suggest for a potential role of the MVB pathway in α-synuclein degradation, besides the UPS and autophagy.

# α-Synuclein Expression in Yeast Leads to Impairment of the UPS, and Vesicular Trafficking

Expression of α-synuclein in yeast leads to UPS impairment. The protein composition of the proteasome is altered, concomitant to a moderate decrease in the chymotrypsin-like enzymatic proteasomal activity (in isolated proteasomes), and to a marked delay in the degradation of short-lived proteins (pulse-chase assay) (Chen et al., 2005). Consequently, the cellular levels of polyubiquitylated proteins increased moderately (Chen et al., 2005). The degradation of the UPS model substrate GFPu, in which the C-terminus of this protein comprises a degron, was delayed upon α-synuclein expression in yeast (Outeiro and Lindquist, 2003). The impairment of UPS-dependent protein degradation upon α-synuclein expression appears to be substrate specific. Whereas, the degradation of the cytosolic proteasome substrate Deg1-β-Gal was unaffected, the degradation of the ER luminal substrate CPY\* but not of the ER membrane substrate sec61-2 was severely impaired (Cooper et al., 2006). Thus, α-synuclein expression impairs the UPS and more specifically the degradation of selective substrates of the ER-associated degradation (ERAD) pathway.

α-Synuclein expression also affects vesicular trafficking, including ER to Golgi transport, endocytosis, vesicular recycling back to the plasma membrane, and vacuolar fusion (Cooper et al., 2006; Gitler et al., 2008; Zabrocki et al., 2008; Basso et al., 2013). Since the MVB pathway and autophagy depend both on vesicular fusion processes, it is very likely that α-synuclein expression also affects the protein degradation via these two vesicle-based pathways. Measuring the degradation rates of substrates of autophagy or the MVB pathway upon α-synuclein expression will help to address this issue.

# α-Synuclein Expression in Yeast Triggers Cell Death, which is Modulated by the Activities of the UPS, Autophagy, and Ubiquitin-Dependent Vesicular Trafficking

Yeast cells overexpressing wild-type and disease-associated αsynuclein demonstrated growth deficits and age-dependent loss of clonogenic cell survival paralleled by the emergence of morphological markers of apoptosis and necrosis (Willingham et al., 2003; Flower et al., 2005; Witt and Flower, 2006; Büttner et al., 2008, 2013a,b; Lee et al., 2008; Su et al., 2010). The cellular accumulation of ROS and mitochondrial dysfunction are pivotal for the execution of α-synuclein-triggered cell death. The use of the antioxidant N-acetyl cysteine (NAC) or the use of yeast strains deleted for mitochondrial DNA (ρ 0 strain) protected from ROS and α-synuclein-triggered cell death (Büttner et al., 2008, 2013a). The translocation of the mitochondrial cell death proteins Nuc1 and cytochrome c, into the nucleus and the cytosol, respectively, was observed to be critical for the execution of cell death (Flower et al., 2005; Büttner et al., 2013b). The ER also contributes to cytotoxicity, because α-synuclein expression results in ER stress and in the induction of the unfolded protein response (UPR) (Cooper et al., 2006).

Since α-synuclein has been proposed to be a UPS substrate and since α-synuclein expression resulted in UPS impairment (especially ERAD), it is likely that the UPS is involved in modulating α-synuclein-triggered cytotoxicity. Moderate α-synuclein expression, which is non-toxic for wild-type yeast cells, resulted in severe growth deficits in yeast cells bearing mutations in the 20S proteasomal barrel (pre1-1001, pre2-1001, doa3-1) (Dixon et al., 2005; Sharma et al., 2006) and in the 19S regulatory particle of the 26S proteasome (sen3-1) (Sharma et al., 2006), or treated with the proteasome inhibitor lactacystin (Lee et al., 2008). Consistently, expression of Rpt5, a component of the 19S regulatory particle, and expression of the ERAD ubiquitin ligase Hrd1 suppressed αsynuclein-triggered cytotoxicity (Liang et al., 2008; Gitler et al., 2009). Bridging high-throughput genetic and transcriptional data with the ResponseNet algorithm predicted the AAA-ATPase Cdc48, also critically involved in ERAD, to be a modulator of α-synuclein-triggered cytotoxicity (Yeger-Lotem et al., 2009). Thus, the UPS in general, and specifically the ERAD pathway appears to be a potent modulator of α-synuclein-triggered cell death.

Since α-synuclein aggregates have been proposed to be substrates of autophagy, it is likely that autophagy, like the UPS, modulates α-synuclein-triggered cytotoxicity. In fact, the in silico combination of high-throughput genetic and transcriptional data predicted the target of rapamycin (TOR) pathway, as a modulator of α-synuclein-triggered cytotoxicity (Yeger-Lotem et al., 2009). Addition of the TOR-inhibitor rapamycin markedly enhanced the growth deficits elicited by α-synuclein (Yeger-Lotem et al., 2009). Since inactivation of the TOR pathway induces autophagy, these data suggested, that enhancing autophagy is harmful but not cytoprotective for cultures expressing α-synuclein. Consistently, pharmacological inhibition of autophagy by treatment with chloroquine markedly extended chronological life span of yeast cells expressing α-synuclein (Sampaio-Marques et al., 2012). Although rapamycin also affects other cellular pathways, it remains possible that autophagy, in contrast to the UPS, plays a detrimental role in α-synuclein-triggered cytotoxicity.

Besides the UPS and autophagy, ubiquitin-dependent vesicle trafficking plays a role in modulating α-synuclein-triggered cytotoxicity. The E3 ligase Rsp5, involved in ubiquitin-dependent vesicle trafficking, was predicted by the ResponseNet algorithm to affect α-synuclein-triggered cytotoxicity (Yeger-Lotem et al., 2009). Indeed, loss-of-function mutations in the gene encoding Rsp5 (rsp5-1 strain) increased α-synuclein-triggered growth deficits, whereas overexpression of Rsp5 was cytoprotective (Tofaris et al., 2011). Chemical genetic screens in wildtype yeast cells established that N-aryl benzimidazole (NAB) promoted endosomal transport and protected cells from cytotoxicity (Tardiff et al., 2013). This was dependent on the deubiquitinase Doa4, the E3 ubiquitin ligase Rsp5, the Rsp5 adaptor Bul1, the DUBs Ubp7, and Ubp11, which can deubiquitylate Rsp5 substrates, the potential Rsp5 substrates (Bap2, Bap3, and Mmp1), and Vps23, which directs Rsp5 substrates for degradation in the vacuole (Tardiff et al., 2013). Notably, promoting ER-Golgi vesicle trafficking had very similar effects: αsynuclein-triggered cytotoxicity was reduced, and this reduction also depended on ubiquitin proteases, namely Ubp3 and its cofactor Bre5 (Cooper et al., 2006; Gitler et al., 2008). Thus, promoting ubiquitin-regulated vesicle trafficking prevents α-synucleintriggered cytotoxicity.

# Yeast α-Synuclein Models are Highly Useful to Elucidate the Role of Diverse Ubiquitin-Related Protein Degradation Pathways in Modulating Cytotoxicity and Neuronal Cell Death

In yeast, α-synuclein is both a substrate and an inhibitor for the UPS, autophagy, and the ubiquitin-dependent vesicular trafficking (**Figure 1**, **Table 1**). The activities of the UPS and of MVB pathways appear to play protective roles, whereas increased autophagy potentially contribute to α-synuclein-triggered cytotoxicity. The relevance of the different pathways in modulating α-synuclein-triggered cytotoxicity might depend on α-synuclein itself, e.g., the expression levels, the post-translational modifications, the cellular localizations, folding or distinct aggregation conditions. Similarly, the chronological and replicative aging of the yeast cultures might be decisive. Systematic analyses of these factors will help to get a better understanding of the pathophysiological effects of α-synuclein expression in yeast with respect to the reciprocal effects to ubiquitin-dependent protein degradation.

These studies are very promising, due to the high consistence of the yeast α-synuclein models with other α-synuclein model systems. In mammalian cells, α-synuclein degradation also depends on the UPS, autophagy, and ubiquitin-dependent vesicular trafficking (more specifically the MVB pathway) (Tofaris et al., 2011; Ebrahimi-Fakhari et al., 2012; Xilouri et al., 2013). Likewise, in mammalian cells, α-synuclein interferes with these ubiquitin-modulated protein degradation pathways, and the activities of these pathways are discussed to influence the cytotoxicity of α-synuclein (Ebrahimi-Fakhari et al., 2012; Xilouri et al., 2013). Yeast α-synuclein models have already been very valuable in deciphering novel cellular mechanisms linking

ubiquitin-dependent pathways with α-synuclein-triggered cytotoxicity. For instance, the influence of the ubiquitin-modulated ER-Golgi vesicle trafficking on α-synuclein-triggered cytotoxicity was first identified in yeast and then confirmed in flies and worms (Cooper et al., 2006). Likewise, the degradation of α-synuclein via the MVB pathway is conserved from yeast to mammalians (Tofaris et al., 2011), and the protective effect of promoting ubiquitin-dependent endosomal trafficking by NAB was first identified in yeast and later on confirmed in worms, rats, and human PD patient-derived neurons (Chung et al., 2013; Tardiff et al., 2013).

# Huntington's Disease (HD)

HD is an autosomal dominant neurodegenerative disorder characterized by a progressive loss of neurons in the striatum and the cortex with a consequent decline of cognitive and motor functions (Ross and Tabrizi, 2011). HD is caused by an abnormal polyglutamine (polyQ) expansion in the protein huntingtin due to an aberrant CAG codon expansion in the exon 1 of the gene encoding huntingtin (Ross and Tabrizi, 2011). This results in an aggregation-prone protein eventually triggering cytotoxicity and neuronal cell loss (Ross and Tabrizi, 2011). Increasing the length of the polyQ expansion accelerates aggregation of huntingtin and strictly correlates with the increase in cytotoxicity and the decrease in disease onset (Ross and Tabrizi, 2011). In order to dissect underlying mechanisms, various HD models have been established, comprising transgenic mouse lines, mammalian cell culture, and yeast (Mason and Giorgini, 2011; Ross and Tabrizi, 2011).

# In Yeast, Huntingtin with Disease-Associated Expanded Glutamine Stretches (polyQ) is a Cytoplasmic Aggregation-Prone Protein, which is Degraded Via the UPS and Autophagy

In yeast, expression of fluorescence protein-tagged huntingtin exon 1 with disease-inducing polyQ expansions (e.g., 103Q) led to very efficient cytoplasmic aggregation, in contrast to fusion proteins with normal glutamine repeats (e.g., 25Q) (Meriin et al., 2002, 2003; Duennwald et al., 2006a,b). Intermediate polyQ length (e.g., 47Q) showed moderate aggregation when expressed in logarithmically growing cells but here strong aggregation occurred delayed upon chronological aging (Cohen et al., 2012). Besides the length of polyQ expansions, aggregation was influenced by the amino acid sequences flanking the polyQ stretch (Duennwald et al., 2006b; Wang et al., 2009), and by the presence of endogenous proteins with prion properties and glutamine repeats (Duennwald et al., 2006a). The N-terminal domain of huntingtin (N17 fragment) which precedes the polyQ stretch is required for recruiting the chaperonin TRiC and the 14-3-3 protein Bmh1 which both promote aggregation in yeast and


TABLE

1


Ubiquitin-dependent

protein

(Continued)


cell-free systems (Tam et al., 2009; Wang et al., 2009; Duennwald, 2011; Crick et al., 2013). PolyQ followed by the endogenous proline-rich region of huntingtin (103QP) tended to form very tight aggregates (one or two per cell), whereas polyQ lacking the proline-rich region (103Q) preferred to form amorphous aggregates dispersed throughout the cytosol (Duennwald et al., 2006b). The yeast protein Rnq1 co-localized with both 103QP and 103Q aggregates, and its prion conformation [RNQ+] was found to be necessary for aggregation of 103QP and 103Q (Duennwald et al., 2006a).

In line with the appearance of different types of polyQ aggregates, these aggregates are directed to at least two cellular compartments. First, misfolded polyQ (103QP) was actively transported to aggresomes (Wang et al., 2009), which are large juxtanuclear aggregates that co-localize with the centrosomes in mammalian cells and the spindle pole bodies in yeast (Johnston et al., 1998; Wang et al., 2009). In mammalian cells, aggresomes are ubiquitylated, highly dynamic and easily accessible to the UPS (Johnston et al., 1998). They are formed from smaller aggregates, which are actively transported to the centrosome via the microtubule cytoskeleton (Johnston et al., 1998). Both the N17 fragment and the proline-rich domain of huntingtin are required for aggresome formation in yeast (Wang et al., 2009). Alternatively, polyQ aggregates (it is unclear whether 103Q or 103QP has been used) could be transported to perivacuolar inclusions, called "insoluble protein deposits" (IPOD) (Kaganovich et al., 2008). IPODs are non-ubiquitylated insoluble protein aggregates covered by the ubiquitin-like autophagy protein Atg8 (Kaganovich et al., 2008). Therefore, it is tempting to speculate that polyQ aggregates in perivacuolar inclusions are substrates of autophagy, whereas polyQ aggregates in aggresomes are primarily substrates of the UPS.

PolyQ aggregates could be ubiquitylated (96QP) (Lu et al., 2014), and physically interacted with proteins of the UPS, including proteasomal subunits (103QP, 96QP) (Wang et al., 2009; Park et al., 2013). Consistently, polyQ aggregate clearance (103Q) was improved in a yeast strain with elevated UPS capacities (1ubr2), confirming that polyQ aggregates are substrates of the UPS (Kruegel et al., 2011). However, the degradation of ubiquitylated polyQ was not limited to the UPS but could also occur by ubiquitin-dependent autophagy (Lu et al., 2014). Here, polyQ (96QP) was ubiquitylated by the E3 ligase Rsp5, and ubiquitylated polyQ was recognized by the ubiquitin-Atg8 adaptor protein Cue5, enabling the targeting of polyQ aggregates (96QP) to the vacuole for their degradation (Lu et al., 2014). In sum, ubiquitylated polyQ aggregates are degraded via the UPS, and ubiquitin-dependent autophagy.

Yeast strains with mutations affecting the formation of endocytic vesicles demonstrated decreased aggregation of polyQ (47Q, 103Q, 103QP) (Meriin et al., 2007). In contrast, lack of proteins which are pivotal for later steps of endocytosis increased polyQ aggregation (103Q), and when present these proteins colocalized with polyQ aggregates (103Q) (Meriin et al., 2007). These data suggest that these ubiquitin-controlled vesicular processes are involved in the formation of polyQ aggregates (Meriin et al., 2007). However, whether they are also involved in the degradation of polyQ remains elusive.

TABLE

1


Continued

# PolyQ Expression Leads to Impairment of Specific Branches of the UPS Pathway in Yeast

PolyQ expression inhibits the UPS. In the presence of polyQ, increased cellular levels of polyubiquitylated proteins were observed (with 103Q but not with 103QP) (Duennwald and Lindquist, 2008), genes involved in ubiquitin cycle and protein ubiquitylation were up-regulated (with 103Q) (Giorgini et al., 2008; Tauber et al., 2011), and the degradation of cytosolic and ER-associated proteins were impaired (with 103Q but not with 103QP) (Duennwald and Lindquist, 2008). Notably, the polyQdependent UPS impairment is substrate specific. Upon polyQ expression (103Q), the degradation of cytosolic ubiquitin-fusion degradation (UFD) substrates (Ub-P-LacZ) was more affected than the degradation of cytosolic substrates of the N-end rule pathway (Ub-R-LacZ) (Duennwald and Lindquist, 2008). Consistently, the degradation of ER-associated proteins via ERAD, which shares many components with the UFD pathway, was drastically affected upon polyQ expression (103Q) (Duennwald and Lindquist, 2008). Here, the degradation of the ER luminal misfolded variant of the carboxypeptidase Y (CPY<sup>∗</sup> ) was impaired, as well as the degradation of the misfolded ER membrane protein sec61-2, and others (Duennwald and Lindquist, 2008). PolyQ (both 103Q and 103QP) physically interacted with the AAA-ATPase Cdc48 and its co-factors Ufd1 and Npl4, which are pivotally involved in UFD and ERAD (Duennwald and Lindquist, 2008; Wang et al., 2009). It has been proposed that sequestration of the Cdc48-Ufd1/Npl4 complex by polyQ (103Q) is the cause for the specific impairment of the UFD and ERAD pathways, culminating in ER stress and UPR induction (Duennwald and Lindquist, 2008). Notably, the interaction of polyQ (103QP) with the Cdc48-Ufd1/Npl4 complex was essential for the formation of the polyQ-containing aggresomes (Wang et al., 2009). The formation of these aggresomes is believed to be a protective cellular mechanism to prevent the evenly spread of misfolded proteins within a cell (Johnston et al., 1998). Therefore, the interaction of the Cdc48-Ufd1/Npl4 complex with polyQ might be both protective (at least for 103QP) and detrimental (for 103Q); it prevents from the accumulation of misfolded polyQ but promotes specific dysfunction of the UFD and ERAD pathways.

Expression of polyQ containing the proline-rich domain (97QP) can also lead to UPS dysfunction via an alternative mechanism, which is based on the sequestration of chaperones. PolyQ (97QP) inhibited the UPS-mediated degradation of the misfolded cytosolic variant of carboxypeptidase Y (1ssCPY\*) lacking the signaling sequence (ss) for entering the secretory pathway (Park et al., 2013). Upon polyQ expression (97QP), the steady-state levels and the turnover rates of 1ssCPY<sup>∗</sup> were increased and delayed, respectively (Park et al., 2013). PolyQ-triggered (97QP) UPS inhibition was not due to direct interference with proteasomal function, because the ubiquitin-independent proteasomal degradation of ornithine decarboxylase was not affected by polyQ expression (Park et al., 2013). UPS dysfunction could also not be explained by interference of polyQ (97QP) with the ubiquitylation of 1ssCPY\* prior its proteasomal degradation (Park et al., 2013). Instead, polyQ (97QP) interfered with the transfer of the misfolded protein to the proteasome, prior proteasomal degradation, which surprisingly took place in the nucleus (Park et al., 2013). The nuclear transfer of misfolded 1ssCPY<sup>∗</sup> was done by binding to the type II Hsp40 chaperone Sis1, which shuttles into the nucleus, and polyQ (97QP) interfered with this process by sequestering Sis1 in the cytosol (Park et al., 2013). Consistently, Sis1 overexpression restored the degradation of 1ssCPY<sup>∗</sup> in the presence of polyQ. Thus, polyQ (both 103Q and 97QP) interferes with the proteasome-dependent degradation of proteins in the nucleus and the cytosol, by sequestering the Sis1 chaperone (for 97QP) and the Cdc48-Ufd1/Npl4 complex (for 103Q), respectively.

Whether polyQ interferes with autophagy and protein degradation based on ubiquitin-controlled vesicular transport remains to be determined. Since polyQ aggregates (103Q) impaired endocytosis in yeast (Meriin et al., 2003, 2007), it is tempting to speculate that polyQ aggregates (103Q) also affect these vesicle-based protein degradation pathways.

# PolyQ Expression in Yeast Triggers Mitochondrial Dysfunction and ER Stress, which is Modulated by Specific UPS Activities

PolyQ constructs encoding 103 glutamine residues (103Q) efficiently triggered growth deficits and morphological markers of apoptosis in yeast (Meriin et al., 2002; Duennwald et al., 2006a; Sokolov et al., 2006; Solans et al., 2006). In contrast, polyQ constructs encoding 25 glutamine residues (25Q) remained nontoxic, and the constructs with intermediate polyQ lengths showed intermediate cytotoxicity (Meriin et al., 2002; Duennwald et al., 2006a; Sokolov et al., 2006; Solans et al., 2006; Ocampo et al., 2010). Besides the number of glutamine expansions, the cytotoxicity of polyQ strongly depended on the proline-rich domain flanking the polyQ stretch (Duennwald et al., 2006b). PolyQ (103QP) remained non-toxic, whereas polyQ lacking the prolinerich region (103Q) induced cytotoxicity (growth deficits) (Duennwald et al., 2006b). In this context, the presence of the prion conformation of the yeast protein Rnq1 was a prerequisite for polyQ-triggered cytotoxicity (103Q), and the presence of other endogenous yeast proteins with glutamine repeats also influenced polyQ-triggered cytotoxicity (103Q) (Duennwald et al., 2006a). Thus, polyQ-triggered cytotoxicity depends on the length of the glutamine repeats, the absence or presence of the proline-rich domain, and the cellular protein interaction network.

Cytotoxic polyQ (103Q) physically interacted with mitochondria and triggers critical mitochondrial dysfunction (Solans et al., 2006; Ocampo et al., 2010). PolyQ expression (103Q but not 103QP) also induced ER stress leading to UPR (Duennwald and Lindquist, 2008), and has also been proposed to lead to lethal impairment of cell cycle progression (Bocharova et al., 2008, 2009). ER and cell cycle impairments are believed to be consequences of the impairment of specific branches of the UPS pathway by polyQ expression (103Q), including the ERAD and anaphase promoting complex (APC) pathways (Bocharova et al., 2008, 2009; Duennwald and Lindquist, 2008). The mitochondrion-associated protein degradation pathway (MAD) shares pivotal components with the ERAD pathway (Heo et al., 2010; Taylor and Rutter, 2011). Therefore, it is tempting to speculate that polyQ expression (103Q) also impairs MAD, contributing to the observed mitochondrial damage.

PolyQ-triggered cytotoxicity was increased by genetic impairment of the ERAD and UPR pathways and by application of ER stress (for 103Q but not for 103QP) (Duennwald and Lindquist, 2008). In contrast, promoting ERAD by expression of Npl4 and Ufd1, which are pivotally involved in ERAD, or constitutive activation of the UPR suppressed polyQ-triggered cytotoxicity (103Q) (Duennwald and Lindquist, 2008). These data are in line with the idea that ERAD dysfunction by polyQ expression (103Q) is critical in the execution of cytotoxicity and cell death. Deletion of the APC substrate ASE1 relieved polyQ-triggered cytotoxicity (103Q), suggesting that preventing the accumulation of Ase1 upon dysfunction of the APC pathway is beneficial (Bocharova et al., 2008). Expression of polyQ flanked by the proline-rich domain (96QP), which is non-toxic under normal conditions, became cytotoxic upon accumulation of the UPS model substrate 1ssCPY<sup>∗</sup> and could be relieved upon Sis1 overexpression (Park et al., 2013). Genetic inactivation of ubiquitin-dependent autophagy (e.g., rsp5-2, 1cue5) also induced cytotoxicity upon polyQ expression (96QP) (Lu et al., 2014). These data suggest that besides ERAD and APC pathways, the Sis1-dependent protein degradation and the ubiquitin-dependent autophagy are pivotal in the modulation of polyQ-triggered cytotoxicity in yeast.

# Yeast PolyQ Models are Highly Useful to Elucidate the Role of Diverse Ubiquitin-Related Protein Degradation Pathways in Modulating Cytotoxicity and Neuronal Cell Death

PolyQ expression in yeast leads to aggregation, and aggregates can be allocated to distinct cellular compartments, including aggresomes and perivacuolar inclusions (IPOD) (**Figure 2**, **Table 1**). Aggregated polyQ is a substrate of the UPS and autophagy. Besides its role of a substrate of protein degradation, polyQ impairs the UPS (especially ERAD), and also affects proper function of vesicular transport (more specifically endocytosis). This leads to ER stress, mitochondrial dysfunction, and cell death. All these features depend on the length of the polyQ stretch, the absence (or presence) of the N-terminal part (N17 fragment) and the proline-rich domain, respectively, and the cellular protein interaction network.

The data obtained in yeast HD models have been very helpful for a better understanding of polyQ-triggered effects in higher model systems, as well as in humans. For instance, the allocation of different aggregates into distinct cellular compartments (aggresomes and IPOD, respectively) was conserved from yeast to mammalians (Johnston et al., 1998; Kaganovich et al., 2008; Wang et al., 2009). As in yeast, polyQ are substrates of the UPS and autophagy (Lu et al., 2014; Margulis and Finkbeiner, 2014; Martin et al., 2014). PolyQ aggregates block specific branches of the UPS (including the ERAD and the Sis1-dependent pathways) in mammalians as in yeast (Duennwald and Lindquist, 2008; Park et al., 2013; Margulis and Finkbeiner, 2014). Like in yeast cells, polyQ impairs vesicle-based protein degradation, including autophagy (Sapp et al., 1997; Aronin et al., 1999; Meriin et al., 2003, 2007; Martin et al., 2014).

Interestingly, in cultured cells and in flies, ubiquitylation (and sumolyation) of the N-terminal part of huntingtin (N17) at specific residues (K6, K9, and K15) turned out to be an efficient modulator of polyQ aggregation and neurotoxicity (Steffan et al., 2004). Thus, distinct (poly)ubiquitylation patterns could on the one hand dictate the fate of polyQ (e.g., aggregation and degradation) and on the other hand determine its effects on ubiquitindependent degradation of other cellular proteins. This issue could further be dissected in yeast HD models.

Of specific interest is the role of p97/VCP, the ortholog of the yeast AAA-ATPase Cdc48. As in yeast, p97/VCP and its cofactors Npl4 and Ufd1 were sequestered by polyQ aggregates in mammalian cells leading to ERAD dysfunction, ER stress, and UPR induction (Duennwald and Lindquist, 2008). Overexpression of Npl4, Ufd1, or p97/VCP rescued from ERAD dysfunction and/or ER stress and UPR (Duennwald and Lindquist, 2008; Leitman et al., 2013). In C. elegans, p97/VCP was involved in the clearance of detrimental polyQ aggregates (Nishikori et al., 2008). Since p97/VCP has also been proposed to play important roles in autophagy and apoptosis (Braun and Zischka, 2008; Krick et al., 2010), it is tempting to speculate that the interaction of polyQ aggregates with this protein complex is critical for the switch from UPS and autophagy dysfunctions to neuronal cell loss.

# Alzheimer's Disease (AD)

AD is the most prevalent form of age-related dementia (Querfurth and Laferla, 2010). Synaptic loss and neuronal decline can be observed in affected brain regions, including the hippocampus and the cortex (Querfurth and Laferla, 2010). The accumulation of extracellular senile plaques and intracellular aggregates, composed of β-amyloid, and the intracellular accumulation of neurofibrillary tangles comprising hyperphosphorylated variants of the protein tau and UBB+<sup>1</sup> , the frameshift variant of ubiquitin B, contribute to AD progression (van Leeuwen et al., 1998; Laferla et al., 2007; Benilova et al., 2012; Mandelkow and Mandelkow, 2012).

# Yeast β-Amyloid Models

Several yeast models expressing β-amyloid in the cytosol and in the secretory pathway have been generated (Caine et al., 2007; Treusch et al., 2011; D'Angelo et al., 2013; Matlack et al., 2014; Mossmann et al., 2014) (**Table 1**). Cytosolic β-amyloid (GFP-Aβ42 or Aβ42-GFP fusion proteins) led to moderate growth deficits and to the induction of the heat shock response in wild-type cells (Caine et al., 2007). Mitochondria isolated from aging-prone yeast cells (1coa6) upon prolonged exposure to cytosolic Aβ42 demonstrated signs of mitochondrial impairment, including inhibition of mitochondrial pre-protein maturation, increased ROS production, decreased mitochondrial membrane potential and reduced oxygen consumption (Mossmann et al., 2014). Recent studies suggest that mitochondrial impairment triggers UPS dysfunction and vice versa (Livnat-Levanon et al., 2014; Maharjan et al., 2014; Segref et al., 2014; Braun et al., 2015). Since mitochondrial and UPS dysfunction are believed to contribute to AD progression, it is of high interest, whether cytosolic Aβ42 interacts with ubiquitin-dependent processes.

Whereas, the detrimental effects of cytosolic β-amyloid remained moderate, β-amyloid directed to the secretory pathway resulted in high cytotoxicity concomitant to β-amyloid oligomerization and aggregation (Treusch et al., 2011; D'Angelo

et al., 2013; Matlack et al., 2014). Here, β-amyloid impaired the clathrin-mediated endocytic trafficking of a plasma membrane receptor (which is a ubiquitin-modulated process), and expression of endocytic genes rescued β-amyloid-triggered cytotoxicity (Treusch et al., 2011; D'Angelo et al., 2013). A genomewide screen for modifiers of β-amyloid-triggered cytotoxicity identified the yeast homolog of phosphatidylinositol binding clathrin assembly protein (PICALM), as suppressor of cytotoxicity (Treusch et al., 2011). A screen of ∼140,000 compounds for β-amyloid rescue in yeast identified the 8-hydroxyquinoline clioquinol as effective in preventing β-amyloid-triggered growth deficits via promoting β-amyloid degradation and restoring endocytic function (Matlack et al., 2014). The outcome of these two unbiased yeast screens validate the usefulness of yeast AD models, because PICALM is one of the most highly validated AD risk factors (Treusch et al., 2011), and clioquinol is effective in both mouse and C. elegans AD models (Matlack et al., 2014).

# Yeast tau Models

Yeast models expressing AD-associated tau have been generated (De Vos et al., 2011) (**Table 1**). Tau forms oligomers and is hyperphosphorylated when expressed in yeast as it is in AD patients (Vandebroek et al., 2005, 2006; Zabrocki et al., 2005; Vanhelmont et al., 2010). Although tau expression per se remained non-toxic in wild-type cells, it increased α-synucleintriggered growth deficits (Zabrocki et al., 2005). Therefore, combining several AD risk factors (e.g., tau, β-amyloid, and/or UBB+<sup>1</sup> ) could lead to more effective yeast AD models. This turned out to be necessary in mouse AD models, because mutations in tau are also not very toxic in transgenic mice (Nisbet et al., 2015). Similarly, AD-associated mutations in the β-amyloid precursor protein (APP) or the γ-secretase components presenilin 1 or 2, which are pivotally involved in β-amyloid generation from APP, do also not fully recapitulate the key pathological events of AD (Nisbet et al., 2015). In order to mimic simultaneously more than one pathological aspect of AD double and triple transgenic AD mouse models or AD patient-derived neuronal stem-cell-derived three-dimensional culture systems have been developed (van Tijn et al., 2012; Choi et al., 2014; Nisbet et al., 2015).

# Yeast UBB+<sup>1</sup> Models

UBB+<sup>1</sup> -expressing yeast models have been established (Tank and True, 2009; Verhoef et al., 2009; Dennissen et al., 2011; Krutauz et al., 2014; Braun et al., 2015) (**Table 1**). As in mammalian cells, UBB+<sup>1</sup> is a substrate of the UPS in yeast. It is ubiquitylated prior proteasomal degradation and truncated by the DUB Yuh1 (or mammalian ubiquitin carboxy-terminal hydrolase UCH-L3) (De Vrij et al., 2001; van Tijn et al., 2007, 2010; Tank and True, 2009; Verhoef et al., 2009; Dennissen et al., 2011; Braun et al., 2015). Accumulation of UBB+<sup>1</sup> impairs the UPS in yeast and mammalian cells, culminating in the accumulation of polyubiquitylated proteins. In both cases, the degradation of UFD substrates (e.g., Ub-G76V-GFP, Ub-P-LacZ), and of N-end rule substrates (e.g., Ub-R-GFP, Ub-R-LacZ) was significantly reduced (Lindsten et al., 2002; van Tijn et al., 2007; Tank and True, 2009; Braun et al., 2015). The inhibitory effect of UBB+<sup>1</sup> on the UPS is at least partially due to inhibition of DUBs, especially Ubp6/USP14, which is associated with the 26S proteasome and involved in the disassembly of polyubiquitin tags from substrate proteins (Krutauz et al., 2014).

Despite of the UBB+<sup>1</sup> -triggered UPS impairment, both yeast and mammalian cells can tolerate UBB+<sup>1</sup> without marked signs of cytotoxicity (Hope et al., 2003; van Tijn et al., 2007, 2012; Tank and True, 2009; Yim et al., 2014), and in some cases UBB+<sup>1</sup> expression was even protective when mammalian cells were treated with chemicals inducing oxidative stress (Hope et al., 2003; Yim et al., 2014). Consistently, transgenic expression of UBB+<sup>1</sup> in mice failed to cause overt neurodegeneration although it did affect spatial reference memory and caused a central dysfunction of respiratory regulation (Fischer et al., 2009; van Tijn et al., 2011; Irmler et al., 2012). In contrast, prolonged expression of high levels of UBB+<sup>1</sup> induced cell death and pivotal mitochondrial impairment in neuronal cells and in yeast (De Vrij et al., 2001; Tan et al., 2007; Braun et al., 2015). In yeast the UBB+<sup>1</sup> -triggered cell death could be prevented by specifically promoting the UPS activity at mitochondria (mitochondrionassociated degradation, MAD), which protected cells from mitochondrial impairment but did not alter the steady-state levels of UBB+<sup>1</sup> (Braun et al., 2015). The accumulation of UBB+<sup>1</sup> increased the cytotoxicity of polyQ (103Q) in yeast (Tank and True, 2009), which is highly comparable to mammalian systems, where UBB+<sup>1</sup> expression accelerates cell death in transgenic HD mouse models (de Pril et al., 2004, 2007, 2010). Thus, on the one hand UBB+<sup>1</sup> is a neurotoxic protein by itself, whose cytotoxicity depends both on the cellular UPS capacity and on mitochondrial (dys)function. On the other hand UBB+<sup>1</sup> is putatively a potent modifier of cytotoxicity of other misfolded neurotoxic proteins. Especially when combined with further AD risk factors, including <sup>β</sup>-amyloid and tau, the yeast UBB+<sup>1</sup> model could be very valuable for elucidating the molecular connections among AD risk factors, UPS (dys)function, mitochondrial activity, and cell survival.

# Amyotrophic Lateral Sclerosis (ALS)

ALS is a frequent degenerative motor neuron disease, resulting in muscle weakness and wasting (Andersen and Al-Chalabi, 2011). Among the most common ALS-associated genes are TARDBP, and FUS, encoding the RNA-binding proteins TDP-43 and FUS/TLS (Andersen and Al-Chalabi, 2011). ALS-associated variants of these proteins demonstrate mislocalization and/or a high tendency for aggregation, and these aggregates are ubiquitylated (Andersen and Al-Chalabi, 2011; Da Cruz and Cleveland, 2011). Yeast models expressing ALS-associated wild-type and mutant TDP-43 and FUS/TLS have been established to further analyze the detrimental roles of these proteins on cell survival (Bastow et al., 2011).

TDP-43 and FUS/TLS efficiently form cytoplasmic aggregates when expressed in yeast cells triggering cytotoxicity and cell death (Johnson et al., 2008; Braun et al., 2011; Kryndushkin et al., 2011; Sun et al., 2011) (**Table 1**). It is very little known via which pathways these neurotoxic proteins and their aggregates are degraded in yeast, whether these proteins impair ubiquitin-modulated protein degradation, and whether the activities of ubiquitin-modulated proteolysis pivotally affect their cytotoxicity. In the case of TDP-43, a temperature-sensitive variant of the AAA-ATPase Cdc48 was found as an enhancer of TDP-43-triggered cytotoxicity (growth deficits) (Armakola et al., 2012). Promoting vesicular trafficking (ER -> Golgi) by treating TDP-43-expressing yeast cells with NAB relieved TDP-43-triggered cytotoxicity (Tardiff et al., 2012, 2013). Thus, the activities of ubiquitin-modulated vesicular transport and Cdc48 dependent cellular processes may be pivotal in modulating TDP-43-triggered cytotoxicity in yeast and neurons. Cdc48 and its mammalian homolog p97/VCP are involved in many different ubiquitin-modulated processes, including ERAD, MAD, and vesicular transport, such as endocytosis, and autophagy (Bug and Meyer, 2012). Indeed, mutations in p97/VCP lead to cytoplasmic TDP-43 aggregation and cell death in neurons of transgenic mice and flies (Gitcho et al., 2009; Johnson et al., 2010; Ritson et al., 2010; Rodriguez-Ortiz et al., 2013). Yeast TDP-43 models could help to further discriminate the role of the distinct Cdc48/p97/VCP pathways in modulating cell death and neuronal loss.

# Conclusions and Outlook

In yeast models expressing neurotoxic proteins, these proteins are on the one hand substrates of distinct protein degradation pathways, and on the other hand the trigger for their impairment. Consistently, the activities of the UPS, autophagy, and other ubiquitin-controlled vesicle-based protein degradation pathways are pivotal for the cytotoxicity of neurotoxic proteins. Best described for yeast PD and HD models, these findings could be confirmed in other neurotoxic model systems, confirming that yeast models expressing neurotoxic proteins are very helpful in elucidating novel paradigms of pathobiology in neurodegenerative disorders. It is very likely, that yeast models for AD (βamyloid, tau, UBB+<sup>1</sup> ), and ALS (e.g., TDP-43, FUS/TLS), will lead to the identification of further ubiquitin-regulated protein degradation pathways that potentially underlie neuronal dysfunction and cell loss. In this respect, the easy and fast combination of different neurotoxic proteins in one yeast model (e.g., different AD-associated proteins) could be a very straightforward approach.

The role of free ubiquitin homeostasis has been largely unattended when analyzing the effects of ubiquitylated neurotoxic proteins and ubiquitin-dependent protein degradation on neuronal survival. For instance, loss of free ubiquitin was found in schizophrenia (Rubio et al., 2013), and down-regulation of free ubiquitin was determined to be causative for p53 accumulation and apoptosis in hippocampal neurons from rats (Tan et al., 2000). In contrast, free ubiquitin levels increase in cancer cells, which turned out to be pivotal for cell growth (Oh et al., 2013). Consistently, high levels of free ubiquitin confers resistance to inhibitors of protein translation in yeast (Hanna et al., 2003). Modulating free ubiquitin levels in yeast models expressing human neurotoxic proteins could address the role of free

# References


ubiquitin homeostasis on the degradation and the cytotoxicity of these proteins.

# Acknowledgments

I am grateful to the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG) for grant BR 3706/3-1. This publication was funded by the University of Bayreuth in the funding program Open Access Publishing.


Tau-4R and Tau-P301L proteins isolated from yeast deficient in orthologues of glycogen synthase kinase-3beta or cdk5. J. Biol. Chem. 281, 25388–25397. doi: 10.1074/jbc.M602792200


in a yeast model for Parkinson. Biochim. Biophys. Acta 1783, 1767–1780. doi: 10.1016/j.bbamcr.2008.06.010


**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 © 2015 Braun. 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.