# ALZHEIMER'S DISEASE: ORIGINAL MECHANISMS AND TRANSLATIONAL IMPACT

EDITED BY : Cesare Mancuso and Silvana Gaetani PUBLISHED IN : Frontiers in Pharmacology and Frontiers in Neuroscience

#### Frontiers eBook Copyright Statement

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

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

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

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

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

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

ISSN 1664-8714 ISBN 978-2-88963-631-0 DOI 10.3389/978-2-88963-631-0

### 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

# ALZHEIMER'S DISEASE: ORIGINAL MECHANISMS AND TRANSLATIONAL IMPACT

Topic Editors: Cesare Mancuso, Università Cattolica del Sacro Cuore, Italy Silvana Gaetani, Sapienza University of Rome, Italy

Citation: Mancuso, C., Gaetani, S., eds. (2020). Alzheimer's Disease: Original Mechanisms and Translational Impact. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-631-0

# Table of Contents


Cristina Angeloni, Maria Cristina Barbalace and Silvana Hrelia

*17 7-Pyrrolidinethoxy-4'-Methoxyisoflavone Prevents Amyloid ß–Induced Injury by Regulating Histamine H3 Receptor-Mediated cAMP/CREB and AKT/GSK3ß Pathways*

Linlin Wang, Jiansong Fang, Hailun Jiang, Qian Wang, Situ Xue, Zhuorong Li and Rui Liu

*33 Monoaminergic System Modulation in Depression and Alzheimer's Disease: A New Standpoint?*

Maria Grazia Morgese and Luigia Trabace


Anna Rita Zuena, Paola Casolini, Roberta Lattanzi and Daniela Maftei


Fabiana Morroni, Giulia Sita, Agnese Graziosi, Gloria Ravegnini, Raffaella Molteni, Maria Serena Paladini, Kris Simone Tranches Dias, Ariele Faria dos Santos, Claudio Viegas Jr., Ihosvany Camps, Letizia Pruccoli, Andrea Tarozzi and Patrizia Hrelia

*81 Sleep and ß-Amyloid Deposition in Alzheimer Disease: Insights on Mechanisms and Possible Innovative Treatments*

Susanna Cordone, Ludovica Annarumma, Paolo Maria Rossini and Luigi De Gennaro

*93 Doxycycline for Alzheimer's Disease: Fighting ß-Amyloid Oligomers and Neuroinflammation*

Claudia Balducci and Gianluigi Forloni

*100 Targeting Synaptic Plasticity in Experimental Models of Alzheimer's Disease*

Dalila Mango, Amira Saidi, Giusy Ylenia Cisale, Marco Feligioni, Massimo Corbo and Robert Nisticò

*108 MicroRNA-200a-3p Mediates Neuroprotection in Alzheimer-Related Deficits and Attenuates Amyloid-Beta Overproduction and Tau Hyperphosphorylation* via *Coregulating BACE1 and PRKACB*

Linlin Wang, Jianghong Liu, Qian Wang, Hailun Jiang, Li Zeng, Zhuorong Li and Rui Liu

*120 Palmitoylethanolamide (PEA) as a Potential Therapeutic Agent in Alzheimer's Disease*

Sarah Beggiato, Maria Cristina Tomasini and Luca Ferraro


Alfonso Grimaldi, Natalia Pediconi, Francesca Oieni, Rocco Pizzarelli, Maria Rosito, Maria Giubettini, Tiziana Santini, Cristina Limatola, Giancarlo Ruocco, Davide Ragozzino and Silvia Di Angelantonio


Ilaria Dal Prà, Ubaldo Armato and Anna Chiarini

*209 Ethanolic Extract of* Orthosiphon stamineus *Improves Memory in Scopolamine-Induced Amnesia Model* Thaarvena Retinasamy, Mohd Farooq Shaikh, Yatinesh Kumari and Iekhsan Othman

# Editorial: Alzheimer's Disease: Original Mechanisms and Translational Impact

Cesare Mancuso1,2\* and Silvana Gaetani <sup>3</sup>

<sup>1</sup> Fondazione Policlinico Universitario A Gemelli IRCCS, Rome, Italy, <sup>2</sup> Institute of Pharmacology, Università Cattolica del Sacro Cuore, Rome, Italy, <sup>3</sup> Department of Physiology and Pharmacology "V. Erspamer," Sapienza University of Rome, Rome, Italy

Keywords: Alzheimer's disease, preclinical studies, drug research and development, neurodegeneration, synaptic plasticity

Editorial on the Research Topic

### Alzheimer's Disease: Original Mechanisms and Translational Impact

Alzheimer's disease (AD) is a neurodegenerative disorder characterized by progressive and irreversible worsening of cognitive functions, inability to perform everyday activities, and mood disorders. Currently, AD is considered the leading cause of dementia and hospitalization of older adults in nursing homes. In the United States, 5.8 million people has been calculated to suffer from AD in 2019, 81% being 75 years or older; the percentage of individuals with AD increases with age, from 3% of people aged 65–74 to 32% of people aged 85 and older. Women are more affected by AD than men (M/F 2/1) probably because of their longer lifespan. Finally, African Americans and Hispanics are about twice likely to develop AD as older Whites (Alzheimer's Association, 2019). The lack of any updated epidemiologic survey about AD in Europe is quite disappointing; the most accurate analysis dates back 2017 and reveals an estimated prevalence at 5.05% (men 3.31% and women 7.13%) increasing with age (Niu et al., 2017). In Europe, about 3 million people was estimated to suffer from AD (Mayer et al., 2018).

From a pathogenetic viewpoint, the early "amyloid cascade hypothesis", which considered fibrillar b-amyloid (Ab) and hyperphosphorylated tau protein (pTau) as the main inducers of the pro-oxidant status and neuroinflammation leading to neuronal death, was definitely challenged (Selkoe and Hardy, 2016). Clinical evidence has clearly demonstrated both the evidence that the amount of senile plaques, containing fibrillar Ab, does not correlate with the severity of AD and the lack of efficacy of therapies targeting fibrillar Ab in terms of improvement of cognitive function (Nelson et al., 2012; Penninkilampi et al., 2016; Wang et al., 2017). Over the last few years, soluble <sup>A</sup>b, mainly in the oligomeric form, has been proposed as the toxic species being responsible for the early impairment of synaptic plasticity and neurotransmission occurring in AD (Abdel-Hafiz et al., 2018; Li et al., 2018). Unfortunately, AD begins several years before the onset of symptoms, which become evident when neurodegeneration reaches the point of no return. This is the reason why drugs currently available, such as acetylcholinesterase inhibitors and the N-methyl-D-aspartate (NMDA) receptor antagonist memantine, have only limited symptomatic effects; regrettably, there is not any class of drugs capable of preventing or contrasting the evolution of the disease (Mancuso et al., 2011; Mhillaj et al., 2017). As recently reported by Cummings et al. (2019), 132 drugs are under clinical development for AD and only 28 of them are in phase III; among these latter, nine are anti-amyloid agents, eight are compounds targeting neuropsychiatric symptoms, and only three are antioxidant/neurotransmitter-based therapies.

#### Edited by:

Agata Copani, University of Catania, Italy

> Reviewed by: Elena Marcello,

University of Milan, Italy

\*Correspondence: Cesare Mancuso cesare.mancuso@unicatt.it

#### Specialty section:

This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology

Received: 15 January 2020 Accepted: 06 February 2020 Published: 26 February 2020

#### Citation:

Mancuso C and Gaetani S (2020) Editorial: Alzheimer's Disease: Original Mechanisms and Translational Impact. Front. Pharmacol. 11:157. doi: 10.3389/fphar.2020.00157

**5**

The aim of this Research Topic is to outline the multifactorial etiology of AD and promising key factors for the development of new and successful therapeutic strategies. The issues addressed in this Research Topic include, among others, the interplay between well-known and novel molecular mechanisms, such as oxidative and neuroinflammatory events leading to synaptic failure, some comorbidities secondary to exaggerated Ab deposition and the potential therapeutic role for medicinal herbs or drugs to slowdown the progression of AD. This Research Topic, in which several leading experts have provided important contributions, is organized in eight original research articles (including a brief research report), four reviews, and six mini-reviews.

Caruso et al. examined the role of stress and the effects due to the hyperactivation of the hypothalamic-pituitary-adrenal axis as determinants of AD. This review is quite interesting because it focuses the attention on potential lifestyle risk-factors whose elimination could drastically reduce the onset of AD. That said, everybody knows that a stress-free life is an unattainable dream, and the possibility to prevent or contrast AD based on an impossible lifestyle is a vain hope. However, the evidence that specific genetic variants, by reducing either the activity of specific enzymes involved in cortisol degradation (e.g., the 11ßhydroxysteroid dehydrogenase) or the sensitivity of glucocorticoid receptor to cortisol, might increase or decrease the risk to develop AD, respectively, opens new avenues about the role of tailored medicine for an early diagnosis of dementia. Chemokines and their receptors are widely distributed in both neurons and glial cells and play a pivotal role in neuroinflammation. Zuena et al. highlighted the contribution of prokineticin 2 and its receptors in the pathogenesis of AD; the authors, after a careful analysis of available preclinical data, strong support the hypothesis that the pharmacological antagonism of prokineticin receptors could reduce neurodegeneration, thus including these chemokines in the arena of novel and promising drug targets in AD. The contribution of glial cells, in particular astrocytes, to neuroinflammation is a blooming field of research. In an interesting original research, Grimaldi et al. described the detection of both Ab and pTau aggregates in the retina of AD patients vis-à-vis with neuronal death and detrimental astrocytes and microglial activation. These data, confirm the role of aberrant glial cell activation as a milestone in the pathogenesis of AD, but also suggest the hypothesis to consider retina as an easily accessible window for an early detection of pathological AD hallmarks. Dal Pra et al. described the role of family C Gprotein-coupled receptors (GPCR), in particular those expressed by astrocytes, in the onset and progression of AD; furthermore, these authors highlight the role of GPCR as possible drug-targets to challenge neurodegeneration. The effect of aging on astrocyte function was explored by Bronzuoli et al. in a transgenic mouse model of AD (3xTg-AD): the authors demonstrated how aging, rather than AD progression, importantly affects morphology and functions of hippocampal glial cells. These results, novel and provocative, should prompt researchers to further study the role of astrocytes and microglia in both physiological and pathological aging. Recent studies have demonstrated how

brain microRNAs participate in multiple aspects of AD pathology: in this regard, Wang, Liu et al. studied microRNA-200a-3p (miR-200a-3p) in transgenic preclinical model of AD (APP/PS1 and SAMP8 mice) and in the blood of AD patients. The authors concluded that this microRNA is neuroprotective through the inhibition of Ab overproduction via suppression of the expression of BACE1 and the synergistic decrease of pTau hyperphosphorylation. The contribution of mitochondriaderived reactive oxygen species in neuronal death is another quite exploited line of research in neurodegenerative diseases. Cenini and Voos provided an updated and exhaustive review about the potential therapeutic efficacy of several agents, including some nutritional antioxidants, to challenge AD by acting at the mitochondrial level. However, the authors concluded that, despite the huge lines of preclinical evince supporting this idea, there is no clinical evidence strong enough to support the hypothesis that mitochondria pharmacological manipulation is currently an option for AD therapy.

The progressive loss of cognitive function in AD subjects was associated to the early impairment in synaptic transmission due to Ab deposition. Long-term potentiation (LTP) and long-term depression (LTD) are the two most characterized forms of durable synaptic strength, particularly in the hippocampal region, and the magnitude of LTP and LTD is considered as an index of cognitive function in many different experimental conditions. In an interesting mini review, Mango et al. described how LTP and LTD are dysfunctional in several preclinical models of AD. Furthermore, these authors discussed the possible beneficial effects of either investigational agents or non-invasive treatments, such as repetitive transcranial magnetic stimulation and transcranial direct current stimulation, to contrast or slow-down dementia by modulating synaptic plasticity. In a preclinical model of early AD amyloidosis, the McGill-R-Thy1-APP transgenic rat, Qi et al. described the effects of soluble Ab on synaptic plasticity. According to this study, pre-plaque Ab mediated an agedependent inhibition of both LTP and novelty explorationinduced depotentiation in these animals, but only at apical synapses in the CA1 area of hippocampus. The differential susceptibility of plasticity at apical and basal synapses suggests a circuit-selective reduction in the dynamic range of synaptic gain and weakening.

An important aspect that healthcare practitioners must deal with, is the onset of comorbidities in AD patients due to the abnormal A<sup>b</sup> deposition in brain. Cordone et al. provided a detailed review about the occurrence of sleep disturbances in AD subjects as early as Ab accumulates in the brain. The authors raised the alarm, based on a restricted number of clinical trials, that sleep disruption could lead to deleterious effects on Ab accumulation in healthy populations. The most useful approach to reduce this risk is to encourage virtuous behavior, such as reducing both the use of psychoactive substances and the time of exposure to light in the evening, practicing physical and social activities, and keeping constant bed and wake times. With regard to pharmacological treatments, melatonin was extensively studied for this purpose, but the final evidence supporting its beneficial role to improve cognitive skills by restoring sleep efficiency is still lacking. Depression is another comorbidity frequently occurring in AD patients, in particular during the preclinical stage, and several lines of evidence have linked soluble <sup>A</sup>b formation with depressive state (Colaianna et al., 2010; Chi et al., 2014). On this regard, Morgese and Trabace summarized the preclinical and epidemiological studies about the role of monoaminergic system impairment as a cause of depression in AD and proposed novel therapeutic approaches based on the modulation of such a neurotransmitter system.

The use of medicinal plants, endowed with antioxidant and neuroprotective features, to contrast neurodegeneration is currently a hot field of research. Angeloni et al. provided a complete mini review on the neuroprotective effects of icariin, a prenylated flavonoid considered as the main bioactive of Herba epimedii (a Chinese herbal medicine), in AD. The authors described the pharmacokinetics of icariin as well as the antinflammatory and antioxidant effects in AD. Similarly, Retinasamy et al. described the neuroprotective and nootropic outcomes of Orthosiphon stamineus, a medicinal plant abundant in Southeast Asia, in scopolamine-treated rats. Beggiato et al. overviewed the neuropharmacology of N-palmitoylethanolamide (PEA), a lipid mediator belonging to the class of the Nacylethanolamides and firstly isolated from soy lecithin, egg yolk, and peanut meal. On these bases, both icariin and Orthosiphon stamineus, as well as PEA, have been proposed as potential adjuvant therapies in AD subjects.

Over the last few years, many drugs, initially authorized and marketed for the treatment of other diseases, have proven to be potentially effective for the treatment of AD. Ono and Tsuji and Balducci and Forloni put under the spotlight cilostazol and doxycycline, respectively. The first is an antiplatelet drug used for the treatment of intermittent claudication and the second is a wide-spectrum antibacterial drug belonging to the tetracycline family. Both cilostazol and doxycycline were mainly tested in preclinical models of AD and they showed neuroprotective

## REFERENCES


properties in terms of inhibition of soluble Ab oligomerization and aggregation as well as improvement of antioxidant defense in the brain. Although the efficacy of these two drugs in AD subjects has not been definitively proven (some clinical trials are still ongoing), a possible reposition strategy should be considered for these two agents. LC1405 (7-pyrrolidinethoxy-40 methoxyisoflavone) is a novel potential H3 receptor antagonist which has been shown to reduce neurodegenerative damage, ameliorate cholinergic dysfunction and improve learning and memory in an APP/PS1 double transgenic mouse model of AD (Wang, Fang et al.). Morroni et al. reported the neuroprotective effects of a novel feruloyl-donepezil hybrid compound able to reduce neural damage and improve spatial cognition in mice. This approach is quite interesting, because these "chimeric" drugs take advantage of the pharmacological activities of each compound providing an efficient synergism in terms of neuroprotection.

## AUTHOR CONTRIBUTION

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

## ACKNOWLEDGMENTS

As Guest-Editors of this Research Topic, we thank all the authors of the contributions that have allowed us to give rise to an issue of great scientific interest and relevance. We also thank the devoted Reviewers who have provided the authors with effective and useful suggestions. Finally, we want to address great appreciation to all the members of the Editorial Offices of Frontiers who have certainly contributed together with authors and Reviewers in making this Research Topic a real success.

therapeutic avenues for Alzheimer's disease. Acta Neuropathol. Commun. 6 (1), 121. doi: 10.1186/s40478-018-0626-x


meta-analysis. Neurologia 32 (8), 523–532. doi: 10.1016/j.nrl.2016.02.016 Penninkilampi, R., Brothers, H. M., and Eslick, G. D. (2016). Pharmacological agents targeting g-secretase increase risk of cancer and cognitive decline in Alzheimer's disease patients: a systematic review and meta-analysis. J. Alzheimers Dis. 53 (4), 1395–1404. doi: 10.3233/JAD-160275

Selkoe, D. J., and Hardy, J. (2016). The amyloid hypothesis of Alzheimer's disease at 25 years. EMBO Mol. Med. 8 (6), 595–608. doi: 10.15252/emmm.201606210

Wang, Y., Yan, T., Lu, H., Yin, W., Lin, B., Fan, W., et al. (2017). Lessons from anti-amyloid-b immunotherapies in Alzheimer disease: aiming at a moving target. Neurodegener. Dis. 17 (6), 242–250. doi: 10.1159/000478741

Conflict of Interest: 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 © 2020 Mancuso and Gaetani. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Icariin and Its Metabolites as Potential Protective Phytochemicals Against Alzheimer's Disease

#### Cristina Angeloni<sup>1</sup> \*, Maria Cristina Barbalace<sup>2</sup> and Silvana Hrelia<sup>2</sup> \*

<sup>1</sup> School of Pharmacy, University of Camerino, Camerino, Italy, <sup>2</sup> Department for Life Quality Studies, University of Bologna, Bologna, Italy

#### Edited by:

Cesare Mancuso, Catholic University of the Sacred Heart, Italy

#### Reviewed by:

Luca Tiano, Polytechnical University of Marche, Italy Angela Maria Rizzo, University of Milan, Italy

\*Correspondence:

Cristina Angeloni cristina.angeloni@unicam.it Silvana Hrelia silvana.hrelia@unibo.it

#### Specialty section:

This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology

Received: 29 January 2019 Accepted: 04 March 2019 Published: 19 March 2019

#### Citation:

Angeloni C, Barbalace MC and Hrelia S (2019) Icariin and Its Metabolites as Potential Protective Phytochemicals Against Alzheimer's Disease. Front. Pharmacol. 10:271. doi: 10.3389/fphar.2019.00271 Alzheimer's disease (AD) is a neurodegenerative disorder affecting more than 35 million people worldwide. As the prevalence of AD is dramatically rising, there is an earnest need for the identification of effective therapies. Available drug treatments only target the symptoms and do not halt the progression of this disorder; thus, the use of natural compounds has been proposed as an alternative intervention strategy. Icariin, a prenylated flavonoid, has several therapeutic effects, including osteoporosis prevention, sexual dysfunction amelioration, immune system modulation, and improvement of cardiovascular function. Substantial studies indicate that icariin may be beneficial to AD by reducing the production of extracellular amyloid plaques and intracellular neurofibrillary tangles and inhibiting phosphodiesterase-5 activity. Moreover, increasing evidence has indicated that icariin exerts a protective role in AD also by limiting inflammation, oxidative stress and reducing potential risk factors for AD such as atherosclerosis. This mini-review discusses the multiple potential mechanisms of action of icariin on the pathobiology of AD including explanation regarding its bioavailability, metabolism and pharmacokinetic.

Keywords: icariin, icaritin, icariside, phytochemicals, Alzheimer's disease, oxidative stress, inflammation

## INTRODUCTION

Alzheimer's disease (AD) is a progressive irreversible neurodegenerative disease that is becoming a population aging-related concern for public health systems all over the world due to its direct and indirect costs (Dos Santos et al., 2018). Clinically, AD is mainly characterized by cognitive and memory decline and accounts for up to 70% of all dementia cases in the elderly (Hebert et al., 2003) affecting more than 35 million people worldwide (Povova et al., 2012). AD possesses a multifactorial etiology that involves different pathophysiological processes like abnormal protein aggregation, neurons and synapses degeneration, neuroinflammation, mitochondrial damage, oxidative stress and excitotoxicity, which interfere with several neurotransmitters signaling pathways (Behl and Ziegler, 2017).

In particular, two major hallmarks characterized AD: extracellular accumulation of amyloid β peptide (Aβ) and intraneuronal aggregation of tau protein also known as

neurofibrillary tangles (NFTs) (Calderon-Garcidueñas and Duyckaerts, 2017). Aβ is synthesized in the brain by the cleavage of the transmembrane amyloid precursor proteins (APP). Two secretases are responsible for Aβ production: β-secretase activity cleaving enzyme (BACE1) and the γ-secretase complex. BACE1 cleaves APP, producing an APP C-terminal fragment, which is subsequently cleaved within the transmembrane domain by γ-secretase at 40 or 42 residues, leading to the release of two different Aβ peptides Aβ1−<sup>40</sup> or the most abundant Aβ1−<sup>42</sup> (Thal et al., 2015) due to the variability in the C-terminus of Aβ (Tamagno et al., 2005). When APP is catabolized by other enzymatic activities (α- and Z-secretase complexes), Aβ is not produced.

In the normal brain, tau has 2 or 3 phosphate groups and binds to microtubules through electrostatic interaction (Jho et al., 2010). In AD, tau becomes hyperphosphorylated and the phosphorylation alters the net charge affecting the conformation of the microtubule binding region, thereby causing detachment of tau from microtubules that accumulates inside the neurons and aggregate to form NFTs (Trojanowski and Lee, 2002).

Beside Aβ plaques and NFTs, more than 50% of AD patients exhibit concurrent α-synuclein pathology (Twohig et al., 2018). α-synuclein is a 140 amino-acid protein abundantly expressed in neuronal presynaptic terminals. Different studies suggest that α-synuclein might be involved in the development of AD from the very early stages of Aβ pathology formation (Uéda et al., 1993; Vergallo et al., 2018).

Two other recognized pathological features of AD are neuroinflammation and oxidative stress (González-Reyes et al., 2017). In the normal brain, microglia does not produce proinflammatory molecules or reactive oxygen species (ROS), but in AD, Aβ induces the activation of astrocyte and microglia with a sustained release of proinflammatory molecules (Yucesoy et al., 2006). Elevated brain concentrations of inflammatory cytokines such as interleukin-1α (IL-1α), IL-β, IL-6, and tumor necrosis factor-α (TNF-α) have been associated with AD (Zilka et al., 2006). It has been shown that brain tissues in AD patients are exposed to oxidative stress (Gella and Durany, 2009), a condition characterized by an imbalance between ROS production and the endogenous antioxidative defense system.

Another very common feature of patients with AD is vascular dysfunction (Iadecola, 2005). It has been observed that a reduction in cerebral blood flow leads to a decline of Aβ clearance from the brain promoting neuronal degeneration and onset of AD (Zlokovic, 2011). On these bases, it is very important to improve endothelial function to prevent/counteract AD. Phosphodiesterase-5 inhibitors might interfere with the pathophysiological processes of AD such as neurovascular dysfunction (Sabayan et al., 2010). In particular, they can exert their positive effect on learning and memory by activating the NO/cGMP pathway (Puzzo et al., 2009) that produces a regulatory effect on endothelial function by relaxing blood vessels (Schulz et al., 2002). Moreover, cGMP could be used as a secondary messenger of the neurotransmitter acetylcholine (de Vente, 2004). Consequently, the inhibition of phosphodiesterase-5 is considered a novel approach to prevent/counteract AD.

Nowadays, several therapeutic strategies are used in clinical practice to counteract AD, however, all the drugs utilized are not able to alt or slow AD progression and possess many side effects (Mancuso et al., 2011). Therefore, there is a great interest in exploring new potential drug candidates for the treatment of AD. Antioxidant and anti-inflammatory activities of phytochemicals have been widely reported (Angeloni and Hrelia, 2012; Tarozzi et al., 2013; Angeloni et al., 2015, 2017). In this background, nutraceuticals are interesting therapeutic compounds to be explored as preventive and beneficial agents for AD. Icariin is a flavonoid present in Herba Epimedii, a traditional Chinese herbal medicine. Icariin has been shown to possess several biological activities. This mini-review focuses on the role of icariin and its metabolites in AD. Substantial studies indicate that icariin and its metabolites may be beneficial to AD by reducing the production of extracellular amyloid plaques and intracellular NFTs and inhibiting phosphodiesterase-5 activity. Moreover, increasing evidence has indicated that icariin exerts a protective role in AD also by limiting inflammation, oxidative stress and reducing potential risk factors for AD such as atherosclerosis.

## ICARIIN

Icariin (molecular formula: C33H40O15, molecular weight: 676.67 g/mol) is a prenylated flavonoid considered as the main bioactive of Herba Epimedii, a traditional Chinese herbal medicine used since thousands of years. Giving its therapeutic effects such as osteoporosis prevention, ameliorating sexual dysfunction, modulation of immune system, and improvement of cardiovascular function, Herba Epimedii has been included into Chinese pharmacopeia, indicating icariin as quality marker (Committee, 2010; Tang and Eisenbrand, 2011).

In tradition, icariin has been evidenced to possess antiinflammatory, antioxidant, antidepressant and aphrodisiac effects (Liu et al., 2004; Tang and Eisenbrand, 2011; Liu et al., 2015). In addition, several in vitro and in vivo reports show many pharmacological activities elicited by icariin.

In different animal models of osteoporosis icariin demonstrated significant osteogenic effects mediated by Wnt/β-catenin and bone morphogenetic protein (BMP) signaling pathways (Wei et al., 2011; Li et al., 2013, 2014). Moreover, a clinical trial, conducted in postmenopausal women, showed a positive effect of icariin on bone mineral density (Zhang et al., 2007). Preliminary research suggests that icariin could be useful for treating erectile dysfunction as it was active on cavernous smooth muscle cells (Ning et al., 2006). The most promising effect of icariin at cardiovascular level is the promotion of stem cell differentiation into beating cardiomyocytes which suggests its likely application in cardiac cell therapy or tissue engineering (Jin et al., 2010; Zhou L. et al., 2013; Zhou et al., 2014). Moreover, icariin has also been evaluated for prevention and treatment of thrombosis in atherosclerosis as it reduces platelet adhesiveness and aggregation besides a decrease in serum cholesterol (Zhang et al., 2013). Multiple studies have indicated that icariin has been found to be beneficial to cancer (Zhang Y. et al., 2014), rheumatoid arthritis (Sun et al., 2013), immune

system (Li et al., 2011), liver disease (Lee et al., 1995), diabetic nephropathy (Qi et al., 2011), sedative (Pan et al., 2005) and so on. Icariin has been found to possess multiple neuroprotective effects: it improves survival and function of neurons (Guo et al., 2010; Li F. et al., 2010) and triggers their self-renewal through neural stem cells (Huang et al., 2014).

## PHARMACOKINETICS OF ICARIIN

Despite the numerous studies on icariin, the main challenge remains its very low oral bioavailability due to the physicochemical characteristics (Chen et al., 2008), and P-glycoprotein-mediated efflux in intestinal mucosa (Zhang Y. et al., 2012). Different studies indicated the importance of icariin hydrolysis by lactase phlorizin hydrolase in the small intestine and by microbiota β-glucosidase to release metabolites before its absorption (Zhao et al., 2010; Chen et al., 2011; Qian et al., 2012). In addition, icariin is a prenylated flavonoid and it has been reported that the prenyl-moiety decreases the bioavailability and plasma absorption of prenylated flavonoids (Chen et al., 2014). In this regard the presence of icariin in urine was less than 0.425% showing that probably the most of icariin is metabolized and excreted as metabolites (**Figure 1**) (Yu et al., 2016). Fortunately, the modern techniques offer a range of methods to overcome this issue. To increase icariin bioavailability, researchers have developed several drug delivery systems such as combining icariin with snailase (an exogenous hydrolase) to improve intestinal hydrolysis (Liu et al., 2017), encapsulating icariin into liposome (Yang et al., 2012), producing icariin/hydroxylpropylbeta-cyclodextrin inclusion complex that enhances intestinal absorption probably through a solubilizing effect and/or the inhibition of P-glycoprotein (Zhang Y. et al., 2012).

Several methods have been used to investigate the pharmacokinetic characteristics of icariin and its metabolites like UFLC-TOF/MS (Qian et al., 2012), HPLC-MS/MS (Cheng et al., 2015; Sun et al., 2018), a liquid chromatographic

method combined with electrospray ionization tandem mass spectrometry (Xu et al., 2017), and GC-MS (Shen et al., 2007).

After oral administration of Epimedium extract, the HPLC-MS/MS analysis of rat plasma revealed a rapid absorption and elimination of icariin, with a t1/<sup>2</sup> ranged from 0.5 to 1 h; meanwhile the elimination of icariside II, which is chemically the monoglycoside form of icariin and in vivo predominant bioactive compound, from plasma takes a longer time from 3 to 18 h (Sun et al., 2018). Another study using a LC-MS method reported icariside II and icaritin (the aglycone form of icariin) as the major metabolites of icariin in rat feces after both oral and intramuscular administration (Xu et al., 2017). Interestingly, analyzing the data from various tissues (liver, heart, spleen, lung, kidney, brain, testicle, uterus and ovary) of male and female rats, the distribution of icariin differs in total tissue concentration (much higher in male rats than female rats) with the exception of genital organs (higher in females) (Xu et al., 2017). Interestingly, the pharmacokinetic profile of pure icariin depends on the route of administration. After oral administration, icariside II is the main form in rat plasma as 91.2% icariin is converted in it, but after intravenous injection only 0.4% of icariin is transformed in icariside II, demonstrating the role of intestinal microbiota in metabolizing icariin administrated per o.s. (Cheng et al., 2015). Indeed, it has been reported that icariin is metabolized to icaritin via icariside I and II by the rat intestinal microbiota (Zhou J. et al., 2013). Recently, a study on icariin metabolism by human microbiota evidenced a different pattern of metabolites depending on bacterial strains, and interestingly icariside I was not detected (Wu et al., 2016). In particular, the metabolites produced by human bacteria were icariside II, icaritin, and desmethylicaritin where the 4<sup>0</sup> -methyl of icaritin is removed. In human serum, the peak of icaritin was observed at 8 h after Epimedium decoction intake, suggesting that the conversion of icariin to icaritin is primarily at intestinal level, differently desmethylicaritin was not observed (Shen et al., 2007). Unfortunately, studies of icariin metabolism and distribution in humans are really few and should be improved to have a comprehensive view of the icariin pharmacokinetics properties. What emerged from these studies is that icariin is scarcely present in plasma because of its rapid elimination, and the tissue distribution of icariin in the brain is scarce, incoherently with the largely literature supporting neuroprotective effects (Xu et al., 2017; Zhang et al., 2017). Therefore, a possible explanation to this controversial issue could be that the observed biological effects are, in part, mediated by icariin metabolites. However, it is important to further improve the knowledge of the possible effects and mechanisms of icariin metabolites at cerebral level and against AD, and to develop and characterize novel delivery systems to increase the uptake and distribution of icariin in the brain.

## ICARIIN AND ITS METABOLITES IN Aβ NEUROTOXICITY

The effects of icariin in counteracting Aβ deposition and Aβ induced neurotoxicity have been largely investigated. The first

study that reported an effect of icariin on Aβ was carried out in rats challenged with Aluminum (Luo et al., 2007). In particular, icariin (60 and 120 mg/kg) administrated by gavage for 3 months significantly attenuated Aβ1−<sup>40</sup> production induced by Aluminum treatment. Besides, icariin counteracted learning and memory deficit, increased SOD activity and decreased MDA levels. These findings were further deepened by the same authors in a different AD model (Nie et al., 2010). In rats treated with Aβ25−35, icariin improved the learning and memory deficits by both a decreased production of insoluble fragments of Aβ due to the downregulation of β-secretase expression (BACE1) and to its antioxidatant activity. In the same experimental model, icariin nearly completely suppressed the abnormal inward calcium currents induced by Aβ25−<sup>35</sup> in a dose dependent manner suggesting a potential neuroprotective effect of icariin on Aβ25−35-induced neurotoxicity via the balance of intracellular calcium homeostasis (Li L. et al., 2010). Urano and Tohda (2010) showed that icariin (50 µmol/kg) administrated for 8 days was effective in improving spatial memory impairment in 5×FAD rats, an AD model characterized by an elevated production of Aβ1−<sup>42</sup> (Oakley et al., 2006). In a rodent APP/PS1 model of cerebral amyloidosis for AD a icariin (100 mg/kg by daily gavage) treatment for 10 days significantly attenuated Aβ deposition and restored impaired nesting behavior (Zhang Z. Y. et al., 2014). Zhang L. et al. (2014) demonstrated that icariin (100 µmol/kg for 6 months) counteracted Aβ burden and deposition in the hippocampus of APPV7171 transgenic mice by reducing the expression of both APP and BACE1. In the same experimental model, icariin (30, 60 mg/kg twice a day for 4 months) improved learning and memory of APP/PS1 mice in Y-maze tasks, reduced Aβ deposition, and down-regulated both APP and (phosphodiesterase-5) PDE5 (Zhang et al., 2010). Of note, the inhibition of PDE5 stimulated the NO/cGMP signaling pathway as evidenced by an increased expression of three nitric oxide synthase (NOS) isoforms, together with increased NO and cGMP levels in the hippocampus and cortex of mice. Similar results were obtained by Li F. et al. (2015) in Tg2576 mice treated with icariin (60 mg/kg) for 3 months. Icariin improved spatial working memory, reduced the levels of both Aβ1−<sup>40</sup> and Aβ1−42, downregulated APP expression and enhanced neurogenesis. These aspects were further investigated using a triple-transgenic mouse model of Alzheimer's disease (3× tg-AD) (Chen et al., 2016). An icariin treatment (65 mg/kg) for 6 months enhanced neuronal cell activity as identified by an increase of brain metabolite N-acetylaspartate and ATP production, preserved the expressions of mitochondrial key enzymes such as cytochrome c oxidase subunit 4 (COX IV) and pyruvate dehydrogenase E1 component subunit alpha (PDHE1α), and postsynaptic density protein 95 (PSD95), reduced Aβ plaque deposition in the cortex and hippocampus, and down-regulated BACE1 expression. Intragastric administration of icariin reversed the decreases in PSD95, brain derived neurotrophic factor (BDNF), pTrkB, pAkt, and pCREB expressions induced by Aβ1−<sup>42</sup> injection in rats suggesting that icariin may improve synaptic plasticity through the BDNF/TrkB/Akt pathway (Sheng et al., 2017). The ability of icariin to increase pCREB was also observed in a senescence accelerated prom mouse model (SAMP8) characterized by early Aβ deposition (Zhang Z. et al., 2012). Moreover, icariin decreased the level of Aβ in rat hippocampus subjected to permanent occlusion of bilateral common carotid arteries (BCCAO) (Li W. X. et al., 2015), a model used to mimic cerebral hypoperfusion that occurs in vascular dementia and Alzheimer's. This reduction of Aβ deposition was related to different effects such as the down-regulation of APP and BACE1, and an increased expression of insulin-degrading enzyme (IDE) and disintegrin and metalloproteinase domain 10 (ADAM10) in rat hippocampus. In an in vitro model, icariin (40–160 µg/mL) was able to dose-dependently protect cortical neurons against Aβ1−<sup>40</sup> induced damage by enhancing the expression of CART and activating ERK signaling pathway (Sha et al., 2009). In addition, icariin 0.01 µM was able to counteract the axon and dendritic shortening induced by Aβ1−<sup>42</sup> in rat cortical neurons (Urano and Tohda, 2010). In cultured rat PC12 cells icariin (20 µM) counteracted apoptosis induced by Aβ25−<sup>35</sup> and this effect appeared to be mediated by the activation of the PI3K/Akt signaling pathway (Zhang et al., 2015). In agreement with these results, Zeng et al. (2010b) observed that icariin (5–20 µM) dose dependently reduced cell death and apoptosis in PC12 cells exposed to Aβ25−35. In addition, the authors demonstrated that this protection is partially due to activation of the PI3K/Akt signaling pathways that induces the inhibition of GSK-3β and, consequently, reduces tau protein hyperphosphorylation. Icaritin, another compound extracted from Epimedium, demonstrated to be neuroprotective against the toxicity induced by Aβ25−<sup>35</sup> in primary rat cortical neurons (Wang et al., 2007). In particular, icaritin increased cell viability and reduced apoptosis by an estrogen receptor dependent mechanism and by activating ERK1/2 MAPK pathway.

## ICARIIN AND ITS METABOLITES IN OXIDATIVE STRESS

Oxidative stress plays a crucial role in the pathogenesis of many neurodegenerative diseases including AD. The antioxidant activity of icariin has been demonstrated in primary cortical neurons exposed to H2O<sup>2</sup> (Zhang et al., 2010). In particular, icariin (1.2 µM) counteracted H2O2-induced neurotoxicity by reducing ROS production, increasing mRNA expression of the antioxidant enzymes catalase and peroxiredoxin 1 (PRX1) by a mechanism mediated by SIRT1 up-regulation. Icariin (5-50 µM) attenuates LPS-induced oxidative stress in primary microglial cells reducing ROS level in a dose dependent manner (Zeng et al., 2010a). In an in vivo study carried out in rats, icariin showed a protective effect against learning and memory deficit induced by aluminum by increasing SOD activity and decreasing malondialdehyde (MDA) levels (Luo et al., 2007). It has been shown that iron overload is involved in the progression of AD (El Tannir El Tayara et al., 2006). Excessive iron levels lead to increased oxidative stress through the Fenton reaction (Rolston et al., 2009). In order to counteract iron overload, APP/PSI mice were treated with icariin (120 mg/kg) for 3 months. Icariin reduced

iron overload and protected mice against oxidative stress reducing lipid peroxidation and increasing the activity of the antioxidants enzymes SOD and glutathione peroxidase (Zhang et al., 2018).

Icariside also demonstrated to be effective in counteracting oxidative stress. Icariside II attenuated Aβ25−35-induced intracellular and mitochondrial ROS generation in PC12 cells (Liu et al., 2018).

## ICARIIN AND ITS METABOLITES IN NEUROINFLAMMATION

As previously underlined, AD is associated with neuroinflammation, which is triggered by microglia activation in the brain (Heneka et al., 2015). It is now widely accepted that these brain cells are likely to contribute to the mechanisms of neuronal damage and cognitive loss (Sarlus and Heneka, 2017). Icariin has been reported to have an anti-inflammatory effect on primary rat microglial cultures activated by LPS (Zeng et al., 2010a). In particular, icariin (5–50 µM) reduced the release of nitric oxide (NO), prostaglandin E (PGE)-2 in a dose dependent manner and down-regulated the expression of proinflammatory cytokines such as tumor necrosis factor (TNF)-α, interleukin (IL)-1β and IL-6. Icariin also inhibited the protein expression of inducible nitric oxide synthase (iNOS) and cyclooxygenase (COX)-2. The authors showed that the mechanisms behind this anti-inflammatory effect is the inhibition of the TAK1/IKK/NF-κB and JNK/p38 MAPK pathways. The ability of icariin (100 mg/kg for 10 days) to counteract microglia activation was also observed in the cortex and hippocampus of APP/PSI mice (Zhang Z. Y. et al., 2014) and these data were corroborated by a recent study of Zhang et al. (2018) that showed that icariin (120 mg/kg for 3 months) reduced neuroinflammation in the cerebral cortex of APP/PSI transgenic mice inhibiting the release of IL-6, IL-1β and TNF-α.

In an AD model obtained by ICV injection of STZ in rats, icariside II (10 mg/kg for 21 days) reduced the expression of TNF-α, IL-1β, COX-2, TGF-b1 by preventing the degradation of IkB-α and NFK-B p65 phosphorylation (Yin et al., 2016). These findings were confirmed by the results of Deng et al. (2017) showing that icariside II (20 mg/kg for 15 days) attenuated Aβ25−35 induced expression of TNF-α, IL-1β, COX-2, and iNOS in rat hippocampus.

## CONCLUSION

fphar-10-00271 March 15, 2019 Time: 17:44 # 6

As reported, icariin, besides its use in complementary and alternative traditional Chinese medicine, is a very promising molecule to counteract many pathophysiological processes of AD, having an impact on Aβ production and removal pathways, on oxidative stress mediated effects and on neuroinflammatory cascade (**Figure 2**).

The practical possibilities of AD prevention and counteraction with this pleiotropic compound should be further investigated in clinical studies and represent a challenge for future researches.

## REFERENCES


## AUTHOR CONTRIBUTIONS

CA conceived the idea. CA and MB prepared the manuscript. SH reviewed the drafts and provided important information for the completion of this manuscript.

## FUNDING

This work was supported by MIUR-PRIN 2015 (No. 20152HKF3Z).

Alzheimer's disease from a neuroinflammatory and oxidative stress perspective. Front. Mol. Neurosci. 10:427. doi: 10.3389/fnmol.2017.00427


depression in rats and is associated with the regulation of hippocampal neuroinflammation. Neuroscience 294, 193–205. doi: 10.1016/j.neuroscience. 2015.02.053



disease mice. Neural Regen. Res. 13, 731–736. doi: 10.4103/1673-5374.23 0302


**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 © 2019 Angeloni, Barbalace and Hrelia. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

## 7-Pyrrolidinethoxy-4<sup>0</sup> - Methoxyisoflavone Prevents Amyloid β–Induced Injury by Regulating Histamine H3 Receptor-Mediated cAMP/CREB and AKT/GSK3β Pathways

Linlin Wang1,2† , Jiansong Fang<sup>3</sup>† , Hailun Jiang<sup>1</sup> , Qian Wang<sup>1</sup> , Situ Xue<sup>1</sup> , Zhuorong Li<sup>1</sup> \* and Rui Liu<sup>1</sup> \*

#### Edited by:

Cesare Mancuso, Università Cattolica del Sacro Cuore, Italy

#### Reviewed by:

Damiana Leo, University of Mons, Belgium Domenica Donatella Li Puma, Catholic University Medical School, Italy

#### \*Correspondence:

Zhuorong Li lizhuorong@imb.pumc.edu.cn Rui Liu liurui@imb.pumc.edu.cn †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Neuropharmacology, a section of the journal Frontiers in Neuroscience

Received: 19 December 2018 Accepted: 21 March 2019 Published: 10 April 2019

#### Citation:

Wang L, Fang J, Jiang H, Wang Q, Xue S, Li Z and Liu R (2019) 7-Pyrrolidinethoxy-4<sup>0</sup> - Methoxyisoflavone Prevents Amyloid β–Induced Injury by Regulating Histamine H3 Receptor-Mediated cAMP/CREB and AKT/GSK3β Pathways. Front. Neurosci. 13:334. doi: 10.3389/fnins.2019.00334 1 Institute of Medicinal Biotechnology, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China, <sup>2</sup> Institute of Materia Medica, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, China, 3 Institute of Clinical Pharmacology, Guangzhou University of Chinese Medicine, Guangzhou, China

In studies on the treatment of Alzheimer's disease (AD), in which cognition is enhanced even modestly or selectively, it has been considered that the histamine H3 receptor (H3R) may be a potential target. In this study, we aimed at evaluating the ability of 7-pyrrolidinethoxy-4<sup>0</sup> -methoxyisoflavone (indicated as LC1405), a novel potential H3R antagonist identified from our H3R antagonist screening system, to ameliorate amyloid β (Aβ)-induced cognitive deficits, and to explore the underlying mechanisms that are related to H3R-modulated signaling. Our results demonstrated that LC1405 effectively reduced the progression of Aβ-associated disorders, such as improved learning and memory capabilities, preserved tissues from suffering neurodegeneration and ultrastructural abnormalities, and ameliorated cholinergic dysfunction in an APP/PS1 double transgenic mouse model of AD. In an in vitro model, LC1405 protected neuronal cells against copper-induced Aβ toxicity, as demonstrated by the improvement in cell viability and decrease in neuronal apoptotic ratio. In addition, treatment with LC1405 resulted in the up-regulation of acetylcholine (ACh) or histamine release and provided neuroprotection through cellular signaling cascades involving H3Rmediated cAMP/CREB and AKT/GSK3β pathways. Furthermore, the beneficial effects of LC1405 on Aβ-mediated toxicity and H3R-mediated cAMP/CREB and AKT/GSK3β axes were reversed after pharmacological activation of H3R. In conclusion, our results demonstrated that LC1405 blocked Aβ-induced toxicity through H3R-modulated signaling transduction both in vitro and in vivo. The results also suggested that LC1405 might have translational potential as a complementary therapy to control disease progression in AD patients who developed cognitive deficits with H3R-related ACh neurotransmission abnormality.

Keywords: acetylcholine, Alzheimer's disease, amyloid beta-peptide, cyclic AMP response element binding protein, histamine H3 receptor

## INTRODUCTION

fnins-13-00334 April 8, 2019 Time: 12:4 # 2

Alzheimer's disease (AD), the most common cause of dementia in elderly population, is characterized by complicated and multifactorial pathophysiological alterations, primarily including senile plaque deposits, Tau protein hyperphosphorylation, high oxidative stress, metal ion dyshomeostasis, and neurotransmitter system irregularities (Ballard et al., 2011; Chiang and Koo, 2014). Currently, the principal treatment to combat AD in clinical practice involves the administration of acetylcholinesterase (AChE) inhibitors and the N-methyl-D-aspartate (NMDA) receptor antagonist memantine, which resulted in limited symptomatic improvement. However no effective treatment strategy is available that results in recovery or even retardation in the progression of the disease. Therefore, it is of utmost importance to develop novel and effective therapies for the treatment of AD.

Over the past decade, preclinical studies and clinical trials identified histamine H3 receptor (H3R), a histamine receptor subtype that is predominantly expressed in neurons of the central nervous system (CNS), as a possible target for cognitionenhancing candidates that may have beneficial effects on mildto-moderate AD (Esbenshade et al., 2006; Bonaventure et al., 2007; Mani et al., 2017). Among the diverse H3R antagonists investigated up to now, two small H3R antagonists, ABT-239 and A-431404, showed procognitive effects in ketamine and MK-801-induced animal models (Brown et al., 2013). Pre-clinical studies demonstrated that several H3R antagonists ameliorate cognitive deficits and related behaviors in a substantial number of animal models characterized by learning or memory dysfunction (Browman et al., 2004).

In AD, H3R in the prefrontal cortex and hippocampus acts as a presynaptic auto-receptor coupled to Gαi/o-proteins that controls the synthesis and release of histamine, and as a heteroreceptor on histaminergic and non-histaminergic neurons in regulating the release of other neurotransmitters, including acetylcholine (ACh), dopamine, glutamate, norepinephrine, and gamma-aminobutyric acid (GABA) (Clapham and Kilpatrick, 1992; Ligneau et al., 1998; Bacciottini et al., 2002; Medhurst et al., 2007). Furthermore, within the neuronal intracellular signal transduction H3R participates in a variety of pathways (Clapham and Kilpatrick, 1992; Esbenshade et al., 2008; Bhowmik et al., 2014). When ligands bind to H3R-activated Gi proteins, adenylyl cyclase (AC) is inhibited, leading to decreased levels of cyclic adenosine monophosphate (cAMP) and reduced phosphorylation-activation of cAMP response element binding protein (CREB), a transcription factor that is closely related to cognitive functions (Bhowmik et al., 2014). In addition, pathological alterations of phosphatidylinositol-3-kinase/protein kinase B/glycogen synthase kinase 3β (PI3K/AKT/GSK3β) signaling transduction are due to β-amyloid (Aβ)-stimuli via a variety of signal transduction pathways, thereby amplifying Aβ-induced pathogenic responses in the brain through abnormal H3R agonistic reactions (Bhowmik et al., 2014). Accordingly, a large number of previous studies demonstrated that targeted activation of H3R in neurons resulted in an accelerated cognitive decline and aggravated Aβ-induced neuronal perturbation in β-amyloid precursor protein (APP) transgenic mice, accompanied by progressive loss of cholinergic neurons and destructive signaling pathways involving cAMP/CREB and AKT/GSK3β cascades (Bhowmik et al., 2014; Bardgett et al., 2011; Bitner et al., 2011). Therefore, the central role of H3R in AD suggests that it may be an attractive target in the development of novel therapies against diseases using H3R antagonists.

In the present study, we have investigated cognitive improvement and neuronal protection using a potential non-imidazole H3R antagonist, 7-pyrrolidinethoxy-4<sup>0</sup> methoxyisoflavone (**Figure 1A**, indicated as LC1405) that has affinity with human H3R and high H3R inhibition power in vitro. We investigated its action against Aβ-induced neurotoxicity, and the underlying mechanisms of action against Aβ toxicity correlated with H3R-modulated signaling both in APP and presenilin 1 (PS1) double transgenic mice and copper-induced Aβ toxicity in APP Swedish mutation overexpressing SH-SY5Y cells.

## MATERIALS AND METHODS

## Animals and Drug Treatment

Heterozygous APPswe695/PSEN1dE9 (APP/PS1) transgenic mice and age-matched wild-type (WT, C57BL/6) littermates were obtained from the Model Animal Research Center of Nanjing University (Nanjing, China). All animal studies were approved by the Institute of Medicinal Biotechnology, Chinese Academy of Medical Sciences (Beijing, China) and performed in accordance with ethical guidelines of the Experimental Animal Care and Use Committee.

Nine-month-old WT mice and APP/PS1 transgenic mice were randomly divided into four groups, including a WT control group (n = 13, seven males and six females), LC1405-treated WT group (n = 13, seven males and six females), APP/PS1 control group (n = 13, seven males and six females), and LC1405 treated APP/PS1 group (n = 13, seven males and six females). Mice in the WT+LC1405 and APP/PS1+LC1405 groups received an intragastric administration of LC1405 for 6 d/week at a dosage of 3 mg/kg. LC1405 was dissolved in 20% hydroxypropylβ-cyclodextrin (CMC-Na), thus, mice in the control group received 20% CMC-Na according to the same modality. Drug treatment was performed for 20 weeks.

After behavioral tests, mice were divided into three groups for the parallel evaluation of different parameters. Three mice per group were transcardially perfused with normal saline solution, followed by 4% paraformaldehyde (PFA). Then, brains were collected and subjected to immunohistochemical staining and transmission electron microscopy. Four mice per group were chosen for the detection of oxidative stress, cholinergic activity, Aβ levels, and cAMP content in the brain. Four mice per group were prepared for quantification of H3R-mediated signaling transduction in the cortex and hippocampus, such as phosphorylated CREB, AKT, and GSK3β using ELISA. The brain of the remaining mice was quickly removed and stored at −80◦C until further experiments. The acetylcholine and histamine levels were evaluated in another two WT mouse groups

(n = 4, two males and two females) that received an intragastric administration of LC1405 at a dosage of 3 mg/kg dissolved in CMC-Na or only CMC-Na as control. Mice were randomly selected for both the initial division into treatment groups and the subsequent selection for the three experiments.

## Behavioral Assessment of Learning and Memory

Behavioral tests were performed when mice were thirteen months old, following 20 weeks of treatment with LC1405 or vehicle control. The Morris Water Maze (MWM) test is a classical visualspatial learning technique used in rodents for assessing learning and memory capabilities hippocampus-dependent (Vorhees and Williams, 2006). Briefly, during the acquisition trial, mice were subjected to four training trials per day for five consecutive days. Both the escape latency (time required to find the platform) and swimming speed were recorded. At the end of the last trial, the platform was removed to proceed for the probe trial, which was carried out at 2 and 48 h post-training. The time the mice spent in the target quadrant and the frequency of passing through the platform were recorded.

Contextual short-term memory of the mice was assessed using a passive avoidance test (Nasehi et al., 2012). The apparatus consisted of illuminated (bright) and non-illuminated (dark) compartments. During the acquisition trial, mice were initially placed in the bright compartment for a maximum of 300 s, and upon entering the dark compartment received an electric foot shock (0.3 mA, 2 s). As a measure of memory retention, the initial latency of mice to enter the dark compartment and error times were recorded at 24 h after acquisition training.

## Immunohistochemical Staining

The brain was embedded in paraffin and dissected into 8 µM sections. Fluoro-Jade B (FJB, Histochem, Jefferson, AR, United States) and Thioflavin S (Thio S, Sigma, St. Louis, MO, United States) were used to evaluate neuronal degeneration and fibrillar Aβ level in hemi-brain tissue sections, respectively, and were performed using standard histological techniques as previously reported (Liu et al., 2014, 2018). Brain pathological features were evaluated on the sections under a fluorescence microscope (Olympus IX70, Olympus, Tokyo, Japan). The number of FJB-positive neurons and Thio S-positive neurons were manually evaluated as the number of neurons in 1 mm<sup>2</sup> of the cerebral cortex and hippocampus region. Cell count was obtained by averaging the counts from 10 sections per mouse.

## Ultrastructural Analysis by Transmission Electron Microscopy

The prefrontal cortex and hippocampus were carefully harvested from the PFA-perfused brain of the experimental mice and placed overnight in the fixative [20 mL of 2.5% glutaraldehyde (Merck, Darmstadt, Germany) and 2.0% PFA (Beijing Chemical Works, Beijing, China) in 0.15 M cacodylate buffer (Merck, Darmstadt, Germany)]. Ultrathin sections were cut as previously described (Yu et al., 2015), and a LEO 906 transmission electron microscope (Zeiss, Oberkochen, Germany) was used for ultrastructural imaging.

## Measurement of Oxidative Stress and Cholinergic Activity in the Brain

The hemi-brains of APP/PS1 and WT mice were homogenized via ultrasonication, and centrifuged at 12,000 × g for 10 min at 4 ◦C. The concentration of malondialdehyde (MDA), superoxide dismutase (SOD) and glutathione peroxidase (GSH-Px) within the homogenates was determined using commercial assay kits (Jiancheng Biotech, Nanjing, Jiangsu, China) in accordance with the corresponding manufacturer's instructions. ACh levels and the AChE activity were determined using the ultrasensitive Amplex@ red ACh/AChE assay kit (Molecular Probes, Paisley, United States) according to the manufacturer's guidelines.

## Acetylcholine Quantification Using High-Performance Liquid Chromatography

For conventional analysis, brain tissue ACh levels were quantified using high-performance liquid chromatography (HPLC) coupled to electrochemical detection. In brief, ACh was separated on a BetaBasic C18 column (Thermo-Hypersil, Waltham, MA, United States; 150 mm × 1.0 mm; particle size, 3 µm) at a temperature of 25 ◦C using a mobile phase consisting of 100 mM Na2HPO4, 2.0 mM sodium octanesulfonic acid, 0.5 mM tetramethylammonium chloride and 100 µL of Reagent MB microbicide (ESA Inc., Chelmsford, MA, United States), adjusting the pH to 8.0. The separated ACh fraction was placed in an enzyme reactor (Bioanalytical Systems) which yielded H2O<sup>2</sup> for the detection with an enzyme-coated glassy carbon electrode with a potential set at 100 mV versus silver/silver chloride (Ag/AgCl). The sensitivity for ACh detection was approximately 0.5 fmol/15 µL of sample. Chromatographic data were collected and quantified by comparison with known standard concentrations.

## Histamine Determination Using High-Performance Liquid Chromatography

Brain histamine levels were determined by HPLC and quantified by fluorometric detection. Brains were homogenized in 3% perchloric acid solution containing 0.5 mM EDTA, and then centrifuged at 5500 × g for 15 min at 4◦C. The supernatant was collected and histamine was separated using a Waters Atlantis C18 HPLC column (Milford, MA, United States; 150 mm × 3.0 mm; particle size, 3 µm) with a mobile phase containing 160 mM KH2PO4, 0.45 mM octanesulfonic acid, 1% methanol and 0.1 mM EDTA (pH 4.5), delivered at 0.5 to 0.7 mL/min. Using a T-piece, the eluent line was connected to the reagent line, through which a 0.02% solution of o-phthaldialdehyde (OPA) was delivered in 0.15 M NaOH at a rate of 0.60 mL/min. The OPA reagent was mixed with the eluent in a mixing coil of metal tubing (outer diameter, 1.1 mm; inner diameter, 0.55 mm; length, 1 m) that enabled the derivatization

reaction at room temperature. The fluorescence of the reaction product was measured using a fluorometric detector (Ex/Em, 350 nm/450 nm). Chromatographic data were collected and quantified by comparison with known standard concentrations. The sensitivity for histamine detection was approximately 20 fmol/20 µL of sample.

## Cell Culture and Treatments

To evaluate the neuroprotective effects of LC1405, the Swedish mutant form of the human APP695 gene was stably transfected into SH-SY5Y cells that were purchased from the ATCC (ATCC <sup>R</sup> CRL-2266, Manassas, VA, United States), to establish an AD in vitro model (named APPsw cells). In this cell model, copper acts as a promotor for triggering Aβ-mediated neurotoxicity when added as a stimulator into the culture medium (Zhang et al., 2006; Zhao et al., 2013). Cells were cultured in DMEM/F12 medium (Invitrogen, Carlsbad, CA, United States) supplemented with 10% fetal calf serum (Invitrogen, Carlsbad, CA, United States) and incubated at 37◦C in a humidified chamber containing 5% CO2. The detailed protocol and groups were as follows. Cells were randomly divided into two groups: one group was supplemented with 300 µM copper, whereas the other was not (indicated as control cells). The group supplemented with copper was divided into subgroups based on LC1405 concentrations as follows: 0 µM (copper-treated APPsw cells), 0.03 µM, 0.1 µM, 0.3 µM, 1.0 µM, 3.0 µM, and 10.0 µM. Different concentrations of LC1405 were added at the start of copper-initiated injury, then cells were incubated for 24 h at 37◦C. To determine whether H3R inhibition was involved in the neuroprotective effects of LC1405 against Aβ toxicity, the specific agonists of H1 to H4, 2-(3-trifluoromethylphenyl) histamine (FMPH), amthamine dihydrobromide, (R)-(α)-(-)-methylhistamine dihydrobromide (RAMH) and VUF-8430 were used. Cells were pretreated with the specific agonists at 1.0 µM for 30 min at 37◦C before being treated with LC1405.

## Cell Viability Assay

Viability was determined by 3-(4,5-dimethylthiazol-2-yl)- 5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium (MTS, Promega, Madison, WI, United States). In brief, cells were incubated for 4 h at 37◦C with an appropriate amount of MTS according to the manufacturer's instructions. The soluble product formazan was detected using a Spark 20M multimode microplate reader (Tecan Group Ltd., Mannedorf, Switzerland) at 490 nm.

## Cell Immunofluorescence Assay

Cell immunofluorescence is a sensitive fluorescence-based multiparametric technology that is used to determine the expression and activity of proteins within a cell. Cells were seeded in a 96-well plate at a density of 8000 cells/well in 200 µL medium/well and subjected to all treatments described in the section 2.8. The apoptotic ratio in vitro was determined by simultaneous staining with acridine orange (AO) and ethidium bromide (EB) (Sigma-Aldrich, St. Louis, MO, United States). After treatment with LC1405 at different concentrations, 100 µg/mL AO and 100 µg/mL EB were added to the cells and incubated for 15 min at 37◦C. The expression of β-APP, p-CREB, and p-PI3K was quantified using cell immunofluorescence routine procedures as previously described (Liu et al., 2012). The following primary polyclonal rabbit antibodies were used: anti-β-APP [1:500, Cell Signaling Technology (CST), Danvers, MA, United States], anti-p-CREB (Ser133) (1:120, CST), and anti-p-PI3K p85α (Y607) (1:80, Abcam, Cambridge, MA, United States). The fluorescent secondary antibodies used were goat anti-rabbit conjugated with Alexa Fluor 488 or Alexa Fluor 546 (1:1000, Invitrogen, Carlsbad, CA, United States). The apoptotic ratio and mean fluorescence intensity were determined and analyzed using a Cellomics ArrayScan VTI HCS Reader (Cellomics Inc., Pittsburgh, PA, United States) running Morphology Explorer BioApplication software for the average of 20 fields of view in each selected well.

## Determination of Aβ and cAMP Concentrations

After the treatment, mouse hemi-brains and APPsw cells were separately homogenized using ultrasonication and centrifuged at 12,000 × g for 10 min at 4◦C. An aliquot of mouse hemibrains or APPsw cells after homogenization was resuspended in 2% sodium dodecyl sulfate (SDS) containing protease inhibitors. After centrifugation, the supernatant was collected for the detection of soluble Aβ. The remaining SDS-insoluble pellet was sonicated, dissolved in 70% formic acid, and centrifuged for 60 min at 100,000 × g, and the supernatant was collected for the detection of insoluble Aβ. The concentration of both soluble and insoluble Aβ was quantified using commercially available ELISA kits (BioSource, Camarillo, CA, United States) specific for the detection of human Aβ1−40/42. In addition, soluble Aβ oligomers (oAβ) were detected using western blot as previously described (Liu et al., 2018). Another aliquot of mouse hemi-brains was homogenized in RIPA buffer (Cell Signaling Technology, Danvers, MA, United States) containing protease inhibitor, phosphatase inhibitor, and phenylmethyl sulfonylfluoride (PMSF). Forty µg protein per lane were run on polyacrylamide gel, transferred onto a polyvinylidenedifluoride membrane, blocked with 5% BSA in Tris-buffer saline containing 0.1% Tween-20 (TBST) for 2 h, and subsequently incubated with the primary antibody overnight, using the rabbit anti-oligomer conformation-specific A11 (pre-fibrillar Aβ oligomer, 1:1000, Invitrogen, Carlsbad, CA, United States) diluted in blocking solution. Membranes were washed with TBST prior to incubation with horseradish peroxidase-labeled (HRP)-linked secondary antibody (1:1000, ZSGB-Bio, Beijing, China) at room temperature for 1 h. The signals were detected using an enhanced chemiluminescence kit. Chemiluminescence image acquisition and densitometric band quantitation were performed using Fusion-FX6 imaging system (Vilber Lourmat, Marne-la-Valle, France). The cAMP concentration was determined using a cAMP assay kit (R&D Systems, Minneapolis, MN, United States) in accordance with the manufacturer's guidelines.

## Quantification of Phosphorylated CREB, AKT, and GSK3β Using ELISA

After the treatment, the cortex and hippocampal tissue of APP/PS1 and WT mice and APPsw cells were homogenized after addition of RIPA buffer (CST, Danvers, MA, United States) containing protease inhibitors, phosphatase inhibitors, and PMSF, then centrifuged at 20,000 × g for 15 min at 4◦C. The protein concentration of the samples was quantified using a commercially available bicinchoninic acid (BCA) kit (Thermo Fisher Scientific, Rockford, IL, United States). ELISA kits were used for quantification of phospho-CREB (Ser133, R&D Systems, Minneapolis, MN, United States), phospho-AKT (Ser473, R&D Systems) and phospho-GSK3β (Ser9, CST, Danvers, MA, United States) activity in accordance with the manufacturers' guidelines.

## In vitro BACE-1 Assay

The ability of LC1405 at concentrations ranging from 0.3 and 300.0 µM to inhibit beta-site APP-cleaving enzyme (BACE-1) activity was determined using a BACE-1 assay kit according to the manufacturer's guidelines.

## Statistical Analysis

Data were analyzed using SPSS software (version 18.0, SPSS, Inc., Chicago, IL, United States), and presented as mean ± standard error of the mean (SEM). Statistical analysis was performed by using different tests as follows: (1) escape latency of the MWM test the acquisition trial was analyzed using the ANOVA for repeated measures (training days and treatment, with treatment as main effect), and one-way ANOVA with Tukey's post hoc analyses were used to analyze treatment differences (treatment as main effect); (2) treatment differences in the probe trials, pathological and biochemical assays, and in vitro studies were performed using a one-way ANOVA, followed by Tukey's post hoc testing to analyze the differences between groups; (3) the area under the curve (AUC) of histamine and ACh levels in the brain after LC1405 administration were analyzed by student t-test. The statistical significance was set at a p-value of less than 0.05.

## RESULTS

## LC1405 Treatment Improves Cognitive Deficits in APP/PS1 Mice

At 13 months of age, APP/PS1 mice underwent behavioral testing using the MWM test, a widely-accepted test for spatial learning and memory capability. During the acquisition trial, the escape latency to the target platform for all mice is illustrated in **Figure 1B**. Significant differences on escape latency within groups [F(4,192) = 138.697, p < 0.001] and a significant treatment effect on escape latency were found [F(3,48) = 20.572, p < 0.001]. Subsequent post hoc comparison illustrated that treatment with 3 mg/kg of LC1405 was considerable in improving the spatial learning ability in APP/PS1 mice compared to APP/PS1 control mice (p < 0.05). Probe trials were carried out to evaluate shortterm memory at 2 h and long-term memory at 48 h following the five-consecutive-day trials. Our data indicated that APP/PS1 control mice stayed for a shorter time and completed fewer crossings in the target quadrant at both time points compared with WT control mice (**Figures 1C,D**) (2 h: 18.77 ± 1.25% vs. 30.14 ± 3.04%, 1.38 ± 0.31 vs. 3.46 ± 0.52, p < 0.001 and 0.01; 48 h: 17.94 ± 0.85% vs. 32.51 ± 1.37%, 1.38 ± 0.29 vs. 3.62 ± 0.24, both p < 0.001), while LC1405-treated APP/PS1 mice stayed in the target quadrant for a longer time (2 h: 27.29 ± 1.47% vs. 18.77 ± 1.25%, p < 0.01; 48 h: 26.51 ± 0.87% vs. 17.94 ± 0.85%, p < 0.001), and performed more crossings at the platform location compared with APP/PS1 control mice (2 h: 3.00 ± 0.29 vs. 1.38 ± 0.31, p < 0.05; 48 h: 2.92 ± 0.28 vs. 1.38 ± 0.29, p < 0.01). Moreover, our results demonstrated no significant differences in swimming speed among treatment groups in the acquisition trial (**Figure 1E**), suggesting that the improvement on learning capability in LC1405-treated APP/PS1 mice was not due to locomotor ability.

During the retention trial in the passive avoidance test, APP/PS1 control mice re-entered the illuminated compartment more frequently with shorter step-through latency compared to WT control mice (**Figures 1F,G**) (151.58 ± 24.22 s vs. 236.57 ± 24.14 s, 3.22 ± 0.82 vs. 1.11 ± 0.45, both p < 0.05,), while LC1405-treated APP/PS1 mice showed less tendency toward the illuminated compartment, involving longer stepthrough latency and fewer numbers of errors compared to APP/PS1 control mice (220.75 ± 26.25 s vs. 151.58 ± 24.22 s, 1.44 ± 0.58 vs. 3.22 ± 0.82, both p < 0.05), indicating that LC1405 treatment ameliorated memory dysfunction.

In WT mice, LC1405 treatment did not affect cognitive capabilities in the behavioral tests performed, therefore we concluded that long-term oral administration of LC1405 plays a role in ameliorating Aβ-dependent cognitive deficits rather than affecting functions that are normal.

## LC1405 Prevents Neuronal Degeneration and Protects Ultrastructure, but Has No Substantial Effect on the Reduction of Aβ Burden or Oxidative Markers in APP/PS1 Mice

FJB is a useful marker for the identification of neuronal degeneration because of its ability to stain the entire neuron, such as the cell body, dendrites, axons, and axon terminals (Ballok et al., 2003). In both the cerebral cortex and the hippocampus of WT control and LC1405-treated WT mice, FJB-positive neurons were rarely detected. However, in the cortical and hippocampal regions of APP/PS1 mice, the number of FJB-positive neurons was significantly increased (**Figures 2A,B**) (cortex: 18.67 ± 2.02 vs. 0; hippocampus: 23.00 ± 2.30 vs. 0.67 ± 0.67; both p < 0.001**)** compared to WT control mice. LC1405 treatment decreased the number of FJB-positive neurons in the cortex and hippocampus of APP/PS1 mouse brains compared to APP/PS1 control group (cortex: 5.67 ± 0.88 vs. 18.67 ± 2.02; hippocampus: 8.33 ± 1.20 vs. 23.00 ± 2.30; both p < 0.001), indicating that LC1405 treatment relieved the degenerative pathology in APP/PS1 mouse brain.

vs. APP/PS1 control.

To further evaluate degenerative changes at the ultrastructural level, sections of the cortex and hippocampus were observed using a transmission electron microscope (**Figure 2C**). The cortical and hippocampal regions of WT control and LC1405 treated WT mice were well-characterized, showing a compact neuropil appearance with normal neurons (N) and astrocytes (As), with no swelling or shrinkage. In APP/PS1 mouse brain regions, neuropils were disrupted and neuronal degeneration was detected as the neurons showed a shrunken appearance. Degenerative neurons were embodied in the ruptured neuronal membranes, with cytoplasm containing dark granules, and condensed nuclei. In addition to degeneration of adjacent neuropils, intracellular vacuolation and astrocyte edema were observed, accompanied with Aβ plaques (SP) deposited in the neuropils nearby. LC1405 treatment attenuated neuropil degeneration in the cortex and hippocampus of LC1405-treated APP/PS1 mice. Signs of swelling and disintegration of astrocytes were disappeared, and Aβ plaque deposition was less apparent

in surrounding neuropils. Together, these findings indicated that LC1405 treatment was effective in preventing Aβ-mediated neuronal degeneration.

Overproduction and aggregation of Aβ are considered key pathological events in the neurodegenerative cascade in AD. Although in this study neuronal degeneration and neuropil ultrastructure were ameliorated by LC1405 treatment, Aβ aggregates/misfolded proteins or similar in the cortex and hippocampus as indicated by Thio S staining, were not reduced after long-term oral administration of LC1405 (**Figures 2D,E**) (cortex: 46.93 ± 1.98 vs. 42.27 ± 1.71; hippocampus: 33.67 ± 1.43 vs. 37.23 ± 3.32). Similarly, the levels of soluble or insoluble Aβ1−<sup>42</sup> or Aβ1−<sup>40</sup> in the brain did not significantly decrease following LC1405 treatment (**Figures 2Fi–iv**) (soluble Aβ1−42: 246.80 ± 28.48 ng/g vs. 233.47 ± 25.05 ng/g; insoluble Aβ1−42: 1236.65 ± 34.57 ng/g vs. 1246.66 ± 87.17 ng/g; soluble Aβ1−40: 79.57 ± 10.45 ng/g vs. 79.18 ± 11.08 ng/g; insoluble Aβ1−40: 221.98 ± 20.29 ng/g vs. 218.87 ± 31.29 ng/g). Furthermore, the increased level of A11-immunoreactive prefibrillar oligomers, which represented the presence of potential neurotoxic Aβ oligomers and then underwent a concerted conformation change from protofibrils to form fibrils (Glabe, 2008; Gulisano et al., 2018), could not be significantly decreased by LC1405 treatment either (**Figures 2Fv,vi**) (ratio of WT control: 2.31 ± 0.20 vs. 2.43 ± 0.25). Combined, these data suggested that LC1405 treatment did not have a substantial effect in reducing Aβ levels in APP/PS1 mouse brain.

Oxidative stress is implicated in the pathology of AD, and in this study, the APP/PS1 mouse model displayed alterations in markers of oxidative stress in the brain, such as MDA, SOD, and GSH-Px compared to WT control group (**Figures 2G–I**) (MDA: 59.63 ± 3.32 nmol/mg protein vs. 27.87 ± 2.69 nmol/mg

protein; SOD: 76.33 ± 2.25 U/mg protein vs. 126.78 ± 8.72 U/mg protein; GSH-Px: 3.62 ± 0.35 U/mg protein vs. 7.78 ± 0.47 U/mg protein; all p < 0.001). However, these markers were not significantly ameliorated in APP/PS1 mouse brain after treatment with LC1405 (MDA: 55.03 ± 3.71 nmol/mg protein vs. 59.63 ± 3.32 nmol/mg protein; SOD: 78.72 ± 6.10 U/mg protein vs. 76.33 ± 2.25 U/mg protein; GSH-Px: 4.06 ± 0.31 U/mg protein vs. 3.62 ± 0.35 U/mg protein).

## LC1405 Treatment Does Not Alter AChE Activity in APP/PS1 Mouse Brain Tissue, but Increases the Level of ACh and Improves the Release of Histamine and ACh

Due to an early and severe depletion of cholinergic innervations in AD pathology, a decrease in AChE activity in the brain is a consistent finding (Dumas and Newhouse, 2011; Carvajal et al., 2013). A slight decrease in AChE activity was observed in APP/PS1 mouse brain, and LC1405 treatment did not alter the AChE activity significantly either in APP/PS1 or in WT mouse brain (**Figure 3A**) (LC1405-treated WT vs. WT control: 1.05 ± 0.10 vs. 1.00 ± 0.06; LC1405-treated APP/PS1 vs. APP/PS1 control: 0.79 ± 0.11 vs. 0.84 ± 0.08). Upregulated ACh levels were observed in LC1405-treated WT and APP/PS1 mice compared to the correspondent LC1405 untreated groups (**Figure 3B**) (LC1405-treated WT vs. WT control: 2285.69 ± 123.96 pg/g vs. 1900.43 ± 82.70 pg/g, p < 0.05; LC1405-treated APP/PS1 vs. APP/PS1 control: 1761.88 ± 59.18 pg/g vs. 1222.23 ± 62.42 pg/g, p < 0.01), suggesting that LC1405 might be effective in increasing the responsiveness of ACh in cholinergic neurons.

In subsequent studies, the concentration of histamine and ACh were compared in the brain of WT mice with and without LC1405 treatment. After LC1405 administration, changes in ACh and histamine levels were expressed in terms of the area under the curve (AUC) for a 0 to 4 h time-period. The time course results of histamine following LC1405 treatment revealed its rapid release that peaked within 30–60 min of dosing and persisted over the 4 h test period. In addition, ACh levels increased over the 60 min period after administration. When comparing the responses with the results of WT control mice, significant increases in AUC in histamine and ACh levels were observed after administration of 3 mg/kg LC1405 (**Figures 3C,D**) (histamine: 14.61 ± 2.05 vs. 3.72 ± 0.72, p < 0.001; ACh: 20.27 ± 5.64 vs. 2.64 ± 0.80, p < 0.05). This was not only the maximal response of ACh after LC1405 treatment, but a sustained increase in the brain after treatment from 60 to 120 min. The response of ACh could be a consequence of increased histamine release. Therefore, these results suggested that LC1405 might act on H3R, thereby leading to an overall increase in histamine and ACh levels.

## LC1405 Treatment Modifies H3R-Mediated Signaling in APP/PS1 Mouse Brain

To identify whether H3R inhibition was a result of the treatment with LC1405, the downstream signaling pathways of H3R coupling were evaluated to assess the potential therapeutic efficacy. Our results showed that transduction of the cAMP/CREB pathway changed significantly in the brains of APP/PS1 mice, which showed a significant decrease in cAMP and p-CREB levels, of 76.12 and 62.78%, respectively in 13-month-old APP/PS1 mice when compared with the WT control (**Figures 3E,F**, all p < 0.001). The Tukey's post hoc comparison demonstrated that oral administration of LC1405 resulted in increased cAMP levels and upregulated p-CREB activity in APP/PS1 mouse brain compared to APP/PS1 control group (cAMP: 1.55 ± 0.17 nmol/mg protein vs. 0.66 ± 0.06 nmol/mg protein; p-CREB: 1.41 ± 0.14 ng/mg protein vs. 0.72 ± 0.09 ng/mg protein, 1.68 ± 0.11 ng/mg protein vs. 0.76 ± 0.09 ng/mg protein; all p < 0.001).

It should be noted that LC1405 treatment also activated cAMP/CREB cascades in WT control mice, indicating that an oral dose of 3 mg/kg of LC1405 affected basal transduction in normal brain (cAMP: 3.66 ± 0.13 nmol/mg protein vs. 2.79 ± 0.25 nmol/mg protein; p-CREB: 2.28 ± 0.15 ng/mg protein vs. 1.93 ± 0.12 ng/mg protein, 2.29 ± 0.08 ng/mg protein vs. 2.04 ± 0.07 ng/mg protein; p < 0.05-0.01). Thus, cAMP/CREB pathway transduction following H3R inhibition might play a role in the action of LC1405.

To investigate H3R downstream signaling, modification of the AKT/GSK3β axis was evaluated. Up-regulation of phosphorylated AKT was accompanied by a high level of GSK3β phosphorylation in the cortex and hippocampus of APP/PS1 mice compared to WT control group (**Figures 3G,H**) (p-AKT: 2.58 ± 0.09 ng/mg protein vs. 0.73 ± 0.06 ng/mg protein, 2.91 ± 0.15 ng/mg protein vs. 0.69 ± 0.04 ng/mg protein; p-GSK3β: 5.54 ± 0.40 ng/mg protein vs. 1.45 ± 0.20 ng/mg protein, 5.82 ± 0.31 ng/mg protein vs. 1.76 ± 0.15 ng/mg protein; all p < 0.001). However, LC1405 treatment suppressed the phosphorylation in APP/PS1 mice (p-AKT: 1.56 ± 0.04 ng/mg protein vs. 2.58 ± 0.09 ng/mg protein, 1.61 ± 0.05 ng/mg protein vs. 2.91 ± 0.15 ng/mg protein; p-GSK3β: 3.44 ± 0.09 ng/mg protein vs. 5.54 ± 0.40 ng/mg protein, 3.22 ± 0.11 ng/mg protein vs. 5.82 ± 0.31 ng/mg protein; all p < 0.001), suggesting that LC1405 might be able to reverse the activated signaling of AKT/GSK3β in AD.

## LC1405 Protects Neuronal Cells Against Copper-Induced Aβ Toxicity in vitro

To better mimic Alzheimer's deficits, copper treatment was used to study metal ion imbalance triggering Aβ neurotoxicity using cells overexpressing the Swedish mutant form of human APP (Zhang et al., 2006; Liu et al., 2012). Before evaluating the protective effects of LC1405, we first established its safe, nontoxic dose in APPsw cells without copper treatment. Our results indicated that LC1405 concentration ranging between 0.03 and 10.0 µM over 24 h did not have any toxic effect (**Figure 4A**). However, in the presence of 300 µM copper, LC1405 significantly increased cell viability at 0.3 µM, 1.0 µM, and 3.0 µM to in a concentration-dependent manner (**Figure 4B**) (74.46 ± 2.99%, 78.15 ± 4.12%, 85.93 ± 3.40% vs. 55.30 ± 3.22%, p < 0.05-0.001).

Similar findings were observed in cells stained with the AO/EB apoptosis-detection dye. The normal morphology of control cells

FIGURE 4 | LC1405 protects APPsw cells against copper-induced Aβ toxicity. (A) Non-toxic concentrations of LC1405 that could be safely used to treat APPsw cells for 24 h without copper treatment (n = 8). (B) Treatment with LC1405 increased cell viability as evaluated by MTS cell proliferation assay (n = 8). (C) Representative images of acridine orange (AO)/ethidium bromide (EB) and β-APP staining (×20 magnification). (D) LC1405 treatment decreased the percentage of apoptotic cells in copper-treated APPsw cells (n = 4). (E) LC1405 treatment affected the mean fluorescence intensity of β-APP in copper-treated APPsw cells (n = 4). (F) LC1405 treatment inhibited the content of Aβ1−40/<sup>42</sup> in copper-treated APPsw cells (n = 4). (G) Higher concentrations of LC1405 resulted in slight BACE-1 inhibition. Data are expressed as mean ± SEM. <sup>∗</sup>p < 0.05, ∗∗∗p < 0.001 vs. control cells, #p < 0.05,##p < 0.01, ###p < 0.001 vs. copper-treated cells.

was characterized by the presence of green-colored nuclei and an intact structure, whereas apoptotic characteristics included shrinkage, membrane blebbing, chromatin condensation, and the formation of apoptotic bodies were found in copper-treated APPsw cells. In the presence of copper, the proportion of APPsw cells undergoing apoptosis was 72.43 ± 3.76% compared to control cells (**Figures 4C,D**, p < 0.001). LC1405 treatment at concentrations of 0.3 µM, 1.0 µM, and 3.0 µM rescued the morphological changes indicative of apoptosis, and reduced the percentage of cells positively stained for the apoptosis-detection dye in a concentration-dependent manner compared with the cells treated with copper (41.41 ± 2.55%, 31.42 ± 3.78%, 26.77 ± 1.92% vs. 72.43 ± 3.76%, all p < 0.001). Combined with these protective effects, LC1405 concentrations at 0.3 µM, 1.0 µM, and 3.0 µM were chosen to investigate the underlying mechanism of LC1405 protection in the AD in vitro model.

## LC1405 Has Limited Effects on Decreasing β-APP Expression or Attenuating Aβ1−40/<sup>42</sup> Levels

Aβ peptides are primarily generated from the cleavage of APP, which represent the major pathological event in the development of AD. In the present study, administration of copper increased the expression of β-APP by 2.11 fold. Similarly, the levels of Aβ1−<sup>40</sup> and Aβ1−<sup>42</sup> in APPsw cells increased respectively by 11.65 and 5.56 fold (**Figures 4C,E,F**, all p < 0.001). Treatment with LC1405 at the tested concentrations did not significantly reduce the expression of β-APP or the levels of Aβ1−<sup>40</sup> and Aβ1−<sup>42</sup> in APPsw cells, except at the highest concentration of 3.0 µM that showed an effect on inhibiting Aβ1−40/<sup>42</sup> levels by the reduction of 21.35% and 23.15% (**Figure 4F**, both p < 0.05). Moreover, LC1405 at higher concentrations (100 and 300 µM) resulted in a slight BACE-1 inhibition (**Figure 4G**, 7.42 ± 0.64% and 9.76 ± 0.83%), what was outside the range of concentrations tested in AD cells. Thus, our results suggested that LC1405 treatment might be independent of Aβ overproduction in the β-amyloidogenic pathway.

## H3R Inhibition Contributes to the Neuroprotective Effects of LC1405 Against Aβ-Induced Toxicity

LC1405 might be identified as a candidate H3R antagonist with neuroprotection through virtual screening, serial cell-based assays, and extensive neuroprotection evaluation. In view of the Bayesian prediction, LC1405, as a potential H3R ligand, serves as a reference compound (**Supplementary Table 2**). In target functional activity assays in vitro, LC1405 showed substantial antagonistic effects combined with a relative high affinity for human and rat H3R, but relatively lower affinities for H1, H2, and H4 subtypes (**Supplementary Table 2**). Furthermore, LC1405 itself blocked the decrease of cAMP induced by RAMH (**Supplementary Table 2**), indicating that LC1405 might interact with H3R, thereby directly increasing cAMP levels. Importantly, LC1405 treatment exerted neuroprotective effects on rat primary cortical neurons against Aβ25−35- and fibrillar Aβ1−<sup>40</sup> (fAβ1−40)-induced toxicity at 0.3 µM, 1.0 µM, and 3.0 µM in a concentration-dependent manner (**Supplementary Figure 2**, p < 0.05-0.001).

To determine whether H3R contributed to LC1405-mediated neuroprotection against Aβ-induced cytotoxicity, specific H3R agonist and other histamine receptor subtype agonists were used. As shown in **Figure 5C**, pharmacological activation of H3R with RAMH blocked the ability of LC1405 to rescue the decreased cell viability observed after exposure to copper at all the tested concentrations (57.84 ± 1.07% vs. 69.80 ± 1.56%, 58.96 ± 1.63% vs. 77.44 ± 5.07%, 60.91 ± 1.15% vs. 86.01 ± 3.27%, p < 0.05- 0.001). Other specific histamine receptor agonists that activated H1, H2, and H4 subtypes did not exert any substantial effect (**Figures 5A,B**) or only partly reduced cell viability that had been increased by LC1405 (**Figure 5D**, 61.40 ± 1.66% vs. 71.87 ± 2.13%, p < 0.05: copper-injured APPsw cells treated with LC1405 and VUF-8430 vs. copper-injured APPsw cells treated with LC1405). Thus, these results suggested that neuroprotection due to LC1405 against copper-induced Aβ injury might be largely attributable to its inhibitory effects on H3R.

## H3R Mediates the Neuroprotective Effect of LC1405 Against Aβ Injury Through cAMP/CREB and PI3K/AKT/GSK3β Signaling Pathways

Histamine H3 receptor signaling caused negative coupling of AC, thereby decreasing the intracellular activity of the cAMP/PKA pathway, subsequently reducing CREB levels (Bhowmik et al., 2014). In APPsw cells subjected to copper, a significant reduction in the intracellular of cAMP level and a significant decrease in phosphorylated CREB level were observed compared to the ones without copper treatment (**Figures 6A–C**) (cAMP: 64.04 ± 2.06 nmol/mg protein vs. 84.92 ± 2.51 nmol/mg protein; p-CREB: 182.82 ± 7.39 vs. 772.98 ± 19.03; both p < 0.001), indicating that the cAMP/CREB signaling pathway was inhibited in response to copper-triggered Aβ toxicity. LC1405 treatment contrasted Aβ-neurotoxicity by increasing intracellular levels of cAMP and up-regulating phosphorylated CREB, both in a concentrationdependent manner (cAMP: 76.27 ± 0.91 nmol/mg protein, 79.97 ± 0.84 nmol/mg protein, 81.71 ± 2.61 nmol/mg protein vs. 64.04 ± 2.06 nmol/mg protein; p-CREB: 280.07 ± 9.85, 302.98 ± 11.49, 359.74 ± 16.54 vs. 182.82 ± 7.39; p < 0.05- 0.001). However, these effects of LC1405 were abolished at all the tested concentrations by the pharmacological activation of H3R treated with RAMH (cAMP: 63.48 ± 2.88 nmol/mg protein vs. 76.27 ± 0.91 nmol/mg protein, 64.10 ± 2.95 nmol/mg protein vs. 79.97 ± 0.84 nmol/mg protein, 65.92 ± 4.11 nmol/mg protein vs. 81.71 ± 2.61 nmol/mg protein; p-CREB: 202.51 ± 6.27 vs. 280.07 ± 9.85, 217.85 ± 7.47 vs. 302.98 ± 11.49, 218.85 ± 14.51 vs. 359.74 ± 16.54; p < 0.05-0.001), suggesting the participation of the cAMP/CREB pathway following H3R signaling in the reduction of Aβ-mediated deficits by LC1405 using copper to trigger neurotoxicity of Aβ.

Histamine H3 receptor is related to PI3K/AKT/GSK3β cascade that play a role in neuronal survival, thereby exerting neuroprotection on lesions due to multiple cytotoxic factors (Provensi et al., 2016). Copper-triggered Aβ exposure

increased the phosphorylated level of PI3K, AKT, and GSK3β (**Figures 6C–E**) (p-PI3K: 853.62 ± 26.34 vs. 347.84 ± 10.75; p-AKT: 3.98 ± 0.16 ng/mg protein vs. 1.23 ± 0.11 ng/mg protein; p-GSK3β: 7.00 ± 0.36 ng/mg protein vs. 1.00 ± 0.07 ng/mg protein; all p < 0.001), whereas LC1405 treatment at the concentration of 0.3 µM, 1.0 µM, and 3.0 µM prevented the increased phosphorylation of PI3K, AKT, and GSK3β in APPsw cells after exposure to copper (p-PI3K: 651.74 ± 26.29, 548.64 ± 12.29, 467.27 ± 18.17 vs. 853.62 ± 26.34; p-AKT: 2.85 ± 0.12 ng/mg protein, 2.20 ± 0.10 ng/mg protein, 1.94 ± 0.08 ng/mg protein vs. 3.98 ± 0.16 ng/mg protein; p-GSK3β: 4.70 ± 0.11 ng/mg protein, 4.25 ± 0.13 ng/mg protein, 3.90 ± 0.23 ng/mg protein vs. 7.00 ± 0.36 ng/mg protein; all p < 0.001). However, in response to RAMH, H3R activation affected the LC1405 effect of preventing the increased phosphorylation of PI3K, AKT, and GSK3β induced by copper (p-PI3K: 810.88 ± 16.18 vs. 651.74 ± 26.29, 778.13 ± 19.69 vs. 548.64 ± 12.29, 801.26 ± 13.68 vs. 467.27 ± 18.17; p-AKT: 3.92 ± 0.15 ng/mg protein vs. 2.85 ± 0.12 ng/mg protein, 3.42 ± 0.21 ng/mg protein vs. 2.20 ± 0.10 ng/mg protein, 3.33 ± 0.06 ng/mg protein vs. 1.94 ± 0.08 ng/mg protein; p-GSK3β: 5.66 ± 0.22 ng/mg protein vs. 4.70 ± 0.11 ng/mg protein, 5.17 ± 0.11 ng/mg protein vs. 4.25 ± 0.13 ng/mg protein, 5.15 ± 0.12 ng/mg protein vs. 3.90 ± 0.23 ng/mg protein; p < 0.05-0.001). These observations indicated that H3R-mediated PI3K/AKT/GSK3β signaling is involved in the protective role of LC1405 against copper-induced Aβ cytotoxicity.

Based on the observations above, LC1405 protected neurons against copper-triggered Aβ-induced toxicity through H3Rdependent cAMP/CREB and PI3K/AKT/GSK3β signaling.

## DISCUSSION

Two major contributions are provided by this study to elucidate the underlying mechanism of action of LC1405. First, LC1405 was identified as a prospective H3R antagonist that prevented Aβ-induced neurotoxicity and as a potential treatment of AD both in vitro and in vivo. Second, a cellular signaling profile of LC1405 in H3R antagonism was provided, characterized by a beneficial H3R-dependent signaling of cAMP/CREB and AKT/GSK3β axes. These findings revealed novel evidence and insights that focus on the role of H3R, which might be a potential therapeutic target in the treatment of AD.

From a previous multi-step screening process, LC1405 was identified as a potential H3R-targeting compound as a result of in silico prediction of H3R ligands, and subsequent cellbased-target assays on H3Rs and in vitro neuroprotective evaluation (see **Supplementary Material**). Furthermore, the ability of LC1405 in reversing the effects of AD and underlying mechanisms were evaluated in APP/PS1 double transgenic mice, and SH-SY5Y cells that express APP with a familial Swedish mutation, both excessively expressing the APP gene

and mimicking β-amyloidogenic disturbance (Zhang et al., 2006; Voss et al., 2014).

Our study found that LC1405 improved learning and memory deficits in AD mice (**Figures 1B–G**). Consistent with these findings, neurodegeneration and ultrastructural abnormalities were rescued in the hippocampal and cortical regions (**Figures 2A–C**). Importantly, these alterations were in line with the in vitro cytoprotection and apoptotic preservation demonstrated after LC1405 treatment (**Figures 4A–D**). Thus, treatment with LC1405 is a promising approach in ameliorating the pathology of AD.

Excessive oxidation has been recognized as a contributor to Aβ-induced neurotoxicity with metal dyshomeostasis being implicated in the Aβ aggregation process of AD. Extremely high ion concentrations, such as copper, have been found colocalized with Aβ deposits in the AD-affected brain (Jakob-Roetne and Jacobsen, 2009; Duce and Bush, 2010). In addition, redox active copper ions lead to ROS overproduction, resulting in oxidative damage that triggers neurodegeneration of the brain (Huang et al., 1999; Jakob-Roetne and Jacobsen, 2009). Here, limited antioxidative effects were obtained after LC1405 treatment in APP/PS1 mice, as indicated by the limited reduction in oxidative biomarkers, such as MDA, SOD, and GSH-Px (**Figures 2G–I**). Thus, LC1405 did not provide a sufficient antioxidant effect through scavenging ROS generation due to Aβ-mediated neurotoxicity related with AD.

The widely accepted amyloid hypothesis suggests that the assembly of Aβ to form aggregates, involving oligomers, protofibrils, fibrils, and even senile plaques, is a central event in the progression of AD (Ferreira et al., 2015; Gulisano et al., 2018). Various reports also confirmed that aggregated Aβ are toxic to neuronal cells, thereby indicating that inhibition of Aβ aggregation as a potential AD therapy is a reasonable strategy in AD treatment. However, the results

from pathological, biological, and western blotting experiments showed that LC1405 treatment did not provide a significant reduction in Aβ burden in APP/PS1 mice (**Figures 2C–F**), which resulted in Aβ plaque deposition as early as 2 months of age and moderate levels of Aβ deposition at the age of 5 months (Blanchard et al., 2003; Ding et al., 2008).

Apart from Aβ aggregation and deposition, the β-amyloidogenic pathway in which Aβ is produced through APP cleavage by BACE1 also plays a role in the evaluation of neuroprotective agents. LC1405 was found to possess a weak inactivation effect on BACE-1 (**Figure 4G**). This result partly explained the limited down-regulation of APP expression in vitro and failure of lowering multiple forms of Aβ levels, including soluble and insoluble Aβ, A11-immunoreactive prefibrillar oligomers, and amyloid aggregates/misfolded proteins or similar (**Figures 2D–F**, **4C,E,F**). Therefore, we initially presumed that the beneficial effect of LC1405 on enhancing learning capacity and reducing memory deficits might be independent of changes in the amyloidogenic processing of APP, or in the process of secretion or deposition of Aβ. Although we found that LC1405 was effective in recuing neuronal cell viability due to fibrillar Aβ1−40/<sup>42</sup> toxicity (**Supplementary Figure 2**), whether LC1405 had the definite effect on altering aggregation profile of Aβ should be further studied on the synthetic peptide and Alzheimer-related animal models.

In the part of the brain that is responsible for learning and memory, H3R, both for integrity and function, is conserved in Aβ over-expressing regions (Foley et al., 2009), indicating that H3R may neither participate in the amyloidogenic processing of APP nor alter the secretion or deposition of Aβ. In our study, LC1405 might preferentially produce improvements in cognition and neuroprotective effects by altering H3R-dependent processes rather than by reducing Aβ deposition or decreasing oxidative stress. Two potential beneficial effects of LC1405 involved in preventing Aβ toxicity were identified in this work: LC1405 (i) improved cholinergic activity by up-regulating ACh and histamine release; (ii) maintained cognitive molecular cascades of H3R-dependent transduction, involving cAMP/CREB and AKT/GSK3β signaling pathways.

Histamine H3 receptor is an auto- and heteroreceptor that negatively regulates the release of histamine and several cognition-related key neurotransmitters, such as ACh, which is recognized as a major neurochemical modulator of cognitive processing, particularly in AD (Bartus, 2000). In this study, we found that oral administration of LC1405 increased the levels of histamine and ACh in mouse brain (**Figures 3C,D**), which was in accordance with the hypothesis that H3R antagonists increased neurotransmitter levels involving ACh and histamine in the brain, thus counteracting AD deficits (Bitner et al., 2011). H3Rs on histaminergic neurons provide tonic inhibition of the firing rate (Jin and Panula, 2005), whereas those on presynaptic histaminergic terminals restrict histamine synthesis and release (Arrang et al., 1983, 1987). Therefore, our hypothesis was that LC1405 administration might inhibit presynaptic H3 autoreceptor activation and increase histamine levels in the synaptic cleft. In addition, long-term administration of LC1405 increased ACh levels in the brain without the influence of AChE activity (**Figures 3A,B**). Since an in vivo micro-dialysis assay suggested that hetero-H3R-mediated regulation of ACh is related to potassium levels in different regions of the brain (Blandina et al., 1996), but not by counteracting the activation of histamine H1 or the H2 subtype (Esbenshade et al., 2008), our conclusion was that in response to Aβ toxicity, LC1405 treatment might increase endogenous ACh synthesis or its release in cholinergic neurons that have been stimulated by increased histamine derived from over-activated histaminergic terminals through the antagonistic action of H3R, rather than by increasing synaptic ACh via a reduction in enzymatic degradation by AChE.

Activation of H3Rs mediates a series of intracellular signaling pathways that are involved in the pathogenesis of AD, including the Gαi/o-protein-coupled inhibition of AC (Lovenberg et al., 1999), downstream exaggeration of cognitive decline through transduction of cAMP/CREB (Moreno et al., 2011), and activation of PI3K/AKT/GSK3β signaling (Rapanelli et al., 2016). Constitutively active H3R inhibits cAMP increase, which would otherwise activate PKA phosphorylation of CREB (Shi et al., 2012), an important signaling molecule located in the cell nucleus and responsible for the synaptic plasticity implicated in the cognitive function. In agreement with this hypothesis, in our study, cAMP/CREB signaling was seriously weakened in Aβ-mediated Alzheimer's pathogenesis, whereas LC1405 rescued aberrant cAMP/CREB signaling both in vitro and in vivo (**Figures 3E,F**, **6A,B**). When APPsw cells that were pretreated with the H3R agonist (RAMH) were treated with LC1405 in the presence of copper, restoration of the cAMP/CREB signaling pathway was completely abolished (**Figures 6A,B**). Therefore,

these results suggested that H3R inactivation of LC1405 plays a pivotal role in neuronal transduction of the cAMP/CREB pathway in response to Aβ neurotoxicity. It is reasonable to believe that the ability of LC1405, as a H3R antagonist, to stimulate the release of histamine and ACh might lead to specific cellular signaling events as a property of auto-receptors in the phosphorylation-activation of CREB that contributed to enhanced cognitive function.

Besides H3R-mediated signaling through Gαi/o-proteins, Gβγ-subunits are known to activate specific signal transduction pathways involving the PI3K/AKT/GSK3β cascade (Rapanelli et al., 2016). Similar to other G protein-coupled receptor (GPCR) signaling transduction, H3R activation results in AKT phosphorylation at Ser 473 and subsequently GSK3β hyperphosphorylation occurring from increased phosphorylation of S9 through PI3K activation via the Gβγ-subunits of Gαi/<sup>o</sup> proteins, previously demonstrated in a neuroblastoma cell line, primary cultures of cortical neurons, and in striatal slices of Sprague-Dawley rats (Bongers et al., 2007). Several studies support the findings that H3R antagonists function as indirect inhibitors of GSK3β, thereby resulting in decreased S9 phosphorylation and exert therapeutic effects in neurodegenerative disorders beyond amelioration of symptoms (Vohora and Bhowmik, 2012). Consistent with these studies, our results showed that the PI3K/AKT/GSK3β pathway was one pathway involved in LC1405 neuroprotection, which could be diminished by a H3R agonist (**Figures 3G,H**, **6C–E**). In this regard, H3R-mediated postsynaptic PI3K/AKT/GSK3β cascade resulted involved in LC1405-induced neuroprotection against Aβ neurotoxicity.

Prior to the identification of the anti-AD effects of LC1405, the safe dosage of LC1405 was evaluated from 0.3 to 10.0 µM in vitro and long-term oral administration at 3 mg/kg in vivo (**Figures 2B–G**, **4A**). The phenomenon that LC1405 alone activated cAMP/CREB cascade was in accordance to the fact that LC1405 targets H3Rs with the specific GPCR characteristics (**Figures 3E,F**). Since PI3K/AKT/GSK3β signaling pathway was regulated by several factors, we concluded that in a physiological state, a single factor mediated by LC1405 influencing PI3K/AKT/GSK3β pathway resulted in well-balanced effects. Therefore, LC1405 could trigger PI3K/AKT/GSK3β pathway and attenuate cognition deficits under physiological, but not pathological, conditions.

In summary, LC1405 had beneficial effects on blocking Aβ-induced neuronal stress both in vitro and in vivo. Anti-H3R therapy using LC1405 did not affect Aβ burden or fully alter the redox balance system, but was effective in increasing the level of histamine and ACh in the brain. LC1405 improved cognition through H3R-dependent cellular

## REFERENCES

Arrang, J. M., Garbarg, M., and Schwartz, J. C. (1983). Auto-inhibition of brain histamine-release mediated by a novel class (H3) of histamine-receptor. Nature 302, 832–837.

signaling cascades involving cAMP/CREB and AKT/GSK3β pathways (**Figure 7**). Taken together, our results demonstrated that LC1405 ameliorated cognitive deficits by blocking H3Rmodulated signaling transduction against Aβ-induced cellular stress. Thus, LC1405 may be a prospective H3R antagonist that holds a potential in controlling disease progression in AD patients, who already developed cognitive deficits with H3Rrelated ACh neurotransmission abnormalities.

## ETHICS STATEMENT

All of the protocols dealing with the maintenance and handling of animals were followed as stated in the Guidelines of the Institute of Medicinal Biotechnology, Chinese Academy of Medical Sciences.

## AUTHOR CONTRIBUTIONS

LW performed the in vivo experiments. JF, HJ, QW, and SX performed the in vitro experiments. JF performed the in silico experiments. LW and JF employed statistical analysis and wrote the manuscript. RL and ZL performed principal investigation, and revised and edited the manuscript. All authors read and approved the final manuscript.

## FUNDING

This study was supported by the National Natural Science Foundation of China (No. U1803281 and 81673411), the Nonprofit Central Research Institute Fund of Chinese Academy of Medical Sciences (2018RC350013), and Chinese Academy of Medical Sciences (CAMS) Innovation Fund for Medical Science (2017-I2M-1-016).

## ACKNOWLEDGMENTS

We thank Prof. Dong-mei Wang, Institute of Materia Medica, Chinese Academy of Medical Sciences and Peking Union Medical College, for providing all the active compounds.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fnins. 2019.00334/full#supplementary-material

Arrang, J. M., Garbarg, M., and Schwartz, J. C. (1987). Autoinhibition of histamine synthesis mediated by presnaptic H3 receptors. Neuroscience 23, 149–157.

Bacciottini, L., Passani, M. B., Giovannelli, L., Cangioli, I., Mannaioni, P. F., Schunack, W., et al. (2002). Endogenous histamine in the medial septumdiagonal band complex increases the release of acetylcholine from the

hippocampus: a dual-probe microdialysis study in the freely moving rat. Eur. J. Neurosci. 15, 1669–1680.



**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 © 2019 Wang, Fang, Jiang, Wang, Xue, Li and Liu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Monoaminergic System Modulation in Depression and Alzheimer's Disease: A New Standpoint?

*Maria Grazia Morgese and Luigia Trabace\**

*Department of Clinical and Experimental Medicine, University of Foggia, Foggia, Italy*

The prevalence of depression has dramatically increased, and it has been estimated that over 300 million people suffer from depression all over the world. Depression is highly comorbid with many central and peripheral disorders. In this regard, depressive states have been associated with the development of neurological disorders such as Alzheimer's disease (AD). Accordingly, depression is a risk factor for AD and depressive symptomatology is common in pre-clinical AD, representing an early manifestation of this disease. Neuropsychiatric symptoms may represent prodromal symptoms of dementia deriving from neurobiological changes in specific cerebral regions; thus, the search for common biological substrates is becoming an imperative and intriguing field of research. Soluble forms of beta amyloid peptide (Aβ) have been implicated both in the development of early memory deficits and neuropsychiatric symptoms. Indeed, soluble Aβ species have been shown to induce a depressive-like phenotype in AD animal models. Alterations in monoamine content are a common feature of these neuropathologies. Interestingly, serotonergic system modulation has been implicated in alteration of Aβ production. In addition, noradrenaline is considered crucially involved in compensatory mechanisms, leading to increased Aβ degradation *via* several mechanisms, including microglia modulation. In further agreement, antidepressant drugs have also been shown to potentially modulate cognitive symptoms in AD and depression. Thus, the present review summarizes the main knowledge about biological and pathological substrates, such as monoamine and related molecules, commonly involved in AD and depression pathology, thus shading light on new therapeutic approaches.

Keywords: Alzheimer's disease, depression, noradrenaline, serotonin, dopamine, beta amyloid

## INTRODUCTION

Many pathologies have been indicated as comorbid with Alzheimer's diseases (AD) and in particular neuropsychiatric disorders such as depression (Ownby et al., 2006; Sun et al., 2008). Indeed, depression is common in pre-clinical AD and may represent an early manifestation of this disease before the appearance of cognitive impairments (Geerlings et al., 2000; Visser et al., 2000). In this regard, much evidence endorses a strong relationship between depression and AD, so much that this mental illness has been proposed as a risk factor for AD or as a prodromic AD phase (Modrego and Ferrandez, 2004). The amyloid cascade hypothesis

#### *Edited by:*

*Cesare Mancuso, Catholic University of the Sacred Heart, Italy*

#### *Reviewed by:*

*Francesco Moccia, University of Pavia, Italy Maria Elisabetta Clementi, Istituto di Chimica del Riconoscimento Molecolare (ICRM), Italy*

> *\*Correspondence: Luigia Trabace luigia.trabace@unifg.it*

#### *Specialty section:*

*This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology*

> *Received: 12 March 2019 Accepted: 16 April 2019 Published: 17 May 2019*

#### *Citation:*

*Morgese MG and Trabace L (2019) Monoaminergic System Modulation in Depression and Alzheimer's Disease: A New Standpoint? Front. Pharmacol. 10:483. doi: 10.3389/fphar.2019.00483*

**33**

postulates that neurodegeneration in AD is related to abnormal accumulation of amyloid beta (Aβ) plaques in various areas of the brain. However, soluble forms of this peptide have been implicated in the development of early memory deficits as well as of neuropsychiatric symptoms (Rowan et al., 2005). Indeed, significant cognitive deficits have been directly attributed to soluble Aβ fragments (Mattson, 2004; Cleary et al., 2005), and increased levels of soluble Aβ oligomers have been linked to synaptic dysfunction (Hardy and Selkoe, 2002; Selkoe and Schenk, 2003). Meanwhile, it has been reported that in depressed patients, Aβ peptide levels are increased (Pomara and Sidtis, 2010). In good agreement, we have previously demonstrated that Aβ, intracerebroventricularly (icv) injected in rats 7 days earlier, evokes a depressive-like profile accompanied by lower cortical serotonin (5-HT) and neurotrophin content (Colaianna et al., 2010). Furthermore, we later reported that such impairment was associated with altered stress response and increased noradrenaline (NA) levels (Morgese et al., 2014, 2015). In addition, in the same model, cognitive impairment was demonstrated either acutely, such as 2 h after Aβ administration, or more enduringly, i.e., 7 days after the peptide central release (Morgese et al., 2014; Tucci et al., 2014; Mhillaj et al., 2018). Although the role of dopamine (DA) was less studied concerning depression and AD, recently, its role has been brought to the fore (Nobili et al., 2017) but is still in need of further evaluation.

The present review is aimed at summarizing the main knowledge related to biological and pathological substrates, such as monoamines and related molecules, commonly involved in AD and depressive pathology, with the scope of shedding light on possible therapeutic approaches.

## MONOAMINE SYSTEM IN DEPRESSION AND ALZHEIMER'S DISEASE

## Serotonergic System

The treatment of affective disorders is mainly based on the enhancement of the noradrenergic and serotonergic systems through selective or nonselective reuptake inhibitors. Such a pharmacological schedule sinks the roots on the catecholaminergic theory of affective disorders stating the crucial role of lower central NA and 5-HT availability in the insurgence of depression (Mann et al., 1986; Schildkraut, 1995; Mann, 1999). Alterations in these neurotransmitter systems have also been linked to neurodegenerative disorders such as AD. Impairment of the serotonergic system has been reported in the very early stages of AD (Versijpt et al., 2003; Egashira et al., 2005; Kepe et al., 2006), and substantial disruption of the serotonergic system in AD has been postulated according to both clinical and *postmortem* studies (Morgan et al., 1987; Lanctot et al., 2001). In this regard, Aβ in its soluble forms, either monomeric or oligomeric, has been associated with the modulation of these systems. In particular, we have previously found that soluble Aβ injected icv in rats caused a significant reduction in 5-HT at the prefrontal cortex level, without interfering with the physiological functioning of other areas such as the striatum or the nucleus accumbens (Colaianna et al., 2010). These results indicated that the prefrontal cortex is an area highly sensitive to Aβ effects, and this area is also crucially involved in the etiopathogenesis of depressive phenomena. Indeed, impairment of 5-HT neurotransmission in the prefrontal area is central to both depressive disorders (Krishnan and Nestler, 2008) and several neurodegenerative diseases (Mattson, 2004; Egashira et al., 2005). Furthermore, we have more recently individuated the vulnerability of the hippocampal area to the action of exogenous Aβ icv injected. Indeed, we have found that this peptide can reduce 5-HT levels in the hippocampus, and this event is associated with a proinflammatory state and higher rate of activated microglia (Mhillaj et al., 2018). In addition, the treatment with a selective COX-2 inhibitor, such as celecoxib, was able to prevent the reduction in 5-HT levels, thus preventing the Aβ-induced depressive-like behavior and restoring Aβ plasma levels to control (Mhillaj et al., 2018; Morgese et al., 2018a). Accordingly, we have recently demonstrated that environmental factors, such as modified dietary factors, can lead to serotonergic impairment associated with increased levels of Aβ. In particular, we found that deficiency in polyunsaturated fatty acids of the omega 3 family, thus corresponding to a condition linked to a pseudoinflammatory state (Solbrig et al., 2010; Graeber et al., 2011), led to a depressive-like phenotype characterized by reduced 5-HT content and higher Aβ levels (Morgese et al., 2017). Accordingly, an anti-inflammatory diet, such as a diet enriched in omega 3 fatty acids, was able to prevent the reduction in 5-HT caused by Aβ injection, preventing the depressive phenomenon (Bove et al., 2018; Morgese et al., 2018b). Likewise, depressed patients showed higher risk for the development of AD (Kessing and Andersen, 2004). On the other hand, *postmortem* studies performed in AD patients revealed low 5-HT and relative receptor content (Reynolds et al., 1995). An *in vitro* model of familiar AD confirmed these observations, since cells overexpressing APP gene with the Swedish mutations associated with familial AD, indicated an altered sensitivity of the serotonergic system and 5-HT1B receptor subtype in particular (Tajeddinn et al., 2016). Furthermore, in a double transgenic model of early AD, fluoxetine, an antidepressant drug acting as serotonin-selective re-uptake inhibitors (SSRIs), ameliorated the impairment of spatial learning by preventing neuronal loss (Ma et al., 2017) and delayed the cognitive decline associated with synaptic changes (Zhou et al., 2018). Accordingly, clinical evidence revealed that SSRIs significantly improve depressant symptoms and daily activities in AD patients (Werner and Covenas, 2015). This point is very intriguing considering that cognitive decline is recognized also as a clinical feature of depressive state. Interestingly, serotonergic system activation was reported to negatively modulate interstitial Aβ content. Indeed, in transgenic animal models of AD, the enhancing of 5-HT signaling, through the administration of SSRI antidepressants, rapidly reduced Aβ production *in vivo via* activation of extracellular regulated kinase (ERK) and the α-secretase-mediated pathway (Cirrito et al., 2011; Fisher et al., 2016). Indeed, the sequential proteolytic cleavage of amyloid precursor protein (APP) can also occur *via* α-secretase, leading to the production of α-CTF later transformed by γ-secretase into AICD and p3 peptides (Chow et al., 2010). This pathway is recognized as the non-amyloidogenic pathway since APP is cleaved by α-secretase in the Aβ region, yielding to lower Aβ production (Chow et al., 2010). This pathway has been described as neurotrophic and neuroprotective (Chow et al., 2010); therefore, therapeutic strategies steered at pushing APP processing toward α-secretase-mediated derivatives are under the spotlight. Furthermore, a PET imaging study carried out in cognitively normal individuals evidenced lower Aβ accumulation in consequence to increased 5-HT signaling (Sheline et al., 2014), and retrospective analysis on patients under antidepressants further confirmed this finding (Vlassenko et al., 2011). In this regard, we have recently demonstrated that fluoxetine treatment not only could restore 5-HT content in animals centrally injected with Aβ characterized by depressive-like phenotype but also reduced Aβ plasma levels (Schiavone et al., 2017). In further agreement, activation of serotonergic receptors, such as 5-HT4, 5-HT6, and 5-HT7, corresponded to lower Aβ content, whereas the opposite effect was retrieved after simultaneous pharmacological blockade of 5-HT4 and 5-HT7 (Cho and Hu, 2007; Fisher et al., 2016). 5-HT4 partial agonists have been proposed as fast-acting antidepressants (Lucas et al., 2007; Vidal et al., 2014) and have been shown to ameliorate cognitive deficit in anxiety/depressive models (Darcet et al., 2016). In good agreement, pharmacological activation of 5-HT4 receptors was shown to enhance short- and long-term memory function (Meneses, 2007), endorsing the hypothesis of a putative role of these drugs for the amelioration of symptomatology of depression in AD. With regard to other receptor subtypes, it has been shown that APP can be released upon activation of 5-HT2A and 5-HT2C, and activation of 5-HT2C receptor promotes the expression of neprilysin, a well-characterized Aβ degrading enzymes (Tian et al., 2015). However, it should be considered that both 5-HT2C agonists and antagonists have been evaluated as antidepressants (Cryan and Lucki, 2000; Steardo et al., 2000; Cryan et al., 2005).

As regard to 5-HT2A receptors, genetic polymorphisms have been described in AD patients affected by major depression (Holmes et al., 2003) and, in AD patients, lower binding to these receptors has been identified (Versijpt et al., 2003). In addition, intra-hippocampal injection of Aβ was associated with a significant reduction in 5-HT2A expression (Christensen et al., 2008). However, the effects of the activation of these receptors may vary depending on the cerebral pathway involved. Indeed, 5-HT2A knocked down mice showed an altered phenotype with depressive-like symptoms (Popa et al., 2005), and 5-HT2A antagonists have been evaluated as antidepressants (Zhang and Stackman, 2015); thus, a better understanding would help the developing of targeted compounds. On the other hand, 5-HT6 receptors represent a novel therapeutic strategy in AD. Indeed, clinical trial for studying the efficacy and tolerability of the 5-HT6 receptor antagonist, SB-742457, in subjects with mildto-moderate and probable AD, revealed a safe profile and possible utility in improving cognitive symptoms of AD (Maher-Edwards et al., 2010). However, antagonists of these receptor subtypes have been indicated as useful also in the treatment of non-cognitive symptoms associated with AD (Garcia-Alloza et al., 2004). However, despite early positive findings, larger phase-III trials have failed to demonstrate any statistically significant impact on cognition for either idalopirdine or intepirdine, two 5-HT6 antagonists, as adjunct to cholinesterase inhibitors. Paradoxically, 5-HT6 receptor agonists also hold cognitive enhancing properties (Khoury et al., 2018). Likewise, polymorphism of these receptors has been associated with altered response to antidepressant treatment in major depressive disorder (Lee et al., 2005), although contrasting results have been reported (Wu et al., 2001); hence, further research is warranted.

## Noradrenergic System

The noradrenergic system is also implicated in the etiopathogenesis of both depression and AD. However, it has been recognized that the cause of depression is more complex than just an alteration in the levels of 5-HT and/or NA, being more directly caused by dysfunction in brain areas or neuronal systems modulated by monoamine systems (Delgado and Moreno, 2000). It has been postulated that antidepressants, by enhancing neurotransmission in normal noradrenergic or serotonergic neurons, can restore lost functions in affected brain areas under monoamine control through a time-dependent process (Delgado and Moreno, 2000). Indeed, noradrenergic and serotonergic systems are strictly interconnected and control each other *via* heteroreceptors. In particular, a negative feedback has been hypothesized considering that increased 5-HT levels correspond to NA release, which in turn inhibits further 5-HT release *via* α2AR activation (Mongeau et al., 1997). This process is mediated through inhibitory α2 receptors (α2AR) at 5-HT terminal levels and 5-HT3 receptors at NA terminals. Interestingly, increased α2AR have been found in *postmortem* brains of depressed patients (Meana et al., 1992; Ordway et al., 1994), and a theory of α2AR supersensitivity in depression was postulated early on Charney et al., 1981. In this regard, increased α2-adrenoceptor density was retrieved in most regions of a rat model of depression, such as the flinders sensitive rat (Lillethorup et al., 2015) and in patients with depressive disorders (Cottingham and Wang, 2012). Interestingly, it has been postulated that tricyclic compounds can bind α2AR, thus functioning as arrestin-based ligands, and such an effect can explain their antidepressant property (Cottingham et al., 2015). Βeta-arrestins are a small family of regulators of G proteincoupled receptors that regulate desensitization, internalization along, and initiation of their own signaling of such receptors (Jiang et al., 2013). Long-term activation of these receptors causes endocytosis and downregulation through the recruitment of α2AR/arrestin complex (Cottingham et al., 2015). The NA system is deeply affected also in neurodegeneration and in early AD (Haglund et al., 2006). Indeed, α2A adrenergic receptors modulate APP endocytic sorting and promote Aβ generation through disrupting APP interaction with a vacuolar protein sorting (Vps10) family protein, a family of receptors that plays a decisive role in controlling the outcome of APP proteolytic processing (Chen et al., 2014). In addition, this study pointed to the use of α2A antagonists as a new direction for AD treatment. In this light, another putative target for the generation of novel AD treatments is targeting β-arrestin. Indeed, increased β-arrestin 1 levels were shown in a transgenic animal model of AD as well as in *postmortem* study (Liu et al., 2013). In keeping in mind a parallel route for depression and AD, β-arrestin signaling has also been associated with antidepressant properties of drugs (Golan et al., 2013). Overexpression of β-arrestin 2 was associated with increased Aβ production. In particular, experimental conditions able to silence the β-arrestin 2 gene corresponded to Aβ rate of production by regulating γ-secretase activity (Thathiah et al., 2013). Accordingly, Pontrello et al. found that the loss of dendritic spine in hippocampal neurons caused by Aβ was prevented by deleting β-arrestin-2 (Pontrello et al., 2012). On the other hand, polymorphisms in the gene encoding for β2 adrenergic receptor have been associated with an increased risk of developing sporadic late onset AD (Yu et al., 2008), while alterations in β adrenergic receptors were reported in depressed patients (Mann et al., 1986). Indeed, much evidence indicates that activation of these receptors yield to antidepressant effects (Overstreet et al., 2008; Gu et al., 2012). Nonetheless, Aβ interacts with the noradrenergic system directly binding to β-adrenergic receptors (Igbavboa et al., 2006; Wang et al., 2011). Aβ may cause desensitization and subsequently internalization of β2 adrenergic receptors in prefrontal cortical neurons (Wang et al., 2011). Furthermore, β2 adrenergic receptor activation mediates phosphorylation of tau after Aβ exposure both *in vivo* and *in vitro* (Wang et al., 2013). On the other hand, we have found that central icv injection of Aβ increases noradrenergic tone after either 2 h or after 7 days from the central injection, probably reflecting a neuroprotective phenomenon (Morgese et al., 2014, 2015), considering that, NA is protective against neuroinflammatory processes. Accordingly, NA is able to modulate glial activation, and pharmacological strategies finalized to increase NA are considered a valid approach for neurodegenerative diseases (Braun et al., 2014). *In vitro* studies have evidenced that neuroprotective effects of noradrenergic locus coeruleus (LC) afferents against Aβ rely on the stimulation of neurotrophic NGF and BDNF autocrine or paracrine loops *via* beta adrenoceptor activation of the cAMP response element binding protein pathway (Counts and Mufson, 2010; Liu et al., 2015). After Aβ exposure, lower NA concentrations in LC projecting areas facilitate the inflammatory reaction of microglial cells, thus impairing microglial migration and phagocytosis, ultimately decreasing Aβ clearance (Heneka et al., 2010). Accordingly, progression of AD is paralleled by the loss of noradrenergic function in LC (Kelly et al., 2017), indicating the crucial role of this system in neurodegeneration.

## Dopaminergic System

As regards the dopaminergic system, impairment of its neurotransmission has been implicated in many diseases including depression (Schmidt et al., 2001), and several pre-clinical studies have indicated the involvement of dopaminergic, either D1, D2, or D3, in antidepressant effects (Pytka et al., 2016). In good agreement, it has been shown that pure dopaminergic drugs, such as pramipexole, DA precursors, and DA reuptake inhibitors, show therapeutic efficacy in depression (El Mansari et al., 2010; Belujon and Grace, 2017). In addition, neurodegenerative diseases associated with the loss of dopaminergic function, such as Parkinson's or Huntington's diseases, have high comorbidities with depression and anxiety (Dale et al., 2016; Schrag and Taddei, 2017; Smeltere et al., 2017).

Concerning AD, it was shown that prefrontal cortical and hippocampal areas showed lower DA receptor expression (Kemppainen et al., 2003; Kumar and Patel, 2007). Interestingly accumbal expression of D2-like receptors, dopaminergic transporter, and tyrosine hydroxylase enzyme was found altered in AD brains (Rinne et al., 1986; Allard et al., 1990; Murray et al., 1995; Joyce et al., 1997). Imaging studies evidenced atrophy of this area in a cohort of AD patients (Pievani et al., 2013). Aβ administration disrupts the cholinergic control of DA release, particularly in the nucleus accumbens (Preda et al., 2008), but we also reported a blunting of DA release in the prefrontal cortex of rat after icv injection of the peptide (Trabace et al., 2007). In addition, the increase in DAnergic tone has been proposed as a possible therapeutic strategy for AD, considering that dopaminergic dysfunction plays a pathogenic role in cognitive decline (Martorana et al., 2009, 2013; Koch et al., 2014; Martorana and Koch, 2014). Furthermore, selective DAnergic neuronal degeneration in ventral tegmental area was demonstrated in AD transgenic mice at pre-plaque stages, suggesting that lower hippocampal and accumbal DA outflow correlate to memory deficits and dysfunction of reward processing (Nobili et al., 2017).

## CONCLUSIONS

It has been reported that depressed individuals are nearly twice as likely to develop dementia, often in the form of AD, compared with non-depressed individuals. Unfortunately, few pharmacological tools are available for dementia; thus, the need for novel therapeutic strategies is very compelling. Future studies aimed at elucidating the mechanisms through which drugs modulating monoamine release may prove helpful in individuating novel strategy for slowing down cognitive impairment in pre-clinical AD phase, often associated with mood alterations, taking into account their effects on Aβ production/clearance, aggregation status, and neuroinflammatoryinduced pathways. Furthermore, some of these molecules are already commercialized; thus, such a novel potential therapeutic approach for AD treatment may become rapidly clinically suitable.

## AUTHOR CONTRIBUTIONS

MM and LT helped in study design, drafting, revising, and accepting of the final version of the manuscript.

## FUNDING

This work was supported by Intervento cofinanziato dal Fondo di Sviluppo e Coesione 2007-2013–APQ Ricerca Regione Puglia "Programma regionale a sostegno della specializzazione intelligente e della sostenibilità sociale ed ambientale– FutureInResearch", Italy to MM (code OC970P6) and by PRIN 2015 code 2015XSZ9A2\_005 to LT.

## REFERENCES


a neuroendocrine mouse model of anxiety/depression. *Neurosci. Lett.* 616, 197–203. doi: 10.1016/j.neulet.2016.01.055


spine plasticity, long-term depression (LTD), and learning. *Proc. Natl. Acad. Sci. USA* 109, E442–E451. doi: 10.1073/pnas.1118803109


**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 © 2019 Morgese and Trabace. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Pharmacological Potential of Cilostazol for Alzheimer's Disease

*Kenjiro Ono1 \* and Mayumi Tsuji <sup>2</sup>*

*1 Division of Neurology, Department of Medicine, Showa University School of Medicine, Tokyo, Japan, 2 Department of Pharmacology, Showa University School of Medicine, Tokyo, Japan*

Alzheimer's disease (AD), a slow progressive form of dementia, is clinically characterized by cognitive dysfunction and memory impairment and neuropathologically characterized by the accumulation of extracellular plaques containing amyloid β-protein (Aβ) and neurofibrillary tangles containing tau in the brain, with neuronal degeneration and high level of oxidative stress. The current treatments for AD, e.g., acetylcholinesterase inhibitors (AChEIs), have efficacies limited to symptom improvement. Although there are various approaches to the disease modifying therapies of AD, none of them can be used alone for actual treatment, and combination therapy may be needed for amelioration of the progression. There are reports that cilostazol (CSZ) suppressed cognitive decline progression in patients with mild cognitive impairment or stable AD receiving AChEIs. Previously, we showed that CSZ suppressed Aβ-induced neurotoxicity in SH-SY5Y cells *via* coincident inhibition of oxidative stress, as demonstrated by reduced activity of nicotinamide adenine dinucleotide phosphate oxidase, accumulation of reactive oxygen species, and signaling of mitogen-activated protein kinase. CSZ also rescued cognitive impairment and promoted soluble Aβ clearance in a mouse model of cerebral amyloid angiopathy. Mature Aβ fibrils have long been considered the primary neurodegenerative factors in AD; however, recent evidence indicates soluble oligomers to initiate the neuronal and synaptic dysfunction related to AD and other protein-misfolding diseases. Further underscoring the potential of CSZ for AD treatment, we recently described the inhibitory effects of CSZ on Aβ oligomerization and aggregation *in vitro*. In this review, we discuss the possibility of CSZ as a potential disease-modifying therapy for the prevention or delay of AD.

Keywords: Alzheimer's disease, amyloid **β**-protein, oligomer, cilostazol, neurotoxicity

## INTRODUCTION

Alzheimer's disease (AD), a progressive neurodegenerative disease, is associated with dementia. The brains of patients with AD are characterized by the occurrence of plaques primarily composed of amyloid β-protein (Aβ) and neurofibrillary tangles composed of tau protein (Selkoe and Hardy, 2016; Gao et al., 2018). Despite the recent advances in symptomatic therapy involving the use of N-methyl-D-aspartate receptor (NMDAR) antagonist and cholinergic drugs, no disease-modifying therapies (DMTs) exist, which directly ameliorate AD-related neurodegenerative processes at the present (Cummings et al., 2016).

#### *Edited by:*

*Silvana Gaetani, Sapienza University of Rome, Italy*

#### *Reviewed by:*

*Luigia Trabace, University of Foggia, Italy Melissa L. Perreault, University of Guelph, Canada*

*\*Correspondence: Kenjiro Ono onoken@med.showa-u.ac.jp*

#### *Specialty section:*

*This article was submitted to Neuropharmacology, a section of the journal Frontiers in Pharmacology*

*Received: 09 February 2019 Accepted: 03 May 2019 Published: 22 May 2019*

#### *Citation:*

*Ono K and Tsuji M (2019) Pharmacological Potential of Cilostazol for Alzheimer's Disease. Front. Pharmacol. 10:559. doi: 10.3389/fphar.2019.00559*

Aβ aggregation is considered one of the most important pathogenic processes, i.e., the amyloid hypothesis; therefore, studies on DMTs have primarily focused on the agents that prevent the accumulation of tau deposits and Aβ in the central nervous system (Cummings et al., 2016). Indeed, *in vitro* and cell studies, human genetics analyses, and neurophysiological studies in animal models strongly implicate Aβ aggregation in AD-associated neurodegeneration *via* the promotion of oxidative stress, inflammation, and apoptosis (Selkoe and Hardy, 2016).

Aβ molecules aggregate to form soluble oligomers and fibrils (Ono, 2018). Subsequently, Aβ aggregates can directly cause neurodegeneration by acting on neurons or indirectly cause it by activating astrocytes and microglia, thereby triggering cytotoxic inflammatory cascades. Hence, to date, several DMTs have been developed targeting different Aβ aggregates (Ono, 2018).

Cilostazol (CSZ) is a selective phosphodiesterase (PDE) 3 inhibitor, which increases intracellular cyclic AMP (cAMP) concentration and activates the cAMP-dependent protein kinase A (PKA), thus causing inhibition of platelet aggregation as well as inducing peripheral vasodilation. In addition, CSZ prevents oxidative stress (Kurtoglu et al., 2014), promotes neurogenesis (Tanaka et al., 2010), acts as an anti-atherogenic agent by enhancing cholesterol elimination from macrophages (Nakaya et al., 2010), inhibits inflammatory cytokine production and signaling (Jung et al., 2010), and improves systemic lymphatic function by inducing the proliferation and stabilization of lymphatic endothelial cells (Kimura et al., 2014).

CSZ is primarily used to prevent cerebral ischemia (Shinohara et al., 2010); however, it also reported slow cognitive decline in patients with mild cognitive impairment (MCI), AD, and cerebrovascular disease (CVD) (Arai and Takahashi, 2009; Sakurai et al., 2013; Taguchi et al., 2013; Ihara et al., 2014; Tai et al., 2017). While the mechanisms of cognitive preservation remain unclear, CSZ has been shown to decrease Aβ25–35 accumulation and to concomitantly reduce cognitive deficits in animal models of AD (Hiramatsu et al., 2010; Park et al., 2011). Using the human-derived neuroblastoma cell line SH-SY5Y cells, we recently reported that CSZ suppressed Aβ1–42-induced neurotoxicity *via* the inhibition of oxidative stress, as demonstrated by coincident reduced reactive oxygen species (ROS) accumulation, mitogenactivated protein kinase (MAPK)-p38 signaling, and nicotinamide adenine dinucleotide phosphate (NADPH) oxidase activity in SH-SY5Y cells (Oguchi et al., 2017).

Although fibrils have long been considered to be the primary neurodegenerative agents, recent evidence indicate that soluble oligomers initiate neuronal and synaptic dysfunctions associated with AD (oligomer hypothesis) (Selkoe and Hardy, 2016; Ono, 2018). Furthermore, different evidence suggests that a tau pathogenesis is mediated by low-molecular-weight (LMW) oligomers of Aβ, e.g., dimers and trimers (Ittner and Gotz, 2011). If this is the case, DMTs should target the neurotoxic activity of these smaller Aβ assemblies to achieve the highest efficacy. Underscoring the potential efficacy of CSZ, we recently demonstrated the inhibitory effects of CSZ on aggregation of Aβ isoforms *in vitro*, including oligomer formation (Shozawa et al., 2018).

In this review, we evaluate the therapeutic possibility of CSZ for AD pathogenesis based on clinical and basic research findings, taking account of the present situation in which no DMTs have been available and some effective combination therapy is seriously sought after.

## PROTECTIVE EFFECTS OF CILOSTAZOL ON NEURONAL CELLS

CSZ has been known to protect various cell types from different stressors, e.g., endothelial cells from H2O2-induced oxidative stress (Ota et al., 2008), vascular smooth muscle cells from endothelin-induced vasoconstriction (Kawanabe et al., 2012), cells constituting the blood-brain barrier (BBB) from collagenaseinduced stroke damage (Takagi et al., 2017), and primary cultured hepatocytes from ethanol-induced damage (Xie et al., 2018). It would be reasonable to expect that CSZ may also be neuroprotective and effective in the treatment of AD or vascular dementia.

Types of neurodegeneration that possibly cause dementia include synaptic transmission dysfunction, neuronal cell death, CREB-related loss of long-term potentiation, and so on. Researches on possible molecular mechanisms of neuroprotection by CSZ will be reviewed in the following.

As mentioned in the Introduction, CSZ has been approved in various countries as an anti-platelet agent, whose inhibition of PDE3 results in PKA activation to suppress platelet aggregation. Some study indicated that neuroprotection by CSZ was associated with the inhibition of PDE3 (Mabuchi et al., 2001), but the molecular mechanisms underlying neuroprotection remain uncertain because PDE3 is abundantly expressed in the heart and vascular smooth muscle cells, but far less in the human brain (Lakics et al., 2010). Thus, it is unlikely that CSZ-induced PDE3 inhibition in neuronal cells is the primary mechanism for improving cognitive impairments in AD. Further, in our experiments using SH-SY5Y cells, CSZ did not reverse the decrease in cAMP concentration induced by Aβ1–42 exposure despite a reduction in neurotoxicity (Oguchi et al., 2017). Thus, CSZ-mediated neuroprotection seems unrelated to PDE3. In addition to its action of selective PDE3 inhibition, CSZ is known to activate other serine/threonine kinase including AMP-activated protein kinase (AMPK) (Park et al., 2016). Neuronal cells treated with CSZ exhibit increased expression of phosphorylated AMPKα, causing upregulation of Aβ autophagy and decreasing intracellular Aβ accumulation (Park et al., 2016).

Another possible protective mechanism involves the modulation of NMDA signaling that is critical in synaptic transmission. Seixas da Silva et al. (2017) recently reported that NMDA receptor (NMDAR) activation mediates the reduced AMPK activity and metabolic deficits in cultured hippocampal neurons exposed to Aβ1–42 oligomers. CSZ suppressed the cognitive deficits caused by an NMDAR antagonist in mice (Hashimoto et al., 2010). In this case, cAMP-response elementbinding protein (CREB) decrease induced by an NMDAR antagonist was counteracted by CSZ treatment and the resulting increase in CREB suppressed the cognitive deficits. CSZ seems to activate AMPK *via* Sir1 in neurons, and this in turn activates CREB (Park et al., 2016).

CSZ appears to suppress oxidative stress through multiple mechanisms. Choi et al. (2002) first reported that CSZ can ameliorate oxidative stress by scavenging hydroxyl and peroxy radicals, thus decreasing ischemic cerebral infarction. In a recent study of mice with permanent focal cerebral ischemia, CSZ suppressed oxidative stress in ischemic neurons by reducing NADPH oxidase (NOX) 2 expression, further resulting in reduced infarct volume (Shichinohe et al., 2015). Moreover, CSZ treatment in SH-SY5Y cells significantly reduced ROS generation during Aβ1–42 exposure by downregulating NOX activation and Nox-4 mRNA expression (Oguchi et al., 2017).

Furthermore, CSZ treatment significantly reduced the expression of the proapoptotic protein Bax and the activation of the apoptosis effector caspases, while significantly increasing the expression of the antioxidant enzyme superoxide dismutase and the antiapoptotic protein Bcl-2 (Oguchi et al., 2017). These results suggest that CSZ attenuates Aβ1–42-induced cytotoxicity in neuronal cells by inhibiting NOX-derived ROS production and mitochondrial damage, resulting in reduced apoptosis.

ROS generated during the early stage of Aβ aggregation also activates the p38-MAPK and JNK signaling pathways in AD brains (Zhu et al., 2002; Tabner et al., 2005). ERK1/2 is activated by neural signals associated with synaptic plasticity and cytoprotection. In the mouse hippocampus, ERK1/2 is activated in postsynaptic neurons by NMDAR activation during long-term potentiation (LTP) induction (Schmitt et al., 2005). Calmodulin-dependent kinase kinase/calmodulin kinase I activity gates extracellular-regulated kinase-dependent LTP. NMDAR activation phosphorylates (activates) ERK1/2, which subsequently regulates the various gene expressions by the CREB phosphorylation. In our recent study, CSZ elevated ERK1/2 and CREB phosphorylation in SH-SY5Y cells treated with Aβ1–42 (Oguchi et al., 2017). In another cell system, that is, mouse neuroblastoma Nm2a cells with overexpression of human mutated amyloid precursor protein (APP) cells, CSZ was shown to increase CREB phosphorylation (Lee et al., 2014).

Recent reports have implicated aberrant CREB signaling in cognitive and neurodegenerative disorders. The hippocampal accumulation of Aβ peptide causes synapse loss and disrupts LTP, which is critical for encoding long-term spatial, associative, emotional, and social memories, through deficient CREB signaling (Saura and Valero, 2011). Further, Qiu et al. (2016) reported that Aβ1–42 oligomers induce apoptosis through decreased Akt and CREB phosphorylation in PC12 cells (Qiu et al., 2016). In addition, the exposure of SH-SY5Y cells to Aβ1–42 decreased phosphorylated CREB, a response prevented by CSZ, and pretreatment with a MEK1/2 inhibitor significantly suppressed CSZ-stimulated CREB phosphorylation (Oguchi et al., 2017).

In summary, Aβ-induced oxidative stress is inhibited by CSZ by scavenging and suppressing NOX activity. Amelioration of oxidative stress by CSZ reduces Aβ-induced activation of p38-MAPK signaling, which is strongly linked to apoptosis and inflammatory responses. Alternatively, CSZ increases ERK1/2 activity in neuronal cells, promoting CREB phosphorylation and transactivation of CRE-controlled genes including Bcl-2 (**Figure 1**). In addition, CSZ protects cells from mitochondrial

dysfunction, which is another ROS source, by inhibiting Aβ-induced increase in Bax and activation of effector caspases. Thus, CSZ may have multiple cytoprotective actions against oxidative stress, impaired synaptic plasticity, mitochondrial dysfunction, and apoptosis and therefore can prevent neuronal damage and associated cognitive deficits in AD.

## CILOSTAZOL INHIBITS A**β** OLIGOMER FORMATION

Several studies have reported the LMW oligomers of Aβ to be particularly toxic (Shankar et al., 2007; Ono et al., 2009; Ono, 2018). LMW oligomers from APP-expressing CHO cells caused progressive dysfunction of synaptic plasticity in rat hippocampal slices (Shankar et al., 2007). Further, LMW oligomers, especially dimers, that were isolated from AD brains exerted synaptic toxicity (Shankar et al., 2008). In our combined structural and cellular studies using pure Aβ oligomers, we revealed that LMW oligomers (dimer, trimer, and tetramer) are more cytotoxic than monomer (Ono et al., 2009); this superior toxicity correlated with the increases in β-sheet content and the seeding activity to facilitate fibrillization (Ono et al., 2009).

Shozawa et al. (2018) recently demonstrated that CSZ significantly inhibited both Aβ1–40 and Aβ1–42 aggregation, but with a stronger inhibitory effect on oligomerization than on fibrillization. Although structural change to β-sheet and fibrillization generally correlate during peptide assembly (Levine, 1999), we reported that LMW oligomers including PICUP-derived oligomers initiated exhibiting β-sheet content at the dimer stage; conversely, a thioflavin fluorescence increase was not observed, which is indicative of fibril formation (Ono et al., 2009). Although Aβ oligomers were initially believed to be positioned on the ON-pathway from monomer to fibrils, some oligomers (e.g., amylospheroids and PICUPderived oligomers) are positioned on the OFF-pathway but exhibit higher toxicity (Hoshi et al., 2003; Ono, 2018). Recently, we reported high-molecular-weight oligomers, e.g., protofibrils to also be positioned on the OFF-pathway using combined thioflavin T assay, electron microscopy, and highspeed atomic force microscopy (Watanabe-Nakayama et al., 2016). Thus, an explanation for the superior inhibitory potency of CSZ against Aβ oligomerization than against fibrillization is the fact that LMW oligomers generated by PICUP are positioned on the OFF-pathway (**Figure 2**; Shozawa et al., 2018).

Until now, we reported that several hydroxyl radical scavengers, e.g., rosmarinic acid, curcumin, and rifampicin, exhibit inhibitory effects on Aβ, tau, and α-synuclein (αS) oligomer formation (Ono et al., 2012; Takahashi et al., 2015; Umeda et al., 2016). Yen and Hsieh (1997) and Tomiyama et al. (1996) reported a phenolic compound with hydroxyl groups, particularly orthoquinone and naphthohydroquinone to be a good hydroxyl radical scavenger. Based on binding assays, we hypothesized that the orthoquinone ring of rosmarinic acid and curcumin and naphthohydroquinone of rifampicin facilitate their specific binding to free Aβ/tau/αS, thereby inhibiting aggregation (Takahashi et al., 2015; Umeda et al., 2016). Regardless of the

and finally mature fibrils. CSZ inhibits ON-pathway formation of Aβ fibrils concurrently with strong prevention of OFF-pathway Aβ oligomers (scale bars = 100 nm). This research was originally published in *Neurosci. Lett.* (Shozawa et al., 2018).

Ono and Tsuji Cilostazol for AD

absence of a quinone ring in CSZ, its quinolone ring with free radical scavenging activity may be associated with Aβ binding and/or inhibition of Aβ oligomerization (**Figure 2**; Orhan Puskullu et al., 2013; Shozawa et al., 2018).

CSZ reportedly suppresses Aβ accumulation-induced tauopathy *via* increased PKA-linked CK2/SIRT1 expression *in vitro* (Lee et al., 2014). Additionally, the oral administration of CSZ to C57BL/6J mice prior to Aβ25–35 injection showed significant improvement in spatial learning and memory, prevented Aβ-induced immunoreactivity of Aβ and phosphorylated tau, and suppressed microglia activation compared with control Aβ25–35-injected mice. Nevertheless, posttreatment with CSZ following Aβ25–35 administration and Aβ accumulation did not reduce Aβ-induced neuropathology. Moreover, CSZ had no effect on neprilysin and insulin-degrading enzyme involved in Aβ peptide degradation (Park et al., 2011). Additionally, in a mouse model of cerebral amyloid angiopathy (CAA), CSZ facilitated soluble Aβ clearance and rescued cognitive deficits (Maki et al., 2014).

Very recently, administration of CSZ was reported to increase proteasome activity and reduce the levels of total and aggregated tau species and cognitive decline in a mouse model of tauopathy (Schaler and Myeku, 2018).

In summary, these findings suggest that CSZ promotes the clearance of Aβ oligomers and blocks Aβ oligomer formation, thereby preventing tau pathogenesis. On the other hand, it does not facilitate the clearance of mature fibrils, possibly limiting its clinical efficacy in advanced AD.

## CILOSTAZOL IMPROVES COGNITIVE DECLINE IN PATIENTS WITH ALZHEIMER'S DISEASE

As an antiplatelet therapy, patients generally use 100 mg CSZ orally twice daily; hence, one potential mechanism for cognitive improvement is anti-thrombotic activity and prevention of focal ischemia. At this dose, the plasma concentration attains a steady state between 1.5 and 3.2 μM. Similarly, rats orally administered 10 mg/kg CSZ had a plasma concentration of 993 ng/ml (2.69 μM) as measured by radioactive carbon; however, the concentration was only 99 ng/g in the cerebrum and 946 ng/g in the hypophysis following an oral administration of 10 mg/kg CSZ, suggesting only a minor fraction of CSZ passes through the BBB (Akiyama et al., 1985). Whether the prevention of Aβ oligomer formation, neurodegeneration, and cognitive impairment can occur in patients with AD at these clinical CSZ doses needs to be clarified.

The concentrations required to prevent Aβ aggregation are 10- to 40-fold higher (25–100 μM) than the effective concentration of 2.5 μM identified in the present Aβ toxicity assay, which is notably within the range of normal plasma concentrations. Further, its brain concentrations may be substantially lower than its plasma concentrations. However, the cerebrospinal fluid concentrations of Aβ were only 200–300 pg/ml (~50 pM) in patients with AD (Hu et al., 2015), which is approximately 1,000,000-fold lower than the Aβ concentrations observed in this aggregation study. Considering the effective Aβ to CSZ ratio, it needs to examine whether a long-term clinical administration of CSZ continues to inhibit Aβ oligomer formation *in vivo*.

In Japan and other Asian countries, CSZ is clinically used to prevent cerebral ischemic diseases (Shinohara et al., 2010), including CAA, because it carries a limited risk of hemorrhage in most elderly patients (Charidimou et al., 2012; Saito and Ihara, 2014). The second CSZ Stroke Prevention Study (CSPS2) for patients with cerebral infarction reported hemorrhagic stroke to be significantly less frequent in a CSZ group than in an aspirin group (Shinohara et al., 2010; Uchiyama et al., 2014). These effects may be explained, at least partially, by an inhibitory effect on matrix metalloproteinase-9 expression and the protection of vascular endothelial cells (Hase et al., 2012; Kasahara et al., 2012).

The efficacy of CSZ in patients with MCI (Taguchi et al., 2013), AChEI-treated patients with clinically probable AD (Arai and Takahashi, 2009; Tai et al., 2017), and patients with AD and CVD (Sakurai et al., 2013; Hishikawa et al., 2017) has been evaluated in several small-scale clinical studies. In a pilot study involving 10 patients with moderate AD who were administered AChEI donepezil, a 5- to 6-month add-on CSZ treatment significantly increased the Mini Mental State Examination score in comparison with the baseline score (Arai and Takahashi, 2009). In a larger pilot study comprising 30 participants, a 12-month CSZ add-on therapy improved cognitive impairments in those with stable AD (Tai et al., 2017). Recently, a pilot study including 101 patients with AD and asymptomatic lacunar infarction reported that combination therapy with CSZ and the AChEI galantamine significantly improved the Geriatric Depression Scale and Abe's behavioral and psychological symptoms of dementia scores and a 6-month CSZ monotherapy significantly improved the Geriatric Depression Scale score (Hishikawa et al., 2017). The effects of a 6-month CSZ treatment on cognition and regional cerebral blood flow (rCBF) were examined in 20 elderly patients with mild-to-moderate AD and CVD (Sakurai et al., 2013). As the results showed, the CSZ group did not show any changes in cognitive function, whereas the control group showed a cognitive decline on the AD Assessment Scale-Cognitive Subscale. Analysis of treatment effect revealed that the CSZ group showed increased rCBF in the right anterior cingulate lobe, whereas the control group showed decreased rCBF in the left middle temporal gyrus. On the other hand, initiated study in 2011 by the Seoul National University Hospital revealed that no difference between CSZ and placebo groups was reported on cognitive measures, which included the MMSE and the cognitive scale of the cognitive part of the AD Assessment Scale in 36 mild-to-moderate AD patients with subcortical white matter hyperintensities treated with donepezil for a 6-month period (Prickaerts et al., 2017).

Furthermore, an approximately 2-year retrospective analysis concluded that CSZ improves cognitive function in patients with MCI (Taguchi et al., 2013). Randomized placebo-controlled clinical phase II trials are currently ongoing for patients with MCI (Saito and Ihara, 2014).

Side effects of CSZ include most commonly headache, diarrhea, abnormal stools, irregular heart rate, and palpitations. It is contraindicated in patients with severe heart failure or severe hepatic/renal impairment (Chapman and Goa, 2003).

## CONCLUSION

CSZ was reported to promote Aβ clearance, inhibit Aβ oligomerization, and suppress Aβ-induced neurotoxicity *in vitro* and *in vivo*. CSZ is reported to suppress cognitive decline progression in some patients with MCI or AD. For examination of these effects in a larger scale, randomized placebo-controlled phase II trials are ongoing for patients with MCI (Saito and Ihara, 2014). As future direction, potential effects of CSZ on AD comorbidities, such as depression or metabolic dysfunctions (e.g., diabetes), will also have to be examined in AD or MCI

## REFERENCES


patients with or without CVD because oxidative stress plays the important role in these diseases as in AD (Novais and Starkstein, 2015; Karki et al., 2017; Morgese et al., 2017).

## AUTHOR CONTRIBUTIONS

KO and MT wrote the manuscript and approved the final version of the manuscript.

## FUNDING

This work was supported by Grants-in-Aid for Scientific Research (Kakenhi) from the Japan Society for the Promotion of Science (JSPS) under Grants JP26461266 and JP19K07965 (KO), and Research and Development Grants from the Japan Agency for Medical Research and Development (16dk0207021h0001) (KO).


in hippocampal neurons as a protective response after exposure to glutamate in vitro and ischemia in vivo. *J. Neurosci.* 21, 9204–9213. doi: 10.1523/ JNEUROSCI.21-23-09204.2001


kinase-dependent long-term potentiation. *J. Neurosci.* 25, 1281–1290. doi: 10.1523/JNEUROSCI.4086-04.2005


dynamics of amyloid β1-42 aggregates. *Proc. Natl. Acad. Sci. USA* 113, 5835–5840. doi: 10.1073/pnas.1524807113


**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 © 2019 Ono and Tsuji. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Chemokines in Alzheimer's Disease: New Insights Into Prokineticins, Chemokine-Like Proteins

### *Anna Rita Zuena1, Paola Casolini1, Roberta Lattanzi1\* and Daniela Maftei2*

*1 Department of Physiology and Pharmacology "Vittorio Erspamer," Sapienza University of Rome, Rome, Italy 2 Department of Biochemical Sciences "Alessandro Rossi Fanelli," Sapienza University of Rome, Rome, Italy*

Alzheimer's disease is the most common neurodegenerative disorder characterized by the presence of β-amyloid aggregates deposited as senile plaques and by the presence of neurofibrillary tangles of *tau* protein. To date, there is a broad consensus on the idea that neuroinflammation is one of the most important component in Alzheimer's disease pathogenesis. Chemokines and their receptors, beside the well-known role in the immune system, are widely expressed in the nervous system, where they play a significant role in the neuroinflammatory processes. Prokineticins are a new family of chemokine-like molecules involved in numerous physiological and pathological processes including immunity, pain, inflammation, and neuroinflammation. Prokineticin 2 (PROK2) and its receptors PKR1 and PKR2 are widely expressed in the central nervous system in both neuronal and glial cells. In Alzheimer's disease, PROK2 sustains the neuroinflammatory condition and contributes to neurotoxicity, since its expression is strongly upregulated by amyloid-β peptide and reversed by the PKR antagonist PC1. This review aims to summarize the current knowledge on the neurotoxic and/ or neuroprotective function of chemokines in Alzheimer's disease, focusing on the prokineticin system: it represents a new field of investigation that can stimulate the research of innovative pharmacotherapeutic strategies.

Keywords: Alzheimer's disease, chemokines, prokineticin receptors, Aß-peptide, prokineticins

## ALZHEIMER'S DISEASE

Alzheimer's disease (AD) is the most common progressive neurodegenerative disorder and is the most frequent cause of dementia, characterized by a progressive and irreversible mental decline with loss of cognitive skills and memory function. The histopathological hallmarks of AD are extracellular senile plaques that are aggregates of amyloid-β (Aβ) peptide, and intracellular aggregation of hyperphosphorylated *tau* protein that forms neurofibrillary tangles (Haass and Selkoe, 2007; Huang and Mucke, 2012). This aggregation causes a neurotoxic cascade, which, in turn, leads to neuronal degeneration and atrophy of the brain regions involved in memory and cognitive impairment (temporal and parietal lobe, pre-frontal cortex, and hippocampus), increasing, in this way, brain neuroinflammation (Raskin et al., 2015; Bronzuoli et al., 2016). It is well known, in fact, that neuronal dysfunction is not the solely cause of AD pathogenesis and progression. There are increasing evidences showing that microglia and astrocytes are implicated in the neuroinflammatory reactions that characterize this pathology. Microglia cells are the innate immune cells of the central nervous system (CNS) and are involved in regulating synaptic plasticity and remodelling neuronal circuits. Astrocytes

#### *Edited by:*

*Cesare Mancuso, Catholic University of the Sacred Heart, Italy*

#### *Reviewed by:*

*Rory R. Koenen, Maastricht University, Netherlands Raffaella Bonecchi, Humanitas University, Italy*

*\*Correspondence: Roberta Lattanzi roberta.lattanzi@uniroma1.it*

#### *Specialty section:*

*This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology*

> *Received: 08 March 2019 Accepted: 15 May 2019 Published: 29 May 2019*

#### *Citation:*

*Zuena AR, Casolini P, Lattanzi R and Maftei D (2019) Chemokines in Alzheimer's Disease: New Insights Into Prokineticins, Chemokine-Like Proteins. Front. Pharmacol. 10:622. doi: 10.3389/fphar.2019.00622*

are the most numerous glial cells in the brain, and they provide nutrients and structural support to neurons. Moreover, microglia and astrocytes are responsible for brain homeostasis, and they react to disease stressors by innate immune responses such as production and release of inflammatory mediators that aim to resolve pathological state. In persistent pathological conditions, such as neurodegenerative diseases, however, microglia as well as astrocytes change their physiological phenotype and, consequently, lose their helpful function. Several studies from post-mortem brains of AD patients and AD animal models have revealed a co-localization of reactive glial cells with senile plaques and neurofibrillary tangles (Parachikova et al., 2007; Hickman et al., 2008; Lopez-Gonzalez et al., 2015). In particular, the early recruitment of microglia around plaques seems beneficial in AD by promoting phagocytosis of Aβ. However, the excessive amount of Aβ occurring with the disease progression overwhelms microglia, which loses its phagocytic capacity in favor of a pro-inflammatory role (Jay et al., 2015). It is known, in fact, that activation of microglia involves the release of several pro-inflammatory molecules (specifically IL-1β, TNFα, and C1q) and induces the activation of astrocytes that consequently lose their neuroprotective activity (Liddelow et al., 2017). Astrocytes' neurotoxic phenotype is abundant in AD patients' brain. Therefore, in these conditions, microglia and astrocytes promote neuroinflammatory response, being responsible for the synthesis of different pro-inflammatory mediators including chemokines and mediators with chemokinelike function as defensins and macrophage migration inhibitory factor (MIF) (Casolini et al., 2002; Sudduth et al., 2013; Williams et al., 2013; Azizi et al., 2014; Guerriero et al., 2017; Chun et al., 2018).

This review aims to summarize the most current knowledge on role of chemokines in AD, focusing on the prokineticins, chemokine‐like molecules that have a role in the amyloid-induced neuronal damage (all the data shown below are summarized in **Table 1**).

## CHEMOKINES

Chemokines are chemotactic cytokines originally identified as factors regulating immune cell migration to sites of inflammation (Luster, 1998). This family, also widely expressed in the CNS, exerts its functions through chemokine receptors that belong to the superfamily of G-protein-coupled receptors. Most chemokines bind to more than one receptor, and several distinct chemokines share common receptor (Rossi and Zlotnik, 2000). Chemokines can be classified into sub-families on the basis of the sequential position of the first two of the four cysteine residues: CXC, CC, CX3C, and C (Bachelerie et al., 2013). The main alteration in chemokines and receptors discriminating pathophysiological inflammatory conditions from physiological ones is their increased expression as demonstrated in plasma, cerebrospinal fluid (CSF), and brain tissue of patients with AD. Microglia, astrocytes, and neurons are believed to be the main source of chemokines and their receptors' production (Liu et al., 2014). In general, most of the chemokines and their receptors contribute to the neuroinflammatory component of AD by recruiting peripheral blood monocytes and promoting glial cell activation, even if emerging data hypothesize for some of them, a neuroprotective role.

**CCL2** or monocyte chemoattractant protein 1 (**MCP-1**), mostly produced by glial cells, seems to have a detrimental role in AD pathogenesis, as its overexpression has been found in brain (Sokolova et al., 2009; Vukic et al., 2009), mature senile plaques, microglia, and microvessels of AD patients (Grammas and Ovase, 2001). Clinical data of AD patients have shown an increase of CCL2 both in CSF and in plasma (Westin et al., 2012; Zhang et al., 2013), and, according to several authors, it correlates with the disease progression and the cognitive decline (Galimberti et al., 2006a; Kimura et al., 2018; Lee et al., 2018). Conversely, other studies report no association between CCL2 plasma levels and AD (Kim et al., 2011; Porcellini et al., 2013). On the other hand, the deficit of CCR2 (CCL2 receptor) may aggravate the disease progression. In transgenic mouse models of AD (Tg2576 mice and APPswe/PSEN1), Ccr2 deficiency accelerates memory deficits and disease progression increasing the Aβ soluble levels in the brain (El Khoury et al., 2007; Naert and Rivest, 2011). This could be due to an impaired macrophage recruitment, microglial accumulation, and Aβ clearance, which seems to be CCR2 dependent.

**CXCL8** (or **interleukin** 8) is produced in CNS by neurons, microglia, and astrocytes in response to proinflammatory signals. It has been found to be increased in serum, CSF, and brains of AD patients (Galimberti et al., 2006b; Ashutosh et al., 2011; Alsadany et al., 2013). Moreover, high levels of its receptors (CXCR2) have been reported in neuritic plaques of AD tissue, as well as in microglia and astrocytes (Xia et al., 1997; Flynn et al., 2003). Bakshi and collaborators have demonstrated *in vitro* that the knock-down or the pharmacological block of CXCR2 with the antagonist SB225002 induces an inhibition in Aβ release,through inhibition of γ-secretase, while the activation of CXCR2, with the exogenous chemokines hrIL8 and hrGRO-α, leads to an increase in Aβ. These data have been confirmed by the same authors in *in vivo* studies, in which Cxcr2 deficient mice show a reduction of Aβ that is associated to γ-secretase decrease (Bakshi et al., 2008, Bakshi et al., 2011). Furthermore, the intra-hippocampal Aβ1–42 injection induces microglial chemotactic response that involves the hippocampal overexpression of CXCL8/CXCR2 in a timedependent manner (Ryu et al., 2015). The hippocampal Aβ1–42 injection also causes an up-regulation of CXCR2 in peripheral T cells associated with an increased T cell entry in the brain. These effects are reduced by intraperitoneal injection with the CXCR2 antagonist SB332235 (Liu et al., 2010a).

**CXCL10** (or **IP-10**). Clinical research in AD patients has demonstrated a positive correlation between the levels of CXCL10 in CSF and cognitive impairment (Galimberti et al., 2006b). CXCL10 is physiologically expressed in astrocytes and elevated in AD patients (Xia et al., 2000) and AD transgenic mice, where it co-localizes with Aβ plaques (Duan et al., 2008; Zaheer et al., 2013). CXCL10 binds to CXCR3 receptor that plays a critical role in the generation of AD pathology: in a transgenic AD mouse model, CXCR3 deficiency significantly reduces Aβ plaque formation and strongly diminishes Aβ peptide in brain tissue; this correlates with the improvement of the behavioral deficit (Krauthausen et al., 2015).


TABLE 1 | Summary of the effects of chemokines and prokineticins in different cellular and animal models of AD and their expression in AD patients.

**CX3CL1** (also named **fractalkine**), unlike other chemokines, is produced as a transmembrane-anchored protein and exerts its functions as a membrane protein or as soluble isoforms. It is produced by neurons and astrocytes, whereas microglia constitutively express CX3CR1, its sole receptor (Mizuno et al., 2003). CX3CR1 is also partly expressed in astrocytes and neurons (Meucci et al., 2000; Cardona et al., 2006), suggesting that CX3CL1/CX3CR1 pathway plays a major role in neuron/ microglia communication and allows neurons to regulate microglia activation (Limatola and Ransohoff, 2014). Clinical studies suggest that CX3CL1/CX3CR1 may participate to the development of AD pathogenesis: serum fractalkine is elevated in patients with mild AD, and its reduction is positively correlated with the cognitive decline (Kim et al., 2008). Moreover, an overexpression of CX3CL1 is present in the hippocampus of AD patients (Strobel et al., 2015). In AD animal models, different studies demonstrate a neuroprotective role of CX3CL1/ CX3CR1 axis in *tau* pathology by favoring an anti-inflammatory context. Transgenic mice lacking CX3CR1 show increased *tau* phosphorylation and aggregation associated with microglial activation and behavioral impairments (Bhaskar et al., 2010; Bolós et al., 2017). In a mouse model of AD, Nash and collaborators have confirmed these results: the overexpression of soluble fractalkine reduces *tau* pathology with a significant reduction of microglial activation and neuronal loss (Nash et al., 2013). Regarding the CX3CL1/CX3CR1 in Aβ pathology, data seem to be divergent (Finneran and Nash, 2019). CX3CR1 deficiency in three different AD mouse models reduces β-amyloid deposition, enhancing Aβ phagocytic ability by microglia (Lee et al., 2010; Liu et al., 2010b). Similarly, upregulation of CX3CR1 in the hippocampus of rats injected with Aβ1−40, induces microglial activation, synaptic dysfunction, and cognitive impairments (Wu et al., 2013). On the other hand, CX3CR1 deficiency in mice overexpressing human amyloid precursor protein (APP) exhibits enhanced *tau* pathology, microglial activation, expression of proinflammatory markers (IL-1β, TNF-α, and IL-6), neuronal death in the dentate gyrus, and worsened learning abilities (Cho et al., 2011). Moreover, in the same mouse model, a decrease in CX3CL1 has been shown in cerebral cortex and hippocampus (Duan et al., 2008).

**CCL5** (also known as **RANTES**) and its receptors CCR5 are found on endothelial cells, glia, and neurons throughout the brain. The functional role of RANTES in AD is not completely clear. Indeed, different *in vitro* studies have shown that treatment of neuronal cultures with RANTES enhances neuronal survival (Tripathy et al., 2010) and protects against Aβ toxicity (Bruno et al., 2000). Conversely, *in vivo* studies have reported that RANTES and CCR5 are increased in transgenic mice brain (Subramanian et al., 2010; Haskins et al., 2016) as well as in microvessels of AD human brain (Tripathy et al., 2010). CCR5 also binds other chemokines such as CCL3, whose role in AD is still not clear. However, mice lacking CCL3 or CCR5 exhibit a reduced glial activation and an improvement of spatial learning deficit induced by the intracerebroventricular (ICV) Aβ1−40 injection (Passos et al., 2009). Moreover, the ICV administration of CCL3 in mice impairs synaptic transmission and spatial memory; these effects are reverted by a CCR5 antagonist (Maraviroc) (Marciniak et al., 2015). Furthermore, CCR5 deficiency results in enhanced longterm potentiation (LTP) and learning/memory performances, while neuronal CCR5 overexpression causes memory deficits (Zhou et al., 2016). In line with these findings, Lee and co-authors have demonstrated that CCR5 deficient mice have higher accumulation of Aβ, associated with astrocyte activation in the brain, and an impairment of memory and learning functions (Lee et al., 2009).

The chemokine **CXCL12** (**SDF-1α**), and its receptor CXCR4, are expressed and widely detected in the developing and adult CNS (Ma et al., 1998; Zou et al., 1998; Banisadr et al., 2002; Schönemeier et al., 2008). In addition to their role in neuroinflammation, they regulate neuronal excitability and synaptic transmission (Limatola et al., 2000) and modulate neuronal firing and neuron/glia communication (Bezzi et al., 2001). An upregulation of Cxcr4 gene and protein levels has been found in the brains of AD (Weeraratna et al., 2007). Furthermore, a decrease in plasma CXCL12 is reported in early AD patients and negatively correlated with CSF *tau* protein levels (Laske et al., 2008). In agreement with these results, Parachikova and Cotman (2007) show an increase of CXCR4 protein expression and a decrease of CXCL12 in the hippocampus of AD patients. However, the authors also demonstrate that transgenic AD mice, in which the mutation induces an overproduction of human APP, exhibit a reduction of CXCL12 and CXCR4 that correlates with deficits in different cognitive tasks. Moreover, chronic administration of a CXCR4 antagonist (AMD3100) results in impaired learning and memory in young non-transgenic mice, thus supporting the hypothesis that low levels of CXCL12/CXCR4 are linked to cognitive deficits (Lu et al., 2002). It has also been demonstrated that pre-treatment with CXCL12 inhibits the deleterious effect induced by the ICV injections of Aβ, suggesting that the beneficial effect of CXCL12/CXCR4 on memory and learning in AD could be linked to the prevention of dendritic regression and neuronal apoptosis induced by Aβ (Raman et al., 2011). In agreement with these data, a recent study demonstrates that CXCL12 mediates the neuroprotective and anti-amyloidogenic actions of human painless NGF (hNGFp) treatment in 5xFAD mice, transgenic mice that co-overexpress five familial AD mutant forms of human APP and presenilin 1 (Capsoni et al., 2017). Noteworthy, CXCL12 also binds CXCR7 that seems to be involved in the progression of various CNS pathologies including AD (Puchert et al., 2017).

## PROKINETICINS

The amphibian Bv8 and mammalian prokineticin 2 (PROK2) are two secreted bioactive peptides that belong to the prokineticin (PK) family. Prokineticins are highly conserved across the species. They are characterized by the presence of a conserved N-terminal sequence AVITGA, 10 cysteine residues, that create five disulphidebridged motifs and a tryptophan residue in position 24. A degree of similarity between prokineticin and defensins, a subclass of cationic antimicrobial peptides involved in innate immunity that, similarly to prokineticins, contain a high number of cysteine residues, was reported. However, their small size (8 kDa), signaling mechanisms, receptor coupling (G protein-coupled receptors, GPCR), as well as chemotactic and immune-modulatory functions classify prokineticins as chemokines (Monnier and Samson, 2008; see also Negri and Ferrara, 2018).

Prokineticins activate two GPCR, prokineticin receptor 1 (PKR1) and 2 (PKR2), widely distributed in different organs and tissues as well as in CNS, in which PKR2 results more expressed. Both receptors are highly conserved sharing 85% amino acid identity and diverging mainly in their N-terminal region (Kaser et al., 2003). The PKRs are coupled to Gq, Gi, and Gs depending on the type of cellular localization and activate different intracellular signal pathways (for an explicative review, see Negri and Ferrara, 2018). It has also been reported that PKR2, at least in human neutrophils, undergoes dimerization (Marsango et al., 2011). Furthermore, in saccharomyces, the dimerization takes place from interactions between transmembrane domains TMs 4 and 5, with a specific role playing by TM5 on PKR2 function (Sposini et al., 2015).

PROK2 is abundantly expressed in CNS and associated with multiple physiological and pathological functions such as circadian rhythm, neurogenesis, angiogenesis, pain, inflammation, and neuroinflammation (Cheng et al., 2006; Zinni et al., 2017; Negri and Ferrara, 2018; Negri and Maftei, 2018).

The first evidence suggesting the potential role of PROK2/ PKRs in AD comes from *in vitro* and *in vivo* studies demonstrating their overexpression following Aβ1–42 exposure. The incubation of primary cortical cell cultures (CNs) with Aβ1–42 increases the PROK2/PKR mRNA and protein expression in a time-dependent way with a maximum increase at 48 h.Immunofluorescence studies revealed that PKR1 only increases in neuronal cell body, while PROK2 and PKR2 increase both in neurons and in astrocytes (Severini et al., 2015; Caioli et al., 2017). The Aβ1–42-induced PROK2 overexpression is strongly reduced by preincubation with PC1, a non-peptide PKR antagonist (Balboni et al., 2008; Congiu et al., 2014). Interestingly, PC1 reverts the Aβ1–42-induced neuronal death (in a concentration-dependent way), suggesting that reducing the activation of the PK system could be beneficial against Aβ1–42– induced neuronal toxicity.In support of these data, we found that the incubation of CNs with Bv8, the amphibian homologue of PROK2, induces neuronal apoptosis, comparable to the one exerted by Aβ peptide. It is also noteworthy that Bv8-induced neurotoxic effects are concentration-dependent, being achieved specifically at picomolar (and not at nanomolar) concentrations, and are blocked by pre-incubation with PC1. These low concentrations of Bv8, which induce the harmful effects, could be correlated with the small amount of PROK2, eventually released during Aβ incubation. Along this line, a harmful effect of PROK2 at picomolar concentration is reported in cerebral ischemia (Cheng et al., 2012).

It is known that Aβ1–42 induces a significant increase of the ionic current through the AMPA receptors (Parameshwaran et al., 2008; Wang et al., 2010). We have demonstrated that in neurons from CNs, Aβ treatment affects the glutamatergic transmission through the involvement of the PK system (Caioli et al., 2017). Indeed, CNs incubated with Bv8, as well as with Aβ1–42, exhibit an alteration of glutamatergic transmission that results in increase in AMPA receptor ionic currents, measured both as evoked and as spontaneous, which are blocked by PC1. The up-modulation of AMPA ionic currents is not due to modifications of the GluR1 or GluR2 AMPA receptor subunit expression; rather, it appears to be mediated by modification in phosphorylation of AMPA

subunits by the activation of protein kinase C (PKC) intracellular pathway. This is confirmed by the antagonist effect of PKC inhibitor (Go6983). Of note, PC1 also reduces neuronal death induced by kainite, thus exerting a protective action.

The involvement of the PK system in AD is also strongly supported by *in vivo* evidences. In the Tg2576 (TG) transgenic mouse model, PC1 exposure prevents LTP impairment in hippocampal slices, indicating that the pharmacological block of PKRs in TG neurons is sufficient to rescue the synaptic plasticity and to protect against the deleterious effect of PROK2. No changes are observed in age-matched wildtype controls, suggesting that PKR blockade does not affect synaptic plasticity in physiological conditions. In a nontransgenic animal model of AD, induced by ICV injection of Aβ1–42, PROK2 and PKR mRNA expression is increased in both cortex and hippocampus (Severini et al., 2015). In addition, PKR2 mRNA levels in hippocampus are found to be increased not only in the early times (24 h after Aβ1–42 injection) but also in the later stage (14 days after Aβ1–42 injection), indicating a significant role of PK system in the progression of pathology (Lattanzi et al., 2019).

In AD, the differential expression and the alternative splicing of genes notoriously involved in the pathology, such as APOE, APP, or *tau*, may contribute to the pathogenesis of the disease (Love et al., 2015). Alternative splicing of GPCR is a common mechanism that allows the formation of cell-specific isoforms with different biological activity in health and disease (Einstein et al., 2008; Wang et al., 2008).

In a recent paper, a PKR2 splice variant has been identified in rat hippocampus and called TM4-7 since the lack of the second exon gives rise to a receptor containing only four transmembrane domains. Interestingly, in the preclinical model of AD induced by ICV injection of Aβ in rat, the expression of TM4-7 receptor isoform is strongly up-regulated in the hippocampus and the expression ratio between the sliced form TM4-7 and the long form PKR2 increases with the progression of the disease, reaching maximum levels 35 days after Aβ injection (Lattanzi et al., 2019). We can hypothesize that, as already observed in yeast (Lattanzi et al., 2019), also in the brain, TM4–7 may generate homodimer or heterodimer functional receptors with PKR2, so adding versatility and complexity to the already complex mechanisms of brain regulation induced by PROK2 and its receptors. A summary of the hypothetical role of PK system in Aβ-induced cell toxicity is shown in **Figure 1**.

## REFERENCES


## CONCLUSIONS

The work summarized in the present review indicates that chemokines and their receptors are crucial neuroinflammatory actors for AD pathogenesis and/or progression. However, many gaps remain in the knowledge on the specific role of neuroinflammation in AD, which impede the development of successful therapeutic strategies. Prokineticins represent a new class of chemokine-like proteins involved in Aβ-induced toxicity and represent an innovative approach for the study of AD pathogenesis. Although in our laboratory many studies are still in progress, the success of the pharmacological blockade of PKRs (that reduces the Aβ-induced neuronal death) leads us to hope for a future promising pharmacotherapeutic strategy of AD. This is particularly relevant considering that, to date, there are no drugs that block or slow down the progression of the disease.

## AUTHOR CONTRIBUTIONS

ARZ and DM made literature search and wrote the first draft of the manuscript. PC and RL designed the aim of the review. DM drew the figure. All authors contributed to reading and approving the final version of the manuscript.


patients with early Alzheimer's disease. *J. Alzheimers Dis.* 15, 451–460. doi: 10.3233/JAD-2008-15311


**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 © 2019 Zuena, Casolini, Lattanzi and Maftei. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Astrocyte Function Is Affected by Aging and Not Alzheimer's Disease: A Preliminary Investigation in Hippocampi of 3xTg-AD Mice

*Maria Rosanna Bronzuoli1, Roberta Facchinetti1, Marta Valenza1,2, Tommaso Cassano3, Luca Steardo1 and Caterina Scuderi1\**

*1 Department of Physiology and Pharmacology "V. Erspamer," Sapienza University of Rome, Rome, Italy, 2 Epitech Group SpA, Saccolongo, Italy, 3 Department of Clinical and Experimental Medicine, University of Foggia, Foggia, Italy*

#### *Edited by:*

*Cesare Mancuso, Catholic University of the Sacred Heart, Italy*

#### *Reviewed by:*

*Nicholas Castello, Gladstone Institutes, United States Marco Milanese, University of Genoa, Italy*

> *\*Correspondence: Caterina Scuderi caterina.scuderi@uniroma1.it*

#### *Specialty section:*

*This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology*

*Received: 12 February 2019 Accepted: 17 May 2019 Published: 06 June 2019*

#### *Citation:*

*Bronzuoli MR, Facchinetti R, Valenza M, Cassano T, Steardo L and Scuderi C (2019) Astrocyte Function Is Affected by Aging and Not Alzheimer's Disease: A Preliminary Investigation in Hippocampi of 3xTg-AD Mice. Front. Pharmacol. 10:644. doi: 10.3389/fphar.2019.00644*

Old age is a risk factor for Alzheimer's disease (AD), which is characterized by hippocampal impairment together with substantial changes in glial cell functions. Are these alterations due to the disease progression or are they a consequence of aging? To start addressing this issue, we studied the expression of specific astrocytic and microglial structural and functional proteins in a validated transgenic model of AD (3×Tg-AD). These mice develop both amyloid plaques and neurofibrillary tangles, and initial signs of the AD-like pathology have been documented as early as three months of age. We compared male 3×Tg-AD mice at 6 and 12 months of age with their wild-type age-matched counterparts. We also investigated neurons by examining the expression of both the microtubule-associated protein 2 (MAP2), a neuronal structural protein, and the brain-derived neurotrophic factor (BDNF). The latter is indeed a crucial indicator for synaptic plasticity and neurogenesis/ neurodegeneration. Our results show that astrocytes are more susceptible to aging than microglia, regardless of mouse genotype. Moreover, we discovered significant agedependent alterations in the expression of proteins responsible for astrocyte–astrocyte and astrocyte–neuron communication, as well as a significant age-dependent decline in BDNF expression. Our data promote further research on the unexplored role of astroglia in both physiological and pathological aging.

Keywords: aging, Alzheimer's disease, 3×Tg-AD mouse, astrocyte, connexin-43, AQP4, S100B, brain-derived neurotrophic factor

## INTRODUCTION

Alzheimer's disease (AD) is currently considered a multifactorial disorder, although aging still remains its greatest risk factor (van der Flier and Scheltens, 2005; Hodson, 2018). Many targets are considered to design novel therapeutics. Glia represents one of them because of its contribution in the regulation of several highly specialized brain functions including glutamate, ions and water homeostasis, excitability and metabolic support of neurons, synaptic plasticity, brain blood flow, and neurotrophic support (Acosta et al., 2017; Bronzuoli et al., 2017). These functions are well integrated since astrocytes tightly communicate one to each other through gap junctions, comprised mainly of connexin-43 (CX43), that provide the structural basis for astrocyte networks (Bruzzone et al., 1996; Theis et al., 2005). These cells quickly respond to brain insults, synergistically working with microglia, the intrinsic immune effector of the brain, to remove injurious stimuli thus restoring brain homeostasis (Gehrmann et al., 1995; Scuderi et al., 2013). For example, during reactive astrogliosis, an event well described in both aged and AD brains (Scuderi et al., 2014a; Steardo et al., 2015; Rodríguez-Arellano et al., 2016), astrocytes modify their structure, usually studied by detecting the cytoskeletal glial fibrillary acidic protein (GFAP) and connexin expression (Peters et al., 2009; Giaume et al., 2010). Connexins form a "honeycomb" organization that creates edges between the end-feet enwrapping blood vessels. Here, astrocytes through aquaporin-4 (AQP4) water channels coordinate water flux and the clearance of interstitial fluid and neurotoxic solutes, including beta amyloid (Aβ). When such a clearance is dysfunctional, Aβ deposition occurs facilitating neurodegeneration (Iliff et al., 2012).

Astrocytes support neurons by releasing several neurotrophins, like S100B and the brain-derived neurotrophic factor (BDNF) (Marshak, 1990; Wiese et al., 2012). In the event of an abnormal production, as in some diseases including AD, such molecules contribute to neuronal damage (Vondran et al., 2010; Scuderi et al., 2014a; Bronzuoli et al., 2016). Glia involvement in the onset and/or progression of AD has been demonstrated by our and other groups (Esposito et al., 2007a; Esposito et al., 2007b; Esposito et al., 2011; Scuderi et al., 2011; Scuderi et al., 2012; Scuderi et al., 2014b; Scuderi and Steardo, 2013; Cipriano et al., 2015; Salter and Stevens, 2017; Taipa et al., 2017). Less is known about their role during healthy aging. Therefore, in this brief research report, we provide a preliminary descriptive investigation of the effects of aging on morphology and functions of hippocampal astrocytes and microglia by comparing young adult (6-month-old) and aged (12-month-old) healthy (Non-Tg) and AD-like (3×Tg-AD) mice to model healthy and pathological aging, respectively. We explored the hippocampal expression of GFAP and S100B for astrocytes, and the ionized calcium binding adaptor molecule-1 (Iba1) and the cluster of differentiation 11b/c (CD11b/c) for microglia. Moreover, we examined deeper astrocyte functions by exploring AQP4 and CX43 expression. Since one of the most important astrocytes functions is the neurotrophic support (Kimelberg and Nedergaard, 2010), we also investigated BDNF production and the dendritic microtubule-associated protein 2 (MAP2).

Collectively, our findings reveal that aging negatively alters astrocytic functions, including their neurotrophic support. A main effect of aging and not of genotype was detected in all astrocytic markers here studied, thus suggesting that observed alterations to astroglia functions were related to aging itself rather than AD.

## MATERIALS AND METHODS

All procedures involving animals were approved by the Italian Ministry of Health (Rome, Italy) and performed in compliance with the guidelines of the Directive 2010/63/EU of the European Parliament, and the D.L. 26/2014 of Italian Ministry of Health.

## Animals

Six- and 12-month-old male 3×Tg-AD mice (homozygous for PS1M146V and homozygous for the co-injected APPswe, tauP301L

transgenes) were used as model of pathological aging. This mutant mouse exhibits, indeed, AD-like plaques and tangles associated with synaptic dysfunction (Oddo et al., 2003a), observed also in our experimental conditions (see **Supplementary Figure S1**). To reproduce a condition of healthy aging, age-matched wild-type littermates (Non-Tg) (C57BL6/129SvJ) were used. Mice were housed in an enriched environment at controlled conditions (22 ± 2°C temperature, 12-h light/12-h dark cycle, 50–60% humidity), with *ad libitum* food and water. Six male mice per group were decapitated, and their brains rapidly isolated and either flash frozen in 2-methylbutane to perform immunofluorescences or dissected to isolate hippocampi for western blot analyses. Tissues were stored at −80°C.

## Western Blot

Hippocampi were processed as previously described (Scuderi et al., 2018a). Briefly, tissues were homogenized in ice-cold hypotonic lysis buffer and then centrifuged. Fifty micrograms of proteins was resolved on 12% acrylamide SDS-PAGE gels and then transferred onto nitrocellulose membranes, which were blocked for 1 h with either 5% bovine serum albumin (BSA) (Fitzgerald, MA) or non-fat dry milk (Bio-Rad, Italy) in tris-buffered saline-0.1% tween-20 (Corning, NY). Membranes were then incubated overnight with one of the following primary antibodies: rabbit anti-GFAP (1:25,000; Abcam, UK), rabbit anti-S100B (1:1,000; Novus Biological, CO), rabbit anti-Iba1 (1:1,000; Abcam), rabbit anti-CD11b/c (1:1,000; Bioss, MA), mouse anti-CX43 (1:500; EMD Millipore, MA), mouse anti-AQP4 (1:500; Santa Cruz, TX), rabbit anti-BDNF (1:1,000; Abcam), mouse anti-β-amyloid (1:200; Millipore, Germany), rabbit anti-p[Ser396]tau (1:1,000; Thermo Fisher Scientific, MA). Rabbit anti-β-actin (1:1,500, Santa Cruz) was used as loading control. After rinses, membranes were incubated with the proper secondary horseradish peroxidase (HRP)-conjugated antibody (1:10,000–1:30,000; Jackson ImmunoResearch, UK), and immunocomplexes detected by an ECL kit (GE Healthcare Life Sciences, Italy). Signals were analyzed by ImageJ.

## Immunofluorescence

As previously described (Bronzuoli et al., 2018), hippocampal coronal slices (12-μm thickness) obtained at a cryostat were postfixed with 4% paraformaldehyde (Sigma-Aldrich). After blockage in 1% BSA dissolved in PBS/0.25% triton X-100, slices were incubated overnight with one of the following primary antibodies: mouse anti-CX43 (1:50, EMD Millipore), mouse anti-AQP4 (1:50, Santa Cruz), rabbit anti-GFAP (1:1000, Abcam), mouse anti-MAP2 (1:250, Novus Biologicals). Sections were rinsed in PBS and incubated for 2 h with the proper secondary antibody [1:200 fluorescine-affinipure goat anti-rabbit IgG (H+L); 1:300- 1:400 rhodamine-affinipure goat anti-mouse IgG (H+L) (Jackson ImmunoResearch)] and DAPI (1:75,000, Sigma-Aldrich). Fluorescence was detected by an Eclipse E600 microscope (Nikon, Japan). To avoid the observation of differences among groups caused by artifacts, the exposure parameters, including gain and time, were kept uniform during image acquisitions. Pictures were captured by a QImaging camera and analyzed by ImageJ. Immunofluorescence quantifications are expressed as ΔF/F0 = [(F− F0)/F0], where F is the mean fluorescence intensity and F0 is the mean background fluorescence. We performed immunofluorescence experiments in the hippocampus, focusing our analyses on the Ammon's horn 1 (CA1) subregion because this area is one of the most vulnerable to AD, in both patients (Rössler et al., 2002; Mueller et al., 2010) and 3×Tg-AD mice (Oddo et al., 2003a). We analyzed three serial coronal sections per animal (between −1.82 and −1.94 mm from bregma) spaced 36 µm apart, analyzing four ROIs in the stratum radiatum of each section (200 × 100 µm).

## Statistics

Data were analyzed by two-way ANOVA using GraphPad Prism6. When applicable, Bonferroni's *post hoc* test was used. Data were expressed as mean ± standard error of the mean (SEM) of percentage of control (6-month-old/Non-Tg mice).

## RESULTS

## Aging Affects Morphology and Functions of Hippocampal Astrocytes, Independent of Genotype

Results from Western blot experiments, performed in homogenates of hippocampi of Non-Tg and 3×Tg AD mice, showed that age significantly affects astrocyte morphology and functions. In fact, we observed a significant reduction of the cytoskeletal protein GFAP and the neurotrophin S100B in 12-month-old mice compared with 6-month-old mice, irrespective of genotype (**Figure 1A**, **B**, **C**). Moreover, we found a significant genotype-by-age interaction on GFAP data (*p =*  0.0357). By immunofluorescence, we observed a significant reduction of GFAP in the hippocampal CA1 subregion of 12-month-old mice compared with 6-month-old mice, independently of genotype (**Figure 1F**, **G**, **H**). In addition, results from immunofluorescence and Western blot revealed that aging impacts on astrocyte functions. Indeed, we found a significant decrease of CX43 expression in 12-month-old mice compared with 6-month-old mice, independently of genotype (**Figure 1A**, **D**, **F**, **I**). Moreover, in the same experimental conditions, we observed an increased expression of AQP4 in the hippocampi of aged mice regardless of genotype (**Figure 1A**, **E**, **G**, **L**). For both AQP4 and CX43, no genotype-by-age interaction was detected.

## Aging Does Not Significantly Affect Hippocampal Microglia

In hippocampal homogenates of Non-Tg and 3×Tg-AD mice, we performed Western blot experiments for Iba1, a calcium-binding protein constitutively expressed by both surveillant and activated microglia, and CD11b/c, a marker of proliferative reactivity. Obtained results showed that the expression of Iba1 was not affected by either age or genotype (**Figure 2A**, **B**). However, we observed a significant increase of CD11b/c in 6-month-old 3×Tg-AD mice in comparison with their age-matched Non-Tg littermates, indicative of a potential microglial activation in young transgenic animals, but not in aged ones (**Figure 2A**, **C**).

## Aging Reduces Hippocampal Expression of BDNF, With No Significant Impact on Neuronal Loss, Independent of Genotype

The neurotrophic factor BDNF is produced by neurons and, only under pathological circumstances, by astrocytes (Parpura and Zorec, 2010; Fulmer et al., 2014). Therefore, we tested whether aging could affect BDNF production and, in turn, cause neuronal loss. Results from Western blot experiments, performed in homogenates of hippocampi of Non-Tg and 3×Tg AD mice, showed a significant age-related decrease in BDNF production, irrespective of genotype (**Figure 3A**, **B**). Despite the observed reduction of this important neurotrophic factor, the hippocampal expression of the dendritic marker MAP2, analyzed by immunofluorescence, was not significantly different between all experimental groups (**Figure 3C**, **D**).

## DISCUSSION

In the present study, we provide the first preliminary evidence of the effect of aging on structure and functions of hippocampal glial cells. Our primary goal was to study the impact of age on glial cells in conditions of physiological and pathological, AD-like, aging. To start addressing this issue, we used young adult and aged Non-Tg and 3×Tg-AD mice to simulate healthy and pathological aging, respectively. Collectively, our results indicate that aging affects astrocytes with no significant differences between the two genotypes.

The importance of glia in maintaining brain homeostasis and cerebral metabolism is well documented (Parpura and Haydon, 2008; Dzamba et al., 2016). Growing evidence demonstrate the fundamental role of these cells in the etiopathogenesis of several neuropsychiatric disorders thus opening new scenarios to the development of glia-targeted drugs (Bronzuoli et al., 2017; Bronzuoli et al., 2018; Bronzuoli et al., 2019; Cartocci et al., 2018; Scuderi et al., 2018a). The role of glia in healthy aging is still poorly investigated. No data are yet available elucidating whether glial abnormalities involved in neurodegeneration were due to disease progression or they were just a consequence of aging itself. Therefore, we compared young adult (6-monthold) and aged (12-month-old) healthy (Non-Tg) and AD-like (3×Tg-AD) mice. 3×Tg-AD mice progressively and hierarchically develop Aβ plaques and neurofibrillary tangles in AD-relevant brain regions (cortex, hippocampus, and amygdala) and show an age-related cognitive decline that closely mimics the human AD progression (Oddo et al., 2003b; Cassano et al., 2011; Cassano et al., 2012; Romano et al., 2014; Coughlan et al., 2018). Here, we demonstrate that aging affects glial cells, especially modifying astrocyte structure and functions. In our experimental conditions, we observed that aging reduces the expression of the cytoskeletal GFAP and the neurotrophin S100B, regardless of mice genotype. The age-dependent increase in astrocyte reactivity has been well documented (Beach et al., 1989; David et al., 1997; Janota et al., 2015).

FIGURE 1 | Effects of aging on morphology and functions of hippocampal astrocytes. (A) Representative western blots for GFAP, S100B, CX43 and AQP4 and (B, C, D, E) densitometric analyses normalized to β-actin loading control. Results are expressed as means ± SEM of percentage of controls (6-month-old/Non-Tg) (N = 3, in triplicate). (F) Representative fluorescent photomicrographs of CX43 (red) and GFAP (green) in the hippocampal CA1 region of both 6- and 12-month-old Non-Tg and 3×Tg-AD mice. White arrows indicate CX43 mainly expressed in astrocytes enveloping blood vessels. (G) Representative fluorescent photomicrographs of AQP4 (red) and GFAP (green) in the hippocampal CA1 region of both 6- and 12-month-old Non-Tg and 3×Tg-AD mice. White arrows indicate AQP4 expressed in astrocyte end-feet surrounding blood vessels. Fluorescence analyses of (H) GFAP, (I) CX43 and (L) AQP4 are expressed as ΔF/F0 = [(F− F0)/F0], where F is the mean fluorescence intensity and F0 is the mean background fluorescence. Nuclei were stained with DAPI (blue). Scale bar 50 µm. Statistical analysis was performed by two-way ANOVA followed by Bonferroni's *post hoc* test (\*\**p* < 0.01; \*\*\**p* < 0.001, 6-month-old vs 12-month-old).

However, data obtained from both human material and animal models demonstrate the existence of a complex and region-specific glial response in AD and aging that can be crudely summarized in glial reactivity or glial degeneration, atrophy, and loss of functions (Rodríguez et al., 2016; Verkhratsky et al., 2016; Scuderi et al., 2018b). Our findings are apparently in contrast with the evidence indicating an age-matched astrogliosis in 3×Tg-AD mice (Oddo et al., 2003a; Zaheer et al., 2013). However, Oddo and colleagues analyzed cortical and hippocampal levels of GFAP in both 3×Tg-AD and Non-Tg mice showing no substantial difference between genotype in the hippocampal levels of GFAP in agreement with the results of the present paper (Oddo et al., 2003a). GFAP expression

showed a trend toward being upregulated with increasing age in their experimental conditions; however, they compared 2-month-old mice with 16-month-old mice. The different age chosen for comparison could explain the discrepancy with the present data. 3×Tg-AD mice show activated astrocytes and microglia as age increases and the majority of the studies detected them near amyloid plaques (Oddo et al., 2003a; Zaheer et al., 2013; Rodríguez-Arellano et al., 2016). Interestingly, numerous investigators reported the concomitant occurrence of astrogliosis and astroglial atrophy, demonstrating that the latter appears as a generalized process, whereas astrogliosis is triggered by senile plaques and Aβ aggregates (Rodríguez et al., 2009; Olabarria et al., 2010; Heneka et al., 2010; Verkhratsky et al., 2010; Yeh et al., 2011). In parallel with activation, the AD progression is also associated with astrodegeneration. Accordingly, atrophic astrocytes have been observed in the hippocampus, prefrontal and entorhinal cortices of mouse models of AD (Verkhratsky et al., 2016). In line with these observations, our group has recently demonstrated a reduction of GFAP in the hippocampus of aged 3×Tg-AD mice (Scuderi et al., 2018a). Interestingly, Hoozemans and colleagues (2011) showed lower GFAP immunostaining in post-mortem brains from old AD patients (>80 years) compared with younger AD cases, concluding that the occurrence of astrocyte activation decreases with increasing age in AD dementia.

In our experimental conditions, aging negatively affected astrocyte functions, with particular regard to some of their homeostatic and neurotrophic roles. We found, in both 3×Tg-AD and Non-Tg mice, a significant reduction of CX43 expression, one of the major gap junction proteins that allow an efficient communication among astrocytes. Our data are in line with those obtained by Cotrina and collaborators (2001) demonstrating an age-dependent CX43 reduction in C57Bl/6 mice. The impairments observed in our experimental conditions seems to be caused by aging; in fact, we did not observe any significant differences between the two genotypes. We speculate that this finding together with the aforementioned astrocyte atrophy could be due to a reduced communication among astrocytes. This communication is usually mediated by endogenous peptides, growth factors, neurotransmitters, bioactive lipids, and structural proteins widely expressed in the astrocytic end-feet enveloping blood vessels (Rouach et al., 2002). Interestingly, some authors demonstrated that CX43 reduction boosts Aβ deposits (Koulakoff et al., 2012). In this context, our data could suggest a mechanism responsible for the observed deposition of a huge amount of Aβ in the brain of healthy subjects not affected by AD (Chételat et al., 2013).

AQP4 is another astrocytic protein implicated in the CNS lymphatic drainage and clearance of interstitial solutes, including Aβ (Cotrina et al., 2001; Iliff et al., 2012; Yang et al., 2016). We found that aging is responsible for the augmented AQP4 expression, independent of genotype. Accordingly, AQP4 is highly expressed near Aβ plaques in amyloidopathies (Moftakhar et al., 2010). The raise in AQP4 expression could be a compensatory process aimed at keeping water and ion homeostasis, and at encouraging the clearance of the interstitial fluid in the CNS in both physiological and pathological aging (Gupta and Kanungo, 2013). Interestingly, our data on CX43 and AQP4 are in line with those we obtained on Aβ(1-42) levels (Supplementary Figure S1). Indeed, we found higher hippocampal Aβ(1-42) levels in aged mice of both genotypes. We also observed a significant increase of Aβ(1-42) expression in 3×Tg-AD mice in comparison to Non-Tg animals, at both ages. Given these results, further experiments are required to look for a correlation between severity of pathology and astrocyte protein expression, also investigating other areas crucially involved in AD.

Quite unexpectedly, we did not observe significant structural modifications in microglia in our experimental conditions, except for a significant activation in young transgenic mice, supposedly indicative of the presence of an early

3×Tg-AD vs 6-month-old/Non-Tg).

pro-inflammatory status which disappears in adult mice. These findings are in line with those previously obtained in 3×Tg-AD mice showing that neuroinflammation occurs early in AD-like pathology (Bronzuoli et al., 2018; Scuderi et al., 2018a), and with some authors suggesting that neuroinflammation becomes less and less evident and spread as the disease progresses, remaining detectable in the proximity of Aβ plaques and neurofibrillary tangles (Yeh et al., 2011; Rodríguez-Arellano et al., 2016; Verkhratsky et al., 2016).

The role of microglia in aging and AD is complex and not fully elucidated. Microglial pro-inflammatory markers, including fractalkine and receptors, are reduced in the brain of aged mice compared to adult controls (Wynne et al., 2010), and microglia in aged animals appear irregularly distributed, with variable morphology of both cell bodies and processes, occupying smaller territories (Tremblay et al., 2012). Moreover, it has been demonstrated that 3×Tg-AD mice at 2, 3, and 6 months progressively show significant microglia activation in the entorhinal cortex but not in the hippocampus (Janelsins et al., 2005), as well as a significant increase of activated microglia at 18 months of age compared to 3×Tg-AD mice at 9 months, but not at 12 months of age (Rodríguez et al., 2013). Conversely, very recently Belfiore and colleagues (2019) have demonstrate an age-dependent activation of hippocampal microglia in female 3×Tg‐AD mice, from 6 up to 20 months of age. The heterogeneity of these results confirms that categorizing microgliosis is still particularly challenging. Nowadays, several authors hypothesize the presence of diverse microglial reactions to different disease stages, suggesting that the complete characterization of these processes may open new avenues for therapeutic intervention (Mosher and Wyss-Coray, 2014; Sarlus and Heneka, 2017). Based on these considerations and keeping in mind that our study is preliminary, we believe that further experiments using more specific markers will be required to better characterize the microglia phenotype and its correlations with the neuroinflammatory process also investigating other brain areas importantly affected by AD.

Astrocytes produce and release several neurotrophins, including S100B and BDNF. S100B is a calcium-binding protein mainly involved in cell cycle progression and differentiation as well as neurite outgrowth (Scotto et al., 1998; Sen and Belli, 2007). BDNF is produced mainly by astrocytes over neurons under pathological conditions (Dougherty et al., 2000). We found a significant reduction of both BDNF and S100B in aged mice, irrespective of genotype. Since BDNF regulates the astrocytic expression of S100B and both together are required to support at least serotonergic neurons (Ye et al., 2011), further studies would elucidate the cross-talk between astrocytes and neuronal cells. Despite the reduced neurotrophic support, we did not detect neuronal impairment, as assessed by MAP2.

In conclusion, in this brief research report, we provide preliminary data demonstrating that aging rather than AD progression importantly affects morphology and functions of hippocampal glial cells. In our experimental conditions, the most vulnerable cells were the astrocytes whose structure and functions appear profoundly modified. Additional studies are required to further reveal the role of astrocytes and microglia in both physiological and pathological aging, using more specific markers for the detection of changes in their morphology and/or functions, and extending such observations in other brain regions. Avoiding any superficial projection to human disease and keeping in mind that human astrocytes are more complex than their murine counterparts, these data open novel perspective in the field of astrocyte functions in health and disease.

## DATA AVAILABILITY STATEMENT

Datasets are available on request. Requests to access the datasets should be directed to caterina.scuderi@uniroma1.it.

## ETHICS STATEMENT

All procedures involving animals were approved by the Italian Ministry of Health (Rome, Italy) and performed in compliance with the guidelines of the Directive 2010/63/EU of the European Parliament, and the D.L. 26/2014 of Italian Ministry of Health.

## REFERENCES


## AUTHOR CONTRIBUTIONS

MB, RF, MV, and CS performed most of the molecular experiments and analyzed the data. TC housed and raised the animals. LS, TC, and CS supervised the experiments and discussed the results. MB, RF, MV, and CS wrote the manuscript. All authors contributed to and approved the final manuscript.

## FUNDING

This work was supported by the Italian Ministry of Instruction, University and Research (MIUR) to LS (PON01-02512, PRIN prot. 2009NKZCNX, PRIN prot. 2015HRE757) and to CS (PRIN prot. 2015KP7T2Y\_002); the SAPIENZA University of Rome to CS (prot. C26A15X58E and prot. MA116154CD981DAE) and MB (prot. C26N15BHZZ).

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2019.00644/ full#supplementary-material

FIGURE S1 | Effects of aging on Aβ(1-42) production and tau protein phosphorylation in hippocampus of 3×Tg-AD and Non-Tg mice. (A) Representative Western blots for Aβ(1-42) and p[Ser396]tau and (B, C) densitometric analyses normalized to β-actin as loading control. Results are expressed as means ± SEM of percentage of controls (6-month-old/Non-Tg) (N = 3, in triplicate). Statistical analysis was performed by two-way ANOVA followed by Bonferroni's *post hoc* test (\*\**p* < 0.01; \*\*\**p* < 0.001, 6-month-old vs 12-month-old; ##*p* < 0.01, 6-month-old/3×Tg-AD vs 6-month-old/ Non-Tg; ###*p* < 0.001, 12-month-old/3×Tg-AD vs 12-month-old/Non-Tg).

model of autism spectrum disorders. *Neuroscience* 372, 27–37. doi: 10.1016/j. neuroscience.2017.12.053


Alzheimer's disease. *CNS Neuro. Disord. Drug Targets* 12 (1), 62–69. doi: 10.2174/1871527311312010011


**Conflict of Interest Statement:** MV is currently employed by company Epitech Group SpA, which had no role in study design, collection, analysis, and interpretation of data, in the writing of the report, and in the decision to submit the paper for publication. The remaining 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 © 2019 Bronzuoli, Facchinetti, Valenza, Cassano, Steardo and Scuderi. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# PQM130, a Novel Feruloyl–Donepezil Hybrid Compound, Effectively Ameliorates the Cognitive Impairments and Pathology in a Mouse Model of Alzheimer's Disease

#### *Edited by:*

*Cesare Mancuso, Catholic University of the Sacred Heart, Italy*

#### *Reviewed by:*

*Maria Laura Giuffrida, Italian National Research Council (CNR), Italy Paulo Cesar Ghedini, Universidade Federal de Goiás, Brazil*

#### *\*Correspondence:*

*Andrea Tarozzi andrea.tarozzi@unibo.it*

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

#### *Specialty section:*

*This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology*

> *Received: 22 March 2019 Accepted: 21 May 2019 Published: 12 June 2019*

#### *Citation:*

*Morroni F, Sita G, Graziosi A, Ravegnini G, Molteni R, Paladini MS, Dias KST, dos Santos AF, Viegas C Jr., Camps I, Pruccoli L, Tarozzi A and Hrelia P (2019) PQM130, a Novel Feruloyl-Donepezil Hybrid Compound, Effectively Ameliorates the Cognitive Impairments and Pathology in a Mouse Model of Alzheimer's disease. Front. Pharmacol. 10:658. doi: 10.3389/fphar.2019.00658*

*Fabiana Morroni1†, Giulia Sita1†, Agnese Graziosi1, Gloria Ravegnini1, Raffaella Molteni2, Maria Serena Paladini 2, Kris Simone Tranches Dias 3, Ariele Faria dos Santos 3, Claudio Viegas Jr.3, Ihosvany Camps4, Letizia Pruccoli5, Andrea Tarozzi5\* and Patrizia Hrelia1*

*1 Department of Pharmacy and BioTechnology–FaBiT, Alma Mater Studiorum–University of Bologna, Bologna, Italy, 2 Department of Medical Biotechnology and Translational Medicine, University of Milan, Milan, Italy, 3 Institute of Chemistry, Federal University of Alfenas, Alfenas, MG, Brazil, 4 Institute of Exact Sciences, Federal University of Alfenas, Alfenas, MG, Brazil, 5 Department for Life Quality Studies-QuVi, Alma Mater Studiorum-University of Bologna, Rimini, Italy*

Alzheimer's disease (AD) is the most frequent type of dementia in older people. The complex nature of AD calls for the development of multitarget agents addressing key pathogenic processes. Donepezil, an acetylcholinesterase inhibitor, is a first-line acetylcholinesterase inhibitor used for the treatment of AD. Although several studies have demonstrated the symptomatic efficacy of donepezil treatment in AD patients, the possible effects of donepezil on the AD process are not yet known. In this study, a novel feruloyl–donepezil hybrid compound (PQM130) was synthesized and evaluated as a multitarget drug candidate against the neurotoxicity induced by Aβ1-42 oligomer (AβO) injection in mice. Interestingly, PQM130 had already shown anti-inflammatory activity in different *in vivo* models and neuroprotective activity in human neuronal cells. The intracerebroventricular (i.c.v.) injection of AβO in mice caused the increase of memory impairment, oxidative stress, neurodegeneration, and neuroinflammation. Instead, PQM130 (0.5–1 mg/kg) treatment after the i.c.v. AβO injection reduced oxidative damage and neuroinflammation and induced cell survival and protein synthesis through the modulation of glycogen synthase kinase 3β (GSK3β) and extracellular signal–regulated kinases (ERK1/2). Moreover, PQM130 increased brain plasticity and protected mice against the decline in spatial cognition. Even more interesting is that PQM130 modulated different pathways compared to donepezil, and it is much more effective in counteracting AβO damage. Therefore, our findings highlighted that PQM130 is a potent multi-functional agent against AD and could act as a promising neuroprotective compound for anti-AD drug development.

Keywords: Alzheimer's disease, amyloid-**β** oligomers, oxidative stress, apoptosis, neuroprotection, multitarget ligand, drug discovery, feruloyl-donepezil hybrid

## INTRODUCTION

The World Health Organization estimated the presence of 47.5 million people worldwide with dementia in 2015 and predicted that the number of patients will be almost tripled by 2050 (https://www.who.int/mental\_health/neurology/dementia/ en/). Mainly owing to significant increases in lifespan, dementia represents one of the major global health crises of the 21st century. The most widespread form of dementia is Alzheimer's disease (AD). AD is a lethal neurodegenerative illness that begins with brain alterations more than 20 years before the clinical symptoms (Mori et al., 2019). This multifaceted and progressive neurodegenerative disease is pathologically characterized by the amyloid-β (Aβ) accumulation in amyloid plaques and the hyperphosphorylation of tau in neurofibrillary tangles, followed by a consistent neuronal loss leading to brain atrophy and dementia. Although scientific research has changed course from fibrillar Aβ, implicated in plaque formation, to soluble Aβ, whose accumulation is probably the cause of the early synaptic dysfunction (Selkoe, 2002), the protein is still considered the keystone of AD. Levels of soluble Aβ oligomers (AβO) have been shown in several experimental models to potently inhibit hippocampal long-term potentiation (LTP), increase dendritic spine loss, and impair cognition in mice (Walsh et al., 2002; Lacor et al., 2007; Morroni et al., 2016; Herline et al., 2018). Although AD progression is tightly connected to Aβ aggregation, the scientific consensus is quite firm in suggesting that several other factors likely contribute to the development of AD. Such factors include loss of cholinergic transmission, mitochondrial dysfunction, progressive oxidative damage, excitotoxicity, and neuroinflammatory processes, which may trigger a "domino" cascade of events leading to manifestation of AD (Macchi et al., 2014; Hampel et al., 2018; Pérez et al., 2018). It is likely that AD begins as a synaptic disorder and decreased synaptic activity is one of the best pathological signal of cognitive decline in AD (Coleman and Yao, 2003). Brain-derived neurotrophic factor (BDNF) is a pleiotropic growth factor in the brain, and it plays a crucial role in the survival and neuronal function (Hu et al., 2019). Indeed, not only can it modulate synapse formation and neurogenesis, but it can also reduce oxidative stress and cell death. In the early stage of AD, the levels of the precursor form of BDNF, mature BDNF, or its mRNA are reduced in the parietal cortex and hippocampus (Phillips et al., 1991; Peng et al., 2005; Song et al., 2015).

There is currently no cure for AD. Unfortunately, the AD clinical trials targeting Aβ to date have been unsuccessful, demonstrating the need to investigate innovative therapeutic approaches beyond Aβ, and trying to focus attention on other early key events, in particular synaptic dysfunction, oxidative stress, or the early events of neuroinflammation (Marttinen et al., 2018). Thus, it is likely reasonable to argue that multifactorial diseases, such as AD, cannot be successfully treated by modulating a single target, but they will require multitarget drug treatment to address the different pathological facets of these diseases.

Acetylcholinesterase (AChE) inhibitors and *N*-methyl-daspartate antagonists are the current therapies for AD-related symptoms with poor efficacy and no evidence of disease modification (Lanctôt et al., 2009). Donepezil is a highly centrally selective, reversible, and non-competitive AChE inhibitor and currently the most frequently prescribed drug for the treatment of AD. Clinical trials with donepezil have highlighted slight but reproducible improvements in cognitive function of the treated patients as compared to placebo. However, these effects were transient because cognitive function continued to decline over time in patients (Doody et al., 2007).

As a consequence of the failure of one target–one ligand approach to provide promising results in AD treatment, new findings suggested that one molecule hitting multiple targets could represent the winning strategy to treat complex diseases (Schmitt et al., 2004). Thus, "the multi-target-directed ligand (MTDL) approach is based on the design of new scaffolds with different pharmacophoric subunits connected in a single molecule, which could modulate multiple molecular targets at the same time" (Dias et al., 2017). Considering the MTDL approach, we studied here the activity of the multitarget ligand PQM130 (**Figure 1**), which is the most promising compound of a new series of molecular hybrids synthesized by the combination of two subunits, the *N*-benzylpiperidine group present in donepezil and responsible for its AchE selectivity, linked to the feruloyl group present in ferulic acid (Dias et al., 2017). Ferulic acid is one of the degradation products of curcumin, which has already shown neuroprotective activities probably due to its ability to modify the kinetics of Aβ fibril formation, as well as to its anti-oxidative and anti-inflammatory activities (Hamaguchi et al., 2010; Sgarbossa et al., 2015). The multitarget ligand PQM130 has already been investigated for its *in vitro* anticholinesterase, metal-chelating, antioxidant, neuroprotective, and anti-inflammatory properties, in different *in vivo* models (Dias et al., 2017). Moreover, PQM130 also highlighted an interesting pharmacokinetic profile from the *in silico* evaluation of the absorption, distribution, metabolism, elimination (ADME) parameters, using the software QikProp 3.1 (Schrödinger, LLC, New York, NY, USA;

**Abbreviation:** Aβ, amyloid-β; AβO, amyloid-β oligomers; AChE, acetylcholinesterase; ACTB, actin; AD, Alzheimer's disease; ADME, absorption, distribution, metabolism, elimination; BDNF, brain-derived neurotrophic factor; DCF, 2′7′-dichlorofluorescein; DCFH-DA, 2′7′-dichlorodihydrofluorescein diacetate; DON, donepezil; ECL, enhanced chemiluminescence; GFAP, glial fibrillary acidic protein; GSH, glutathione; GR, glutathione reductase; H&E, hematoxylin/eosin; i.c.v., intracerebroventricular; i.p., intraperitoneal; LTP, long-term potentiation; MWM, Morris water maze; MTDL, multi-target-directed ligand; Nrf2, nuclear factor (erythroid-derived 2)-like 2; OD, optical density; pNA, p-nitroaniline; ROS, reactive oxygen species; TBS, Tris-buffered saline; TP53, tumor protein 53; UF, fluorescence intensity arbitrary units; VH, vehicle.

see **Supplementary Material 1**) (Dias Viegas et al., 2018). Interestingly, ADME data of PQM130 showed a good human absorption and blood–brain barrier penetration in accordance with the software reference parameters (Dias Viegas et al., 2018). A similar *in silico* approach was adopted to evaluate the PQM130 safety, using the VEGA platform (https://www. vegahub.eu/; Mario Negri Institute for Pharmacological Research, Milan, Italy), which includes various QSAR models. In particular, mutagenicity (CONSENSUS) and carcinogenicity (IRFMN/ANTARES) models reported the absence of mutagenic and carcinogen effects of PQM130 (see **Supplementary Material 2** and **3**).

In the current study, we have further examined the neuroprotective effects of the multitarget ligand PQM130 in comparison also to donepezil in a mouse AD model generated by intracerebroventricular (i.c.v.) injection of Aβ1-42 oligomers (Aβ1-42O) and discussed the molecular mechanisms with particular attention to its nootropic, neuroprotective, and neurotrophic activities.

## MATERIALS AND METHODS

## Reagents

Aβ1–42 peptides were purchased by AnaSpec (Fremont, CA, USA). Aprotinin, bovine serum albumin (BSA), CHAPS, 2'7'-dichlorodihydrofluorescein diacetate (DCFH-DA), dimethyl sulfoxide, 5,5′-dithiobis (2-nitrobenzoic acid), dithiothreitol, donepezil hydrochloride, EDTA, eosin, ethanol, glycerol, hematoxylin, Hepes pH 7.4, hexafluoroisopropanol, leupeptin, β-mercaptoethanol, sodium chloride, sodium fluoride, sodium orthovanadate, sucrose, sulfosalicylic acid, Triton-X 100, tris pH 7.5, xylen, and primary antibodies anti-synaptophysin and anti-β-actin were provided by Sigma-Aldrich (St Louis, MO, USA). Paraformaldehyde solution (4%) was provided by Santa Cruz Biotechnology (Dallas, TX, USA) and NP-40 was from Roche Diagnostic (Risch, Switzerland). Caspase substrates were purchased from Alexis Biochemicals (San Diego, CA, USA). Primary antibodies phospho-GSK3α/β (Ser21/9) and GSK3α/β, phospho-p44/42 MAPK (ERK1/2, Thr202/Tyr204) and p44/42 MAPK, and anti-GFAP were provided by Cell Signaling Technologies Inc. (Danvers, MA, USA). Secondary anti-mouse and anti-rabbit antibodies were purchased from GE Healthcare (Piscataway, NJ, USA) and fluorescein was from Life Technologies (Carlsbad, CA, USA). Bradford assay solution, enhanced chemiluminescence (ECL) solution, Trisbuffered saline (TBS), and Tween 20 were purchased from Bio-Rad Laboratories S.r.L. (Hercules, CA, USA). Normal goat serum (NGS) was provided by Wako Pure Chemical Industries (Osaka, Japan). All experiment reagents were reagent grade and commercially available.

## Animals

Adult male C57Bl/6 mice (9 weeks old, 25–30 g body weight; Harlan, Milan, Italy) were utilized. The mice were housed in a temperature-controlled room (23–24°C) with free access to food and water and presented with 12 h light/12 h dark cycles. Briefly, procedures on the mice were carried out according to the European Communities Council Directive 2010/63/EU and the current Italian Law on the welfare of the laboratory animal (D.Lgs. n.26/2014). The animal protocol was approved by the Italian Ministry of Health (Authorization No. 291/2017-PR) and by the corresponding committee at the University of Bologna. The number of experimental animals was minimized and care was taken to limit mice suffering.

## Experimental Design

The animals were randomized into five groups (*n* = 10/group): Sham/VH, Aβ/VH, Aβ/DON, Aβ/PQM130 0.5 mg/kg, and Aβ/ PQM130 1.0 mg/kg. Four groups were treated with Aβ1-42O by a unilateral i.c.v. injection, while the other received a unilateral i.c.v. injection of saline solution (sham group). One hour after the brain lesion, mice received intraperitoneal (i.p.) treatment of 1 mg/kg of donepezil hydrochloride (DON, Sigma-Aldrich), 0.5 or 1 mg/kg of PQM130, or vehicle (VH, saline). The dose injected was selected according to the literature (Furukawa-Hibi et al., 2011; Dias et al., 2017). We treated the mice daily for 10 days. At the conclusion of the treatment period, the mice underwent behavioral assessment. After the behavioral analysis, the animals were deeply anesthetized before being sacrificed by cervical dislocation to collect the samples for immunohistochemical and neurochemical analysis (for experimental design, see **Figure 2**).

## A**β**1-42 Oligomers Preparation and Injection

Aβ1–42 peptides (AnaSpec) were solubilized to 1 mg/ml in hexafluoroisopropanol before being sonicated and lyophilized at room temperature. The unaggregated Aβ1–42 film obtained was dissolved to a final concentration of 1 mM with sterile dimethyl sulfoxide and stored at −20°C until use. The Aβ1–42O were prepared according to the protocol of Tarozzi et al. (2008). Briefly, to enhance oligomer formation, the Aβ1-42 stock was diluted in saline buffer at 40 μM and incubated for 48 h at 4°C (Hong et al., 2007; Maezawa et al., 2008). Six microliters of Aβ1-42O (40 μM) were injected i.c.v., using a stereotaxic mouse frame (myNeuroLab, Leica-Microsystems Co., St. Louis, MO, USA) and a 10-µL Hamilton syringe, at a rate of 0.5 ml/min. After the injection, the needle was left in place for a few minutes before being retracted slowly and the wound was cleaned and sutured. The sham mice received the corresponding volume of saline. The following coordinates were used: anteroposterior: +0.22, mediolateral: +1.0, dorsoventral: −2.5, with a flat skull position.

## Donepezil Hydrochloride and PQM130 Preparations

Donepezil hydrochloride was purchased from Sigma-Aldrich and PQM130 (purity 98% by HPLC) was synthesized and provided by Professor Claudio Viegas Jr from the PeQuiM-Laboratory of Research in Medicinal Chemistry, Institute of Chemistry, Federal University of Alfenas (Alfenas, MG, Brazil). Briefly, the powders were solubilized and aliquoted in sterilized saline (donepezil) or in dimethyl sulfoxide (PQM130). The work solutions were prepared at a concentration of 0.1 mg/ml (donepezil and PQM130) and

0.05 mg/ml (PQM130) in sterilized saline. Animals were daily i.p. injected with 1 mg/kg solution (donepezil and PQM130) or 0.5 mg/kg (PQM130) for 10 days.

## Behavioral Analysis

All the tests were performed between 9.30 a.m. and 3.30 p.m. All scores were attributed by a blinded observer.

## Morris Water Maze (MWM)

The test was performed as described previously (Morroni et al., 2018a). Briefly, the apparatus was a circular plastic tank (1.0 m diameter, 50 cm height) filled with water and milk (22°C), and a submerged platform (1.5 cm under the water surface) positioned in the center of one of the four quadrants of the maze. A camera was placed to register mice's movements and send data to an automated tracking system (EthoVision, Noldus, The Netherlands). For each training trial, animals were placed into the pool at one of the four positions selected randomly, and the latency to find the hidden platform was recorded. Mice that could not reach the platform within 60 s were guided to it by the experimenter. After the trial, each mouse was placed under a warming lamp in a holding cage for 25 s until the next trial. Training trials were conducted four times a day for 5 days. On day 6, the platform was removed and animals were allowed to swim freely for 60 s. The parameters measured during the probe trial were escape latency, frequency in the platform zone, and time spent in the opposite quadrant to the platform zone.

## Y-Maze Test

The spatial working memory was evaluated by recording spontaneous alternation behavior in the Y-maze as described earlier (Sarter et al., 1988). Briefly, each arm of the maze [Ugo Basile® S.r.L., Gemonio (VA), Italy] was 35 cm long, 15 cm high, and 5 cm wide and converged to a 120° angle. The mice were positioned at the end of the A arm and allowed to move freely through the maze for 5 min. The entry in all three arms consecutively was counted as an alternation. Thus, the number of maximum alternations was

calculated as the total number of arm entries minus two and the percentage of alternation was calculated as (actual alternations/ maximum alternations) × 100 (Lopes et al., 2018).

## Tissue Preparation for Immunohistochemistry and Neurochemical Analysis

At the end of behavioral tests, the mice were deeply anesthetized and sacrificed by cervical dislocation. The brains were quickly removed and one hemisphere of each mouse was fixed in 4% paraformaldehyde (Santa Cruz Biotechnology) for 48 h. The other hemispheres were immediately removed, and the hippocampi were isolated on ice and transferred to liquid nitrogen.

For the protein extraction, the tissues were homogenized in lysis buffer and the cytoplasmic protein concentration was determined by the Bradford method (Bradford, 1976).

## Determination of Caspase-9 and -3 Activations

Caspase-9 and -3 enzyme activities were measured according to Movsesyan et al. (2002). Briefly, the tissue lysates were incubated with the assay buffer and a 50 mmol/L concentration of chromogenic *p*-nitroaniline (pNA) substrate (caspase-9, Ac-Leu-Glu-His-Asp-pNA; caspase-3, Z-Asp-Glu-Val-Asp-pNa; Alexis Biochemicals). Each sample was incubated for 3 h at 37°C and the amount of pNA released was measured with a microplate reader (GENios, TECAN®, Mannedorf, Switzerland) at 405 nm. The values were expressed as the mean ± SEM of optical density (OD) of each experimental group.

## Determination of Cellular Redox Status

The redox status, in terms of reactive oxygen species (ROS) formation, was evaluated by measuring the oxidation of DCFH-DA to 2′7′-dichlorofluorescein (DCF) (Morroni et al., 2014). The samples (60 μl) were incubated for 30 min with 2 mg/ ml of DCFH-DA, and the conversion into the fluorescent product DCF was measured (excitation at 485 nm, emission at 535 nm) using a microplate reader (GENios, TECAN®). The values were normalized to protein content and expressed as the mean ± SEM of fluorescence intensity arbitrary units (UF) of each experimental group.

## Determination of Glutathione Content

Glutathione (GSH) content was assessed using the protocol described earlier (Morroni et al., 2018b). Briefly, samples were deproteinized with 4% sulfosalicylic acid, and the supernatants were added to 5,5′-dithiobis (2-nitrobenzoic acid) (4 mg/ml). The developed coloration was read quickly at 412 nm (GENios, TECAN®) and the results were calculated using a standard calibration curve. The values were normalized to protein content and expressed as the mean ±SEM of GSH mmol/mg protein of each experimental group.

## Western Blotting

The samples (30 μg proteins) were run on 4–15% SDS polyacrylamide gels (Bio-Rad Laboratories S.r.L.) and electroblotted onto 0.45 μm nitrocellulose membranes. The membranes were incubated at 4°C overnight with primary antibody recognizing phospho-GSK3α/β (Ser21/9), phospho-p44/42 MAPK (ERK1/2, Thr202/Tyr204) (1:1,000; Cell Signaling Technology Inc), or anti-synaptophysin (1:1,000; Sigma-Aldrich). After washing with TBS-T (TBS + 0.05% Tween20), the membranes were incubated with secondary antibodies (1:2,000; GE Healthcare). ECL was used to visualize the bands (Bio-Rad Laboratories). The membranes were then reprobed with GSK3α/β, p44-42 MAPK (1:1,000; Cell Signaling Technology Inc.), or anti-β-actin (1:1,000; Sigma-Aldrich). The data were analyzed by densitometry, using Quantity One software (Bio-Rad Laboratories® S.r.L.). The values were normalized and expressed as the mean ± SEM of the densitometry in each experimental group.

## Immunohistochemistry

The fixed brains were sliced on a vibratome (Leica Microsystems, Milan, Italy) at 40 μm thickness, and the slices were stained as described earlier (Morroni et al., 2016).

## Hematoxylin/Eosin Staining

Hematoxylin/eosin (H&E) staining was assessed as previously illustrated (Fischer et al., 2008). Briefly, the selected sections were rehydrated by a graded series of alcohols (Sigma-Aldrich). Then, the slices were counterstained in hematoxylin for 8 min and then rinsed for 10 min in tap water. Subsequently, the slices were immersed in distilled water and then in 80% ethanol before being stained in 25% eosin solution (in ethanol 80%) for 1 min. Finally, the slices were dehydrated with graded alcohol before being fixed in xylen.

## Anti-Glial Fibrillary Acidic Protein (GFAP) Staining

The immunofluorescence staining was assessed according to our previous study (Morroni et al., 2018a). Selected slices were rinsed in phosphate buffer and then incubated in TBS-A (TBS 0.1% Triton-X 100) and TBS-B (TBS-A 2% BSA) to reduce a specific absorption. The sections were then incubated with anti-GFAP primary antibody (1:300; Cell Signaling Technology Inc.) in TBS-B with 3% NGS (Wako Pure Chemical Industries) at 4°C overnight. After 24 h, the slices were washed with TBS-A and TBS-B before being incubated with secondary antibody (1:200; Fluorescein, Life Technologies) in TBS-B with 3% NGS. To verify the binding specificity, some sections were incubated with only primary or secondary antibody. In these conditions, we did not find any positive staining.

## Quantitative Images Analysis

Image analysis was conducted by an investigator unaware of the treatment groups, using a microscope (AxioImager M1, Carl Zeiss, Oberkochen, Germany) and an image analysis system (AxioCam MRc5, Carl Zeiss) equipped with dedicated software (AxioVision Rel 4.8, Carl Zeiss). The hippocampal region was defined at low magnification (2.5× objective), and the H&E or GFAP staining was evaluated by densitometry of five different sections for each sample analyzed at a higher magnification (10×, 20×, or 40× objective). Quantification and morphological analysis were assessed with the ImageJ software.

## RNA Preparation and Gene Expression Analysis

Total RNA was isolated from hippocampus using the Pure link RNA mini kit (Ambion, Thermo Fisher Scientific, Carlsbad, CA, USA), as illustrated earlier (Morroni et al., 2018b). Briefly, the samples were lysed on ice with 1% β-mercaptoethanol by using a homogenizer SHM1 (Stuart, Bibby Scientific LTD, Staffordshire, UK). The samples were then added to an equal volume of 70% ethanol. The solution was filtered using a cartridge containing a clear silica-based membrane to which the RNA binds. RNA was finally eluted with RNase-free water and stored at −80°C. RNA was quantified by spectrophotometric analysis and reversetranscribed using High Capacity cDNA Reverse Transcription kit (Applied Biosystems, Thermo Fisher Scientific).

The mRNA encoding for the mouse nuclear factor (erythroidderived 2)-like 2 (Nrf2), GSH reductase (GR), tumor protein 53 (TP53), and the actin (ACTB) as internal reference were quantified by Taqman RT-PCR with a 7900HT Fast Real-Time PCR system (Applied Biosystems). The samples were run in 96-well format in triplicate. The specific Taqman gene expression assays (Applied Biosystems) were Nrf2 (Mm0047784\_m1), GSTP1 (Mm04213618\_ gH), GR (Mm00439154\_m1), TP53 (Mm01731290\_g1), and ACTB (Mm00607939\_s1).

To assess mRNA levels of different BDNF transcripts (total form, long 3′UTR form, exon IV, exon VI) and synaptophysin, samples were processed for RT-PCR reaction and subsequently analyzed by qRT-PCR instrument (CFX384 Real-Time system, Bio-Rad Laboratories S.r.l.) using the iScript one-step RT-PCR kit for probes (Bio-Rad Laboratories S.r.l.). The samples were run in 384-well format in triplicate as multiplexed reactions with a normalizing internal control (ACTB). The primers and probe sequences, respectively, were as follows: total BDNF (Fwd: AAGTCTGCATTACATTCCTCGA, Rev: GTTTTCTGAAAGA GGGACAGTTTAT, Probe: TGTGGTTTGTTGCCGTTGCCA AG), long 3′UTR BDNF (Fwd: GTTGTCATTGCTTTACTGGCG, Rev: AATTTTCTCCATCCCTACTCCG, Probe: AATCTACCCC TCCCATTCCCCGT), BDNF exon IV (Fwd: AGCTGCCTTGAT

GTTTACTTTG, Rev: CGTTTACTTCTTTCATGGGCG, Probe: AGGATGGTCATCACTCTTCTCACCTGG), BDNF exon VI (Fwd: GGACCAGAAGCGTGACAAC, Rev: ATGCAACCGAAG TATGAAATAACC, Probe: ACCAGGTGAGAAGAGTGATGAC CATCC), Synaptophysin (Fwd: CCTGTCCGATGTGAAGATGG, Rev: AGGTTCAGGAAGCCAAACAC, Probe: ACACATGCAAG GAACTGAGGGACC), and ACTB (Fwd: ACCTTCTACAATGA GCTGCG, Rev: CTGGATGGCTACGTACATGG, Probe: TCTG GGTCATCTTTTCACGGTTGGC).

Each RT-PCR run followed the manufacturer's conditions: an incubation at 50°C for 10 min (RNA retrotranscription), followed by a step at 95°C for 5 min (TaqMan polymerase activation). Subsequently, 39 cycles of PCR were performed (95°C for 10 s, and then 30 s at 60°C). A comparative cycle threshold (Ct) method was used to determine the relative target gene expression versus the sham group (Rossetti et al., 2016). Specifically, a fold change for each target gene relative to ACTB was determined by the 2−Δ(ΔCt) method, where ΔCt = Ct, target – Ct, β-actin; Δ(ΔCt) = Ct, exp. group – Ct, control group and Ct is the threshold cycle. For graphical clarity, the obtained data were then expressed as percentage versus the Sham/VH, which has been set at 100%.

## Statistical Analysis

The data were analyzed with the PRISM 5 software (GraphPad Software, La Jolla, CA, USA) and expressed as mean ± SEM of each experimental group. The difference between the groups was analyzed by one-way ANOVA with Bonferroni *post hoc* test. The results were considered statistically significant when a *p* value was less than 0.05.

## RESULTS

## PQM130 Ameliorated A**β**1-42O-Induced Cognitive Deficits in Mice

The i.c.v. injection of Aβ1-42O induced cognitive impairment as shown in the MWM and Y-maze tests. During the MWM training phase, all the mice learned the platform location, as clearly highlighted by the decreased latency and the distance traveled to find the platform. However, the Aβ/VH mice needed more time and traveled a longer distance to locate the platform than the sham mice, which undoubtedly highlighted a shortterm memory impairment in these mice. From the fourth day of training, the treated groups (Aβ/DON and Aβ/PQM130)

showed a significantly lower escape latency than those in the Aβ/VH group (*p* < 0.05; **Figure 3A**). The swimming speed was not significantly different among the groups during the training (data not shown). In the probe trial, the mice in the Aβ/VH group revealed difficulties in locating the original position of the removed platform (longer latency to first enter the target zone, less frequency crossing the platform, and more time spent swimming in the opposite quadrant; **Figure 3B**–**D**). Interestingly, the Aβ/DON and Aβ/PQM130 mice performed better than the Aβ/VH, even though significantly only with regard to time spent in the opposite quadrant (donepezil *p* < 0.01; PQM130 *p* < 0.05 and *p* < 0.01, respectively). In the Y-maze test, which assesses spatial working memory, the spontaneous alternation behavior of Aβ/VH group was significantly lower than the sham group (*p* < 0.05, **Figure 4**), confirming the difficulty in remembering which arm has already been visited. This behavioral impairment was significantly improved in the Aβ/DON and Aβ/PQM130 groups (donepezil *p* < 0.001; PQM130 *p* < 0.05 and *p* < 0.001, respectively), demonstrating that DON and PQM130 could effectively increase spatial working memory in the early stage of AD development.

## PQM130 Prevented A**β**1-42O-Induced Neuronal Death in Mice

We next observed the pathologic changes in different hippocampal areas through H&E-stained sections from the sham, the Aβ/VH group, and the mice under different treatments (DON and PQM130 treatment groups: 1 and 0.5 mg/kg). In the Aβ/VH mice, H&E staining exhibited irregular and sparse neuronal arrangements in the CA1, CA3, and DG regions of the hippocampus. We also observed many unhealthy neurons (**Figure 5A**). Interestingly,

on the performance in the Y-maze test in the Aβ1-42O-injected mice. The spontaneous alternation percentage was recorded in a 5 min trial. The values are expressed as mean ± SEM (*n* = 10) (#*p* < 0.05 vs. Sham/VH, \**p* < 0.05 and \*\*\**p* < 0.001 vs. Aβ/VH; ANOVA, *post hoc* test Bonferroni).

PQM130 treatment but not donepezil ameliorated neuronal injury compared with the saline-treated Aβ group (*p* < 0.01, **Figure 5B**). In AD, increased p53 level was detected in various parts of patient brains (Cenini et al., 2008) when compared to the brains of healthy individuals. Likewise, data from animal AD models showed an increase in p53 level in affected neurons (Ohyagi et al., 2005). As could be expected, the Aβ treatment induced the up-regulation of p53 at gene level. On the contrary, PQM130 but not donepezil significantly down-regulated p53 expression (*p* < 0.05, **Figure 6A**). Subsequently, to elucidate the underlying mechanisms of the PQM130 improvement on Aβ-induced neuronal damage, the activations of caspase-9 and -3 were detected. Once activated, the caspase-9 cleaves and activates the effector procaspase-3 triggering the apoptotic pathway. As shown in **Figure 6B** and **C**, the caspase-9 and -3 were markedly activated in the hippocampal samples of the Aβ1-42O-treated group, when compared to the sham group (*p* < 0.05). However, PQM130 treatment was able to inhibit the activation of both caspases induced by Aβ1-42O, especially at the highest dose (*p* < 0.05 and *p* < 0.01, respectively), while donepezil was effective to counteract the activation of the caspase-3 but not caspase-9 (*p* < 0.05).

## PQM130 Antagonized A**β**1-42O-Induced Oxidative Stress in Mice

As shown in **Figure 7A** and **B**, the Aβ1-42O injection induced a predictable oxidative stress to the mice brain, as underlined by significant increased ROS formation (*p* < 0.001) and decreased GSH levels in the hippocampal samples compared to the sham group. However, the administration of PQM130, but not donepezil, resulted in the significant decrease of ROS compared with the Aβ/VH group (*p* < 0.001 and *p* < 0.01, respectively). Moreover, PQM130 treatment increased GSH levels in the hippocampi of the Aβ/VH mice close to the sham group levels, particularly with the 0.5 mg/kg dose group (*p* < 0.01). In addition, we carried out gene expression profiling as an effective biomarker to detect cellular stress. In this study, the gene expression analysis for GR enzyme and Nrf2 demonstrated that Aβ treatment decreased GR mRNA expression levels, while donepezil and PQM130 (0.5 mg/kg) significantly increased GR mRNA levels (*p* < 0.01 and *p* < 0.05, respectively; **Figure 7C**). As expected, the expression of Nrf2 was found to be significantly decreased in the hippocampi of the Aβ/VH mice (*p* < 0.001); conversely, PQM130 (1 mg/kg) treatment markedly up-regulated the mRNA levels of Nrf2, compared to the Aβ/VH mice (*p* < 0.001).

## PQM130 Regulated GSK3**β** and ERK1/2 Protein Expressions in Mice

Because glycogen synthase kinase 3β (GSK3β) played a pivotal role in the pathogenesis of AD (Llorens-Martín et al., 2014), we examined the phosphorylation levels of GSK3β (Ser9) to investigate its potential involvement in the PQM130 mechanism of neuroprotection (**Figure 8A**). As shown in **Figure 8A**, the levels of phosphorylated GSK3β was decreased, although not significantly, in the Aβ/VH group. However, the treatment with PQM130 at a dose of 0.5 mg/kg significantly increased the levels

of phosphorylated GSK3β protein (*p* < 0.05). In addition, the phosphorylation of ERK1/2 was also detected in our model, since the MAPK/ERK1/2 signaling pathway is involved in the modulation of neuronal apoptosis and may contribute to AD pathogenesis (Morroni et al., 2016). Results found the Aβ1-42O injection increased the phosphorylation of ERK1/2 compared with the sham group (*p* < 0.001). However, treatment with PQM130 and donepezil markedly repressed the phosphorylation of ERK1/2 induced by Aβ1-42O (*p* < 0.05, **Figure 8B**), indicating that the dephosphorylation of ERK1/2 concurred to the antiapoptotic effect of PQM130.

## PQM130 Reduced A**β**1-42O-Induced Astrocytic Activation in Mice

To examine the effects of PQM130 on neuroinflammation induced by Aβ1-42O, we performed immunohistochemical staining for the astrocyte marker GFAP. The quantitative analysis showed that the percentages of the GFAP-stained hippocampal areas were markedly increased in the Aβ/VH group compared with the Sham/VH group (*p* < 0.01). However, in the PQM130 treated mice (1 mg/kg), GFAP-positive areas decreased (*p* < 0.01, **Figure 9B**) compared to those in the vehicle-treated Aβ1-42O mice. These results suggested that PQM130 treatment alleviated the neuroinflammation induced by Aβ1-42O in the AD brain.

## PQM130 Modulated Synaptic Plasticity in Mice

Firstly, we analyzed the total BDNF gene expression in our samples and the results did not show any significant difference among the different experimental groups (**Figure 10A**). In order to clarify the different responsiveness to PQM130, the expression profile of some neurotrophin transcripts, namely, long 3′UTR BDNF and exons IV and VI, were investigated (**Figure 10B**–**D**). In deep, PQM130 (1 mg/kg) increased significantly the expression of long 3′UTR BDNF (*p* < 0.05, **Figure 10B**) and isoform IV (*p* < 0.05, **Figure 10C**), whereas no changes were found in the other experimental groups. Classic effects of BDNF consist of promoting differentiation, migration, and dendritic arborization, and enhancing neuronal viability. In addition to these recognized actions, recent findings highlighted that BDNF affects development, function, and plasticity in the synapse (Kuczewski et al., 2009). Thus, we next investigated the effect of PQM130 on the pre-synaptic protein synaptophysin. As shown in **Figure 11A**, there is a slight decrease in synaptophysin mRNA levels in the Aβ/VH and Aβ/DON groups, while the values of the PQM130 groups were maintained at the sham group levels. Even more interesting, the Western blot analysis (**Figure 11B**) revealed a more pronounced reduction of synaptophysin expression in the Aβ/VH and Aβ/DON hippocampal samples.

However, after PQM130 treatment (1 mg/kg), the expression of synaptophysin was significantly increased as compared to the Aβ/VH group (*p* < 0.05).

## DISCUSSION

The inhibition of AChE activity is the most realistic approach in the symptomatic treatment of mild to moderately severe AD. Patients are currently treated with AChE inhibitors, and among these, the first-line symptomatic drug is donepezil. In the light of the increasingly accepted conception of AD as a complex pathological network, intensive efforts are being made in the search of new drugs that can simultaneously hit several key biological targets of the network, including AChE. Moreover, AD has decades-long preclinical period (Jack and Holtzman, 2013), which suggests the need to find early therapeutic agents with efficacy at initial stages of the AD pathology. Taking into account all these considerations, the present study aimed to assess the efficacy of the feruloyl– donepezil hybrid PQM130 on AD neurodegenerative processes and on cognitive outcomes, trying to make also a comparison with donepezil activity. In our previous study, PQM130 had already shown an interesting *in vivo* anti-inflammatory activity and *in vitro* metal chelator activity, as well as neuroprotective activity against oxidative damage (Dias et al., 2017). Here, we have elucidated the multifaceted activities of PQM130, like decreasing neuronal death and oxidative stress, improved neurotrophic effect, counteracted inflammation, and ameliorated spatial memory functions as compared to the Aβ1-42O lesioned group. It is clear that a successful neuroprotective and neurotrophic strategy could not only delay the

progression of neurodegeneration but also provide improvements in the disease condition.

In the MWM test, two main parameters are necessary to locate the hidden platform. Firstly, the mice should develop skills needed to handle the stressful condition, like swimming and recognizing the hidden platform as the only escape route. The second parameter is the spatial learning component, which implies that the mice have to learn exactly the platform's position and reach it within a minute from the different starting position (Broadbent et al., 2004; Ghumatkar et al., 2015). Here, we found a progressive improvement in the spatial memory as shown by the significant reduction in escape latency time in the PQM130-treated mice as compared to the Aβ/VH mice when evaluated on the 4th and 5th days. This improvement may be ascribed to PQM130's ability to reduce oxidative stress and AChE activity to finally enhance cholinergic neuronal transmission. The Aβ/DON group showed a swimming performance comparable to the mice treated with the same dose of PQM130. This effect of donepezil may be related to its AChE inhibition (Ghumatkar et al., 2015). However, the probe trial was not implemented significantly in this study, only the time spent in the opposite quadrant markedly decreased after PQM130 and donepezil treatments. Thus, the reduced escape latency time in the PQM130-treated group demonstrates its interesting effect on spatial learning ability. Working memory has been previously reported to be negatively involved in the early stages of AD (Kim et al., 2014; Okamoto et al., 2018), and spontaneous alternation behavior in the Y-maze test may be considered as a reflection of this kind of short-term memory. The continuous spontaneous

alternation in the Y-maze test can both elude stressful handling of animals and provide memory and locomotor evaluation (Kirshenbaum et al., 2015). Interestingly, we showed that PQM130 counteracted the negative effect of Aβ1-42O on working memory in a dose-dependent manner. This was highlighted by the significant enhancement in percent of alternation behavior in the Y-maze, and the effects of the highest dose of PQM130 are comparable to those of donepezil. Our data are in agreement with previous studies showing that donepezil significantly improves alternation deficits in this test (Meunier et al., 2006; Hu et al., 2012) in the Aβ-injected mice.

Although it is not clearly known how Aβ injection can induce memory impairment in mice, we previously established that Aβ directly caused apoptosis leading to neuronal cell loss and, ultimately, neurodegeneration (Yao et al., 2005; Morroni et al., 2016). The memory and cognitive decline in AD are strongly related to the apoptotic pathway (Obulesu and Lakshmi, 2014; Xu et al., 2017), which involves mitochondrial dysfunction, caspase activation, and DNA fragmentation (Ramalho et al., 2008). Our data showed that the hippocampal damage and caspase-9 and -3 activations in lesioned mice were markedly reversed by PQM130. Meanwhile, donepezil did not show the same effectiveness in counteracting apoptosis and neuronal damage. Moreover, increased p53 level is infallibly detectable in brain areas attained by AD, in the corresponding brain areas of animal models, and in neuronal cells isolated from AD brains (Szybińska and Leśniak, 2017). Interestingly, PQM130 substantially reduced the expression of p53, which corroborated its antiapoptotic activity. It is known that p53 directly binds to and increases the activity of GSK3β while inhibition of nuclear GSK3β attenuated p53-dependent transcription (Watcharasit et al., 2002). The link between p53 and GSK3β (i.e., between p53 and tau phosphorylation) may be more complex; however, in this study, we found that the decrease in p53 expression levels after PQM130 treatment is most likely reflected in a phosphorylation (and thus deactivation) of GSK3β, leading to protection against neuronal death induced by Aβ1-42O.

ANOVA, *post hoc* test Bonferroni).

Several studies demonstrated that oxidative stress precedes the rise of senile plaques and neurofibrillary tangles, therefore leading to dementia's symptoms (Wang et al., 2014; Tian et al., 2019). Indeed, the increase of Aβ and oxidative stress, causing neuronal cell death, are common mechanisms in the progression of AD (Lee et al., 2011). Here, we found that PQM130 and not donepezil significantly ameliorates oxidative damage as demonstrated by the increase of GSH levels and GR and Nrf2 expressions in the Aβ1-42O-treated mice, confirming the similar evidences recorded by PQM130 in neuronal SH-SY5Y cells (Dias et al., 2017). The role of Nrf2 in Aβ-induced oxidative stress is controversial (Rong et al., 2018). Ramsey and colleagues described that in AD brains, Nrf2 is mostly found in the cytoplasm in its inactive form, which means that Nrf2 does not trigger the expression of antioxidant enzymes (Ramsey et al., 2007). Sarkar and colleagues demonstrated that Aβ25-35 increased oxidative stress and suppressed Nrf2 activation (Sarkar et al., 2017). Moreover, Branca et al. showed that reducing Nrf2 levels exacerbated cognitive impairments in a transgenic model of AD. They also speculated that "Nrf2 might act as a molecular link between brain aging and AD" (Branca et al., 2017). Numerous laboratories supported this hypothesis showing that Nrf2 activity decreased with aging (Suh et al., 2004; Zhang et al., 2015; Li et al., 2018). Moreover, Nrf2 activity is strictly related to tau pathology, enhancing the link between Nrf2 and AD (Lastres-Becker et al., 2014). In our model, exposure to Aβ1-42O caused a marked decrease in Nrf2 activation, and only PQM130 significantly increased its expression, probably due to the presence of ferulic acid in this hybrid molecule. In this regard, the presence of α, β-unsaturated carbonyl system in PQM130 suggests the ability of this molecule to activate Nrf2 through a Michael addition reaction (de Freitas Silva et al., 2018). Thus, the effect of PQM130 on Aβ1-42O-induced oxidative injury could explain the ability of PQM130 to counteract apoptotic cell death and cognitive impairment observed in our model.

Activity of ERK1/2 is modulated by ROS, and several studies demonstrated its activation in different AD models (Zhu et al., 2002; Chong et al., 2006; Gan et al., 2014; Chang et al., 2018). Moreover, inhibition of ROS formation decreased ERK1/2 activation in an AD model (Kim et al., 2009). The ERK pathway is fundamental to memory consolidation and synaptic plasticity in the hippocampus. Moreover, the fine regulation of ERK is crucial for the hippocampal functions (Goedert and Spillantini, 2006). Notably, in our model, Aβ1-42O contributed to the abnormal activation of ERK1/2 and there was an obvious decrease of p‐ERK1/2 levels by PQM130 administration.

ERK activation is also found in reactive astrocytes, affecting Aβ production through ROS formation (Kim and Wong, 2009). Therefore, compounds inactivating astrocytes and MAPK pathways could reduce Aβ formation and thus prevent or counteract neuronal injury in the AD brain (Butterfield, 2002; Lee et al., 2011). Additionally, glial cells and their resident protein GFAP are able to combine neuronal input, control synaptic activity, and translate signals tightly linked to learning and memory by the formation of cytoskeletal filaments (Konar et al., 2011; Ghumatkar et al., 2015). Our results showed that PQM130 and not donepezil might alleviate reactive gliosis, by reason of the ability of this treatment to reduce levels of GFAP in the hippocampus of the Aβ1-42O-lesioned mice.

BDNF belongs to the neurotrophin family of survivalpromoting molecules. It exerts significant protective effects on fundamental neuronal pathways altered in AD (Nagahara et al., 2009). The transcription of the BDNF gene is very elaborated. At least eight promoters encode to different mRNA transcripts, each containing a 5′ exon spliced to a common 3′ coding exon, and all of which generate the same BDNF protein (Aid et al., 2007; Chapman et al., 2012). We examined the expression of total BDNF mRNA, two 5′ exon-specific transcripts (IV and VI), and BDNF mRNA transcripts with a long 3′ untranslated region (3′UTR) in the hippocampal samples. BDNF mRNA transcripts with long 3′UTRs play essential roles in dendritic spine morphology and long-lasting synaptic plasticity (An et al., 2008; Chapman et al., 2012). In our model, the expression of these transcripts was not reduced by Aβ1-42O injection; however, PQM130 markedly increased long 3′UTR and exon IV. Intriguingly, the increased level of BDNF in the hippocampus was accompanied by an up-regulated expression of synaptophysin. Therefore, these neurochemical results lead us to assume that PQM130 may activate the BDNF signaling pathway and thus control the expression of its downstream signaling components and the structural proteins associated to synaptic plasticity in the hippocampus, improving cognitive deficits in mice.

In conclusion, the results of this study demonstrated the nootropic, neuroprotective, and neurotrophic activities of the multi-target drug PQM130 in our AD experimental model. The nootropic effect could be related to the inhibition of AChE activity and the modulation of neuronal survival pathways, and consequently ameliorating the spatial memory formation. Neuroprotection might be attributed to its high potential as antioxidant, and to its ability to counteract apoptotic death and inflammation. Neurotrophicity might be ascribed to its increased BDNF and synaptophysin levels in the hippocampus. Compared to the first-line treatment donepezil, PQM130 appears a more attractive multipotent therapeutic molecule. Thus, our research findings prospect PQM130 as a promising candidate to be further investigated in AD therapy.

## DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

## REFERENCES


## ETHICS STATEMENT

Procedures on the mice were carried out according to the European Communities Council Directive 2010/63/EU and the current Italian Law on the welfare of the laboratory animal (D.Lgs. n.26/2014). The animal protocol was approved by the Italian Ministry of Health (Authorization No. 291/2017-PR) and by the corresponding committee at the University of Bologna.

## AUTHOR CONTRIBUTIONS

AT, FM, and PH contributed to the conception and design of the study. KD and CV designed and synthesized the molecule PQM130. LP, AS, and IC performed *in silico* studies. GS and AG performed behavioral, biomolecular, and immunohistochemical analysis. GR, MP, and RM performed gene expression analysis. GS performed the statistical analysis. FM and GS wrote the first draft of the manuscript. LP contributed to data analysis. PH contributed reagents/materials/analysis tools. All the authors contributed to the manuscript revision and read and approved the submitted version.

## FUNDING

This work was supported by Ministero dell'Istruzione, dell'Università e della Ricerca (MIUR), PRIN 2015 (Prot. 20152HKF3Z and 2015SKN9YT003), and Fondazione del Monte di Bologna e Ravenna.

## ACKNOWLEDGMENTS

The authors thank Dr. Silvia Zanasi for the linguistic revision of the manuscript.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2019.00658/ full#supplementary-material


brain-derived neurotrophic factor MRNAs in hippocampus. *Neurobiol. Aging* 33 (4), 832.e1–832.e14. doi: 10.1016/j.neurobiolaging.2011.07.015


ferulic acid improves cognition and reduces Alzheimer-like pathology in mice. *J. Biol. Chem.* 294 (8), 2714–2731. doi: 10.1074/jbc.RA118.004280


not inverse agonist and agonist beta-carbolines. *Psychopharmacology* 94 (4), 491–495. doi: 10.1007/BF00212843


**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 © 2019 Morroni, Sita, Graziosi, Ravegnini, Molteni, Paladini, Dias, dos Santos, Viegas, Camps, Pruccoli, Tarozzi and Hrelia. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Sleep and **β**-Amyloid Deposition in Alzheimer Disease: Insights on Mechanisms and Possible Innovative Treatments

*Susanna Cordone1, Ludovica Annarumma1, Paolo Maria Rossini2,3 and Luigi De Gennaro1\**

*1 Department of Psychology, University of Rome "Sapienza," Rome, Italy, 2 Department of Neurological, Motor and Sensory Sciences, IRCCS San Raffaele Pisana, Rome, Italy, 3 Institute of Neurology, Catholic University of The Sacred Heart, Rome, Italy*

#### *Edited by:*

*Filippo Caraci, University of Catania, Italy*

#### *Reviewed by:*

*Elena Marcello, University of Milan, Italy Sylvain Chauvette, Centre Intégré Universitaire de Santé et de Services Sociaux de la Capitale-Nationale (CIUSSSCN), Canada*

> *\*Correspondence: Luigi De Gennaro luigi.degennaro@uniroma1.it*

#### *Specialty section:*

*This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology*

*Received: 07 March 2019 Accepted: 28 May 2019 Published: 20 June 2019*

#### *Citation:*

*Cordone S, Annarumma L, Rossini PM and De Gennaro L (2019) Sleep and β-Amyloid Deposition in Alzheimer's Disease: Insights on Mechanisms and Possible Innovative Treatments. Front. Pharmacol. 10:695. doi: 10.3389/fphar.2019.00695*

The growing interest in the preclinical stage of Alzheimer's disease (AD) led investigators to identify modifiable risk and predictive factors useful to design early intervention strategies. The preclinical stage of AD is characterized by β-amyloid (Aβ) aggregation into amyloid plaques and tau phosphorylation and aggregation into neurofibrillary tangles. There is a consensus on the importance of sleep within this context: the bidirectional relationship between sleep and AD pathology is supported by growing evidence that proved that the occurrence of sleep changes starting from the preclinical stage of AD, many years before the onset of cognitive decline. Hence, we review the most recent studies on sleep disturbances related to Aβ and the effects of sleep deprivation on Aβ accumulation in animal and human models. We also discuss evidence on the role of sleep in clearing the brain of toxic metabolic by-products, with original findings of the clearance activity of the glymphatic system stimulated by sleep. Furthermore, starting from new recent advances about the relationship between slow-wave sleep (SWS) and Aβ burden, we review the results of recent electroencephalographic (EEG) studies in order to clarify the possible role of SWS component disruption as a novel mechanistic pathway through which Aβ pathology may contribute to cognitive decline and, conversely, the eventual useful role of SWS in facilitating Aβ clearance. Finally, we discuss some promising innovative, effective, low-risk, non-invasive interventions, although empirical evidence on the efficacy of sleep interventions in improving the course of AD is at the very beginning.

Keywords: sleep, **β**-amyloid, Alzheimer's disease, glymphatic system, slow-wave activity

## INTRODUCTION

Alzheimer's disease (AD) is the most common cause of dementia and represents one of the most dramatic challenges of modern society. The increase in elderly population and life expectations and the public health and economic challenges led investigators to develop sensitive biomarkers, risk and predictive factors able to facilitate early detection and effective intervention strategies (Sperling et al., 2011).

The relationship between sleep and AD is well known: a high percentage of AD patients complained sleep disturbances along the entire course of the disease, increasing in severity with the progression of AD (Prinz et al., 1982; Vitiello et al., 1990; Moe et al., 1995).

Assuming that AD pathophysiology occurs many years before the manifestation of cognitive decline, recent literature data show that research in neuroscience is focusing on the preclinical stage of AD. This stage is characterized by deposition of extracellular amyloid-β (Aβ) into insoluble plaques in the brain associated with the aggregation of protein tau into intracellular neurofibrillary tangles (e.g., Lucey and Bateman, 2014). Amyloid deposition can be measured *in vivo* in humans by cerebrospinal fluid (CSF) Aβ concentration levels and by positron emission tomography with the amyloid tracer Pittsburgh compound B (PET-PiB). Through the use of both these measures, it has been observed that amyloid deposition was present approximately in 25–30% of cognitively intact individuals in their eighth decade (Morris et al., 2009).

Starting from the assumption that the intervention strategies are most effective in the preclinical stage of the disease, new research lines are investigating the modifiable factors occurring during this stage, together with the neuropathological events. In this context, the role of sleep in relation with Aβ is increasing in importance: sleep disturbances are present since the occurrence of Aβ accumulation, denoting a strict bidirectional relationship between sleep and Aβ. The subjective and objective measures of fragmented sleep are associated with the degree of Aβ accumulation and Aβ levels in the CSF (Lim et al., 2013; Spira et al., 2013; Mander et al., 2015).

For both animal and human models, it has been observed that sleep deprivation (SD) causes the augmentation of soluble Aβ (Kang et al., 2009; Ooms et al., 2014). Furthermore, in the mouse model, SD provokes also an increase in amyloid plaque deposition (Roh et al., 2014).

With the aim of investigating which aspects of sleep could be responsible for modulation of Aβ, an increasing number of electroencephalographic (EEG) studies (e.g., Ju et al., 2017) explored the role of slow-wave sleep (SWS) and specific non rapid eye movements (NREM) SWS components—with particular reference to slow-wave activity (SWA)—as candidates in the clearance of Aβ, promoting glymphatic system activity. The glymphatic system is a perivascular network diffused in the brain that has the role of achieving the exchange in interstitial fluid and CSF (Boespflug and Iliff, 2018) and is mainly implied in the clearing process of Aβ and other interstitial solutes. Growing evidence demonstrated that the glymphatic system mainly is active during sleep and impaired with aging and post-traumatic brain (Iliff et al., 2014; Kress et al., 2014; Zeppenfeld, 2017).

Here, we briefly describe the bidirectional relationship between sleep and Aβ. Then, we review recent literature with particular reference to the last decade, reporting empirical evidence on the relationship between sleep disturbances and Aβ in elderly populations, and the most recent SD experimental data in animal and human models. We also discuss a new concept "beyond amyloid," underlying the importance of other factors related directly and indirectly to sleep—that has been receiving growing interest as contributors in the pathophysiology and progression of AD. On the basis of the most recent advances, we also discuss findings on the role of sleep in clearing the brain of toxic metabolic by-products, providing the results of new studies about the clearance activity of the glymphatic system stimulated by sleep, with particular reference on the role of SWA.

On the basis of the reviewed data, we also report a recent promising research line, describing innovative early sleep intervention strategies.

## SLEEP AND A**β**: A BIDIRECTIONAL RELATIONSHIP

For over 25 years, sleep disorders have been associated with AD, with a 25–66% of AD patients that exhibit sleep disturbances being considered one of the leading causes of patient institutionalization (Moran et al., 2005; Guarnieri et al., 2012).

During the last years, with the growing interest in the preclinical stage of AD, the role of sleep in association with AD has radically changed. Sleep changes occur many years before the appearance of cognitive symptoms, together with the early pathophysiological events. The presence of sleep disturbances, since the preclinical stage of the disease, underlines a possible crucial role of sleep in AD pathology and progression.

In 2009, Kang and colleagues showed in the AD mouse model that Aβ levels in the interstitial fluid increased during wakefulness and decreased during sleep. After this pioneering finding, many observational studies investigated on poor sleep as a potential human AD biomarker. Two physiological mechanisms could explain how poor sleep could promote AD: i) during SWS, the brain could be able to better clean metabolic waste, and Aβ clearance would be more effective during SWS (Xie et al., 2013); and ii) another mechanism was based on evidence that increased neuronal firing could promote Aβ production, and the firing is reduced in SWS as compared with wakefulness or REM sleep, and, consequently, sleep loss could lead to increase neuronal activity resulting in Aβ increase (Vyazovskiy et al., 2009; Ju et al., 2014). At this purpose, it is important to underline that literature data remain controversial. In particular, the open issue seems to be related to the different frequencies of neuronal firing: it is known that the majority of cortical neurons fire at low frequency (Chauvette et al., 2010; Barth and Poulet, 2012). Starting from this assumption, while sleep seems to reduce the firing of highfiring-rate neurons, in case of very-low-frequency-firing (around 1 Hz) neurons, it seems to augment the firing as underlined in evidence conducted in both rodents and cats (Watson et al., 2016). Moreover, Grosmark et al., (2012) examined the firing rates of hippocampal CA1 neurons and found that only REM episodes were related to a decreased firing rates at hippocampal level, revealing also an increase in firing during NREM sleep episodes.

During the last 15 years, many studies assessed the relations between subjective and objective sleep measures and increasing Aβ levels and lower cognitive performance in elderly populations, suggesting that poor sleep could increase the risk of obtaining low cognitive outcomes (Scullin and Bliwise, 2015). The focus on healthy elderly population represents a crucial methodological advance from earlier studies conducted in already-impaired older adults (Spira and Gottesman, 2017). Indeed, in the light of developing the most effective and early interventions, it becomes necessary to evaluate a series of age-related characteristics in a healthy population and, ideally, following across time any change in different physiological, structural, functional, and behavioral aspects.

Hence, current research lines are focusing their attention on the relationship between Aβ, sleep, and cognitive function in different experimental conditions, with particular reference to healthy elderly populations by investigating sleep disturbances and sleep deprivation related to Aβ burden.

## A**β** AND SLEEP DISTURBANCES: A NEW PERSPECTIVE

Growing evidence supports the notion that insomnia, excessive daytime sleepiness (EDS), sleep-disordered breathing (SDB), and circadian sleep–wake alterations all seem to increase the risk of AD (for a review, see Yaffe et al., 2014). It has been shown that sleep disorders modify neurotransmitter activity that could cause consequent dysfunction of the "default mode network," which has a crucial role in the pathophysiology of AD (Yulug et al., 2017).

Following the new perspective of Aβ and sleep bidirectional relationship, many studies investigated whether sleep disruption could lead to deleterious effects on Aβ accumulation in healthy populations (e.g., Cross et al., 2015).

Recently, Chen and colleagues (2018) assessed the CSF Aβ levels in 23 patients with chronic insomnia, to reveal the potential effects of chronic sleep lack on the pathogenesis of AD. The authors found that CSF Aβ42 levels are significantly increased in insomniac patients. Furthermore, Aβ levels significantly correlated with the Pittsburgh Sleep Quality Index (PSQI) scores (i.e., the most used self-report measure that assesses sleep quality).

Carvalho and co-workers (2018) conducted a longitudinal study to assess the association between EDS and Aβ levels, through the use of PiB-PET scans across time (3 years' follow-up). The results show that cognitive normal elderly (*n* = 283) with a high score of EDS at baseline condition were subjected to a higher risk of developing changes related to AD, as shown by the progressive increase of their Aβ levels over 3 years.

A similar longitudinal experimental design has been applied by Sharma and colleagues (2017) to investigate the association between obstructive sleep apnea (OSA) and Aβ levels in a cognitively intact elderly population (*n* = 208). The results show that high OSA indexes were related to higher Aβ levels measured by PiB-PET scans and the increasing in Aβ levels progressively augmented across time (2 years' follow-up).

The strong limitation of these studies is, however, that their assessment of sleep disturbances was based only on a self-report questionnaire, without the use of objective or clinical measures.

Some of these observations were, however, confirmed in studies conducted using actigraphy to examine sleep disorders related to preclinical AD. For example, Lim and co-workers (2013), through a prospective actigraphic study with 10 consecutive sleep recordings, show that participants with higher sleep fragmentation have a risk of developing AD symptoms in the successive 3 years, which is 1.5 times higher if compared with that of participants having lower fragmented sleep.

It is interesting to note that sleep fragmentation characterized, directly or indirectly, all these sleep disturbances: insomnia, EDS, and OSA are representative of poor sleep quality, restricted duration, and difficulty in maintaining sleep continuity.

The need for exploring sleep disturbances associated with Aβ with objective measures seems mandatory at this stage of evidence. Objective measures should include sleep EEG recordings, to obtain important details concerning macrostructural and micro-structural EEG measures.

Particular attention should be given to the changes in the sleep–wake cycle. Aging *per se* leads to many changes at the physiological level, and these modifications concern also sleep and circadian rhythm and could derive from hypothalamic functional alterations (Monk et al., 2011). Indeed, lateral hypothalamus contains also neurons that impact wakefulness, in terms of initiating and maintaining wakefulness state (Chemelli et al., 1999). This role seems to be accomplished by orexin (hypocretin) neuropeptides that could be considered as a sort of substrate connecting dysfunctional homeostatic and cognitive processes in case of aging or AD and has been receiving growing attention also concerning new intervention strategies based on sleep restoration (Guarnieri et al., 2014).

The hypothesis of a relevant role of orexin neuropeptide in AD has also been confirmed by a study conducted with transgenic 2567 (TG2567) mice. This type of mouse model is commonly used in AD experimental procedures because it presents a mutation of the amyloid precursor protein but does not contract signs of AD at the behavioral level. Results show that orexin intracerebroventricular administration during inactive periods (that probably correspond to sleep) lead to an augmentation of wakefulness periods and Aβ levels in interstitial fluid (Kang et al., 2009).

Many evidence shows that alterations in circadian rhythm and sleep–wake cycle are strictly related to Aβ pathology: high Aβ levels are associated with fluctuations in alertness (e.g., Musiek et al., 2015), which contribute to the successive occurrence of "sundowning" with progressive neurodegeneration across time.

Considering animal models, the major concentrations of Aβ occur during wakefulness in TG2567 mice and wild type (WT; Kang et al., 2009), and, similarly, human models denote significant differences in Aβ levels measured during wakefulness (maximum concentration) and during normal sleep (minimum concentration; Huang et al., 2012).

The changes concerning sleep–wake cycle alterations include nocturnal sleep fragmentation, increased wakefulness, and functional impairment in daytime activity with diurnal napping (Vitiello and Prinz, 1989; van Someren et al., 1996). The specific sleep alterations in the preclinical stage of AD regard SWS, while REM sleep seems to be affected in later stages, with the progression of the disease (Vitiello and Prinz, 1989). Although brain remains electrically and metabolically active during sleep, a reduction of functional connectivity occurs at sleep onset (Vecchio et al., 2017) and with increasing depth of NREM sleep, with the maximal reduction observed during stage 3 (N3) of SWS (Horovitz et al., 2009). It has been hypothesized that the decrease during N3 could be due to decreased neuronal activity in this sleep stage. In support of these findings in humans, Fernandez and colleagues in a study conducted on mice found similar results regarding the decreased connectivity during sleep (Fernandez et al., 2017).

A recent interesting contribution investigated the impact of Aβ on the sleep–wake cycle, demonstrating that Aβ disturbs the synchronization of the central nervous system clocks, which are crucial for many processes. The worsening in synchronization is typical of normal aging and could further worsen, leading to neurodegeneration (Cedernaes et al., 2017).

This hypothesis is also supported by 24-h fluctuations in CSF Aβ levels measured by PET that decrease with age in "amyloid positive" healthy adults (Kang et al., 2009; Huang et al., 2012).

## THE EFFECTS OF SLEEP DEPRIVATION ON THE **β**-AMYLOID ACCUMULATION

SD studies have been providing crucial advancements in the understanding of the mechanisms underlying the intriguing bidirectional relationship between sleep and Aβ.

It is well known that, in general, SD is strictly associated with cognitive impairment. Both clinical and experimental studies show that sleep loss, even for a few hours, provokes cognitive impairments. A wide range of empirical evidence demonstrates impairments in memory, learning, attention, decision making, and emotional reactivity in healthy human subjects after sleep loss (Chee and Chuah, 2008; Goel et al., 2009; McCoy and Strecker, 2011).

SD experimental contributions derive from both animal and human models, and although animal models represent a fundamental approach in the field of translational research in AD, the findings on the animal model do not completely overlap those in humans.

Studies manipulating SD make a distinction between total sleep deprivation (lack of sleep for a specific period extending to a longer period of sleep), mild sleep restriction (longterm shortening of the usual duration of sleep), and sleep fragmentation (interrupted sleep in different stages).

Starting from the assumption that sleep loss has an important negative implication in cognitive impairment, the most recent contributions of animal and human models in the field of AD are extending their focus on other interesting methodological and conceptual issues: i) scheduling different long-term SD periods; ii) investigating different subtypes of cognitive impairments; iii) testing the irreversibility of these impairments, by utilizing longitudinal experimental protocols; iv) increasing the use of non-genetically predisposed mouse model to better replicate adult-elderly individuals; v) considering sleep loss as a "stressor"; vi) investigating possible neuronal mechanisms implied in sleep– Aβ relationship; and vii) considering the possible role of glia in AD pathogenic mechanisms.

## Animal Models

Firstly, Rothman and colleagues (2013) tested the hypothesis that long-term mild restriction could worsen AD progression using the TG model with plaques and tangles, the 3×TgAD mouse model (*n* = 10). Mice were subjected to 6 h of sleep restriction per day for 6 weeks. The results showed that after chronic sleep restriction, there was an accentuation of Aβ accumulation in the cerebral cortex, demonstrating for the first time that longterm mild sleep restriction could worsen AD progression. Furthermore, behavioral data also reveal a worsening of memory loss in sleep-restricted mice if compared with controls. This study also analyzed corticosterone circulating levels, which were elevated 1 week after the start of the SD period and lasted for 4 weeks. The increase in corticosterone levels led to consider SD also in relation to stress intrinsically inherent to the procedures to sleep deprive, indicating that mice could not be able to acclimate to a stressor.

The notion of SD acting as a potential stressor was confirmed in other mice model studies. In particular, a successive study by Di Meco and co-workers (2014), utilizing the same 3×TgAD mouse model (*n* = 18), evaluated the functional and biological consequences of 4-h SD per day for 8 weeks. Compared with controls, mice subjected to SD show impaired cognitive performance in learning and memory abilities, but there are no significant differences in Aβ accumulation.

Interestingly, SD has an impact also in reducing postsynaptic density protein 95 levels with parallel augmentation of glial fibrillary acidic protein levels. This result confirms the importance of SD as a chronic stressor, able to influence also cognitive functions AD neuropathology.

Aiming to investigate the effects of chronic SD (mice underwent to 2-month SD; they were awakened from 12:00 PM to 8:00 AM of the next day) on cognition and Aβ accumulation in mice, Qiu and colleagues (2016) used a model of familial AD transgenic mice (*n* = 40) and their WT (*n* = 40). Results showed a worsening in cognition (learning and memory) in mice subjected to SD if compared with non-SD mice for both TG and WT models. Furthermore, the augmentation in Aβ accumulation and typical senile plaques was observed after SD in the cortex and hippocampus. It is interesting to note that the effects related to SD lasted for 3 months and remained stable also in normal, nonexperimental conditions.

These findings underline that chronic SD could be considered as a potential risk factor for AD.

This is the first study demonstrating that reversal learning ability is defected by chronic SD. The authors assume that SD as a stressor can cause damage to both the hippocampus and cortex and it could aggravate dementia. This study explores the effects of chronic SD on both familial AD model and sporadic one, to facilitate the understanding of the association between chronic SD and AD. On the basis of those results, it is possible to suggest that chronic SD is not only a risk factor for familial AD but also contributes to sporadic AD.

Familial (i.e., early-onset AD) AD typically accounts for a quite small percentage of all AD cases (e.g., Bali et al., 2012), and it becomes fundamental to evaluate whether chronic SD could facilitate Aβ accumulation and make vulnerable the nongenetically predisposed mice to the risk of dementia, specifically investigating sporadic AD (i.e., late-onset AD, which is the most common AD form). In a recent study of Zhao and co-workers (2017), non-genetically predisposed mice (adult and WT C57BL/6 mice) were submitted to chronic SD for 2 months (equivalent to 6–8 years in humans). Mice usually sleep 10–12 h each day, and experimental SD protocol allowed mice to sleep restriction (4 h per day). This study had the aim of verifying alterations in cognitive function and Aβ levels after a period of SD. Results showed that cognitive function worsened for both learning and memory domains. Concerning Aβ accumulation, the most affected brain areas were at the prefrontal and temporal lobe levels, differing from above-mentioned findings, suggesting that it could be possibly due to variant genetic background.

Beyond the traditional mice models, the fruit fly *Drosophila melanogaster* has been established as a model for AD (e.g., Bonner and Boulianne, 2011) because it recapitulates some key pathophysiological and cognitive characteristics of AD (Tabuchi et al., 2015). A recent study of Tabuchi and colleagues (2015) hypothesized a functional mutual interaction that includes sleep, neuronal excitability, and Aβ levels also in animals different from mice model. To this purpose, using mechanical deprivation, authors subjected flies to SD for 1 week, starting from the assumption that sleep functions downscale synaptic strength and SD could have a deleterious effect on neuronal excitability related to AD pathology. The results of this study underline an augmentation in neuronal excitability due to abundant Aβ accumulation and a worsening of this effect after SD. These changes in electrical activity are also linked to cognitive impairment: mild cognitive impairment (MCI) patients showed augmented hippocampal activation by functional magnetic resonance imaging (fMRI), while its reduction improved memory performance (Bakker et al., 2012). The findings of this study raise the important issue of the strong interaction between sleep and Aβ, adding the negative relevance of neuronal hyperexcitation in case of sleep loss and, on the contrary, suggesting a possible beneficial role of intact sleep in AD pathophysiology.

## Human Models

The first human study by Ooms and colleagues (2014) investigated the effects of one night of total SD on CSF Aβ42 levels in healthy middle-aged men (*n* = 26). A night of undisturbed sleep led to a 6% decrease in Aβ42 levels, whereas this decrease was not observed in case of sleep restriction. Furthermore, Aβ42 levels in a different time of the day were significantly different between the two groups in contrast to Aβ40 and tau protein levels that remained stable. These results suggest that SD could interfere with the physiological decrement in Aβ42 levels measured in the morning, and it elevates the risk of AD. Another study (Wei et al., 2017) tested the effects of short-term total sleep deprivation on plasma Aβ levels. Twenty healthy volunteers underwent 24 h of SD. Results show that SD can provoke Aβ40 plasma level augmentation and a decrement in Aβ42/Aβ40 ratio. As mentioned above for animal models, these results underline a possible mechanism that also implies oxidative stress in the impairment of peripheral Aβ clearance as a pathophysiological mechanism of AD. The plasma Aβ40 level was linearly correlated with the time of wakefulness, and the changes were reversed after sleep recovery. These sleep-related effects were not observed for plasma Aβ42.

In contrast, Lucey and colleagues (2017) found that disrupted sleep increases AD risk *via* increased Aβ38, Aβ40, and Aβ42 levels by 25% to 30% in cognitively normal adults (*n* = 8; aged 30 to 60 years).

Using PET, Shokri-Kojori and colleagues (2018) showed that acute SD impacts the Aβ burden in brain regions implicated in AD. In particular, the authors evaluated the effects of one-night SD on Aβ burden in healthy controls (*n* = 22). The increase in Aβ burden was found in the hippocampus and thalamus, confirming that SD has an impact on brain regions strictly connected with AD pathology starting from the first stages of the disease.

The most recent studies are trying to increase in the extent of SD manipulation. Olsson and co-workers (2018) investigated the effect of five consecutive nights of partial sleep deprivation (PSD) on CSF biomarkers in healthy adults (*n* = 13). After baseline condition (8 h per night for five nights), PSD protocol consisted in reducing sleep to 4 h for the successive eight nights. CSF biomarker included Aβ, tau, orexin, monoamine metabolites, neuron-derived biomarkers, and astro- and microglia-derived biomarkers. Results showed that five to eight consecutive nights of PSD only affected CSF orexin levels, in terms of augmentation. No impact was found on the other CSF biomarkers.

The above-mentioned empirical evidence related to animal and human models, with particular reference to new interesting methodological and conceptual issues, has provided new important insights going towards a "beyond amyloid" concept of AD pathologic mechanisms.

## AD AS A MULTIFACTORIAL DISEASE: BEYOND AMYLOID

The growing evidence on AD pathogenesis is depicting a new conceptual and experimental framework that overcomes the mere "amyloid cascade hypothesis," leading to even more complex considerations on other factors that could contribute, reciprocally interact, and have a role in AD pathogenesis and progression. Clinicians and researchers are increasingly considering AD as a multifactorial disease syndrome, trying to identify the roles of these different factors in relation to AD pathogenesis and progression. In recent literature, an open issue concerns the importance and the weight of the different "contributors" and their mutual interactions. A recent commentary of Mullane and Williams (2018) underlines the new consideration of AD as a "complex cellular phase consisting of feedback and feedforward responses of astrocytes, microglia, and vasculature" and a "holistic understanding of the spatial, temporal and cellular aspects of the disease process" (De Strooper and Karran, 2016) is required. In their commentary, authors did not question the already-known notions about Aβ burden as a key pathological feature in AD, but they suggest a complex framework that goes "beyond amyloid" in the understanding of AD pathogenesis. This concept also depends by clinical evidence that shows substantial independence of neurodegenerative signs (such as Aβ accumulation) from cognitive status: it has been shown that the removal of Aβ from AD brain could not have positive effects on cognition. This consideration also suggests that the amyloid cascade hypothesis of AD could be rejected (Egan et al., 2018).

In the perspective of AD as a multifactorial disease, recent studies present many causal factors implicated in AD. These factors include inflammation (McGeer et al., 1990), neurotoxic protein accumulation in the brain (Boland et al., 2018) that can be associated in part with sleep deprivation, disruption of the glymphatic system and blood–brain barrier (BBB) dysfunction (Jessen et al., 2015; Sweeney and Zlokovic, 2018), oxidative stress, and microglia dysfunction. None of these factors could provoke, independently, AD occurrence, but their combinations and interactions could have a role in facilitating its occurrence. The role of these contributing factors could clarify the fact that a high percentage of individuals that present signs of AD, such as amyloid plaques, do not show impairment in cognition (Aizenstein et al., 2008); and, on the contrary, it is possible to develop AD without observed amyloid signs in the brain (Nelson et al., 2011; Herrup, 2015).

Within this new framework, the most relevant issue concerns how sleep is inserting in the mechanistic pathways underlying AD pathogenesis, to develop new preventive and early intervention strategies.

It is important to underline that many recent contributions investigated sleep also in relation to tau protein with dissociated findings for changes of Aβ and tau (e.g., Holth et al., 2017). Also, for this reason, the relationship between sleep and tau is not the subject of the present review.

## THE ROLE OF SLEEP IN CLEARING THE BRAIN OF TOXIC METABOLIC BY-PRODUCTS

Recent reports indicate a strict relation between disrupted sleep, brain glymphatic system, and AD (O'Donnel et al., 2015; Krueger et al., 2016). Removing waste from the central nervous system is crucial for the maintenance of brain homeostasis during life span, and, at this purpose, the role of the glymphatic system has been profoundly investigated. The glymphatic system includes a perivascular network for CSF transport (Iliff et al., 2012) connected to a downstream lymphatic system (Aspelund et al., 2015; Louveau et al., 2015). Xie and colleagues (2013) conducted the first pioneering studies that showed as Aβ protein was transported from the interstitial fluid space and out of the brain through the glymphatic system. The association between sleep and glymphatic system derived from evidence that demonstrate as sleep (with particular reference to SWS) augmented the action of the glymphatic system in clearing Aβ from the brain if compared with wakefulness state, suggesting a possible beneficial role of sleep also in intervention strategies with the aim of ameliorate cognitive status (Benveniste, 2018). The role of SWS in the clearance of Aβ *via* glymphatic system was tested by Xie and co-workers in an animal model: Aβ in rodent brain was cleared significantly faster in the cortex during SWS if compared with wakefulness. Although the presence of a glymphatic system in the human brain has not been proven, several correlation studies based on the investigation of the relationship between Aβ levels in the CSF and sleep/wake variables—together with SD deprivation evidence—hypothesized a plausible presence of a glymphatic system also in the human brain (Volkow et al., 2012; Xie et al., 2013). Furthermore, other evidence refers, specifically, to i) SWS disruption with a parallel increase in CSF Aβ levels (Ju et al., 2017) and ii) correlation of subjective measures regarding sleep duration and Aβ levels in the brain (Spira et al., 2013).

Again, the importance of sleep in Aβ clearance has also been confirmed in relation with a significant increase in interstitial fluid space volume at the cortex level in sleeping rodents than in wakefulness periods (Xie et al., 2013; Ding et al., 2016).

Animal and human models showed that Aβ levels in the interstitial fluid undergo diurnal oscillations (Musiek et al., 2015) that seem to be due to decreased neural activity in SWS during NREM sleep in some brain areas, which could be linked to a decrement in Aβ production. Considering the specificity of SWS in clearing Aβ, a series of new experimental protocols have been conducted to clarify the role of SWS on the light of the new findings mentioned above.

## THE RELATION BETWEEN SLOW-WAVE ACTIVITY (SWA), NREM OSCILLATION COMPONENTS, **β**-AMYLOID BURDEN, AND BRAIN STRUCTURAL AND FUNCTIONAL DIFFERENCES

By reviewing recent literature on sleep and AD relationship, it is evident that most experimental contributions have been conducted in healthy samples, in line with the concept of investigating the earliest preclinical pathological events related to AD and the possibility to target preventive and early-based sleep intervention.

Furthermore, the most recent innovative framework in the field of sleep and AD relationship derives from another new methodological perspective. The most important findings of the last decade show as the combination of different measurements represent the best overview to provide a complete vision of anatomical, electrophysiological, metabolic, and behavioral aspects underlying this relationship (**Supplementary Table 1**). In the light of SD animal and human models mentioned above, it emerges the need for examining specific brain regions—related to sleep—involved in AD pathogenesis. It becomes essential to investigate the complex interactions among the features related to AD with particular reference to the mechanistic pathways that could link specific Aβ accumulation in the brain, sleep state, and cognitive impairment (memory, at first).

Concerning the electrophysiological measurement of brain activity, the investigation of changes in electrical oscillations of sleep through EEG, with particular reference to NREM constituent oscillations—such as slow oscillations and sleep spindles—acquires great importance in this research line.

In particular, one quantification of slow waves entails the measurement of spectral EEG power in the 0.5- to 4.5-Hz range during SWS, also known as SWA. Studies conducted in healthy populations showed significant SWA power reductions with aging progression (Dijk et al., 1989; Landolt et al., 1996; Mander et al., 2013). In particular, the highest decreases in SWA are observed within the prefrontal cortex (PFC) during the first part of the night, in correspondence with the first NREM cycle (Landolt et al., 1996; Mander et al., 2013).

Slow wave is strictly related to gray matter within specific PFC regions: atrophy in these brain areas seems to predict the degree of changes in NREM slow-wave characteristics in elderly populations (Mander et al., 2013; Varga et al., 2016).

At this purpose, a relevant contribution of Mander and co-workers (2013) that combined MRI, fMRI, EEG, and memory measurements demonstrated that atrophy in the prefrontal gray matter predicted NREM SWA disruption, and the interaction between those measures predicted cognitive performance in overnight episodic memory retention. fMRI scans obtained during memory task showed that memory impairment was related to continuous hippocampal activation and reduced connectivity between the hippocampus and prefrontal cortex (Mander et al., 2013).

Considering the importance of the early Aβ accumulation in the brain, SWA has been investigated in relation to Aβ levels in CSF and memory consolidation. It is well known that Aβ's earliest occurrence appears at the cortical level and also includes the medial Prefrontal Cortex (mPFC) (Buckner et al., 2005). Indeed, subcortical structures are affected successively. For this reason, Mander et al. (2015) hypothesized involving the hippocampus since the preclinical stage of AD (clearly observed with early memory impairment) with regards the indirect pathways in which Aβ could impact the hippocampal–neocortical functioning. The hypothesized pathway is a sort of loop that includes NREM sleep disruption, SWA, and memory destruction (Mander et al., 2013).

The same research group conducted a successive study (Mander et al., 2015) on older adult population (*n* = 26), to clarify whether the extent of Aβ accumulation at mPFC level was related to decreased NREM SWA. Furthermore, the authors had the aim of proving whether NREM SWA correlated with the degree of memory impairment. Memory evaluation referred, specifically, to overnight hippocampus-dependent memory consolidation. Experimental protocol combined PET-PiB scanning to measure regional Aβ burden for one night of sleep EEG recording of behavioral and functional fMRI for sleepdependent memory consolidation. This elegant study is the first evidence to demonstrate the correlation between cortical Aβ and impaired generation of NREM SWS, also related to the prediction of successive decline in the hippocampal–neocortical memory transformation and related overnight memory retention. The results of this study extend prior studies that showed the association among the accumulation of Aβ within mPFC, the NREM SWA impairment, and the further correlation of disrupted NREM SWA with impaired hippocampal–neocortical memory transformation and related overnight memory retention. Taken together, these results provide important insights at anatomical and electrophysiological levels, as discussed by the authors (Mander et al., 2015). Anatomically, through the analyses of source localization, a correspondence of mPFC regions impacted early by Aβ burden and slow-wave generator has been observed in healthy young adults (Sepulcre et al., 2013; Mander et al., 2015). At the electrophysiological level, an important distinction within the delta frequency range (1–4 Hz) emerges: the relation

between Aβ and NREM SWA regards only the low-frequency range of SWA (0.6–1 Hz). So it suggests that only this specific frequency range is associated with Aβ pathology.

Considering microstructural features of NREM sleep, both amplitude and density of slow waves are significantly reduced across aging (Dubè et al., 2015), showing the largest changes over the frontal lobe and be maximal during the greatest expression of NREM oscillation, in the first one to two NREM cycles. These changes in the morphology of waveforms could be due to the diminishing of synchronized neuronal firing that causes sleep oscillations, provoked by a disruption in the slow wave depolarized or hyperpolarized down states that shape the slow wave (Beenhakker and Huguenard, 2009; Mander et al., 2017). Within the same frequency range of slow waves, AD patients compared with healthy controls show a 40% decrease of frontal spontaneous K complex during NREM sleep, and this decrease was positively related with Mini Mental State Examination (MMSE) scores (De Gennaro et al., 2017). The lack of a similar decrease in the amount of SWA suggests that the <1-Hz slow activity could be the potential sleep EEG marker, even though it does not differentiate AD from MCI patients (Reda et al., 2017).

Concerning sleep spindles, these NREM characteristic features reflect transient bursts of oscillatory activity in the 12- to 15-Hz range and are generated through an interaction between cortico-thalamic networks and reticular nucleus of the thalamus (De Gennaro and Ferrara, 2003). Rauchs and colleagues (2008), for the first time, reported a specific decrease in spindle in AD patients in association with learning abilities, underlying that these changes involved only fast spindles (13–15 Hz). Other studies demonstrated that the spectral power in this frequency range is decreased in middle-aged and older adults relative to young adults and is maximal over frontal EEG derivations (De Gennaro and Ferrara, 2003; Mander et al., 2014). A decreased rate of sleep spindles in older adults is also negatively associated with structural brain integrity: empirical evidence points to subcortical reductions in the gray matter within the hippocampus, and this decrement seems to predict the extent of decreased sleep spindle density at frontal lobe level in older adults (Fogel et al., 2017). In 2012, Wersterberg and colleagues investigated sleep physiology and memory in amnestic MCI patients (aMCI; *n* = 10) during stage 2 (i.e., the sleep stage where spindle activity is maximally expressed). The results showed that aMCI patients had fewer stage 2 spindles than aged-matched healthy adults (*n* = 18).

Furthermore, aMCI patients, if compared with controls, show less time spent in SWS and lower delta and theta power. Importantly, sleep deficiency in aMCI was also implicated in declarative memory consolidation: altered sleep patterns seem to contribute to memory impairments by contrasting with sleepdependent memory consolidation. In this study, reduced stage 2 spindles regarded fast (typically 13–15 Hz) but not slow spindles, in line with previous contributions that showed that fast spindles are most disrupted in AD (Rauchs et al., 2008), and these reductions were observed at frontal recording sites (Bliwise, 1993). A decrease of parietal fast spindle density in AD and MCI patients compared with healthy controls was confirmed by others (Gorgoni et al., 2016). This decrease positively correlated with MMSE scores.

The role of NREM activity, including the expression of sleep spindle EEG oscillations, has been further investigated by Mander and colleagues (2014). The authors started from the hypothesis that sleep could have a role in improving learning ability, in terms of better restoring next-day encoding learning abilities (e.g., Yoo et al., 2007). This role could be due, specifically, to the activity of fast-frequency sleep spindles over the left prefrontal cortex, as previously observed in healthy young adults (Mander et al., 2011). Experimental protocol combined fMRI and EEG recordings to i) evaluate the role of prefrontal fast sleep spindles in predicting next-day episodic encoding ability and ii) assess whether disruption in this EEG activity decreases the effectiveness of this positive benefit on hippocampal-dependent episodic learning ability.

The results showed that decreased prefrontal sleep spindle in older adults statistically mediated the effect of old age on next-day episodic learning and that the extent of this impairment in learning ability seems to depend on the degree of decreased prefrontal sleep spindles. Prefrontal spindle activity also seems to be related to hippocampal activation and could have an impact on learning ability in post-sleep phase. These results contribute to confirming the hypothesis that disrupted sleep physiology has a role in the agerelated cognitive decline in later life (Mander et al., 2014).

It is interesting to note that there is a strong resemblance concerning source generators of NREM SWS oscillations, which preponderate at mPFC level, and brain areas in which Aβ mainly accumulate in both cognitively healthy older adults and AD patients (e.g., Murphy et al., 2009).

NREM SWS disruption is exacerbated early in the course of AD, and this decrease predicts the severity of observed memory impairment (Wersterberg et al., 2012; Liguori et al., 2014). Finally, recent studies in human and rodents demonstrate that interstitial Aβ levels rise and fall with the brain states of wake and NREM sleep, respectively: shorter NREM duration and greater NREM fragmentation have been reported in mice over-expressing Aβ proteins (Kang et al., 2009; Roh et al., 2012), while human subjective reports of reduced sleep duration and diminished sleep quality correlate with cortical Aβ burden in healthy older adults (Spira et al., 2013). Moreover, direct manipulations of sleep and Aβ production in rodent models of AD have established bidirectional relationships between both factors (Kang et al., 2009; Roh et al., 2012; Xie et al., 2013).

The role of disrupted NREM sleep in contributing to the impairment of new hippocampal-based memory has also been evaluated through an interesting experimental method. Evidence shows that disturbing deep NREM sleep by applying acoustic stimulation during SWS coupled with spindle activity [when clicks are presented in synchrony with upcoming slow oscillation (SO) up states] had an impact on next-day cognitive performance (Ngo et al., 2013).

The experimentally induced increase in NREM SWA (in particular in the slow <1-Hz frequency range) seems causally enhancing subsequent consolidation and, thus, long-term memory retention in young adults (Marshall et al., 2006). In sleeping humans, Ngo and colleagues (2013) applied auditory stimulation in the phase with the ongoing slow oscillatory EEG activity, in order to enhance train of slow-wave oscillation (SO; <1 Hz) and assess next-day memory performance. Results show that declarative memory improved after stimulation, but only after in-phase stimulation with the ongoing slow oscillation rhythm (Ngo et al., 2013), underlying the possible role of SO in getting better memory performance.

These findings suggest that the effectiveness of acoustic stimulation in enhancing SO and improving memory performance could be mainly due to the timing in phase with slow oscillatory activity, as shown also in transcranial direct current stimulation: the experimental augmentation of SO intensity improved cognitive abilities, in terms of better performance in post-sleep hippocampal-dependent learning capacity (Antonenko et al., 2013).

To summarize, the role of SWS NREM components discussed above could inspire new experimental intervention approaches, based on non-invasive, low-cost, and effective preventive strategies.

## POSSIBLE INNOVATIVE TREATMENTS

The importance of developing sleep preventive and intervention strategies derives from the assumption that sleep could be considered as a modifiable and treatable risk factor for AD.

Some behavioral practices are well known and of concern, at first, due to the use of sleep hygiene in AD. In general, sleep hygiene guidelines suggest some preventive behaviors in terms of i) limiting the use of psychoactive substances (e.g., caffeine, alcohol, or smoking); ii) avoiding light exposure from television or computer in the evening; iii) practicing regular physical exercise; and iv) keeping constant bed and wake times with adequate light exposure upon waking (e.g., McCurry et al., 2012).

Some evidence suggests that physical and social activities could improve sleep quality, and the major benefits derived from multimodal treatments that combine sleep hygiene education, light physical exercise (walking), and bright light therapy have been observed (McCurry et al., 2005).

Concerning pharmacological treatment of sleep deficiency in AD, three drugs have been tested as alternatives to traditional hypnotics: melatonin, trazodone, and ramelteon (Mccleery et al., 2014), but the reported efficacy only concerned mild, moderate, and severe AD patients, without being proved in preclinical stages of AD.

In general, the effectiveness of melatonin is the most investigated. A recent review (Xu et al., 2015) evaluated the effectiveness of melatonin assumption for sleep disturbance and cognitive function in dementia patients observed in seven studies (*n* = 520). Results showed that the use of melatonin protracted total sleep time (TST) and, marginally, sleep efficiency, but cognitive function did not change significantly.

Going beyond the already known non-pharmacological strategies, in the light of the new findings on the role of SWS related to AD, it is possible to propose a possible intervention based on NREM SWS. In particular, Mander and colleagues (2016) suggest that NREM sleep enhancement in midlife to late life may lead to a preventive positive effect that could decrease AD risk, probably improving Aβ clearance or combating oxidative stress. Consequently, sleep restoration could also improve memory consolidation. Those authors suggest several promising tools for achieving NREM SWS enhancement benefit, with particular reference to <1-Hz NREM SWA. Currently, the experimental results associated with <1-Hz NREM SWA are contrasting: transcranial Direct Current Stimulation (tDCS) in the <1-Hz range seems to be effective in memory consolidation in young and older adults, in patients with schizophrenia, in individuals with attention-deficit hyperactivity disorders, and in patients with lobe epilepsy (Prehn-Kristensen et al., 2014; Del Felice et al., 2015). The results of other contributions show no benefits in memory consolidation in young and older adults (Eggert et al., 2013; Sahlem et al., 2015).

Enhancement of <1-Hz NREM SWA was also tested through auditory closed-loop stimulation during NREM SWS with promising positive outcomes on hippocampus-dependent memory consolidation (Ngo et al., 2013; Papalambros et al., 2017). Closed loop in phase auditory stimulation at low intensity might be considered a novel implement to improve restorative aspects of sleep rhythms, also in case of sleep disturbance (Riemann et al., 2011).

It is not possible to assume the real effectiveness of these methods, because, in our knowledge, there is no evidence of long-term benefits of this <1-Hz NREM SWA enhancement.

## CONCLUDING REMARKS

This review recapitulates recent contributions on the relationship between sleep and Aβ. Evidence from the last decade, deriving from sleep disturbances and sleep deprivation in relation with Aβ, raises important conceptual and methodological issues in the field of AD research. The concept of AD as a multifactorial

## REFERENCES


disease is growing up, and, consequently, the need for exploring different contributing factors in the pathogenesis and progression of the disease becomes fundamental in future perspectives.

The current findings on the relationship between sleep and Aβ have been providing important contributions i) to increase the understanding of the mechanisms underlined this relationship; ii) to understand the role of the different sleep stages in the pathogenesis and progression of AD; and iii) to target innovative, non-invasive intervention strategies based on sleep restoration.

Reviewing recent literature, the major contributions derive from studies that combine more measurements and foresee longitudinal experimental designs, extending their focus on many contributing factors, overlapping the mere amyloid cascade hypothesis. Currently, these studies are few, and further investigations are needed to confirm and extend the new promising finding in the field of sleep and AD.

## AUTHOR CONTRIBUTIONS

SC wrote the manuscript with the contribution of LA in the bibliography search and in phase of writing. PR and LG have contributed and supervised to all the writing phases of the review. All authors agree to the final submitted version.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2019.00695/ full#supplementary-material


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

*Copyright © 2019 Cordone, Annarumma, Rossini and De Gennaro. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Doxycycline for Alzheimer's Disease: Fighting **β**-Amyloid Oligomers and Neuroinflammation

*Claudia Balducci\* and Gianluigi Forloni*

*Istituto di Ricerche Farmacologiche Mario Negri, IRCCS, Milan, Italy*

Alzheimer's disease (AD) is the most widespread form of dementia, affecting about 45 million people worldwide. Although the β-amyloid peptide (Aβ) remains the most acknowledged culprit of AD, the multiple failures of Aβ-centric therapies call for alternative therapeutic approaches. Conceivably, the complexity of the AD neuropathological scenario cannot be solved with single-target therapies, so multiple-target approaches are needed. Core targets of AD to date are soluble oligomeric Aβ species and neuroinflammation, in an intimate detrimental dialogue. Aβ oligomers, the most neurotoxic species, appear to induce synaptic and cognitive dysfunction through the activation of glial cells. Anti-inflammatory drugs can prevent the action of Aβ oligomers. Neuroinflammation is a chronic event whose perpetuation leads to the continuous release of pro-inflammatory cytokines, promoting neuronal cell death and gross brain atrophy. Among the possible multi-target therapeutic alternatives, this review highlights the antibiotic tetracyclines, which besides antimicrobial activity also have pleiotropic action against amyloidosis, neuroinflammation, and oxidative stress. A particular focus will be on doxycycline (Doxy), a second-generation tetracycline that crosses the blood– brain barrier more easily and has a safer clinical profile. Doxy emerged as a promising preventive strategy in prion diseases and gave compelling pre-clinical results in mouse models of AD against Aβ oligomers and neuroinflammation. This strongly supports its therapeutic potential and calls for deciphering its exact mechanisms of action so as to maximize its effects in the clinic.

#### *Edited by:*

*Cesare Mancuso Catholic University of the Sacred Heart, Italy*

#### *Reviewed by:*

*Igor Klyubin Trinity College Dublin, Ireland Paolo Tucci University of Foggia, Italy*

*\*Correspondence: Claudia Balducci claudia.balducci@marionegri.it*

#### *Specialty section:*

*This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology*

*Received: 30 April 2019 Accepted: 07 June 2019 Published: 03 July 2019*

#### *Citation:*

*Balducci C and Forloni G (2019) Doxycycline for Alzheimer's Disease: Fighting β-Amyloid Oligomers and Neuroinflammation. Front. Pharmacol. 10:738. doi: 10.3389/fphar.2019.00738*

Keywords: Alzheimer's disease, beta-amyloid oligomers, neuroinflammation, tetracycline, memory

## INTRODUCTION

Alzheimer's disease (AD) is a subtle and so far incurable neurodegenerative disease that makes patients completely unable to run their daily life activities, remember their past and relatives (Selkoe, 2011). It affects about 45 million people worldwide with an enormous socio-economic burden, likely to increase further because of longer life expectancy, and aging as a major risk factor (Garre-Olmo, 2018).

The brains of AD patients present two main lesions: extracellular senile plaques and intracellular neurofibrillary tangles. Senile plaques, rich in aggregates of the β-amyloid peptide (Aβ), act both as a reservoir of the most neurotoxic Aβ soluble species, namely Aβ oligomers (AβOs), and a determinant of neuritic dystrophy and neuronal network interruption (Mucke and Selkoe, 2012). Neurofibrillary tangles are rich in hyperphosphorylated tau protein, which dissociates from microtubules, causing their destabilization. Mitochondria are also compromised (Desler et al., 2018). Neuroinflammation is a chronic neurotoxic event (Heneka et al., 2015), and the vascular system is damaged due to the accumulation of Aβ on the vessel wall (Daulatzai, 2017). On the functional level, synaptic activity is severely impaired and responsible for the onset of cognitive deficits (Mucke and Selkoe, 2012). Progressive neuronal loss culminates in gross brain atrophy (Pini et al., 2016).

The drama of AD lies in the fact that when cognitive deficits arise bringing patients to clinical attention, their brain is already severely compromised and the pathology has progressed for about 10–15 years. This makes the identification of an efficacious therapy a very hard challenge and suggests that the complexity of AD means we must abandon single-target therapies and move on to multi-level approaches. All the single-target "Aβ-centric" clinical trials so far have failed to produce significant benefits (Panza et al., 2019).

The present review will focus on two vital therapeutic AD targets, AβOs and neuroinflammation, that have recently attracted much attention among scientists fighting AD. We highlight the possibility of counteracting AβOs' detrimental activities and neuroinflammation with doxycycline (Doxy), a second generationtetracycline antibiotic, which has anti-AβOs, anti-inflammatory activities (Balducci et al., 2018), a good blood–brain barrier (BBB) penetration, and a safe pharmacological profile.

## A**β** OLIGOMERS: THE MOST DANGEROUS SYNAPTIC ENEMY

The amyloid cascade hypothesis was first put forward by Hardy and Higgins in 1992, stating that: "Deposition of the Aβ peptide, the main component of senile plaques, is the primary event in the pathogenesis of AD" (Hardy and Higgins, 1992). About 10 years later this theory underwent a significant revision, with Aβ plaques overtaken by smaller and soluble AβOs (Hardy and Selkoe, 2002).

AβOs are the first species originating from the amyloidogenic process, by which the Aβ peptide, with its remarked hydrophobicity and overload in AD brains, aggregates, leading to the formation of different-sized polymers including soluble oligomers, protofibrils, and insoluble fibrils.

AβOs are the small, soluble aggregates and the most potent toxic conformers of AD (Haass and Selkoe, 2007), as well as the best correlate of disease severity compared to plaques (Kuo et al., 1996; Lue et al., 1999; McLean et al., 1999). The number of Aβ plaques detectable many years before the onset of clinical symptoms (Perrin et al., 2009) does not correlate with the severity of the cognitive loss in patients (Lue et al., 1999; Naslund et al., 2000) and Aβ deposits are also found in cognitively healthy subjects.

Many experimental data have supported the important role of AβOs in synaptic dysfunction. From transgenic mouse models of AD, it emerged that the onset of synaptic and cognitive dysfunction preceded plaque deposition (Holcomb et al., 1998; Hsia et al., 1999; Mucke et al., 2000; Balducci et al., 2010b). Ultrastructural examination of AD mouse brains revealed the presence of AβOs in the synaptic compartment before plaque deposition (Balducci et al., 2010b). *In vitro* and *in vivo* data also indicated that the application of well-characterized synthetic AβO-enriched solutions, as well as oligomeric species extracted from patient brains or derived from AD mutated cell lines, abolished the formation of new dendritic spines, inhibiting synaptic plasticity and remodeling, thus impairing learning and memory when delivered in the brain of naive mice or rats (Cleary et al., 2005; Lesne et al., 2006; Poling et al., 2008; Shankar et al., 2008; Balducci et al., 2010a; Freir et al., 2011).

Also in humans, biochemical and morphological analyses indicate that AD represents, at least at the more initial stages of the pathology, an attack at the synapses. Indeed, the degree of cognitive decline has been correlated with a decrease of the presynaptic marker synaptophysin in the hippocampal area and associated cortices. Notably, a 25% reduction in the expression of synaptophysin was described in the cortex of MCI or very mild AD subjects compared to aged-matched healthy controls (Selkoe, 2002).

In a virtual scenario, AβOs must be visualized as undisturbed dynamic entities, either newly formed or traveling in and out of plaques, perturbing the CNS at many functional levels (Benilova et al., 2012; Forloni et al., 2016).

For many years, synapses have been considered the main AβO target. Initial work supported this, by describing the ability of AβOs to interfere with post-synaptic receptors such as N-methyl D-aspartate receptors (NMDARs) (Balducci et al., 2010b; Yamin, 2009), affecting calcium current, and α-amino-3-hydroxy-5 methyl-4-isoxazolepropionic acid receptors (AMPARs), with the ultimate outcome of inhibition of synaptic plasticity and the induction of memory impairment through the prevention/ abolition of new dendritic spine formation where new memories are stored (Chidambaram et al., 2019). Action in the pre-synaptic compartment was also described, through an interaction with the nicotinic acetylcholine receptors α7-nAcChR (Dineley et al., 2001; Puzzo et al., 2008). The cellular prion protein (PrPC) has been suggested to mediate AβO effects, although this remains controversial since we and others have confirmed direct binding between PrPC and AβOs, but not a functional contribution (Forloni and Balducci, 2011). Activation of the apoptotic machinery was also described as an AβO-mediated mechanism for synaptic loss (Jo et al., 2011).

However, compelling new theories have emerged in more recent years, bringing to light an intimate mutual interaction between AβOs and glial cells, responsible for synaptic perturbation and loss.

## NEUROINFLAMMATION: THE OTHER SPECIAL CULPRIT TO WATCH OUT FOR

Neuroinflammation has re-emerged as a driving force of neurodegeneration (McManus and Heneka, 2017). Microglia are important in brain tissue homeostasis, secreting neurotrophic factors, and patrolling the microenvironment through the release of cytokines and chemokines that influence astrocytes and neurons, particularly after infection or cell injury. This triggers inflammatory events normally calling a transient immune response followed by tissue repair. Under pathological conditions, resolution (Serhan et al., 2007; Serhan, 2010) can fail, promoting chronic neuroinflammation and neurodegeneration.

Microglia are found in a chronic activated state in AD around senile plaques, and through the continuous release of proinflammatory cytokines they drive neuropathology from the very early disease stages (Heneka et al., 2015). Neuroinflammation in the AD brain is chronic presumably because it is never resolved, as indicated by studies showing low levels of specialized proresolving mediators (Wang et al., 2015).

A relation between the detrimental action of AβOs and microglial cells is illustrated by the fact that microglia are also crucial in the control of synapse modelling and activity, and consequently of cognitive functions. Resting-state microglia survey neuronal activity by establishing intimate contacts with neurons. They are highly dynamic and plastic cells which continuously extend and retract their processes and contact synapses in an activity- and experience-dependent manner (Morris et al., 2013). However, under pathological conditions, when microglial cell activation is chronic as in AD, synaptic surveillance is lost and cognitive functions are perturbed.

Substantial data have corroborated the existence of a "dangerous liaison" between AβOs and microglial cells, by which they mutually sustain their detrimental effects on synapses. A series of studies comparing the effect of AβOs with that of Aβ fibrils demonstrated that AβOs foster microglial activation to a greater extent and apparently in a conformation-dependent manner (Heurtaux et al., 2010), with the lightest oligomeric species more likely to induce neuroinflammation. Michelucci et al. (2009) reported that AβOs are stronger inducers of the M1 microglial pro-inflammatory phenotype than fibrils. Then, He et al. (2012) described a more pronounced pro-inflammatory action of AβOs after chronic delivery inside the hippocampus of C57BL/6 mice, closely correlated with more severe cognitive deficit, altered neuronal organization, and ultrastructural changes. They also showed that AβOs increased the expression of toll-like receptor-4 (TLR4) and TNFα.

TLR4 belong to a well-known family of pattern recognition receptors initiating the innate immune response and are critically involved in AD (Drouin-Ouellet and Cicchetti, 2012). Using an AβO-induced acute mouse model, we also demonstrated that a single intracerebroventricular injection (ICV) of AβOs in C57BL/6 naive mice induced a transient memory impairment in the novel object recognition test (Balducci et al., 2010a), associated with transient activation of glial cells and an increase in the expression of pro-inflammatory cytokines in the hippocampus within a 2–24 h time window (Balducci et al., 2017), a crucial interval in the elaboration, and consolidation of long-term memory (Sutton and Schuman, 2006). Pre-treatment with anti-inflammatory drugs abolished the AβO-mediated memory impairment. While seeking further insight on the molecular mechanisms linking AβOs and microglial activation toward memory impairment, we also confirmed that TLR4 are vital, since neither memory impairment nor glial activation was observed in TLR4 null mice receiving AβOs ICV (Balducci et al., 2017).

Beside synaptic dysfunction and cognitive deficits, microglia activation also comes on stage to explain synapse loss. In a very elegant paper, Hong et al. (2016) demonstrated that microglia mediates abnormal synapse engulfment through C1q and C3 complement factors. C1q is the initiating protein of the classical complement pathway and, together with C3, localizes on synapses to mediate synaptic pruning through microglial phagocytosis (Presumey et al., 2017). In pathological conditions, such as AD, their expression is increased and localized on postsynaptic proteins, exacerbating synapse loss. The fact that this phenomenon was detectable at very early pre-plaque ages in AD-mutated mice suggested that most likely soluble Aβ species were involved. This was confirmed by specifically injecting AβOs ICV in wild-type mice and demonstrating an increase in C1q synaptic deposition, as well as the C1q-mediated C3 opsonization marking synapses for their elimination (Hong et al., 2016).

## TETRACYCLINES IN THE THERAPY OF AD: NOT ONLY ANTIBIOTICS

One of the most difficult challenges in AD is identifying an efficacious therapy to delay the onset, halt its progression, and prevent or reverse cognitive dysfunction. To date most attempts have focused on the Aβ peptide, with scarce or no beneficial effects (Panza et al., 2019). There might be several reasons to explain these multiple failures: wrong treatment timing, inappropriate treatment regimen, and poor or inadequate selection of patients. Although they are all valid possibilities, one of the main problems limiting therapeutic success may lie in the multi-factorial nature of AD, probably requiring multi-target therapies.

We therefore propose the antibiotic tetracyclines as a promising multi-target therapeutic approach, with a special focus on Doxy, a second-generation tetracycline with a safer pharmacological profile and a better passage across the BBB.

Interest in the tetracyclines in AD raised around the early 2000s when it was found, using cell-free approaches, that tetracyclines could inhibit the aggregation of both the synthetic PrP residues 106–126 and 82–146 of human PrP and the Aβ peptide. Facilitation of PrP and Aβ disaggregation as well as the sensitivity of their aggregates to proteases were also described (Tagliavini et al., 2000; Forloni et al., 2001). These anti-amyloidogenic effects were later confirmed for a series of other misfolding proteins responsible for neurodegenerative disorders, including Huntington's and Parkinson's disease (reviewed in Stoilova et al., 2013).

Neuroprotective activities of tetracyclines were first demonstrated against PrP *in vitro*, and *in vivo* again using the synthetic PrP residues 82–146 and 106–126 and through infection with the pathological form of the PrP, namely PrP scrapie (PrPsc). *In vitro*, tetracycline prevented the PrP 106–126-mediated neurotoxicity and astroglial proliferation (Tagliavini et al., 2000); *in vivo* pretreatment with either tetracycline or Doxy in experimental scrapie reduced infectivity, delayed the onset of pathology, and increased survival when intracerebrally injected in Syrian hamsters (De Luigi et al., 2008). Incubation of 263K PrPscinfected brain homogenate with 1 mM tetracycline or doxycycline resulted in more than 80% reduction in the PK-resistant core of PrPsc. It was also reported that these compounds can interact with partially purified PrPsc from patients with the new variant of Creutzfeldt–Jakob disease (CJD) (Forloni et al., 2002).

At the clinical level, Doxy gave positive results in the initial observational studies in CJD patients, which, however, were not confirmed in a double blind against placebo trial in subjects with a diagnosis of definitive or probable sporadic CJD or genetic forms of the disease (Haik et al., 2014). Later it was reported that an asymptomatic CJD patient given Doxy for 4 years survived longer (Assar et al., 2015; Pocchiari and Ladogana, 2015), and Varges and coworkers (2017) showed a longer survival in earlystage CJD patients, suggesting a preventive action of the drug.

In AD, Doxy has been tested in two clinical trials in mild to moderate patients, yielding both positive and negative results (Loeb et al., 2004; Molloy et al., 2013). In the first study there was less decline in cognitive abilities and functional behavior, whereas no benefits were obtained in the second one. The two trials were comparable in terms of patients' stage of disease at enrolment. Doxy was given orally at the dose of 200 mg/day together with rifampin 300 mg/day in the first study. The same doses were used in the second trial, with the sole difference that Doxy was given at the dose of 100 mg twice a day, rather than one. The main difference lies in the fact that in the former, patients were treated for 3 months, whereas in the latter one, treatment continued for 12 months. As stated by the authors, one possible explanation for the failure in the later study is that Doxy might have some negative properties that become evident when treatment is longer than 3 months. Unfortunately, the therapeutic effects were investigated only at the behavioral and functional levels, with no assessment of Aβ and tau levels in plasma and/or CSF, or of the inflammatory state.

Despite these controversies at the clinical level, preclinical studies indicated the therapeutic potential of Doxy. Initial investigation was done in two simple *in vivo* models. Diomede et al. (2010) tested Doxy in *Caenorhabditis elegans* (*C. elegans*), a simplified invertebrate model of AD where intracellular Aβ deposits caused *C. elegans* paralysis. Doxy protected against this damage by directly interacting with the Aβ aggregates and reducing the load of AβOs. Subsequently, Costa and colleagues (2011) demonstrated that Doxy treatment of Aβ42-expressing *Drosophilae melanogaster* did not improve their lifespan but slowed the progression of their locomotor deficits and partially rescued the toxicity of Aβ in the developing eye.

We recently found that Doxy had beneficial effects in acute and chronic mouse models of AD (Balducci et al., 2018). Chronic treatment with 10 mg/kg Doxy for 20 days or 2 months, injected intraperitoneally in APP/PS1dE9 transgenic mice, significantly restored memory independently of plaque reduction, but lowered the expression of the 18-mer oligomeric species. Interestingly, an acute treatment also led to memory recovery. On the basis of this evidence, and the lack of changes in plaque number, we assumed that Doxy was restoring memory by interfering with the oligomeric species. This was confirmed in the AβO-induced acute mouse model described above (Balducci et al., 2010a; Balducci and Forloni, 2014), which demonstrated that C57BL/6 naive mice treated ICV specifically with AβOs and pre-treated with Doxy were no longer impaired in their recognition memory. Moreover, because of the close relation between microglial activation and AβO detrimental cognitive effects (He et al., 2012; Balducci et al., 2017), we further show that the memory protection was associated with abolition of AβO-mediated microglial activation. The anti-inflammatory effect of Doxy together with memory recovery was also proved in the APP/PS1dE9 mice chronically treated with Doxy, and in LPS-treated mice, which present an AβO-independent inflammatory context (Balducci et al., 2018). The anti-inflammatory effect of Doxy has been demonstrated in a series of other pathological contexts (reviewed in Stoilova et al., 2013). **Figure 1** depicts all Doxy effects described above in our AD mouse models.

Although we did not find direct AβO-Doxy binding, we assume that—as described for AβOs and tetracycline—an atypical supramolecular interaction might occur, which will result in the formation of colloid structures sequestering and abolishing AβO toxicity *in vitro* (Airoldi et al., 2011). Accordingly, Costa et al. (2011), using transmission electron microscopy, dynamic light scattering, and thioflavin T binding, demonstrated that Doxy leads to the formation of smaller, non-amyloid and non-toxic Aβ aggregates. **Figure 2** summarizes the expected Doxy-mediated changes in the brain of AD patients.

FIGURE 1 | Doxy-mediated effects in AD mouse models. AβOs are the most powerful toxic species in AD brain, which are responsible for the memory impairment. Such detrimental effect is associated with microglial cell activation, a chronic event in AD responsible for both cognitive dysfunction, synaptic loss, and neurodegeneration. Doxy apparently interferes with either the action of AβOs by directly neutralizing their effects at both neuronal and glial level, and/or exerting a direct anti-inflammatory effect. All these actions culminate in a positive outcome at the cognitive level by restoring memory to normal.

Beside their anti-amyloidogenic and anti-inflammatory effects, tetracyclines also have anti-oxidative and anti-apoptotic activities (Stoilova et al., 2013; Santa-Cecília et al., 2019). Oxidative stress and apoptosis are typical features of AD, whose resolution in the intricate pathological scenario will help to better restore brain physiology.

This evidence and the favorable pharmacological features of Doxy in a translational prospect suggest that this drug holds a considerable therapeutic potential for AD and other neurodegenerative diseases. A recently published comprehensive review describes well the protective effect of Doxy also in Parkinson's disease and multiple sclerosis (Santa-Cecília et al., 2019).

Despite the beneficial effects of Doxy, clinical trials tell us that not all treatment protocols are effective, or stages of disease adequate for patient enrollment. Patients with too advanced

## REFERENCES


disease are apparently unlikely to respond (Assar et al., 2015; Pocchiari and Ladogana, 2015). This does not necessarily imply that the drug is ineffective, just that it must be used more appropriately. Because of this, Doxy deserves one more chance in AD therapy, more likely with application at a prodromal stage, and a "precision medicine" approach. The latter is strongly recommended, since it will enable us to define the clinical profile (i.e., inflammatory profile) of responders compared to non-responders.

## AUTHOR CONTRIBUTIONS

CB wrote the manuscript and created figures. GF revised the manuscript.


**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 © 2019 Balducci and Forloni. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Targeting Synaptic Plasticity in Experimental Models of Alzheimer's Disease

*Dalila Mango1†, Amira Saidi1†, Giusy Ylenia Cisale2, Marco Feligioni1,3, Massimo Corbo3 and Robert Nisticò1,4\**

*1 Laboratory of Neuropharmacology, EBRI Rita Levi-Montalcini Foundation, Rome, Italy, 2 Department of Physiology and Pharmacology, Sapienza University of Rome, Italy, 3 Department of Neurorehabilitation Sciences, Casa Cura Policlinico, Milan, Italy, 4 School of Pharmacy, Department of Biology, University of Rome Tor Vergata, Rome, Italy*

Long-term potentiation (LTP) and long-term depression (LTD) of hippocampal synaptic transmission represent the principal experimental models underlying learning and memory. Alterations of synaptic plasticity are observed in several neurodegenerative disorders, including Alzheimer's disease (AD). Indeed, synaptic dysfunction is an early event in AD, making it an attractive therapeutic target for pharmaceutical intervention. To date, intensive investigations have characterized hippocampal synaptic transmission, LTP, and LTD in *in vitro* and in murine models of AD. In this review, we describe the synaptic alterations across the main AD models generated so far. We then examine the clinical perspective of LTP/LTD studies and discuss the limitations of non-clinical models and how to improve their predictive validity in the drug discovery process.

#### *Edited by:*

*Cesare Mancuso, Catholic University of the Sacred Heart, Italy*

### *Reviewed by:*

*Filippo Caraci, University of Catania, Italy Ai Na Ng, University of Bristol, United Kingdom*

#### *\*Correspondence:*

*Robert Nisticò robert.nistico@uniroma2.it*

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

#### *Specialty section:*

*This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology*

> *Received: 09 May 2019 Accepted: 17 June 2019 Published: 16 July 2019*

#### *Citation:*

*Mango D, Saidi A, Cisale GY, Feligioni M, Corbo M and Nisticò R (2019) Targeting Synaptic Plasticity in Experimental Models of Alzheimer's Disease. Front. Pharmacol. 10:778. doi: 10.3389/fphar.2019.00778*

Keywords: long-term potentiation, long-term depression, synaptic plasticity, Alzheimer's disease, predictive validity

## INTRODUCTION

Long-term synaptic plasticity is considered the neural basis of learning and memory process (Bliss and Collingridge, 1993). Long-term potentiation (LTP) and long-term depression (LTD) are the major forms of durable synaptic strength changes in central nervous system abundantly studied in the hippocampal region (Malenka and Bear, 2004). The magnitude of LTP and LTD is largely used in many different experimental conditions and animal models as an indicator of cognitive function; on the other hand, dysregulation of synaptic plasticity underlies a large number of neurodegenerative disorders such as Alzheimer's disease (AD) (Selkoe, 2002).

AD is a multifaceted neurodegenerative disorder typified by a progressive and irreversible memory deficits and cognitive decline. To date, AD can only be diagnosed *post-mortem*, through two characteristic neuropathological lesions in the brain: senile plaques, consisting of β-amyloid protein oligomers aggregates (Aβo, residues 1–40/42), and intracellular neurofibrillary tangles (NFT), constituted of abnormally hyperphosphorylated tau protein accumulation predominantly

**Abbreviations:** Aβ, amyloid β protein; Aβo, β-amyloid protein oligomers aggregates; AD, Alzheimer's disease; AMPAR, L-αamino-3-hydroxy-5-methyl-4-isoxazole propionate receptor; APP, amyloid precursor protein; BST, basal synaptic transmission; E-LTP, early-LTP; FAD, familiar Alzheimer's disease; LFS, low frequency stimulation; LTD, long-term depression; LTP, long-term potentiation; L-LTP, late- LTP; mGlu, metabotropic glutamate receptor; NFT; neurofibrillary tangles; PPF, paired pulse facilitation; Tg, transgenic; WT PS1, wild-type human PS1; NA, not assessed; rTMS, repetitive transcranial magnetic stimulation; sAD, sporadic Alzheimer's disease; tDCS, transcranial direct current stimulation.

in hippocampal and cortical regions. The "amyloid cascade hypothesis" is so far the prominent theory to describe the timecourse of AD neurodegeneration (Hardy and Higgins, 1992). Impaired synaptic function of the hippocampus is an early event leading to defective hippocampal-dependent memory appearing long before the buildup of amyloid plaques and neuronal cell death (Selkoe, 2002; Tanzi, 2005). Therefore, synaptic plasticity is often used to evaluate part of the phenotype. Accordingly, many electrophysiological studies on the different models have been performed to delineate such changes.

Early impairments in synaptic transmission were highlighted in different mouse models of AD and are caused, among other factors, by Aβ which leads to impairment of LTP *via* tau protein (Shipton et al., 2011). Notably, many studies investigated the correlation between age and synaptic dysfunction in order to describe the onset and development of pathology in a specific mouse model. Discrepancy in the results obtained by the different researchers across the various models of AD may reflect the type of mutation studied, in addition to several other sources of variations such as experimental design, age, or strain.

## LTP AND LTD IN NORMAL CONDITIONS

Several studies indicate that the hippocampus plays a crucial role in higher cognitive functions and in information-storage (reviewed by Neves et al., 2008). LTP was first studied in the hippocampus and has been widely characterized using biochemical, electrophysiological, and molecular techniques (reviewed by Bliss et al., 2007). LTP is characterized by a shortterm phase (early- or E-LTP) and a subsequent long-term phase of potentiation (late- or L-LTP) (Reymann et al., 1989). Importantly, distinct forms of N-methyl-D-aspartate (NMDA) receptor LTP coexist at synapses (Park et al., 2014) and these can also be distinguished based on their responsiveness to protein kinase A (PKA) inhibitors (Park et al., 2016). E-LTP and L-LTP can be induced in hippocampal slices by different induction protocols and are sustained by distinct cellular and molecular pathways. E-LTP (<1 h) is characterized by the recruitment of postsynaptic 2-amino-3-(3-hydroxy-5-methyl-isoxazol-4-yl)propionic acid (AMPA) receptors, either from neighboring extra synaptic receptors or from intracellular reserve pool by exocytosis (Penn et al., 2017). On the other hand, L-LTP (>3 h) involves *de novo* protein synthesis promoting structural and functional changes (Frey et al., 1988). Among the key players facilitating the transition from E-LTP to L-LTP, brain-derived neurotrophic factor (BDNF) (Panja and Bramham, 2014) and transforming growth factor β1 (TGF-β1) (Caraci et al., 2015; Caraci et al., 2018) are noteworthy.

The classical form of LTD can be experimentally elicited using specific electrical low-frequency stimulation (LFS) protocols in slices (Dudek and Friedlander, 1996). Most of the LTD forms studied imply activation of NMDA receptors (Dudek and Friedlander, 1996) and/or metabotropic glutamate (mGlu) receptors (Fitzjohn et al., 1999). Other chemical forms of LTD are obtained either by exogenous application of NMDA or muscarinic receptor agonists (reviewed by Collingridge et al., 2010), as well as through activation of microglia (Zhang et al., 2014).

## LTP AND LTD IN EXPERIMENTAL AD

## *In Vitro* Models

The peptide amyloid beta is released endogenously during physiologic neuronal activity and causes enhancement of synaptic plasticity and memory formation when administered at picomolar concentrations that likely resemble the physiological level in the brain (Puzzo et al., 2008; Morley et al., 2010; Puzzo et al., 2011; Lawrence et al., 2014). Conversely, a prolonged exposure to the same amount is able to impair synaptic plasticity by glutamate-induced excitotoxicity (Koppensteiner et al., 2016). Specifically, it has been demonstrated that NR2B-containing NMDA receptors and mGlu5 receptors mediate the synaptotoxic effects of Aβo (Rammes et al., 2017).

The effects of acute application of exogenous Aβ oligomers (Aβo) obtained from synthetic, secreted from AD transgenic cells, or extracted from AD patients' brain, on synaptic transmission, have been widely studied in *ex vivo* hippocampal slices. All studies suggest that treatment of hippocampal slices with Aβo 200–500 nM induces alteration in LTP and LTD, generally manifested as loss of LTP and enhancement of LTD (Lambert et al., 1998; Wang et al., 2002; Shankar et al., 2008; Li et al., 2009; Jo et al., 2011; Cavallucci et al., 2013; Mango et al., 2016). Additionally, another study has shown that over-expression of Aβ in organotypic slices reduces the number of surface L-α-amino-3-hydroxy-5-methyl-4-isoxazole propionate receptor (AMPAR) similar to what occurs in mGlu receptor-dependent LTD (Kamenetz et al., 2003). Recently, we have demostrated that hippocampal mouse slices treated with Aβo display an enhancement of mGlu receptordependent LTD (Mango and Nisticò, 2018).

## AD Mouse Models

Animal models of AD should fully model human features of disease including gradual cognitive decline, synaptic dysregulation and spine loss, plaque load and NFT accumulation, inflammation, neurodegeneration, and atrophy of the central nervous sistem (CNS).

The breakthrough of amyloid precursor protein (APP) and PS human mutations has led to the generation of transgenic (Tg) animal models which strictly replicate the cardinal features of AD. AD models are largely used to explore in a spatiotemporal manner the pathogenic mechanisms of AD and the benefit of therapeutic approaches.

Several lines of transgenic models of AD have been generated so far; each recapitulates specific aspects of the disease. They exemplify more integrated approaches to examine the complex effects of Aβo on brain integrity and network function. Several mouse models have been so far analyzed for hippocampal synaptic function by means of electrophysiological techniques (see **Table 1**). To simplify, we divide them in APP-derived, PS1 derived, APP/PS1, 3xTg, and 5xTg models. Most of these models are constructed on the overexpression of familial AD (FAD) linked mutated genes into the mouse genome, and magnitude of LTP and LTD has been used to study synaptic plasticity alterations. Several exhaustive reviews on this topic have been published in the literature (Morrissette et al., 2009; Ashe and Zahs, 2010; Elder et al., 2010; Marchetti and Marie, 2011; Spires-Jones and Knafo, 2012; TABLE 1 | The table summarizes relevant data relative to experimental Alzheimer's disease (AD) models for which hippocampal electrophysiological analyses were performed.


*Models were grouped into in vitro models, APP-derived, PS1-derived, APP/PS1, 3xTg, and 5xTg models. Electrophysiological readouts include basal synaptic transmission (BST), paired pulse facilitation (PPF), long-term potentiation (LTP), and long-term depression (LTD). For each model, we also report the principal references to the electrophysiological studies.*

Peineau et al., 2018 or visit the Alzheimer forum at http://www. alzforum.org/res/com/tra/default.asp).

### APP-Derived Models

These mice models over-express the human APP, which is mutated in one or more sites. The single mutations introduced in the APP gene represent mutations associated with FAD, which are termed the Swedish (swe, K670N & M671L, Mullan et al., 1992), the Indiana (ind, V717F, Murrell et al., 1991; Hsia et al., 1999), the London (Ld, V171I, Goate et al., 1991), and the Arctic (E693G, Nilsberth et al., 2001) mutations. Other mouse models show a double mutation, such as the Swe mutation together with either the Indiana or Arctic mutation. These models manifest progressive Aβ accumulation and plaques similar to those found in humans. Aβ plaques observed in AD Tg mice brain appear structurally comparable to those discovered in the human brain; they start as diffuse plaques consisting mainly of Aβ 42 with a dense Aβ 42 core that contains Aβ 40 with many other non-Aβ components, among which are ubiquitin and synuclein (Yang et al., 2000). Moreover, these mice models show hyperphosphorylated tau and hippocampal-dependent memory deficits similar to human AD pathology, but do not show NFTs, cholinergic deficits, or neuronal death (Morrissette et al., 2009).

APP23 mice (Sturchler-Pierrat et al., 1997) display normal LTP in the hippocampus and prefrontal cortex at all ages tested (Roder et al., 2003). Tg2576 and the hAPPJ20 mouse models present an age-dependent reduction in LTP in the CA1 area (Nalbantoglu et al., 1997; Chapman et al., 1999; Fitzjohn et al., 2001; Balducci et al., 2011; D'Amelio et al., 2011) and in the dentate gyrus (Palop et al., 2007).

LTD has not been largely explored in APP-derived mouse models; only few studies have reported LTD measurement (D'Amelio et al., 2011; Cavallucci et al., 2013; Lanté et al., 2015). Both groups performed field recordings in hippocampal slices from Tg2576 mice and have shown enhanced NMDA receptordependent LTD starting already from 3 to 4 months of age.

In addition, a recent study described alteration of mGludependent LTD in 4-month-old Tg2576 mice, elicited by perfusion of the group I mGlu agonist DHPG (Mango and Nisticò, 2018).

## PS1-Derived Models

These models over-express the human presenilin gene (PS1) encoding an FAD mutation. Being part of the secretase complex, this gene is involved in the APP proteolysis. Presenilin variants do not produce neuropathology, but potentiate plaque deposition in APP transgenic mice. Electrophysiological studies have been performed on M146L, A246E, and L286V mutants (Parent et al., 1999; Barrow et al., 2000; Schneider et al., 2001; Auffret et al., 2009). A mouse model was also generated harboring the PS1ÆE9 FAD mutation, which results in a functional and non-cleavable modification of PS1 (Zaman et al., 2000). One knock-in mouse was engineered in which mouse PS1 was exchanged by its mutant M146V (Sun et al., 2005). These presenilin FAD mutants consistently exhibit an age-dependent increase of Aβ42 with minor effect on Aβ40; however, they do show amyloid plaques, tau pathology, cholinergic alterations, or neurodegeneration and present only a mild cognitive deficit (Games et al., 2006). Most studies report that young adult (up to 6 months) transgenic mice over-expressing PS1M146L, PS1M146V, PS1A246E, PS1ΔE9, or PS1L286V display enhanced CA1-LTP elicited using different conditioning protocols (Schneider et al., 2001; Oddo et al., 2003; Dewachter et al., 2008; Auffret et al., 2009). So far, no study described LTD in these mouse models.

### APP/PS1 Models

To accelerate the brain Aβ accumulation and plaque aggregation, researchers crossed APP- and PS1-derived animal models. Most electrophysiological studies focused on double transgenic models harboring the human APPswe transgene together with the PS transgene.

APP/PS1 double mutant mice develop rapid and extensive Aβ plaque accumulation, tau pathology, and cognitive defects, even though they lack cholinergic deficits, neuronal loss, and NFTs (Morrissette et al., 2009).

Most studies report a reduction of LTP in APPswe/PS1M146L (Gong et al., 2004; Trinchese et al., 2004, Trinchese et al., 2008), APPswe/PS1P246L (Chang et al., 2006), APPswe/PS1L166P (Calella et al., 2010), and APPswe/PS2N141I (Richards et al., 2003) mice. Using a standard LFS protocol, Chang and coauthors (2006) report a linear decline in CA1-LTD expression between 9 and 20 months of age in the APPswe/PS1P246L model. Moreover, loss of LTD was described in the APPswe/PS1M146V mouse model (Song et al., 2014). Also, a recent study has shown mGlu receptor-LTD impairment in the APPswe/PSEN1/ΔE9 (Yang et al., 2016), whereas no alteration in the basal transmission was found in this model (Volianskis et al., 2010).

## 3xTg Model

The 3xTg model over-expressing human APPswe and tau MAPTP301L and encoding a knock-in of PS1M146V was generated for the first time by Oddo et al. (2003). These mice display both Aβ aggregates and NFT and show hippocampal-dependent memory decline during aging. Also, they show cholinergic alterations and neuronal loss in the cortex (Oddo et al., 2003; Perez et al., 2010). This model is advantageous since it presents a significant intracellular Aβ deposition before the occurrence of extracellular plaques, which become evident around 12 months of age, and notably, it develops NFTs comparable to humans.

Concerning the functional aspects, 6-month-old 3xTg mice display impairment of LTP (Oddo et al., 2003), which correlates with intracellular Aβ well before plaque and tangle pathology. Recently, other triple transgenic mice have been generated harboring APP, PS2, and tau mutations (Rhein et al., 2009; Grueninger et al., 2010) but electrophysiological investigations have not been performed so far.

## 5xFAD Model

5xFAD mice (Tg6799 line) harbor three APP and PS2 (M146V and L286V) mutations that are causally related to FAD (Oakley et al., 2006). They exemplify one of the most early-onset mouse models with robust amyloid pathology (Oakley et al., 2006; Ohno et al., 2006, Ohno et al., 2007). Indeed, 5xFAD mice start developing amyloid deposition already from 2 months of age and show early hippocampal dysfunction, as evidenced by reduced basal synaptic transmission and LTP (Kimura and Ohno, 2009; Crouzin et al., 2013). This mouse model exhibits a strong pathology: at 1.5 months of age mice already express intracellular Aβ42, which massively progresses at 2 months of age with extracellular Aβ accumulation, senile plaques, and lack of specific neuronal populations. Cognitive impairment is reported at 4–6 months of age (Oakley et al., 2006). LTD has not yet been investigated in this mouse model.

## CLINICAL RELEVANCE OF LTP/LTD

Many of the mechanisms underlying LTP and LTD in the rodent hippocampal slice preparation are shared also in hippocampal tissue from patients undergoing surgery for intractable temporal lobe epilepsy. Indeed, LTP is induced in the temporal lobe and in the dentate gyrus in humans using similar protocols of stimulation (Chen et al., 1996; Beck et al., 2000) and is modulated by the different pharmacological approaches just as in nonclinical models. These studies further support the notion that also the human brain manifests LTP- and LTD-like events even though linking synaptic plasticity to human learning and memory remains a challenge (Bliss and Cooke, 2011). It also should be kept in mind that the human tissue investigated in electrophysiological studies is found in a pathological state, deriving from patients with an epileptic focus in the temporal lobe.

Notably, LTP- and LTD-like events are nowadays exploited in humans for therapeutic purposes. These plastic changes can be induced through several noninvasive techniques such as repetitive transcranial magnetic stimulation (rTMS) and transcranial direct current stimulation (tDCS) and are used for the treatment of a variety of neurological and psychiatric conditions, such as epilepsy, drug addiction, depression, Parkinson disease, neuropathic pain, tinnitus, and stroke (Schulz et al., 2013).

We can therefore hypothesize that stimulation of neuroplasticity in the early stages of AD, through pharmacological and noninvasive approaches, can attenuate disease progression. Even though numerous therapeutic interventions reverse synaptic alterations and improve behavior in the non-clinical AD models, so far, there has been no successful translation into disease-modifying compounds in humans (Nisticò et al., 2012).

## PREDICTIVE VALUE OF AD MODELS

Considering the recent clinical trial failures in AD, there has been considerable discussion as to whether results obtained from non-clinical models are predictive or simply misleading. There are numerous reasons why non-clinical studies may have failed to predict clinical trial outcome. One of the main issues it that Tg mice carry FAD mutations, accounting for only 5–10% of all AD cases, while the vast majority of AD cases are sporadic (sAD). As a consequence, these models have a low face and predictive validity for the sporadic form of AD. Moreover, transgenic models normally overexpress APP with consequent overproduction several APP fragments. Of note, a knock-in mice with a phenotype more similar to human was recently generated (Saito et al., 2014). Another limitation is that each mouse model develops only specific characteristics of AD (i.e. Aβ *vs*. tau pahology) and does not recapitulate the complexity of the human disease. It has also to be taken into account whether animal models display a similar spatiotemporal profile of disease progression when compared to AD patients. For example, cognitive deficits usually precede plaque load in mice, whereas the opposite occurs in patients. An important issue to be considered is that the majority of AD models lack neuronal cell death, while a substantial neurodegeneration is observed in the human AD brain.

It can be argued that the various non-clinical models typify specific disease-related targets and pathways, a potential advantage for testing candidate molecules on selected targets involved in AD pathogenesis. Indeed, this target-driven approach in non-clinical models has been translated over the years into numerous clinical studies (Nisticò et al., 2012).

In addition to intrinsic limitations of animal models, experimental bias is another crucial factor. For example, gender- and litter-dependent differences, variability in transgene expression, and the different genetic background among models and even between active treatment and placebo groups should all be considered in translation. Also, diversities in brain anatomy, neuronal physiology, metabolism, and disease susceptibility play a central role. Moreover, given the complex dynamics of drugtarget interactions, *in vivo* studies in non-clinical models should include a complete pharmacokinetics/pharmacodynamics profile in order to ensure that the dose range and timing are specific to the target (e.g. based on receptor occupancy and/or correlative biomarker data). On the other hand, the therapeutic window and the off-target effects that may lead to adverse effects should also be identified especially for behavioral experiments.

To improve clinical translation, non-clinical endpoints need to be accurately selected and should employ a combination of disease-relevant approaches such as integrated neurophysiological/behavioral paradigms. Electrophysiological techniques record neural activity from large neural populations down to single cells and are ideal to measure synaptic plasticity, as well as firing activity and neural oscillations. A limitation is that this approach does not address spatiotemporal information and is not suitable for noninvasively detecting activity from deep brain regions. In contrast, functional neuroimaging provides a noninvasive spatiotemporal readout of changes in brain function, making it an invaluable tool for most clinical studies. Unfortunately, the use functional imaging has been limited in non-clinical setting due to the restricted applicability to animal models and the relatively high cost.

## CONCLUSIONS

We have previously hypothesized that alterations in normal synaptic function are not only a key feature but also a leading cause of disease (Nisticò and Collingridge, 2012). In this respect, LTP and LTD can serve as synapse survival and death signals, respectively. Thus, conditions that promote LTD, i.e. following excessive Aβ load in the early-onset forms of disease, can lead to loss of synapses. On the other hand, promoting LTP, which is known to inhibit LTD (Peineau et al., 2007), can represent a protective mechanism to preserve synaptic plasticity and brain connectivity.

It seems important to investigate the molecular mechanisms that influence plasticity in the human brain and to determine whether its vulnerability to aging and neurodegeneration can be modified by pharmacological intervention. Considering that AD is a complex disease affecting multiple signaling pathways, therapeutic strategies should not be directed to a single target rather to a combination of targets. To ensure a successful outcome, therapy should start at an early stage of disease. In addition, highly sensitive and specific biomarkers should identify susceptible individuals at the onset of disease (Hampel et al., 2014).

Generally, the predictive value of non-clinical models in the drug discovery process has been largely debated independently of the therapeutic area (McGonigle and Ruggeri, 2014; Mullane and Williams, 2019). Accordingly, several compounds showing robust efficacy in experimental models of AD have failed so far in clinical trials. Once a lead compound is selected, selection of nonclinical endpoints through integrated approaches should reflect the clinical endpoints in phase I studies. Correct design of nonclinical studies can be a long, complex, and expensive process that may slow down the course of drug development (Mohs and Greig, 2017); nonetheless, the probability of successful approval and hence time saving and return on investment is certainly increased.

## AUTHOR CONTRIBUTIONS

DM, AS and GC prepared the manuscript and reviewed the drafts, MF and MC reviewed the drafts, RN conceived the idea, prepared the manuscript and reviewed the drafts. All authors contributed to the writing and final approval of the manuscript.

## FUNDING

This work was supported by Fondazione Turano.

## REFERENCES


fear conditioning by a N-terminal beta amyloid fragment. *J. Neurosci.* 34, 14210–14218. doi: 10.1523/JNEUROSCI.0326-14.2014


protein 23 transgenic mice. *Neuroscience* 120, 705–720. doi: 10.1016/S0306- 4522(03)00381-6


**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 © 2019 Mango, Saidi, Cisale, Feligioni, Corbo and Nisticò. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# MicroRNA-200a-3p Mediates Neuroprotection in Alzheimer-Related Deficits and Attenuates Amyloid-Beta Overproduction and Tau Hyperphosphorylation *via* Coregulating BACE1 and PRKACB

#### *Edited by:*

*Silvana Gaetani, Sapienza University of Rome, Italy*

#### *Reviewed by:*

*Ana Luisa Carvalho, University of Coimbra, Portugal Yun Ling, Fudan University, China*

#### *\*Correspondence:*

*Zhuorong Li lizhuorong@imb.pumc.edu.cn Rui Liu liurui@imb.pumc.edu.cn*

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

#### *Specialty section:*

*This article was submitted to Neuropharmacology, a section of the journal Frontiers in Pharmacology*

*Received: 10 December 2018 Accepted: 21 June 2019 Published: 19 July 2019*

#### *Citation:*

*Wang L, Liu J, Wang Q, Jiang H, Zeng L, Li Z and Liu R (2019) MicroRNA-200a-3p Mediates Neuroprotection in Alzheimer-Related Deficits and Attenuates Amyloid-Beta Overproduction and Tau Hyperphosphorylation via Coregulating BACE1 and PRKACB. Front. Pharmacol. 10:806. doi: 10.3389/fphar.2019.00806*

*Linlin Wang1,2†, Jianghong Liu3†, Qian Wang1, Hailun Jiang1, Li Zeng1,4, Zhuorong Li1\* and Rui Liu1\**

*1 Institute of Medicinal Biotechnology, Chinese Academy of Medical Science and Peking Union Medical College, Beijing, China, 2 Institute of Materia Medica, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China, 3 Department of Neurology, Xuan Wu Hospital, Capital Medical University, Beijing, China, 4 Department of Biopharmaceutics, Shenyang Pharmaceutical University, Shenyang, China*

Alzheimer's disease (AD) is characterized by two landmark pathologies, the overproduction of amyloid-beta peptides (Aβ), predominated by the β-amyloid protein precursor cleaving enzyme 1 (BACE1), and hyperphosphorylation of the microtubule protein, tau, because of an imbalance in a kinase/phosphatase system that involves the activation of the protein kinase A (PKA). Current evidence indicates that brain microRNAs participate in multiple aspects of AD pathology. Here, the role and underlying molecular mechanisms of microRNA-200a-3p (miR-200a-3p) in mediating neuroprotection against AD-related deficits were investigated. The expression of miR-200a-3p was measured in the hippocampus of APP/PS1 and SAMP8 mice and in an AD cell model *in vitro*, as well as in blood plasma extracted from AD patients. The targets of miR-200a-3p were determined using bioinformatics and dual-luciferase assay analyses. In addition, cell apoptosis was detected using flow cytometry, and related protein levels were measured using Western blot and enzyme-linked immunosorbent assay (ELISA) techniques. miR-200a-3p was confirmed to be depressed in microarray miRNA profile analysis *in vitro* and *in vivo*, suggesting that miR-200a-3p is a potential biomarker of AD. Subsequently, miR-200a-3p was demonstrated to inhibit cell apoptosis accompanied by the inactivation of the Bax/ caspase-3 axis and downregulation of Aβ1-42 and tau phosphorylation levels *in vitro*. Further mechanistic studies revealed that miR-200a-3p reduced the production of Aβ1-42 and decreased hyperphosphorylation of tau by regulating the protein translocation of *BACE1* and the protein kinase cAMP-activated catalytic subunit beta (*PRKACB*) associated with the three prime untranslated regions, respectively. Importantly, the function of miR-200a-3p was reversed by overexpression of BACE1 or PRKACB in cultured cells. This resulted in an elevation in cell apoptosis and increases in Aβ1-42 and tau hyperphosphorylation levels, involving the epitopes threonine 205 and serine 202, 214, 396, and 356, the favorable phosphorylated sites of PKA. In conclusion, our study suggests that miR-200a-3p is implicated in the pathology of AD, exerting neuroprotective effects against Aβ-induced toxicity by two possible mechanisms: one involving the inhibition of Aβ overproduction *via* suppression of the expression of BACE1 and synergistically decreasing the hyperphosphorylation of tau *via* attenuation of the expression of PKA.

Keywords: Alzheimer's disease, miR-200a-3p, BACE1, PRKACB, tau protein, apoptosis

## INTRODUCTION

Alzheimer's disease (AD) is the most common form of dementia affecting people aged 65 and older, which is characterized by two landmark pathologies, extracellular senile plaques consisting of amyloid-beta peptides (Aβ) and intracellular neurofibrillary tangles (NFTs) composed of hyperphosphorylated tau proteins (Ulep et al., 2018). AD has a long preclinical latency and is difficult to diagnose and prevent at early stages. Currently, there are no effective drugs or treatment modalities to stop or reverse the progression of the disease process. Thus, it is of vital importance to explore the underlying mechanisms and potential targets of AD.

Among the complex etiology of AD, Aβ peptides are derived from the successive cleavage of the amyloid protein precursor (APP) by the β-APP cleaving enzyme 1 (BACE1), or β-secretase, and gamma (γ)-secretase enzyme to form Aβ aggregates that have the potential to develop into Aβ plaques (Li et al., 2000). Similarly, the abnormally hyperphosphorylated tau protein in NFTs is thought to be generated from an imbalance in the kinase/phosphatase system, indicated by a series of activity-altered enzymes involving cyclic AMP (cAMP)-dependent protein kinase/protein kinase A (PKA), cyclin-dependent protein kinase 5 (CDK5), glycogen synthase kinase 3β (GSK3β), protein phosphatase 1 (PP1), and protein phosphatase 2A (PP2A), which are key players in the progression of AD (Wang et al., 2014). Furthermore, other pathological factors associated with AD development include oxidative imbalance, neuroinflammation, and calcium homeostasis disturbance, which lead to an overproduction of Aβ peptides by activating BACE1 *via* a set of tau-associated phosphorylated kinases (Chami and Checler, 2012). Considering key AD pathogenic mechanisms, simultaneous interference with two or more causes associated with Aβ-induced tau hyperphosphorylation may achieve better therapeutic efficacy with multiple benefits by combining collaborative mechanisms.

MicroRNAs (miRNAs) are short, single-stranded RNAs of about 20–25 base pairs (bp) in length, which regulate posttranscriptional expression of target messenger RNA (mRNA) by associating with the mRNA three prime-untranslated region (3'-UTR) (Geekiyanage et al., 2012). In addition, miRNA expression has been proven to have tissue, cell, and disease specificity (Ratnadiwakara et al., 2018). Abundance of miRNAs has been illustrated specifically in gene expression changes related to AD and can be also found in the cerebrospinal fluid (CSF) and blood plasma (Fransquet and Ryan, 2018; Zendjabil, 2018). Thus, miRNAs are excellent candidates as noninvasive biomarkers and potential regulators of associated target genes in AD.

MicroRNA-200a-3p (miR-200a-3p), belonging to the miR-200 family of miRNAs and located on chromosome 1p36, plays an important role in human cancers and modulates cell apoptosis and proliferation (Feng et al., 2014; Wu et al., 2017). Accumulating evidence is available illustrating that miR-200a-3p may be involved in AD pathology; however, there is evident controversy about miR-200a-3p levels detected in different Alzheimer's models. Some studies elucidated that miR-200a-3p was downregulated in K595N/M596L (APPswe)/presenilin 1 (PS1) deltaE9 (APP/PS1) mice during the progression of AD (Liu et al., 2014), whereas some other experiments showed that in the brain of AD patients and in the hippocampus of APP/PS1 mice, the expression of miR-200a-3p was increased (Lau et al., 2013; Zhang et al., 2017). In line with the finding that miR-200a-3p was downregulated in AD, miR-200a-3p was shown to inhibit apoptosis in the SH-SY5Y neuroblastoma cell line *via* the modulation of one of the targets of sirtuin-1 (Salimian et al., 2018). Nevertheless, miR-200a-3p appears to have multiple downstream targets, characterized by the aberrant expression involved in the gene regulation networks among diseases; therefore, the specific roles and underlying molecular mechanisms of miR-200a-3p in AD remains unexplored.

In this study, we investigated the expression of miR-200a-3p in the hippocampus of APP/PS1 and senescence-accelerated mouse prone 8 (SAMP8) mice and in an AD cell model *in vitro*, as well as in blood plasma extracted from AD patients. We further explored the roles of miR-200a-3p and its potential molecular mechanisms in the AD cell model. Collectively, this study revealed that miR-200a-3p supplementation might play a neuroprotective role in AD, which highlights potential future research avenues and novel therapeutic targets for AD.

## MATERIALS AND METHODS

## Animals and Treatments

APP/PS1 mice and age-matched wild-type (WT) littermates were purchased from the Jackson Laboratory (Bar Harbor, ME). SAMP8 and senescence-accelerated mouse resistance 1 (SAMR1) mice were provided by the Institute of Genetics and Developmental Biology of the Chinese Academy of Sciences. The animals had *ad libitum* access to food and water at stable room temperature and humidity environment according to the Guide for the Care and Use of Laboratory Animals. The experiment was approved by the ethical committee of the Institute of Medicinal Biotechnology (IMB-D8-2018071102).

The mice were then divided into the following groups: 1-monthold, 3-month-old, 6-month-old, or 9-month-old mice, with the inclusion of age-matched control mice (WT or SAMR1). Each group was composed of four mice (two males and two females per group). Depending on the different ages of each group, brains were collected and then evaluated by real-time polymerase chain reaction (PCR) analysis during the course of the disease.

## Human Blood Sample Data and Collection

Blood samples of seven AD patients and five normal agematched volunteers (NAVs) were acquired from the Xuanwu Hospital Capital Medical University, and the study was approved by the ethics committee of Xuanwu Hospital Capital Medical University, China (**Table 1**). The peripheral blood was collected from each patient after fasting for 12 h. The serum was separated by centrifugation at 1,000 × g for 10 min at room temperature, followed by centrifugation at 130,000 × g for 5 min at 4°C. The samples were stored at 80°C until required.

## Cell Culture and Plasmid Transfection

Human neuroblastoma SH-SY5Y cells (ATCC; Manassas, VA) and human embryonic kidney (HEK)293 cells (ATCC) were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (FBS) (Gibco/Invitrogen, Grand Island, NY) at 37°C in a humidified 5% CO2 incubator. SH-SY5Y cells transfected with the Swedish mutant form of human APP (referred to as "APPswe cells") is an established AD cell model, in which copper can trigger the neurotoxicity of Aβ, leading to cell apoptosis (Liu et al., 2012; Zhao et al., 2013). APPswe cells were then cultured in DMEM/F-12 supplemented with 10% FBS and 500 μg/ml G418 (Invitrogen). miR-200a-3p mimics, miR-200a-3p inhibitor, and respective negative controls (NCM or NCI), as well as BACE1-siRNA, PRKACB-siRNA, and respective NCs were synthesized by GenePharma (Shanghai, China). The oligonucleotides were transfected in a final concentration of 50 nΜ using the Lipofectamine 3000 reagent (Invitrogen; Carlsbad, CA). The BACE1 and PRKACB expression vectors were constructed by inserting human BACE1 and PRKACB cDNA into pCMV6 vector, and their promoters were CMV and tagged with Myc-DDK. The pCMV6-BACE1-Myc-DDK and pCMV6-PRKACB-Myc-DDK were purchased from ORIGENE (Beijing, China) and transfected into APPswe cells in a final concentration of 2 μg/ml using the Lipofectamine 3000 reagent.

## Quantitative Reverse Transcription Polymerase Chain Reaction Analysis

The total RNA of neuronal cells and mouse brain tissue were extracted using TRIZOL (Invitrogen) according to the manufacturer's instructions. The TaqMan microRNA Reverse Transcription reagent (Invitrogen) was used to reverse transcript 10 ng of total RNA to complementary DNA (cDNA). The following

TABLE 1 | Clinical data of AD patients compared to normal age-matched volunteers (NAVs).


*NAVs, normal age-matched volunteers; AD, AD patients; M, males; F, females.*

reactions were performed in a total volume of 20 μl containing the following: 1 μl TaqMan small RNA assay, 1.3 μl cDNA sample, 10 μl TaqMan universal PCR Master Mix, and 10 μl nuclease-free water. Each sample was run in triplicate. Small nuclear RNA U6 was used as normalization. The thermo cycle conditions were set as follows: enzyme activation at 50°C for 2 min, denaturation at 95°C for 10 min, followed by 40 cycles of denaturation at 95°C for 15 s and extension at 60°C for 1 min. For mRNA analysis, 2 μg of total RNA was mixed with Maxima H Minus cDNA Synthesis Master Mix (Invitrogen) for reverse transcription. Subsequently, a qPCR assay was used using the SYBR Green Master Mix (Invitrogen) on an ABI-7500 Fast Real-Time PCR System (Applied Biosystems, Foster City, CA). The primers used are listed in **Table 2**. The data were analyzed using the 2-ΔΔCT method.

## Dual-Luciferase Reporter Assay

The 3'-UTR of *BACE1* and *PRKACB* containing the binding site of miR-200a-3p were cloned into the luciferase reporter plasmid (Promega; Madison, WI), and the binding site mutants were synthesized by a commercial company (Sangon Biotechnology, Shanghai, China). The WT or mutant luciferase plasmid together with the pRT-TK Renilla luciferase vector (Promega) were cotransfected with miR-200a-3p mimics or NCs into HEK293 cells. After 48 h, the corresponding vector's luminescence was detected using the GloMax Multi luminometer (Promega) with a Dual-Luciferase Reporter Assay system. The Renilla luminescence was used to normalize the signal. All of the experiments were repeated four times independently.

## Western Blotting Analysis

Protein samples were extracted from differently treated cells using the M-PER Mammalian protein Extraction Reagent (Pierce Biotechnology; Rockford, IL) at appropriate time points, according to the manufacturer's instructions. Then, sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) was used to separate target proteins and were subsequently transferred to a polyvinylidene fluoride (PVDF) membrane (Merck Millipore, Billerica, MA). The membranes were then blocked in 5% nonfat milk dissolved in tris-buffered saline (TBS) containing 0.1% Tween-20 (TBST) for 1 h and then probed with the corresponding antibody as listed in **Table 3** overnight at 4°C. The next day, membranes were washed three times with TBST and incubated with the appropriate secondary antibodies (Abcam, Cambridge, MA) for 1 h at room temperature. The protein bands were visualized and quantified by a protein imaging system (Biorad, Munich, Germany).

#### TABLE 2 | PCR primer sequences.


*F, forward; R, reverse.*

#### TABLE 3 | Primary antibodies used in this study.


## A**β**1-42 Assay

APPswe cells were seeded in 24-well plates and transfected with miR-200a-3p, BACE1, or PRKACB related plasmids. After 48 h, cells were lysed and detected using the Human Aβ1-42 enzyme-linked immunosorbent assay (ELISA)Kit (Invitrogen) according to the manufacturer's instructions. The concentration of Aβ1-42 was determined by the absorbance value detected using a microplate reader (Tecan Group Ltd., Mannedorf, Switzerland) at 450 nm.

## Cell Apoptosis Assay by Flow Cytometry

APPswe cells were transfected with miR-200a-3p mimics, miR-200a-3p inhibitor, pCMV6-BACE1-Myc-DDK, pCMV6- PRKACB-Myc-DDK, BACE1 siRNA, PRKACB siRNA, and their relative controls separately or in combination. After transfection, copper was added to trigger the Aβ toxicity. The Annexin BD Pharmingen FITC-Annexin V/propidium iodide (PI) Apoptosis Detection Kit (BD Biosciences, San Jose, CA) was used to detect cell apoptosis of APPswe cells after different plasmid transfections, according to the manufacturer's protocol. Briefly, cells were washed in cold phosphate buffer saline (PBS) and stained with a mixture of FITC-labeled Annexin-V and PI on ice for 15 min and analyzed in a FACSCalibur flow cytometry (BD Biosciences).

## Caspase-3 Activity Assay

APPswe cells were transfected with miR-200a-3p mimics, miR-200a-3p inhibitor, and the relative controls, as well as pCMV6- BACE1-Myc-DDK and pCMV6-PRKACB-Myc-DDK separately or in combination. After different treatments, APPswe cells were used to detect the activity of caspase-3 using the Human Active Caspase-3 (Asp175) SimpleStep ELISA Kit (Abcam, Cambridge, MA) according to the manufacturer's instructions.

## Bioinformatics and Heatmap Analysis

The miR-200a-3p target predicted by computer-aided algorithms was obtained from TargetScan. The miRNA expression profiling was performed as described before (Wang et al., 2017). RNA extraction was performed with TRIzol regent and synthesized into cDNA. After purification, the cDNA was labeled with Hy3/Hy5 and hybridized to the microarray (Exiqon, Vedbæk, Denmark) according to the Exiqon's instruction. The slides were scanned by Axon GenePix 4000B microarray scanner (Exiqon), and heatmap analysis of the expression of the selected miRNAs at different stages of AD development was accomplished using the color gradation function of Microsoft Excel 2016 (Microsoft Corporation, Redmond, WA).

## Statistical Analysis

Data are represented as mean ± standard error of the mean (SEM). All of the experiments were repeated at last three times. Data were analyzed using Student's t-test or one-way ANOVA followed by Tukey's *post hoc* tests where appropriate. Comparisons between two groups were performed using Student's t-test. *P* values of less than 0.05 were considered statistically significant. All of the analyses were performed using the GraphPad Prism Version 7.0 (GraphPad Prism Software, La Jolla, CA).

## RESULTS

## The Expression of miR-200a-3p is Decreased During AD Progression

It has been described that the APP/PS1 double transgenic mouse model develops AD pathology and cognitive impairment with increasing age (Trinchese et al., 2004). Before exploring the role of miR-200a-3p in the pathological processes of AD, we determined the manifestation of miR-200a-3p in miRNA profiles of APP/PS1 transgenic mice using microarray analysis. As shown in **Figure 1A**, the levels of miR-200a-3p were significantly downregulated in 6-month-old and 9-month-old mice. Moreover, to assess the changes of miR-200a-3p expression during AD progression, SH-SY5Y cells overexpressing the APPswe plasmid and two mouse models, the APP/PS1 transgenic and the mutant SAMP8 strains, were used because they have been previously shown to express deficits closely similar to AD (Ding et al., 2008; Liu et al., 2012; Takagane et al., 2015). Our results demonstrated that the expression of miR-200a-3p was significantly reduced in APPswe cells when compared to normally cultured SH-SY5Y cells (**Figure 1B**, *P* < 0.01). This was in line with the depressed levels of miR-200a-3p found in the hippocampus of APP/PS1 and SAMP8 mice at 3, 6, or 9 months of age when compared to their corresponding WT control counterparts (**Figures 1C**, **D**, *P* < 0.05–0.01). Because miRNAs can circulate in the blood and the CSF, they are attractive candidates as biomarkers of AD. To this end, we explored the expression of miR-200a-3p in the plasma of AD patients and NAVs and found that miR-200a-3p levels were significantly downregulated in the blood plasma of AD patients when compared to those in NAVs (**Figure 1E**, *P* < 0.05). Therefore, these data confirmed a reduced tendency of miR-200a-3p expression levels in the pathological processes of AD and also indicated that miR-200a-3p might participate in the regulation of this process.

## miR-200a-3p Inhibits A**β**1-42 Production, Attenuates Tau Phosphorylation, and Suppresses Apoptosis in APPswe Cells

Aβ1-42 peptides, derived *via* processing of APP by BACE1, contribute to AD pathogenesis (Li et al., 2000). We determined the effect of miR-200a-3p on Aβ1-42 production and found

(*n* = 3). (C, D) Decreased expression of miR-200a-3p in the hippocampus of APP/PS1 (C) and SAMP8 mice (D) (*n* = 4). (E) Reduced levels of miR-200a-3p in the plasma of AD patients compared with normal age-matched volunteers (NAVs) (*n* = 5–7). Data are shown as the mean ± SEM.\**P* < 0.05. \*\**P* < 0.01 versus relevant control.

that overexpression of miR-200a-3p inhibited the production of Aβ1-42 in miR-200a-3p mimics-transfected APPswe cells, whereas miR-200a-3p knockdown led to an Aβ1-42 overproduction in miR-200a-3p inhibitor-transfected APPswe cells (**Figure 2A**, *P* < 0.05).

The hyperphosphorylation of tau is associated with neural apoptosis and plays an important role in the development of AD. When we overexpressed miR-200a-3p in APPswe cells, the phosphorylation of tau protein at serine 202/threonine 205 (AT8), serine 214 (S214), serine 396 (S396), and serine 356 (S356) epitopes was significantly decreased (**Figures 2B**, **C**, *P* < 0.01– 0.001), whereas the inhibition of miR-200a-3p led to opposite effects (*P* < 0.05–0.001).

Furthermore, flow cytometry using PI and Annexin V staining was performed to analyze the involvement of apoptotic pathways in APPswe cells transfected with miR-200a-3p mimics, miR-200a-3p inhibitor, or NCs. Our results revealed that the ratios of early apoptosis, late apoptosis, and total apoptosis in APPswe cells overexpressing miR-200a-3p were all significantly decreased (**Figures 2D**, **E**, all *P* < 0.001), whereas inhibition of miR-200a-3p significantly increased the apoptotic ratios (*P* < 0.01–0.001). Among apoptotic pathways, the expression levels of the proapoptotic protein Bax were found to be downregulated when miR-200a-3p was overexpressed in APPswe cells and changed in an opposite manner when miR-200a-3p was inhibited (**Figures 2F**, **G**, *P* < 0.01, *P* < 0.001), accompanied by the activity and protein level of caspase-3, which were affected in the same manner in all transfected APPswe cells (**Figures 2F–H**, *P* < 0.05– 0.001). The apoptotic ratios, expression of Bax, and activity and protein level of caspase-3 in NC-transfected APPswe cells were not altered compared with those in the non-transfected groups. Collectively, these observations indicated that the upregulation of miR-200a-3p exerted a neuroprotective effect against AD deficits in APPswe cells.

inhibitors (200aI), corresponding negative controls (NCM/NCI), and nontransfected controls (control) in APPswe cells, as demonstrated by representative images (D) and quantitative analysis (E) by flow cytometry (*n* = 3). (F, G) miR-200a-3p mimics decreased the expression of caspase-3 and Bax, as demonstrated here by representative images (F) and quantitative analysis (G) by Western blotting. (H) miR-200a-3p mimics decreased the activity of caspase-3 in APPswe cells (*n* = 4). Data are shown as the mean ± SEM. *\*P* < 0.05, \*\**P* < 0.01, *\*\*\*P* < 0.001 versus relative control.

## The *BACE1* mRNA is a Direct Target of miR-200a-3p

To investigate the role of miR-200a-3p in AD development, it was necessary to find targets correlated with signaling transduction associated with AD pathology. For this, we first performed computational analyses to identify potential binding sites of miR-200a-3p using the miRNA target prediction database TargetScan. Our results indicated that miR-200a-3p had a potential target site in the 3'-UTR of the *BACE1* mRNA (**Figure 3A**). Subsequently, the systematic diagrams

is significantly decreased and increased after miR-200a-3p mimics (200aM) and inhibitor (200aI) transfection, respectively, relative to the corresponding negative controls (NCM/NCI), as demonstrated here by representative images (E) and quantitative analysis (F) by Western blotting (*n* = 3). The data are presented as the

of miR-200a-3p and *BACE1* mRNA 3'-UTR were established (**Figure 3B**).

mean ± SEM. \**P* < 0.05, \*\**P* < 0.01, \*\*\**P* < 0.001 versus relevant control.

Furthermore, a dual-luciferase reporter assay was used to investigate the manner in which miR-200a-3p regulated *BACE1*. For this, the 3'-UTR of *BACE1* mRNA containing WT or mutation binding site of miR-200a-3p was cloned in a luciferase vector, and miR-200a-3p mimics or NCs together with Renilla plasmid were cotransfected into HEK293 cells. The luminescence activity was significantly decreased in the cells that were cotransfected with miR-200a-3p mimics plus *BACE1* mRNA 3′-UTR WT (**Figure 3C**, *P* < 0.001). However, there was no effect on the luciferase activity in the cells cotransfected with luciferase plasmids containing mutation binding site of miR-200a-3p. Thus, we concluded that miR-200a-3p was specifically binding to the 3'-UTR of *BACE1*.

Because a direct relationship between miR-200a-3p and *BACE1* expression was established, Western blot and qRT-PCR analysis after the transfection of miR-200a-3p mimics, miR-200a-3p inhibitor, or NCs were performed to identify the target function of miR-200a-3p on the *BACE1* gene. As shown in **Figures 3E**, **F**, overexpression of miR-200a-3p downregulated the expression of the BACE1 protein (*P* < 0.05), whereas the inhibitory expression of miR-200a-3p upregulated BACE1 protein levels (*P* < 0.01). Because of the result that miR-200a-3p mimics or miR-200a-3p inhibitor did not influence the expression of *BACE1*mRNA at the transcriptional level significantly (**Figure 3D**), we concluded that the interaction manner of miR-200a-3p with BACE1 relied on the regulation of the protein translation process, rather than influencing the stability of *BACE1* mRNA.

## The PRKACB is Another Target of miR-200a-3p

Our bioinformatic analysis predicted that there was another candidate target of miR-200a-3p, *PRKACB*, a universally conserved gene that encodes one of the paralogous catalytic subunits of PKA that increases the levels of phosphorylated tau at Ser214, Ser356, and Ser396 epitopes by a number of kinases, widely reported in AD brains and mouse models (Wang et al., 2013; Wang et al., 2015). Importantly, the 3'-UTR of the *PRKACB* transcript contains a putative miR-200a-3p binding site (**Figure 4A**). Subsequently, the luciferase reporters were constructed containing the WT or binding site mutations of *PRKACB* 3'-UTR (**Figure 4B**). Our results indicated that transfection of miR-200a-3p mimics significantly inhibited the activity of the luciferase reporter for WT 3'-UTR *PRKACB* (**Figure 4C**, *P* < 0.001), but not for binding site mutations of *PRKACB* 3'-UTR, indicating the potential interaction between miR-200a-3p and *PRKACB*. We further examined the regulatory manner of miR-200a-3p in PRKACB at the protein and mRNA levels in APPswe cells by transfecting miR-200a-3p mimics and miR-200a-3p inhibitors using Western blot and qPCR analysis. Our data illustrated that miR-200a-3p mimics significantly downregulated both the mRNA and protein expressions of PRKACB (**Figures 4D–F**, both *P* < 0.001), whereas miR-200a-3p inhibitors upregulated PRKACB at these two levels (*P* < 0.001 and *P* < 0.01). Therefore, we suggested that PRKACB may act as another target of miR-200a-3p by associating with the 3'-UTR of *PRKACB*.

negative controls (NCM/NCI), as demonstrated here by representative images (E) and quantitative analysis (F) by Western blotting (*n* = 3). Data are shown as the mean ± SEM. \**P* < 0.05, \*\**P* < 0.01, \*\*\**P* < 0.001 versus relevant control.

## miR-200a-3p Displays a Neuroprotective Role in APPswe Cells by Regulating BACE1 and PRKACB, Decreasing A**β** Overproduction and tau Hyperphosphorylation, Respectively

During our study of miR-200a-3p on the pathological changes in AD associated with BACE1 and PRKACB, we explored the cell apoptosis, levels of Aβ1-42, and phosphorylation of the tau protein at several phosphorylated sites that PKA favors after overexpression of BACE1 and PRKACB cotransfected with miR-200a-3p mimics into APPswe cells. Apoptosis-related assays indicated that cotransfection of BACE1 and PRKACB with NCM increased the early, late, and total apoptotic ratios of APPswe cells and the activity of caspase-3, an apoptotic marker, as well (**Figures 5A–D**, *P* < 0.01–0.001). Meanwhile, the neuroprotection of miR-200a-3p mimics involving the suppression of apoptotic ratios and inhibition of active caspase-3 release in APPswe cells (*P* < 0.05–0.001) was reversed *via* cotransfection with BACE1 and PRKACB, respectively (*P* < 0.05–0.001). Additionally, when BACE1 and PRKACB were knockdown using BACE1 siRNA and PRKACB siRNA, a decreased tendency of miR-200a-3p neuroprotection was seen in APPswe cells (**Supplementary Figure 1**). These results indicated that there might be a direct neuroprotective relationship between miR-200a-3p and BACE1 and PRKACB.

As for Aβ overproduction and tau hyperphosphorylation, cotransfection of BACE1 and PRKACB with NCM increased the production of Aβ1-42 and the expression of tau phosphorylation at

AT8, Ser214, Ser396, and Ser356 (**Figures 5E–G**, *P* < 0.05–0.001). Importantly, cotransfection of BACE1 and PRKACB together with miR-200a-3p mimics abolished the beneficial effects of miR-200a-3p, increasing Aβ1-42 production and enhancing the levels of tau phosphorylation at detected phosphorylated sites (*P* < 0.05– 0.01). Collectively, these findings suggest that miR-200a-3p may protect neural cells by coregulating BACE1 and PRKACB *in vitro*.

## DISCUSSION

AD is a complex neurodegenerative disease affected by multigene activity and layers, which to date has no effective therapeutic treatments. As a result, a multigene regulatory approach is sought out as an AD therapeutic modality. Markedly, miRNAs are considered to be attractive candidates for their modulatory role in mRNA transcription and protein translation. In the present study, we deduced three primary findings regarding the function of one identified miRNA and its role in the pathological progression of AD. First, miR-200a-3p was found to demonstrate the consistent depression observed in the brain tissue of AD mice, in cell culture models of AD, and in the blood of AD patients. Second, miR-200a-3p was verified as a participant in the pathogenesis of AD directly *via* the regulation of *BACE1* and *PRKACB* expression levels using a target prediction database and through a dualluciferase reporter assay. Third, miR-200a-3p was found to have neuroprotective effects by suppressing the overproduction of Aβ and the hyperphosphorylation of the tau protein *via* regulating the

FIGURE 5 | Neuroprotection of miR-200a-3p by regulating the participation of BACE1 and PRKACB in Aβ overproduction and tau hyperphosphorylation, respectively, in APPswe cells. (A, B) Effects of miR-200a-3p, BACE1, and PRKACB overexpression on apoptosis of APPswe cells detected using flow cytometry assay (A) and demonstrated by quantitative analysis (B) (*n* = 3). (C, D) The release of active caspase-3 in APPswe cells when miR-200a-3p and BACE1 (C) or PRKACB (D) overexpression measured using ELISA (*n* = 4). (E) Effects of miR-200a-3p and BACE1 overexpression on the production of Aβ1-42 peptides in a cell model of AD (*n* = 4). (F, G) Effects of miR-200a-3p and PRKACB overexpression on tau phosphorylation at AT8, Ser214 (pTS214), Ser396 (pTS396), and Ser356 (pTS356) sites in APPswe cells, as demonstrated by representative images (F) and quantitative analysis (G) by Western blotting (*n* = 3). Data are shown as the mean ± SEM. *\*P* < 0.05, \*\**P* < 0.01, *\*\*\*P* < 0.001 versus NCM. *\$P* < 0.05, *\$\$P* < 0.01, *\$\$\$P* < 0.001 versus 200aM.

protein expression of BACE1 and PRKACB in the AD cell model, respectively. Here, we provide an alternative strategy by targeting miR-200a-3p for the prevention and/or treatment of AD.

Although AD develops as a multifactorial process, the abnormal overproduction and accumulation of Aβ are still considered to be key events (Moore et al., 2014). BACE1 is highly expressed in neurons and is a crucial protease that functions in the first step of the pathway leading to the majority of the Aβ production in the pathology of AD (Marques et al., 2012). BACE1 has been proposed as a viable therapeutic target for AD; however, challenges with the currently investigated BACE1 inhibitors, involving low oral bioavailability, long serum half-life, and low blood–brain barrier (BBB) penetration ratio (Vassar, 2014), have not been fully overcome yet. Thus, investigators have turned toward miRNA replacement therapy in light of their specific interactions. There are several miRNAs that have been proven to target BACE1, such as miR-124, miR-29c, and miR-195 (Zong et al., 2011; Fang et al., 2012; Zhu et al., 2012; Du et al., 2017). Following deep data mining analysis combined with target prediction and confirmatory experiments, we found that the downregulation of miR-200a-3p was implicated in AD etiology and targeted the 3'-UTR of BACE1 mRNA. We also verified that there was a negative correlation between miR-200a-3p and BACE1 in the blood of AD patients and that the imposed upregulation of miR-200a-3p significantly downregulated the expression of BACE1 at the protein level. As BACE1 is known to be significantly increased in the brains of AD patients, as well as in their CSF (Stozická et al., 2007; Evin and Hince, 2013), we suggest that miR-200a-3p could act as a potential peripheral biomarker together with BACE1 for AD diagnosis and/or treatment.

miR-200a-3p is a member of the miR-200 family of miRNAs that plays a vital role in various cancers, including bladder cancer, breast cancer, colorectal cancer, and gastric cancer, among others (Feng et al., 2014; Muralidhar and Barbolina, 2015). Recently, the miR-200 family of proteins was found to be crucial for neural differentiation and proliferation, along with their participation in Aβ production in AD (Liu et al., 2014; Pandey et al., 2015). In the present study, we used the established AD *in vitro* model in APPswe cells and found through apoptotic and cytoactive analyses that Aβ led to high apoptotic ratios of neurons, accompanied by the activation of the Bax/caspase-3 axis, whereas upregulation of miR-200a-3p rescued neural apoptosis and recovered the apoptotic caspase-3 pathway. In addition, miR-200a-3p relieved the production of Aβ1-42, manifesting that miR-200a-3p protected against the injury derived from the overexpression of APP in the neuroblastoma SH-SY5Y cell line. Importantly, when miR-200a-3p and BACE1 expression were collaboratively established, the inversed effects of BACE1 on miR-200a-3p function were observed in APPswe cells, indicating an increase in cell apoptosis, an increase in caspase-3 activities, and an overproduction of Aβ1-42 peptides. Collectively, our findings may illuminate a potential mechanism indicating that miR-200a-3p contributed to the neuroprotective effects in AD *via* the regulation of BACE1.

In addition to the Aβ pathology, the role of tau hyperphosphorylation is another widely appreciated etiology in AD development. The tau protein is encoded by the *MAPT* gene, which consists of six isoforms in the central nervous system (CNS) generated by alternatively splicing and containing numerous sites that can be phosphorylated *via* the activity of various enzymes, including PKA, CDK5, GSK3β, and ΜΑPK (Lee and Leugers, 2012). The development of NFTs is composed of three stages involving phosphorylated tau proceedings, preneural NFTs (pre-NFTs), intraneural NFTs, and extraneural NFTs. Phosphorylation epitopes at serine 199, 202, and 409 are associated with pre-NFT stages, whereas serine 396, 404, and threonine 231 occur in intraneuronal NFT stages (Kimura et al., 1996). The phosphorylation of tau at serine 396 and 404 sites was observed in early and late stages of AD and has been demonstrated to have greater preference in the earliest formation of NFTs, whereas the serine 214, serine 202, threonine 205 sites have been shown to be mostly associated with mature NFTs (Mondragon-Rodriguez et al., 2014). PKA is an important kinase that phosphorylates multiple sites of the tau protein, including serine 214, 396, and 356, alone or by sequential phosphorylation *via* cooperation with other kinases, such as GSK3β or CDK5, in the progression of AD (Jensen et al., 1999; Ksiezak-Reding et al., 2003; KyoungPyo et al., 2004; Wang et al., 2007). Essentially, prephosphorylated tau by PKA has been suggested to be the better substrate than the phosphorylation by other kinases, such as CDK5 and GSK3β (Liu et al., 2006). Thus, we chose the most representative phosphorylation sites of tau involved in AD and also PKA-preferred, such as serine 396, serine 214, serine 356, serine 202, and threonine 205, to perform our research.

As demonstrated in the present study *via* miRNA target prediction and validation using a dual-luciferase reporter assay, PRKACB, the catalytic units of PKA, was also identified as a direct target of miR-200a-3p, which negatively regulates the expression of PRKACB, thereby inhibiting PKA protein expression and activation. Moreover, miR-200a-3p demonstrated a consistent inhibitory effect on the phosphorylation of tau at serine 202/threonine 205 (AT8) in APPswe cells, including the AKT-preferable phosphorylation site serine 214, 396, and 356 (Jensen et al., 1999; Ksiezak-Reding et al., 2003; KyoungPyo et al., 2004). These observations provide further evidence of the miR-200a-3p-modulated pathological alterations of tau in AD. Moreover, our investigation showed that PKA overexpression blocked the inhibitory effects of miR-200a-3p on the modulation of tau phosphorylation at the examined phosphorylation sites of tau, including the epitopes that were mainly phosphorylated by PKA, *in vitro*. Taken together, we propose the existence of a regulatory event between miR-200a-3p and PKA that could help explain the mediation of tau hyperphosphorylation leading to the pathology of AD.

## CONCLUSIONS

In summary, our study suggests that miR-200a-3p is implicated in AD progression and can exert neuroprotective effects against Aβ-induced toxicity by two possible mechanisms: first, by directly and/or indirectly inhibiting the overproduction of Aβ *via*  suppressing the expression of BACE1 and second, by simultaneously decreasing the hyperphosphorylation of tau through attenuating

the expression of PKA (**Figure 6**). In addition to this study, further mechanisms need to be explored, particularly those associated with the regulatory axis of miR-200a-3p. In addition, local delivery of miR-200a-3p into certain brain areas, such as the cortex and hippocampus, *via* proper and novel carriers may also shed a light on preventing detrimental effects of miRNA treatment in AD.

## ETHICS STATEMENT

The animals had *ad libitum* access to food and water at stable room temperature and humidity environment according to the Guide for the Care and Use of Laboratory Animals. The experiment was proved by the Ethical Committee of the Institute of Medicinal Biotechnology (IMB-D8-2018071102).

## AUTHOR CONTRIBUTIONS

LW and JL carried out the experiments, analyzed the results, and wrote the manuscript. JL provided the plasma. QW, HJ, and LZ

## REFERENCES


helped design the experiments, prepare the figures, analyze the data, and review the manuscript. RL and ZL designed the study and the experiments, interpreted the results, and wrote the manuscript. All authors have read and approved the final manuscript.

## FUNDING

This study was supported by the National Natural Science Foundation of China (No. U1803281 and 81673411), China; the Non-profit Central Research Institute Fund of Chinese Academy of Medical Sciences (2018RC350013), China; and the Chinese Academy Medical Sciences (CAMS) Innovation Fund for Medical Science (2017-I2M-1-016), China.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fphar.2019.00806/ full#supplementary-material


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

*Copyright © 2019 Wang, Liu, Wang, Jiang, Zeng, Li and Liu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Palmitoylethanolamide (PEA) as a Potential Therapeutic Agent in Alzheimer's Disease

*Sarah Beggiato1,2,3, Maria Cristina Tomasini1,2 and Luca Ferraro1,2,3\**

*1 Department of Life Sciences and Biotechnology, University of Ferrara, Ferrara, Italy, 2 Technopole of Ferrara, LTTA Laboratory for the Technologies for Advanced Therapies, Ferrara, Italy, 3 IRET Foundation, Bologna, Italy*

*N*-Palmitoylethanolamide (PEA) is a non-endocannabinoid lipid mediator belonging to the class of the *N*-acylethanolamine phospolipids and was firstly isolated from soy lecithin, egg yolk, and peanut meal. Either preclinical or clinical studies indicate that PEA is potentially useful in a wide range of therapeutic areas, including eczema, pain, and neurodegeneration. PEA-containing products are already licensed for use in humans as a nutraceutical, a food supplement, or a food for medical purposes, depending on the country. PEA is especially used in humans for its analgesic and anti-inflammatory properties and has demonstrated high safety and tolerability. Several preclinical *in vitro* and *in vivo* studies have proven that PEA can induce its biological effects by acting on several molecular targets in both central and peripheral nervous systems. These multiple mechanisms of action clearly differentiate PEA from classic anti-inflammatory drugs and are attributed to the compound that has quite unique anti(neuro)inflammatory properties. According to this view, preclinical studies indicate that PEA, especially in micronized or ultramicronized forms (i.e., formulations that maximize PEA bioavailability and efficacy), could be a potential therapeutic agent for the effective treatment of different pathologies characterized by neurodegeneration, (neuro) inflammation, and pain. In particular, the potential neuroprotective effects of PEA have been demonstrated in several experimental models of Alzheimer's disease. Interestingly, a singlephoton emission computed tomography (SPECT) case study reported that a mild cognitive impairment (MCI) patient, treated for 9months with ultramicronized-PEA/luteolin, presented an improvement of cognitive performances. In the present review, we summarized the current preclinical and clinical evidence of PEA as a possible therapeutic agent in Alzheimer's disease. The possible PEA neuroprotective mechanism(s) of action is also described.

Keywords: neuroinflammation, preclinical studies, animal models, 3xTg-AD, ultramicronized formulation

## INTRODUCTION

Neuroinflammation and synaptic dysfunction in Alzheimer's disease (AD) have been originally considered as epiphenomena with inflammation and altered neurotransmission occurring when damaged neurons provoke glia activation and changes in neuron biology. However, the growth of knowledge about the molecular mechanisms underlying AD converted this earlier view and points to a causal role of these events in the pathology (Overk and Masliah, 2014; Heneka et al., 2015; Steardo et al., 2015; Van Eldik et al., 2016; González-Reyes et al., 2017; Ahmad et al., 2019). Specifically, it

#### *Edited by:*

*Cesare Mancuso, Catholic University of the Sacred Heart, Italy*

#### *Reviewed by:*

*Andrea Armirotti, Istituto Italiano di Tecnologia, Italy Marco Feligioni, European Brain Research Institute, Italy*

> *\*Correspondence: Luca Ferraro luca.ferraro@unife.it*

#### *Specialty section:*

*This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology*

*Received: 29 April 2019 Accepted: 25 June 2019 Published: 24 July 2019*

#### *Citation:*

*Beggiato S, Tomasini MC and Ferraro L (2019) Palmitoylethanolamide (PEA) as a Potential Therapeutic Agent in Alzheimer's Disease. Front. Pharmacol. 10:821. doi: 10.3389/fphar.2019.00821*

Beggiato et al. PEA and Alzheimer's Disease

is now well established that the pathogenesis of AD includes also interactions with immunological mechanisms/responses in the brain. Neuroinflammation in AD is predominantly linked to central nervous system (CNS)-resident microglia, astroglia, and perivascular macrophages, which have been implicated at the cellular level (Zádori et al., 2018). Regional inflammatory responses characterize the CNS in AD, with deposits of β-amyloid (Aβ) as foci, associated with increased expression of pro-inflammatory cytokines, acute phase proteins, and complement components, along with signs of activated microglia and reactive astrocytes (Skaper et al., 2018). According to this scenario, neuropathological studies in human brains, demonstrating the activation of glial cells, mainly microglia and astrocytes (Zimmer et al., 2014; Chaney et al., 2018; Edison et al., 2018; Knezevic and Mizrahi, 2018), have been corroborated by studies in animal models of AD in which an overproduction of pro-inflammatory signals by glial cells triggers a neurodegenerative cascade (Birch et al., 2014; Heneka et al., 2015; Chun et al., 2018; Saito and Saido, 2018). On the other hand, mounting evidence indicates that also oxidative stress and synaptic dysfunction are early events in AD (Overk and Masliah, 2014; Wirz et al., 2014; Kamat et al., 2016; Cai and Tammineni, 2017). Changes in neuronal activity/signaling in AD can promote the β-amyloidogenic pathway of amyloid precursor protein (APP) processing, leading to increased Aβ levels and thus creating a sort of a positive feedback or a vicious cycle to accelerate AD pathogenesis (Herrup, 2010; Wirz et al., 2014; Cai and Tammineni, 2017). These findings indicate that neuroinflammation, oxidative stress, and synaptic dysfunction are integral parts of AD pathogenesis, and not solely consequences of Aβ-induced CNS damage. Thus, the relationship between neurodegeneration and neuroinflammation is strictly interdependent, suggesting that compounds able to simultaneously target these processes might be effective therapeutic agents in AD. In this context, endocannabinoid signaling and endocannabinoid-related compounds have been demonstrated to modulate the main pathological processes during early AD, including protein misfolding, neuroinflammation, excitotoxicity, mitochondrial dysfunction, and oxidative stress (Aso and Ferrer, 2014; Bedse et al., 2015; Fernández-Ruiz et al., 2015). Among these compounds, *N*-palmitoylethanolamide (PEA) has attracted much attention because it exerts a local anti-injury function through a down-modulation of mast cells and protects neurons from excitotoxicity through several mechanisms (Mattace Raso et al., 2014; Petrosino and Di Marzo, 2017).

PEA is a non-endocannabinoid lipid mediator belonging to the class of the *N*-acylethanolamine (NAE) phospolipids, which also includes the first endocannabinoid to be discovered, *N*-arachidonoyl-ethanolamine (anandamide; AEA) and the anorectic mediator *N*-oleoyl-ethanolamine (OEA). PEA was firstly isolated from soy lecithin, egg yolk, and peanut meal (Ganley et al., 1958; Petrosino and Di Marzo, 2017). Either preclinical or clinical studies indicate that PEA is potentially useful in a wide range of therapeutic areas, including eczema, pain, and neurodegeneration. PEA-containing products (Normast®, Glialia®, Nevamast®, Adolene®, Visimast®, Mastocol®, and Pelvilen®) are already licensed for use in humans (generally 1,200 mg/day) as a nutraceutical, a food supplement, or a food for medical purposes, depending on the country. PEA is especially used in humans for its analgesic and antiinflammatory properties (Petrosino and Di Marzo, 2017; Tsuboi et al., 2018) and has demonstrated high safety and tolerability (Gabrielsson et al., 2016; Nestmann, 2016; Petrosino and Di Marzo, 2017). In the last decade, several studies suggested that PEA might exert protection against neuroinflammation and neurodegeneration, thus indicating that the compound possesses exceptional potential as a novel treatment for neurodegenerative disorders (Hansen, 2010; Skaper et al., 2014; Iannotti et al., 2016; Brotini et al., 2017; D'orio et al., 2018; Scuderi et al., 2018).

In this review, we initially briefly discuss the main molecular targets of PEA and its pharmacological properties, including the available pharmacokinetic data. Successively, we report the *in vivo* and *in vitro* findings, along with clinical results, supporting the possible role of PEA as a therapeutic agent in AD.

## PHARMACOLOGY OF PEA

PEA attracted the interest of the scientific community mainly after the discovery by an Italian Nobel Prize laureate Rita Levi Montalcini and co-workers that some acylethanolamides, initially termed ALIA-amides (autacoid local injury antagonist; ALIA) are endogenously synthesized lipids exerting interesting antiinflammatory properties (Levi-Montalcini et al., 1996). PEA (C16:0 *N*-acylethanolamine; **Figure 1**) is a lipid mediator biologically synthetized in many plants as well as in cells and mammal tissues. It belongs to the class of non-endocannabinoid NAE, which also includes stearoylethanolamide (C18:0 *N*-acylethanolamine), oleoyl-ethanolamide (OEA, C18:1 *N*-acylethanolamine), and linoleoylethanolamide (C18:2 *N*-acylethanolamine). These compounds are much more abundant than the endocannabinoid anandamide in several animal tissues and endowed with important biological actions. The biosynthesis and metabolism of PEA have been deeply described elsewhere (Petrosino et al., 2010; Tsuboi et al., 2013; Petrosino and Di Marzo, 2017; Tsuboi et al., 2018), and we refer to those reviews for their description.

## Mechanisms of Action of PEA: Focus on Neuroinflammation

Several preclinical *in vitro* and *in vivo* studies have demonstrated that PEA can induce its biological effects by acting on several molecular targets in both central and peripheral nervous systems (Mattace Raso et al., 2014; Iannotti et al., 2016; Petrosino and Di Marzo, 2017; Tsuboi et al., 2018). As reported above, it has been initially suggested that PEA, belonging to the class of acylethanolamides, exerts its anti(neuro)inflammatory effects by acting as an "autacoid local injury antagonist" (ALIA) leading to a down-regulation of mast cell activation (Levi-Montalcini et al., 1996). However, subsequent preclinical studies strongly supported the view that PEA can directly activate at least two different receptors: the peroxisome proliferator-activated receptoralpha (PPAR-α; Lo Verme et al., 2005) and the orphan GPCR 55 (GPR55; Pertwee, 2007).

PPAR-α actually seems to be the main molecular target involved in the anti(neuro)inflammatory effects of PEA. PPAR-α is, in fact, known for its protective role against (neuro)inflammation, and PPAR-α ligands are recognized as possible anti-inflammatory compounds (Devchand et al., 1996; Straus and Glass, 2007). When activated by a ligand, PPAR-α forms a heterodimer with 9-*cis*-retinoic acid receptor (RXR) able to interact with specific DNA sequences in the promoter regions of selective genes, thus leading to complex anti-inflammatory responses (Daynes and Jones, 2002). *In vitro*, PEA is able to activate PPAR-α with a half-maximal effective concentration (EC50) of 3.1 ± 0.4 μM (De Gregorio et al., 2018). Numerous studies demonstrated that PPAR-α antagonist or the genetic ablation of this receptor counteracts/prevents the protective effects of PEA against neuroinflammation and neurodegeneration in cellular or animal models of different pathologies (Scuderi et al., 2011; D'Agostino et al., 2012; Esposito et al., 2012; Paterniti et al., 2013a; Avagliano et al., 2016; Cristiano et al., 2018), thus supporting the relevance of this target in the mechanism of action of PEA.

PEA has shown agonist activity towards the orphan receptor GPR55 (Baker et al., 2006), which was proposed as a third cannabinoid receptor (Pertwee, 2007; Yang et al., 2016). In fact, cannabinoids are able to interact with GPR55, thus inducing some behavioral, immunological, and neuroinflammatory activities (De Gregorio et al., 2018; Balenga et al., 2014). However, the limited sequence similarity between GPR55 and cannabinoid receptors does not support this concept (Baker et al., 2006). At the present, the relevance of this receptor activation in the anti-inflammatory/neuroprotective PEA-induced effects remains to be clarified. It has been reported that PEA improves murine experimental colitis and that this effect is, at least partially, mediated by GPR55 activation (Borrelli et al., 2015). Furthermore, PEA protects against atherosclerosis by promoting an anti-inflammatory and proresolving phenotype of lesional macrophages, and this effect involves GPR55 activation (Rinne et al., 2018). The expression of GPR55 was protective against the insult exerted by MPP+ in a cellular model of Parkinson's disease, but an agonist of GPR55 did not enhance neuroprotection in GPR55-expressing cells (Martínez-Pinilla et al., 2019). However, the GPR55 agonist abnormal-cannabidiol (Abn-CBD), a synthetic cannabidiol isomer, displayed beneficial properties when chronically administered (5 weeks) to a murine model of Parkinson's disease (Celorrio et al., 2017). Moreover, a neuroprotective role of GPR55 activation on neural stem cells *in vitro* and *in vivo* has been recently proposed, thus suggesting that GPR55 could provide a novel therapeutic target against negative regulation of hippocampal neurogenesis by inflammatory insult (Hill et al., 2019). Finally, a selective agonist for GPR55 protected dentate gyrus granule cells and reduced the number of activated microglia after NMDA induced lesions in an *in vitro* model of rat organotypic hippocampal slice cultures (Kallendrusch et al., 2013). Taken together, these data suggest that the beneficial antineuroinflammatory effects of PEA might be mediated, at least in part, by GPR55 activation. However, other data suggested anti-inflammatory properties of GPR55 blockade. For example, a GPR55 antagonist diminished inflammation in experimental colitis by reducing the levels of pro-inflammatory cytokines, tumor necrosis factor-alpha (TNF-α), interleukin (IL)-1β, and IL-6 and impairing leukocyte activation and migration (Stančić et al., 2015). In addition, anti-neuroinflammatory effects have been observed after the treatment of LPS-activated primary microglial cells with a GPR55 inverse agonist (Saliba et al., 2018). Thus, the precise role of GPR55 in the anti-inflammatory/neuroprotective PEA action remains to be elucidated.

Besides its direct action on PPAR-α and GPR55, compelling evidence indicates that PEA could produce several indirect receptor-mediated actions, through the so-called entourage effect (Mattace Raso et al., 2014; Petrosino and Di Marzo, 2017). Given its weak affinity for CB1 and CB2 receptors, cannabinoid receptors are not considered direct targets of PEA. However, PEA can indirectly activate cannabinoid receptors through different indirect mechanisms. In particular, PEA may indirectly activate CB1 and CB2 receptors by acting as a false substrate for fatty acid amide hydrolase (FAAH), the enzyme involved in the degradation of the endocannabinoid AEA (Petrosino et al., 2016; Petrosino and Di Marzo, 2017), thus leading to a reduced degradation of AEA. This action leads to increased levels of AEA and, in turn, an increased activation of cannabinoid receptor-mediated signaling. Furthermore, quite recent studies have demonstrated that PEA increases the levels of CB2 receptor mRNA and protein as a result of PPAR-α activation, and this effect is involved in PEA-induced microglia changes associated with increased migration and phagocytic activity (Guida et al., 2017). Finally, the discovery that GPR55 forms receptor heteromer with either CB1 or CB2 receptors (Balenga et al., 2014; Martínez-Pinilla et al., 2014; Martínez-Pinilla et al., 2019) raises the exciting possibility that PEA might modulate CB1- and/or CB2-mediated intracellular signaling by targeting the GPR55 protomer in these putative GPR55/CB1 or GPR55/CB2 heterodimers. PEA can also indirectly activate the transient receptor potential vanilloid type 1 (TRPV1) channel, which is also a target for the endocannabinoids (Zygmunt et al., 2013), *via* different mechanisms. In particular, PEA-induced increase of endocannabinoid levels can modulate inflammation and other immune functions *via* TRPV1 channel (Ross, 2003). In addition, putative allosteric properties of PEA at TRPV1 channels have been proposed to possibly explain the ability of the compound to increase the endocannabinoidinduced activation and desensitization of TRPV1 channels (Petrosino and Di Marzo, 2017). Finally, as the existence of a direct biochemical interaction has been proposed, it seems likely that PEA can also indirectly activate TRPV1 channels *via* PPAR-α activation. The possible relevance of these mechanisms in the anti-neuroinflammatory/neuroprotective effects of PEA remains to be clarified. In fact, TRPV1 channel activation has been linked to either anti-neuroinflammatory or pro-neuroinflammatory signaling (Kong et al., 2017). Interestingly, it has been recently reported that TRPV1 activation reduces central inflammation in multiple sclerosis (Stampanoni Bassi et al., 2019). Accordingly, neuroprotective effects of TRPV1 activation in animal models of Parkinson's and Alzheimer's diseases have been reported (Jiang et al., 2013; Nam et al., 2015; Jayant et al., 2016; Xu et al., 2017; Zhao et al., 2017; Balleza-Tapia et al., 2018).

Taken together, the above findings strongly suggest that PEA by activating multifactorial pharmacological targets and by mediating several cellular mediators could play promising protective roles in contrasting neuroinflammation and neurodegeneration. The ability of PEA to synergistically interact *via* several mechanisms is attributed to the compound's quite unique properties in respect to the traditional antiinflammatory drugs.

## Pharmacokinetic

Given its poor water solubility, large particle size in the native state, and, possibly, short‐lived action, PEA might have limitations in terms of solubility and bioavailability. In fact, PEA is almost insoluble in water, while its solubility in most other aqueous solvents is very poor with a partition coefficient (log *P*) estimated to be > 5 (Lambert et al., 2001). Published data on PEA bioavailability are still scarce, but recent findings are contributing to better understand the pharmacokinetic of the compound and the possible relevance of new oral formulations.

It was originally reported that in rats, following its intraperitoneal (i.p.) administration, *N*‐[1‐14C]‐PEA was mainly distributed in some peripheral organs, and the lower concentrations were found in the brain, plasma, and erythrocytes (Zhukov, 1999). Moreover, orally administered *N*‐[9,10‐3 H]‐PEA (100mg/kg of body weight) was able to penetrate through the blood–brain barrier (BBB), but only in small amounts with a brain bioavailability corresponding to 0.95% of the oral dose (Artamonov et al., 2005; Gabrielsson et al., 2016). It has also been reported that PEA administration to humans leads to a two- to nine-fold increase in plasma baseline concentrations, depending on the dose (Balvers et al., 2013). The poor pharmacokinetic of PEA prompted the development of different formulation strategies, especially aimed at ameliorating the compound distribution. For instance, it has been demonstrated that when PEA was formulated as an emulsion in corn oil and administered subcutaneously (s.c.) to young DBA/2 mice (10mg/kg of body weight), the compound was more extensively distributed in several organs, including the brain (Grillo et al., 2013).

In addition, a PEA suspension in corn oil administered to rats by gastric gavage (100mg/kg of body weight) led to an about 20-fold increase in basal PEA plasma levels (Vacondio et al., 2015).

The highest PEA plasma concentration was observed after 15min (*C*max = 420 ± 132nM); PEA plasma levels returned to the baseline ones ~2h after the compound administration. The formulation of PEA as micronized or ultramicronized particles (m‐PEA and µm‐PEA, respectively) has been more recently proposed as a strategy to possibly increase PEA bioavailability, also in the CNS, without affecting the pharmacological efficacy of the compound (Impellizzeri et al., 2014; Petrosino and Di Marzo, 2017; Petrosino et al., 2018). It has been firstly reported that the oral administration of µm‐PEA (30mg/kg of body weight) to a beagle dog led to a five-fold increase in blood PEA concentration. The peak of plasma PEA levels (~55–60pmol/ml) was observed 1 and 2h after the administration of the compound (Cerrato et al., 2012). Subsequently, another study confirmed this finding (Petrosino et al., 2016). Another pharmacokinetic profile of m‐PEA and µm‐PEA after a single oral administration (15mg/kg of body weight) to beagle dogs is reported in a US patent (Della Valle et al., 2013). In this case, blood samples have been taken at time 0 (immediately before the administration of PEA) and at times (*t*) 1, 2, and 3h; the administration of m‐PEA and µm‐PEA leads to similar peak concentration values of PEA in serum (22.2 and 22.4pmol/ml, respectively; ~2 times higher than the baseline values) measured in the blood samples taken 1h after the compound administration, with PEA concentrations returning to basal values at *t* = 2h. Petrosino et al. (2016) reported the first preliminary pharmacokinetic data in humans. In particular, the authors measured blood PEA concentrations after the oral administration of m-PEA (300mg) to human volunteers. Blood sample collection was carried out immediately before, and after 2, 4, and 6h after PEA assumption; under this conditions, the peak of plasma PEA levels (~22pmol/ml) was observed 2h after the compound assumption, with a drop to baseline levels within the following 2h. Very recently, Petrosino et al. (2018) demonstrated by orally administering µm‐[13C]4-PEA or a naïve [13C]4-PEA (30mg/kg of body weight) formulation to healthy and carrageenaninjected rats, that ultramicronization increases the ability of PEA to reach peripheral and central tissues under either healthy or local inflammatory conditions. In particular, the plasma concentrations of [13C]4-PEA were measured at 5, 15, 30, and 60min after the oral administration of the compound in ultramicronized and naïve formulations to healthy rats. Rats receiving µm-[13C]4-PEA showed higher mean plasma levels of the compound than rats receiving naïve [13C]4-PEA. In rats receiving µm‐[13C]4-PEA, the peak concentration of [13C]4-PEA (5.4 ± 1.87pmol/ml) was found after 5min, and it was five times higher than the concentration measured in rats administered with the naïve formulation (1.1 ± 0.35pmol/ml), in which no significant peak plasma concentrations were found.

Collectively, the above findings suggest that micronized or ultramicronized formulations of PEA maximize the compound bioavailability and efficacy, although further studies are necessary to undoubtedly confirm this hypothesis. Other strategies have been proposed to improve PEA bioavailability. For instance, PEA ester derivatives, prepared by conjugating PEA with various amino acids, have been synthetized as PEA prodrug and allowed to modulate the kinetics of PEA release in plasma and stability in liver homogenate (Vacondio et al., 2015). Two derivatives, l-Val-PEA, with suitable PEA release in plasma, and d-Val-PEA, with high resistance to hepatic degradation, were orally Beggiato et al. PEA and Alzheimer's Disease

administered to rats, and plasma levels of prodrugs and PEA were measured at different time points, in comparison with naïve PEA (equimolar doses corresponding to 100mg/kg of PEA). Both prodrugs showed significant release of PEA but provided lower plasma concentrations than those obtained with equimolar doses of naïve PEA. The highest PEA plasma concentrations were observed after 15min following PEA, l-Val-PEA, or d-Val-PEA (420 ± 132, 56.4 ± 13.5, or 53.9 ± 19.7nM, respectively). It has also been reported that the loading of the compound in nanostructured lipid carriers (NLCs) enhances the ocular bioavailability of PEA (Puglia et al., 2018) and that polyethylene glycol esters of PEA proved to be able to delay and prolong the pharmacological activity of the compound (Tronino et al., 2015), thus suggesting that these formulations might also ameliorate systemic PEA pharmacokinetic.

## PEA AND ALZHEIMER'S DISEASE

Several preclinical and some clinical indications support the view of PEA as a therapeutic tool with high potential for the effective treatment of different pathologies characterized by neurodegeneration and neuroinflammation (Calabrò et al., 2016; Brotini et al., 2017; Holubiec et al., 2018; Impellizzeri et al., 2019). In this context, the potential beneficial effects of PEA have been demonstrated in several *in vitro* and *in vivo* experimental models of AD.

## Preclinical Evidence

### *In Vitro* Studies

To our knowledge, the first experimental indication of PEA as a possible therapeutic agent in AD has been published by Scuderi et al. (2011). In their pioneering work, the authors evaluated the ability of PEA (10−7M) to mitigate Aβ (Aβ1–42; 1μg/ml)-induced astrogliosis in primary cultures of rat astrocytes. The results indicate that PEA treatment attenuated Aβ-induced astrocyte activation, as proven by its effects in reducing astrocyte hypertrophied cell bodies and thickened processes, along with the expression of glial fibrillary acidic protein (GFAP) and S100 calcium-binding protein B (S100B), two specific markers of astrocyte activity also linked to AD pathogenesis. Furthermore, PEA was able to blunt Aβ-induced neuroinflammation by significantly diminishing either the altered expression of pro-inflammatory molecules, such as cyclooxygenase-2 (COX-2) and inducible nitric oxide synthase (iNOS), or the enhanced release of prostaglandin PGE2, nitric oxide, IL-1β, and TNF-α. Interestingly, the PPAR-α antagonist MK886 was able to partly blunt the PEA-induced effects against Aβ-induced astrogliosis and neuroinflammation, thus suggesting a significant, but not exclusive, involvement of the PPAR-α in mediating the above-mentioned PEA actions. Concerning the possible intracellular signaling involved in PEA-induced effects, it has also been demonstrated that PEA critically diminished the Aβ-induced activation of p38 and Jun N-terminal kinase (JNK), as well as the subsequent activation of nuclear transcription factors, such as nuclear factor kappaB (NF-kB) and activator protein 1 (AP-1) (Scuderi et al., 2011). Later, the same group demonstrated that PEA treatment exerted protective effects against Aβ-induced toxicity also in primary rat mixed neuroglial co-cultures and organotypic hippocampal slices (Scuderi et al., 2012; Scuderi and Steardo, 2013). In particular, in mixed neuroglial co-cultures PEA prevented the increase in astrocyte number and the quantity of apoptotic nuclei in microtubule-associated protein 2 (MAP2)-positive neurons induced by Aβ challenge. Under these experimental conditions, the PEA antigliosis and neuroprotective effects were completely ascribed to PPARα activation, since MK886, the selective PPARα antagonist, almost completely abolished the PEA-induced effects. On the contrary, GW9662, a selective PPARγ antagonist, did not exert any significant influence. Furthermore, PEA decreased Aβ-induced astrocyte and microglia activation in organotypic cultures of rat hippocampi, an effect associated with a rescue of neuronal CA3 damage caused by Aβ challenge. PEA treatment also rescued neuron integrity and reduced the levels of neuroinflammation markers in this preparation. Once more, these effects were completely abolished by the pretreatment with a PPARα antagonist (Scuderi and Steardo, 2013). Finally, PEA was also evaluated for its possible effects in AD angiogenesis and neuroinflammation by using Aβ-treated C6 rat astroglioma cells and human umbelical vein endothelial cells (HUVEC) (Cipriano et al., 2015). In line with the previous findings, under these experimental conditions, PEA concentration-dependently reduced the expression of pro-inflammatory and pro-angiogenic markers in Aβ (1μg/ml)-stimulated C6 cells. Interestingly, the medium aspired from PEA-treated C6 cells was able to reduce the HUVEC proliferation induced by their exposure to the conditioned medium from Aβ-treated C6 cells. The possible anti-angiogenic properties of PEA were also supported by the demonstration that the compound inhibited the nuclear levels of mitogen-activated protein kinase 1, which is associated with the main pro-angiogenic pathway, as well as the cytoplasmic vascular endothelial growth factor in HUVEC exposed to the medium from Aβ-treated C6 rat astroglioma cells. Once again, these effects were blocked by the treatment with the PPAR-α antagonist GW6471. As the release of proangiogenic factors during astrogliosis has been suggested as a key step in controlling AD progression, these findings further support the role of PEA as therapeutic agents for AD (Cipriano et al., 2015).

During the last years, other groups confirmed the protective action of PEA against the *in vitro* toxic effects of Aβ. For instance, in a very elegant study, it has been demonstrated that in wild-type (WT) mice, the addition of several acylethanolamides (including PEA) partially reverted Aβ-induced inflammation. However, the genetic deletion of FAAH (i.e., the enzyme involved in the degradation of the endocannabinoid AEA) in astrocytes induced an increased sensitivity to the pro-inflammatory Aβ-induced action, and this effect involved PPAR-α, PPAR-γ, and TRPV1 receptors, but not CB1 or CB2 receptors (Benito et al., 2012). Based on these findings, the authors raised the possibility that an excessively prolonged enhancement of the endocannabinoid tone may have harmful consequences, instead of the beneficial effects exerted by an acute increased tone.

Moving from the above data, we evaluated the protective role of PEA against Aβ-induced toxicity on cell viability and glutamatergic transmission in primary cultures of cerebral cortex neurons and astrocytes from the triple-transgenic murine model of AD (3xTg-AD) and their WT littermates (non-Tg mice; Tomasini et al., 2015). 3xTg-AD mice were selected because these animals harbor three mutant human genes (APPSwe, PS1M146V, and tauP301L) and closely mimic many aspects of AD in humans. In fact, these animals are characterized by agedependent build-up of both plaques and tangles in the cerebral cortex, hippocampus, and amygdala regions, along with early synaptic dysfunction and cognitive decline, thus constituting a widely used and validated AD model. The results indicated that Aβ1-42 fragment (0.5 μM; 24h) treatment induced a reduction of cell viability and an increase in glutamate levels in cultured cortical neurons and astrocytes from non-Tg mice, but not in those from the genetic model of AD. The Aβ-induced effects in non-Tg cell cultures were counteracted by a pretreatment with PEA (0.1 μM). Based on these findings, it has been hypothesized that exogenous Aβ treatment failed to induce deleterious effects in 3xTg-AD mice-derived cortical neurons, as these cells at 8days *in vitro* were already exposed to a quite high concentration of endogenous peptide fragment. In fact, Aβ levels were observed in these cell cultures after 6days *in vitro* (Vale et al., 2010), and we demonstrated that control cultured cortical neurons obtained from 3xTg-AD mice displayed morphological alterations similar to those observed in Aβ-exposed cultured cortical neurons obtained from non-Tg mice (Tomasini et al., 2015). However, the treatment with PEA prevented the effects of Aβ in cultured cortical neurons and astrocytes from non-Tg mice but failed to affect the morphological alterations and glutamate levels in 3xTg-AD mice-derived cell cultures. This suggests that the compound may be effective in the early AD or when Aβ is accumulating, thus initiating to damage the CNS. Later, Bronzuoli et al. (2018), by using a different *in vitro* protocol, demonstrated that PEA did not display toxic effects in both astrocytes and neurons from 3xTg-AD mice, at the tested concentrations (0.01, 0.1, and 1μM), but it promoted neuron viability and counteracted reactive astrogliosis in mature 3xTg-AD primary astrocytes.

In a further study, we evaluated whether astrocytes could participate in regulating the Aβ-induced neuronal damage, by using primary mouse astrocyte cell cultures and mixed astrocyte–neuron cultures (Beggiato et al., 2018). The results indicated that in the presence of astrocytes pre-exposed to Aβ1-42 fragment (0.5 μM; 24h), there was a reduction of neuronal viability, an increase in the number of neuronal apoptotic nuclei, a decrease in the number of MAP-2-positive neurons, and an increase in the number of neurite aggregations/100 μm as compared with control (i.e., untreated) astrocyte–neuron co-cultures. Taken together, these data indicate that astrocytes contribute to Aβ-induced neurotoxicity and neuroinflammation. Interestingly, these effects were not observed when neurons were cultured in the presence of astrocytes pre-exposed to PEA (0.1 μM), applied 1h before and maintained during Aβ treatment. Thus, it has been concluded that PEA, by blunting Aβ-induced astrocyte activation, improved neuronal survival in mouse astrocyte–neuron co-cultures.

Finally, other researchers investigated the possible anti-AD action of co-ultraPEALut, a co-ultramicronized formulation of PEA in combination with the vegetable flavonoid luteolin (Lut),

which demonstrated antioxidant properties. Previous studies, in fact, indicated that the association of these two molecules, in a fixed ratio of 10:1 in mass, induced a strong neuroprotective activity (Paterniti et al., 2013b). The exposure of human neuronal cells obtained by differentiating SH-SY5Y neuroblastoma cells to Aβ1-42 (1µM; 24h) induced a reduction of cell viability and neuroinflammatory responses. These effects were counteracted by the pre-treatment with co-ultraPEALut (reference concentrations: 27, 2.7, and 0.27µM OF PEA) for 2h (Paterniti et al., 2014). Similar results were obtained from an *ex vivo* organotypic model of AD. In particular, hippocampal slice cultures were prepared from mice at postnatal day 6, and after 21days of culturing, the slices were pre-treated with co-ultraPEALut and then incubated with Aβ1-42 fragment (1µM; 24h). Under these experimental conditions, the pre-treatment with co-ultraPEALut significantly reduced iNOS and GFAP expression, restored neuronal iNOS and brain-derived neurotrophic factor (BDNF), and reduced the apoptosis (Paterniti et al., 2014). In line with these data, co-ultraPEALut reduced the expression of mRNA codifying serum amyloid A (SAA) in oligodendrocyte precursor cells subjected to TNF-α treatment. The relevance of this finding is supported by the evidence that SAA immunoreactivity is found in axonal myelin sheaths of cortex in AD (Barbierato et al., 2017).

### *In Vivo* Studies

The promising *in vitro* results prompted the development of *in vivo* studies aimed at evaluating the neuroprotective properties of PEA in animal models of AD. Firstly, D'Agostino et al. (2012) tested PEA against the learning and memory dysfunctions induced in mice by the intracerebroventricular injection of Aβ25–35 peptide (9nmol). To this purpose, PEA was administered once a day (3–30mg/kg, s.c.), starting 3h after Aβ25–35, for 1 or 2 weeks, while water-maze, water-maze working memory, and novel object recognition tests were used to assess cognitive performances. The authors demonstrated that, depending on the dose, PEA reduced (10mg/kg of body weight) or prevented (30mg/kg of body weight) the cognitive impairments induced by Aβ25–35 peptide injection. In line with previous *in vitro* findings, the beneficial effects of PEA appear mediated by PPAR-α, as the compound failed to rescue memory deficits induced by Aβ25–35 peptide injection in PPAR-α null mice, and GW7647 (a synthetic PPAR-α agonist) mirrored the effects of PEA. These encouraging behavioral results were corroborated by the evidence that in the same animals used for cognitive tests, PEA reduced brain lipid peroxidation, protein nitrosylation, iNOS induction, and caspase3 activation (D'Agostino et al., 2012). Comparable results have been obtained following the intrahippocampal injection of Aβ1–42 combined with PEA treatment in adult male rats (Scuderi et al., 2014). Immunofluorescence analysis of the hippocampal CA3 area ipsilateral to the injection site revealed that injection of Aβ1–42 induced astrocyte activation, as demonstrated by the fact that these cells showed a stellate shape and multiple branched processes. An increased expression of GFAP and S100B mRNA and protein, as well as increased densities of S100B-positive astrocytes, was also observed. Finally, intrahippocampal injection of Aβ1–42 was also associated with an upregulation of inflammatory markers, such as iNOS, COX-2, IL-1β, and TNF-α in homogenates of hippocampi ipsilateral to the injection site. When PEA (10mg/kg of body weight) was intraperitoneally administered once a day for seven consecutive days, starting from the day of Aβ1–42 injection, it was able to partially or completely antagonize Aβ1–42-induced toxic effects. Again, the effects of PEA were prevented by the treatment with GW6471 (2mg/kg), thus demonstrating the involvement of PPAR-α. Moreover, the authors demonstrated the PPAR-α-dependent ability of PEA to restore the alteration in the Wnt signaling pathway caused by Aβ1–42 hippocampal infusion. This is relevant since Wnt signaling pathway plays different roles in the development of neuronal circuits and also in the adult brain, where it regulates synaptic transmission and plasticity and has been also implicated in various diseases including neurodegenerative diseases (Inestrosa and Varela-Nallar, 2014). Finally, PEA reduced phosphorylated tau protein overexpression and rescued cognitive functioning, further strengthening the potential properties of the compound as a therapeutic agent in AD (Scuderi et al., 2014).

More recently, *in vivo* studies demonstrated that PEA displays beneficial effects also in a genetic model of the pathology. In a first study, the effects of chronic administration (3-month treatment) of µm‐PEA in 3xTg-AD mice, at two different stages of the pathology, were evaluated by administering the compound *via* a subcutaneous delivery system to groups of 3-month-old and 9-month-old animals (Scuderi et al., 2018). The animals were then tested at the end of the 3-month treatment and thereafter at the age of 6months (i.e., early-symptomatic stage) and 12months (i.e., clearly symptomatic stage), respectively. A battery of cognitive and non-cognitive tasks, followed by biochemical assessments of neuropathology, have been performed. Under these experimental conditions, µm‐PEA rescued cognitive functions in 6-month-old 3xTg-AD mice as evaluated by means of novel object recognition test (short- and long-term memory), inhibitory passive avoidance task (contextual learning and memory), and Morris water maze (spatial learning). In 12-month-old animals, µm‐PEA significantly improved the short-term memory in 3xTg-AD mice, with no significant effects on long-term memory. Furthermore, the compound did not exert significant effects on learning or memory in aged non-Tg mice. Interestingly, µm‐PEA also reduced depressive-like behaviors, measured by the tail suspension test and forced swim test, in early-symptomatic, but not in clearly symptomatic, 3xTg-AD mice, while it counteracted anhedonia-like phenotype of both young (6-month-old) and aging (12-month-old) 3xTg-AD mice. Overall, these findings indicate that µm‐PEA induces either beneficial cognitive or other non-cognitive effects that might be relevant to AD. Moreover, biochemical data also demonstrated that chronic µm‐PEA treatment reduced Aβ formation and phosphorylation of tau protein and promoted neuronal survival in the CA1 subregion of the hippocampus. These effects were associated with a normalization of the astrocytic function, a rebalancing of glutamatergic transmission, and a general reduction of neuroinflammatory conditions. The evidence that these biochemical/neurochemical effects were particularly manifest when the treatment was performed at a precocious stage of the pathology suggests the therapeutic potential of µm‐ PEA as an early treatment in AD.

In a recent study (Bronzuoli et al., 2018), the same treatment protocol as utilized in the above research was used to evaluate the effects of the chronic µm‐PEA treatment on reactive astrogliosis and neuronal function in the frontal cortex of 6-month-old 3xTg-AD mice, compared with their age-matched non-Tg littermates. Once again, µm‐PEA demonstrated beneficial effects in reducing pathology-related biochemical alterations in this animal model of AD. In fact, 3-month µm‐PEA treatment markedly reduced astrocytic activation in 3xTg-AD mice, as demonstrated by the decrease in GFAP mRNA and protein expression and the trend toward a decrease of S100B protein expression levels. Furthermore, chronic treatment reduced iNOS levels, slightly dampened the expression of Aβ, and increased the expression of BDNF in 3xTg-AD mice. Taken together, these findings indicate that early-symptomatic 3xTg-AD mice display signs of reactive gliosis in the frontal cortex and that the chronic µm‐PEA may counteract such phenomenon, also improving the trophic support to neurons, in the absence of astrocytes and neuronal toxicity.

## CLINICAL EVIDENCE

To the best of our knowledge, current clinical studies of PEA are mostly related to pain or peripheral inflammatory-related conditions, while there are very few studies aimed at evaluating the possible beneficial effects of PEA on CNS-related pathologies in human beings. This could be due to the fact that very little is known about the pharmacokinetics of PEA in humans (*please see* Pharmacokinetic). In fact, the bioavailability and apparent volume of distribution have not been clearly evaluated; and blood PEA levels, at least in animals, do not accurately reflect levels in the CNS (Davis et al., 2019). The micronized or ultramicronized forms of PEA increased bioavailability in animals compared with naïve forms, but there are very few and very recent clinical data to confirm that this is true for humans. Thus, while it seems likely that the new PEA formulations improve the compound bioavailability, complete pharmacokinetics data are urgently necessary to assess the precise tissue distribution and site of metabolism of PEA. These data will possibly allow to overcome the major difficulties in setting up clinical studies focused at evaluating the possible therapeutic role of PEA against CNS disorders.

In line with the above information, there are no clinical data concerning the possible beneficial effects of PEA in AD patients. However, Calabrò and colleagues (2016) in a case report described the case of a patient affected by mild cognitive impairment (MCI) who was treated for 9 months with highdose PEALut. As MCI may be symptomatic of normal aging or of a transition to early AD, the results of this observation are here reported. A 67-year-old patient presented, at the onset of the observational period, a mild memory impairment, as demonstrated by the specific neuropsychological assessment, including attentive matrices (AM), Babcock Story Recall Test (BSRT), Mini-Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), Rey Auditory Verbal Learning Test (RAVLT), Trail Making Test (TMT), and verbal fluency tests (VFTs). After a 3-month treatment with PEALut, the patient reported a non-significant cognitive amelioration, whereas her neuropsychological evaluation was almost normal after a 9-month treatment (significant improvement of RAVLT, AM, and TMT in comparison with those in the pre-treatment period).

To support the possible beneficial effect of PEA in neurodegenerative disorders, a study involving 30 Parkinson's disease patients receiving levodopa demonstrated that uµ-PEA (600mg for 1 year) slowed down disease progression and disability (Brotini et al., 2017).

## CONCLUSIONS

To summarize, preclinical either *in vitro* or *in vivo* data (**Tables 1** and **2**) strongly suggest that PEA, especially in its ultramicronized formulation, exerts quite robust therapeutic effects in several animal models of AD. In particular, published findings indicate that µm‐PEA treatment ameliorates both cognitive deficits and a range of neuropathological features of AD. A correlation between PEA anti-inflammatory, neuroprotective, neurobehavioral, and neurovascular effects might be suggested from the results in animal AD models, thus attributing to the compounds' unique properties, especially compared with those of classic anti-inflammatory drugs. Despite the obvious limits of the mentioned preclinical studies and by avoiding any simplistic extrapolation of data from the animal model to the human condition, the results of these intensive preclinical experiments propose µm‐PEA as a potential therapeutic agent, which could have an impact on the progression of AD, especially when the pathology is at an early stage. This hypothesis is also supported by studies demonstrating the PEA treatment efficacy in ameliorating the symptomatology of other neurodegenerative conditions such as Parkinson's disease (Esposito et al., 2012; Avagliano et al., 2016; Brotini et al., 2017; Crupi et al., 2018) and multiple sclerosis (Loría et al., 2008; Orefice et al., 2016).

Based on the available results and for translational purposes, it becomes now urgent to evaluate the possible beneficial effects of orally administered µm‐PEA in animal models of AD. In the event of positive results, these studies would help to rapidly define

TABLE 1 | Summary of *in vitro* preclinical studies supporting the role of palmitoylethanolamide (PEA) as a possible therapeutic agent in Alzheimer's disease (AD).


*Aβ1–42, β amyloid 1–42 peptide; BDNF, brain-derived neurotrophic factor; co-ultraPEALut, ultramicronized formulation of PEA/luteolin combination; GFAP, glial fibrillary acidic protein; iNOS, inducible nitric oxide synthase; MAP-2, microtubule-associated protein 2; PPARα, peroxisome proliferator-activated receptor-alpha.*


TABLE 2 | Summary of the available *in vivo* preclinical studies supporting the role of PEA as a possible therapeutic agent in AD.

*Aβ1–42 = β amyloid 1–42 peptide; Aβ25–35 = β amyloid 25–35 peptide; BDNF, brain-derived neurotrophic factor; µm-PEA, ultramicronized PEA formulation; PPARα, peroxisome proliferator-activated receptor-alpha.*

adaptive clinical trials and will hopefully allow to speed up the development of an innovative therapy for AD. In this context, it is worth noting that PEA-containing products (Normast®, Glialia®, Nevamast®, Adolene®, Visimast®, Mastocol®, and Pelvilen®) are actually used for certain medical indications, especially inflammatory pain. Moreover, as an endogenous compound, PEA has a safely profile at pharmacological doses. Relevant PEAinduced side effects were not seen in humans at oral doses up to 1,800 mg/day. Finally, PEA has proven efficacious in humans in a number of clinical settings, and none of the clinical trials with PEA to date have reported treatment-related adverse events (Skaper et al., 2018).

## REFERENCES


## AUTHOR CONTRIBUTIONS

LF conceptualized the review content, wrote part of the manuscript, and contributed with funding acquisition. SB conceptualized the review content and wrote part of the manuscript. MT wrote part of the manuscript and edited the final version.

## FUNDING

This work has been supported by a grant from Alzheimer's Drug Discovery Foundation (ADDF) to LF (grant # 20151001) and from the University of Ferrara (FAR 2018).


PPAR-α, PPAR-γ and TRPV1, but not CB1 or CB2 receptors. *Br. J. Pharmacol.* 166 (4), 1474–89. doi: 10.1111/j.1476-5381.2012.01889.x


biochemical and structural damage in experimental models of Alzheimer's disease. *Brain Res.* 1642, 397–408. doi: 10.1016/j.brainres.2016.04.022


inflammation. *Neurogastroenterol. Motil.* 27 (10), 1432–1445. doi: 10.1111/ nmo.12639


presymptomatic phase? *J. Alzheimers Dis.* 38 (4), 719–740. doi: 10.3233/ JAD-130920


**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 © 2019 Beggiato, Tomasini and Ferraro. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Pre-plaque Aß-Mediated Impairment of Synaptic Depotentiation in a Transgenic Rat Model of Alzheimer's Disease Amyloidosis

Yingjie Qi<sup>1</sup> , Igor Klyubin<sup>1</sup> , Neng-Wei Hu1,2, Tomas Ondrejcak<sup>1</sup> and Michael J. Rowan<sup>1</sup> \*

<sup>1</sup> Department of Pharmacology & Therapeutics, Institute of Neuroscience, Trinity College Dublin, Dublin, Ireland, <sup>2</sup> Department of Physiology and Neurobiology, Zhengzhou University School of Medicine, Zhengzhou, China

#### Edited by:

Cesare Mancuso, Catholic University of the Sacred Heart, Italy

#### Reviewed by:

Thomas Behnisch, Fudan University, China Satoshi Fujii, Yamagata University, Japan Graham L. Collingridge, University of Toronto, Canada

> \*Correspondence: Michael J. Rowan mrowan@tcd.ie

#### Specialty section:

This article was submitted to Neuropharmacology, a section of the journal Frontiers in Neuroscience

Received: 17 May 2019 Accepted: 31 July 2019 Published: 14 August 2019

#### Citation:

Qi Y, Klyubin I, Hu N-W, Ondrejcak T and Rowan MJ (2019) Pre-plaque Aß-Mediated Impairment of Synaptic Depotentiation in a Transgenic Rat Model of Alzheimer's Disease Amyloidosis. Front. Neurosci. 13:861. doi: 10.3389/fnins.2019.00861 How endogenously produced soluble amyloid ß-protein (Aß) affects synaptic plasticity in vulnerable circuits should provide insight into early Alzheimer's disease pathophysiology. McGill-R-Thy1-APP transgenic rats, modeling Alzheimer's disease amyloidosis, exhibit an age-dependent soluble Aß-mediated impairment of the induction of long-term potentiation (LTP) by 200 Hz conditioning stimulation at apical CA3-to-CA1 synapses. Here, we investigated if synaptic weakening at these synapses in the form of activitydependent persistent reversal (depotentiation) of LTP is also altered in pre-plaque rats in vivo. In freely behaving transgenic rats strong, 400 Hz, conditioning stimulation induced stable LTP that was NMDA receptor- and voltage-gated Ca2<sup>+</sup> channeldependent. Surprisingly, the ability of novelty exploration to induce depotentiation of 400 Hz-induced LTP was impaired in an Aß-dependent manner in the freely behaving transgenic rats. Moreover, at apical synapses, low frequency conditioning stimulation (1 Hz) did not trigger depotentiation in anaesthetized transgenic rats, with an agedependence similar to the LTP deficit. In contrast, at basal synapses neither LTP, induced by 100 or 200 Hz, nor novelty exploration-induced depotentiation was impaired in the freely behaving transgenic rats. These findings indicate that activity-dependent weakening, as well as strengthening, is impaired in a synapse- and age-dependent manner in this model of early Alzheimer's disease amyloidosis.

Keywords: soluble amyloid beta, synaptic plasticity, hippocampus, novelty exploration, depotentiation, apical dendrites, basal dendrites, Alzheimer's disease

## INTRODUCTION

There is great interest in understanding how different forms of synaptic plasticity contribute to normal brain function and how disruption of these physiological processes may underlie key aspects of the pathophysiology of neurodegenerative diseases. Although much research has elucidated the physiological significance of long-term potentiation (LTP) in memory mechanisms and have implicated impairments of LTP in cognitive impairment, in particular in Alzheimer's disease (AD) models [for review see Spires-Jones and Knafo (2012), Puzzo et al. (2015)], much less is known regarding activity-dependent persistent LTP reversal (depotentiation).

Usually, LTP decay/reversal is an active process and mechanisms underlying depotentiation are believed to play important roles in the time- and state- dependent erasure of certain forms of memory. Behaviourally induced depotentiation at CA3-to-CA1 hippocampal synapses occurs within a specific time-window when an animal acquires new information (Xu et al., 1998; Manahan-Vaughan and Braunewell, 1999; Abraham et al., 2002; Straube et al., 2003; Collingridge et al., 2010; Qi et al., 2013). Similar depotentiation can be induced with low frequency electrical conditioning stimulation both in vivo and in vitro (Staubli and Lynch, 1990; Fujii et al., 1991; Bashir and Collingridge, 1994; Migues et al., 2016).

Exogenous application of certain soluble aggregates of the AD protein amyloid ß (Aß), in particular certain soluble aggregates/oligomers of Aß (Aßo), potently inhibit LTP in wildtype (WT) rodents (Cullen et al., 1997; Lambert et al., 1998; Hu et al., 2008; Klyubin et al., 2014). Recently, we reported that prior to the deposition of fibrillar Aß in plaques in APP transgenic (TG) rats, endogenously produced Aß causes an age-dependent disruption of LTP induced by 200 Hz at CA1 apical synapses, the deficit being transiently rescued by subacute administration of agents that lower soluble Aß (Qi et al., 2014). An accumulation of Aß oligomers accompanies the impairment of NMDA receptordependent LTP (Qi et al., 2014; Zhang D. et al., 2017) and selective reduction in NMDA, but not AMPA, receptor-mediated baseline synaptic transmission (Qi et al., 2014). The LTP deficit in pre-plaque transgenic rats appears to be mediated by an agedependent pro-inflammatory milieu in the hippocampus (Leon et al., 2010; Hanzel et al., 2014; Iulita et al., 2014; Qi et al., 2018) driven by Aß oligomer binding to cellular prion protein and glutamate acting at metabotropic glutamate receptor 5 (Zhang D. et al., 2017). Somewhat similarly, an early intracellular buildup of Aß oligomers correlates with impairment of hippocampusdependent memory (Leon et al., 2010; Hanzel et al., 2014; Iulita et al., 2014) in the absence of observable synaptic structural change (Martino Adami et al., 2017).

Whether or not other forms of synaptic plasticity such as depotentiation are also affected in an age-dependent manner is unknown. Based on our previous finding that exogenous application on an Aß-containing APP fragment induced depotentiation in a narrow time window (Kim et al., 2001) and Aß facilitates low frequency stimulation-induced long-term depression (LTD) in WT rats (Li et al., 2009; Hu et al., 2014), we predicted that the induction of depotentiation would be facilitated in TG rats. Furthermore, because the acute disruption of synaptic plasticity by exogenously applied Aßo is synapseselective, with preferential vulnerability of apical over basal synapses to LTP inhibition (Hu et al., 2009), we expected that the disruption of LTP and depotentiation would be selective to apical, as opposed to basal, synapses in TG rats.

Consistent with a key role for endogenous Aßo in mediating LTP inhibition at apical synapses, an antibody that preferentially binds soluble aggregates of Aß over monomer reversed the deficit in freely behaving TG animals. Using a strong 400 Hz conditioning protocol we induced robust LTP, thereby allowing us to study depotentiation in these rats. To our surprise, we found that depotentiation was impaired both in freely behaving and anaesthetized TG animals. Interestingly, endogenous Aß-mediated inhibition of both LTP and depotentiation was restricted to apical synapses, with neither LTP nor depotentiation in TG rats being significantly altered at basal synapses. These findings indicate that in addition to synapse-selective deficits in LTP induction, the synaptic plasticity mechanisms for timedependent weakening of previously strengthened synapses are also disrupted by Aßo in early pre-plaque AD amyloidosis.

## MATERIALS AND METHODS

## Animals

Male TG rats (2.5–6 months old) expressing human APP751 with Swedish and Indiana mutations under the control of the murine Thy1.2 promoter (McGill-R-Thy1-APP) (Leon et al., 2010) and their age-matched WT littermates were genotyped commercially by Transnetyx (Cordova, TN, United States) using real time PCR. All experiments were carried out in accordance with the approval of the Health Products Regulatory Authority, Ireland, using methods similar to those described previously (Qi et al., 2014). Animals had free access to food and water and a 12-h lights on/off cycle.

## In vivo Surgery and Electrophysiology

For non-recovery experiments the rats were anaesthetized with urethane (1.5 g/kg, i.p.) and core body temperature was maintained at 37.5 ± 0.5◦C. For recovery experiments the implantation procedure was comparable but carried out under anaesthesia using a mixture of ketamine and xylazine (80 and 8 mg/kg, respectively, i.p.) according to methods similar to those described previously (Qi et al., 2013). For the recovery experiments the rats were allowed at least 14 days after surgery before recordings began. These rats were housed individually in their home cages post-surgery between recording sessions.

Recording electrodes (Teflon-coated tungsten wire; external diameter 75 µm bipolar or 112 µm monopolar) were positioned in the stratum radiatum of area CA1. Similar wire electrodes were placed either in the stratum radiatum or stratum oriens to selectively stimulate either apical or basal synapses, respectively. Screw electrodes located over the contralateral cortex were used as reference and earth. The stimulation and recording electrodes were optimally located using a combination of physiological and stereotactic indicators. Field excitatory post-synaptic potentials (EPSPs) were recorded in the stratum radiatum of the dorsal hippocampus in response to stimulation of the ipsilateral stratum radiatum (apical synapses) or stratum oriens (basal synapses) (**Figure 1**). The recording site was located 3.8 mm posterior to bregma and 2.5 mm lateral to midline, and the stimulating site was located 4.6 mm posterior to bregma and 3.8 mm lateral to midline. The final depths of the electrodes were adjusted to optimize the electrically evoked EPSP and confirmed by postmortem analysis. With the stimulation electrode in stratum oriens, the far-field EPSP from basal synapses was reversed in polarity because the recording electrode was located in the stratum radiatum (Leung et al., 2003).

Where necessary a stainless steel guide cannula (22 gauge, 0.7-mm outer diameter, length 13 mm) was implanted above the right lateral ventricle before the electrodes were implanted ipsilaterally. Injections were made via a Hamilton syringe which was connected to the internal cannula (28 gauge, 0.36 mm outer diameter). The injector was removed 1 min post-injection and a stainless steel plug was inserted. The position of the cannula was verified post-mortem by investigating the spread of ink dye after i.c.v. injection.

Test stimuli were delivered to the Schaffer-collateral/ commissural pathway every 30 s to evoke field EPSPs that were 45–60% maximum amplitude. To induce potentiation the following high frequency stimulation (HFS) protocols were used: w100 Hz, consisting of a single series of 10 trains of 10 stimuli at test pulse intensity with an inter-train interval of 2 s; 100 Hz, consisting of a single series of 10 trains of 20 stimuli at test pulse intensity with an inter-train interval of 2 s; 200 Hz, consisting of a single series of 10 trains of 20 stimuli at test pulse intensity with an inter-train interval of 2 s; s200 Hz HFS, consisting of a single series of 10 trains of 20 stimuli at high intensity (75% maximum) with an inter-train interval of 2 s; 400 Hz, consisting of a single series of 10 trains of 20 stimuli at test pulse intensity with an inter-train interval of 2 s; s400 Hz, consisting of a single series of 10 trains of 20 stimuli at high intensity (75% maximum) with an inter-train interval of 2 s; and 3 sets of 10 trains of 20 high intensity (75% maximum) pulses at 400 Hz with an inter-train interval of 2 s and an inter-set interval of 5 min (3 × s400 Hz). To induce depotentiation with electrical low frequency stimulation (LFS), 900 very high intensity pulses (95% maximum) were applied at 1 Hz.

Hippocampal electroencephalogram (EEG) was monitored between recordings of the evoked EPSPs from the same electrodes as described previously (Qi et al., 2013). The power (mV · ms) frequency spectrum of theta EEG in the 6–8 Hz theta band was calculated using the modulus of the amplitude (PowerLab Chart version 7 for Windows, ADInstruments Ltd., Oxford, United Kingdom).

Recovery animal experiments were carried out in a welllit room. The recording compartment consisted of the base of the home cage, including normal bedding and food/water, but the sides were replaced with a translucent Perspex plastic box (27 × 22 × 30 cm) with an open roof. The rats had access to food and water throughout the whole recording session from the same position as in the home cage. All animals were first habituated to the recording procedure over the post-surgery recovery period.

## Novelty Exploration

The novelty exploration protocol used to trigger depotentiation was similar to that described previously (Qi et al., 2013). Briefly, novelty exposure was begun by placing one small elastic ball very gently near the nose of the animal. Once the attention of the animal was drawn to the ball, then the ball was placed on the floor out of immediate reach of the animal. If the animal did not move to explore the ball, the attention-drawing procedure was repeated until the animal moved actively to explore it. Three minutes after the ball was placed, another small object was introduced near the animal using the same attention-drawing method. In order to encourage the rats to continue undisturbed exploration, after four objects, including the ball, had been explored for 12 min in total, clean dark blue tissue paper was inserted gently, totally covering the four walls of the recording compartment and the whole floor including the animal and the objects, leaving animal partly hidden from direct view for another 18 min. The tissue paper and objects were gently removed from the recording compartment at 30 min. In pilot experiments we confirmed (Qi et al., 2013) that such novelty exploration does not persistently affect baseline transmission in WT rats (n = 2).

## Drugs

The monoclonal antibody 3A1 was generated by Dr. Brian O'Nuallain against dityrosine cross-linked Aß1–40 with no detectible binding to APP and an ∼700 fold preference for soluble cross-linked Aß aggregates over Aß monomers in Capture/Sandwich ELISA (Frost et al., 2017), and mouse IgG1 isotype control antibody (Biolegend, United Kingdom) were administered in 5 i.c.v. injections (20 µg in 5 µl per injection) over 3 days with the last injection 2 h prior to HFS. We chose this regimen because we found a similar protocol was effective for other anti-Aß strategies in this model. A 20 µg dose of 3A1 was selected because a 10 µg treatment regimen did not reverse the LTP deficit in pilot experiments (n = 2, data not shown). Mibefradil (50 nmol in 5 µl i.c.v., Sigma) and (R)-3-(2 carboxypiperazin-4-yl) propyl-1-phosphonic acid (CPP, 7 mg/kg, i.p., Ascent Scientific) were dissolved in distilled water and administered 30 min and 2 h prior to HFS, respectively. The doses were chosen based on their ability to inhibit LTP with different HFS protocols in WT rats (Doyle et al., 1996; Ryan et al., 2010).

## Data Analysis

fnins-13-00861 August 12, 2019 Time: 16:42 # 4

Unless otherwise stated, the magnitude of potentiation and the power of 6–8 Hz EEG frequency are measured as a percentage of the baseline recordings made during the initial 30-min period, and expressed as the mean ± standard error of the mean. For statistical analysis, EPSP amplitudes were grouped into 10-min epochs. We used standard one-way ANOVA to compare the level of potentiation between multiple groups and one-way ANOVA with repeated measures to compare multiple times within groups. Two-way ANOVA with repeated measures was used to analyse the EEG. A significant overall ANOVA was followed by post hoc Bonferroni-corrected t-tests. Paired and unpaired Student's t-tests were used to compare potentiation within one group and between two groups, respectively. A P < 0.05 was considered statistically significant.

## RESULTS

## Targeting Aß Oligomers Reverses the LTP Deficit in Freely Behaving TG Rats

Previously we reported that an antibody that recognizes all conformations of Aß, including monomers, soluble aggregates and fibrils, reversed the LTP deficit in TG rats (Qi et al., 2014). Here, we directly examined the involvement of soluble Aß aggregates in mediating the inhibition of LTP at apical synapses between CA3 and CA1 pyramidal cells in 4–6-monthold, pre-plaque, TG rats using the conformation-selective anti-Aß monoclonal antibody 3A1 (Frost et al., 2017). We followed the same 3-day antibody injection protocol that we previously employed to study the Aß-dependence of impaired LTP in these animals (Qi et al., 2014) whereby TG animals received i.c.v. injections of 3A1 or an isotype control antibody IgG1 (5 × 20 µg in 5 µl). Consistent with published findings (Qi et al., 2014), whereas our standard, "200 Hz" induction protocol, consisting of a single set of 200 Hz trains at test pulse intensity, triggered LTP in vehicle-injected freely behaving 4–6-month-old WT rats (3 h post-HFS, 120.6 ± 3.9%, p = 0.02 compared with pre-HFS baseline, n = 4), the same protocol failed to induce LTP in their TG littermates that had received the control antibody (94.9 ± 2.7%, p = 0.42 compared with pre-HFS baseline, n = 5, **Figure 2A**). Unlike the isotype control-treated TG rats, repeated treatment with the conformation-selective anti-Aß antibody 3A1 (5 × 20 µg in 5 µl) reversed the LTP deficit in the TG rats (114.8 ± 2.8%, n = 6, p = 0.009 compared with baseline, p = 0.0007 compared with TG animals injected with IgG1, p = 0.25 compared with WT littermates, **Figure 2A**). When these TG animals were followed longitudinally (**Figures 2B,C**) it was clear that the recovery of the ability to induce LTP by 3A1 was transient. Thus, LTP was strongly inhibited in these rats when tested again, a week after ceasing treatment with 3A1 (p = 0.385, compared with pre-treated animals, p = 0.0028, compared with animals immediately after treatment, n = 5, **Figure 2B**). In TG animals treated with the control antibody LTP was inhibited at all time points over the same period (**Figure 2C**). These findings provide convincing evidence of a requirement for Aßo in

FIGURE 2 | An aggregate conformation-selective anti-Aß antibody, 3A1, transiently rescues the LTP deficit in freely behaving transgenic rats. (A) In a cross-sectional analysis, repeated i.c.v. treatment of 4–6-month-old TG rats with 3A1 (TG 3A1) but not isotype control antibody IgG1 (TG IgG1) restored LTP to a level indistinguishable from WT littermates injected with vehicle (WT veh). Left hand panel shows LTP time course. Inserts show representative EPSP traces at the times indicated. Calibration bars: vertical, 1 mV; horizontal, 10 ms. Summary bar chart of LTP (last 10 min post-HFS) is in the right-hand panel. (B,C) A longitudinal analysis revealed the transient nature of the recovery of LTP in TG animals injected with 3A1. The LTP time course from the same TG animals the week before (TG pre), immediately after (TG 3A1 or TG IgG1), and 1 week after (TG post) treatment is shown in the left panels. Right hand panels show longitudinal data from individual rats. Arrows indicate the time point of application of a single set of 200 Hz HFS at test pulse intensity (200 Hz). The # symbol stands for a statistical comparison between pre- and 3 h post-HFS values within one group (paired t-test) whereas an <sup>∗</sup> indicates a comparison of 3 h post-HFS values between groups (one-way ANOVA (A) or one-way ANOVA with repeated measures (B,C) followed by post hoc Bonferroni test). One symbol, p < 0.05; two symbols, p < 0.01; three symbols, p < 0.001; ns, p > 0.05. Values are mean ± S.E.M.% pre-HFS baseline EPSP amplitude.

mediating LTP inhibition in the TG rats and are consistent with our previous report of an age-dependent accumulation of Aßo in the brains of these animals (Zhang D. et al., 2017).

## Strong High Frequency Stimulation Is Required to Induce Robust LTP in Freely Behaving TG Rats

In order to study depotentiation in TG rats, we needed to generate similar LTP to that induced in WT animals. Previously, we found that increasing the strength of the HFS protocol, using three sets of trains of high intensity pulses at 400 Hz, overcomes the LTP deficit (Qi et al., 2014). We questioned if such a strong protocol was necessary, or if other, intermediate strength, protocols might be sufficient to induce robust LTP in these TG rats. Therefore, we assessed the HFS-dependence of the LTP inhibition in TG rats further by increasing the frequency and intensity. Like 200 Hz HFS, a single train of 400 Hz tetanization at test pulse intensity triggered LTP in 4–6-month-old WT rats (118.6 ± 2.0%, p < 0.0001 compared with pre-HFS baseline, n = 8) and not significant potentiation in TG littermates (110.6 ± 3.9%, p = 0.093, compared with pre-HFS baseline, n = 8, **Figure 3A**). Somewhat similar results were obtained when the same protocol was applied but with high intensity pulses during the tetanus (75% maximum, s400 Hz). However, in this case a significant LTP was induced in both WT (125.3 ± 3.6%, p = 0.004 compared with pre-HFS baseline, n = 5) and TG rats (114.1 ± 4.5%, p = 0.019, compared with pre-HFS baseline, n = 7, **Figure 3B**). In contrast, and consistent with our previous observations (Qi et al., 2014), three sets of 400 Hz at high intensity (3 × s400 Hz) induced large and stable LTP in 4–6-month-old TG rats (147.8 ± 11.0%, p = 0.007, compared with pre-HFS baseline, n = 6) and their WT littermates (153.8 ± 11.5%, p = 0.004 compared with pre-HFS baseline, n = 6, **Figure 3C**).

We wondered if the mechanism underlying the induction of LTP by 3 × s400 Hz in TG rats was similar to what we had found previously in WT rats which required activation of voltage-gated voltage-gated Ca2<sup>+</sup> channels (VGCCs) in addition to NMDA receptors (Ryan et al., 2010). Indeed, pretreatment of TG rats with the NMDA receptors antagonist CPP (7 mg/kg, i.p.) alone only partly reduced the magnitude of LTP (119.0 ± 5.8%, p = 0.019, compared with pre-HFS baseline, n = 5, p = 0.009, compared with vehicle-injected TG, 149.3 ± 6.6%, n = 5, **Figure 4A**). In contrast, LTP was completely blocked when CPP (7 mg/kg, i.p.) and the VGCC inhibitor mibefradil (50 nmol in 5 µl, i.c.v.) were administered in combination (99.4 ± 5.0%, p = 0.48, compared with pre-HFS baseline, n = 5, **Figure 4B**). We also confirmed that this combination completely inhibited LTP induced by the 3 × s400 Hz protocol in WT littermates (99.8 ± 4.2%, p = 0.43, compared with pre-HFS baseline, n = 5, not illustrated).

These data indicate that in order to induce robust LTP in TG rats it was necessary to use a strong HFS protocol that engages VGCCs in addition to the NMDA receptors, in contrast to LTP induced by our standard 200 Hz protocol in WT rats which only requires NMDA receptors (Hu et al., 2008).

FIGURE 3 | Tetanus strength-dependent potentiation in freely behaving TG rats. To develop a conditioning protocol that would induce robust LTP in TG rats comparable to that found in WT littermates a range of HFS paradigms were applied: (A) a single set of 400 Hz trains at 50% of maximal EPSP amplitude (400 Hz); (B) a single strong set of 400 Hz trains at 75% of maximal EPSP amplitude (s400 Hz); (C) three strong sets of 400 Hz trains at 75% of maximal EPSP amplitude (3 × s400 Hz). Left-hand panels show LTP time course. Inserts show representative EPSP traces at the times indicated. Calibration bars: vertical, 1 mV; horizontal, 10 ms. The time point of HFS application is indicated by arrow, arrow head and three arrow heads, respectively. Summary bar charts of LTP (last 10 min post-HFS) are in right hand panel. The # symbol stands for a statistical comparison between preand 3 h post-HFS values within one group (paired t-test) whereas an "ns" above the line indicates a comparison of 3 h post-HFS values between groups (unpaired t-test). One symbol, p < 0.05; two symbols, p < 0.01; four symbols, p < 0.0001; ns, p > 0.05. Values are mean ± S.E.M.% pre-HFS baseline EPSP amplitude.

## Novelty Exploration-Induced Depotentiation Is Impaired in an Aßo-Dependent Manner in Freely Behaving TG Rats

fnins-13-00861 August 12, 2019 Time: 16:42 # 6

Just like memories, LTP is transiently susceptible to active erasure due to the ability of additional experience or electrical LFS to reverse this form of synaptic potentiation (Kim et al., 2007; Clem and Huganir, 2010; Diaz-Mataix et al., 2011). Given the potential importance of such activity-dependent persistent reversal of previously established synaptic LTP in brain function and our previous findings with exogenous Aß-containing APP fragments (Kim et al., 2001), we wondered if behaviourally or electrically induced depotentiation is also disrupted in TG rats. Prolonged exploration of a non-aversive novel environment can trigger a rapid depotentiation of LTP at CA3-CA1 synapses that is prevented by prior acquisition of information about the new environment (Xu et al., 1998; Manahan-Vaughan and Braunewell, 1999). To compare the ability of novelty exploration to instigate depotentiation in 4–6-month-old WT and TG rats, first we applied the 3 × s400 Hz protocol to induce robust LTP (**Figures 5A,B**). One h after the induction of LTP the animals were allowed to actively explore novel objects that were placed in the recording box for 30 min. Whereas novelty exploration in the WT rats strongly reversed the previously established LTP (3 h post-HFS, 116.2 ± 5.3%, p = 0.006, compared with a pre-novelty potentiation, 157.3 ± 10.5%, p > 0.05, compared with pre-HFS baseline, n = 5), similar novelty was much less effective in TG rats (3 h post-HFS, 149.9 ± 10.7%, p = 0.29, compared with prenovelty potentiation, 145.3 ± 9.6%, p = 0.006, compared with pre-HFS baseline, n = 6). In contrast, brief novelty exploration was effective in triggering depotentiation in TG rats that had received repeated i.c.v. injections of the anti-Aßo antibody 3A1 (5 × 20 µg in 5 µl over 3 days) (117.0 ± 3.8%, n = 5, p = 0.005, compared with pre-novelty potentiation, 161.2 ± 7.3%, p = 0.008, compared with pre-HFS baseline, p = 0.032 compared with TG animals, p = 0.90 compared with WT littermates).

Consistent with a widespread activation of the hippocampus during novelty exploration, theta power of the local EEG was increased, particularly during the first 20 min (**Figure 5C**) and there was no significant difference in the magnitude of the increase in theta power between WT and TG rats (p = 0.33, for the group × time interaction, two-way ANOVA with repeated measures). This indicates that the extent of hippocampal engagement was similar in both groups and therefore unlikely to underlie the difference in the magnitude of depotentiation induced in the two groups. Moreover, similar to our previous finding in WT animals (Qi et al., 2013), novelty exploration did not affect baseline synaptic transmission in TG rats (99.4 ± 3.0%, p = 0.304, compared with pre-novelty values, n = 4, **Figure 5D**).

## Age-Dependent Inhibition of Electrically Induced Depotentiation in Anaesthetized TG Rats

In addition to novelty exploration, the application of electrical LFS triggers an NMDA receptor-dependent depotentiation

in vivo that can be studied under anaesthesia (Doyle et al., 1997). This enabled us to evaluate depotentiation under conditions where possible confounding effects of behavioral phenotype are unlikely to affect the outcome.

FIGURE 5 | A deficit in novelty exploration-induced depotentiation in freely behaving 4–6-month-old TG rats is reversed by the conformation-selective anti-Aß antibody 3A1. (A) 1 h after LTP induction by 3 × s400 Hz the rats were allowed to continuously explore a novel environment for 30 min (Novelty, solid bar). Whereas novelty exploration failed to reverse LTP in TG rats, depotentiation was induced in WT littermates and TG rats previously treated for 3 days with the monoclonal antibody 3A1 (TG 3A1). Inserts show representative EPSP traces at the times indicated. Calibration bars: vertical, 1 mV; horizontal, 10 ms. (B) Summary bar chart of the magnitude of potentiation both pre-novelty (50–60 min post-HFS epoch) and post-novelty (170–180 min post-HFS epoch). The # symbol stands for a statistical comparison with pre-HFS values within one group and the & symbol stands for a statistical comparison with pre-novelty values within one group (one-way ANOVA with repeated measures followed by post hoc Bonferroni test). An <sup>∗</sup> indicates a comparison of post-novelty values between groups (one-way ANOVA with post hoc Bonferroni test). One symbol, p < 0.05; two symbols, p < 0.01; ns, p > 0.05. The time point of HFS application is indicated by three arrow heads. (C) Increased theta band EEG power (6–8 Hz frequency range) during novelty exploration in both WT and TG rats. Top panels show representative examples of EEG band-pass filtered at 5–15 Hz. Calibration bars: vertical, 0.1 mV; horizontal, 1 s. Bottom panel shows time course of EEG changes. (D) No discernable change in baseline synaptic transmission after novelty exploration (Novelty, solid bar) in TG rats. Insert shows representative EPSP traces at the times indicated. Calibration bars: vertical, 1 mV; horizontal, 10 ms. Values are mean ± S.E.M.% either pre-HFS baseline EPSP amplitude (A,B,D) or theta power (C).

First, we confirmed (Qi et al., 2014) that LTP induced by 200 Hz at high pulse intensity (75% maximum, s200 Hz) was inhibited in anaesthetized 4–6-month-old TG rats (1 h post-s200 Hz HFS, 123.6 ± 5.1%, n = 14, p = 0.009, compared with WT littermates, 143.1 ± 4.3%, n = 13, **Figure 6A**). Following this, we applied the 3 × s400 Hz HFS protocol, which induced

potentiation at the same three time points (s200 Hz, last 10 min post-s200 Hz HFS; 3 × s400 Hz, last 10 min post-3 × s400 Hz HFS; 1 Hz, last 10 min post-1 Hz LFS) are in the right-hand panels. The # symbol stands for a statistical comparison with pre-HFS values within one group and the & symbol stands for a statistical comparison with pre-LFS values within one group (one-way ANOVA with repeated measures followed by post hoc Bonferroni test). An <sup>∗</sup> indicates a comparison of potentiation values between groups (one-way ANOVA with post hoc Bonferroni test). One symbol, p < 0.05; two symbols, p < 0.01; three symbols, p < 0.001; ns, p > 0.05. Values are mean ± S.E.M.% pre-HFS baseline EPSP amplitude.

LTP that was similar in magnitude in both groups (1 h post-3xs400 Hz HFS, TG, 147.8 ± 7.6%, p = 0.47, compared with WT, 155.0 ± 6.3%). One h later we applied LFS consisting of 900 very high intensity pulses (95% maximum) at 1 Hz. As expected, LFS completely and persistently reversed LTP in WT rats (3.5 h post-s200 Hz HFS, 90.3 ± 10.1%, p = 0.23, compared with pres200 Hz HFS baseline). In contrast, the same 1 Hz protocol in the TG littermates only caused a transient reversal of LTP (3.5 h post-s200 Hz HFS, 131.1 ± 9.1%, p = 0.008, compared with pre-s200 Hz HFS baseline).

To determine the possible age-dependence of the deficit in depotentiation in the TG rats, we examined the efficacy of LFS to reverse LTP triggered by either the 3 × s400 Hz HFS or the s200 Hz protocol in 2.5–3.5-month-old TG rats, an age when there is no apparent LTP deficit (Qi et al., 2014). Thus, one h after the induction of LTP with the 3 × s400 Hz HFS protocol in the younger rats the application of LFS triggered a strong reversal of LTP (2.5 h post-3 × s400 Hz HFS, 111.5 ± 12.3%, p = 0.41 compared with pre-HFS baseline, n = 7, **Figure 6B**). Similar findings were observed for depotentiation after the s200 Hz protocol (2.5 h post-s200 Hz HFS alone, 96.4 ± 8.1%, p = 0.79 compared with pre-HFS baseline, n = 6, data not illustrated), indicating that the deficit in depotentiation is agedependent, with a time of onset similar to the impairment in LTP induction by the 200 Hz HFS protocol at these synapses (Qi et al., 2014).

## LTP and Depotentiation at Basal Synapses Are Resistant to Disruption in Freely Behaving TG Rats

Most research has focused on the disruptive effects of Aß on plasticity at apical synapses between CA3 and CA1 hippocampal pyramidal cells. At a circuit level, different CA1 pyramidal cells have different inputs and outputs and perform multiple tasks in parallel (Slomianka et al., 2011; Soltesz and Losonczy, 2018). This diversity is reflected in the expression of different receptors and types of plasticity at different input synapses (Roth and Leung, 1995; Colavita et al., 2016; Brzdak et al., 2019). Moreover, the susceptibility of LTP at the different synaptic inputs to disruption by Aß varies. Thus, we (Hu et al., 2009) and others (Zhao et al., 2018) found that NMDA receptor-dependent LTP at basal synapses, unlike apical synapses, is resistant to inhibition by exogenously applied Aß.

First, we compared the ability of the standard 200 Hz conditioning stimulation protocol to trigger LTP at basal synapses in 4–6-month-old WT and TG freely behaving rats, an age when LTP is inhibited at apical synapses. With the stimulation electrode in stratum oriens, the far-field EPSP from basal synapses was reversed in polarity because the recording electrode was located in the stratum radiatum (Leung et al., 2003; see **Figure 1**). Different from apical synapses, the application of HFS at basal synapses induced similar magnitude of LTP in both groups (WT, 152.6 ± 12.4, p = 0.004, compared with pre-HFS baseline, n = 6, TG, 156.9 ± 13.6, p = 0.017, compared with pre-HFS baseline, n = 5, WT versus TG, p = 0.82, **Figure 7A**). Similarly, there was no evidence of inhibition of LTP when we tested rats longitudinally between 3.5 and 6 months (p > 0.05 for all ages tested, **Figure 7D**). Because we were primarily interested in the effects of pre-fibrillar soluble Aß aggregates, we did not investigate animals older than 6 months, when Aß plaques start to be detectible in some TG rats (Leon et al., 2010; Hanzel et al., 2014; Iulita et al., 2014).

Amongst other differences, the threshold for LTP induction at basal synapses has been reported to be lower at basal

#### FIGURE 7 | Continued

pulse intensity consisting 20 pulses; w100 Hz, a weak single set of 100 Hz HFS at test pulse intensity consisting 10 pulses) in 4–6-month-old rats. Left hand panels show LTP time course. Inserts show representative EPSP traces at the times indicated. Calibration bars: vertical, 1 mV; horizontal, 10 ms. The time point of HFS application is indicated by an arrow. Summary bar charts of LTP (last 10 min post-HFS) are in the right-hand panel. (D) Summary longitudinal data for the magnitude of 200 Hz HFS induced LTP tracked repeatedly in the same WT and TG rats every 2nd week between 3.5 and 6 months of age. The # symbol stands for a statistical comparison between pre- and 3 h post- HFS values within one group (paired t-test) whereas an "ns" above the line in (A–C) and throughout (D) indicates a comparison of 3 h post-HFS values between groups (unpaired t-test). One symbol, p < 0.05; two symbols, p < 0.01; ns, p > 0.05. Values are mean ± S.E.M.% pre-HFS baseline EPSP amplitude.

compared with apical synapses (Leung et al., 2003). Because the LTP deficit at apical synapses was associated with an apparent increase in threshold, we decided to assess if the threshold for LTP was altered at basal synapses using weaker HFS protocols. When the frequency of the conditioning stimulation was reduced from 200 Hz to 100 Hz, similar magnitude LTP was induced both in 4–6-month-old TG and WT littermates (WT, 139.0 ± 13.5, p = 0.04, compared with pre-HFS baseline, n = 6, TG, 149.5 ± 14.7, p = 0.028, compared with pre-HFS baseline, n = 6, **Figure 7B**). Moreover, in order to determine if LTP might be facilitated in stratum oriens in the TG rats, we reduced the number of pulses per train from 20 to 10 (w100 Hz). This very weak HFS didn't induce LTP in either group (WT, 108.7 ± 10.6, p = 0.59, compared with pre-HFS baseline, n = 5, TG, 113.0 ± 6.2, p = 0.24, compared with pre-HFS baseline, n = 4, **Figure 7C**).

Finally, we wondered if, like LTP, novelty exploration-induced depotentiation at the basal synapses (Qi et al., 2013) is preserved in freely behaving 4–6-month-old TG rats. Consistent with previous studies (Hu et al., 2009; Qi et al., 2013), the initial post-HFS potentiation appears decremental. Nevertheless, as seen in **Figure 7A**, stable LTP was recorded in both WT and TG rats during the subsequent 2 h post-200 Hz HFS. In contrast, LTP reverted back to baseline when rats were allowed explore novel objects for 30 min, starting 1 h after inducing LTP with the 200 Hz protocol. Thus, the magnitude of depotentiation was similar in both groups (WT, 106.4 ± 6.5, p = 0.30, compared with pre-HFS baseline, n = 6, TG, 119.4 ± 9.8, p = 0.16, compared with pre-HFS baseline, n = 5, **Figure 8**).

## DISCUSSION

Here we provide evidence that, prior to plaque deposition, Aßo mediate an age-dependent inhibition of both LTP and depotentiation at apical synapses in the CA1 area of APP TG rats. To allow us to study depotentiation in these rats we used a strong protocol to induce an additional LTP that was blocked by combined treatment with an NMDA receptor antagonist and a VGCC inhibitor. Novelty exploration in freely behaving animals and electrical LFS under anaesthesia failed to trigger depotentiation at apical synapses at the pre-plaque

FIGURE 8 | Depotentiation is normal at basal dendrites in freely behaving TG animals. One hour after LTP induction by 200 Hz conditioning stimulation 4–6-month-old rats were allowed to continuously explore a novel environment for 30 min (Novelty, solid bar). Left hand panels show the time course of LTP and depotentiation. Inserts show representative EPSP traces at the times indicated. Calibration bars: vertical, 1 mV; horizontal, 10 ms. Summary bar chart of pre-novelty (pre, 50–60 min post-HFS epoch) and post-novelty (post, 170–180 min post-HFS epoch) potentiation is in right hand panel. The # symbol stands for a statistical comparison with pre-HFS values within one group and the & symbol stands for a statistical comparison with pre-novelty values within one group (one-way ANOVA with repeated measures followed by post hoc Bonferroni test). An "ns" above the line indicates a comparison of post-novelty values between groups (one-way ANOVA with post hoc Bonferroni test). One symbol, p < 0.05; two symbols, p < 0.01; ns, p > 0.05. The time point of HFS application is indicated by an arrow. Values are mean ± S.E.M.% either pre-HFS baseline EPSP amplitude.

stage. In contrast, neither LTP nor novelty exploration-induced depotentiation was altered at basal synapses in similarly aged TG rats. Thus, the age-dependent deficit in LTP and depotentiation is selective for apical synapses. This differential vulnerability of plasticity at apical and basal synapses strongly indicates a circuit-selective reduction in the dynamic range of synaptic gain and weakening.

The ability of repeated dosing with an Aßo-selective antibody, 3A1 (Frost et al., 2017) provides evidence that the age-dependent LTP deficit in pre-plaque TG rats is mediated by Aßo. This finding extends our previous reports that (a) soluble Aß is necessary for the LTP inhibition (Qi et al., 2014), and (b) there is an age-dependent increase in Aßo in the brain of TG rats starting around the time of the onset of the LTP deficit (Zhang D. et al., 2017). Similar to the beneficial action of the non-conformationselective anti-Aß antibody McSA1 (Qi et al., 2014), when followed longitudinally in individual rats, the LTP impairment re-emerged within 1 week of ceasing treatment with 3A1.

The inhibition of LTP induction by our standard 200 Hz conditioning protocol, which is NMDA receptor-dependent (Doyle et al., 1996; Hu et al., 2008), was hypothesized to be due to an increase in the threshold for LTP induction consequent to a reduction in NMDA receptor-mediated synaptic transmission in TG rats (Qi et al., 2014). Consistent with this proposal, in the present studies whereas intermediate strength protocols were ineffective, repeated high intensity 400 Hz HFS triggered robust LTP in the TG rats. This finding contrasts with our previous report that acute exogenously applied Aßo potently inhibited LTP induced by comparable 200 and 400 Hz conditioning protocols (Klyubin et al., 2014). Similar to WT rats (Doyle et al., 1996; Hu et al., 2008; Ryan et al., 2010), a combination of an NMDA receptor antagonist and VGCC blocker fully prevented LTP induction by repeated high intensity 400 Hz tetanus in TG rats. Since this protocol, unlike the weaker protocols, triggered similar magnitude LTP in both sets of animals, it is possible that VGCC-dependent LTP is relatively spared compared to NMDA receptor-dependent LTP in TG rats. Thus, the inhibition of LTP by endogenous Aßo in TG rats may depend on the source of the initial Ca2<sup>+</sup> entry trigger for plasticity induction. Future in vitro studies with saturating concentrations of selective antagonists will be required to evaluate this possibility.

Contrary to our predictions based on the synaptic weakeningpromoting acute effects of exogenously applied Aß (Li et al., 2009; Hu et al., 2014; O' Riordan et al., 2018) and Aß-containing APP fragments (Kim et al., 2001), depotentiation was strongly inhibited at apical synapses in TG rats. This was the case for both novelty exploration and LFS-induced depotentiation in freely behaving and anaesthetized 4–6-month-old TG rats, respectively. It appears that, in addition to LTP inhibition, Aßo mediate the inhibition of depotentiation in the TG rats since the monoclonal antibody 3A1 also reversed this deficit. The age-dependent increase in Aßo (Zhang D. et al., 2017) and similar age-dependence of the depotentiation and LTP deficits supports a similar role of Aßo in both deficits. In view of the known NMDA receptor-dependence of both novelty exploration- and electrical LFS-induced depotentiation (Doyle et al., 1996; Qi et al., 2013), it seems likely that the deficit in depotentiation in TG rats is, like that suggested for the LTP deficit, mediated by a reduction in NMDA receptor-mediated transmission (Qi et al., 2014).

In order to study depotentiation in TG animals, we needed to first apply the strong, repeated train of 400 Hz protocol to induce robust LTP. Although LTP has been reported to be resistant to activity-dependent reversal when induced by repeated stimulus trains in vitro (Woo and Nguyen, 2003), we found that high intensity LFS in vivo induced persistent reversal of LTP triggered either by our standard 200 Hz or the strong 400 Hz protocols in younger TG rats. Further studies are required to determine if depotentiation depends on the LTP induction and/or LTP reversal protocols in vivo. Based on available evidence, LTP may be susceptible to depotentiation over a longer time period in vivo (Doyle et al., 1997; Xu et al., 1998). Given our finding that exploration-induced depotentiation was associated with enhancement of theta power, it would be particularly interesting in future studies to examine if LFS with different frequencies in the theta range are differentially affected in TG rats.

In apparent contrast to the present findings, depotentiation has been reported to be normal in hippocampal slices from young pre-plaque APP TG (Tg2576) mice (Huh et al., 2016). In that study depotentiation induced either by a 5 Hz electrical stimulation protocol (applied 5 min after HFS) or the receptor

kinase ErbB4 ligand neuregulin 1, was impaired only in plaqueladen mice. The deficit in the older mice was attributed to damage to certain interneurons that express ErbB4. Whether or not similar changes are present in pre-plaque TG rats is not known. Moreover, because Aß may bind ErbB4 and its ablation prevents exogenously applied Aß-mediated inhibition of LTP (Zhang H. et al., 2017), future studies should further examine its role in depotentiation deficits at the pre-plaque stage of TG rats.

A corollary of the widely accepted theory that LTPlike persistent synaptic strengthening provides an essential component of memory formation is that depotentiation will promote memory erasure. Indeed, interventions, including novelty exploration, that relatively selectively induce depotentiation can trigger the erasure of newly formed memories or habits (Hayashi-Takagi et al., 2015; Medina, 2018; Ge et al., 2019). Investigating whether patients with early AD have a deficit in memory interference from novel information is a topic worth pursuing (Muecke et al., 2018; Thomas et al., 2018).

Our finding that LTP at basal, as opposed to apical, synapses appear to be unchanged in the pre-plaque TG rats is consistent with previous reports that exogenously applied Aßo fails to inhibit LTP in stratum oriens (Hu et al., 2009; Zhao et al., 2018). Recently, it has become clear that CA1 pyramidal neurons with cell bodies either near the stratum radiatum or stratum oriens generally form different networks, with the latter having much more extensive basal dendritic trees with a strong input from CA2 (Graves et al., 2012; Soltesz and Losonczy, 2018). The known different signaling pathways mediating LTP at these synapses (Roth and Leung, 1995; Colavita et al., 2016; Brzdak et al., 2019) and the finding that spines have high turnover rates in stratum oriens (Pfeiffer et al., 2018), may help explain the relative resistance of synaptic plasticity at basal synapses to disruption of both LTP and depotentiation. A significant but relatively poorly explored question for AD research is to understand why only certain pathways are affected early in the disease process (Fu et al., 2018). Understanding the mechanisms underlying the pathway selectivity of the plasticity disrupting action of endogenously

## REFERENCES


generated Aßo, as reported here, may help clarify the early pathophysiology of Alzheimer's disease.

## DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

## ETHICS STATEMENT

The animal study was reviewed and approved by Animal Research Ethics Committee, Trinity College Dublin and Health Products Regulatory Authority, Ireland.

## AUTHOR CONTRIBUTIONS

YQ and IK performed the experiments. IK and MR wrote the manuscript. All the authors contributed to study design, and read and approved the final manuscript.

## FUNDING

MR was supported by the Science Foundation Ireland (14/IA/2571) and the Irish Health Research Board (HRA-POR-2015-1102), and N-WH was supported by the National Natural Science Foundation of China (81471114).

## ACKNOWLEDGMENTS

We thank Prof. A. C. Cuello (McGill University, Montreal) for providing McGill-R-Thy1-APP breeding stock and Dr. Brian O'Nuallain (Verseau Therapeutics, Bedford, MA, United States) for providing the monoclonal antibody 3A1.


Proc. Natl. Acad. Sci. U.S.A. 96, 8739–8744. doi: 10.1073/pnas.96.15. 8739


fnins-13-00861 August 12, 2019 Time: 16:42 # 12

studies. Neuropharmacology 121, 231–246. doi: 10.1016/j.neuropharm.2017. 03.036


**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 © 2019 Qi, Klyubin, Hu, Ondrejcak and Rowan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Mitochondria as Potential Targets in Alzheimer Disease Therapy: An Update

### *Giovanna Cenini\* and Wolfgang Voos*

*Institut für Biochemie und Molekularbiologie, Rheinische Friedrich-Wilhelms-Universität Bonn, Bonn, Germany*

Alzheimer disease (AD) is a progressive and deleterious neurodegenerative disorder that affects mostly the elderly population. At the moment, no effective treatments are available in the market, making the whole situation a compelling challenge for societies worldwide. Recently, novel mechanisms have been proposed to explain the etiology of this disease leading to the new concept that AD is a multifactor pathology. Among others, the function of mitochondria has been considered as one of the intracellular processes severely compromised in AD since the early stages and likely represents a common feature of many neurodegenerative diseases. Many mitochondrial parameters decline already during the aging, reaching an extensive functional failure concomitant with the onset of neurodegenerative conditions, although the exact timeline of these events is still unclear. Thereby, it is not surprising that mitochondria have been already considered as therapeutic targets in neurodegenerative diseases including AD. Together with an overview of the role of mitochondrial dysfunction, this review examines the pros and cons of the tested therapeutic approaches targeting mitochondria in the context of AD. Since mitochondrial therapies in AD have shown different degrees of progress, it is imperative to perform a detailed analysis of the significance of mitochondrial deterioration in AD and of a pharmacological treatment at this level. This step would be very important for the field, as an effective drug treatment in AD is still missing and new therapeutic concepts are urgently needed.

Keywords: Alzheimer disease, therapeutic strategy, mitochondria, mitochondrial dysfunction, mitochondrial therapy

## INTRODUCTION

Alzheimer disease (AD) is a complex and heterogeneous disorder strongly affecting the cognitive functions and the memory of seniors.

Many risk factors were proposed to be significant contributors for the AD onset such as senescence, autophagy defects, genetic factors [i.e., ApolipoproteinaE-allele4 (APOE4), Triggering receptor expressed on myeloid cells 2 (Trem2)], microbiota alterations, lifestyle choices, cardiovascular and traumatic brain injury, as well as environmental factors (level of education, hypertension, obesity, diabetes, smoking, hearing loss, depression, physical inactivity, social isolation) (Livingston et al., 2017). It is now well accepted that important cellular pathways are compromised in AD. Together with intraneuronal neurofibrillary tangles (NFT) made of hyperphosphorylated tau protein and the extraneuronal senile plaques (SP) made of beta-amyloid (Aβ) peptides, synaptic failure, vascular

#### *Edited by:*

*Cesare Mancuso, Catholic University of the Sacred Heart, Italy*

#### *Reviewed by:*

*Amandine Grimm, University of Basel, Switzerland Cristina Carvalho, University of Coimbra, Portugal Sónia C. Correia, University of Coimbra, Portugal*

> *\*Correspondence: Giovanna Cenini gcenini@gmail.com*

#### *Specialty section:*

*This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology*

> *Received: 10 April 2019 Accepted: 18 July 2019 Published: 23 August 2019*

#### *Citation:*

*Cenini G and Voos W (2019) Mitochondria as Potential Targets in Alzheimer Disease Therapy: An Update. Front. Pharmacol. 10:902. doi: 10.3389/fphar.2019.00902*

damage, increased oxidative stress, neuronal and axonal injury, microglia-regulated neuroinflammation, and mitochondrial dysfunction are hallmarks of the disease (**Figure 1**).

Along the past years, Aβ peptides have been considered one of the most promising therapeutic targets for AD. However, many clinical studies based on the Aβ cascade hypothesis failed, and the idea that Aβ pathology is not anymore the leading primary cause of AD has risen (Morris et al., 2018). Instead, nowadays the belief that AD is a multi-factorial disease is growing steadily, and mitochondrial dysfunction is one of the factors that may actively contribute to the disease onset and progression (Iturria-Medina et al., 2017; Veitch et al., 2019). Despite that, a logical temporal order of the events in AD, as well as a valid and effective therapy, is still missing. However, our society urgently requires medical interventions to counteract this deleterious disease because of the severe negative impact on the quality of lives of the afflicted patients as well as on the health system as a whole due to a rapidly aging population.

This review focuses on the description of the role of mitochondrial dysfunction and the status of mitochondrial therapy in AD. The main question addressed here is: could the mitochondrial organelle be a valid pharmacologic target to prevent or delay the AD onset or to block the AD progression?

## MITOCHONDRIA

The mitochondrion is a cellular organelle with a characteristic and unique structure formed by two membranes, respectively called outer mitochondrial membrane (OMM) and inner mitochondrial membrane (IMM) that surround the matrix. Mitochondria are defined as the powerhouse of the cell because every cell in the human body relies on the energy provided by these organelles to sustain their vital functions. Mitochondrial energy production via the so-called process of oxidative phosphorylation takes place at the IMM through the activity of respiratory chain complexes (RCC), generating an inner membrane potential (mtΔΨ) that is used by the ATP-synthase enzyme complex to synthesize adenosine triphosphate (ATP). This process depends on the supply of reducing equivalents by the end-oxidation of nutrients *via* the Krebs cycle or β-oxidation in the mitochondrial matrix compartment (Stock et al., 2000). Mitochondria contain their own DNA (mtDNA) located in the matrix that encodes mainly 13 protein subunits of the RCC. All other mitochondrial protein components are encoded in the nuclear DNA (nuDNA) and are imported into the organelle after the translation at cytosolic ribosomes. Hence, the maintenance of an entire and functional mitochondrial proteome requires a fine-tuned and well-coordinated sequence of many reactions and a close integration of organellar and cellular biogenesis processes (Pfanner et al., 2019).

Neurons are strictly dependent on the presence of mitochondria in particular at the synapses where these organelles produce ATP and buffer Ca2+-ion concentration, both fundamental processes for the implementation of neurotransmission and generation of membrane potential along the axon (Li et al., 2004; Verstreken et al., 2005; Gazit et al., 2016). This justifies the high amount of mitochondria at the synaptic area, higher than any other part of the neurons. Linked to that, a correct and efficient transport of neuronal mitochondria at the synaptic terminals is fundamental

compromised in AD are on focus.

for their correct function. Both non-synaptic and synaptic mitochondria are usually synthesized in the neuronal soma and then transported in the other area of the neurons where they are required. The transport of mitochondria along the axons is guaranteed *via* microtubules and requires motor proteins such as kinesin, dynein, as well as the OMM protein Mitochondrial Rho GTPase (Miro). Axonal transport of mitochondria is also influenced by the metabolic demand and the Ca2+ status at the synaptic level (Yi et al., 2004; Glater et al., 2006; Russo et al., 2009; Sheng and Cai, 2012).

The enzymatic activity of the mitochondrial RCC results essentially in two "side effects." First, the generation of the mtΔΨ along the IMM is essential also for the execution of mitochondrial import of nuclear-encoded proteins and overall it is a parameter that reflects the health status of mitochondria and cells (Shariff et al., 2004). Second, a leakage of electrons from the RCC contributes significantly to the formation of reactive oxygen species (ROS). Therefore, ROS are considered a typical by-product of bioenergetic pathways (Quinlan et al., 2013). However, under normal physiological conditions, ROS production is well balanced by the presence of adequate antioxidant systems, and the damage to the diverse cellular constituents is contained. However, during aging, as well as during several pathological conditions, in particular in neurodegenerative diseases, this equilibrium becomes unbalanced. Increased ROS concentrations result in molecular damage at the site where they are produced or, through diffusion, in surrounding areas, leading to the generation of the so-called oxidative stress condition. ROS targets essentially comprise all cellular macromolecules, ranging from proteins, lipids, carbohydrates, up to nucleic acids (Cipak Gasparovic et al., 2017). The hippocampus region, the cortex, and more generally the brain are particularly vulnerable to oxidative stress because of their high consumption of oxygen and dependence on mitochondrial energy production. This susceptibility is increased by low levels of antioxidant defenses and a high content of polyunsaturated fats, which are especially vulnerable to oxidative alterations (Cobley et al., 2018).

Mitochondria form a dynamic tubular network extended throughout the cytosol, a behavior that is often misrepresented by the cell biology textbooks. Two crucial processes, fusion and fission, regulate the entire morphology and structure of this mitochondrial network (Mishra and Chan, 2016). During the fission reaction, a part of the mitochondrial tubule is divided into fragments, a process that is regulated by a member of the dynamin family, Dynamin-1-like protein (Drp1), together with the OMM fission factors Mitochondrial fission 1 protein (Fis1) and Mitochondrial dynamics protein MID49 [Mitochondrial elongation factor 2 (MIEF2)]. Fusion, where two or more pieces of mitochondria are fused together to one structure, happens through joint activity of the proteins Dynamin-like 120 kDa protein [or Optic atrophy protein 1 (OPA1)] and Mitofusin 1 and 2 (Mfn1 and Mfn2). Fusion/fission processes together with the precursor proteins import and internal proteins translation are part of the mitochondrial biogenesis in which the cells increase their mitochondrial mass (Sanchis-Gomar et al., 2014). A master regulator of mitochondrial biogenesis is Peroxisomeproliferator-activated receptor γ coactivator-1α (PGC-1α) (Scarpulla, 2011) that activates a series of transcriptional factors, including the Mitochondrial transcription factor A (TFAM), which regulates transcription and replication of mtDNA (Kang et al., 2018), and Nuclear respiratory factor 1 (NFR-1) and 2 (NFR-2), which control the mitochondrial protein-encoded nuclear genes (Scarpulla, 2011).

The buffer of intracellular Ca2+ is mediated mainly by the cooperation between endoplasmic reticulum (ER) and mitochondria through the formation of contact sites (Krols et al., 2016) that permit the Ca2+ uptake from the cytosol and the exchange of the ion between the two organelles (Rizzuto and Pozzan, 2006). Ca2+ regulates important mitochondrial metabolic enzymes (McCormack et al., 1990). The mitochondria contain two types of Ca2+ channels: the Mitochondria calcium uniporter (MCU) with high selectivity for this ion and localized in the IMM (De Stefani et al., 2011) and the Voltage-dependent anion channel (VDAC) localized in the OMM that regulates the release of the Ca2+ from the mitochondria (Krols et al., 2016). Furthermore, VDAC cooperates with the adenine nucleotide transporter in the IMM and the cyclophin D (CypD) in the matrix on the formation of the mitochondrial permeability transition pore (mPTP) (Bernardi, 1999). An mPTP opening leads to activation of apoptosis and then cell death (Green and Kroemer, 2004). As already mentioned above, at the synaptic level, mitochondria regulate the amount of Ca2+ fundamental for neurotransmission and in general for the exertion of synaptic functions (Werth and Thayer, 1994; Billups and Forsythe, 2002).

Mitochondrial functions and eventually cellular homeostasis are guaranteed by a dedicated mitochondrial quality control system (mtQCS). The mtQCS comprises a multitude of different biochemical mechanisms that act at different levels, affecting individual polypeptides as well as the whole organelle. While the folding state and activities of mitochondrial proteins are controlled by endogenous chaperones and proteases (Voos, 2013), damaged mitochondria may be removed by a selective autophagy pathway, termed mitophagy (Youle and Narendra, 2011). The primary regulator of the mitophagy is a specialized signaling system consisting of the protein PTEN-induced kinase 1 (Pink1) and the ubiquitin ligase Parkin that is activated after the loss of mtΔΨ (Rüb et al., 2017). An accumulation of Pink1 at the OMM of damaged mitochondria is thought to recruit Parkin that leads to a labeling of the mitochondria for the subsequent mitophagy process. This is followed by the formation of an autophagosomal membrane engulfing the mitochondria followed by its fusion with the lysosomes where ultimately the digestion of the mitochondrial material takes place.

## MITOCHONDRIAL DYSFUNCTION IN AD

In AD brain, the alteration of energetic pathways, also linked to the reduction of glucose consumption, is a well-established feature of the disease (Gibson and Shi, 2010). The glucose uptake in the brain is usually measured with the positron emission tomography (PET) tracer 18-fluorodeoxyglucose (fDG). In subjects with AD, PET studies have consistently demonstrated a low rate of glucose metabolism (between 20% and 30% lower than healthy individuals) in brain regions involved in processing memory (e.g., the hippocampus, posterior cingulate, temporal, and parietal lobes) (Kapogiannis and Mattson, 2011). Furthermore, it was proposed that the metabolic changes appeared earlier than the onset of the histopathological markers and symptoms (Gibson and Shi, 2010). Although the real cause is still unclear, the defective metabolism that characterizes AD could be easily linked to mitochondrial dysfunction.

Since its formulation in 1992 (Hardy and Higgins, 1992), the "amyloid cascade hypothesis" has dominated the AD field in the past 30 years. This hypothesis was based on two clear evidences: Aβ peptides constitute the extraneuronal senile plaques and mutation of Aβ peptides precursor, amyloid-β precursor protein (APP), leads to an early onset of AD. However, due to the fails in all Phase III clinical trials in human AD, this hypothesis has substantially lost ground and needed to be strongly revised or integrated with other hypotheses (Karran et al., 2011). In 2004, a new hypothesis was proposed to explain the onset of sporadic AD. The hypothesis, called "mitochondrial cascade hypothesis," described that each human genetic heritage influences mitochondrial functions with a primary repercussion on the onset of AD pathology. In other words, according to this hypothesis, the mitochondrial dysfunction is the primary process to trigger all the cascade of events that lead to sporadic late-onset AD (Swerdlow and Khan, 2004; Swerdlow et al., 2014).

Despite the fact that the validity of the mitochondrial cascade hypothesis has yet to be demonstrated in different AD models as well as human patients, the following mitochondrial functions were found severely compromised in the AD context (Hauptmann et al., 2009): mitochondrial morphology (Johnson and Blum, 1970) and number (Hirai et al., 2001), oxidative phosphorylation, mtΔΨ, Ca2+ buffering, ROS production (Butterfield and Halliwell, 2019), mtDNA oxidation and mutation (Wang et al., 2006), mitochondrial-ER contact sites (Area-Gomez et al., 2018), mitochondrial biogenesis, mitochondrial transport along the neuronal axon (Calkins and Reddy, 2011), and mitophagy (**Figure 1**). In a neuronal context, any of these dysfunctional processes could lead to synaptic deficits and critical consequences not only for single neurons but also for a more complex structure like the brain (Cai and Tammineni, 2017).

In AD brains, the activities of the enzymes involved in mitochondrial energy production, such as complex IV cytochrome c oxidase (COX), pyruvate dehydrogenase complex, mitochondrial isocitrate dehydrogenase, α-ketoglutarate dehydrogenase (αKGDH), and ATP synthase complex were found decreased, while the succinate dehydrogenase (complex II) and malate dehydrogenase activities were increased (Maurer et al., 2000; Cardoso et al., 2004; Gibson and Shi, 2010; Wojsiat et al., 2015). This definitely compromises the maintenance of the mtΔΨ and eventually of the mitochondrial ATP production (Beck et al., 2016).

In line with that, the imbalance between ROS production and antioxidant power was observed in AD brains, cerebrospinal fluid (CSF), and blood (García-Blanco et al., 2017). Since the 1990s, the ROS-induced oxidative stress has received considerable attention as one of the main factors contributing to the AD pathogenesis (Mark et al., 1997). Already the mild cognitive impairment (MCI), an early stage in the AD chronology, is characterized by the significant increase of oxidative stress markers, such as lipid peroxidation and protein oxidation products, and the decrease of antioxidants in the brain and peripheral compartments (Praticò et al., 2002; Rinaldi et al., 2003; Butterfield et al., 2006).

The analysis of the samples from different AD experimental models and AD patients showed a strong link between the oxidative stress and mitochondrial dysfunction. In the transgenic mice over-expressing human APP (Tg mAPP mice), an early and progressive accumulation of Aβ peptide in synaptic mitochondria led to a mitochondrial synaptic dysfunction such as damaged mitochondrial respiratory activity, increased mPTP and oxidative stress, and impaired mitochondrial axonal transport (Du et al., 2010). Data from the 3xTg-AD mice showed that the compromised mitochondria bioenergetics together with elevated oxidative stress levels are early phenomena appearing before the development of observable Aβ plaques (Hauptmann et al., 2009; Yao et al., 2009). Oxidation of one of the mitochondrial enzymes involved in the oxidative phosphorylation, ATP synthase, was found in isolated lymphocytes from AD peripheral blood as well as in MCI and AD brains (Sultana et al., 2006; Reed et al., 2008; Tramutola et al., 2018). This may explain the compromised activity of the ATP synthase and the reduction of ATP levels in AD. Another paper showed a correlation between the reduction of the mitochondrial enzyme Aconitase (ACO2) activity and the plasma antioxidant levels in peripheral lymphocytes from MCI and AD patients proving again the strong association between the oxidative stress and the mitochondrial dysfunction in AD (Mangialasche et al., 2015). Interestingly, the new and innovative technology for AD modeling obtained with the human induced pluripotent stem cells (iPSCs) directly from AD patients demonstrated further that AD-relevant mitochondrial aberrations, including oxidative stress, have a causative role in the developments of the disease. Indeed, neurons and astrocytes from AD-iPSCs presented increased ROS production and RCC levels and enhanced susceptibility to the stressors (Ochalek et al., 2017; Oksanen et al., 2017; Birnbaum et al., 2018).

The mitochondrial dynamics such as fusion and fission processes were found unbalanced in AD, potentially leading to i) compromised distribution and morphology of mitochondria in the neurons (Hirai et al., 2001) and ii) fragmented mitochondria observed in fibroblasts and brains from AD patients (Wang et al., 2008a; Wang et al., 2009). The mitochondrial fusion and fission proteins were differentially expressed in AD hippocampus with an increase of the mitochondrial fission protein Fis1 alongside with a significant downregulation of Drp1 and fusion proteins Mfn1, Mfn2, and OPA1 (Wang et al., 2009). Similar results were found in a AD cybrids model, together with bleb likeand shorter mitochondria compared to control samples (Gan et al., 2014). Furthermore, increased phosphorylation at Ser 616 site and S-nitrosylation of Drp1, which both facilitate the mitochondrial fission (Taguchi et al., 2007; Cho et al., 2009), were higher in a AD brains compared to control (Wang et al., 2009). Beside that, the protein Drp1 was seen interacting with Aβ and phosphorylated tau in brain homogenates from AD patients (Manczak et al., 2011; Manczak and Reddy, 2012). A recent study performed in samples from AD and healthy control subjects showed the significant association between a specific polymorphism in *MFN2* gene and AD suggesting that genetic polymorphism of fusion process regulation might be involved in the AD pathogenesis (Kim et al., 2017). In addition, mfn2 protein act as a tether between mitochondria and ER membranes (de Brito and Scorrano, 2008). In this regard, mfn2 influences the Presenilin 2 (PS2), whose mutation is linked to the familial AD (FAD), in the modulation of the mitochondria-ER contact sites (Filadi et al., 2016).

Several experimental AD models linked to APP overexpression or Aβ peptides treatments are characterized as well by mitochondrial fragmentation and abnormal mitochondrial distribution along the neurons due to an alteration of mitochondrial fusion and fission proteins levels (Wang et al., 2008b; Du et al., 2010; Zhao et al., 2010; Calkins and Reddy, 2011; Wang et al., 2017). All these results lead to two critical remarks: i) the altered balance between fusion and fission that interferes with mitochondrial transport contributes actively to the AD pathogenesis and ii) the mitochondrial dynamics impairment could be a new therapeutic target in AD.

Another key mitochondrial function, the mitochondrial biogenesis, was impaired in AD. The significant reduction of the number of mitochondria in AD human hippocampus and in cell culture models already suggests that the mitochondrial biogenesis is compromised (Hirai et al., 2001; Wang et al., 2008b). Furthermore, the level of protein regulating the mitochondrial biogenesis such as PGC-1α, NRF1 and 2, and TFAM was significantly reduced in human AD hippocampus and cellular models overexpressing APP Swedish mutation (Qin et al., 2009; Sheng et al., 2012). In the AD mouse model harboring mutant human transgenes of APP and Presenilin-1 (PS1), the mitochondrial biogenesis markers were found again declined in particular in the hippocampus region, and the use of melatonin brought beneficial effects (Song et al., 2018).

Interestingly, on one side, mitophagy was able to reverse the memory impairment, to prevent the cognitive deterioration and the Aβ peptide/tau pathology in several AD models (Fang et al., 2019). However, on the other side, mitophagy was also strongly affected in AD, leading to the accumulation of damaged mitochondria and consequently to dysfunctional neurons. One cause may be the impairment of the fusion between the autophagosome and lysosomes. This was observed in cultured cells overexpressing mutant APP, in AD mouse models, and also in neurons from AD patients' brain (Boland et al., 2008; Lee et al., 2010; Coffey et al., 2014). In AD brains, the somatic mutations found in mtDNA are higher than in healthy brains, potentially triggering other neuropathological consequences such as the increased ROS production in neurons and the promotion of amyloidogenic processing of APP (Lin et al., 2002).

The two major and typical histopathological markers of AD, Aβ peptide and tau, harmfully accumulate in or interact nonspecifically with mitochondria (Eckert et al., 2010). Aβ peptide and abnormal tau negatively affect axonal transport and consequently the transport of mitochondria along the axon from the neuronal soma to the synapses. AD mouse models, overexpressing Aβ peptides, have damaged mitochondria usually characterized by impaired axonal transport of mitochondria, a reduced mtΔΨ, and inhibited RCC with a compromised ATP production (Rui et al., 2006). The accumulation of Aβ peptides or of the precursor APP inside the mitochondria (Anandatheerthavarada et al., 2003; Hansson Petersen et al., 2008) and even the interaction of Aβ peptides with some component of the mitochondrial matrix (Lustbader et al., 2004) would be the most straightforward and rational explanations to justify the mitochondrial dysfunctions in the animal models of AD. However, mitochondria lack APP and the set of the enzymes required for Aβ peptide generation, making a mitochondria-localized production of Aβ peptides unlikely. Furthermore, a solid mechanism that explains the mitochondrial import of Aβ peptides and the direct negative effects of Aβ peptides on mitochondria is still missing, suggesting that the mitochondrial dysfunctions identified in all these AD models are indirect effects of Aβ peptides. In support of this point, a recent study showed that Aβ peptides impaired mitochondrial import of nuclear-encoded precursor proteins due to an extra mitochondrial co-aggregation process (Cenini et al., 2016).

Tauopathies including AD are also characterized by mitochondrial dysfunction. Tau influences, directly and indirectly, the mitochondrial transport along the neuronal axon and the mitochondrial functions. This leads to the reduction and impairment of mitochondria at the presynaptic terminals with obvious deleterious consequences (Dubey et al., 2008; DuBoff et al., 2012). In AD brains, phosphorylated tau was found interacting with VDAC1 leading as well to mitochondrial dysfunction (Manczak and Reddy, 2012). Hyperphosphorylation of tau negatively affects complex I activity with a decrease of ATP production, an increase of oxidative stress, dissipation of mtΔΨ, induction of the mitochondrial fission, and excessive mitochondrial fragmentation in postmortem brains from AD patients and in murine models (Manczak et al., 2011; Eckert et al., 2014). In addition, mitochondrial stress was shown to promote tau-hyperphosphorylation in a mouse model (Melov et al., 2007). These observations argue for a prominent role of tau pathology in the mitochondrial dysfunction of AD.

The Translocase of outer membrane 40 kDa submit homolog (Tomm40) is a mitochondrial channel localized in OMM that is fundamental for the import of nuclear-encoded mitochondrial preproteins (Chacinska et al., 2009). Aβ peptides affected directly or indirectly the mitochondrial import machinery including Tomm40, and this may also contribute to the mitochondrial dysfunction observed in AD (Devi et al., 2006; Anandatheerthavarada and Devi, 2007; Cenini et al., 2016). *TOMM40* gene is contained in a tight gene cluster together with APOE gene in the chromosome 19 (Gottschalk et al., 2014; Subramanian et al., 2017). APOE is one of the most significant genetic risk factors for late-onset sporadic AD (LOAD) with the ε4/ε4 isoform linked to the highest risk (Saunders et al., 1993). It seems that also a variable-length, deoxythymidine homopolymer polymorphism in intron 6 of the *TOMM40* gene represents a genetic risk for LOAD. However, different groups showed that *TOMM40* SNPs (single-nucleotide polymorphisms) are associated with the LOAD (Martin et al., 2000; Takei et al., 2009; Kim et al., 2011; Davies et al., 2014). In a Caucasian ethnic group three variants of the *TOMM40* polymorphisms were identified, and the variant rs10524523 has received particular attention since it lowered the age of LOAD onset by 7 years in APOE3/4 carriers (Roses et al., 2010). Furthermore, this variant was associated with impaired cognition and the gray matter volume in the brain area susceptible to AD (Johnson et al., 2011). Different groups also demonstrated the strong influence of *TOMM40* "523" variant on *TOMM40* and *APOE* genes transcription (Linnertz et al., 2014; Payton et al., 2016).

The integration of all these facts into a significant biological context like neuronal cells in AD, suggests that the accumulation of dysfunctional mitochondria at the synapses and the lack of their replacement would contribute substantially to the neurons degeneration and consequently to the worsening of the AD condition.

## MITOCHONDRIAL THERAPIES IN AD

AD is still without a cure and also essentially lacks a rational understanding of the primary event triggering the disease. Nevertheless, an improved comprehension of this deleterious disorder and the development of effective treatments are essential not only to heal the disease but also eventually to prevent or postpone the onset of the symptoms in the patients.

The traditional cures used nowadays to treat the AD patients are so far the cholinesterase inhibitors (**donepezil**, **rivastigmine**, and **galantamine**) and **memantine** that block the N-methyl-Daspartate (NMDA) receptor and the excess of glutamate activity. NMDA receptors and acetylcholin (Ach) are fundamental in memory and learning processes and their concentration and function are compromised in AD (Francis, 2005). However, these treatments improve the cognitive and memory functions, without really slowing down the progression of the disease.

As described above, mitochondrial dysfunctions and a compromised energetic metabolism are two prominent aspects of AD pathology. Therefore, mitochondria should be seriously considered as pharmacological targets. In the course of history, nevertheless, different compounds affecting mitochondria were already tested in AD without a successful outcome. However, as the idea of AD as a multifactorial disease gained more ground in the last years, a reconsideration of mitochondria as a valid therapeutic target together with other medications is strongly recommended.

Mitochondria could be targeted through two ways: i) by pharmacologic approaches acting on mitochondria directly or ii) by action on the lifestyle that indirectly hits this organelle (**Figure 2**). In the following section, we describe the most popular mitochondrial treatments that have been used until today on AD patients, and in **Table 1**, we summarize specifically the beneficial effects of these compounds on mitochondria in different experimental AD models. The table is also a proof that these treatments are able to act effectively and positively on mitochondria, and therefore a revision and improvement of their use in AD would be worthy.

More information about the ongoing clinical trials concerning mitochondria in AD are summarized in Wilkins et al. and in Perez Ortiz et al. (Perez Ortiz and Swerdlow, 2019; Wilkins and Morris, 2017), and they can also be found in www. clinicaltrials.gov.

## Antioxidants

Since the increased oxidative stress accompanied by the reduction of the antioxidant power was measured in the brain, CSF, and blood from AD patients, treatments with antioxidant compounds were tested to counteract this oxidative unbalance and slow down the progression of the AD symptoms.

Typical antioxidants were the **vitamins**, **E** and **C**, but their effects in the context of AD remain questionable. For example, in two studies with vitamin E, some markers of lipid peroxidation were found decreased in AD patients' CSF, with no consistent effect on or even a deterioration of cognitive functions (Arlt et al., 2012; Galasko et al., 2012). Vitamin E was also administered in combination with **selenium**. However, high levels of selenium were found toxic with a pro-oxidant effect, glial activation, and neuronal death (Vinceti et al., 2014). There is an important study called **PREADViSE** that was performed to see the longterm effect of anti-oxidant supplements (Vitamin E, selenium, Vitamin E + selenium or placebo) on dementia incidence among asymptomatic men. However, the supplement did not prevent dementia occurrence (Kryscio et al., 2017).

TABLE 1 | List of compounds and lifestyle activities effects on mitochondria in experimental models for AD.


(*Continued*)

#### TABLE 1 | Continued


(*Continued*)

#### TABLE 1 | Continued


*AD, Alzheimer's disease; ETC, electron transport chain; RCC, respiratory chain complexes, mtΔΨ: mitochondrial membrane potential; OCR, oxygen consumption rates; ATP, adenosine triphosphate; mPTP, mitochondrial permeability transition pore; mtDNA, mitochondrial deoxyribonucleic acid; nuDNA, nuclear deoxyribonucleic acid; APE1, apurinic/ apyrimidinic endonuclease 1; MnSOD, manganese superoxide dismutase; OGG1, oxoguanine DNA glycosylase-1; αKGDH, α-ketoglutarate dehydrogenase; COX, cytochrome c oxidase or complex IV; TIM, translocase inner membrane; TOM, translocase outer membrane; Mfn1, mitofusin-1; Drp1, dynamin-1-like protein; PGC-1α, peroxisome-proliferatoractivated receptor γ coactivator-1α; NAD, nicotinamide adenine dinucleotide; NADH, reduced nicotinamide adenine dinucleotide; ROS, reactive oxygen species; RNS, reactive nitrogen species; GSH, glutathione; GSSG, oxidized glutathione; 3-NT, 3-nitrotyrosine; MDA, malondiaaldehyde; SelM, selenoprotein M; 6-OHDA, 6-hydroxydopamine; OA, okadaic acid; H2O2, hydrogen peroxide; NMN, nicotinamide mononucleotide; Aβ, β-amyloid peptide; AβPP, β-amyloid precursor protein; PS1, presenilin 1; BACE1, β-secretase-1; HEK293, human embryonic kidney 293 cell lines; HUVEC, human umbilical vein endothelial cell line; M17, human neuroblastoma cell line; N2a, mouse neuroblastoma cell line; LUHMES, Lund human mesencephalic cell line; SH-SY5Y, human neuroblastoma cell lines; IMR-32, human neuroblastoma cell lines; PC12, pheochromocytoma of rat adrenal medulla-derived cell lines; OHCs, organotypic hippocampal slice cultures; NARP, cybrid cell lines bearing mtDNA mutation T8993G; CGN, cerebellar granule neurons; 5xFAD, mice expressing human APP and PSEN1 genes with a total of five AD-linked mutations, the Swedish, Florida, and London mutations in APP, and the M146L and L286V mutations in PSEN1; APP/PSEN1, mice contain human APP gene bearing the Swedish mutation and PS1 gene containing L166P mutation; TgP301S, mice expressing mutant human microtubule-associated protein tau (MAPT); Tg19959, mice expressing human APP gene bearing the Swedish mutation and Indiana mutation; TgCRND8, mice expressing human APP695 gene with the Swedish mutation and Indiana mutation; 3xTg-AD, mice contain three mutations (Swedish, MAPT, PS1) associated with familial AD; Tg2676 mice, mice expressing mutant human form of APP (isoform 695) with Swedish mutation; APP751SL, mice expressing the human APP bearing both Swedish and the London mutation; ApoE4 Tg mice, mice expressing human apolipoprotein E (APOE) gene; OXYS rats, senescence-accelerated rats; MCAT, mitochondria-targeted catalase; C. elegans, Caenorhabditis elegans.* 

Targeting directly the mitochondria with antioxidant compounds was always one of the most considered therapeutic strategies in AD. In this regard, an antioxidant directed to mitochondria that has been tried was the **coenzyme Q10**  (**CoQ10**). CoQ10 has a quinone structure and is a component of the mitochondrial RCC. In a rat model for AD, CoQ10 prevented the cognitive decline (Dehghani Dolatabadi et al., 2012). Still, due to a low bioavailability in the brain (Kwong et al., 2002), CoQ10 has never been successful in humans. To overcome this issue, the **mitoquinone mesylate** (**MitoQ**) was optimized. MitoQ is an antioxidant compound made of ubiquinone conjugate with triphenylphosphonium (TPP). The TPP is necessary to target the molecule to the mitochondria because it helps to cross the lipid bilayers accumulating on the negative site of mitochondrial membranes (Kelso et al., 2001; Smith et al., 2003). MitoQ behaved as ROS scavenger and was tested in different AD model systems (see **Table 1**). Here, MitoQ shown to prevent oxidative damage,

to protect RCC activity, to reduce Aβ peptide levels, synaptic loss, and astrogliosis, and to improve cognitive functions (McManus et al., 2011; Ng et al., 2014). As reported in the review from Ortiz (Perez Ortiz and Swerdlow, 2019), at the moment, MitoQ is tested in a small clinical trial to check its effect on cerebrovascular blood flow in AD. Similarly to MitoQ, other antioxidant compounds (**SkQ1**, **MitoApo**, **astaxanthin**) affect positively the mitochondrial functions (see **Table 1**) and could be potentially used to treat AD (Lobos et al., 2016; Stefanova et al., 2016; Brenza et al., 2017).

Another group of antioxidant molecules such as **melatonin**, **α-lipoic acid (LA)**, **N-Acetyl-cysteine (NAC)**, and *Ginkgo biloba* were tested *in vivo* and *in vitro* and showed protective effects on Aβ peptide accumulation and mitochondrial toxicity as well as on cognitive functions (Dong et al., 2010; Rosales-Corral et al., 2012a). **Melatonin** is a neurohormone produced by the pineal gland with neuroprotective functions in AD pathogenesis (Shukla et al., 2017). Melatonin is a ROS scavanger and showed some anti-amyloidogenic properties (Dong et al., 2010; Rosales-Corral et al., 2012a). At mitochondrial level, melatonin prevented the ROS production, the cardiolipin oxidation, and the mPTP opening, restored the Ca2+ balance, and reduced the caspase-3 and -9 levels (Feng and Zhang, 2004; Jou et al., 2004; Petrosillo et al., 2009; Espino et al., 2010). Treatments with **α-lipoic acid**, a cofactor for many RCC enzymes, exhibited a positive effect on cognitive functions in clinical trials on AD patients and in murine models of aging and AD, α-lipoic acid affected also the formation and the stabilization of Aβ peptide fibril as well as the protection against the Aβ peptide toxicity in cultured hippocampal neurons (Liu et al., 2002; Lovell et al., 2003; Ono et al., 2006; Hager et al., 2007; Quinn et al., 2007; Sancheti et al., 2013). **N-Acetyl-cysteine** (**NAC**) is the precursor of the endogenous antioxidant glutathione (GSH), a key molecule for the maintenance of mitochondrial functions (Traber et al., 1992). *In vitro* and *in vivo*, NAC had beneficial effects on Aβ peptide and phosphorylated tau levels with improvement of cognitive functions, protection against memory decline, and reduction of oxidative stress markers (see also **Table 1**) (Studer et al., 2001; Fu et al., 2006; Huang et al., 2010; Costa et al., 2016). In two clinical trials, subjects with MCI, AD, or early memory loss were treated for a long time with a nutraceutical formulation that also included NAC. Improvement of cognitive and behavioral functions was observed (Remington et al., 2015; Remington et al., 2016). *G. biloba* is a natural antioxidant already used in the Chinese traditional medicine. **Table 1** shows all the effects of *G. biloba* on mitochondrial functions. Two clinical trials were performed to test the effect of *G. biloba* in the prevention against memory and cognitive decline in older adults and AD subjects. Unfortunately, no positive effects were observed in these tests (Snitz et al., 2009; Vellas et al., 2012).

The **Szeto-Schiller** (**SS**) **tetrapeptides** are a group of small peptides that due to their structure act as antioxidants and can reach the mitochondrial matrix and the IMM (Szeto, 2006). In one of AD murine models, the **SS31** reduced Aβ peptide production, mitochondrial dysfunction, and enhanced mitochondrial biogenesis and synaptic activity (Calkins et al., 2011; Reddy et al., 2017). Recently, a combination of SS31 and the mitochondrial division inhibitor 1 (Mdivi1) was tested in cultured AD cells with positive effects, suggesting that a combined treatment of mitochondria-targeted antioxidants could have higher effectiveness (Reddy et al., 2018).

An interesting preclinical study proposed to target the antioxidant enzyme **catalase** to the mitochondria. Catalase catalyzes the decomposition of hydrogen peroxide (H2O2) in water (H2O) and oxygen (O2) and is typically localized in the peroxisome. A double transgenic mouse with mitochondriatargeted catalase (MCAT) and APP was created, and the protective effects against abnormal APP processing, Aβ peptide pathology, and lifespan extension were tested. Mitochondrial catalase showed beneficial outcomes in this highly artificial model. Although most of the antioxidant clinical trials were not entirely successful, this study proved that a direct target of an antioxidant to the mitochondria might still have a chance as a therapeutic approach in AD (Mao et al., 2012).

Despite the oxidative stress unbalance is an evident hallmark in AD and some mitochondrial-targeted antioxidant strategies showed promising effect on cognitive functions, none entered so far in the market as a valid AD treatment. There are different reasons to justify the failures (summarized in Persson et al. paper; Persson et al., 2014). The antioxidants at certain concentrations and conditions could behave as pro-oxidants and therefore they are more harmful than useful. The antioxidant administration during the clinical trials was probably started too late during the development of the disease suggesting that an early intervention could be more effective. Last, the antioxidant bioavailability in the brain could be low due to the difficulty of these molecules to cross the blood–brain barrier (BBB) requiring a rational modification of their structure to overpass this issue.

## Phenylpropanoids

The **phenylpropanoids** are natural compounds that exert many physiological functions crucial for the survival of plants. In this heterogeneous group of substances, many subclasses have been identified such as stilbenoids, flavonoids, curcuminoids, phenolate esters, and lignans. These compounds showed an effect against the Aβ peptide and tau pathologies, on the activation of the inflammation response, on the oxidative stress, and also on the mitochondrial dysfunction (Kolaj et al., 2018). Between others, **resveratrol**, **quercetin**, **wogonin**, **epigallocatechin-3-gallate** (**EGCG**), and **curcumin** were already tested and showed to promote mitochondrial biogenesis, to impede apoptotic pathways through inhibition of DNA fragmentation, ROS formation, and caspase-3 activation, and to reduce perturbation of mtΔΨ and ATP levels (see also **Table 1** for the effects of phenylpropanoids on mitochondria in AD models) (Lagouge et al., 2006; Davis et al., 2009; Im et al., 2012; Valenti et al., 2013; Reddy et al., 2016). Furthermore, these compounds were able to restore the mitochondrial functions in a transgenic mouse model of AD (Dragicevic et al., 2011b). In particular in an *in vitro* study, **EGCG**, a major flavonoid component of the green tea, accumulated in mitochondria and exerted a strong influence on the mitochondrial functions proposing it as pharmacological treatment in AD (Schroeder et al., 2009; Dragicevic et al., 2011b). However, phenylpropanoids have a dual effect on mitochondrial function, depending on the concentration. For example, EGCG could increase apoptosis in cultured neurons at specific concentrations, while quercetin protected cultured hippocampal cells against Aβ peptide-induced apoptosis only in low concentrations (Chung et al., 2007; Ansari et al., 2009). **Curcumin** is an antioxidant compound with massive potential for the prevention and treatment of AD. It showed beneficial effects on Tg2576 AD model mice, such as reduction of the brain oxidative stress and the neuroinflammation, but no effect in AD patients, probably due to a low bioavailability (Lim et al., 2001; Baum et al., 2008; Ringman et al., 2012). New strategies have been implemented to overpass this limitation and improve the curcumin pharmacokinetics, such as the nanotechnology-based delivery system, new pharmaceutical formulations, and the change in the way of administration (Reddy et al., 2014; Serafini et al., 2017).

Like the antioxidant, the use of the phenylpropanoids in AD treatment needs to be considered with caution and none of them has become a real therapy yet. The new AD clinical trials based on this group of molecules definitely require a broad design, a substantial revision, and a careful implementation.

## Action on the Lifestyle Calories Restriction, Diet, Exercises

Lifestyle activities, in particular **exercise** and **diet**, have been known to act at the mitochondrial level and should therefore be considered as possible interventions to treat AD. **Table 1** reports the effects of the compounds and activities strictly related to the lifestyle on mitochondria from AD models.

A **Mediterranean diet** has been correlated to the reduction of the incidence of AD (Scarmeas et al., 2006; Karstens et al., 2019). The Mediterranean diet is mainly composed of fruits, vegetables, and omega-3 fatty acids, which are enriched in olive oil. It was observed that, for example, **polyphenol-rich extra-virgin oil** reduced mitochondria-generated oxidative stress and insulin resistance in high-fat diet fed rats (Lama et al., 2017). Another polyphenol component of olive oil called **oleuropein aglycone** (**OLE**) promoted autophagy, decreased aggregated proteins levels, and reduced the cognitive impairment in AD patients' brain (Grossi et al., 2013; Cordero et al., 2018). **Hydroxytyrosol** (**HT**), another bioactive compound of olive oil, ameliorated mitochondrial dysfunction in an animal model of AD (Peng et al., 2016). On the other side, higher consumption of **fructose** affected negatively the mitochondrial function in hippocampus from adult rats, suggesting that fructose consumption should be actively avoided (Cigliano et al., 2018). **Ketones** are another source of energy for the brain when there is a limited amount of available glucose (Owen et al., 1967). The ketone ester diet in a model of AD (3xTgAD) had positive effects also on mitochondrial functions (Pawlosky et al., 2017). The therapeutic ketosis was suggested to reduce the AD brain pathology including the accumulation of Aβ plaques and NFT (Kashiwaya et al., 2013). Of course, the results obtained in AD murine models have to be proven in humans through clinical trials (Puchowicz and Seyfried, 2017). In this regard, there are experiments going on at the University of Kansas about the effect of a ketogenic diet (KD) on participants with AD, but no definitive results are available yet (Taylor et al., 2018; Taylor et al., 2019).

An extreme form of diet is represented by **calorie restriction** (**CR**). CR is a strong limitation on calorie intake without facing a lack of nutrients. It is well known that CR is an excellent way to extend lifespan, to increase insulin sensitivity, and to prevent agerelated diseases (Mattison et al., 2017). At the mitochondrial level, CR showed positive effects by affecting mitochondrial biogenesis through the induction of NO synthetase (eNOS) (Nisoli et al., 2005). Newly synthesized mitochondria led to an increase of mitophagy, reduction of ROS, increased ATP levels, and overall improvement of the mitochondrial quality and cell bioenergetics (López-Lluch et al., 2006). Furthermore, CR affected the mtDNA content as well as the amount of TFAM-bound mtDNA in rats (Picca et al., 2013). There are ongoing clinical studies around the world concerning the effect of CR and dietary intervention on MCI (Wilkins and Morris, 2017).

**Physical exercise** (**PE**) has been demonstrated to generally benefit the health of the body and mind, affecting properties such as brain plasticity and cognitive function. Hence, it could be a good prevention for age-related diseases (Hernández et al., 2015; Paillard et al., 2015). It is well known that PE targets mitochondria and improved mitochondrial function (see **Table 1** to check the effects of PE on mitochondria in AD models). A study showed that PE increased mtDNA repair, ameliorated mitochondria respiratory function through the increase of RCC activity, attenuated ROS generation capacity together with a reduction of Aβ1-42 peptide levels, and correlated with an amelioration of cognitive function in the hippocampus from the APP/PS1 transgenic mouse model of AD (Bo et al., 2014). However, data obtained in another AD mouse model (3xTg-AD) demonstrated that short-term exercise did not augment the critical gene expression of mitochondrial biogenesis, even if the glucose metabolism was overall improved (Do et al., 2018). Maternal exercise during pregnancy resulted in a positive effect on mitochondrial function concerning the onset of AD. In this study, a protective effect against Aβ oligomer-induced neurotoxicity in the adult offspring brain rats was shown (Klein et al., 2019). Clinical trials with PE were performed in older adults with healthy as well as impaired cognitive function. Aβ1-42 concentration in plasma and CSF was modified. In the brain, improvements of cognitive and executive functions, and even a change of hippocampal volume and memory, were observed, together with a reduced brain atrophy (Baker et al., 2010; Erickson et al., 2011; Vidoni et al., 2015; Yokoyama et al., 2015). Of course, in these human studies, neither a direct effect of PE on mitochondria nor the molecular mechanisms of PE benefits have been proved. However, all the studies performed in animal models positively supported the hypothesis that PE may have a beneficial effect on mitochondrial functions and glucose metabolism also in humans.

Diet, CR, and PE can also be combined to improve the quality of human aging and to prevent neurodegenerative disease (Rege et al., 2017). These approaches were shown to affect mitophagy, the cellular removal mechanism for damaged mitochondria, indicating the mitophagy as a new and promising therapeutic target to prevent the progression of the diseases. Experimental evidences from rodent studies showed that fasting and exercises could have a beneficial effect not only on mitophagy but also on mitochondrial biogenesis, reduction of oxidative stress, and overall neuronal plasticity (Alirezaei et al., 2010). Other strategies to boost mitophagy in order to delay AD are the use of compounds like **2-deoxyglucose**, which protects neurons and enhances mitochondrial functions (**Table 1**) (Duan and Mattson, 1999; Yao et al., 2011). Additional molecules that promote autophagy/mitophagy are **rapamycin**, **spermidine**, **urolithins**, and the antibiotic **actinonin** (Spilman et al., 2010; Morselli et al., 2011; Ryu et al., 2016; Fang et al., 2019). The mTOR inhibitor rapamycin was already demonstrated to have beneficial effects on a mouse AD model (Spilman et al., 2010). Testing these molecules in clinical AD might be worth it.

## Other Mitochondria-Based AD Therapy Oxaloacetate

Treatment with **oxaloacetate** (**OOA**), an intermediate of the Krebs cycle and gluconeogenesis, has been proposed as a new therapeutic approach for AD, and it was already tested in some AD subjects (Swerdlow et al., 2016). Studies involving OOA performed in mice showed positive effects on glycolysis, respiratory fluxes, mtDNA and mtDNA-encoded proteins, activation of mitochondrial biogenesis, hippocampal neurogenesis activity, neuroinflammation, and change in brain insulin signaling (Wilkins et al., 2014). Despite there are no studies about the direct efficacy of OOA treatment on mitochondria in AD models, clinical trials with OOA in AD are ongoing.

### NAD

Nicotinamide adenine dinucleotide (NAD) is an intermediate common to several mitochondrial metabolic pathways such as glycolysis, TCA cycle, and oxidative phosphorylation. Studies on *in vitro* and *in vivo* AD models proved that NAD treatments acted directly on mitochondrial functions and were beneficial (**Table 1**). In the past, the effect of a stabilized oral NAD formulation on cognitive functions in AD patients was also tested. The rationale behind this testing was based on the enhancement of the cellular bioenergetic to improve brain performance in the fight against neurodegenerative diseases. Interestingly, after 6 months of treatment, the subjects with probable AD showed no cognitive deterioration suggesting that NAD could be an excellent method to prevent the AD progression (Demarin et al., 2004). However, further studies are needed to prove NAD as an effective treatment to slow down AD.

### Pioglitazone

The **pioglitazone** is a peroxisome proliferator-activated receptor gamma (PPARγ) agonist. PPARγ is a ligand-activated nuclear transcription factor that has a role in regional transcriptional regulation of chr19q13.32 (Subramanian et al., 2017). This region contains the *TOMM40-APOE-APOC1* genes and, as already mentioned, *TOMM40* and *APOE4* genes are risk factors for the LOAD development. Pioglitazone was able to decrease the transcription of *TOMM40*, *APOE*, and *APOC1* genes making this molecule an interesting candidate in the AD therapy (Subramanian et al., 2017). In CHO cell line overexpressing APP695 isoform, pioglitazone lowered the Aβ1-42 level and restored the mitochondrial activity (Chang et al., 2015). These results were then confirmed *in vivo* in APP/PSEN1 mice (**Table 1**) (Chang et al., 2019).

Pioglitazone is usually used to treat diabetes mellitus type 2. Some years ago, the pharmaceutical company Takeda used this compound in a large and global Alzheimer's prevention study called TOMMORROW to slow down the progression from MCI to AD. The people involved were selected based on their *APOE* and *TOMM40* genotype without considering Aβ status. In 2018, phase III of this prevention trial, unfortunately, closed down because the results against symptomatic AD were negative, despite some improvement in brain metabolism.

### Dimebon

Another compound that affects mitochondria but failed the AD clinical trial was **dimebon** (**latrepirdine**). Dimebon (latrepirdine) is an old antihistaminic drug (first generation of H1-antagonist) used against allergies that was selected in an AD clinical trial because it demonstrated cognition and memoryenhancing properties in rats treated with neurotoxin (Bachurin et al., 2001). Moreover, dimebon showed a substantial effect on mitochondria from different AD models (**Table 1**). Anyway, dimebon lacked reproducibility in the AD clinical trials and showed opposite effects on neuropsychiatric and cognitive symptoms, and daily activities (Bachurin et al., 2001; Doody et al., 2008). In a review from 2018, Eckert et al. asked the scientific community to reevaluate the drug dimebon as a potential treatment of AD since one of the clinical trials was able to show a slight improvement of mitochondrial functions after using dimebon in respect of the substantial effect on cognition and behavior (Eckert et al., 2018).

## CONCLUSION

In a multitude of studies, mitochondrial dysfunction has been demonstrated to be a crucial feature of AD. Several experimental results suggested that a decline of mitochondrial activity happens during aging and may get worse at early stages of the disease, contributing to disease onset. However, more thorough investigations are needed to properly address this point. The suitability of the mitochondria as a target in AD treatment is still under discussion, considering that some pharmacological trials were not successful and others were more promising, but none led to a real marketable AD drug. Nevertheless, the current understanding of AD indicates that a complete cure may not be reachable yet. Future research efforts should be invested to i) understand the real chronology of events, ii) collocate correctly the mitochondrial dysfunction inside this temporal sequence, and iii) establish if the mitochondrial dysfunctions are a primary cause or a secondary event. Only when these three key points will be correctly settled, it will be easier to intervene pharmacologically and no more time and money will be wasted for futile therapeutic studies. The failures of the respective drugs or clinical trials often happened because the underlying scientific background was not always very robust or because the models and the tools used to prove the basal hypothesis were not always well defined or validated. Therefore, a more rational approach to a complex human disease like AD is needed as well as an improvement of communication between the different scientific disciplines in order to achieve a better understanding of the disease etiology and to develop new and more effective drugs.

## AUTHOR CONTRIBUTIONS

GC conceived the idea and prepared the manuscript. WV reviewed the draft and provided important information for the completion of this manuscript.

## FUNDING

The Deutsche Forschungsgemeinschaft (Grant No VO 657/5-2 to WV) supported the work in our laboratory.

## REFERENCES


to autophagic pathology in Alzheimer's disease. *J. Neurosci.* 28, 6926–6937. doi: 10.1523/JNEUROSCI.0800-08.2008


Alzheimer's amyloid-induced mitochondrial dysfunction. *J. Alzheimers Dis.* 26, 507–521. doi: 10.3233/JAD-2011-101629


tuning the antagonistic effect of mitofusin 2. *Cell. Rep.* 15, 2226–2238. doi: 10.1016/j.celrep.2016.05.013


accumulates with age in AD transgenic mice. *Neurobiol. Aging* 30, 1574–1586. doi: 10.1016/j.neurobiolaging.2007.12.005


R-alpha -lipoic acid. *Proc. Natl. Acad. Sci. U.S.A.* 99, 2356–2361. doi: 10.1073/ pnas.261709299


peptide SS31 in Alzheimer's disease. *J. Alzheimers Dis.* 62, 1549–1565. doi: 10.3233/JAD-170988


of Alzheimer's disease (GuidAge): a randomised placebo-controlled trial. *Lancet Neurol.* 11, 851–859. doi: 10.1016/S1474-4422(12)70206-5


**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 © 2019 Cenini and Voos. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Neuroinflammatory Processes, A1 Astrocyte Activation and Protein Aggregation in the Retina of Alzheimer's Disease Patients, Possible Biomarkers for Early Diagnosis

Alfonso Grimaldi<sup>1</sup> , Natalia Pediconi<sup>1</sup> , Francesca Oieni<sup>2</sup> , Rocco Pizzarelli<sup>1</sup> , Maria Rosito<sup>1</sup> , Maria Giubettini<sup>3</sup> , Tiziana Santini<sup>1</sup> , Cristina Limatola2,4, Giancarlo Ruocco1,5 , Davide Ragozzino2,4 and Silvia Di Angelantonio1,2 \*

#### Edited by:

Cesare Mancuso, Catholic University of the Sacred Heart, Italy

### Reviewed by:

Ileana Soto, Rowan University, United States Claes Wahlestedt, University of Miami, United States

#### \*Correspondence:

Silvia Di Angelantonio silvia.diangelantonio@uniroma1.it; silvia.diangelantonio@iit.it

#### Specialty section:

This article was submitted to Neuropharmacology, a section of the journal Frontiers in Neuroscience

Received: 15 May 2019 Accepted: 19 August 2019 Published: 04 September 2019

#### Citation:

Grimaldi A, Pediconi N, Oieni F, Pizzarelli R, Rosito M, Giubettini M, Santini T, Limatola C, Ruocco G, Ragozzino D and Di Angelantonio S (2019) Neuroinflammatory Processes, A1 Astrocyte Activation and Protein Aggregation in the Retina of Alzheimer's Disease Patients, Possible Biomarkers for Early Diagnosis. Front. Neurosci. 13:925. doi: 10.3389/fnins.2019.00925 <sup>1</sup> Center for Life Nanoscience, Istituto Italiano di Tecnologia, Rome, Italy, <sup>2</sup> Department of Physiology and Pharmacology, Sapienza University, Rome, Italy, <sup>3</sup> Crestoptics S.p.A., Rome, Italy, <sup>4</sup> IRCCS Neuromed, Pozzilli, Italy, <sup>5</sup> Department of Physics, Sapienza University, Rome, Italy

Alzheimer's disease (AD), a primary cause of dementia in the aging population, is characterized by extracellular amyloid-beta peptides aggregation, intracellular deposits of hyperphosphorylated tau, neurodegeneration and glial activation in the brain. It is commonly thought that the lack of early diagnostic criteria is among the main causes of pharmacological therapy and clinical trials failure; therefore, the actual challenge is to define new biomarkers and non-invasive technologies to measure neuropathological changes in vivo at pre-symptomatic stages. Recent evidences obtained from human samples and mouse models indicate the possibility to detect protein aggregates and other pathological features in the retina, paving the road for non-invasive rapid detection of AD biomarkers. Here, we report the presence of amyloid beta plaques, tau tangles, neurodegeneration and detrimental astrocyte and microglia activation according to a disease associated microglia phenotype (DAM). Thus, we propose the human retina as a useful site for the detection of cellular and molecular changes associated with Alzheimer's disease.

Keywords: Alzheimer's disease, microglia, astrocytes, retina, beta-amyloid, tau, human, neurodegeneration

## INTRODUCTION

Alzheimer's disease (AD) is a neurodegenerative disorder leading to dementia during elderly. This pathology is manifested with cognitive and psychiatric symptoms such as memory and cognitive impairments, behavioral abnormalities, disorientation and circadian rhythms disturbances (Hart et al., 2016). Due to population aging, AD cases are constantly increasing and it is estimated that by the year 2050 AD will globally affects about 115 million people (Alzheimer's Disease International World Alzheimer Report, 2009; Scheltens et al., 2016) thus posing a growing concern on public health.

Studies from post-mortem brains revealed that, at the neuropathological level, distinctive features of AD include aggregation of amyloid beta protein (Aβ) and tau protein hyperphosphorylation (pTau), which cause synapses loss and neuronal degeneration. These plaques and tangles are generally associated with activated microglia and reactive astrocytes in AD brain (Ahmed et al., 2017; Henstridge et al., 2019). Over the years several candidate drugs have been proposed as a putative therapy for AD. However, despite many attempts, a definitive cure has not been found yet and only symptomatic treatments are available.

A possible explanation for this arises from the fact that AD is a complex multifactorial disorder, while pharmacological treatments are preferentially directed against a specific cellular target. Moreover, proteins aggregation can occur decades before the appearance of neurological symptoms, thus making interventions at later stages less effective.

It is therefore tempting to speculate that AD diagnosis at earlier stages, possibly when synapses and neuronal functions are not yet compromised, could give better results in terms of pharmacological and clinical intervention (Frisoni et al., 2017).

The search for early biomarkers in AD has become a very attractive area of study. Imaging brain changes in Aβ with techniques such as computed tomography (CT), magnetic resonance index (MRI) and positron emission tomography (PET) in combination with the analysis of cerebrospinal fluid (CSF) have established Aβ and tau as an indicator of the AD disease.

In particular, both PET and CSF biomarkers allow to diagnose AD during the prodromal phase of the disease (Rowe et al., 2013; Petersen et al., 2016), but pitfalls such as high cost and invasiveness make these tests unsuitable for a populationwide screening.

Recently, researchers focused their interests on the retina in order to find new biomarkers for AD. The rationale behind this choice rely on the fact that, the retina and the brain share a common embryological origin therefore they could share pathological mechanisms as well (Sernagor et al., 2001; Kavcic et al., 2011; Chang et al., 2014; Lee et al., 2014; Bambo et al., 2015; Javaid et al., 2016). Recent studies have shown that anatomical alterations such as thinning of the ganglion cell and retinal nerve fiber layers can be detected already during AD early stages (Thomson et al., 2005; Paquet et al., 2007) thus strengthening the idea that the retina could be used in early AD diagnosis. However, far from reaching a consensus, studies on retinal tissues from AD patients and mouse models produced contrasting results. For instance, amyloid plaques have been found in AD post-mortem retinas (but not in control cases) by Löffler et al. (1995). Similar results have been confirmed by other research groups (Hoh Kam et al., 2010; Alexandrov et al., 2011; Koronyo-Hamaoui et al., 2011; Koronyo et al., 2017), in addition a positive correlation between inclusion body and cortical amyloid burden has been observed (Snyder et al., 2016). On the contrary, den Haan and collegues detected a signal for phosphorilated Tau (pTau) in the inner and outer plexiform layer of the retina (IPL and OPL respectively), but they did not find any signal for Aβ plaques and neurofibrilary tangles (den Haan et al., 2018). Most likely, these discrepancies may be attributed to differences in staining methods, sections preparation and time of tissue harvesting. Moreover, restricting retinal analysis to the presence of protein aggregates and vascular alterations solely may be poorly specific for AD diagnosis, as Aβ deposits have been found also in other pathological conditions such as macular degeneration (Zhao et al., 2015). Beyond classical neuropathological hallmarks, proteomic analysis of cerebrospinal fluid from AD patients revealed the upregulation of proteins related to microglia and astrocytes activation. For instance, Aβ plaques aggregation can activate astrocytes and microglia thus inducing the release of mediators of inflammation such as interleukin-1 beta (IL-1β) and ultimately cellular and neuronal apoptosis both in brain and retina (Liu et al., 2014).

In neurodegenerative disease, microglia response consists in migrating to sites of damage or injury, secreting numerous inflammatory molecules, and phagocytizing debris and aggregated proteins (Heneka et al., 2015; Song and Colonna, 2018). However, microglia, astrocytes and immune signaling are not just secondary players in disease processes but actually contribute to synaptic and neuron loss and buildup of pathogenic proteins even at the earliest stages of disease (Hong et al., 2016; Shi et al., 2017; Sosna et al., 2018). Aggregation of Aβ plaques can lead to inflammatory reactions, astrocytes and microglia activation, release of inflammatory cytokines such as interleukin-1 beta (IL-1β) and cellular and neuronal apoptosis in both the brain and retina (Liu et al., 2014). Moreover, the misregulated expression of glia released soluble factors in AD has been linked to disease associated microglia, and to microglia induced NLRP3 inflammasome and complement activation (Heneka et al., 2013; Sheedy et al., 2013; Lian et al., 2015; Iram et al., 2016) and synaptic loss (Hong et al., 2016). Moreover, a specific microglia phenotype has been described in neurodegenerative diseases characterized by the downregulation of microglia homeostatic genes, the upregulation of specific degeneration associated markers (DAM), and sustained by the upregulation of TREM2 (Yeh et al., 2017; Deczkowska et al., 2018).

To gain a better understanding and to identify new molecular targets for a more complete panel of putative biomarkers for AD diagnosis, we performed immunostaining on human retinas obtained from both AD patients and age matched controls. We report the presence of Aβ and pTau protein aggregates together with neuronal loss. We also report, for the first time, the upregulation of IL1-β on microglia and the presence of neurotoxic A1 astrocytes in AD retina. These findings highlight protein aggregates and cellular markers as targets to be considered for AD diagnosis.

## MATERIALS AND METHODS

## Human Samples

Human retinal slices from AD patients and age matched controls were purchased from Human Eye Biobank for Research, St Michael Hospital, Toronto, Canada. Prior to death, donors signed informed consent for autopsy and use of tissue and medical records for research purposes. No documented history of eye disease was reported for the AD or control cases. The use of human tissue has been approved by the Ethics Committee of Fondazione Santa Lucia I.R.C.C.S. to GR.

## Immunofluorescence

fnins-13-00925 August 31, 2019 Time: 18:20 # 3

Slices were deparaffined with absolute Xylene (3 × 5 min) and rehydrated by scaling down from 100% to 50% ethanol, 2 × 10 min for each concentration. After PBS washes (3 × 5 min), antigen retrieval has been performed by soaking slices for 40 min in a warm solution containing (in mM): 10 Na-citrate, 0.05% Tween 20, pH 6.0, 90◦C. Subsequently, slices were incubated for 45 min in a blocking solution (3% goat serum and 0.3% Triton X-100 in PBS). Primary antibodies diluted in blocking solution were incubated over-night at 4◦ [anti-βTubulin III, T2200, Sigma, clone Aa441-450, 1:500; anti-GFAP, MAB360, Millipore, clone GA5; 1:400; anti-Iba1, 019-19741, Wako, 1:400; anti-OPN, 691302, Biolegend, clone A15059B, 1:100; Anti-βAmyloid, 8243s, CELL SIGNALING, clone D54D2, 1:100; anti-Cleaved-Casapase3, 9661s, CELL SIGNALING, clone Asp175, 1:100; anti-PhosphoTau (Thr212, Ser214), MN106, INVITROGEN clone AT100 1:50; anti-PhosphoTau (Ser 202, Thr205), INVITROGEN, clone AT8, 1:40; anti-C3d, A0063, Dako, 1:500; anti-IL-1β, sc-32294, Santa Cruz Biotechnology, clone E7-2-hIL1β, 1:50]. After three washes in PBS, sections were stained for 45 min with a secondary antibody and Hoechst in order to visualize nuclei, then coverslips were mounted with Diamond Antifade Mountant (Molecular Probes) and images acquired with a confocal microscopy.

## Confocal Spinning Disk and VCS Microscopy

The acquisition of the images was performed through a Nikon Eclipse Ti equipped with a X-Light V2 spinning disk combined with a VCS (Video Confocal Super resolution) module (CrestOptics) based on structured illumination and with a LDI laser source (89 North). The images were acquired by using Metamorph software version 7.10.2. (Molecular Devices) with a 60x PlanApo λ oil objective (1.4 numerical aperture) and sectioning the slice in Z with a step size of 0.2 µm for spinning disk and 0.15 µm for VCS to obtain a total Z-stack of about 10 µm. In order to achieve super-resolution, raw data obtained by the VCS module have been processed with a modified version of the joint Richardson-Lucy (jRL) algorithm (Ingaramo et al., 2014; Ströhl and Kaminski, 2015; Chakrova et al., 2016), where the out of focus contribution of the signal has been explicitly added in the image formation model used in the jRL algorithm, and evaluated as a pixel-wise linear "scaled subtraction" (Heintzmann and Benedetti, 2006) of the raw signal. The acquisitions obtained were transformed into a z-projection and then analyzed using the ImageJ software.

## Laser Scanning Acquisitions

For conventional confocal laser scanning analysis of retinal slices, images acquisition has been performed through a confocal laser scanning microscope (microscope (FV10i Olympus) equipped with a 60x water immersion objective. Acquired images were processed and analyzed off line using ImageJ. For each retina six images were acquired. For every image the maximum intensity projections of z-series stacks was created.

## Microglia Density Analysis

The number of Iba1<sup>+</sup> microglia cells has been reported as number of somas per acquired area (400 µm<sup>2</sup> ). To quantify microglia density, the same images were analyzed by Metamorph software. A z-projection based on the maximal intensity signal was obtained and after threshold setting, fluorescence intensity value has been recorded. Data are expressed as area occupied by fluorescent cells versus total slice area.

## Astrogliosis Analysis

Astrogliosis analysis has been performed by staining retinas section for GFAP and C3. Following threshold adjustment astrogliosis was quantified as fluorescence intensity above threshold. Data are expressed as area occupied by fluorescent cells versus total slice area. For the measurement of GFAP/C3 colocalization, the value of GFAP-overlapping the C3 signal was considered.

## Analysis of Neurodegeneration

Confocal images of cells positive for cleaved caspase 3, an effector enzyme of the apoptotic pathway, were analyzed by means of Metamorph software. We manually counted the number of cells positive for the signal of the cleaved caspase 3 within each image. Thereafter, this number has been divided by the total number of ganglion neurons counted in the region of interest (ROI) and thus expressed as percentage of degenerating ganglion cells in each slice.

## FISH

Sections were deparaffined with absolute Xylene (3 × 5 min) and partially rehydrated with 70% ethanol (2 × 10 min each) before proceeding with RNA FISH staining. RNA in situ hybridization was performed as described previously (Raj and Tyagi, 2010) with minor modifications: briefly, sections were incubated in Wash Buffer A (Stellaris, Biosearch Technologies) for 5 min and then, in the Hybridization Buffer supplemented with 2 mM VRC complex (Sigma, R3380) and 125 nM TREM2 FISH probes 3<sup>0</sup> -end labeled with Quasar 670 fluorophore (Biosearch Technologies, SMF-1063-5). Incubation was performed overnight at 37◦C in Top Brite automatic slide hybridizer (Resnova). After two washes in Wash buffer A for 30 min and one wash with Wash Buffer B for 5 min at 37◦C, Hoechst was added for 15 min at room temperature. Finally, slices were mounted with ProLong Diamond Antifade Mountant (Thermo Fischer Scientific Pm36961) and analyzed with a confocal laser scanning microscope. Immunofluorescence for IBA1 was performed sequentially to RNA FISH staining.

## Protein Aggregates

The size of the amyloid plaques and tau tangles was studied on images acquired with the confocal microscope and subsequently analyzed with the Metamorph program (version 7.6.5.0). The "Trace Region" function of this program makes it possible to surround the deposits of these proteins and to obtain their volume expressed in µm<sup>3</sup> . Their density has been studied with the ImageJ program through which it was possible to count the number of these aggregates within each slice analyzed.

## Statistics and Data Analysis

fnins-13-00925 August 31, 2019 Time: 18:20 # 4

Data are shown as the Mean ± SEM. Statistical significance between controls and AD patients was assessed with the nonparametric Mann–Whitney test or T-test as indicated. A pvalue < 0.05 was considered significant. All statistical analyses were done using Sigma Plot 11.0, Origin or Clampfit software.

## RESULTS

## Typical Aβ and pTau Protein Aggregates in AD Human Retina

The accumulation of Aβ and pTau aggregates in the retinal tissue has been reported on both AD human tissue (Koronyo-Hamaoui et al., 2011; Koronyo et al., 2017) and mouse models (Criscuolo et al., 2018; Grimaldi et al., 2018), thus suggesting a putative use of the eye as a valuable structure for the study and the diagnosis of AD and other neurodegenerative disease. However, due to the complex nature of the disease, and to the positive Aβ staining in macular degeneration (Lynn et al., 2017), a more comprehensive panel of biomarkers needs to be defined to avoid misleading AD diagnosis. Using human retinal slices from AD patients and controls (samples were obtained from the St. Michael's Hospital Human Eye Biobank), we first confirmed the presence of Aβ and pTau aggregates in the retinal layers.

We analyzed retinal cross sections from 10 clinically and neuropathologically confirmed AD patients (mean age ± SEM: 85.7 ± 2.7 years; range 71–98 years; 8 females and 2 males) and 10 healthy controls (mean age ± SEM: 75.3 ± 2.9 years; range: 65– 93 years; 6 females and 4 male). In all AD patients we detected retinal Aβ immunoreactivity and deposits both in the inner and outer layers (minimum diameter: 110 nm; **Figures 1A,B**). Aβ staining, evaluated as Aβ plaques number in a region of 400 µm<sup>2</sup> , was significantly higher in AD patients (10.4 ± 1.8 n = 45/5 fields/patients) compared to controls (3.8 ± 0.7 n = 44/5 fields/controls; p < 0.01; **Figure 1C**). Sparse and diffuse retinal Aβ deposits were found occasionally also in control slices (**Figure 1A**, left), as previously reported (Koronyo et al., 2017). Plaques volume measured on 3D stacks was higher in AD patients respect to controls as reported in the histogram in **Figure 1D**.

In order to assess the presence of hyperphosphorylated tau isoforms we performed immunofluorescence analysis using two different monoclonal antibodies directed against the hyperphosphorylated tau protein (pTau; clone AT-8 and clone AT-100, **Figure 1E**). Following the staining with the AT-100 clone we observed a strong and diffuse staining for pTau both in the IPL and OPL in the majority of AD patients (6/10). Indeed, pTau immunoreactivity was significantly higher in the AD retina compared to age matched controls as quantified by fluorescence intensity in each field of view (AD: 0.76 ± 0.11, n = 38/6 fields/patients; CTRL: 0.22 ± 0.06, n = 32/6 fields/controls; p < 0.005; FOV = 400 µm<sup>2</sup> ; **Figure 1F**), thus confirming the accumulation of neurofibrillary tangles deposits in AD patients retina (**Figure 1E**, right). Using the AT-8 clone we found discrete pTau immunofluorescence in the IPL and in the OPL only in two out of 10 patients analyzed; in these AD patients the number of pTau tangles was significantly higher respect to control (AD: 4.4 ± 1.7, n = 15/2 fields/patients; CTRL: 0.9 ± 0.4; n = 15/2 fields/controls; p < 0.05; not shown). Co-labeling the retina with AT-100 and TUJ1 as neuronal marker indicated that in AD retina pTau expression was mainly in the Retinal Ganglion Cells (RGC) of the inner layer (IL; **Figure 1G**).

These findings clearly demonstrate an increase in Aβ and pTau aggregates in the retina of AD patients respect to controls thus pointing at them as putative biomarkers for AD diagnosis.

## AD Human Retina Displays Neurodegeneration in the Ganglion Cell Layer

An increased cleavage of proteins such as APP and presenilins operated by caspase-3, has been associated with neurodegeneration in AD (Louneva et al., 2008). In a previous paper from our laboratory, we reported immunoreactivity for the cleaved caspase-3 at the level of the RGC layer of the 3xTg-AD mouse model (Grimaldi et al., 2018). A positive staining for caspase-3 is also present in the IL of human retinas (**Figure 2A**, green dots) in TUJ1 positive cells (red). In particular we found that the number of caspase-3 positive retinal ganglion cells in each field of view (FOV) examined was increased compared to age matched controls (AD: 11.4 ± 2.2%, n = 17/6 fields/patients; CTRL: 6.0 ± 1.0%, n = 18/6 fields/control; p < 0.05; **Figure 2B**) These results indicated that the AD retinal IL is more subject to neuronal death thus suggesting that visual defects and optic nerve thinning observed in AD may rely on retinal ganglion cells neurodegeneration.

## Increased Astrocytosis and Microglia Reactivity in AD Patients Retina

In AD brain, astrocytes have been found closely associated with fibrillar amyloid plaques suggesting that Aβ accumulation may serves as a cue for the activation of this cell type (Henstridge et al., 2019). Astrocytes activation can have a neuroprotective effect but it may trigger the release of pro-inflammatory species such as cytokines and chemokines (Stadelmann et al., 2002; Farina et al., 2007). To evaluate putative astrogliosis in the retina of AD patients we performed immunostaining for the GFAP. Analysis of acquired images showed a marked astrogliosis localized at the level of the ganglion cell layer. Astrogliosis may arise also as a consequence of aging however, the amount of astrocyte activation was more pronounced in the AD retina compared to controls (**Figure 3A**), as quantified by fluorescence intensity in each field of view (AD: 5.5 ± 1.1, n = 44/6 fields/patients; CTRL: 1.9 ± 0.3, n = 44/6 fields/controls; p < 0.001; **Figure 3B**).

As for astrocytes, also microglia activation can show a dual role in inflammatory processes. Increased microglia reactivity in the retina has been observed in post-mortem tissues from AD patients (Beach et al., 1989) and in the 3xTg-AD mouse model (Grimaldi et al., 2018). Immunostaining against Iba1 a

FIGURE 1 | Human AD retina displays β-amyloid and pTau aggregates. (A) Representative images of retinal slices from AD patients and control cases immunolabeled with anti-β-amyloid antibody (green) and Hoechst for nuclei visualization (blue); bar 20 µm. (B) Insert representing β-amyloid staining in AD retina at higher magnification; bar 20 µm (C) Number of β-amyloid plaques/field of view (∗∗p < 0.01 AD vs. Ctrl; t-test; n = 45/5 fields/patients). (D) Distribution of β-amyloid plaque volume measured in AD patients and control cases. (E) Representative images of retinal slices from AD patients and control cases immunolabeled with anti-p-tau AT-100 antibody (green) and Hoechst for nuclei visualization (blue); bar 20 µm. (F) Quantification of p-tau AT-100 area covered by fluorescent signal/field of view (∗∗∗p < 0.001 AD vs. Ctrl; t-test; n = 38/6 fields/patients). (G) Representative images of retinal slices from AD patients and control immunolabeled with anti-p-tau AT-100 (green) and anti-TUJ1 as RGC marker (red) at higher magnification (Hoechst for nuclei visualization in blue; bar 20 µm). IL, inner layer; OL, outer layer.

microglia marker, revealed that this cell type was mainly present in two layers: the inner plexiform and the outer plexiform layers (**Figure 3C**). Microglia cell density was increased in AD patients retina compared to age matched controls (AD: 3.9 ± 0.3; n = 73/10 fields/patients; CTRL: 2.5 ± 0.3, n = 73/10 fields/controls; p < 0.005; **Figure 3D**).

These results, confirming the presence of altered astrocyte and microglia density in human AD retina, suggest the possibility to add markers for glial activation in the set of retinal biomarkers for AD diagnosis.

## AD Human Retina Displays Upregulation of Neurotoxic Microglia and Astrocytes

Several different proteins have been demonstrated to be involved in the interplay among neurons, astrocytes and microglia in neurodegenerative diseases. Particularly, altered expression of these proteins in AD has been linked to disease associated microglia, to microglia induced detrimental astrocytes activation (Heneka et al., 2013; Sheedy et al., 2013) and synaptic loss (Hong et al., 2016). Here we examined and compared the expression of IL-1β, TREM-2 (triggering receptor expressed on myeloid

n = 73/10 fields/patients). IL, inner layer; OL, outer layer.

cells-2), complement component C3 and OPN (Osteopontin) between AD and control samples.

Immunofluorescence analysis of IL-1β expression on AD retinal slices showed enhanced IL-1β cytoplasmic staining in the inner nuclear layer (INL) of AD patients compared to age matched controls (**Figure 4A**, left). IL-1β immunostaining in AD patients retina was found to be colocalized with Iba1-expressing cells, a marker of microglia, but not with the astrocytes marker GFAP (not shown). Indeed, co-labeling with Iba1 (**Figure 4A**, right) revealed that the number of microglia cells positive for IL-1β was significantly higher compared to control (AD: 3.5 ± 0.4, n = 31/5 fields/patients; CTRL: 2.2 ± 0.4, n = 29/5 fields/controls; p < 0.05; **Figure 4B**; average number of Iba1/IL-1β + cells for each FOV), with almost all microglia cells expressing IL-1β in AD.

Astrocytic C3 upregulation has been demonstrated to be downstream of the microglia induced IL1-β activation. Moreover, its expression seems to be regulated by β-amyloid in both AD mouse models and human brain (Lian et al., 2015). C3 upregulation is now considered to be a marker for detrimental A1 astrocytic phenotype taking part in AD related synaptic loss. We examined C3 expression in human retina by immunofluorescence analysis. We found C3 to be expressed in the ganglion cell layer. Co-staining with GFAP revealed that C3 immunoreactivity was confined in GFAP positive cells as shown in **Figure 4C**. Fluorescence intensity quantification showed that the percentage of C3 positive astrocytes was strongly upregulated in AD retina respect to control (AD: 26 ± 4, n = 45/6 fields/patients; CTRL: 7.8 ± 1.6, n = 46/6 fields/controls; p < 0.001; **Figure 4D**).

Upregulation of OPN in AD patients CSF may arise from neurons, microglia or both. Indeed, while the OPN encoding gene SPP1 (secreted phosphoprotein 1) is now considered a good DAM microglia marker (Ulland et al., 2017), AD brains display OPN upregulation in CA1 pyramidal neurons (Wung et al., 2007). We analyzed the expression level of OPN in human retinal slices of AD patients and age matched controls by immunofluorescence analysis using a monoclonal antibody against the full length OPN peptide (**Figure 5A**, left). We found significant upregulation of OPN expression in AD retinas, quantified as fluorescence intensity over threshold in each field of view (400 µm<sup>2</sup> ) (AD: 3.0 ± 0.6, n = 24/3 fields/patients) compared to age matched controls (CTRL: 0.7 ± 0.2, n = 27/4 fields/patients; p < 0.001) (**Figure 5B**). OPN expression was localized in the retinal ganglion cell layer as demonstrated by the co-staining against tubulin isoform βIII (TUJ1; **Figure 5A**, middle), a selective marker for RGC neurons. Indeed, the percentage of RGC neurons positive for OPN (**Figure 5A**, left) was significantly higher in AD respect to control retinas (AD: 84 ± 5%, n = 36/6 fields/patients; CTRL: 63 ± 6%, n = 34/6 fields/patients; p < 0.01; **Figure 5C**). Conversely, co-labeling with anti-Iba1 did not show upregulated OPN expression on microglia (data not shown).

TREM2 is now considered a crucial regulator in promoting microglia responses to Aβ in AD. TREM2 transcript has been shown to be over-expressed by microglia both in the brain of AD mouse models as well as human patients (Frank et al., 2008; Lue et al., 2015; Wang et al., 2015), suggesting that TREM2 upregulation could mirror with AD progression. Moreover, in the 3xTg-AD mouse model, we reported the upregulation of TREM2 mRNA in sorted retinal microglia cells (Grimaldi et al., 2018). Due to the lack of specific antibodies against human TREM2, FISH experiments have been performed in order to evaluate the expression level of TREM2 transcript in retinal slices of AD patients and age matched controls (**Figure 5D**). The analysis of TREM2 RNA levels obtained with fluorescence threshold analysis did not show, on average, any difference between AD and control retinas (AD: 0.18 ± 0.03, n = 31/4 fields/patients; CTRL: 0.22 ± 0.06, n = 37/4 fields/controls; p = 0.3; data not shown). However, the number of TREM2 positive cells in each field of view (400 µm<sup>2</sup> ) was higher in AD retina respect to control (AD: 8.5 ± 0.8, n = 31/4 fields/patients; CTRL: 6.2 ± 0.6, n = 37/4 fields/controls; p < 0.05; **Figure 5E**), indicating an increased TREM-2 expression. When TREM2 FISH analysis was performed together with immunofluorescence for Iba1, we observed Iba1 positive cells expressing TREM2 in AD retina (**Figure 5D**, left). It should be noticed, however, that in our experiments TREM2 staining was not exclusively confined to Iba1-positive cells. This could be ascribed to protein denaturation taking place during FISH protocol, preventing an absolute quantification of TREM2 expressing retinal microglia cells.

These data indicate that detrimental astrocytic and microglia activation can be detected in AD patients retina with increased release of pro-inflammatory compounds that could be considered as possible biomarkers for AD diagnosis.

## DISCUSSION

As of now, a definitive AD diagnosis is made possible only after post-mortem examination of brain tissue (Elahi and Miller, 2017). Despite many attempts there is an urgent need to find new easy-to-acquire, cost-effective strategies for AD diagnosis, even at early stages in order to allow a timely and effective therapeutic program (Frisoni et al., 2017).

The retinal tissue is now considered as a very promising structure to be used in searching of new AD biomarkers as it is thought that it could mirror the pathological changes happening in the brain during the disease progression. Even more, the retina is an accessible structure and this anatomical feature could be of extreme importance in the development of new imaging methods for its examination.

However, Aβ and pTau protein aggregates could not be considered specific markers to AD therefore, a more comprehensive panel of biomarkers is needed.

For this reason, we used post-mortem retinal slices from AD patients to investigate the presence of classical AD features and the expression of specific neuron-to-glia signaling proteins found both in the CSF of AD patients and in the brain of AD mouse models (Doens and Fernández, 2014). We here report that AD patients retina show, in addition to the presence of Aβ plaques and pTau tangles, ganglion neuron degeneration, astrogliosis, microglia activation, and up-regulation of specific disease associated neuron-to-glia signaling proteins, such as IL-1β, C3, OPN and TREM2.

In agreement with previous results (Koronyo et al., 2017; den Haan et al., 2018), we observed increased Aβ deposits and diffuse spreading of pTAu signals in the inner retinal layer of AD patients compared to age-matched controls. However, it has to be noticed that only a subset of AD retinal slices here analyzed were positive for pTau staining with AT-8 and AT-100 antibodies, this could be ascribed to pTau specific topographic distribution in human retina; indeed pTau staining was clearly reported in the anterior part of the superior retina (Koronyo et al., 2017; den Haan et al., 2018), and using transverse retinal slices obtained from the Human Eye Bank we could not asses which part of the retina we were analyzing. Due to this technical issue it is plausible that we are underestimating the amount of pTau in the AD patients retina. Also, we show that neuronal apoptosis is significantly higher in the retina of AD patients compared to age matched controls, predominantly in RGC layer.

As it is well known that glial cells are activated in AD associated inflammatory states, we performed immunostaining to

evaluate putative difference between controls and AD samples. Although GFAP and Iba1 reactivity was present both in AD and aged controls, the area of GFAP positive staining and the number of microglia cells were significantly higher in AD retinas thus indicating that glia (both astrocytes and microglia) activation is significantly more pronounced in AD retinas compared to controls.

We show here for the first time that retinal microglia of AD patients display, respect to their age matched controls, higher expression of IL-1β a typical marker of pro-inflammatory and DAM microglia (Keren-Shaul et al., 2017) suggesting microglia response to Aβ and pTau accumulation. These results are in line with the increased expression of IL-1β found in brain microglia in human patients and mouse models (as reviewed in Shaftel et al., 2008). Conversely, while in the brain of 5xFAD mouse model IL-1β was overexpressed by astrocytes (Rosenzweig et al., 2019), in human AD and control retina we did not observe IL-1β expression in GFAP positive cells. It should be noted that we found enhanced IL-1β staining also in other cells in the INL of AD retina, probably due to the NLRP3 inflammasome activation and monocytes infiltration.

We also report an increase of TREM2 mRNA level in the retina of AD patients respect to aged controls. The importance of TREM2 in the central nervous system (CNS) is widely recognized, as TREM2 mutations are linked to an increased risk of developing several neurodegenerative diseases (Ulland and Colonna, 2018), and TREM2 RNA has been found to be upregulated in the brain of patients and mouse models (Keren-Shaul et al., 2017). Moreover, we previously reported TREM2 upregulation in retinal microglia sorted from early symptomatic 3xTg-AD mice (Grimaldi et al., 2018). However, while it is evident that TREM2 expression on microglia cells plays a prominent role in driving microgliosis in the brain of AD mouse models and patients (Keren-Shaul et al., 2017), we did not find increased expression of TREM2 transcript on Iba1 positive cells in the retina of AD patients. It can be speculated that in our samples the expression of TREM2 on microglia cells is underestimated due to technical reasons. Indeed, RNA FISH experiments on post-mortem paraffinembedded tissue is per se technically challenging, requiring the use of formamide to increase the hybridization efficiency (Cattoretti et al., 1993; Gilbert et al., 2007). This procedure might

induce protein denaturation causing the failure of subsequent immunofluorescence analysis. Another possible explanation may rely on the age of AD patients here analyzed. Indeed, while microglia TREM2 expression is important in triggering and sustaining the DAM phenotype in response to protein aggregates during disease progression (Song and Colonna, 2018; Ulland and Colonna, 2018), downregulation of TREM2 has been reported in human post-mortem retinal tissues of patients suffering from Age-Related Macular Degeneration (Bhattacharjee et al., 2016) where Aβ deposits are present. In this framework it is likely that post mortem retinal tissue from aged patients is not the ideal specimen for TREM2 analysis. Moreover, the observed increase in TREM2 transcript level could be due to infiltrating monocytes negative for Iba1, as reported in the brain of AD patients (Fahrenhold et al., 2018).

The analysis of osteopontin expression in retinal tissue revealed increased OPN staining in AD retinal ganglion cells with evident colocalization with tubulin β-III. This result is partially in contrast with previous findings in the AD brain. Indeed, although OPN transcript (SPP1) expression has been shown to be upregulated in microglia in AD models (Ulland et al., 2017) and thus considered a marker for DAM phenotype, we did not find OPN upregulation on Iba1 positive cells. On the other hand our result is in line with data reporting that the protein osteopontin is upregulated on CA1 hippocampal pyramidal neurons in human AD brain (Wung et al., 2007). Moreover, in mouse retina, OPN is reported to be expressed on RGC (Ju et al., 2000; Duan et al., 2015). These considerations make OPN a good candidate biomarker for retinal AD diagnosis despite its cellular localization.

Here, for the first time we report the presence of A1 astrocytic phenotype in the retina of AD patients, as revealed by the strong upregulation of C3 protein on retinal astrocytes. This result is consistent with the reported presence of A1 reactive brain astrocytes in neurodegenerative diseases (Liddelow et al., 2017). The importance of reactive A1 astrocytes in neurodegenerative diseases rely on the evidence that synaptic loss can be favored by element of the complement cascade released by the A1 astrocytes (Stevens et al., 2007; Hong et al., 2016; Sekar et al., 2016). It is noteworthy that A1 activation may be initiated by microglia released IL-1β as well as by extracellular Aβ accumulation (Lian et al., 2015; Liddelow and Barres, 2017), both found and here reported in retina of AD patients.

These observations further support the possibility that ocular biomarkers could be used for early detection of AD associated neurodegeneration. It should be noted, however, that other ocular pathologies, such as glaucoma shares histopathological hallmarks with AD including increased levels of tau protein and microglial activation (Ramirez et al., 2017). It is therefore desirable that a

## REFERENCES

Ahmed, M. E., Iyer, S., Thangavel, R., Kempuraj, D., Selvakumar, G. P., Raikwar, S. P., et al. (2017). Co-localization of glia maturation factor with nlrp3 inflammasomeand autophagosome markers in human Alzheimer's disease brain. J. Alzheimers. Dis. 60, 1143–1160. doi: 10.3233/JAD-170634

complete panel of biomarkers able to discriminate between agerelated and disease-related retinal changes would be available and could be used as a target for in vivo imaging through a retinal scan. However, this will requires the development of both specific AD biomarkers ligands and long working distance high resolution imaging techniques, in order to achieve a non-invasive and inexpensive diagnosis of AD through the retinal scan.

## DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

## ETHICS STATEMENT

Human Subject Research: The studies involving human participants were reviewed and approved by the Ethics Committees Fondazione Santa Lucia IRCCS. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

## AUTHOR CONTRIBUTIONS

AG, NP, and FO designed, carried out, and analyzed the immunofluorescence experiments on protein aggregates and neurodegeneration. MR and RP designed, carried out, and analyzed the immunofluorescence experiments on astrocytes and microglia. MG designed, carried out, and analyzed the VCS and spinning disks acquisition. TS and AG designed, performed, and analyzed the FISH experiments. SD wrote the manuscript with the help of RP, GR, DR, and CL. SD conceived the project.

## FUNDING

This work was supported by the CrestOptics-IIT JointLab for Advanced Microscopy to AG, RP, GR, and SD, the MARBEL Life2020 grant to SD, and the SynaNet H2020 Program to AG, CL, SD, and DR.

## ACKNOWLEDGMENTS

The authors wish to thank the Imaging Facility at Center for Life Nano Science (IIT@Sapienza) for the support and technical advice.

Alexandrov, P. N., Pogue, A., Bhattacharjee, S., and Lukiw, W. J. (2011). Retinal amyloid peptides and complement factor H in transgenic models of Alzheimer's disease. Neuroreport 22, 623–627. doi: 10.1097/WNR.0b013e3283497334

Alzheimer's Disease International World Alzheimer Report, (2009). Alzheimer's Disease International World Alzheimer Report. London: Alzheimer's Disease Internationa.



**Conflict of Interest Statement:** MG is currently employed at CrestOptics. This study received funding from the JointLab between Istituto Italiano di Tecnologia and CrestOptics.

The remaining 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 © 2019 Grimaldi, Pediconi, Oieni, Pizzarelli, Rosito, Giubettini, Santini, Limatola, Ruocco, Ragozzino and Di Angelantonio. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Risk Factors for Alzheimer's Disease: Focus on Stress

*Alessandra Caruso1, Ferdinando Nicoletti1,2, Alessandra Gaetano1 and Sergio Scaccianoce1\**

*1 Department of Physiology and Pharmacology, Sapienza Università di Roma, Rome, Italy, 2 Neuropharmacology Research Unit, I.R.C.C.S. Neuromed, Pozzilli, Italy*

In vulnerable individuals, chronic and persistent stress is an established risk factor for disorders that are comorbid with Alzheimer's disease (AD), such as hypertension, obesity and metabolic syndrome, and psychiatric disorders. There are no disease-modifying drugs in the treatment of AD, and all phase-3 clinical trials with anti-amyloid drugs (e.g., β- or γ-secretase inhibitors and monoclonal antibodies) did not meet the primary endpoints. There are many reasons for the lack of efficacy of anti-amyloid drugs in AD, the most likely being a late start of treatment, considering that pathophysiological mechanisms underlying synaptic dysfunction and neuronal death begin several decades before the clinical onset of AD. The identification of risk factors is, therefore, an essential step for early treatment of AD with candidate disease-modifying drugs. Preclinical studies suggest that stress, and the resulting activation of the hypothalamic–pituitary–adrenal axis, can induce biochemical abnormalities reminiscent to those found in autoptic brain samples from individuals affected by AD (e.g., increases amyloid precursor protein and tau hyperphosphorylation). In this review, we will

#### *Edited by:*

*Cesare Mancuso, Catholic University of the Sacred Heart, Italy*

#### *Reviewed by:*

*Bruno P. Imbimbo, Chiesi Farmaceutici (Italy), Italy Fabricio Ferreira de Oliveira, Elysian Clinic, Brazil*

#### *\*Correspondence:*

*Sergio Scaccianoce sergio.scaccianoce@uniroma1.it*

#### *Specialty section:*

*This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology*

*Received: 24 May 2019 Accepted: 31 July 2019 Published: 10 September 2019*

#### *Citation:*

*Caruso A, Nicoletti F, Gaetano A and Scaccianoce S (2019) Risk Factors for Alzheimer's Disease: Focus on Stress. Front. Pharmacol. 10:976. doi: 10.3389/fphar.2019.00976*

critically analyze the current knowledge supporting stress as a potential risk factor for AD.

Keywords: stress, glucocorticoids, Alzheimer's disease, risk factor, animal model

## INTRODUCTION

According to the World Health Organization, "a risk factor is any attribute, characteristic or exposure of an individual that increases the likelihood of developing a disease or injury" (www.who. int/topics/risk\_factors). Alzheimer's disease (AD) is a neurodegenerative disorder characterized by progressive impairments in cognitive functions (Hampel et al., 2015). AD is characterized by loss of neurons and synapses in the cerebral cortex and hippocampus (Nisticò et al., 2012). Formation of aggregates of the β-amyloid peptide (Aβ1-42) and neurofibrillary tangles resulting from tau protein hyperphosphorylation are the major hallmarks of AD. These histopathological processes occur in brain regions that are involved in memory formation and emotional regulation (Gómez-Isla et al., 1996; Murray et al., 2006; Holtzman et al., 2011). The hippocampus is particularly vulnerable to AD-associated neuronal damage (Mu and Gage, 2011; Hollands et al., 2016). Genetic studies of early-onset familial AD (eFAD) have demonstrated that the formation of Aβ1-42 aggregates, rather than tau hyperphosphorylation, lies at the core of AD. eFAD is caused by mutations in the genes encoding for amyloid-ß precursor protein (APP) (Goate, 2006), presenilin 1 (PSEN1) (Sherrington et al., 1995), and presenilin 2 (PSEN2) (Levy-Lahad et al., 1995; Rogaev et al., 1995) inherited as an autosomal dominant trait (Guerreiro et al., 2012). PSEN1 mutations account for most eFAD, while APP and PSEN2 are rarer. However, these findings have constituted the bases that led to the proposal of the so-called "amyloid cascade hypothesis," which posits that dysregulation of amyloid-ß (Aß) peptide production and/or proteolytic degradation plays a key role in triggering the pathological and behavioral changes observed in AD patients (Selkoe and Hardy, 2016). Although our knowledge of neuropathological and neurochemical alterations associated with AD has impressively increased in the last decades, the current treatment is limited to cholinesterase inhibitors and the N-methyl-D-aspartate (NMDA) receptor channel blocker, memantine. None of these drugs can slow the progression of AD. Several putative disease-modifying drugs have been developed and continue to be developed with the hope of restraining the progression of the disease. Most of these drugs target either the production or the aggregation process of Aβ1-42 (Anand et al., 2017). Results of clinical studies with all these drugs have been highly disappointing. For example, a recently concluded randomized clinical trial with an inhibitor of β-secretase (BACE1), the enzyme that cleaves APP to uncover the N-terminus domain of Aβ1-42, did not show any reduction in cognitive or functional decline in AD patients, suggesting that either disease progression does not rely exclusively on amyloid formation or, alternatively, that anti-amyloid drugs should be administered several years prior to the onset of AD to be effective (Egan et al., 2018). The entire AD community was frustrated by the lack of efficacy of aducanumab, an anti-amyloid monoclonal antibody that was considered as highly promising based on a phase 1b clinical trial (Sevigny et al., 2016). If these drugs fail because treatment starts too late, i.e. when pathophysiological mechanisms of AD are already established, research should be directed to the identification of risk factors that can reliably predict the development of AD. As highlighted above, a minority of patients has eFAD with autosomal dominant transmission. Children have 50% chance to inherit the same mutation, and they are natural candidates for early treatment with candidate disease-modifying drugs. Apolipoprotein E4 (ApoE4) is the most established risk factor for sporadic AD (besides age), and subjects who are homozygous for ε4 (the gene encoding for ApoE4) and showed brain amyloidosis by PET scanning at an early age are also candidates for early treatment. The presence of ApoE4 may also predict responses to drug treatment in AD. For example, inhibitors of angiotensin-converting enzyme (ACE) improve cognition in patients affected by AD carrying ApoE4 and certain ACE polymorphisms (de Oliveira et al., 2014; de Oliveira et al., 2018). However, only about half of AD patients are ApoE4-positive, and the presence of cerebral amyloidosis is only suggestive of later development of AD (old individuals may have cerebral amyloidosis without AD).

Cardiovascular and metabolic disorders, such as hypertension, type-2 diabetes, metabolic syndrome, hypercholesterolemia, unhealthy dietary pattern, poor physical and cognitive activity, and smoking may increase the vulnerability to develop AD (Barnard et al., 2014; Xu et al., 2015). This review aims to comment on preclinical and clinical data on stress and glucocorticoids as risk factors for AD. Stress activates the hypothalamic–pituitary– adrenal (HPA) axis, with an ensuing increase in blood levels of glucocorticoid hormones (cortisol in humans and corticosterone in rodents). Hypothalamic corticotrophin-releasing hormone (CRH) is the main secretagogue of adrenocorticotropic hormone (ACTH) from the pituitary gland. ACTH, in turn, stimulates the production of glucocorticoids from the adrenal cortex. Glucocorticoids exert a crucial role in the adaptive physiological and behavioral responses to stress. Moreover, glucocorticoid hormones exert a negative feedback signal capable of inhibiting the activation of the HPA axis: the main targets of glucocorticoid-induced negative feedback are the anterior pituitary, the hypothalamus, and the hippocampus. Glucocorticoid binds to two receptors: the mineralocorticoid receptor (MR) and the glucocorticoid receptor (GR). Both are ligand-dependent transcription factors. Of the two receptors, MRs have one order of magnitude higher affinity for glucocorticoids than GRs. At low levels of circulating glucocorticoids, e.g., during the circadian nadir, MRs are fully occupied; in contrast, GR activation occurs at the circadian peak of glucocorticoids or in response to stressful events. Interestingly, both MRs and GRs are highly expressed in pyramidal neurons of CA1 and CA2 and in granule cells of the dentate gyrus of the hippocampus **(**Han et al., 2005), which is a vulnerable brain region in AD (Henneman et al., 2009). It has been hypothesized that long-lasting stress and the resulting sustained hypocortisolemia could be a potential neurodegenerative factor for the hippocampus (Angelucci, 2000). However, recent findings have depicted a more complex relationship between stress and neurodegeneration.

## HYPOTHALAMIC–PITUITARY–ADRENAL AXIS DYSFUNCTION IN ALZHEIMER'S DISEASE

Clinical reports of hypercortisolism in AD patients suggest a causal role for glucocorticoids in AD (Bruno et al., 1995; Hatzinger et al., 1995; Greenwald et al., 1986; Peskind et al., 2001; Rasmuson et al., 2001; Wilson et al., 2003; Hoogendijk et al., 2006; Johansson et al., 2010; Curto et al., 2017; Ouanes and Popp, 2019). However, it should be considered that some degree of stress could be present in a condition involving bodily or psychic suffering, especially when patients are cognitively able to perceive memory impairment, which is among the first symptoms reported by patients suffering from AD (Saydak et al., 1987). Dysregulation of the corticotropic axis is present in individuals suffering from depression, diabetes, and metabolic syndrome. These clinical conditions have been hypothesized to increase the risk to develop AD later in life (Ownby et al., 2006; Huang et al., 2014; Rojas-Gutierrez et al., 2017). In particular, it has been reported that patients who experienced late-life, but not early- or mid-life, depression had a two-fold increased risk for AD (Barnes et al., 2012; Singh-Manoux et al., 2017). Single nucleotide polymorphism (SNP) analysis in patients affected by AD supports the hypothesis that elevated glucocorticoid levels increase the risk to develop AD. de Quervain et al. (2004) analyzed SNPs in 10 glucocorticoid-related genes in 814 AD patients. They found an association between AD and a rare haplotype in the 5' regulatory region of the gene encoding for type-1 11ß-hydroxysteroid dehydrogenase (11ß-HSD1). 11ß-HSD1, also known as cortisone reductase, catalyzes the conversion of cortisol into the biological inert 11-keto derivative (cortisone). Thus, subjects carrying this rare haplotype with reduced 11ß-HSD1 transcription show less inactivation of glucocorticoids, which, in turn, is associated with an increased vulnerability to the clinical manifestation of AD. On the contrary, subjects bearing a polymorphism of the GR Caruso et al. Stress and Alzheimer's Disease

gene (NR3C1) are characterized by a reduced risk to develop AD (van Rossum et al., 2008). More precisely, carriers of the ER22/23EK allele (approximately 7% of the entire population) were associated with a decreased risk of developing dementia. The presence of the ER22/23EK allele leads to a decreased sensitivity of GRs to glucocorticoids (Russcher et al., 2005).

Several lines of evidence suggest a tight connection between neuroinflammation and AD (see Nichols et al., 2019 for a recent review). In a double-blind, placebo-controlled trial, 138 AD patients received prednisone (10 mg daily for 1 year). Glucocorticoid treatment not only failed to ameliorate cognitive decline as assessed by the Alzheimer's Disease Assessment Scale but also caused a greater behavioral decline, as measured by the Brief Psychiatric Rating Scale (Aisen et al., 2000). These findings suggest a detrimental effect of glucocorticoids in AD. Some clinical trials have investigated the effects of the glucocorticoid receptor antagonist, mifepristone, in AD patients (Belanoff et al., 2002; DeBattista and Belanoff, 2005). Although a significant improvement of cognitive function was observed in AD patients after a 6-week treatment with 200 mg of mifepristone (Pomara et al., 2002), there are no ongoing clinical trials with glucocorticoid antagonists in AD.

## PRECLINICAL STUDIES ON STRESS AS RISK FACTOR FOR ALZHEIMER'S DISEASE

Preclinical studies aimed at elucidating the role of glucocorticoids as a risk factor for AD have been mostly conducted in transgenic (Tg) mice, such as Tg2576 mice expressing human APP carrying the Swedish mutation (KM670/671NL), mice with a double mutations of APP and PSEN1, and 3xTgAD mice, characterized by a triple mutations of APP (Swedish mutations), PSEN1 (M146V), and the P301L mutation in the gene encoding tau protein (MAPT) (Götz et al., 2018). However, it is important to note that transgenic AD mice recapitulate features of eFAD that represents only 3% of AD (Bird, 1999), with the important limitation of the limited lifespan of mice. Rats have been used for the induction of "AD-like" pathology using i.c.v. or intrahippocampal injection of Aβ1-42 oligomers, tau protein, or excitotoxins (see Shree et al., 2017 for a recent review). In both Tg and non-Tg models, the effect of stress was investigated by either exposing animals to stress of variable duration or administering glucocorticoids (the natural hormone, corticosterone, or the synthetic, long-acting, and GR-selective glucocorticoid, dexamethasone). Some studies have investigated the role of CRH independently of its function in the regulation of the HPA axis and the potential use of CRH receptor antagonists as disease-modifying drugs in AD (Rissman et al., 2012). One of the first demonstrations that stress hormones are linked to AD-like neuropathology was provided by the evidence that kainic acid-induced tau hyperphosphorylation was amplified by repeated (i.e., 7 days) corticosterone administration in rats (Elliott et al., 1993). Dexamethasone treatment in rats was also found to increase the expression of APP in the cerebral cortex, cerebellum, and brain stem (Budas et al., 1999). The effect of glucocorticoids on APP processing and Aβ1-42 production was also investigated in 3xTgAD mice, in which dexamethasone treatment for 7 days caused a significant increase in soluble and insoluble Aβ1-42 in the hippocampus, cortex, and amygdala, and also leads to the mislocalization of tau to the somatodendritic compartment (Green et al., 2006). Moreover, neuroblastoma N2A cells incubated with dexamethasone or corticosterone showed an increased expression of both APP and BACE leading to enhanced production of Aβ1-42. Interestingly, 3xTgAD mice showed an age-dependent increase in serum corticosterone levels, which is observable already at 9 months of age (Green et al., 2006). Although glucocorticoid administration mimics only partially the hormonal endpoint of stress-induced HPA activation, the above findings paved the way to explore the effect of stress (of different intensity and duration) on AD neuropathology. One of the most popular Tg mouse models of AD expresses human APP with the London mutation (V717I). Using this model, it has been demonstrated that exposure to longterm (8 months) restraint stress caused learning and memory deficits as well as an increase in extracellular amyloid plaque deposition and intraneuronal APP and Aβ1-42 immunoreactivity, and neurodegeneration in the hippocampus and cerebral cortex (Jeong et al., 2006). Tg2576 transgenic mice expressing human APP with the Swedish mutation (K670M/N671L) were used to study the effects restraint stress (2 h daily for 16 consecutive days) (Lee et al., 2009). Stress caused a rapid increase in plaque formation, insoluble Aβ accumulation, and dendritic atrophy of cortical neurons (Lee et al., 2009). Besides, restraint stress caused a down-regulation of matrix metalloproteinase-2 (MMP-2), which, similarly to MMP-9, is involved in the clearance of Aβ (Roher et al., 1994). In the same study, the authors have demonstrated that MMP-2 down-regulation and Aβ pathology were completely prevented by the administration of the CRH receptor antagonist, NBI 27914, reinforcing the hypothesis that over-activation of the HPA axis contributes to the development of stress-induced AD-like pathology. The hypothesis that a down-regulation of MMP-2 is a linking bridge between stress and AD pathology is supported by the evidence that i) MMP-2 expression was reduced in cortical neurons treated with corticosterone (Lee et al., 2009); ii) infusion of the MMP inhibitor, GM6001, increases Aβ formation in Tg2576 mice (Yin et al., 2006); and iii) MMP-2 was reduced in the parietal cortex of Tg2576 mice (Lee et al., 2009).

Tau hyperphosphorylation is a molecular hallmark present in both the hereditary and sporadic forms of AD. Hyperphosphorylated tau protein has a key pathogenic role in AD neuronal dysfunction because it accumulates in form of insoluble aggregates and neurofibrillary tangles with a consequent malfunction of axonal transport (Iqbal et al., 2010; Fitzpatrick et al., 2017). When exposed to dexamethasone, neuronal cell lines bioengineered to express the human homolog of the protein tau (PC12-htau) showed a greater degree of susceptibility to the neurotoxic actions of Aβ1-42 as well as marked increases in tau hyperphosphorylation at specific epitopes implicated in AD neuropathology. More specifically, exposure to dexamethasone reduced tau turnover and, consequently, increased cytoplasmic accumulation of tau. These effects were abolished by pharmacological blockade of GRs with mifepristone, indicating that activation of GRs mediates the effects of glucocorticoids on tau protein. Tau hyperphosphorylation was ultimately mediated by a GR-dependent activation of cyclin-dependent kinase 5 (CDK5) and glycogen synthase kinase-3β (GSK3β) (Sotiropoulos et al., 2008). The effect of stress on tau hyperphosphorylation has been extensively studied in recent years. Sotiropoulos et al. (2011) found that in Wistar rats, exposure to unpredictable chronic (1 month) stress induced tau hyperphosphorylation in the hippocampus and prefrontal cortex. In line with the hypothesis that glucocorticoids mediate the action of stress, these authors also demonstrated that treatment with dexamethasone for 14 days mimicked the amplifying effect of stress on Aβ-induced tau hyperphosphorylation. Both stress and glucocorticoid administration activated GSK3β and CDK5, as well as calciumcalmodulin-dependent protein kinase-II, the MAP kinase pathway, and the JUN kinase pathway in the hippocampus and prefrontal cortex, and caused an impairment in the hippocampus- and prefrontal cortex-dependent memory. Based on these findings, the authors concluded that sustained stress, *via* glucocorticoid hypersecretion, could influence the onset and progression of AD pathology and highlighted the role of tau hyperphosphorylation in the effect of stress on AD. CRF is the principal driving force, which controls both tonic and phasic activation of the HPA axis. However, the hypothesis that CRF might have a causative role in AD independently of ACTH and glucocorticoid secretion has been addressed in several studies. Rissman and colleagues have demonstrated that stress-induced tau hyperphosphorylation was not prevented by adrenalectomy, while it was absent in type-1 CRF receptor (CRFR1)-deficient mice and mice treated with a selective CRFR1 antagonist (antalarmin). This suggested that CRF induced tau pathology through a central mechanism independent of the activation of the HPA axis (Rissman et al., 2007). They used two mouse models of AD, i.e., Tg2576 mice, which express APPK670/671L, and PS19 mice, which express human P301S mutant tau. They also used two different stress protocols: chronic restraint stress (CRS) and chronic unpredictable stress (CUS), both delivered for 1 month. In both Tg2576 and PS19 mice, CRS, but not CUS, induced an increase in Aβ1-42 and hyperphosphorylated tau in the hippocampus and frontal cortex. Moreover, CRS, but not CUS, caused deficits in hippocampus-dependent memory. In apparent contrast with the glucocorticoid-centric hypothesis of stress and AD, PS19 mice implanted with a corticosterone pellet did not show increases in the levels of hyperphosphorylated tau. In contrast, injection of the CRF antagonist, NBI 27914, 15 min before the onset of restraint stress abolished tau accumulation and prevented memory impairment. The hypothesis of a central action of CRF in causing AD-like neuropathology was further supported by the demonstration that transgenic mice overexpressing CRF showed an increase in tau phosphorylation in the hippocampus, and CRFR1 ablation in Tg mice carrying a double mutation of APP and PS1 reduced Aβ accumulation in several brain regions (Campbell et al., 2015). Intriguing findings were reported by Kvetnansky et al. (2016), who used CRF knockout mice showing that CRF potentiated tau phosphorylation during acute stress, but inhibited phosphorylation in response to repeated stress. Although the precise mechanism(s) by which CRF may exacerbate AD neuropathology remains to be determined, studies in neuronal cultures have demonstrated that CRF-induced tau phosphorylation hampers neuronal energetics and interferes with axonal transport of mitochondria (Le et al., 2016).

Tau mislocation has recently been proposed as a relevant pathophysiological mechanism in AD (Hoover et al., 2010; Tai et al., 2012; Zempel et al., 2013; Le et al., 2016). A large body of evidence suggests that hyperphosphorylated tau causes a derangement of synaptic function with a resulting impairment of excitatory synaptic transmission (Ittner et al., 2010; Crimins et al., 2013; Xie et al., 2017), leading to deficit in learning and memory (Kimura et al., 2007). In mice overexpressing APP (APP23 mice) crossed with tau transgenic mice, a redistribution of hyperphosphorylated tau from axons to dendrites increased the localization of Fyn in the postsynaptic density. Fyn, in turn, phosphorylates the GluN2B subunit of NMDA receptors at Y1472, leading to excitotoxic downstream signaling (Ittner et al., 2010). A direct effect of glucocorticoids on tau mislocation has been studied by Pinheiro and colleagues (Pinheiro et al., 2016). In male Wistar rats, prolonged (14 days) dexamethasone exposure led to cytosolic and dendritic tau accumulation in the hippocampus, but, interestingly, Fyn levels were not altered. Additional evidence of a relationship between stress-induced glucocorticoid hypersecretion

Alzheimer's disease.

and synaptic tau missorting was provided by Lopes et al. (2016), who used wild-type and tau knockout mice. In wild-type mice, exposure to CUS for 6 months caused behavioral disturbances as well as synaptic tau missorting and enhanced levels of Fyn in hippocampal postsynaptic density fractions. None of these effects were observed in mice lacking tau. Interestingly, as opposed to wild-type mice, tau knockout mice did not show changes in plasma corticosterone levels in response to CUS as well as following an acute restraint stress application. Collectively, these findings suggest that the phosphorylation status of the tau protein exerts an important role in the relationship between sustained stress and AD synaptic pathology. If chronic glucocorticoid elevation represents a causative factor contributing or exacerbating the development of AD (see also Lopes et al., 2016; Fitzpatrick et al., 2017), this would offer a pharmacotherapeutic target for AD and other tauopathies. Several studies have shown that negative experiences during childhood not only increase the probability to develop anxiety, depression, and substance use disorder but also increase the vulnerability to several clinically relevant diseases (Dich et al., 2015; Berg et al., 2017). Experimental findings support the hypothesis that early life experiences can affect adrenocortical stress response in adult life, which in turn may cause cognitive dysfunction (Chen and Baram, 2016). For the relationship between early life events and AD, see the excellent review by Lesuis et al. (2018). Here, we focus on the principal findings related to perinatal stress and AD. Exposure of pregnant APPswe/PS1dE9 mice to restraint stress during the first week of gestation caused gender-dependent behavioral and histopathological changes in the offspring. Adult male offspring showed impairment in spatial memory, while females exhibited a better performance in a spatial memory task and, interestingly, a reduced plaque load in the hippocampus (Sierksma et al., 2013). Accordingly, the effects of early life stress on the developmental trajectory of the CNS have been often reported to be gender-dependent (Naninck et al., 2015; Loi et al., 2017). In male APPswe/PS1dE9 mice, early postnatal stress (from postnatal day 2 to 9), in the form of reduced availability of bedding and nesting material, increased plaque load and impairs synaptic plasticity in the adult life (Lesuis et al., 2019). Riluzole, a drug that reduces glutamate release, prevented the effects of early life stress when added to the drinking water from weaning onwards. The

## REFERENCES


effects of early life stress were also evaluated in wild-type rodents. In Wistar rats, daily maternal separation during the first 3 weeks of life induced in the adult male offspring cognitive deficits as well as increases in both Aβ40 and Aβ42 hippocampal levels in the adult male offspring. These effects were paralleled by an increased expression of BACE1 and hyperphosphorylated tau (Martisova et al., 2013). In contrast, exposure to an enriched and 'positive' environment during early postnatal life exerts protective effects against AD-related neuropathology and cognitive functions. In these studies, neonatal handling has been the most used experimental paradigm. Neonatal handling increases maternal care causing permanent neurochemical and behavioral alterations in the adult progeny (Meaney, 2001). Lesuis et al. (2017) have studied the effects of neonatal handling from postnatal days 2 to 9 in APPswe/PS1dE9 mice. In adulthood (11 months) mice subjected to neonatal handling showed a reduced amyloid load in the hippocampus paralleled by increased performance in learning paradigms (e.g., t-maze and contextual fear memory). In APP-V717I x Tau-T301P (biAT) bigenic mice, neonatal handling was shown to reduce hippocampal Aß accumulation and to prolong lifespan (Lesuis et al., 2016). Finally, 3xTg-AD mice daily handled from birth to weaning (postnatal day 21) showed reduced deficits in spatial learning and exploratory behavior (Cañete et al., 2015).

## CONCLUSIONS

In the last years, our knowledge on the pathogenetic mechanisms of AD has dramatically improved. Several preclinical studies have demonstrated that stress is a potential risk factor for AD (**Figure 1**). However, the marked individual difference in perceiving and coping with stress makes any generalization difficult at the moment. Nevertheless, we suggest that behavioral, psychological, or pharmacological strategies aimed at increasing resilience to stress might delay the onset or slow the progression of AD.

## AUTHOR CONTRIBUTIONS

SS, AG, and AC prepared the draft of the manuscript. FN reviewed the manuscript.

dementia: differential effects for Alzheimer disease and vascular dementia. *Arch. Gen. Psychiatry* 69, 493–498. doi: 10.1001/archgenpsychiatry.2011.1481


modulate long term survival and amyloid protein levels in a mouse model of Alzheimer's disease. *Oncotarget* 7, 39118–39135. doi: 10.18632/oncotarget.9776


microtubule severing by TTLL6 and spastin. *EMBO J.* 32, 2920–2937. doi: 10.1038/emboj.2013.207

**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 © 2019 Caruso, Nicoletti, Gaetano and Scaccianoce. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Family C G-Protein-Coupled Receptors in Alzheimer's Disease and Therapeutic Implications

#### *Ilaria Dal Prà\*, Ubaldo Armato and Anna Chiarini\**

*Human Histology and Embryology Unit, University of Verona Medical School, Verona, Italy*

#### *Edited by:*

*Silvana Gaetani, Sapienza University of Rome, Italy*

## *Reviewed by:*

*Fabio Tascedda, University of Modena and Reggio Emilia, Italy Wenhan Chang, University of California, United States*

*\*Correspondence:*

*Ilaria Dal Prà ippdalpra@gmail.com Anna Chiarini anchiari@gmail.com*

#### *Specialty section:*

*This article was submitted to Experimental Pharmacology and Drug Discovery, a section of the journal Frontiers in Pharmacology*

*Received: 11 April 2019 Accepted: 07 October 2019 Published: 28 October 2019*

#### *Citation:*

*Dal Prà I, Armato U and Chiarini A (2019) Family C G-Protein-Coupled Receptors in Alzheimer's Disease and Therapeutic Implications. Front. Pharmacol. 10:1282. doi: 10.3389/fphar.2019.01282*

Alzheimer's disease (AD), particularly its sporadic or late-onset form (SAD/LOAD), is the most prevalent (96–98% of cases) neurodegenerative dementia in aged people. AD's neuropathology hallmarks are intrabrain accumulation of amyloid-β peptides (Aβs) and of hyperphosphorylated Tau (p-Tau) proteins, diffuse neuroinflammation, and progressive death of neurons and oligodendrocytes. Mounting evidences suggest that family C G-protein-coupled receptors (GPCRs), which include γ-aminobutyric acid B receptors (GABABRs), metabotropic glutamate receptors (mGluR1-8), and the calcium-sensing receptor (CaSR), are involved in many neurotransmitter systems that dysfunction in AD. This review updates the available knowledge about the roles of GPCRs, particularly but not exclusively those expressed by brain astrocytes, in SAD/ LOAD onset and progression, taking stock of their respective mechanisms of action and of their potential as anti-AD therapeutic targets. In particular, GABABRs prevent Aβs synthesis and neuronal hyperexcitability and group I mGluRs play important pathogenetic roles in transgenic AD-model animals. Moreover, the specific binding of Aβs to the CaSRs of human cortical astrocytes and neurons cultured *in vitro* engenders a pathological signaling that crucially promotes the surplus synthesis and release of Aβs and hyperphosphorylated Tau proteins, and also of nitric oxide, vascular endothelial growth factor-A, and proinflammatory agents. Concurrently, Aβs•CaSR signaling hinders the release of soluble (s)APP-α peptide, a neurotrophic agent and GABABR1a agonist. Altogether these effects progressively kill human cortical neurons *in vitro* and likely also *in vivo*. Several CaSR's negative allosteric modulators suppress all the noxious effects elicited by Aβs•CaSR signaling in human cortical astrocytes and neurons thus safeguarding neurons' viability *in vitro* and raising hopes about their potential therapeutic benefits in AD patients. Further basic and clinical investigations on these hot topics are needed taking always heed that activation of the several brain family C GPCRs may elicit divergent upshots according to the models studied.

Keywords: Alzheimer's disease, G-protein-coupled receptors, amyloid-beta, calcium-sensing receptor, GABAB receptors, metabotropic glutamate receptors

## INTRODUCTION

Alzheimer's disease (AD), particularly its sporadic or late-onset form (SAD/LOAD), is by far the most prevalent cause of senile dementia in humans (Alzheimer's Association, 2018). Typically, multiple neurotoxic factors accumulate in the AD brain, such as soluble amyloid-β oligomers (sAβ-os) and insoluble Aβ fibrils (fAβs), the latter aggregating into senile plaques (Gouras et al., 2015); hyperphosphorylated soluble Tau oligomers (p-Tau-os) that collect into insoluble neurofibrillary tangles (NFTs) (Bloom, 2014); overproduced reactive oxygen species (ROS) (Butterfield and Boyd-Kimball, 2018); nitric oxide (NO); vascular endothelial growth factor-A (VEGF-A), and proinflammatory agents (Dal Prà et al., 2014a; Chiarini et al., 2016). Altogether, these neurotoxins cause a spreading neuroinflammation, progressive synaptic losses, and cortical human neurons and oligodendrocytes deaths with the consequent breaking up of neural circuits. The clinical counterparts of AD neuropathology are steadily worsening losses of memories and cognitive abilities, which inexorably lead to patients' demise (Dal Prà et al., 2015a; Dal Prà et al., 2015b; Calsolaro and Edison, 2016).

Amyloid precursor protein (APP), a multifunctional protein widely expressed in the central nervous system (CNS), represents the source of the neurotoxic sAβ-os and fAβs that progressively accumulate in AD brains. Transmembrane APP holoprotein can undergo alternative enzymatic handling: (i) *nonamyloidogenic processing* (*NAP*) by α-secretases that leads to the production of the soluble (s)APP-α while obstructing Aβs synthesis (Chiarini et al., 2017b; Rice et al., 2019) (**Figure 1**); and (ii) *amyloidogenic processing* (*AP*) by β-secretase (BACE1) and γ-secretase liberating Aβs (**Figure 2**). Notably, sAPP-α's physiological roles are multifaceted, and to-date only partly understood. The available evidence reveals that sAPP-α promotes the neural differentiation of human embryo stem cells (Freude et al., 2011) and protects hippocampal neurons from the harm due to ischemia (Smith-Swintosky et al., 1994), glucose deficiency (Furukawa et al., 1996), brain trauma, and excitotoxicity (Mattson et al., 1993; Goodman and Mattson, 1994). In addition, sAPP-α complexes with and inhibits the activity of BACE1/β-secretase protein thus hindering any excess production of toxic Aβ42/Aβ42-os (Stein and Johnson, 2003; Obregon et al., 2012; Peters-Libeu et al., 2015). Moreover, sAPP-α stimulates axonal outgrowth (Ohsawa et al., 1997), synaptogenesis, and synaptic plasticity (Hick et al., 2015; Habib et al., 2016). Remarkably, sAPP-α also curbs the activity of glycogen synthase kinase (GSK)-3β and the hyperphosphorylation and overrelease of neurotoxic p-Tau/p-Tau-os, the main components of NFTs (Deng et al., 2015). And an increased activity of GSK-3β has been linked to hyperphosphorylation of Tau in the brains of AD patients. Typically, in AD Tau is phosphorylated at over 30 serine/threonine residues by various protein kinases, including GSK-3β (Pei et al., 1999). The D1 and D6a domains of sAPP-α are the locations of its neuroprotective and neurotrophic activities since they stimulate axons outgrowth when added as separate fragments to *in vitro* hippocampal neurons (Jin et al., 1994; Qiu et al., 1995; Ohsawa et al., 1997). In keeping with such findings, sAPP-α upholds cognition and memory integrity in animal models of physiological aging and of AD (Roch et al., 1994; Meziane et al., 1998; Bour et al., 2004; Ring et al., 2007; Corrigan et al., 2012; Xiong et al., 2016) (**Figure 1**).

SAD/LOAD, which comprises ~98–96% of the cases, starts from neuronal nests in the layer II of the lateral entorhinal cortex (LEC) in the temporal lobe (Khan et al., 2014) where small ischemic areas may occur in aged subjects (Ishimaru et al., 1996). Thence, in the course of 20–40 years (*asymptomatic stage*) SAD/ LOAD silently spreads to wider and wider upper cerebral cortex areas, particularly to those involved in storage and retrieval of memories and in handling complex cognitive activities (Khan et al., 2014). When the unremitting attrition depletes the cortical human neurons' functional reserve, SAD/LOAD's first clinical symptoms start manifesting as amnesias. This marks the onset of the *amnestic minor cognitive impairment* or *aMCI stage* that lasts 3–5 years while its symptoms progressively worsen. Eventually, the *full symptomatic stage* takes over, whose exacerbating symptoms include permanent losses of short-term (first) and long-term (later) memories, changes in personality and behavior, loss of the several language-related abilities, failure to cope with daily tasks and needs, motor problems, cognitive shortfalls, dementia, and eventually death. However, it is still hard to diagnose the earliest asymptomatic stage of AD because specific biomarkers are few and the highly neurotoxic, synapsedestroying sAβ42-os are hardly detectable when senile plaques and NFTs are still absent (Selkoe, 2008a; Selkoe, 2008b; Ferreira and Klein, 2011; Klein, 2013; Dal Prà et al., 2015a). Even so, the ghostly sAβ42-os eventually cause a noticeable accumulation of Aβ42 as fibrils and senile plaques, and of p-Tau-os as NFTs (Medeiros et al., 2013). Presently, the diagnosis of SAD/LOAD is based upon detecting brain deposits of insoluble Aβs (senile plaques) *via* PET imaging and specific changes in Aβ42/Aβ40 and Tau/p-Tau ratios values in the cerebrospinal fluid (CSF), which are deemed to be pathognomonic (McKhann 2011). PET imaging can also detect the brain accumulation of NFTs (Hall et al., 2017). The quest of blood biomarkers of AD is still ongoing with some preliminary promising results (Nabers et al., 2018; Nakamura et al., 2018; Palmqvist et al., 2019).

Presently, no drug therapy modifies or mitigates AD's relentless course (Jessen et al., 2014). This unsatisfactory situation still lingers because of various reasons. First, SAD/ LOAD pathogenesis remains unclear and, hence, an open to speculation topic. Second, animal models closely mirroring human SAD/LOAD are as yet not available (Ameen-Ali et al., 2017). Transgenic (*tg*) animal (mostly rodent) AD-models only partially and imperfectly emulate the early-onset familial (EOF) AD variety, which comprises at most 2–4% of AD cases. It is well established that EOFAD results from mutations in the amyloid precursor protein (*APP*) or presenilin1 (*PSEN1*) or presenilin2 (*PSEN2*) genes. These mutations drive a constitutive, diffuse intrabrain overproduction and overload of sAβ-os, insoluble Aβ fibrils, and hence Aβ-heaped senile plaques and concurrently of p-Tau-os and NFTs. Conversely, no genetic mutations underlie SAD/LOAD pathogenesis, although *APOE* (Huang and Mahley, 2014) and *TREM2* (Gao et al., 2017) gene variants could increase AD proclivity. While Aβ-os appear as the first main AD drivers, in such *tg* AD-model animals p-Tau accumulates as NFTs later and only when a mutated *MAPT* transgene is inserted

holoprotein the soluble (s)APP-α peptide, whose multiple neurotrophic and neuroprotective effects are summarized in this figure. Recent evidence indicates that as a GABAB1aR agonist sAPP-α also constitutively moderates neuronal excitability thus preventing neurons' harm. In summary, APP holoprotein's *NAP* hinders the development of AD and preserves neuronal viability, trophism, and function.

too (Oddo et al., 2003; Cohen et al., 2013). *Third*, AD-model rodent brains substantially differ under several cytological, structural, and functional standpoints from the human brain. Some differences are obvious, such as brain size and weight, the cerebral cortex limited extent, the largely prevailing primitive olfactory cortex, and so on. Other differences are more subtle,

but still exceedingly important, as they critically regard a genetic homology of only 80% and structural and functional features of cortical neurons and neuroglial cells--e.g. the total absence of some human cortical astroglial subtypes from rodent brain cortices--and the dissimilar extension of the astrocytes' domains, and the unlike reactions (e.g. Ca2+ fluxes) of each neural cell

FIGURE 2 | The *amyloidogenic processing (AP)* of amyloid precursor protein (APP) holoprotein. In this pathway β-secretase/BACE1 and γ-secretase sequentially cleave APP holoprotein yielding several Aβ peptide isoforms. The two most prevalent Aβ isoforms are the 40- and 42-amino acid-long residues, the length of which is determined by the cleavage site of the γ-secretase. Under physiological conditions the synthesis of monomeric neurotrophic Aβ peptides is very limited. However, when over produced Aβ peptide monomers end up aggregating first into soluble oligomers (Aβ-os), the first Alzheimer's disease (AD) drivers, next into insoluble fibrils, and eventually into senile plaques. The latter can both take up and release the neurotoxic Aβ-os. The Aβ42 isoform is the main component of senile plaques as is it highly prone to oligomeric and polymeric (fibrillar) aggregation. The Aβ-os interact with several nerve cell membrane receptors, including the calcium-sensing receptor (CaSR). Notably, CaSR-bound Aβ-os trigger a complex set of intracellular signals that promote the development and progression of AD neuropathology (see Figure 3 for further details).

type once exposed to AD-driving neurotoxins. As a revealing example, in *tg* AD-model animals, the cortical astrocytes die sooner than neurons, whereas cortical neurons die earlier than astrocytes in human AD brains. Altogether, such a complex set of divergences has suggested that *tg* rodent AD-model animals may not be the ultimate means to identify therapeutic approaches benefiting human AD patients (Ransohoff, 2018). Hitherto, all drugs that resulted advantageous when given to *tg* AD-model animals have failed to act as beneficial therapeutics in human AD patients (Cummings, 2017). It should be noted that this is a currently recognized general problem affecting the quest for and successful trial of novel drugs preclinically tested in animal models of other human diseases besides AD (Gabrielczyk, 2019). Unquestionably, novel rodent and non-human primate (NHP) AD models are under development (Podlisny et al., 1991; Sasaguri et al., 2017) but their potential worth for human AD therapeutic research remains to be assessed.

So, are there acceptable alternatives to AD-model animals? Given the just mentioned species-related differences, one should take stock of untransformed human neural cells making up "*preclinical AD models in Petri dishes*". An example of this kind may be neural cells differentiated from human induced pluripotent stem cells (iPSCs) isolated from normal subjects and/or from EOFAD and SAD/LOAD patients and set into 2D or 3D cultures (Kim et al., 2015). Human iPSC models are easier to handle than NHP models and may also be integrated into mouse AD models (Espuny-Camacho et al., 2017).

Still another suitable preclinical human AD model *in vitro* consists of untransformed cortical adult human astrocytes and/ or neurons. Human astrocytes from the temporal lobe cerebral cortex exhibit a stable ("*locked-in*") differentiated phenotype and give reproducible responses when exposed to fAβs and/ or sAβs. The experimental exploitation of the latter cells has revealed that exogenous Aβs specifically bind the calciumsensing receptor (CaSR) (Dal Prà et al., 2014a; Dal Prà et al., 2014b; Dal Prà et al., 2015b), a member of family C G-protein coupled receptors (GPCR), and activate a pathological signaling that could drive human LOAD/SAD onset and progression and also worsen EOFAD's course. These findings have clearly pointed out to a class of therapeutic agents, the CaSR negative allosteric modulators (NAMs), which effectively block all such AD's pathogenetic mechanisms in untransformed cortical human neurons and astrocytes *in vitro* and could stop the progression of AD neuropathology in the patients (Armato et al., 2013; Chiarini et al., 2017a; Chiarini et al., 2017b).

Family C GPCRs also include the metabotropic glutamate (mGlu) and GABAA/B receptors (Bräuner-Osborne et al., 2007; Urwyler, 2011). Results of studies in animal model suggest that mGlu-Rs and GABA-Rs might also be involved in AD pathophysiology because AD concurs with alterations of glutamatergic transmission (Caraci et al., 2018). Therefore, we discuss here the roles of family C GPCRs in AD progression and hence their potential relevance to human AD therapy.

## FAMILY C G-PROTEIN-COUPLED RECEPTORS

In general, GPCRs are among the most numerous groups of transmembrane proteins of the mammalian genome. To date, about 800 of these proteins have been identified in humans (Fredriksson et al., 2003). The relevance of their manifold functions has made them therapeutically attractive as shown by the fact that they are the targets of ~34% of United States Food and Drug Administration-approved drugs (Hauser et al, 2017). Currently, GPCRs are distinguished into six classes (families A–F) (**Table 1**) based upon amino acid sequence homologies, elected signal transduction pathways, and pharmacological outlines.

Details concerning the structure of family C GPCRs are known for the mGluRs (Kunishima et al., 2000; Tsuchiya et al., 2002; Muto et al., 2007; Doré et al., 2014; Wu et al., 2014), GABABR (Geng et al., 2013), and CaSR (Gama et al., 2001; Geng et al., 2016) extracellular domains (ECDs), and for the mGluRs and CaSR transmembrane domains (Doré et al., 2014). Family C GPCRs share a common general structure characterized by a huge bilobed N-terminal extracellular domain (ECD) or TABLE 1 | G-Protein-coupled receptors (GPCRs) Families.


*\*These receptors abound in rodents but are absent from humans (Niswender and Conn 2010; Alexander et al., 2017).*

*\*\*Not considered in this review.*

"Venus Flytrap" (VFT) (Fredriksson et al., 2003; Lagerström and Schlöth, 2008; Rosenbaum et al., 2009). A cysteine-rich region (CR) links the ECD/VFT to the 7TM domain including seven transmembrane helical hydrophobic regions (TM1–TM7) connected extracellularly by three loops (ECL1–ECL3) and intracellularly by three loops (ICL1–ICL3). The CR domain is extant in all family C GPCRs save for GABABRs. Finally, the 7TM domain is linked to the intracellular C-terminal domain (ICD), whose tail interacts with G proteins to activate downstream signaling pathways.

Family C GPCRs function as mandatory dimers (El Moustaine et al., 2012) joined by a disulfide bond topping the two VFTs. GPCRs can be formed into homodimers or into heterodimers with other members of the same group or family (Goudet et al., 2005; Doumazane et al., 2011; Kammermeier, 2012; Sevastyanova and Kammermeier, 2014) or with extraneous GPCRs (Ciruela et al., 2001; Gama et al., 2001; Cabello et al., 2009; Kniazeff et al., 2011). The orthosteric ligands bind the pockets placed in the slit between the two VFT's lobes causing the active closure of both slits (closed-closed conformation) or of only one slit (open-closed conformation) (Parnot and Kobilka, 2004). Conformational changes inside the VFT domains are conveyed through the cysteine-rich and 7TM regions to the ICD domain to regulate G-proteins binding and activate intracellular signals (Rondard et al., 2006).

The family C GPCRs subtype specificity of orthosteric agonists and antagonists varies because the amino acid sequence of the VFT binding pocket may or may not be extensively conserved. Group III mGluRs are an example of the former case, as their orthosteric agonists and antagonists are possessed of a mostly unchanging broad-spectrum activity (MacInnes et al., 2004; Austin et al., 2010). To overcome this obstacle to therapeutic applications, an active quest has been and still is going on for drugs that bind on particular amino acid sequences defined as allosteric sites or pockets (e.g. on the ECL1–ECL3 of the 7TM domain) and placed well outside the VFT-inside orthosteric site. A number of allosteric sites located within the 7TM domain were identified by investigations using site-directed mutagenesis and allosteric modulator cocrystal methods (Gregory et al., 2012;Doré et al., 2014; Wu et al., 2014; Christopher et al., 2015). These novel drugs selectively modify the receptor signaling triggered by an orthosteric ligand acting as either positive allosteric modulators or PAMs or negative allosteric modulators or NAMs (Engers and Lindsley, 2013). PAMs favor whereas NAMs hinder the binding affinity/activity (the so-called cooperativity) of orthosteric ligands. Finally, neutral allosteric ligands (NALs) bind their sequence on a receptor but do not change the cooperativity of its orthosteric ligand (Wootten et al., 2013; Christopoulos, 2014). The otherwise steady structural conformation of orthosteric ligand-bound GPCRs undergoes transient changes, which impact on the interactions with G proteins or other transducers, when the receptors also bind allosteric modulators (Canals et al., 2011). Lipophilic PAMs and NAMs cross the blood–brain barrier (BBB) (Ritzén et al., 2005). However, extremely lipophilic PAMs and NAMs may exhibit a lesser ability to reach the neural cells while their unwanted side effects may become stronger (Goudet et al., 2012). Present-day methods used to evaluate allosteric interactions, i.e. functional and radio-ligand binding assays, need proper probes to reveal the receptor's response and fully disclose the allosteric ligand's properties (Price et al., 2005; Hellyer et al., 2018). The affinity and effectiveness of allosteric modulators can be quantified *via* the allosterism's operational model that associates the allosteric ternary complex model (ATCM) with Black–Leff 's operational model of pharmacological agonism (Black and Leff, 1983; Ehlert, 1988; Leach et al., 2007; May et al., 2007). The latter allows to quantify the modulator's effectiveness and its impact on orthosteric agonist affinity vs. efficacy (Black and Leff, 1983; Leach et al., 2007; Keov et al., 2011; Gregory et al., 2012). Classically, NAMs and PAMs are effective only when the natural orthosteric ligands (or probes) are present (Conn et al., 2009). At times, surrogate orthosteric probes are required for functional assays. Probes of different chemical nature may affect the cooperativity in opposite directions or leave it unchanged (Koole et al., 2010; Valant et al., 2012; Sengmany et al., 2017). Remarkably, allosteric modulators evoke saturable effects, i.e. no further activity obtains when they reach saturating doses (May et al., 2007; Klein, 2013). Specific PAMs and NAMs may or sometimes may not elicit the internalization of their receptors, and hence may or may not desensitize the cells to corresponding orthosteric ligands (Conn et al., 2009). Some PAMs and NAMs are of a mixed type acting as both orthosteric agonists and PAMs, while others act as both agonists for one receptor subtype and as antagonists for another receptor subtype. Preferred PAMs or NAMs should not permanently activate or inactivate the involved receptors because this could elicit harmful effects (Célanire and Campo, 2012). Clearly, receptor subtype-specific PAMs and NAMs are indispensable tools for basic and preclinical pharmacological research and, when beneficial, might be transformed into drugs apt for clinical trials. For example, one chemokine receptor-5 NAM, i.e. Maraviroc, successfully reached the clinical use to treat late-stage HIV disease (Dorr et al., 2005).

Here, a cautionary note is in order about the interspecies translatability of experimental results related to family C GPCRs. In fact, brain locations of the same family C GPCRs can significant vary between animal species (e.g. mouse vs. rat) and between animals and humans. Such species-related divergences could explain the inconsistent observations one may make when experimenting with more than one animal species. This is a further drawback to translating the beneficial effects evoked by orthosteric agonists/antagonists or allosteric PAMs/NAMs in animal models of diseases into beneficial therapeutic upshots in disease-matching human patients. Moreover, one should not overlook another caveat concerning the extrapolation to postnatal life of results gained by administering PAMs or NAMs to animal embryo-fetal cellular models. In fact, during development mGluRs expression undergoes divergent changes in distinct types of cells. Finally, on a positive note, the combined administration of orthosteric agonists or antagonists with NAMs or PAMs can provide additive or synergistic neuroprotection, which might have future therapeutic applications (Vernon et al., 2005; Bennouar et al., 2013).

## GABABRs, APP, A**β**s, AND AD

There are two classes of GABARs, i.e. GABAARs and GABABRs. While GABAARs are fast-acting ionotropic receptors functioning as ligand-gated ion channels, GABABRs are metabotropic family C GPCRs, whose structure comprises the subunits R1a, R1b, and R2 (Chang and Shoback, 2004). Receptor dimerization or oligomerization is obligatory for GABARs as it impacts on function. Only GABAB R1a/R2 or R1b/R2 heterodimers do reach the cell surface, bind GABA to R1 subunit, and activate G-protein-mediated intracellular signals *via* R2 subunit (White et al., 1998; Marshall et al., 1999; Margeta-Mitrovic et al., 2001). Of note, GABABR1 and R2 subunits form heterodimers also with the CaSR (this topic will be further discussed below). Two "*sushi domains* (SDs)" abut from the N-terminus of GABABR1a VFT, which are required for the receptor's trafficking to the cell surface and for its presynaptic inhibitory activity (Gassmann and Bettler, 2012; Hannan et al., 2012). Adaptor proteins link the GABABR1a SD1 domains to axoplasmic kinesin-1 motors. Recently, proteomic methods permitted to identify three adaptor proteins playing this role, i.e. APP, adherence-junction associated protein 1 (AJAP-1), and PILRα-associated neural protein (PIANP). Thus, axonal trafficking cargo vesicles carry at least three distinct types of GABABR1a/adaptor protein/kinesin-1 complexes. It is noteworthy that the formation of any of such GABABR1a/APP/ kinesin-1 complexes obstructs the amyloidogenic processing (*AP*) of the involved APP molecules into Aβ42/Aβ42-os, the AD main drivers. This could be a novel AD-preventing mechanism involving GABABR1a axonal trafficking (Dinamarca et al., 2019).

Neurons express GABABRs on both presynaptic and postsynaptic membranes. Neuronal activity controls GABABRs presynaptic expression (Guetg et al., 2010; Terunuma et al., 2010; Orts-Del'Immagine and Pugh, 2018). Conversely, model animals of AD (Chu et al., 1987; Iwakiri et al., 2005) and of various other diseases, like chronic epilepsy (Thompson et al., 2006–2007), fragile X syndrome (Kang et al., 2017), and Parkinson disease (Borgkvist et al., 2015), downgrade GABABRs expression and inhibitory function causing neuronal hyperexcitability. Human astrocytes express both GABAARs and GABABRs and constitutively synthesize and release GABA, therefore being GABAergic cells. GABA release from human astrocytes is dosedependently increased by glutamate or by NMDAR coagonists like D-serine and glycine. Conversely, inhibitors of kainic acid receptors and of NMDARs decrease GABA release from human astrocytes. Interestingly, the administration of exogenous GABA suppresses the proinflammatory responses of activated astrocytes and microglia to noxious stimuli (Lee et al., 2011a; Lee et al., 2011b).

The GABABRs classical ligand is γ-aminobutyric acid (GABA), which is the chief inhibitory neurotransmitter of the mature central nervous system (CNS) of vertebrates. GABAreleasing (GABAergic) neurons are ubiquitous in the CNS. Nearly half of all CNS synapses express some GABABRs and thus respond to GABA (Li et al., 2016). The signaling activated by the GABA•GABABR complexes inhibits the release of neurotransmitters from the targeted postsynaptic neurons (Benes and Berretta, 2001; Bettler et al., 2004). But the signaling activated by the presynaptic GABA•GABABR complexes of the GABAergic neurons blocks their further GABA release thus acting as a physiological self-controlling mechanism exerting an indirect excitatory effect on postsynaptic neurons. An astrocyte → neurons metabolic cycle upkeeps the neuronal stores of GABA (Losi et al., 2014; Mederos et al., 2018). The released GABA moieties are taken up by both pre- and postsynaptic neurons and by the astrocytes enveloping tripartite synapses. In their mitochondria, the astrocytes convert the uptaken GABA into glutamine and forward the glutamine to adjoining neurons. In presynaptic neurons, two sequentially acting enzymes synthesize GABA from glutamine: first, glutaminase, which converts glutamine into glutamate; and second, glutamic acid decarboxylase (GAD), which transforms glutamate into GABA. Next, the vesicular transporter of GABA (VGAT) transfers it into cargo/synaptic vesicles, which release GABA into the synaptic cleft after the neuron's membrane depolarization. GABA signaling regulates several physiological aspects of CNS activity, like neurogenesis, neuronal development, sexual maturation, circadian rhythms, motor functions, learning, and memory (Petroff, 2002). GABAergic interneurons are the chief inhibitory neurons in the CNS. Abundant reports in the literature stress how dysfunctions of the activity of GABAergic interneurons disrupt the glutamatergic excitatory/GABAergic inhibitory (E/I) balance in neural circuits of the cerebral cortex, hippocampus, and subcortical structures (e.g. amygdala) causing declining cognitive capabilities and worsening memory losses. Therefore, it is generally held that E/I imbalance significantly impacts on the pathogenesis of AD (Govindpani et al., 2017; Villette and Dutar, 2017) and of various other CNS disorders, such as major depression (Fee et al., 2017), schizophrenia, autism's spectrum, bipolar disorder (Benes and Berretta, 2001; Lehmann et al., 2012; Gao and Penzes, 2015; Xu and Wong, 2018), and anxiety (Babaev et al., 2018).

Unfortunately, the results of postmortem studies on AD patients did not throw much light onto GABABRs role(s) in AD and this for good reasons--one being the wide-ranging variability of the patients' terminal lifetime. Different alterations of biological parameters--such as expression of RNA, proteins, and enzymes--can be induced by co-morbidities, medications, aging, and leading death causes and are all directly linked to the duration and severity of the full-blown or symptomatic AD phase (Govindpani et al., 2017). Therefore, it is not surprising that divergent findings abound concerning GABABRs' possible role(s) in AD. Thus, recent postmortem studies of human brains and of *tg* AD-model animals reported that GABAergic neurons and GABABRs were unaffected by AD neuropathology (Li et al., 2016). Conversely, earlier studies had conveyed the opinion that GABABRs signaling undergoes profound changes in AD (Takahashi et al., 2010). Significantly lowered GABA levels were detected in the temporal cortex of AD patients (Gueli and Taibi, 2013) and in cerebrospinal fluid (CSF) samples from both AD patients and the cognitively normal elderly (Grouselle et al., 1998). Conversely, raised GABA levels turned up in the hippocampus and CSF samples of AD patients and were ascribed to the impairment of synaptic plasticity (Jo et al., 2014). These studies also noted that the reactive astrocytes surrounding Aβs senile plaques overproduced GABA *via* monoamine oxidase-B (MAO-B) activity and abnormally released it through the bestrophin-1 channels. Under physiological conditions, bestrophin-1 channels are mostly localized at the microdomains of hippocampal astrocytes nearby glutamatergic synapses and mediate glutamate release. A switch from the glutamate-releasing normal astrocyte to the reactive astrocyte releasing GABA *via* bestrophin-1 channels is a common phenomenon occurring in various pathological conditions coupled with astrogliosis, such as traumatic brain injury, neuroinflammation, neurodegeneration, and hypoxic and ischemic insults. In AD, bestrophin-1 channels are redistributed to the soma and the processes of hippocampal reactive GABA-containing astrocytes. Bestrophin-1 channelsmediated GABA release from reactive astrocytes hinders synaptic plasticity and transmission and spatial memory by reducing dentate granule cell excitability (Oh and Lee, 2017). It was claimed that suppressing GABA overproduction by monoamine oxidase-B (MAO-B) or GABA overrelease through bestrophin 1 channels from the dentate gyrus reactive astrocytes fully restored learning and memory in AD-model mice (Jo et al., 2014). However, the long-term administration of selegiline, an irreversible MAO-B inhibitor, did not improve AD neuropathology in a clinical trial (Park et al., 2019). To explain this unforeseen upshot, the authors suggested that multiple factors, like age, sex, and differences in brain regions could impact on the GABA release from astrocytes and neurons and should not be ignored when planning therapeutic drug attempts. Indeed, different brain regions of Tg2576 (human APP695 plus the Swedish double mutation K670N, M671L) AD-model mice released dissimilar amounts of GABA in relation to their actual age and sex. Cortical GABA levels were higher in older than 6 months female than male mice; however, at a more advanced age, this difference vanished in the parietal cortex but became more pronounced in the prefrontal cortex. Moreover, at 12–18 months of age, hippocampal levels of GABA were lower in female than in male mice. Altogether, these data revealed that with advancing age the functional disruption of GABA signaling turns out to be more intense in AD-model female mice (Hsiao et al., 1996). Conversely, under or up to 9 months of age, hippocampal GABA levels were higher in female than in male mice, likely because the former enjoyed the protective activity of estrogens (Roy et al., 2018). By extrapolating these data from animals to humans one can infer that a single therapeutic strategy addressing GABABRs modulation might not be so easily feasible in AD. In fact, any drug inhibiting the GABABRs residing in one brain region might exacerbate the dysregulation of GABAergic signaling in other brain areas.

Hence, the role(s) of GABA/GABABRs signaling in the pathogenesis of AD has(ve) hitherto remained unclear if not confusing (Govindpani et al., 2017). However, the recent studies we mention just below performed on wild-type and AD-model animals have thrown some more light on the contribution of GABAergic remodeling to the pathogenesis of both early and late stages of AD.

First, we recall here that the ε4 allele of apolipoprotein E (APOE) is the main known genetic risk factor for LOAD/SAD. Notably, in the brains of aged APOEε4 mice, an attenuation of GABAergic inhibitory inputs on associated excitatory neurons drives a specific neuronal hyperactivity phenotype. Hence, an APOEε4-driven hippocampal neuronal excitatory hyperactivity might be among the causative factors underlying the increased risk of AD among APOEε4 carriers (Nuriel et al., 2017). In addition, several AD-mouse models exhibit an early and marked neuronal hyperactivity in the hippocampus (Busche et al., 2012; Busche et al., 2015). Moreover, functional magnetic resonance imaging (fMRI) studies have revealed that humans with mild cognitive impairment (MCI), as well as presymptomatic carriers of EOFAD mutations show enhanced neuronal activity in the same brain region, the hippocampus (Quiroz et al., 2010; Haberman et al., 2017). Therefore, when the increase in brain activity takes place early in the pathogenetic process it may be rightly considered a driving factor in AD development.

Second, according to a recent report, Aβs, AD's main drivers, are intensely degraded by endothelin-converting enzyme-2 (ECE-2) and neprilysin (NEP) in the somatostatinergic and parvalbuminergic synapses of GABAergic interneurons residing in the neocortex and hippocampus. These observations support the view that under physiological conditions Aβs may partake in the regulation of interneurons' inhibitory signaling in AD-relevant brain areas (Pacheco-Quinto et al., 2016). However, it must be stressed here that a reduction of Aβs catabolism at the synapses of these two distinct populations of GABAergic neurons is not the unique GPCRs-mediated mechanism favoring Aβs accumulation in AD brains.

Third, the exciting results of very recent studies have revealed that GABABR1a receptors bind three novel orthosteric agonists besides GABA, i.e. soluble (s)APP-α, sAPP-β, and sAPP-η proteins. The α-, β-, and η-secretases, respectively, shed them from the extracellular domain of APP into the brain environment. Next, each of these peptides can bind GABABR1a receptors and block the release of neurotransmitters from hippocampal presynaptic excitatory axonal terminals thus silencing synaptic transmission. Most interesting, a 17-mer peptide of the ExD flexible portion of sAPP-α, which binds the extracellular Sushi1 domain of the GABABR1a could replicate the squelching effect on neurotransmission brought about by the whole sAPP-α molecule. These results explain, at least in part, the synaptic dysfunction affecting some APP-overexpressing AD-model animals. Moreover, they suggest that this 17-mer peptide could therapeutically counteract the excitatory hyperactivity of neuronal synaptic function brought about by Aβs (Rice et al., 2019; Tang, 2019).

## mGLURs AND AD

The seven-transmembrane-spanning mGluRs physiologically control synaptic transmission and neuronal excitability in the CNS and influence behavioral output processes. These receptors are assigned to three groups according to their G-protein coupling and signal transduction pathways. Group I encompasses mGluR1 and mGluR5; group II includes mGluR2 and mGluR3; and group III embraces mGluR4, mGluR6, mGluR7, and mGluR8. In general, group I receptors are coupled to the phospholipase (PL) C/InsP3/Ca2+ release cascade, whereas groups II and III receptors are linked up to the adenylyl cyclase/cyclic AMP/PKA release cascade (Niswender and Conn, 2010). Initial studies performed with agonists (or antagonists), which bind the intragroup-shared extracellular orthosteric sites, indicated that activation of group II or group III mGluRs brought about neuroprotection, whereas activation of group I mGluRs elicited either neuroprotection or neurotoxicity according to experimental models and conditions employed (Nicoletti et al., 1999; Bruno et al., 2001). More recent studies using PAMs or NAMs, which are receptor subtypespecific, brought to light a somewhat different picture (see for references: Gregory and Conn, 2015). Indeed, the allosteric modulation of mGluRs is a major area of interest for Basic and Clinical Pharmacology (Stansley and Conn, 2019). In the CNS, mGluRs are involved in the regulation of glutamate uptake, cell proliferation, neurotrophic support, and proinflammatory responses. Accordingly, the potential therapeutic spectrum of mGluRs allosteric modulators embraces AD, and also covers PD, stress, anxiety, autism, depression, and schizophrenia (Stansley and Conn, 2019).

## Group I mGluRs (-1 and -5)

Group I mGluR1 and mGluR5 are expressed at postsynaptic membranes, couple to Gαq, and positively modulate neuronal excitability through the interaction with scaffolding proteins such as Homer or Shank. The consequent activation of phospholipase C leads to an increase in [Ca2+]i . Activation of group I mGluRs may set off a multiplicity of neurons' and astrocytes' signaling pathways variously modulating synaptic plasticity and, likely, synaptic protein synthesis (D'Antoni et al., 2014). These transduction mechanisms form a highly complex network including polyphosphoinositide hydrolysis, mitogen-activated protein kinase/extracellular signal-regulated kinase (MAPK/ ERK), phospholipase D, phospholipase A2, phosphoinositide 3-kinase (PI3K), mammalian target of rapamycin (mTOR), and the endocannabinoid 2-arachidonoylglicerol synthesis. Activation of ERK and mTOR by group I mGluRs is especially linked to *de novo* protein synthesis in neurons, a process that underlies longterm changes in activity-dependent synaptic plasticity. Group I mGluRs also enhance postsynaptic excitability thus exacerbating neuronal damage (Nicoletti et al., 1996). It is also noteworthy to recall here that in preclinical studies *antagonists* of mGluR1 and mGluR5 exhibited anxiolytic-like properties just as did *agonists* of group II/III mGluRs (Stachowicz et al., 2007).

By interacting with NMDARs, mGluR1 and mGluR5 regulate neuronal developmental plasticity. The interaction between group I mGluRs and NMDARs is reciprocal (Alagarsamy et al., 1999). Moreover, the expression of these receptors is developmentally regulated (Nicoletti et al., 1986; Schoepp and Johnson, 1989; Minakami et al., 1995; Romano et al., 1996; Casabona et al., 1997). Group I mGluRs are cross-linked with NMDARs through a chain of anchoring proteins (Tu et al., 1999), and their activation amplifies NMDA currents (Aniksztejn et al., 1995; Awad et al., 2000; Pisani et al., 2001; Skeberdis et al., 2001; Heidinger et al., 2002; Kotecha and MacDonald, 2003). In addition, activation of mGluR1 accelerates NMDARs trafficking (Lan et al., 2001). The NMDA component of long-term potentiation (LTP) is abolished in mice lacking mGluR5 (Jia et al., 1998). In cultured neurons and developing brains the interaction between mGluR5 and NMDAR is amplified by EphrinB2 (Calò et al., 2005), a ligand for EphB receptor tyrosine kinases playing a role in activity-dependent synaptic plasticity in the CNS (Slack et al., 2008).

In the developing brain, mGluR5 contributes to the functional maturation of astrocytes since mGluR5 ablation leads to serious deficits in arborization of astroglial processes and in the expression of glutamate transporters (Morel et al., 2014; Verkhratsky and Nedergaard, 2018).

To date, the roles of group I mGluRs in the pathogenesis of AD are poorly understood and the object of controversy.

*In vitro* cultured fetal (E15) Sprague–Dawley rat neurons expressed mGluR1 whereas neonatal astrocytes did not. These findings limited to neurons the investigation of an alleged neuroprotective effect of the mGluR1/R5 *agonist*, 3,5-dihydroxyphenylglycine (DHPG), against Aβ neurotoxicity, which was instead suppressed by the mGluR1 *antagonist* JNJ 16259685 (3,4-dihydro-2H-pyrano(2,3-b)quinolin-7-yl)-(cis-4-methoxy-cyclohexyl)-methanone). Interestingly, estrogen-α receptors (E-αRs) could activate the same neuroprotection against Aβ toxicity and cell survival pathways as mGluR1 did. Indeed, E-αRs and mGluR1 were colocalized in cultured cortical neurons and were interdependent in activating the PI3K/Akt pathway that favors cell survival in pure neuronal cultures (Spampinato et al., 2012).

As regards mGluR5s, they are expressed by both astrocytes and neurons in all the CNS areas, signal through Gq protein (Vanzulli and Butt, 2015), and partake in synaptic plasticity, assembly of neuronal circuitry, and neuronal viability (Ballester-Rosado et al., 2010; Purgert et al., 2014).

Data gained from both *in vitro* and animal models suggest that the synaptic dysfunction of mGluR5s might favor the development of AD (Kumar et al., 2015). mGluR5s are overexpressed by astrocytes as a reactive response induced by stimulation with growth factors (i.e. FGF, EGF, and TGF-β1) or by exposure to soluble Aβ oligomers (Aβ-os) *in vitro* (Casley et al., 2009; Grolla et al., 2013; Lim et al., 2013). Aβ-os exposure also raises the expression of type I InsP3Rs, which are placed downstream from mGluR5, and strengthens Ca2+ responses mediated through the mGluR5/InsP3R cascade in hippocampal astrocytes (Grolla et al., 2013). Notably, astrocytes surrounding Aβ senile plaques overexpressed mGluR5, which was associated with Ca2+ signaling dysregulation and abnormal ATP release in APPswe/PS1 transgenic AD-model mice (Shrivastava et al., 2013). Reportedly, Aβ-os exposure caused an excessive clustering and widely reduced diffusion of Aβ-os/mGluR5 complexes on the plasma membrane of *in vitro* rat embryo astrocytes. These effects were coupled with an augmented Ca2+ influx altogether damaging synapses (Renner et al., 2010; Shrivastava et al., 2013). Activation of mGluR5s by the allosteric agonist DHPG increased ATP release from Aβ-os-exposed astrocytes, which delayed mGluR5 diffusion in cultures of astrocytes plus/minus neurons *in vitro* (Renner et al., 2010)—an effect mGluR5's selective antagonist MPEP counteracted thus preventing Aβ-os/mGluR5 driven synaptotoxicity (Shrivastava et al., 2013).

Interestingly, proinflammatory cytokines like interleukin-1β (IL-1β) and tumor necrosis factor-α (TNF-α) downregulated the expression of mGluR5 while upregulating that of mGluR3 in cortical astrocytes isolated from the hSOD1(G93A) rat model of amyotrophic lateral sclerosis (entailing like AD an intense neuroinflammation) and cultured *in vitro*. These findings suggested the existence of a protective antiexcitotoxic adaptive mechanism (Berger et al., 2012). In fact, the mGluR5 selective antagonist MPEP hampered the astrocytes' secretion of two proinflammatory cytokines, IL-6 and IL-8 (Shah et al., 2012). Therefore, the activation of astrocytes' mGluR5 advances the release of proinflammatory cytokines, which then downregulate mGluR5 expression. This indicates that under physiological conditions a reciprocal feed-back mechanism controls the expression levels of mGluR5 in astrocytes and in microglia too (Berger et al., 2012). This mechanism might be offset by the Aβ-os-forced overexpression of mGluR5 in AD, thus potentiating the release of toxic amounts of proinflammatory cytokines and glutamate. Next, the latter increases the production/release of p-Tau-os and of NO and the activity of apoptotic caspase-3 (Talantova et al., 2013; Lee et al., 2014).

Another noteworthy study showed that the activation of mGluR5 stimulated the α-secretase-mediated extracellular shedding of neurotrophic and neuroprotective sAPP-α (Sokol et al., 2011), also an agonist of GABABR1a receptors (Rice et al., 2019). But mGluR5 forms complexes with the Homer proteins that interact with and activate NMDARs (Tu et al., 1999; Awad et al., 2000; Attucci et al., 2001; Moutin et al, 2012). Aβ1-42-os can bind mGluR5s and enhance their clustering together, causing mGluR5 signaling overactivation, intracellular Ca2+ accumulation, impaired calcium homeostasis, and synaptic disruptions (Renner et al., 2010; Zhang et al., 2015b). In greater detail, mGluR5s act as coreceptors for Aβ-os bound to prion protein (PrPc ). Next PrPc activates the mGluR5, which elicits the loss of synapses through *Fyn* tyrosine kinase activation and eukaryotic elongation factor 2 (eEF2) phosphorylation (Um et al., 2013). *Fyn* phosphorylates NR2A and NR2B subunits of NMDA receptors thus altering the receptors' intracellular trafficking that is essential for synaptic plasticity. Moreover, interactions between *Fyn* tyrosine kinase and Tau proteins play a role in regulating the synapse function and the postsynaptic toxic signaling pathways driven by Aβ-type excitotoxicity, causing the loss of dendritic spines. Notably, Aβ-os exposure also induces the eEF2 phosphorylation by eEF2 kinase that is known to associate with mGluR5. Aβ-os-induced impairment of LTP is dependent on eEF2 phosphorylation that is increased in brains from both *tg* AD-model mice and AD patient autopsies (Nygaard, 2018).

Altogether these data support the view that mGluR5 activation by specific PAMs facilitates excitotoxic mechanisms causing the death of neurons (Parmentier-Batteur et al., 2014), whereas mGluR5-specific NAMs act neuroprotectively in AD-model animals (Bruno et al., 2017). Administering MPEP, a mGluR5 selective antagonist, prevents this synaptic loss in *tg* AD-model mice (Rammes et al., 2011; Hu et al., 2014; Kumar et al., 2015). In addition, deletion of mGluR5 prevents memory loss in AD-model mice (Hamilton et al., 2014). But to gain these benefits there is a price to pay, which is the negative impact of mGluR5 selective antagonists on activity-dependent synaptic plasticity mechanisms in brain regions that are not affected by AD (Bruno et al., 2017). Of course, the translatability of these interesting results to human AD patients remains a topic worth exploring.

## Group II mGluRs (-2 and -3)

Group II mGluR2 and mGluR3 are mostly localized presynaptically. Depending on the nature of the ligand, mGluR2s signal *via* Gi/o or Gq11 proteins (González-Maeso et al., 2008; Fribourg et al., 2011) and negatively modulate neuronal excitability (Conn and Pin, 1997). Thus, activation of group II mGluRs is endowed with potential neuroprotective properties as it may curtail glutamatergic signaling and mitigate neuronal hyperexcitability (Nicoletti et al., 1996). Stimulation of group II mGluRs inhibits adenylyl cyclase (AC), activates K+ channels, and blocks presynaptic voltage-gated calcium channels, thus hindering intracellular Ca²+ fluxes and synaptic neurotransmitters release (Benarroch, 2008; Niswender and Conn, 2010). Groups II mGluRs also team with the MAPK and PI3K pathways to confer neuroprotection (D'Onofrio et al., 2001). As mentioned also below, neuroprotection is mediated by transforming growth factor-β1 (TGF-β1) released through astrocytes' mGluR3 signaling. TGF-β1 binds and activates its membrane receptors coupled with serine/threonine kinase activity thereby inducing the *Smad* signaling cascade. It also synergistically operates with other neurotrophins such as nerve growth factor (NGF), brain-derived neurotrophic factor (BDNF), and glial-derived neurotrophic factor (GDNF) (Caraci et al., 2011b). In the rodents' thalamus, the selective activation of mGluR2 modulates the inhibition at synapse level of sensory neurons functionally linked to information processing, attention, and cognition (Copeland et al., 2017). Conversely, the selective activation of mGluR2 increases the incidence of neuronal deaths *in vitro* (Corti et al., 2007; Caraci et al., 2011b). Accordingly, a mGluR2-specific NAM hindered the death of ischemiasensitive neurons in the hippocampal CA1 area, whereas a mGluR2-specific PAM promoted the death both of ischemiasensitive CA1 neurons and of ischemia-resistant CA3 neurons (Motolese et al., 2015). Recent investigations have revealed the formation of intragroup and intergroup heterodimers between different mGluRs (Doumazane et al., 2011; Rondard et al., 2011; Kammermeier, 2012). New allosteric modulators capable of differentiating homodimers from heterodimers have disclosed the assembly of mGluR2/mGluR4 heterodimers in corticostriatal fibers (Yin et al., 2014).

The mGluR3, whose activation inhibits AC activity and hence cyclic AMP production, is the most abundant astroglial receptor along all the lifetime (Sun et al., 2013). Mounting evidence indicates that mGluR3 upkeeps synaptic homeostasis, including synaptic plasticity and synaptogenesis (see for references: Durand et al., 2013). In addition, activated mGluR3 plays major neuroprotective roles in AD and other neuropathologic conditions. Once added to pure cultures of newborn rat astrocytes, the orthosteric agonists LY379268 or LY354740 specifically activated mGluR3 (rodent astrocytes do not express mGluR2), promoting the production and release of TGF-β and of GDNF (see also above and Caraci et al., 2011a). The same agonists increased the expression of α-secretase, whose activity is essential for APP's physiological *NAP*. The upshot is an amplified extracellular shedding of the neurotrophic and neuroprotective and GABABR1a agonist sAPP-α (**Figure 1**) (Bruno et al., 1997; Bruno et al., 1998; Corti et al., 2007; Battaglia et al., 2009; Di Liberto et al., 2010; Battaglia et al., 2015; Rice et al., 2019). Moreover, an indirect role for mGluR3 in AD is denoted by the progressive decrease with aging in mGluR2 and mGluR3 expression and, consequently, in their antiamyloidogenic action in hippocampal astrocytes from PDAPP-J20 AD-model mice (Durand et al., 2014). In subsequent studies, the same authors showed that the LY379268-elicited activation of astrocytes' and neurons' mGluR3 suppressed or mitigated the Aβ-driven neurotoxicity and death of both neurons and astrocytes. In both cell types, agonist-activated mGluR3 increased the shedding of neuroprotective sAPP-α and the expression of BDNF. In addition, LY379268-activated mGluR3s induced astroglia- and microglia-mediated phagocytosis and removal of Aβs from the extracellular environment. Finally, mGluR3 orthosteric agonists LY379268 or LY404039 suppressed the nitric oxide (NO) induced death of cultured rat astrocytes *via* the inhibition of AC, which reduced intracellular cAMP levels, the activation of Akt, and the formation of antiapoptotic p65 and c-Rel complexes of the NF-κB family (Durand et al., 2011; Durand et al., 2017).

Conversely, Caraci et al. (2011a) showed that mGluR2 and mGluR3 enhanced neurotoxicity in pure cultures of rat brain neurons challenged with Aβ1–42 or with its neurotoxic fragment Aβ25–35. However, if the neurons were cocultured with the astrocytes, the activation of mGluR2 and mGluR3 brought about neuroprotective effects through the release of TGF-β1 from the astrocytes. TGF-β1 is a well-known agent endowed with neuroprotective and anti-inflammatory activities (see also above) in experimental AD-models (Chen et al., 2015) as it also stimulates microglia to scavenge Aβs (Tichauer and von Bernhardi, 2012).

## Group III mGluRs (-4, -6, -7, and -8)

Group III mGluRs (-4, -6, -7, and -8) are mainly localized presynaptically, couple to Gαi/o, and negatively modulate neuronal excitability (Conn and Pin, 1997). They are likely to act as autoreceptors on glutamatergic synaptic terminals and as heteroceptors on GABAergic and other neurotransmitter terminals (Cartmell and Schoepp, 2000; Ferraguti and Shigemoto, 2006). Group III mGluRs stimulation results in AC inhibition, K+ channels activation, and block of presynaptic voltage-gated calcium channels, thus decreasing Ca²+ flow into cells and neurotransmitters release from synapses (Benarroch, 2008; Niswender et al., 2008). Therefore, their activation elicits potential neuroprotective effects that dampen glutamatergic signaling and inhibit neurotransmitters release thereby mitigating neuronal excitability (Nicoletti et al., 1996). As a particularity, activated mGluR7 stimulates protein kinase C (PKC) or phospholipase C resulting in the inhibition of neuronal calcium channels (Perroy et al., 2000; Pelkey et al., 2007). Brain expression of mGluR4, -7, and -8 is proper of cortical and hippocampal neurons and of synapses located in the basal nuclei (striatum, pallidum), subthalamic nucleus, and substantia nigra (both pars compacta and pars reticularis) (Bruno et al., 1996; Faden et al., 1997; Hovelsø et al., 2012). Instead, mGluR6 expression is exclusive of the retina (Nakajima et al., 1993). The main subcellular location of mGluR7 is at the central area of presynaptic terminals just where the membrane coalesces with synaptic vesicles: this suggests its involvement in the modulation of neurotransmitter release. Conversely, mGluR4 and mGluR8 are placed at the periphery of presynaptic terminals (Shigemoto et al., 1997; Schoepp, 2001; Palucha and Pilc, 2007). Group III mGluRs also cooperate with MAPK and PI3K signaling pathways to impart neuroprotection (Iacovelli et al., 2002). Recently, these receptors have become the focus of therapeutic attempts because they (i) can modulate defective neurotransmission yielding symptomatic improvements through the neuroprotective hindering of multiple neurodegenerative mechanisms and (ii) have more favorable safety and tolerability profiles (Hovelsø et al., 2012). Activation of group III mGluRs by glutamate and/or other agonists is neuroprotective as it inhibits glutamate release from neurons' presynaptic terminals and from microglia, thus mitigating excitotoxicity; concurrently, astrocytes intensify the uptake of glutamate and microglia increase neurotrophic factors synthesis (see Williams and Dexter, 2014 for an in-depth review on this topic).

Rather few studies exist about the effects on a neurodegenerative disease like AD exerted by group III mGluRs activation *via* broad spectrum agonists and PAMs or inactivation *via* NAMs. Copani et al. (1995) reported that broad-spectrum group III mGluRs agonists L-serine-Ophosphate (L-SOP) and l-2-amino-4-phosphono-butanoate (L-AP4) could lessen the apoptotic death rate of neurons exposed to Aβs. The authors suggested that such agonists would exert neuroprotective effects in AD. Similarly, group III agonist RS-PPG, which activates preferentially mGluR8 and likely also mGluR4, exerted neuroprotective actions on neurons exposed to harmful hypoxic or hypoglycemic conditions (Bruno et al., 2000; Sabelhaus et al., 2000). Notably, acute hypoxia can induce neurons to overproduce lethal amounts of Aβs *via* a mechanism. involving another family C GPCR, the CaSR (Kim et al., 2014; Bai et al., 2015). Besides, PHCCC, a specific mGluR4 PAM, and also a partial antagonist of group I mGluRs, protected cultured cortical mouse neurons against the Aβs-elicited cytotoxicity and NMDAR excitotoxicity (Maj et al., 2003).

But what about the astrocytes? Under basal conditions, rodent (rat and mouse) astrocytes in primary cultures express mGluR4, but neither mGluR7 nor mGluR8 (Phillips et al., 1998; Janssens and Lesage, 2001). However, mGluR8 is expressed by reactive astrocytes adjacent to chronic inflammatory lesions (Geurts et al., 2005). Besong et al. (2002) provided evidence that broad spectrum orthosteric agonists activating group III mGluRs, like L-AP4, 4-phosphonophenylglycine (4-PPG), or L-SOP hindered the expression and secretion of the proinflammatory chemokine RANTES in astrocyte cultures. These beneficial effects of the mGluR4 broad spectrum agonists were counteracted by pretreating the astrocytes cultures with the selective group III mGluRs NAM (R,S)-α-methyl-serine-O-phosphate or with pertussis toxin. Altogether, these findings suggest that such agonists might mitigate neuroinflammation in conditions like AD, multiple sclerosis, and experimental allergic encephalomyelitis.

Extracellular glutamate homeostasis, which is essential for physiological glutamatergic neurotransmission and excitotoxicity prevention, depends on the activity of astrocytes' transporters like GLT-1 and GLAST (Anderson and Swanson, 2000). Neuroinflammatory conditions associated with a neurodegenerative disease like AD or experimental treatments (e.g. with LPS, MPTP, etc.) reduce astrocytes' GLT-1- and GLAST-mediated glutamate uptake due to a fall in endogenous antioxidant glutathione (GSH) activity. Broad spectrum group III mGluRs agonists rescue GSH normal levels and restore astrocytes' GLT-1- and GLAST-mediated glutamate uptake alleviating neuronal excitotoxicity (Yao et al., 2005; Zhou et al., 2006; Foran and Trotti, 2009). Thus, activation of astrocytes' group III mGluR3 and mGluR5 and also of group II mGluRs by broad spectrum agonists increases GLT-1- and GLAST protein expression and glutamate uptake activity as the signaling of both groups likely involves Gi/o, MAPKs, and PI3K pathways (Aronica et al., 2003; Beller et al., 2011; Williams and Dexter, 2014). The activation of group III mGluRs by wide spectrum agonists also curtails the release of proinflammatory cytokines from activated microglia (Combs et al., 2000). Wide spectrum group III mGluRs agonists also hinder proinflammatory cytokines release, RANTES included, from the astrocytes exposed to neurotoxic agents (Mennicken et al., 1999; Besong et al., 2002), thereby helping mitigate neuroinflammation and reduce neuronal demise. Therefore, one might surmise that the effects of these agonists on astrocytes and microglia would likely impact on the course of AD and perhaps also of other neurodegenerative diseases.

## CALCIUM SENSING RECEPTOR

The CaSR is a (poly)cationic receptor, as its evolutionary history shows (Riccardi and Kemp, 2012). This is why CaSR's preferred yet not unique orthosteric agonist is Ca2+. A CR region, necessary for receptor activation (Huang et al., 2011; Hendy et al., 2013) connects CaSR's huge (~612 amino acids) ECD, the bilobed (LB1 and LB2) VFT to the 7TM domain whose seven transmembrane α-helices (TM1–TM7) are joined by three extracellular and three intracellular loops. Two domains of the CASR's intracellular C-terminal tail are necessary for CaSR expression at the cell surface and its composite signaling functions *via* G-proteins (see below). The VFT contains the binding pockets for the orthosteric (type I) agonists (Hendy et al., 2013), which besides extracellular Ca2+ (Hofer and Brown, 2003), include various divalent and trivalent cations, polyamines, aminoglycoside antibiotics, and cationic polypeptides (Silve et al., 2005; Saidak et al., 2009; Magno et al., 2011; Zhang et al., 2015a). The CaSRs of human cortical astrocytes also specifically bind Aβs, likely at the VFTs (Dal Prà et al., 2014a; Dal Prà et al., 2014b, Dal Prà et al., 2015b). Moreover, X-ray crystallography studies (Geng et al., 2016) have revealed that in the resting state the 3D structure of CaSR's ECD exhibits an open conformation kept up by PO4 3− anions. Independently of the presence or absence of Ca2+ ions, CaSR activation occurs when an L-α-amino acid closes the VFT groove, triggering the formation of a new homodimer interface between the membrane-proximal LB2 and the CR domains. Ca2+ ions stabilize the active state to fully activate the receptor. Indeed, CaSR's ECD is endowed with four Ca2+-binding sites, of which the Ca2+ ion at site #4 stabilizes, upon orthosteric agonist binding, the CaSR homodimer's active conformation (Geng et al., 2016). Importantly, orthosteric agonists also induce the dissociation of inhibitory PO4 3− anions from the arginine residues acting as their relatively weak binding sites. Thus, the CaSR-inactivating action of bound PO4 3− anions is overturned (Quinn et al., 1997; Cheng et al., 2004; Geng et al., 2016). As the other GPCRs do, CaSR swings between conformation-varying active and inactive states (Rosenbaum et al., 2011). The changes in conformation due to activation include a rearrangement of the 7TM and ICD domains. The CaSR's 7TM helical domains can modulate signal transduction. The 7TM's intracellular loops 2 and 3 are crucially involved in the activation of downstream effectors (Goolam et al., 2014). Besides, various CaSR's 7TM sites bind allosteric (type II) ligands. The latter include both the aromatic L-α-amino acids and the highly selective *allosteric agonists* or *PAMs*, shorttermed *calcimimetics*, and *allosteric antagonists* or *NAMs*, shorttermed *calcilytics* (Nemeth, 2002). As will be discussed later, these pharmacological agents offer exciting perspectives in the field of clinical therapeutics. In response to orthosteric ligand binding, the CaSR's ICD tails interact with Gs or Gq/11 or G12/13, or Gi/o, proteins (Chang et al., 2001; Hofer and Brown, 2003; Conigrave and Ward, 2013), and with β-arrestin 1/2 (Thomsen et al., 2012). Such interactions turn on several signaling pathways (Saidak et al., 2009; Magno et al., 2011), which underlie the receptor's complex actions and comprise: (i) second messengerproducing enzymes (e.g., AC); (ii) phospholipases A2, C, and D; (iii) protein kinases (e.g. PKCs, MAPKs, AKT); (iv) Ca2+ influx *via* TRPC6-encoded receptor-operated channels; and (v) transcription factors (reviewed in Zhang et al., 2015a). Moreover, the intracellular adaptor-related protein complex (AP2) binds the CaSR's ICD promoting the receptor's clathrinmediated endocytosis (Nesbit et al., 2013). Finally, CaSR's ICD ubiquitylation and phosphorylation modulate the receptor's recycling, degradation, and desensitization (Zhuang et al., 2012; Breitwieser 2013).

In general, the CaSR preserves systemic Ca2+ homeostasis by promptly sensing any changes in the extracellular calcium concentration [Ca2+]e and, accordingly, by modulating the amounts of parathyroid hormone (PTH) released from parathyroid glands as well as the reabsorption of Ca2+ from kidneys and its deposition in bones (Hofer and Brown, 2003). Dysfunctions of the CaSR severely alter systemic Ca2+ homeostasis (Brown, 2007; Hendy et al., 2009). Gain-of-function CaSR mutations result in autosomal dominant hypocalcemia, whereas loss-of-function CaSR mutations cause severe neonatal primary hyperparathyroidism (Hendy et al., 2009; Ward et al., 2012; Hannan et al., 2018).

But, what about the CaSR in the brain? All types of brain neural and cerebrovascular cells express the CaSR, with particular intensely in the hippocampus, an AD-relevant area (Chattopadhyay, 2000; Yano et al., 2004; Noh et al., 2015). Dal Prà et al. (2005) showed that untransformed astrocytes isolated from the adult human temporal cortex and cultured *in vitro* express functional CaSRs, less intensely when proliferating but more strongly when mitotically quiescent. Notably, changes in the growth medium [Ca2+]e did not impact on CaSR expression levels by adult human astrocytes. But preservation of systemic Ca2+ homeostasis is not the CaSR's main task in the brain. In fact, fluctuations in [Ca2+]e physiologically modulate, *via* corresponding adaptations of CaSR signaling, a variety of neural cells activities like CaSR's L-amino acid sensing (Conigrave and Hampson, 2006), K+ fluxes (Chattopadhyay et al., 1999), proliferation, differentiation, migration of both neurons and oligodendrocytes during growth, and synaptic plasticity and neurotransmission during postnatal life (Bandyopadhyay et al., 2010; Riccardi et al., 2013; Ruat and Traiffort, 2013; Kim et al., 2014; Noh et al., 2015; Tharmalingam et al., 2016).

Remarkably, CNS diseases, such as AD and ischemia/ hypoxia/stroke, change the CaSR's expression levels and hence alter the cellular processes CaSR signaling regulates (Armato et al., 2013; Dal Prà et al., 2014a; Dal Prà et al., 2014b; Bai et al., 2015; Dal Prà et al., 2015b). The first hint that the CaSR might play a role in AD pathogenesis stemmed from the observation that Aβs-elicited peaks of cytosolic [Ca2+]i had a killing effect on hippocampal neurons (Brorson et al., 1995). A second clue was the opening of Ca2+-permeable nonselective cation channels (NSCCs) by fibrillar Aβ1–40 or Aβ25–35 in hippocampal neurons of wild type (WT) *CaSR*+/+ rats; notably, this effect could not be replicated in *CaSR*−/− rats. The authors speculated that Aβs might bind the CaSR because they have, just like polyamines, orderly spaced arrays of positive charges (Ye et al., 1997).

In this regard, the specific formation of plasma membrane Aβs•CaSR complexes and their subsequent endocytosis in cultured cortical untransformed adult human astrocytes could be proven by using the *in situ* proximity ligation assay (isPLA), which reveals the specific formation of stable complexes between two molecules placed within a 30 nm range (Dal Prà et al., 2014a; Dal Prà et al., 2014b; Dal Prà et al., 2015b). The latter results implied that since all types of human neural and cerebrovascular cells express the CaSR, they are vulnerable to the neurotoxic effects driven by pathological Aβ•CaSR signaling (Chiarini et al., 2016). However, it remains to be ascertained whether at the level of CaSR•GABABR1 heterodimers of human cortical astrocytes and neurons Aβs•GABABR1 complexes also form and what their functional roles would be under both physiological and pathological conditions: topics worth investigating further.

Moreover, a genetic analysis study on cohorts of 435 healthy controls and 692 SAD patients showed that an intron 4 polymorphic dinucleotide repeat marker of the *CASR* gene associated with an AD susceptibility, while three nonsynonymous SNPs of exon 7 were linked with an AD propensity only in non-*APOEε4* allele carriers. Hence, variations in the *CASR* gene sequence may impact on SAD susceptibility especially in subjects having no *APOEε4* allele (Conley et al., 2009).

The *CASR* gene P1 and P2 promoters regulate its transcription by binding several transcription factors, including SP1, SP3, STAT1, STAT3, CREB, and NFκB, which concurrently control the expression of other AD-related genes (see for details and references: Chiarini et al., 2016). Therefore, the transcription factors regulation of CaSR expression is tightly linked to the pathophysiology of AD.

It is well known that Aβ42-os simultaneously bind to several other CNS cells surface receptors besides the CaSR (see for details and references: Chiarini et al., 2016). Therefore, Aβ42 os•CaSR signaling triggers a throng of cellular responses sosme authors include in the so called "*calcium dyshomeostasis*", such as toxic ROS overrelease from mitochondria, and intracellular Ca2+ surges *via* NMDARs' activation driving further mitochondrial ROS releases (Kam et al., 2014; Jarosz-Griffiths et al., 2016).

However, it must be stressed here that the pathological Aβ42-os•CaSR signaling performs much more AD-specific upstream feats than those just mentioned. In fact, it drives the overproduction and overrelease of Aβ42-os and p-Tau-os, the two main AD culprits, from human cortical neurons and astrocytes. Moreover, it also induces the production and release of surpluses of other neurotoxic agents, such as NO and VEGF-A, and likely others more, from the adult human cortical astrocytes. Additionally, the pathological Aβ42-os•CaSR signaling profoundly suppresses sAPP-α extracellular shedding from human astrocytes and neurons (Chiarini et al., 2016; Chiarini et al., 2017a; Chiarini et al., 2017b). These Aβ-os-elicited noxious effects associate with concurrent upsurges in the expression of APP, BACE1, and CaSR proteins. Remarkably, the crucial upshot of all the mentioned effects of Aβ42-os•CaSR signaling is the death of human cortical neurons both *in vitro* (Armato et al., 2013) and in the *in vivo* brain. In the latter, the progressive disconnections of neural circuits—a cause of advancing cognitive decline—and a chronic diffuse reactive neuroinflammation eventually lead to full blown or symptomatic AD (Crimins et al., 2013; Kayed and Lasagna-Reeves, 2013; Medeiros et al., 2013).

Moreover, a study using *3xTg* AD-model mice showed that the amount of brain CaSR immunoreactivity progressively increased with age, particularly in areas where Aβ42 fibrils accumulate most, such as the hippocampi. Thus, local fibrillar Aβ42 buildup and CaSR expression raise in parallel in both Aβ-exposed human cortical neurons and astrocytes cultured *in vitro* and in the hippocampi of *3xTg* AD-model mice (Armato et al., 2013; Chiarini et al., 2016; Gardenal et al., 2017). This soaring expression of neural cells' CaSRs associates with a declining expression of inhibitory GABABR1as (Chang et al., 2007; Kim et al., 2014).

Whereas GABAB and taste receptors obligatorily function as heterodimers (Jones et al., 1998; Nelson et al. 2002), mGluRs and CaSR function both as disulfide-linked homodimers (Zhang et al., 2001; Pidasheva et al., 2006) and as CaSR/GABABRs, CaSR/mGlu1αR and CaSR/mGlu5R heterodimers (Gama et al., 2001). Ectopic overexpression and coimmunoprecipitation studies revealed that CaSR/GABABR1a heterodimers do affect CaSR protein expression in opposing ways. The total and cell surface expression and signaling of the CaSRs were suppressed by coexpressing GABABR1as, being instead increased (i) by co-expressing GABAB2Rs; (ii) by knocking out GABABR1a in mouse brains; and (iii) by deleting GABABR1a in cultured hippocampal neurons. The GABABRs and CaSRs form heterodimers as soon as they are synthesized, since these protein complexes are already detectable around the cells' nuclei and in the endoplasmic reticulum. In such early complexes GABABRs bind an immature form of the CaSR. Clearly, GABABR1a and GABABR2 subunits compete for the CASRs. The CaSR/GABABR heterodimers appear to have altered pharmacological properties with respect to the prevailing CaSR homodimers. Results gained using (i) the GABABRs agonists baclofen and GABA, (ii) the GABABR1a antagonist CGP-3548, and (iii) GABABR1a expression knockdown in cultured mouse growth plate chondrocytes indicated that GABABR1a can elicit both CaSR-independent and CaSR-mediated actions. However, divergent results gained from different experimental models suggested that an endogenous expression or a targeted overexpression of one or more of these receptors, coexisting differences in ligands and in their relative quantities and in downstream intracellular signaling pathways could elicit unlike upshots under various physiological and/or pathological conditions (Gama et al., 2001; Chang et al., 2007). During the initial phases of disease progression in AD-model animals, the decline of GABABR1s' availability, which concurs with CaSR's overexpression, induced a *neuronal hyperactivity* in hippocampal and cerebrocortical circuits, whose upshot was functional impairment (Busche and Konnerth 2015). The mechanism(s) underlying this loss of neuronal working capability remain(s) unclear: an overconsumption of O2 on the part of the hyperactive neurons might be a contributory factor. Nothing is so far known about the existence and pathophysiological roles of CaSR heterodimers in cortical human untransformed astrocytes and neurons. Therefore, to-date the impacts (if any) the CaSR/ GABABRs and CaSR/mGluRs heterodimers might exert on human AD's course and on anti-AD therapeutic approaches remain to be assessed.

Notably, in cortical adult human astrocytes the pathological Aβ•CaSR signaling heavily affects the APP holoprotein metabolism significantly deflecting it from its physiological *NAP* (**Figure 3**). APP's *NAP* typically obstructs the *de novo* production of Aβ42s/Aβ42-os since the α-secretases (mainly ADAM 10) cut the APP molecule just within the Aβ42 amino acid sequence (Kuhn et al., 2010) (**Figures 1** and **3**). Notably, APP's *NAP* prevails over APP's *AP* in untreated (control) cortical adult human astrocytes, which directly shed all the sAPP-α they produce into the environment while secreting only tiny amounts of monomeric Aβ42 (Chiarini et al., 2017b). Hence, it has been posited that by constitutively releasing substantial amounts of sAPP-α, which is an agonist of GABABR1as (Rice et al., 2019), human astrocytes could continually abate any noxious neuronal hyperexcitability.

(or calcilytics). Under physiological conditions, the *NAP* of APP largely prevails in cortical human astrocytes and neurons. Conversely, the pathological Aβ•CaSR signaling hugely enhances the APP holoprotein's *AP* at the expense of *NAP* in both human cell types. This leads to a surplus synthesis, intracellular accumulation, and extracellular release of Aβ42-os. The latter spread extracellularly to bind and activate the signaling of the CaSRs of adjoining teams of astrocytes and neurons (Chiarini et al., 2017a). Such self-sustaining vicious cycles amplify and propagate the pathological Aβ•CaSR signaling and its neurotoxic effects to wider and wider cortical areas. The Aβ•CaSR signaling also increases the activity of the glycogen synthase kinase-3β (GSK-3β), which strongly phosphorylates Tau proteins at amino acid sites typical of Alzheimer's disease (AD). The thus hyperphosphorylated Tau proteins also form oligomers (p-Tau-os) that are next released extracellularly within exosomes (not shown), thereby starting the tauopathy typical of AD. Other noxious effects of Aβ•CaSR signaling, such as increases in the synthesis and release of nitric oxide (NO) and vascular endothelial growth factor-A (VEGF-A), and other proinflammatory agents are not shown here for the sake of clarity. The crucial upshot of the harming effects of pathological Aβ•CaSR signaling is the progressive death of the cortical human neurons crucially involved in memories and cognition processing. In a most striking fashion, highly selective CaSR NAMs (calcilytics) suppress *all* the just mentioned neurotoxic effects brought about by pathological Aβ•CaSR signaling thus restoring the APP's *NAP*, Tau, NO, and VEGF-A to their physiological settings and consequently preserving the viability and function of human neurons notwithstanding the presence of Aβ peptides. Hence, NAMs could stop AD progression, safeguard the survival of the cortical human neurons, and preserve the memories, cognitive and coping capabilities of the patients. — blocking effects; +++, stimulating effects.

Under the same basal conditions, CaSR signaling is only modulated by extracellular cations levels, particularly by the [Ca2+]e. On the other hand, a dramatic change in this mechanism occurs when increased quantities of exogenous Aβs bind the human astrocytes' (and neurons') CaSRs and activate their pathological signaling that strongly promotes APP's *AP* over APP's *NAP* (**Figure 3**). This leads to an excess production, accumulation, and secretion of neurotoxic Aβ42/Aβ42-os from the cortical astrocytes and from the neurons in which an alike APP's *AP* mechanism operates (Armato et al., 2013). Concurrently, the astrocytes' and neurons' intense extracellular shedding of sAPP-α is curtailed by ~70%, while sAPP-α abnormally accumulates within the cells (Chiarini et al., 2017a). On the basis of such results the authors posited that an ongoing Aβ•CaSR signaling that would spread in vicious cycles from teams to teams of "*master*" astrocytes' and "*client*" neurons could cause a substantial loss of the neurotrophic and neuroprotective effects otherwise brought about by extracellularly shed sAPP-α, including its agonistic action on GABABR1as, thereby favoring a harmful neuronal hyperexcitability. In addition, the Aβ42/Aβ42-os-exposed human astrocytes and neurons could simultaneously release increasing amounts of neurotoxic Aβ42-os (Armato et al., 2013), p-Tau-os (within exosomes) (Chiarini et al., 2017a), NO, VEGF-A (Dal Prà et al., 2005; Chiarini et al., 2010; Dal Prà et al., 2014a; Dal Prà et al., 2014b; Chiarini et al., 2016). and likely other noxious agents. Therefore, it would not be surprising that under such dire circumstances cortical human neurons keep losing synapses and consequently die. Interestingly, in line with the just mentioned findings, CSF levels of sAPP-α significantly decrease in LOAD/ SAD patients (Lewczuk et al., 2010, which indirectly confirms the substantial fall of its extracellular shedding from human astrocytes (Chiarini et al., 2017b).

## CaSR NAMs as Potential Anti-AD Therapeutics

As mentioned above, several PAMs and NAMs of the CaSR are available. L-α-amino acids with an aromatic ring and positively charged amino groups (NH3+) are naturally occurring CaSR PAMs (Lee et al., 2007). Synthetic phenylalkylamine CaSR PAMs ("*calcimimetics*"; e.g. AMG 416, Cinacalcet, and NPS R-568) having two-to-four aromatic rings and NH3+ groups have been synthesized. PAMs augment the CaSR's sensitivity to activation by [Ca2+]e and hence lower the EC50 for [Ca2+]e. Notably, three CaSR PAMs, i.e. Evocalcet, Etelcalcitide, and Cinacalcet, have successfully reached the clinical use to mitigate primary and secondary hyperparathyroidism and tumor-elicited hypercalcemias (Nemeth and Goodman, 2016).

CaSR NAMs are small amino-alcohol molecules like NPS 2143 (Nemeth et al., 2001), Calhex 231 (Kessler et al., 2006), NPSP795 (Gafni et al., 2015), or quinazolinones like ATF936 and AXT914 (Wildler et al., 2010). CaSR NAMs right-shift the [Ca2+]e response curve decreasing the CaSR sensitivity to [Ca2+] e and thus increasing the EC50 for [Ca2+]e (Ferry et al., 1997; Huang and Breitwieser, 2007). As previously anticipated, both PAMs and NAMs bind the 7TM domain of the CaSR. The CaSR binding pockets of NAMs and PAMs partially overlap but are not identical. NAMs bind between the TM3 and TM5 loops, whereas both PAMs and NAMs attach between the TM6 and TM7 loops (Petrel et al., 2003; Petrel et al., 2004; Miedlich et al., 2004). It has been shown that point mutated residues of the 7TM helices (i.e. Phe668, Phe684, Trp818, Phe821, Glu837, and Ile841) lessen the antagonism of CaSR NAM NPS 2143 (Petrel et al., 2003; Petrel et al., 2004; Miedlich et al., 2004). CaSR's allosteric agonism and antagonism are modulated *via* the involvement of distinct amino acids and mechanisms (see for further details Leach et al., 2016; Keller et al., 2018). The identification of synthetic allosteric modulators of the CaSR has prompted searches for their therapeutic applications in diseases in which the CaSR signaling is dysfunctional (Hannan et al., 2018). However, hitherto the therapeutic potentials of both PAMs and NAMs of the CaSR have been only modestly exploited (Nemeth, 2013; Saidak et al., 2009; Widler, 2011; Ward et al., 2012). Like other GPCRs, CaSRs exhibit the "*ligand-biased signaling*" feature, i.e. in a certain type of cell a signaling pathway may be steadily picked up over the others according to the specific ligand involved (Leach et al., 2015). Interestingly, NAMs and PAMs too can induce this biased signaling, which in future might allow to therapeutically target a specific cell type over others (see for details: Davey et al., 2012; Leach et al., 2015; Hannan et al., 2018).

Let's zero in on an important NAMs' feature: they enhance parathyroid hormone (PTH) secretion from the parathyroid glands and increase blood calcium levels (calcemia) (Nemeth, 2004; Nemeth, 2013). Several phase II clinical trials were undertaken to assess CaSR NAMs potential therapeutic efficacy in women with postmenopausal osteoporosis based on the assumption that the released PTH would stimulate osteogenic processes. However, these trials failed because NAMs induced a several hour-lasting oversecretion of PTH that stimulated both osteogenic and osteolytic processes in the osteoporotic bones. These failures prompted to search for new CaSR NAMs inducing a lesser and shorter-lasting PTH release (Nemeth, 2013; Riccardi and Kemp, 2012; Davey et al., 2012; Ward et al., 2012). The same failures also demoted CaSR NAMs from the drugs potentially beneficial in humans even because of the modest hypercalcemia (hyperparathyroidism) they induced. However, attempts were performed to treat hypoparathyroidism and autosomal dominant hypocalcemia (ADH) driven by gain-of-function CaSR mutations with CaSR NAMs (Nemeth, 2013; White et al., 2009; Letz et al., 2010; Park et al., 2013; Nemeth and Goodman, 2016). The use of CaSR NAMs was also considered in cases of

breast and prostate carcinomas to prevent bone metastases, which are established through a CaSR-mediated signaling (Liao et al., 2006; Mihai et al., 2006). Other potential therapeutic uses of CaSR NAMs have included asthma attacks (Yarova et al., 2015); pulmonary artery idiopathic hypertension (Yamamura et al., 2012; Yamamura et al., 2015); stroke (Kim et al., 2014); and, last but not least, LOAD/SAD and EOFAD (Armato et al., 2013; Chiarini et al., 2016; Chiarini et al., 2017a; Chiarini et al., 2017b; Chiarini et al., 2017c).

The use of CaSR NAMs as therapeutics in SAD/LOAD and EOFAD is supported by the results gained from preclinical AD models "in Petri dishes" made up by untransformed human cortical astrocytes and/or neurons. In fact, a 30-min administration of a CaSR NAM, be it NPS 2143 or NPS 89696, completely suppressed all the above-mentioned neurotoxic responses evoked by the pathological Aβ•CaSR signaling (Chiarini et al., 2010; Armato et al., 2013; Dal Prà et al., 2014b; Dal Prà et al., 2015a; Chiarini et al., 2016; Chiarini et al., 2017c). Therefore, the authors posited that *in vivo* administered CaSR NAMs would (A) preserve the shedding of neurotrophic and neuroprotective and GABABR1a agonist sAPP-α from the plasma membranes of astrocytes and likely neurons, thereby (i) obstructing the amyloidogenesis from APP and hence the cerebral accumulation of neurotoxic soluble Aβ42-os and fibrillar Aβ42 polymers, and (ii) abating the noxious neuronal hyperexcitability *via* sAPP-α•GABABR1a signaling; (B) suppress the surplus synthesis and exosomal intrabrain dissemination of neurotoxic p-Tau-os and the consequent hypertoxic effects elicited by combined actions of the Aβ42 os/p-Tau-os duet (Ittner and Götz, 2011; Chiarini et al., 2017a); (C) reduce the increased synthesis and secretion of neurotoxic amounts of NO, VEGF-A, and likely other neurotoxic agents; (D) suppress any other harmful effects elicited by Aβ•CaSR signaling in oligodendrocytes, microglia, cerebrovascular cells of any kind; and (E) safeguard the blood–brain barrier (BBB) functional integrity. The above *in vitro* results also indicate that NAM efficacy persists notwithstanding a continued presence of soluble Aβ-os, fibrillar Aβs, and p-Tau-os (Chiarini et al., 2017a; Chiarini et al., 2017b; Chiarini et al., 2017c). Therefore, it is likely that CaSR NAMS could safeguard *in vivo*, as they do *in vitro* (Armato et al., 2013), the viability and functions of the cortical human neurons preserving the integrity of critical cognition-essential upper cerebral cortical regions (Choi et al., 2013; Lee et al., 2013; Barateiro et al., 2016). In brief, CaSR NAMs would uphold the patients' ability to record and recover memories and to deal with their daily needs. Most important, the relatively cheap to synthesize CaSR NAMs appears to be the so far unique class of anti-AD therapeutics capable of concurrently targeting the multiple noxious effects triggered by pathological Aβs•CaSR signaling in human neurons, astrocytes, and the other brain cell types (Chiarini et al., 2010; Dal Prà et al., 2011; Armato et al., 2013; Dal Prà et al., 2014b; Dal Prà et al., 2015a; Chiarini et al., 2016).

Ischemic neuronal injury is known to locally generate Aβs surpluses (Ishimaru et al., 1996). More recent studies showed that the intraventricular administration of CaSR NAMs did decrease the death of neurons in the cortical *penumbra* zone of animal models of ischemia/hypoxia/stroke by effectively suppressing the concurrent *acute* increase in the local Aβ-os production (Kim et al., 2014; Bai et al., 2015). These results further strengthen the idea that pathological Aβ•CaSR signaling is crucially involved in both acute (ischemia/stroke) and chronic (AD) conditions causing neuronal death (Armato et al., 2013; Dal Prà et al., 2015a; Chiarini et al., 2016).

*Fyn* kinase inhibitor Saracatinib (AZD0530) and NMDAR inhibitors Memantine and Nitromemantine were endorsed as drugs to counteract the neurotoxicity driven by the extracellular accumulation of Aβ42-os (Kaufman et al., 2015). However, CaSR NAMs act on Aβ•CaSR signaling well upstream of *Fyn* and NMDARs. In addition, CaSR NAMs obstruct any cytotoxic effects and likely also any impediments to proliferation and differentiation of neural stem cells in the dentate gyrus subgranular zone (Unger et al., 2016).

Concerning their most salient pharmacological characteristics, because of their lipid-soluble chemical structures and limited numbers of electrical charges, CaSR NAMs traverse the BBB. They can be administered by any route: hitherto the oral route has been the preferred one for clinical trials. Rodents could endure NAM NPS 2143 administration with no serious off-target effects being reported (Nemeth, 2002; Kim et al., 2014). During the failed phase I and phase II clinical trials assessing NAMs antiosteoporosis activity, human subjects also satisfactorily tolerated the administration of novel NPS 2143 derivatives, which affected PTH release less intensely (no record was taken concerning any brain-related effects). In general, the safety data collected from the clinical trials of CaSR NAMs did not record any major side-effect. Obviously, the calcemia levels had to be checked periodically due to NAM-elicited increases in plasma PTH levels (Nemeth and Shoback, 2013).

## CONCLUSIONS

This survey--necessarily short given the huge amount of literature concerning this verily fascinating topic--leads us to several closing considerations. First and foremost, a lot of data from AD-model animals had to be forcibly mentioned because analogous human data are not available. Therefore, there is still quite a lot to discover and learn about the physiological roles of family C GPRS particularly in relation to human CNS and other viscera. Second, it is undeniable that some of these GPCRs could play central roles on human AD. Our work has been mainly, but not exclusively, based upon the experimental exploitation of human cortical astrocyte cultures and has focused on the pivotal role pathological Aβ•CaSR signaling exerts on the onset and progression of AD

## REFERENCES


and on the potentially beneficial therapeutic effects CaSR NAMs could exert in LOAD/SAD patients. The interactions of CaSR heterodimers with other family C GPCRs, e.g. GABABRs and group I mGluRs, still constitute a mostly unexplored field of endeavor and their impact on AD onset and progression (if any) needs to be clarified. Notably, even in the gene mutations-driven EOFAD, CaSR NAMs could bring to bear mitigating and life-lengthening upshots by suppressing the additional aggravating consequences brought about by the concurrent Aβ•CaSR signaling adding up to those stemming from the mutated genes. Third, any possible AD-promoting effects of CaSR PAMs (calcimimetics) in humans should be thoroughly investigated since in our preclinical *in vitro* AD model PAM NPS R-568 significantly increased Aβ42/Aβ42-os release from untransformed human adult cortical astrocytes (Armato et al., 2013). Fourth, we wish to add a last comment about CaSR NAMs as candidate therapeutics for human AD. For reasons pertaining to normal physiology, CaSR NAMs failed their initial task as antiosteoporosis therapeutics (Nemeth, 2004; Nemeth, 2013; Nemeth and Shoback, 2013; Nemeth and Goodman, 2016). Moreover, the induction of a mild hypercalcemia by CaSR NAMs has been a bit too much stressed as "hyperparathyroidism" creating a prejudice against their use in humans. Nevertheless, one should remember that no drug is devoid of unwanted and/or off-target effects: the chemotherapeutics administered to oncological patients are a striking example of this. Therefore, CaSR NAMs' rather slight off-target effects, chiefly the mild controllable hypercalcemia, should be objectively weighed against CaSR NAMs' crucial capability of averting the worsening loss of memories and cognitive abilities, including recognition of the self, and the later unavoidable demise AD would inexorably deliver.

## AUTHOR CONTRIBUTIONS

All the authors i.e. IDP, UA, and AC, equally contributed to the bibliographic searches and to the conception, discussion, and writing of the manuscript.

## FUNDING

This work was supported by the FUR (Fondo Universitario per la Ricerca) 2018 of the Ministry of Italian University and Research to IDP and AC.


the production of RANTES in glial cell cultures. *J. Neurosci.* 22, 5403–5411. doi: 20026585


proinflammatory cytokines stimulate VEGF-A secretion by cultured, early passage, normoxic adult human cerebral astrocytes. *J. Alzheimer's Dis.* 21, 915– 926. doi: 10.3233/JAD-2010-100471


receptor activation. *Neuropharmacology* 121, 100–110. doi: 10.1016/j. neuropharm.2017.04.019


Hick, M., Herrmann, U., Weyer, S. W., Mallm, J. P., Tschäpe, J. A., Borgers, M., et al. (2015). Acute function of secreted amyloid precursor protein fragment APPsα in synaptic plasticity. *Acta Neuropath.* 129, 21–37. doi: 10.1007/s00401-014-1368-x

Hofer, A. M., and Brown, E. M. (2003). Extracellular calcium sensing and signaling. *Nat. Rev. Mol. Cell Biol.* 4, 530–538. doi: 10.1038/nrm1154


L-amino acids enhances ERK1/2 phosphorylation. *Biochem. J.* 404, 141–149. doi: 10.1042/BJ20061826


and patterning. *Trends Pharmacol. Sci.* 20, 73–78. doi: 10.1016/ S0165-6147(99)01308-5


**Conflict of Interest:** 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 © 2019 Dal Prà, Armato and Chiarini. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Ethanolic Extract of *Orthosiphon stamineus* Improves Memory in Scopolamine-Induced Amnesia Model

*Thaarvena Retinasamy, Mohd Farooq Shaikh\*, Yatinesh Kumari and Iekhsan Othman\**

*Neuropharmacology Research Laboratory, Jeffrey Cheah School of Medicine and Health Sciences, Monash University Malaysia, Bandar Sunway, Malaysia*

#### *Edited by:*

*Jacob Raber, Oregon Health and Science University, United States*

#### *Reviewed by:*

*Boyer D. Winters, University of Guelph, Canada Santiago J. Ballaz, Yachay Tech University, Ecuador*

#### *\*Correspondence:*

*Mohd Farooq Shaikh farooq.shaikh@monash.edu Iekhsan Othman iekhsan.othman@monash.edu*

#### *Specialty section:*

*This article was submitted to Neuropharmacology, a section of the journal Frontiers in Pharmacology*

*Received: 21 December 2018 Accepted: 23 September 2019 Published: 29 October 2019*

#### *Citation:*

*Retinasamy T, Shaikh MF, Kumari Y and Othman I (2019) Ethanolic Extract of Orthosiphon stamineus Improves Memory in Scopolamine-Induced Amnesia Model. Front. Pharmacol. 10:1216. doi: 10.3389/fphar.2019.01216*

Alzheimer's disease (AD) is a chronic neurodegenerative brain disease which is characterized by impairment in cognitive functioning. *Orthosiphon stamineus* (OS) Benth. (Lamiaceae) is a medicinal plant found around Southeast Asia that has been employed as treatments for various diseases. OS extract contains many active compounds that have been shown to possess various pharmacological properties whereby *in vitro* studies have demonstrated neuroprotective as well as cholinesterase inhibitory effects. This study, therefore aimed at determining whether this Malaysian plant derived flavonoid can reverse scopolamine induced learning and memory dysfunction in the novel object recognition (NOR) test and the elevated plus maze (EPM) test. In the present study, rats were treated once daily with OS 50 mg/kg, 100 mg/kg, 200 mg/kg and donepezil 1 mg/kg *via* oral dosing and were given intraperitoneal (ip) injection of scopolamine 1 mg/kg daily to induce cognitive deficits. Rats were subjected to behavioral analysis to assess learning and memory functions and hippocampal tissues were extracted for gene expression and immunohistochemistry studies. All the three doses demonstrated improved scopolamineinduced impairment by showing shortened transfer latency as well as the higher inflexion ratio when compared to the negative control group. OS extract also exhibited memoryenhancing activity against chronic scopolamine-induced memory deficits in the longterm memory novel object recognition performance as indicated by an increase in the recognition index. OS extract was observed to have modulated the mRNA expression of CREB1, BDNF, and TRKB genes and pretreatment with OS extract were observed to have increased the immature neurons against hippocampal neurogenesis suppressed by scopolamine, which was confirmed by the DCX-positive stained cells. These research findings suggest that the OS ethanolic extract demonstrated an improving effect on memory and hence could serve as a potential therapeutic target for the treatment of neurodegenerative diseases like AD.

Keywords: Alzheimer's disease, *Orthosiphon stamineus*, scopolamine, animal model, learning and memory

## INTRODUCTION

Neurodegenerative diseases have emerged to become a globally critical burden with the aging population. The number of Alzheimer's patients have steadily increased over the years with currently more than 46 million people living worldwide with the disease and the number is expected to increase to 131.5 million by 2050 (Mat et al., 2017). Alzheimer's disease (AD) is the most common cause of cognitive impairment in the elderly population and is characterized by various symptoms that include learning and memory impairment, cognitive dysfunction, language impairment and behavioral dysfunction like depression, agitation and psychosis that continue to become more severe with the disease progression (Brookmeyer et al., 2007; Mahdy et al., 2012). Thus, due to the debilitating nature of the disease, it continues to exist as a huge societal social and economic burden.

One of the key neuropathological features underlying the symptoms associated with Alzheimer's is neuronal loss and when examined microscopically, the presence of senile plaques and neurofibrillary tangles (NFTs) serves as the main features of the disease. A number of mechanisms have been conjectured to further elucidate the pathogenesis of Alzheimer's like cholinergic dysfunction, oxidative damage, beta amyloid toxicity, hyperphosphorylation of tau protein, and inflammation of senile plaques (Mahdy et al., 2012; Von Bernhardi et al., 2015). The cholinergic system that comprises of the cholinergic neurotransmitters play a vital role in memory processing whereby loss of the cholinergic neurons and its subsequent decrease results in learning and memory dysfunction characteristic of Alzheimer's (Watanabe et al., 2009; El-Marasy et al., 2012; Blake et al., 2014). Thus far there are yet to be diseasemodifying drugs approved for Alzheimer's. The medications available are only capable of temporarily alleviating the symptoms of cognitive impairment; however, they do not halt the inevitable progression of the disease. To date, four cholinesterase inhibitors or ChEI (tacrine, rivastigmine, donepezil and galantamine) and a partial NMDA receptor antagonist (memantine) are the only approved treatment options for AD. However, these drugs fail to completely cure the disease, which warrants a search for newer class of targets that would eventually lead to effective drugs for the treatment of AD (Obulesu and Rao, 2011). Thus, major therapeutic research is underway to explore the memory enhancing activities of natural products.

The pharmacological and therapeutic effects of traditional medicinal plants have been associated with the various chemical constituents isolated from their crude extract whereby in particular, active constraints that demonstrate antioxidant activity have been linked to play a central role in various neurodegenerative diseases (Silva et al., 2005; Mahdy et al., 2012). *Orthosiphon stamineus* (OS) Benth. (Lamiaceae) is an Asian folklore medicinal plant that has been employed as treatments for various diseases like influenza, inflammation, urinary tract infections, and angiogenesis related conditions like cancer (Geng et al., 2013; Yehya et al., 2018). OS have been reported to demonstrate anti-inflammatory, antioxidant, antibacterial and hypoglycemic effects (Awale et al., 2003; Akowuah et al., 2004; Ho et al., 2010; Abdelwahab et al., 2011; George et al., 2015). Additionally, several scientific studies have also reported the safety profile of 50% ethanol extract of OS in *in vivo* rat models and the LD50 have been said to be more than 5000 mg/kg (Chin et al., 2008; Mohamed et al., 2011; Yehya et al., 2018). Phytochemical studies have demonstrated that OS leaves extracts contain more than 20 phenolic bioactive compounds like rosmarinic acid, 2,3-dicaffeoyltartaric acid, eupatorine, sinesitin, oleanolic acid, ursolic acid, pentacyclic triterpenes, and b-sitosterol (Awale et al., 2003; Shin et al., 2015; Yehya et al., 2018). Among these active compounds, rosmarinic acid has been reported to be the main flavonoid present in the 50% ethanol extract of OS extract and plays a central role for the various pharmacological activities exerted by the OS extract. Flavonoids which are the principal group of polyphenols are also reported to be efficacious in decreasing oxidative stress and are said to promote various physiological benefits, particularly in learning and memory, scavenging free radicals and cognitive impairment (Bhullar and Rupasinghe, 2013; Bhullar and Rupasinghe, 2015; Ghumatkar et al., 2015). Besides that, standardized ethanolic extract of OS were also found to be able to reverse age-related deficits in short-term memory as well as prevent and reduce the rate of neurodegeneration (George et al., 2015).

Additionally, preliminary studies of OS extract have also demonstrated neuroprotective and choline esterase inhibitory effects; this in turn further indicates OS extract's potential in prompting CNS related reactions. Although OS extract possesses various uses, there are yet no studies on its neuropharmacological activities against AD-like conditions. Therefore, this present study aimed at distinguishing the anti-amnesic potential of this plant derived flavonoid memory deficits in a rat model of cognitive impairment caused by scopolamine.

## MATERIALS AND METHODS

## Plant Materials

The 50% ethanolic OS extract was procured from NatureCeuticals Sendirian Berhad, Kedah DA, Malaysia. The extract from leaves of OS was prepared under GMP-based environment using DIG-MAZ technology by Natureceuticals Sdn. Bhd., Malaysia. The DIG-MAZ is an extraction system which involves all of the key extraction processes like percolation, digestion, maceration, and distillation. The extract was kept in an airtight container until further experimentations. The OS extract was dissolved in distilled water and filtered using a membrane filter unit (0.22 lm) before being administered to the rats for the study.

## Experimental Animals

In-house-bred adult male Sprague Dawley rats weighing between 200–300 g and between 6 and 8 weeks old were acquired from the animal facility of School of Medicine and Health Sciences of Monash University Malaysia. The rats were kept and maintained in cages under standard husbandry conditions (12:12 h light/ dark cycle, at controlled room temperature (22 ± 2°C), stress free, water *ad libitum*, standard diet, and sanitary conditions). Prior to the experiment, the rats were allowed to acclimatize for a period of 1 week to reduce environmental stress. All the experimental protocols were approved and conducted according to the approval of the Monash Animal Research Platform (MARP), Australia with the reference number MARP/2016/028.

## Experimental Design

The range of OS extract doses was determined following the pre-screening results. OS extract, donepezil and scopolamine were prepared by dissolving it in saline. Normal control rats were administered with saline throughout the experiment. The treatments were given both orally and intraperitoneally (i.p.) at a volume corresponding to 0.1 ml/100 g of bodyweight. All experiments were performed in a balanced design (eight animals per group) to avoid being influenced by order and time. The behavioral studies were divided into two categories namely the nootropic and the scopolamine models.

## Nootropic Model

Group 1: Normal control (saline)

Group 2: Positive control [Donepezil (DPZ) 1 mg/kg]

Group 3: Low dose of OS (50 mg/kg OS)

Group 4: Medium dose of OS (100 mg/kg OS) Group 5: High dose of OS (200 mg/kg OS)

## Scopolamine Model


Group 3: Negative control (Scopolamine 1 mg/kg)


For the nootropic activity, all the groups received pretreatment orally for 6 consecutive days before being subjected to a battery of behavioral tests from Day 6 until Day 8 for the novel object recognition (NOR) and the elevated plus maze (EPM) tests, respectively, as observed in **Figure 1**. For the scopolamine model, amnesia was induced in all the groups except the control group by daily intraperitoneal injections of scopolamine (1 mg/ kg) for 9 days after OS extract pre-treatment (Day nine to Day 17). Thirty minutes prior to the administration of scopolamine, NOR was conducted on Day 10 and Day 15, and EPM was carried out on Days 11 and 12, and Days 16 and 17 of the study as seen in **Figure 1**. At the end of the experiment, the rats were sacrificed, and their brains were isolated for further biochemical and immunohistochemistry analysis.

## Novel Object Recognition

In the object recognition task, the experimental apparatus consisted of an open field box (40 × 40 × 40 cm) made of black acrylic material. The method used was the same as described by (Ennaceur and Delacour, 1988), with slight modifications. The behavior test was conducted between 9:00 AM and 6:00 PM under dim red-light illumination conditions. The objects to be discriminated were two similar transparent cultured flasks filled with water and a toy Lego of the same height (new object). One day prior to the experiment, each rat was habituated to the open field box without any object for 10 min. On the experiment day, during the first trial, each rat was placed in the open field for 5 min and allowed to freely explore the two identical objects (transparent cultured flasks with water). After 90 min of post-training session, one old object used during the training session was replaced by a novel object and the rat was left to explore the objects for 2 min. The time spent with each object was recorded and evaluated using SMART software version 3.0 (Panlab, Harvard Apparatus). Both objects presented during the test session were different in texture, color, and size. The open field box was cleaned with 70% ethanol between runs to minimize scent trails. The recognition index was computed using the formula [TB/(TA + TB) \* 100] where TA and TB are time spent exploring familiar object A and novel object B respectively (Batool et al., 2016). Exploration of an object was deemed when a rat sniffed or touched the object with its nose and/or forepaws.

## Elevated Plus Maze Test

The elevated plus maze test measures anxiety in animals but a significant parameter measured in EPM called the transfer latency (the time taken for the animal to move from an open arm to the closed arm) has been shown to be noticeably reduced if the animal has had prior experience of entering into the open and closed arm

and this reduced transfer latency has been demonstrated to be associated with memory process and an increase in inflexion ratio indicates nootropic activity. Additionally, several studies of various nootropics and amnesic agents on EPM have further reiterated this model as a widely accepted paradigm to study learning and memory processes in rodents (Itoh et al., 1990; Yadav et al., 2017). In EPM, transfer latency on Day 1 is deemed as acquisition (learning) and memory retention is then examined after 24 h.

The EPM apparatus consists of four arms of equal dimensions, i.e., two open arms (50 × 10 cm) that are crossed with two closed arms, enclosed by high walls of 40 cm high. These arms are connected with the help of a central square (10 × 10 cm) that gives an appearance of a plus sign to the maze. This maze is elevated from the ground by 50 cm. The method used was the same as described by (Vasudevan and Parle, 2006). The behavior test was conducted between 9:00 AM and 6:00 PM under dim red-light illumination conditions. The memory was assessed in EPM in two sessions, 24 h apart. During the training session, the rats were placed at the end of the open arm, facing away from the central platform. With the help of the stopwatch, the transfer latency (TL1) was noted, i.e., the time taken by rat with all its four legs to move into any one of the enclosed arms. If the rat failed to enter any one of the enclosed arms within 90 s, it was gently pushed into one of the two enclosed arms and the TL was assigned as 90 s. The rat was allowed to explore the maze for the next 10 s and then returned to its home cage. The maze was cleaned with 70% ethanol between runs to minimize scent trails. To assess memory, the retention test phase was carried out 24 h after the training session whereby a decrease in time latency (TL2) during the test session was deemed as an index of memory improvement. The cut-off time for each rat to explore the maze in both the phases (training and test) was 90 s.

The transfer latency was expressed as inflexion ratio, calculated using the formula:

$$IR = \frac{\left(L1 - L0\right)}{L0}$$

L0: Initial TL (s) on the 1st day and L1: TL (s) on the 2nd day.

## Tissue Processing

All the rats were sacrificed 1 hour after the behavioral test under ketamine-xylazine anesthesia. In each group half the rats (n = 4/group) were fixed with 4% paraformaldehyde (PFA) for immunohistochemistry analysis while the remaining half of the rats (n = 4/group) were used for gene expression analysis. The hippocampal region from the whole brain was isolated immediately and were homogenized on ice cold 200 µL Trizol and stored at −80°C for real-time PCR analysis.

## Gene Expression

Total RNA from the rat brain's hippocampal region was extracted following the method employed by (Bhuvanendran et al., 2018), with some minor modifications. The single-step method, phenolchloroform extraction and Trizol reagent (Invitrogen) was used to isolate the total RNA from the hippocampal region. Briefly, the tissues were homogenized in 200 µL of Trizol solution. The mixture was then extracted using chloroform and centrifuged at 135,000 rpm at 4°C. The alcohol was removed, and the pellet was washed twice with 70% ethanol and resuspended in 20 µL of RNase free water. RNA concentration was determined by reading absorbance at 260 nm using Nanodrop. A 500 ng amount of total RNA was reverse transcribed to synthesize cDNA using Quantitect® Reverse Transcription Kit according to the manufacturer's protocol. Then the mRNA expression of genes encoding cAMP response elementbinding protein (CREB1), brain-derived neurotrophic factor (BDNF), tropomyosin receptor kinase B (TrkB), and IMPDH2 in the hippocampus was measured *via* real-time PCR using the StepOne Real-Time PCR system. Subsequently, the cDNA from the reverse transcription reaction was subjected to Real-Time PCR using QuantiNova™ SYBR® Green PCR kit according to manufacturer's protocol. The comparative threshold (CT) cycle method was used to normalize the content of the cDNA samples, which consists of the normalization of the number of target gene copies versus the endogenous reference gene, IMPDH2.

## Immunohistochemistry

Immunohistochemical analysis was performed by assessing neurogenesis using Doublecortin (DCX) in the hippocampus. Four brain tissues from each group were immersed in the fixative solution, 4% paraformaldehyde overnight and were methodically cryoprotected in 10%, 20%, and 30% sucrose solution respectively for 24 h. The brains were then embedded in 15% Polyvinypyrrolidone, frozen using dry ice and cut into coronal frozen sections (40 µm) using a Leica CM3050 cryostat. The sections were stored in an anti-freeze buffer. The free-floating sections were subjected to endogenous peroxidase quenching with 1% H2O2 in methanol for 30 min. After washing with phosphate buffered saline, PBS, the tissues were treated with blocking buffer (1% Bovine Serum Albumin and 0.3% Triton X-100) for 1 hour followed by incubation with primary DCX (1:250, Abcam) antibodies overnight at 4°C. After washing with PBS, the tissues were then biotinylated with goat anti-rabbit secondary antibody (Abcam) for 2 h. The tissues were then subsequently washed with PBS and exposed to an avidin biotin peroxidase complex (Vectastain ABC kit, Vector) for another 2 h. The peroxidase activity was then visualized using a stable diaminobenzidine solution (DAB, Sigma). All immunoreactions were observed under a microscope (BX41, Olympus) and these results were quantified using DigiAcquis 2.0 software.

## Statistical Analysis

Data obtained from all studies were expressed as mean ± SEM. The data were analyzed using one-way analysis of variance (ANOVA) followed by Dunnett *post hoc* test. The P-values of \*P < 0.05, \*\*P < 0.01, and \*\*\*P < 0.001 were considered as statistically significant. All the experimental groups were compared with the Scopolamine (SCP) 1 mg/kg group except for the nootropic model where the experimental groups were compared with the control group.

## RESULTS

## Scopolamine Induced Model Behavioral Analysis

### Effect of OS Extract in Nootropic Model

The NOR test was used to evaluate whether OS treatment could reverse scopolamine-induced recognition impairment whereby the effect of OS extract at different doses were assessed following 7 days of pretreatment. Based on the results obtained in **Figure 2A**, pretreated group of OS extract demonstrated an increase in recognition index [F (4, 32) = 1.096, P = 0.3752] for novel object particularly observable in animals treated with dose of 50 mg/kg and 200 mg/kg OS extract. Overall, it can be said that the preference for the novel object was more or less the same among the OS treated groups when compared to the controls.

The memory function was also assessed using the EPM test to gauge the spatial long-term memory. Based on the results obtained in **Figure 2B**, the time taken for each rat to move

FIGURE 2 | Behavioral analysis for novel object recognition (NOR) and elevated plus maze (EPM). (A) represent the graph plot for the recognition indices in NOR for the nootropic model (B) represents the graph plot for the inflection ratios in EPM for the nootropic model. The behavioral analysis for (A,B) were compared to control group. Data are expressed as Mean ± SEM, n = 8 and statistical analysis by one-way ANOVA followed by Dunnett test \*P < 0.05 and \*\*P < 0.01.

from the open arm to either of the enclosed arms on the first trial (familiarization session), termed transfer latency 1, did not significantly differ between groups. However, during the test session, termed transfer latency 2, a decrease in time for transfer latency 2 was observed within the groups. Significant results were observed in rats administered with 100 mg/kg and 200 mg/kg OS extract thus indicating improvement in inflexion ratio [F (4, 35) = 3.713, P = 0.0127] observed among all groups. These results demonstrate that supplementation of OS extract significantly improved memory function in rats thus demonstrating optimal nootropic effects.

### Effect of OS Extract in Acute Scopolamine Model

In the acute scopolamine-induced memory impairment in rats as depicted in **Figure 3 (A1),** the percentage of recognition index for the positive control (donepezil 1 mg/kg), negative control (scopolamine 1 mg/kg) and the OS extract treated groups, were found to be unchanged, indicating that the acute scopolamineinduced memory was not impaired [F (5, 38) = 0.691, P = 0.6333].

For the EPM test, the scopolamine administered group (negative group) demonstrated a decrease in inflexion ratio when compared to the control and the other treated groups as depicted in **Figure 3 (B1)**. When donepezil (positive group), a well-established standard drug for Alzheimer's disease was administered, a significant increase in inflexion ratio was observed when compared to the scopolamine treated rats. Similarly, significant improvement in inflexion ratio [F (5, 42) = 23.32, P < 0.0001] was observed in all the 3 doses of OS extract with both 50 mg/kg and 100 mg/kg demonstrating a notable increase in inflexion ratio when compared to the scopolamine treated group indicating that the memory impairment induced by scopolamine was reversed. These results further reiterate that OS extract were able to improve retention memory.

### Effect of OS Extract in Chronic Scopolamine Model

In the chronic scopolamine model, the NOR test showed that there was a relatively large drop in the percentage of recognition index for the scopolamine treated group as shown in **Figure 3 (A2)**. The percentage of recognition index for all the OS extract groups were observed to have a significant increase [F (5, 42) = 13.74, P < 0.0001] when compared to the scopolamine treated group indicating improved memory retention.

For the EPM test as observed in **Figure 3 (B2)**, when chronic exposure of scopolamine was given to the rats, a notable decrease in inflexion ratio was observed in the scopolamine treated group whereas both 50 mg/kg and 100 mg/kg dose of OS extract demonstrated a significant improved inflexion ratio [F (5, 42) = 23.32, P < 0.0001] when compared to the scopolamine treated group indicating that the increase could be due to the repeated exposure of scopolamine. However, the 200 mg/kg OS extract demonstrated a significant increase in inflexion ratio when compared to the negative group, but when compared between the OS extract doses, it was much lower compared to the other two doses. Based on these results, we can say that OS extract does improved memory retention when exposed repeatedly to scopolamine.

scopolamine model respectively (B1 and B2) represents the graph plot for inflection ratios in EPM for both the acute and chronic scopolamine model. All the behavioral analysis was compared to the negative control (SCP 1 mg/kg). Data are expressed as Mean ± SEM, n = 8 and statistical analysis by one-way ANOVA followed by Dunnett test \*\*P < 0.01 and \*\*\*\*P < 0.0001.

## Scopolamine Induced Model Gene Expression Analysis

### Gene Expression in the Hippocampal Region

In the hippocampal region, the BNDF mRNA levels were found to be significantly down regulated [F (6, 49) = 3.948, P = 0.0027] when injected with scopolamine as compared to the control group, as depicted in **Figure 4A**. Similarly, even the CREB1 [F (6, 49) = 7.517, P < 0.0001] and TRKB [F (6, 49) = 10.23, P < 0.0001] mRNA levels were observed to be down regulated when given scopolamine as shown in **Figures 4B**, **C**, respectively. This down regulation was ameliorated significantly by OS extract pretreatment as compared with the negative (scopolamine 1 mg/kg) group. Moreover, in all the three mRNA expression levels, namely CREB1, BDNF, and TRKB were observed to be significantly higher when the rats were treated with 100 mg/kg OS extract. Similar up regulation in the mRNA levels were also observed in the positive group rats that were treated with donepezil.

## Scopolamine-Induced Model Immunohistochemistry Analysis

In the immunohistochemical studies, scopolamine injection was observed to have suppressed adult neurogenesis, shown as distributed dendrites and neuron bodies in the dentate gyrus (DG) region by DCX staining, particularly in the sub-granular zone (SGZ). Additionally, pretreatment with OS extract were observed to have ameliorated [F (5,18) = 12.74, P < 0.0001] the adult neurogenesis by enhancing immature neurons in the SGZ compared to the scopolamine treated group, as depicted in **Figure 5**.

## DISCUSSION

The present study demonstrated that pretreatment with OS extract improved memory retention as evident by the improved inflexion ratio observed in the EPM test as well as the increase in the recognition index observed in the OS treated rats. Previous studies have demonstrated scopolamine to show profound amnesic effects in various learning paradigms through the disruption of the cholinergic neurotransmission whereby when given acutely, scopolamine was said to produce spatial memory deficit (Goverdhan et al., 2012; Ghumatkar et al., 2015). In our study, similar results were observed whereby the scopolamine treated group in both the acute and chronic model for the EPM test demonstrated decreased inflexion ratio indicating impairment of spatial memory. Similarly, in the NOR test, the recognition index was decreased in the scopolamine treated group in the chronic model indicating cognitive deficit. However similar results were not observed in the acute model further corroborating that the NOR test were not influenced by the acute scopolamine treatment. Donepezil is a well-established drug used to treat dementia associated with AD (Sumanth et al., 2010; Sumanth et al., 2010) and was hence used as the positive control in our *in vivo* study as it was said to be able to reverse scopolamine induced memory impairment in previous studies (Cachard-Chastel et al., 2008; Ghumatkar et al., 2015). The

rats that were treated with 50 mg/kg OS extract showed a decrease in transfer latency in which the rats were able to remember and enter the closed arm quickly compared to the training session, which was observable by the improved inflexion ratio, however the rats that were treated with 200 mg/kg did show improved memory retention as compared to the scopolamine treated group but not as that observed in the 50 mg/kg OS treated group. A ceiling effect was observed with higher doses. Therefore, based on the behavioral analyses, we can conclude that the spatial memory was improved in both the acute and chronic scopolamine model and this improved performance may be attributed to its enhanced cholinergic neuronal transmission.

Scopolamine is a non-selective muscarinic cholinergic receptor antagonist that inhibits the central cholinergic neuronal activity which in turn leads to impairment in spatial learning and memory in rodents and humans (Konar et al., 2011). The central cholinergic system is also found to be closely associated with neurogenesis and/or cell proliferation in the hippocampus (Yoo et al., 2011). In this study, we illustrated the nootropic and neuroprotective effects of OS extract in a scopolamine induced amnesia model. The medicinal value of OS extract has been well recognized, particularly in regard to its anti-oxidant and anti-inflammatory activities (Arafat et al., 2008) In particular, rosmarinic acid which is the main flavonoid component of OS extract has demonstrated various pharmacological properties that may potentially hinder neurodegeneration and improve memory and cognitive functioning (Essa et al., 2012). Thus, this cocktail of flavonoids could be in turn responsible for the positive behavioral results observed.

The underlying mechanism for the improvement in memory retention observed in the behavioral studies was further explored by evaluating the biochemical parameters like expression of CREB1, BDNF. and TrkB genes in rats treated with OS extract and scopolamine. Adult hippocampal neurogenesis and neuroplasticity are modulated by many neurotrophic factors such as BDNF (Begni et al., 2017; Begni et al., 2017). BDNF is a small dimeric protein which is one of the

neurotrophic factors that play a vital role in regulating not only the neuronal development, maintenance and survival, but also in the cognition, formation and storage of memory. In 1991, reduced expression of BDNF were first seen in hippocampus samples from AD donors suggesting that this decrease may contribute to the progressive cell death characteristic of AD (Phillips et al., 1991). Furthermore, BDNF was also found to promote the survival of all major types of neurons related to functional changes in AD, and has been suggested as an essential contributor of the etiology of neurodegenerative disorders (Schindowski et al., 2008). As stated earlier, BDNF is involved in neuronal survival and plasticity that binds to high-affinity receptors, TrkB (tropomyosin receptor kinase B) (Givalois et al., 2004). Previous studies have also demonstrated that both BDNF and TrkB play a critical role in long-term synaptic plasticity in the adult brain (Schinder and Poo, 2000; Dwivedi, 2009). BDNF-TrkB interaction promotes the survival and differentiation of neurons and synaptic plasticity of the central nervous systems (Lu et al., 2008; Kim et al., 2010). Thus, a decrease in BDNF and its receptor, TrkB may lead to synaptic and cellular loss and memory deficits characteristic of AD. In the present study, the induction of scopolamine-induced amnesia showed suppression of BDNF and TrkB expressions in the hippocampus. Similar results were also observed in the prefrontal cortex whereby scopolamine reduced mRNAs of BDNF and TrkB. OS extract was found to have increased both the BDNF and TrkB stained cells in the hippocampus and the prefrontal cortex region. In the hippocampus, all the OS extract doses were found to be effective and showed maximum protection by increasing the BDNF and TrkB levels.

On the other hand, CREB1 is a co-factor of CREB and is essential for memory and synaptic plasticity in the central nervous system whereby disruption of phosphorylated CREB within the hippocampal region triggers the progression of neurodegenerative diseases like AD, Parkinson's disease and Huntington's disease (Lee et al., 2015). Previous studies have demonstrated that the activation of CREB ameliorated cognitive impairment *via* the cholinergic system (Kotani et al., 2006; Lee et al., 2015). This was congruent with our results whereby the expression of the CREB1 gene was reduced by scopolamine and pretreatment with OS extract markedly increased the CREB1 mRNA levels. So, it can be suggested that OS extract could be a potent treatment for neurodegenerative diseases and its possible mechanism might be modulating the cholinergic activity *via* the CREB-BDNF pathway.

The hippocampus is a pivotal region of the brain that is critical for learning and memory function and is highly susceptible to neuronal injury produced by scopolamineinduced cholinergic activity dysregulation, which can in turn trigger impairment of synaptic plasticity and loss of spatial learning memory (Mattson et al., 2002; Heo et al., 2014a). DCX is a marker of neuroblasts, neuronal precursor cells, and immature neurons. It is associated with structural plasticity in the adult mammalian brain, and has been used as a marker of newly formed neurons in the DG of the adult hippocampus (Bonfanti 2006; Heo et al., 2014b; Heo et al., 2014b). DCX is involved in neuronal migration and development, and it is continuously expressed during adult neurogenesis thus enabling it to be used to measure neurogenesis (Knoth et al., 2010; Heo et al., 2014b). Previous studies have reported decreased DCX expression during aging and thus decrease in neurogenesis (Brown et al., 2003; Hwang et al., 2008; Heo et al., 2014b). In our study, similar results were observed whereby the number of DCX-positive cells in the hippocampal DG was decreased in scopolamine induced rats, whereas the OS treated rats were observed to have increased number of DCX-positive cells. However, further research is necessary to verify its mechanism. Based on the behavior results for EPM, improved inflexion index was observed for both 50 and 100 mg/kg which was equivalent to the positive group, Donepezil indicating that OS at these doses were able to completely reverse the scopolamine induced memory impairment. This was further reiterated by the increase in dendrites and neuron bodies observed. For the 200 mg/kg dose, behavioral studies did show a slight improvement in inflexion index compared to the only scopolamine induced group but when compared to the positive group, Donepezil and the other 2 doses, 50 and 100 mg/kg groups, the improvement

## REFERENCES


was not that convincing. This result was further supported as there were dendrites and neuron bodies observed during the cell counting but not as much as the other 2 doses.

## CONCLUSION

In conclusion, the present work demonstrated that OS extract was able to revert the scopolamine induced amnesia in the rats thus further distinguishing its anti-amnesic effects. Additionally, we also established that the positive effects of OS extract could be mediated *via* the BDNF-TrKB pathway, CREB-BDNF pathway and also the hippocampal neurogenesis. This suggests that the OS extract could be a promising candidate as a memory enhancer or as a therapeutic treatment for neurodegenerative diseases like AD.

## ETHICS STATEMENT

All the experiments were conducted according to the approved protocols by the Monash Animal Research Platform (MARP) animal ethics committee in Australia.

## AUTHOR CONTRIBUTIONS

TR performed all the experiments and was responsible for the writing of the manuscript in its entirety. YK helped in designing of gene expression and immunohistochemistry study. MS and IO were involved in conceptualizing, designing of the study, result analysis, and manuscript editing. All authors gave their final approval for the submission of the manuscript.

## FUNDING

Authors are thankful to NKEA Research Grant Scheme (NRGS), Ministry of Agriculture and Agro-Based Industry Malaysia (Grant No. NH1014D066) for providing funding support.


food-borne bacteria. *Food Chem.* 122 (4), 1168–1172. doi: 10.1016/j. foodchem.2010.03.110


in athymic nude mice model. *J. Adv. Res.* 15, 59–68. doi: 10.1016/j.jare. 2018.05.006

Yoo, D. Y., Kim, W., Yoo, K. Y., Lee, C. H., Choi, J. H., Kang, I. J., et al. (2011). Effects of Nelumbo nucifera rhizome extract on cell proliferation and neuroblast differentiation in the hippocampal dentate gyrus in a scopolamine-induced amnesia animal model. *Phytother. Res.* 25 (6), 809–815. doi: 10.1002/ptr.3337

**Conflict of Interest:** 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 © 2019 Retinasamy, Shaikh, Kumari and Othman. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*