# GOLGI DYNAMICS IN PHYSIOLOGICAL AND PATHOLOGICAL CONDITIONS

EDITED BY : Jaakko Saraste, Vladimir Lupashin, Yanzhuang Wang and Kristian Prydz PUBLISHED IN : Frontiers in Cell and Developmental Biology

#### 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 978-2-88963-539-9 DOI 10.3389/978-2-88963-539-9

#### About Frontiers

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

#### Frontiers Journal Series

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

#### Dedication to Quality

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

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

#### What are Frontiers Research Topics?

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

# GOLGI DYNAMICS IN PHYSIOLOGICAL AND PATHOLOGICAL CONDITIONS

Topic Editors: Jaakko Saraste, University of Bergen, Norway Vladimir Lupashin, University of Arkansas for Medical Sciences, United States Yanzhuang Wang, University of Michigan, United States Kristian Prydz, University of Oslo, Norway

Citation: Saraste, J., Lupashin, V., Wang, Y., Prydz, K., eds. (2020). Golgi Dynamics in Physiological and Pathological Conditions. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-539-9

# Table of Contents

*05 Editorial: Golgi Dynamics in Physiological and Pathological Conditions* Kristian Prydz, Vladimir Lupashin, Yanzhuang Wang and Jaakko Saraste

*09 Identification of CDC42 Effectors Operating in FGD1-Dependent Trafficking at the Golgi*

Mikhail Egorov and Roman Polishchuk

*20 Golgi Complex Dynamics and its Implication in Prevalent Neurological Disorders*

Mario O. Caracci, Luz M. Fuentealba and María-Paz Marzolo


Christian Makhoul, Prajakta Gosavi and Paul A. Gleeson

*90 Defects in COG-Mediated Golgi Trafficking Alter Endo-Lysosomal System in Human Cells*

Zinia D'Souza, Jessica Bailey Blackburn, Tetyana Kudlyk, Irina D. Pokrovskaya and Vladimir V. Lupashin

*110 CREB3 Transcription Factors: ER-Golgi Stress Transducers as Hubs for Cellular Homeostasis*

Luciana Sampieri, Pablo Di Giusto and Cecilia Alvarez


Shaheena Shaik, Himani Pandey, Satish Kumar Thirumalasetti and Nobuhiro Nakamura

*166 The Structure and Function of Acylglycerophosphate Acyltransferase 4/ Lysophosphatidic Acid Acyltransferase Delta (AGPAT4/LPAAT*d*)* Mikhail A. Zhukovsky, Angela Filograna, Alberto Luini, Daniela Corda and Carmen Valente

	- Peter M. Luo and Michael Boyce

Sarah Oddoux, Davide Randazzo, Aster Kenea, Bruno Alonso, Kristien J. M. Zaal and Evelyn Ralston

*316 BML-265 and Tyrphostin AG1478 Disperse the Golgi Apparatus and Abolish Protein Transport in Human Cells*

Gaelle Boncompain, Nelly Gareil, Sarah Tessier, Aurianne Lescure, Thouis R. Jones, Oliver Kepp, Guido Kroemer, Elaine Del Nery and Franck Perez

*329 Protein Amphipathic Helix Insertion: A Mechanism to Induce Membrane Fission*

Mikhail A. Zhukovsky, Angela Filograna, Alberto Luini, Daniela Corda and Carmen Valente

# Editorial: Golgi Dynamics in Physiological and Pathological Conditions

#### Kristian Prydz <sup>1</sup> \*, Vladimir Lupashin<sup>2</sup> \*, Yanzhuang Wang<sup>3</sup> \* and Jaakko Saraste<sup>4</sup> \*

<sup>1</sup> Department of Biosciences, University of Oslo, Oslo, Norway, <sup>2</sup> Department of Physiology and Biophysics, University of Arkansas for Medical Sciences, Little Rock, AR, United States, <sup>3</sup> Department of Molecular, Cellular and Developmental Biology, University of Michigan, Ann Arbor, MI, United States, <sup>4</sup> Department of Biomedicine and Molecular Imaging Center, University of Bergen, Bergen, Norway

Keywords: Golgi apparatus, Golgi dynamics, Golgi structure, Golgi function, membrane trafficking

#### **Editorial on the Research Topic**

#### Edited and reviewed by:

Georg Haase, INSERM U1106 Institut de Neurosciences des Systèmes, France

#### \*Correspondence:

Kristian Prydz kristian.prydz@ibv.uio.no Vladimir Lupashin vvlupashin@uams.edu Yanzhuang Wang yzwang@umich.edu Jaakko Saraste jaakko.saraste@uib.no

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 08 December 2019 Accepted: 10 January 2020 Published: 29 January 2020

#### Citation:

Prydz K, Lupashin V, Wang Y and Saraste J (2020) Editorial: Golgi Dynamics in Physiological and Pathological Conditions. Front. Cell Dev. Biol. 8:7. doi: 10.3389/fcell.2020.00007 **Golgi Dynamics in Physiological and Pathological Conditions**

The internal reticular apparatus, first reported in 1898, and later called the Golgi apparatus, was initially observed by Camillo Golgi after his refinement of the silver nitrate technique for staining cells in the nervous system, for which he received the Nobel Prize in Physiology or Medicine in 1906. The true existence of the Golgi apparatus was, however, disputed for decades, and many scientists regarded it as a staining artifact (Palade and Claude, 1949). Only after the electron microscope became a more generally available tool for cell biologists was the Golgi apparatus accepted as an authentic cellular structure, and its existence demonstrated in practically all eukaryotic cells, from yeast to man. However, for reasons that are not completely understood, the Golgi assumes various forms in different cell types, ranging from tubular networks or individual cisternae (budding yeast S. cerevisiae) to separate cisternal stacks (e.g., invertebrates and plants) and a ribbon-like structure of interconnected stacks (vertebrates). As reflected in this collection, much of the recent efforts to define the mechanism of Golgi structure formation have been put on Golgi structural proteins including Golgins and GRASPs, actin and microtubule cytoskeletons, Rabs, and other GTPases, as well as certain other proteins (Ahat et al.; Egorov and Polishchuk; Lowe; Phuyal and Farhan; Satoh et al.; Shaik et al.). Although these proteins have their own characteristic functions, they coordinate with each other to maintain the structural and functional integrity of the Golgi.

The fascinating structure of the Golgi stacks is partly determined by GRASP proteins which also participate in a number of additional cellular processes, such as unconventional secretion and autophagy (Ahat et al.). In vertebrate cells, the multiple stacks are further organized into the Golgi ribbon, a continuous structure where the cisternal stacks are joined together by lateral connections (Saraste and Prydz). Several proteins have been shown to be involved in the regulation of ribbon structure, among these GRASP55 and GRASP65 (Ahat et al.) and the tethering protein giantin (Satoh et al.). Although the ribbon-like organization of the Golgi remains enigmatic, it is highly interesting with respect to certain pathological situations. Indeed, fragmentation of the Golgi ribbon is encountered in a number of neurological diseases (Caracci et al.; Makhoul et al.). The consequences of Golgi structural abnormalities on its function and cellular activities may vary in different experimental or disease conditions. For example, knocking down the GRASP proteins accelerates intra-Golgi trafficking (Xiang et al., 2013; Lee et al., 2014), while inhibition of GM130 and p115 results in the accumulation of COPI vesicles and reduced membrane trafficking (Seemann et al., 2000). Similarly, Aβ (amyloid β) induced Golgi reorganization in Alzheimer's disease has been proposed to increase APP (amyloid precursor protein) trafficking and Aβ production (Joshi et al., 2014), while GM130 knockout slows down ER-Golgi trafficking, resulting in Purkinje neuron loss and ataxia in mice (Liu et al., 2017). In addition, both the morphology (Makhoul et al.) and the internal environment (such as lumenal pH) of the Golgi apparatus are altered in cancer cells, leading to changes in glycosylation (Kellokumpu). Future investigations of the cause and effect of Golgi defects in disease will undoubtedly yield exciting findings essential for the understanding of both Golgi function and disease development.

Structural reorganization of the Golgi is also an integral part of naturally occurring processes, as best demonstrated in the case of cell division. During the late G2 phase, as cells prepare for mitosis, the Golgi ribbon is unlinked into individual stacks, which subsequently undergo further disassembly and vesiculation. In the course of these events, repositioning of Golgi elements is typically observed in the perinuclear region of many vertebrate cells, controlled by the duplicated centrosome and centrosome-nucleated microtubules of the forming mitotic spindle (Mascanzoni et al.). Notably, the pulling apart of the centrosomes during the late G2/early mitosis is accompanied by the evidently equal partitioning of the intermediate compartment (IC), a permanent membrane network that–unlike the Golgi stacks—keeps its properties during mitosis (Saraste and Marie, 2018).

The role of the Golgi apparatus as an important way station in anterograde trafficking along the secretory pathway was established during the 1960s and 1970s (see Farquhar and Palade, 1998, for a review), while retrograde transport via this organelle was described much later (Sandvig et al., 1992). In addition, the role of the Golgi apparatus as an "educational" site for glycoproteins, proteoglycans, and glycolipids is welldescribed and generally accepted. Accordingly, the majority of newly synthesized proteins that enter the Golgi apparatus at its cis-side carry N-linked glycans of identical structure, but leave the trans-Golgi region equipped with highly diverse glycans, specific for the actual species, the cell type, as well as the cell's developmental stage or degree of differentiation (Fisher et al.; Akintayo and Stanley). The ability of various transiting cargo molecules to obtain a healthy output of these and other Golgi modifications, however, requires mechanisms of membrane homeostasis and transport that are still subject to active investigation and dispute (Mironov and Beznoussenko; Saraste and Prydz; Makhoul et al.; Pantazopoulou and Glick). Not only the evaluation of competing Golgi models (Mironov and Beznoussenko), but also performing mathematical modeling (Fisher et al.) can advance our understanding of how cargo molecules that enter the cis-face of the Golgi apparatus are modified during Golgi transit, and how their final structures will turn out. The ongoing attempts to correlate the structural and functional dynamics of the Golgi apparatus are still absolutely required to achieve this goal. While the mechanisms of how different cargo molecules traffic through the Golgi stacks are still under debate, it remains even less clear how precise localization of Golgi resident proteins is achieved within the polarized stacks.

The swarms of COPI vesicles observed at the outskirts of the Golgi membranes are essential for normal Golgi function, although their engagement is still not fully understood due to partial knowledge of their exact cargo selection and composition, places of origin and destination, delivery mechanisms, interaction partners and regulatory modes (Pantazopoulou and Glick; Luo and Boyce; D'Souza et al.). The Rho GTPase Cdc42 is involved in the regulation of actin filament- and microtubuledependent Golgi positioning, in addition to interacting with COPI vesicles or tubules, thus potentially promoting anterograde transport toward the leading edge of migrating cells (Phuyal and Farhan). The trafficking capacity of these transport intermediates can be regulated in a number of ways (Luo and Boyce). For example, defects in the octameric COG complex that functions as a tether in COPI vesicle-mediated retrograde transport, not only affect traditional Golgi functions like glycosylation and sorting, but also exert effects elsewhere in the cell, in particular within the endo-lysosomal system (D'Souza et al.). Knockdown of the COG3 subunit of COG–or the ZW10 subunit of NRZ/Dsl1, another member of the CATCHR family of multisubunit tethering complexes–leads to the dispersal of the Golgi apparatus throughout the cytoplasm of metazoan cells (Zolov and Lupashin, 2005; Sun et al., 2007) in a process that requires both Rab GTPases and kinesin motor proteins (Liu et al.). The dynamic nature of the Golgi apparatus is underlined by the fact that while a large number of proteins are required to maintain its normal organization, treatments affecting a single structural or machinery component are often sufficient to destabilize its structure, as exemplified by the EGFR tyrosine kinase inhibitors BML-265 and AG1478 (Boncompain et al.).

In addition to the COPI transport machineries, the dynamic nature and maintenance of the Golgi apparatus crucially depend on efficient mechanisms of membrane fusion and fission. Both processes are influenced by lipid modifying enzymes, such as acylglycerophosphate acyltransferases (Zhukovsky et al.), membrane curvature-sensing proteins, and fission inducing-proteins (Zhukovsky et al.). To understand Golgi function completely it will be important to reconstitute Golgi fusion and fission in vitro using purified components and endogenous cargo.

A question that has been touched upon, but is far from being solved, is whether all the stacks in a Golgi ribbon handle the same cargo and have identical enzymatic contents to carry out the same protein and lipid modifications. While the nonlinked, wide-spread Golgi stacks in Drosophila cells differ in their enzymatic repertoire (Yano et al., 2005), the apical and basolateral routes in mammalian epithelial MDCK cells have also been shown to treat the same cargo molecule differently (Prydz et al., 2008). A related question is whether there may be lateral segregation of cargo within cisternae of the same Golgi stack. Interestingly, this was recently shown to be the case for HeLa cells, where two proteins heading for the endolysosomal system gradually underwent lateral segregation while passing through the Golgi (Chen et al., 2017). Here, the paper by Ernst et al. discusses recent findings showing that acylation can influence the lateral positioning of proteins in Golgi cisternae, and as a consequence, their anterograde transport efficiency (Ernst et al.). While anterograde transport of cargo also takes place after the stacks of the Golgi ribbon have been dispersed, for instance, due to breakdown of microtubules, an intact ribbon does seem to be important for the transport of large cargo molecules (Lavieu et al., 2014). In a new view of the Golgi ribbon, the non-compact zones linking the cisternal stacks are jointly occupied by the permanent IC elements and recycling endosomes (RE). As mentioned above, as a prelude for cell division, the stacks that undergo reversible break-down disconnect from these centrosome-linked compartments, which as a consequence would be allowed move to the spindle poles for partitioning into the forming daughter cells (Saraste and Marie, 2018; Saraste and Prydz).

Indeed, the Golgi apparatus is in intimate communication with both pre- and post-Golgi compartments. A number of important proteins recycle between the endoplasmic reticulum (ER) and the Golgi apparatus. YIPFα1A, a member of the YIPF protein family, functions at ER exit sites and interacts with COPII components, but can also localize to the IC and cis-Golgi, interact with YIPβ1A, and recycle back to the ER (Shaik et al.). Members of the CREB3 family of transcription factors move from the ER to the Golgi apparatus when the cell receives a proper signal. The activation of these proteins in the Golgi apparatus is based on two sequential proteolytic cleavage events. Upon cleavage, the N-terminal portions of the proteins, which are localized to the cytoplasmic side of the Golgi membrane, are released and become free to move into the nucleus (Sampieri et al.).

At the trans-side of the Golgi stacks, the trans-Golgi network (TGN) is an important site for Golgi exit of cargo molecules destined for various organelles of the endomembrane system and different plasma membrane domains. With the continuous improvement of the resolution of fluorescence-based light microscopes, it is now possible to observe the segregation of

#### REFERENCES


various cargo molecules and the membrane carriers that exit the Golgi at the level of the TGN (Huang et al.).

To fully (or at least better) understand the Golgi apparatus in various physiological and pathological conditions, one has to examine this organelle in a variety of tissues at different stages of differentiation, development or degeneration, for instance in neuronal (Caracci et al.; Rabouille and Haase, 2016) and muscle cells (Oddoux et al.). This is important to understand the adaptability of the Golgi apparatus to the cellular requirements, ensuring a healthy glycan output, as exemplified in this collection by the analysis of glycans of oocytes and sperm cells (Akintayo and Stanley). Equally important is to study a growing number of genetic diseases—such as the Aarskog-Scott syndrome (Egorov and Polishchuk), a faciogenital dysplasia caused by mutations in a GEF protein (FGD1) regulating the Rho GTPase Cdc42–that are found to affect Golgi structure and function.

Altogether, topical reviews, hypothesis and theory articles, and original studies in this collection illustrate our recent progress in understanding Golgi biology, and also outline a specific set of yet unanswered Golgi-related questions. For example, exactly how do cargo and resident proteins travel to, through and out of the Golgi? What are the exact modes, carriers and molecular machineries of bi-directional Golgi trafficking? How are Golgi structure and its various functions modified during normal (differentiation, development, etc.) and abnormal (diseases, drugs, pathogens, etc.) circumstances?

#### AUTHOR CONTRIBUTIONS

KP, VL, YW, and JS prepared and wrote the manuscript.

#### FUNDING

This work was supported by the NIH Grant GM083144 to VL and by the National Institute of Health (Grants GM112786, GM105920, and GM130331), MCubed and the Fastforward Protein Folding Disease Initiative of the University of Michigan to YW.

Proc. Natl. Acad. Sci. U.S.A. 111, 1849–1854. doi: 10.1073/pnas.132 3895111


cells. Proc. Natl. Acad. Sci. U.S.A. 102, 13467–13472. doi: 10.1073/pnas.050 6681102

Zolov, S. N., and Lupashin, V. V. (2005). Cog3p depletion blocks vesiclemediated Golgi retrograde trafficking in HeLa cells. J. Cell Biol. 168, 747–759. doi: 10.1083/jcb.200412003

**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 Prydz, Lupashin, Wang and Saraste. 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.

# Identification of CDC42 Effectors Operating in FGD1-Dependent Trafficking at the Golgi

#### Mikhail Egorov\* and Roman Polishchuk\*

Telethon Institute of Genetics and Medicine, Naples, Italy

Loss of function mutations in the FGD1 gene cause a rare X-linked disease, faciogenital dysplasia (FGDY, also known as Aarskog-Skott syndrome), which is associated with bone and urogenital abnormalities. The FGD1 gene encodes à CDC42-specific guanine nucleotide exchange factor. The mutations are frequently located in the DH module of FGD1 preventing its transformation to the active form. We previously reported that Golgi-associated FGD1 regulates post-Golgi transport of some conventional and bonespecific proteins in a CDC42-dependent manner. However, the downstream targets of FGD1/CDC42 signaling that operate to support transport from the Golgi remain elusive. Here, we demonstrate that Golgi-localized CDC42 effectors might be involved in FGD1 mediated post-Golgi transport, probably through coordination of Golgi membrane and cytoskeleton dynamics. Overexpression of effector-specific CDC42 mutants (exhibiting preferential affinities for PAK1, IQGAP1, N-WASP, or PAR6) only partially rescue membrane trafficking in FGD1-deficient cells, indicating that the orchestrated activities of several downstream targets of CDC42 are required to support FGD1-mediated export from the Golgi. Our findings provide new insights into understanding the molecular mechanisms behind FGD1/CDC42-dependent transport events and uncover new targets whose potential might be explored for correction of membrane trafficking in FGDY.

Keywords: Aarskog-Skott syndrome, FGD1, post-Golgi transport, signaling, GEF

# INTRODUCTION

Faciogenital dysplasia (FGDY) is a rare X-linked disorder that manifests in defects of bone development such as disproportional acromelic short stature, abnormal face shape, as well as cardiac, ocular, urogenital abnormalities and mental retardation (Aarskog, 1970; Orrico et al., 2015). FGDY is caused by loss-of-function mutations in the FGD1 gene that encodes a 961 amino acid FGD1 protein that acts as a specific GEF for the small Rho GTPase CDC42 (Pasteris et al., 1994; Zheng et al., 1996). The predicted frequency of clinical manifestation of FGDY is about 1:25,000 (Orrico et al., 2015). The FGD1 protein is expressed in regions of active osteogenesis in developing long bones (Gorski et al., 2000). In humans, the highest levels of FGD1 expression have been observed in bone tissue, kidney, liver, lung, heart and brain (Pasteris et al., 1994; Olson, 1996). Thus, the pattern of postnatal FGD1 expression strongly correlates with clinical manifestations of FGDY. Moreover, the FGD1/CDC42 signaling machinery has an important role in osteogenetic differentiation in hMSC and may persist throughout adult life (Gao et al., 2011). Similarly to other

#### Edited by:

Roberto Weigert, National Institute of Dental and Craniofacial Research (NIDCR), United States

#### Reviewed by:

Vladimir Lupashin, University of Arkansas for Medical Sciences, United States Catherine Jackson, UMR7592 Institut Jacques Monod (IJM), France

#### \*Correspondence:

Mikhail Egorov mihvegorov@gmail.com Roman Polishchuk polish@tigem.it

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 12 July 2018 Accepted: 16 January 2019 Published: 04 February 2019

#### Citation:

Egorov M and Polishchuk R (2019) Identification of CDC42 Effectors Operating in FGD1-Dependent Trafficking at the Golgi. Front. Cell Dev. Biol. 7:7. doi: 10.3389/fcell.2019.00007

genetic diseases, there is no specific treatment for patients with FGDY, surgical intervention being the only option to correct some abnormalities and increase the quality of life.

Although FGD1 was detected at the Golgi (Estrada et al., 2001; Egorov et al., 2009), the mechanism by which the FGD1/CDC42 machinery regulates the transport of cargo at the Golgi apparatus remains unclear. We reported previously that FGD1 silencing strongly suppressed CDC42 activity at the membranes of trans-Golgi network (TGN). As a result, formation of post-Golgi transport intermediates was compromised, apparently due to an impairment of TGN membrane extension along microtubules (Egorov et al., 2009). In particular, this interaction is important for docking of nascent carriers to the microtubules (MTs) that trigger trafficking events. The failure of FGD1 to coordinate membrane and cytoskeletal dynamics at the Golgi could explain the membrane transport delay induced by loss-of-function FGD1 mutations and might be considered as one of the key mechanisms of FGDY pathogenesis.

Given the importance of the FGD1/CDC42 machinery in regulating post-Golgi membrane transport, we investigated whether and to what extent the downstream targets of Golgilocalized CDC42 mediate FGD1-dependent signal transduction at the Golgi (McCallum et al., 1998; Matas et al., 2004; Cau and Hall, 2005; Valente et al., 2012; Matsui et al., 2015). Our findings indicate that suppression of either PAK1 or IQGAP1 genes resembles the inhibitory effects of FGD1 silencing on post-Golgi transport, while their activation only partially overrides the TGN transport block induced by FGD1 deficiency. Expression of effector-specific CDC42 mutants revealed that other downstream components of the FGD1/CDC42 signaling pathway including N-WASP and PAR6 might be involved in FGD1-mediated trafficking events. Further characterization of the proposed signaling pathways may help to uncover the key druggable molecules with therapeutic potential for FGDY.

#### RESULTS

To determine whether Golgi-associated CDC42 targets such as IQGAP1, PAK1 and N-WASP operate in FGD1/CDC42 dependent trafficking events, we assessed the ability of these proteins to influence post-Golgi transport by monitoring constitutive export. We reasoned that CDC42 effectors whose suppression would cause the aberrant FGDY-like "secretory" phenotype (Egorov et al., 2009) are likely to be involved in FGD1/CDC42-medited regulation of trafficking events at the Golgi. To address this issue, we silenced genes of interest in HeLa cells with specific siRNAs and analyzed the impact of gene suppression on post-Golgi transport of VSVG, as previously described (Polishchuk et al., 2003). VSVG was synchronized within the TGN using a 20◦C block. We found that VSVG strongly co-localized with the trans-Golgi marker TGN46 in both control and IQGAP1-, PAK1-, or N-WASP-depleted cells at the end of the 20◦C temperature block (**Figure 1A**, upper row; see **Figure 1C** for silencing efficiency), indicating that suppression of these CDC42 effectors did not impact on the delivery of cargo proteins to the TGN. Release from the 20◦C block triggered the formation of numerous VSVG-containing carriers from the Golgi apparatus in control cells. To block the fusion of carriers with the plasma membrane (which facilitates quantification), 0.5% tannic acid treatment was employed (Egorov et al., 2009). Unlike control cells, where numerous post-Golgi carriers formed within 1 h after the release of the 20◦C block, IQGAP1- or PAK1-depleted cells exhibited a significant reduction in the number of TGNderived VSVG carriers (**Figure 1A**, bottom row). Morphometric analysis confirmed that suppression of either IQGAP1 or PAK1 resulted in a significant decrease of VSVG-containing post-Golgi carriers (**Figure 1B**). In contrast, inhibition of N-WASP did not significantly affect the post-Golgi transport of VSVG (**Figures 1A,B**) indicating that N-WASP may operate mainly in a different transport route from the Golgi (Matas et al., 2004; Egea et al., 2013).

The inhibitory impact of IQGAP1 and PAK1 depletion on export from the TGN was confirmed by electron microscopy analysis. The general morphology of the TGN in HeLa cells lacking either PAK1 or IQGAP1 proteins were assessed using TGN38-HRP transfection (Egorov et al., 2009). The inspection of thin sections revealed a TGN38-HRP-associated DAB precipitate, which decorated the lumen of both tubular and round membrane profiles in the TGN area of control HeLa cells (**Figure 1D**). The silenced cells exhibited much longer and larger TGN profiles with 2–3 labeled cisternae in some Golgi stacks (**Figures 1D,E**). This phenotype indicates significantly reduced consumption of TGN membranes in both IQGAP1- and PAK1-deficient cells due to a delay of protein export at the level of the TGN. In summary, our observations suggest that suppression of the CDC42 downstream targets IQGAP1 and PAK1 inhibits export from the Golgi in a manner similar to what is observed in FGD1-deficient cells (Egorov et al., 2009).

To further test the impact of IQGAP1/PAK1 proteins on the FGD1-dependent mechanism of post-Golgi transport, we investigated their ability to rescue VSVG trafficking in FGD1 deficient cells. To this end, FGD1-silenced HeLa cells were co-transfected with VSVG-EGFP and plasmids encoding either PAK1-wt or PAK1-Thr423 (a constitutively active mutant) (Jaffer and Chernoff, 2002), and VSVG transport was analyzed upon a release from the 20◦C block as described above. The loss of FGD1 function led to a delay of VSVG-GFP transport, while expression of PAK1-wt and PAK1-Thr423 resulted in a moderate increase in the number of post-Golgi VSVG-positive transport carriers in FGD1-deficient cells (**Figures 2A,B**). It appears that overexpression of PAK1 rescues post-Golgi trafficking to some extent but cannot completely overcome the transport block at the TGN induced by FGD1 knockdown.

We also evaluated the ability of IQGAP1 to facilitate protein export from the TGN in FGD1-silenced cells using the same experimental strategy. Expression of IQGAP1-WT induced a modest but significant recovery of VSVG export from the Golgi in FGD1-deficient cells (**Figures 2A,B**). To show that the IQGAP1-mediated rescue of trafficking in FGD1-deficient cells indeed requires CDC42, we used the IQGAP1 mutant (IQGAP1- T1050AX2), which fails to interact with its upstream activator CDC42/RAC1 (Fukata et al., 2002). We observed that expression of this particular IQGAP1 mutant was unable to rescue VSVG

FIGURE 1 | Down-stream targets of CDC42 regulate post-Golgi transport in HeLa cells. Cells were treated with either control non-targeting duplex or specific siRNAs of IQGAP1, PAK1, and N-WASP for 72 h. Cells were then processed for evaluation of post-Golgi trafficking (A,B), efficiency of silencing (C), or TGN morphology (D,E). (A) After the 20◦C block, VSVG accumulated at the Golgi level where it overlapped with the TGN46 marker (top row). Release of the secretory material from the Golgi (32◦C 60 min; see bottom row) in the presence of 0.5% tannic acid resulted in the accumulation of VSVG carriers (arrows) in control and N-WASP-depleted cells, while most VSVG still remained in the Golgi (empty arrows) in IQGAP1- and PAK1-silenced cells. Scale bar, 15 µm. (B) Quantification of VSVG-containing carriers after 60 min of release from the 20◦C block (n = 30 cells; ∗∗∗p < 0,001, t-test). (C) Western blotting showed the efficient reduction in the expression of corresponding proteins after treatment with siRNAs for N-WASP, IQGAP1, and PAK1. The level of actin was used as a loading marker. (D) Control, IQGAP1 or PAK1-silenced cells were transfected with TGN38-HRP, fixed and subsequently processed for electron microscopy analysis (see methods). Control HeLa cells showed the TGN-38-HRP-associated DAB precipitate in several tubular and rounded profiles at the trans side of the Golgi stack. Unlike control cells, the silencing of either the IQGAP1 gene or the PAK1 gene resulted in a much larger TGN, comprised of numerous additional labeled cisternae and tubular profiles. Scale bar: control and siRNA PAK1 200 nm, siRNA IQGAP1 500 nm. (E) Quantification of the increase in the area of the TGN compartment in IQGAP1- and PAK1-silenced cells (n = 20 stacks; ∗∗∗p < 0.001, t-test).

FIGURE 2 | CDC42 effectors partially rescue post-Golgi transport of VSVG in FGD1-deficient HeLa cells. HeLa cells were treated with targeted FGD1 siRNA and co-transfected with VSVG-GFP and a plasmid encoding CDC42 effector (indicated at the corresponding panel). Cells were subsequently exposed to the 20◦C block for 2 h and then shifted to 32◦C for 60 min in the presence of tannic acid. (A) Expression of either the PAK1 or the IQGAP1 mutant partially rescued the VSVG-GFP transport in FGDY cells; (B) Quantification of post-Golgi VSVG-GFP- containing carriers per cell in the experiments shown in panel A (n = 30 cells; <sup>∗</sup>p < 0.005, t-test). (C) Co-expression of VSVG-GFP and different constitutively active CDC42 mutants (see Table 1) partially rescued transport of VSVG-GFP in FGD1-silenced cells. (D) Quantification of post-Golgi VSVG-GFP-containing carriers per cell in the experiments shown in panel C (n = 30 cells; ∗∗∗p < 0.001, <sup>∗</sup>p < 0.005, t-test). Scale bars, 20 µm.

transport in FGD1-silenced cells even after 60 min of release from the TGN (**Figures 2A,B**). These findings demonstrate that IQGAP1 might potentially operate as a FGD1/CDC42 signal conductor to regulate transport from the Golgi. However, none of these downstream CDC42 effectors were sufficient alone to completely restore trafficking in FGD1-deficient cells.

To further characterize the FGD1 signaling pathway that is mediated by CDC42 effectors, we used constitutively active CDC42 mutants, which are defective in binding to different downstream targets of CDC42 (Li et al., 1999; Guo et al., 2010; See **Table 1**). We first tested whether expression of the CDC42/A37 mutant, which is defective in interacting with PAK1, could overcome the transport block induced by FGD1 lossof-function using the traffic pulse protocol (**Figures 2C,D**). We noted a significant (but not full) recovery of VSVG-EGFP post-Golgi transport in FGD1-deficient cells expressing this CDC42 mutant. Thus, PAK1 does not appear to operate as an indispensible effector in post-Golgi trafficking events that are regulated through the FGD1/CDC42 axis. This observation is in line with a previous report showing a limited ability of FGD1 to regulate the activity of PAK1 (Nagata et al., 1998).

Another mutant, CDC42/H63, which is defective in binding to IQGAP1, had a similar impact. Overexpression of this mutant in FGD1-deficient cells showed a moderate improvement in VSVG-EGFP transport (**Figures 2C,D**), indicating that CDC42 can sustain post-Golgi trafficking through other effectors. Indeed, this mutant binds PAR6, N-WASP and to some extent PAK1 (**Table 1**).

Next we tested whether expression of the CDC42 mutants, which do not interact with N-WASP/PAR6 (K40) or with PAR6 (AA173-174), could restore transport in FGD1-deficient cells (**Figures 2C,D**). We observed no statistically significant recovery of VSVG-GFP post-Golgi transport in FGD1-silenced cells after K40 expression and only a slight increase (10.3%) in trafficking in the case of CDC42-AA173-174 expression. This suggests that FGD1/CDC42-dependent post-Golgi trafficking might require N-WASP and PAR6, whose roles in this process have to be further investigated.

Finally, we tested CDC42 constitutively active mutants that have no binding limitations toward any of the effectors (L-28 and QQ186-187). Overexpression of these mutants resulted in significant recovery of VSVG trafficking in FGD1-depleted cells.


+, binding to the indicated effectors; −, no binding activity; Li et al. (1999) and Guo et al. (2010).

Interestingly, expression of the QQ186-187 mutant, which has a lower binding affinity for PAR6, was less efficient in recovering VSVG transport compared to CDC42/L28. This implies that only a CDC42 form that is fully active toward all downstream effectors is able to completely rescue the Golgi export block caused by FGD1 loss-of-function (**Figures 2C,D**). Thus, the recovery post-Golgi trafficking in FGD1-deficient cells appears to require the synergistic action of several CDC42 effectors at a certain ratio. Alternatively, the limited effect on the transport by some of the mutants might be due to their weak association with the Golgi apparatus. Additional experiments will be necessary to fully characterize the consequences of these interactions.

To further explore possible mechanisms downstream of FGD1/CDC42 signaling at the Golgi, we assessed the levels of CDC42 effectors (PAK1/IQGAP1) under conditions of FGD1 silencing. Surprisingly, Western blot revealed that both PAK1 and IQGAP signals increased in FGD1-deficient cells by 78 and 19.5%, respectively (**Figures 3A–C**). Using an antibody that specifically recognizes the phosphorylated forms of PAK1- 3, we found that Golgi-associated levels of phospho-PAK were elevated as judged by immunofluorescence inspection (**Figures 4A–F**). Taking into account that only PAK1 (but not PAK2/3) resides at the Golgi, we assume that the increase in phospho-PAK in the TGN46 compartment reflects activation of PAK1 at the Golgi in FGD1-deficient cells (**Figures 4G,H**). It appears that PAK1 becomes more active at the Golgi in the absence of functional FGD1. Thus, the up-regulation/activation of CDC42 might explain a weak compensatory response supporting cell homeostasis in conditions of a mutated FGD1 protein. This response, however, does not appear to be sufficient for recovery of normal post-Golgi trafficking rates.

Driven by the hypothesis that FGD1 could function in the signaling machinery that regulates the interaction between nascent TGN membranes and MTs (Egorov et al., 2009), we investigated the involvement of FGD1 and CDC42 effectors in both the nucleation and growth of the Golgi-associated subset of MTs in HeLa cells. To this end, we employed a well-known approach based on the regrowth of MTs in interphase cells after their nocodazole-induced depolymerization (Efimov et al., 2007). Immunofluorescence revealed that control cells quickly recovered the Golgi-associated pool of MTs 2–10 min after nocodazole washout (**Figures 5A,E**). By contrast, there was a significant delay of MT recovery in the case of FGD1 suppression since only a few MT profiles emerged from the Golgi elements in FGD1-silenced cells (**Figures 5B,E**). This effect of FGD1 depletion on Golgi-associated MT growth might contribute to inhibiting the formation of post-Golgi carriers, which exit from the TGN along MT tracks (Polishchuk et al., 2003; Egorov et al., 2009). Similarly to FGD1 knockdown, the silencing of either PAK1 or IQGAP1 by specific siRNAs affected the efficiency of MT regrowth from the Golgi units (**Figures 5C–E**) suggesting that both CDC42 effectors regulate the dynamics of MTs emerging from the Golgi. Importantly, in these experiments we found that the rate of MT growth from the MTOC was not affected, implying the involvement of FGD1, PAK1 and IQGAP in MT dynamics operates only at Golgi stacks.

The function, stabilization and dynamics of MTs are tightly regulated by numerous MT-associated proteins. One of these is the well-characterized MT non-motor plus-end-binding protein CLASP (cytoplasmic linker associated protein (1) that has been shown to drive nucleation of non-centrosomal MTs originating from the TGN (Efimov et al., 2007). We reasoned that the dynamics of CLASP1 at the Golgi might be affected by FGD1 loss-of-function. Confocal microscopy showed a very clear but faint CLASP1 signal at the Golgi membranes in control cells (**Figures 6A–C**). By contrast, FGD1-silenced cells exhibited a strong increase in CLASP1 levels at the Golgi (**Figures 6D–F**) indicating that molecular coordination of Golgi membranes and MT dynamics are likely to be impaired by the loss of FGD1.

# DISCUSSION

Our findings indicate that the impact of FGD1 on post-Golgi trafficking is indeed mediated by CDC42 effectors. However, none of the CDC42 effectors appears to be absolutely indispensible: rather, it is the orchestrated activity of several such effectors that is required to support export of cargo proteins from the TGN. We have shown that direct Golgi-associated downstream targets of CDC42, such IQGAP1 and PAK1, are involved in regulating constitutive post-Golgi membrane transport. PAK1 has been already reported to operate in the fission of post-Golgi carriers from the TGN (Valente et al., 2012), while the involvement of IQGAP1 in post-Golgi trafficking has not yet been observed despite a well-documented association of IQGAP with the Golgi (McCallum et al., 1998). Our observations suggest that FGD1, IQGAP1 and PAK1 are required (i) to support formation of post-Golgi carriers and (ii) to allow nucleation of Golgi-associated MTs, which are used by post-Golgi carriers as highways to reach their target destinations (Polishchuk et al., 2003; Egorov et al., 2009). However, both IQGAP and PAK1 exhibited a limited ability to restore the impaired TGN export in FGD1-deficient cells when expressed alone. Indeed, the impact of their overexpression on post-Golgi trafficking in the absence of FGD1 appears to be incomplete, probably due to the lack of activity of other effectors coordinated by the FGD1/CDC42 machinery or by their weaker association with TGN membranes when CDC42 is missing. On the other hand, we found that the levels of both IQGAP1 and PAK1 were increased in cells lacking FGD1.

This indicates that cells might try to compensate for the loss of FGD1 through the up-regulation of the CDC42 effectors PAK1 and IQGAP1. These compensatory mechanisms may target PAK1 and IQGAP1 through FGD1 independent signaling pathways in an attempt to circumvent the FGD1 deficiency. In this regard, FGD1 has been shown to play a limited role in PAK1 activation (Nagata et al., 1998) indicating that additional mechanisms might be involved to stimulate PAK1 activity at the Golgi. However, such activation does not appear to be sufficient to recover post-Golgi transport, probably because other components of the FGD1/CDC42-controlled molecular network remain inactive or cannot be recruited to the Golgi.

The complexity of FGD1-regulated post-Golgi transport might imply the coexistence of several CDC42-dependent signaling pathways, some of which may include N-WASP as an effector. This hypothesis is supported by our findings showing that CDC42 mutants with impaired N-WASP binding have a limited capacity to support VSVG export from the Golgi in FGD1-deficient cells. N-WASP regulates actin dynamics and the formation of transport carriers at the Golgi (Egea et al., 2013). We cannot rule out that actin-based mechanisms of post-Golgi carrier formation are also controlled by the FGD1/CDC42 machinery. WASP, as a CDC42 target, might be a part of this mechanism as well, but our RNAi results suggest that WASP is probably indispensable for post-Golgi trafficking.

Our findings also suggest that one of the plausible mechanisms by which FGD1 regulates post-Golgi transport requires the stabilization of a Golgi-emanating subset of MTs, which serve as tracks for the formation of post-Golgi carriers and for their further translocation to the target membrane (Polishchuk et al., 2003; Egorov et al., 2009). Indeed, nucleation of MTs at the Golgi membranes was strongly inhibited by silencing

of FGD1 or by suppression of the CDC42 effectors IQGAP and PAK1. MT assembly at Golgi membranes is driven by MT-associated proteins such as CLASPs (Efimov et al., 2007; Miller et al., 2009). It has been shown recently that the dynamics of CLASP2 association with the Golgi is regulated by a novel protein complex, PAR6/PAR3/aPKC, that has CDC42 and PAR6

as targets (Matsui et al., 2015). This protein complex controls CLASP2 phosphorylation at sites responsible for its interaction with GCC185 at the TGN (Matsui et al., 2015). Disruption of this PAR6 complex results in aberrant CLASP accumulation at the Golgi (Matsui et al., 2015). Notably, FGD1 knockdown causes a similar increase in CLASP association with Golgi membranes (see **Figure 6**). Thus, it is tempting to speculate that the missing link between the FGD1/CDC42 machinery and MTs could be constituted by the PAR/PKC-CLASP complex (**Figure 7**). In this scenario, the lack of FGD1 activity might impact on the

bar, 28 µm.

fcell-07-00007 January 31, 2019 Time: 18:45 # 9

PAR6 complex and, hence, further affect the dynamics of CLASP. This, in turn, would cause either MT instability or a low rate of MT regrowth, thus affecting the formation and translocation of Golgi-derived transport carriers and post-Golgi trafficking in general. Indeed, abnormal CLASP accumulation and impaired nucleation of MTs at the Golgi in FGD1-deficient cells supports this hypothesis. We also found that loss of interaction with PAR6 reduces the ability of constitutively active CDC42 mutants to rescue transport in FGD1-deficient cells. This suggests that PAR6 is involved in the FGD1/CDC42 pathway that regulates MT dynamics at TGN membranes.

Further studies of the molecular players involved in FGD1/CDC42-dependent regulation of post-Golgi trafficking represents a challenging task in finding a cure for FGDY. Although we found that activation of some CDC42 effectors may partially compensate for FGD1 deficiency, none of these effectors alone circumvent the consequences of FGD1 loss. Therefore, drugs or treatments that stimulate multiple FGD1/CDC42 targets have to be considered for eventual therapeutic approaches. Furthermore, considering the abundance of FGD1 in bones and its importance for the transport of bone proteins (Egorov et al., 2009), pharmacological activation of FGD1/CDC42 effectors deserves serious evaluation as a potential strategy to facilitate bone regeneration after damage or surgery. In this context, significant efforts will be required to identify key druggable nodes in the FGD1/CDC42 signaling pathway(s) operating in the bone-specific post-Golgi secretory pathway(s).

# MATERIALS AND METHODS

#### Antibodies and Reagents

The following antibodies were used: anti-TGN46 1:1000 (IF) (AbD Serotec, United Kingdom); anti-Giantin 1:500 (IF) (Abcam, United Kingdom); anti-myc 1:250 (IF); anti-flag

1:250 (IF); anti-HA 1:500 (IF) and P5D4 Cy3-conjugated anti-VSVG 1:500 (IF) from Sigma-Aldrich (Italy); anti-N-WASP 1:1000(WB); anti-IQGAP1 1:1000(WB); phosphoPAK1- 3 1:500 (IF), 1:1000 (WB); and PAK1 (N-20) 1:500 (IF), 1:1000 (WB) were purchased from Santa Cruz Biotechnology (United States); anti-FGD1 1:1000 (WB) made by G. Di Tullio (CMNS, Italy); GAPDH 1:1000 (WB), CLASP1 1:300 (IF) provided by I. Kaverina (Vanderbilt University, United States).

The following plasmids were used: IQGAP1-T1050AX2 flag (provided by K. Kaibuchi, Japan); IQGAP1-WT-myc (Columbia University); PAK1-WT-myc, PAK1-Thr423-myc (provided by M. Gimona); CDC42 mutants were tagged with HA (a gift form Y. Zheng, United States); VSVG-EGFP (provided by J. Lippincott-Schwarz, NIH/NICHD, United States).

# Cell Culture, Transfections, and Infection With VSV

HeLa cells (ATCC, United States) were cultured in DMEM (Invitrogen, Italy) supplemented with 10% FBS. Both non-targeted (control) and target-specific siRNAs were obtained from Sigma Aldrich (Italy) and transfected by Oligofectamine (Invitrogen, Italy). The efficiency of silencing was assessed for each experiment by Western blotting. Plasmid transfection was performed using Lipofectamine 2000 (Invitrogen, Italy). The infection of HeLa with VSV was performed as previously described (Polishchuk et al., 2003). The nododazole-washout assay was performed according to a published protocol (Efimov et al., 2007).

### Western Blotting

RNAi-treated HeLa cells were lysed in 50 mM Tris–HCl, pH 8.8, containing 0.2% SDS and protease inhibitor cocktail (Roche, France). Sixty micrograms of cell lysates were run on 7% SDS– PAGE and subsequently transferred to nitrocellulose. Blots were probed with specific antibodies of interest using the ECL-based detection method (Amersham Pharmacia Biotech, Piscataway, NJ, United States).

#### Immunofluorescence and Confocal Microscopy

For immunofluorescence analysis, HeLa cells were fixed with 4% paraformaldehyde in phosphate-buffered saline at room temperature for 30 min, permeabilized with 0.2% saponin for 30 min, and then blocked with 2% bovine serum albumin (BSA) for 30 min. The cells were labeled with the primary antibodies and secondary antibodies of interest conjugated to Alexa Fluor 488 and 568. Cells were mounted with mowiol and analyzed on LSM 510 META or LSM710 confocal microscopes (Carl Zeiss, Germany) with the 63 × Apo NA 1.4 objective. Numbers of VSVG carriers and numbers of MTs emerging from the Golgi were quantified in confocal images using ImageJ software (NIH, United States) as described (Egorov et al., 2009). Colocalization of p-PAK or CLASP1 with TGN46 was evaluated using either ImageJ (NIH, United States) or Zen Black (Zeiss, Germany) software.

#### Electron Microscopy

For electron microscopy analysis, cells of interest expressing TGN38-HRP were directly fixed with 2% glutaraldehyde

(pH 7.4) and incubated with buffered 3,3<sup>0</sup> -diaminodenzidine DAB, according to Egorov et al. (2009). After dehydration the cells were embedded in Epon812. Embedded cells were sectioned with a 45◦ degree diamond knife (Diatome, Switzerland) using a Leica ultramicrotome (Leica, Austria). The thin 70 nm sections were imaged with a Tecnai-12 electron microscope (FEI, The Netherlands) equipped with an Ultra View CCD digital camera (Soft Imaging System, Munich, Germany).

#### Statistical Analysis

All statistical data are presented as a mean value ± standard deviation. The unpaired Student t-test was used to calculate differences between two sets of data. Statistical significance between different sets of data are indicated in the text and figures as follows: <sup>∗</sup>p < 0.05; ∗∗p < 0.01; and ∗∗∗p < 0.001. All results are representative of three independent experiments.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

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

#### FUNDING

This work was supported by Telethon, Italy (Grant #TGM16YCBDM09 to RP) and by AIRC Italy (Grant #IG17118).

#### ACKNOWLEDGMENTS

The authors thank Cathal Wilson for critical reading of the manuscript.

phosphorilation to generate cell polarity. Mol. Biol. Cell. 26, 751–761. doi: 10.1091/mbc.E14-09-1382


**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 Egorov and Polishchuk. 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.

# Golgi Complex Dynamics and Its Implication in Prevalent Neurological Disorders

#### Mario O. Caracci, Luz M. Fuentealba and María-Paz Marzolo\*

Departamento de Biología Celular y Molecular, Facultad de Ciencias Biológicas, Pontificia Universidad Católica de Chile, Santiago, Chile

Coupling of protein synthesis with protein delivery to distinct subcellular domains is essential for maintaining cellular homeostasis, and defects thereof have consistently been shown to be associated with several diseases. This function is particularly challenging for neurons given their polarized nature and differential protein requirements in synaptic boutons, dendrites, axons, and soma. Long-range trafficking is greatly enhanced in neurons by discrete mini-organelles resembling the Golgi complex (GC) referred to as Golgi outposts (GOPs) which play an essential role in the development of dendritic arborization. In this context, the morphology of the GC is highly plastic, and the polarized distribution of this organelle is necessary for neuronal migration and polarized growth. Furthermore, synaptic components are readily trafficked and modified at GOP suggesting a function for this organelle in synaptic plasticity. However, little is known about GOPs properties and biogenesis and the role of GOP dysregulation in pathology. In this review, we discuss current literature supporting a role for GC dynamics in prevalent neurological disorders such as Alzheimer's disease, Parkinson's disease, Huntington's disease, and epilepsy, and examine the association of these disorders with the wide-ranging effects of GC function on common cellular pathways regulating neuronal excitability, polarity, migration, and organellar stress. First, we discuss the role of Golgins and Golgi-associated proteins in the regulation of GC morphology and dynamics. Then, we consider abnormal GC arrangements observed in neurological disorders and associations with common neuronal defects therein. Finally, we consider the cell signaling pathways involved in the modulation of GC dynamics and argue for a master regulatory role for Reelin signaling, a well-known regulator of neuronal polarity and migration. Determining the cellular pathways involved in shaping the Golgi network will have a direct and profound impact on our current understanding of neurodevelopment and neuropathology and aid the development of novel therapeutic strategies for improved patient care and prognosis.

Keywords: Golgins, GOPs, CLASP2, neurodegeneration, epilepsy, LRRK2, Reelin, synaptic activity

# INTRODUCTION

The Golgi complex (GC) is composed of distinct compartments or cisternae: the cis-, medialand trans-Golgi and the trans-Golgi network (TGN) (Chen et al., 2017). The cis-Golgi, located immediately after the endoplasmic reticulum (ER), is where proteins are sorted based on carboxyl terminal signal peptides to either continue moving through the Golgi cisternae or return to the

#### Edited by:

Vladimir Lupashin, University of Arkansas for Medical Sciences, United States

#### Reviewed by:

Leslie K. Climer, Baylor University, United States Juan S. Bonifacino, National Institutes of Health (NIH), United States

> \*Correspondence: María-Paz Marzolo mmarzolo@bio.puc.cl

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 18 March 2019 Accepted: 18 April 2019 Published: 07 May 2019

#### Citation:

Caracci MO, Fuentealba LM and Marzolo M-P (2019) Golgi Complex Dynamics and Its Implication in Prevalent Neurological Disorders. Front. Cell Dev. Biol. 7:75. doi: 10.3389/fcell.2019.00075

ER, or, in the case of lysosomal enzymes, be tagged with mannose-6-phosphate to be directed to the lysosomes. The medial-Golgi is enriched in glycosyl-transferase enzymes and conducts several post-translational modifications of recently translated proteins. The trans-most Golgi cisterna branches toward a tubular and reticulating compartment called the TGN, considered the central cargo sorting station of the cell (Guo et al., 2014). Both clathrinand non-clathrin coated vesicles, as well as tubules, are formed at the TGN to deliver lipid and proteins to their final destination, mainly plasma membrane and endosomal compartments (Lowe, 2011; Guo et al., 2014). Golgi cisternae are laterally linked to form the Golgi ribbon structure. The stability of this complex structure and the maintenance of cisternae-specific protein makeup largely depends on resident structural proteins and their interaction with both microtubules and the actin cytoskeleton (Lowe, 2011).

Excellent reviews have addressed the biology of the GC in an array of cell types (Hanus and Ehlers, 2008; Nakano and Luini, 2010; Lowe, 2011; Ramírez and Couve, 2011; Muschalik and Munro, 2018). However, little attention has been given to the role of GC in pathological conditions associated with the central nervous system (CNS). Increased GC fragmentation has been reported in several neurological disorders, including Alzheimer's disease (AD) (Joshi et al., 2015), amyotrophic lateral sclerosis (ALS) (Sundaramoorthy et al., 2015), Creutzfeldt-Jakob disease (Sakurai et al., 2000), and Parkinson's disease (PD) (Gosavi et al., 2002), among others. Compellingly, this fragmentation is often observed at early stages of the disease thus making it unlikely that GC fragmentation is a result of apoptotic cell death (Gosavi et al., 2002; Liazoghli et al., 2005; van Dis et al., 2014).

In the present review, we summarize relevant information regarding the regulation of GC dynamics in neurons and associate it with common cellular and molecular defects identified in neurodevelopmental and neurodegenerative disorders. First, we describe the distinct role and organization of the GC in neurons, where it participates of relevant developmental processes such as neuronal polarity and migration. Afterward, we discuss the role of Golgins in CNS development and disease. We then address the convergence of well-known neurological disorders with abnormal GC architecture observed therein. Finally, we link all cellular pathways discussed in the previous sections with known cellular outcomes of Reelin signaling activation, a cell signaling pathway consistently associated with CNS development that includes neuronal polarity and Golgi/cytoskeletal dynamics (Santana and Marzolo, 2017).

#### ARCHITECTURE AND FUNCTION OF THE GC IN NEURONS

The role of the GC in polarized cells such as neurons is particularly challenging given the differential protein composition of each neuronal compartment and the long distances that newly synthesized and modified proteins need to traverse to get to their subcellular destinations (Bentley and Banker, 2016). The polarized distribution of proteins in the dendritic and axonal compartments is essential to maintaining neuronal homeostasis, which is highly relevant for neuronal chemical and electrical communication throughout the CNS (Bentley and Banker, 2016). Furthermore, during CNS development, neuronal polarization is also essential, as neuronal migration and positioning within the developing brain ultimately impact neuronal connectivity (Bentley and Banker, 2016). One of the first relevant processes in neuronal polarization is evidenced by the change in localization and polarity of the GC, in a highly regulated process that is essential for axonal specification and dendritic growth. The Serine/Threonine Kinase 25 (STK25) is a crucial mediator of GC polarity; STK25 downregulation inhibits axonal specification in hippocampal neurons, and conversely, overexpression of STK25 leads to the development of multiple axons (Santana and Marzolo, 2017). Golgi morphology is also significantly disrupted in STK25 silenced neurons, consistent with loss of dendritic polarity of the GC (Matsuki et al., 2010). STK25 expression is also a determinant for establishing neuronal polarity and appropriate migration in developing mouse neocortex and hippocampus (Matsuki et al., 2013; Rao et al., 2018). Additionally, the knockdown of STK25-interacting partners, the pseudokinase STE20-Related Kinase Adaptor (STRAD), Liver kinase B1 (LKB1) and Golgi matrix protein 130 kDa (GM130) reproduces such polarity and migration deficits (Matsuki et al., 2010, 2013; Orlova et al., 2010), suggesting a pivotal role for the STK25-STRAD-GM130-LKB1 complex, in GC polarity and neuronal development.

As Golgi positioning determines neuronal polarity, dendritic, and axonal development show differential reliance on secretory trafficking (Ye et al., 2007). For instance, limiting ER-to-Golgi trafficking decreases membrane supply toward the dendrites without affecting the axon (Ye et al., 2007). More importantly, secretory membrane trafficking and dendritic development are highly dependent on discrete units of ER and GC located within the dendritic compartment which fulfill essential functions in protein synthesis, maturation and sorting to target membranes (Hanus and Ehlers, 2008).

Endoplasmic reticulum-related structures referred to as ER exit sites (ERES) and ER-Golgi intermediate compartments (ERGICs) are widely distributed throughout the soma and dendritic compartment, in stark contrast to discrete Golgi units referred to as Golgi outposts (GOPs) which are mostly found in proximal dendrites of cultured hippocampal neurons and in apical dendrites of pyramidal neurons in vivo (Ramírez and Couve, 2011). Several protein markers forcis-, medial-, and trans-Golgi have been identified in GOPs. However, their abundance and colocalization with one another are highly dynamic, suggesting that GOPs form by discrete Golgi compartments which only transiently interact with one another (Zhou et al., 2014), distinct from the classical GC ribbon morphology observed in non-polarized cells or neuronal soma.

Dendrites containing GOPs exhibit higher complexity compared to the ones lacking these structures. Furthermore, dendritic polarization of the somatic Golgi precedes the increase of dendritic complexity (Horton et al., 2005). The frequent location of GOPs at dendritic branch points, in conjunction with time-lapse analysis, showed that dendritic branches dynamically extend as GOPs move toward a developing

branch (Ye et al., 2007). Finally, disruption of Golgi polarity by overexpression of Golgi reassembly-stacking protein of 65 kDa (GRASP65), a binding partner of GM130, leads to loss of dendritic arbor polarity and uniform growth of dendritic branches (Horton et al., 2005).

Golgi outposts are not only important for dendritic development; in fact, proteins associated with synaptic function also traffic through these distal GC structures. For instance, the kainate receptor GluK2 readily colocalizes to GOPs 30 min after synchronous release from the ER, and activation of protein kinase C (PKC) reduces GluK2 ER exit (Evans et al., 2017). In contrast, PKC-dependent phosphorylation of synaptic scaffold synapse-associated protein 97 (SAP97) protein is necessary for interacting with ADAM10 and facilitating its exit from GOPs into the synaptic compartment (Saraceno et al., 2014). Trafficking of N-methyl-D-aspartate (NMDA) receptors also bypasses somatic GC and localizes to GOPs where it colocalizes with SAP97 (Jeyifous et al., 2009). Finally, specialized GOPs enriched in glycosylation machinery and positive for the TGN resident protein calneuron-2 (Mikhaylova et al., 2009) referred to as Golgi satellites (GS), have been described to be widely distributed in the dendritic arbor and only occasionally associated with GC cisternae markers (Mikhaylova et al., 2016). Interestingly, the motility of GS is substantially reduced upon treatment with the GABA receptor antagonist bicuculine and with BDNF treatment, suggesting a regulatory role for these GS in activity-dependent neuronal plasticity (Mikhaylova et al., 2016). Indeed, GS colocalized with synaptic transmembrane proteins including synaptic cell adhesion proteins Neuroligin-1 and Neuroplastin-55, synaptic scaffold protein Homer1 and ionotropic receptor subunit GluA1 (Mikhaylova et al., 2016). The role of GOPs in synaptic plasticity has been controversial as they are mostly found in the largest dendrite of pyramidal neurons; however, the description of specialized and widely distributed GS which readily colocalize with synaptic components provides a fresh impulse to this idea and warrants further investigation.

Despite the pivotal role of GOPs in neuronal function and development, we know little about their biogenesis. Livecell imaging experiments have shown that GOP biogenesis at apical dendrites requires GC deployment from the soma and subsequent tubule fission which involves the activation of GClocalized RhoA GTPase and downstream effectors, LIM motifcontaining protein kinase 1 (LIMK1) and cofilin (Quassollo et al., 2015). These data suggest that GOPs generate from an existing GC cisterna rather than synthesized de novo at the dendritic compartment and depends on the regulation of the actin cytoskeleton. More recently GOPs biogenesis was determined to be highly dependent on the secretory pathway, as silencing of Sec23 and Sec31, components of the COPII complex mediating anterograde transport from the ER to the cis-Golgi (Fromme et al., 2008), decreases the number of GOPs in Drosophila Dendritic arbor (Da) neurons (Chung et al., 2017). Moreover, expression of these subunits is enhanced by CrebA expression (CREB3L in mammals), a master transcriptional regulator of secretory trafficking-associated genes (Fox et al., 2010). Accordingly, CrebA overexpression in Drosophila Da neurons increased GOP abundance in dendrites (Chung et al., 2017).

To sum up, the highly polarized nature of neurons is highly dependent on GC positioning. Furthermore, long-range protein synthesis and transport are mediated by discrete units of GC such as GOPs and GS. Over the next sections, we will discuss how the disruption of physiological GC function directly impacts in neurological disorders.

### STRUCTURAL PROTEINS OF THE GC AND THEIR ROLE IN NEURODEVELOPMENT AND DISEASE

Golgins are a set of proteins characterized by the presence of coiled-coil domains that play a substantial role in maintaining GC morphology (Muschalik and Munro, 2018). Golgins associate with several proteins including small GTPases of the Rab and Arl families, that control their tethering function and membrane recruitment (Cheung and Pfeffer, 2016). Loss-offunction approaches targeting several Golgins have proven their role in maintaining GC architecture in cell culture systems, but little is known of their role in the development and function of the nervous system or neuropathology.

Animal knockout (KO) models for Golgins such as Giantin or GMAP 210 have shown them to be essential for healthy development, as these animals exhibit defects in craniofacial and skeletal development (Smits et al., 2010; Stevenson et al., 2017). On the other hand, KO of Golgin-84, a protein closely related to Giantin, does not show any significant developmental abnormalities, and compound mutants for both of these Golgins do not show any additional defects (McGee et al., 2017). While these models highlight the importance of Golgins in embryonic development, some Golgins might be dispensable; indeed, several Golgins are associated with human diseases (Toh and Gleeson, 2016), but only a few have been linked to obvious neurological defects.

In this regard, neuronal GM130 KO mice showed severe motor defects similar to ataxia (Liu et al., 2017). GM130 is one of the most studied Golgins and is known to maintain Golgi ribbon morphology (Lowe et al., 1998). Furthermore, GM130 is necessary for maintaining the interaction of multi-cisternae structures within distal dendrites in Drosophila (Zhou et al., 2014). GM130 KO in mice leads to Purkinje cell degeneration in the cerebellum and impaired secretory trafficking. The latter ultimately affects dendritic growth, as previously described in Drosophila (Zhou et al., 2014; Liu et al., 2017), and these defects were both associated with Golgi fragmentation and abnormal positioning (Liu et al., 2017). On the other hand, GM130 overexpression is observed in in vitro models for mucopolysaccharidosis type IIIB (MPSIIIB) (Roy et al., 2011), a lysosomal storage disorder featuring strong neurological symptoms such as intellectual disability and progressive dementia (Kan et al., 2014). Most notably, overexpression of GM130 alone mimicked MPSIIIB cellular defects observed in HeLa cells (Roy et al., 2011). In conclusion, it appears both gain- and loss-of-function of GM130 drastically impact the morphology and function of the GC and could be a promising target for pharmacological intervention.

Mutations in the Golgin bicaudal D homolog 2 (BICD2) has been described in spinal muscular atrophy related disorders (Neveling et al., 2013). BICD2 participates in vesicular trafficking through direct interaction with dynactin and dynein motor proteins (Splinter et al., 2012) in association with active Rab6 (Cheung and Pfeffer, 2016). Overexpression of BIDC2 missense mutants affecting the coiled-coil domain, in HeLa cells, leads to severe GC fragmentation (Neveling et al., 2013). Most notably, the expression of these mutants in neurons leads to increased motility of BICD2/dynein–dynactin complexes and reduced neurite outgrowth (Huynh and Vale, 2017).

In conclusion, GC structural proteins play important roles in development and disease; however, there remains much to elucidate regarding the direct role of Golgins in inheritable neurological disorders and neurodevelopment (**Figure 1**).

#### DISRUPTION OF GC ARCHITECTURE IN NEUROLOGICAL DISORDERS

#### Alzheimer's Disease

Alzheimer's disease is a neurodegenerative disorder featuring progressive neuronal deterioration and cognitive impairment

trafficking through microtubules; mutations are related to Spinal muscular atrophy disorders. Giantin tethers COPI vesicles to the GC; the knockout animals manifest craniofacial defects. GM130 localizes in cis-Golgi and GOPs, participates in Golgi positioning and is involved in the maintenance of Golgi ribbon morphology and multi-cisternae structures; knockout animals display motor defects, cerebellar Purkinje cell degeneration, and aberrant dendritic growth, while overexpression is linked with a model of the lysosomal storage disorder Mucopolysaccharidosis type IIIB (MPSIIIB).

(Verheijen and Sleegers, 2018). It is widely accepted that AD arises from abnormal proteolytic processing of the amyloid precursor protein (APP), which leads to the generation of Aβ peptides, whose aggregation and deposition enhances neuronal cytotoxicity (Ittner and Götz, 2011). Aside from the amyloid cascade hypothesis, the appearance of neurofibrillary tangles of hyperphosphorylated Tau, a protein commonly associated with the microtubule network, is also a common histopathological observation in AD (Ittner and Götz, 2011).

Interestingly, disruption of GC architecture is observed in neuropathological processes associated with AD. There is growing interest in studying these phenomena in neurodegenerative disorders, as GC fragmentation is thought not to be the result of apoptosis or cell death associated pathways but rather precede them (Nakagomi et al., 2008).

The integrity of GC morphology regulates APP processing and has been extensively reviewed elsewhere (Joshi et al., 2015). Aβ triggers GC fragmentation through activation of cyclindependent kinase-5 (CDK5), which phosphorylates GRASP65. In turn, loss of Golgi integrity increases Aβ production by enhancing amyloidogenic protein cleavage (Joshi et al., 2014). Interestingly, activation of CDK5 by itself increases GC fragmentation through direct phosphorylation of GM130 (Sun et al., 2008). Moreover, overactivation of this kinase is common in AD and other neurodegenerative disorders including PD and ALS (Qu et al., 2007; Shukla et al., 2012; Klinman and Holzbaur, 2015). Overexpression of GRASP65 rescues GC morphology and increases accumulation of APP in the GC compartment, which, strikingly, is accompanied by an increase in non-amyloidogenic processing (Joshi et al., 2014). Indeed, more recently, the TGN has been identified as a major site for α-secretase cleavage and inhibition of APP trafficking from the TGN has been found to increase sAPPα levels (Andersen et al., 2005; Tan and Gleeson, 2019). Thus, GC integrity regulates proteolytic cleavage of APP in physiological and pathological settings (**Figure 2A**).

Golgi Fragmentation is induced in cell culture by overexpression of wild type (WT) and AD-associated mutant Tau (Liazoghli et al., 2005). Besides, the presence of neurofibrillary tangles in AD patients' brains correlates with abnormal GC morphology and decreased surface area in the neocortex and hippocampus (Antón-Fernández et al., 2017). The increased neuronal activity triggered by KCl-mediated neuronal depolarization induces GC fragmentation by a CDK5 dependent mechanism (Mohamed et al., 2017). Concomitantly, KCl-induced depolarization increases secretion of Tau protein into the extracellular medium in WT neurons, which is inhibited by pharmacological blockade of CDK5, indicating that GC maintenance is a crucial regulator of Tau secretion (Mohamed et al., 2017). In this context, knockdown of the small GTPase Rab1A, which mediates ER to GC trafficking and whose deficiency is known to induce Golgi fragmentation (Haas et al., 2007), also increases Tau secretion independently of hyperexcitability (Mohamed et al., 2017). This correlation between neuronal hyperexcitability and Tau deposition is of particular interest since seizures are often comorbid with AD. Furthermore, hyperphosphorylated Tau can be detected in brains of patients with refractory epilepsy and AD patients with

seizures show an accelerated cognitive decline (Tai et al., 2016; Vossel et al., 2016).

Altogether these data highlight the importance of Golgi morphology maintenance and its overall impact on the deposition of both Aβ peptide and Tau in the extracellular medium which, in turn, may enhance the spreading of cytotoxicity throughout the brain (Pignataro and Middei, 2017; Yamada, 2017). Thus, understanding the cellular basis of GC fragmentation could inform the development of drugs to slow down AD progression.

#### Parkinson's Disease

Parkinson's Disease (PD) is the second most common neurodegenerative disorder after AD and is diagnosed based on motor abnormalities such as rigidity, bradykinesia, abnormal posture, and tremors (Błaszczyk, 2016). Determining the genetic basis for PD has been an elusive task. Nevertheless, a few genes have been linked to familial type PD including the encoding of the extensively studied α-synuclein (SCNA) and leucine-rich repeat kinase 2 (LRRK2) (Wood-Kaczmar et al., 2006).

Overexpression of α-synuclein in rat substantia nigra leads to GC fragmentation, a phenotype that is reversed by co-expression of Rab1A, which enhances ER to GC trafficking (Haas et al., 2007) (**Figure 2B**). While Rab1A expression is not able to rescue the neurodegeneration completely, it improves motor function in affected animals (Coune et al., 2011). Most notably, LRRK2 interacts and phosphorylates Rab1A, a phosphorylation event that is significantly induced by LRRK2 G2019A, a PD-associated mutant with increased kinase activity (Jeong et al., 2018). Additionally, overexpression of phospho-dead Rab1A induces neurodegeneration in cultured neurons (Jeong et al., 2018). However, despite these observations, the regulatory role of Rab1A phosphorylation remains unclear. LRRK2 also phosphorylates and directly interacts with Rab7L1, a TGN resident Rab GTPase that has also been identified as a candidate gene for PD (Fujimoto et al., 2018). LRRK2-mediated phosphorylation of Rab7L1 increases TGN fragmentation and PD-associated LRRK2 mutants demonstrate highly increased Rab7L1 phosphorylation rates (Fujimoto et al., 2018).

The retromer complex (VPS35-VPS26-VPS29), which participates in retrograde cargo retrieval from endosomes back to the TGN and recycling cargo back to the plasma membrane, has also been associated with PD (Zhang et al., 2018). Knockdown of VPS26 induces GC fragmentation (Seaman, 2004) and nonsynonymous mutations in this subunit have been reported in PD patients (Gustavsson et al., 2015). This fragmentation is reduced by downregulation of the Golgin GCC88, which finely regulates GC morphology through interaction with the actin cytoskeleton (Makhoul et al., 2019). The retromer complex also appears to have a direct correlation with LRRK2 function, for instance, the brain tissue of patients featuring LRRK2 PD-associated mutations show reduced expression of retromer complex component VPS35 (Zhao et al., 2017). More interestingly, expression of PD-associated VPS35 D620N mutant elevates phosphorylation of Rab GTPases (Namely Rab8A, Rab10, and Rab12) and increases overall kinase activity of LRRK2 to a greater degree than PD pathogenic mutations (Mir et al., 2018).

Overall, these data point to an extensive regulatory network where proper functioning of LRRK2 kinase activity can modulate traffic between the GC and the ER through regulation of Rab GTPases. Furthermore, retrograde trafficking toward the GC in association with the Retromer complex is also affected by LRRK2 activity, thus regulating membrane protein availability and affecting the overall organellar organization.

Overexpression of LRRK2 G2019A is known to reduce dendritic arborization (Lin et al., 2010), which suggests that this protein might play a role in GOP biogenesis or function. Lrrk (Drosophila homolog of LRRK2) colocalizes with medial Golgi marker α-mannosidase II-GFP compartments in both the soma and dendrites of Drosophila Da neurons (Lin et al., 2015). Notably, Lrrk-positive GOPs are mostly stationary while other GOPs exhibit dynamic anterograde and retrograde movement within dendrites (Lin et al., 2015). The direct interaction of Lrrk with Lva, a golgin in Drosophila with no known orthologs (Sisson et al., 2000), prevents the association of GOPs to dynein–dynactin complexes, thus reducing motility. Consistently, decreased expression of Lrrk reduced GOPs anterograde movement and increased dendritic development (Lin et al., 2015). Remarkably, expression of LRRK2 G2019A is associated with an enhancement of retrograde GOPs movement, which is consistent with a suppressed dendritic arborization seen in neurons expressing this variant (Lin et al., 2015). In conclusion, LRRK2 function not only modulates GC structure but also plays an essential role in regulating GOPs positioning in maturing dendrites, which could underlie dendrite degeneration seen in animal models expressing mutant LRRK2.

The dysregulation of LRRK2 has been extensively studied and directly linked to mitochondrial dysfunction, oxidative stress, and synaptic dysfunction among others (Li et al., 2014). On the other hand, over this section, we have enlisted a wide array of protein interactors and phosphorylation targets which point out to an LRRK2 signaling hub residing at the GC that can directly impact on the organelle dynamics.

#### Huntington's Disease

Huntington's disease (HD) is a neurodegenerative disorder characterized by progressive motor and cognitive impairment (Munoz-Sanjuan and Bates, 2011). It is genetically linked to an abnormal CAG repeat expansion in the first exon of HTT, a gene encoding the protein huntingtin (HTT), which participates in several cellular pathways including secretory and endosomal trafficking, transcriptional regulation, autophagy, and ciliogenesis (Saudou and Humbert, 2016). CAG repeats result in a poly-glutamine extension in the translated protein product, which is prone to aggregation and generates cellular toxicity (Labbadia and Morimoto, 2013). HD is one of several so-called poly-Q diseases, each one associated with a different gene locus but sharing similar pathological outcomes (Nath and Lieberman, 2017).

The Golgi adaptor acyl-coenzyme A binding domain protein 3 (ACBD3) is a GC resident protein which interacts directly with giantin and regulates GC morphology (Sohda et al., 2001). ACBD3 is overexpressed in the striatum of HD patients and in

HD model mice where it interacts directly with mutant HTT (**Figure 2C**). The treatment of Q7 and Q111 striatal cell lines with monensin, an ionophore that disrupts acidic organelles including the GC and lysosomes used to model the Golgi stress response, increases ACBD3 levels in both cell lines with a greater induction in Q111, the cell line with higher polyglutamine repeats in HTT (Sbodio et al., 2013).

ACBD3 has also been implicated in glycosphingolipid synthesis in the GC through direct interaction with the transporter of glucosylceramide precursors, human four phosphate adaptor protein 2 (FAPP2) (Liao et al., 2018). Knockdown of ACBD3 leads to mislocalization of FAPP2 from the TGN to the cytoplasm, fragmenting the GC and severely affecting cellular lipidic profile (Liao et al., 2018). Glycosphingolipid metabolism has received considerable attention regarding neurological diseases such as HD, ALS, hereditary spastic paraplegia, lysosomal storage diseases, among others (Desplats et al., 2007; Robert et al., 2009; Boukhris et al., 2013; Dodge et al., 2015).

Metabolic stress related to amino acid deprivation is associated with HD. Low levels of cysteine γ-lyase (CSE), the enzyme directing cysteine biosynthesis has been documented in HD. Cysteine depletion triggers a stress response that enhances ATF4 expression, which is also reduced in HD and ultimately leads to increased oxidative stress and cytotoxicity (Sbodio et al., 2016). ATF4 is a stress-induced transcription factor that directs the expression of adaptative genes necessary to withstand stressors such as hypoxia, amino acid deprivation and organelle stress (Wortel et al., 2017). Q111 cells are highly sensitive to low cysteine medium and show reduced expression of both ATF4 and CSE. However, pre-treatment with low dose monensin rescues these expression levels and increases cell survival rates relative to control Q7 cells, without affecting GC morphology (Sbodio et al., 2018). Neurodegeneration in HD is a result of multiple stress pathways which can be successfully modulated with monensin and are also related to the expression of GC structural proteins, thus suggesting an important role for GC integrity in HD.

More recently, expression of poly-Q-containing Machado-Joseph's protein associated with spinocerebellar ataxia type 3 (Matos et al., 2018) in Da neurons led to reduced dendritic branching and severe reduction of Mannosidase II-EGP positive GOPs in dendrites (Chung et al., 2017). This effect correlates with diminished expression of genes linked to the secretory pathway, including a master transcriptional regulator of secretory trafficking genes, CrebA (Fox et al., 2010; Chung et al., 2017). Indeed, upregulation of CrebA rescues GOP deficiency in poly-Q expressing Da Neurons (Chung et al., 2017). Nuclear localization of poly-Q containing proteins is known to disrupt normal transcriptional programming and perturb nuclear protein turnover (Mohan et al., 2014). However, its potential role in regulating secretory trafficking in neurons through transcriptional repression opens a new door for understanding the cellular pathways involved in poly-Q diseases.

As described above, HD and poly-Q diseases feature a wide array of pathological pathways ranging from metabolic regulation toward basic transcriptional activity issues. Nevertheless, the convergence of these pathways toward GC dynamics regulation offers a novel approach for understanding and ultimately treat these neurodegenerative disorders.

#### Epilepsy

Epilepsy is a condition associated with recurrent, unprovoked seizures which result from a hypersynchronous discharge of neurons in the brain (Stafstrom and Carmant, 2015). The delicate balance between inhibitory and excitatory transmission is often disrupted, and most of the genes associated with epilepsy are known to regulate neuronal connectivity, synaptic discharge, and signaling, as well as ion channel function and trafficking (Meisler and Kearney, 2005; Poduri et al., 2013; Maljevic et al., 2019).

Neuronal hyperexcitability has been shown to regulate the GC, as chronic exposure to slightly elevated concentrations of KCl (5 nM) inducing neuronal depolarization for 2 days leads to increased GC fragmentation (Thayer et al., 2013). Additionally, pharmacological hyperactivation of neuronal activity through treatment with bicuculine or APV, GABAA, and NMDA receptor antagonists induces Golgi fragmentation (Thayer et al., 2013) (**Figure 2D**). Remarkably these changes in GC architecture could be reversed after returning neurons to the normal culture medium (Thayer et al., 2013), showing that GC structure is highly dynamic and remodels itself upon external pharmacological signals. Bicuculine-induced Golgi fragmentation can be prevented by pre-treatment with Ca2+/calmodulin-dependent protein kinase II/IV (CAMKII/IV) inhibitors, suggesting a high dependence of calcium signaling pathways in activity-induced GC fragmentation (Thayer et al., 2013). Several voltage-gated calcium channels associate with epilepsy (Rajakulendran and Hanna, 2016) and calcium signaling mediated by CAMKII has been consistently associated with neurological disease including epilepsy, schizophrenia, and autism spectrum disorders (Robison, 2014). Calcium homeostasis also relies on GC integrity in concert with ER and mitochondrial reservoirs (Lissandron et al., 2010; Yang et al., 2015). However, little is known of the role of GC calcium storage in cellular homeostasis or disease.

Golgi positioning defects could also play a role in the onset of seizure disorders, as the proliferation, migration, and differentiation of neural progenitors are ultimately responsible for shaping neuronal connectivity in the developing brain. Mutations in STRAD, a core protein in the Golgi positioning complex, were identified through microarray analysis in patients with polyhydramnios megalencephaly and symptomatic epilepsy (PMSE) syndrome (Puffenberger et al., 2007). Expression of a PMSE-associated 180-amino acid, C-terminal truncating mutation of STRAD in adult neuronal progenitor cells leads to the disassembly of the GC, reduced expression of GM130 and severe defects in dendritic development (Rao et al., 2018). The reported defects in dendritic growth were similar to the ones observed in STK25-deficient neurons. Furthermore, the knockdown of GM130 in adult neuronal progenitors reproduced this observation (Rao et al., 2018), suggesting

an underlying role of GC integrity in the onset of this rare neurodevelopmental disorder.

The mammalian target of rapamycin (mTOR)-related signaling pathway has acquired increasing importance in seizure disorders, as somatic mutations of upstream and downstream signaling components, underlie cortical lamination deficits and the onset of focal cortical dysplasia and hemimegalencephaly, both neurodevelopmental disorders associated with seizures (Poduri et al., 2013). Most notably, a recent study pointed out that GC fragmentation associated with slight overexpression of Golgin GCC88 decreases mTOR activity and increases autophagy in HeLa cells (Gosavi et al., 2018), suggesting that disruption of GC morphology could lead to reduced mTOR activity in neurological disorders. Conversely, human PMSE brains and brains from STRAD knockdown mice show an enhanced activity of mTORC1 and cortical heterotopias resembling mTORC1 deficient lamination defects (Orlova et al., 2010). Increased mTORC1 activity was associated with LKB1 mislocalization to the nucleus (Orlova et al., 2010), a protein kinase which inhibits mTOR activity through activation of AMPK (Shaw et al., 2004).

Epileptic encephalopathies are highly heterogeneous disorders, and there is little understanding of the impact of genetically associated variants on phenotypical severity (Niestroj et al., 2018). Moreover, the availability of several pharmacological and genetic approaches both in vivo and in vitro can often lead to conflicting interpretations; however, we can conclude that both approaches ultimately lead to GC fragmentation associated to increased synaptic activity. Additionally, the direct role of the GC in regulating mTOR activation, a protein kinase consistently associated with epilepsy, gives us new insight for treating so-called "mTORopathies" (Crino, 2015). In this context, mTOR inhibitor, rapamycin has been successfully used to attenuate seizures and cortical dysplasia in animal models featuring enhanced mTOR activity (Hsieh et al., 2016).

#### REELIN SIGNALING REGULATES GC DYNAMICS AND MORPHOLOGY

In the next section, we will analyze the role of Reelin signaling in the regulation of these processes commonly affected in neurological disorders, and how this pathway plays an important role in the structuring and dynamics of the CG. Indeed, Reelin signaling dysfunction has been documented in several neurodevelopmental and neurodegenerative disorders including AD, PD, HD, and epilepsy (Haas et al., 2002; McCullough et al., 2012; Bodea et al., 2014; Baek et al., 2015; Cuchillo-Ibanez et al., 2016). Furthermore, Reelin signaling plays a central role in regulating neuronal migration and polarity during development and also regulates neuronal plasticity in the adult brain (Santana and Marzolo, 2017; Wasser and Herz, 2017).

In this context, Reelin was first described as a protein absent in the mouse mutant reeler which features defective neuronal migration in the neocortex and hippocampus, resulting in the inversion of cortical layering (Frotscher, 2010). Most notably, reeler recapitulates phenotypes observed in GM130 KO mice (Liu et al., 2017) including ataxia and cerebellar atrophy (Yuasa et al., 1993; Magdaleno et al., 2002) suggesting that Reelin participates in pathways associated with Golgi structural regulation during early brain development.

Reelin is a large, secreted glycoprotein that binds to the ApoE receptor 2 (ApoER2) and/or very low density lipoprotein receptor (VLDLR) leading to Src-family kinases Fyn- and Src-dependent tyrosine phosphorylation of the adaptor protein Dab1, which in turn regulates several downstream pathways, including PI3K-Rac1/Cdc42, PI3K-AKT-GSK3β, and Crk/CrkL-Rap1 among others (**Figure 3**) (Santana and Marzolo, 2017).

Through the PI3K-Cdc42/Rac1 pathway, Reelin is an upstream activator of LIMK1 and cofilin, which, as previously discussed, participate in GC tubule fragmentation (Quassollo et al., 2015). However, it is not yet known if Reelin induces the formation of GOPs. What has been established is that Reelin signaling plays a role in GC deployment and positioning (**Figure 3**). The GC appears condensed near the nucleus in Dab1 and Reelin KO neocortical pyramidal neurons as opposed to extended toward the apical layer in control neurons (Matsuki et al., 2010). Strikingly, acute treatment with Reelinconditioned medium in cultured hippocampal neurons leads to rapid deployment of the GC into the largest neurite. Furthermore, GC deployment is significantly reduced by expression of STK25, suggesting opposing roles for Reelin- and STK25 mediated GC modification (Matsuki et al., 2010). In nocodazoletreated hippocampal neurons, Reelin specifically promotes the reconstitution of disrupted dendritic GC, but not somatic GC. Reelin also induces the formation of plus-end dynamic microtubules labeled with the microtubule plus end binding protein 3 (EB3) (Meseke et al., 2013a). Other studies have shown that the GTPases Rac1 and Cdc42 (via its GEF aPIX) are involved in GC translocation and deployment (Jossin and Cooper, 2011; Meseke et al., 2013b). Cdc42 is found in the plasma membrane and the GC where it is known to participate in determining cell polarity and regulation of the actin cytoskeleton. Activation of Cdc42 also leads to reducing retrograde transport from the GC and increased intra-GC anterograde trafficking via COPI vesicles (Farhan and Hsu, 2016). It remains unclear whether Reelin activation of Cdc42 impacts the processes of cisternal maturation, and retrograde or anterograde trafficking.

More recently, Reelin signaling modulation of the microtubule nucleating protein CLASP2 has further strengthened the role of this pathway in the regulation of neuronal cell polarity and migration through regulation of GC morphology (Dillon et al., 2017). CLASP proteins are found in the GC and act as acentrosomal nucleation sites for the microtubule network (Efimov et al., 2007). CLASP2 associates with the TGN resident Golgin GCC185, a protein known to be necessary for maintaining ribbon formation of the Golgi stack (Brown et al., 2011). The overexpression of CLASP2 enhances dendritic branching and Golgi condensation (Beffert et al., 2012), thus supporting an important role for CLASP2 in GC dynamics. In line with this evidence, CLASP2 function has been consistently associated with dendritic and axonal development, synaptic function, neuronal progenitor proliferation, and migration (Beffert et al., 2012; Dillon et al., 2017).

releasing it from GCC185. Unphosphorylated CLASP2 can bind pDab1 and GCC185, not bound to the microtubules, leads to dismantling of Golgi ribbon morphology. pDab1 via PI3K and ITSN activates Cdc42. The Cdc42/Rac1 GEF αPix/Cool2 has been involved in Reelin-induced Golgi translocation after axon specification (not shown in the figure) and deployment to the apical dendrite. (2) Active Cdc42 and Rac1 induce the positioning of microtubule-associated Golgi vesicles possibly via regulation that involves dynein and microtubule-associated Lis1, necessary for Golgi translocation and deployment. Besides, Reelin signaling promotes the formation of dynamic plus end microtubules labeled with EB3. (3) Also, Cdc42/Rac1 activate LIMK1 that phosphorylates and inactivates cofilin, leading to actin fiber formation, required for Golgi deployment and possibly GOPs formation. (4) Furthermore, active Cdc42 could regulate Golgi transport by reducing COPI retrograde trafficking and promoting tubular Golgi formation at the TGN. (5) Src also phosphorylates NMDAR, increasing Ca2<sup>+</sup> influx and activating CAMKII, which phosphorylates CREB and, together with CBP regulates gene expression, thus, conceivably inducing genes to promote enhanced secretion and possibly GOP biogenesis.

CLASP2 directly interacts with Dab1 and triggers phosphorylation by GSK3β (Dillon et al., 2017). CLASP2 features multiple GSK3β phosphorylation sites, allowing for a delicate tuning of microtubule binding activity which differentially affects dendritic and axonal growth (Hur et al., 2011). Furthermore, Reelin signaling increases inhibitory phosphorylation of GSK3β mediated by AKT (Beffert et al., 2002), adding to a complex regulation of microtubules through regulation of this kinase. GSK3β can be found in the TGN of HeLa cells, and its depletion causes Golgi fragmentation and deregulation of anterograde trafficking, diverting the cation-independent mannose-6-phosphate receptor, CI-M6PR, from transport to the pre-lysosomal compartment into the exocytic pathway (Adachi et al., 2010). GSK3β localization at the GC is dependent on TGN Golgin p230 and also determines CLASP2 phosphorylation status and localization, as GSK3β

depletion leads to increased peripheral localization of CLASP2 (Adachi et al., 2010).

An exciting and recently described Reelin signaling modulator is the multi-scaffold protein Intersectin (ITSN), which acts as a guanine nucleotide exchange factor for Cdc42 and also functions as an essential scaffold for endocytic and exocytic processes at the pre-synapse, making it an interesting protein to study in pathological conditions associated with neurotransmitter release such as epilepsies (Gerth et al., 2017). Pharmacological inhibition of the interaction between ITSN and Cdc42 reduces GC motility and disrupts GC organization through modification of the actin cytoskeleton (Friesland et al., 2013). ITSN1 also interacts directly with Golgin GCC88 in HeLa cells and localizes to the TGN. Most notably, GCC88-induced GC fragmentation is significantly reduced in ITSN1-silenced HeLa cells, further suggesting a role for this protein in actin regulation (Makhoul et al., 2019).

ITSN1 and ITSN2 interact directly with VLDLR and Dab1, and Reelin-mediated Dab1 phosphorylation is significantly reduced in ITSN1 KO neurons abrogating Reelin-mediated long term potentiation (LTP) enhancement (Jakob et al., 2017) thus directly affecting learning and memory processes (Nicoll, 2017). Moreover, double KO of ApoER2 and ITSN1 show laminar defects reminiscent of ApoER2/VLDLR double KO (Jakob et al., 2017). While these findings directly implicate ITSN in Reelin signal transduction associated with neuronal migration and plasticity, the recent description of ITSN residing at the TGN opens the possibility for novel functions of this protein in Reelin-associated GC regulation. In conclusion, Reelin signaling regulates both the actin and microtubule cytoskeleton and in turn these accessory proteins directly impact on Reelin pathway activation.

Throughout this article, we have also argued for an essential role of neuronal activity-dependent modulation of GC architecture. Hyperexcitability is a major driver of GC fragmentation and may account for common morphological GC features observed in neuropathological conditions (Thayer et al., 2013). It is also interesting that increased synaptic activity modulates Reelin signaling. As already mentioned, it is wellestablished that Reelin signaling enhances LTP in hippocampal slices through binding of ApoER2 and increased recruitment of AMPAR to the cell surface (Weeber et al., 2002; Beffert et al., 2005; Qiu et al., 2006) and downstream activation of cAMP response element-binding protein (CREB) transcriptional activator (Telese et al., 2015). More recently, stimulation of synaptic activity by a short incubation with 10 mM KCl was shown to induce dendritic clustering of ApoER2 in hippocampal neurons without modifying its total expression (Pfennig et al., 2017). Increased activity also enhances Dab1 phosphorylation and localization at synaptic sites (Pfennig et al., 2017) suggesting that alterations in synaptic activity modulate Reelin signaling activation through ApoER2.

On the other hand, the secretion of Reelin itself appears independent of calcium-mediated exocytosis that is modulated by synaptic activity, although it is sensitive to brefeldin A, suggesting that this process requires a fully functioning GC (Lacor et al., 2000). Interestingly, the incubation of neurons with kainic acid, as an in vitro model for epileptic seizures, leads to GC dispersion and decreased proteolytic processing of Reelin evidenced by an increase in the full length 400 kDa protein and a decrease in the 320 kDa secreted Reelin. Furthermore, the glycosylation level of the 320 and 400 kDa secreted Reelin is significantly decreased in kainic acid-treated neurons (Kaneko et al., 2016). Altogether, these data suggest, that Reelin post-translational modifications and secretion might be sensitive to glutamate and kainate receptor agonists, and further suggests a regulatory role for neuronal excitability in Reelin pathway signal transduction.

Overall, Reelin signaling appears to impact GC dynamics during development directly and might play a role in the interplay between cell signaling and activity-dependent plasticity which also regulates the GC. Nevertheless, how Reelin signaling directly impacts secretory trafficking remains to be determined. In this regard, a few clues can be found through the examination of transcriptional regulatory pathways. Several lines of evidence point out that efficient secretory pathway functioning is essential for the establishment of GOPs in dendrites. Indeed, overexpression of CrebA in Da neurons increased the number of GOPs by upregulating genes associated with ER-Golgi anterograde trafficking (Chung et al., 2017). Most notably, CREBbinding protein (CBP) has been shown to interact directly with CrebA in Drosophila (Anderson et al., 2005) and CBP modulation also regulates GOP abundance (Chung et al., 2017).

CREB-binding protein and related transcriptional partners including CREB and p300 have been associated with Rubinstein– Taybi syndrome, a neurodevelopmental disorder within the autism spectrum (Roelfsema et al., 2005). CBP-dependent transcription is highly dependent on calcium influx, induces transcription of early genes associated with synaptic plasticity and is important for LTP in the hippocampus (Wood et al., 2005). Nevertheless, CBP also controls transcription in other cell signaling pathways. For instance, the recent identification of LRP8-Reelin-regulated neuronal enhancers for genes activated by acute treatment with Reelin showed that these are modulated by a transcriptional complex, which including CREB and CBP (Telese et al., 2015). Similarly, Reelin treatment enhances the transcription of early genes associated with neuronal plasticity in cortical neurons (Telese et al., 2015). While gene ontology analysis of Reelin-induced genes did not show enrichment in genes associated with protein trafficking, likely due to the acute nature of the stimuli, the convergence of related transcriptional activators further strengthens a possible role for Reelin signaling in the modulation of Golgi dynamics. Particularly interesting genes enhanced by Reelin treatment include the stress response effectors ATF4, PERK, and ETS2 (Telese et al., 2015). These genes are associated with transcriptional regulation induced by Golgi stress (Baumann et al., 2018; Sbodio et al., 2018), suggesting that Reelin-mediated transcriptional activation might precede major changes in Golgi protein composition and directly impact the secretory pathway.

Altogether we can conclude that Reelin signaling impacts GC dynamics through 4 distinct mechanisms: (1) The activation of LIMK1, Cdc42, and GSK3β all well-known downstream effectors of Reelin signaling. (2) The direct interaction and regulation of cytoskeletal scaffolds with core receptor components ApoER2, VLDLR, Dab1 which could suggest a structural role for these

proteins at the GC, distinct from their role as Reelin receptors at the plasma membrane. (3) The regulation of synaptic plasticity which modulated GC dynamics in both physiological and pathological conditions (4) and finally, while there is a robust description of several protein kinases that become activated upon induction of the Reelin pathway, we have only scratched the surface of the transcriptional regulatory network activating downstream. In this context, future transcriptomic and proteomic approaches could unveil new components associating Reelin signaling with GC dynamics.

### CONCLUDING REMARKS

One hundred and twenty years have passed since the Golgi complex was first described, and a formidable body of information regarding its structure, protein composition, homeostatic dynamics, and trafficking has since been elucidated. Different from other cellular pathways, our basic knowledge of the dynamic regulation of the GC and its function in protein trafficking has preceded our knowledge of how defects in this organelle's homeostasis affect health and disease. In this article, we have discussed the current information linking GC dysfunction with neurodegenerative and neurodevelopmental disorders. It is clear that GC dynamics is directly linked to common neuronal defects observed in neurological disorders including abnormal proteolytic processing, dendritic arborization, neuronal migration, and synaptic plasticity. While each of these processes encompasses a broader and more complex protein network, our review should prompt a more thorough investigation of the GC in the onset, diagnosis, and treatment of neurological disorders.

Expanding our knowledge in this area will significantly aid clinical investigation as genome-wide association studies and large-scale exome sequencing initiatives have consistently unveiled variants and protein networks directly associated with both endosomal and secretory trafficking pathways which ultimately converge in the GC (Nuytemans et al., 2013; Novarino et al., 2014; Raghavan et al., 2018; Striano and Nobile, 2018). More importantly, a large body of data directly points to a central role for cell signaling cues in the regulation of Golgi dynamics in both development and disease. Indeed, Reelin

#### REFERENCES


signaling has been implicated in neuronal polarity, migration and synaptic function, and its direct interaction with downstream effectors linked with the regulation of GC architecture places this pathway as a potential master regulator of GC biology and a pharmacological target. Indeed, chemical modulators for Cdc42, LIMK1, and GSK3β are currently available (Hong et al., 2013; Petrilli et al., 2014) and several FDA-approved GSK3β inhibitors are already being used in the treatment of neurological diseases (Caracci et al., 2016).

Finally, we have extensively discussed changes in GC architecture. However, there is little and often conflicting information about how Golgi disassembly orfragmentation affects integral trafficking of transmembrane proteins and ionotropic channels towards the cell surface or synaptic compartment. Protein translation and ER stress are well-known to be associated with neurological diseases. Here, we have comprehensively described the GC alterations in neuropathology. Future work connecting the role of the GC in neuropathology with its established participation in delivering protein cargo to organelles, endosomes and the plasma membrane, will consolidate our understanding of the secretory pathway and unveil novel cellular paradigms for understanding prevalent neurological diseases.

### AUTHOR CONTRIBUTIONS

MOC wrote and designed the first draft of the manuscript and made a relevant contribution to the general idea of the review article. LMF made the figures and wrote the legend section of the manuscript. M-PM supervised all steps of the writing process, including the structure of the article and figures, the list of references, and wrote and edited the final version of the manuscript. All authors contributed to manuscript revision read and approved the submitted version.

# FUNDING

This work was supported by the CONICYT-PCHA/National Graduate Program/2016- 21160410 to MOC and 2017-21171004 to LMF and Fondo Nacional de Ciencia y Tecnología, FONDECYT of Chile, project 1150444 to M-PM.

related to tau hyperphosphorylation. Neurobiol. Dis. 97, 11–23. doi: 10.1016/j. nbd.2016.10.005



induces neurodegeneration. Mol. Neurodegener. 13:8. doi: 10.1186/s13024-018- 0240-1




**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 Caracci, Fuentealba and Marzolo. 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.

# Roles for Golgi Glycans in Oogenesis and Spermatogenesis

#### Ayodele Akintayo and Pamela Stanley\*

Department of Cell Biology, Albert Einstein College of Medicine, New York, NY, United States

Glycosylation of proteins by N- and O-glycans or glycosaminoglycans (GAGs) mostly begins in the endoplasmic reticulum and is further orchestrated in the Golgi compartment via the action of >100 glycosyltransferases that reside in this complex organelle. The synthesis of glycolipids occurs in the Golgi, also by resident glycosyltransferases. A defect in the glycosylation machinery may impair the functions of glycoproteins and other glycosylated molecules, and lead to a congenital disorder of glycosylation (CDG). Spermatogenesis in the male and oogenesis in the female are tightly regulated differentiation events leading to the production of functional gametes. Insights into roles for glycans in gamete production have been obtained from mutant mice following deletion or inactivation of genes that encode a glycosylation activity. In this review, we will summarize the effects of altering the synthesis of N-glycans, O-glycans, proteoglycans, glycophosphatidylinositol (GPI) anchored proteins, and glycolipids during gametogenesis in the mouse. Glycosylation genes whose deletion causes embryonic lethality have been investigated following conditional deletion using various Cre recombinase transgenes with a cell-type specific promoter. The potential effects of mutations in corresponding glycosylation genes of humans will be discussed in relation to consequences to fertility and potential for use in contraception.

Keywords: glycosylation, glycans, Golgi, spermatogenesis, oogenesis, fertility

#### INTRODUCTION

The mammalian glycome is defined by the genes that encode activities required for the synthesis of glycosylated proteins and lipids. These include genes that encode glycosyltransferases, nucleotide sugar synthases, nucleotide sugar modifiers, nucleotide sugar transporters and glycosidases required to prune glycans during synthesis. Also required for optimal glycosylation are proteins that maintain the structure and environment of the secretory pathway. A summary of the classes of glycan synthesized by mammals is given in **Figure 1**. The actual complement of glycan structures expressed in the endoplasmic reticulum (ER), in Golgi compartments, at the cell surface, or secreted by a mammalian cell will depend on the glycosylation-related genes that are active in that cell, and that spectrum will likely vary at different stages of development or differentiation. With few exceptions, all glycoproteins, proteoglycans and glycolipids are synthesized in the secretory pathway, although N-glycans and glycosylphosphatidylinositol (GPI) anchors begin their synthesis on the cytoplasmic side of the ER membrane, before an immature glycan is flipped to the luminal side. Maturation continues in the ER and Golgi compartments. Thus, effects on the structure or biochemical environment of the ER and Golgi compartments may affect glycosylation and alter the nature of the glycans produced. The glycans on mature glycoproteins and glycolipids expressed

#### Edited by:

Vladimir Lupashin, University of Arkansas for Medical Sciences, United States

#### Reviewed by:

Taroh Kinoshita, Osaka University, Japan Michael Tiemeyer, University of Georgia, United States

> \*Correspondence: Pamela Stanley pamela.stanley@einstein.yu.edu

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology Received: 17 March 2019

Accepted: 21 May 2019 Published: 07 June 2019

#### Citation:

Akintayo A and Stanley P (2019) Roles for Golgi Glycans in Oogenesis and Spermatogenesis. Front. Cell Dev. Biol. 7:98. doi: 10.3389/fcell.2019.00098

at the cell surface or secreted from a cell are the functional glycans involved in recognition by glycan binding proteins on other cells, and in the extracellular matrices of a tissue. Glycoprotein exceptions to synthesis in the secretory pathway are proteins modified on Ser or Thr by O-GlcNAc in the cytoplasm or nucleus (Zachara et al., 2015), glycogen that is also synthesized in the cytoplasm, and hyaluronan synthesized at the plasma membrane and extruded to the extracellular matrix (Vasconcelos et al., 2013).

A commonly used strategy to determine if a particular glycan or sugar residue is necessary for the development or differentiation of a specific cell type is to inactivate or delete a glycosylation gene. However, this has the drawback that many glycosylated molecules of different functions will be altered. To address functions for glycans in individual glycoproteins, one or more known glycosylation sites can be mutated to preclude their modification. However, relatively few glycans are added at a known amino acid, making this approach arduous, although useful once glycosylation sites have been established. Usually it is important not to eliminate glycosylation of a glycoprotein altogether since this often leads to defective folding, aggregation and degradation and/or inability to exit the ER.

This review will describe the consequences for oogenesis (**Figure 2**) and spermatogenesis (**Figure 3**) of altering glycan synthesis by targeted inactivation of glycosylation genes responsible for the synthesis of Golgi glycans in the mouse. These glycans are identified in **Figures 1**, **4**. Until recently, glycans in the ovary and testis have been investigated by immunohistochemistry and other histochemical assays and by using glycan binding proteins such as plant lectins (Lee and Damjanov, 1984; Wu et al., 1984; Lohr et al., 2010). However, the application of matrix-assisted laser desorption mass spectrometry imaging (MALDI-MSI) has begun to reveal the molecular nature of glycans along with their location in complex tissues (Drake et al., 2018).

#### GOLGI GLYCANS IMPORTANT FOR OOGENESIS IN MAMMALS

Deletion or inactivation of glycosyltransferases that function in the Golgi compartment may have profound or relatively mild effects (Stanley, 2016). For example, inactivation of the Mgat1 gene which blocks synthesis of all complex and hybrid N-glycans, leads to embryonic lethality (Ioffe and Stanley, 1994; Metzler et al., 1994). By contrast, inactivation of Mgat5, which prevents the addition of a single β1,6 branch of complex N-glycans, has much milder consequences (Dennis et al., 2002; Partridge et al., 2004). Conditional deletion of Mgat1 in oocytes using the ZP3-Cre transgene (Lewandoski et al., 1997) was used to identify roles for complex and hybrid N-glycans in differentiation of oocytes from the primordial follicle stage (**Figure 2**). Loss of MGAT1 and hybrid and complex N-glycans allowed progression of mutant oocytes to ovulation and fertilization (Shi et al., 2004). Heterozygous embryos from mutant egg and wild type sperm give rise to embryos that develop normally. Mgat1 null oocytes fertilized by Mgat1 null sperm also give Mgat1[-/-] embryos that develop to ∼E9.5 as observed in a global knockout (Ioffe and Stanley, 1994; Metzler et al., 1994). However, closer examination revealed that females with Mgat1 null oocytes have ∼25% more empty deciduae than control females, and the zona pellucida of mutant oocytes is very thin. Zona pellucida (ZP) glycoproteins lacked complex N-glycans showing that deletion of Mgat1 was efficient. While all embryos from Mgat1 null oocytes implant, not all continue to develop, suggesting that a proportion of oocytes do not give rise to fully functional embryos. Defective oogenesis became more apparent when females were stimulated to ovulate by hormonal treatment. Females producing Mgat1 null oocytes gave ∼50% fewer ovulated eggs, even though they generated equivalent numbers of antral follicles to controls. In addition, embryos arising from Mgat1 null eggs were slow in developing through the early stages of embryogenesis prior to implantation. Thus, MGAT1 and complex N-glycans (**Figure 4A**) are required for the development of fully competent oocytes and ovulated eggs. In addition, these experiments clearly demonstrated that complex or hybrid N-glycans are not required for sperm binding or fertilization. Further investigations revealed that oocytes lacking MGAT1 have aberrant development of preovulatory follicles (retarded folliculogenesis), and structural alterations of the cumulus in cumulus-oocyte complexes (COC; Williams and Stanley, 2009). The reason for sub-optimal development of oocytes that lack MGAT1 may be related to a reduction in growth factor signaling since complex N-glycans regulate the retention of growth factor receptors at the cell surface, and thereby regulate signaling duration (Boscher et al., 2011).

Whole body deletion of Man2a1 which encodes an alpha-mannosidase II that acts after MGAT1 to prepare the substrate of MGAT2 (**Figure 4A**), does not result in obvious defects in female fertility (Chui et al., 1997). Global deletion of both Man2a1 and Man2a2 (previously called alpha-mannosidase IIX) causes embryonic or perinatal lethality (Akama et al., 2006) and oogenesis in conditional mutants has not been examined. Mgat2 null mice also die perinatally (Ye and Marth, 2004) in an inbred mouse model, and only 30% of the surviving females in an outbred mouse background were fertile. However, no further description of the basis for the fertility defect, and no targeted deletion in oocytes have been reported. The sugar added after GlcNAc to complex N-glycans in the trans Golgi compartment is Gal (**Figure 4A**). Deletion of B4galt1 also causes perinatal lethality in one mouse strain (Lu et al., 1997) but not in another (Asano et al., 1997). Females lacking B4GALT1 were fertile but had a slight reduction in litter size when compared to wild type (Asano et al., 1997).

The deletion of another Golgi glycosyltransferase by ZP3-Cre that affects oogenesis gives rise to a phenotype distinct from the loss of Mgat1. The glycosyltransferase C1GALT1 (also known as T synthase) transfers Gal to GalNAc attached at Ser/Thr residues, and initiates the synthesis of core 1 and 2 O-GalNAc glycans in the Golgi (**Figures 1**, **4B**). Deletion of C1galt1 is embryonic lethal at ∼E13.5 (Xia et al., 2004). Conditional deletion by ZP3-Cre in primary follicles (**Figure 2**) is efficient and strikingly gives rise to an increase in the number of eggs and pups produced by females that lack C1GALT1 in oocytes (Williams and Stanley, 2008). The increase of ∼30–50% in ovulated eggs leads to sustained increases

FIGURE 1 | Cell Surface Glycans in Mammals. The diagram depicts one or more glycans from each class of mammalian glycan. The diagram is modified from Figure in Stanley (2016) with permission. Sugar symbols are according to the Symbol Nomenclature for Glycans (Varki et al., 2015).

in litter size. While follicular development is enhanced, there is no increase in apoptosis of mutant oocytes (Williams and Stanley, 2008). In addition, oocytes lacking C1GALT1 generate more multiple oocyte follicles (MOF) at late stages of folliculogenesis. The formation of MOFs may be due to the compromised follicular basal lamina apparent in follicles containing C1galt1 null oocytes (Christensen et al., 2015; Grasa et al., 2016). In fact, robust fusion of mutant follicles is observed in in vitro cultures containing follicles with C1galt1 mutant oocytes, whereas no MOF arose in cultured control oocytes (Christensen et al., 2017). The enhanced folliculogenesis and fertility in females with oocytes lacking C1GALT1 reflects the enhanced sensitivity

of follicles containing mutant oocytes to stimulation by follicle stimulating hormone (FSH), reduced apoptosis and altered BMP15-GDF9 signaling (Grasa et al., 2015).

Interestingly, when both Mgat1 and C1galt1 were deleted by ZP3-Cre, the results were significantly worse for folliculogenesis, ovulation and fertility than deletion of either one alone (Williams and Stanley, 2011). Removing N-glycans and core 1 and 2 O-GalNAc glycans together abolishes the high fertility phenotype observed in C1galt1 mutant females and worsens the folliculogenesis defects observed in Mgat1 null mice, resulting in an overall reduction in egg production and fertility. Nevertheless, eggs lacking both complex and hybrid N-glycans as well as core 1 and 2 O-glycans (and therefore all Gal residues on N- and O-GalNAc glycans) were fertilized, showing that all these glycans are dispensable for fertilization. However, follicle production was greatly reduced, as were the number of ovulated double mutant eggs, and females had, at most, one litter (Williams et al., 2007; Grasa et al., 2012). The extracellular matrix of the cumulus in the cumulus-oocyte complexes (COCs) with double mutant oocytes was also altered (Lo et al., 2018). Ultimately, the combined inactivation of Mgat1 and C1galt1 in oocytes led to premature ovarian failure (POF), a novel model for investigation of this syndrome in human premature ovarian insufficiency (POI; Williams and Stanley, 2011; Grasa et al., 2016). In vitro culture of double mutant follicles has shown that a subset retains the potential to develop to antral follicles in vitro if obtained prior to the development of POF (Kaune et al., 2017). In summary, the generation of complex or hybrid N-glycans as well as core 1 and 2 O-GalNAc glycans in the Golgi is essential for fully functional oogenesis and the production of a complete complement of developmentally-competent eggs and embryos.

Another class of O-glycans is found only on Ser/Thr of epidermal growth factor-like (EGF) repeats at specific consensus sequences (**Figure 1**; Haltiwanger et al., 2017). Notch receptors and canonical Notch ligands are highly modified by these O-glycans because of the large number of EGF repeats in their extracellular domains. O-fucose, O-glucose and O-GlcNAc residues are transferred to proteins with EGF repeats in the ER by POFUT1, POGLUT1 and EOGT respectively, and further elongated in Golgi compartments (Takeuchi and Haltiwanger, 2014). Deletion of the initiating glycosyltransferase genes is embryonic lethal for O-fucose and O-glucose glycans but not for O-GlcNAc glycans (Varshney and Stanley, 2018). Female mice lacking secretory pathway O-GlcNAc glycans are fertile. Surprisingly, females with ZP3-Cre-mediated deletion of Pofut1 leading to the loss of all O-fucose glycans in oocytes, also have normal fertility, and thus apparently normal oogenesis (Shi et al., 2005), despite proposed roles for Notch signaling in oogenesis (Xu and Gridley, 2013; Feng et al., 2014). By contrast, the global deletion of Lfng, a Golgi GlcNAc-transferase which adds GlcNAc to O-fucose-modified EGF repeats, leads to infertility in female mice due to defective follicular development and meiotic maturation of oocytes (Hahn et al., 2005). Because the loss of O-fucose on EGF repeats has no effect on female fertility, the Lfng null phenotype must not be due to defect(s) in oocytes but in some other cell(s) required for normal oogenesis (**Figure 4E**). No effects on female fertility were reported in mice lacking the related GlcNAc-transferases Mfng or Rfng (Moran et al., 2009).

Only in a few cases have specific roles in oogenesis and/or female fertility been reported for mice with disrupted synthesis of glycosaminoglycans (GAG), glycosylphosphatidylinositol (GPI) anchors or glycolipids. While female mice lacking dermatan-4-O-sulfotransferase 1 (CHST14) are fertile when crossed with wild type males and Chst14 null males are also fertile when crossed to wild type females, homozygous null females crossed with homozygous null males give no progeny (Akyuz et al., 2013; **Figure 4F**). GAGs as well as glycolipids and other glycans of glycoproteins are main components of the extracellular matrix and, it is known that the composition and structure of the extracellular matrix is critical for the development of follicles and eggs in the ovary (Nagyova, 2018). The extracellular matrix and COC are significantly compromised in mice with oocytes lacking MGAT1 (Williams and Stanley, 2009) or both MGAT1 and C1GALT1 (Lo et al., 2018). ZP3 promoter-directed conditional deletion of Piga, the enzyme that initiates GPI-anchor synthesis in the endoplasmic reticulum (ER), gives rise to females in which folliculogenesis and egg production appear normal but mutant eggs lacking PIGA and thus GPI-anchored proteins at the cell surface are unable to fuse with sperm (Alfieri et al., 2003). One reason for this may be the absence of the GPI-anchored protein JUNO at the egg membrane. JUNO is essential for sperm binding to the egg via its ligand Izumo1 (Bianchi et al., 2014).

Despite the indications from mouse mutants that fertility would be expected to be affected in women with altered

glycosylation, a literature search did not reveal known genetic bases associated with glycosylation genes that correlate with defective oogenesis, fertility or POI in humans. Nor do the results in mice to date suggest that disruption of glycosylation would be an effective female contraceptive.

#### GOLGI GLYCANS IMPORTANT FOR SPERMATOGENESIS IN MAMMALS

Spermatogenesis in mammals is a multi-step differentiation process in which spermatogonia (Sg), spermatocytes (Sc), spermatids (St) and ultimately spermatozoa are generated in close contact with Sertoli cells (**Figure 3**). The first Golgi glycosylation gene knockout in the mouse to reveal a role for N-glycans in spermatogenesis was global deletion of Man2a2 (Akama et al., 2002; Fukuda and Akama, 2003), an alpha-mannosidase II which removes two mannose residues from the N-glycan product of MGAT1 and generates the substrate for MGAT2 (**Figure 4A**). The protein is enriched in germ cells except spermatogonia and condensing spermatids, and is not prominent in somatic cells. The N-glycan product of alpha-mannosidase IIx, detected in testis sections using the lectin Griffonia simplicifolia (GSA), was detected on spermatogenic cells (except spermatogonia) but not somatic cells (Sertoli, Leydig), suggesting that MAN2A2 activity is mainly restricted to germ cells in the seminiferous tubule (Akama et al., 2002). Germ cells of the Man2a2 whole body knockout form multi-nuclear cells (MNC or syncytia) and the mice are infertile. Complex N-glycans are absent based on the lack of GSA binding to germ cells. Hybrid N-glycans are expected to be present after deletion of Man2a2 (**Figure 4A**). Thus, neither hybrid nor oligomannose N-glycans on glycoproteins were able to support spermatogenesis or the production of normal numbers of mature sperm. Testes from the Man2a2 global knockout were smaller, contained fewer spermatids and produced immature sperm. Adhesion of Man2a2 mutant germ cells to Sertoli cells was reduced and a N-glycan terminating in GlcNAc was proposed to be necessary for germ-Sertoli cell adhesion. Global deletion of the other alpha mannosidase II encoded by the Man2a1 gene does not affect viability or male fertility, but whole body knockout of both Man2a1 and Man2a2 is neonatal lethal (Akama et al., 2006; Hato et al., 2006). Unusual hybrid N-glycans observed in tissues from E15.5 Man2a1/Man2a2 double knockout embryos (Hato et al., 2006) were not reported in germ cells lacking MAN2A2 (Akama et al., 2002). Disruption of the Man1a2 gene that encodes an alpha mannosidase involved in the pruning of oligomannose N-glycans, is perinatal lethal (Tremblay et al., 2007), and conditional deletion in germ cells has not been reported.

A marked spermatogenic phenotype is obtained when Mgat1 is specifically deleted in spermatogonia using Stra8-iCre (Batista et al., 2012). In males with germ cells lacking both complex and hybrid N-glycans, testes are small, MNCs are very prominent in

testis tubules, apoptosis is increased, and there are no mature sperm in epididymis. Investigation into potential mechanisms of the block in spermatogenesis identified ERK1/2 signaling as markedly reduced in the absence of MGAT1 (Biswas et al., 2018). A physiological inhibitor of MGAT1 termed GnT1IP or MGAT4D, and expressed most highly in male germ cells (Huang and Stanley, 2010; Huang et al., 2015), is being investigated as a potential regulator of complex N-glycan production during spermatogenesis. Whole body disruption of Mgat2 gene expression is lethal in the early days postpartum in inbred mice. On an "outbred" background, the few survivors have spermatogonia and spermatocytes but no round spermatids or mature spermatozoa in the lumen of the seminiferous tubules (Wang et al., 2002). Inactivation of Fut8 encoding the fucosyltransferase responsible for N-glycan core fucosylation, induces early postnatal death (Wang et al., 2005) and no targeted deletion in germ cells has been reported.

The absence of core 1 and 2 O-glycans due to deletion of C1galt1 by Stra8-iCre, has no effect on spermatogenesis (Batista et al., 2012). Interestingly, neither does deletion of Pofut1, and thus the absence of O-fucose glycans does not disrupt spermatogenesis (Batista et al., 2012). Deletion of Pofut1 in Sertoli cells using Amh-Cre also has no deleterious effect on spermatogenesis (Hasegawa et al., 2012). Nor does the absence of NOTCH1 in germ cells (Batista et al., 2012). However, global deletion of the Golgi GlcNAc-transferase encoded by Lfng, which acts after POFUT1, gives rise to a defective rete testis (Hahn et al., 2009). This results in few sperm in the epididymis, although spermatogenesis progresses normally. Since this was not observed in Pofut1 conditional germ cell knockout males, the Lfng knockout rete testis phenotype is probably due to defective development of non-germ cells of the testis. Global deletion of B4galt1 leads to delayed spermatogenesis in the few males that do not die perinatally (Lu et al., 1997). Seminiferous tubules are smaller and spermatids are fewer in B4galt1 null testes. However, fertility and viability are not affected in B4galt1 null males on a different genetic background (Nishie et al., 2007).

In summary, hybrid and complex N-glycans (**Figure 4A**) are required in germ cells for spermatogenesis to progress to mature sperm but core 1 and core 2 O-GalNAc glycans (**Figure 4B**) or O-fucose glycans (**Figure 4E**) are dispensable. Interestingly, however, some GalNAc residues on Ser/Thr, potentially unmodified further or extended to make core 3 or core 4 O-GalNAc glycans (Brockhausen and Stanley, 2017), are required for spermatogenesis. Deletion of a GalNAc transferase termed Galnt3, which is one of approximately 20 polypeptide GalNAc transferases that initiate the synthesis of O-GalNAc glycans in the Golgi, leads to disrupted spermatogenesis and male infertility (Ichikawa et al., 2009; Miyazaki et al., 2013). GALNT3 localizes to early and medial Golgi compartments of spermatids and spermatocytes (Miyazaki et al., 2013). GALNT3 seems to be important for O-GalNAc addition to proteins in testis because its deletion can be easily detected by lectins that bind to GalNAc-O-Ser/Thr. Deletion leads to defects in acrosome formation, increased apoptosis in the seminiferous tubule, reduced production of sperm, and oligoasthenoteratozoospermia (rare and immotile sperm, deformed round head spermatozoa). A related member of the GALNT family termed GALNTL5, which is lacking a portion of the usual GALNT C-terminal domain and has no known GalNAc-transferase activity, is exclusively expressed in testis germ cells. The protein is detected in the cytoplasm of round spermatids, around the acrosome of elongated spermatids, and in the neck region of spermatozoa. Haploinsufficiency causes spermatogenic defects in the mouse due to immotile spermatozoa (Takasaki et al., 2014). Heterozygosity of GALNTL5 also leads to immotile sperm in men (Hagiuda et al., 2019). It will be interesting to know the function of this novel glycosyltransferase-like protein.

Glycolipids synthesized in the Golgi (**Figure 4C**) are also important in spermatogenesis. This was initially noted in mice lacking GM2/GD2 synthase (B4GALNT1) which synthesize only Glc-ceramide (Glc-Cer), Lactosyl-Cer, GM3 and GD3 Golgi glycolipids (Takamiya et al., 1998). Male mice lacking B4GALNT1 are sterile and do not produce sperm. Spermatogenesis proceeds to the stage of round spermatids but maturing spermatids fuse into MNCs and sperm are not produced. Investigations of Glycosphingolipids (GSL) in wild type testes compared to Galnt1[-/-] testes identified novel fucosylated neutral and monosialylated GSL with very long polyenoic ceramides termed FGSL (Sandhoff et al., 2005). However, the absence of sialylated GSL following deletion of GM3 synthase (ST3GAL5/SIAT9) or GD3 synthase (ST8SIA1) and therefore the loss of sialylated FGSL has no effect on male fertility (Kawai et al., 2001; Yamashita et al., 2003, 2005). Thus, the neutral Golgi FGSL found in germ cells but not Leydig or Sertoli cells (Sandhoff et al., 2005) appear to be essential for spermatogenesis and male fertility. This result suggests that O-series gangliosides are sufficient for male spermatogenesis and sialic acid residues are not necessary on these glycolipids during germ cells maturation (**Figure 4C**). The loss of seminolipid (HSO3-3-monogalactosylalkylacylglycerol or SO4-Gal-Eag; **Figure 4D**) by deletion of the galactosyltransferase CGT also disrupts spermatogenesis (Fujimoto et al., 2000). Cgt is expressed in later-stage spermatocytes and deletion does not disrupt the formation of spermatogonia or early spermatocytes. However, mutant males have no spermatids or sperm. Deletion of Cgt precludes the addition of Gal to the precursor of seminolipid (Eag). However, subsequent deletion of the sulfotransferase Cst gives a similar phenotype to deletion of Cgt (Honke et al., 2002) showing that Gal-Eag is not able to fulfill the functions of SO4-Gal-Eag in testis. Cst is expressed in germ cells and germ cells of Cst null males do not proceed through the first meiotic division (Honke et al., 2002; Zhang et al., 2005).

The GPI-anchor on the subset of GPI-anchored proteins in testis is initiated on the ER membrane and additional sugars are added in the Golgi compartment (Ferguson et al., 2017; **Figure 1**). While numerous GPI-anchored proteins are important in the formation of sperm, effects specific to GPI anchor Golgi glycosyltransferases have not been reported. The ER resident protein PGAP1 (post GPI attachment to proteins 1) is a GPI inositol-deacylase involved in the maturation of the GPI-anchor by removing palmitate from inositol. A whole body deletion of the gene does not affect spermatogenesis but induces a severe reduction in male fertility. The ability of sperm to ascend into

the oviduct and to attach to the zona pellucida of the oocyte is severely impaired (Ueda et al., 2007). Inactivation of the Piga gene which blocks the formation of GPI anchors altogether is embryonic lethal, but chimeric mice generated by injecting Piga null ES cells into wild type blastocysts allowed mice with Piga null and wild type cells to be born and spermatogenesis to be investigated (Lin et al., 2000). Chimeric males were sterile. While spermatogenesis was markedly reduced, sperm were made and sperm lacking PIGA were detected in the epididymis. However, Piga null sperm were not transmitted.

In searches for connections between reduced fertility in men and defective glycosylation, only one mutation in a putative glycosyltransferase gene has been identified - GALNTL5 (Hagiuda et al., 2019). However, many more glycosyltransferase defects may ultimately be tied to altered spermatogenesis and fertility in men if results from gene deletions in mouse are a guide. In addition, the mechanisms of defective spermatogenesis in mice vary depending on the glycosylation pathway affected. Glycosyltransferases whose loss appears to spare spermatogonia and spermatocytes but result in fused spermatids and no mature sperm are good candidates for reversible contraception in men. Thus, MGAT1 (**Figure 4A**) and B4GALNT1 (**Figure 4C**) make good inhibitor targets for development as male contraceptives.

#### CONCLUSION AND FUTURE DIRECTIONS

Although relatively few mice with germ-cell specific deletion or inactivation of a glycosylation gene have been investigated, it is clear from these experiments that different classes of glycans are required for different aspects of oogenesis and spermatogenesis in mammals. Thus, complex N-glycans are necessary for optimal oogenesis and the production of fully functional ovulated eggs, and they are essential for the production of sperm. This is not due to general defects in protein folding or degradation

#### REFERENCES


since MGAT1 acts in the medial Golgi, long after glycoproteins have exited the ER. By contrast, core 1 and core 2 O-GalNAc glycans are not required for spermatogenesis but are important for the regulation of oogenesis and female fertility and for protection from POF. O-fucose glycans are dispensable for both oogenesis and spermatogenesis when deleted in germ cells. The requirement for LFNG in both female and male fertility is predicted to be due to its functions in non-germ cells. Complex glycolipids are necessary for spermatogenesis but appear to be dispensable for oogenesis. Glycosaminoglycans have not been sufficiently investigated to know if they may be important in fertility. Inhibiting synthesis of all GPI anchors does not prevent sperm from being produced but mutant sperm are not transmitted. In females, the absence of GPI anchors impairs the ability of oocytes to fuse with sperm. There is only one human mutation in a putative O-GalNAc-transferase, GALNTL5, which is associated with infertility. However, other genetic bases of defective oogenesis, spermatogenesis or fertility are sure to be detected in humans based on results to date in mutant mice. The sensitivity of spermatogenesis to defects in all these pathways clearly indicates the potential to give rise to idiopathic cases of infertility.

### AUTHOR CONTRIBUTIONS

PS and AA wrote the manuscript. PS modified the figures. AA designed the **Figures 3**, **4**.

#### FUNDING

The writing of this review was supported by funds from grant RO1 105399 to PS from the National Institute of General Medical Sciences. Funding was also provided by grant RO1 36434 to PS from the National Cancer Institute.

are dispensable, for mammalian spermatogenesis. Biol. Reprod. 86:179. doi: 10.1095/biolreprod.111.098103




**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 Akintayo and Stanley. 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.

# Golgi pH, Ion and Redox Homeostasis: How Much Do They Really Matter?

#### Sakari Kellokumpu\*

Faculty of Biochemistry and Molecular Medicine, University of Oulu, Oulu, Finland

Exocytic and endocytic compartments each have their own unique luminal ion and pH environment that is important for their normal functioning. A failure to maintain this environment – the loss of homeostasis – is not uncommon. In the worst case, all the main Golgi functions, including glycosylation, membrane trafficking and protein sorting, can be perturbed. Several factors contribute to Golgi homeostasis. These include not only ions such as H+, Ca2+, Mg2+, Mn2+, but also Golgi redox state and nitric oxide (NO) levels, both of which are dependent on the oxygen levels in the cells. Changes to any one of these factors have consequences on Golgi functions, the nature of which can be dissimilar or similar depending upon the defects themselves. For example, altered Golgi pH homeostasis gives rise to Cutis laxa disease, in which glycosylation and membrane trafficking are both affected, while altered Ca2<sup>+</sup> homeostasis due to the mutated SCPA1 gene in Hailey–Hailey disease, perturbs various protein sorting, proteolytic cleavage and membrane trafficking events in the Golgi. This review gives an overview of the molecular machineries involved in the maintenance of Golgi ion, pH and redox homeostasis, followed by a discussion of the organelle dysfunction and disease that frequently result from their breakdown. Congenital disorders of glycosylation (CDGs) are discussed only when they contribute directly to Golgi pH, ion or redox homeostasis. Current evidence emphasizes that, rather than being mere supporting factors, Golgi pH, ion and redox homeostasis are in fact key players that orchestrate and maintain all Golgi functions.

#### Edited by:

Vladimir Lupashin, University of Arkansas for Medical Sciences, United States

#### Reviewed by:

Martin Lowe, The University of Manchester, United Kingdom Francois Foulquier, Université de Lille, France Daniel Ungar, University of York, United Kingdom

\*Correspondence:

Sakari Kellokumpu sakari.kellokumpu@oulu.fi

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology Received: 21 March 2019 Accepted: 16 May 2019 Published: 11 June 2019

#### Citation:

Kellokumpu S (2019) Golgi pH, Ion and Redox Homeostasis: How Much Do They Really Matter? Front. Cell Dev. Biol. 7:93. doi: 10.3389/fcell.2019.00093 Keywords: homeostasis, Golgi pH, Golgi redox state, glycosylation, protein sorting, cancer

#### WHAT IS GOLGI HOMEOSTASIS?

Compartmentalization is a key feature of eukaryotic cells, and it allows cells to complete various tasks with amazing speed and specificity. One drawback to compartmentalization is its need for the extra energy required to generate the unique luminal environments of each compartment. Unlike the ER, the other secretory pathway compartments, such as the ERGIC, Golgi apparatus and secretory vesicles, are all mildly acidic (Kim et al., 1996; Demaurex et al., 1998; Palokangas et al., 1998; Schapiro and Grinstein, 2000; Wu et al., 2001; Paroutis et al., 2004). Each of them has its own unique resting pH (pH set point) that facilitates their efficient functioning, be it membrane trafficking, cargo selection, glycosylation, proteolysis, or protein sorting. Nevertheless, these compartments also have common properties with the ER, with which they all communicate at

least to some extent. The ER and the Golgi apparatus are the most closely associated compartments, sharing several common properties including high calcium concentration and oxidative potential, as well as the ability to synthesize glycans. This review, focusing on the Golgi apparatus, will discuss ER-related processes for comparative purposes only.

In general, organelle acidity is driven by the ATP-mediated proton pump, the V(vacuolar)-ATPase, which is counterbalanced by anion influx or cation efflux, and proton leak back to the cytoplasm via a "H<sup>+</sup> leak channel" whose identity still remains elusive (Paroutis et al., 2004). Many other energy consuming pumps and leak channels, which are needed to maintain balanced Cl−, Ca2+, Mn2<sup>+</sup> and K<sup>+</sup> levels, are also present in the Golgi membranes. These include the Golgi pH regulator (GPHR, a chloride channel), a mid-1-related chloride channel (MClC) and voltage-gated chloride channels ClC-3B in mammalian cells, and Gef-1 in yeast (Schwappach et al., 1998; Jentsch et al., 1999; Nagasawa et al., 2001; Gentzsch et al., 2003; Maeda et al., 2008). Golgi membranes also contain two different isoforms of the Na+/H<sup>+</sup> exchanger (NHE7 and NHE8), of which NHE7 seems to mediate the influx of Na<sup>+</sup> or K<sup>+</sup> in exchange for H<sup>+</sup> (Numata and Orlowski, 2001; Lin et al., 2005). Although the exact physiological roles of many of these transporters remain unclear, they are known to contribute to Golgi resting pH, membrane potential, vesicular trafficking and protein sorting in the organelle. The use of fluorescent redox probes has recently revealed the Golgi redox state to be important for Golgi homeostasis and functions. Samoylenko et al. (2013) showed that the oxidative potential of the Golgi is higher than that of the endoplasmic reticulum (ER), the main site of disulfide bond formation in the cells. Indeed, earlier observations had shown that disulfide bonds can also form "late," i.e., in the Golgi compartment, and facilitate disulfide bondmediated oligomerization of some secretory products before their secretion to the extracellular space (Wagner, 1990; Walter et al., 2009; Chiu and Hogg, 2019). Cellular oxygen levels also regulate the level of Golgi nitric oxide (NO), a free radical that has been shown to be important for maintaining Golgi morphology and for ensuring continued membrane trafficking to the cell surface (Galkin et al., 2007; Nakagomi et al., 2008; Lee et al., 2011, 2013). Taken together, these examples highlight the complexity of factors needed to maintain the unique Golgi environment and the functions that depend on it. The following paragraphs summarize the molecular machineries involved, before focusing on why their failures result in organelle dysfunction and disease.

#### REGULATION OF GOLGI pH, ION AND REDOX HOMEOSTASIS

#### Transport of Protons and Golgi pH Homeostasis

The acidity of the Golgi lumen was first demonstrated in 1983 by using electron microscopy and a compound (DAMP) that accumulated in acidic cellular compartments (Glickman et al., 1983). Currently, several different fluorescence-based approaches have been used to identify proteins that contribute to Golgi acidity and its resting pH (Glickman et al., 1983; Kim et al., 1996, Kim et al., 1998; Demaurex et al., 1998; Llopis et al., 1998; Miesenbock et al., 1998; Schapiro and Grinstein, 2000; Wu et al., 2000, 2001; Machen et al., 2001; Paroutis et al., 2004). These studies have shown that different Golgi sub-compartments have distinct pH set points, decreasing along the cis–trans axis of the Golgi stack from pH 6.7 (cis-Golgi) to pH 6.0 at the trans-Golgi network (TGN). A pertinent question is how this pH gradient is established and maintained along the Golgi stack, given its dynamic nature resulting from the continuous flow of incoming and leaving vesicular carriers. Another related issue is whether a similar gradient also applies to other ions that are uniquely concentrated in the Golgi lumen and thus contribute to its ion homeostasis.

The resting pH of the Golgi lumen is now known to be determined mainly by three different ion transport systems that include the vacuolar (V)-ATPase-mediated proton pump, counter ion (Cl−) transport, and proton "leak" across the Golgi membranes back to the cytoplasm (Wu et al., 2001; Demaurex, 2002; Paroutis et al., 2004). In brief, the V-ATPase uses ATP as an energy source to pump protons into the Golgi lumen. Due to proton pumping, the membrane potential starts to increase (inside positive) and must be counterbalanced by Cl<sup>−</sup> influx. This very likely takes place via the GPHR Cl<sup>−</sup> channel (or a cation efflux channel). Once the Golgi pH is sufficiently acidic (pH < 6.3), proton efflux via an elusive "proton leak channel" prevents further acidification of the Golgi lumen. The resting pH, or the pH set point, is established once the rate of proton pumping matches its leak rate across Golgi membranes. Because of continuous H<sup>+</sup> pumping by the V-ATPase, it is the rate of H<sup>+</sup> leakage that dictates the resting pH of the organelle (Wu et al., 2001). The authors showed that the rate of proton efflux decreases between successive secretory compartments. They also suggested that the higher density of H<sup>+</sup> pumps in the later secretory compartments may also contribute to their lower resting pH. These two factors are likely responsible also for the decreasing pH gradient along the cis- to trans-axis of the Golgi stacks, even though direct proof for this does not yet exist.

The V-ATPase itself is a multi-subunit protein complex (Drory and Nelson, 2006; Jefferies et al., 2008), whose composition may vary between different compartments. For example, the Golgi localized V-ATPase appears to possess a subunit a different from that found in other V-ATPases (the Stv1p instead of the Vph1p in yeast) (Jefferies et al., 2008). The V-ATPase activity is also regulated by glucose or nutrient levels, yet under normal conditions (i.e., at least when counter-ion conductance is sufficient and, therefore, does not restrict proton pumping), it is assumed to be constantly active (Schapiro and Grinstein, 2000; Wu et al., 2001). In support of this, the Golgi lumen in intact cells starts to alkalinize when the V-ATPase activity is shut down by using concanamycin A (**Figure 1**, green dots).

Cl<sup>−</sup> influx seems to be normally required to prevent membrane potential increase due to proton pumping by the V-ATPase (Glickman et al., 1983; Schapiro and Grinstein, 2000; Paroutis et al., 2004). Under normal conditions, it is

considered to be high enough and mediated by the GPHR protein channel termed the Golgi pH Regulator (Maeda et al., 2008). Mutation of the protein was shown to increase Golgi resting pH (by 0.4–0.5 pH units), alter glycosylation, delay transport to the plasma membrane, and induce Golgi fragmentation. These findings thus provide strong support for the view that H<sup>+</sup> pumping is dependent on Cl<sup>−</sup> influx and is needed to maintain a constant membrane potential. The extent to which other Golgi-localized chloride channels, such as the voltage-gated chloride channels ClC-3B (Gentzsch et al., 2003) and Gef1p in yeast (Schwappach et al., 1998) regulate Golgi resting pH remains unclear.

Other studies have suggested that continuous H<sup>+</sup> pumping may be facilitated by passive K<sup>+</sup> efflux rather than by Cl<sup>−</sup> influx (Howell and Palade, 1982). This may relate to a high permeability of the Golgi membranes to K<sup>+</sup> ions (Schapiro and Grinstein, 2000), and could perhaps be mediated by Na<sup>+</sup> and K<sup>+</sup> conductive channels or transporters such as the Na+/K+-ATPase (Poschet et al., 2001). In support of the latter possibility, acetylstrophanthidin (an inhibitor of the Na+/K+- ATPase) was proposed to increase luminal acidity by inhibiting electrogenic Na+/K<sup>+</sup> exchange (3 Na<sup>+</sup> for 2 K+), thereby reducing the accumulation of other cations (relative to H+) in the Golgi lumen. Alternatively, the Na+/H<sup>+</sup> exchanger NH7 could also facilitate the acidification of the Golgi lumen by transporting H<sup>+</sup> into the Golgi lumen in exchange for luminal K<sup>+</sup> ions (Numata and Orlowski, 2001). However, recent data indicates that NH7 does not transport K<sup>+</sup> ions (Milosavljevic et al., 2014), thus leaving open whether Na<sup>+</sup> ions may suffice for an acid loading function of this exchanger in the Golgi compartment.

# Proton Leak Across the Golgi Membranes

Despite its importance, the identity of the "proton leak channel" still remains elusive. It may involve exchange of luminal protons for cytosolic cations via a proton conductive channel, or via import of base equivalents. Physiological measurements indicate that proton efflux in the TGN is voltage-sensitive and inhibited by Zn2+, suggesting the involvement of a regulated channel (Cherny and DeCoursey, 1999; Schapiro and Grinstein, 2000). Other studies suggest that the molecular characteristics of a putative H<sup>+</sup> channel mimic those of the plasma membrane H<sup>+</sup> channels (Numata and Orlowski, 2001; Nakamura et al., 2005). Therefore, the two ubiquitously expressed, Golgi-localized Na+/H<sup>+</sup> exchanger isoforms, NHE7 and NHE8, are good candidates for this channel, because Na+/H<sup>+</sup> exchange is normally driven by existing ion gradients, and high amounts of H<sup>+</sup> in the Golgi lumen will drive influx of Na+. In support of this, overexpression of both NHE7 and NHE8 were found to increase Na<sup>+</sup> and K<sup>+</sup> influx to the Golgi lumen (Numata and Orlowski, 2001), and both also raised the Golgi resting pH (Nakamura et al., 2005). However, changes in sodium concentration during Golgi pH measurements did not markedly alter the Golgi resting pH (Demaurex et al., 1998), leaving some doubts about the possible roles of these NHEs in mediating proton leakage across Golgi membranes. In accordance with this, Milosavljevic et al. (2014) recently showed that NHE7, at least when expressed at the plasma membrane, acts as an acid loader rather than as a "H<sup>+</sup> leak" pathway in the cells. The NHE8 isoform also seems to have more pronounced effects on endosomes than it has on the Golgi (Lawrence and Bowers, 2010). Therefore, further work is needed to clarify the exact roles of the NHEs in the Golgi membranes.

Soluble buffering molecules may also be used for regulating the Golgi resting pH. In support of this view, a homolog of the erythrocyte anion exchanger 1 (Band 3, AE1, SLC4A1) was identified as the AE2a isoform (SLC4A2a) of the SLC4A gene family in the Golgi membranes in a number of cell types (Kellokumpu et al., 1988; Holappa et al., 2001). All members of this gene family are electroneutral HCO<sup>3</sup> <sup>−</sup>/Cl<sup>−</sup> exchangers, regulating cytosolic pH, chloride concentration and cell volume through an obligatory one to one exchange of chloride for bicarbonate (Romero et al., 2004; Alper, 2006). Our recent highthroughput Golgi pH measurements indicate that it is involved in Golgi pH regulation, as its overexpression increased, and knockdown decreased, the Golgi resting pH (Khosrowabadi et al., unpublished). The potential involvement of various channels, pumps and transporters in the maintenance of Golgi homeostasis is summarized in **Figure 2**.

# Transport of Ca2<sup>+</sup> and Mn2<sup>+</sup> Ions and Golgi Homeostasis

Ca2+, Mg2+, or Mn2<sup>+</sup> ions are present at high concentrations in the Golgi lumen (Van Baelen et al., 2004; Pizzo et al., 2010). Their presence is important for cargo concentration and sorting (Chanat and Huttner, 1991) and glycosylation (Leach, 1971; Vanoevelen et al., 2007). Golgi membranes also possess relevant pumps for Ca2<sup>+</sup> and Mn2<sup>+</sup> uptake (SERCA2, SPCA1/2),

channels for Ca2<sup>+</sup> release (IP3R, RyR) and luminal proteins that bind Ca2<sup>+</sup> with high affinity (Van Baelen et al., 2004; Brini and Carafoli, 2009; Lin et al., 2009; Vangheluwe et al., 2009; Zampese and Pizzo, 2012). Mn2<sup>+</sup> ions are important cofactors for Golgi-resident glycosyltransferases. The DXD motif conserved in many glycosyltransferases appears to have a key role in Mn2+ mediated donor substrate binding and catalytic activity (Breton et al., 2006). Mn2<sup>+</sup> ions also act as scavengers for reactive oxygen species (ROS) (Coassin et al., 1992). Of these, SERCA and SPCA type pumps are responsible for the maintenance of the low cytosolic and high luminal Ca2<sup>+</sup> concentrations typical of many secretory pathway compartments. These two Ca2<sup>+</sup> pumps seem to contribute differentially to Ca2<sup>+</sup> uptake into the Golgi, as SERCA2 is enriched in the cis-Golgi, while SPCA1 is mainly present in the trans-Golgi (Wong et al., 2013). The localization of Ca2<sup>+</sup> release channels, the inositol-1,3,5-trisphosphate receptor (IP3Rs) and the ryanodine receptor (RyR), also seems to be different, as IP3 did not release Ca2<sup>+</sup> in the trans-Golgi, while activation by caffeine did so (Vanoevelen et al., 2004; Lissandron et al., 2010; Wong et al., 2013). Contrasting with SERCAs, SPCAs are also engaged in Mn2<sup>+</sup> transport, and thus can provide this essential trace metal supply to Golgi glycosyltransferases (Vangheluwe et al., 2009).

Recent evidence also suggests that mutations of TMEM165 cause a type II congenital disorder of glycosylation in humans by interfering with Mn2<sup>+</sup> and, perhaps, also Ca2+/H<sup>+</sup> transport (Foulquier et al., 2012; Dulary et al., 2017, 2018; Thines et al., 2018) and, therefore, also with Golgi ion and pH homeostasis and glycosylation. However, it is not yet fully clear what role this multi-spanning membrane protein plays in Golgi ion homeostasis. Recent evidence indicates that, unlike the Golgilocalized SPCA1, the ER-associated SERCA pump 2b isoform partially rescued TMEM165 KO-induced glycosylation defect by Mn2<sup>+</sup> (Houdou et al., 2019). Moreover, those authors also recently showed that the TMEM165 KO can also be rescued by galactose supplement in HEK293 cell culture media, or when given to patients (Morelle et al., 2017). Further studies are needed to reveal the exact role of the TMEM165 transporter in the maintenance of Golgi ion homeostasis.

Another multi-spanning membrane protein capable of transporting Ca2<sup>+</sup> has been identified in the Golgi membranes. This protein, named Golgi-associated anti-apoptotic protein (GAAP), has recently gained increasing attention due to its role in tumorigenesis (Rojas-Rivera and Hetz, 2015; Carrara et al., 2017). It is a member of the Transmembrane Bax Inhibitor-1 Motifcontaining (TMBIM) protein family regulating Ca2<sup>+</sup> levels and fluxes in intracellular stores, confering resistance to a broad range of apoptotic stimuli and promoting cell adhesion and migration via the activation of store-operated Ca2<sup>+</sup> entry (SOCE) (Saraiva et al., 2013; Carrara et al., 2015).

The Golgi lumen also harbors several Ca2+-binding proteins, including Cab45, CALNUC, p54/NEFA and calumenin, all of which, except for Cab45, are distinct from their ER counterparts. Of these, the most abundant is CALNUC, an EF-hand, Ca2+ binding protein resident in the CGN and cis-Golgi cisternae. It plays a major role in Ca2<sup>+</sup> buffering and secretion through the Golgi (Lin et al., 1999, 2009). Recent studies of Cab45 have demonstrated its importance in Ca2+-mediated protein sorting. Cab45 is the core component of this oligomerizationdriven sorting mechanism, also involving the cytoplasmic actin cytoskeleton, and the Ca2<sup>+</sup> ATPase SPCA1 (Pakdel and von Blume, 2018). The system relies first on the local synthesis of sphingomyelin at the TGN membrane enhancing Ca2<sup>+</sup> import by SPCA1, which then drives secretory protein sorting and export, thereby coupling lipid synthesis to protein sorting and secretion (Deng et al., 2018).

#### Golgi Redox Homeostasis

Genetically encoded and targeted fluorescent probes such as roGFP and HyPer have been used to determine organelle redox states (Meyer and Dick, 2010; Lukyanov and Belousov, 2014). By using roGFP2 as a probe, we have previously shown that mitochondria have a less oxidizing environment than that of the ER (Samoylenko et al., 2013). Intriguingly, it was also found that the Golgi lumen is more oxidizing than the ER despite being considered as the most oxidizing compartment in eukaryotic cells. One possibility for its higher oxidizing power is that it serves for "late" disulfide bond formation, as indicated in studies showing that the assembly of von Willebrand factor oligomers to multimers, or other secretory products, requires tail-to-tail disulfide bond formation in the Golgi (Wagner, 1990; Chiu and Hogg, 2019). Such disulfide bond formation in the Golgi is likely assisted by members of the Quiescin-Sulfhydryl Oxidase (QSOX) gene family (Codding et al., 2012) that all display PDI-like thioredoxin (Trx) domains and ERVlike oxidase domains. These domains allow QSOX proteins to efficiently couple disulfide bond formation with the reduction of molecular oxygen to hydrogen peroxide. QSOX1 is widely expressed within the secretory pathway compartments, while one of the splice variants, QSOX1a, mainly localizes to the Golgi (Chakravarthi et al., 2007; Heckler et al., 2008). This suggests its involvement in disulfide bond formation, possibly related to the maturation of ECM components, or to the formation of higher order structures in the Golgi. Other proteins that can regulate Golgi redox homeostasis include the glutaredoxins Grx6 and Grx7 (Mesecke et al., 2008). They both belong to an ubiquitous family of proteins that catalyze the reduction of disulfide bonds with the help of reduced glutathione. Grx6 and Grx7 represent the first glutaredoxins found in the cis-Golgi in baker's yeast (Mesecke et al., 2008). They both show a high glutaredoxin activity in vitro, and yeast cells lacking both proteins exhibit growth defects and a strongly increased sensitivity toward oxidizing agents. Grx6 and Grx7 are probably important for counteracting oxidation-driven disulfide bond formation in the Golgi.

The availability of oxygen is also intimately linked to the Golgi redox state. When low, it causes hypoxia, a condition that affects multiple cellular compartments including mitochondria and the ER. Recent evidence indicates that hypoxia also modulates Golgi functions and, in particular, those related to membrane trafficking and glycosylation events. Accordingly, hypoxia has been shown to alter expression levels of both glycosyltransferase and nucleotide sugar transporter genes, and to inhibit membrane trafficking between the ER and the Golgi (Koike et al., 2004; Shirato et al., 2011; Belo et al., 2015; Bensellam et al., 2016; Taniguchi et al., 2016). We recently showed that hypoxia modulates the Golgi redox state and glycosylation without markedly affecting Golgi pH homeostasis (Hassinen et al., 2019; see also below). Oxygen levels not only affect the redox state of the Golgi lumen, but also the production of NO levels by modulating the activity and expression of various NO synthase isoforms, including neuronal nitric oxide synthase (nNOS), inducible NOS (iNOS), and endothelial NOS (eNOS) (Jeffrey Man et al., 2014). Of these, only eNOS is located to the Golgi membranes (Iwakiri et al., 2006) via myristylation or palmitylation of its N-terminus. NO is a lipophilic compound and can readily pass through membranes. Both NO and superoxide (O<sup>2</sup> <sup>−</sup>), another possible product of eNOS activity, are highly reactive free radicals and increase ROS load.

#### ALTERED GOLGI HOMEOSTASIS IN GOLGI DYSFUNCTION AND DISEASE

#### Membrane Trafficking and Protein Sorting Defects

Failure to maintain Golgi pH, ion, and redox homeostasis is commonly associated with membrane trafficking and protein sorting defects. Monensin, a Na+/H<sup>+</sup> ionophore, was the first compound shown to block intra-Golgi transport between the medial- and trans-Golgi cisternae (Griffiths et al., 1983a,b). Kuismanen et al. (1985) reported that intracellular transport of the Uukuniemi virus membrane glycoproteins (G1 and G2) was not inhibited by monensin. Whether this discrepant behavior is cell or virus type-dependent remains unclear. Protein sorting in the Golgi is also dependent upon existing pH gradients. In line with these observations, Schaub et al. (2006) showed that monensin induces the relocalization of B4GalT1 galactosyltransferase (but not ST6Gal-I) and alpha-1,3-fucosyltransferase 6 (Schaub et al., 2008) in swollen vesicles derived from the TGN based on their colocalization with TGN46, a specific TGN marker. This relocalization was also found to be signal-mediated, involving a short sequence in its cytoplasmic tail, which, when present in ST6Gal-I, was able to relocate the latter into the TGN-derived swollen vesicles from the trans-Golgi cisternae or the TGN. However, the signals were not needed for the steady state localization of these enzymes in the trans-Golgi cisternae.

A better example of pH-sensitive membrane trafficking steps is the retrograde transport from the Golgi back to the ER. This was demonstrated using bafilomycin A (a V-ATPase specific inhibitor), which inhibited retrograde, but not anterograde, transport from the intermediate compartment (IC)/cis-Golgi

back to the ER (Palokangas et al., 1998). This preferential effect on retrograde trafficking may relate to the more acidic environment at the IC/cis-Golgi interface than that of the ER and to the pHdependent retrieval system mediated by the KDEL-receptor (see below). Based on these observations, it seems that cargo selection and membrane fission is more sensitive to a pH change than membrane fusion is, thereby inhibiting or delaying transport between successive secretory compartments., However, due to the pH-sensitivity of viral protein-mediated membrane fusion events with endosomal membranes (White and Whittaker, 2016; Desai et al., 2017), this scenario needs further testing. Indeed, these viral fusion events typically occur via "inside to outside" fusions with organelle membranes and are therefore topologically opposite from fusions that take place between vesicular carriers and their target membranes.

Another well-characterized example of a pH-dependent protein sorting step is the KDEL receptor, which returns escaped ER resident proteins from the cis-Golgi back to the ER. The receptor is a key component of a homeostatic control system that regulates trafficking between the ER and the Golgi compartments and within the Golgi itself (Lewis and Pelham, 1992; Scheel and Pelham, 1996; Cancino et al., 2014). The receptor binds the peptide sequence KDEL (or a similar sequence motif), leading to interaction with two different Golgi-associated heterotrimeric G-proteins, which regulate the transport machineries via phosphorylation (Giannotta et al., 2012; Cancino et al., 2014). Of notice here is that both cargo binding and its release are regulated by the pH gradient between the two organelles (see Brauer et al., 2019, and references therein). In the more acidic environment of the cis-Golgi, the receptor recognizes the motif and binds to it, while at the neutral pH of the ER lumen, it releases the motif and the associated cargo. Moreover, p58/ERGIC-53/LMAN1, a receptor protein involved in the export of soluble glycoproteins from the ER, employs a similar pH gradient for its oligomerization and accessory protein-mediated binding with specific cargo and its release in the low pH-high calcium environment at the ER-Golgi interface (Appenzeller-Herzog et al., 2004; Appenzeller-Herzog and Hauri, 2006).

A third well-known example is the mannose-6-phosphate receptor, which binds lysosomal enzymes carrying the man-6-P tag in the Golgi and releases them in the lower pH environment of the endosomes (Ghosh et al., 2003). In line with this, we showed that, in some cancer cell lines with problems in lysosomal acidification, the ligand-bound receptor cannot unload its ligand in lysosomes and accumulates in endosomal/lysosomal compartments (Kokkonen et al., 2004). This suggested that further lysosomal enzyme cargo sorting at the TGN is impossible and can result in their aberrant secretion into the extracellular space, a phenomenon that is often associated with tumorigenesis and likely helps cancer cells to invade and metastasize to adjacent tissues (Mohamed and Sloane, 2006; Gocheva and Joyce, 2007; Kallunki et al., 2013).

Golgi pH homeostasis is also important for the sorting of apical and basolateral proteins in polarized epithelial cells. Caplan et al. (1987) showed that laminin and heparan sulfate proteoglycan (HSPG) are normally actively sorted to the basolateral surface of polarized canine renal tubule cells (MDCK) in a pH-dependent manner. By increasing the pH of the Golgi and other cellular compartments in MDCK cells with NH4Cl (**Figure 1**, red dots), the authors were able to divert the two abovementioned secretory proteins to both the apical and basolateral transport vesicles, with the outcome that roughly equal amounts were sorted to both surfaces in the treated cells. Since the TGN is the main sorting station for these two surface domains (Guo et al., 2014), it is likely that the cargo recognition and sorting at the TGN may not depend only on specific sorting signals but also on the existence of an environment suitable for their recognition by the sorting machinery in each case. We recently showed that the apical targeting of the CEAMCAM5 (carcinoembryonic antigen, CEA), a well-known follow-up marker for colorectal cancer, is also a pH-sensitive process (Kokkonen et al., 2018). CEA is a typical GPI-anchored apical protein present in gut epithelial cells. For an unknown reason, CEA exhibits a non-polarized distribution in cancer cells, such as in CaCo-2 cells. Guided by the notion that the Golgi resting pH is ∼0.5 pH units higher in CaCo-2 cells than in non-malignant cells, we treated MDCK cells stably expressing CEA with various compounds, including concanamycin A (CMA: a proton pump inhibitor, see **Figure 1**). We showed that, in contrast to drugs affecting the redox state, CMA attenuated apical targeting of CEA without affecting its trafficking to the cell surface. In the presence of the drug, CEA was delivered equally to apical and basolateral domains of MDCK cells due to inhibition of its GPI anchor-mediated association with lipids rafts.

Autosomal recessive Cutis Laxa type II is the first inherited disease identified thus far that is tightly linked to altered Golgi pH homeostasis. The skin of these patients shows excessive wrinkling at an early age. It is caused by mutations in the gene encoding the a2 subunit of the Golgi-localized V-ATPase (ATP6V0A2) (Kornak et al., 2007). However, patients belonging to a closely related disease group suffering from Wrinkly skin syndrome are heterogeneous, in that only some patients carry the same mutation and show no symptoms of elastin deficiency (Morava et al., 2009). Although the Golgi resting pH has not yet been directly measured, it is expected that this V-ATPase defect perturbs Golgi pH homeostasis, because the patients' cells exhibited glycosylation and membrane trafficking defects (Morava et al., 2005; Kornak et al., 2007; Hucthagowder et al., 2009). Further studies are still needed, as Golgi membranes also seem to co-express another subunit, the "a1" isoform, of the V-ATPase (see Kornak et al., 2007). Nevertheless, the "a2" subunit mutations impair retrograde trafficking from the Golgi back to the ER, but here again, the mechanistic details remain unclear. One complicating factor in gaining an understanding of the cutis laxa phenotype is the fact that the "a2" subunit appears to localize also in early endosomes (Hurtado-Lorenzo et al., 2006), suggesting that altered endosomal pH and dysfunction may also contribute to the disease etiology.

Khayat et al. (2019) recently described a new pH homeostasisassociated disease with multigenerational non-syndromic intellectual disability (ID). The disease is caused by missense mutations in the alkali cation/proton exchanger NHE7 (SLC9A7). The variant protein localized correctly in the

TGN/post-Golgi vesicles, but its N-linked glycans were abnormal likely due to less acidic pH of the TGN/post-Golgi compartments in patient's cells. Membrane trafficking, however, was unaffected. These observations are consistent with a role for NHE7 in the regulation of TGN/post-Golgi pH homeostasis and suggest that abnormal Golgi pH homeostasis may be the cause of neurodevelopmental defects associated with this disease.

Other luminal ions, and particularly Ca2+, also contribute to membrane trafficking defects. The depletion of cellular Ca2<sup>+</sup> stores in NRK cells using thapsigargin abolished KDEL receptormediated retrieval of ER chaperones GRP94/endoplasmin and GRP78/BiP, resulting in their appearance in the culture medium (Ying et al., 2002). Accordingly, thapsigargin was found to inhibit Brefeldin A-induced retrograde transport from the Golgi back to the ER in HeLa cells (Ivessa et al., 1995). Calcium depletion also selectively inhibited proteolytic cleavage of pro-somatostatin or proinsulin, without affecting their secretion (Austin and Shields, 1996). It also interfered with the sorting of secretogranin II into immature granules in semi-intact PC12 cells (Carnell and Moore, 1994), even though high Ca2<sup>+</sup> and low pH have been reported to facilitate the concentration of cargo proteins in regulated secretory vesicles (Chanat and Huttner, 1991).

SPCA1 is a Golgi-localized Ca2<sup>+</sup> ATPase that transports both Ca2<sup>+</sup> and Mn2<sup>+</sup> into the Golgi lumen and, therefore, plays an important role in Golgi cation homeostasis (Van Baelen et al., 2004; Missiaen et al., 2007). In humans, allelic mutations of the SPCA1 gene (Vanoevelen et al., 2007; Brini and Carafoli, 2009) are the cause of Hailey–Hailey disease, in which patients' keratinocytes exhibit an increased cytosolic Ca2<sup>+</sup> concentration, and defects in protein sorting and Ca2<sup>+</sup> signaling (Missiaen et al., 2004; Ramos-Castaneda et al., 2005; Vanoevelen et al., 2007). Lowered levels of Ca2<sup>+</sup> and Mn2<sup>+</sup> cations in the Golgi lumen in patients' cells lead to defects in protein folding, trafficking and sorting or proteolytic cleavage of prohormones (Missiaen et al., 2004; Grice et al., 2010). These defects could explain why the affected patients are unable to maintain structurally intact desmosomes and epidermis. The fact that Mn2<sup>+</sup> is an important cofactor for many glycosyltransferases (Kaufman et al., 1994) suggests that glycosylation is altered in affected cells and may also contribute to the disease etiology.

Golgi Ca2<sup>+</sup> (and Mn2+) homeostasis is also dependent on cellular oxygen levels. A good example of this is the fact that intermittent hypoxia upregulates the expression of both SPCAs in HCT116 cells (Jenkins et al., 2016), suggesting that Ca2<sup>+</sup> and/or Mn2<sup>+</sup> transport from the cytosol to the Golgi lumen via SPCAs likely increases in hypoxic cells. Oxygen also regulates nitrogen oxide (NO) levels in the Golgi by modulating eNOS activity and thus, NO production, thereby locally enhancing S-nitrosylation of Golgi proteins, especially of the N-ethylmaleimide-sensitive factor (NSF) (Iwakiri et al., 2006). Since NSF is involved in membrane fusion events, this modification delays protein transport from the ER to the plasma membrane and, thus, can at least partially explain why hypoxia inhibits ER-Golgi vesicular trafficking. On the other hand, compounds that can scavenge NO (such as c-PTIO, N-acetylcysteine and hemoglobin) induced Golgi fragmentation (Lee et al., 2011, 2013), which was accompanied by the depletion of α-soluble NSF acceptor protein (α-SNAP) from Golgi membranes, in accordance with the observed delay in ER-Golgi trafficking.

# Golgi pH Homeostasis and Glycosylation Defects

Glycosylation is likely the most pH-sensitive process of the Golgi functions. For example, monensin was shown to prevent processing of Uukuniemi viral G proteins into endo-H-resistant and under-sialylated species without affecting membrane trafficking (Kuismanen et al., 1985). Campbell et al. (2001) in turn were able induce the expression of oncofetal Thomsen-Friedenreich (TF- or T-) antigen in LS174T goblet-differentiated cells by increasing Golgi pH with bafilomycin A and monensin. Axelsson et al. (2001) were the first to provide a mechanistic link for these pH-induced glycosylation changes by using prolonged NH4Cl treatment in HeLa and LS 174T cells. They showed that inhibition of O-glycan synthesis by NH4Cl was accompanied by mislocalization of N-acetylgalactosaminyltransferase 2, b-1,2-Nacetylglucosaminyltransferase I and b-1,4-galactosyltransferase 1 into endosomal compartments, while the drug had no effect on Golgi morphology. However, because most of the enzymes that elongate O-glycans were not addressed in the study, it remains unclear whether enzyme re-localization is solely responsible for the observed glycosylation defect(s). Later, by using increasing concentrations of chloroquine, we (Rivinoja et al., 2006) demonstrated that only a 0.2 pH unit increase in Golgi luminal pH is needed to interfere with mucin type O-glycosylation and terminal a-2,3-sialylation of N-linked glycans without causing any changes to overall Golgi morphology. The latter defect correlated well with the observed mislocalization of the relevant sialyltransferase (ST3Gal-III) into endosomal compartments, while no such redistribution was observed with ST6Gal-I (or B4GalT-I), i.e., the enzyme that adds sialic acid to terminal galactose residue via an a-2,6-linkage. Cutis laxa type II patients also display defects in sialylation of both N-linked and O-linked glycans (Morava et al., 2005; Wopereis et al., 2005; Kornak et al., 2007). These observations indicate that glycosylation in general is highly sensitive to changes in Golgi luminal pH and, if altered, can be due to mislocalization of a selected set of glycosyltransferases. Other known causes are changes in the expression levels of the enzymes, yet these do not strictly correlate with glycan profiles displayed by the cells.

Most glycosyltransferases show an intrinsic tendency to form oligomeric complexes with each other. Typically, such complexes include enzymes that successively add sugar residues to a glycan chain (Kellokumpu et al., 2016). All enzymes also form homomers in the ER (Hassinen and Kellokumpu, 2014), perhaps facilitating their folding or transport to the Golgi, or both. On the other hand, enzyme heteromers only form after the enzymes arrive in the Golgi compartment. This switch from enzyme homomers to enzyme heteromers is dependent on the pH gradient or Golgi redox homeostasis (see section "Golgi Redox Homeostasis and Altered Glycosylation" last paragraph) between the ER and the Golgi. Thus, heteromer formation of N-glycan processing enzymes B4GalT-I and ST3Gal-III, and of enzymes

that synthesize mucin type O-glycan core structures (ppGalNacT-6, C1GalT-1, C2/3GNT), is prevented by increasing Golgi pH with chloroquine (Hassinen et al., 2011). Intriguingly, the same enzymes (except B4GalT-I) were also found to mislocalize in chloroquine treated cells (unpublished observations), suggesting that heteromer formation may contribute to their retention or retrieval in the Golgi membranes. However, in other cases such as B4GalT-I and ST6Gal-I, heteromer formation was not affected by an increase in Golgi luminal pH (Hassinen et al., 2011), but rather by an altered Golgi redox state (Hassinen et al., 2019; see also Section "Golgi Redox Homeostasis and Altered Glycosylation" last paragraph). These observations point to fundamental differences in the way enzyme heteromers form in the Golgi lumen, and perhaps reflecting the high specificity of the interactions needed to prevent irrational interactions that could otherwise lead to the synthesis of mixed or irrelevant glycan structures (Kellokumpu et al., 2016). The high specificity for the interactions could also explain a failure to identify any consensus Golgi retention motif(s) in Golgi enzymes, except those needed for their retrieval from later compartments to earlier ones via GOLPH3-mediated binding to the COPI complex (Rabouille and Klumperman, 2005; Tu et al., 2008; Sechi et al., 2015; Liu et al., 2018). Whether such Golgi retention motifs involving relevant enzyme interactions in different Golgi subcompartments indeed exist remains to be tested. However, their existence is supported by the pH-dependent mislocalization of a set of glycosyltransferases (Rivinoja et al., 2009; unpublished observations); otherwise, it would be difficult to understand how luminal alkalization can interfere with the recognition of retrieval motifs on the cytosolic side of the Golgi membranes. In addition, oligomerization that inherently involves enzyme interactions, has been considered to be important for the retention of resident glycosyltransferases in the Golgi (Nilsson et al., 2009).

The loss of enzyme heteromers is generally accompanied by changes in O- and N-linked glycosylation, but likely concerns other glycosylation pathways as well. One reason for these changes is that heteromer formation significantly increases the activity of the complex constituents (Kellokumpu et al., 2016). For example, both enzyme activities of B4GalT-I/ST6Gal-I heteromers were 2.5-fold higher than their respective homodimers (Hassinen et al., 2011). How this activation is achieved is currently unclear but may involve substrate channeling or conformational changes brought about by the interaction. In other words, the formation of the enzyme heteromers from enzyme homomers in the Golgi could simply serve to keep enzymes silent until they arrive in the Golgi. Such a system would increase both the speed and fidelity of glycan synthesis, as it would also prevent the intervention of competing enzymes that can use the same acceptor sugar as a substrate. This view is in line with the pH-independent α-2,6-sialylation and formation of B4GalT-I/ST6Gal-I heteromers (in contrast to α-2,3-sialylation and the formation of ST3Gal-III/B4GalT-I heteromers; see Rivinoja et al., 2009; Hassinen et al., 2011). The difference in pH-dependency can also explain why the carcinoembryonic antigen (CEACAM5) extracted from colon cancer tissue carries α-2,6-linked sialic acid instead of the α-2,3-linked sialic acid found in normal tissues (Yamashita et al., 1987; Kobata et al., 1995). The extent to which the loss of enzyme heteromers contributes to glycosylation remains to be determined when enzyme interaction mutants become available.

#### Cancer-Associated Glycosylation Changes

Altered glycosylation is one of the hallmarks of cancers. Such alterations can involve changes in the elongation of O-glycans, the branching of N-glycans, sulfation, O-acetylation of sialic acid, fucosylation and the expression of blood group antigens (Jass et al., 1994; Kuhns et al., 1995; Capon et al., 1997; Taylor-Papadimitriou et al., 1999; Hakomori, 2002; Roth, 2002; Lau and Dennis, 2008; Ungar, 2009; Reis et al., 2010; Radhakrishnan et al., 2014; Vajaria and Patel, 2017; Rodrigues and Macauley, 2018). Some of these changes are used as cancer markers, while others also have verified roles in promoting tumorigenesis (Hakomori, 1991, 1996; Ono and Hakomori, 2004; Peixoto et al., 2016). Based on existing data, several factors that can cooperatively contribute to the above cancer-associated glycosylation changes have been put forward. These include the altered expression of glycosyltransferases or nucleotide sugar transporter genes (Yang et al., 1994; Brockhausen et al., 1995; Hanisch et al., 1996; Lloyd et al., 1996; Kumamoto et al., 2001), and perhaps also a loss of activity of Cosmc, a specific molecular chaperone needed for the folding and catalytic activation of C1GalT-I (Schietinger et al., 2006). The C1GalT-I enzyme normally adds galactose to the Tnantigen (GalNAc-Ser), forming a mucin type O-glycan core 1 structure (the T-antigen). Thus, the loss of its activity may result in an increased expression of the Tn-antigen in cancer cells.

Cancer-associated glycosylation changes can also result from a more general defect related to altered Golgi ion or pH homeostasis. The first observations suggesting this came from studies in which the treatment of cells with pH gradient dissipating drugs increased the expression of cancer-associated Tn- and T-antigens (Thorens and Vassalli, 1986; Gawlitzek et al., 2000; Axelsson et al., 2001; Campbell et al., 2001; Kellokumpu et al., 2002). At high concentrations, these same compounds also induced Golgi fragmentation typically seen also in cancer cells (Kellokumpu et al., 2002). Direct Golgi pH measurements with fluorescent probes in breast and colorectal cancer cells (MCF-7, HT-29, SW-48) showed that the Golgi resting pH is indeed more alkaline (∼0.2–0.4 pH units) than that of non-malignant cells (Rivinoja et al., 2006). These early observations strongly supported the view that abnormally high Golgi resting pH is responsible for the increased expression of cancer-associated glycan antigens.

An important question at the time was why abnormal Golgi pH is detrimental to glycosylation. Although one still cannot exclude possible effects on the synthesis or transport of nucleotide sugars, we believe that either the loss of the O-glycosyltransferase heteromers or enzyme mislocalization, or both, are the two main reasons for the pH-dependent glycosylation changes seen in cancer cells. These two factors might in fact also be interlinked, given that oligomerization appears to be important for Golgi retention (Nilsson et al., 1993, 1994, 2009). Thus, at elevated Golgi resting pH, the enzymes responsible for synthetizing the

O-glycan core structure are unable to form heteromers (Hassinen et al., 2011). Their loss then abrogates their retention in the Golgi, whereby they mislocalize to endosomal compartments and are therefore unable to elongate the core GalNAc residue with other sugar residues in the Golgi. In support of this view, O-glycosyltransferases seem to have altered distribution in cancer cells in vivo (Egea et al., 1993). Similar relocalization of the initiating ppGalNAcT-1/2 to the ER was seen after growth factor-induced activation of the src kinase, or by transfecting cells with constitutively active src (Gill et al., 2010, 2011). This relocalization was linked to a COP-I-dependent trafficking event, as a dominant-negative Arf1 isoform, Arf1(Q71L), blocked ppGalNacT redistribution. Note however, that the pH-induced relocalization of the enzyme involved transport to the endosomal compartments via bulk flow (Rivinoja et al., 2009), suggesting that pH-induced relocalization of the enzymes is associated with impaired Golgi retention, rather than with activated transport from the Golgi to the ER.

Another issue related to organelle acidification defects in cells is its association with the multidrug resistance (MDR). In certain MDR cancer cell lines, chemotherapeutic drugs (often weak bases) become protonated and sequestered in acidic organelles (Schindler et al., 1996; Altan et al., 1998). Sequestration in resistant cells allows the removal of cytotoxic drugs from the cytoplasm via secretory and recycling pathways. In contrast, drug-sensitive cells were shown to have defects in organelle acidification, whereby a similar sequestration of the drugs does not occur, exposing the cells to high concentrations of the drugs. However, these acidification defects are not universal in all MDR cells, suggesting that other mechanisms for MDR exist (Simon, 1999).

#### Golgi Redox Homeostasis and Altered Glycosylation

Reactive oxygen species and hypoxia (low oxygen environment) are key modulators of the cellular redox state (Adler et al., 1999). ROS and hypoxia also modulate Golgi-associated vesicular trafficking, protein sorting and glycosylation events (Regoeczi et al., 1991; Koike et al., 2004; Yin et al., 2006; Shirato et al., 2011; Ermini et al., 2013; Lehnus et al., 2013; Belo et al., 2015). Most often, this is thought to be mediated mainly by hypoxia-inducible factors (HIF-1-3) that regulate the expression of hundreds of genes, including a variety of proteins involved in glycosylation. Specifically, hypoxia has been shown to down- or up-regulate enzymes that synthetize nucleotide sugars in the cytoplasm, Golgi-localized glycosyltransferases (Mgat2, Mgat-3 and Mgat5 and 5b, fucosyltransferases 1, 2 and 7, sialyltransferases ST3Gal-I and ST6Gal-1) and transporters of UDP-Galactose, CMP-sialic acid (Sialin) and UDP-N-acetylglycosamine (UGT1) (Koike et al., 2004; Shirato et al., 2011; Belo et al., 2015; Taniguchi et al., 2016). Some of these are involved in the synthesis of cancerassociated sialyl Lewis A/X carbohydrate epitopes typically found in selectins on O-linked glycans and glycolipids (Kumamoto et al., 2001; Shirato et al., 2011). HIF-1α in the Pa-Tu-8988S and Pa-Tu-8988T pancreatic cancer cell lines have also been shown to suppress the expression of the UDP-glucuronosyltransferase (Kato et al., 2016), cytosolic O-GlcNAc transferase (OGT) (Liu et al., 2014) and glucosylceramide synthase (GCS) (Zhao et al., 2003). Finally, Jenkins et al. (2016) showed that expression of the SPCA2 Ca2<sup>+</sup> pump in HCT116 colon cancer cells was upregulated by hypoxia, and by reactive oxygen and nitrogen species. The authors suggested that this upregulation is associated with Mn2+-dependent cell cycle arrest, but whether these changes relate to increased Ca2<sup>+</sup> or Mn2<sup>+</sup> transport to the Golgi lumen, Ca2+-mediated protein sorting, glycosylation, or detoxification from excess Mn2+, remains unclear.

Based on the above observations, Taniguchi et al. introduced the term "Glyco-redox" to link altered glycosylation with oxidative stress generated by hypoxia or ROS (Taniguchi et al., 2016), and to emphasize their close association with Parkinson disease, Alzheimer's disease, amyotrophic lateral sclerosis (ALS), and chronic obstructive pulmonary disease (COPD). These changes may also partly involve cleavage of cell surface glycosaminoglycans and N-linked glycans, thereby affecting interactions of cells with the extracellular matrix (Eguchi et al., 2002; Eguchi et al., 2005), as has been observed using hypoxia-mimicking agents such as CoCl<sup>2</sup> (Taniguchi et al., 2016). Oxidative stress also provides a link between altered glycosylation, high-fat diet and the onset of type II diabetes (Ohtsubo and Marth, 2006; Ohtsubo, 2010; Ohtsubo et al., 2011). High levels of free fatty acids were shown to inhibit the activity of two transcription factors (Foxa2 and Hnf1a) that normally positively regulate the expression of Mgat4a, a glycosyltransferase needed for β1,4-GlcNAc branching of N-glycans. This modification is needed for insulin-stimulated transport of the GLUT-4 transporter to the cell surface, whereby it is able to bind and import glucose for further use by the pancreatic β-cells. In the absence of β1,4-GlcNAc branching, the GLUT-4 transporter remains intracellular, leading to an accumulation of glucose in the blood.

We have recently shown that hypoxia also modulates glycosylation in a HIF-independent manner by reducing Golgi redox state (Hassinen et al., 2019). Specifically, we demonstrated that even moderate hypoxia (5% O2) lowers the Golgi oxidizing potential to the level found in the ER of normoxic cells, consistent with possible problems in disulfide bond formation. Based on lectin microarray glycan profiling, this decrease was accompanied by an attenuated sialylation of N-glycans, an elongation of O-linked glycans, the loss of pH-independent interaction between the B4GalT-I and ST6Gal-I, and the loss of ST6Gal-I activity in hypoxic cells. These findings can explain the reduced a-2,6-sialylation in hypoxic cells. Given that sialylation was not the only change in glycosylation, we expect that other enzymes are similarly affected by a lowered Golgi oxidizing potential.

#### CONCLUDING REMARKS

The above examples highlight the critical roles of Golgi pH, ion and redox homeostasis in the maintenance of Golgi functions and its unique architecture. The existing data also emphasizes that Golgi acidity, high cation concentrations and redox state are not

just background components in the Golgi, but rather, they emerge as the key players in orchestrating various membrane trafficking events, keeping the Golgi resident enzymes correctly localized and active, and facilitating their cooperative interactions in glycosylation. In addition, they are important for cargo selection by post-Golgi vesicular carriers and for protein sorting to the apical surface in polarized epithelial cells. Further support of their importance is provided by several human diseases, including Cutis Laxa, Hailey–Hailey disease and congenital disorder of glycosylation 2K (CDG2K), which all are caused by altered ion homeostasis in the Golgi lumen. Other Golgi homeostasisassociated diseases are certainly waiting to be identified.

Although many mechanistic details remain incompletely understood, the current evidence indicates that each of the Golgi pH, ion or redox (oxygen) regulatory systems can perturb all Golgi functions simultaneously or may specifically impair only one of them. For example, a small increase in Golgi luminal pH (0.2 pH units) will perturb glycosylation with no detectable effect on other Golgi functions. A failure to maintain Golgi oxidative potential causes defects in both membrane trafficking and glycosylation, with no detectable changes in enzyme localization in the Golgi. High Ca2<sup>+</sup> concentrations seem to contribute to membrane trafficking and protein sorting events at the Golgi, but also to glycosylation, since many enzymes need Mn2<sup>+</sup> ions to remain catalytically active.

The above examples highlight the complexity and mutual interplay of the regulatory systems needed to establish and maintain Golgi homeostasis and the many different cellular phenotypes encountered by manipulating homeostatic

#### REFERENCES


machineries with drugs or mutations. For example, it is well known that perturbed membrane trafficking often results in changes in Golgi architecture and glycosylation. The interdependence of these phenomena is probably the biggest obstacle for better understanding the role of each of these regulatory systems in the Golgi. Nevertheless, such studies are needed to provide us with new insights into how the various Golgi tasks are executed and to identify the molecular machineries that act in the background to keep these tasks ongoing. These studies will eventually unveil how the Golgi compartment functions as an organelle and what purpose(s) its unique architecture with stacked and flattened cisternae actually serves for.

#### DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

#### AUTHOR CONTRIBUTIONS

SK wrote the text and made the figures.

### FUNDING

This work was financially supported by the Academy of Finland.


and sorting of secretogranin II. J. Cell Biol. 127, 693–705. doi: 10.1083/jcb.127. 3.693



aberrant glycosylation. Hum. Mol. Genet. 28, 598–614. doi: 10.1093/hmg/ ddy371



the glycosylation defects. J. Clin. Endocrinol. Metab. 102, 1375–1386. doi: 10. 1210/jc.2016-3443



lessons from research on glutathione, reactive oxygen and nitrogen species to glycobiology. Arch. Biochem. Biophys. 595, 72–80. doi: 10.1016/j.abb.2015. 11.024


in intracellular Ca2+ storage. Eur. J. Cell Biol. 819, 469–483. doi: 10.1078/0171- 9335-00266


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

Copyright © 2019 Kellokumpu. 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.

# Acylation – A New Means to Control Traffic Through the Golgi

Andreas M. Ernst<sup>1</sup> \*, Derek Toomre<sup>1</sup> and Jonathan S. Bogan1,2

<sup>1</sup> Department of Cell Biology, Yale School of Medicine, Yale University, New Haven, CT, United States, <sup>2</sup> Section of Endocrinology and Metabolism, Department of Internal Medicine, Yale School of Medicine, Yale University, New Haven, CT, United States

The Golgi is well known to act as center for modification and sorting of proteins for secretion and delivery to other organelles. A key sorting step occurs at the trans-Golgi network and is mediated by protein adapters. However, recent data indicate that sorting also occurs much earlier, at the cis-Golgi, and uses lipid acylation as a novel means to regulate anterograde flux. Here, we examine an emerging role of S-palmitoylation/acylation as a mechanism to regulate anterograde routing. We discuss the critical Golgi-localized DHHC S-palmitoyltransferase enzymes that orchestrate this lipid modification, as well as their diverse protein clients (e.g., MAP6, SNAP25, CSP, LAT, β-adrenergic receptors, GABA receptors, and GLUT4 glucose transporters). Critically, for integral membrane proteins, S-acylation can act as new a "self-sorting" signal to concentrate these cargoes in rims of Golgi cisternae, and to promote their rapid traffic through the Golgi or, potentially, to bypass the Golgi. We discuss this mechanism and examine its potential relevance to human physiology and disease, including diabetes and neurodegenerative diseases.

#### Edited by:

Yanzhuang Wang, University of Michigan, United States

#### Reviewed by:

Luke Chamberlain, University of Strathclyde, United Kingdom Nicholas Davis, Wayne State University, United States Paul Anthony Gleeson, University of Melbourne, Australia

\*Correspondence:

Andreas M. Ernst andreas.ernst@yale.edu

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology Received: 10 April 2019 Accepted: 29 May 2019

Published: 12 June 2019

#### Citation:

Ernst AM, Toomre D and Bogan JS (2019) Acylation – A New Means to Control Traffic Through the Golgi. Front. Cell Dev. Biol. 7:109. doi: 10.3389/fcell.2019.00109 Keywords: Golgi, palmitoylation, acylation, anterograde transport, Golgi bypass, membrane traffic

# INTRODUCTION

A major function of the Golgi is to receive, sort and modify proteins and lipids and to deliver these cargoes to new locations [e.g., the plasma membrane (PM), endo-lysosome, or back to endoplasmic reticulum (ER)]. Impaired coordination of this highly orchestrated assembly line can result in abnormal glycosylation (Climer et al., 2015; Fisher and Ungar, 2016) or a virtual "traffic jam." In neurons, Golgi dysfunction is associated with numerous diseases such as Parkinson's, Alzheimer's, and Huntington's diseases, as well as with cognitive disorders (Glick and Nakano, 2009; Bexiga and Simpson, 2013; Rabouille and Haase, 2015). Altered Golgi function in other cell types may also contribute to disease. Yet, how the mammalian Golgi sorts proteins and lipids remains poorly understood.

# ORGANIZATION OF THE MAMMALIAN GOLGI

In mammalian cells the Golgi looks like a perinuclear crescent or "ribbon" by light microscopy and as a "stack" of pancake-like cisternae by three-dimensional electron microscopy (Ladinsky et al., 1999; Huang and Wang, 2017; Martinez-Martinez et al., 2017), with the cis side closest to the ER. The cisternae are stacked by GRASP proteins (Zhang and Wang, 2015) and individual

Golgi "mini"-stacks are laterally linked to form an extensive Golgi ribbon (Wei and Seemann, 2017; Makhoul et al., 2018). The crescent-shaped ribbon is often several micrometers long and highly convoluted. It contains, on average, five cisternae with a thickness of 50 nm each, which are separated by intercisternal spaces of 5–15 nm (Ladinsky et al., 1999). This organization of the Golgi differs drastically from that found in lower eukaryotes: yeast Golgi is neither stacked nor linked and within the cisternae, the Golgi enzymes exchange dynamically to cause maturation of the enclosed cargo (Losev et al., 2006; Glick and Nakano, 2009). These differences in Golgi structure underscore one of the biggest and most debated riddles in cell biology: How does the Golgi function to sort anterograde traffic?

Whereas for retrograde traffic there is a clear consensus that COPI vesicles mediate retrograde flux, for anterograde traffic multiple models have been proposed, including those of vesicular transport, cisternal maturation, and others (Patterson et al., 2008; Jackson, 2009; Pfeffer, 2010; Glick and Luini, 2011). Central to this longstanding debate is the question of whether cisternae within the mammalian Golgi ribbon mature (i.e., dynamically exchange their enzymes), or whether the enzyme composition of a given cisterna is stable, with cargo passing through static successive layers via vesicular/tubular carriers (Glick and Luini, 2011). Two longstanding challenges in addressing this question are that (1) current imaging cannot visualize live Golgi dynamics at sufficient resolution to unambiguously distinguish between these models (which may require live cell imaging at tens of nanometers in 3D), and (2) the machinery used for anterograde traffic is debated (e.g., Do COPI vesicles carry only retrograde traffic or do they act in both directions?). Arguably, the only current consensus is that there is no consensus. But might there also be other means to drive cargo forward?

## A CIS-FACE PROBLEM: HOW TO DISTILL ANTEROGRADE CARGO FROM RETROGRADE PROTEINS AND LIPIDS?

A remarkable requirement of Golgi function upstream of intra-Golgi transport lies in the segregation of anterograde from retrograde cargo and lipids (Glick and Nakano, 2009). Indeed, one of the largest remaining questions in Golgi biology is exactly how sorting of anterograde from retrograde cargo (ER resident proteins, lipids) is achieved upon Golgi entry. While glycosylation is a highly coordinated and critical function of the Golgi, there is little evidence that it is a driving force for sorting (Gomez-Navarro and Miller, 2016). But how much sorting is there? This has been addressed by comparing the total amount of anterograde membrane that leaves the ER via COPII vesicles to the amount of new membrane actually required to sustain cell growth (Barlowe and Helenius, 2016; Bottanelli et al., 2017). Notably, about 90% of the membrane has to be recycled to the ER and, in light of this large retrograde backflow, the incoming cargo has to be extensively concentrated to achieve an efficient rate of anterograde net movement.

The question remains as to whether anterograde sorting of cargo at the Golgi is an active process (i.e., mediated by unknown signals), or whether a net anterograde flux is achieved simply by the selective retrograde retrieval of lipids and proteins upon entry into the Golgi. More simply, is active sorting required to move cargo forward, or not? More is understood about the retrograde traffic machinery, which has been extensively characterized and appears to consist mainly of COPI vesicles and tubular Rab6- connections. Both act in conjunction with the selective retrieval of retrograde cargo and depend on basic amino acids that serve as concentration and retrieval signals (White et al., 1999; Duden, 2003; Barlowe and Helenius, 2016). How sorting of anterograde cargo is achieved at the cis face of the Golgi is much less clear. One obvious concept is that there may be an anterograde amino acid sorting signal. Here, it was recently proposed that acidic residues on the cytoplasmic tail of a model transmembrane cargo, vesicular stomatitis virus glycoprotein (VSV-G) could promote anterograde routing at the Golgi (Fossati et al., 2014). Nevertheless, the lack of such residues in the cytoplasmic tails of multiple viral spike proteins (HA, NA, and GP), as well as mammalian integral PM resident proteins, suggests that such signals do not act generally in anterograde sorting, but rather are specific to VSV-G at COPII exit sites in the ER (Votsmeier and Gallwitz, 2001).

# The Cis-Golgi Is a Hot Spot for Protein Palmitoylation

Hints that some type of lipid-based anterograde sorting signal might exist appeared in the 1980s, when it was shown that efficient intra-Golgi cargo transport requires a specific type of activated lipid, palmitoyl-CoA (Glick and Rothman, 1987). Furthermore, it was discovered that the hydrolysis and transfer of palmitoyl-CoA is necessary for both vesicle budding and fission, and that its presence is needed in donor Golgi membranes (Pfanner et al., 1989; Ostermann et al., 1993). These data positioned the acylation reaction at the cis-Golgi. However, despite uncovering a potential role of palmitoyl-CoA in anterograde Golgi sorting, the underlying molecular mechanism could not be elucidated.

# PALMITOYLATION – LIPIDATION OF PROTEINS AS A MOLECULAR SWITCH?

In recent years, palmitoyl-CoA has increasingly been recognized for its importance in the fields of developmental and cell biology, in microbiology, and in neuroscience (El-Husseini Ael et al., 2002; Linder and Deschenes, 2007; Fukata and Fukata, 2010; Demers et al., 2014). In addition to its role in the synthesis of membrane lipids (Blom et al., 2011), palmitate is attached covalently and post-translationally to several hundreds of proteins (Blanc et al., 2015). These protein lipidation (or fatty acylation) reactions can occur on three types of different amino acids: (1) on cysteine residues (forming thioester linkages, S-palmitoylation), (2) on serine/threonine residues (forming oxyester linkages, O-palmitoylation), or (3) on primary amino groups (forming amide linkages, N-palmitoylation) (Jiang et al., 2018). Among these distinct types of acylation, S-palmitoylation stands out, as the thioester-bond is easily reversible and can

thus act as a two-way toggle. Further, acylation has an inherent biophysical propensity to dynamically alter the properties of the modified protein, including the oligomerization state within a membrane or the targeting of soluble proteins to membranes (Smotrys and Linder, 2004). As such, S-acylation is well-positioned to function as a molecular and biophysical switch. For clarity, while the fatty acid S-acylation is typically palmitate, other fatty acids such as stearate (C18) and unsaturated variants can be used.

### DHHC S-PALMITOYLTRANSFERASES ARE FOUND IN KEY POSITIONS ACROSS THE SECRETORY PATHWAY

In mammals, the enzymes responsible for protein S-acylation are a family of 23 proteins, termed protein-acyl transferases or PATs. These integral membrane proteins typically contain four to six transmembrane domains and a cysteine-rich domain (CRD) within a cytoplasmic loop. A conserved Asp-His-His-Cys (DHHC) tetrapeptide motif is located within the CRD and is the active site of each "DHHC" enzyme (Fukata and Fukata, 2010; Greaves and Chamberlain, 2011; De and Sadhukhan, 2018; Rana et al., 2019). The fact that the active site is positioned at the cytoplasmic face dictates that unlike N-palmitoylation, which takes place in the lumen of the Golgi (Buglino and Resh, 2008; Jiang et al., 2018), S-acylation occurs exclusively in the cytoplasm. Acylation of a substrate protein is thought to be mediated by a two-stage mechanism: (1) auto-acylation of a single or multiple Cys residue(s) within the CRD of the DHHC enzyme by Acyl-CoA to form an enzyme-acyl thioester moiety, and (2) the Cys residue from a substrate protein attacks the enzyme-acyl intermediate to form a substrate-acyl product (Rana et al., 2019). Notably, integral membrane proteins frequently have Cys residues immediately adjacent to their transmembrane anchors (Aicart-Ramos et al., 2011), and recent crystallographic data position the active site of the PAT near the cytoplasmic leaflet (Rana et al., 2018).

Unlike most other posttranslational modifications, S-acylation does not seem to require a consensus motif. Rather, it appears that the juxtamembrane positioning of the substrate Cys residue close to the PAT active site is required for acylation of the substrate (Rodenburg et al., 2017). How such a spatial positioning is achieved to promote the acylation of soluble proteins is even less clear.

A separate set of enzymes, called acyl-protein thioesterases (APTs), mediates the de-acylation of S-palmitoylated proteins (Vartak et al., 2014; Lin and Conibear, 2015; Won et al., 2018). Interestingly, APTs are soluble proteins that are recruited to Golgi membranes, in part by S-palmitoylation, and that are further able to undergo auto-depalmitoylation. Recent work shows that one isoform, APT1, resides on mitochondria and that the S-palmitoylation of mitochondrial proteins is dynamically regulated (Kathayat et al., 2018). Thus in distinct cellular membranes, acylated cargo proteins are positioned in vicinity to enzymes that remove this modification, so that the abundance of acylated proteins may be maintained under homeostatic control (Vartak et al., 2014).

Given that there are 23 DHHC PAT isoforms in mammals, several important questions naturally arise. Where are these acylation/de-acylation machineries localized? Do PATs exhibit differential expression patterns across different tissues? Are they present in different stages of development or stages of cell proliferation? Ohno et al. set out to address these points and examined both the subcellular location of overexpressed PAT isoforms as well as their expression levels in different tissues (Ohno et al., 2006). Strikingly, the majority of DHHC isoforms appeared to localize exclusively or partially to the Golgi, while others exhibited a distinct localization to the ER, the PM, or to vesicular-tubular structures. Furthermore, a tissue-specific expression pattern was observed for multiple isoforms, e.g., DHHCs 11, 19, and 20 appear exclusively in the testis, while many were ubiquitously expressed and only appear to be absent in distinct tissues.

## S-Palmitoylation Is a Committed Step for Anterograde Transport at the Cis-Golgi

A recurring theme in the literature is the correlation between S-palmitoylation and the affinity of membrane proteins for sphingolipid/cholesterol-containing membranes (Levental et al., 2010; Salaun et al., 2010; Ernst R. et al., 2018). The observations that DHHC isoforms localize to distinct positions within the secretory pathway and are present in distinct tissues at specific levels suggest that S-palmitoylation serves a purpose beyond that of targeting proteins to cholesterol-rich microdomains. However, the strong concentration of DHHC isoforms in the Golgi could be interpreted as a requirement of protein lipidation to achieve compatibility with the complex (and cholesterol-sphingolipid rich) membranes encountered at the trans-Golgi network and beyond. A gradient of sphingolipids and cholesterol exists within the Golgi, with the trans-Golgi/TGN exhibiting the highest concentrations of both lipid classes (Van Meer et al., 2008). If S-palmitoylation serve the purpose of sorting anterograde cargo to "raft-like" membranes, this is presumably where DHHC enzymes should be localized.

#### CONCENTRATION OF DHHC ISOFORMS IN THE CIS-GOLGI

Recently, the intra-Golgi localization of DHHC PATs was investigated in a quantitative manner using exogenously expressed tagged constructs (Ernst A.M. et al., 2018). 17 out of 23 DHHC PATs exhibited significant overlap with endogenous Golgi markers, and 9 localized to the Golgi exclusively: DHHCs 3, 7, 9, 11, 13, 15, 17, 21, and 22. Strikingly, 6 out of 9 DHHCs exhibited strong co-localization with endogenous cis-Golgi markers (DHHCs 3, 7, 13, 17, 21, and 22), while the remaining 3 were positioned at the trans-Golgi (DHHCs 9, 11, and 15). Interestingly, all cis-Golgi-localized DHHCs are expressed in most tissues, while those detected at the trans-Golgi are highly tissue-specific (Ohno et al., 2006). If sorting into "raft-type" microdomains might be a main function of protein

S-palmitoylation, then the observations prompt the question of why the entire ubiquitously expressed pool of Golgi DHHC PATs is found in the cis [a subcompartment with very low cholesterol and sphingolipid contents (Jackson et al., 2016)]. Notably, cell-free system studies of minimal components needed for anterograde Golgi transport demonstrated that palmitoyl-CoA in cis Golgi donor membranes greatly facilitated anterograde cargo transport and budding (Glick and Rothman, 1987; Pfanner et al., 1989; Ostermann et al., 1993). Together, this suggests that palmitoylation may do more than only sorting proteins to "raft" membrane domains.

### S-PALMITOYLATION INDUCES ANTEROGRADE SORTING OF MEMBRANE CARGO

Independently, Ernst A.M. et al. (2018) employed a clickable analog of palmitate, alkyne-palmitate, to identify in mammalian cells the major site of S-acylation activity. Pulse-chase based metabolic labeling revealed a rapid and specific incorporation of palmitate into the cis-Golgi. This incorporation depended on activation of the probe with CoA and resulted in thioester linkages to proteins within Golgi membranes other than the aforementioned DHHC PATs; specificity was validated by Triacsin C (a competitive inhibitor of Acyl CoA synthetase) and hydroxylamine-sensitivity of the palmitate labeling in situ and of proteins on SDS-PAGE (neutral hydroxylamine cleaves thioester linkages present of S-acylated proteins, but not oxyester linkages formed from incorporation of palmitate into lipids). In pulse-chase experiments, the S-palmitoylated proteins partitioned over time from the cis- to the trans-Golgi, in a strictly anterograde fashion, with no signal appearing in the ER. The cargo then appeared at the PM, but this did not occur if DHHC enzyme activity was impaired. To identify the cis-Golgi PAT responsible for the apparent anterograde cargo routing, candidate Golgi-localized PATs (DHHCs 3, 7, 9, 11, 13, 15, 17, 21, 22) were overexpressed and probed for catalysis of anterograde routing of S-palmitoylated proteins. Only the closely related DHHCs 3 and 7 were capable of catalyzing the rate and extent of anterograde transport of bulk S-acylated proteins. Concordantly, the model S-acylated substrates VSV-G and transferrin receptor were probed for a differential partitioning through the Golgi as a function of their acylation status (importantly, both cargoes are classical "non-raft" markers, ruling out a sphingolipid/cholesterol-dependency of the sorting). Strikingly, while the rate of entry into the Golgi was identical for wildtype and mutant cargoes, transport through the Golgi was slowed when these cargo proteins were not acylated, strongly suggesting that S-acylation represents a sorting event, routing them efficiently along an anterograde track. These data suggest that the biochemical requirement for palmitoyl-CoA for in vitro reconstitution trafficking assay detected in reports 30 years prior (Glick and Rothman, 1987; Pfanner et al., 1989) stemmed from S-palmitoylation of the model cargo VSV-G employed in the cell-free system, and that no other unknown cofactors were involved in modulating partitioning of VSV-G form donor to acceptor Golgi membranes. In search for an explanation of how S-palmitoylation modulated the anterograde routing of cargo, alkyne-palmitate-based metabolic labeling was combined with electron tomography. In agreement with an earlier observation that indicated VSV-G preferentially accumulated at the cisternal rims (Orci et al., 1997), the authors found that bulk S-palmitoylated proteins are indeed strongly enriched in the highly curved perimeters of the cisternal rim, which consists of tubules and fenestrated sheet-like elements (Ladinsky et al., 1999). In order to test whether sorting to areas of high curvature results directly from S-acylation of the cargo, model acylated transmembrane peptides were probed for a partitioning between flat and curved membranes in vitro. Strikingly, acylation of the peptides resulted in a strong partitioning into highly curved membranes, to an extent in line with the increase in anterograde transport observed for trafficking of model acylated cargoes through the Golgi. Together, the data strongly support a model whereby S-acylation directly modulates the biophysical property of the cargo, resulting in its increased partitioning to the cisternal rim of cis-Golgi membranes, which in turns facilitates its anterograde routing (see **Figure 1**).

## ADDITIONAL EXAMPLES FOR ANTEROGRADE ROUTING OF S-ACYLATED MEMBRANE CARGO

#### Linker for Activation of T Cells (LAT)

Hundt et al. (2009) investigated the trafficking of LAT, a dually S-palmitoylated protein that is a crucial signaling molecule for T-cell receptor-based stimulation of T-cells (Hundt et al., 2009). Anergic T cells lack palmitoylation of LAT, resulting in a reduction in Tyr phosphorylation and activation of PLCγ1 (Ladygina et al., 2011). Hundt et al. demonstrated that when LAT is not S-palmitoylated, its coupling to sphingolipid/cholesterol-rich membranes is not affected, but rather results in LAT's accumulation in the Golgi, with negligible levels detected at the PM. Further, LAT is in the family of transmembrane adaptor proteins (TRAPs), and the additional family members LIME and NTAL/LAB also were shown to require S-palmitoylation for efficient export to the PM (Stepanek et al., 2014). Thus, anterograde routing of TRAPs via S-palmitoylation at the Golgi emerges as a requirement for T-cell function.

# β-Adrenergic Receptors (AR)

β-adrenergic receptor (AR) isoforms 1–3 are all S-palmitoylated proteins, but these isoforms exhibit different sites of acylation. Recently, Adachi and colleagues investigated mammalian β3AR, and observed that a distinct site (Cys-153) is crucial for proper targeting of the receptor to the PM, while other sites (Cys-361/363) impact its stability at the PM (Adachi et al., 2019). These observations prompt the hypothesis that within the AR family, different sites of S-palmitoylation are employed to toggle between states, and may control AR targeting to different

cargo and reduces the inclusion of anterograde cargo in the retrograde carriers. Thioesterase activity allows for cycles of lateral diffusion back to the cisternal center

cellular locations, modulating its surface expression and hence downstream signaling.

# GABA Type a Receptors (GABAARs)

and re-acylation by DHHCs.

GABA receptors are well-established S-acylated PM residents. Based on studies using overexpressed DHHCs, they were postulated to be palmitoylated by DHHCs 3 and 7 (Fang et al., 2006). Subsequently, Kilpatrick et al. (2016) identified DHHC3 as a specific PAT of the γ2 subunit of GABAARs through knockout of either DHHC3 or 7 in mice (Kilpatrick et al., 2016). Whereas knockout of DHHC7 had no effect on GABAAR trafficking in neurons, DHHC3 KO neurons exhibited drastically reduced levels of GABAAR γ2 at synapses, impacting synaptic function. Most remarkably, and in line with the identification of DHHCs 3 and 7 as ubiquitous regulators of protein sorting at the cis-Golgi (Ernst A.M. et al., 2018), knockout of the individual PATs resulted in only marginal effects, while a double knockout of DHHCs 3 and 7 resulted in perinatal lethality of mouse embryos, emphasizing the importance of these PATs and sorting at the cis-face of the Golgi for general cell function.

# Glucose Transporter 4 (GLUT4)

Insulin stimulates glucose uptake in fat and muscle cells by causing the exocytic translocation of GLUT4 to the PM. Although this had previously been considered exclusively as a post-Golgi process, more recent data make clear that recycled GLUT4 accumulates in a pool of small (∼50 nm diameter) vesicles that reside near the ERGIC and cis-Golgi compartments, in association with TUG, Golgin-160, and ACBD3 (Acyl-CoA Binding domain-containing protein 3, also known as GCP60) (Xu et al., 2011; Bogan, 2012; Orme and Bogan, 2012; Belman et al., 2015). Upon insulin stimulation, these vesicles are mobilized by TUG cleavage, and they are proposed to traffic to the cell surface by an unconventional secretion pathway that bypasses the Golgi stack (Xu et al., 2011; Bogan, 2012; Habtemichael et al., 2018). GLUT4 is S-acylated and the PAT responsible for this modification was recently identified (Du et al., 2017).

Both DHHCs 3 and 7 bound to GLUT4, but only silencing of DHHC7 abolished GLUT4 acylation, leading to the conclusion that DHHC7 acts as a specific PAT for GLUT4. Other proteins that cotraffic with GLUT4 or that regulate this process are also palmitoylated in adipocytes (Ren et al., 2013a). Importantly, the ability of insulin to stimulate GLUT4 translocation was impaired by knockdown or knockout of DHHC7 and by mutation of the palmitoylated Cys residue in GLUT4 (Ren et al., 2015; Du et al., 2017). Thus, it may be that S-acylation is required to concentrate the GLUT4 in the highly curved membranes of the insulin-responsive vesicles.

#### S-PALMITOYLATION CONFERS MEMBRANE PROTEINS A "SORTING-COMPETENT" STATE

How might S-palmitoylation of proteins induce partitioning to the highly curved rims of Golgi cisternae? As noted above, S-palmitoylation occurs exclusively at the cytoplasmic leaflet of endomembranes, frequently on one or two adjacent sites and at juxtamembrane positions of integral membrane proteins. It is known that the cis-Golgi cisternae are stacked by the action of GRASP65 proteins, which leads to a significant flattening of membranes in the stacked area (Ladinsky et al., 1999). Upon S-palmitoylation, the acylated proteins would exhibit a local mass excess in the cytoplasmic leaflet of the membrane, concomitant with an asymmetric hydrophobic Z-profile (i.e., an increase in spontaneous curvature/transition from cylindrical to conical profile); see top right panel of **Figure 1**. In support of this concept, early studies on erythrocytes demonstrated that drugs intercalating into the cytoplasmic leaflet induced a morphological change from a flat and disc-like to spherical morphology (Sheetz and Singer, 1974). In the Golgi, the tightly stacking proteins in the central disk region would inhibit such a morphological transition – therefore, the local mass excess in the cytoplasmic leaflet due to S-palmitoylation is expected to induce curvature stress, forcing acyl chains in the cytoplasmic leaflet to potentially form locally curved clusters. This stress could presumably induce the observed partitioning toward the positively curved cisternal rims, where S-palmitoylation-induced curvature stress would be released (**Figure 1**). This reasoning suggests that S-palmitoylation serves as a biophysical switch and sorting signal at the cis Golgi, to extract cargo proteins from planar membranes and concentrate them at the cisternal rims.

The cisternal rim is fenestrated and comprises a network of tubular-vesicular elements, referred to as the "non-compact zone/region" (Ladinsky et al., 1999). A significant fraction of PM resident proteins (>15%) are predicted to be S-acylated (Ernst A.M. et al., 2018). Yet, the data nonetheless raise the question how anterograde cargo that lack this sorting signal can be efficiently sorted to the cisternal rim. The diffusive flux along the gradient generated by S-palmitoylated proteins would have the potential to drag along non-acylated proteins. In line with this hypothesis, overexpression of cis-Golgi DHHCs resulted in an increased flux of (soluble) secretory cargo, and even impacted non-acylated cargo to a lower but significant extent (Ernst A.M. et al., 2018). Multiple additional scenarios can also be envisioned, and could contribute to spontaneous curvature-based sorting of proteins at the Golgi: (1) conicity of the hydrophobic moiety (the membrane anchor) acquired through hetero-oligomerization or protein folding, as observed for the polytopic membrane channel KvAP, which exhibits affinity for areas of high curvature without being lipidated (Aimon et al., 2014), (2) heterooligomerization of palmitoylated and non-palmitoylated proteins, and (3) coagulation of integral S-palmitoylated proteins and luminal proteins via lectins, e.g., the mannose-specific lectin VIP36 that is present at the cis Golgi (Fullekrug et al., 1999).

A further consideration is that S-acylation-induced sorting of membrane proteins may apply to intracellular organelles other than the Golgi. At the ER, notably, S-palmitoylation of the ER-resident Calnexin alters its localization toward mitochondria-associated membranes (MAM), which exhibit a high extent of positive curvature (Lynes et al., 2012). Supportively, access of the ER-thioredoxin TMX to MAM also requires S-acylation (Lynes et al., 2012). However, S-acylation of Calnexin was also suggested to regulate access to sheet-like structures (Lakkaraju et al., 2012), which might not be a contradiction given that a recent study detected nanoholes within ER-sheets (that would give rise to curvature, Schroeder et al., 2019). Additionally, the endosome-localized DHHC15 was suggested to concentrate cargo for efficient uptake by retromer (Mccormick et al., 2008). Thus, S-palmitoylation may be a ubiquitous mechanism to cause the partitioning of integral membrane proteins into regions of high curvature.

From an evolutionary standpoint, S-palmitoylation represents a widespread post-translational modification that is conserved from yeast to mammals, plants, and even encountered in multiple parasites (Lam et al., 2006; Roth et al., 2006; Brent et al., 2009; Hemsley et al., 2013; Hodson et al., 2015). The fact that virtually all spike proteins of enveloped viruses are S-palmitoylated, regardless of whether or not their viral exit sites are enriched in sphingolipids/cholesterol (e.g., Influenza A NA, VSV-G, Ebola GP), further emphasizes the evolutionary pressure toward maintaining this putative anterograde sorting signal to maximize the efficiency of their anterograde routing at the Golgi. Possibly, palmitoylation may also induce curvature at the PM, which may be relevant for budding of membrane, in addition to targeting soluble and membrane proteins (Li et al., 2017; Rana et al., 2019; Wang et al., 2019).

### Controlled Access of Soluble Proteins to the Cis-Golgi S-Palmitoylation Machinery

Two members within the DHHC PAT family, the huntingtin-interacting proteins DHHC 13 and 17, particularly stand out because of their architecture. They contain long amino-terminal ankyrin-repeats, which were shown to bind to specific soluble substrate bearing corresponding ankyrin-binding motifs (Lemonidis et al., 2015). Among their substrates are the critical neuronal proteins MAP6 and CSP, but also the ubiquitous t-SNAREs in the SNAP25 family (Greaves et al., 2010; Lemonidis et al., 2014). Interestingly, Lemonidis et al. (2014) demonstrated

that binding of these substrates to DHHCs 13 and 17 does not (13) or only marginally (17) induces their acylation, whereas DHHCs 3 and 7 are not able to bind the proteins but efficiently mediate their acylation. The fact that DHHCs 3, 7, 13, and 17 all localize to the cis-Golgi strongly suggest cooperativity among the PATs in this subcompartment, with DHHCs 13 and 17 serving to recruit these proteins to the cis-Golgi, where they are subsequently S-palmitoylated by DHHCs 3 and/or 7 in a concerted, two-stage process.

As noted above, several proteins involved in the regulation of GLUT4 translocation are palmitoylated, and some of these may also be recruited as soluble proteins onto highly curved membranes (Ren et al., 2013a,b; Du et al., 2017). In unstimulated cells, the insulin-responsive vesicles are trapped in association with ACBD3, which binds palmitoyl-CoA (Belman et al., 2015; Yue et al., 2019). The related protein ACBD6 facilitates N-myristoylation of substrate proteins by cooperating with N-myristoyltransferase enzymes (Soupene and Kuypers, 2019). Whether ACBD3 may participate in palmitoylation of soluble proteins is not known, although in general ACBD family members have diverse roles in lipid-modification and metabolism pathways (Chen et al., 2012; Nazarko et al., 2014; Soupene et al., 2016) and in Golgi structure (Yue et al., 2017; Liao et al., 2019).

Data also suggest that the accumulation of a pool of small, insulin-responsive vesicles may be impaired during the development of type 2 diabetes, which may result from excess membrane diacylglycerols and sphingolipids (Garvey et al., 1998; Maianu et al., 2001; Czech, 2017; Habtemichael et al., 2018; Petersen and Shulman, 2018). We speculate that these excess lipids might potentially disrupt the palmitoylation-based sorting mechanisms discussed above and thus contribute to attenuated insulin action.

#### Outlook

#### Enrichment of S-Palmitoylated Cargo in Vesicular-Tubular Carriers Versus Entrapment in Maturing Cisternae at the Golgi

S-palmitoylation-induced sorting of cargo to the cisternal rim fits well with the long-proposed role of COPI vesicles as anterograde carriers (Rothman, 2014) in addition to their well-established role as retrograde carriers (Brandizzi and Barlowe, 2013). Artificially introducing stable adhesions ("staples") between Golgi cisternae does not impair anterograde transport, showing that anterograde transport primarily occurs from the highly curved rims (Lavieu et al., 2013), where COPI vesicles bud (Orci et al., 1986). When mitochondria are engineered to invade the Golgi by affording them adhesion to its cisternae, these organelles dissect the cisternae apart and immobilize them (Lee et al., 2014). Nonetheless, anterograde transport continues (Dunlop et al., 2017), implying a small diffusible carrier. Independently, COPI vesicles, visualized by super-resolution microscopy, carry VSV-G protein between separated Golgi areas in fused cells and account for most of the cargo transported by this route (Pellett et al., 2013). We propose that palmitoylation-driven anterograde sorting bears the potential to help resolve the long-standing conundrum of how one coat could mediate two fates (Pelham, 1994). Retrograde cargo is selected into COPI vesicles by classic receptor-dependent binding (such as KDEL – KDEL receptor or KKXX – coatomer (White et al., 1999; Duden, 2003; Barlowe and Helenius, 2016)). We can envision a physical-chemical, membrane-intrinsic process by which S-palmitoylated anterograde cargo spontaneously clusters within curved regions to the exclusion of other proteins, including retrograde cargo and their receptors. The same COPI coat could then pinch off distinct anterograde and retrograde vesicles (Orci et al., 1997) from these differentiated regions of membrane.

An alternative interpretation, along the lines of a cisternal maturation model for intra-Golgi transport (Casler et al., 2019; Kurokawa et al., 2019), is that S-palmitoylation-induced sorting of cargo to the rim serves the purposes of redistributing the cargo while the cisternae mature (i.e., enzyme contents change). Potentially through the action of APT-mediated acylation-deacylation cycles, the de-palmitoylated cargo would lose their stable anchoring at the cisternal rim and partition back to the (flat) center of the cisterna. There, the cargo would engage in additional rounds of acylation by DHHCs, and through its partitioning back and forth across the cisterna, its chances of being processed by glycosyltransferases might increase. However, a challenge here is to explain how palmitoylated cargo would traffic through the maturing Golgi faster than non-palmitoylated cargo, as observed (Ernst A.M. et al., 2018). One would need to invoke that only the non-palmitoylated cargo partially traffic retrograde, for which there is currently scant evidence.

A third option would be a "fast-track" through the Golgi mediated by intercisternal continuities. VSV-G translocation through the Golgi was reported to induce intercisternal connections (Trucco et al., 2004), and additional types of GTPase-dependent tubules were detected on the Golgi (Park et al., 2015; Bottanelli et al., 2017). S-palmitoylation could here putatively provide a mechanism to allow partitioning into these highly curved connections between cisternae (**Figure 1**). To address this will require visualization of sufficiently large areas of the Golgi, with specificity, at very high resolution. This may be possible with new advances in super-resolution imaging (Huang et al., 2016) and with large volume electron microscopy approaches such as FIB-SEM (Kizilyaprak et al., 2019).

#### FUTURE STEPS: VISUALIZING DHHC PLATFORMS

Acylation at the cis-face of the Golgi emerges as an important mechanism for routing of several proteins to the PM. Given that multiple DHHC PATs are capable of cooperating to achieve S-palmitoylation of specific substrates (Lemonidis et al., 2014; Abrami et al., 2017), coupled with the fact that several of these isoforms are present only in distinct tissues (Ohno et al., 2006), indicates that screening DHHC libraries to identify "specific" PATs for a protein of interest is not adequate. PATs are emerging as finely adjusted enzyme assemblies rather than isolated entities, and thus their spatial context (location within

the same cisterna and across cisternae) has to be studied – ideally at endogenous levels and by means of super-resolved light and electron microscopy. These nano-assemblies are expected to vary in composition across different tissues and during different developmental stages/states of cell proliferation. We hypothesize that DHHC platforms have the potential to provide a mechanism for fine-tuning the flux of specific substrates from the Golgi to the PM, thereby controlling the availability and timing of receptors, channels, and other types of cargo at the cell surface.

With S-acylation as a marker for anterograde cargo in hand, live-cell and super-resolution microscopy-compatible orthogonal labeling strategies need to be developed to better visualize trafficking of anterograde cargo as a class (as opposed to individual proteins with varying properties) – on the nanoscale and on timescales that allow to observe complete transport/maturation cycles across the mammalian Golgi. These advances would allow investigations to reopen the exploration of putative differences in the mechanism of intra-Golgi transport between lower and higher eukaryotes, and furthermore would enable detailed studies of acylation-controlled protein export from the Golgi in response to external stimuli. Such studies could elucidate important

#### REFERENCES


fundamental principles, as well as insights into mammalian physiology and disease.

#### AUTHOR CONTRIBUTIONS

AE, DT, and JB researched the relevant literature, and conceived and wrote the manuscript.

## FUNDING

AE and DT thank the Mathers Foundation. JB is grateful to the NIH (DK092661) and the American Diabetes Association for research support.

#### ACKNOWLEDGMENTS

We thank Dr. Kirill Grushin for help with preparation of the space-filling structures shown in **Figure 1** and Dr. James E. Rothman for his support and insightful discussion.



transporter 4 and insulin-regulated aminopeptidase in type 2 diabetes mellitus: implications regarding defects in vesicle trafficking. J. Clin. Endocrinol. Metab. 86, 5450–5456. doi: 10.1210/jc.86.11.5450



**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 Ernst, Toomre and Bogan. 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.

# The Physiological Functions of the Golgin Vesicle Tethering Proteins

#### Martin Lowe\*

Faculty of Biology, Medicine and Health, University of Manchester, Manchester, United Kingdom

The golgins comprise a family of vesicle tethering proteins that act in a selective manner to tether transport vesicles at the Golgi apparatus. Tethering is followed by membrane fusion to complete the delivery of vesicle-bound cargo to the Golgi. Different golgins are localized to different regions of the Golgi, and their ability to selectively tether transport vesicles is important for the specificity of vesicle traffic in the secretory pathway. In recent years, our mechanistic understanding of golgin-mediated tethering has greatly improved. We are also beginning to appreciate how the loss of golgin function can impact upon physiological processes through the use of animal models and the study of human disease. These approaches have revealed that loss of a golgin causes tissuerestricted phenotypes, which can vary in severity and the cell types affected. In many cases, it is possible to attribute these phenotypes to a defect in vesicular traffic, although why certain tissues are sensitive to loss of a particular golgin is still, in most cases, unclear. Here, I will summarize recent progress in our understanding of golgins, focusing on the physiological roles of these proteins, as determined from animal models and the study of disease in humans. I will describe what these in vivo analyses have taught us, as well as highlight less understood aspects, and areas for future investigations.

Keywords: Golgi apparatus, golgin, vesicle traffic, tether, animal model, secretion, glycosylation, extracellular matrix

# INTRODUCTION

The Golgi apparatus lies at the heart of the secretory pathway, serving to modify newly synthesized cargo proteins and to sort and transport these proteins to their final destination, which may be inside or outside the cell. It is comprised of flattened membrane compartments called cisternae that are layered on top of one another to form the Golgi stack. In non-vertebrates the Golgi stacks exist as discrete entities within the cytoplasm, whereas in vertebrates the stacks are laterally connected to form a single-copy Golgi ribbon which is located adjacent to the centrosome (**Figure 1**). Newly synthesized cargo proteins arriving from the endoplasmic reticulum (ER) enter the Golgi apparatus at the cis-face, which in vertebrates comprises a tubulo-vesicular compartment called the cis-Golgi network (CGN). Cargo then transits the Golgi stack before arriving at the exit station of the Golgi, the trans-Golgi network (TGN), where it is sorted into carriers for delivery to its post-Golgi destination. As cargo transits the Golgi stack, numerous resident enzymes carry out post-translational modifications, most notably at the level of glycosylation, allowing for cargo maturation. There is extensive recycling of the resident enzymes between Golgi cisternae, which is required to maintain the distinct identity of the cisternae in the face of the cisternal migration that is thought to carry cargo forwards, and to ensure that cargo is correctly modified. In addition to the forward transport of newly synthesized cargo proteins, and the recycling of Golgi residents,

#### Edited by:

Vladimir Lupashin, University of Arkansas for Medical Sciences, United States

#### Reviewed by:

Elizabeth Sztul, University of Alabama at Birmingham, United States Sean Munro, Medical Research Council, United Kingdom

> \*Correspondence: Martin Lowe martin.lowe@manchester.ac.uk

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 12 March 2019 Accepted: 16 May 2019 Published: 18 June 2019

#### Citation:

Lowe M (2019) The Physiological Functions of the Golgin Vesicle Tethering Proteins. Front. Cell Dev. Biol. 7:94. doi: 10.3389/fcell.2019.00094

the Golgi also receives recycling proteins from the endolysosomal system, which arrive at the TGN. We can therefore think of the Golgi apparatus as a hub for protein traffic. The golgins act in a selective manner to capture cargo protein-containing transport vesicles at different regions of the Golgi, in a process referred to as vesicle tethering. In this review, I will describe recent progress in understanding how golgins mediate vesicle tethering, and go on to describe how loss of golgin function can impact upon physiological processes, as determined from the use of animal models and the study of human diseases attributable to the loss of golgin function.

#### GOLGINS AS VESICLE TETHERING PROTEINS

The golgins comprise a family of Golgi-localized coiled-coil proteins with a similar topology (Munro, 2011; Witkos and Lowe, 2015). They are anchored to the Golgi membrane by their carboxy-terminus, and extend into the cytoplasm to facilitate vesicle capture, which is in most cases is mediated by the extreme amino-terminus of the protein (Cheung et al., 2015; Gillingham and Munro, 2016; Wong et al., 2017; Gillingham, 2018). In humans, there are at least 11 golgins, with varying degrees of conservation between different eukaryotes depending upon the particular golgin. The different golgins are localized to distinct regions of the Golgi apparatus, consistent with their ability to tether different vesicle types (Wong and Munro, 2014; Gillingham and Munro, 2016; Gillingham, 2018). For example, golgins localized at the cis-Golgi are competent to selectively tether vesicles arriving from the ER and intra-Golgi vesicles mediating recycling from later Golgi cisternae, whereas those at the trans-Golgi tether vesicles arriving from the endolysosomal system (**Figure 2**). In contrast, golgins residing within the Golgi stack are able to tether only intra-Golgi transport vesicles. Golgins therefore play a major role in dictating the specificity of vesicle traffic within the secretory pathway. In addition, the elongated nature of the golgins, coupled with their ability to tether vesicles via their membrane-distal amino-termini, giving a greater radius of capture, allows for increased efficiency of

intra-Golgi vesicles. Golgin-84 and TMF can only tether intra-Golgi vesicles, while the trans-Golgi golgins are able to ether endosome-derived vesicles. traffic. Following vesicle capture, golgins are thought to cooperate with other proteins, including Rab GTPases and multi-subunit tethering complexes such as COG and GARP, to mediate the transition from tethering to membrane fusion, which operates over a relatively short distance and is mediated by SNARE

trans-Golgi/trans-Golgi network (TGN) (blue: GCC88, GCC185, golgin-97, golgin-245). For most golgins, the ability to tether vesicles has been demonstrated, indicated by arrows), but for some this ability remains to be shown (CASP, giantin, golgin-160). ER-derived vesicles are tethered at the cis-Golgi by GM130 and GMAP-210, which can also tether recycling

proteins (Witkos and Lowe, 2017). Although the majority of mammalian golgins have been shown to mediate vesicle tethering, for a few members of the family this ability has yet to be formally demonstrated. This may reflect the nature of the cell-based assay that has been used to assess golgin-mediated tethering, which may have missed tethering interactions mediated by certain golgins (Wong and Munro, 2014; Wong et al., 2017), or it could indicate that some golgins have functions other than vesicle tethering. Golgin-160, for example, has been shown to mediate binding to the microtubule motor protein dynein for Golgi morphology and positioning (Yadav et al., 2012). This does not preclude an additional vesicle tethering activity for golgin-160, but such an activity has so far not been demonstrated. Other golgins have also been shown to recruit factors to the Golgi membrane in order to control downstream cellular processes. GM130 is particularly interesting in this regard. In addition to acting as a tether for

ER-derived vesicles (Wong and Munro, 2014), it also interacts with the microtubule nucleation factor AKAP450 to promote microtubule nucleation at Golgi membranes (Rivero et al., 2009), the protein kinase STK25, which is important for cell migration (Preisinger et al., 2004), RasGRF, a Cdc42 GEF also important for cell migration (Baschieri et al., 2014), and WAC, which functions in autophagy induction (Joachim et al., 2015). It is therefore possible for golgins to have cellular functions other than, or in addition to, vesicle tethering, although for most golgins it would appear that vesicle tethering is their primary role.

#### ANIMAL MODELS FOR GOLGIN FUNCTION

As might be expected, numerous cell-based studies have been carried out to investigate the roles of golgins in vesicle trafficking. Loss of function studies have typically been performed in generic cell lines, using model cargo proteins, to give us a picture of what trafficking steps different golgins participate in (Ramirez and Lowe, 2009; Munro, 2011; Cheung and Pfeffer, 2016). The use of generic cell lines and model cargoes for these studies is certainly justified considering that the golgins are widely expressed throughout tissues. However, in many cases it has not been possible to unambiguously determine the functional requirement for a golgin using this type of approach, with relatively mild phenotypes reported, or in some cases no trafficking phenotypes at all. A possible explanation is functional redundancy between the golgins. For example, the trans-Golgi golgins share overlap in their ability to tether vesicles containing the same endosomal cargo proteins, while GMAP-210, golgin-84, and TMF can tether vesicles with the same intra-Golgi cargo (Wong and Munro, 2014; Wong et al., 2017). This is consistent with these golgins sharing similar vesicle tethering motifs in their amino-termini, which are distinct from those found in the trans-Golgi golgins (Wong and Munro, 2014; Wong et al., 2017). Hence, loss of a single golgin may not manifest as a phenotype due to redundancy with another golgin. The use of model cargo proteins may also fail to reveal the requirement for a golgin in the trafficking of a particular cargo, which may have a greater dependency on one golgin compared to others. Moreover, the organization of the secretory pathway can vary between cell types, which again may allow for differential requirements for golgins in trafficking within cells. Finally, golgins may only enhance trafficking efficiency, such that loss of a golgin may only have a minimal impact upon rates of transport as measured in the standard trafficking assays used in cultured cell models. Hence, the inherent limitations of model cell lines and cargo proteins make it difficult to appreciate the functional requirement for golgins in trafficking in a broader context, and do not allow for assessment of the physiological consequences of impaired golgin-dependent transport. One way to address these issues is to use animal models, allowing the analysis of the myriad different cell types throughout the body, and the cargo proteins expressed by these cells, as well as determining the consequences upon organismal physiology upon loss of golgin function. Such approaches can also reveal degrees of functional redundancy in an in vivo context. The established animal models for golgins, and the information that has been gained from them, are discussed below (see also **Table 1**).

#### Golgins at the cis-Golgi GM130

One of the best-studied golgins is GM130, which is anchored to the cis-Golgi via binding to the Golgi stacking protein GRASP65 (Nakamura et al., 1995; Barr et al., 1997), although interactions with the related GRASP55 and Rab1 may also contribute to its Golgi targeting (Moyer et al., 2001; Short et al., 2001; Weide et al., 2001; Bekier et al., 2017). GM130 is able to tether ER to Golgi vesicles at the cis-Golgi (Wong and Munro, 2014), and, as described above, can anchor a number of other proteins, namely AKAP450, STK25, RasGRF, and WAC, to the Golgi membrane (Preisinger et al., 2004; Rivero et al., 2009; Baschieri et al., 2014; Joachim et al., 2015). Depletion of GM130 in cultured cells has given variable results, with some studies reporting no effects upon transport (Puthenveedu and Linstedt, 2001; Kondylis and Rabouille, 2003; Puthenveedu et al., 2006), whereas others have seen reduced rates of ER to Golgi transport (Marra et al., 2007; Diao et al., 2008), although in most cases a loss of Golgi ribbon architecture is observed. GM130 has been knocked out in the fruit fly Drosophila melanogaster (Zhou et al., 2014). The flies appear viable, although a detailed phenotypic analysis has yet to be reported. As seen in mammalian hippocampal neurons (Horton et al., 2005), the Golgi can exist as "outposts" within the dendritic arbor of Drosophila larval neurons (Zhou et al., 2014). Loss of GM130 within the larval neurons causes loss of Golgi outpost organization, and a reduced ability to nucleate microtubules from the outposts, accompanied by reduced dendritic branching, indicating a role for the protein in neuronal Golgi and microtubule organization.

Morpholino-induced depletion of GM130 in zebrafish larvae causes reduced brain size, skeletal muscle disorganization, and altered mobility, suggesting a neural or neuromuscular defect (Shamseldin et al., 2016), although the mechanistic basis for these phenotypes remains to be determined. Loss of mammalian GM130 also leads to a major neurological phenotype (Liu et al., 2017). Although viable at birth, GM130 knockout mice fail to thrive and die by 1–2 months of age. Loss of GM130 globally, or by tissue-specific knockout within the central nervous system, results in progressive neurodegeneration due to death of Purkinje neurons within the cerebellum, which causes ataxia. Purkinje neurons have a large and extensively branched primary dendrite, which requires constant delivery of protein and lipid for its growth and maintenance, which continues into postnatal development (Dvorak and Bucek, 1970; Tanaka, 2009). The Golgi is particularly well developed in Purkinje neurons and is positioned at the base of the dendrite within these cells, allowing for extensive and polarized delivery of secretory cargoes into the dendrite (Tanabe et al., 2010; Liu et al., 2017). Upon loss of GM130, the Purkinje cell Golgi is mis-localized, possibly through loss of binding to AKAP450, and the Golgi ribbon is fragmented (Liu et al., 2017) (**Figure 3**). These changes


#### TABLE 1 | Animal models and human diseases associated with loss of golgin function.

∗ In all cases the phenotypes are those seen upon loss of golgin expression in the whole organism. In some cases, where indicated, the same phenotypes were also reported using conditional tissue-specific knockout. The type of manipulation used to abrogate expression of the golgin is indicated. Pel: P-element excision; MO: morpholino; KO: total organism knock-out by gene targeting; cKO: conditional tissue-specific knock-out by gene targeting; ENU: N-ethyl-N- nitrosourea generated mutant; Genetrap: insertional mutagenesis by gene trap; CRISPR: CRISPR-Cas9 generated knockout; TALEN: TALEN-generated knockout; Spon: spontaneous mutation.

are accompanied by reduced secretory traffic into the dendrite, accounting for the atrophy of the dendritic arbor seen in these cells. Whether impairment of microtubule nucleation at the Golgi contributes to the Purkinje cell phenotype is currently unclear, as is any potential contribution of Golgi outposts, which appear to be absent from this type of neuron. Nevertheless, the mouse studies point to an important role for GM130 in neuronal Golgi organization and trafficking, which is important for the maintenance of Purkinje cells. Whether other neuronal types in mammals are also affected by loss of GM130, as suggested by the study of Drosophila and zebrafish GM130, remains to be seen.

Loss of mouse GM130 affects Golgi morphology in other tissues, with fragmentation of the Golgi ribbon seen in two major secretory cell types, the pancreatic acinar cell and the type II alveolar cell within the lung (Liu et al., 2017). Further work will be required to determine the physiological consequences of GM130 loss in these cells. The study of a second GM130 knockout mouse model, which also has reduced viability and growth, has revealed changes in liver and lung pathology, with

FIGURE 3 | Golgi fragmentation in Purkinje neurons upon loss of GM130. The Golgi apparatus in Purkinje neurons is located within the cell soma, close to the centrosome, at the base of the elaborate apical dendritic tree that forms in these neurons. This positioning allows for polarized delivery of newly synthesized secretory cargo to the dendrite, which is required for its growth and maintenance. Loss of GM130, either in the whole organism, or selectively with the central nervous system, causes fragmentation of the Purkinje Golgi apparatus and its mislocalization away from the centrosome. Golgi fragmentation is shown both schematically and in electron microscope (EM) images of the Purkinje neuron Golgi apparatus. These changes result in reduced cargo delivery into the dendrite, which causes atrophy of the dendritic tree. This ultimately causes death of the Purknije neurons, which control movement, and as a consequence the GM130 knockout mice suffer from severe ataxia.

increased numbers of autophagosomes and increased fibrosis within these tissues (Park et al., 2018). How the loss of GM130 results in these phenotypes remains to be determined. Finally, loss of GM130 in mice, both globally or using a germ-cell tissue specific knockout, causes defects in spermatogenesis, resulting in male infertility (Han et al., 2017). During spermatogenesis, the Golgi plays an important role in biogenesis of the acrosome, which is a lysosome-related organelle located at the head of the sperm (Moreno et al., 2000). The acrosome contains digestive enzymes that, upon release, break down the outer layers of the ovum to allow fertilization to occur. Hence defective acrosome biogenesis often results in male infertility. GM130 knockout mice are completely deficient in acrosome biogenesis, explaining the penetrant male infertility phenotype seen in these animals (Han et al., 2017). Interestingly, acrosome biogenesis is particularly sensitive to perturbation of Golgi function, and this process is defective in animals lacking several other golgins (see below).

#### GMAP-210

GMAP-210 is bound to the cis-Golgi membrane through interaction of its carboxy-terminal GRAB domain with the small GTPase Arf1 (Gillingham et al., 2004). It is competent to tether both ER-derived transport vesicles and retrograde intra-Golgi transport vesicles via two overlapping vesicle-tethering motifs in its amino-terminus (Drin et al., 2008; Wong and Munro, 2014; Wong et al., 2017). It can also anchor the ciliary protein IFT20, which is a component of the IFT (intraflagellar transport) complex, to the Golgi (Follit et al., 2006). Through this interaction it is likely that GMAP-210 participates in the trafficking of cargo proteins from the Golgi to the cilium (Monis et al., 2017), although it has also been proposed that IFT20 participates in ER to Golgi transport (Noda et al., 2016). In both cases the mechanistic details remain to be dissected. Loss of GMAP-210 at the cellular level results in extensive Golgi fragmentation and reduced rates of secretory traffic (Smits et al., 2010; Roboti et al., 2015; Sato et al., 2015; Wehrle et al., 2018). Several animal models for GMAP-210 have been generated, and the phenotypes are consistent with defects in Golgi and/or cilia function. Loss of the GMAP-210 ortholog in Caenorhabditis elegans, called SQL-1, does not affect health or viability, but there is mild disruption of Golgi organization, and defective IFT transport, most likely as a consequence of reduced delivery of components required for IFT transport from the Golgi to the cilium (Broekhuis et al., 2013).

Both global and tissue-specific knockout mouse models have been generated to study mammalian GMAP-210, also called TRIP11. Systemic loss of GMAP-210 results in neonatal lethality, indicating that GMAP-210 is an essential protein (Follit et al., 2006; Smits et al., 2010). In the first knockout model, which utilized a gene trap to disrupt GMAP-210 expression in all tissues, there is a pleiotropic phenotype with defects in the heart and lung, as well as reduced overall growth (Follit et al., 2006). Although the underlying cause of these phenotypes remains to be determined, it has been proposed that impaired cilium-based signaling during development may be responsible. The most prominent phenotype in the second knockout model, which was an N-ethyl-N- nitrosourea (ENU) generated mutant, was skeletal dysplasia, with shortening of the bones and reduced ossification (Smits et al., 2010). These changes were accompanied by abnormal chondrocyte differentiation and increased cell death. The chondrocyte Golgi apparatus is extensively fragmented upon loss of GMAP-210, and there is a deficit in glycosylation and secretion of extracellular matrix proteins. The reduced ability to modify and secrete matrix proteins is likely to account for the bulk of the cartilage and bone phenotype seen upon in the GMAP-210 knockout mice. However, it also possible that altered ciliary signaling could contribute to the phenotypes considering that signaling from the cilium is also important for various aspects of bone morphogenesis, including chondrocyte differentiation and formation of the growth plate (Braun and Hildebrandt, 2017). Interestingly, the Golgi was fragmented in other embryonic tissues e.g., kidney, but not in others e.g., lung and intestine, indicating differences in the requirement for GMAP-210 to maintain Golgi organization between cell types (Smits et al., 2010). More recently, GMAP-210 has been conditionally knocked out in various cell types associated with skeletal development, which showed that it is the loss of GMAP-210 in chondrocytes,

as opposed to fibroblasts, osteoclasts or osteoblasts, that is important for the skeletal phenotype seen in GMAP-210 deficient mice (Bird et al., 2018). This study also showed that secretion of certain matrix proteins was more sensitive to loss of GMAP-210 that others, supporting the view that GMAP-210 may promote the trafficking of certain cargoes over others. More recently, it has been shown that loss of Gmap-210 from zebrafish gives a skeletal phenotype, with a reduction in bone density (Daane et al., 2019). This suggests the physiological role of GMAP-210 is conserved between vertebrates, although further characterization will be required to confirm this.

#### Golgin-160

Golgin-160 (also known as GOLGA3) is enriched towards the cis-side of the Golgi apparatus (Hicks et al., 2006). It has been shown to recruit the dynein microtubule motor protein to Golgi, which is important for Golgi positioning and ribbon maintenance (Yadav et al., 2012), but whether it participates in vesicle tethering has yet to be demonstrated. Loss of golgin-160 in two different mouse models has no effect upon viability (Matsukuma et al., 1999; Bentson et al., 2013), which is perhaps surprising considering the strong Golgi positioning defect seen when golgin-160 is depleted in cultured cells (Yadav et al., 2012). The only overt phenotype in vivo is male sterility (Matsukuma et al., 1999; Bentson et al., 2013). There is increased death of male germ cells upon loss of golgin-160, and those sperm that are generated have deficient acrosome biogenesis (Bentson et al., 2013). Further analysis is required to ascertain the cause of death attributable to loss of golgin-160, which has previously been implicated in proapoptotic signaling, and to determine how its loss impairs acrosome biogenesis, although it is likely via defective Golgi positioning or trafficking.

#### Golgins at the Cisternal Rims Golgin-84 and CASP

Golgin-84 is present at the cisternal rims of the Golgi apparatus, where it is enriched towards the cis-side (Diao et al., 2003; Satoh et al., 2003). Golgin-84 is able to tether intra-Golgi transport vesicles, and as a transmembrane protein, is also recycled to earlier cisternae in intra-Golgi transport vesicles, in which case it is a cargo protein (Wong and Munro, 2014). There are orthologs of golgin-84 in D. melanogaster and C. elegans, although they have yet to be analyzed at the functional level. Mice lacking golgin-84, also known as GOLGA5, are viable, and there is no overt phenotype in these mice, which develop and grow normally (McGee et al., 2017). There is also no male infertility, indicating that golgin-84 is dispensable for acrosome biogenesis. The golgin CASP is also anchored by a carboxy-terminal transmembrane domain and present at the cisternal rims (Gillingham et al., 2002). CASP can bind golgin-84, and although this interaction has been implicated in vesicle tethering (Malsam et al., 2005), it remains to be demonstrated whether CASP is sufficient to tether vesicles in its own right. Interestingly, and in contrast to golgin-84, CASP is absent from D. melanogaster. No animal knockout for CASP has been generated as of yet so the in vivo importance of CASP remains to be determined.

#### Giantin

Another golgin present at the cisternal rims is giantin, the largest golgin, which, like golgin-84 and CASP, is anchored to the Golgi membrane by a carboxy-terminal transmembrane anchor (Linstedt et al., 1995). It has yet to be shown that giantin can tether transport vesicles in cells. If it does, then it is likely to tether certain types of intra-Golgi vesicles (Sonnichsen et al., 1998), although it may also play a role in laterally linking Golgi cisternae within the Golgi ribbon (Koreishi et al., 2013). Loss of giantin in cultured cell models gives a mild cilia and secretory trafficking phenotype, with little apparent effect on Golgi morphology (Koreishi et al., 2013; Bergen et al., 2017). Giantin is absent from D. melanogaster and C. elegans, but several vertebrate models have been generated to study the functions of the protein in vivo. Loss of giantin in zebrafish manifests as changes in cilia number and morphology (Bergen et al., 2017). Although the underlying mechanism remains to be determined, it is possible defective trafficking to the cilium may be responsible for this phenotype. There is also a mild skeletal phenotype in giantin-deficient zebrafish, which is reminiscent of a congenital disorder of glycosylation known as hyperphosphatemic familial tumoral calcinosis (HFTC) in humans (Stevenson et al., 2017). This phenotype may be explained by the loss of expression of the glycosyltransferase GALNT3 in the giantindeficient zebrafish, whose mutation in humans causes HFTC (Topaz et al., 2004). Interestingly, GALNT3 is one of 22 glycosyltransferases whose expression is altered upon loss of giantin, which is indicative of an adaptive or compensatory response to chronic loss of the protein. This result lends support to the idea that giantin is important for Golgi function, possibly in enzyme recycling within the Golgi stack, and also indicates that chronic loss of a Golgi protein can induce adaptive or compensatory responses to alleviate the phenotype. It also raises the possibility that adaptive or compensatory responses can account for lack of phenotypes seen in other golgin knockout models, either within certain tissues and/or throughout the organism.

Loss of mammalian giantin, also called GOLGB1, in mouse knockout models, has revealed craniofacial defects, including a cleft palate (Lan et al., 2016; McGee et al., 2017). There is reduced accumulation of the matrix glycosaminoglycan (GAG) hyaluronan and reduced protein glycosylation, consistent with giantin functioning to maintain cargo protein glycosylation and GAG synthesis at the Golgi, which in turn is important for proper matrix assembly and formation of the palate (Lan et al., 2016). Thus, although giantin is widely expressed throughout the body, the knockout phenotype is restricted to certain tissues. The combined knockout of giantin and golgin-84 gives the same phenotype as loss of giantin alone, arguing against the possibility that these golgins function in a redundant manner in vivo (McGee et al., 2017). In contrast to knockout mouse models, rats lacking giantin display a much more severe phenotype, manifesting as

an embryonic lethal osteochondrodysplasia, with systemic oedema and shortening of the long bones in addition to craniofacial defects and a cleft palate (Katayama et al., 2011). The chondrocytes from these animals have a swollen ER and disrupted Golgi, reduced GAG production, and altered production of extracellular matrix proteins (Kikukawa et al., 1990, 1991; Katayama et al., 2018). Thus, giantin would appear to be important for proper glycosylation and possibly secretion of extracellular matrix proteins and GAGs, which in turn is important for skeletal development, albeit with varying degrees of phenotypic severity in the different vertebrate models.

#### TMF

TMF is also present at the rims of the Golgi cisternae, but is more enriched towards the trans-side of the Golgi compared to golgin-84 and giantin, consistent with its recruitment to the membrane by Rab6 (Fridmann-Sirkis et al., 2004; Yamane et al., 2007). TMF is competent to tether intra-Golgi transport vesicles (Wong and Munro, 2014), which is mediated by an amino-terminal tethering motif, and unlike other golgins, there is also a separate tethering motif lying within the central region of the protein (Wong et al., 2017). Depletion of TMF from cells results in altered Golgi morphology and displacement of the Golgi enzyme GalNac-T2, consistent with a role in retrograde intra-Golgi transport (Fridmann-Sirkis et al., 2004; Yamane et al., 2007). There are orthologs of TMF in D. melanogaster and C. elegans, although they have yet to be analyzed at the functional level. TMF knockout mice are viable and healthy, although male mice are sterile (Lerer-Goldshtein et al., 2010). Loss of TMF does, however, affect the composition of mucus within the colon, which is comprised of the heavily glycosylated mucin proteins. There is altered post-translation modification and secretion of the MUC2 mucin from goblet cells, which in turn affects interaction between the colonic epithelium and gut microbiome (Bel et al., 2012, 2014). The mechanism by which loss of TMF alters MUC2 processing and trafficking awaits further investigation, but is likely to involve disrupted recycling of Golgi enzymes within the Golgi stack. The male sterility phenotype in TMF knockout mice arises from defective acrosome biogenesis. The Golgi is mis-positioned in developing spermatids upon loss of TMF, and there is reduced tethering of pro-acrosomal vesicles, explaining the lack of a functional acrosome in these cells (Lerer-Goldshtein et al., 2010; Elkis et al., 2015). The results are consistent with a role for TMF in membrane delivery into the forming newly forming acrosome during spermatogenesis.

#### Golgins at the Trans-Golgi

There are four golgins at the trans-Golgi that function in vesicle tethering at this compartment, namely golgin-245, golgin-97, GCC185 and GCC88 (Cheung and Pfeffer, 2016; Gillingham and Munro, 2016). They are all anchored to the trans-Golgi membrane via a carboxy-terminal GRIP domain, which binds to membrane-associated Arl1 (Munro, 2011; Witkos and Lowe, 2015). GCC185, in addition to a vesicle trafficking function, also contributes to microtubule nucleation at the Golgi through binding to the microtubule binding proteins CLASP1 and CLASP2 (Efimov et al., 2007). There appears to be overlapping functionality in vesicle trafficking between the GRIP domain golgins in that they can tether vesicles carrying the same endosome-derived model cargo protein CI-MPR (Wong and Munro, 2014; Cheung et al., 2015). However, it should be noted that the same cargo protein may be carried by more than one type of transport vesicle. Indeed, it was recently demonstrated that GCC88 tethers only a sub-population of vesicles carrying CI-MPR, which are presumably different to the CI-MPR containing vesicles tethered by the other trans-Golgi golgins (Cui et al., 2019). This finding is consistent with the different vesicle-tethering motif found in GCC88 compared to that in the other golgins. In this regard, golgin-245 and golgin-97 share a common tethering motif, binding to the RabGAP-like protein TBC1D23, which in turn binds the WASH complex on endosome-derived vesicles to mediate tethering (Shin et al., 2017). Although all of the GRIP-domain golgins have an ortholog in D. melanogaster, and all except golgin-97 have an ortholog in C. elegans, the functional requirement for these golgins in either model organism has yet to be tested. As of yet, no vertebrate knockout models for any of the GRIP-domain golgins have been generated.

#### LOSS OF GOLGIN FUNCTION AS A CAUSE OF HUMAN DISEASE

With the advent of "next-generation sequencing" (NGS), there has been an explosion in the discovery of rare genetic variants responsible for human disease (Boycott et al., 2013). Amongst the many pathogenic variants that have been discovered are those within genes encoding Golgi proteins (Zappa et al., 2018). The identification of these rare variants, and the study of patients carrying them, are important not only from a clinical perspective, but can also prove extremely informative with regard to deciphering the physiological roles of the encoded proteins. We may expect the phenotypes to mirror those seen in animal models, but this is not always the case, which can be due to a variety of reasons, including differences in organismal physiology, the expression of compensatory genes, or adaptive responses in the different organisms. To date, mutations in only two of the golgins, GMAP-210 and GM130, have been linked to disease in humans, as described further below (see also **Table 1**).

#### GMAP-210

Mutations in GMAP-210 that result in loss of protein expression cause the severe neonatal lethal autosomal recessive disorder achondrogenesis type 1A (ACG1A) (Smits et al., 2010). ACG1A is manifest as skeletal dysplasia, with shortening of the bones and reduced ossification, accompanied by craniofacial abnormalities and other skeletal defects, which is similar to the phenotype seen in knockout mice (Molz and Spycher, 1980; Smits et al., 2010). Within ACG1A chondrocytes, there is swelling of the ER and Golgi fragmentation, accompanied

by reduced secretion of extracellular matrix (Smits et al., 2010; Wehrle et al., 2018, 2019) (**Figure 4**). This is further compounded by reduced differentiation of the chondrocytes, resulting in reduced cell numbers, such that matrix protein secretion is greatly reduced in the bone growth plates (Wehrle et al., 2019). The results are consistent with human GMAP-210 playing an important role in the modification and trafficking of extracellular matrix proteins, as seen in the mouse. It is also likely that ciliary impairment might contribute to the ACG1A phenotype, possibly via control of chondrocyte differentiation, although this remains to be confirmed. Recently, a second genetic cause of ACG1A has been identified, with mutations in the lamin B receptor (LBR) shown to cause an identical clinical phenotype to that seen upon loss of GMAP-210 (Wehrle et al., 2018). LBR is localized to the ER and nuclear envelope, where it not only binds to lamin B, but is also a key enzyme in the production of cholesterol (Tsai et al., 2016). The loss of LBR results in reduced synthesis of cholesterol (Tsai et al., 2016), which is known to be important for secretory protein traffic (Stuven et al., 2003; Ridsdale et al., 2006). Hence, depletion of cholesterol upon LBR deficiency impairs transport within the secretory pathway, pointing to a common pathogenic mechanism to that seen upon loss of GMAP-210 (Wehrle et al., 2018). Mutations in LBR are also associated with a different skeletal disorder known as Greenberg syndrome (Greenberg et al., 1988). The differences in phenotype between ACG1A and Greenberg syndrome may be explained by residual expression of truncated LBR in the latter, which, although it remains to be demonstrated, could induce additional cellular phenotypes to those seen upon the complete loss of LBR in ACG1A. Nevertheless, the demonstration of altered secretory trafficking in LBR-associated ACG1A suggests the same defect may also account for, or contribute to, the Greenberg dysplasia phenotype.

Hypomorphic mutations in GMAP-210 result in a milder skeletal disorder known as odontochondrodysplasia (ODCD) (Wehrle et al., 2019). ODCD manifests as skeletal abnormalities including shortening of the tubular limb bones and scoliosis, accompanied by dental abnormalities (Goldblatt et al., 1991). The milder phenotype of ODCD compared to ACG1A is due to the residual expression of GMAP-210 in the former, allowing for a less severe impairment of secretory protein traffic, and hence a less severe effect on extracellular matrix deposition during bone development (Wehrle et al., 2019). At the cellular level, the severity of Golgi disruption and trafficking deficiency correlates well with the amount of GMAP-210 lost from cells, consistent with the notion that ODCD simply reflects a less severe form of GMAP-210 deficiency.

#### GM130

Mutation of GM130 in humans appears to manifest as a neuromuscular disorder, with developmental delay, progressive microcephaly, and muscular dystrophy (Shamseldin et al., 2016). The mechanisms underlying these phenotypes remain to be determined, but based upon studies in zebrafish and mouse models it is likely defects in secretory trafficking, and possibly microtubule organization, are responsible. It should be noted that only a single patient with this disorder has been identified, and although the mutation in this patient resulted in a loss of GM130 protein, the identification of additional patients will be required to establish this condition as a GM130-dependent disorder. Reduced expression of GM130 has also been implicated in tumorigenesis (Baschieri et al., 2014; Baschieri and Farhan, 2015). GM130 levels are reduced in breast cancer, and depletion of GM130 in cancer cells increases cell migration, which may be relevant for the human disease. It has been proposed increased cell migration arises from altered Cdc42 activity downstream from RasGRF, to which GM130 normally binds.

## Other Golgi Trafficking Proteins and Human Disease

Mutations in many Golgi proteins have been identified as the cause of disease in humans (Zappa et al., 2018). Of interest here are those within the COG proteins, which form the multisubunit COG vesicle-tethering complex that acts at several Golgi cisternae to mediate recycling of Golgi-resident enzymes within the Golgi stack (Willett et al., 2013). COG interacts with several golgins, and can also interact with Rab GTPases and SNAREs, and likely acts downstream of the golgins to facilitate the transition from golgin-mediated vesicle capture to SNAREdependent membrane fusion (Willett et al., 2013; Witkos and Lowe, 2017). Mutation of COG subunits causes defects in the glycosylation of cargo proteins within the Golgi, which is a consequence of impaired recycling of Golgi enzymes within the Golgi stack. This manifests as various tissue specific defects grouped under the umbrella of type II congenital disorders of glycosylation (CDGs) (Zeevaert et al., 2008). The nature and severity of the phenotype depends upon the COG subunit mutated, and the nature of the mutation itself. The COG CDGs indicate that impairment of intra-Golgi trafficking can manifest as disease, raising the possibility that additional CDGs may be caused by mutation of golgins mediating intra-Golgi trafficking. The trans-Golgi protein GORAB, although originally described as a golgin (Hennies et al., 2008), is in fact a scaffolding protein for the COPI vesicle coat, and likely participates in vesicle formation as opposed to vesicle tethering (Witkos et al., 2019). Loss of GORAB causes the skin and bone disorder gerodermia osteodysplastica, which can be attributed to defective glycosylation of extracellular matrix proteins in these tissues, and as such can also be classified as a CDG (Hennies et al., 2008; Chan et al., 2018).

# SUMMARY AND FUTURE DIRECTIONS

As can be seen from the study of golgins, it is not trivial to extrapolate findings generated in cell models to whole organism physiology. Nevertheless, in many cases, it is possible to derive explanations for the organismal phenotype based upon knowledge gained from in vitro cell biological studies. Hence, in most cases we see defective secretory traffic and/or cargo protein glycosylation in the affected cell types in in vivo models and human patients. What is less clear is why certain

tethering at the cis-Golgi results in a deficit in both transport and modification of extracellular matrix proteins, which in turn causes a defective cell matrix to form outside the cells. This manifests as the severe skeletal dysplasia achondrogenesis Type 1A. There is also a loss of IFT20-dependent transport of cargo proteins to the cilium, resulting in defective ciliary function and impaired cell differentiation, such that matrix production is also decreased, further compounding the phenotype.

cell types or cargo proteins within the organism have a greater sensitivity to loss of a particular golgin compared to another. A trivial explanation is that tissue- or cell-type specific differences in golgin expression are responsible. So, for golgins with overlapping specificity, we may envisage differences in their expression to account for differences in phenotypic severity between different tissues or cell types. This may be further complicated by the presence of different golgin transcripts in different cell types, which may vary during development and also between species. For example, GMAP-210 is expressed as several transcripts in human cells as a result of alternative splicing, with some of the variants lacking regions encoding functionally important regions of the protein (Ramos-Morales et al., 2001; Wehrle et al., 2019). Splicing of GMAP-210 varies between cell types and also during chondrocyte differentiation, consistent with it being functionally important (Wehrle et al., 2019). Interestingly, there are several GM130-related transcripts that are expressed in primates, but absent from non-primate species, probably as a consequence of genomic rearrangements that occurred during primate evolution (discussed in Munro, 2011). Hence, this golgin may have evolved to have several isoforms, whose functions have yet to be investigated.

Redundancy in golgin function is likely to occur in most cells and during most developmental stages, accounting for the lack or restricted nature of most phenotypes seen upon golgin knockout. However, the presence of an overt phenotype upon loss of many of the golgins, indicates that in vivo, redundancy is rarely complete. To better understand the degree of redundancy between golgins in an in vivo context, it will be interesting to perform combinatorial knockouts of the golgins. To date this has only been performed in mice with golgin-84 and giantin, which failed to reveal any functional redundancy between these golgins (Bird et al., 2018). This

fcell-07-00094 June 14, 2019 Time: 17:27 # 9

type of combinatorial approach will be easier to perform in more tractable species such as D. melanogaster or C. elegans, and based upon the findings in these species, it will also be important to explore further redundancy in vertebrate models. A good starting point might be the golgins with similar aminoterminal vesicle tethering motifs (Wong et al., 2017), which would be predicted to tether vesicles carrying similar cargo proteins. However, it is also important to consider that several of the golgins participate in functions other than vesicle tethering, and hence, in certain cases, the phenotypes observed in vivo will also depend upon cellular defects that are unrelated to vesicle tethering.

Another factor to consider is the nature of the cargo proteins themselves, and also the rate of their production, modification, and secretion. It is interesting that the extracellular matrix is particularly sensitive to loss of two of the golgins, GMAP-210 (Smits et al., 2010; Wehrle et al., 2018, 2019) and giantin (Kikukawa et al., 1990, 1991; Lan et al., 2016; Katayama et al., 2018). Matrix proteins tend to be bulky, complex, and highly modified within the Golgi, and they are produced at very high levels during skeletal development. It is therefore reasonable to propose that their production is intrinsically more sensitive to loss of golgin function, although why loss of only GMAP-210 and giantin manifests as a matrix phenotype is still unclear. This might suggest tethering of vesicles containing matrix proteins, or the enzymes that modify them, is mediated by these golgins. Defective trafficking to the cilium is also likely to contribute to the phenotype considering that both golgins appear to function in this process (Follit et al., 2006; Bergen et al., 2017; Monis et al., 2017), and alterations in ciliary signaling are known to impact upon skeletal development (Braun and Hildebrandt, 2017). Similarly, loss of GM130 manifests as a phenotype in Purkinje neurons, even though the other cis-Golgi golgins are abundantly expressed in these cells (Liu et al., 2017). Again, this cell-type sensitivity may reflect the high secretory capacity of Purkinje neurons, especially during dendritic growth and maintenance, but also suggests that tethering of vesicles carrying Purkinje cell dendrite-specific cargoes is particularly reliant upon GM130. However, it could also reflect roles for GM130 in other cellular processes such as microtubule nucleation (Rivero et al., 2009), which is known to be important for dendrite morphogenesis (Kapitein and Hoogenraad, 2015). Further studies will be required to address these possibilities, and to determine the degree to which impaired tethering of vesicles carrying different cargoes can account for the in vivo phenotypes seen upon loss of different golgins. An attractive approach would be to recapitulate the vesicle-tethering assay that has been used so effectively in vitro (Wong and Munro, 2014; Shin et al., 2017; Wong et al., 2017), in an in vivo context, to identify the spectrum of cargoes in different tissues that are tethered by the different golgins.

Another factor to consider when it comes to studying golgins in vivo is the propensity for cells and organisms to adapt to the chronic loss of expression of a particular protein. This has been clearly observed in giantin knockout models, where there are compensatory changes in the expression of numerous Golgi enzymes, which is likely to ameliorate the cellular and organismal phenotype (Stevenson et al., 2017). Similarly, analysis of GMAP-210-deficient chondrocytes has revealed changes in the expression levels of both secreted extracellular matrix proteins and Golgi resident trafficking machinery, which includes other golgins (Bird et al., 2018). Hence, the manifestation of the phenotype is a consequence of not only the trafficking defect directly induced by loss of the golgin itself, but also the cell's response to this defect. This complicates experiments to directly analyze golgin function in vivo, but is less of a concern when it comes to modeling human disease, which is nearly always due to chronic loss of protein function. Although adaptive or compensatory responses can complicate interpretation of phenotypes, they can also provide new insight into golgin function. For example, changes in expression of trafficking machinery, Golgi enzymes or cargo proteins can allow for a more global understanding of how golgins function within the secretory pathway (Stevenson et al., 2017; Bird et al., 2018). Other changes within the transcriptome more generally can also provide a way to better understand the cellular consequences downstream from the loss of a particular golgin, which may lead to the development of new hypotheses regarding golgin function within cells and in the organism. In terms of more directly analyzing golgin function in vivo, methods to acutely remove the protein, such as the auxin-inducible degron system should prove to be informative (Nishimura et al., 2009). This approach is now more tractable given the advent of CRISPR-Cas9 technology to knock-in the degron tag.

It is interesting that spermatogenesis is particularly sensitive to loss of golgin function. Three golgins have been shown to be important for acrosome formation (GM130, golgin-160, TMF) (Lerer-Goldshtein et al., 2010; Bentson et al., 2013; Han et al., 2017), underlining the importance of Golgi-dependent trafficking for the biogenesis of this organelle. It is possible that other golgins also contribute to this process. Golgin-84 is highly enriched in the testis (Bascom et al., 1999), consistent with a role in spermatogenesis, although it does not appear to be absolutely required for this process as male golgin-84 knockout mice are fertile (McGee et al., 2017). GMAP-210 has been localized to pro-acrosomal vesicles during spermatogenesis, suggesting a role in acrosome biogenesis (Kierszenbaum et al., 2011), but the functional requirement for GMAP-210 in this process has yet to be tested in an animal model. Interestingly, germline specific loss of the GMAP-210 interaction partner IFT20 causes male infertility, although whether acrosome biogenesis is affected in these mice has not been determined (Zhang et al., 2016). For remainder of the golgins, we await the development of suitable animal models to assess their involvement in spermatogenesis. For those golgins that function in spermatogenesis, it will be important to better define the mechanisms by which they act. This may help explain why some golgins appear dispensable for this process (golgin-84 and giantin) (McGee et al., 2017), whereas others are not (GM130, golgin-160, TMF) (Lerer-Goldshtein et al., 2010;

Bentson et al., 2013; Han et al., 2017). It would also provide new mechanistic insight into the process of acrosome formation. In addition to differences in testis-specific expression, we may expect pro-acrosomal vesicle trafficking to have a functional requirement for certain golgins over others, and there is also the possibility of redundancy in golgin function during this process. Further investigation will be required to distinguish between these possibilities.

#### AUTHOR CONTRIBUTIONS

ML devised and wrote the manuscript.

#### REFERENCES


#### FUNDING

Work in the Lowe lab is funded by the BBSRC (BB/S014799/1) and Lowe Syndrome Trust (MU/ML/2016 and ML/MU/LST NOV/18).

#### ACKNOWLEDGMENTS

I would like to acknowledge Hardy Chan, Uwe Kornak, Anika Wehrle, and Ekkehart Lausch for interesting discussions, and Anika Wehrle and Tomasz Witkos for critically reading the article.

model for gerodermia osteodysplastica. PLoS Genet. 14:e1007242. doi: 10.1371/ journal.pgen.1007242


transcription factor, is a Golgi membrane protein related to giantin. Mol. Biol. Cell 13, 3761–3774. doi: 10.1091/mbc.E02-06-0349


reticulum and then transported to the Golgi apparatus. Proc. Natl. Acad. Sci. U.S.A. 92, 5102–5105. doi: 10.1073/pnas.92.11.5102


fcell-07-00094 June 14, 2019 Time: 17:27 # 12


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

Copyright © 2019 Lowe. 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.

fcell-07-00094 June 14, 2019 Time: 17:27 # 13

# Golgi Dynamics: The Morphology of the Mammalian Golgi Apparatus in Health and Disease

#### Christian Makhoul, Prajakta Gosavi and Paul A. Gleeson\*

The Department of Biochemistry and Molecular Biology, Bio21 Molecular Science and Biotechnology Institute, The University of Melbourne, Melbourne, VIC, Australia

#### Edited by:

Vladimir Lupashin, University of Arkansas for Medical Sciences, United States

#### Reviewed by:

Roman Polishchuk, Telethon Institute of Genetics and Medicine, Italy Benjamin S. Glick, The University of Chicago, United States Nobuhiro Nakamura, Kyoto Sangyo University, Japan

\*Correspondence:

Paul A. Gleeson pgleeson@unimelb.edu.au

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 16 March 2019 Accepted: 03 June 2019 Published: 03 July 2019

#### Citation:

Makhoul C, Gosavi P and Gleeson PA (2019) Golgi Dynamics: The Morphology of the Mammalian Golgi Apparatus in Health and Disease. Front. Cell Dev. Biol. 7:112. doi: 10.3389/fcell.2019.00112 In vertebrate cells the Golgi consists of individual stacks fused together into a compact ribbon structure. The function of the ribbon structure of the Golgi has only begun to be appreciated (De Matteis et al., 2008; Gosavi and Gleeson, 2017; Wei and Seemann, 2017). Recent advances have identified a role for the Golgi in the regulation of a broad range of cellular processes and of particular interest is that the modulation of the Golgi ribbon is associated with regulation of a number of signaling pathways (Makhoul et al., 2018). Various cell responses, such as inflammation, and various disorders and diseases, including neurodegeneration and cancer, are associated with the loss of the Golgi ribbon and the appearance of a dispersed or semi-dispersed Golgi. Often the dispersed Golgi is referred to as a "fragmented" morphology. However, the description of a dispersed Golgi ribbon as "fragmented" is inadequate as it does not accurately define the morphological state of the Golgi. This issue is particularly relevant as there are an increasing number of reports describing Golgi fragmentation under physiological and pathological conditions. Knowledge of the precise Golgi architecture is relevant to an appreciation of the functional status of the Golgi apparatus and the underlying molecular mechanism for the contribution of the Golgi to different cellular processes. Here we propose a classification to define the various morphological states of the nonribbon architecture of the Golgi in mammalian cells as a guide to more precisely define the relationship between the morphological and functional status of this organelle.

#### Keywords: Golgi ribbon, Golgi morphology, cell sensing, Golgi stacks, signaling

# BACKGROUND

The basic unit of the Golgi apparatus is usually considered to be a stack of cisternae which is highly polarized, with the cis-face receiving cargo from the ER and the trans-face of the stack, the TGN, associated with sorting cargo for post-Golgi export (Boncompain and Perez, 2013). However, the organization of these mini-stacks in the cell varies amongst different organisms. In plants and invertebrates individual Golgi stacks are scattered independently throughout the cytoplasm whereas in most vertebrate cells during interphase, individual Golgi stacks are fused

**Abbreviations:** GM130, cis-Golgi matrix protein; GMAP210, Golgi microtubule associated protein; GOLPH3, Golgi phosphoprotein 3; GRASP, Golgi reassembly stacking protein; MTOC, microtubule organizing center; mTOR, mechanistic target of rapamycin; PI4P, phosphatidylinositol-4-phosphate; TGN, trans-Golgi network.

together into a compact ribbon structure located in close proximity to the MTOC (Wei and Seemann, 2017). The structure of the Golgi ribbon in mammalian cells is best revealed by electron microscopy (Rambourg and Clermont, 1997); high resolution optical microscopy using cis and trans markers can also detect the ribbon organization (Gosavi et al., 2018). An important question under investigation in the field is the relevance of the Golgi ribbon structure and the functional differences of the Golgi ribbon compared with "isolated Golgi mini-stacks" or other states of Golgi architecture. In other words what is the evolutionary advantage of the more complex ribbon morphology of the Golgi apparatus in vertebrate cells and what functions may be regulated by a transition to a nonribbon structure?

Studies over the past few years are revealing that Golgi membranes provide a platform for the regulation of a range of cellular processes including cell polarization (Kupfer et al., 1983), directed migration (Millarte and Farhan, 2012), stress (Sasaki and Yoshida, 2015), DNA repair (Farber-Katz et al., 2014), mitosis (Rabouille and Kondylis, 2007), metabolism (Abdel Rahman et al., 2015), pro-inflammatory responses (Chen and Chen, 2018) and autophagy (Yamamoto et al., 2012; Lamb et al., 2013). Indeed, in mammalian cells there is now considerable evidence that the Golgi, like other intracellular organelles, can act as a cell sensor (Farhan and Rabouille, 2011; Mayinger, 2011; Millarte and Farhan, 2012; Sasaki and Yoshida, 2015; Luini and Parashuraman, 2016; Gosavi and Gleeson, 2017; Makhoul et al., 2018). It is also becoming clear that the precise morphology of the Golgi is relevant to the regulation of a number of these cell processes (Makhoul et al., 2018). The association of Golgi morphology with signaling was borne out of a genome wide kinome and phosphatome screen which identified a large cohort of kinases and phosphatase (20% of the total in the genome) that influenced the morphology of the Golgi (Chia et al., 2012). The changes in Golgi morphology included either fragmentation of the Golgi (loss of Golgi ribbon) or the formation of a very compact and condensed Golgi in the perinuclear location. The relevance of actin in the regulation of Golgi morphology was highlighted in this study by the identification of a number of kinases, for example ROCK1 and PAK1, which regulate actin dynamics and modulate Golgi structure (Chia et al., 2012). Other genome wide analyses have also highlighted the likelihood that the Golgi can receive and transmit a wide variety of signals that could influence, not only membrane transport pathways, but also other processes, apoptosis, mitosis, autophagy and stress responses (Farhan et al., 2010; Millarte et al., 2015).

The Golgi ribbon structure is highly dynamic and can undergo very rapid remodeling during a range of different conditions. For example, during mitosis the disassembly of the Golgi ribbon is an early event in G2/M transition and plays an important role as a cell cycle checkpoint in promoting mitotic entry (Wei and Seemann, 2010; Corda et al., 2012). The regulation of Golgi dynamics is mediated by interactions between molecular scaffolds located on the Golgi membrane and the cytoskeleton. MT dynamics can regulate the location of the Golgi ribbon at the centrosome and the repositioning of the Golgi to facilitate polarized trafficking and directed secretion (Millarte and Farhan, 2012; Sanders and Kaverina, 2015). In addition, membrane components of the Golgi can also nucleate and stabilize microtubules at both the cis- and trans-Golgi, and therefore the Golgi itself is also a MTOC (Efimov et al., 2007; Wu et al., 2016). Actin-mediated processes also contribute to the form and function of the Golgi and at least nine Golgi-localized molecular scaffolds have been identified which interact with the actin cytoskeleton [see review (Gosavi and Gleeson, 2017)]. Enhancement of actin polymerization at the Golgi results in dispersal of the ribbon, whereas inhibition of actin polymerization with specific drugs such as latrunculin A results in compaction of the Golgi (Lazaro-Dieguez et al., 2006; Makhoul et al., 2018, 2019). Given these regulatory networks that modulate the Golgi ribbon structure, we consider it very plausible that the balance between the Golgi ribbon and Golgi mini-stacks may define both the qualitative and quantitative responses of signaling pathways. The connection between Golgi morphology and signaling also has important ramifications on understanding the molecular basis of a number of diseases that are associated with the loss of the Golgi ribbon and the appearance of a dispersed Golgi. For example, the survival of some cancer cells has been shown to be associated with a dispersed Golgi which reduces the level of apoptosis (Farber-Katz et al., 2014; Petrosyan, 2015).

#### WHAT DOES GOLGI FRAGMENTATION MEAN?

The term "Golgi fragmentation" is commonly used to describe the morphological status of a dispersed Golgi in mammalian cells, as detected by optical microscopy stained with Golgi markers. A dispersed Golgi is often observed in experimental systems for example when cells are treated with drugs to perturb the cytoskeleton, e.g., nocodazole (Wei and Seemann, 2010), when membrane flux is perturbed, or when components of Golgi transport machinery, or the cytoskeletal interaction system, are knocked down, knocked out, or overexpressed (Zappa et al., 2018). In various physiological states, for example stress (Serebrenik et al., 2018) and pathological conditions, particularly cancer and neurodegeneration, the Golgi has often lost the typical compact juxtanuclear location and is observed by confocal microscopy as dispersed structures throughout the cytoplasm [see reviews by Gosavi and Gleeson (2017), Wei and Seemann (2017); in these pathological conditions the Golgi apparatus is also referred to as fragmented]. There are an increasing number of reports describing Golgi fragmentation under physiological and pathological conditions (**Figure 1A**). However, a problem with the use of the term "fragmented" is that it implies that the structural integrity of the Golgi is lost and that the morphology associated with the "fragmented" structure represents a disintegrated, abnormal or destroyed organelle. In a number of cases this is clearly misleading as the individual Golgi stacks may remain intact and can maintain the classical functions of the organelle, namely glycosylation and membrane transport. Indeed, dispersed Golgi

mini-stacks occur in some specialized cells, such as differentiated myoblasts (Lu et al., 2001), differentiated neurons which contain individual Golgi stacks or "outposts" along dendrites (Lasiecka and Winckler, 2011), gastric parietal cells (Gunn et al., 2011), and uroepithelial cells of the urinary bladder (Kreft et al., 2010), without apparent deficiency in membrane transport and glycosylation. Rather, the dispersal of the Golgi ribbon associated with experimental and pathological conditions may reflect a shift of dynamic balance between the compact ribbon morphology and the individual Golgi mini-stacks or can result in perturbation of the ribbon architecture as well as the integrity of the Golgi stack. This is a relevant issue as the precise morphological status of the Golgi will very likely influence, in some cases the efficiency of transport and glycosylation (Puthenveedu et al., 2006), and in other cases a variety of signaling networks but not necessarily transport or glycosylation. It is important to differentiate between intact Golgi mini-stacks and loss of the integrity of the Golgi stacks in defining what is meant by a "fragmented" Golgi. Here we review the different Golgi morphologies that have been detected and characterized in experimental, physiological and pathological setting.

# RETHINKING THE TERMINOLOGY OF GOLGI MORPHOLOGICAL STATES

The structures of the Golgi fragments differ depending on the nature of pathway involved to perturb or modulate the Golgi ribbon. It is important to have a better ultrastructural characterisation of the Golgi "fragments" following loss of the Golgi ribbon as the functional outcome is likely to be very different depending on the precise Golgi structures. Aside from conditions that result in an elongated Golgi ribbon, we can identify from the literature at least 4 different scenarios associated with Golgi ribbon "fragmentation." These are depicted in **Figure 2** and described as follows:


We do not infer that **Figure 2** represents the only morphologies of Golgi "fragments" and it is possible that additional scenarios will be identified as the structures of Golgi "fragments" are investigated more extensively.

the dispersed Golgi stacks is compromised with shortened cisternae, swelling of cisternae and increase in Golgi associated tubules and vesicles; (3) A scenario where there is dispersal of one Golgi compartment. Here the TGN is selectively dispersed throughout the cytoplasm whereas the remainder of the stack remains in a ribbon structure; (4) Scenario where there is loss of ribbon and stacks with Golgi membranes dispersed predominantly as tubules and vesicles. Numbers refer to the classification of the Golgi morphologies given in text.

## EXAMPLES OF A RELATIONSHIP BETWEEN GOLGI MORPHOLOGY AND CELL PROCESS

The co-ordination of changes in Golgi morphology and various cell processes has received considerable attention. For a more detailed summary of the background information of the processes influenced by the changes in Golgi morphology, such as trafficking, glycosylation, stress, DNA repair, the reader is referred to a number of recent reviews (Farhan and Rabouille, 2011; Millarte and Farhan, 2012; Sasaki and Yoshida, 2015; Gosavi and Gleeson, 2017; Makhoul et al., 2019). Below are some examples to highlight the range of cell processes that are regulated or co-ordinated by different morphological states of the Golgi. In a number of other instances, such as some cancers and stress responses, the Golgi ribbon is dispersed as fragments, however, the morphology of these Golgi fragments has not been well characterized. The discussion herein will be focused on the examples where the Golgi morphology is well defined.

#### DNA Repair and Cancer

There is an intimate association between Golgi morphology and the DNA damage response (Farber-Katz et al., 2014). The Golgi membrane tether, GOLPH3, is an oncogene and overexpression of GOLPH3 results in enhanced cell survival following DNA damage (Scott et al., 2009; Farber-Katz et al., 2014). Conversely, loss of GOLPH3 prevents the dispersion of the Golgi ribbon, enhances the Golgi ribbon and promotes apoptosis after DNA damage. The DNA damage response is mediated by the kinase DNA-PK, which phosphorylates GOLPH3 and promotes Golgi "fragmentation" by enhancing actin polymerization at Golgi membranes (Dippold et al., 2009). mTOR is modulated by the changes in Golgi morphology mediated by GOLPH3 (Scott et al., 2009) and is likely to contribute to the outcome of the DNA response. Hence, the precise Golgi structure is tied to cell survival and apoptosis. The identity of the Golgi structures following the dispersal of the Golgi ribbon by phospho-GOLPH3 remain to be characterized. Clearly future studies examining the relationship between Golgi fragmentation and mTOR signaling in cancer cells will be well worthwhile.

### GENETIC DISORDERS ASSOCIATED WITH ALTERED GOLGI MORPHOLOGY

Many diseases have been identified with monogenic disorders caused by inherited mutations of either components associated

with transport machinery or of the enzymes resident in the Golgi. Many of these diseases are associated with fragmentation of the Golgi ribbon, for example, defects in the conserved oligomeric Golgi complex (COG) in congenital disorders of glycosylation (Miller and Ungar, 2012). In many cases the disorders are associated with pathologies restricted to a limited number of organs or tissues. The basis for tissue specificity is poorly understood, but likely due to deficiencies in glycosylation and secretion and also alterations in signaling networks associated with the loss of the Golgi ribbon, such as Golgi stress responses. The Golgi in all COG subunit knockout cell lines show moderate to severe change in morphology as characterized by EM, associated with loss of the ribbon, dilated cisternae and in some cases disruption of the mini-stacks. The application of EM was fundamental in defining the morphological changes (Blackburn and Lupashin, 2016).

#### Neurodegenerative Diseases

The status of the Golgi in neurodegenerative diseases has recently received considerable attention (**Figure 1B**). Loss of the Golgi ribbon is a common feature of many neurodegenerative diseases including Alzheimer's disease, Huntington disease, amyotrophic lateral sclerosis and Parkinson's disease (Gonatas et al., 2006; Haase and Rabouille, 2015; Rabouille and Haase, 2015; Sundaramoorthy et al., 2015). It is very likely that the perturbations in the Golgi architecture in these diseases contributes to the pathological processes. In most cases the precise morphological structure of the Golgi fragments in these neurodegenerative diseases have not been defined. However, two experimental systems have recently investigated changes in Golgi structure and neuronal degeneration in some detail. Firstly, conditional knock out of GM130, a structural golgin which regulates the Golgi ribbon, in the central nervous system was demonstrated to cause Golgi fragmentation, atrophy of dendrites and neuronal degeneration in mice (Liu et al., 2017). EM analysis of the GM130 KO Purkinje cells showed a reduction of Golgi cisternal length and stacking and, in addition, the typical Golgi dendrite outposts were absenting in these GM130 -/- Purkinje cells (Liu et al., 2017). In a second study, hippocampal neurons from mice transgenic for the Swedish mutation of amyloid precursor protein (APP) and a mutant presenilin 1 subunit of γ-secretase, where both mutations are associated with early onset Alzheimer's disease, revealed extensive Golgi fragmentation by optical microscopy (Joshi et al., 2014). Quantitative EM of neurons in these mice showed a reduction in the number and length of the cisternae in the stacks compared with neurons from wild type mice. In addition, cisternae were swollen. The changes in Golgi morphology in these primary neurons was shown to be directly associated with the elevated level of amyloid-β production. Primary neurons treated with synthetic amyloid-β also showed similar fragmentation of the Golgi as well as an increase in tubulovesicular structures associated with Golgi cisternae compared with untreated cells. The loss of the Golgi ribbon was due to phosphorylation of GRASP65, a Golgi structural protein

which plays a key role in disassembly of the Golgi ribbon and stacks in mitosis (Joshi et al., 2014). Comparison of these two studies above is informative as the pathways mediating the changes in Golgi morphology differ in each case leading to differences in the morphology of the Golgi "fragments." Consideration needs to be given as to how these different pathways affecting Golgi morphology influence the downstream responses.

# mTOR Signaling

Our lab has established an experimental approach to perturb the balance between the Golgi ribbon and Golgi mini-stacks by modulating the dose of GCC88, a golgin located at the TGN. This strategy allowed a stable cell line, HeLa-B6, to be established that lack a Golgi ribbon. We have shown that GCC88 regulates the balance between Golgi ribbons and mini-stacks by an actin dependent process (Makhoul et al., 2019) and identified intersectin-1 (ITSN1), a guanine nucleotide exchange factor for cdc42 (Hussain et al., 2001), as an interactor of GCC88 responsible for the loss of the Golgi ribbon (Makhoul et al., 2019). Analyses of HeLa B6 cells, which lack a Golgi ribbon, demonstrated reduced mTOR activity and an associated increase in autophagosome biogenesis (Gosavi et al., 2018). mTOR is one of the major signaling pathways of eukaryotic cells and known to be a negative regulator of autophagy (Wullschleger et al., 2006). Hence, the balance of Golgi stacks to Golgi ribbon has a direct effect on the mTORC1 pathway. The use of cis- and trans- compartment specific markers, and EM tomography, was critical in revealing the morphological changes in the Golgi ribbon.

### Inflammation

Inflammasomes of the innate immune system act as a scaffold for caspase 1-dependent activation of pro-inflammatory cytokines (Broz and Dixit, 2016). The NLRP3 (nucleotidebinding domain, leucine-rich-containing family, pyrin domaincontaining-3) is a versatile inflammasome which can be activated by a range of microbial and non-microbial stimuli resulting in secretion of pro-inflammatory cytokines interleukin 1β (IL-1β) and interleukin 18 (IL-18) and programmed cell death by pyroptosis. A recent study has demonstrated that the activation of the cytosolic NLRP3 by stimuli involves recruitment of NLRP3 to dispersed TGN membranes to facilitate NLRP3 scaffold assembly (Chen and Chen, 2018). The dispersed TGN, but not the underlying other compartments, specifically recruits NLRP3, via PI4P, to assemble the downstream adaptor complex ASC which undergoes polymerization in the perinuclear region before recruiting caspase-1 to activate the downstream signaling pathway (Chen and Chen, 2018). Hence this important finding demonstrates that the modulation of the architecture of the TGN selectively, is critical in the activation of this pathway. The use of compartment specific markers (TGN and cis-Golgi) together with optical and EM analysis was critical in revealing the morphological changes in the Golgi ribbon.

#### CONCLUDING REMARKS

fcell-07-00112 July 2, 2019 Time: 17:43 # 6

The precise architecture of Golgi morphology is defined by the high resolution optical microscopy and by EM. The inclusion of this information in future studies in the field will provide a considerable insight into the dynamics of the Golgi, the pathways for perturbation of the ribbon structure and the functional consequences associated with these different pathways.

In summary, we propose that the use of the term "fragmented Golgi" is inadequate to describe Golgi structures associated with many treatments and conditions and the differences in these Golgi structures are likely to be relevant physiologically. Given the dynamic nature of the Golgi apparatus, it is possible that there may be a balance between different morphological states of the Golgi at any given time i.e., mini-stacks and ribbon structures. Understanding the dynamic balance between the different Golgi morphologies in molecular detail is critical for a full appreciation of this organelle during normal cell processes and also under pathological conditions. It will be fascinating to see what unfolds as we learn more about the cell sensing functions of this complex

#### REFERENCES


organelle and the relationship between these functions and Golgi structures.

#### DATA AVAILABILITY

All datasets analyzed for this study are included in the manuscript and the supplementary files.

#### AUTHOR CONTRIBUTIONS

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

#### FUNDING

This work was supported by funding from the Australian Research Council (DP160102394).

kinase/phosphatase functional screening. J. Cell Biol. 189, 997–1011. doi: 10. 1083/jcb.200912082


Joshi, G., Chi, Y., Huang, Z., and Wang, Y. (2014). Abeta-induced Golgi fragmentation in Alzheimer's disease enhances Abeta production. Proc. Natl. Acad. Sci. U.S.A. 111, E1230–E1239. doi: 10.1073/pnas.1320192111



**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 Makhoul, Gosavi and Gleeson. 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.

#### Edited by:

J. Christopher Fromme, Cornell University, United States

#### Reviewed by:

Roman Polishchuk, Telethon Institute of Genetics and Medicine, Italy John George Lock, University of New South Wales, Australia Juan S. Bonifacino, Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), United States

> \*Correspondence: Vladimir V. Lupashin vvlupashin@uams.edu

#### †Present address:

Jessica Bailey Blackburn, Division of Allergy, Pulmonary and Critical Care Medicine, Department of Medicine, Vanderbilt University Medical Center, Nashville, TN, United States

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology Received: 09 April 2019

Accepted: 11 June 2019 Published: 03 July 2019

#### Citation:

D'Souza Z, Blackburn JB, Kudlyk T, Pokrovskaya ID and Lupashin VV (2019) Defects in COG-Mediated Golgi Trafficking Alter Endo-Lysosomal System in Human Cells. Front. Cell Dev. Biol. 7:118. doi: 10.3389/fcell.2019.00118

# Defects in COG-Mediated Golgi Trafficking Alter Endo-Lysosomal System in Human Cells

Zinia D'Souza, Jessica Bailey Blackburn† , Tetyana Kudlyk, Irina D. Pokrovskaya and Vladimir V. Lupashin\*

Department of Physiology, University of Arkansas for Medical Sciences, Little Rock, AR, United States

The conserved oligomeric complex (COG) is a multi-subunit vesicle tethering complex that functions in retrograde trafficking at the Golgi. We have previously demonstrated that the formation of enlarged endo-lysosomal structures (EELSs) is one of the major glycosylation-independent phenotypes of cells depleted for individual COG complex subunits. Here, we characterize the EELSs in HEK293T cells using microscopy and biochemical approaches. Our analysis revealed that the EELSs are highly acidic and that vATPase-dependent acidification is essential for the maintenance of this enlarged compartment. The EELSs are accessible to both trans-Golgi enzymes and endocytic cargo. Moreover, the EELSs specifically accumulate endolysosomal proteins Lamp2, CD63, Rab7, Rab9, Rab39, Vamp7, and STX8 on their surface. The EELSs are distinct from lysosomes and do not accumulate active Cathepsin B. Retention using selective hooks (RUSH) experiments revealed that biosynthetic cargo mCherry-Lamp1 reaches the EELSs much faster as compared to both receptor-mediated and soluble endocytic cargo, indicating TGN origin of the EELSs. In support to this hypothesis, EELSs are enriched with TGN specific lipid PI4P. Additionally, analysis of COG4/VPS54 double KO cells revealed that the activity of the GARP tethering complex is necessary for EELSs' accumulation, indicating that protein mistargeting and the imbalance of Golgi-endosome membrane flow leads to the formation of EELSs in COG-deficient cells. The EELSs are likely to serve as a degradative storage hybrid organelle for mistargeted Golgi enzymes and underglycosylated glycoconjugates. To our knowledge this is the first report of the formation of an enlarged hybrid endosomal compartment in a response to malfunction of the intra-Golgi trafficking machinery.

#### Keywords: COG complex, golgi apparatus, CRISPR, endosomes, glycosyltransferase, GARP complex, endocytosis

**Abbreviations:** Baf A1, bafilomycin A1; Tf, transferrin; CDG, congenital disorder of glycosylation; COG, conserved oligomeric golgi; EELSs, enlarged endolysosomal structures; ER, endoplasmic reticulum; GARP, golgi-associated retrograde protein; MTC, multi-subunit tethering complex; PM, plasma membrane; RUSH, retention using selective hooks; SNARE, SNAP (soluble NSF attachment protein) receptor; TGN, trans-golgi network; VPS, vacuolar protein subunit.

## INTRODUCTION

fcell-07-00118 July 1, 2019 Time: 17:2 # 2

Membrane trafficking is a conserved and tightly controlled process in all eukaryotic cells transporting about 30–50% of total proteins synthesized in a cell (Bonifacino and Glick, 2004). Trafficking events, modulated by components of the trafficking machinery which include coat proteins, adaptors, small GTPases, coiled-coil tethers, MTCs, SNAREs and SNAREassociated proteins, must act in concert to ensure proper synchronization of all the steps involved. MTCs achieve this through their many orchestrated interactions with the trafficking machinery (Oka and Krieger, 2005; Yu and Hughson, 2010). The major MTC at the Golgi is called the COG complex and is conserved from yeast to humans, as its name implies (Miller and Ungar, 2012; Climer et al., 2018a). This octameric complex made of two distinct subcomplexes with four subunits in each sub-complex (so called lobes A and B) and involved in Golgi retrograde trafficking (Lupashin and Ungar, 2008; Willett et al., 2013b; Ha et al., 2016). Mutations in 7 out of 8 individual subunits give rise to congenital disorders of glycosylation (CDG) called COG-CDG, a type-II CDG associated with high morbidity and mortality (Wu et al., 2004; Foulquier et al., 2006; Kranz et al., 2007; Paesold-Burda et al., 2009; Reynders et al., 2009; Lubbehusen et al., 2010; Kodera et al., 2015). In agreement with observed CDG human disease phenotype, siRNA-modulated depletion of COG subunits in human cells resulted in both N- and O-glycosylation defects (Zolov and Lupashin, 2005; Shestakova et al., 2006; Pokrovskaya et al., 2011). Similar glycosylation defects and depletion of multiple Golgi glycosyltransferases have been observed in HEK 293T cells completely depleted for each of the individual COG subunits using a CRISPR-Cas9 approach (Bailey Blackburn et al., 2016). It was also noted that defects in glycosylation are only one important facet of COG deficiency. The COG complex functionally interacts with multiple components of the cellular trafficking machinery, including SNAREs (Shestakova et al., 2007; Laufman et al., 2011; Kudlyk et al., 2013; Willett et al., 2013a; Willett et al., 2016), Rabs (Miller et al., 2013), coat proteins (Suvorova et al., 2002), coiled-coil tethers (Sohda et al., 2007; Sohda et al., 2010; Miller et al., 2013), the Biogenesis of Lysosome-Related Organelles Complex 1 (BLOC-1) (Gokhale et al., 2012) and copper transporter ATP7A (Comstra et al., 2017). Disruption of specific protein-protein interactions in COG-deficient cells is likely to produce multiple mutant phenotypes. Indeed, in addition to glycosylation defects, COG deficient cells exhibited a disrupted Golgi morphology, altered secretion, mislocalization of various protein and lipids and the formation of vacuole-like enlarged-endolysosomal structures (EELSs) (Bailey Blackburn et al., 2016; Blackburn et al., 2018). Importantly, all these defective phenotypes appeared to be independent from well-described glycosylation defects and therefore require special investigation (Blackburn et al., 2018).

In this report, we have used microscopic, biochemical and genetic approaches to characterize in detail one of the most prominent phenotypes in COG-depleted cells – EELSs.

# MATERIALS AND METHODS

#### Reagents and Antibodies

Reagents were as follows: LysoSensor Yellow/Blue DND-160 (Life Technologies; L7545), LysoTracker Red DND-99 (Life Technologies; L7528) GNL-Alexa 647 (Willett et al., 2013c), Baf A1 (RP1, B40500), Filipin (Sigma-Aldrich; F4767), TopFluor-Cholesterol (Avanti Polar Lipids). Primary antibodies used for western blotting (WB) or immunofluorescence microscopy (IF) were made in the lab or commercially purchased. Antibodies and their dilutions were as follows: rabbit monoclonal anti-MGAT1 (Abcam; ab180578; WB 1:500), goat affinity purified polyclonal anti-B4GALT1 (R&D Biosystems; AF3609; WB 1:500), goat affinity purified polyclonal anti-ST6GAL1 (R&D Biosystems; AF5924; WB 1:500), β-Actin (Sigma; WB 1:1000), rabbit anti-COG4 (Ungar et al., 2002), rabbit polyclonal anti-VPS54 (St. John's Laboratory; STJ115181; WB 1:1000), mouse anti-Lamp2 (DHSB; IF 1:100). Secondary antibodies used for WB or IF were as follows: fluorescent dye conjugated AffiniPure Donkey anti-mouse, anti-rabbit, or anti-sheep (IF 1:1000, Jackson Laboratories) and infrared dye IRDye 680RD or IRDye 800CW anti-mouse, anti-rabbit or anti-goat (WB 1:20,000, LI-COR).

#### Cell Culture

HEK293T cells (ATTC) were grown in DMEM/F12 medium (Thermo Fisher Scientific) supplemented with 10% FBS (Atlas Biologicals, Lot #F07J17A1; Cat #F-0500-A) with or without antibiotic/antimycotics where indicated. Cells were grown at 37◦C and 5% CO<sup>2</sup> in a 90% humidified incubator. For passaging cells, they were briefly trypsinized (0.25% trypsin EDTA, Gibco) for 3–4 min and resuspended in media.

HEK293T COG1 through COG8 knockout clones were described previously (Bailey Blackburn et al., 2016; Blackburn and Lupashin, 2016; Climer et al., 2018b).

Both VPS54 knockout (KO) and VPS54/COG4 double knockout (DKO) cell lines were generated using CRISPR (Jinek et al., 2012; Cong et al., 2013; Mali et al., 2013) in a similar fashion in HEK293T or HEK293T COG4 KO cells. First, HEK293T-Cas9 stable cell line was created by lentiviral transduction with FLAG tagged Cas9. HEK 293 FT cells were used to produce lentiviral particles with lentiCas9-Blast and helper plasmids according to the protocol described by the manufacturer. Prior to transfecting HEK 293 FT cells, they were placed in serum free Opti-MEM with 25 µM Chloroquine and GlutaMAX. The next day, the media was replaced with Opti-MEM supplemented with GlutaMAX. After 48 h of transfection, the media was collected and cell centrifuged at 400 g for 10 min. The cell free supernatant was then filtered using a 0.45 µm PES filter. 1 ml of this filtrate was added to HEK293T or HEK293T COG4 KO cells seeded on 6 cm dishes. 24 h later the media was replaced with DMEM/F12 supplemented with 10% FBS and 10 µg/ml blasticidin. To knockout VPS54 in HEK293T-Cas9 cells, CRISPR dual gRNAs were purchased from Transomics (TEDH-1088059, TEDH-1088060, TEDH-1088062). HEK293T-Cas9 cells were transfected with a cocktail of these three gRNAs

effectively inducing six cuts in VPS54 gene at the following target sequences:

Guide# TEDH-1088059 target sequences: grna-a: ACAAATATTCCTGAAACAGGCAGAAGGAAC grna-b: ATCTAGAAAGTGTTATGAATTCCATGGAAT Guide# TEDH-1088060 target sequences: grna-a: CAAAAGATAATTCACTGGACACAGAGGTGG grna-b: CATTCTACCTCCCACAGATCAGCAAGGAAC Guide# TEDH-1088062 target sequences: grna-a: CTTAACTCTGTAGCCACAGAAGAAAGGAAA grna-b: GTAAGCATGTCAGTAGTAACAGATGGGATG

Lipofectamine 3000 (Thermo Fisher Scientific) was used for transfecting cells according to the manufacturer's protocol. VPS54 KO cells were selected with puromycin (3 µg/mL) for 2 days post-transfection. Surviving cells were then single cell plated onto a 96-well plate. Clonal populations of VPS54 KO were screened by western blot for absence of the targeted protein.

VPS54/COG4 DKO cells were created in HEK293T COG4 KO-Cas9 cells by transfecting them with the same VPS54-specific dual gRNAs used to create the single VPS54 KOs. VPS54/COG4 DKO cells were selected with puromycin (3 µg/mL) for 2 days post-transfection followed by single-cell plating onto 96-well plates to obtain clonal populations that were screened for absence of VPS54 gene product by western blotting.

#### Plasmid Preparation and Transfection

Mammalian expression constructs were generated using standard molecular biology techniques or obtained as generous gifts. See **Table 1** for complete list of plasmids. Plasmids were isolated from bacteria using the QIAprep Spin Miniprep Kits (Qiagen). Plasmid transfections were performed using Lipofectamine 3000 (Thermo Fisher Scientific) according to the manufacturer's instructions.

### Transfection and Live Cell Immunofluorescence Microscopy

Cells were plated on collagen (50 µg/ml) coated 35 mm dishes glass bottom dishes with no. 1.5 coverglass (MatTek corporation), Lipofectamine 3000 was used to transfect cells as per the protocol described by the manufacturer. Briefly, the DNA and lipid reagent were separately diluted in Opti-MEM. Prior to transfection they were combined and added to dishes. Cells were incubated overnight with the DNA-lipid complexes and imaged the next day. Prior to imaging, the media was replaced with warm FluoroBrite DMEM Media (Life Technologies) supplemented with 10% FBS. Cells were imaged on an LSM880 Zeiss inverted microscope outfitted with confocal optics with the 63× oil 1.4 numerical aperture (NA) objective and Airyscan. The environment throughout imaging was controlled at 37◦C, 5% CO2, and 90% humidity.

#### BSA and Transferrin Labeling and Uptake

BSA (Sigma, #A7906) and Tf (EMD Millipore, #616397) were labeled using LI-COR's IRDye 650 Protein Labeling Kit and VRDye 549 Protein Labeling Kit, respectively, according to the manufacturer's protocol. Briefly, BSA or Tf was dissolved in azide free phosphate buffer, pH 8.5 to obtain a final concentration of TABLE 1 | Plasmids used in this study.


1 mg/ml. The labeling dye was dissolved in ultrapure water and added to the protein solution and the labeling was carried out for 2 h at room temperature in the dark. Pierce Polyacrylamide Spin Desalting Columns included in the kit were used to separate unconjugated dye after the labeling reaction was complete. The dye to protein ratio was calculated using the formula,

$$\frac{D}{P} = \frac{Abs max}{\varepsilon Dye} \Bigg/ \frac{A280 - \left(CF \ast Abs max\right)}{\varepsilon Protien} \Bigg/$$

and was found to be 0.96 and 0.77 for BSA and Tf, respectively. HEK 293T COG4 KO cells were plated on 35 mm collagen coated glass bottom dishes. Cells were transfected with Lamp2- GFP to label the EELSs. The next day, they were incubated with 100 µg/ml BSA-650 and Tf-550 diluted in 10% FBS supplemented media for 30 min or overnight. After incubation, the cells were washed with DPBS six times, the media was replaced with 10% FBS and GlutaMAX supplemented FluoroBrite DMEM Media and the cells were imaged on the LSM880 as described above.

#### Immunofluorescence Microscopy

12 mm glass coverslips (#1, 0.17 mm thickness) were collagen coated. HEK293T WT and COG4 KO cells were plated to

be 60–70% confluent at the time of processing. Cells were washed with Dulbecco's phosphate buffered saline (DPBS) and stained according to the protocol described previously (Zolov and Lupashin, 2005). Briefly, freshly prepared 4% paraformaldehyde (PFA) (16% stock solution diluted in DPBS; Electron Microscopy Sciences) was used as a fixative. After fixation, cells were permeabilized with 0.1% Triton X-100 for 1 min followed by treatment with 50 mM ammonium chloride for 5 min to quench free aldehydes. Cells were then washed three times with DPBS and blocked twice for 10 min each with 1% BSA, 0.1% saponin in DPBS. Antibodies were diluted in DPBS with 1% cold fish gelatin and 0.1% saponin. Cells were incubated with the primary antibody for 1 h at room temperature. Cells were washed four times with DPBS and incubated for 30 min with fluorescently tagged secondary antibodies diluted in antibody buffer. Cells were washed four times with DPBS. Hoechst diluted 1:10000 in DPBS was used to stain the nucleus. Coverslips were then washed four times with DPBS, rinsed with ddH2O, and mounted on glass microscope slides using Prolong Gold antifade reagent (Life Technologies). Cells were imaged with the 63× oil 1.4 numerical aperture (NA) objective of the LSM880 described above.

# High Pressure Freezing, Freeze Substitution, and Electron Microscopy

Electron microscopy was performed as previously described (Bailey Blackburn et al., 2016). Briefly, cells were plated on sapphire disk to be at 100% confluency next day. Disks were placed in PBS with 2% agarose, 100 mM D-mannitol, and 2% FBS then subjected to high pressure freezing using a Leica EM PACT2 with rapid transfer system. Samples were placed in acetone with 2% Osmium tetroxide, 0.1% Glutaraldehyde, and 1% ddH2O in liquid nitrogen then transferred to a freeze substitution chamber at −90◦C. Cells were slowly warmed to 0◦C and stained with a 1% Tannic acid/1% ddH2O solution in acetone for 1 h followed by staining with a 1% osmium tetroxide/1% ddH2O solution in acetone. Samples were then embedded in Araldite 502/Embed 812 resin (EMS) with DMP-30 activator added and processed in a Biowave at 70◦C under vacuum for 3 min per embedding step. Samples were then baked at 60◦C for 48 h before holders were removed and samples were cut. Postcutting staining was done with uranyl acetate and lead citrate prior to imaging.

Ultrathin sections were imaged at 80 kV on a FEI Technai G2 TF20 transmission electron microscope. Images were taken with a FEI Eagle 4 k× USB Digital Camera.

#### Retention Using Selective Hooks (RUSH) Assay

The Lamp1 RUSH reporter was a kind gift from Dr. Juan Bonifacino. The RUSH assay was performed as previously described (Boncompain et al., 2012; Chen et al., 2017). Briefly, cells were grown on collagen coated 35 mm glass dishes to a confluency of 60–70%. Prior to transfection, the media was supplemented with 100 µg/ml avidin to prevent biotin in the media from interfering with the RUSH reporter. Cells were co-transfected with the Lamp1 RUSH reporter and Lamp2-GFP using Lipofectamine 3000 as described above. After overnight transfection, the media was replaced with the chase mix which consisted of biotin and cycloheximide diluted to a final concentration of 40 µM and 50 µM, respectively, in FluoroBrite DMEM Media supplemented with 10% FBS and GlutaMAX.

To block the exit from the TGN, cells were incubated at 20◦C for 90 min immediately after the addition of the chase mix. After 90 min incubation at 20◦C, GNL-Alexa 647 (1:500) (Willett et al., 2013c) was added to the media and incubation was continued for another 30 min at 20◦C effectively resulting in a total of 2 h incubation at 20◦C which led to the retention of the RUSH reporter in the trans-Golgi upon its exit from the ER. The cells were washed 4–5 times with DPBS then, warm FluoroBrite DMEM Media supplemented with 10% FBS and GlutaMAX was added and the cells were imaged at 37◦C.

# Drug Treatment, Cell Lysis and Western Blotting

Cells were grown to a confluency of 80% on 6-well plates and treated with 200 nM Baf A1 for 24 h. Following drug treatment, the cells were stained with Hoechst for 20 min and imaged on a Zeiss Axiovert 200 microscope with phase-contrast. For western blot analysis, cells were untreated or treated with 200 nm Baf A1 for 24 h. Before lysing the cells, media was removed, and the wells were washed three times with DPBS. Hot 2% SDS was added to the wells and the lysates were collected in microcentrifuge tubes which were immediately placed in a heating block at 70◦C for 5 min. Amount of protein was determined using the Pierce BCA Protein Assay Kit as per the protocol. 6× sample buffer containing 5% β-mercaptoethanol was added to the lysates and 10 µg protein was loaded onto BioRad 4–15% gradient gels and the gel was run at 160 V.

An Invitrogen Power Blotter was used to transfer proteins onto a nitrocellulose membrane by semi-dry transfer. After transfer was complete, the membrane was washed with PBS and incubated for 30 min with Odyssey Blocking Buffer (LiCor). Primary antibodies were diluted in the same blocking buffer and were added to the membrane. The membrane was incubated overnight with primary antibody. The next day the membrane was washed three times and incubated with Donkey anti-Mouse IRDye 680RD or Donkey anti-Rabbit or Donkey anti-Goat IRDye 800CW secondary antibodies diluted in 5% evaporated milk in 1× PBS. The membrane was washed four times with 0.1% Tween-20 in PBS. The membrane was then air-dried and imaged using the Odyssey CLx Imaging System (LiCor). Analysis was done in Image Studio Version 5.2.

### Data and Statistical Analysis

The EELSs in phase contrast images were counted using the Particle Analysis plugin in ImageJ after setting an appropriate threshold to mask the background. Bar graphs representing the number of EELSs per cells were plotted in Microsoft excel 2010. The diameter of Lamp2 positive compartments in HEK293T WT and COG4 KO cells was manually estimated using Zen blue lite 2.3. Colocalization analysis was performed

using the RGB color plugin in ImageJ which measured the signal intensities of mCh and GFP along a line of interest. For statistical analysis, Student's t-test. was performed using the online "QuickCalcs" calculator of GraphPad Prism. Graphs were generated in Microsoft excel 2010.

# RESULTS

## COG Depletion Results in the Accumulation of EELSs

During the initial characterization of major morphological phenotypes in HEK293T cells completely depleted for individual COG subunits we observed a significant accumulation of large vacuole-like intracellular structures that were positive for both endosomal and lysosomal markers (Bailey Blackburn et al., 2016). A majority (95%) of Lamp2-positive membranes in wild-type (WT) HEK293T cells are ≤0.91 µm in diameter (**Supplementary Figure S1**). We called enlarged vacuole-like compartment (≥1 µm in diameter) observed in COG KO cells enlarged endo-lysosomal structures (EELSs). The formation of EELSs was COG-dependent since stable expression of GFP-tagged wild-type copies of the missing COG subunit completely abolished EELS formation (**Figure 1A**). A significant accumulation of EELSs was also observed in fibroblasts obtained from COG7-CDG patients (Blackburn et al., 2018) as well as in recently created COG4 KOs in human retinal pigment epithelial diploid RPE-1 cells (data not shown). Staining of live HEK293T COG4 KO cells with both LysoSensor (**Figure 1B**) and LysoTracker (**Figure 4B** and **Supplementary Figure S3**), which only fluoresce or are preferentially sequestered in acidic environments, indicated that the lumen of EELSs' is acidic. Importantly, this pH dependent fluorescence of LysoSensor is lost upon short treatment (4 h) with Baf A1, an inhibitor of vacuolar ATPase (vATPase) (**Figure 1B**). Interestingly, the EELSs completely disappear after prolonged treatment (24 h) of both COG4 KO and COG7 KO cells with Baf A1 suggesting that the activity of vATPases and the resulting low pH in the lumen of EELSs is necessary for their maintenance (**Figures 1C,D**).

Bafilomycin A1-induced removal of EELSs from COG4 KO cells renders these cells less viable than WT cells treated with Baf A1 for 72 h (**Figure 1E**), indicating that the formation of EELSs could be an adaptive feature in cells depleted for COG complex activity.

# Golgi Glycosyltransferases Are Targeted to EELSs

To look at whether EELSs are accessible to Golgi enzymes we transiently expressed RFP-tagged ST6GAL1 (ST-RFP) in COG4 KO cells. This construct is usually strictly Golgi localized in wild-type cells (**Figure 2A**; Lavieu et al., 2014) but in COG4 KO cells it is rapidly (6 h after expression) relocated to Lamp2-GFP-positive small and enlarged structures (**Figure 2B**), indicating that Golgi enzymes are mistargeted to endolysosomal compartments in COG KO deficient cells. Even more strikingly, after overnight expression ST-RFP is entirely off the Golgi and the RFP fluorescence is seen within the lumen of Lamp2 positive compartments, including EELSs (**Figure 2C**), indicating active cleavage/degradation of the transmembrane ST-RFP construct. Supporting this notion, we have previously shown that both medial (MGAT1) and trans-Golgi (B4GALT1 and ST6GAL1) enzymes are unstable in COG-deficient cells (Blackburn et al., 2018). To confirm that Golgi enzymes were being degraded in the acidic compartment we used Baf A1 to inhibit vATPasedependent degradation and re-assessed the stability of these enzymes. After 24 h of Baf A1 treatment we saw improved stability of the underglycosylated Golgi enzymes (**Figure 2D**). This observation combined with the loss of EELSs upon Baf A1 treatment (**Figure 1C**) suggested that an acidic lumen within the EELSs may be contributing to pronounced degradation of Golgi enzymes in COG-deficient cells.

## Late Endocytic Markers Are Present on EELSs

To determine which markers are present on the EELSs we used a superresolution microscopy approach in combination with staining for specific markers of the ER (ER tracer), Golgi (Giantin, GOSR1), PM (VAMP3) and endolysosomal (Rab5, Rab7, Lamp2, and STX8) system (**Supplementary Figure S3**). EELSs were positively stained with late-endosomal/lysosomal marker Lamp2 (Blackburn et al., 2018) and our initial staining attempts revealed that Lamp2-positive EELSs were also positive for Rab7a and STX8 and negative for Golgi markers Rab6, GOSR1, and Giantin (**Supplementary Figure S3**). Unfortunately, the EELSs do not preserve well upon PFA fixation, which causes the structures to partially collapse. To work around this, we used transient transfections of fluorescently tagged proteins. To avoid overexpression artifacts, only cells weakly expressing FP-tagged proteins were analyzed (**Figure 3** and **Supplementary Figure S2**). Analysis of HEK293T COG4 KO cells that co-express Lamp2-mCherry and GFP-tagged markers revealed that another endolysosomal marker CD63 (Lamp3) completely co-localized with EELSs (**Figure 3A**). We also found co-localization of two late-endosomal Rabs; Rab7a, and Rab9a (**Figures 3B,C**) and Golgi/endosomal Rab39a with Lamp2 on the EELSs' membrane (**Figure 3D**). Other tested GFPtagged Rabs (Rab1, 3, 4, 5, and 6) failed to co-localize with Lamp2-mCherry-positive EELSs (**Supplementary Figure S3**). Analysis of GFP-tagged SNAREs revealed that only two late endosomal SNAREs, Qa-SNARE STX8 and R-SNARE VAMP7, were present on membranes of EELSs (**Figures 3E,F**). Other tested GFP-tagged SNAREs (STX16, STX17, Sec22b, GOSR1, Vamp3, Vamp4, Vamp8, SNAP25, and SNAP29) were not localized on the membrane of Lamp2-mCherry-positive EELSs (**Supplementary Figure S3**).

Filipin, which forms complexes with cellular cholesterol, labels the Golgi apparatus of fixed cells (Pagano et al., 1989). We found that this fluorescent compound co-localizes with Lamp2 positive EELSs (**Figure 4A**), indicating that the membrane of EELSs is cholesterol-rich. In support of this finding, incubation of live COG4 KO cells with TopFluor cholesterol revealed a significant distribution of the fluorescent cholesterol analog to the membrane of acidic EELSs (**Figure 4B**).

FIGURE 1 | EELSs in COG KO cells are highly acidic and the activity of vacuolar ATPase is necessary for their long-term stability. (A) Specific accumulation of EELSs in COG KO cells. DIC images of HEK293T WT, COG4 KO, and stably rescued COG4 KO cells. (B) EELSs are highly acidic. HEK293T COG KO cells were incubated with LysoSensor Yellow Blue DND-160 as described in "Materials and Methods," treated with vacuolar ATPase inhibitor Baf A1 and imaged with Zeiss LSM880. Note that before Baf A1 treatment, LysoSensor fluorescence is seen in the EELSs (arrows) due to a low pH environment within their lumen. After treatment with Baf A1 LysoSensor fluorescence is diminished. Scale bars are 10 µm. (C) Phase–contrast images of COG4 KO cells before and after drug treatment. EELSs disappear after 24 h of treatment with 200 nm Bafilomycin A1. Scale bars are 50 µm. (D) Bar graph indicates the average number of EELSs per cell before and after treatment with Baf A1. (E) COG depleted cells are more sensitive to Baf A1 treatment. WT and COG4 KO cells were seeded at 40% confluence and incubated in culture media with or without with Baf A1. 72 h later dishes were rinsed with PBS to remove dead detached cells and remaining cells were counted. Arrows point to EELSs that are between 1 to 10 µm in diameter. Three fields were imaged and error bars indicate SD for n = 3. ∗∗p < 0.01.

To better understand why some molecules from Golgi and endosomal origins are preferentially targeted to the EELSs we decided to look as phosphatidylinositol (PI) species, which are known to be lipid determinants that identify different compartment membranes from one another in the cell. To analyze PI distribution in relation to EELSs, we used lipid biosensors, generously gifted by Dr. Sergio Grinstein (Bohdanowicz et al., 2012; Levin et al., 2016). Interestingly, the EELSs' membranes were found to be enriched in PI4P (**Figure 4C**), a lipid normally localized in the trans-Golgi and PM (**Supplementary Figure S4**; Behnia and Munro, 2005; Levin et al., 2017). The EELSs were negative for PI(4,5)P2, PI(3,4,5)P3, and PI3P (**Figure 4D**). Taken together, this data suggests that EELSs acquire late endocytic markers, cholesterol and PI4P.

To better understand the EELSs structure and contents, we turned to transmission electron microscopy (**Figure 5**). Large electron-sparse "empty" vacuole-like membrane structures were visible in every COG-deficient clone analyzed (**Figures 5B–F**), but were absent in wild-type HEK293T cells (**Figure 5A**). A small fraction of peripheral vacuoles contain either membrane-like

(**Figure 5B**) or even cell-like (**Figure 5C**) internalized cargo which may represent phagosomal structures. The EELSs were often located close to the fragmented Golgi (**Figures 5D,E**) and multi-vesicular bodies (MVBs) (**Figure 5F**) but were morphologically distinct from both organelles. Additionally, there was no apparent contact between the EELSs and the ER.

Lamp2 is localized on both late endosomal and lysosomal membranes (Chen et al., 2010). We have previously shown that lysosomal enzyme Cathepsin D is partially missorted and secreted in COG-deficient cells (Blackburn et al., 2018) leading us to wonder if lysosomal enzymes might be mistargeted to these structures. Therefore we looked for the activity of lysosomal the enzyme Cathepsin B within the EELSs using the Magic Red Assay (**Figure 6**). In this assay, the Magic red substrate fluoresces upon cleavage by Cathepsin B making it a read out for localization of active Cathepsin. Overall intracellular activity of Cathepsin

B was similar in HEK293T WT and COG4KO cells indicating that function of lysosomes was not dramatically disturbed in COG-depleted cells. Interestingly, enzyme activity was absent in EELSs indicated by the absence of red fluorescence which is robust in normal sized lysosomes of both WT and COG4 KO cells suggesting that the EELSs, though acidic and positive for lysosomal markers such as Lamp2 and Rab7a are distinct from mature lysosomes (**Figure 6**).

FIGURE 5 | Characterization of the EELSs by Electron microscopy. (A) Wild-type and (B–F) COG KO HEK293T cells grown on sapphire disks were high pressure frozen, fixed, embedded in Araldite 502/Embed 812 resin and analyzed by transmission electron microscopy. Note that the EELSs (labeled with star) are very prominent vacuole-like compartments in every COG KO cell.

We next asked how long it would take for newly synthesized biosynthetic cargo to get to EELSs. Since our initial experiments utilizing VSVG-ts45 system (Hirschberg et al., 1998) did not reveal any significant transport of plasma membrane-localized transmembrane cargo to or through the EELSs (data not shown), we used an alternative assay with endolysosome-specific

cargo, mCherry-Lamp1 using the RUSH system. The RUSH system was designed for synchronous biotin-dependent release of the reporter from the ER allowing visualization of the fate of the reporter as it passes through the secretory pathway (Boncompain et al., 2012). **Figure 7A** shows a schematic depiction of the RUSH assay. We chose to use Lamp2-GFP in addition with the Lamp1 RUSH construct as endogenous Lamp1 colocalizes with Lamp2 in a steady-state (Raposo et al., 1997). HEK293T COG4 KO cells were co-transfected with the Lamp1 RUSH reporter and Lamp2-GFP overnight in biotin-free media to allow for the accumulation of mCherry-Lamp1 in the ER and delivery of Lamp2-GFP to the EELSs. Biotin and cycloheximide were added 16 h after transfection to release mCherry-Lamp1 and to block any additional protein synthesis. 30 min after biotin addition mCherry fluorescence of the Lamp1 RUSH reporter was mostly perinuclear and distinct from Lamp2-GFP, indicating that RUSH reporter had reached the Golgi (**Figure 7B**). About 90 min later, mCherry signal starts to appear within Lamp2-GFP positive EELSs. At 2 h clear localization is seen within EELSs (**Figure 7C**) confirming that selective biosynthetic cargo is delivered to this compartment. Interestingly, majority of mCherry signal was localized not on the EELSs' membrane labeled with Lamp2-GFP, but instead in the EELSs' lumen indicating that either whole mCherry-Lamp1 molecule was internalized within EELSs or that mCherry was cleaved from the mCherry-Lamp1 construct. It is important to note that GFP in the Lamp2-GFP construct is localized on the cytoplasmic side of the membrane, while mCherry in mCherry-Lamp1 construct is predicted to be on the

luminal side, thus potentially exposing it to cleavage in the acidic environment of the EELSs.

### Endocytic Cargo Accumulates Within EELSs

Having established that newly synthetized Lamp1 is rapidly delivered to EELSs we turned to analysis of the endocytic cargo. HEK293T COG4 KO cells were fed with fluorescently labeled BSA-650 and Tf-549 to see whether receptor mediated (Tf), (Harding et al., 1983) or bulk (BSA) (Harding et al., 1985) endocytic cargo gets delivered to EELSs. Prior to feeding with endocytic tracers, these cells were transfected with Lamp2-GFP to label the EELSs. After 1 h incubation with BSA-650 and Tf-549, both markers appear in intracellular puncta that are likely to represent normal endosomes/lysosomes but do not yet appear in or on EELSs (**Figure 8A**). However, after 24 h accumulation of both fluid-phase and receptor-mediated tracers can be seen within EELSs (**Figure 8B**), indicating that EELSs are accessible to endocytic cargo, however, with a slow rate of delivery. Interestingly, at this time point the distribution of endocytic markers in EELSs was heterogeneous (**Figure 8C**). Majority (more than 90%) of EELSs were positive for Tf, while

only half of them contained both BSA and Tf and a very small percentage was either empty or only BSA-positive. Though Tf and BSA are internalized via different mechanisms since COG4 KO cells were simultaneously fed with BSA and Tf, some fraction of Tf would also enter via bulk phase endocytosis/pinocytosis making the heterogeneous distribution into EELSs rather surprising and raising the question of whether there is any selectivity that directs the delivery of endocytic cargo to the EELSs. In conclusion, EELSs are accessible to both fluid-phase and receptor-mediated cargo, but the delivery of both types of cargo is delayed in comparison to the rate of delivery of biosynthetic cargo (ST6Gal1 and Lamp1 RUSH reporter).

### The EELSs Preferentially Originate From the Golgi

In order to compare the kinetics of the delivery of biosynthetic and endocytic cargo to EELS directly we modified the RUSH system by introducing an additional TGN exit block. COG4 KO cells were co-transfected with the mCherry-Lamp1 RUSH reporter and Lamp2-GFP. The next day, immediately after adding biotin and cycloheximide, the cells were incubated at 20◦C for 2 h to induce a TGN exit block essentially trapping the RUSH construct in the TGN (Griffiths et al., 1989). Additionally, all plasma-membrane glycoproteins were decorated with fluorescently labeled GNL-647, a lectin which binds to immature surface glycans in in COG KO cells (Bailey Blackburn et al., 2016; Blackburn and Lupashin, 2016). This set up essentially enabled simultaneous tracking of biosynthetic and endocytic cargo as well as the comparison of the kinetics of delivery. **Figure 9A** schematically depicts the experimental set up. At the beginning of the chase, GNL-647 signal is localized on the plasma membrane as well as in endosomes and mCherry signal of Lamp1 RUSH reporter is perinuclear in the TGN (**Figure 9B**). The TGN localization of the RUSH reporter was confirmed by immunofluorescence with a TGN marker, TGN46 (data not shown). 3 h after transferring the cells to 37◦C, all mCherry signal is seen within the lumen of Lamp2 positive EELSs but, these are mostly negative for GNL-647 (**Figure 9C**) confirming that anterograde delivery of Golgi cargo is faster than retrograde delivery of plasma membrane cargo to the EELSs and suggests that the TGN is the dominant supplier of membrane and cargo to the EELSs.

#### An EELS Is a Stable Hybrid Compartment

The appearance of EELSs is a very prominent feature of COGdepleted cells; at the same time, they disappear after Baf A1 treatment and during chemical fixation, suggesting their fragile and/or transient nature. Live cell analysis indicated that EELSs exhibit slow movement within cells and are not as dynamic as endosomes/lysosomes, but both membrane and soluble cargo may be exchanged between EELSs and other organelles of secretory and endocytic pathways. To test this we employed a fluorescent recovery after photobleaching (FRAP) approach on cells transfected with Lamp2-GFP (membrane cargo) and mCherry-Lamp1 (soluble cargo) to analyze the rate of the material exchange between EELSs and other organelles. Both GFP and mCherry fluorescence within one EELS in HEK293T COG4 cells co-expressing Lamp2-GFP (membrane cargo) and mCherry-Lamp1 (soluble cargo) was bleached and the time course of fluorescence recovery was analyzed (**Figure 10**). Strikingly, no significant fluorescence recovery of both membrane and soluble markers was detected even after 2 h indicating that the membranes of EELSs do not physically connected to other membranes of secretory and endocytic pathways and that fusion with upcoming membrane transport carriers happens at a slow rate. A similar result was observed while performing FRAP analysis only with luminal mCherry (**Supplementary Figure S5**). It is important to note, that in this experiment Lamp2-GFP-positive EELSs remained intact ruling out the possibility of the EELS collapsing upon photobleaching. We concluded that the EELS is a stable membrane compartment with restricted communication with other secretory and endocytic compartments.

### GARP Activity Is Necessary for EELSs Formation

To test if the deficiency of another Golgi vesicle tethering complex, GARP, will result in accumulation of EELSs we knocked out VPS54 (**Figure 11**) and VPS53 (data not shown) in HEK293T cells. VPS54 is a unique subunit of the GARP complex, an MTC that operates between late endosomes and TGN (Bonifacino and Hierro, 2011) while VPS53 belongs to both GARP and EARP (Endosome-Associated Recycling Protein) complexes (Schindler et al., 2015). In both KO cells lines (**Figure 11A** and data not shown) no accumulation of enlarged vacuole-like structures was detected, indicating that EELSs are COG KO specific and are formed due to trafficking defects caused by loss of COG's function. Interestingly, knock out of VPS54 in COG KO cells (VPS54/COG4 DKO) resulted in a complete disappearance of the EELSs and the cells appeared like HEK293T WT or COG rescued cells in shown in **Figures 1**, **11C**). Similar results were obtained with VPS53/COG4 DKO cells (**Supplementary Figure S6**). This led us to conclude that the EELSs' formation requires the activity of the GARP complex in COG deficient cells.

# DISCUSSION

In this study, we continue our investigation of a specific phenotypic defect that arises upon complete depletion of individual COG subunits. Using microscopy and biochemistry approaches, EELSs were characterized to understand how they are formed in COG KO HEK293T cells. This phenotype is not unique to HEK293T cells and is present in COG4 KO RPE1 (data not shown) cells and in a subset of fibroblasts obtained from COG7-CDG patients (Blackburn et al., 2018).

Our studies revealed that formation of EELSs is COG specific and this phenotype is rescued upon stable re-expression of the WT copy of the deleted COG subunit. Importantly, formation and/or stability of EELSs were also severely diminished in cells treated with vacuolar-ATPase inhibitor Baf A1 and upon inhibition of another MTCs GARP. Based on these observations we hypothesized that formation of EELSs depends on the activity

COG4 KO cells. After that, plasma membrane glycoproteins were labeled with GNL-647 and delivery of both endocytic and biosynthetic cargo to the EELSs was analyzed using AiryScan microscopy. (A) Schematic representation of the RUSH assay with a 20◦C TGN exit block. (B) Fluorescent images of GNL-647, Cherry-Lamp1 and Lamp2-GFP (EELS marker) immediately after shifting the cells to 37◦C and (C) after 3 h of shifting the cells to 37◦C. Scale bars are 10 µm.

of vATPase and that GARP is playing an essential role in delivery of membranes to EELSs because knocking out VPS54 results in the loss of EELSs (**Figures 11**, **12**). In agreement to this prediction we found that the lumen of EELSs is acidic due to the activity of vATPase. vATPases pump protons into the lumen of organelles thus maintaining an appropriate pH in their lumen (Oot et al., 2017). The decreasing pH gradient from cis to trans-Golgi is also maintained by the activity of vATPases (Marshansky and Futai, 2008). This pH gradient is crucial for functions within the Golgi like glycosylation and cargo sorting to be properly executed.

Impairment in one of the subunits of v-ATPase, the a2 subunit, causes a type II-CDG, the same subtype of CDG that COG depletion causes (Kornak et al., 2008; Bahena-Bahena et al., 2014). This subunit anchors vATPases into membranes and also provides a channel for protons to move from the cytosol to the lumen. One possibility is that the COG complex machinery is responsible for proper targeting/retention of vATPase in the Golgi sub-compartments and COG malfunction alters vATPase's intracellular distribution creating a hybrid post-Golgi compartment with very low pH. Our preliminary mass-spectrometry analysis of COG complex membrane partners indicates that several subunits of vATPase may indeed directly or indirectly interact with COG4 and COG8 (JBB, VL, personal communication).

We have also observed that EELSs are lost in cells under stress induced by various triggers such as overnight treatment with cycloheximide or starvation (ZD, VL, personal communication). One possibility is that the loss of EELSs could be due to the disassembly of vATPase under stress to conserve cytosolic ATP (Kane, 2000). However, this is only conjecture and a detailed analysis of vATPases in COG KOs will shed more light. At this point it is clear that low pH with the lumen of EELSs is crucial for their maintenance. The relatively large size (up to 10 µm in diameter) and "empty" appearance of the EELSs (**Figure 5**) indicate that in addition to vATPase other ion and water transporters may also be mistargeted to EELSs in COG deficient cells. It will be important to test if aquaporins that reside constitutively at the plasma membrane in most cell types are redistributed to EELSs. Several studies have demonstrated that the aquaporins are present in intracellular vesicles in liver and kidney, implying that aquaporins in post-Golgi compartments could be involved in their volume regulation (Sugiya et al., 2008). If aquaporins are present on EELSs, it would also explain why EELSs are sensitive to disruption of H<sup>+</sup> homeostasis

which is lost upon BafA treatment. In this scenario, aquaporins may constantly pump in water to maintain osmotic balance between EELSs' acidic lumen and the cytosol. Baf A1 induced vATPase inactivation makes the EELSs hypotonic with respect to the cytosol and so the EELSs collapse because of outward movement of water.

What is the origin of EELSs? Appearance of the EELSs in COG deficient cells is a very unusual and an unexpected cellular phenotype. However, formation of enlarged endocytic compartments has been observed previously as a result of malfunctioning of endocytic sorting/trafficking machinery (Choudhury et al., 2005; Kamalesh et al., 2017; Kaur et al., 2018). An example of this is the appearance of enlarged early endosomes in cells expressing GTP-restricted Rab5 (Stenmark et al., 1995). To our knowledge, our lab is the first to implicate altered Golgi trafficking machinery in the formation of the enlarged post-Golgi acidic compartments. We believe that the EELSs actually originate from the trans-Golgi in COG-deficient cells as a result of protein and lipid mis-targeting. Cargo delivery studies (**Figures 7**, **8**) provide the evidence that EELSs are dominantly fed by anterograde trafficking and the delivery of endocytic cargo is much slower in comparison. We have found that EELSs are enriched in cholesterol, similar to the TGN and plasma membrane and typical TGN lipid, PI4P (**Figure 4**). Interestingly, the localization of yeast vATPases was shown to be regulated by its interaction with a PI4P (Banerjee and Kane, 2017). The EELSs are enriched with PI4P and this interaction may be the reason EELSs acquire vATPases. Besides PI4P no other Golgi marker co-localized with EELSs by IF, suggesting that Golgi proteins are either rapidly degraded as in the case of ST6GAL1 (**Figure 2A**), within the acidic lumen of EELSs or is efficiently sorted out of EELSs. Efficient sorting out of resident proteins should include robust communication between the EELSs and other cellular compartments, but our FRAP experiments indicate very limited exchange for both soluble and transmembrane cargo. Interestingly, transmembrane protein mCherry-Lamp1 was efficiently delivered to EELSs, but mCherry fluorescence was detected mostly in the EELS's lumen, indicating degradation or specific cleavage of this hybrid molecule. It will be interesting to investigate which enzyme actually cleaves the mCherry-Lamp1 hybrid since tested lysosomal proteases like Cathepsin B were mostly absent from EELSs (**Figure 6**). EELSs are positive for a subset of late endosomal markers (Rabs7, 9, 39, Lamp1, 2, CD63, STX8, and Vamp7) and we hypothesized that these proteins are delivered to nascent EELSs in a GARP-dependent mechanism (**Figure 12**). In this scenario defects in GARPdependent delivery would abolish "feeding" and enlargement of nascent EELSs, essentially rescuing EELSs' formation. Based on distribution of endocytic cargo and EM analysis we also propose that the EELSs are heterogeneous and include large phagocytic vacuoles directly originating from plasma membrane and containing large extracellular cargo and even neighboring cells (**Figure 5C**). In the future it will be important to figure out what triggers the increase of phagocytic activity in COG deficient cells.

What is the function of EELSs in COG deficient cells? The localization of a COG sensitive glycosyltransferase ST6Gal1 to enlarged vacuoles (**Figure 2A**) and the rescue of stability of several Golgi enzymes upon treatment with BafA1 (**Figures 2B,C**) suggests that these Golgi resident proteins could be degraded in the EELSs. We propose that the formation of EELSs is an adaptive function in COG deficient cells. In favor of

TGN feeding it with cargo and membrane. This results in a bottle neck effect at the TGN and the consequent enlargement of this compartment manifests as EELSs. Hence, when the GAPR complex is no longer active in COG4 KO cells, the EELSs are no longer formed. This figure was made using BioRender.

this hypothesis, we have observed that Baf A1 induced abolition of the EELSs makes COG4 KO cells less viable that WT treated cells (**Figure 1E**). The EELSs may serve to isolate and degrade both underglycosylated and missorted proteins and lipids to avert their negative impact on cell physiology.

# AUTHOR CONTRIBUTIONS

ZD, JB, and VL designed and conducted the experiments and wrote the manuscript. TK and IP designed and conducted the experiments and edited the manuscript.

#### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the **Supplementary Files**.

#### FUNDING

This work was supported by the National Institutes of Health (R01GM083144) (VL).

#### ACKNOWLEDGMENTS

fcell-07-00118 July 1, 2019 Time: 17:2 # 18

We are thankful to Juan Bonifacino, Sergio Grinstein, Mitsunori Fukuda, Rainer Duden, Marc Coppolino, Ghanshyam Swarup, Thierry Galli, Grégory Lavieu, George Banting, and Santiago M. Di Pietro as well as others who have provided the reagents. We would also like to thank the UAMS for permitting us to use their core facilities and expertise particularly in sequencing, flow cytometry, and microscopy.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2019.00118/ full#supplementary-material

FIGURE S1 | Scatter plot indicating size distribution of Lamp2 positive compartments in WT and COG4 KO cells. The diameter of Lamp2 positive structures was analyzed in 31 WT cells and 43 COG4 KO cells, transiently transfected with fluorescent protein tagged Lamp2. Of the 724 Lamp2 positive structures in WT cells, 95% of the vacuoles had a diameter ≤0.91 µm while in the case of COG4 KO cells, of the 885 vacuoles counted, 58% of the vacuoles had a diameter ≥0.91 µm. Out of the 58% of EELSs with a diameter ≥0.91 µm, 88% of them had a diameter ≥1 µm which we defined as EELSs.

#### REFERENCES


FIGURE S2 | Localization of GFP-CD63 (A), GFP-Rab7a (B), GFP-Rab9a (C), GFP-Rab39a (D), GFP-Stx8 (E), and GFP-Vamp7 (F) in WT and COG4 KO cells. Graphs generated using the RGB Profiler plug-in in ImageJ represent the quantification of mCh and GFP signal intensities along the white line (left to right) drawn. Overlapping peaks indicate colocalization of the two markers. Scale bars are 10 µm.

FIGURE S3 | ER, Golgi and early endosomal markers do not colocalize with the EELSs. Live cell imaging of COG4 KO cells indicate that EELSs, labeled with LysoTracker Red DND-99 do not colocalize with ER tracker. Immunostaining with Lamp2 antibody or transient transfection with Lamp2-mCh in COG4 KO cells to label EELSs show that EELSs are negative for Golgi (Giantin, TGN38) and Rabs (Rab1, 3, 4, 5, and 6). EELSs are also negative for the SNARES- Stx16, Stx17, GOSR1, Sec22b, SNAP23, SNAP29, Vamp3, Vamp4, and Vamp8. Scale bars are 10 µm.

FIGURE S4 | PI4P localization in HEK293T WT cells. GFP fluorescence of the PI4P biosensor, GFP-2×P4M is seen on the PM and has perinuclear/Golgi localization. Scale bars are 10 µm.

FIGURE S5 | EELSs are stable after photobleaching. Bleaching of luminal mCh does not result in collapse of the EELS. The cells were co-transfected with the mCh-Lamp1 RUSH reporter construct and Lamp2-GFP. Bleaching only luminal mCh does not affect the EELS's stability and the EELS can be seen with only GFP on its membrane even 50 min after bleaching. In this time period, there is no recovery of mCh fluorescence. Scale bars are 2 µm.

FIGURE S6 | VPS53 KO recues the formation of EELSs in COG4 KO cells. Knocking out VPS53 in COG4 KO cells using a CRISPR-Cas9 mediated approach results in DKO cells with no EELSs. Scale bars are 50 µm.


protein in SNARE complexes of the apical plasma membrane of epithelial cells. Mol. Biol. Cell 9, 1437–1448. doi: 10.1091/mbc.9.6.1437



**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 D'Souza, Blackburn, Kudlyk, Pokrovskaya and Lupashin. 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.

# CREB3 Transcription Factors: ER-Golgi Stress Transducers as Hubs for Cellular Homeostasis

Luciana Sampieri1,2, Pablo Di Giusto1,2 and Cecilia Alvarez1,2 \*

<sup>1</sup> Centro de Investigaciones en Bioquímica Clínica e Inmunología (CIBICI-CONICET), Córdoba, Argentina, <sup>2</sup> Departamento de Bioquímica Clínica, Facultad de Ciencias Químicas, Universidad Nacional de Córdoba, Córdoba, Argentina

CREB3 family of transcription factors are ER localized proteins that belong to the bZIP family. They are transported from the ER to the Golgi, cleaved by S1P and S2P proteases and the released N-terminal domains act as transcription factors. CREB3 family members regulate the expression of a large variety of genes and according to their tissue-specific expression profiles they play, among others, roles in acute phase response, lipid metabolism, development, survival, differentiation, organelle autoregulation, and protein secretion. They have been implicated in the ER and Golgi stress responses as regulators of the cell secretory capacity and cell specific cargos. In this review we provide an overview of the diverse functions of each member of the family (CREB3, CREB3L1, CREB3L2, CREB3L3, CREB3L4) with special focus on their role in the central nervous system.

#### Edited by:

Jaakko Saraste, University of Bergen, Norway

#### Reviewed by:

Hiderou Yoshida, University of Hyogo, Japan Carolyn Machamer, Johns Hopkins University, United States

\*Correspondence: Cecilia Alvarez calvarez@fcq.unc.edu.ar

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 15 April 2019 Accepted: 17 June 2019 Published: 03 July 2019

#### Citation:

Sampieri L, Di Giusto P and Alvarez C (2019) CREB3 Transcription Factors: ER-Golgi Stress Transducers as Hubs for Cellular Homeostasis. Front. Cell Dev. Biol. 7:123. doi: 10.3389/fcell.2019.00123 Keywords: CREB3 transcription factors, ER stress response, Golgi stress response, secretory pathway, secretory capacity, cellular homeostasis, central nervous system

#### INTRODUCTION

Eukaryotic cells have different ways of achieving homeostasis and coping with cellular requirements. The secretory pathway plays a fundamental role in maintaining homeostasis since it needs to adapt to endogenous and exogenous stimuli to regulate the cellular capacity for secretion. Most of the studies aimed to understanding the adaptation of the secretory pathway have been carried out inducing some type of organelle-specific stress, such as cargo overload, structural damage or perturbation of an enzymatic activity. Consequently, the signaling pathways activated to achieve homeostasis are considered a response to stress. In this sense, the ER stress response (or the unfolded protein response) and the Golgi stress response have been described. In addition, lysosomes and peroxisomes have their own stress responses, and a mitochondrial unfolded protein response (UPRmt) has also been reported (Sasaki and Yoshida, 2015; Melber and Haynes, 2018). In general, a stress response is triggered by a sensor protein, that detects the insufficiency of organelle function and activates one or more transcription factors which in turn induce the transcription of genes involved in the modulation of organelle function. For instance, to upregulate the capacity of the ER, sensor molecules located on the ER membrane, such as ATF6, IRE1 and PERK, activate transcription factors which increase the transcription of ER-related genes (Walter and Ron, 2011).

The Golgi stress response is triggered when its capacity to handle protein processing is overloaded and the cell needs to increase the expression of modification enzymes. This response, which has been less studied than the ER stress response, is associated with the following pathways: TFE3 (Taniguchi et al., 2015), HSP47 (Miyata et al., 2013), proteoglycans (PG)

(Sasaki et al., 2019), ETS (E26 transformation-specific, Baumann et al., 2018), and CREB3 (Reiling et al., 2013). Oku et al. (2011) and Taniguchi et al. (2015, 2016) have identified that TFE3 and MLX transcription factors regulate expression of some Golgirelated genes after inhibiting Golgi function with monensin, among other treatments. TFE3 and MLX bind to the cis-acting element called the Golgi apparatus stress response element (GASE) and modulate the TFE3 pathway. MLX competes with TFE3 for GASE sites resulting in the attenuation of TFE3 induction. The HSP47 pathway was characterized using a GALNAc structural analog (named BG or GalNAc-bn) which is a competitive inhibitor of mucin type O-glycosylation. This treatment induces expression of the ER chaperone HSP47 that prevents Golgi stress-induced apoptosis (Miyata et al., 2013). It is not clear how this ER-localized chaperone evades BG-induced Golgi stress. Moreover, cellular treatments performed to reduce PG glycosylation in the Golgi (Sasaki et al., 2019) contributed to characterize the PG pathway, which regulates the expression of PG-induced Golgi stress genes. Furthermore, Baumann et al. (2018) have recently identified three ETS transcription factor family members, ELK1, ETS1 and GABPA/B that respond to pharmacological Golgi disruption, suggesting that they operate in parallel. In addition to the TFE3, HSP47, PG, and ETS pathways, Reiling et al. (2013) reported the CREB3 pathway that leads to Golgi stress response, inducing apoptosis through ARF4 transcriptional activation after Brefeldin A treatment. CREB3 belongs to the CREB3 family of transcription factors that were shown to regulate numerous genes involved in secretory capacity and structure of the Golgi complex, including ER chaperones and various transport factors (Bailey and O'hare, 2007; Murakami et al., 2009; Fox et al., 2010; Barbosa et al., 2013). The study of transcription factors that respond to Golgi stress as well as the resulting signaling pathways are essential to understand the role of this organelle in physiological and pathological contexts such as tumor and neurodegenerative processes. In this review we provide an overview of the CREB3 family, and details of the individual roles of all members (CREB3, CREB3L1, CREB3L2, CREB3L3, CREB3L4) with special focus on CREB3, CREB3L1, and CREB3L2 functions in the central nervous system (CNS).

# OVERVIEW OF THE CREB3 FAMILY

The CREB3 family of transcription factors is comprised of five members in mammals: CREB3, CREB3L1, CREB3L2, CREB3L3, and CREB3L4. They belong to the large bZIP family, which is one of the mayor families of transcription factors. CREB3 family is highly related to the Sterol Regulatory Element-Binding Proteins (SREBPs) and the Activating Transcription Factor 6, ATF6, families. They also have leucine zipper domain and undergo regulated intramembrane proteolysis (RIP). SREBPs and ATF6 are prototypical ER-bound transcription factors that act in response to different signals recognized by the ER. SREBPs regulate fatty acid and cholesterol metabolism. The stimulus for SREBPs activation is the absence of sterols, and its N-terminal domain promotes transcription of many genes involved in cholesterogenesis and lipogenesis (Eberle et al., 2004). ATF6 is best known for its role in transducing signals linked to ER stress (Haze et al., 1999). Recent reports, however, have described novel functions for ATF6 related to organogenesis and tissue homeostasis (Wang et al., 2015; Naranjo et al., 2016; Jin et al., 2017). ATF6 is retained in the ER through interactions between its luminal tail and the ER chaperone GRP78/BiP. ER stress induces GRP78/BiP dissociation from ATF6, resulting in the exposure of its Golgi-localization sequences (Shen et al., 2002). Following translocation to the Golgi Complex, ATF6 is cleaved by site-1 protease (S1P) and site-2 protease (S2P) to release its N-terminal active fragment. This active portion of ATF6 is then transported to the nucleus, where it binds to ER stress-response elements which results in the expression of ER stress proteins including GRP78/BiP and XBP1 (Hillary and Fitzgerald, 2018).

The overall range of functions of the CREB3 family includes development, metabolism, secretion, survival, differentiation, tumorigenesis and cell division, among others. One important finding regarding the role of CREB3 transcription factors in secretion was reported by Fox et al. (2010). They identified in the Drosophila salivary gland factors required for its secretory function and found that dCREB-A (the only CREB3 family member encoded by Drosophila) is required to enhance the expression of genes encoding components of the secretory pathway. Moreover, they showed that dCREB-A also targets genes encoding cell specific proteins that require the secretory pathway to reach their destination.

CREB3 transcription factors are single-pass membrane proteins localized in the ER with their N-terminus facing the cytoplasm and the C-terminus the ER lumen (**Figure 1**). Details of the sequence homologies between the different members of the group have been previously reviewed (Chan et al., 2011; Fox and Andrew, 2015). In summary, CREB3 members share the following functional domains (named from N- to C-terminus, **Figure 1A**): the transactivation domain (TAD) that mediates sequence specific DNA binding, a conserved domain of approximately 30 residues called ATB (adjacent to bZIP), a basic region (Basic) next to the leucine zipper domain (Zip) called together bZIP DNA-binding domain, and a transmembrane domain (TMD). The ATB domain is not part of the bZIP, but a distinct feature of this family and may consequently indicate special functions for these proteins (**Figure 1A**, Bailey and O'hare, 2007). In response to different signals, including ER stress, CREB3 proteins are transported from the ER to the Golgi complex where they are cleaved (activated) through RIP (Brown et al., 2000) by S1P and S2P proteases sequentially. The first cleavage is performed by S1P, a membrane-bound serine protease of the subtilisin family (Sakai et al., 1998). After that, the resulting CREB3 protein is cleaved by S2P, a membrane-embedded zinc metalloprotease to release the N-terminal fragment, which translocates into the nucleus and activates the transcription of target genes (**Figure 1B**). Also, the N-terminal form of CREB3 transcription factors can form homo- and hetero-dimers with differential transcriptional activity (Vinson et al., 2006).

CREB3 transcription factors can be regulated by their proteolytic cleavage (induced by both ER and Golgi signals) and by increase of their transcription. Transcription of dCREB-A is

cooperatively regulated by the transcription factors CBP (CREBbinding protein) and Cut (Chung et al., 2017). Interestingly, in dendritic cells, expression of constitutively active form of CREB3 increases its own transcript levels (Sanecka et al., 2012). In chondrocytes, CREB3L2 is a direct target gene of the master regulator for chondrogenesis, Sox9 (Hino et al., 2014) and orphan nuclear receptor Nr4a1 controls the expression of CREB3L1 in Arginine vasopressin neurons (Greenwood et al., 2017). Ceramide also acts as a CREB3L1 upstream regulator by promoting its proteolytical activation (Denard et al., 2012). Despite the identification of some signals and molecules that lead to the activation of members the CREB3 family, their regulation in different tissues remains poorly studied. The dual localization of CREB3 transcription factors (ER and Golgi complex) and their

ability to sense ER stress and to respond to Golgi disruption posit them as molecules involved in both the ER and Golgi stress response.

#### CREB3

CREB3, also called LZIP or Luman (after a legendary hero in ancient China; Lu et al., 1997), was identified through yeasttwo hybrid assays for its interaction with the transcriptional co-activator HCF (Host Cell Factor, Freiman and Herr, 1997; Lu et al., 1997). CREB3 mRNA is detected in many rat and human tissues, but liver and nervous system have the highest levels of expression (Ying et al., 2015).

CREB3 levels increase in an age-specific manner in Leydig cells, which are responsible for the synthesis of testosterone in testicles. CREB3 knock-down in mouse testis leads to an increase in steroidogenesis genes expression and testosterone synthesis (Wang et al., 2019). In contrast, in ovarian mouse granulosa cells, CREB3 depletion induces a decrease of estradiol and progesterone synthesis and promotes cell proliferation (Zhao et al., 2016). Steroidogenic enzymes and cell cycling factorsencoding genes were found down-regulated and up-regulated in CREB3 knock-down cells, respectively. These results highlight an important role of CREB3 in male and female reproduction.

In agreement with data found in Drosophila CREB3 gene, dCREB-A, expression of the constitutive active CREB3 form in dendritic cells induces up-regulation of secretory pathway genes, including COPII components as well as Golgi proteins such as GBF1 and Arf4 (Sanecka et al., 2012). In line with these results, a CREB3 isoform lacking the transmembrane domain induces ARF4 transcription by binding to the cAMP response element motif in the ARF4 promoter (Jang et al., 2012). Moreover, the CREB3- ARF4 signaling is implicated in the Golgi stress response induced by BFA treatment and supports the survival of Chlamydia trachomatis and Shigella flexneri (Reiling et al., 2013). Moreover, the Chlamydia pneumoniae-specific inclusion membrane protein, Cpn0147, interacts with CREB3 to mediate the interaction with the host cell endoplasmic reticulum (Zhao et al., 2017).

It has been reported that in human osteogenic sarcoma cells (HOS) CREB3 binds to the CC chemokine receptor 1 (CCR1) and participates in Leukotactin-1-induced cell migration by enhancing the NF-κB activation pathway (Ko et al., 2004; Jang et al., 2007a,b). Furthermore, Sung et al. (2008) showed that CREB3 regulates the expression of the chemokine receptors CCR1 and CCR2 in a human monocyte cell line (THP-1). Interestingly, these two receptors are involved in the early stages of atherogenesis. Altogether, these data account for an important function of CREB3 in the process of cell migration with high impact on the pathogenesis of atherosclerosis. Additionally, CREB3 expression has been linked to breast cancer. In both MCF-7 and MD-MB-231 breast cancer cell lines, CREB3 binds specifically to Histone Deacetylase 3 (HDAC3). In this context, HDAC3 is a co-repressor of CREB3-mediated CXCR4 gene expression (Kim et al., 2010). CXCR4 is a chemokine receptor, a target of CREB3 and a crucial mediator of cell migration in both leukocytes and tumor cells. Interestingly, CXCR4 is highly expressed in primary and metastatic human breast cancer cells. Howley et al. (2018) found that ARF4, COPB1 and USO1, ER-Golgi trafficking proteins regulated by CREB3, are associated with an invasive phenotype in a mouse model of metastatic progression. This evidence points out to the relevance of the regulation of components of the secretory pathway, such as the Golgi Complex, by members of the CREB3 family not only in physiological contexts but also in a pathologic environment such as cancer.

#### CREB3L1

CREB3L1 was originally named OASIS (for old astrocyte specifically induced substance) because it was identified in a screening for genes induced in long-term cultured astrocytes (old astrocytes) obtained from newborn mice brains as an in vitro model to study gliosis (Honma et al., 1999). Northern blot assays performed in multiple human tissues indicated that heart, placenta, pancreas, prostate, lung, and colon express higher CREB3L1 levels than brain, testis, and skeletal muscle (Omori et al., 2002). CREB3L1 is also highly expressed in osteoblasts. In fact, CREB3L1-deficient mice exhibited severe osteopenia caused by a decrease in the levels of type I collagen, the major component of the bone matrix (Murakami et al., 2009). In agreement with that, CREB3L1 activates transcription of type I collagen a1 gene, Col1a1, by directly binding to a CRE-like sequence in its promoter region. Other targets of CREB3L1 are Xbp1 and the chaperone protein GRP78/BiP, genes typically up-regulated during ER stress (Murakami et al., 2009). Moreover, CREB3L1 regulates bone angiogenesis during bone development regulating the expression of hypoxia-inducible factor-1α (HIF-1α) target genes (Cui et al., 2015). CREB3L1 critical contribution in bone formation was also confirmed by its role as a genetic cause of autosomal recessive osteogenesis imperfecta in humans (Symoens et al., 2013; Keller et al., 2018; Guillemyn et al., 2019). CREB3L1 mutant forms identified in these patients also down-regulate the expression of COPII components, Sec23A and Sec24D.

CREB3L1 expression was detected in pancreatic beta-cell lines and rodent islets and is highly active during pancreas development. Transfection of active form of CREB3L1 in a pancreatic β-cell line induced expression of genes involved in protein transport and implicated extracellular matrix production (Vellanki et al., 2010). Results from our group indicate that, in thyroid cells, CREB3L1 levels are up-regulated by thyrotropin and CREB3L1 is sufficient to increase transport proteins levels and induce Golgi enlargement (Garcia et al., 2017). We also show that expression of CREB3L1 dominant negative hampers the TSH-induced Golgi enlargement. Moreover, CREB3L1 expression is frequently altered in many cancer types and in some of them, like in breast and bladder cancer, is epigenetically silenced through DNA methylation (Rose et al., 2014; Ward et al., 2016). Interestingly, a retrospective study performed on biopsy samples analysis from triple negative breast cancer indicated

that CREB3L1 levels in tumors responsive to doxorubicin chemotherapy were significantly higher than in those resistant to this treatment (Denard et al., 2018). It has been postulated that CREB3L1 is a metastasis suppressor and that it may function in a similar way to p53 as a regulator of cell proliferation (Denard et al., 2011). In contrast, CREB3L1 is up-regulated in a metastatic subtype of triple-negative breast cancer cells that have activated both PERK signaling and the epithelial-to-mesenchymal transition program (Feng et al., 2017). In this tumor subtype CREB3L1 expression promotes invasion through the activation of extracellular matrix genes such as Col1a and FN1.

#### CREB3L2

CREB3L2 also known as BBF2 human homolog on chromosome 7 (BBF2H7) was identified as a novel human protein whose C-terminal region is fused to the FUS (fusion) genes in low-grade fibromyxoid sarcoma as a result of chromosomal translocation. Changes in CREB3L2 levels were first described in C6 glioma, HEK293 and MEF cells treated with thapsigargin (Kondo et al., 2007). Moreover, CREB3L2 mRNA levels were detected in different cell types and tissues (Kondo et al., 2007; Panagopoulos et al., 2007).

One of the findings that link CREB3L2 to the regulation of the secretory pathway is its participation in the differentiation of hepatic stellate cells (HSCs) to myofibroblast-like cells, a critical event in hepatic fibrosis. This process is characterized by enlargement of the ER and Golgi complex. Interestingly, another feature of this process is the up-regulation of Sec23A and Sec24D, components of the coat protein complex II (COPII), mediated by CREB3L2 (Tomoishi et al., 2017).

CREB3L2 mRNA is enriched in the developing notochord of Xenopus laevis embryos, where it also regulates genes of the secretory pathway (Tanegashima et al., 2009). Moreover, during early embryonic development of medaka fish, CREB3L2 is required for transcriptional regulation of a complete set of genes (Sec23a/24d/13/31a, Tango1, Sedlin, and KLHL12) essential for the enlargement of COPII vesicles to accommodate type II collagen for export from the ER (Ishikawa et al., 2017). In agreement with this, the well described zebrafish feelgood mutation, which disrupts head skeleton and notochord development through loss of secretory capacity, consists of a missense mutation in the DNA-binding domain of CREB3L2 and this results in decreased expression of sec23a and sec24d genes (Melville et al., 2011). An interesting fact about CREB3L2 is that, in developing cartilage, both the N- and C-terminal of the protein have important roles. After CREB3L2 cleavage, the N- terminus exerts its activity as transcription factor by promoting secretion of extracellular matrix proteins through induction of Sec23a expression (Saito et al., 2009; Hino et al., 2014). The C- terminal part of the protein, on the other hand, promotes the proliferation of chondrocytes and inhibits hypertrophic differentiation via regulating the Indian hedgehog (Ihh)/parathyroid hormonerelated protein signaling pathway (Saito et al., 2014). CREB3L2, as well as some cartilage matrix genes like Col2a1, are targets of Sox9 (Hino et al., 2014). The result of this transcriptional activation axis is the acceleration of cartilage matrix protein secretion during chondrocyte differentiation. In chondrocytes, CREB3L2 also functions as a target of the F-Box protein Fbxw7, a component of SKP1-CUL1-F-box protein type ubiquitin ligase which contributes to stem cell maintenance and cell differentiation. Fbxw7 targets the nuclear form of CREB3L2 for degradation in mesenchymal cells, thereby contributing to chondrogenesis (Yumimoto et al., 2013). Data obtained by Al-Maskari et al. (2018) showed that CREB3L2 expression increases during human B-cell transition to antibody secreting cells, which is a logic finding considering that this event implies great secretory overload (Shi et al., 2015) and that CREB3L2 has already been reported to up-regulate genes involved in the secretory pathway in similar cellular contexts.

The role of CREB3L2 in cancer is mainly represented by the specific translocation t(7;16)(q33;p11) that results in the creation of the chimeric gene FUS-CREB3L2, which is responsible for low-grade fibromyxoid sarcoma (LGFMS); a rare, slow-growing type of cancer that usually forms in the deep soft tissues of the legs or trunk (chest and abdomen) (Panagopoulos et al., 2007; Bartuma et al., 2010). In this context, the chimeric gene is believed to regulate CD24 (Moller et al., 2011). It was also described that, in malignant glioma, an FRS2/PAK1 activated RAS/MAPK signaling cascade up-regulates CREB3L2, which directly binds to the ATF5 promoter resulting in ATF5 transcription, an anti-apoptotic factor which plays a role in cell survival (Sheng et al., 2010).

#### CREB3L3

CREB3L3 (also known as CREB-H) was originally isolated as a transcription factor expressed in a liver-specific manner (Omori et al., 2001). CREB3L3 is also expressed in the stomach and small intestine. The roles for this transcription factor include triglyceride metabolism in the liver (Lee et al., 2011), reduction of cholesterol absorption (Kikuchi et al., 2016), glucose and lipid metabolism (Nakagawa et al., 2016b; Nakagawa and Shimano, 2018), acute phase response activation (Zhang et al., 2006) and hepcidin-mediated iron metabolism (Vecchi et al., 2009). CREB3L3 expression is regulated by a number of nuclear receptors, including PPARα (Danno et al., 2010), HNF4α (Luebke-Wheeler et al., 2008), GR (Lee et al., 2010), and ERRγ (Misra et al., 2014).

Surprisingly, CREB3L3 knock-out mice are viable, fertile, have a normal lifespan and do not display any gross physical or behavioral abnormalities. No anatomical or histological differences were found between liver and gastrointestinal tract of CREB3L3−/− mouse embryos respect to the controls. This information argues in favor of the notion that CREB3L3 is not essential for hepatogenesis and hepatocyte differentiation in the mouse. However, CREB3L3−/− mice showed a strong decrease in the transcript levels of acute phase genes compared to control animals when tunicamycin was administered to induce ER stress (Luebke-Wheeler et al., 2008). When intestinal CREB3L3 knock-out mice were compared to floxed mice, there were no

apparent differences in metabolic parameters. On the other hand, the liver CREB3L3 knock-out mice showed hyperlipidemia due to increased expression levels of genes related to cholesterol synthesis relative to floxed mice (Nakagawa et al., 2016a). Moreover, a whole genome expression analysis performed recently on liver samples from CREB3L3−/− mice subjected to ketogenic diet underscored the relevance of CREB3L3 in regulating apolipoprotein metabolism (Ruppert et al., 2019).

It has been demonstrated that CREB3L3 is required for the acute inflammatory response by regulating transcription of CRP and SAP genes which encode C-reactive protein and serum amyloid P-component proteins, respectively. Moreover, CREB3L3 and ATF6 interact with each other to synergistically activate expression of their target genes upon ER stress (Zhang et al., 2006). CREB3L3 has also been strongly linked to the regulation of cell proliferation. Interestingly, it is significantly underexpressed in hepatocellular carcinoma tissues and cells. Also, the loss of CREB3L3 function in hepatocellular carcinoma might contribute to the initiation and/or progression of cancer (Chin et al., 2005).

# CREB3L4

CREB3L4, also known as AIbZIP, CREB4 or TISP40, was first described in 2002 by two independent research groups. One of these groups was originally interested in identifying androgenregulated genes in human prostate cancer cells, and so CREB3L4 cDNA was isolated from LNCaP human prostate cancer cells treated with the synthetic androgen R1881. Because the cDNA analysis revealed that it contains a region with extensive similarity to the bZIP domain of CREB/ATF transcription factors, the protein was designated AIbZIP (Androgen-Induced bZIP protein; Qi et al., 2002).

The main function of CREB3L4 appears to be related to the tissue where it is mostly expressed; the prostate. This transcription factor has been shown to be involved in the proliferation of prostate cancer cells promoted by the Androgen Receptor (AR) and IRE1α (Kim et al., 2017). Interestingly, it was described that CREB3L4 can interact with CREB3L1 to inhibit its nuclear translocation in LNCaP cells. This results in p21 suppression and, consequently, increase in cell proliferation (Cui et al., 2016). Also, CREB3L4 has been linked to the process of cellular differentiation into adipocytes (Kim et al., 2014). Another group has reported its participation in male germ cell development, underlying that whereas CREB3L4 knockdown moderately impairs spermatogenesis, it is not sufficient to produce infertility in mice (Adham et al., 2005). Downstream target genes of CREB3L4 are tightly associated with prostate cell proliferation. In this context, the direct interaction between CREB3L4 and AR has been reported. Ben Aicha et al. (2007) demonstrated that a number of different genes with diverse functions are induced by CREB3L4: transcription factors, genes involved in protein processing, genes encoding channels and transporters, genes in charge of lipid and sugar metabolism and signal transduction genes, among others. These data imply once again the concept that the functions of CREB3 proteins, in this case CREB3L4, might not only be limited to the response to ER stress.

CREB3L4 is predominantly expressed in prostatic tissue and in breast cancer and prostate cancer cell lines. Interestingly, its expression is higher in cancerous prostate cells compared with non-cancerous prostate cells. Two independent research groups generated CREB3L4 knock-out mice, and both reported these mice to be healthy and fertile. Their findings using these murine models are slightly different between each other, but they both point to mild defects in spermatogenesis. More specifically, it was found that acetylated H2A and H4 histones are abnormally retained in epididymal sperm, implying that CREB3L4 might be regulating sperm head nuclei maturation in the mouse (Nagamori et al., 2006).

## CREB3 FAMILY IN THE CENTRAL NERVOUS SYSTEM

An RNA-sequencing transcriptome study performed with mouse brain cells (Zhang et al., 2014) indicated that CREB3L3 and CREB3L4 are minimally detected in different cell types of the CNS, while CREB3, CREB3L1 and CREB3L2 are co-expressed in most of them (**Figure 2**). In this section we review their role in physiological and pathological processes of the CNS.

#### Sampieri et al. CREB3 Family in Cellular Homeostasis

#### CREB3

CREB3 is the family member with the highest levels of expression in different cell types of the nervous system (Zhang et al., 2014, **Figure 2**). As we mentioned above, CREB3 was identified for its interaction with the transcriptional co-activator HCF (Freiman and Herr, 1997; Lu et al., 1997). During herpes virus infection, CREB3 and HCF together with the virion transactivator factor, VP16 induce expression of virion genes. Moreover, it has been postulated that, in neurons of the trigeminal ganglia, CREB3 is required for the establishment of herpes virus latency (Lu and Misra, 2000). Interestingly, CREB3 is also able to activate promoters of genes critical for herpes virus reactivation suggesting a complex role of the CREB3-HCF interaction in this process. The complexity of the CREB3-HCF interaction is harder to understand in dorsal root ganglion neurons where HCF and CREB3 poorly colocalized and exhibited Golgi and ER patterns, respectively (Kolb and Kristie, 2008). Another link between CREB3 and viral infection in the CNS is provided by the CREB3-Herp (homocysteine-induced ER protein) pathway in the Poliovirus (PV)-induced apoptosis. Herp is a CREB3 direct transcriptional target (Liang et al., 2006) involved in Ca2+ regulation in neurons. Down-regulation of CREB3 or Herp expression in IMR5 cells, a neuroblastoma cell line, increased PVinduced apoptosis (Mirabelli et al., 2016). Further analysis will be necessary to understand the participation of CREB3 and Herp in maintaining the balance of pro and antiapoptotic signals during the PV-induced neuropathogenesis of poliomyelitis.

Furthermore, CREB3 regulates intrinsic elongating form of axonal growth linked to injury-associated axonal ER responses (Ying et al., 2014). In dorsal root ganglion sensory neurons CREB3 localizes to the soma and the axonal ER. In response to nerve injury, CREB3 is synthesized in axons and transported to the nuclei of injured neurons via importin-dependent retrograde transport (Hasmatali et al., 2019). Interestingly, immunoprecipitation assays indicated that CREB3 interacts with importin, but CREB3-target genes involved in axon growth were not identified.

CREB3 knock-out mouse model (Penney et al., 2018) used to study the hypothalamic pituitary adrenal (HPA) axis and the glucocorticoid (GC) response linked CREB3 function in CNS to the secretory pathway. These animals have low levels of corticosterone and high levels of the glucocorticoid receptor (GR). Chromatin immunoprecipitation assays performed with mouse embryonic hippocampal cells indicated that CREB3 binds to the promoter region of genes that contain GC response elements (GRE). Interestingly, CREB3 also acts as a GR co-factor since it interacts with GR and enhances GR activity. Furthermore, in hippocampal cells, CREB3 binds to the promoter of genes encoding COPII components, regulating their expression.

#### CREB3L1

CREB3L1 expression is transiently up-regulated in the brain of mouse embryos and becomes weaker in the adult (Saito et al., 2012). Up-Regulation of its expression was also demonstrated in reactive astrocytes proximal to a spinal cord injury (Nikaido et al., 2002). Studies performed in CREB3L1 knock-out mice indicated that, in astrocytes, CREB3L1 promotes glial scars formation, impedes axon growth and functional recovery after spinal injury (Sumida et al., 2018). Furthermore, CREB3L1 protein expression was detected in astrocytes and in neuronal primary cultures obtained from hippocampi of mice, but its mRNA up-regulation was detected only in astrocytes after treatment with kainic acid (KA). However, pyramidal neurons in the hippocampi of CREB3L1−/− mice were more susceptible to the toxicity induced by KA than those of wild-type mice (Chihara et al., 2009) suggesting a protective role of astrocytes against the KA-induced neuronal damage. One CREB3L1 target identified in astrocytes is the chondroitin 6-O-sulfate transferase 1 (C6ST1) gene, which encodes the major sulfotransferase of proteoglycan chondroitin sulfate (CSPG, Okuda et al., 2014). In vitro luciferase assays indicated that CREB3L1 binds to the first intron region of mouse C6ST1 gene. In contrast to the protective role that astrocytes play against KA-induced neuronal damage, astrocytes derived from wild-type CREB3L1 mice inhibited neurite outgrowth of cultured hippocampal neurons, whereas astrocytes from CREB3L1 knock-out mice did not. Therefore, CREB3L1 induction in reactive astrocytes in the injured brain may help to establish a non-permissive microenvironment for regenerating axons (Okuda et al., 2014). Another example of changes in the levels of CREB3L1 that modulate the function of the nervous system is given in the human retinal pigment epithelial cells, ARPE-19 (Miyagi et al., 2013). In these cells CREB3L1 transcriptionally regulates the vascular endothelial growth factor-A (VEGFA) which not only acts as a mediator of angiogenesis but also as a trophic and protective factor of retinal neurons (Foxton et al., 2013).

Increased CREB3L1 expression was also shown in the hypothalamus, specifically in the supraoptic and paraventricular nuclei of dehydrated and salt-loaded rats. The CREB3L1 upregulation occurs in arginine vasopressin (AVP) neurons, where CREB3L1 activates AVP gene transcription in vivo (Greenwood et al., 2014, 2015). In the mouse pituitary cell line AtT20, CREB3L1 expression is up-regulated by cAMP in vitro, and orphan nuclear receptor Nr4a1 is the transcription factor controlling the expression of CREB3L1. Furthermore, the ability to activate CREB3L1 by Nr4a1 is related to the level of methylation of the CpG island within the CREB3L1 proximal promoter (Greenwood et al., 2017).

Moreover, a direct link between CREB3L1 levels and changes in neurons was shown in Drosophila class IV dendritic arborization (C4da) neurons where nuclear polyglutamine (polyQ) toxicity reduced CREB3L1/CREBA levels. Additionally, polyQ led to down-regulation of genes involved in the secretory pathway and to the loss of Golgi outposts (GOPs). Furthermore, C4da neurons exhibited defective terminal dendrite elongation and decreased supply of plasma membrane (Chung et al., 2017). In C4da neurons, CREB-binding protein directly regulates CrebA transcription in cooperation with the Drosophila transcription factor Cut (Chung et al., 2017). Consistent with the data obtained in Drosophila, rat hippocampal primary neurons exhibited reduced number of dendrites spines and GOPs due to polyQ toxicity. Also, CREB3L1/CrebA overexpression restored the loss of GOPs and the down-regulation of COPII-related genes induced by polyQ. However, no significant changes exist in

the branching and elongation of terminal dendrites for the overexpression of CREB3L1 alone.

#### CREBL2

Expression of CREB3L2 in vivo was analyzed by immunohistochemistry in a mouse model of permanent focal brain ischemia where CREB3L2 was detected in the region closer to the infarction region in the striatum, especially in neurons (labeled with MAP2). In contrast to CREB3L1, no expression of CREB3L2 was detected in astrocytes in this model. The role of CREB3L2 in the brain was studied using SK-N-SH cells, a human neuroblastoma cell line. Overexpression and siRNA transfection assays indicate that CREB3L2 reduces and increases the sensitivity to ER stress-induced cell death, respectively (Kondo et al., 2007). Although CREB3L2-depleted cells die more than control cells by the action of thapsigargin, the levels of typical ER-stress markers such as GRP78/BiP, XBP1, CHOP, and PDI were not modified. CREB3L2-target genes that protect against cell death induced by thapsigargin were not explored in SK-N-SH cells. Participation of CREB3L2 in cell survival was also reported in mouse malignant glioma GL261 cells (Sheng et al., 2010), where CREB3L2 was identified as one of the 12 genes required for expression of ATF5 (activating transcription factor 5), an anti-apoptotic factor which plays a role in cell survival (Monaco et al., 2007). Sheng et al. (2010) also showed that expression of CREB3L2 in human malignant glioma was higher than in normal brain and that individuals with ATF5-positive glioblastomas had shorter survival times than those with ATF5-negative glioblastomas. Moreover, expression of CREB3L2, ATF5 and its target, the oncogene MCL1 (myeloid cell leukemia sequence 1, Le Gouill et al., 2004) decreased after serum-induced GS9-6 differentiation indicating that these proteins are enriched in undifferentiated cells. ATF5 is required for terminal differentiation and survival of olfactory

TABLE 1 | CREB3 family members in the central nervous system, list of their functions, upstream regulators and downstream target genes.


N/D, not defined.

sensory neurons according to homozygous Atf5 knock-out mice (Wang et al., 2012).

CREB3L2 levels also increase during oligodendrocytes (OL) maturation where Chd7 (chromodomain-helicase-DNA-binding protein 7) and Sox10 activates a transcriptional program for OL differentiation (He et al., 2016). CREB3L2 and Osterix are Chd7 targets and their down-regulation in OL precursor cells inhibited expression of genes related to myelination (He et al., 2016). Moreover, in the dorsal-root ganglion neurons-like F11 cell line, CREB3L2 was identified as a direct target gene of the transcription factors nuclear factor IL-3 (NFIL3) and the tree isoforms of the CCAAT-enhancer-binding proteins (C/EBPα, C/EBPβ, and C/EBPδ; Macgillavry et al., 2011). It has been postulated that these transcription factors co-regulate CREB3L2 expression during forskolin-induced neurite outgrowth model. NFIL3 and C/EBPs integrate a complex transcriptional regulatory network that fine-tunes the expression of neuronal outgrowthrelated genes.

# OUTLOOK

CREB3 transcription factors are widely expressed in different tissues and they regulate a broad range of developmental, physiological and pathological processes playing fundamental roles in cellular homeostasis. Most of them show a tissuespecific preferential expression, however, CREB3, CREB3L1, and CREB3L2 co-express in different cells of the CNS (**Figure 2**) where they participate in essential processes (**Table 1** and **Figure 3**). For example, CREB3L1 and CREB3L2 are involved in neurite outgrowth (Macgillavry et al., 2011; Okuda et al., 2014) while CREB3 and CREB3L1 modulate (directly or indirectly) axonal growth after an injury (Ying et al., 2014; Sumida et al., 2018). CREB3 regulates pathogenic mechanisms of herpes and polio virus (Lu and Misra, 2000; Liang et al., 2006), and in non-CNS cells a CREB3-ARF4 signaling pathway mediates the susceptibility to pathogens (Reiling et al., 2013). As we mentioned above, CREB3 transcription factors can also modulate cell survival by activating expression of anti-apoptotic factors (Sheng et al., 2010). Also, CREB3 and CREB3L1 contribute to neuroendocrine regulation of the hypothalamic/pituitary/adrenal axis modulating the GR activity and the AVP gene transcription (Greenwood et al., 2014; Penney et al., 2018). Most of these processes require the adaptation of the secretory pathway which is regulated by CREB3 family members in multiple cell types (Fox and Andrew, 2015). In line with this, CREB3 regulates expression of genes encoding COPII components and formation of Golgi outposts in hippocampal cells (Chung et al., 2017; Penney et al., 2018).

CREB3 transcription factors emerge as signaling hubs for the regulation of Golgi homeostasis, integrating stimuli from multiple sources to control secretion, protein post-translational modification and trafficking, impacting on membrane expansion and composition. Their contribution to the CNS is crucial since neurons are specially sensitive to Golgi stress and Golgi fragmentation, events tightly connected to neurodegenerative diseases (Gonatas et al., 2006; Machamer, 2015; Lopez et al., 2017). Although some upstream regulators and downstream target genes of CREB3 transcription factors have been identified

(**Table 1**), many questions remain open: What are the specific stimuli that trigger the activation of these factors and how are they sensed? How are CREB3 family genes regulated in each cell type of the CNS? What are the consequences of lacking one or more CREB3 transcription factors in a cell? Can they replace each other? How do they participate in neuronal development? Comprehensive understanding of how CREB3 transcription factors function promises not only to explain fundamental biological questions, but also to provide new options for therapeutic intervention.

#### AUTHOR CONTRIBUTIONS

CA, LS, and PD wrote the manuscript. CA contributed with funding support.

#### REFERENCES


#### FUNDING

LS and PD are both CONICET doctoral fellows. CA is an investigator of CONICET and a Professor at Universidad Nacional de Córdoba. Work in the Álvarez lab is supported by the grants PICT2016-0042-Prestamo BID from the Ministry of Science and Technology of Argentina, SECyT-UNC Tipo A (2016 and 2018) from Universidad Nacional de Córdoba.

#### ACKNOWLEDGMENTS

The authors thank Susana Genti-Raimondi (Universidad Nacional de Córdoba, Córdoba, Argentina) and Hector A. Saka (Universidad Nacional de Córdoba, Córdoba, Argentina) for their comments after reading the manuscript, special thanks to HAS for his contribution with figure design.



concentration increase and apoptosis induced by poliovirus. J. Gen. Virol. 97, 2194–2200. doi: 10.1099/jgv.0.000544


regulator of secretion. Front. Mol. Neurosci. 11:352. doi: 10.3389/fnmol.2018. 00352



with high-grade metastatic breast cancers with poor prognosis and is prevalent in triple negative breast cancers. Breast Cancer Res. 18:12.


**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 Sampieri, Di Giusto and Alvarez. 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.

# New Insights Into the Golgi Stacking Proteins

Erpan Ahat<sup>1</sup> , Jie Li<sup>1</sup> and Yanzhuang Wang1,2 \*

<sup>1</sup> Department of Molecular, Cellular and Developmental Biology, University of Michigan, Ann Arbor, MI, United States, <sup>2</sup> Department of Neurology, University of Michigan School of Medicine, Ann Arbor, MI, United States

The Golgi stacking proteins, GRASP55 and GRASP65, are best known for their roles in Golgi structure formation. These peripheral Golgi proteins form trans-oligomers that hold the flat cisternal membranes into stacks. Depletion of both GRASP proteins in cells disrupts the Golgi stack structure, increases protein trafficking, but impairs accurate glycosylation, and sorting. Golgi unstacking by GRASPs depletion also reduces cell adhesion and migration in an integrin-dependent manner. In addition to Golgi structure formation and regulation of cellular activities, GRASPs, in particular GRASP55, have recently drawn attention in their roles in autophagy, and unconventional secretion. In autophagy, GRASP55 senses the energy level by O-GlcNAcylation, which regulates GRASP55 translocation from the Golgi to the autophagosome-lysosome interface, where it interacts with LC3 and LAMP2 to facilitate autophagosome-lysosome fusion. This newly discovered function of GRASP55 in autophagy may help explain its role in the stress-induced, autophagosome-dependent unconventional secretion. In this review, we summarize the emerging functions of the GRASP proteins, focusing on their roles in cell adhesion and migration, autophagy, unconventional secretion, as well as on novel GRASP-interacting proteins.

#### Edited by:

Daniel Ungar, University of York, United Kingdom

#### Reviewed by:

Antonino Colanzi, Istituto di Biochimica delle Proteine (IBP), Italy Paul Anthony Gleeson, The University of Melbourne, Australia Adam Graham Grieve, University of Oxford, United Kingdom

> \*Correspondence: Yanzhuang Wang yzwang@umich.edu

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

Received: 18 April 2019 Accepted: 03 July 2019 Published: 16 July 2019

#### Citation:

Ahat E, Li J and Wang Y (2019) New Insights Into the Golgi Stacking Proteins. Front. Cell Dev. Biol. 7:131. doi: 10.3389/fcell.2019.00131 Keywords: Golgi, stacking, GRASP65, GRASP55, O-GlcNAcylation, autophagy, unconventional secretion

# INTRODUCTION

The Golgi apparatus is an essential membrane-bound organelle in the cell that functions as a "post station" in the secretory pathway (Klute et al., 2011). In mammalian cells, Golgi membranes are organized as stacks of multiple flat cisternae, which are further linked into a ribbon-like structure located in the perinuclear region (Klumperman, 2011). The Golgi functions as a protein modification and sorting center in the secretory pathway, with different modification enzymes residing in different subcompartments, including cis-Golgi network (CGN), cis-, medial-, and trans-cisternae, and trans-Golgi network (TGN) (Goldfischer, 1982). The Golgi receives newly synthesized proteins and lipids from the endoplasmic reticulum (ER), sequentially modifies, and dispatches them to distinct destinations by protein sorting at the TGN (Marsh and Howell, 2002; Brandizzi and Barlowe, 2013).

The highly ordered Golgi stack structure, which is the functional unit of the Golgi, is believed to facilitate sequential protein modification, and processing in mammals. Using in vitro assays mimicking Golgi disassembly and reassembly that occur during the cell cycle, two Golgi peripheral membrane proteins, GRASP65 and GRASP55 (Golgi ReAssembly and Stacking Protein), were identified as Golgi stacking factors (Barr et al., 1997; Shorter et al., 1999; Wang et al., 2003). Both GRASPs were further characterized and confirmed to control Golgi stacking and ribbon linking in vivo (Wang et al., 2003; Puthenveedu et al., 2006; Feinstein and Linstedt, 2008; Xiang and Wang, 2010). GRASP65 is mainly targeted to cis-Golgi, whereas GRASP55 localizes

to medial- and trans-cisternae. GRASP65 and GRASP55 have similar domain structures. The conserved GRASP domain at the N-terminus contains a membrane anchor and can form dimers and trans-oligomers. The more divergent serine proline-rich (SPR) domain at the C-terminus contains multiple phosphorylation sites, whose phosphorylation inhibits GRASP oligomerization in mitosis (Wang et al., 2005; Vielemeyer et al., 2009; Tang et al., 2012) and perhaps also in stress and pathological conditions (**Figures 1**, **2A**; Joshi et al., 2014, 2015; Joshi and Wang, 2015). In coordination with GRASPs, GRASPs interacting proteins, including the GRASP65 binding partner GM130, and GRASP55 binding protein Golgin-45, may also be involved in Golgi structure formation (Lee et al., 2014).

Since depletion or inhibition of GRASP55 and GRASP65 impairs Golgi structure formation, these proteins have been recently used as tools to disrupt the Golgi structure and thereby determine the functional consequence of Golgi structural disruption. In addition, a number of novel GRASP-interacting proteins have been identified and GRASPs have been linked to autophagy, unconventional secretion and other cellular activities, such as cell adhesion, migration, and growth. In this review, we attempt to summarize these new discoveries on GRASP functions and discuss the potential links between these new findings.

# GRASP55 AND GRASP65 AS TOOLS TO PROBE THE BIOLOGICAL SIGNIFICANCE OF GOLGI STRUCTURE FORMATION

The role of GRASPs in Golgi stack formation and the impact of GRASP depletion or inhibition on Golgi functions have been explored using a number of experimental approaches. Inhibition of GRASP65 by microinjecting inhibitory antibodies, knocking down (KD) GRASPs by siRNA, or knocking out (KO) GRASPs by CRISPR/Cas9, all significantly impair Golgi stack formation (Wang et al., 2003; Tang et al., 2010; Bekier et al., 2017). Research from the Rothman and Wang labs demonstrated that depletion of GRASPs, which results in Golgi fragmentation, increases the trafficking of selected cargo molecules, including vesicular stomatitis virus G-protein (VSVG), α5 integrin, and CD8 (Wang et al., 2008; Xiang et al., 2013; Lee et al., 2014). A plausible explanation for this result is that Golgi unstacking increases the membrane surface for vesicle formation and thus accelerates protein trafficking (Wang et al., 2008; Zhang and Wang, 2015; Bekier et al., 2017; Ahat et al., 2019). Need to mention, in contrast to the results described above, D'Angelo et al. (2009) reported that GRASP55/65 bind the C-terminal hydrophobic tail of specific transmembrane proteins such as CD8, and this interaction is required for CD8 trafficking through the Golgi stack. These controversial results could be caused by the different constructs used in the studies; the first two labs used full length CD8 (or a fusion protein with full length CD8), and while the last one used a VSVG-CD8α (C-terminal tail) chimera. Other factors, such as the knockdown efficiency, may also be involved. Considering that there is currently no other way to disrupt Golgi stacking other than manipulating GRASPs, this subject requires further investigation.

Glycomic analysis by mass spectrometry showed that GRASP depletion, especially GRASP55 single depletion or GRASP55/65 double-depletion, results in a reduction in the overall glycan abundance, complexity, and glycoprotein composition at the plasma membrane. Interestingly, GRASP depletion-mediated Golgi unstacking also causes mis-sorting of lysosome enzymes such as cathepsin D (Xiang et al., 2013; Zhang et al., 2015). Therefore, it has been proposed that cisternal stacking impedes the intra-Golgi trafficking speed by reducing the accessibility of coat proteins to Golgi membranes, which ensures accurate glycosylation, and sorting (Zhang and Wang, 2015, 2016; Huang and Wang, 2017). Recently, GRASP55 and GRASP65 single knockout mice have been reported, with only limited defects in Golgi structure and function (Veenendaal et al., 2014; Chiritoiu et al., 2019). One possibility is that the knockout effect of one GRASP may be compensated by the redundancy of the other GRASP protein. It has been demonstrated that when one GRASP is depleted in cells, the level of the other GRASP protein may increase to compensate for the knockout effect (Bekier et al., 2017). GRASP55 and GRASP65 double knockout mice have not been reported so far.

# EFFECTS OF GOLGI DESTRUCTION ON CELL ATTACHMENT, MIGRATION, AND GROWTH

The effect of Golgi unstacking induced by GRASP depletion on cellular activities such as cell attachment, migration, and growth have recently been investigated. GRASP KD or KO in HeLa cells reduces cells adhesion to fibronectin-coated dishes (Ahat et al., 2019). GRASP depletion also reduces cell migration in HeLa and MDA-MB-231 cells. While the effect was significant when a single GRASP was depleted, it was more robust when both GRASPs were removed. Because cell attachment and migration are mediated by cell adhesion molecules, in particular integrins, the level of a variety of integrins in GRASP KO cells was assessed. Among the 10 integrin subunits tested, which can form 8 different heterodimers, α5β1 integrins, the major and well-characterized integrin complex in HeLa, and MDA-MB-231 cells that utilizes fibronectin as its ligand (Mierke et al., 2011), exhibited the most robust reduction upon GRASP depletion. Further analysis showed that GRASP depletion reduces α5β1 integrin levels not only in the cell, but also at the cell surface, providing a reasonable explanation how GRASP depletion reduces cell attachment, and migration (Ahat et al., 2019). Consistently, exogenous expression of α5β1 integrins rescues the attachment and migration defects in GRASP-depleted cells. The effects of GRASP depletion on α5β1 integrins are specific for Golgi unstacking, as disruption of the Golgi ribbon by knocking down Golgin-84 or by nocodazole treatment, or destroying the Golgi structure by brefeldin A treatment, does not reduce integrin levels in cells (Xiang et al., 2013).

There are three possibilities to reduce the α5β1 integrin protein levels in cells by GRASP depletion: decreased synthesis,

accelerated degradation, or both. As Golgi structural defects caused by GRASP-depletion may impair protein glycosylation, which plays an important role in protein stability (Shental-Bechor and Levy, 2008), it was initially speculated that Golgi unstacking may impair α5β1 glycosylation and thus reduces their stability. However, the results demonstrated that the reduction of α5β1 integrin is due to decreased protein synthesis rather than increased degradation (Ahat et al., 2019). Interestingly, GRASPdepletion significantly increases total protein synthesis and accelerates cell proliferation (Ahat et al., 2019), consistent with the previous report that GRASP65 depletion accelerates cell cycle progression (Tang et al., 2010). How GRASP depletion selectively decreases the synthesis of α5β1 integrin while increasing overall protein synthesis remains a mystery. One possibility is that the overall protein synthesis is increased due to accelerated cell proliferation and enhanced protein trafficking. In addition, it has been proposed that the Golgi functions as a signal hub, as many signaling molecules are docked on the Golgi membranes and respond to different cellular stresses (Mayinger, 2011; Makhoul et al., 2018). So it is also possible that α5β1 integrin synthesis is regulated by signaling pathways on the Golgi in response to Golgi structural changes caused by GRASP depletion.

# NEW DISCOVERIES ON GRASP55 AND GRASP65 INTERACTING PROTEINS

Recently, some new discoveries have been made on the known interacting partners of GRASPs. Golgin-45 is one of the earliest identified GRASP55-binding proteins, which is involved in vesicle tethering and Golgi structure regulation (Short et al., 2001; Lee et al., 2014). The structural basis of GRASP55 interaction with Golgin-45 has recently been revealed. The last C-terminal residues of Golgin-45, QGELIAL, insert into the canonical PDZpeptide binding pocket in the PDZ1 domain of GRASP55, while the upstream residues of the C-terminal sequence of Golgin-45, TRYENITFNCCNHC, interacts with both PDZ domains by inserting into the cleft between them. Furthermore, the C-terminus of Golgin-45 also binds the PDZ2 domains of the two neighboring GRASP55 molecules, which enhances GRASP55 oligomerization. This is thought to play an important role in Golgi stacking. The third interaction site between Golgin-45 and GRASP55 is a unique zinc finger-like structure formed between Cys393/Cys<sup>396</sup> of Golgin-45 and His<sup>18</sup> (β1)/Cys<sup>103</sup> (β2) of GRASP55 (Zhao et al., 2017). Similar to the Golgin-45 and GRASP55 interaction, GM130 interacts with GRASP65 via its C-terminal KITVI sequence that binds PDZ1, and via the IPFFY sequence that interacts with both PDZ domains by inserting into the hydrophobic cleft between them. But unlike the Golgin-45 and GRASP55 interaction, GRASP65 undergoes conformational change on PDZ domain upon GM130 interaction but does not form a zinc-finger structure on GM130-GRASP65 interaction (Hu et al., 2015; Zhao et al., 2017).

In the past several years, a number of new GRASP binding proteins have been identified (**Figure 1** and **Table 1**). Two interacting proteins have recently been discovered for GRASP65, the actin elongation factor Mena (mammalian enabled homolog), and the Hsc70 (heat shock cognate 71 kDa protein)

co-chaperone DjA1 (DnaJ homolog subfamily A member 1). Mena is recruited to the Golgi membranes by GRASP65 to facilitate actin polymerization and GRASP65 oligomerization, and thus functions as a bridging protein in Golgi ribbon linking (Tang et al., 2016). DjA1 binds to GRASP65 and promotes GRASP65 oligomerization in a Hsc70-independent manner (Li et al., 2019b).

Most recently, several novel GRASP55 binding partners have been identified that are related to the newly discovered function of GRASP55 in autophagy. GRASP55 not only facilitates autophagosome-lysosome fusion via the interactions with LC3 on autophagosomes and LAMP2 on lysosomes (Zhang and Wang, 2018a,b; Zhang et al., 2018), but also directly binds Beclin-1 (BECN1) and UVRAG to facilitate the assembly



and membrane association of the phosphatidylinositol 3-kinase (PtdIns3K, or PI3K) complex (Zhang et al., 2019), and therefore plays an important role in autophagosome maturation during both glucose depletion and amino acid starvation. Moreover, GRASP55 binds cystic fibrosis transmembrane conductance regulator (CFTR) and transforming growth factor beta 1 (TGFβ1) to facilitate their unconventional secretion (Gee et al., 2011; Nüchel et al., 2018). These new findings reveal novel roles for GRASPs in cellular activities outside of the Golgi. While some of these findings have been recently reviewed elsewhere (Li et al., 2019a), several key advances are discussed below in detail.

### GRASP55 REGULATES AUTOPHAGOSOME-LYSOSOME FUSION

How the Golgi copes with different stresses and whether there is a Golgi stress sensor have not been systematically studied. In an effort to explore how the Golgi responds to energy deprivation, a number of Golgi proteins were examined for O-GlcNAcylation, a cytosolic glycosylation that serves as an energy sensor to regulate cellular pathways (Zhang et al., 2018). In this study, Zhang et al. discovered that GRASP55, but not other Golgi matrix proteins examined, including GRASP65, GM130 and Golgin-45, is O-GlcNAcylated under growth condition. Upon glucose starvation, GRASP55 is de-O-GlcNAcylated and forms puncta outside of the Golgi area. Since glucose starvation induces autophagy, the function of GRASP55 in autophagy was then tested. Indeed, depletion of GRASP55, but not GRASP65, increased the number of autophagosomes but decreased the autophagic flux, indicating a defect in autophagosome-lysosome fusion.

How does a Golgi protein like GRASP55 regulate autophagy? The study by Zhang et al. (2018) revealed that GRASP55 de-O-GlcNAcylation upon glucose deprivation allows some of the GRASP55 molecules to colocalize with autophagosomes. GRASP55 targeting to autophagosomes is regulated by O-GlcNAcylation, as it is enhanced by mutating the O-GlcNAcylation sites on GRASP55 as well as by glucose starvation which reduces GRASP55 O-GlcNAcylation. Further biochemical studies demonstrate that GRASP55 interacts with LC3-II on autophagosomes, and this interaction is enhanced by GRASP55 de-O-GlcNAcylation. In addition, GRASP55 also interacts with LAMP2 on lysosomes. These together suggest a possibility that GRASP55 may function as a bridging protein to facilitate LC3-LAMP2 interaction as well as autophagosomelysosome fusion. Indeed, this possibility was subsequently confirmed using in vivo and in vitro approaches (**Figure 2B**; Zhang et al., 2018). In cells, GRASP55 depletion reduces LC3 and LAMP2 colocalization as well as autophagosome-lysosome fusion. In vitro, GRASP55 facilitates autophagosome-lysosome fusion in an in vitro fusion assay. Furthermore, the addition of recombinant GRASP55 enhances LC3 and LAMP2 coimmunoprecipitation from cell lysates (Zhang et al., 2018). Thus, like in Golgi stacking, GRASP55 oligomers serve as membrane tethers to facilitate autophagosome-lysosome fusion.

The role of GRASP55 in autophagosome-lysosome fusion is not limited to glucose starvation, but also in amino acid starvation. Upon amino acid starvation, GRASP55 not only physically interacts with LC3 and LAMP2, but also regulates the formation of the PI3K UVRAG complex that is known to facilitate autophagosome-lysosome fusion (Zhang et al., 2019). Here, GRASP55 directly interacts with Beclin-1, induces the UVRAG PI3K complex formation and increases its membrane association. These reports identified GRASP55 as a specific energy and nutrient sensor on the Golgi to regulate autophagy.

Since GRASP55 still appears on autophagosomes in the presence of the protein synthesis inhibitor cycloheximide, it is speculated that GRASP55 is targeted to autophagosomes from an existing pool upon autophagy induction (Zhang et al., 2018). One remaining interesting question concerns how GRASP55 is targeted to autophagosomes. GRASP55 is unlikely translocated to the autophagosomes with the entire Golgi as cargo for autophagic degradation, since no other Golgi markers are found in the newly formed autophagosomes, and GRASP55 is localized on the outer membrane of autophagosomes instead of the lumen (Zhang et al., 2018, 2019). One possibility is through vesicular transport, similar to the transmembrane protein Atg9 that is normally enriched in the Golgi and is translocated to autophagosomes upon autophagy induction. However, unlike Atg9, GRASP55 does not have a transmembrane domain, and how the translocation occurs selectively on de-O-GlcNAcylated GRASP55 remains unknown. Alternatively, it is possible that a small pool of GRASP55 constantly shuttles between the Golgi, cytosol, and autophagosomes; and the equilibrium is regulated by GRASP55 O-GlcNAcylation. Nevertheless, how GRASP55 is targeted to autophagosomes under stress conditions requires further investigation.

While the discovery of GRASP55 as a membrane tether in autophagosome-lysosome fusion is exciting, it is unknown how GRASP55 interplays with other known tethering proteins, in particular Rab7 and the HOPS complex, as well as the STX17- SNAP29-VAMP7/8 SNARE complex that mediates the fusion (Jager et al., 2004; Itakura et al., 2012; Jiang et al., 2014). It is possible that these proteins function sequentially during autophagosome-lysosome fusion. Alternatively, GRASP55 may function as an independent mechanism in autophagosomelysosome fusion.

#### GRASP55/65 AND UNCONVENTIONAL SECRETION

It was long believed that only proteins with canonical ER signal peptides at the N-terminus can be secreted out of the cell. Since the discovery of the Golgi-independent unconventional secretion of the cytokine interleukin 1 beta (IL-1β) (Muesch et al., 1990), more proteins without ER signal sequences (leaderless proteins) and some integral membrane proteins have been reported to be transported or secreted in a Golgi-independent manner (Kinseth et al., 2007; Schotman et al., 2008; Gee et al., 2011). Collectively, the non-canonical Golgi-independent secretion is referred to as unconventional protein secretion (UPS). Interestingly, although

UPS itself is Golgi-independent, it requires two important Golgi proteins, GRASP55 and GRASP65, in mammalian cells and their homologs in other model organisms. Whether unconventional secretion is a by-product of loss of Golgi functions, or viceversa, has not been ruled out due to the lack of molecular tools to manipulate Golgi structure, and function without affecting GRASPs. In a previous review (Rabouille, 2017), unconventional secretion is generally classified into four categories, type I is direct translocation of cargo molecules across the plasma membrane via pore formation, type II is unconventional secretion through ATPbinding cassette (ABC) transporters, type III is vesicle-mediated secretion of cytosolic proteins, and type IV is the Golgi bypassing transportation of integral membrane proteins. GRASPs and their homologs have been reported to play a role in type III and type IV UPS, which is the focus of this review in terms of unconventional secretion.

#### GRASPs and Unconventional Trafficking of Integral Membrane Proteins CFTR

The Phe508 deletion from CFTR results in the inhibition of its trafficking to the plasma membrane and ER-associated degradation. It has been shown that GRASP55 and GRASP65 are required for the unconventional secretion of both WT and DeltaF508 CFTR when ER-to-Golgi trafficking is inhibited by expressing a dominant negative Sar1 or Arf1, or by inhibiting vesicle fusion via overexpression of syntaxin 5 (Gee et al., 2011). GRASP55 is phosphorylated at Ser441 during certain ER stresses, monomerized, and located near the ER to regulate unconventional secretion of CFTR, which requires CFTR-GRASP55 interaction (**Figure 2C**; Kim et al., 2016; Gee et al., 2018). Similar to GRASP55, GRASP65 overexpression rescued the secretion defect of mutant CFTR, although GRASP65 was not extensively tested in the same studies.

One contradiction in these reports is that overexpression of the GRASP55-G2A mutant inhibits CFTR secretion, whereas expression of N-terminally tagged GRASP55, which has been shown to abolish its membrane targeting similar to the GRASP-G2A mutant, increases CFTR secretion (Gee et al., 2011; Heinrich et al., 2014; Kim et al., 2016). Considering that GRASP55 regulates CFTR secretion by direct binding, GRASP55 may serve as a cytosolic chaperone to regulate CFTR sequestration by autophagosomes or by multivesicular bodies (MVBs) (Noh et al., 2018). Here, GRASP55 phosphorylation and re-localization in response to ER stress is necessary for CFTR secretion. A recent report showed that GRASP55 regulates the IRE1 ER unfolded protein response (UPR) pathway (Chiritoiu et al., 2019). It is unclear why the medial/trans-Golgi protein GRASP55, but not the cis-Golgi protein GRASP65, is the main player in regulating this activity as GRASP65 is localized on the Golgi compartments closer to the ER.

#### Integrin

dGRASP, the Drosophila homolog of GRASP, regulates the Golgi bypassing, non-canonical trafficking of the alpha subunit of integrin during Drosophila wing development (Schotman et al., 2008). Unlike normal situations in which dGRASP is localized on the Golgi, during wing development, dGRASP localizes near the plasma membrane to regulate non-canonical secretion of alpha-integrin. It was proposed that dGRASP regulates integrin secretion via facilitating the fusion of integrin-containing vesicles with the plasma membrane (Schotman et al., 2008). Alternatively, GRASP depletion may cause mis-sorting of an unknown, plasma membrane destined vesicle fusion protein (i.e., SNARE), which subsequently affects the non-canonical secretion of integrin. Unlike CFTR, it is unclear if dGRASP binding is required for integrin secretion. Considering that GRASP55 translocates to the ER in CFTR secretion while dGRASP translocates to near the plasma membrane area in integrin secretion, GRASP may regulate the secretion of these two transmembrane cargoes through different mechanisms.

## GRASP55 and Endosome/ Autophagosome-Dependent Unconventional Secretion of Cytosolic Proteins

#### IL-1β

Interleukin 1 beta, the major cytosolic regulator of inflammation, was one of the first identified cargoes of UPS (Muesch et al., 1990). IL-1β and another cytokine, TGFβ1, are secreted via a non-canonical secretory pathway distinct from the conventional ER-Golgi pathway (Nüchel et al., 2018). IL-1β is exported either through vesicle mediated secretion or via Gasdermin-Dmediated pore formation at the plasma membrane (Kayagaki et al., 2015). More recently, it was revealed that IL-1β is first restrained in the intermembrane space of autophagosomes and then secreted by the fusion of autophagosomes with the plasma membrane (**Figure 2D**). This fusion process is mediated by SNAREs including Sec22b on autophagosomes, syntaxin 3 and syntaxin 4 on the plasma membrane, and SNAP-23 and SNAP-29 from the cytosol (Dupont et al., 2011; Kimura et al., 2017). Most recently, a question was raised on whether autophagy is indeed involved in IL-1β secretion (Chiritoiu et al., 2019).

Interestingly, depletion of GRASP55 or GRASP65 reduces IL-1β secretion (Zhang et al., 2015; **Figure 2D**). GRASP55 may regulate the secretion of IL-1β through its role in autophagy or by modulating the IREα/XBP-1 UPR pathway (Dupont et al., 2011; Zhang and Wang, 2018a; Chiritoiu et al., 2019; van Ziel et al., 2019). Considering the function of GRASP55 in autophagosomelysosome fusion, it is reasonable to speculate that GRASP55 may also be involved in autophagosome-plasma membrane fusion in this scenario.

#### AcbA/Acb1

GRASP65 and GRASP55 were originally identified in mammalian cells to regulate Golgi stacking and ribbon linking. Kinseth et al. (2007) provided the first evidence that GRASP regulates unconventional secretion in Dictyostelium. This unexpected finding came from the observation that knockout of GrpA, the GRASP homolog in Dictyostelium, results in a secretion defect of AcbA, a protein that is processed in the extracellular environment to produce the spore differentiation

factor-2 (SDF 2) (Kinseth et al., 2007; Levi and Glick, 2007). Later, it was reported that Acb1, the budding yeast homolog of ACBP, is also unconventionally secreted, which depends on Grh1 (the GRASP homolog in the budding yeast), core Atg genes, ESCRT machinery and SNAREs, but is independent of COPII coated vesicles (**Figure 2D**; Duran et al., 2010; Manjithaya et al., 2010).

Although GRASP is required for Acb1 secretion, it is not clear at which step GRASP is involved. It seems that the vesicles carrying Acb1 are distinct from autophagosomes because GFP-Grh1 colocalizes with neither the autophagosome marker Atg8 nor the phagophore marker Ape1 (Bruns et al., 2011). Instead, a novel membrane-bound compartment called CUPS (compartment for UPS) is required for Acb1 secretion. However, the formation and identity of this membrane structure are largely unknown (Ye, 2018).

#### OUTLOOK

It is clear that GRASP65 and GRASP55 have both Golgidependent and Golgi-independent functions. On the Golgi, GRASP trans-oligomers are the primary machineries for Golgi stack formation. The recent finding of reduced cell adhesion and migration under GRASP depletion strengthens the essential role of Golgi stacking in protein trafficking, modification, and signaling.

Under certain stress conditions or at certain stages of development, GRASPs also function outside of the Golgi, likely as membrane tethers as within the Golgi stacks. The findings are interesting, but several outstanding questions remain. For example, how is GRASP55 targeted to different locations outside of the Golgi? Do GRASPs function as tethers in the regulation of unconventional secretion? The recent findings of GRASP55 in autophagy may help address these questions. Furthermore, the role of GRASPs in unconventional secretion has been confirmed in a variety of systems, but not all types of unconventional secretion require GRASPs. For example, unconventional secretion of a misfolding-associated protein secretion (MAPS) cargo GFP1-10 and a cilia transmembrane protein Peripherin/rds is reported to be GRASP-independent (Tian et al., 2014; Lee et al., 2016). This indicates a heterogeneity of unconventional secretion and the specificity of GRASPs in the regulation of unconventional

#### REFERENCES


secretion of certain substrates. Future studies by comparing different cargo molecules may help understand the exact roles of GRASPs in unconventional secretion.

It is clear that GRASP proteins play important roles in unconventional secretion, but the underlying mechanism remains largely unknown. Both unconventional secretion of certain cargoes and autophagy are augmented under stress conditions. The newly uncovered roles of GRASP55 as an energy sensor in the Golgi and a membrane tether in autophagy indicate that it may serve as a stress sensor and an effector in stress response; and these roles may be linked to unconventional secretion of certain cargo molecules. GRASP55 may coordinate Golgi-dependent and Golgi-independent trafficking pathways in the cell under different conditions. In addition to autophagy, the emerging role of GRASP55 in the regulation of ER stress and UPR indicates that GRASP55 may affect unconventional secretion through UPR (Chiritoiu et al., 2019; van Ziel et al., 2019). Systematic studies on Golgi response to different stress stimuli and identification of novel GRASP interacting proteins under normal and stress conditions may shed light on the mechanism of GRASP55 and GRASP65 in Golgi-dependent and Golgi-independent functions.

# AUTHOR CONTRIBUTIONS

All authors contributed to the design, conception and manuscript preparation. All authors approved the publication of this study.

# FUNDING

This work was supported by the National Institutes of Health (Grants GM112786, GM105920, and GM130331), M-Cubed, and the Fastforward Protein Folding Disease Initiative of the University of Michigan to YW.

# ACKNOWLEDGMENTS

We thank Wang Lab members for their stimulating discussions. The figures are created using BioRender.




**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 Ahat, Li and Wang. 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.

# Multifaceted Rho GTPase Signaling at the Endomembranes

#### Santosh Phuyal\* and Hesso Farhan\*

Department of Molecular Medicine, Institute of Basic Medical Sciences, University of Oslo, Oslo, Norway

The Rho family of small GTPases orchestrates fundamental biological processes such as cell cycle progression, cell migration, and actin cytoskeleton dynamics, and their aberrant signaling is linked to numerous human diseases and disorders. Traditionally, active Rho GTPase proteins were proposed to reside and function predominantly at the plasma membrane. While this view still holds true, it is emerging that active pool of multiple Rho GTPases are in part localized to endomembranes such as endosomes and the Golgi. In this review, we will focus on the intracellular pools and discuss how their local activation contributes to the shaping of various cellular processes. Our main focus will be on Rho signaling from the endosomes, Golgi, mitochondria and nucleus and how they regulate multiple cellular events such as receptor trafficking, cell proliferation and differentiation, cell migration and polarity.

#### Edited by:

Jaakko Saraste, University of Bergen, Norway

#### Reviewed by:

Stéphanie Miserey-Lenkei, Centre National de la Recherche Scientifique (CNRS), France Heike Folsch, Northwestern University, United States

#### \*Correspondence:

Santosh Phuyal santosh.phuyal@medisin.uio.no Hesso Farhan hesso.farhan@medisin.uio.no

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology Received: 19 April 2019 Accepted: 28 June 2019

Published: 16 July 2019

#### Citation:

Phuyal S and Farhan H (2019) Multifaceted Rho GTPase Signaling at the Endomembranes. Front. Cell Dev. Biol. 7:127. doi: 10.3389/fcell.2019.00127 Keywords: Rac1, Cdc42, RhoA, RhoB, RhoD, Golgi, endocytosis, membrane trafficking

#### THE RHO GTPase FAMILY

The founding member of the Rho GTPase, termed Rho for Ras homolog, was identified in 1985 (Madaule and Axel, 1985). Shortly thereafter, two back-to-back papers elegantly demonstrated the functional importance of Rho and Rac in actin cytoskeleton assembly (Ridley and Hall, 1992; Ridley et al., 1992) driving the expansion of Rho GTPase biology. Since then, the Rho family of small GTPases has grown to include 20 separate proteins divided into seven subfamilies: Rho, Rac, Cdc42, Rnd, RhoD, RhoBTB, and RhoH (Hodge and Ridley, 2016). The majority of Rho family members undergo conformational switching between GTP-bound active and GDP-bound inactive states. This GTP/GDP cycling is tightly regulated by guanine nucleotide exchange factors (GEFs) and GTPase-activating proteins (GAPs). GEFs function as Rho GTPases activator by catalyzing the exchange of GDP for GTP, whereas GAPs facilitate hydrolysis of GTP leading to their inactivation. An additional layer of complexity is provided by guanine-nucleotide dissociation inhibitors (GDIs) that bind to the inactive pool of Rho GTPases and sequester them in cytosol (Garcia-Mata et al., 2011). Additional factors contributing to the complexity of Rho GTPase signaling is the crosstalk between its family members, distinct subcellular distribution of their GEFs and GAPs, and the post-transcriptional modifications such as phosphorylation, ubiquitination, and palmitoylation that regulate stability and spatial distribution of Rho GTPases.

The switch I and switch II domains of Rho GTPases undergo conformational change upon GTP-binding. Once active, Rho GTPases associate with membranes and selectively interact with downstream effectors and other scaffolding proteins to mediate a myriad of biological processes including reorganization of the actin cytoskeleton, regulation of membrane trafficking, cell motility and polarity. A large body of literature deals with signaling initiated by Rho GTPases and their effectors at or around the plasma membrane although intracellular pool of Rho GTPases and their

regulators are increasingly apparent. It is becoming clear that coordination of functionally active Rho GTPases, their GEFs and GAPs, as well as their effectors extend beyond the plasma membrane to multiple intracellular organelles. Understanding how cells decode these spatio-temporal signals to generate biologic outputs is a major area of investigation.

Here, we review emerging knowledge on biological processes and signaling events mediated by intracellular pools of Rho GTPases, with special emphasis on signaling patterns that are triggered or initiated by endomembrane localized Rho GTPases. In particular, we summarize recent studies that provide direct evidence for endosomal, Golgi, mitochondrial and nuclear pools of Rho GTPase signaling.

## RHO GTPase SIGNALING FROM ENDOSOMES

The endosomal system consists of pleomorphic membranous carriers that processes and transports a range of cargoes including active signaling receptors. The ability of endocytic organelles in generating compartmentalized signaling patterns by directional shuttling and/or retaining of signaling molecules into specific locations within the cell is well documented and has been extensively reviewed elsewhere (Palfy et al., 2012). In this review, we will summarize Rho GTPase signaling originating directly from the endosomal system and elaborate on how Rho family members utilize endosomes for signal propagation along the endocytic pathway (**Figure 1**).

From the Rho GTPase family, RhoB was among the first ones to be predominantly detected in the endocytic compartments (Adamson et al., 1992; Gampel and Mellor, 2002). Today, the list of endosome localized Rho GTPases has grown to include RhoD, Rac1, Cdc42, TCL, and TC10, although the underlying mechanism of their recruitment to endosomes remains to be clarified in the majority of cases. In the case of RhoB, the type of prenylation was suggested to be the major determinant of its precise localization. The geranylgeranylated form of RhoB was localized to late endosomes, while the farnesylated form was detected predominantly at the plasma membrane (Wherlock et al., 2004). Whether the kind of lipid modification is a deciding factor for all other endosome localized Rho GTPases remains to be addressed. Furthermore, the identity, localization and regulation of the prenylation factors responsible for differential RhoB prenylation is also poorly characterized. Additionally, it is also unclear how GEFs/GAPs for Rho GTPases are targeted to the endosomes. Nevertheless, the recruitment of Rho GTPases to endosomes generates spatially restricted signals that, as we will discuss in the following sections, has consequences for numerous cellular processes.

It can be expected that endosome-specific Rho GTPases represent a functionally distinct subset from those existing at the plasma membrane. Indeed, the farnesylated pool (plasma membrane localized) of RhoB appeared functionally distinct from the geranylgeranylated (endosome localized) pool in that treatment of cells with farnesyl-transferase inhibitors, which abolishes the farnesylated pool of RhoB, resulted in an increased recycling of endocytosed epidermal growth factor (EGF) receptor (Wherlock et al., 2004). In agreement with this, EGF-triggered signaling was prolonged in cells treated with farnesyl-transferase inhibitors. Endosomal RhoB was also shown to recruit the Ser/Thr kinase PRK1 to this compartment, resulting in its activation. Active PRK1 on endosomes then regulated the trafficking of EGFR in a way leading to prolonged signaling and preventing its degradation (Mellor et al., 1998; Gampel et al., 1999). Interestingly, PRK1 has been reported to phosphorylate the intermediate filament proteins vimentin and neurofilament as well as interact with α-actinin (Mukai, 2003). Whether RhoB-PRK1 regulated kinetics of EGF signaling shows actindependency remains to be explored in greater details. The fact that RhoB promotes endosome recycling appears to be true in several cell types. While RhoB controls EGFR recycling in epithelial cells, it also controls recycling of the integrin LFA-1 in T-lymphocytes, which regulates their migration (Samuelsson et al., 2017). These studies imply that RhoB targeted to endosomal compartment is an autonomous signaling entity. Supporting this view, cytoplasmic endosomes harboring RhoB were found to pick up inactive Src kinase and stimulate its activity en route to the plasma membrane (Sandilands et al., 2004). This type of translocation and activation of Src required the presence of RhoB containing endosomes since inactive pool of Src kinase accumulated around the perinuclear region in RhoB−/<sup>−</sup> cells growing on fibronectin (Sandilands et al., 2004). The exact molecular mechanism underlying RhoB mediated endosomal motility, however, remains elusive. Proteins central to actin dynamics such as Scar1, Dia1, and mDia2 are also associated with RhoB-positive endosomes (Sandilands et al., 2004; Fernandez-Borja et al., 2005; Wallar et al., 2007) suggesting that RhoBpositive cytoplasmic entities could regulate endosomal trafficking in a manner dependent on rearrangement of actin cytoskeleton. It is currently unclear which GEFs activate RhoB on endosomes and which GAPs control this activity. In addition, it is unknown whether endocytic Rab GTPases contribute to the regulation of the endosomal RhoB pool. A possible candidate is the Rhospecific GAP, DLC3 which has been localized to Rab8-positive membrane tubules, reminiscent of the endocytic recycling compartment (Braun et al., 2015). Depletion of DLC3 impairs transferrin receptor endocytosis and this effect was neutralized by simultaneous depletion of RhoA and RhoB.

Several Rho family members regulate some aspects of actin cytoskeleton dynamics, and accordingly, multiple studies have demonstrated the importance of endosomal pool of Rho GTPases for actin-based endocytic vesicle movement. For example, a role for RhoD in controlling the endocytic vesicle movement has been documented. It was shown that cells possessing active RhoD display reduced velocity of early endosome movement, thereby slowing down the membrane trafficking events (Murphy et al., 1996). This effect of RhoD likely involves actin-based mechanisms since RhoD-mediated recruitment of its effector hDia2C to early endosomes aligns them along actin filaments (Gasman et al., 2003). RhoD has also been reported to bind to the Rab5 effector Rabankyrin-5 on early endosomes (Nehru et al., 2013). Thereby, Rab5 and RhoD may cooperate to regulate internalization of EGFR. It remains to be tested if RhoB and

FIGURE 1 | Schematic illustration depicting Rho GTPase signaling from endosomes. A number of Rho family members residing at endosomes generate localized signaling output to regulate a wide array of biological functions.

RhoD might control bidirectional endosomal movement by directly regulating each other's activity. Such a crosstalk between Rho GTPases has already been demonstrated for RhoB and Rac1 (Marcos-Ramiro et al., 2016). In cytokine pre-treated vascular and microvascular endothelial cells, endosomal RhoB retains Rac1 in endosomes and negatively affects its activity to prevent endothelial barrier reformation (Marcos-Ramiro et al., 2016).

While RhoD and Rab5 cooperate to regulate endocytic trafficking, Rac1 required Rab5 for its endosomal recruitment (Palamidessi et al., 2008). Rac1 was detected by several groups on endosomes in different eukaryotic organisms (Strehle et al., 2006; Palamidessi et al., 2008; Menard et al., 2014). It was shown that formation of Rab5-positive early endosomes was a pre-requisite for endosomal recruitment of Rac1 and its GEF Tiam1 leading to Rac1 activation. This active, endosomal pool of Rac1 was then delivered to specific plasma membrane domains turning them into regions of localized actin cytoskeleton remodeling. The process of endocytic Rac1 delivery required another small GTPase, Arf6 (**Figure 1**). The Rab5-Rac1-Arf6 signaling circuit was crucial not only for localized actin dynamics and the morphology of cancer cells, but also for directed cell migration (Palamidessi et al., 2008). An analogous mechanism was found to account for concentration of Cdc42 at the leading edge of astrocytes (Osmani et al., 2010). In migrating astrocytes, Cdc42 and its GEF β-PIX were shown to co-localize on endosome-like structures. The localization of Cdc42 on endosome-like vesicles required Rab5, and the directed delivery of Cdc42 to the leading edge depended on Arf6, as was observed for Rac1. Hence, the concept of coupling of multiple small GTPases might be more universal than currently anticipated. While these studies further provide evidence for the intriguing possibility that endosomes serve as a hub for Rho GTPase activation and spatiotemporal signal generation, they also raise the question about how this mechanism is controlled. The kinase LRRK2 is unusual in that it harbors a GTPase (ROC) domain. LRRK2 was found to localize to endosomes and to play an important role in negatively controlling Rac1 activity on this site (Schreij et al., 2015). Loss of LRRK2 resulted in hyperactive Rac1 and loss of dendritic spines in neuronal dendrites. It is unclear whether LRRK2 regulates Rac1 through phosphorylation, or through a scaffolding effect. There is evidence for phosphorylation-dependent regulation of Rho GTPases (Kwon et al., 2000; Schoentaube et al., 2009; Schwarz et al., 2012). LRRK2 also possesses GTPase activity (Liu and West, 2017) and it remains to be tested whether this activity is involved in Rac1 regulation. Notably, LRRK2 is mutated in familial Parkinson's disease and was very recently shown to be activated by the Rab29 on endosomes as well as the trans-Golgi network (Purlyte et al., 2018). This provides new avenues for

future investigations of crosstalk of Rac1 or Cdc42 with Rab29 and linking this to the pathogenesis of Parkinson's disease.

Rac1 and Cdc42 signaling from endosomes extends beyond Rab5-positive endosomal entities. For instance, Rab11 positive recycling endosomes were shown to traffic Rac1 to immunological synapses (Bouchet et al., 2016). The interaction of Rac1 with Rab11 on recycling endosomes was facilitated by a Rab11 effector FIP3. This Rab11-FIP3-Rac1 tripartite complex controlled T-cell spreading, cortical rigidity and immunological synapse symmetry in a manner dependent on Rac1 (Bouchet et al., 2016). The effects of FIP3 on T-cell spreading and synapse symmetry were Rac1 dependent, because this effect was abolished in the presence of Rac1 inhibitor, NSC23766. We note here that this inhibitor disrupts the interaction of Rac1 with its GEF Tiam/Trio. Thus, it is possible that FIP3 recruits Tiam/Trio to endosomes to mediate localized Rac1 activation. Alternatively, it is possible that Rac1 is recruited to recycling endosomes in its active form. In some cases, endocytosis of certain cell surface receptors may initiate endosomal Rac1 activation. Indeed, c-Met internalization and its trafficking to perinuclear endosomes in response to HGF-stimulation was essential for Rac1 activation (Menard et al., 2014). Notably, optimal Rac1 activation required interaction of c-Met with Rac1 specific GEF Vav2 in the perinuclear endosomes to initiate Rac1-driven cell migration (Menard et al., 2014). Late endosomal Rac1 together with Cdc42 also responds to growth factor independent signals. Using fluorescence resonance energy transfer (FRET)-based biosensors, it was shown that constitutively active Rac1 and Cdc42 promoted actin-nucleation events to drive sorting of cargo into intraluminal vesicles (Kajimoto et al., 2018). Interestingly, this process was initiated by sphingosine 1-phosphate signaling to multivesicular endosomes localized Rac1 and Cdc42, and also required activity from GEFs PLEKHG2 and P-Rex1 (Kajimoto et al., 2018).

Finally, the Rho GTPase family members TCL and TC10 have both been described to reside in early endosomal compartments to control endocytic pathway (de Toledo et al., 2003; Kawase et al., 2006). Apart from that, TCL and TC10 have been relatively underexplored and a clear picture of signaling originating from endosomal pools of these two Rho GTPases is lacking. The activity of TC10 on exocytic vesicles was reported using a FRETbased probe. It was shown that TC10 activity drops sharply just before fusion of the vesicle with the plasma membrane (Kawase et al., 2006). This result indicates that GTP hydrolysis by TC10 is a critical step for vesicle exocytosis. This role of TC10 appears to be applicable to several types of vesicles such as delivery of EGFR to the cell surface (Kawase et al., 2006) as well as translocation of the glucose transporter GLUT4 to the cell surface upon insulin stimulation (Chiang et al., 2001). Because TC10 was also shown to localize to subdomains of the plasma membrane (Liu and West, 2017) it remains to be determined which pool of TC10 regulates exocytic vesicle trafficking.

Taken together, these studies establish that Rho GTPases and their GEFs/GAPs assemble into signaling entities at early/late endosomes (**Figure 1**). However, it is unclear whether active Rho GTPases at the early/late endosomal compartments stem from plasma membrane internalized active pool. So far, the requirement of GEFs for optimal endosomal Rho GTPase signaling output points toward active recruitment of Rho GTPases to the endosomes. However, such a demonstration by selective manipulation of local pools of GEFs in contrast to depletion of total cellular pools of GEFs is currently missing.

# RHO GTPase SIGNALING FROM THE GOLGI

Over two decades ago, Cdc42 was already noticed to localize to the Golgi (Erickson et al., 1996). Since then a number of other Rho GTPase family members, their GEFs and GAPs, and interaction partners of Rho GTPases have been detected at the Golgi apparatus (**Figure 2**). At the Golgi, Cdc42 interacts with components of the COPI coat, which was originally shown to be important for cellular transformation (Wu et al., 2000). A major function of coatomer is to mediate formation of COPI vesicles for retrograde transport from the Golgi back to the ER (Letourneur et al., 1994). Accordingly, Cdc42 was shown to regulate membrane deformation by coatomer components (Park et al., 2015). However, rather than affecting COPI transport back to the ER, Cdc42 appeared to promote intra-Golgi anterograde transport. Thus, active Cdc42 might introduce a bias toward anterograde versus retrograde COPI trafficking (Park et al., 2015). It should be mentioned that these results were primarily obtained with a fast cycling mutant of Cdc42 (Cdc42-F28L) and that it remains to be shown whether Cdc42 can introduce this traffic bias under physiologic conditions. Notably, the fact that several tumors exhibit secretion of ER chaperones, might be interpreted as a failure of COPI-based retention. Future work should test the possibility whether higher Cdc42 activity at the Golgi is involved in this phenomenon.

An important question is whether Cdc42 is active at the Golgi, which is best answered by live cell imaging. The first detection of active Cdc42 at the Golgi was made possible by FRET-based reporters (Nalbant et al., 2004). However, it remained unclear whether this pool is functionally relevant for canonical Cdc42 related processes such as regulation of the cytoskeleton, cell polarity or cell migration. An early indication that this pool is functionally relevant for these processes was the finding that coatomer-bound (i.e., Golgi localized) Cdc42 stimulates actin assembly, but inhibits recruitment of the molecular motor dynein to the nascent vesicles (Chen et al., 2005). This observation suggests a model where Cdc42 disassociates from the coatomer once the vesicle formation is completed, allowing dynein to bind and transport vesicles. In addition, the Golgi pool of Cdc42 was reported to mediate dynein and microtubule-dependent Golgi positioning in directionally migrating cell (Hehnly et al., 2010). Later, the Golgi matrix protein GM130 was proposed to act as a factor that selectively regulates the Golgi pool of Cdc42 without affecting the plasma membrane (Baschieri et al., 2014). GM130 binds to RasGRF, which was previously shown to inhibit Cdc42 activity (Calvo et al., 2011). Using GM130 to specifically modulate Cdc42 activity at the Golgi, it was demonstrated that this pool acts as a reservoir that supplies the leading edge plasma membrane of directionally migrating cells with active Cdc42 (Baschieri et al., 2014). This was in agreement with previous reports showing that

intra-Golgi trafficking and actin dynamics at the Golgi. The figure highlights function of some of the Rho GTPases at the Golgi.

cytoplasmic vesicles carrying Cdc42 are delivered to the leading edge (Osmani et al., 2010).

Another interesting question is whether the Golgi harbors GAPs and GEFs that would mediate Cdc42 activity at this organelle. A GAP for Cdc42, ARHGAP10, was reported to be recruited to the Golgi by active Arf1 (Dubois et al., 2005). It was revealed that active Arf1 via its interaction with ARHGAP10 controls Cdc42 activity at the Golgi, and in this way regulates Golgi structure and actin cytoskeleton dynamics at the Golgi (Dubois et al., 2005). Earlier report suggested that the GEF Tuba might play a role in modulating Cdc42 activity at the Golgi (Kodani et al., 2009). However, four independent groups have failed to localize Tuba to the Golgi, but rather found it to be confined to cytoplasmic vesicles or the cell surface (Salazar et al., 2003; Kovacs et al., 2006; Baschieri et al., 2014; Bruurs et al., 2018). Very recently, the Golgi matrix protein GCC88 was found to interact with the long form of intersectin-1, a Cdc42 GEF (Makhoul et al., 2019). Intersectin-1 was convincingly localized to the Golgi providing us with an excellent candidate that might mediate local activation of Cdc42. Further supporting a role for intersectin-1 at the Golgi is an earlier report showing that a small molecule (ZCL278) that inhibited Cdc42-intersectin1 interaction disrupted Golgi structure (Friesland et al., 2013). It will be interesting to see whether this drug could be used to test some of the aforementioned effects of Golgi-based Cdc42 functions such as COPI transport or cell transformation.

In addition to Cdc42, active RhoA have also been localized to Golgi (Quassollo et al., 2015). Using FRET-based biosensors, active RhoA was detected at Golgi outposts (GOPs) in neuronal dendrites (Quassollo et al., 2015). Reportedly, the biogenesis of GOPs occurs from dendrite-localized ER exit sites (Horton et al., 2005). However, it remained unclear whether GOPs could emanate through fission from the somatic Golgi. RhoA at the Golgi was activated downstream of lysophosphatidic acid initiating the activation of a cascade involving ROCK, LIMK1 and PKD1, thereby resulting in tubulation of the somatic Golgi and elongation of these tubules into dendrites (Quassollo et al., 2015). RhoA was also involved in activating dynaminmediated fission events in dendrites leading to separation of the Golgi tubules and the formation of GOPs. Whether this is a main pathway for GOP formation has to be tested in the future. Because GOPs are important for dendrite formation and branching, it is conceivable that they play an important role in synaptic integration and plasticity. Thus, potential roles

of Golgi-based signaling of RhoA in these processes merits future investigations.

A very recent example of an active role of RhoA at the Golgi is the identification of a cascade triggered by protease activated receptors at the cell surface that activate a RhoA GEF called GEF-H1. GEF-H1 mediates RhoA activation at the trans-Golgi, which in turn activates PKD, a Ser/Thr kinase of great importance for the biogenesis of post-Golgi carriers. Thereby, RhoA was shown to regulate cargo delivery for localized exocytosis at focal adhesions (Eisler et al., 2018).

More recently, Golgi localization of RhoD and its role in maintaining Golgi homeostasis was reported (Blom et al., 2015). Endogenous RhoD colocalized with the Golgi resident proteins, whereas ectopic expression of constitutively active and inactive RhoD alone or with its binding partner WHAMM (WASP homolog associated with actin, membranes, and microtubules) led to Golgi fragmentation (Blom et al., 2015). Using temperature sensitive VSV-G to monitor trafficking, it was found that RhoD deregulates anterograde vesicular transport from the endoplasmic reticulum (ER) to the plasma membrane. In cells where RhoD activity was perturbed, VSV-G was scattered in vesicular structures positive for GM130 that were sensitive to endoglycosidase-H cleavage, indicating that the VSV-G containing Golgi derived vesicles were not fully functional (Blom et al., 2015). This suggests a role for RhoD in the secretory pathway.

The recently identified RhoBTB-1,-2, and -3 are much larger than classical small GTPases, possess additional domains and are not regulated by the conventional GTPase cycle (Aspenstrom et al., 2007; Ji and Rivero, 2016). In fact, RhoBTB3 – the only RhoBTB family member that was reported to localize to Golgi – functions as an ATPase (Espinosa et al., 2009). RhoBTB3 is anchored at the trans-side of the Golgi, where it functions as an effector for Rab9, a GTPase localized on late endosomes that traffics to the trans-Golgi. Arrival of these Rab9-positive carriers to the Golgi, induces activation of the ATPase activity of RhoBTB3, which is required to remove the coating off these vesicles, thereby preparing them to fuse with the Golgi (Espinosa et al., 2009). Corroborating this, it was further demonstrated that Golgi-residing RhoBTB3 is important for maintaining Golgi architecture since its depletion resulted in Golgi fragmentation (Lu and Pfeffer, 2013). At the Golgi RhoBTB3 is part of a Cul3-RING-E3 ubiquitin ligase complex, which binds Cyclin E and targets its proteasomal degradation (Lu and Pfeffer, 2013). This finding highlights that a Golgilocalized signaling molecule plays a role in the cell cycle by regulating G1-S-phase entry, a phase where Golgi has not been involved previously.

A major function of Rho GTPases is to regulate the cytoskeleton as well as molecular motors. The actin-nucleating Arp2/3 complex was shown to localize to the Golgi and to play a role in its polarization during cell migration (Magdalena et al., 2003). The dynamics of Arp2/3 at the Golgi were later shown to be regulated by the Golgi-pool of Cdc42 (Dubois et al., 2005). Cdc42 was also shown to regulate actin-assembling formin family members FMNL -2 and -3 at the Golgi and to thereby regulate the architecture of this organelle (Kage et al., 2017). Further evidence linking Cdc42 and Golgi architecture via regulation of actin dynamics was provided by identifying the Cdc42 exchange factor ITSN-1 as an interaction partner for the Golgi matrix protein GCC88 (Makhoul et al., 2019). Importantly, this interaction was suggested to play a role in Golgi dispersal that is observed in neurodegeneration.

## RHO GTPase SIGNALING FROM MITOCHONDRIA

The first ever description of mitochondria localized Rho GTPase was provided by Boivin and Beliveau (1995). The authors subjected outer cortical region of kidney from Sprague–Dawley male rats to subcellular fractionation and probed for RhoA, Cdc42 and Rac1 in different subcellular fractions. While all three Rho GTPases were detected in plasma membrane and cytosolic fractions, only Rac1 was detected in mitochondriaenriched fractions (Boivin and Beliveau, 1995). Corroborating this, Velaithan et al. (2011) showed direct physical interaction between mitochondrial Rac1 and Bcl-2 in human cancer cell lines and clinical biopsies from B-cell lymphoma patients. A year later, Rac1 was also detected in mitochondria in alveolar macrophages isolated from asbestosis patients (Osborn-Heaford et al., 2012). This study further revealed that the C-terminal cysteine (Cys-189) residue of Rac1 is required for its mitochondrial import. Subsequent studies have since included neuronal cells in the repertoire of cell types exhibiting mitochondrial Rac1 (Natsvlishvili et al., 2015; Pan et al., 2018).

In addition to describing localization of Rac1, these studies also provide compelling evidence for Rac1 signaling on mitochondria. The direct interplay of Rac1 and Bcl-2 at mitochondria maintained a mild pro-oxidant intracellular milieu through increased intracellular superoxide levels, successively promoting the death inhibitory activity of Bcl-2 (Velaithan et al., 2011; **Figure 3**). Interestingly, inhibition of Rac1-Bcl-2 interaction in lymphoma cells restored death signaling in response to common chemotherapeutic agents (Velaithan et al., 2011). Elevated mitochondrial Rac1 activity in alveolar macrophages via its interaction with cytochrome c, increased oxidative stress and contributed to the development of pulmonary fibrosis (Osborn-Heaford et al., 2012). Furthermore, increased Rac1 signaling was proposed to play important roles in the regulation of neuroplasticity and prevention of apoptosis and autophagy via its association with sigma-1 receptor, inositol 1,4,5-trisphosphate receptor and Bcl-2 at the mitochondrial membrane (Natsvlishvili et al., 2015). In contrast to these findings, interference of Rac1-Bcl-2 complex either by inhibition or siRNA mediated depletion of Rac1 relieved mitochondrial oxidative stress and promoted neuronal survival in a focal cerebral ischemia in vivo in a diabetic rat model and a hyperglycemia-exposed PC-12 cell in vitro model (Pan et al., 2018). Hence, Rac1 signaling at the mitochondria could produce cell-type specific outcomes under different conditions.

Generally, mitochondria have largely been overlooked as sites for Rho GTPase signaling. Whether members of Rho GTPase

family other than Rac1 also localize to mitochondria and might have been missed because of their rapid shuttling in and out of the mitochondria remains a subject for future studies.

# RHO GTPASE SIGNALING FROM NUCLEUS

Because the outer nuclear membrane is part of the ER and it also is a membrane enclosed organelle, we will briefly review evidence for Rho GTPase signaling in and from this location. The C-terminal polybasic region of Rac1 harbors functional nuclear localization signals (K-K-R-K and K-R-K-R) that promote its nuclear translocation (Lanning et al., 2004; Navarro-Lerida et al., 2015). Reportedly, Rac1 also contains two internal nuclear export motifs (Navarro-Lerida et al., 2015). It was revealed that interaction of Rac1 with nucleophosmin-1 mediates its efficient nuclear export (Navarro-Lerida et al., 2015; **Figure 3**). Consequently, wealth of data describe nucleocytoplasmic shuttling of Rac1 and postulate functional implications of nuclear Rac1 signaling (Kraynov et al., 2000; Michaelson et al., 2008; Menard et al., 2014; Navarro-Lerida et al., 2015; Woroniuk et al., 2018). Such a nuclear localization signal is also present in the C-terminal polybasic region of several other Rho GTPase family members (Williams, 2003). Whether other Rho GTPases also shuttle in and out of the nucleus in a manner similar to Rac1 remains to be investigated. Rac1 cycling in and out of the nucleus seemingly depends on the cell cycle, with increased nuclear Rac1 during the late G2 phase (Michaelson et al., 2008). In addition to Rac1, its GEF Tiam2 is also present at the outer nuclear membrane (Woroniuk et al., 2018), providing the basis for Rac1 signaling at the nuclear envelope (**Figure 3**). The question next arises is what functional consequences does Rac1 signaling in nucleus or on its envelope might have. Given its pivotal role in actin dynamics, Rac1 signaling in the nucleus could lead to changes in nuclear shape and position in actin-dependent fashion. This could have consequences for cancer cell invasion since structural changes in shape and size, and deformity of the nucleus are decisive factors for invasion through tight gaps in the extracellular matrix (Friedl et al., 2011). An additional effect of Rac1 signaling was altered nuclear membrane fluidity and order (Navarro-Lerida et al., 2015), which might further be an important factor regulating the ability of the nucleus to deform during invasion. The actin mesh around the nucleus might also play a role in positioning of this organelle, because depletion of Rac1 GEF Tiam2 resulted in a failure to position the nuclei with cellular axis of migration (Woroniuk et al., 2018). Moreover, aggressive tumors display higher nuclear Rac1, a phenomenon that results in increased invasiveness in vitro by potentiating cytoplasmic RhoA signaling (Navarro-Lerida et al., 2015). Alternatively, Rac1 may directly interact with nuclear proteins to induce actin-independent changes in cells. On one hand, active Rac1 directly interacts with STAT3 and regulates its activity by promoting its phosphorylation (Simon et al., 2000). On the other hand, Rac1 interaction with nucleophosmin-1 attenuates Rac1 signaling and inhibits cell spreading (Zoughlami et al., 2013). Moreover, increased Rac1 shuttling into the nucleus accelerates cell division (Michaelson et al., 2008). Finally, a fraction of active monomeric Rac1 segregated in the nucleus from dimeric and inactive Rac1 in the cytoplasm upon induction of DNA damage (Hinde et al., 2014).

Despite lacking a canonical nuclear localization signal, RhoA has been reported to translocate to the nucleus (Baldassare et al., 1997; Dubash et al., 2011; **Figure 3**). When quiescent fibroblasts were stimulated with α-thrombin, RhoA was found to translocate to the nucleus and stimulate enzymatic activity of PLD leading to the production of phosphatidic acid and diacylglycerol (Baldassare et al., 1997). The precise biological consequences of these lipid species in the nucleus remains to be fully understood. Later, Garcia-Mata and co-workers also demonstrated that a pool of RhoA is present in the nucleus with a subset of its GEFs (Net1, Ect2) and GAPs (DLC1, p190 RhoGAP). Notably, RhoA and Net1 activity selectively increased in the nucleus upon DNA damage implicating nuclear Net1/RhoA activity in DNA damage signaling (Dubash et al., 2011).

Another Rho GTPase with potential nuclear effects, but no clear nuclear localization signal is RhoB, which was shown to localize to the nuclear envelope, and to interact with and regulate the transcription factor DB1 (Lebowitz and Prendergast, 1998; Gerald et al., 2013; **Figure 3**). The RhoB-DB1 interaction was important for sprouting and proliferation in primary human blood endothelial cells (angiogenesis) in favor of lymphatic endothelial cells (lymphangiogenesis). Accordingly, RhoB knockout mice exhibited reduced angiogenesis but enhanced lymphangiogenesis in response to wounding (Gerald et al., 2013). Neither the stimulus that triggers RhoB translocation, nor the temporal dynamics of this subcellular pool were determined. Elucidating these details will be important to gain a full understanding of spatiotemporal RhoB signaling.

A recent study has described a biologic role for nuclear Cdc42 (Liu et al., 2018). When cultured under stiff mechanical environment, Cdc42 translocated from cytoplasm to the nucleus in tumor repopulating cells (i.e., cancer cells with stem-like properties) (Liu et al., 2018). The nuclear translocation of Cdc42 elevated expression of Tet2, an epigenetic modifier involved in chromatin methylation. Tet2 expression leads to increased expression of p21 and p27, which induce a G1 phase arrest and thus dormancy. The fact that stiffnessmediated dormancy was observed in vivo suggests that nuclear Cdc42 activity might also be relevant in cancer (Liu et al., 2018). Future work is needed to test how essential Cdc42 is for this process and whether identifying drugs that inhibit nuclear translocation of Cdc42 might be useful for cancer therapeutics.

#### REFERENCES


# CONCLUDING REMARKS

It is becoming increasingly apparent that signals emanating from distinct subcellular pools of Rho GTPases is spatially and temporally regulated to generate diverse physiological outcomes. Owing to the recent advancements in Rho GTPase biosensors and microscopy techniques, our knowledge of the complexity of Rho GTPase signaling in time and space has significantly advanced. Nevertheless, many new and interesting questions are yet to be addressed. For example, our understanding of any potential Rho GTPase signaling originating at the ER, the largest organelle in the cell, remains primitive. ER extends throughout the cytoplasm and forms membrane contact sites with endosomes, Golgi, mitochondria and plasma membrane. Membrane contact sites are known to play well-defined roles in bidirectional transport of molecules as well as signal transmission. Since Rho GTPases are able to signal from endosomes, Golgi, mitochondria and plasma membrane, it is tempting to speculate that they may participate in the biogenesis of membrane contact sites, or use them as passages to change subcellular localization. Such speculations are supported by the observation in yeast that the Rho-like GTPase Gem1 (the homolog of mammalian Miro1) is localized to the ER-mitochondrial contact sites (Kornmann et al., 2011). The molecular machinery that recruits Rho GTPases and their GEFs/GAPs to distinct endomembranes has yet to be thoroughly characterized. Finally, we need to incorporate the diverse Rho GTPase signaling units together to better understand their crosstalk and biological outcome as a whole. This makes it clear that using drugs or tools that globally activate or inhibit Rho GTPases is unlikely to be of great use. Future efforts should focus on identifying tools and methods to specifically modulate subcellular pools.

### AUTHOR CONTRIBUTIONS

SP and HF researched the relevant literature, conceived, and prepared the manuscript.

#### FUNDING

This research in the Farhan lab was supported by funding from the Norwegian Research Council, the Norwegian Cancer Society, and the Anders Jahre Foundation.



following injury by regulating VEZF1-mediated transcription. Nat. Commun. 4:2824. doi: 10.1038/ncomms3824


in Golgi polarity in scratch wound models. Mol. Biol. Cell 14, 670–684. doi: 10.1091/mbc.e02-06-0345


**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 Phuyal and Farhan. 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.

# Organelle Inheritance Control of Mitotic Entry and Progression: Implications for Tissue Homeostasis and Disease

Fabiola Mascanzoni, Inmaculada Ayala and Antonino Colanzi\*

Institute of Biochemistry and Cell Biology, National Research Council, Naples, Italy

The Golgi complex (GC), in addition to its well-known role in membrane traffic, is also actively involved in the regulation of mitotic entry and progression. In particular, during the G2 phase of the cell cycle, the Golgi ribbon is unlinked into isolated stacks. Importantly, this ribbon cleavage is required for G2/M transition, indicating that a "Golgi mitotic checkpoint" controls the correct segregation of this organelle. Then, during mitosis, the isolated Golgi stacks are disassembled, and this process is required for spindle formation. Moreover, recent evidence indicates that also proper mitotic segregation of other organelles, such as mitochondria, endosomes, and peroxisomes, is required for correct mitotic progression and/or spindle formation. Collectively, these observations imply that in addition to the control of chromosomes segregation, which is required to preserve the genetic information, the cells actively monitor the disassembly and redistribution of subcellular organelles in mitosis. Here, we provide an overview of the major structural reorganization of the GC and other organelles during G2/M transition and of their regulatory mechanisms, focusing on novel findings that have shed light on the basic processes that link organelle inheritance to mitotic progression and spindle formation, and discussing their implications for tissue homeostasis and diseases.

Keywords: cell cycle, mitosis, golgi complex, organelles, mitotic spindle

# INTRODUCTION

Entry into mitosis requires major cell reorganization to allow the proper inheritance of the genetic material between the daughter cells (Champion et al., 2017). The cells follow a specific and coordinated series of events to complete a successful division cycle. Mitosis is triggered by a complex regulatory circuit that controls the activation of the Cyclin-dependent kinase 1 (CDK1)/CyclinB complex, which is the master regulator of the mitotic onset (Nigg, 2001). Entry into mitosis and its proper completion are under the surveillance of checkpoints, which arrest the progression of the cell cycle if DNA damage or spindle failures are detected (Nigg, 2001). The irreversible commitment to mitotic entry is associated with a rapid and profound reorganization of cell shape. The cells become progressively round, centrosomes separate, the microtubules (MTs) organize the spindle apparatus, chromatin is condensed, and the nuclear envelope is disassembled. The latter event allows the spindle to capture and segregate the chromosomes (Champion et al., 2017).

#### Edited by:

Jaakko Saraste, University of Bergen, Norway

#### Reviewed by:

Nobuhiro Nakamura, Kyoto Sangyo University, Japan Catherine Jackson, UMR7592 Institut Jacques Monod (IJM), France

> \*Correspondence: Antonino Colanzi a.colanzi@ibp.cnr.it

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 18 April 2019 Accepted: 04 July 2019 Published: 23 July 2019

#### Citation:

Mascanzoni F, Ayala I and Colanzi A (2019) Organelle Inheritance Control of Mitotic Entry and Progression: Implications for Tissue Homeostasis and Disease. Front. Cell Dev. Biol. 7:133. doi: 10.3389/fcell.2019.00133

Interestingly, it is now evident that also the proper reorganization and segregation of cellular organelles during mitosis is indispensable to ensure a correct cell division. Several studies have demonstrated that perturbations of the redistribution of specific organelles result in profound alterations of the cell division process (Jongsma et al., 2015), with potential consequences on tissue development and homeostasis. For example, epithelia are specialized animal tissues that form protective barriers lining the organs and the body. Each epithelium is characterized by a specific structural organization and cell composition. In response to constant turnover and environmental insults, epithelia maintain homeostasis through a tight balance of cell duplication, differentiation, and death (Tai et al., 2019). Moreover, cell divisions must also follow a predetermined orientation to preserve correct tissue architecture. Thus, errors in the control of the number and orientation of cell divisions can have detrimental consequences, compromising tissue development and/or function, and potentially leading to tumor progression (Ragkousi and Gibson, 2014).

Here, we review new findings related to the mechanisms underlying the mitotic segregation of subcellular organelles, and discuss how they impact on the proper mitotic progression, with an emphasis on those involved in the coordination of Golgi complex (GC) inheritance with the cell cycle.

#### PREPARATION OF CELL ENTRY INTO MITOSIS

The primary event of cell division is the segregation of sister chromatids between the daughter cells. This essential task is achieved by the spindle apparatus, which is a very complex system that is organized by the centrosomes and composed of MTs and hundreds of regulatory proteins (Heald and Khodjakov, 2015). The recruitment of the γ-Tubulin Ring Complex (γ-TuRC) to the pericentrosomal region stabilizes MT minus ends and creates anchor points associated to the microtubule-organizing center (MTOC). The γ-TuRC act as a template for the polymerization of α/β tubulin heterodimers into MTs, which then undergo rapid growth and catastrophe phases (Petry and Vale, 2015; Petry, 2016). The centrosomes organize three types of MT fibers: kinetochore, polar, and astral MTs (**Figure 1**). The kinetochore MTs are directed toward the kinetochores and are responsible for the traction forces required to separate the sister chromatids. The polar MTs form antiparallel fibers that exert sliding forces mediated by kinesins, and thus, accomplish both separation and elongation of the spindle poles in mitosis. The astral MTs associate with specific domains of the cell cortex (**Figure 1**; Petry, 2016).

The assembly of the spindle is regulated by phosphorylation events. The kinase Aurora A associates with the centrosomes and is a crucial regulator of mitotic entry and centrosome maturation, which consists of the recruitment of proteins involved in MT nucleation and anchoring (Marumoto et al., 2002). Aurora B is a fundamental player in chromosome segregation because it controls the chromosome-microtubule attachment and sister chromatid cohesion (Hindriksen et al., 2017). Pololike kinase 1 (PLK1) is involved in spindle maintenance by regulating MT dynamics through the phosphorylation of Microtubule-associated proteins (MAPs) and kinesin motors (**Figure 1**; Bruinsma et al., 2012). The concerted lengthening and shortening of the kinetochore, and of the polar spindle MTs, control the proper alignment of chromosomes at the spindle midzone (Heald and Khodjakov, 2015; Petry, 2016). The correct attachment of the chromosomes to the mitotic spindle is monitored by a spindle assembly checkpoint (SAC), which is activated in the case of incorrectly attached kinetochores, resulting in a delay of anaphase onset. The impairment of this checkpoint can lead to aneuploidy, aging and cancer progression (Lara-Gonzalez et al., 2012).

An additional crucial event during cell division is the orientation of the spindle apparatus, as it determines the proper orientation of the cell division axis (Lu and Johnston, 2013). After its formation, the spindle apparatus is subjected to a rotation process. The astral MTs associate with specific cues at the cell cortex, from where they are subjected to pulling forces (**Figure 1**). Thus, complex signaling pathways coordinate MTmediated forces that are originated at the cell cortex and are applied on the spindle MTs (Lu and Johnston, 2013).

A critical preparatory step for spindle formation is the progressive cell reshaping from a flat to a spherical geometry that occurs during the G2/M transition (Champion et al., 2017). The rounding process becomes detectable during late prophase, but the preparatory steps begin during G2, when the focal adhesions (FA) undergo selective disassembly. This process is essential to achieve the progressive retraction of the cell margins and the formation of the actomyosin cortex, which are necessary for the cell rounding in metaphase (Cadart et al., 2014; Champion et al., 2017). Cell rounding is necessary to create a symmetric cell organization that allows the kinetochore MTs to capture the chromosomes. In fact, a reduced rounding can cause the scattering of chromosomes, which decreases the probability of being captured by the kinetochore MTs. Besides, an asymmetric cell geometry can induce spindle deformations, resulting in the splitting of the spindle poles and formation of acentriolar spindles, thus increasing the likelihood of multipolar cell division and aneuploidy. Therefore, alterations of mitotic rounding can affect spindle morphology, chromosome segregation, and timely mitotic progression (Champion et al., 2017). In addition, the geometry of the cell during mitosis also influences the orientation of the spindle and, as a consequence, the direction of the cell division axis, and thus tissue morphogenesis and differentiation (Morin and Bellaiche, 2011). Collectively, these findings emphasize that correct cell geometry and the controlled redistribution of cell components are crucial for the daughter cells fate, and errors in the process can lead to the development of diseases (Lancaster et al., 2013).

Entry into mitosis is also accompanied by an extensive reorganization of subcellular organelles, which undergo stereotyped reorganization of the structure and localization (Champion et al., 2017). For instance, during G2 the centrosomes move from the perinuclear area to the center

of the nucleus in order to be disengaged and separated (Sutterlin and Colanzi, 2010). The reorganization of organelle morphology can range from the full disassembly of the GC to the subtle modifications of the endosomes (Lowe and Barr, 2007; Jongsma et al., 2015). In the next paragraphs, we will summarize the current knowledge about the mitotic fate of the intracellular organelles, with a focus on the GC, and referring to several recent reviews for more mechanistic details.

# THE GOLGI COMPLEX

# Structural Reorganization of the Golgi Complex During Mitosis

The GC has a pivotal role in the secretory pathway, as it is involved in the modification and sorting of cargoes (Wei and Seemann, 2010). In mammalian cells, the GC is characterized by a ribbon structure, which is composed of several polarized stacks of cisternae that are laterally connected by tubules (Lowe, 2011). The structure of the GC is maintained by the Golgi Reassembly Stacking Protein of 55 and 65 kDa (GRASP55 and GRASP65) (Zhang and Wang, 2015) and by the members of the golgin family (Gillingham and Munro, 2016). All these proteins act as membrane tethers and concur in the stacking of the cisternae and in directing the formation of the membranous tubules connecting the stacks (Xiang and Wang, 2010; Witkos and Lowe, 2015). The perinuclear position of the GC is maintained through the association with MTs (Maia et al., 2013).

During late G2, the Golgi ribbon is divided into individual stacks (unlinking) (**Figures 2A,B**). The pro-ribbon role of the GRASPs is inhibited by phosphorylation. The basic regulatory elements have been identified, and they include the kinase PKD that leads to a RAF1/MEK1/ERK1-mediated phosphorylation of GRASP55 (Valente and Colanzi, 2015). In addition, a major inducer of ribbon unlinking is phosphorylation of GRASP65 by JNK2 and PLK1 (Acharya et al., 1998; Colanzi et al., 2000; Feinstein and Linstedt, 2007; Sengupta and Linstedt, 2010; Tang and Wang, 2013; Rabouille and Linstedt, 2016). Moreover, Golgi unlinking requires the fissioninducing protein BARS to cleave the tubules connecting the stacks (Hidalgo Carcedo et al., 2004; Colanzi et al., 2007). For more mechanistic details the reader is referred to several reviews (Lowe and Barr, 2007; Corda et al., 2012; Tang and Wang, 2013; Ayala and Colanzi, 2017; Wei and Seemann, 2017). During prophase (**Figure 2C**), the activation of CDK1 leads to the phosphorylation of additional sites on golgins and GRASPs, resulting in complete inhibition of the membrane tethering processes. Furthermore, CDK1 also induces the phosphorylation of adaptor proteins involved in membrane fusion, such as p47/VCIP135 and p37, causing inhibition of membrane fusion events (Uchiyama et al., 2003; Kaneko et al., 2010; Tang and Wang, 2013; Totsukawa et al., 2013). An additional contribution to the inhibition of the membrane fusion machineries is provided by the

centrosome; GC, Golgi complex; NU, nucleus. Adapted with permission from Ayala and Colanzi (2017).

HACE1-mediated monoubiquitination of the SNARE protein syntaxin 5 (Huang et al., 2016). As a result, the Golgi membranes are consumed by an extensive vesiculation during metaphase, when they become dispersed into vesicular/tubular clusters, with some intermediate compartment (IC) and Golgi proteins redistributed into the Endoplasmic Reticulum (ER) (**Figure 2D**; Saraste and Marie, 2018). Then, during telophase and cytokinesis, the dispersed Golgi proteins and membranes are gradually reassembled into a GC in each of the daughter cells (**Figure 2E**; Shorter and Warren, 2002; Altan-Bonnet et al., 2004; Colanzi and Corda, 2007).

Importantly, the disassembly of the GC is also a requirement for mitotic entry. Indeed, blocking the unlinking step induces a potent G2 block of the cell cycle, pointing out that a mitotic "Golgi checkpoint" oversees the correct premitotic cleavage of the GC (Sutterlin et al., 2002; Hidalgo Carcedo et al., 2004). Although the existence of a mechanism that controls the correct partitioning of the organelle could be surprising, recent findings are starting to reveal the general framework, and the evidence suggests that a physical/functional connection of the GC with the centrosomes and the MT network is at the basis of this novel checkpoint.

## Role of the Centrosome in the Localization and Structure of the Golgi Complex in Interphase

It is already known that the structure and localization of the GC can be modulated by the centrosome through multiple mechanisms (Sutterlin and Colanzi, 2010; Rios, 2014). The centrosome is composed of two centrioles enclosed by pericentriolar material (PCM), which consists of a thick shell of multiprotein complexes. In most cell types, the centrosome represents the major MTOC, and is involved in the formation of radial MT fibers (Sanchez and Feldman, 2017; Muroyama and Lechler, 2017). The template for MT nucleation is the γ-TuRC, which is recruited to the centrosome. The centrosome is positioned at the cell center, close to the nucleus. Following a polarization stimulus, it is reoriented in the direction of the leading edge of the cell (Pouthas et al., 2008).

The centrosomal MTs form radial fibers that guide the positioning of the Golgi membranes toward the cell center thanks to dynein, which is a minus end-directed motor complex (Rios, 2014) that is recruited at the GC by Golgin160 (Yadav et al., 2012). Also, the actin cytoskeleton contributes to the maintenance of the ribbon, as it forms tracks for actin-based motors (Valderrama et al., 1998). Besides, a subset of MTs is nucleated at the GC (Chabin-Brion et al., 2001). Nucleation of MTs from the GC is driven by a multiprotein complex that is organized by the scaffold protein AKAP-450, which is recruited at the cis-Golgi by GM130 (Zhu and Kaverina, 2013). The newly nucleated MTs are then stabilized and anchored at the trans-Golgi network (TGN) through the MT binding proteins CLIP-associated proteins 1 and 2 (CLASP1/2), due to their ability to stabilize MT plus ends by suppressing MT catastrophes (Efimov et al., 2007; Wu et al., 2016). The simultaneous knockdown of both CLASPs reduces MT nucleation at the GC and causes mitotic defects, such as multipolar spindle formation, and consequent cytokinesis failure (**Table 1**; Mimori-Kiyosue et al., 2005; Lansbergen et al., 2006; Pereira et al., 2006).

The GC-based MT nucleation is crucial not only for the structural integrity of the GC, but also for the formation of asymmetric MTs that are essential for the orientation of the GC toward the leading edge during migration (Vinogradova et al., 2009; Kaverina and Straube, 2011; Wu and Akhmanova, 2017), and for the polarized delivery of cargoes (Miller et al., 2009; Sanders and Kaverina, 2015). Yet, the significance of the GC-centrosome proximity and of the ribbon organization are not completely understood. In particular, the knockdown of the golgin GMAP210 or Golgin160 induces the unlinking into separated stacks, which are still able to transport cargoes to the cell surface but that become unable to direct the secretion toward specific domains of the plasma membrane (PM) at the leading edge. As a result, the directional persistence of cell migration is reduced (Yadav et al., 2009). In agreement with these observations, depletion of GM130 using siRNA in UO2S or HeLa cells results in ribbon unlinking and reduced efficiency of cell migration (Kodani and Sutterlin, 2008). Additionally, experiments based on the expression of various N-terminal fragments of AKAP-450 led to the conclusion that the proximity of the GC to the centrosome, but not the presence of an intact ribbon, is the crucial factor for optimal directional cell migration (Hurtado et al., 2011). However, experiments based on RPE1 cells in which GM130 was knocked out led to the conclusion that a close association of the GC with the centrosome is not required for cell migration or protein transport (Tormanen et al., 2019). Therefore, more investigations are needed to better understand the functional consequences of perturbations of the GC structure, or of its proximity to the centrosome.

# Physical and Functional Relationships Between the Golgi Complex and the Centrosome During Mitosis

The structural reorganization of the GC during the cell cycle appears to be coordinated with those of the centrosome (Sutterlin and Colanzi, 2010). The centrosomes are duplicated during S-phase; then, during G2, they are pulled apart, in coincidence with the severing of the Golgi ribbon (**Figure 2B**; Persico et al., 2010). Thus, the GC is segregated into two groups of stacks, each of which is localized in proximity to a separated centrosome (**Figure 2C**). Furthermore, during this phase, the membranes of the IC remain closely associated with the centrosomes and become detached from the bulk of the GC, suggesting that the IC maintains its identity during mitosis and provides an intermediate station for Golgi dispersal (Marie et al., 2012; Saraste and Marie, 2018). In concomitance with these events, the composition and size of the PCM material are profoundly modified, involving the recruitment of other components, like Cep192/SPD-2, PCNT/PLP, and Cep215/Cnn, in a process that is defined as "centrosome maturation"(Joukov et al., 2014). The "mature" centrosomes reach their final position and orientation in metaphase, when they direct MT nucleation for the formation of the spindle, which is fundamental for correct segregation of the chromosomes into the daughter cells (Meraldi and Nigg, 2002). Defects in assembly and duplication of the centrosomes, and the consequent problems in MT nucleation, are the primary cause of the formation of aberrant spindles. In support of a functional GC-centrosome relationship, the G2-specific Golgi ribbon unlinking acts as a controller of the centrosomal recruitment of Aurora A (Persico et al., 2010), which is a major regulator of G2/M transition, centrosome maturation, and spindle formation (**Figures 2B,C**; Marumoto et al., 2002). In particular, Barretta et al. (2016) demonstrated that upon unlinking, Src is activated at the GC, then Src interacts with Aurora A and phosphorylates the residue Y148, increasing the kinase activity of Aurora A, which then is recruited to the centrosomes to induce their maturation (**Figure 2a**). Aurora A is a pivotal switch of spindle formation, as its inhibition or ablation causes formation of multipolar and/or fragmented spindles (Marumoto et al., 2002; Hoar et al., 2007; Malumbres and Perez de Castro, 2014). An additional line of evidence of the correlation between mitotic Golgi disassembly and centrosomebased functions, has been revealed thanks to an assay designed to prevent the disassembly of the Golgi stacks during mitosis through the controlled formation of 3,3<sup>0</sup> -diaminobenzidine (DAB) polymers in the Golgi lumen (Guizzunti and Seemann, 2016). Cells containing DAB polymers in the Golgi stacks entered into mitosis normally, but they arrested in metaphase with intact Golgi clusters associated with monopolar spindles, which caused SAC activation. Artificial disassembly of the GC relieved this block, suggesting that the disassembly of the Golgi stacks is required for progression through mitosis (Guizzunti and Seemann, 2016; Wei and Seemann, 2017).

In addition, several reports have shown a direct role of Golgi matrix proteins in assisting spindle formation. For example, the N-terminal domain of GM130 includes a nuclear localization signal (NLS) that has been shown to be essential for proper spindle assembly (Wei et al., 2015). During interphase, the NLS is masked by the interaction with the Golgi matrix protein p115. CDK1-mediated phosphorylation of GM130 dissociates p115 from GM130, and this triggers a crucial pathway of mitotic disassembly of the Golgi stacks (Nakamura et al., 1997; Nakamura, 2010; Wei et al., 2015), as it unmasks the NLS, which


TABLE 1 | Golgi located proteins involved in spindle formation.

becomes able to bind and sequester at the Golgi the nuclear pore component importin-α. Of note, before mitosis onset, importinα is bound to the spindle assembly factor TPX2 (**Figure 2b**), keeping this protein inactive. Accordingly, as a consequence of the binding of importin-α to GM130, TPX2 becomes free to interact with Aurora A, resulting in the increase of its kinase activity and local stimulation of MT nucleation required for the assembly of the spindle. Furthermore, once the spindle fibers are formed, they become "stabilized" by GM130 (**Figure 2c**), which directly binds and bundles MTs, thus linking Golgi membranes to the spindle (**Figures 2C,D**; Wei et al., 2015; Wei and Seemann, 2017). Probably correlated to this function, depletion of GM130 causes the formation of over duplicated centrosomes and multipolar spindles during mitosis (**Table 1**), resulting in metaphase arrest and cell death (Kodani and Sutterlin, 2008).

To further support the functional connection of the GC with the spindle, several reports have shown evidence of Golgi-associated proteins that influence spindle formation and mitotic progression. In this regard, the GM130 interactor p115 becomes associated with the mitotic spindle throughout mitosis. A specific armadillo-like fold of the N-terminus of p115 was responsible for its interaction with γ-tubulin and centrosomal targeting. Strikingly, p115 depletion causes spindle abnormalities, chromosome defects, and cytokinesis failure (**Table 1**; Radulescu et al., 2011). Also, the other GM130 interactor, GRASP65, has been endowed with mitosis-specific roles. Indeed, GRASP65 depleted cells show multiple disorganized and non-functional spindle asters (**Table 1**), indicating that GRASP65 regulates MT dynamics during entry into mitosis (Sutterlin et al., 2005). The list of Golgi-associated proteins with roles in spindle formation is not limited to the GM130-based protein complex. For instance, depletion of the Golgi-associated phosphoinositide phosphatase SAC1 causes perturbations of Golgi architecture and spindle abnormalities (**Table 1**; Liu et al., 2008). In addition, tankyrase-1 is an ADP-ribosyltransferase that is associated with the Golgi in interphase, and relocates to the spindle poles during mitosis. Its depletion causes mitotic arrest with abnormal chromosome segregation, bipolar spindle formation, and failure of telomere separation (**Table 1**; Chang et al., 2005). Also, the tankyrase-1 substrate Miki translocates from the GC to the centrosomes during the late G2/M phase. Depletion of Miki induces a pseudometaphase state that leads to the formation of multinucleated cells (**Table 1**; Ozaki et al., 2012). Another Golgiassociated protein implicated in cell cycle control is the Rad50 interacting protein RINT-1, whose depletion causes partial Golgi fragmentation, centrosome amplification during interphase, and increased formation of multiple spindle poles that culminate in frequent chromosome missegregation (**Table 1**; Lin et al., 2007).

More in general, the spindle recruits and directs the inheritance of Golgi matrix proteins that are involved in the formation of the Golgi ribbon, while a minimal set of proteins and membranes sufficient to reassemble functional Golgi stacks are inherited independently of the spindle (Wei and Seemann, 2009). It could be speculated that the Golgi matrix proteins recruited by the spindle are not simple passengers, but acquire different mitosis-specific functions. In support of this possibility, during mitosis, the small GTPase Arf1 becomes inactive and dissociates from the Golgi membranes (Altan-Bonnet et al., 2003), and this correlates with the dispersal of several peripheral Golgi proteins. If Arf1 is artificially kept active, Golgi membranes do not fragment, and the peripheral proteins remain associated with the GC throughout mitosis. These cells enter mitosis, but exhibit gross defects in chromosome segregation and cytokinetic furrow formation, resulting in multinucleation (Altan-Bonnet et al., 2003).

Thus, there is a substantial amount of evidence to conclude that an active functional interplay between the GC and the centrosome is crucial for spindle formation and, hence, for accurate segregation of the genetic material.

#### STRUCTURAL REORGANIZATION OF OTHER ORGANELLES DURING MITOSIS

#### Endoplasmic Reticulum

The ER is a large continuous membranous organelle that is responsible for the synthesis of the majority of the integral membrane proteins and lipids. It comprises three different

domains: the smooth ER (SER), the rough ER (RER), and the nuclear envelope. The ER constitutes a vast network of cisternae and tubules spread across the cytosol, and establishes contacts with several subcellular compartments (Phillips and Voeltz, 2016).

The ER undergoes marked structural modifications during mitosis. Specifically, the cisternae are transformed into mixed populations of tubules, the extent of which varies among cell lines (Puhka et al., 2007, 2012). In addition, during late prophase, the nuclear envelope is disassembled and its membranes reabsorbed into the bulk of the ER to expose the chromatin to the spindle apparatus. In prometaphase, the ER is split into two large pools of membranes that maintain continuity throughout mitosis (Lu et al., 2009; Schwarz and Blower, 2016). During anaphase and telophase, after chromosomal segregation, the nuclear envelope reassembles, and this marks the beginning of the reorganization of the ER compartment, although the underlying molecular mechanisms are poorly understood (Schwarz and Blower, 2016).

The ER is also a major hub for intracellular organization and signaling (Jongsma et al., 2015; Schwarz and Blower, 2016). Indeed, throughout interphase it forms multiple contact sites (CS) with the PM, endosomes, GC, and cytoskeleton (Phillips and Voeltz, 2016). Membrane CSs have crucial functions in interorganelle signaling and lipid transfer. Considering its role as a major organizing compartment, it is likely that the inheritance of the ER is regulated by yet unknown control mechanisms. Recently, it has been shown that the ER–PM CSs undergo significant changes in morphology and function during mitosis (Yu et al., 2019). One of the major functions of CSs is to control Ca2<sup>+</sup> signaling through the store-operated Ca2<sup>+</sup> entry (SOCE), which depends on the integrity of ER–PM CS to allow contact between the proteins STIM1 and Orai1 (Yu et al., 2019). During mitosis, the density of these specific CSs is decreased. Also, the average distance between the PM and the closest ER in mitosis is increased. Therefore, the down-regulation of ER–PM junctions in mitosis induces SOCE inhibition by preventing the interaction of STIM1 with Orai1. This inhibition could affect not only Ca2<sup>+</sup> signaling but also lipid metabolism and membrane structure (Yu et al., 2019).

#### Endosomes

The membranous endocytic system mediates the traffic of lipids, proteins, and other molecules among various intracellular locations. These features have an essential impact on signal transduction and nutrient acquisition. Even if the various cellular functions of the endolysosomal system are extensively investigated, the mechanisms of endosome inheritance are marginally known. The current view is that endosomes and lysosomes remain intact during mitosis, and that during cytokinesis these organelles accumulate in the proximity of the MTOC (Bergeland et al., 2001; Jongsma et al., 2015). Despite this limited knowledge, an interesting aspect is that during prophase, MT-dependent motors induce the clustering of Rab11 positive endosomes around the centrosomes (**Figures 2C,c**). This localization is important to prevent their transport to the PM and to bring MT-nucleating proteins to the centrosome (Hehnly and Doxsey, 2014). For this reason, Rab11 relocation to the centrosome is necessary for the formation of a functional and properly oriented mitotic spindle. The clustered Rab11 endosomes are then segregated by the mitotic spindles between the daughter cells (**Figures 2D,E**; Hehnly and Doxsey, 2014). Moreover, in line with the proposed role of membrane reservoir in mitotic cells, during anaphase, the recycling endosomes are transported toward the cleavage furrow (**Figure 2E**; Bergeland et al., 2001), where they are believed to provide the membranes required to complete the cytokinesis (Simon and Prekeris, 2008).

## Peroxisomes

The peroxisomes are organelles involved in fatty acid and energy metabolism. During prophase, they are associated with the MT network and remain clustered around the spindle poles (**Figures 2C,D,c**). Throughout telophase, they are repositioned around the reforming nucleus of each daughter cell (**Figure 2E**; Hettema and Motley, 2009). Importantly, an intriguing connection of peroxisome inheritance with tissue development has been demonstrated. The peroxisome-associated protein PEX11b has been found to be essential for the differentiation of the skin, which undergoes a continuous renewal that is required to ensure a healthy tissue turnover (Asare et al., 2017). Specifically, the peroxisomes of PEX11bdepleted cells are functional, indicating a marginal role of this protein for the most classical peroxisome roles. However, while in control cells the peroxisomes localize at the spindle poles during mitosis, in PEX11b knockdown cells they fail to localize properly, resulting in a mitotic delay and SAC activation (Asare et al., 2017). Furthermore, PEX11b deficiency is associated to uncontrolled rotations of the spindle, which should normally be oriented perpendicularly to the basal membrane. The localization of peroxisomes at the spindle poles is the crucial factor, as their artificial relocalization to the cell cortex, or the spindle midzone, is sufficient to cause the alterations of spindle orientation. Importantly, in mouse embryos that are knockdown of PEX11b, the inability of the basal stem cells to orient their spindle perpendicularly to the basal membrane also led to alterations of their differentiation. These perturbations had severe effects, as the epidermis showed hyperproliferation and increased expression of terminal differentiation markers in basal cells, which is a feature typically associated with cancer (Asare et al., 2017). Thus, these data further support the notion that correct organelle inheritance/positioning is crucial for spindle formation and alignment, and they also revealed for the first time that these defects can have direct consequence on the development and maintenance of a stratified epithelium, where spindle orientation and cell differentiation are critical factors for establishing tissue structure and maintain homeostasis.

#### Mitochondria

Mitochondria are the energy factories of cells and are characterized by the presence of a double membrane. The inner membrane is folded into numerous cristae, which increase the surface area, and contain circular DNA together with the protein components needed for transcription and

translation (Friedman and Nunnari, 2014; Suarez-Rivero et al., 2016). Mitochondria are also major stores of Ca2<sup>+</sup> ions. During cell proliferation, the mitochondria have to be segregated into daughter cells (Friedman and Nunnari, 2014). Upon entry into mitosis, the mitochondrial network is cleaved into fragments that are uniformly dispersed within the cytoplasm by an ordered mechanism of inheritance mediated by MTs. During cytokinesis, the mitochondria are recruited to the cleavage furrow, where they remain localized until completion of cell abscission (**Figures 2D,d**; Lawrence and Mandato, 2013a,b).

An important feature of this organelle is that it undergoes a dynamic equilibrium of fusion and fission processes. The fusion of mitochondria is operated by the dynamin-like GTPases mitofusins 1 and 2 (Mfn1 and Mfn2), and optic atrophy 1 (Opa1). Mfn1 and Mfn2 are localized on the mitochondrial outer membrane, while Opa1 resides on the inner membrane (Friedman and Nunnari, 2014). The fission process is mediated by the dynamin-related protein 1 (DRP1). Knockout of any of these proteins is embryonic lethal in mice (Friedman and Nunnari, 2014; Suarez-Rivero et al., 2016). During mitosis, Aurora A phosphorylates the small Ras-like GTPase RALA, which localizes to mitochondria and triggers the formation of a complex with RALBP1 and CDK1/CyclinB, inducing the phosphorylation of DRP1 to stimulate mitochondrial fission (Kashatus and Counter, 2011). The knockdown of either RALA or RALBP1 leads to the inhibition of mitochondrial division. As a result, the cells become unable to evenly distribute mitochondria between the daughter cells, resulting in cytokinesis defects (Kashatus et al., 2011). Interestingly, experiments performed in mammalian stem-like cells revealed that they have the capability of asymmetrically sorting young and old mitochondria. The daughter cell that maintains stem cell features is the one that receives most of the new mitochondria (Katajisto et al., 2015). This is important for maintaining stemness, as the block of mitochondrial fission impedes the asymmetric distribution of mitochondria, and results in loss of stem cell properties in the daughter cells (Katajisto et al., 2015).

#### CONCLUDING REMARKS

Emerging evidence indicates that intracellular organelles undergo coordinated changes in shape and/or localization during mitosis. The preliminary steps for these reorganizations begin during G2, when the FAs are selectively dismantled through a mechanism induced by the G2-specific expression of the CDK1/CyclinB complex (Jones et al., 2018) and DEPDC1B, which is a scaffold protein that localizes to FAs, where it inhibits RhoA signaling (Marchesi et al., 2014). The selective dismantling of FAs generates a specific pattern of residual active integrin-based adhesive structures, which could drive the pulling forces that are exerted on MT fibers to direct the repositioning of the centrosome and, as a consequence, of the GC and the nucleus (Champion et al., 2017). These steps are required for cell rounding, which in turn is fundamental for correct spindle formation and orientation. Proper orientation of the division axis is crucial for correct tissue development and the homeostatic processes that maintain the steady state composition and organization of a tissue (Morin and Bellaiche, 2011).

The drastic and rapid modifications of the structure and localization of subcellular organelles during mitosis could be necessary not only to allow their inheritance, but also to remove possible steric interferences with the assembly of the spindle. Furthermore, it is also emerging that the endomembrane machineries have an active role in the formation of the spindle apparatus. The most investigated example is offered by the GC, whose mitotic partitioning can be schematically divided into consecutive steps (Persico et al., 2009). Specifically, the G2-specific unlinking of the Golgi ribbon stimulates the recruitment and activation of Aurora A at the centrosome, which is a necessary step for cell entry into mitosis (**Figures 2B,a**; Ayala and Colanzi, 2017; Barretta et al., 2016). Then, the disassembly of the Golgi stacks, which occurs after mitosis onset, regulates events that are necessary for the correct formation of the mitotic spindle (**Figures 2C,b,c**; Wei et al., 2015; Guizzunti and Seemann, 2016). Similar requirements are also emerging for other organelles: the recycling endosomes are relocated around the centrosome to be divided in daughter cells (**Figures 2D,c**) (Hehnly and Doxsey, 2014); the peroxisomes have to be correctly localized at the spindle to allow its correct orientation (**Figures 2D,c**) (Kashatus et al., 2011; Asare et al., 2017); and mitochondria fission is necessary for cytokinetic furrow closure (**Figures 2E,d**; Kashatus et al., 2011).

As a consequence, these events have to be coordinated by signaling pathways. A potential integrator of such signals is Aurora A, which is activated by GC unlinking during G2 to induce centrosome maturation (Barretta et al., 2016), and is also activated by a GM130-based pathway to induce spindle formation after mitosis onset (Wei et al., 2015). In addition, the activity of Aurora A is regulated by FA dismantling, which allows the relocation of some FA-associated scaffolds to the centrosomes, where they interact with Aurora A to protect this kinase from the action of phosphatases (Pugacheva and Golemis, 2006; Pugacheva et al., 2007). Interestingly, Aurora A is also necessary for the fission of mitochondria (Kashatus and Counter, 2011). Then, during mitosis, protein machineries that ordinarily operate at the organelle level are repurposed to direct the proper spindle formation and orientation, and the cleavage of the mitotic furrow (**Table 1**).

Thus, with this review we highlight that in addition to DNA segregation, the accomplishment of a correct cell division also requires the proper segregation of intracellular organelles, with likely important implications for organism development and tissue homeostasis. Until now, the only evidence of the importance of correct organelle inheritance for organism development has been provided by the effects of peroxisome misplacement, which results in structural alterations of the epidermis (Asare et al., 2017), and of the lack of

DEPDC1B expression, which leads to severe defects of zebrafish morphogenesis (Marchesi et al., 2014). We anticipate that additional important functional roles will be revealed when novel strategies to specifically perturb the mitotic inheritance of other organelles will be developed.

# AUTHOR CONTRIBUTIONS

All authors contributed to the critical reading of the literature and discussion, writing the text, collecting the references, and preparing the figures.

#### REFERENCES


## FUNDING

The authors would like to thank the Italian Foundation for Cancer Research (AIRC, Milan, Italy; IG 2017 id. 20095 to AC), and the P.O.R FESR Campania SATIN (European Community; MIUR, Italy; Campania Region, Italy) for financial support.

### ACKNOWLEDGMENTS

The authors apologize to all the colleagues whose work has not been cited in this review, due to space limitations.



Ragkousi, K., and Gibson, M. C. (2014). Cell division and the maintenance of epithelial order. J. Cell Biol. 207, 181–188. doi: 10.1083/jcb.201408044


**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 Mascanzoni, Ayala and Colanzi. 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.

# Characteristics and Functions of the Yip1 Domain Family (YIPF), Multi-Span Transmembrane Proteins Mainly Localized to the Golgi Apparatus

#### Shaheena Shaik<sup>1</sup> , Himani Pandey<sup>2</sup> , Satish Kumar Thirumalasetti1,3 and Nobuhiro Nakamura1,2 \*

<sup>1</sup> Graduate School of Life Sciences, Kyoto Sangyo University, Kyoto, Japan, <sup>2</sup> Faculty of Life Sciences, Kyoto Sangyo University, Kyoto, Japan, <sup>3</sup> Department of Biotechnology, Vignan's University, Guntur, India

#### Edited by:

Vladimir Lupashin, University of Arkansas for Medical Sciences, United States

#### Reviewed by:

Bruno Goud, Centre National de la Recherche Scientifique (CNRS), France Martin Lowe, The University of Manchester, United Kingdom

> \*Correspondence: Nobuhiro Nakamura osaru3@cc.kyoto-su.ac.jp

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 15 April 2019 Accepted: 02 July 2019 Published: 30 July 2019

#### Citation:

Shaik S, Pandey H, Thirumalasetti SK and Nakamura N (2019) Characteristics and Functions of the Yip1 Domain Family (YIPF), Multi-Span Transmembrane Proteins Mainly Localized to the Golgi Apparatus. Front. Cell Dev. Biol. 7:130. doi: 10.3389/fcell.2019.00130 Yip1 domain family (YIPF) proteins are multi-span, transmembrane proteins mainly localized in the Golgi apparatus. YIPF proteins have been found in virtually all eukaryotes, suggesting that they have essential function(s). Saccharomyces cerevisiae contains four YIPFs: Yip1p, Yif1p, Yip4p, and Yip5p. Early analyses in S. cerevisiae indicated that Yip1p and Yif1p bind to each other and play a role in budding of transport vesicles and/or fusion of vesicles to target membranes. However, the molecular basis of their functions remains unclear. Analysis of YIPF proteins in mammalian cells has yielded significant clues about the function of these proteins. Human cells have nine family members that appear to have overlapping functions. These YIPF proteins are divided into two sub-families: YIPFα/Yip1p and YIPFβ/Yif1p. A YIPFα molecule forms a complex with a specific partner YIPFβ molecule. In the most broadly hypothesized scenario, a basic tetramer complex is formed from two molecules of each partner YIPF protein, and this tetramer forms a higher order oligomer. Three distinct YIPF protein complexes are formed from pairs of YIPFα and YIPFβ proteins. These are differently localized in either the early, middle, or late compartments of the Golgi apparatus and are recycled between adjacent compartments. Because a YIPF protein is predicted to have five transmembrane segments, a YIPF tetramer complex is predicted to have 20 transmembrane segments. This high number of transmembrane segments suggests that YIPF complexes function as channels, transporters, or transmembrane receptors. Here, the evidence from functional studies of YIPF proteins obtained during the last two decades is summarized and discussed.

Keywords: membrane traffic, interactome, Rab/Ypt proteins, ER–Golgi transport, vesicle budding, vesicle fusion

**Abbreviations:** CDD, Conserved Domain Database; ERGIC, ER–Golgi intermediate compartment; HUGO, Human Genome Organization; SGD, Saccharomyces Genome Database; TGN, trans-Golgi network; YIPF, Yip1 domain family.

# YIPF PROTEINS OF Saccharomyces cerevisiae AND THEIR PROPOSED FUNCTIONS

Yip1 domain family proteins are multi-span transmembrane proteins localized mainly to the Golgi apparatus. Yip1p, a founding member of the YIPF proteins, was found to interact with Ypt1p and Ypt31p, which are homologs of mammalian Rab1 and Rab11, respectively (Yang et al., 1998). Ypt1p and Ypt31p are Ypt/Rab family small GTPases localized at the Golgi apparatus, and essential for ER to Golgi and intra-Golgi transport in Saccharomyces cerevisiae (Lipatova et al., 2015). Yip1p was shown to be required for ER to Golgi and intra-Golgi transport consistent with the proposed functions of Ypt1p and Ypt31p (Yang et al., 1998). Furthermore, a temperature sensitive mutant of Yip1p showed synthetic lethality with Ypt1p and Ypt31p, strongly suggesting that the interactions of Yip1p with Ypt1p and Ypt31p are essential for ER to Golgi transport. Later, Yif1p was found to interact with Yip1p (Matern et al., 2000). Yif1p formed a complex with Yip1p, also interacted with Ypt1p and Ypt31p, and was similarly shown to be essential for ER to Golgi transport. Based on these results, it was proposed that the Yip1p–Yif1p complex binds Ypt1p and Ypt31p to play an essential role(s) in ER to Golgi transport (Matern et al., 2000).

Yip1p and Yif1p both have paralogs, Yip4p and Yip5p, respectively. Yip4p and Yip5p were found to interact with Yip1p and Yif1p and also with Ypt/Rab GTPases (Calero et al., 2002). In contrast with Yip1p and Yif1p, Yip4p and Yip5p were found to be non-essential for viability (Giaever et al., 2002; Pearson and Schweizer, 2002), but the loss of Yip4p and Yip5p did produce some notable phenotypes, including abnormal vacuolar morphology and decreased endocytosis, suggesting a significant role in the membrane trafficking pathway, likely at the late Golgi compartment (Burston et al., 2009; Michaillat and Mayer, 2013). Yip5p strongly interacted with Yip4p, suggesting that these two proteins function together in a complex, similar to Yip1p and Yif1p (Calero et al., 2002). This idea is supported by our biochemical analysis of human homologs, YIPF1, YIPF2, and YIPF6 (Soonthornsit et al., 2017), and a phylogenetic analysis, which is described later. The loss of Yip1p could be compensated for by overexpression of Yif1p, but not by Yip4p, suggesting that Yip1p–Yif1p and Yip4p–Yip5p have non-overlapping function(s) (Calero et al., 2002).

It has been shown that Ypt1p functions in ER to Golgi and intra-Golgi transport at the vesicle docking/fusion step, while Ypt31p functions in trans-Golgi to plasma membrane transport and in endosome to trans-Golgi transport at the vesicle budding step (Lipatova et al., 2015). Therefore, YIPF proteins are proposed to function in vesicle budding and/or fusion at the Golgi apparatus. This is supported by the results of interactome analyses, which showed that YIPF proteins (Yip1p, Yif1p, Yip4p, Yip5p) form a core physical interaction network with selections of Ypt/Rab GTPases (**Figure 1**). This core network is connected with other gene products that function in membrane trafficking, including SNAREs, COPII components (**Table 1**; Ito et al., 2000; Uetz et al., 2000; Heidtman et al., 2003; Chen et al., 2004; TABLE 1 | Physical interactors of budding yeast YIPF proteins.


Physical interactors of all YIPF proteins were combined and categorized according to functional significance. The genetic interactors for YIPF genes are indicated by bold letters. Refer SGD for the full dataset and additional information.

Vollert and Uetz, 2004; Inadome et al., 2007; Lorente-Rodríguez et al., 2009). Furthermore, genes that function in membrane trafficking have been shown to interact genetically with YIP1, YIF1, YIP4, and YIP5 (**Supplementary Table S1**). These include COPI and COPII coats, vesicle tethering factors (COG, TRAPP, and GARP complexes), proposed cargo receptors (p23 members), and regulators of Ypt/Rab and ARF family GTPases (ARFGAPs, GEFs for Ypt/Rab) (**Supplementary Table S1**; refer SDG; Saccharomyces Genome Database<sup>1</sup> ).

Involvement of Yip1p and Yif1p in vesicle budding/fusion was evaluated by an in vitro vesicle budding and fusion assay by two independent groups. Ferro-Novick's group reported that antibodies for Yip1p and Yif1p inhibited vesicle fusion to the Golgi apparatus, but neither vesicle budding from the ER nor vesicle packaging of Yip1p and Yif1p were affected (Barrowman et al., 2003). The antibody must be added at the vesicle budding step for inhibition, suggesting that the Yip1p– Yif1p complex is involved in establishing fusion competence of ER to Golgi transport vesicles at the vesicle budding step. In contrast, Barlowe's group, using a similar in vitro assay, reported that antibodies against Yip1p inhibited COPII vesicle budding from the ER, but not tethering or fusion of the vesicles to Golgi membranes (Heidtman et al., 2003). yip1-4, a temperature

<sup>1</sup>https://www.yeastgenome.org/

sensitive mutant of YIP1, did not cause vesicle accumulation at a restrictive temperature, supporting their in vitro assay results.

Analysis of Yip1A, mammalian homolog of Yip1p, supported its function in vesicle budding (discussed later) (Tang et al., 2001). In addition, YIP1 genetically interacts with GOT1, which is a tetra-spanning small membrane protein that is predicted to function in ER to Golgi transport at the vesicle budding step (Lorente-Rodríguez et al., 2009). GOT1 was identified as a multicopy suppressor of yip1-2, which is a temperature sensitive mutant of YIP1. Got1p cycles between the ER and the Golgi apparatus, and its overexpression causes a complex extension of the ER membrane and simultaneous disruption of the Golgi apparatus, suggesting it functions at either the ER export step or vesicle budding. However, Got1p was not found to form a stable complex with Yip1p, and deletion of GOT1 did not affect the localization of Yip1p. Therefore, the mechanism for the suppression of Yip1p function appears to be indirect, and the relationship of Got1p and Yip1p still requires further clarification.

Because Yip1p and Yif1p are efficiently packaged into COPII vesicles, they must recycle between the ER and the Golgi apparatus (Otte et al., 2001; Barrowman et al., 2003). Considering their interaction with SNAREs, COPI and COPII coats, and other associated factors that regulate the ER to Golgi transport as described above, it is reasonable to assume that Yip1p and Yif1p play roles in both vesicle budding and fusion to coordinate vesicle flow between the ER and the Golgi apparatus.

Interestingly, YIPF proteins also interacted with Atg20p, Ypt35p, Vam7p, and Vps17p, which are PX domain containing proteins (**Figure 1**; Vollert and Uetz, 2004). The PX domain binds phosphatidylinositol 3-phosphate and functions to recruit the proteins to endosomal membranes (Teasdale and Collins, 2012). Most of the PX domain containing proteins are classified as sorting nexins (SNXs). SNXs bind to retromer components that function in the recycling of cargo proteins from endosomes to TGN or plasma membrane (Gallon and Cullen, 2015). Therefore, it is possible that YIPF proteins function also in the endosome to TGN transport. However, the significance of the binding between YIPF proteins and SNXs has not been analyzed so far.

# CONSERVATION OF YIP1P, YIF1P HOMOLOGS IN EUKARYOTES

Our early BLAST search analysis identified mammalian homologs of YIP1, YIF1, YIP4, and YIP5 from the protein sequence and EST databases (Shakoori et al., 2003). Multiple sequence alignment revealed that these proteins, now called the YIPF proteins, commonly have multiple hydrophobic segments with scattered hydrophilic residues on their C-terminal

side (a revised protein sequence alignment result is shown in **Supplementary Figure S1**). Many of these proteins were predicted to have five transmembrane helices by membrane topology analyses using TMpred or SOSUI (Hofmann and Stoffel, 1993; Hirokawa et al., 1998), although some were predicted to have three transmembrane helices, likely because of a rather high number of hydrophilic residues in the hydrophobic segments. Yeast two hybrid analyses indicated that the N-termini of yeast and human YIPF proteins are exposed to the cytoplasm because N-terminal tagging of Gal4 domains showed interaction with similarly tagged cytoplasmic Ypt proteins (Yang et al., 1998; Matern et al., 2000; Calero et al., 2002; Shakoori et al., 2003). Biochemical analyses showed that all of the examined human YIPF proteins exposed their N-terminal hydrophilic regions to the cytosol and short C-terminal hydrophilic regions to the lumen of the Golgi apparatus (**Figure 2**; Tang et al., 2001; Shakoori et al., 2003; Yoshida et al., 2008; Tanimoto et al., 2011). From these results, it was predicted that YIPF proteins have an odd number of transmembrane segments, most probably five, with an N-terminal cytoplasmic region exposed to the cytoplasm and a short C-terminal region exposed to the lumen of the Golgi apparatus (Shakoori et al., 2003). The transmembrane region is composed of multiple hydrophobic segments and is well conserved within YIPF proteins, while the N- and the C-terminal regions are less conserved (**Supplementary Figure S1**). The conservation of the transmembrane region was confirmed by bioinformatics and is now annotated as "Yip1 domain" in CDD (Marchler-Bauer et al., 2017). Thus, human homolog proteins were called the "YIPF" by the HUGO gene nomenclature committee, except YIF1A and YIF1B which are homologs of S. cerevisiae Yif1p (Yoshida et al., 2008; Marchler-Bauer et al., 2017).

Yip1 domain family protein sequences were found in virtually all eukaryotes including protists, fungi, animals, and plants

FIGURE 2 | The structure of YIPF proteins. Schematic representation of the structure of a YIPF protein. The light blue band indicates the lipid bilayer. Brown squares connected by solid lines indicate transmembrane segments. The regions of the three conserved motifs are shown by yellow squares and the consensus sequences were shown (refer the text for the explanation of the motifs).

(**Table 2** and **Supplementary Table S2**). To our surprise, proteins containing a domain similar to YIPF are even found in prokaryotes belonging to the phylum euryarchaeota (COG2881), and bacteria, including Escherichia coli (pfam06930: DUF1282) (**Supplementary Table S3**; Makarova et al., 2015; Marchler-Bauer et al., 2017). Many of these proteins are now annotated as "Yip1 family protein," although these prokaryotic protein sequences were distantly related to eukaryotic family members (**Supplementary Figures S1**, **S2**). No function has been reported for these prokaryotic family members, and the significance of their similarity to the eukaryotic family members requires further investigation.

Phylogenetic analysis after multiple sequence alignment using CLUSTALW revealed that YIPF proteins were divided into two large subfamilies represented by S. cerevisiae Yip1p and Yif1p, respectively (**Supplementary Figure S2**). Each of these large subfamilies was further divided into three small subfamilies. The three small subfamilies in Yip1p subfamily are represented by YIPF5 (Yip1A), YIPF4, and YIPF6, while those in the Yif1p subfamily are represented by YIF1A, YIPF3, and YIPF1 (**Figure 3** and **Table 3**). For these six smaller subfamilies, Yip1p, Yip4p, Yif1p, and Yip5p were grouped with YIPF5 (Yip1A), YIPF6, YIF1A, and YIPF1, respectively (**Supplementary Figure S2** and **Table 3**). Orthologs for Yip1p, Yif1p, Yip4p, and Yip5p were found in all eukaryotes (**Table 2** and **Supplementary Table S2**) except in diplomonads (e.g., Giardia intestinalis) and foraminiferans (e.g., Reticulomyxa filosa), in which only a part of orthologs were found at present (**Supplementary Figure S2**; indicated in white characters). This result strongly suggests that Yip1p, Yif1p, Yip4p, and Yip5p play a fundamental function(s) that is conserved in most eukaryotes. Interestingly, orthologs of YIPF3 and YIPF4 were only found in holozoa, which includes animals (**Table 2** and **Supplementary Figure S2**), but not in holomycota, which includes S. cerevisiae and other fungi (**Supplementary Table S2** and **Supplementary Figure S2**) although both holozoa and holomycota are grouped in uniconta. Orthologs of YIPF3 and YIPF4 were found in filasterea and choanoflagellatea, which are single cell organisms closely related to metazoa (**Table 2** and **Supplementary Figure S2**), suggesting that YIPF3 and YIPF4 were evolved in a common ancestor of holozoans that later evolve into metazoa. Curiously, in Ecdysozoa, orthologs of YIPF3 and YIPF4 were found in nematoda, e.g., Caenorhabditis elegans, but not in arthropoda, including many insects, i.e., Drosophila melanogaster (**Table 2** and **Supplementary Figure S2**). Therefore, it is tempting to speculate that the emergence of YIPF4 and YIPF3, probably by gene duplication from Yip1p and Yif1p or Yip4p and Yip5p, respectively, once played a role in the evolution of metazoans, but those proteins were later lost during the evolution of arthropods.

In most of the mammals and fishes belonging to Teleostomi, e.g., zebrafish (Danio rerio), two close paralogs were found for Yip1p (YIPF5/Yip1A and YIPF7/Yip1B), Yif1p (YIF1A and YIF1B), and Yip5p (YIPF1 and YIPF2), suggesting that gene duplication and subsequent functional divergence occurred to fulfill the needs of vertebrates (**Table 2** and **Supplementary Figure S2**). Curiously, some of these paralogs have not been found in amphibia or aves (birds) at present

TABLE 2 |

Conservation of YIPF proteins in holozoan species.


Orthologs for each YIPF family member were BLAST searched in the NCBI database (March 2019), limiting the indicated taxa, and multiple alignment of the selected sequences was performed with mammalian family members to identify an ortholog for each family member. +, the presence of the orthologs in most of the species in the indicated taxa; an asterisk, the presence of the orthologs in only limited species in the taxon; −, no ortholog found.

TABLE 3 | Summary of nomenclature of YIPF members.


(**Table 2** and **Supplementary Figure S2**). This may indicate the loss of those paralogs in those organisms, although genome/transcriptome analyses of these organisms may still remain incomplete.

# PROPOSAL OF A NEW SYSTEMATIC NOMENCLATURE OF YIPF PROTEINS

Human Genome Organization adopted the nomenclature of human YIPF family members based on the original gene names from S. cerevisiae, YIP1 and YIF1. HUGO also partly adopted our YIPF family numbering system that is based on the order of the cloning of corresponding cDNAs in our laboratory, except YIF1A and YIF1B (Shakoori et al., 2003). This nomenclature is now out of date and confusing, as it does not reflect the localization or complex forming behavior of the YIPF proteins. In addition, S. cerevisiae has a similarly named protein, Yip3p (the mammalian homolog is PRA1), which is also an integral membrane protein with an N-terminal cytoplasmic domain, albeit with only two transmembrane segments (Calero and Collins, 2002; Sivars et al., 2003).

Therefore, we propose renaming the YIPF proteins to clarify the relationship of family members with consideration for their distinct complex formation and localization (**Figure 3** and **Table 3**). (1) The Yip1p homolog is named a subunit (YIPFα) and the Yif1p homolog is named b subunit (YIPFβ), because a pair from each homolog form a complex. (2) Three distinct complexes are formed from special pairs of YIPFα and YIPFβ, and those complexes localize at distinct compartments. Therefore, they are numbered according to their primary localization. The early Golgi (ERGIC) residents are Complex 1 (YIPFα1 and YIPFβ1), the middle Golgi (cis-Golgi) residents are Complex 2 (YIPFα2 and YIPFβ2), and the late Golgi (medial-/trans-Golgi/TGN) residents are Complex 3 (YIPFα3 and YIPFβ3). (3) Two closer paralogs found in vertebrates are differentiated by adding A and B, e.g., YIPFβ3A and YIPFβ3B, because these proteins share a partner YIPF protein and show similar localization, suggesting that they are more similar in their function(s) than other family members (Soonthornsit et al., 2017).

# CONSERVATION OF THE PRIMARY STRUCTURE OF YIPF PROTEINS

When focusing on eukaryotic YIPF proteins, three conserved motifs are found in and around the transmembrane region (**Figure 2** and **Supplementary Figure S1**). They are at (1) the N-terminal cytoplasmic region near the first predicted transmembrane segment [K/R-φ/T-x-x-φ-φ-x-P] (alphabet: single character code of amino acid, f: hydrophobic

amino acid, x: any amino acids, alphabets connected with "/" indicated selection of amino acids at that position), (2) the N-terminal side of the first predicted transmembrane segment [D-L/F-x-G/I-P], and (3) the center of the third predicted transmembrane segment [φ-φ-G-Y-x-φ-φ-P/G/φ-φ-φ-P/φ] (Shakoori et al., 2003). The importance of these conserved motifs is supported by mutation analyses from S. cerevisiae. (1) The proline in the second motif and a partially conserved glycine residue found downstream were simultaneously mutated to leucine (P114L, G129E) in the temperature sensitive mutant yip1-1 (Yang et al., 1998). (2) The glycine at the third position of the third motif was mutated to glutamic acid (G175E) in the temperature sensitive mutant yip1-2 (Yang et al., 1998). (3) The proline at the eighth position of the third motif was mutated to glycine (P180G) in the lethal mutant yip1-19 (Chen et al., 2004). These results strongly suggest that the conserved residues in these motifs are functionally relevant. Importantly, all the three conserved motifs consist of proline residue(s). Proline is found in transmembrane segments of many transport proteins and receptors (Brandl and Deber, 1986; Williams and Deber, 1991; Wess et al., 1993). In striking contrast to its structural breaking nature in hydrophilic environment, in a hydrophobic environment mimicking membrane lipid, proline did not largely affect the helical structure of a model peptide (Li et al., 1996; Jacob et al., 1999), on the contrary, it stabilized the helical structure of the peptide in a higher temperature (Li et al., 1996). It was proposed that proline in transmembrane segments function in conformational change of transporters and/or in cationic ligand binding to the transmembrane segments (Brandl and Deber, 1986; Williams and Deber, 1991). Therefore, a primary interest for the future research is the evaluation of the significance of these prolines.

In addition to the motifs conserved in all YIPF members, there are several regions or motifs that are conserved only in a subset of YIPF members (indicated by purple blankets in **Supplementary Figure S2**). Among these is the [E-P-P-L-E-E] motif, which is conserved in the YIPFα1 (Yip1p) subfamily (indicated by a green blanket). The glutamic acid at the first position of this motif was mutated to lysine (E70K) in the temperature sensitive mutant yip1-4 (Calero et al., 2003). Upon temperature shift, secretion and growth were blocked concurrent with a massive proliferation of ER membrane, indicating the importance of this motif for the function of the YIPFα1 (Yip1p) subfamily (Heidtman et al., 2003). Similarly, mutation of the glutamic acid at the fifth position of the motif caused lethality in yip1-6 (E76K) (Chen et al., 2004). In support of this finding, mutation of the corresponding glutamic acid to lysine in human YIPFα1 (Yip1A) (E95K) showed a functional defect in suppressing the formation of multilamellar clustered ER membranes or ER whorls that were caused by the depletion of YIPFα1 (Yip1A) in HeLa cells, which will be discussed again later (Dykstra et al., 2013). Curiously, mutation of the glutamic acid at the first position of this motif to glycine (E89G) did not show a similar functional defect. It is possible that acidic to basic amino acid mutation may be necessary for inducing the observed phenotype, and this possibility and the significance of other motifs should be addressed in future research.

# COMPLEX FORMATION OF YIPF PROTEINS

As stated above, Yip1p was reported to form a complex with Yif1p for proper function (Matern et al., 2000; Barrowman et al., 2003). Similarly, Yip4p and Yip5p form a complex (Calero et al., 2002). In addition, analysis by us and others revealed that YIPFα1 (Yip1p) sub-φamily proteins form a complex with partner YIPFβ1 (Yif1p) sub-φamily proteins in human cells, namely, YIPF5 (Yip1A) with YIF1A (Yif1) (Jin et al., 2005; Yoshida et al., 2008), YIPF4 with YIPF3 (Tanimoto et al., 2011), and YIPF6 with YIPF1 or YIPF2 (Soonthornsit et al., 2017). Therefore, it is predicted that, as a general rule, YIPFα1 and YIPFβ1 form a paired complex in order to function (**Figure 3** and **Table 3**).

Analysis in human cells revealed that there are at least three distinct complexes of YIPF proteins; Complex 1 (YIPFα1A– YIPFβ1A) (Jin et al., 2005; Yoshida et al., 2008), Complex 2 (YIPFα2–YIPFβ2) (Tanimoto et al., 2011), and Complex 3 (YIPFα3–YIPFβ3A and YIPFα3–YIPFβ3B) (Soonthornsit et al., 2017). The localization and dynamics of these three complexes are significantly different (**Figure 3**). Complex 1 localized mainly in the early compartment of the Golgi apparatus (ERGIC) and recycled between the ER and the cis-Golgi (Yoshida et al., 2008). Complex 2 localized mainly in the middle compartment of the Golgi apparatus (cis-Golgi) and appeared to recycle between the cis- and trans-Golgi (Tanimoto et al., 2011). Complex 3 localized in the late compartment of the Golgi apparatus (medial-/trans-Golgi and TGN) and may travel to endosomes (Soonthornsit et al., 2017). These results suggest that YIPFα1–YIPFβ1, YIPFα2– YIPFβ2, and YIPFα3–YIPFβ3 form distinct complexes to carry out their functions. Consistent with this hypothesis, neither YIPFα2 (YIPF4) nor YIPFα1A (YIPF5) co-immunoprecipitated with YIPFα3 (YIPF6) (Soonthornsit et al., 2017).

Interestingly, si-RNA knockdown of a YIPFα protein specifically reduced the presence of their special partner YIPFβ protein(s) in the same complex, strongly suggesting an exclusive relationship between these family members. Namely, the knockdown of YIPFα1A (YIPF5) reduced YIPFβ1A (YIF1A) (Yoshida et al., 2008), the knockdown of YIPFα2 (YIPF4) reduced YIPFβ2 (YIPF3) (Tanimoto et al., 2011), and the knockdown of YIPFα3 (YIPF6) reduced YIPFβ3A (YIPF1) and YIPFβ3B (YIPF2) (Soonthornsit et al., 2017). Our results showed that knockdown of YIPFα1, YIPFα2, and YIPFα3 only affected YIPFβ1, YIPFβ2, and YIPFβ3, respectively, but not the other family members (Soonthornsit et al., 2017). These results suggest that the expression of a YIPFα protein is specifically and exclusively regulated by the expression of a partner YIPFβ protein. Taken together, this strongly suggests that a YIPFα protein has a specific partner YIPFβ, and these two molecules form a complex that serves as the basic unit for their function. This is supported by the finding that two pairs of YIPF proteins (or more precisely, transcripts or genes coding those proteins) (YIPFα1 and YIPFβ1 or YIPFα3 and YIPFβ3) are found in virtually all eukaryotes (**Table 2** and **Supplementary Table S2**). Exceptionally, two close homologs were shown to share a partner protein, i.e., YIPFβ3A (YIPF1) and YIPFβ3B (YIPF2) with YIPFα3 (YIPF6) (Soonthornsit et al., 2017).

This result suggests that YIPFα1A and YIPFα1B or YIPFβ1A and YIPFβ1B may also share partner proteins, but this possibility must be confirmed in future research. In addition, whether S. cerevisiae Yip1p and Yif1p and Yip4p and Yip5p are localized to the early and late Golgi compartments, respectively, must also be confirmed.

Yip1 domain family proteins were observed in higher order oligomers in mild detergent extract of HeLa cells. YIPFα1A– YIPFβ1A were estimated to form ∼4–8 mer complexes (Yoshida et al., 2008) while YIPFα2–YIPFβ2 form ∼4–16 mer complexes (Tanimoto et al., 2011). These results suggest that a tetramer consisting of YIPFα and YIPFβ, which is most probably formed from two of each molecule, is the minimum functional unit of the YIPF complex. It is possible that a tetramer subcomplex mainly associates homogeneously to form higher order oligomers, although oligomers consisting of mixed subcomplexes may also exist.

#### FUNCTION OF YIPF PROTEINS IN ANIMAL CELLS

Soon after the identification of S. cerevisiae Yip1p, human YIPFα1A (Yip1A) was cloned (Tang et al., 2001; Kano et al., 2004). The protein was shown to localize at the ER exit site and its cytoplasmic domain to interact with COPII components Sec23 and Sec24. YIPFα1A was efficiently incorporated into COPII vesicles and the overexpression of the cytoplasmic domain of YIPFα1A induced the disruption of the Golgi apparatus and inhibited the transport of VSV-G, a transmembrane marker protein, to the Golgi apparatus (Tang et al., 2001). These results suggest that YIPFα1A functions at ER exit sites and is involved in the recruitment of selected soluble and membrane cargo proteins to COPII vesicles. Under mitotic conditions, YIPFα1A was delocalized from the ER exit sites and spread diffusely throughout the ER in parallel with the dissociation of COPII components from the ER exit sites (Kano et al., 2004). This result was consistent with the close relationship of YIPFα1A to COPII components.

Later, we found that YIPFα1A (YIPF5) formed a complex with YIPFβ1A (YIF1A) and mainly localized at the early Golgi compartment (ERGIC and some in cis-Golgi) (Yoshida et al., 2008). We also found that YIPFα1A and YIPFβ1A were recycled between the ER and the Golgi apparatus. Our observation is consistent with former reports, because a part of YIPFα1A had to exist at the ER exit site during its trafficking between the ER and the Golgi apparatus. Knockdown of YIPFα1A or YIPFβ1A induced significant fragmentation of the Golgi apparatus with an accumulation of vesicles around the shortened stacked cisternae without significantly affecting anterograde transport (Yoshida et al., 2008; Kano et al., 2009). These results argue against the role of YIPFα1A in COPII-dependent ER to Golgi transport. In contrast, Lee's group reported that the depletion of YIPFα1A by si-RNA reduced ER to Golgi transport and induced an abnormal multi-lamellar ER structure, supporting the original idea that YIPFα1A functions during COPII vesicle budding from the ER (Dykstra et al., 2010).

The key to understanding this discrepancy may be the expression levels of YIPFα1B/Yip1B and/or YIPFβ1B/Yif1B. It is probable that YIPFα1B and/or YIPFβ1B compensated for the function of YIPFα1A and/or YIPFβ1A because YIPFα1B and YIPFβ1B are the closest homologs of YIPFα1A and YIPFβ1A, respectively. Because the expression levels of YIPFα1B and YIPFβ1B were not determined in any of the above discussed analyses, it is possible that higher expression of YIPFα1B or YIPFβ1B compensated for the loss of YIPFα1A or YIPFβ1A in our hands. However, this possibility does not necessarily mean YIPFα1A and YIPFα1B, YIPFβ1A and YIPFβ1B have completely overlapping roles. The functional difference between YIPFα1A and YIPFα1B is of particular interest because YIPFα1B/Yip1B has been reported to be expressed in muscle cells with a parallel loss of YIPFα1A/Yip1A (Barone et al., 2015). Similarly, YIPFβ1B may have a special role in cargo transport in neuronal cells (Carrel et al., 2008; Alterio et al., 2015).

Aside from these problems, there was an interesting report showing that the knockdown of YIPFα1A/Yip1A induced the dissociation of Rab6 from the Golgi membrane and reduced COPI independent Golgi to ER retrograde transport (Kano et al., 2009). However, there are difficulties in interpreting the results; firstly, because there are three isoforms of Rab6, Rab6A, Rab6A<sup>0</sup> , and Rab6B, it is unknown which of them were affected in the study. Secondly, the localization of Rab6, which was likely at the trans-side of the Golgi apparatus (Antony et al., 1992), was clearly different from YIPFα1A, which was at the ERGIC (Yoshida et al., 2008). Therefore, the mechanism by which knockdown of YIPFα1A affected the localization of the isoform(s) of Rab6 must be clarified to evaluate the significance of this study. Nevertheless, the finding gives us a clue toward clarifying the molecular mechanism and function of the Rab6 dependent, COPI independent retrograde transport pathway which remains poorly understood (Liu and Storrie, 2012).

Brucella is a pathogen that invades and replicates inside cells (Taguchi et al., 2015). It has been shown that Brucella induced the formation of, and resided in, membrane-bound structures called Brucella-containing vacuole (BCV) in the cytoplasm, which eventually fused with ER exit sites to form the replication machinery. It was shown that the IRE1-dependent unfolded protein response was necessary for the induction of BCV. Intriguingly, YIPFα1A/Yip1A was shown to be required for the activation of IRE1 following BCV formation and Brucella replication. In addition, knockdown of YIPFα1A inhibited the oligomer formation and activation of IRE1 induced by tunicamycin, which is a general stress inducer. Therefore, it is hypothesized that YIPFα1A is involved in stress-induced IRE1 activation in the ER, which is also induced by Brucella infection. A subsequent study revealed that YIPFα1A was also involved in the activation of stress adaptations and survival in cancer cells (Taguchi et al., 2017). Knockdown of YIPFα1A reduced the activation of IRE1 and also PERK, which are upstream regulators of the ER stress response and promote cell survival. It is possible that YIPFα1A functions as a chaperone for transmembrane proteins and promotes oligomerization and activation of IRE1 and PERK. This possibility must be evaluated in future studies.

Interestingly, YIPFα2/YIPF4 was shown to interact with E5 proteins from several different types of human papillomaviruses, including HPV-16 and HPV-18 (Müller et al., 2015). Papillomavirus E5 protein is a small multi-span transmembrane protein that has oncogenic activities in epidermal cells (Venuti et al., 2011). Among human papillomaviruses, types 16 (HPV-16) and 18 (HPV-18) have attracted interest because of their high risk for inducing cancer after infection. E5 of HPV-16 was shown to localize at the ER and Golgi apparatus under lower level expression, and also at the plasma membrane under high level expression. YIPFα2 and E5 of HPV-16 were shown to interact at the transmembrane region (Müller et al., 2015). However, these results must be interpreted with caution, because two opposing membrane topologies have been proposed for E5: one predicts the N-terminus to be exposed to the cytosol with the C-terminus exposed to the lumen of the Golgi apparatus (Hu and Ceresa, 2009), while the other predicts the N-terminus to be exposed to the lumen of the Golgi apparatus with the C-terminus exposed to the cytosol (Krawczyk et al., 2010). The tagging of proteins in the above study may not reproduce the native topology of the E5 protein. Intriguingly, YIPFα2 was found to decrease in calcium differentiated human foreskin keratinocytes, while the presence of viral genome in the cells rescued the expression of YIPFα2 and the presence of E5 was not necessary for this effect. The significance of the change in YIPFα2 expression during keratinocyte differentiation and its relationship with HPV infection should be a subject for future investigation.

A mutation in YIPFα3/YIPF6 that caused truncation of the coding sequence, deleting the entire transmembrane region, was shown to render mice susceptible to colitis when the mice were fed a non-toxic dose of dextran sodium sulfate, a model for inflammatory bowel disease (Brandl et al., 2012). Pathological and biochemical analyses revealed that the number of Paneth cells and goblet cells was reduced in the mutant mice. In addition, the secretory granules were smaller and more disorganized in the Paneth cells. Similarly, mucin granules were smaller and more irregular in size in the goblet cells. In accordance, the mucin content of the colon was reduced. These results suggest that YIPFα3 is involved in the synthesis and/or exocytotic transport of mucin in goblet cells in the mutant mice. It is possible that the loss of YIPFα3 damaged either export of mucin from the trans-Golgi or maturation of mucin granules at TGN in the goblet cells. A similar defect in the secretory pathway is hypothesized to occur in Paneth cells. Importantly, and rather surprisingly, the mutant mice were viable and grew more or less normally under normal feeding conditions up to 7 days after the birth, suggesting the loss of YIPFα3 did not cause severe developmental or functional defects (Brandl et al., 2012). This result suggests that the function of YIPFα3 is cell type specific, or the effect of the loss of YIPFα3 was somehow compensated for in most other cell types. In other words, it is likely that differentiated goblet cells and Paneth cells, but not other cells, are highly dependent on the function of YIPFα3. Consistently, the depletion of YIPFα3 by si-RNA did not induce a significant effect either on the ER and Golgi structures or on ER to plasma membrane transport in HeLa cells or HT-29 cells that secrete mucin (Soonthornsit et al., 2017). We hope for a more detailed analysis of the YIPFα3 mutant mice to be carried out in order to obtain further clues about the function of YIPFα3.

It was recently reported that the expression of YIPFα3 was increased in prostate cancer cells that showed bone metastasis and became resistant to castration (Djusberg et al., 2017). In these cells, the androgen receptor gene was amplified together with the YIPFα3 gene, which is located near the androgen receptor gene on the X chromosome (Xq12). The increase in these expression levels was thought to contribute to the malignant phenotype of the cancer cells (Vainio et al., 2012; Djusberg et al., 2017). Curiously, over-expression of YIPFα3 in 22Rv1 cells, which expresses high androgen receptor activity, reduced, instead of increased, cell proliferation. Therefore, whether and how the increase of YIPFα3 affects the malignancy of cancer cells remain unclear. Intriguingly, an increase in the number of ∼83 nm diameter extracellular vesicles, which may function as exosomes, was also induced by the over expression of YIPFα3. Whether this effect is also found in other cell types and how this increase in presumed exosomes is induced are of special interest for elucidating the function of YIPFα3.

YIPFβ1A/YIF1A was shown to interact with VAPB, a mutant of which (VAPB–P56S) has been linked to motor neuron degeneration in amyotrophic lateral sclerosis type 8 (ALS8) (Kuijpers et al., 2013). VAPB is a type II ER membrane protein, which interacts with lipid exchange and lipid-sensing proteins that have FFAT motifs. VAPB is thus thought to be involved in the organization of lipid metabolism and nonvesicular lipid transfer to and from the ER. The chronic expression of VAPB–P56S induced small inclusions scattered around the cytoplasm where the mutant protein accumulated. These inclusions were shown to be formed from and connected to the ER (Fasana et al., 2010). YIPFβ1A was shown to bind to both the wild type and the mutant form of VAPB (VAPB–P56S). Interestingly, YIPFβ1A and YIPFβ1B/YIF1B accumulated in the inclusions induced by VAPB–P56S. Furthermore, overexpression of VAPB induced a dispersal of YIPFβ1A throughout the neuron while the knockdown of VAPB induced an accumulation of YIPFβ1A around the Golgi area. These results suggest that the interactions of VAPB with YIPFβ1A and possibly also with YIPFβ1B have a significant role in the pathology of VAPB–P56S (Kuijpers et al., 2013).

YIPFβ1B/YIF1B was found to interact with 5-HT1AR, one of the serotonin receptors localized at the plasma membrane of soma and dendrites of neurons in the central nervous system (Carrel et al., 2008). 5-HT1AR is a G protein coupled receptor and a major target of anti-depressant drugs, and how it is delivered to a specialized region of the neuron has attracted a medical interest (Barnes and Sharp, 1999). 5-HT1AR showed yeast two-hybrid interaction with YIPFβ1B and the depletion of YIPFβ1B in primary neurons specifically prevented the delivery of 5-HT1AR to distal portions of dendrites (Carrel et al., 2008). YIPFβ1B was hypothesized to support the delivery of 5-HT1AR specifically because the loss of YIPFβ1B did not affect the delivery of other receptors, such as sst2AR, P2X2R, and 5-HT3AR. YIFPβ1B was shown to localize mainly in the ERGIC, similar to YIPFβ1A. Yoshida et al. (2008) and Alterio et al. (2015) raised the question of how an ERGIC protein determines the delivery

of cargo proteins to specific regions of the plasma membrane. One possibility is that the loss of YIFPb1B perturbed the proper processing of 5-HT1AR at the Golgi apparatus, including glycosylation, leading to mis-sorting of the 5-HT1AR at the trans-Golgi/TGN. It is possible that YIPFβ1B is directly involved in the processing of 5-HT1AR by delivering 5-HT1AR to processing enzymes. Alternatively, YIPFβ1B may be indirectly involved in the processing of 5-HT1AR, supporting the proper localization of processing enzymes in the Golgi apparatus. It is also possible that YIFPβ1B functions as a molecular chaperone helping 5-HT1AR to form a proper conformation that is necessary for its correct delivery, or helping 5-HT1AR assemble with other factors that support the proper delivery of target molecules.

It was recently reported that YIPFβ1/Yif1 and YIPFα1/Yip1 have an essential role in dendrite pruning in Drosophila (Wang et al., 2018). As in mammalian cells, Drosophila YIPFα1 and YIPFβ1 form a complex and localize mainly in the ERGIC and the Golgi apparatus. Interestingly, the Golgi apparatus was fragmented in ddaC sensory neurons with truncated mutants of YIPFα1 (Yip1) or YIPFβ1 (Yif1), while these proteins were dispensable for viability or apoptosis. How YIPFα1 and YIPFβ1 induce dendrite pruning in Drosophila remains unclear.

### MECHANISMS OF YIPF PROTEIN FUNCTION AND FUTURE DIRECTIONS FOR RESEARCH

After two decades of studies following discovery of the first YIPF proteins (Yip1p) in S. cerevisiae, the molecular mechanisms of YIPF protein function still remain obscure. It has been difficult to analyze the family in mammalian cells, most likely because of the overlapping functions of YIPF family members. S. cerevisiae only has four family members and is thought to be better suited for the analysis for YIPF proteins, although a decade without any reports on the function of yeast YIPF proteins implies that things have not been so easy, even in S. cerevisiae. In this aspect, the analysis of YIPF proteins in any other available model organisms will face difficulty resulting from the expected overlapping functions, because virtually all eukaryotes have at least four family members. Therefore, even D. melanogaster may not be an ideal model organism because it also has four family members.

Four family members and two distinct complexes were identified in S. cerevisiae; Complex 1; YIPFα1 (Yip1p) and YIPFβ1 (Yif1p) and Complex 3; YIPFα3 (Yip4p) and YIPFβ3 (Yip5p). From the analysis of mammalian family members, Complex 1 is predicted to function in ER to Golgi transport (YIPFα1, YIPFβ1) and Complex 3 in transport between the Golgi and downstream compartments such as endosomes (YIPFα3, YIPFβ3). This idea is supported by the finding that abnormal vacuolar morphology was observed with the null mutants of YIP4 and YIP5 (Michaillat and Mayer, 2013), and endocytosis decreased in the null mutant of YIP4 (Burston et al., 2009). A more detailed analysis of these null mutants will provide more clues toward understanding the function of these complexes.

An earlier study suggested that YIPF proteins were candidates for classical yeast two hybrid analysis when tagged on their N-termini (Yang et al., 1998; Ito et al., 2000; Matern et al., 2000; Uetz et al., 2000; Calero et al., 2001, 2002; Calero and Collins, 2002; Shakoori et al., 2003). However, several trials in our laboratory at library screening using yeast two hybrid analysis to identify binding partners of human YIPF proteins produced no likely candidates. In particular, neither Rab1 nor Rab11, which are orthologs of Ypt1p and Ypt31p that interact with yeast Yip1p, Yif1p, Yip4p, and Yip5p, was picked up either by classical yeast two hybrid analysis or by pull down analysis. Therefore, no direct evidence that YIPF proteins function with Rab/Ypt GTPases has been obtained in mammalian cells so far. It was proposed that Yip1p binds the GDP bound form of Ypt1p or Ypt31p because Yip1p did not show yeast two hybrid interaction with GTPase deficient mutants of Ypt1p or Ypt31p (Yang et al., 1998). To test this possibility in mammalian cells, we tried to detect the interaction of human YIPF proteins with a human or rat GDP binding mutant of Rab1 by yeast two-hybrid analysis, but this has again been unsuccessful to date. It is possible that design of bait and/or pray constructs was not adequate to allow the access of mammalian Rab proteins and YIPF proteins in our system. Therefore, redesigning of the bait and/or pray constructs or use of other analytical system is necessary to evaluate the interaction of mammalian Rab proteins and YIPF proteins. However, we cannot exclude the possibility that YIPF proteins evolved to dispense with the interaction to Rab proteins for their functions in mammalian cells.

The only significant interactions of human proteins found by classical yeast two hybrid analysis were between ArfGAP1 and YIPFβ1A (YIF1A/FinGER7) or YIPFβ1B (YIF1B/FinGER8) (Akhter et al., 2007). Interestingly, this interaction was mediated by the first ALPS motif of ArfGAP1, which was hypothesized to sense curvature of the lipid bilayer (Bigay et al., 2005). However, experiments in our laboratory using detergent extract of HeLa cells have not been able to confirm these interactions. It has been proposed that the ALPS motif binds highly curved small liposome membranes by inserting hydrophobic bulky residues into a loosely packed lipid layer. The ALPS motif was unstructured in its membrane unbound soluble state, but formed an amphipathic helix in the membrane bound state (Bigay et al., 2005). Therefore, it is possible that the conformation of ArfGAP1 could not be reproduced in the presence of detergent, disrupting the interaction of ArfGAP1 and YIPFβ1A or YIPFβ1B. Future work using reconstituted liposomes may overcome this limitation.

As we discussed above, YIPF proteins form higher order oligomers consisting of at least four YIPF protein molecules. Because one YIPF protein has five transmembrane segments, a tetramer is predicted to have 20 transmembrane segments. There are many hydrophilic amino acid residues in these transmembrane segments suggesting that YIPF proteins function as channels, transporters, or transmembrane receptors. If so, the existence of distant homologs in prokaryotic cells, in which no membrane trafficking pathway has been developed, may be informative. It is possible that these prokaryotic homologs function in the transport of some hydrophilic solute (s), such as ions and organic molecules. Alternatively, they may function as membrane receptors

responding to extracellular molecules. It is likely that YIPF proteins have a common function with their prokaryotic homologs in this respect, aside from their function in membrane trafficking pathways. It is even possible that their function as a transporter, a channel, or a receptor is their main function in eukaryotes and the observed interactions between YIPF proteins and Ypt/Rab GTPases and other molecules involved in membrane trafficking serve to control the proper localization of these proteins.

Indeed, transmembrane segments of YIPF proteins consist of conserved proline residues (**Figure 2**), which is found or even conserved in many transporters (Brandl and Deber, 1986; Wess et al., 1993). Furthermore, bioinformatics analysis using SCOOP showed that the Yip1 domain has significant similarity to known transporters (Bateman and Finn, 2007). The Yip1 domain is now grouped as clan Yip1 (CL0112), which contains the following seven family members in the Pfam database: DUF1048, DUF1129, DUF1189, DUF1282, DUF1700, YIF1, and Yip1 (El-Gebali et al., 2019). Among these, DUF1282 (pfam06930), which includes bacterial Yip1p domain families, has similarity with the sulfate permease family (PF00916). On the other hand, DUF1129 (pfam06570) has similarity to Sugar\_tr (PF00083), which includes transporters responsible for moving various carbohydrates, organic alcohols, and acids in a wide range of prokaryotic and eukaryotic organisms. DUF1700 has similarity to VIT1 (PF01988), which includes a group of putative vacuolar ion transporters. These results support the possibility that YIPF protein plays a role as a transporter.

The Golgi apparatus is a factory where glycosylation, sulfation, and phosphorylation of secretory and membrane proteins occur. Substrates and by-products have to be transported into and out of the lumen of the Golgi apparatus to sustain many reactions. Many transporters that support these function have already been identified (Berninsone and Hirschberg, 2000; Hirschberg, 2013). However, many remain undiscovered, including those responsible for the transport of inorganic phosphate, which is produced from nucleotide di-phosphate to reinforce the one-way reaction of sugar transfer to substrates. Physiologically identified transporters GOLAC-1 and GOLAC-2 are possible candidates to fulfill this function (Nordeen et al., 2000; Thompson et al., 2006). An anion channel that functions for the acidification of the Golgi lumen (GPHR) was proposed to be GOLAC-2 (Maeda et al., 2008). On the other hand, a molecule responsible for the function of GOLAC-1 has not yet been identified. Is it possible that YIPF proteins function as GOLAC-1?

One other possibility is that YIPF proteins function as membrane protein chaperones, assisting with the conformational maturation and/or complex formation of transmembrane proteins. The results that YIPFα1A was involved in the oligomerization and activation of IRE1 support this possibility (Taguchi et al., 2015, 2017). Finally, there is a possibility that YIPF proteins function to specify the identity of domains of the Golgi apparatus. Namely, the presence of YIPF Complex 1 (YIPFα1–YIPFβ1), Complex 2 (YIPFα2–YIPFβ2), and Complex 3 (YIPFα3–YIPFβ3) determine the identity of early (ERGIC), middle (cis-Golgi), and late Golgi (medial- /trans-Golgi/TGN) compartments, respectively (**Figure 3**). Golgi resident transmembrane proteins may be anchored to YIPF protein complexes through interactions at transmembrane regions. The dynamics of YIPF proteins support this possibility. After Brefeldin A treatment, most Golgi resident proteins are transported back to the ER, while TGN proteins are transported to the endosomal compartment (Lippincott-Schwartz et al., 1991). YIPF proteins, however, did not travel with these Golgi proteins but remained in distinct cytoplasmic vesicular structures after treatment with Brefeldin A (Yoshida et al., 2008; Tanimoto et al., 2011; Soonthornsit et al., 2017). Therefore, YIPF proteins have the capacity to resist vesicular flow, and this capacity is wellsuited for determining compartment identities. To evaluate this possibility, the mechanisms which determine the localization of the three YIPF subcomplexes must be better understood.

# AUTHOR CONTRIBUTIONS

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

# FUNDING

This work was supported by Grants-in-Aid for Scientific Research (C) (#17K07393) from the Japan Society for Promotion of Science, the Takeda Science Foundation, and the Kyoto Sangyo University for NN.

# ACKNOWLEDGMENTS

We would like to thank all the past and present members of my laboratory for providing the underlying data and scientific discussions, and Prof. Tim Levine (UCL Institute of Ophthalmology) for notifying us of the existence of bacterial Yip1 family proteins and critically reading the manuscript.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2019.00130/ full#supplementary-material

FIGURE S1 | Multiple sequence alignments produced by CLUSTAL W are shown. The five predicted transmembrane segments are indicated on the top (TM1–TM5). Orthologs of family members are grouped and indicated by colored background on the left. YIPF proteins are grouped in two subfamilies (a and b) that are further sub-grouped into three (1–3). In higher chordata, YIPFα1, YIPFβ1, and YIPFβ3 were further sub-divided into A and B (refer text and Table 3 for the definition). The conserved motifs are indicated by red (conserved in all YIPF members) or green (conserved in YIPFα1/Yip1p) lines on the bottom and blankets in the same color on the aligned sequences. Other conserved regions with unknown significance are indicated by purple blankets. The YIPF protein sequences were identified by BLAST search using human and S. cerevisiae YIPF protein sequences. A representative species for each phylum or class was selected for the analysis to simplify the results. Duplicated data sets and divergent isoform sequences were omitted.

FIGURE S2 | Multiple sequence alignments were produced by CLUSTAL W, as described in Supplementary Figure S1 and a phylogenetic tree with boot strap values was drawn by NJplot version 2.3. Orthologs of family members are grouped and indicated by colored background as in Supplementary Figure S1.

### REFERENCES

fcell-07-00130 July 26, 2019 Time: 12:10 # 12


TABLE S1 | Genetic interactors of budding yeast YIPF proteins.

TABLE S2 | Eukaryotic species in which all the ortholog of Yip1p, Yif1p, Yip4p, and Yip5p were found.

TABLE S3 | Prokaryote species in which YIPF homologs are found.


ER-to-Golgi vesicular transport during mitosis. Mol. Biol. Cell 15, 4289–4298. doi: 10.1091/mbc.e03-11-0822


network-localized Yip domain family proteins, which play a role in the Golgi reassembly and glycan synthesis. Exp. Cell Res. 353, 100–108. doi: 10.1016/j. yexcr.2017.03.011


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

Copyright © 2019 Shaik, Pandey, Thirumalasetti and Nakamura. 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.

# The Structure and Function of Acylglycerophosphate Acyltransferase 4/ Lysophosphatidic Acid Acyltransferase Delta (AGPAT4/LPAATδ)

#### Mikhail A. Zhukovsky\*, Angela Filograna, Alberto Luini, Daniela Corda and Carmen Valente\*

Institute of Biochemistry and Cell Biology and Institute of Protein Biochemistry, National Research Council, Naples, Italy

#### Edited by:

Vladimir Lupashin, University of Arkansas for Medical Sciences, United States

#### Reviewed by:

Mitsuo Tagaya, Tokyo University of Pharmacy and Life Sciences, Japan Robin Duncan, University of Waterloo, Canada

#### \*Correspondence:

Mikhail A. Zhukovsky m.zhukovsky@ibp.cnr.it Carmen Valente c.valente@ibp.cnr.it

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 07 May 2019 Accepted: 16 July 2019 Published: 02 August 2019

#### Citation:

Zhukovsky MA, Filograna A, Luini A, Corda D and Valente C (2019) The Structure and Function of Acylglycerophosphate Acyltransferase 4/ Lysophosphatidic Acid Acyltransferase Delta (AGPAT4/LPAATδ). Front. Cell Dev. Biol. 7:147. doi: 10.3389/fcell.2019.00147 Lipid-modifying enzymes serve crucial roles in cellular processes such as signal transduction (producing lipid-derived second messengers), intracellular membrane transport (facilitating membrane remodeling needed for membrane fusion/fission), and protein clustering (organizing lipid domains as anchoring platforms). The lipid products crucial in these processes can derive from different metabolic pathways, thus it is essential to know the localization, substrate specificity, deriving products (and their function) of all lipid-modifying enzymes. Here we discuss an emerging family of these enzymes, the lysophosphatidic acid acyltransferases (LPAATs), also known as acylglycerophosphate acyltransferases (AGPATs), that produce phosphatidic acid (PA) having as substrates lysophosphatidic acid (LPA) and acyl-CoA. Eleven LPAAT/AGPAT enzymes have been identified in mice and humans based on sequence homologies, and their localization, specific substrates and functions explored. We focus on one member of the family, LPAATδ, a protein expressed mainly in brain and in muscle (though to a lesser extent in other tissues); while at the cellular level it is localized at the trans-Golgi network membranes and at the mitochondrial outer membranes. LPAATδ is a physiologically essential enzyme since mice knocked-out for Lpaatδ show severe dysfunctions including cognitive impairment, impaired force contractility and altered white adipose tissue. The LPAATδ physiological roles are related to the formation of its product PA. PA is a multifunctional lipid involved in cell signaling as well as in membrane remodeling. In particular, the LPAATδ-catalyzed conversion of LPA (invertedcone-shaped lipid) to PA (cone-shaped lipid) is considered a mechanism of deformation of the bilayer that favors membrane fission. Indeed, LPAATδ is an essential component of the fission-inducing machinery driven by the protein BARS. In this process, a proteintripartite complex (BARS/14-3-3γ/phosphoinositide kinase PI4KIIIβ) is recruited at the trans-Golgi network, at the sites where membrane fission is to occur; there, LPAATδ directly interacts with BARS and is activated by BARS. The resulting formation of PA is essential for membrane fission occurring at those spots. Also in mitochondria PA

formation has been related to fusion/fission events. Since PA is formed by various enzymatic pathways in different cell compartments, the BARS-LPAATδ interaction indicates the relevance of lipid-modifying enzymes acting exactly where their products are needed (i.e., PA at the Golgi membranes).

Keywords: acyltransferase, AGPAT, LPAAT, phosphatidic acid, lysophosphatidic acid, Golgi complex, membrane fission, BARS

#### INTRODUCTION

Lysophosphatidic acid acyltransferases are an emerging family of enzymes that catalyze the production of PA using LPA and acyl-CoA (Korbes et al., 2016). The product of this enzymatic reaction, PA is involved in several essential cellular functions based on its unique properties as: (i) precursor for the biosynthesis of all glycerophospholipids and triacylglycerol (TAG); (ii) important membrane remodeling metabolite involved in the intracellular transport; and (iii) precursor of bioactive lipid mediators implicated in cell survival, proliferation, and tumor progression (Kume and Shimizu, 1997; Lu et al., 2005; Eto et al., 2014). The LPA is first synthetized, in the de novo-Kennedy pathway, via acylation of glycerol 3-phosphate (G3P) by the glycerol-3-phosphate acyltransferase (GPAT) enzymes that use acyl-CoAs as donors. Then, another fatty acid moiety (often an unsaturated one in eukaryotes) is incorporated at the sn-2 position on the LPA glycerol backbone to form PA by the 1 acylglycerol-3-phosphate acyltransferase enzyme (AGPAT, also known as LPA acyltransferase: LPAAT).

Eleven AGPAT enzymes, named AGPAT 1-11, have been identified in mice and humans based on the homology of their primary sequences (Kume and Shimizu, 1997). All AGPATs display highly conserved structural motif (Coleman and Mashek, 2011) and exhibit acyltransferase activity, using acyl-CoA and lysophospholipid as acyl-donor and acyl-acceptor, respectively (Kume and Shimizu, 1997). AGPAT enzymes have been named based on their substrate specificities and order in which their cloning have been reported. AGPATs 1, 2, 3, 4, and 5 specifically prefer LPA to form PA and, as such, are also known as LPAATα, β, γ, δ and ε, respectively, while AGPATs 6-11 are classified as lysophospholipid acyltransferases (LPLATs) or GPATs based on their substrate specificities (Yamashita et al., 2014).

Of note, recent studies indicate that each LPAAT enzyme is responsible for the production of a distinct and specific pool of PA required to affect downstream lipid metabolism that mediates specific cellular and organelle membrane lipid composition, which, in turn, controls physiological functions (Bradley and Duncan, 2018). Loss of function of a single LPAAT influences different downstream lipid biosynthetic pathways generating distinct pathophysiological consequences, including embryonic lethality, lipodystrophy, impaired spatial learning and memory (Bradley and Duncan, 2018). Moreover, LPAAT enzymes show differential subcellular localization together with a differential expression profiles within tissues or organs as well as preference toward specific fatty acyl-CoA donor moieties (Shindou et al., 2009). These aspects may explain the existence of different LPAAT isoforms in nature able to specifically modulate downstream glycerolipid pathways in different tissues and organs for maintenance of normal physiology. As a consequence, loss of specific LPAAT cannot be functionally and biochemically replaced by other LPAATs (Bradley and Duncan, 2018). Tissue distribution patterns reveal an ubiquitous expression of LPAATα and LPAATγ, while LPAATβ, δ, and ε display distinct tissuespecific profiles (Takeuchi and Reue, 2009).

Following a brief overview on phylogenetic tree of LPAAT genes and on the unique properties of PA, the product of the reaction catalyzed by these enzymes, we analyze the conserved acyltransferase motifs required for enzymatic activity with the spotlight on LPAATδ member. We first discuss the membrane topology of human LPAATδ based on the recently resolved structure of bacterial LPAAT. We then highlight the role of the specific LPAATδ-produced PA pool in lipid metabolism and tissue functions. Finally, we report on recent advances in our understanding of the membrane fission event involving LPAATδ and occurring at the trans-side of the Golgi complex. Specifically, we discuss how this enzyme assembles with other proteins in a protein complex and how this machinery is regulated and operates in the formation of the basolaterally directed post-Golgi carriers.

# LPAAT FAMILY

The LPAAT enzymes are an ancient gene family that were first functionally described in 1956 (Kennedy and Weiss, 1956). It belongs to the subgroup 3 of the membrane bound O-acyltransferase (MBOAT) superfamily (Chang et al., 2011;

**Abbreviations:** AGPAT, acylglycerophosphate acyltransferase; AMPA, α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid; ARF, ADP ribosylation factor; BARS, brefeldin A ADP-ribosylated substrate; C14:0, myristic acid; C16:0, palmitic acid; C18:0, stearic acid; C18:1, oleic acid; C20:4, arachidonic acid; CGL, congenital generalized lipodystrophy; CL, cardiolipin; CoA, coenzyme A; DAG, diacylglycerol; DAGKs, diacylglycerol kinases; DHA, docosahexaenoyl acid; DHA-CoA, docosahexaenoyl-CoA; DHAP, dihydroxyacetone phosphate; DLCL, dilysocardiolipin; ER, endoplasmic reticulum; FAPP, four-phosphate-adaptor protein; G3P, glycerol 3-phosphate; GPAT, glycerophosphate acyltransferase; LPA, lysophosphatidic acid; LPAAT, lysophosphatidic acid acyltransferase; LPC, lysophosphatidylcholine; LPE, lysophosphatidylethanolamine; LPG, lysophosphatidylglycerol; LPI, lysophosphatidylinositol; LPS, lysophosphatidylserine; MIM, mitochondrial inner membrane; MLCL, monolysocardiolipin; MMPE, monomethyl-phosphatidylethanolamine; MOM, mitochondrial outer membrane; NCS-1, neuronal calcium sensor-1; NMDA, N-methyl D-aspartate; PA, phosphatidic acid; PAK1, p21-activated kinase; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol; PI4KIIIβ, phosphatidylinositol 4-kinase IIIβ; PKD, protein kinase D; PLA2s, phospholipases A2; PLD, phospholipase D; TAG, triacylglycerol; TGN, trans-Golgi network; TMD, transmembrane domain; TmPlsC, Thermotoga maritima PlsC; WAT, white adipose tissue.

Korbes et al., 2016). LPAAT genes were found in all three domains of life: Archaea, Bacteria and Eukarya where the LPAAT enzymes play key metabolic roles. At least one LPAAT gene was found in each of the eukaryotic genome examined (Korbes et al., 2016). Some eukaryotic species contain a few LPAAT genes (e.g., thirteen in the soya bean Glycine max), whereas yeast Saccharomyces cerevisiae possesses only one LPAAT gene, SLC1 (Korbes et al., 2016). Five LPAAT enzymes namely LPAATα-LPAATε are present in mammals (Korbes et al., 2016; Bradley and Duncan, 2018).

The phylogenetic tree analyses of LPAAT genes from prokaryotic to eukaryotic species based on phosphate acyltransferase (PlsC) domain identified three distinct clusters designed as cluster I, II, and III (Korbes et al., 2016). Cluster I is the most ancient and contains plant LPAATα and LPAATβ, prokaryotic and fungal LPAATs, as well as animal LPAATα and LPAATβ. Cluster II consists of animal AGPAT6, AGPAT10, AGPAT7, AGPAT9, and AGPAT11. Cluster III is divided into subclusters: IIIa contains animal LPAATγ, LPAATδ and plant LPAATβ, LPAATγ; IIIb is composed of plant LPAATδ, LPAATε, fungal LPAAT and animal AGPAT8; and IIIc includes animal LPAATε. Of note, two different proteins LPAATγ and LPAATδ in animals appeared due to a duplication event (Korbes et al., 2016). The sequence of human LPAATδ is more similar to the sequence of human LPAATγ (ca. 60%) than to that of other three human LPAATs, see **Supplementary Tables S1, S3**. A few residues are typical of the members of subcluster IIIa: they are highly conserved in LPAATγ and LPAATδ orthologs, but are rare among other animal LPAATs. A good example of such residues is tryptophan (W, at position 106 in human LPAATδ), five residues downstream of aspartic acid (D) belonging to the conserved NHxxxxD sequence, see **Supplementary Tables S1–S4**.

Lysophosphatidic acid acyltransferase γ, as LPAATα, is ubiquitously expressed with high mRNA levels in adipose tissue, liver and heart (Yuki et al., 2009). LPAATγ, in addition of being localized, like LPAATα and LPAATβ, at the ER (Agarwal et al., 2011; Yamashita et al., 2014), is a Golgiresident enzyme that controls Golgi structure and retrograde transport from the Golgi complex to the ER. Depletion of LPAATγ causes Golgi membrane fragmentation and severe impairment of COPI-coated vesicle formation (Schmidt and Brown, 2009; Yang et al., 2011). Indeed, inhibition of LPAATγ enzyme impairs the release of COPI buds as vesicles from the Golgi complex. This is due to the role of LPAATγ activity in controlling the membrane fission of these buds from the Golgi membranes (Yang et al., 2011). Of note, LPAATγ prefers C20:4 and C22:6 fatty acyl-CoA donors (Yuki et al., 2009; Koeberle et al., 2010, 2012) and this might explain the original evidence of acyl-CoA's involvement in COPI-coated vesicle formation (Pfanner et al., 1989).

Here we selectively focus on LPAATδ because of its emerging role in membrane remodeling (fission and fusion) and related diseases caused by defective lipid metabolism and/or altered membrane composition, while for an extensive and accurate description of the other LPAAT enzymes the reader is referred to excellent reviews (Yamashita et al., 2014; Bradley and Duncan, 2018).

# PHOSPHATIDIC ACID: THE PRODUCT OF THE LPAAT-CATALYZED ENZYMATIC REACTION

Phosphatidic acid is composed of a three-carbon glycerol backbone, to which two fatty acyl chains are ester-linked at positions C-1 and C-2, and a phosphate is ester-linked at position C-3. On the basis of their shapes, lipids can be divided into three classes (Cullis and De Kruijff, 1979; Corda et al., 2002; Chernomordik and Kozlov, 2003; and references therein): coneshaped lipids, inverted-cone-shaped lipids, and cylindrical lipids. Lipids whose tails are wider than headgroups are cone-shaped, lipids whose headgroups are wider than tails are inverted-coneshaped, and lipids whose headgroups are approximately as wide as tails are cylindrical. PA is a cone-shaped lipid, whereas LPA possesses an inverted-cone shape (Kooijman et al., 2005b; **Figure 1**). Thus, the interconversion from LPA to PA mediated by LPAAT enzymes destabilizes the organization of the lipid bilayer causing a distortion of this bilayer, a process that supports membrane fission (Barr and Shorter, 2000; Shemesh et al., 2003; Pagliuso et al., 2016). Moreover, both PA and LPA are charged negatively (Kooijman et al., 2005a). Hence, PA possesses an unique combination of cone shape and negative charge (Tanguy et al., 2019). Such unique property of PA allows it to recruit various proteins to the membrane (Stace and Ktistakis, 2006). According to the electrostatic/hydrogen bond switch mechanism (Kooijman et al., 2007), the PA charge is able to change from −1 to −2 due to the deprotonation of PA headgroup, caused by formation of the hydrogen bond of positively charged amino acid residue (lysine or arginine) with this headgroup. This change can stabilize the protein-lipid interaction (Kooijman et al., 2007). Due to its phosphomonoester headgroup whose pK<sup>a</sup> value is within physiological range, PA acts as a pH biosensor (Young et al., 2010; Shin and Loewen, 2011). Eventhough the content of PA in the cellular membrane is usually low, ∼1–4% (Zinser et al., 1991; Young et al., 2010; Zegarlinska et al., 2018), this lipid is involved in many important biological processes such as induction of membrane curvature, membrane trafficking, recruitment of proteins to membranes, regulation of catalytic activity of various enzymes, signal transduction, cytoskeletal organization and regulation of gene expression (Kooijman and Burger, 2009; Raben and Barber, 2017; Kameoka et al., 2018; Zegarlinska et al., 2018; Tanguy et al., 2019; and references therein). Various phospholipids, such as PC, PE, PI and CL are synthesized through the common precursor, PA (Yamashita et al., 2014). In the Lands' cycle, these phospholipids can be deacylated and reacylated at the sn-2 position in a coordinated and concerted control cycle by phospholipases A<sup>2</sup> (PLA2s) and LPLATs actions, respectively (Bankaitis, 2009; Ha et al., 2012). This cycle is essential to generate the membrane asymmetry and diversity that support membrane fluidity and curvature required for fundamental biological functions. Thus, PA is involved, directly or indirectly, in the biosynthesis of most phospholipids

(Kassas et al., 2017). Hence, also enzymes that catalyze the synthesis of PA play very important role in biological processes.

In all organisms, PA can be produced by one of three major routes (Foster et al., 2014; Vodicka et al., 2015; Kassas et al., 2017): de novo synthesis in which the final step is acylation of LPA by LPAAT enzymes, phosphorylation of DAG by DAGKs, and hydrolysis of phospholipids by PLD (**Figure 2**). LPAATs are members of the family of AGPAT enzymes that specifically use LPA as acyl acceptor (Yamashita et al., 2014).

#### ACYLTRANSFERASE MOTIFS IN LPAATs

The LPAAT enzymes contain four catalytic motifs I-IV (Takeuchi and Reue, 2009; Rottig and Steinbuchel, 2013; Yamashita et al., 2014), and here we report the sequence alignment of LPAATs from sixty species of different biological kingdoms, see **Supplementary Tables S1**, **S3**.

Motifs I and III are the most conserved, whereas motif IV is the least conserved. Motifs I and III are suggested to be involved in the interaction with the acyl acceptor, whereas motif IV is suggested to interact with acyl donor (Yamashita et al., 2014; Robertson et al., 2017; and references therein).

Catalytic motif I contains a conserved NHxxxxD sequence. In this sequence, the residue following histidine (H) is usually hydrophilic, whereas the 5th residue upstream this H and the residue preceding aspartic acid (D) are almost always hydrophobic, see **Supplementary Tables S1**, **S3**. The H belonging to this sequence acts as a general base to abstract proton from the hydroxyl group which has to be acylated (Pagac et al., 2011; Yamashita et al., 2014; and references therein). The role of D belonging to the NHxxxxD sequence is to maintain the lone pair of electrons on the Nε2 nitrogen of catalytic H so as to abstract a proton (Robertson et al., 2017; and references therein). Among sixty LPAATs from different species presented in the

FIGURE 2 | Biosynthetic pathways for phosphatidic acid production. Phosphatidic acid (PA) can be generated by three major routes: (a-b) The de novo synthesis via lysophosphatidic acid (LPA) formation that occurs via two different acylation pathways. (a) The first and main synthesis route is the glycerol 3-phosphate (G3P) pathway. G3P is acylated by a glycerol 3-phosphate acyltransferase (GPAT) to form LPA. (b) The second pathway of LPA formation involves acylation of dihydroxyacetone phosphate (DHAP) by DHAP acyltransferase (DHAP AT) via the 1-acylDHAP (Ac-DHAP) pathway followed by Ac-DHAP reductase-mediated reduction. LPA can then be further acylated by the addition of an unsaturated fatty acid (generally arachidonate) to form PA, via a lysophosphatidic acid acyltransferase (LPAAT). The inverse reaction is mediated by phospholipase A<sup>2</sup> (PLA2), which thus converts PA into LPA. Both of these reactions are of particular significance for the geometry of the membrane bilayers, since cone-shaped PA is converted to inverted-cone-shaped LPA (and vice versa), thus facilitating rapid changes in membrane curvature. (c) PA is also formed by the breakdown of other phospholipids, and in particular by the activity of phosphatidylcholine (PC)-specific phospholipase D (PLD). (d) Finally, PA can be dephosphorylated by PA phosphatases (PAPs), to form diacylglycerol (DAG), a strongly conical component of the bilayer, which due to its small and uncharged headgroup, has spontaneous transbilayer movement (flip-flop). The opposite reaction is catalyzed by diacylglycerol kinases (DAGKs). PA and DAG have been shown to be in dynamic equilibrium, and this mechanism can affect the composition and curvature of both leaflets of the bilayer.

**Supplementary Tables S1**, **S3**, the H belonging to the catalytic motif I is absolutely conserved, whereas the D is conserved almost absolutely (present in 59 out of 60 sequences).

Catalytic motif II contains a conserved arginine (R). The residue that precedes this R is always (or almost always) hydrophilic, whereas amino acid residue that precedes this hydrophilic residue is usually hydrophobic. In LPAATs from animals and fungi, this R often belongs to the (F/Y)xxR pair, see **Supplementary Tables S1**, **S3**.

Catalytic motif III contains a conserved EGTR sequence. In most LPAAT enzymes, although not in mammalian LPAATδ, amino acid residue that is two residues upstream of glutamic acid (E) belonging to the EGTR sequence is phenylalanine (F), see **Supplementary Tables S1**, **S3**. In PlsC, a LPAAT from bacterium Thermotoga maritima, the only LPAAT for which the crystal structure is available, conserved R belonging to this EGTR sequence is ideally positioned to bind the 3<sup>0</sup> -phosphate of LPA

(Robertson et al., 2017). Among sixty LPAATs from different species presented in **Supplementary Tables S1**, **S3**, E and glycine (G) of the catalytic motif III are also almost absolutely conserved. They are present, respectively, in 58 and 59 of the 60 sequences shown in **Supplementary Tables S1**, **S3**.

Catalytic motif IV contains a conserved proline (P), see (Yamashita et al., 2014; Korbes et al., 2016) and **Supplementary Tables S1**, **S3**. In most LPAAT enzymes, two residues preceding this P are hydrophobic, see **Supplementary Tables S1**, **S3**. Sometimes the residue following this P also plays a role in catalysis (Yamashita et al., 2014).

In mammalian LPAATα, mutations of conserved residues belonging to the calatylic motifs lead to the strong inhibition of catalytic activity. Among these mutations are the following: mutation of H of the motif I to alanine (A), mutations of D of the motif I to E or asparagine (N), mutation of R of the motif II to A, mutation of E of the motif III to D or glutamine (Q), mutation of G of the motif III to leucine (L), mutation of R of the motif III to A or lysine (K) (Yamashita et al., 2007, 2014).

Naturally occurring mutations in human LPAATβ, the best studied and characterized of all mammalian LPAATs, are associated with type one Berardinelli-Seip CGL (Agarwal et al., 2002; Magre et al., 2003; Agarwal, 2012; Subauste et al., 2012) and Brunzell syndrome (Fu et al., 2004). Some of the mutations causing the CGL affect residues belonging to the catalytic motifs and/or were shown to result in partial or complete inhibition of LPAATβ enzymatic activity. Specifically: (i) mutation of E to K at position 172 (E172K) belonging to the conserved EGTR sequence, within the catalytic motif III (Magre et al., 2003; Haghighi et al., 2012); (ii) deletion mutation 140delF affects F that belongs to the conserved FxxR pair within the catalytic motif II (Haque et al., 2005); and (iii) mutation of serine (S) to N at position 100 (S100N) changes the sequence within the catalytic motif I (Cortes et al., 2009).

Finally, single mutations H96A, D101A, and E176A in mouse LPAATγ (Yuki et al., 2009) and quintuple mutation N95A/H96A/D101A/E176A/G177A in human LPAATγ (Schmidt and Brown, 2009) lead to complete inhibition of LPAAT activity. Residues N95, H96, and D101 belong to the catalytic motif I, whereas residues E176 and G177 belong to the catalytic motif III.

#### ANIMAL LPAATδ ENZYME

Lysophosphatidic acid acyltransferase δ is a member of the LPAAT family (Lu et al., 2005; and references therein) that exhibits catalytic activity with LPA, but not with most other major lysophospholipids, such as LPC, LPE, LPS, LPI, LPG, MLCL, DLCL or with glycerol 3-phosphate (G3P) (Eto et al., 2014; Bradley et al., 2015). The unsaturated acyl-CoAs C22:6, C20:4, C18:1 followed by C16:0 are the preferred substrates of this enzyme with Vmax = 23.2 ± 2.4 nmol/min/mg and K<sup>m</sup> = 42.9 ± 2.9 µM for 18:0-LPA (Eto et al., 2014) and Vmax = 38 ± 1 nmol/min/mg, K<sup>m</sup> = 29 ± 1 µM for 18:1-LPA (Pagliuso et al., 2016). The highest LPAATδ catalytic activity was observed at pH 7.4 using 18:1-LPA and 18:1-acyl-CoA, with a minor decrease when pH rose to 7.6 and a sharp decrease when pH was lowered to 7.2 (Bradley et al., 2015).

The human Lpaatδ gene is on chromosome 6. Both human Lpaatδ and mouse Lpaatδ possess seven introns and five exons while Arabidopsis and rice Lpaatδ possess two introns and three exons with similar size (Korbes et al., 2016). Mammalian LPAATδ (residue numbers are for the human ortholog, UniProt accession number Q9NRZ5) contains a few conserved residues belonging to four catalytic motifs, including N95, H96 and D101 of the motif I, F143 and R146 of the motif II, E176 and G177 of the motif III, P206 of the motif IV. The H96A mutant of mouse LPAATδ shows much lower catalytic activity as compared to the wild type enzyme (Eto et al., 2014), and H96V mutant of human LPAATδ is inactive (Pagliuso et al., 2016), demonstrating that H96 residue is essential for catalysis. Moreover, this H96V mutation in human LPAATδ leads to the inhibition of the fission step during post-Golgi carrier formation indicating that LPAATδ catalytic activity is required for this process (Pagliuso et al., 2016). Based on the published results of the mutagenesis experiments with LPAATα (Yamashita et al., 2007, 2014) and LPAATγ (Yuki et al., 2009), we expect that the following mutations will also inhibit the catalytic activity of LPAATδ: D101A, D101E, D101N, R146A, E176A, E176D, E176Q, G177L, R179A, and R179K.

A crystal structure of PlsC, a LPAAT from bacterium Thermotoga maritima (T. maritima), was reported with the studies of the functional role of various PlsC residues (Robertson et al., 2017). Based on comparison among TmPlsC and other LPAATs, we can derive some conclusions concerning the structure of mammalian LPAATδ.

In TmPlsC, the phosphate group of LPA was suggested to interact with the highly conserved residues R159 (within catalytic motif III) and K105 (Robertson et al., 2017). This K105 is located between the catalytic motif I and a highly conserved P (P112 in TmPlsC). This K is highly conserved across all biological kingdoms except Archaea, see **Supplementary Tables S1**, **S3**. In human LPAATδ, K123 followed by K124 correspond to K105 from TmPlsC, see **Supplementary Tables S1**, **S3**. Based on this sequence alignment, we can conclude that in LPAATδ, K123 or, perhaps, K124 interact with the phosphate group of LPA. In the orthologs of animal LPAATδ, K corresponding to K123 of human LPAATδ is conserved, see **Supplementary Tables S2**, **S4**. We expect that in human LPAATδ, the catalytic activity of the K123A/K124A double mutant will be negligible.

### MEMBRANE TOPOLOGY OF HUMAN LPAATδ

The TmPlsC crystal structure is consistent with the organization in two domains: the N-terminal two-helix motif and the αβ-domain that contains all four catalytic motifs (Robertson et al., 2017). This molecule does not contain any TMDs, but its α1 helix (belonging to the N-terminal two-helix motif) enters and exits on the same side of the membrane, due to the presence of the G <sup>25</sup>G <sup>26</sup> kink within this helix.

In **Supplementary Tables S1**, **S3** the sequence alignment of TmPlsC, human LPAATδ and many other LPAATs from different biological kingdoms is indicated. One of TmPlsC aromatic residues suggested to interact with the apolar interior of the lipid bilayer, tryptophan W116 (Robertson et al., 2017) is highly conserved, see **Supplementary Tables S1**, **S3**. This W is located between catalytic motifs I and II and, more precisely, between highly conserved P (P112 in TmPlsC) and catalytic motif II. The W in this location is present in the LPAAT enzymes from many species. In human LPAATδ, W134 and W136 are located between highly conserved P (P130 in LPAATδ, part of highly conserved PxxG motif) and catalytic motif II. In LPAATδ, tryptophans at these positions are highly conserved throughout evolution, see **Supplementary Tables S2**, **S4**. We hypothesize that W134 and W136 from LPAATδ, like W116 from TmPlsC, interact with the apolar interior of the lipid bilayer.

The membrane topology of a few LPAATs, such as human LPAATα (Yamashita et al., 2007), human LPAATγ (Schmidt et al., 2010), Saccharomyces cerevisiae SLC1 (Pagac et al., 2011) and peanut LPAT4 (Chen et al., 2012), was studied experimentally as well as using bioinformatics. All enzymes reported in these studies are localized to the ER or Golgi membranes and contain a TMD between catalytic motifs I and II, while no TMD is foreseen between catalytic motifs II and III (see Figure 3 of Yamashita et al., 2014). This topological organization is quite unusual (motif I on one side of the membrane, and motifs II and III on the other side) and it can be explained by the need to bring in close proximity catalytic motifs I–IV that may penetrate into the membrane from the cytosolic or luminal side to act in concert (Yamashita et al., 2007, 2014; Chen et al., 2012).

Eto et al. (2014) predicted six TMDs in LPAATδ by HMMTOP transmembrane topology prediction server (Tusnady and Simon, 1998, 2001).

We would like to point out that no long hydrophobic stretches between catalytic motifs II and III and between catalytic motifs III and IV are present in LPAATδ (see **Supplementary Tables S2**, **S4**). Accordingly, it is reasonable to propose that no TMD characterizes the organization of this LPAATδ segment, and, hence, the three catalytic motifs II, III, and IV are located on the same side of the membrane. It should be noted, however, that the stretch L126AYVPIIGWMWYF<sup>138</sup> between catalytic motifs I and II of human LPAATδ is very hydrophobic and that similar hydrophobic stretches are present in this location in LPAATδ orthologs from other animal species (**Supplementary Tables S2**, **S4**). It may be hypothesized that this stretch is a TMD and, hence, in LPAATδ, as in all four eukaryotic LPAATs mentioned above, a TMD is present between catalytic motifs I and II. However, this stretch (13 residues) is unusually short, and the vast majority of TMDs are longer (Singh and Mittal, 2016). As a consequence, if this stretch is indeed a TMD, the location of catalytic motifs I and II of LPAATδ on the different sides of the membrane would not favor catalysis (Yamashita et al., 2007; Schmidt et al., 2010; Pagac et al., 2011; Chen et al., 2012).

In order to bring some clarity, we decided to use CCTOP server<sup>1</sup> , a web-based application providing prediction of membrane protein topology. This server utilizes ten different state-of-the-art topology prediction methods, including HMMTOP (Dobson et al., 2015). We studied membrane topology of human LPAATδ using CCTOP. We found that, according to this prediction, with reliability 81.1, human LPAATδ contains only three TMDs: one TMD is upstream of the catalytic motif I, whereas two more TMDs are downstream of the catalytic motif IV. Hence, according to this prediction, all four catalytic motifs are located on the same side of the membrane (**Figure 3**).

We hypothesize that human LPAATδ, like TmPlsC, does not contain any TMDs between catalytic motifs I and II, and all four catalytic motifs are located on the same side of the membrane (**Figure 3**). Moreover, we hypothesize that in human LPAATδ, the L126AYVPIIGWMWYF<sup>138</sup> stretch forms an α-helix, and this helix, like the α1 helix in TmPlsC, enters and exits on the same side of the membrane. In the α1 helix in TmPlsC, such topology is facilitated by the presence of the G <sup>25</sup>G <sup>26</sup> kink in the middle of this helix. We believe that in human LPAATδ, the motif P130xxG<sup>133</sup> plays a role of such kink. This PxxG motif is highly conserved among various LPAATs, see **Supplementary Tables S1**, **S3**, and is conserved across animal LPAATδ orthologs, see **Supplementary Tables S2**, **S4**. In proteins, kinks often coincide with the hinges (flexible regions that decouple pre-hinge and post-hinge portions of protein segment) (Cordes et al., 2002; Bright and Sansom, 2003; Hall et al., 2009). Molecular hinges often contain P residues (Sansom and Weinstein, 2000; Cordes et al., 2002; Bright and Sansom, 2003; Hall et al., 2009). The flexibility introduced by a P residue can be increased in the presence of G (Sansom and Weinstein, 2000; Cordes et al., 2002; Bright and Sansom, 2003) close to P, usually not more than four residues between P and G (Bright and Sansom, 2003).

The function of such hypothetical conformation of the L <sup>126</sup>AYVPIIGWMWYF<sup>138</sup> stretch (entering and exiting on the same side of the membrane) might be, as suggested to the twohelix motif of TmPlsC (Robertson et al., 2017), to be a "fishing bobber" that suspends the LPAATδ active site close to the lipid molecules involved in reaction. Similar function of the short hydrophobic stretch between catalytic motifs I and II can be hypothesized also in other LPAATs.

## PHYSIOLOGICAL FUNCTIONS OF LPAATδ ENZYME IN MOUSE MODEL

Lysophosphatidic acid acyltransferase δ is most highly expressed in brain (Eto et al., 2014; Bradley et al., 2015, 2016, 2017), but also in muscle (Bradley and Duncan, 2018) and, to a lesser extent, in other tissues, such as lungs, intestines, epidermis, and spleen. At the level of the mouse central nervous system, LPAATδ is abundant at the brain stem, cortex, hippocampus, cerebellum and olfactory bulbs (Bradley et al., 2015).

The characterization of the physiological functions of LPAATδ is based on the analysis of the dysfunctions associated with Lpaatδ gene knockout in mouse model. The absence of this enzyme results in a wide range of alterations that include: cognitive

<sup>1</sup>http://cctop.enzim.ttk.mta.hu

FIGURE 3 | Proposed membrane topology of human LPAATδ. The transmembrane domains (TMD; in turquoise) were predicted by CCTOP prediction server (http://cctop.enzim.ttk.mta.hu) (Dobson et al., 2015) and indicated as TMD1 (amino acids 15-38), TMD2 (amino acids 308-327), and TMD3 (amino acids 333-352). The four Catalytic Motifs of LPAATδ are indicated as I-IV in purple, and their amino acid sequences are reported in the sequence alignment as Motif I (amino acids 93-103), Motif II (amino acids 140-146), Motif III (amino acids 173-182), and Motif IV (amino acids 204-209). Highly conserved proline P130 and glycine G133 in the PxxG motif between catalytic motifs I and II are highlighted. Lysines K123 and K124 are highlighted. The N-terminus of the protein is predicted to be located in the lumen while the C-terminus is predicted to be located in the cytoplasm (as indicated).

impairment, impaired force contractility and altered visceral white adipose tissue depots.

The level of PI, PE, and PC is significantly lower (by 52, 32, and 38%, respectively) in the brain of Lpaatδ <sup>−</sup> mice (Bradley et al., 2015). However, the loss of LPAATδ did not influence the level of brain PA (Bradley et al., 2015) and this can be explained by redundancies in the LPAAT family. Indeed, an up-regulation of LPAATε, LPAATα and, even more, of LPAATβ has been found in the brain of LPAATδ <sup>−</sup> mice indicating a degree of compensation in PA synthesis (Bradley et al., 2015). Of note, this adaptation in PA brain content was not able to compensate the reduced levels of the other phospholipid species (PI, PC, and PE) observed in the brain of Lpaatδ <sup>−</sup> mice. This implies that the LPAATδ enzyme produces a specific pool of PA that is used as precursor to support the biosynthesis of PI, PC, and PE. The levels of the other major brain phospholipids were not affected indicating an unique function of LPAATδ in regulating the PI, PC and PE as downstream Kennedy pathway derivatives (Bradley et al., 2015).

Lpaatδ <sup>−</sup> mice have significant impairments in spatial learning and memory (Bradley et al., 2017). This phenomenon can be partly attributed to the drastically lower brain content of the NMDA receptor subunits (namely NR1, NR2A, and

NR2B), and of the GluR1 subunit of the AMPA receptor. NMDA receptor and AMPA receptor are two glutamate-gated transmembrane proteins involved in synaptic plasticity and memory (Bradley et al., 2017). Such a significantly reduced neural content of these subunits, in turn, might be explained by the noticeable decrease in PI, PE, and PC in the brain of Lpaatδ <sup>−</sup> mice. The presence of these glycerophospholipids might be required to preserve the biophysical properties of neuronal membranes in order to generate the conditions for the correct assembly and function of these receptors (Bradley et al., 2017). Specific lipids have been reported to have a modulatory role in the structure and function of many membrane proteins (Opekarova and Tanner, 2003; Lee, 2004; Bogdanov et al., 2008; Zhukovsky et al., 2013; and references therein). We expect that phospholipids whose synthesis depends (directly or indirectly) on LPAATδ are required for the native conformation and normal function of a few other membrane proteins, and the absence of LPAATδ might lead to the disruption of the function of these proteins.

Lysophosphatidic acid acyltransferase δ possesses catalytic activity for DHA-CoA (Eto et al., 2014). DHA is required for neurite outgrowth in hippocampal neurons (Calderon and Kim, 2004), and its reduced levels is associated with the development of cognitive and neurodegenerative disorders. Hence, the high level of LPAATδ expression in the brain is consistent with DHA being abundant among brain phospholipids (Yamashita et al., 2014), suggesting an important role of LPAATδ in maintaining DHA in neural membranes (Eto et al., 2014).

Lysophosphatidic acid acyltransferase δ was found in various muscle types. However, it was detected at highest levels in soleus, a red oxidative fiber-type that is rich in mitochondria (Bradley and Duncan, 2018). This is consistent with the localization of LPAATδ to the outer mitochondrial membrane (Bradley et al., 2015). Lpaatδ <sup>−</sup> mice showed increased PA and PE contents on fiber-type composition that in turn is suggested to impair the force contractility in soleus. These effects seem not associated to LPAATδ-related mitochondria dysfunction; as such, the Lpaatδ <sup>−</sup> mice did not exhibit impaired mitochondrial function or reduced mitochondrial content (Bradley et al., 2015). In Bradley et al. (2018), a compensatory mechanism in PA synthesis was also reported. Indeed, LPAATβ and LPAATε are specifically upregulated in soleus of Lpaatδ <sup>−</sup> mice, but not LPAATγ and LPAATα.

Lysophosphatidic acid acyltransferase δ is also highly expressed in white adipose tissue (WAT), particularly in epididymal and perirenal WAT (Prasad et al., 2011; Mardian et al., 2017). Male mice deficient in Lpaatδ gene have significant (by 40%) increase in the epididymal WAT weight with no effects on perirenal and inguinal WAT, as well as brown adipose tissue. The high PA and TAG levels in the epididymal WAT of Lpaatδ − mice is associated with an increase in adipocyte size rather than in adipocyte number. This is explained by an impaired lipolysis process due to reduced expression levels of adipose triglyceride lipase and phosphorylated hormone-sensitive lipase (Mardian et al., 2017). Here, a compensatory upregulation of LPAATα, LPAATβ, LPAATγ, and LPAATε occurs only in the perirenal WAT and not in the epididymal WAT. This adequate compensation mechanism is associated with normal tissue glycerolipid contents and, in turn, with normal tissue function (Mardian et al., 2017).

Interestingly, the loss of Lpaatδ gene pointed to the functional role of the specific pool of PA generated by LPAATδ enzyme. Indeed, although total PA level may be compensated in Lpaatδ − mice tissue by the induction of other LPAATs, the above studies indicated that the pool of PA generated by adaptive mechanisms is not able to functionally replace the LPAATδ-mediated production of PA and the downstream phospholipid derivatives that support specific cellular and tissue demands (as indeed shown in brain, soleus muscle and epididymal WAT in Lpaatδ − mice) (Bradley and Duncan, 2018; and references therein).

# SUBCELLULAR LOCALIZATION OF LPAATδ

According to Eto et al. (2014), murine LPAATδ localizes to the ER. Other authors reported that in mouse brain, LPAATδ resides on the MOM, but not on the MIM (Bradley et al., 2015). We recently demonstrated that both human and murine LPAATδ are targeted to both trans-Golgi membranes and mitochondria (Pagliuso et al., 2016).

In accordance with the endosymbiotic hypothesis, mitochondria of eukaryotes evolved from aerobic bacteria (Gray, 2012). As expected from this hypothesis, the MOM has similar composition to the plasma membrane and/or ER that may have surrounded symbiotic bacteria (Kuroda et al., 1998; and references therein), and it is not surprising that same or similar proteins are present in all these membranes (Kuroda et al., 1998; Colombo et al., 2005; Bhatt et al., 2008; Tamir et al., 2013; Marchi et al., 2014). We thus concluded that LPAATδ might localize both to the ER (from which it is transported to Golgi) and to the mitochondria (Pagliuso et al., 2016).

In addition to LPAATδ, there are other multipass transmembrane proteins that are targeted to both mitochondria and ER/Golgi. Mammalian diacylglycerol acyltransferase-2 (DGAT2) is a good example of such protein. Like LPAATδ, this enzyme is an acyltransferase, and reaction catalyzed by this enzyme, like LPAATδ-catalyzed reaction, belongs to the Kennedy pathway (Mcfie et al., 2011). DGAT2 contains two TMDs (Mcfie et al., 2014) and is localized to the ER, to the mitochondrial outer membrane, and to the lipid droplets (Mcfie et al., 2011). Mitochondrial targeting sequence contains few residues in the cytosolic portion of the protein, just upstream of the first TMD (Stone et al., 2009). This is a typical location of mitochondrial targeting signals (Rapaport, 2003; Stone et al., 2009). ER targeting signal resides within the first TMD (Mcfie et al., 2011), whereas lipid dropet targeting sequence is in the C-terminal cytosolic region of DGAT2 (Mcfie et al., 2018).

We carefully hypothesize that LPAATδ, like DGAT2, might contain separate targeting signals for mitochondria and for Golgi apparatus. The balance of LPAATδ amount between MOM and ER/Golgi membranes might be determined by comparative affinity of two signals for their respective organelles (Yogev and Pines, 2011). We expect that mutagenesis experiments will allow

to specify the localization of these putative targeting signals. Possibly, as in the case of DGAT2, positively charged residues in the cytosolic portion of LPAATδ, just upstream of the putative second TMD, belong to the mitochondrial targeting sequence.

### ROLE OF LPAATδ ENZYME IN MEMBRANE FISSION OF GOLGI MEMBRANES

Brefeldin A ADP-ribosylated substrate is the shorter splice isoform of CtBP1 protein, a member of the C-terminal binding protein (CtBP) family, known as CtBP1-S/BARS (Spano et al., 1999; Nardini et al., 2003; Valente et al., 2005; Corda et al., 2006). BARS (from here on) is a dual-function protein that in its dimeric NADH-bound conformation acts as transcriptional regulator in the nucleus, whereas in its p21-activated kinase 1 (PAK1)-phosphorylated monomeric conformation mediates membrane fission in the cystoplasm (Nardini et al., 2003, 2009; Yang et al., 2005; Colanzi et al., 2007, 2013; Liberali et al., 2008; Valente et al., 2013). BARS is a key member of a protein complex that is required for various membrane fission processes including basolaterally-directed post-Golgi carrier formation (Bonazzi et al., 2005; Valente et al., 2012), COPI-coated vesicle formation (Yang et al., 2005, 2008; Valente et al., 2012), macropinocytosis (Liberali et al., 2008; Valente et al., 2012), fluid-phase endocytosis and Golgi partitioning in mitosis (Hidalgo Carcedo et al., 2004; Bonazzi et al., 2005; Colanzi and Corda, 2007; Colanzi et al., 2007).

We and others have previously shown that the BARSinduced fission on isolated Golgi membranes correlates with PA production starting from LPA and acyl-CoA and that this LPAAT catalytic reaction supports membrane fission (Weigert et al., 1999; Kooijman et al., 2003; Shemesh et al., 2003; Pagliuso et al., 2016). This LPAAT activity is associated with, rather than intrinsic to BARS (Gallop et al., 2005) as shown by the fact that: (i) the minimal BARS domain able to support COPI-coated vesicle fission does not incorporate this activity (Yang et al., 2005); and (ii) during purification of recombinant BARS from E. coli, bacterial LPAAT, known as PlsC, specifically binds BARS (Gallop et al., 2005; Pagliuso et al., 2016). These data prompted the search for evolutionary conserved interaction between BARS and the LPAAT enzymes from bacteria to mammals. We have shown that BARS, at the TGN, is incorporated in a well-defined protein complex (Valente et al., 2012), where it binds to and activates the LPAATδ enzyme catalyzing the production of a PA pool required to support membrane fission of the basolaterallydirected post-Golgi carriers (Pagliuso et al., 2016; **Figure 4**). Specifically, as the cargo protein reaches the TGN membranes, BARS upon PAK1-mediated phosphorylation at serine 147, that induces its monomeric fission-prone conformation, assembles in a complex where it binds to the 14-3-3γ adaptor (Valente et al., 2012, 2013). Through a 14-3-3γ dimer, BARS is in a tripartite core complex with the phosphoinositide kinase PI4KIIIβ and binds proteins implicated in post-Golgi carrier formation, such as ARF, NCS-1 (also known as frequenin) and PKD (Valente et al., 2012, 2013; **Figure 4**). This complex allows to spatially and temporally couple the budding/tubulation of post-Golgi carriers with their fission. The reversible and regulated formation of this complex enhances the efficiency of the fission machinery that assembles along the tubular carrier precursor emerging out of the TGN on the site where then fission will take place (Valente et al., 2012, 2013; **Figure 4**).

Recently we showed that, when incorporated into this complex, BARS binds to and activates LPAATδ and that this LPAATδ-mediated production of PA is required for fission of post-Golgi carriers (Pagliuso et al., 2016).

Many membrane fission processes require the presence of specific lipids. These lipids could be named lipid ligands (Gopaldass et al., 2017), or lipid factors (Danne et al., 2017), or lipid cofactors (Ramachandran, 2018). In some cases, fission might be energized via the energy used in the synthesis of lipid cofactors (Gopaldass et al., 2017). We claim that PA produced in the reaction catalyzed by LPAATδ is a lipid cofactor for BARS-mediated fission.

Lysophosphatidic acid acyltransferase δ localizes to the Golgi and to the MOM (Bradley et al., 2015; Pagliuso et al., 2016), see above. PA has been shown to play a role in mitochondrial dynamics by regulating membrane fusion and fission (Frohman, 2015; Kameoka et al., 2018; and references therein). We can hypothesize that in mitochondria, the role of a PA-producing enzyme as LPAATδ is also related, among other things, to membrane fission, similar to the role of LPAATδ at the Golgi complex. Such possible dual role of LPAATδ can be somewhat analogous to the function of hFis1 that is dual targeted to mitochondria and peroxisomes and regulates membrane fission of both organelles (Delille and Schrader, 2008).

Differently from BARS, another member of the CtBP family RIBEYE possesses acyltransferase activity (Schwarz et al., 2011). In mammals, CtBPs are encoded by two genes, CtBP1 and CtBP2. CtBP1 has two splicing variants, CtBP1-L and BARS (CtBP1- S/BARS), whereas CtBP2 has three splicing variants, CtBP2-L, CtBP2-S, and RIBEYE. RIBEYE contains a large N-terminal domain that is unrelated to the CtBPs (Maxeiner et al., 2016; Schwarz and Schmitz, 2017) and a C-terminal domain. This C-terminal domain is very similar to CtBP2 and BARS in the NAD(H)-binding domain and substrate-binding domain (ca. 88.5%; Kumar et al., 2002; Nardini et al., 2003) but with relatively different residues in the C-terminal region (ca. 50.6%; Kumar et al., 2002; Nardini et al., 2003). These differences in the C-terminal region could explain the intrinsic LPAAT activity owned by RIBEYE and not by BARS (Schwarz et al., 2011). Indeed, as reported, this is the region responsible for LPAAT activity and substrate binding (Schwarz et al., 2011). Moreover, this RIBEYE enzymatic activity was not due to contaminating proteins copurifying with RIBEYE (Schwarz et al., 2011). It has been proposed that PA generated by RIBEYE at synaptic ribbons facilitates synaptic vesicle trafficking (Schwarz et al., 2011). We hypothesize that in RIBEYE (UniProt accession number Q9EQH5-2), the N792H793xxxxD<sup>798</sup> segment belongs to the LPAAT catalytic motif I, whereas E701GTR<sup>704</sup> segment belongs to the LPAAT catalytic motif III and, perhaps, F366xxR<sup>369</sup> pair belongs to the LPAAT catalytic motif II. However, we should consider that in LPAATs, the residue preceding D is almost

of LPA into PA. At the lipid level this enzymatic conversion of LPA into PA is central for the completion of the fission event. This figure is created using the web-based

always hydrophobic, see **Supplementary Tables S1**, **S3**, whereas in rat RIBEYE, this residue is hydrophilic N797. Moreover, the localization of two catalytic motifs in RIBEYE (motif III upstream of motif I) is very unusual, because in all (or almost all) other LPAATs, including LPAATδ, motif III is downstream of motif I (Korbes et al., 2016). Hence, most probably, LPAAT activity of RIBEYE evolved independently of the other LPAATs. We suppose that in RIBEYE dimer (whose crystal structure is not available yet), N792H793xxxxD<sup>798</sup> motif of one protomer might be close to the E701GTR<sup>704</sup> motif of another protomer. In the literature (Yamashita et al., 2007, 2014; Yuki et al., 2009; Eto et al., 2014; Pagliuso et al., 2016; and references therein), the catalytic activity-disrupting mutants of residues belonging to the catalytic motifs of LPAATs are reported, see above. Based on these data, we expect that in some of the mutants H793A, H793V, D798A, D798E, D798N, R369A, E701A, E701D, E701Q, G702L, R704A, R704K, the LPAAT activity of RIBEYE will be inhibited.

Following on the above considerations, we hypothesize that an evolutionary ancestor of BARS also possessed acyltransferase activity. BARS residues N232, H233 and D238 might form a vestige of an LPAAT catalytic motif I, whereas the E141GTR<sup>144</sup> stretch might be a vestige of an LPAAT catalytic motif III. Possibly, an ancestor of BARS was simultaneously an LPAAT and a fission-inducing protein, it catalyzed production of PA and used this lipid as a cofactor in membrane fission reaction, similar to Agrobacterium tumefaciens PmtA that is an enzyme producing MMPE and a membrane fission-inducing protein that uses this lipid as a cofactor in fission reaction (Danne et al., 2017). Perhaps, later BARS gradually lost its LPAAT activity and simultaneously acquired an ability to bind and activate LPAATδ, and to use PA produced by LPAATδ as a cofactor in membrane

tool BioRender (https://biorender.com/library/).

fission reaction. Interestingly, the BARS residue R144 belonging to the putative vestigial catalytic motif E141GTR<sup>144</sup> also belongs to the Rxx(pS) motif (pS is phosphorylated serine) that is involved in the interaction with 14-3-3γ adaptor protein needed for the formation of fully functional fission-inducing protein complex (Valente et al., 2012). Hypothetically, an ancestor of BARS gradually acquired the ability to bind 14-3-3γ, but in the course of this, the sequence motif E141GTR144, overlapping with the 14-3-3γ-binding site, lost the ability to be involved in LPAAT reaction. That is why, we hypothesize that simultaneously with acquiring the ability to bind 14-3-3γ, an ancestor of BARS gradually lost acyltransferase activity and acquired an ability to bind and activate LPAATδ.

#### CONCLUSION

Lipid-modifying enzymes such as those of the LPAAT family discussed above are now recognized as central actors in diverse cell functions. Their central role is also testified by the neuronal and muscle pathologies, among others, that are linked to mutations in their structure. This is rather unexpected since the numerous isoforms of the different LPAATs could suggest that redundancy is there to protect the organism from the lack of a given membrane component, PA in this case.

We can assume that redundancy cures several defects, but still there are very specific functions in cell compartments or tissues that are finely controlled by the activity of a single member of the LPAAT family.

The challenge in this field is to build a complete picture of the lipid-modifying enzyme localization, function and regulation. This knowledge will help designing the approaches and tools that will allow complementing their activity; with these procedures,

### REFERENCES


we may anticipate the treatment of diseases caused by defective lipid metabolism and/or altered membrane composition.

#### AUTHOR CONTRIBUTIONS

All authors wrote the manuscript. AF prepared the figures. MZ, CV and AF prepared the tables.

# FUNDING

Research work in the authors' laboratories was supported by the Italian Association for Cancer Research (AIRC) (to DC IG10341 and IG18776; to AL IG20786 and IG15767), the AIRC-Fondazione Cariplo TRansforming IDEas in Oncological research project (TRIDEO) (to CV IG17524), the PRONAT project, the PRIN project No 20177XJCHX, the SATIN POR project 2014–2020 and the Italian-MIUR Cluster project Medintech (CNT01\_00177\_962865).

#### ACKNOWLEDGMENTS

We thank the funding agencies listed above for supporting our research.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2019.00147/ full#supplementary-material



suggests important modifier effects. J. Clin. Endocrinol. Metab. 89, 2916–2922. doi: 10.1210/jc.2003-030485



vesicles from Golgi cisternae. Cell 59, 95–102. doi: 10.1016/0092-8674(89)90 872-6



insights into Golgi maintenance. Nat. Cell Biol. 10, 1146–1153. doi: 10.1038/ncb 1774


**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 Zhukovsky, Filograna, Luini, Corda and Valente. 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.

# Epistatic Analysis of the Contribution of Rabs and Kifs to CATCHR Family Dependent Golgi Organization

Shijie Liu<sup>1</sup>† , Waqar Majeed<sup>1</sup>† , Pranas Grigaitis<sup>2</sup> , Matthew J. Betts<sup>2</sup> , Leslie K. Climer<sup>1</sup> , Vytaute Starkuviene2,3,4‡ and Brian Storrie<sup>1</sup> \* ‡

<sup>1</sup> Department of Physiology and Biophysics, University of Arkansas for Medical Sciences, Little Rock, AR, United States, <sup>2</sup> Centre for Quantitative Analysis of Molecular and Cellular Biosystems (BioQuant), Heidelberg University, Heidelberg, Germany, <sup>3</sup> Institute of Pharmacology and Molecular Biotechnology (IPMB), Heidelberg University, Heidelberg, Germany, 4 Institute of Biosciences, Vilnius University Life Sciences Centre, Vilnius, Lithuania

#### Edited by:

Yanzhuang Wang, University of Michigan, United States

#### Reviewed by:

Robert Z. Qi, The Hong Kong University of Science and Technology, Hong Kong Irina Kaverina, Vanderbilt University, United States Mitsuo Tagaya, Tokyo University of Pharmacy and Life Sciences, Japan

> \*Correspondence: Brian Storrie StorrieBrian@uams.edu †Co-first authors ‡Co-senior authors

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology Received: 08 April 2019 Accepted: 26 June 2019 Published: 02 August 2019

#### Citation:

Liu S, Majeed W, Grigaitis P, Betts MJ, Climer LK, Starkuviene V and Storrie B (2019) Epistatic Analysis of the Contribution of Rabs and Kifs to CATCHR Family Dependent Golgi Organization. Front. Cell Dev. Biol. 7:126. doi: 10.3389/fcell.2019.00126 Multisubunit members of the CATCHR family: COG and NRZ complexes, mediate intra-Golgi and Golgi to ER vesicle tethering, respectively. We systematically addressed the genetic and functional interrelationships between Rabs, Kifs, and the retrograde CATCHR family proteins: COG3 and ZW10, which are necessary to maintain the organization of the Golgi complex. We scored the ability of siRNAs targeting 19 Golgi-associated Rab proteins and all 44 human Kifs, microtubule-dependent motor proteins, to suppress CATCHR-dependent Golgi fragmentation in an epistatic fluorescent microscopy-based assay. We found that co-depletion of Rab6A, Rab6A', Rab27A, Rab39A and two minus-end Kifs, namely KIFC3 and KIF25, suppressed both COG3- and ZW10-depletion-induced Golgi fragmentation. ZW10-dependent Golgi fragmentation was suppressed selectively by a separate set of Rabs: Rab11A, Rab33B and the little characterized Rab29. 10 Kifs were identified as hits in ZW10-depletioninduced Golgi fragmentation, and, in contrast to the double suppressive Kifs, these were predominantly plus-end motors. No Rabs or Kifs selectively suppressed COG3 depletion-induced Golgi fragmentation. Protein-protein interaction network analysis indicated putative direct and indirect links between suppressive Rabs and tether function. Validation of the suppressive hits by EM confirmed a restored organization of the Golgi cisternal stack. Based on these outcomes, we propose a three-way competitive model of Golgi organization in which Rabs, Kifs and tethers modulate sequentially the balance between Golgi-derived vesicle formation, consumption, and off-Golgi transport.

Keywords: Golgi analysis, rab, KIF, tether, genetic screen, epistasis analysis

# INTRODUCTION

In most mammalian cells, the Golgi apparatus is organized into a dynamic, ribbon-like structure, which is generated by laterally linked Golgi stacks consisting of several cisternae aligned in parallel (for review, see Wei and Seemann, 2010). Over 1000 proteins, mainly discovered in proteomics studies, are known to play a role in biogenesis and maintenance of this organelle

(Wu, 2004; Gilchrist et al., 2006; Takatalo et al., 2006). Additional regulators of the Golgi apparatus were identified in largescale RNA interference (RNAi) microscopy-based screens. This technique is based primarily on visualization and quantitative analysis of Golgi appearance in light micrographs of fluorescently tagged Golgi resident or cargo proteins. Novel regulators of trafficking-related Golgi function have been identified in screens that analyzed Drosophila tER-Golgi units (Kondylis et al., 2011), the morphology of the mammalian Golgi apparatus (Chia et al., 2012), the early secretory pathway (Farhan et al., 2010; Simpson et al., 2012; Millarte et al., 2015), TGN-tolysosome trafficking (Bickle et al., 2014), or endocytic trafficking (Liberali et al., 2014). In some studies, candidate protein downregulation or up-regulation by RNAi or cDNA over-expression, respectively, was combined with brefeldin A (BFA) induced redistribution of Golgi complex to ER in order to identify regulators of retrograde Golgi-to-ER trafficking (Lisauskas et al., 2012; Galea and Simpson, 2015). The list of hits derived in these studies identifies several novel individual regulators, but fails to derive their functional interactions. Combinatorial knock-downs provide a solution to this problem and, indeed, Galea and Simpson (2015) demonstrated the synergistic effects between Rab1a and, particularly, Rab1b and a number of other Golgi-related Rabs in controlling retrograde Golgi-ER trafficking.

Because many of the membrane trafficking events associated with the Golgi apparatus occur in the immediate vicinity of the organelle, it is difficult to resolve one from another. However, effects can be more readily determined in the case where the Golgi is scattered in a membrane traffickingdependent manner. Inhibition/redirection of Golgi membrane trafficking can lead to disruption of the interphase Golgi apparatus with a classic example being the disruption of the Golgi ribbon into scattered mini-stacks due to druginduced microtubule depolymerization (Rogalski et al., 1984; Thyberg and Moskalewski, 1985; Cole et al., 1996; Yang and Storrie, 1998). Another case is retrograde Golgi tether-dependent membrane trafficking in which tether depletion results in organelle scattering. For instance, COG complex (conserved oligomeric Golgi complex subunit), a member of the CATCHR family (Complex Associated with Tethering Containing Helical Rod) tethers vesicles mediating intra-Golgi membrane trafficking (Lee et al., 2010). The RNAi depletion of the COG3 protein causes Golgi fragmentation accompanied by accumulation of scattered glycosyltransferase-positive vesicles (Zolov and Lupashin, 2005; Shestakova et al., 2007). Knockdown of yet another CATCHR family tether, NRZ complex, through depletion of ZW10 (centromere/kinetochore protein ZW10 homolog) or RINT-1 (Rad50-interacting protein 1) also leads to fragmentation of Golgi ribbon into a cluster of punctate Golgi elements (Hirose et al., 2004; Sun et al., 2007). In interphase cells, ZW10 and RINT-1 mediate retrograde transport between the Golgi apparatus and ER (endoplasmic reticulum) while in mitotic cells the same proteins function to tether the linkage of MAD1 (mitotic spindle assembly checkpoint protein) to chromosomes in a process essential to chromosome segregation (Williams et al., 1992, 1996; Chan et al., 2000; Wainman et al., 2012; Défachelles et al., 2015).

Under the circumstances of double knockdown of tether and a second protein, the suppressive effect of co-depletion becomes a readout to test whether the co-depleted proteins are required for tether-dependent Golgi trafficking. The double knockdown protocol has shown tether-dependent Golgi apparatus organization to be regulated by at least two Rab proteins, Rab6 (Sun et al., 2007; Majeed et al., 2014) and Rab33b (Starr et al., 2010). In double-knockdown experiments, co-depletion of Rab6 strongly inhibited Golgi ribbon fragmentation induced by ZW10 or COG3 knockdown (Sun et al., 2007). Furthermore, of the 15 or more Rab6 effectors, a small subset has been found to be crucial to ZW10 and COG3-dependent Golgi organization. These are motor proteins, MyoIIA or Kif20A, and the dynein motor adaptor, BicD (Majeed et al., 2014). Depletion of BicD by RNAi or overexpression of truncated BicD C-fragment suppresses both ZW10 and COG3-dependent Golgi fragmentation. Here, we designed a phenotype-rescue approach in order to define functional interactions of key Rab proteins and Kifs in tether-dependent Golgi organization. We tested the epistatic effect of RNAi treatments directed against 19 different Golgi associated Rab proteins and all 44 human Kifs. We demonstrated that 4 Rabs and 2 Kifs suppressed both ZW10- and COG3-depletion-induced Golgi fragmentation (ZDI- and CDI-Golgi fragmentation, respectively) strongly indicating that these two pathways share common initial steps. A set of Rab and Kif proteins selectively suppressed ZDI-Golgi fragmentation, whereas, no selective suppressors of CDI-Golgi fragmentation was identified. Based on protein-protein interaction (PPI) network analysis we speculate that the observed results are the outcome of testable direct and indirect protein interactions. These outcomes and approach should be generalizable to large scale studies of protein interactions that define spatiotemporal Golgi organization.

# MATERIALS AND METHODS

# Cell Culture

HeLa cells stably expressing GalNAcT2-GFP were cultured in DMEM supplemented with 10% fetal bovine serum (FBS) and 0.45 mg/ml Geneticin (Jiang and Storrie, 2005). Cells were grown in a humidified incubator at 37◦C and 5% CO2. All cell culture media, sera and associated reagents were obtained from Life Technologies, Sigma-Aldrich or Atlas Biologicals.

#### Antibodies

Mouse monoclonal β-tubulin antibody was purchased from Sigma-Aldrich. Rabbit polyclonal antibodies were purchased from Abcam (KifC3, Rab27a) and Santa Cruz (Kif25 and Rab6). Mouse monoclonal antibody directed against Rab33B was purchased from the Frontier Institute, Shinko-nishi, Ishikari, Japan (Rab33bd5-Mo-Tk02).

## siRNA Treatment

fcell-07-00126 July 31, 2019 Time: 20:5 # 3

Both single siRNAs and SMARTPool siRNAs were synthesized by Thermo Fisher Scientific. The accession numbers of all siRNAs are shown in the **Supplementary Tables 1**, **2**. Individual siRNAs targeting specific proteins are indicated by a suffix to the protein name, e.g., −01, −02, etc. For screening experiments, HeLa-GalNAcT2-GFP were grown and transfected in µ-Plates 96 Well (Ibidi). For validation experiments at high numerical aperture, cells were grown and transfected in 35 mm tissue culture dish containing 12 mm diameter glass coverslips. After overnight culture, cells were transfected using DharmaFECT 1 (Thermo Fisher Scientific) according to the manufacturer's protocol. The next day the transfection was repeated with total incubation post initial transfection being 4 days (Majeed et al., 2014). Similar outcomes were observed at transfection concentrations of either 50 or 100 nM.

#### Light Microscopy

For screening, 25 images/well were taken using a Zeiss AxioObserver Z1 inverted microscope fitted with an automated Ludl stage and a Zeiss Definite Focus system. A 20x/0.80 numerical aperture plan apochromat objective magnified to 32× via a 1.6× Optovar was used. All screening experiments were repeated twice. Golgi organization effects as assayed at 32x objective wide field imaging were validated by collection of confocal image stacks with a 63×/1.40 numerical aperture objective and a BD CARV II spinning disk confocal accessory mounted on a Zeiss 200M inverted microscope. Images were processed with iVision-MACTM software.

### Quantification of Changes in Golgi Organization

Confocal images were first deconvolved using Huygens Professional software to sharpen the distinction between Golgi apparatus and diffuse cytoplasmic fluorescence. To speed image analysis, the deconvolved image stacks were compressed into single plane images using a maximum intensity projection algorithm (MIP, iVision-MACTM software). The resulting images were then segmented between Golgi apparatus and cytoplasm based on the intensity of GalNAcT2-GFP fluorescence and the number of Golgi fragments and Golgi area determined by segmentation analysis using iVision-MACTM software. At least 30 cells were analyzed per data point.

#### Western Blot and qRT-PCR Analysis of Knockdown Extent

For Western blotting, cells transfected with corresponding siRNA were lysed in hot 2% SDS, followed by standard SDS-PAGE (Sun et al., 2007). Western blotting was performed using primary antibodies and appropriate secondary antibodies conjugated with IRDye 800 dyes (LI-COR). Blots were scanned and analyzed using a LI-COR Odyssey system (LI-COR).

For qRT-PCR analysis of siRNA knockdown, HeLa cells treated sequentially with 100 nM siRNA on day 0 and day 1 and after 4 days total, RNA was isolated using the RNeasy Mini Kit (Qiagen) and 3 µg of RNA was converted to cDNA using the High Capacity cDNA Reverse Transcription Kit (Thermo Fisher Scientific). 100 ng of cDNA template was amplified using Rabspecific primers, or scramble control, via the SYBR <sup>R</sup> Green PCR Master Mix Kit (Thermo Fisher Scientific). mRNA expression was quantified relative to GAPDH using the 11CT method. Results are reported as the average of 3–4 replicates.

### High-Pressure Freezing, Freeze-Substitution and Electron Microscopy

Cells grown on sapphire disks coated with a 10 nm carbon layer were transfected with siRNA as described above. Highpressure freezing was performed using a Leica EMPACT2 high-pressure freezing unit (Leica Microsystems). 100 mM mannitol and 2% Type IX ultra-low temperature gelling agarose (Sigma-Aldrich) in DPBS supplemented with 2% FBS was used as cryoprotectant. All solutions and sample holders (Swiss Precision Instruments) were pre-warmed to 37◦C. All manipulations were carried out on a heating block warmed to 37◦C and monitored with a dissecting microscope (Leica Microsystems). Frozen cells were stored in liquid nitrogen.

Specimens were freeze-substituted with anhydrous acetone containing 2% OsO4/0.1% glutaraldehyde/1% H2O at -90◦C for 16–22 h using a Leica AFS unit (Leica Microsystems). Specimens were warmed to 0◦C over 2 days, and then moved to the cold room (4◦C). In the cold room, the specimens were incubated with acetone containing 1% tannic acid/1% H2O for 1 h, then replaced with acetone containing 1% OsO4/1% H2O and incubated for 1 h. The disks were rinsed repeatedly with acetone before and after each of incubation. After that, samples were warmed to room temperature and then plastic embedded essentially as described previously (Marsh et al., 2001).

50 nm thin sections cut with a Leica UltraCut-UCT microtome were collected and post-stained with aqueous uranyl acetate and Reynold's lead citrate (Electron Microscopy Sciences) to enhance contrast. Images were taken using a Tecnai F20 intermediate-voltage electron microscope operated at 80 keV (FEI Co.).

## Construction and Analysis of Protein-Protein Interaction (PPI) Networks

We filtered the IntAct database (Orchard et al., 2014) for high-confidence interactions based on biochemical and cell biology assays for physical interactions. Highconfidence interactions were then filtered for self-loops, orphaned nodes and redundant interaction pairs. Firstdegree interactions were visualized with Cytoscape software (Shannon et al., 2003) using an Edge-weighted Forcedirected layout and a Prefuse Force-directed layout was used for second-degree interaction networks; both using default weighing schemes. In addition, second-degree interaction partners were also filtered according to their cellular localization (either "Golgi" or "vesicle") as retrieved from UniProt (Bateman et al., 2017). Scoring of betweenness

centrality was used to identify the most crucial elements in respective networks.

# RESULTS AND DISCUSSION

# Co-depletion of Rab6 and CATCHR Tethers: Comparative Analysis of Golgi Organization by Fluorescence and Electron Microscopy

We initially tested strategies for robust quantitative analysis of the Golgi complex when imaged by fluorescence and electron microscopy in three states: native compact, fragmented and restored after the double knockdown of Rab6 and retrograde Golgi tethers. We used stably transfected GalNAcT2-GFP distribution as a marker for Golgi ribbon organization in HeLa cells. In these cells, GalNAcT2-GFP localizes normally to the Golgi apparatus and distributes across the entire juxtanuclear ribbon (Storrie et al., 1998). As expected, in cells transfected with negative control siRNA, the Golgi ribbon was unaffected, presenting as a compact, juxtanuclear Golgi complex in the fluorescence microscope having 4.5 detectable fragments on average (**Figure 1A**). Consistent with previous results (Storrie et al., 2012), the control Golgi apparatus by EM was composed of closely arrayed cisternal stacks of ∼900 nm in length and

containing ∼4 cisternae. There were only a small number of vesicles (∼6 vesicles/stack) associated with the control Golgi apparatus (**Figure 1G** and **Table 1**).

The fluorescent Golgi ribbon in GalNAcT2-GFP cells transfected with ZW10 or COG3 siRNA for 2–4 days was fragmented into clustered punctate Golgi elements that were nearly 5-fold more numerous than control (**Figures 1A–C**, **2**). Under these conditions, the tethers were down-regulated by 80% (Sun et al., 2007; Storrie et al., 2012; Majeed et al., 2014). By electron microscopy (EM), the clustered, perinuclear, dilated vesicles, putative Golgi elements (see. also Zolov and Lupashin, 2005) resulting from depletion of COG3 were larger (**Figure 1L**) than the small, stacked cisternal elements resulting after ZW10 depletion (**Figure 1K**). We suggest that in the absence of rapid CATCHR tether-dependent vesicle capture, Golgi-derived vesicles should "wander" producing Golgi ribbon fragments that are detectable even by fluorescence microscopy. By EM, the number of the Golgi-proximal vesicles was increased ∼2-fold with ZW10 siRNA exposure (**Figure 1** and **Table 1**). The maximum diameter of such vesicles was ∼70 nm in crosssections of control and ZW10 knockdown cells, suggesting that we scored the same vesicle class in both conditions (**Supplementary Figure 1**).

Rab6 siRNA alone did not change the number of Golgi fragments significantly when analyzed by fluorescence microscopy (**Figures 1D**, **2**). However, by EM, as previously reported (Storrie et al., 2012), significant structural alterations can be observed: the number of Golgi cisternae increased by 1 to 2 per stack, cisternal length increased by about 3-4-fold, and the number of Golgi-associated vesicles increased nearly 10-fold (**Figure 1H** and **Table 1**). Less striking effects were observed when Rab6A and Rab6A' isoforms were depleted individually (**Figures 1I,J**). Rab6 protein levels as antibody detected were higher, 40% versus 25% (**Supplementary Table 3A**). By EM, Rab6A and Rab6A' knockdowns individually produced an elongate Golgi stack, about half the size of the Rab6 knockdown, with ∼half the vesicle accumulation of the Rab6 knockdown and no increase in cisternal number (**Figures 1I,J** and **Table 1**). Interestingly, the frequency of U-shaped Golgi stacks was high with Rab6A knockdown, but not Rab6A' and what appear to be tubular-vesicular carriers were detected in the Rab6A knockdown (**Figure 1H**). In sum, by EM the two isoforms produced less strong effects and differed somewhat in their phenotypic consequences. Interestingly, by fluorescence microscopy, Golgi area was increased by ∼25% when RNAi depletion of Rab6 is performed in an isoform-dependent manner (**Figure 2**), but is more compact (by 25%) when both isoforms were depleted. We suggest that this outcome is consistent with the higher Rab6 protein levels and indicates that a single Rab6 isoform can support sufficient trafficking to lead to modest Golgi expansion. Our data show, that Rab6 variants recruit factors needed for

TABLE 1 | Quantitative analysis of the effect of selected siRNA induced protein depletions on Golgi cisternal dimensions and associated vesicles (thin section electron microscopy).


ND, not detectable. A stacked Golgi apparatus was not observed in COG3 knockdown cells.

normal trafficking through the Golgi apparatus and/or off Golgi trafficking.

or –4 following the Rab and Kif corresponds to the siRNA sequence used in the experiment. Bars are plotted ± SEM.

We have previously demonstrated by fluorescence microscopy that RNAi-mediated down-regulation of Rab6 efficiently rescues fragmentation of the Golgi complex, caused by the depletion of both retrograde ZW10 and COG3 tethers (Sun et al., 2007; Majeed et al., 2014) and, as shown in (**Figures 1E,F**), this result is obvious at the resolution of a 32x objective screening magnification. Furthermore, by EM, we now tested for the first time whether Golgi cisternal organization was restored. We

normalized to control. The siRNA sequences directed against the proteins labeled on the X-axis are listed in Supplementary Tables 1, 2. The suffix number –1, –2

found that the extent of restoration was more complete for ZW10 than COG3 depletion (**Figures 1M,N**). Using our standard protocol in which siRNAs targeting the tethers and either the Rab6 A or A' isoform were transfected simultaneously at equimolar concentrations, we could now show that both isoforms rescued tether-induced fragmentation. However, in contrast to Rab6 co-depletion, downregulation of Rab6A and Rab6A' separately restored ZDI-Golgi fragmentation slightly more efficiently than that of CDI-fragmentation in terms of the number of the Golgi fragments visible by fluorescence microscopy (**Figure 2**). By EM, we demonstrate that suppression resulted in a normalization of Golgi cisternae length accompanied by a decrease in the number of Golgi proximal vesicles (**Figures 1M,N** and **Table 1**) suggestive of a rebalancing of membrane trafficking toward an organized Golgi cisternal stack. Interestingly, in the case of ZW10 co-depletion, vesicle size was skewed to a slightly larger diameter (**Supplementary Figure 1**) and the number of cisternae displayed across the Golgi stack frequently varied across the length of the stack (**Figure 1M**). In the control Golgi stack, cisternal length is fairly constant going across the stack while, in the ZW10-/Rab6 co-depletion case, a group of cisternae may extend only a 1/3rd of the way along the length of the stack giving the approach of a shorter stack piggy backing on top of a longer stack, one that contains fewer cisternae (**Figure 1M**).

# Combinatorial RNAi-Mediated Screen of Rab Proteins Involved in CATCHR Mediated Maintenance of the Intact Golgi Complex

We next applied epistatic RNAi-based experimentation to screen Rab GTPases in retrograde tether-dependent Golgi trafficking pathways. We tested the epistatic rescue effect of siRNAs directed against 19 Golgi-related Rab proteins chosen on the basis of literature evidence (e.g., Goud et al., 2018; **Supplementary Table 1**). The microscopy-based primary screen was performed in a wide-field modus with a 32x objective configuration, 0.80 numerical aperture, and the hits then validated by confocal microscopy with a 63× objective (see section "Materials and Methods").

Eight out of 19 chosen Rabs were knock-downed by four individual siRNAs and Rab4A was targeted with 2 individual siRNAs (**Supplementary Table 1**). The remaining eight Rabs (Rab1A, Rab1B, Rab2A, Rab10, Rab14, Rab30, Rab34, and Rab43) were targeted with one siRNA only that had been validated in previous studies. Depletions were confirmed by antibody blotting or by qRT-PCR (**Supplementary Table 3B**) as appropriate. As expected, Rab1A and Rab2A induced a strong fragmentation of the Golgi complex when depleted (**Supplementary Table 1** and **Table 2**). None of the other Rabs targeted by a single siRNA impaired the Golgi complex in a fluorescence microscopy read-out or rescued both ZDIand CDI-fragmentation.

All suppressive phenotype hits were found among Rabs targeted by multiple siRNAs. A Rab was considered as a suppressor when a strong rescue phenotype was achieved with 50% of siRNAs or a weak rescue was obtained by three out of four siRNAs. Six Rabs – Rab11A, Rab22A, Rab27A, Rab29, Rab33B, and Rab39A were suppressive for ZDI-fragmentation. For Rab27A and Rab39A, at least one siRNA out of four also showed the suppression for CDI-fragmentation in both replicates of the primary screen (**Supplementary Table 1**). We note that, as previously shown (Starr et al., 2010), siRNA directed against Rab33B at a high concentration (200 nM) suppressed Golgi fragmentation induced by COG3 knockdown. The latter effect was not reproduced with the much lower concentrations of siRNAs used in our assay, which turned out to be sufficient to cause more than 60% of the protein depletion (**Supplementary Table 3B**). Therefore, we scored Rab33B as a suppressor of ZDI-fragmentation only. None of the other Rabs were selective strong suppressors of CDIfragmentation only.

These data were also reflected in the PPI network (**Figure 3**) we constructed, using the tethers, Rabs and Kifs screened in this study (comprising 723 nodes, interconnected by 886 edges) and the tethers, effector Rabs and their direct interactors (99 nodes and 110 edges). All five hit Rabs present in the network were experimentally shown to suppress ZDI-fragmentation, but none of them were a direct interactor with this tether complex in the PPI network. Instead, they all are positioned closer to ZW10 as indicated by the lesser number of edges, required to connect Rabs with ZW10/RINT proteins, compared to the subunits of COG complex.

Interestingly, four out of seven suppressing Rabs (Rab6, Rab11A, Rab33B and Rab39A) are direct interactors to each other in the PPI network, suggesting a cooperative, non-redundant Rab-mediated control of Golgi organization. Two other Rabs, Rab8A and Rab10, also scored in the top ten with respect to their betweenness centrality value in the network analysis, (**Figure 3A** and **Supplementary Figure 2A**). However, these did not score as hits in our experiments. Based on this analysis, we suggest that the literature-guided choice of a single Rab10 siRNA should be widened. On the other hand, Rab8A which was depleted with four siRNAs failed to be included, because the two weak hits did not consistently replicate (**Supplementary Table 1**). We also observed that double knock-down of Rab2A and the tethers aggravated the fragmentation of the Golgi apparatus. However, that was not observed in the case of RNAi targeted against Rab1A that acts one step upstream from Rab2A. Possibly, unique interaction partners of these two spatially related, but sequentially acting Rabs produce the observed difference (**Supplementary Figure 3A**).

As indicated by the analysis of a second-degree interaction network for the screened Rabs (**Supplementary Figure 3**), the rescue of CDI-fragmentation by Rab27A and Rab39A may be an indirect effect. Additionally, the analysis suggests that indirect interactions through a BECN1-dependent ZW10-regulatory pathway (Frémont et al., 2013) may explain the differential effects of siRNA targeting Rab39A. Rab39A depletion induced reduction of the Golgi area, which was further aggravated by co-transfection with siRNA, targeting COG3 (**Figure 2B**). In contrast, the rescue of ZW10 fragmentation by Rab39A downregulation was complete in terms of area and numbers of the Golgi fragments at the level of fluorescence.

For the majority of our hits (**Figures 4A–L** and **Table 2**), we provide the first evidence associating them with the retrograde Golgi trafficking. Only Rab11A and Rab2A along with our positive control Rab6 were previously shown to act in the retrograde trafficking of Golgi membranes to the ER upon addition of BFA (Galea and Simpson, 2015). Interestingly, the Rabs called as hits in both studies are evolutionary conserved from yeast to humans (Pereira-Leal and Seabra, 2001), indicating the role of these regulators in the most basic and, possibly, ancient routes of membrane trafficking. In contrast, human specific proteins (Rab22A and Rab29) or Rabs expressed in higher eukaryotes (Rab27A, Rab33B and Rab39A) were identified only in our study focused on the specific CATCHR-dependent retrograde trafficking, possibly indicating evolutionary coevolvement of these Rabs and tether proteins.

### Validation of Rab27A and Rab33B as Regulators of CATCHR Mediated Post-cisternal Golgi Trafficking

We next repeated the assay with the siRNAs that showed the strongest suppression effect and acquired the images in a confocal modus. In all cases, the phenotypic data were reproduced (**Figures 5A–L**). All effector Rabs induced a little fragmentation of the Golgi complex (by a factor of 1.2–1.6) when observed by the confocal microscopy. We selected Rab27A and Rab33B, in particular, for EM studies as cases representing dual tetherand ZW10 only-specific suppressors, respectively. Rab27A is a non-Golgi protein, but its role in the transport of Golgi-derived vesicles was recently shown (Stinchcombe et al., 2001; Handley et al., 2007; Bello-Morales et al., 2012; Matsunaga et al., 2017). Furthermore, the efficient RNAi-mediated downregulation of them both could be tested on a protein level (**Supplementary Table 3B** and **Supplementary Figures 4C–F**). The knockdown of these two Rabs caused little increase in the cisternal length (**Table 1**), in contrast to fairly large changes observed by RNAi of Rab6. Nonetheless, the impairment of Golgi trafficking was obvious by 5-fold increase of the Golgi-proximal vesicles (**Figures 6A,C** and **Table 1**). Knockdown of Rabs may lead to local vesicle accumulation that limits Golgi fragmentation. Our data suggest, that the depletion of Rab27A and Rab33B impairs off-cisternal/post-cisternal effects, like release of the vesicles. In contrast, Rab6 may be necessary for the organization of cisternae itself as indicated by longer cisternal lengths in depletion experiments.

Rab27A and Rab33B similarly reconstituted the normal length of the cisternae in the double knockdowns. In case of Rab33B, nearly 30% less vesicles were also observed in the double knockdown (**Table 1** and **Figures 6B,D**). All in all, close to normal cisternal organization was restored and at the same time there was a decrease in the number of Golgi associated vesicles, suggesting a rebalancing of vesicle transport under these

TABLE 2 | Summary of Rabs and Kifs knockdown screening against ZW10 or COG3 knockdown.


Rabs highlighted in red were further tested by electron microscopy. Kifs highlighted in red were further tested by electron microscopy.

conditions toward normal Golgi cisternae organization. It is likely that Rab knockdowns suppress the fragmentation by inhibiting motor recruitment.

### Kif Proteins Involved in CATCHR-Dependent and Independent Golgi Organization

We have previously shown, that depletion of a known Rab6 effector, Kif20A, suppresses tether-induced Golgi fragmentation (Majeed et al., 2014). Here, we expanded the screen and the hit scoring procedure with the aim of determining all Kifs that play a role in CATCHR-mediated Golgi trafficking events. SMARTPool siRNAs directed against all 44 known human Kifs were screened and changes in GalNAcT2-GFP distribution analyzed visually (**Supplementary Table 2**). As a first step, we tested which of the Kifs altered the organization of the Golgi complex when down-regulated alone. Incubation of individual SMARTPools directed against 10 Kifs fragmented the Golgi apparatus into clustered punctuated Golgi elements in the interphase cells suggesting that the vesicle trafficking in these cells was altered (**Table 2**). A few of these (e.g., Kif12) enhanced Golgi fragmentation in the epistatic assay, similar to Rab2A. The relatively large number of Kifs fragmenting the Golgi when depleted may reflect the overlapping activities of the motors. Partial crosstalk between the individual SMARTPool siRNAs can be ruled out as multiple sequence database comparisons between siRNAs and the corresponding mRNA sequences failed to show overlaps in siRNA sequences. Nine Kifs showed little to no effect in the epistatic experiments and only Kif14 selectively suppressed ZW10-induced fragmentation (**Table 2**). All 10 Golgi fragmenting Kifs are plus-end motors (Hirokawa et al., 2009). Finally, Kif20A and Kif23 inhibited cytokinesis and thus caused the accumulation of multinucleate cells and long networked strands of Golgi apparatus. SMARTpool directed against Kif18A was toxic to HeLa cells.

Of the 12 Kifs that were effective suppressors of tetherdependent Golgi fragmentation in either a ZW10- or COG3 knockdown background (**Supplementary Table 2**, **Table 2**, and **Figures 4M–R**), 10 were plus end directed motors. The double suppressors were Kif25 and KifC3, both minus-end motors (Noda et al., 2001) that have common binding partners at the centrosomes (e.g., CEP170, WDR62), (**Supplementary Figure 2B**). A third minus-end motor, KIFC2, showed the suppression of COG3-induced fragmentation in one of two replicates (**Supplementary Table 2**) and hence may deserve further consideration. Rescue effects were tested further by using four individual siRNAs/gene. Of the four individual siRNAs directed against Kif25 and KifC3, two were suppressive, and caused 60–70% decrease of the respective protein expression (**Figures 5M–U** and **Supplementary Figures 4G–J**). The best performing siRNAs were used for EM studies. Similar to the suppressive Rabs, Kif25 and KifC3 shifted the morphology to the Golgi complex toward the normal state (**Table 2** and **Figures 6E– H**). Furthermore, not only the vesicle numbers, but also the distance of these vesicles from the Golgi cisternae were reduced in double ZW10/Kif knockdowns by ∼30% as compared with

possess. See also Supplementary Table 5.

Golgi-fragmenting Rabs and Kifs or suppressing Kifs for (A–C), respectively)

FIGURE 4 | Wide field microscopy (32× objective) revealed suppressive effects of Rab or Kif co-depletion on CATCHR protein-dependent Golgi fragmentation. Reference single siRNA treatments: (A) siControl, (B) siZW10, and (C) siCOG2. (D–L) Tests of the suppressive effects of various Rab siRNAs: (D–I) Rab27A and Rab33B as positive examples of suppression of Golgi fragmentation and (J–L) Rab30 as a negative example of suppression. (M–R) Tests of the suppressive effects of Kif siRNA SMARTPool examples.

FIGURE 6 | EM validation of the suppressive effects of siRNAs directed against Rab27A (A,B), Rab33B (C,D), Kif25 (E,F), and KifC3 (G,H) on CATCHR-protein dependent Golgi fragmentation.

ZW10 depletion only (**Supplementary Table 1**). As with Rab6 co-depletion, a slight increase in vesicle diameter was observed (**Supplementary Figure 1**). KifC3 was previously shown to be necessary for Golgi positioning and maintenance (Xu et al., 2002). In contrast to the depletion of other minus-end motors, like cytoplasmic dynein (Harada et al., 1998; Majeed et al., 2014) or KifC1 (She et al., 2017), no fragmentation of the Golgi complex was observed in our study as well as under the conditions of KifC3 being completely removed by targeted promoter trapping strategy under normal levels of cholesterol (Xu et al., 2002). That may reflect the varying input of each Kif for Golgi positioning or their different expression levels in different experimental systems. Also, our data suggest an intriguing possibility, that minusend motors that do or do not fragment the Golgi complex when depleted may interact with different subpopulations of microtubules (MTs), namely, centrosome-derived and Golgiderived MTs, respectively (Chabin-Brion et al., 2001; Zhu and Kaverina, 2013).

Protein-protein interaction network analysis provided little further evidence on the role of these Kifs (**Figure 3C**). In contrast to Rabs (**Figure 3A**), no clear PPI distribution pattern and/or closer positioning (in terms of the degree of interaction) of Kifs to ZW10 in the respective PPI network was observed: the suppressive Kifs were quite evenly distributed over the whole

PPI network. On the other hand, evolutionary tree comparison was more informative. Notably, nearly all suppressive Kifs were distributed within families 1, 2, 4, 5, 6, 9 of the Kif evolutionary tree (Hirokawa et al., 2009) and largely clustered away from the evolutionary tree families containing Golgi-fragmenting Kifs, distributed over the families 3, 7, 8, 10, 11, 12, 13, 14. Few exceptions were nevertheless found: e.g., Kif11 caused Golgi fragmentation when downregulated, but belongs to the family 5, which is more typical for the suppressive Kifs. The clustering of Kifs into two evolutionary based groups may reflect processivity and, thus, spatial labor distribution among these molecular motors, that drive vesicle release, anchor newly formed vesicle, and mediate off-stack vesicle trafficking. Kifs, that are able to travel over long distances rapidly, would propel vesicles away from the organelle. For instance, KIF3A operates over long distances between the Golgi and ER when mediating retrograde COPI vesicle trafficking (Stauber et al., 2006). Similarly, Kif5A is proposed to regulate motility of nearly 50% of Golgi-derived vesicles (Wozniak and Allan, 2006 ´ ) and is characterized by a remarkably high processivity (Toprak et al., 2009; Milic et al., 2014). Both of these Kifs were measured to move with the velocity of 600–800 nm/sec (Svoboda et al., 1993; Guo et al., 2019). Consequently, RNAi of such Kifs should induce vesicle accumulation around the Golgi cisternae and little changes to the overall Golgi morphology. The effect is maintained if capturing of the vesicles by the tether is impaired – we suggest that could be an explanation of a suppression phenotype. The picture is different when Kifs having lower velocity (100 nm/sec) are depleted, such as Kif11 (Guo et al., 2018, 2019) or loaded Kif15 that readily detaches from MTs under these conditions (McHugh et al., 2018). Then, the balance is shifted toward the distribution of vesicles away from the Golgi complex, likely, along the tracts of microtubules, followed by Golgi reassembly at distant sites ("fragmentation phenotype"). Consequently, little chance of suppressing loss-of-tether phenotype occurs in this situation.

#### CONCLUSION

As summarized in **Figure 7** and **Supplementary Table 4**, we have identified Rabs and Kifs in an epistatic retrograde tether-dependent visual assay that are essential for the proper cisternal organization of the Golgi complex. We validated the screening hits via electron microscopy and propose a model

# REFERENCES


of interactions among Rabs, Kifs and tether protein complexes to maintain a morphologically intact Golgi complex. Our data suggest that interphase Golgi apparatus is dynamically unstable due to a tug-of-war between retrograde tether proteins and Kifs acting at long- and short-range distances. The role of the tether would be as a timer or switch ensuring that onstack vesicle consumption is normally quick. Notably, parallel single and double knockdowns of Kifs and the tethers in our study enabled us to distinguish between tether-dependent and -independent groups of Kifs - information that was not available before. Finally, these studies provide a starting point for future detailed mechanistic experiments to characterize the functional links between individual Rabs and Kifs as well as their redundant or singular activities in spatiotemporally determining Golgi organization.

# DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the **Supplementary Files**.

# AUTHOR CONTRIBUTIONS

SL, WM, PG, MB, and LC performed the experiments, data analysis, figure and table preparation, and materials and methods portions of the manuscript. VS and BS directed the experiments, analyzed the data, prepared most of the manuscript, and edited earlier versions of the manuscript.

# FUNDING

Research in the Storrie Laboratory was supported by grants from the NIH: R01 GM092960 and U54 GM105914 and in the Starkuviene Laboratory by a grant from the Heinz Goetze Memorial Fellowship Programm der Athenaeum Stiftung.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2019.00126/ 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 Liu, Majeed, Grigaitis, Betts, Climer, Starkuviene and Storrie. 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.

# A Kinetic View of Membrane Traffic Pathways Can Transcend the Classical View of Golgi Compartments

Areti Pantazopoulou and Benjamin S. Glick\*

Department of Molecular Genetics and Cell Biology, The University of Chicago, Chicago, IL, United States

#### Edited by:

Yanzhuang Wang, University of Michigan, United States

#### Reviewed by:

Sean Munro, Medical Research Council, United Kingdom Alexandre A. Mironov, IFOM – The FIRC Institute of Molecular Oncology, Italy Alberto Luini, Italian National Research Council (CNR), Italy

> \*Correspondence: Benjamin S. Glick bsglick@uchicago.edu

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 10 May 2019 Accepted: 22 July 2019 Published: 06 August 2019

#### Citation:

Pantazopoulou A and Glick BS (2019) A Kinetic View of Membrane Traffic Pathways Can Transcend the Classical View of Golgi Compartments. Front. Cell Dev. Biol. 7:153. doi: 10.3389/fcell.2019.00153 A long-standing assumption is that the cisternae of the Golgi apparatus can be grouped into functionally distinct compartments, yet the molecular identities of those compartments have not been clearly described. The concept of a compartmentalized Golgi is challenged by the cisternal maturation model, which postulates that cisternae form de novo and then undergo progressive biochemical changes. Cisternal maturation can potentially be reconciled with Golgi compartmentation by defining compartments as discrete kinetic stages in the maturation process. These kinetic stages are distinguished by the traffic pathways that are operating. For example, a major transition occurs when a cisterna stops producing COPI vesicles and begins producing clathrin-coated vesicles. This transition separates one kinetic stage, the "early Golgi," from a subsequent kinetic stage, the "late Golgi" or "trans–Golgi network (TGN)." But multiple traffic pathways drive Golgi maturation, and the periods of operation for different traffic pathways can partially overlap, so there is no simple way to define a full set of Golgi compartments in terms of kinetic stages. Instead, we propose that the focus should be on the series of transitions experienced by a Golgi cisterna as various traffic pathways are switched on and off. These traffic pathways drive changes in resident transmembrane protein composition. Transitions in traffic pathways seem to be the fundamental, conserved determinants of Golgi organization. According to this view, the initial goal is to identify the relevant traffic pathways and place them on the kinetic map of Golgi maturation, and the ultimate goal is to elucidate the logic circuit that switches individual traffic pathways on and off as a cisterna matures.

Keywords: Golgi, cisternal maturation, compartments, recycling, COPI, COPII, AP-1, clathrin

#### THE CURRENT STATUS OF THE GOLGI APPARATUS

Cell biologists agree that the Golgi apparatus performs essential functions, but they cannot even draw a diagram of the organelle before uncertainties arise. How many compartments exist within the Golgi, and what do we actually mean by the term "compartment"? How do the different cisternae of the Golgi exchange material, and how are these cisternae connected in space and time? Where and how do resident Golgi proteins localize? Which characteristics are fundamental for the

operation of the Golgi, and which ones are particular to certain cell types? Here, we address some of these questions while proposing a conceptual framework to guide ongoing research.

#### ROLES OF THE GOLGI APPARATUS

Several functions are attributed to the Golgi in a range of cell types (Farquhar and Palade, 1981; Mironov and Pavelka, 2008; Wilson et al., 2011). The Golgi receives proteins and lipids from both the endoplasmic reticulum (ER) and the endolysosomal system, and it orchestrates the sorting and distribution of these cargoes. Thus, the Golgi acts as a crossroads in the biosynthetic and endocytic traffic routes. As newly synthesized cargo molecules transit through the Golgi, they meet protein and lipid modifying enzymes, most of which modify glycans. Indeed, owing to its numerous glycosyltransferases and glycosidases, the Golgi is a major carbohydrate factory of the eukaryotic cell, responsible for the biosynthesis of glycosphingolipids, glycoproteins, and extracellular polysaccharides (Mellman and Simons, 1992; Driouich et al., 1993; Stanley, 2011; D'Angelo et al., 2013). These functions are broadly conserved.

Other functions of the Golgi may be restricted to particular organisms. For example, the Golgi has emerged as a signaling hub that coordinates secretion with environmental cues and the cell cycle (Farhan and Rabouille, 2011; Chia et al., 2012; Luini and Parashuraman, 2016). These phenomena are probably specific to animal cells, although some signaling mechanisms might be universally important for regulating Golgi traffic. In vertebrates but not in other organisms, the Golgi promotes the nucleation of microtubules that are important for directional post-Golgi trafficking, cell polarization, and migration (Rios, 2014; Sanders and Kaverina, 2015). In plant and yeast cells but not in mammalian cells, Golgi cisternae serve as early endosomes that receive endocytic cargoes destined for recycling or degradation (Dettmer et al., 2006; Viotti et al., 2010; Day et al., 2018). Perturbation of conserved or organism-specific Golgi processes can result in disease (Zappa et al., 2018).

#### CLASSICAL DEFINITIONS OF GOLGI COMPARTMENTS

In most studied cells, Golgi cisternae are arranged in polarized stacks, with a cis side that receives traffic from the ER and a trans side that delivers secretory cargoes to the plasma membrane and the endolysosomal system (Farquhar and Palade, 1981). The trans–most cisterna and its associated tubular projections form the trans–Golgi network (TGN) (Klumperman, 2011). Vertebrate cells contain multiple Golgi stacks that are connected laterally by membrane tubules to form the juxtanuclear Golgi ribbon (Klumperman, 2011), but other organisms such as plants, insects, and the yeast Pichia pastoris contain individual or paired Golgi stacks that are distributed throughout the cell (Faso et al., 2009; Kondylis and Rabouille, 2009; Papanikou and Glick, 2009). In many fungi such as Saccharomyces cerevisiae and Aspergillus nidulans, cisternae do not form stacks, and instead are scattered in the cytoplasm (Papanikou and Glick, 2009; Pantazopoulou, 2016). The functional consequences of these differences in Golgi architecture are poorly understood (Lowe, 2011), but an evolutionary analysis indicated that stacking has been lost in multiple eukaryotic lineages and is probably not integral to the operating mechanism of the Golgi (Mowbrey and Dacks, 2009).

For both stacked and non-stacked Golgi organelles, evidence of compartmentation has been obtained. Each cisterna is a membrane-bound compartment in a literal sense, but the term "Golgi compartment" is usually taken to mean a set of functionally equivalent cisternae. Golgi cisternae differ in multiple characteristics including buoyant density (Goldberg and Kornfeld, 1983; Dunphy and Rothman, 1985), morphology (Driouich and Staehelin, 1997; Rambourg and Clermont, 1997), content of specific glycosylation enzymes (Farquhar, 1985; Rabouille et al., 1995; Stanley, 2011), content of specific membrane traffic components (Munro, 2005), membrane thickness (Sharpe et al., 2010), and lipid composition (Bigay and Antonny, 2012; Holthuis and Menon, 2014). Such observations led to an early proposal that the Golgi could be viewed as two organelles in tandem (Rothman, 1981). Later studies suggested that the Golgi could be divided into three compartments that are termed cis, medial, and trans (Dunphy and Rothman, 1985; Mellman and Simons, 1992). The TGN can be described as a sorting station that produces clathrin-coated vesicles and secretory vesicles at the exit face of the Golgi stack (De Matteis and Luini, 2008), and it is often listed as a fourth Golgi compartment. It was proposed that the yeast Golgi consists of four compartments that might correspond to cis, medial, trans, and TGN (Brigance et al., 2000). In mammalian cells, an ER-Golgi intermediate compartment (ERGIC) is also present (Appenzeller-Herzog and Hauri, 2006). Golgi compartments have been envisioned as sequential stations in an assembly line (Kleene and Berger, 1993). According to this model, vesicular transport would carry secretory cargoes from one stable Golgi compartment to the next (Glick and Luini, 2011).

A problem with these concepts is that Golgi compartments have not been defined in a precise and general way. For example, morphological distinctions between cisternae are specific to certain cell types (Driouich and Staehelin, 1997; Rambourg and Clermont, 1997). The classification of Golgi compartments is often based on the localizations of glycosylation enzymes, but different glycosylation enzymes show partially overlapping distributions, so the original idea of cleanly separated Golgi compartments is no longer valid (Nilsson et al., 1993; Velasco et al., 1993; Rabouille et al., 1995; Harris and Waters, 1996). Moreover, in the Golgi, glycosylation enzymes and glycan modifications vary substantially between organisms (Munro, 2001). For stacked Golgi organelles, the number of cisternae per stack also varies (Mollenhauer and Morré, 1991), leading to ambiguity about how to group the cisternae into compartments. There are no firm guidelines for assigning a Golgi resident protein to a specific compartment. The vague definitions of Golgi compartments have arguably

been more of a hindrance than a help in attempts to understand this organelle.

### CARGO TRANSPORT IN MATURING CISTERNAE

The idea of a compartmentalized Golgi fits with the historical assumption that Golgi cisternae are long-lived structures, but many researchers now believe that Golgi cisternae are transient structures that form de novo, progressively mature, and then fragment into secretory vesicles and other types of carriers (Glick and Nakano, 2009; Glick and Luini, 2011). This cisternal maturation model is the basis for the discussion that follows. A caveat is that alternative models continue to be put forth, based on data that are seen as being inconsistent with a simple maturation mechanism (Patterson et al., 2008; Pfeffer, 2010; Mironov et al., 2013; Pellett et al., 2013; Dunlop et al., 2017). Moreover, cisternal maturation may be augmented in some cases by specialized Golgi traffic routes, such as intercisternal tubules in mammalian cells (Beznoussenko et al., 2014). Despite such complicating factors, the support for cisternal maturation is strong.

According to the cisternal maturation model, Golgi cisternae turn over on a time scale of minutes. COPII-dependent carriers that emerge from the ER are thought to generate a new cisterna – or in mammalian cells, a new ERGIC element – which recycles transport components to the ER in retrograde COPI vesicles. The cisterna then matures by recycling some of its Golgi proteins to younger cisternae while receiving other Golgi proteins from older cisternae. Intra-Golgi recycling mechanisms include vesicle-mediated traffic of transmembrane proteins, coupled with dissociation and reassociation of peripheral membrane proteins (Munro, 2005; Papanikou et al., 2015). Finally, the cisterna dissolves into secretory carriers. In this scheme, secretory cargoes largely remain within the maturing cisternae.

The original evidence for cisternal maturation came from electron microscopy (Mollenhauer and Morré, 1991). For example, large cargoes such as algal scales can be visualized in Golgi cisternae, which apparently act as forward transport carriers (Becker et al., 1995). The generality of this mechanism was established by a rigorous morphological study of procollagen secretion in mammalian cells (Bonfanti et al., 1998). Cisternal maturation was then directly observed by video fluorescence microscopy of individual Golgi cisternae in S. cerevisiae (Losev et al., 2006; Matsuura-Tokita et al., 2006). Two-color imaging revealed that early Golgi proteins depart from a cisterna as late Golgi proteins arrive. Further evidence for Golgi maturation came from studying hyphal cells of A. nidulans, in which late Golgi cisternae ultimately dissipate into secretory carriers that move to the growing apex (Pantazopoulou et al., 2014). Recently, three-color imaging of yeast indicated that secretory cargo proteins are continuously present within the maturing cisternae as resident Golgi proteins come and go (Casler et al., 2019; Kurokawa et al., 2019). Based on the similarities between the secretory traffic machineries of fungi and mammals (Duden and Schekman, 1997; Papanikou and Glick, 2009), the maturation pathway seen in fungi is likely to be a conserved feature of the Golgi.

Cisternal maturation is thought to be driven by COPI vesiclemediated intra-Golgi recycling of resident transmembrane proteins (Schnepf, 1993; Glick and Malhotra, 1998; Pelham, 1998; Rabouille and Klumperman, 2005). COPI vesicles also mediate retrograde traffic from the Golgi (and mammalian ERGIC) to the ER (Szul and Sztul, 2011; Barlowe and Miller, 2013), and some researchers have divided COPI vesicles into two categories: COPIa vesicles that mediate recycling to the ER, and COPIb vesicles that mediate intra-Golgi traffic (Donohoe et al., 2007). The mechanisms that generate two types of COPI vesicles are still unclear, but the distinction is useful, and we will employ the COPIa and COPIb nomenclature here.

Recent work revealed that COPI vesicles are not the only drivers of cisternal maturation. A functional study of yeast indicated that COPI mediates recycling of early but not late Golgi proteins (Papanikou et al., 2015). Intra-Golgi recycling of late Golgi proteins apparently involves clathrin-coated vesicles generated with the aid of the AP-1 adaptor. Yeast AP-1 has long been implicated in the recycling of certain late Golgi proteins (Valdivia et al., 2002; Liu et al., 2008; Spang, 2015), and now yeast AP-1 has been found to be restricted to terminally maturing Golgi cisternae, implying that AP-1 recycles a subset of resident Golgi proteins within this organelle (Day et al., 2018). Interestingly, a secretory cargo protein can also be recycled from older to younger cisternae in an AP-1-dependent manner, suggesting that AP-1 vesicles are capable of transporting diverse contents (Casler et al., 2019). AP-1-dependent retrograde traffic within the secretory pathway exists in mammalian cells as well (Hinners and Tooze, 2003; Hirst et al., 2012; Matsudaira et al., 2015). Therefore, intra-Golgi recycling seems to involve the successive actions of COPI and AP-1 vesicles.

#### A CONCEPTUAL FRAMEWORK THAT LINKS CISTERNAL MATURATION, MEMBRANE TRAFFIC, AND GOLGI TRANSMEMBRANE PROTEIN LOCALIZATION

Traffic pathways at the Golgi are now broadly characterized, but fundamental questions remain. We still lack a detailed understanding of intra-Golgi recycling. Even more uncertain are the mechanisms that allow resident Golgi proteins to be concentrated in particular sets of cisternae (Banfield, 2011). Finally, as described above, imprecise definitions of compartments have led to ambiguity about the functional subdivisions of the Golgi. We propose that these issues are all related, and that an updated conceptual framework can shed light on long-standing mysteries.

### Concept 1: The Golgi Performs Multiple Functions in an Ordered Way

Golgi stacks have a polarity that reflects progressive changes in the functional properties of the cisternae

(Dunphy and Rothman, 1985; Rambourg and Clermont, 1997). This organelle is an intermediate between the ER and the plasma membrane, and a variety of lipid modification reactions – including sphingolipid and phosphoinositide synthesis, sterol traffic, and lipid flipping – transform the biosynthetic membranes of the early secretory pathway (thin, loose lipid packing, low surface charge) to the barrier membranes of the late secretory pathway and plasma membrane (thick, tight lipid packing, negative surface charge) (Bigay and Antonny, 2012; Holthuis and Menon, 2014). Meanwhile, as glycolipids and newly synthesized glycoproteins move through the Golgi, their carbohydrate side chains are modified by a series of glycosylation enzymes (Kornfeld and Kornfeld, 1985). Various glycosylation enzymes are concentrated in different cisternae, and their intra-Golgi distributions tend to reflect their order of action (Shorter and Warren, 2002). An additional order-dependent function of Golgi cisternae is sorting. Early Golgi cisternae produce COPI vesicles that recycle resident transmembrane proteins, while late Golgi cisternae produce several types of carriers that sort proteins and lipids for delivery either to the plasma membrane, or to the endolysosomal system, or to younger Golgi cisternae (Myers and Payne, 2013; Papanikou and Glick, 2014; Day et al., 2018).

By changing the functional properties of a Golgi cisterna over time, the cell acquires options that might not otherwise be available. An example is the sorting of lysosomal hydrolases in mammalian cells (Hasanagic et al., 2015). Glycans on those hydrolases are modified by addition of mannose 6 phosphate sorting tags in the youngest Golgi cisternae, before mannosidases arrive. Another example is the formation of secretory vesicles. Those vesicles are generated from the oldest Golgi cisternae, ensuring that secretory proteins are maximally processed before being delivered to the plasma membrane. The general principle is that the ordered pathway of cisternal maturation allows the Golgi to be an efficient and flexible machine for processing and sorting. Because different organisms harness these capabilities in myriad ways, the Golgi has been described as a "factory for evolvability" (Shorter and Warren, 2002).

## Concept 2: Various Membrane Traffic Pathways Operate at Different Times During Cisternal Maturation

We propose that to characterize Golgi organization, the emphasis should be not on glycosylation enzymes, which differ between organisms, but rather on membrane traffic components, which show conserved distributions in the Golgi. In mammalian and plant cells, the TGN produces clathrin-coated vesicles while earlier cisternae in the stack produce COPI-coated vesicles (Mogelsvang et al., 2004; Staehelin and Kang, 2008). Similarly, in the dispersed fungal Golgi, clathrin labels late cisternae while COPI labels early cisternae (Papanikou et al., 2015; Kim et al., 2016; Schultzhaus et al., 2017; Hernández-González et al., 2019). Another example of differential localization is the Arf1 guanine nucleotide exchange factors (GEFs): the GBF/Gea family acts mainly at the early Golgi, while the BIG/Sec7 family acts at the late Golgi or TGN (Gillingham and Munro, 2007). In yeast, the Rab proteins Ypt1 and Ypt31/32 mark the early and late Golgi, respectively (Kim et al., 2016). A general rule is that any given traffic component operates at a specific time during Golgi maturation.

This perspective suggests a possible way to reconcile the ideas of Golgi compartmentation and cisternal maturation. In the maturation model, Golgi compartments could be defined as sequential kinetic stages in the maturation process (Day et al., 2013; Papanikou and Glick, 2014). The relevance of this approach is illustrated by considering the TGN. As traditionally defined, the TGN differs from early Golgi cisternae in prominent ways – it mediates the sorting of biosynthetic cargoes into transport carriers (Griffiths and Simons, 1986; De Matteis and Luini, 2008), it remains separate from the ER when mammalian cells are treated with brefeldin A (Chege and Pfeffer, 1990; Lippincott-Schwartz et al., 1991; Wood et al., 1991), and it often peels off partially or completely from a stacked Golgi (Mollenhauer and Morré, 1991; Mogelsvang et al., 2003; Uemura and Nakano, 2013). When these properties are seen through the lens of Golgi dynamics and membrane traffic pathways, cisternal maturation

FIGURE 1 | Membrane traffic and the localization of resident transmembrane proteins in the maturing Golgi. (A) Diagram of the core membrane traffic pathways that operate at the Golgi. The thick arrow represents the time course of maturation. An individual cisterna evolves along the time axis. In a stacked Golgi, the cis-to-trans spatial axis would also map onto the time axis. The border of the cisterna changes with time to reflect progressive changes in the lipid bilayer. Thin arrows represent vesicular transport pathways, with relevant coats or adaptors labeled. Bars indicate the approximate residence times for COPI and clathrin on the maturing cisterna. Colored ovals represent Golgi transmembrane proteins that follow different recycling pathways. ER, endoplasmic reticulum; LE, late endosome. (B) Predicted kinetic signatures for four different classes of Golgi transmembrane proteins. The colors correspond to those of the colored ovals in (A). See the text for details.

can be divided into an early stage when the cisternae produce COPI vesicles, and a late or TGN stage when the cisternae produce clathrin-coated vesicles (**Figure 1A**). This distinction is valuable.

However, the behavior of additional traffic components would require these two stages to be divided further. After Sec7 recruits Arf1 to the late Golgi in yeast, the GGA clathrin adaptor arrives significantly before the AP-1 clathrin adaptor (Daboussi et al., 2012; Day et al., 2018). In A. nidulans, the GEF for the Ypt31/32 homolog RabE arrives just before the Sec7-labeled late Golgi dissipates into secretory carriers (Pantazopoulou et al., 2014; Pinar et al., 2015). In the yeast Golgi, the Tlg1 SNARE arrives and departs somewhat earlier than Sec7 (Day et al., 2018), and the Golgi residence time of the Rab protein Ypt6 partially overlaps with the residence times of Ypt1 and Ypt31/32 (Suda et al., 2013). Similarly, certain mammalian membrane traffic proteins are concentrated in medial/trans cisternae rather than being restricted to the early or late Golgi (Volchuk et al., 2004; Liu and Storrie, 2012). Morphological analysis of plant cells implied that the early Golgi first undergoes an assembly phase in which COPII vesicles are received while COPIa vesicles are produced, and then undergoes a biosynthetic phase in which COPIb vesicles are received and produced (Staehelin and Kang, 2008). Attempts to assign all of these events to discrete kinetic stages become increasingly contrived. We conclude that it would be unproductive to enumerate all of the kinetic stages of maturation in an effort to define a full set of Golgi compartments.

## Concept 3: Golgi Traffic Pathways Undergo Switch-Like Transitions During Cisternal Maturation

As an alternative to classifying Golgi compartments, we propose that the Golgi should be viewed as a maturing structure controlled by a logic circuit that turns various membrane traffic pathways on and off at different times. During cisternal maturation, each traffic pathway has either a membrane import activity that must be switched on and off at the Golgi, or a membrane export activity that must be switched on and off at the Golgi, or both activities in the case of intra-Golgi recycling. Certain traffic pathways are connected by functional links. These links can be direct – e.g., if a GTPase for one traffic pathway recruits an activator or deactivator of a GTPase for another traffic pathway – or indirect – e.g., if an intra-Golgi traffic pathway recycles an activator of an earlier traffic pathway. Maturation is driven by the traffic pathways. Because the Golgi is a flow system that exchanges membrane both internally and with other organelles, a constraint on the logic circuit is that it must maintain the homeostasis of the endomembrane system.

There is evidence that transitions between Golgi traffic pathways are rapid. A pioneering electron tomography study of the mammalian Golgi indicated that clathrin-coated vesicles bud exclusively from the trans–most cisterna while COPI vesicles bud exclusively from earlier cisternae (Ladinsky et al., 1999). Subsequent kinetic analyses of yeast cisternal maturation revealed abrupt changes in the levels of Golgi resident proteins (Losev et al., 2006; Papanikou et al., 2015; Day et al., 2018). We suggest that the transitions that turn each Golgi traffic pathway on and off are switch-like, triggered when the cisterna crosses a threshold for activation or deactivation. During the intervals between transitions, "micro-maturation" may occur in the form of smaller-scale changes in cisternal properties such as membrane lipid composition (Day et al., 2013).

In the Golgi logic circuit, the order of transitions is expected to be fixed for a given cell type and largely conserved in evolution, but the intervals between transitions might vary. By changing the interval between an "on" transition and the corresponding "off " transition, cells could modulate the number of cisternae in a Golgi stack (Bhave et al., 2014). The implication is that aspects of Golgi architecture can be understood in terms of maturation dynamics.

### Concept 4: Resident Golgi Protein Localization Reflects the Operation of Traffic Pathways

Different resident Golgi proteins are known to be concentrated in distinct sets of cisternae, and some of the signals that confer localization to specific parts of the Golgi have been characterized (Banfield, 2011), but the mechanisms that establish this polarity are obscure. For Golgi peripheral membrane proteins, GTPases are typically involved in membrane recruitment (Munro, 2005). Golgi-localized GTPase systems exhibit crosstalk that can drive sequential association and dissociation of a series of GTPases and their effectors (Segev, 2011; McDonold and Fromme, 2014). For Golgi transmembrane proteins, vesicular traffic pathways determine localization. In the simplest scenario, a given Golgi transmembrane protein arrives at a cisterna when the import activity of a traffic pathway is switched on, and then departs from the cisterna when the export activity of the same or a different traffic pathway is switched on. If two Golgi transmembrane proteins do not follow the same traffic pathway, those two proteins will not always be present at the same time in a maturing cisterna. The traditional view would be that the two proteins reside in different Golgi compartments, whereas the maturation-based view is that the two proteins have different kinetic signatures that reflect their traffic pathways.

According to this updated conceptual framework, the challenge of learning how the Golgi works comes down to the following tasks. First, we need to obtain a robust picture of the traffic pathways that operate at the Golgi. Second, we need to understand how the Golgi logic circuit switches those traffic pathways on and off. Third, we need to determine which traffic pathways are used by particular Golgi proteins. The remainder of our review provides a brief discussion of these tasks.

# GOLGI TRAFFIC PATHWAYS

Even though the components that drive vesicle budding, targeting, and fusion in the secretory pathway have been characterized in biochemical and structural detail, the

physiological roles of those vesicles are still being debated. **Figure 1A** depicts a working model for Golgi traffic based on an interpretation of existing knowledge. For clarity, this model incorporates a simplification: the traffic pathways are illustrated as operating sequentially, but in fact some of them probably overlap. Other uncertainties in the model are pointed out below.

## COPII-Mediated ER-to-Golgi Anterograde Traffic

New Golgi cisternae form through COPII-dependent export from ER exit sites (ERES) (Barlowe and Miller, 2013). In fungi and plants, the ER export pathway seems to involve spherical COPII vesicles, which fuse homotypically to make a Golgi cisterna (Mogelsvang et al., 2003; Staehelin and Kang, 2008). In animal cells, the ER export pathway is less well defined and may involve COPII-dependent extrusion of large membrane carriers (Raote and Malhotra, 2019). Regardless of the detailed mechanism, secretory cargo proteins are exported in a COPIIdependent manner from the ER and are encapsulated in a newly assembled Golgi cisterna.

#### COPIa-Mediated Golgi-to-ER Retrograde Traffic

The best characterized function of COPI vesicles is to recycle proteins to the ER (Barlowe and Miller, 2013). Recycling occurs mainly from the youngest Golgi cisternae (or the mammalian ERGIC) (Klumperman, 2011). In fungal and plant cells, COPI vesicles have been visualized at the interface between ERES and the first Golgi cisterna (Mogelsvang et al., 2003; Staehelin and Kang, 2008). The location of these COPI vesicles, referred to here as COPIa vesicles, suggests that they correspond to the Golgi-to-ER retrograde carriers that have been characterized functionally.

A number of proteins are recycled in a manner suggesting that they are transported in COPIa vesicles. These recycling proteins include soluble ER proteins bound to the KDEL/HDEL receptor, transmembrane ER proteins with C-terminal KKxx retrieval signals, transmembrane ER proteins bound to the Rer1 receptor, and the p24 family of transmembrane proteins (Cosson and Letourneur, 1997; Sato et al., 2001; Aguilera-Romero et al., 2008; Bräuer et al., 2019). Although not explicitly shown in **Figure 1A**, COPII vesicles might continue to fuse with a newly assembled cisterna after COPIa vesicles begin to recycle proteins to the ER.

# COPIb-Mediated Intra-Golgi Retrograde Traffic

COPI vesicles were initially discovered as intra-Golgi carriers (Orci et al., 1986; Malhotra et al., 1989; Rothman and Wieland, 1996), but the details of this pathway are still murky. Intra-Golgi COPI vesicles, referred to here as COPIb vesicles, are often abundantly present in the vicinity of Golgi cisternae (Ladinsky et al., 1999; Klumperman, 2011). COPIb vesicles have been proposed to travel in the anterograde direction or the retrograde direction or both, or to percolate non-directionally, and they have been proposed to carry either secretory cargo proteins or resident Golgi proteins or SNARE proteins (Rothman and Wieland, 1996; Orci et al., 2000; Rabouille and Klumperman, 2005; Fusella et al., 2013; Pellett et al., 2013). Our assumption is that COPIb vesicles function mainly to recycle Golgi transmembrane proteins from older to younger cisternae (Papanikou and Glick, 2014).

**Figure 1A** depicts a hypothetical pathway for COPIb vesicles. They are postulated to bud while a cisterna undergoes the changes that will lead to clathrin recruitment. Budding might be triggered by the arrival of COPI-interacting factors that are not present earlier in the maturation process (Witkos et al., 2019; Zhao et al., 2019). After budding, COPIb vesicles are postulated to fuse with younger cisternae. This fusion event probably occurs after cisternal assembly, as suggested by evidence that glycosylation enzymes are not yet active in the earliest Golgi cisternae (Donohoe et al., 2013; Casler et al., 2019). A consequence of the scheme shown in **Figure 1A** is that between the times of COPIa vesicle budding and COPIb vesicle budding, a cisterna may experience an interval when traffic is minimal. This interval would allow for glycosylation and lipid metabolism, and it might be prolonged in cell types that produce elaborate carbohydrate structures (Glick and Malhotra, 1998). We emphasize that while the model illustrated in **Figure 1A** is appealing, the COPIb vesicle pathway is still incompletely characterized, and other membrane flow patterns could support intra-Golgi recycling (Rabouille and Klumperman, 2005; Day et al., 2013).

Presumably, COPIb vesicles differ from COPIa vesicles with regard to both the cargoes that are packaged and the specificity of vesicle targeting. COPIa and COPIb vesicles were defined by morphological criteria, based on the locations and staining intensities of vesicles visualized by electron microscopy in plant and algal cells (Donohoe et al., 2007; Staehelin and Kang, 2008). Additional morphological studies of mammalian cells suggested the existence of two populations of COPI vesicles (Orci et al., 1997, 2000). This distinction evidently does not reflect differences in the subunit composition or structure of the COPI coat (Bykov et al., 2017; Adolf et al., 2019). Instead, biochemical studies have suggested that COPI vesicles can package two alternative classes of cargo, together with two alternative sets of vesicle tethers (Lanoix et al., 2001; Malsam et al., 2005). A possible explanation is that a single COPI machinery generates two types of vesicles based on the protein and lipid compositions of the parental cisternae.

# Bidirectional Traffic Between the Golgi and Late Endosomes

Traffic of acid hydrolases from the Golgi to late endosomes (or yeast prevacuolar endosomes) is mediated by sorting receptors that are recognized by the GGA clathrin adaptors (Bonifacino, 2004; Hirst et al., 2012; Myers and Payne, 2013). Kinetic studies of yeast cells indicated that GGA is recruited as soon as Sec7 and clathrin arrive, and significantly before the final phase of cisternal maturation (Daboussi et al., 2012). Therefore, **Figure 1A** depicts GGA-dependent traffic to late endosomes as taking place before the formation of secretory vesicles.

Acid hydrolase sorting receptors recycle from late endosomes to the Golgi by pathways that involve sorting nexins (Chi et al., 2015; Wang et al., 2018). These sorting

receptors must be returned to the Golgi to bind acid hydrolases prior to GGA recruitment, so recycling carriers from late endosomes are depicted as arriving shortly before the COPI-to-clathrin transition.

### AP-1-Mediated Intra-Golgi Retrograde Traffic

The functions of AP-1 have long been a puzzle, but as described above, evidence from mammalian and yeast cells indicates that an important role of AP-1 is to mediate retrograde traffic in the late secretory pathway (Bonifacino, 2014; Spang, 2015; Day et al., 2018). Yeast AP-1 is recruited to Golgi cisternae after GGA, and persists longer than any other characterized Golgi marker (Daboussi et al., 2012; Day et al., 2018; Casler et al., 2019). AP-1 is inferred to mediate intra-Golgi recycling during a late phase of cisternal maturation. Based on kinetic studies of the recycling of Golgi transmembrane proteins and of a secretory cargo in yeast, **Figure 1A** depicts AP-1 vesicles as arriving at about the same time that COPI is being replaced with clathrin.

Even though AP-1 arrives and departs after GGA, the residence times of these two adaptors overlap (Daboussi et al., 2012). Thus, GGA might continue to transport acid hydrolases after AP-1 has begun to recycle Golgi transmembrane proteins.

## Additional Golgi Traffic Pathways

Golgi traffic pathways beyond those already listed have been described. For example, in yeast cells, endocytic vesicles are targeted directly to the Golgi and arrive just before Sec7 (Day et al., 2018). The AP-3 adaptor is responsible for delivering certain newly synthesized transmembrane proteins to the lysosome or vacuole (Odorizzi et al., 1998; Llinares et al., 2015), and in the yeast Golgi, AP-3 probably operates soon after Sec7 arrives (Day et al., 2018). Although these pathways are undoubtedly important, we propose that **Figure 1A** is an adequate working model for the conserved core traffic machinery of the Golgi.

#### CONTROL OF GOLGI TRAFFIC PATHWAYS

To understand how Golgi traffic pathways are controlled – i.e., to describe the underlying logic circuit – we need to determine which components act in a given traffic pathway. Most of the relevant components are probably known. In addition to the coats and adaptors that have been mentioned, other players are crucial. An in-depth treatment is beyond the scope of this review, but we will highlight some of the key proteins.


Golgi tethers into three categories based on the types of cargo-containing vesicles that were captured (Wong and Munro, 2014; Gillingham and Munro, 2016). This tether classification scheme can be provisionally integrated with **Figure 1A**, as follows. (1) The first class of tethers operates at newly assembled cisternae to capture COPII vesicles as well as COPIb vesicles. (2) The second class of tethers captures vesicles arriving at the Golgi from late endosomes. (3) The third class of tethers captures intra-Golgi AP-1 vesicles. A test of this interpretation will require further molecular analysis of the vesicle types recognized by the different tethers.

• SNARE proteins drive vesicle fusion, and also contribute to vesicle targeting specificity (Malsam and Söllner, 2011). Several SNARE complexes operate within the Golgi. The SNARE complex that mediates fusion of COPII vesicles is functionally well characterized (Liu and Barlowe, 2002), but less is known about the SNAREs and SNARE complexes that mediate the other three pathways shown in **Figure 1A** for incoming vesicle fusion at the Golgi.

To fill in missing pieces of the membrane traffic story, it will be essential to determine when each traffic component arrives and departs during cisternal maturation. Such kinetic mapping can be performed by video confocal microscopy of tagged yeast Golgi proteins (Losev et al., 2006; Matsuura-Tokita et al., 2006; Papanikou et al., 2015; Ishii et al., 2016; Day et al., 2018). An analogous approach for mammalian cells is high-resolution spatial mapping of resident proteins in Golgi stacks (Tie et al., 2016). Although these sorts of mapping studies cannot prove a particular mechanism, they can constrain possible mechanisms.

Perhaps the most interesting challenge is to elucidate the functional links that define the Golgi logic circuit. Maturation can arise readily in vesicular transport systems, but multiple traffic networks are possible (Mani and Thattai, 2016). An individual traffic pathway is presumably switched on and off through the actions of other traffic pathways. Insight into these functional links is gradually emerging. We will highlight a few examples:


AP-1-dependent and COPIb-dependent pathways for intra-Golgi recycling.

Based on **Figure 1A**, we can speculate about additional functional links. For example, the yeast Vps74 protein has been implicated in COPIb-dependent recycling of resident Golgi proteins, but Vps74 recruitment requires PI4P, which is synthesized late in maturation after GGA arrival (Tu et al., 2008; Graham and Burd, 2011; Daboussi et al., 2012). It seems plausible that AP-1-dependent recycling delivers PI4P to younger cisternae, thereby helping to trigger COPIb vesicle budding. This idea should be experimentally testable. In general, functional links between Golgi traffic pathways can be revealed by examining how targeted perturbations alter Golgi maturation.

#### TRAFFIC-DEPENDENT LOCALIZATION OF GOLGI TRANSMEMBRANE PROTEINS

In both stacked and non-stacked Golgi organelles, resident transmembrane proteins are concentrated in different parts of the Golgi. We propose that this phenomenon reflects the multiple traffic pathways that mediate Golgi recycling. In **Figure 1A**, the blue, green, orange, and red ovals represent Golgi transmembrane proteins, and in **Figure 1B**, the colored curves depict the corresponding kinetics of arrival and departure for each protein. The model predicts the following:


2002; Day et al., 2018). It is also possible that some late acting glycosylation enzymes, such as yeast Mnn1 and mammalian β1,4-galactosyltransferase 1 (Graham et al., 1994; Schaub et al., 2006), recycle in AP-1 vesicles from the late Golgi/TGN to earlier cisternae.

Some transmembrane Golgi proteins may follow more complex itineraries, in which case they will not fit into any of these four categories. For example, the yeast SNARE Sed5 – homologous to mammalian syntaxin-5 – cycles through the ER, presumably via COPIa vesicles, but is also present and active later in the Golgi, suggesting that a subset of the Sed5 molecules recycle via COPIb vesicles (Wooding and Pelham, 1998; Kurokawa et al., 2019). Moreover, the mammalian cation-independent mannose 6-phosphate receptor (CI-MPR) travels to late endosomes in GGA vesicles (Bonifacino, 2004), but is also present in AP-1 vesicles and in cisternae near the trans face of the stack (Doray et al., 2002; Tie et al., 2016), suggesting that a subset of the CI-MPR molecules recycle within the Golgi via AP-1 vesicles. Despite these subtleties, the long-standing problem of determining the localization mechanism for a Golgi transmembrane protein can potentially be solved by describing the protein's traffic itinerary.

# CONCLUDING REMARKS

A vast amount of detailed information has been gathered about elements of the Golgi machine, but the field has lacked a compelling framework for integrating this information. Here, we present a new attempt to construct such a framework. Cisternal maturation is proposed to be controlled by a logic circuit that switches a series of conserved traffic pathways on and off in a choreographed manner. This hypothesis has the virtue of making testable predictions that will advance our understanding. Sooner or later, the Golgi will give up its secrets.

# AUTHOR CONTRIBUTIONS

The ideas were developed through a dialogue between AP and BG. AP wrote the initial draft of the manuscript and generated the figure. BG revised the manuscript.

# FUNDING

This work was supported by the National Institutes of Health (NIH) Grant R01 GM104010.

# ACKNOWLEDGMENTS

We thank members of the Glick Lab for useful discussion.

# REFERENCES



suggest a general mechanism for regulating organelle structure and membrane traffic. Cell 67, 601–616. doi: 10.1016/0092-8674(91)90534-6


Munro, S. (2001). What can yeast tell us about N-linked glycosylation in the Golgi apparatus? FEBS Lett. 498, 223–227. doi: 10.1016/s0014-5793(01)02488-7



**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 Pantazopoulou and Glick. 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.

# Models of Intracellular Transport: Pros and Cons

#### Alexander A. Mironov\* and Galina V. Beznoussenko\*

Department of Cell Biology, The FIRC Institute of Molecular Oncology, Milan, Italy

Intracellular transport is one of the most confusing issues in the field of cell biology. Many different models and their combinations have been proposed to explain the experimental data on intracellular transport. Here, we analyse the data related to the mechanisms of endoplasmic reticulum-to-Golgi and intra-Golgi transport from the point of view of the main models of intracellular transport; namely: the vesicular model, the diffusion model, the compartment maturation–progression model, and the kiss-and-run model. This review initially describes our current understanding of Golgi function, while highlighting the recent progress that has been made. It then continues to discuss the outstanding questions and potential avenues for future research with regard to the models of these transport steps. To compare the power of these models, we have applied the method proposed by K. Popper; namely, the formulation of prohibitive observations according to, and the consecutive evaluation of, previous data, on the basis on the new models. The levels to which the different models can explain the experimental observations are different, and to date, the most powerful has been the kiss-and-run model, whereas the least powerful has been the diffusion model.

Keywords: Golgi complex, intracellular transport, COPI, COPII, ER-Golgi transport

# INTRODUCTION

The structure of the ER–Golgi interface and the Golgi complex (GC) is well-known and has been described many times (Mironov et al., 1998c, 2017; Polishchuk and Mironov, 2004; Mironov and Pavelka, 2008; Klumperman, 2011). Furthermore, most of the molecular machines involved in intracellular transport have now been deciphered (**Table 1**). Currently, there are four main models of intracellular transport: (1) the vesicular model (VM); (2) the compartment (cisterna) maturation—progression model (CMPM); (3) the diffusion model (DM; **Supplementary Figure S1D**); and (4) the kiss-and-run model (KARM), which exists as symmetric and asymmetric variants (**Figure 1**). These models have been well-described in the past (Rabouille et al., 1995; Bannykh and Balch, 1997; Glick et al., 1997; Mironov et al., 1997, 1998a,b; Pelham and Rothman, 2000; Beznoussenko and Mironov, 2002; Luini et al., 2008; Mironov and Beznoussenko, 2008, 2011, 2012; Glick and Nakano, 2009; Pfeffer, 2010; Glick and Luini, 2011; Mironov, 2014). These models are based on three main principles: dissociation, progression, and diffusion (**Figure 2**; **Supplementary Figures S1A–C**, **S2**). Existence of several completely opposite models indicates that there are too many contradictions within this field. In order to solve this problem we used the principle of falsifiability proposed by Popper (1994): any scientific model after its maximal formulation should have a clear description of its so-called prohibitive observations.

#### *Edited by:*

Vladimir Lupashin, University of Arkansas for Medical Sciences, United States

#### *Reviewed by:*

José A. Martínez-Menárguez, University of Murcia, Spain Frederic A. Bard, Institute of Molecular and Cell Biology (A∗STAR), Singapore

#### *\*Correspondence:*

Alexander A. Mironov alexandre.mironov@ifom.eu Galina V. Beznoussenko galina.beznusenko@ifom.eu

#### *Specialty section:*

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> *Received:* 15 April 2019 *Accepted:* 16 July 2019 *Published:* 07 August 2019

#### *Citation:*

Mironov AA and Beznoussenko GV (2019) Models of Intracellular Transport: Pros and Cons. Front. Cell Dev. Biol. 7:146. doi: 10.3389/fcell.2019.00146 TABLE 1 | Main molecular machines involved in the ER-to-Golgi and intra-Golgi transport steps (summarized from current reviews).


which then fuses with the distal compartment (the upper circle).

For the VM, the main falsifying (prohibitive) observation is IGT of megacargoes. For example, the VM poses that IGT is carried out by COPI vesicles. However, COPI vesicles have a diameter of 52 nm, and thus megacargoes cannot be transported by COPI vesicles. However, they are transported (Bonfanti et al., 1998). In order to solve this problem, the megavesicle hypothesis was proposed (Volchuk et al., 2000). In 2001, we showed that megavesicles are not involved in IGT of PCI (Mironov et al., 2001). The second falsifying experiment for the VM of IGT is the depletion of cargo from COPI vesicles.

The prohibitive observations for the DM (see **Supplementary Figure S1D**) are the following: (1) the concentrating of diffusible cargoes; (2) on each transport step, SNAREs are important; (3) the rarity of connections; (4) the presence of stacks without connections during transport (Trucco et al., 2004); and (5) the deviation from a negative exponential regression line during evacuation of cargo from the Golgi zone. Thus, the DM has significant difficulties to explain the increasing concentrations of cargo proteins (including megacargoes) during EGT and IGT (see below). Also, the DM cannot rationalize the necessity for SNAREs.

Increased concentrations (augmentation of the numeric density) of any cargo, and especially of megacargoes, are the prohibitive observation for the CMPM. Also, the speed of the cargo delivery from cis-side to trans-side of the GC should be equal. According to the CMPM, all resident Golgi proteins should undergo recycling with the help of COPI vesicles. Thus, depletion of even a few of the resident proteins from COPI vesicles represents the prohibitive observation for the CMPM. The second important restriction of the CMPM is that Golgiresident proteins should not be depleted in COPI-dependent 52 nm vesicles, because if the concentration of these proteins in the vesicles were lower than in the Golgi cisterna, the recycling of

**Abbreviations:** CMPM, compartment (cisterna) maturation–progression model; COP, coatomer; DM, diffusion model; EGT, ER-to-Golgi transport; ER, endoplasmic reticulum; ERES, ER exit site; GC, Golgi complex; IGT, intra-Golgi transport; KARM, kiss-and-run model; PCI, procollagen I; VLDL, very low-density lipoprotein; VM, vesicular model; VSVG, G protein of the VSV virus.

distal compartment (2).

help of SNAREs it undergoes fusion with a distal compartment (2). (C) The lateral diffusion mode of intracellular transport between a proximal compartment (1) and a

these proteins would be very slow (Glick et al., 1997). Finally, if the Golgi cisternae are immobilized, transport should not be blocked, because of the dynamic nature of Golgi cisternae.

The main principle of the KARM is fusion before fission, and even that fusion results in fission. Fusion/ fission might occur at the same site (i.e., symmetrical variant) or at different sites (i.e., asymmetrical variant). Within the framework of the asymmetrical KARM, fusion would be between the edges of the proximal and distal compartments whereas fission would be somewhere within the proximal compartment where rows of pores or thin tubules should be localized and SNAREs should be concentrated over the cargo domains (Mironov et al., 2013). The KARM does not deny the process of cargodomain maturation (Mironov et al., 2001). The symmetrical KARM suggests what the concentration mechanism should be; i.e., narrow tubules and asymmetry of ionic composition (Mironov and Beznoussenko, 2012). However, not only narrow, but also relatively thick connections can induce increased cargo concentrations in one of two compartments. If the delivery of protons is asymmetric because the diffusion of protein aggregates backwards is slower than in the anterograde direction. Fusion and fission should have molecular mechanisms for their realization (**Supplementary Figure S2**).

The prohibitive observations for the KARM are the following:


# ER-GOLGI TRANSPORT

#### The Vesicular Model

The VM of EGT poses that the exit of cargo proteins occurs in COPII-coated buds. These buds undergo fission and form COPII-coated spherical vesicles (Antonny et al., 2003; Figure 3A in Lee et al., 2005; Brandizzi and Barlowe, 2013; Zanetti et al., 2013; Saito and Katada, 2015). After separation of their coating, these vesicles are transported to the GC as individual vesicles or vesicle aggregates (Bannykh et al., 1996). There are several variants of the VM of EGT (**Figure 3**). Cargo receptors such as TANGO1 are important for some cargoes.

The main support for the VM of EGT was the study by Kaiser and Schekman (1990). However, their interpretation already contained contradiction, because during intracellular transport, as well as COPII vesicles being generated, COPI vesicles should also be generated (according to the VM of IGT, these execute IGT). So COPI vesicles should appear, and their numbers should be higher than that COPII vesicles because their volume is smaller. We found another explanation of their results (see **Supplementary Materials**).

The second corner-stone study assumed to be in favor of the VM is that of Barlowe et al. (1994). They isolated COPIIcoated vesicles after incubation of yeast microsomes with purified component of COPII in the presence of GTP. Importantly, after incubation of microsomes with COPII subunits and GTP, they obtained a mixture of tubules and vesicles. They then filtered this membrane fraction through a gel with small pores. They concluded that these vesicles are formed by COPII and that these vesicles contain cargo proteins. Finally, Bednarek et al. (1995) described so-called COPII-coated buds on the ER of S. cerevisiae after cell permeabilization and incubation with COPII subunits.

There is no doubt that COPII is important for the exit of several cargoes from the ER (Aridor and Balch, 2000; Lee and Linstedt, 2000; Mironov et al., 2003; Omari et al., 2018). Cargo moving from the rough ER to the GC passes through several stages: COPII-, COPII/COPI-, and COPI-co-localization, to finally undergo centralization (Scales et al., 1997). COPII concentrated cargo localized within the artificial membrane (Tabata et al., 2009). Recently, Kurokawa and Nakano (2019) showed that ERES are specialized ER zones for transport of cargo proteins from the ER to the GC. Of interest, in our chapter in a Golgi book (see **Figure 1**; page 18 of Mironov and Pavelka, 2008), round profiles could be seen localized within ERES, with indications that this profile represents COPII vesicles. Thus, previously we believed that COPII vesicles exist.

In spite of the importance of COPII, EGT occurs even in the absence of COPII (reviewed by Mironov, 2014). For example, Bard et al. (2006) showed that EGT still occurs after elimination of COPII subunits, although at a slightly lower rate (54% of the control). Insect cells lacking Sec13 can divide at a normal rate (Tsarouhas et al., 2007). Caco-2 cells lacking Sec13 (in humans, this protein has only one isoform) can grow (Townley et al., 2012). That these cells can divide suggests that the delivery of membrane to the plasma membrane necessary for membrane duplication during cell division was normal. Thus, in the absence of Sec13, Caco-2 cells can transport membrane proteins. There is a lack of detectable defects in the transport or secretion of small, soluble, freely diffusible proteins or transmembrane proteins in Sec13-suppressed cells (Townley et al., 2008) and Caco-2 cells (Townley et al., 2012). Deletion of COPII subunits (i.e., Sar1A, B isoforms, or both isoforms of Sar1p, together with both isoforms of Sec23) did not inhibit EGT of soluble and membrane cargoes (Cutrona et al., 2013). Elrod-Erickson and Kaiser (1996) demonstrated that in S. cerevisiae, deletion of Sec13 together with one of the proteins "bypass of Sec13" (BST)1, BST2 (also known as EMP24, the p24 family protein) or BST3 did not cause cell death, although the removal of just Sec13 was lethal for the cells. In the absence of Sec23 and also of p24, which regulates the behavior of COPI, transport occurs in yeast. Also, in the absence of cargo transport, the GC in yeast disappears (Ayscough and Warren, 1994; Morin-Ganet et al., 2000). The VM, CMPM, and DM cannot explain this phenomenon. In contrast, the KARM can (Mironov et al., 2013). Moreover, to date, nobody has demonstrated that under normal conditions, such as at steady-state, membrane buds on the granular ER and presumably coated, contain Sec13 in their coats. Also, nobody has showed how COPII vesicles or aggregates of COPII vesicles (as was proposed by Bannykh et al., 1996) move toward the GC. We did not find membrane buds on the ER in S. cerevisiae (Beznoussenko et al., 2016). Other studies have also not demonstrated the presence of these buds (reviewed by Beznoussenko et al., 2016). Importantly, according to the most precise electron cryo-microscopy (Bacia et al., 2011), the diameter of COPII vesicles is 65–80 nm. Importantly, COPI vesicles in S. cerevisiae have a diameter of 50 nm (Beznoussenko et al., 2016). The vesicles accumulated by Kaiser and Schekman (1990) also had a diameter of 50 nm.

However, the main contradiction to the VM is the exit of megacargoes, such as pre-chylomicrons [in enterocytes (Sabesin and Frase, 1977; Siddiqi et al., 2003)], very lowdensity lipoprotein (VLDL) [in hepatocytes (Claude, 1970)] and procollagen aggregates (Figures 9–11 in Karim et al., 1979; Bonfanti et al., 1998; Mironov et al., 2003; Patterson et al., 2008). The list of megacargoes includes not only PCI (**Supplementary Figure S1**, on the right), chylomicrons and VLDL, but also relatively large viruses, virions of which are formed within the nuclear envelope or inside the ER and have diameter of up to 200 nm. For instance, Herpes virus is formed within the nuclear envelope and then is transported toward the GC. Using high-resolution electron microscopy, COPII coated buds were not seen on the granular ER, with no COPII-coat in the ER domain where virions were localized (Wild et al., 2017). Also, separated vacuoles with a virion inside their lumen were not detected. They suggested that transport is realized according to a bolus-like mechanism along the ER tubule (see Figure 11 of Wild et al., 2017). It was shown that aggregates of PC are formed already inside the lumen of the ER cisternae (Figures 9–11 of Karim et al., 1979; see also Mironov et al., 2003). Chylomicrons and VLDLs are also formed inside the lumen of the ER (Claude, 1970; Sabesin and Frase, 1977). Several virions are formed inside the lumen of the nuclear envelope.

Megacargoes cannot be inserted into 65–80 nm COPIIdependent vesicles. It is also highly unlikely that megacargoes disassemble into smaller subunits that can be packaged into conventional transport vesicles. For instance, the HSP47 protein helps PCI to form rigid 300-nm trimers already in the ER, and the environment in the GC is not suitable for their disassembly (Bruckner and Eikenberry, 1984; Bonfanti et al., 1998; Patterson et al., 2008). Similarly, chylomicrons and VLDL

FIGURE 3 | Scheme of ER-to-Golgi transport according to the VM and CMPM. Upper section: (A) Formation of COPII-coated bud. (B) Formation of a COPII-dependent protrusion, which might be partially coated with COPII, for a megacargo inside this protrusion. A COPI vesicle recycles the resident ER proteins. Middle section: (A) Movement of COPII-dependent vesicles one after another toward the Golgi complex. (B) Formation of a vesicle aggregate that moves toward the Golgi complex. Fusion of a COPI vesicle with the formation of a ER-to-Golgi carrier. (D) Delivery and fusion of the COPII vesicle to the intermediate compartment. (E) Movement of the ER protrusion. (F) Fusion of the ER protrusion, with the formation of common carrier. (G) Fusion of the protrusion-dependent ER-to-Golgi carrier with the intermediate ERGIC compartment. Lower section: Different possibilities for the final delivery of ER-to-Golgi carriers to the Golgi complex. Left: According to the pure VM: (A) The additional round of COPII-dependent vesicle formation from the ER-to-Golgi carrier from the intermediate compartment. (B) Delivery of a new vesicle to the Golgi complex. (C) According to the CMPM: COPII vesicles or ER protrusions form the new cis-Golgi cisterna. (D) According to the bolus-like mechanism within the framework of the KARM: Movement of the carrier along the ER-Golgi tubule.

are formed inside the smooth ER and cannot be fragmented in the GC. Moreover, at the electron microscopy level, there is no coat visible on the distensions of the ER that contain megacargoes (Mironov et al., 2003). Of interest, Sec13 depletion impaired the deposition of large ECM components such as collagen (Townley et al., 2008). In the absence of COPII, exit of procollagen from the ER is inhibited (Cutrona et al., 2013). Furthermore, not only the absence of a COPII coat, but also the slowdown (after depletion of Sedlin; Venditti et al., 2012) or acceleration of COPII turnover (the Sec23A-M702V mutation, Kim et al., 2010) inhibited procollagen exit from the ER.

To solve these contradictions, it has been proposed that large cargoes are transported by "megavesicles," or "megacarriers" that are formed by unusual combinations of isoforms of COPII subunits (Fromme and Schekman, 2005; Venditti et al., 2012; Malhotra et al., 2015; Santos et al., 2016; Gorur et al., 2017; Raote et al., 2017, 2018). It was proposed that mono-ubiquination of Sec31 can enlarge the COPII coat to accommodate collagen fibrils (4 × 300 nm; Jin et al., 2012). According to the megavesicles model, large cargo aggregates form in megabuds coated with COPII. It has been shown that the membrane protein TANGO1 binds PCVII, and that TANGO1 binds the COPII coat proteins Sec23/Sec24 (Saito et al., 2009). Knockdown of TANGO1 inhibits export of the bulky PCVII (but not of PCI) from the ER (Saito et al., 2009; Nogueira et al., 2014). However, to demonstrate that megabuds and megavesicles exist, it is necessary to show buds coated with COPII or separated megavesicles coated with COPII (preferably using correlative light-electron microscopy or immune electron microscopy). If megavesicles exist, Sec13 should form a cap over the procollagen aggregate, VLDL or chylomicron. Of interest, to date, this requirement has not been fulfilled. For instance, Santos et al. (2016) did not demonstrate co-localization between Sec13 and lipids or Apo B (lipid and Apo proteins of VLDL). There is not any coat visible at on the distensions of the ER filled with procollagen (Mironov et al., 2003). Claude (1970) and Sabesin and Frase (1977) have not observed pre-chylomicrons in the ER megabuds. Of interest, in Figure S1f (by Subramanian et al., 2019), PCI aggregates are larger than TANGO1- positive spots. TANGO1-positive spots do not form rings around PC aggregates and not significant overlapping of TANGO1 and PC was observed. This suggests that TANGO1 could function as the center of PC aggregation (crystallization) and not as a mechanism involved into the formation of COPII coat. Importantly, Santos et al. (2016) did not discuss the contradiction of their data by the results of Siddiqi et al. (2003). It was shown that aggregates of procollagen are formed inside the lumen of the ER cisternae (Figures 9–11 of Karim et al., 1979). However, no coat was visible over these distensions of the ER. Raote et al. (2018) do not provide any single image that directly confirms the scheme of the formation of megabuds proposed by the authors, namely, the COPII ring, then the more external ring of TANGO1, and procollagen-positive spots inside these rings. Thus, at steady-state, megabuds that contain procollagen and are coated with a COPII-like coat were also not detected (Leblond, 1989).

To demonstrate that megavesicles exist, Gorur et al. (2017) engineered cells to stably overexpress the human pro-α1 (I) collagen. Using correlative light-electron microscopy based on serial sections with a thickness of 70–100 nm, they demonstrated a structure filled with PCI with a diameter of 900 nm (Figure 2C: Z7 of Gorur et al., 2017). The thickness of the coat over this structure was more than 40 nm, whereas the typical thickness of COPII coats is 12 nm (Bannykh et al., 1996; Bacia et al., 2011). Also, the significant thickness of the serial sections indicates that the resolution along the Z-axis was 140–200 nm. Therefore, it is not possible to judge whether this structure was connected with the ER or not. Moreover, in the vast majority of studies, the diameter of the procollagen-containing ER-to-Golgi carriers or Golgi cisterna distensions filled with procollagen never exceeds 350 nm (Leblond, 1989; Bonfanti et al., 1998; Patterson et al., 2008; Perinetti et al., 2009). Importantly, in Figure 2ii of Gorur et al. (2017), the labeling for Sec31a does not form a ring, as it should do to be in agreement with the hypothesis of COPIIcoated megavesicles. Of interest, the area of the labeling for procollagen is wider than the labeling for Sec31A; namely, near the border the green intensity is higher than the red intensity, whereas in the center the intensities of red and green are equal. Gorur et al. (2017) used super-resolution light microscopy to detect the ring- or cap-like labeling for Sec31A, which should surround the collagen aggregate. Moreover, in their Figures 3A v–x, the thickness of the COPII coat is >100 nm, whereas under normal conditions, the thickness of the COPII coat is only 12 nm (Bannykh et al., 1996). Moreover, the diameter of the procollagen aggregate was only 100 nm, although under normal conditions their diameter is 300 nm (Mironov et al., 2003). Also, there is an empty space between the PCI spot and the Sec31A-positive cap (see Figure 3A viii of Gorur et al., 2017). Such a space has never been observed under normal conditions. On the other hand, in their Figures S5B:i–iii (Gorur et al., 2017), the diameter of the vesicles is about 200 nm. Only in **Figure S5B**:iv do the vesicle have a diameter of 350 nm, although this vesicle was not coated. Similar large procollagen-positive immobile dots were observed by Omari et al. (2018). Also, McCaughey et al. (2019) observed huge procollgen-positive spots (their Figure 2), which did not go to the GC.

Recently, McCaughey et al. (2019) provided direct evidence that they suggested was in favor of the very minor (if any) role of megavesicles for EGT of procollagen. They demonstrated that EGT of procollagen occurs without the formation of COPIIcoated 200–300 nm carriers. These observations contradict the megavesicle hypothesis proposed by the proponents of the VM of EGT. However, these authors did not discuss this issue, and simply claimed that their "data are consistent with COP IIdependent trafficking of procollagen" (McCaughey et al., 2019, page 12). Also, they did not cite two important papers by Patterson et al. (2008) and by Mironov et al. (2003), where the mechanisms of EGT of PCI were described and the role of COPII was questioned. Moreover, Patterson et al. (2008) presented data on PCI transport in live cells, whereas we had already demonstrated the rarity of the arrival of ER-to-Golgi carriers with a diameter of 300 nm at the GC. Finally, in contrast to McCaughey et al. (2019), we observed rare ER-to-Golgi carriers filled with procollagen III that arrived at the Golgi area after GFP bleaching (Beznoussenko et al., 2014). In McCaughey et al. (2019), the procollagen-containing dots grew inside the Golgi area. In our studies, these dots have acquired their high brightness at the periphery, and then moved to the Golgi area.

Also, Patterson et al. (2008) demonstrated ER-to-Golgi carriers containing procollagen and moving toward the GC. The size here was <350 nm. Analysis of **Movie S1** by Omari et al. (2018) revealed that after bleaching of the ER, which was filled with procollagen, within the area near the GC the spots containing concentrated procollagen moved toward the GC, which was labeled with GM130. Omari et al. (2018) claimed that they observed procollagen-positive spots initially coated with COPII, which after uncoating, moved toward the GC However, careful analysis of **Movie S2** revealed that the procollagenpositive spot is formed not within the domains that co-localize with Sec23, but near the Sec23-positive blob at a distance of about 200 nm. In **Movie S2**, the Sec23-positive rather big spot shown with a back and forth movement, when suddenly a spot containing concentrated labeling for procollagen appeared near the Sec23-positive spot. This uncoated spot then starts to move toward the GC. Sec23-positive rings that surround procollagen-positive spots were not shown. This event occurred exactly as it was described in our study (Mironov et al., 2003). There, we showed that procollagen spots are formed not within ERES, but nearby. No image is shown that demonstrates procollagen-positive spots surrounded with Sec23-positive ring, as derived from the VM of EGT. Significantly, the overall numbers decreased for both LC3-positive autophagic structures and FP-LC3–positive autophagic structures that contained FPproα2G610C(I) (Omari et al., 2018).

#### The Diffusion Model

Within the framework of the DM, EGT occurs by diffusion along constant connections between the ER and the GC. The precise characteristics of the DM of EGT have not yet been specified in the literature. Direct membrane continuities between the ER and the GC have been described many times (Flickinger, 1969, 1973; Claude, 1970; Maul and Brinkley, 1970; Bracker et al., 1971; Holzman, 1971; Morre et al., 1974; Franke and Kartenbeck, 1976; Novikoff and Yam, 1978; Uchiyama, 1982; Broadwell and Cataldo, 1983; Sasaki et al., 1984; Lindsey and Ellisman, 1985; Williams and Lafontane, 1985; Lockhausen et al., 1990; Krijnse-Locker et al., 1994; Sesso et al., 1994; Mironov et al., 2017; also reviewed in Mironov and Pavelka, 2008). These observations were based on high voltage electron microscopy (Lindsey and Ellisman, 1985), scanning electron microscopy (Tanaka et al., 1986), three-dimensional reconstruction of serial sections (Sesso et al., 1994), functional analysis of transport (Krijnse-Locker et al., 1994), and electron microscopy tomography (Ladinsky et al., 1999). For instance, Ladinsky et al. (1999) described a connection between the ER and a small cisterna that showed all of the features of Golgi cisternae; namely, buds and small pores (see Figure 3: C6′ of Ladinsky et al., 1999). They interpreted this structure as "the specialized domain of the ER." However, such small pores were never observed in the ER cisternae (see Figure 3: cis-ER,:trans-ER of Ladinsky et al., 1999). Moreover, to date, nobody has confirmed the possibility that the ER cisterna can be inserted into a Golgi stack. Fixative cannot generate membrane tubules. Moreover, the fixative usually disrupts pre-existing tubules (Duman et al., 2002). Importantly, connections between the ER and pre-Golgi carriers have been described even after application of quickfreezing (Mironov et al., 2003). Previously, we also reported such connections after depletion of both isoforms of Sar1 (Cutrona et al., 2013).

Trucco et al. (2004) (see their Figures 1d,e) demonstrated connectivity between the ER and the cis-most cisterna (CMC) after a 15◦C temperature block. However, the existence of ER– Golgi connections was not confirmed by Koga and Ushiki (2006). Finally, at 15◦C, ERGIC53/58, the KDEL receptor, SNAREs operating at the level of the intermediate compartment, members of the p24 family, and even Man I, redistribute from the pericentral GC to the peripheral spots labeled for COPI/COPII (see above). However, at this temperature, vesicular transport is blocked (Saraste and Kuismanen, 1984; Kuismanen and Saraste, 1989), which suggests that these molecules diffuse along the membrane continuity from the pericentral Golgi to ERES.

### The Cisternal Maturation–Progression Model

According to the CMPM, immature ER-to-Golgi carriers are formed by protrusion from the ER, whereas ER-resident proteins are eliminated from the ER-to-Golgi carriers by retrograde COPI-dependent vesicles (Mironov et al., 2003; see their Figures 2, 3B, upper part). The main argument in favor of the CMPM of EGT is the data of Oprins et al. (2001). They observed a significant (57.6-fold) higher aggregation of chymotrypsinogen and proposed "the concentration by exclusion" (**Supplementary Figure S1**, lower part). However, COPI vesicles that presumably operate as carriers for retrograde transport at the level of ERES cannot be used because COPI vesicle have a very high ratio between their surface and volume, which means that these vesicles are not suitable to explain such high levels of cargo concentration. Recycling of COPI vesicles would eliminate mostly the surface area but not the volume of the immature ER-to-Golgi carriers. Moreover, the simultaneous increase in the amylase aggregation was only 3.7-fold suggesting that several types of COPI vesicles (for each cargo) operate at the level of ERES. However, this has not yet been demonstrated. On the other hand, spots carrying fluorescence cargoes toward the GC do not increase the intensity of their fluorescence during the centripetal movement, and no small spots have been seen to detach from them during this movement (Presley et al., 1997; Scales et al., 1997; Stephens et al., 2000). These suggest that the elimination of membranes not containing cargo proteins does not occur.

If COPI-coated vesicles mediate retrograde, Golgi-to-ER, transport, the concentrating of proteins with KKXX motifs would be expected, such as ERGIC53/58 or p24, in COPI-coated buds. However, to date, there has been no convincing evidence that demonstrates the concentrating of either ERGIC53/58 or p24 in COPI-coated buds on ERES. Moreover, alpha 2 protein of the p24 family is not enriched in Golgi buds (Dominguez et al., 1998). ERGIC53 (Gilchrist et al., 2006) and Hsp47 are depleted in COPI-dependent vesicles that are formed also within ERES (Bannykh et al., 1996). ERGIC53/58 and proteins of p24 family are not concentrated in small, coated round profiles observed near the GC (Cole et al., 1996; Jäntti et al., 1997) or isolated in the presence of GTP (Stamnes et al., 1995, 1998; Rojo et al., 1997, 2000; Bremser et al., 1999). In the single report where ERGIC53 was found in round profiles in the Golgi area (Palokangas et al., 1998), they did not present serial sections. The microinjection of the Sar1p:GTP-restricted mutant induces redistribution of ERGIC proteins to the ER. However, after the microinjection of Sar1p:GTP-restricted mutant together with the inhibitory antibody against ßCOP, the ERGIC53/58, KDEL receptor, cholera toxin, and p24 proteins did not redistribute to the ER, whereas Shiga toxin was shifted to the ER. To explain these observations, two pathways for Golgi-to-ER transport were proposed; namely, COPI-dependent and COPI-independent (Girod et al., 1999). We think that careful analysis of the concentrating of the resident ER and ERES proteins in COPI vesicles derived from ER-to-Golgi carriers might solve this prohibitive observation (this concentration should be higher than in the ER-to-Golgi carriers) or confirm the CMPM of EGT.

#### The Kiss-and-Run Model

The characteristics of the KARM of EGT have not yet been specified in the literature. Here, we proposed our variant of the KARM of EGT (**Figure 4**). The KARM assumes that EGT is realized by a fusion-fission mechanism: initially the membrane protrusion filled with a cargo is formed, and then this protrusion fuses with a tubule that emanates from the GC, and in particular from the CMC. If the distance between the proximal and distal compartments is large, the distal compartments are extended to the proximal, and capture the cargo domain. This was shown by Casler et al. (2019). After this fusion, the fission occurs near the neck that connects the protrusion and the ER. The tubule delivers dynein to the immature ER-to-Golgi carriers. This motor moves the ER-to-Golgi carriers toward the GC. The arrival of this carrier at the GC generates flux of Ca2<sup>+</sup> from ERES (Micaroni et al., 2010a,b) and stimulates fusion of the carrier with the medial GC. When two consecutive compartments are localized far away from each other, the KARM suggests the need for a bolus-like mode of transport. The bolus model was originally proposed for exocytosis (Ayala, 1994). It implies participation of active peristaltic movement of membrane varicosities using mechanical forces generated by membrane coats located at the proximal side of the bolus. The main postulate of the asymmetric variant of the KARM, which is the only one useful for membrane transport (for

FIGURE 4 | Scheme of ER-to-Golgi transport according to the KARM. Protrusions containing small cargoes (A) and megacargoes (B) are formed within the ERES area. (C–F) A tubule is formed from the cis-most Golgi cisterna. It moves toward the ERES area along a microtubule (dashed line), with the help of kinesin (green dot). (G,H) This tubule fuses with ERES with the help of SNAREs (short black lines), and delivers dynein (red dots) to the ER-to-Golgi carrier. (I,J) The bolus moves toward the Golgi complex using dynein and microtubules. (K) Now, the ER-to-Golgi carrier is within the cis-most cisterna. Finally, the tubular connection undergoes rupture (K).

soluble cargo, the symmetric variant of the KARM is suitable), is that fusion between two consecutive compartments occurs before the fission, which takes place in the area where thin tubules (or thinning of the proximal compartment) should be present. Thus, when the ERES is far away from the point of entrance into the Golgi, there should be a specific mechanism that ensures the main KARM postulate. Therefore, the KARM assumes that a tubule emanates from the CMC. This moves toward the ERES and hits this target. This tube might use kinesin for its movement along a microtubule. Indeed, the microtubule motor, kinesin, is present on membranes that cycle between the ER and the GC. At 37◦C, kinesin was most concentrated on peripherally distributed ERES. The finding that kinesin is present on ERGIC structures is hard to reconcile with the VM, because the transport of carriers toward the GC is minus-end directed. Upon temperature reduction or nocodazole treatment, the kinesin distribution shifted onto the GC, while with brefeldin A treatment, kinesin is found in both Golgi-derived tubules and in the ER. This suggested that kinesin associates with membranes that constitutively cycle between the ER and the GC. The role of kinesin on these membranes was examined by microinjection of an anti-kinesin antibody. Golgito-ER, but not ER-to-Golgi, membrane transport was inhibited by the microinjected anti-kinesin antibody (Lippincott-Schwartz et al., 1995). Simultaneously, this peripherally moving tube delivers dynein to ERES for the consecutive centralization of the ER-Golgi carrier, which is formed within ERES. Such tubes have been described (Sciaky et al., 1997; Marra et al., 2007). This consequence of events explains why inhibition of kinesin blocks centralization of ER-Golgi carriers (Lippincott-Schwartz et al., 1995). On the other hand, Sec23p directly interacts with the dynactin complex. Co-localization of COPII and p150Glued was observed and turnover of Sec23 was increased after depolymerization of microtubules with nocodazole (Watson et al., 2005). This observation explains why in the absence of Sar1A and B (and as a consequence, in the absence of binding of Sec23p to ERES), nocodazole-dependent de-polymerization of microtubules does not induce fragmentation of the GC (Cutrona et al., 2013). Recently, Raote and Malhotra (2019) proposed that TANGO1 forms channels for procollagen that connect the ER and the GC.

Thus, the VM, DM, and CMPM cannot explain all of the data (i.e., prohibitive observations) that are contradictory to their logic, whereas the KARM should explain corner-stone observations that support the VM, DM, and CMPM.

# INTRA-GOLGI TRANSPORT

#### The Vesicular Model

The main problem for the VM of IGT (Palade, 1975; Rothman et al., 1980, 1984) is the large cargo aggregates that are incompatible in size with COPI vesicles that cannot be transported by COPI vesicles (**Figure 5**; see also Mironov et al., 1997). The first evidence in favor of the existence of IGT of megacargoes in dynamic experiments was obtained by Bonfanti et al. (1998). The data by Becker et al. (1995) was presented not in the original paper but in a review. Moreover, it was then shown that this experimental model was incorrect (Perasso et al., 2000). The diameter of pre-chylomicrons is greater that the diameter of the internal volume of COPI-dependent vesicles. This also does not support the VM. Transport of VLDL particles through the GC of hepatocytes was demonstrated by Taylor et al. (1997). Transport of PCI through the GC was shown by Bonfanti et al. (1998). Transport of chylomicrons through the GC was suggested by Sabesin and Frase (1977). Similarly, secretory casein submicelles, which are transported through the GC in lactating mammary glands, are larger than COPI vesicles (Clermont et al., 1993). To solve this contradiction, the Rothman group proposed that such large cargoes are transported according to the CMPM, whereas VSVG is transported by vesicles (Orci et al., 2000b; Pelham and Rothman, 2000). Then, to adapt the VM to megacargoes (**Figure 6**), megavesicles were proposed (Volchuk et al., 2000; see their Figure 5). However, our analysis demonstrated that at the level of the GC, megavesicles do not exist (Mironov et al., 2001).

It is established now that the vast majority of cargoes are absent from COPI vesicles. We previously presented a list of cargo proteins that are excluded from COPI vesicles (Mironov et al., 2005). (Martinez-Menárguez et al., 2001) showed that the concentration of amylase in COPI-dependent vesicles is lower than in Golgi cisterna. Even Orci et al. (1986), the main proponents of VM, demonstrated that the VSVG aggregation inside 52–56 nm COPI-dependent vesicles was two-thirds of that in Golgi cisterna (see Table 1 of Orci et al., 1986). To support the VM, Orci et al. (1997) proposed that pro-insulin is transported by COPI vesicles. However, careful analysis of their study reveals that in Table 3 of Orci et al. (1997), the concentration of insulin in Golgi-associated round profile was one-third of that in the Golgi cisterna. To try to solve this contradiction, they proposed that there might be two populations of COPI vesicles: for anterograde and retrograde transport. According to (Martinez-Menárguez et al., 2001), amylase and chymotrypsinogen are excluded from the cisternal rims (and also from buds), whereas the KDEL receptor is not.

Furthermore, to provide additional support for the VM, the Rothman group (Pellett et al., 2013) transfected one population of cells with fluorescently tagged cargo tagged with one fluorophore, and another population of cells with Golgi-resident proteins tagged with another fluorophore. Next, heterokaryons were generated. These cargoes and enzymes were seen in small dots in the cytosol. A small portion of these particles contained COPI. Pellett et al. (2013) measured the diameters of these spots in the cytoplasm and isolated COPI vesicles from the cytosol using

super-resolution light microscopy. Comparison of the diameters of these particles measured at the level of light microscopy with the diameter of isolated COPI vesicles, they concluded that these particles were COPI-dependent vesicles. They stated that the resolution of their super-resolution method is 80 nm. However, in reality (not in model experiments), resolution of stimulated emission depletion microscopy is about 100 nm (Sesorova et al., 2018). The real diameter of COPI vesicles is 52 nm (Marsh et al., 2001). This means that the resolution of their method is lower than the size of the structures they measured. The resolution of super-resolution microscopy depends on the refractory index of the medium, and in-vitro this parameter differs from that of cytosol. Also, the method used by Pellett et al. (2013) was very sensitive to the refractive indices of the media (Sesorova et al., 2018). Therefore, under these conditions, there could be significant systematic errors. Furthermore, they did not take into consideration that COPI-derived vesicles are on strings. This was discovered by Orci et al. (1998), and then confirmed by Marsh et al. (2001). Indeed, in mammalian cells, nobody has demonstrated coated or non-coated 52-nm vesicles at a distance of more than 200 nm (Martinez-Menarguez et al., 2001). Finally, it is known that diffusion of particles with a diameter of more than 50 nm is strongly restricted (Luby-Phelps, 1994).

Another problem with the interpretations presented by Pellett et al. (2013) is the volume-to-surface-area ratio of the transport carriers observed. According to Pellett et al. (2013), over 30 min, 25% of the membrane protein was transported from the GC of one cell to the GC of other cells. During this time, they observed 20,000 such particles. The rate of transport of soluble cargoes is identical to that of the membrane cargoes (see Figure 3 of Pellett et al., 2013). The diameter of COPI-dependent vesicles is 52 nm (Marsh et al., 2001). Their internal volume is 0.000045 µm<sup>3</sup> and the surface area is 0.074 µm<sup>2</sup> . The Golgi volume is 1,500 µm<sup>3</sup> (Mironov and Mironov, 1998). If we take into consideration that the ratio between the volume and the surface area of the GC is 140 (Ladinsky et al., 1999), the surface area of the GC would be 210,000 µm<sup>2</sup> . If these dots were COPI vesicles, 20,000 such vesicles would transport 1,480 µm<sup>2</sup> of surface area (0.7% of the total), and 0.9 µm<sup>3</sup> of Golgi volume (0.06% of the total). These considerations suggest that they observed the movement of carriers, that were much larger than COPI vesicles. Also in Figure 2f by Dunlop et al. (2017), the number of vesicles is not sufficient for IGT of soluble cargoes.

There are also other problems with the VM. There is a significant decrease in the number of COPI vesicles during synchronous IGT (Rambourg and Clermont, 1990; Rambourg et al., 1993; Fusella et al., 2013). Also in Figure 2f (by Dunlop et al., 2017), the number of vesicles is significantly lower than it is necessary for IGT of soluble cargo. There are no COPI vesicles in the microsporidia Paranosema grylli and Paranosema locustae (Beznoussenko et al., 2007), and very few in Ostreococcus tauri (Henderson et al., 2007), Plasmodium falciparum (Hohmann-Marriott et al., 2009) and Tripanosoma cruzi (see movies and Figure 2i of Girard-Dias et al., 2012). The VM cannot explain maturation of Golgi compartments in yeast in the absence of functional COPI (Matsuura-Tokita et al., 2006). Most anterograde cargoes are depleted in COPI vesicles (reviewed by Mironov et al., 2013). Albumin is depleted in COPI vesicles (Beznoussenko et al., 2014 see also Figure 7B of Dahan et al., 1994). S-Palmitoylation of anterograde cargoes at the Golgi membrane interface is an anterograde signal, and it results in the concentrating in curved regions at the Golgi rims, by simple physical chemistry (Ernst et al., 2018).

On the other hand, they did not observed vesicles on strings, which they had described earlier (Orci et al., 1998). The strings were considered as a mechanism that can prevent diffusion of COPI within the cytoplasm. The existence of megavesicles was not shown convincingly (an analysis of images of Volchuk et al., 2000, shows this). There is no COPI-like coat on the megabuds. There is no well-organized mechanism for fission.

# The Diffusion Model

There are several observations that favor the DM. To be relevant, the DM should be based on structures that are interconnected. Tubular connections between Golgi cisternae have been demonstrated by Marsh et al. (2004), Trucco et al. (2004), Beznusenko et al. (2006), and Bouchet-Marquis et al. (2008). Griffiths et al. (1994) described the bending of Golgi cisternae. Inter-cisternal connections are formed when a cargo arrives at the GC because Ca2<sup>+</sup> is liberated and leaks from the Golgi compartments and the ER (Micaroni et al., 2010a,b); this leads to the fusion of COPI vesicles enriched in Qb SNAREs with Golgi cisternae, and the restoration of the Golgi SNARE complex. These connections between the Golgi cisternae are more abundant in transporting Golgi stacks and after stimulation of cell signaling (Clermont et al., 1994; Marsh et al., 2001; Trucco et al., 2004; Mironov and Beznoussenko, 2012; Mironov et al., 2017). These connections are permeable to albumin (Beznoussenko et al., 2014) and lipids (Pagano et al., 1989; Trucco et al., 2004). Moreover, dicumarol destabilizes Golgi tubules and delays IGT (Mironov et al., 2004), whereas after activation of protein kinase A, when the cisternae of the GC become interconnected, IGT is accelerated (Mavillard et al., 2010). This suggests an important role for these connections.

Some lipids can be easily transported along the secretory pathway when the formation of vesicles is inhibited (Sleight and Pagano, 1983; Pagano and Longmuir, 1985). In living cells, spots filled with fluorescent cargoes can move through the prebleached Golgi ribbon while they gradually lose their intensity (Presley et al., 1997). Finally, Patterson et al. (2008) reported that a cargo that exits the Golgi area shows exponential kinetics. This type of kinetics indicates that all of the compartments within the GC are interconnected. Patterson et al. (2008) also proposed that large cargoes that diffuse slowly might even exit from the cis-side of the GC. However, this last explanation is not valid, because PCI always exits from the trans-side of the GC (Bonfanti et al., 1998). Another problem of this study is the following: Patterson et al. (2008) did not examine the GC that is empty before the restoration of IGT. They examined only the GC that was already filled with cargoes. However, the process of Golgi filling with cargoes might take a significant amount of time, and under such conditions the exit kinetics might be different. The main problem of the DM is the protein, lipid and ionic gradients across the Golgi stacks, the presence of SNAREs within all steps of the secretory pathway, and the concentrating of cargoes (including megacargoes) during IGT. Megacargoes cannot diffuse along narrow intercisternal connections (Beznoussenko et al., 2014). Also, the concentrating of albumin (Beznoussenko et al., 2014) and large cargo aggregates that cannot diffuse along the intermediate compartment (Claude, 1970; Sabesin and Frase, 1977; Bonfanti et al., 1998; see below) do not support the DM (Mironov and Beznoussenko, 2008, 2012; Mironov et al., 2013).

# The Cisternal Maturation–Progression Model

The cisternal maturation–progression model (Mironov et al., 1997, 1998b) poses that during IGT, each Golgi compartment undergoes maturation, by gradually transforming into the form of a more distal compartment as its resident proteins undergo recycling in COPI vesicles (**Figure 7**). At the level of the GC, the main prohibitive observation of the problems of the CMPM is the concentrating of soluble cargoes, regulated secretory cargoes and cargo aggregates during IGT (Oprins et al., 2001; Mironov and Arvan, 2008; Beznoussenko et al., 2014). We have reported the concentrating of albumin during IGT (Beznoussenko et al., 2014). Therefore, we proposed that two different mechanisms of IGT could function simultaneously; namely, one for albumin and a1-antitrypsin, and another for PCI (Beznoussenko et al., 2014). Also, Oprins et al. (2001) demonstrated the concentrating of regulatory secretion cargoes during their journey through the GC, before their precipitation within secretory granules. Further, the concentrating within megacargoes is evident from images presented in different studies.

We demonstrated the concentrating of albumin (Beznoussenko et al., 2014), but to be politically correct, we need to explain this discrepancy on the assumption that PCI and albumin use different modes of IGT. McCaughey

the cisterna during its progression is not changed, and that COPI vesicles (COPI, black dots; Golgi-resident proteins, colored dots) should be concentrated in COPI vesicles. (A) Formation of ER-to-Golgi carriers (top). (B) Delivery of ER-to-Golgi carrier to the Golgi complex. (C) Fusion of ER-to-Golgi carriers and formation of the new cis-Golgi cisterna. (D) Formation of COPI-(black dots) coated buds on the Golgi cisternae. (E) Detachment of buds and their uncoating. (F) COPI-dependent vesicles fuse with the proximal Golgi cisternae. (G–I) A new round of cis-cisterna formation, COPI-dependent budding, formation of vesicles, and their uncoating and fusion. (I) Departure of the most-trans cisterna in the form of post-Golgi carriers. (J–N) Additional rounds of similar events. (O) After step-wise departure of the post-Golgi carriers, the cis-Golgi cisterna formed after re-initiation of IGT becomes the trans-cisterna.

et al. (2019) also observed the concentrating of a specifically designed fusion protein based on procollagen in the last medial Golgi compartment (their Figure 1). Although they did not determine whether their chimeric protein behaved exactly as the natural procollagen, their observations with this cargo protein contradicts to the main prohibitive observation of the CMPM. In any case, stably transfected cells did not show any sign of accumulation of their protein within the Golgi area after the release of the RUSH-dependent procollagen from the ER. The concentrating during IGT has not only been shown for albumin (Beznoussenko et al., 2014) and the regulated secretory proteins (Oprins et al., 2001; Mironov and Arvan, 2008), but also for the megacargoes PCI (see Figure 4B of Bonfanti et al., 1998), chylomicrons and VLDL (see above).

The main prohibitive observation for the CMPM is the concentrating of cargo during intra-Golgi transport. We have shown that albumin is concentrated during IGT (Beznoussenko et al., 2014). The concentrating of PCI in cisternal distensions was shown by Bonfanti et al. (1998 see their Figure 4). The concentrating of chylomicrons in cisterna distensions at the trans-side of the GC in enterocytes was shown by Sabesin and Frase (1977). The concentrating of VLDL in cisternal distensions of the GC in hepatocytes was shown by Claude (1970). The concentrating of lipid particles in cisterna distensions at the trans-side of the GC was also demonstrated by Glaumann et al. (1975) (their Figure 7b) and Matsuura and Tashiro (1979) (their Figures 1, 9, 14). During IGT, the number of chylomicrons in one cisternal distension increased (Sabesin and Frase, 1977). This does not support the CMPM (Mironov et al., 2013). Also, in Figures 1, 6 of Dahan et al. (1994) the levels of Apo E, a marker of lipid particles, were seen to increase at the trans-side of the GC, in comparison to the cis-side. The concentrating of large cargo aggregates that cannot diffuse along intercompartmental connection does not support the CMPM, and also does not support the DM. According to the CMPM, different cargoes should not move across the Golgi stacks with different speeds in the cis-to-trans direction. However, soluble cargoes reached the trans-side of the Golgi faster than VSVG and procollagen (Beznoussenko et al., 2014). Also aggregates of cargo proteins inside cisternal distensions of Golgi cisterna moved faster than the specific cisterna domains where the opposing membranes were connected by protein bridges (Lavieu et al., 2013).

The problem of the concentrating of Golgi-resident proteins in COPI-dependent vesicles is very serious (Glick et al., 1997). Golgi-resident proteins should not be depleted in COPIdependent 52-nm vesicles, because if the concentration of these proteins in the vesicles were lower than in the Golgi cisterna, the recycling of these proteins would be very slow. Mathematical modeling based on the CMPM assumptions demonstrated that significant concentrating of the Golgi enzymes in the vesicles is necessary for the CMPM (Glick et al., 1997). Importantly, significant depletion of Golgi-resident proteins in COPI vesicles was discovered in S. cerevisiae (Beznoussenko et al., 2016). Depletion of Golgi-resident proteins in COPI vesicles (Volchuk et al., 2000; Kweon et al., 2004; Gilchrist et al., 2006; Fusella et al., 2013; Beznoussenko et al., 2016) would not support the CMPM. However, Golgi glycosylation enzymes (Velasco et al., 1993; Cosson et al., 2002; Kweon et al., 2004), nucleotide sugar transporters (Fusella et al., 2013), syntaxin 5 (Orci et al., 2000a,b; Gilchrist et al., 2006), p24, ERGIC58, TGN38/46 (Gilchrist et al., 2006) are depleted in COPI vesicles. In spite of this, several groups have reported the concentrating of some Golgiresident proteins in COPI vesicles. A single vesicle filled with the Golgi enzyme galactosyltransferase was shown in the study of Grabenbauer et al. (2005). However, the DAB reaction is not quantitative, and it is not possible to judge whether the concentration of galactosyltransferase in the COPI vesicle was higher or lower than in the corresponding cisterna. Gilchrist et al. (2006) demonstrated that the concentration of Golgi enzymes in the light membrane fraction obtained after incubation of isolated Golgi membranes with cytosol and GTP is higher than in the Golgi cisternae, whereas anterograde cargoes were depleted there. Electron microscopy has revealed that this fraction is composed of 52-nm vesicles. Gilchrist et al. (2006) concluded that COPIdependent vesicles are retrograde transport carriers for the Golgi enzymes. However, careful analysis of their data revealed that when the light fraction was prepared for electron microscopy, it was additionally pelleted onto a sucrose cushion (50% [w/w]) at 45,000 rpm. This procedure was not used for the biochemical measurement of the Golgi enzyme concentrations in this fraction. This sucrose-based centrifugation was also not used in their previous study (Lanoix et al., 1999), and the purity of these 52-nm vesicles was significantly lower. Therefore, it could be proposed that in their light fraction, there were perforated fragments of Golgi cisternae enriched in Golgi enzymes (Kweon et al., 2004), whereas after the additional centrifugation, only 52-nm vesicles remained in the samples prepared for electron microscopy. This might explain why the concentrations of the enzymes in the light fraction were higher than in the isolated Golgi membranes. However, in the same study, several other resident Golgi proteins had lower concentrations in the light fraction than in the isolated GC (Gilchrist et al., 2006).

Further, (Martinez-Menárguez et al., 2001) revealed that in situ, mannosidase II is 1.6-fold more concentrated in COPIcoated peri-Golgi round profiles. However, these data do not demonstrate that this Golgi enzyme was enriched in really separated COPI vesicles because according to the quick-freezing data of Marsh et al. (2001), the vast majority of 52-nm vesicles are not coated. Moreover, on cryosections, it is not possible to distinguish a section of a COPI vesicle from a section of a COPIcoated tube, and sections of tangential tubules can give 52-nm round profiles coated with COPI. Indeed, it was demonstrated that tangential tubules are coated with COPI (Weidman et al., 1993; Yang et al., 2011). Importantly, the vast majority of the 52 nm vesicles within the Golgi area are not coated (Marsh et al., 2001). Moreover, Cosson et al. (2002) showed depletion of this enzyme in these round profiles.

Recent observations that demonstrate that mannosidase I can be recycled by COPI vesicles are not convincing (Rizzo et al., 2013). Rizzo et al. (2013) demonstrated that after the polymerization of the chimeric mannosidase I (ManI-FM), this protein shifted from the cis-side to the trans-side of the GC. Although the monomer form of this chimera is depleted in near-Golgi round profiles, after the depolymerization, ManI-FM quickly appeared in round profiles. According to this study, 50% of these round profiles were coated with COPI. These data were interpreted in favor of the CMPM. However, Rizzo et al. (2013) did not use serial cryosections to distinguish between the round profiles as projections of cross-sections of tubules and the projections of actual vesicles. On random cryosections, this distinction is not possible (Kweon et al., 2004). Moreover, on cryosections, a tubular network can appear as round profiles, which were considered in this study as COPI-dependent vesicles. Also, Rizzo et al. (2013) did not perform the obvious control experiment based on inhibition of the formation of vesicles by COPI (i.e., microinjection of cells with an antibody against

ßCOP). Also, their tomography data presented in favor of the augmentation of the number of COPI-dependent vesicles after the depolymerization are not convincing.

Indeed, they stated, "All reconstructions indicated that the morphology of the carriers and the structure and size of the Golgi stack were similar under all experimental conditions. Moreover, tomography confirmed that most round, 50- to 80-nm structures were indeed vesicles, and that the relative frequency of vesicles and tubules was similar to that seen in thin sections (not depicted)." To demonstrate that round profiles represented separated vesicles, they showed very small sized serial electron microscopy tomography images of only one round profile. However, this round profile showed a visible neck that connected it with a Golgi cisterna. This neck is visible on frames 45–55 of Rizzo et al. (2013). If we take into consideration that the thickness of their tomography slice was 3 nm and the resolution of the presented the images is 10 nm, the obvious conclusion is that this neck represents a membranous structure, and that this round profile actually represents a COPIcoated bud. The statement that 50% of the round profiles were coated with COPI also favors our explanation because a vast majority of free vesicles near the GC are not coated (Marsh et al., 2001). No electron microscopy tomography images that showed the effects of the ManI-FM depolymerization were presented (see the analysis of the paper by Rizzo et al., 2013, in the Supplementary Materials). Thus, these data that show the concentrating of the Golgi-resident proteins in COPI vesicle are not convincing.

During synchronous IGT, the number of COPI vesicles should be sufficient for the recycling of all of the resident proteins. However, when a large amount of cargo moves across the GC, even if we take into consideration the maximal possible speed of COPI vesicle formation (Mironov et al., 2001; Trucco et al., 2004; Fusella et al., 2013), the number of COPI vesicles is less than one-tenth of that necessary for the transport of membranes across the GC. Thus, even when COPI vesicles were generated at maximal speed, the rate of their generation can support only 10% of the vesicles necessary for IGT (Fusella et al., 2013).

Sialyltransferases and fucosyltransferases are present within the trans-most cisterna (TMC). However, there are no buds coated with COPI on the TMC (Ladinsky et al., 1999; Marsh et al., 2001; Mironov et al., 2017). In principle, it is possible that clathrin-dependent vesicles can function as retrograde carriers. Indeed, Velasco et al. (1993) observed labeling of mannosidase II with DAB in clathrin-coated buds within the GC. However, DAB labeling is not quantitative. Moreover, Rothman et al. (1980) reported that the clathrin-dependent vesicles isolated from the GC contained not Golgi-resident proteins, but the cargoes. Ladinsky et al. (1999) and Grabenbauer et al. (2005) showed that within the Golgi area there were only eight clathrin-coated vesicles. This is the diameter of COPI-dependent vesicles. It is important to underline that the diameter of COPI-dependent vesicles is very uniform (Marsh et al., 2001). If we take into consideration that the surface area of these Golgi clathrindependent vesicles is only 1.3-fold higher than that of COPIdependent vesicles, the obvious suggestion is that the number of clathrin-dependent vesicle is about one twentieth of the number necessary for synchronous recycling of the Golgiresident proteins.

Moreover, although there have been several statements that COPI-coated buds can be found within the TMC, or even the TGN, in reality this has not been completely established. There was no convincing evidence in the images that (Martínez-Menárguez et al., 1999) provided that these structures were really TMC/TGN. For instance, in Figure 5B of (Martinez-Menárguez et al., 2001), the coated bud labeled for ßCOP is localized at the trans-side of the Golgi. However, there is no clear membrane continuity of this bud with the TMC or with structures of the TGN. Thus, this image might be the section of a COPI-coated bud of the medial cisterna, and this might be an effect of the section plane. Its diameter is 60 nm. This diameter is too high for COPI vesicles. Figure 5C of (Martínez-Menárguez et al., 1999) shows a membrane bud with a protein coat on the immature secretory granule within the GC of an acinar epithelial cell from the rat pancreas. However, this granule is immature and might represent a distension of the last medial cisterna. Examples of such distensions of the medial Golgi cisternae with no completely dense content can be easily found in images presented on the website "nanotomy." Moreover, we also detected COPI-coated buds on the cargo domains, which appeared on the distension of the TMC, but only during cargo synchronization according to the maxi-wave protocol, when a large amount of cargo moves simultaneously through the GC. Thus, everything should be considered in terms of probability, and the probability of finding COPI-coated buds on the TMC is relatively low. As such, it is not clear how the resident proteins undergo recycling from the TMC. The variability of the diameters of the COPI vesicles described by Martínez-Menárguez et al. (1996) might be a result of chemical fixation or of the methods of measurement. At least in our hands, when we inhibited SNAREs and most of the vesicles formed within the Golgi area derived from the activity of the COPI machine, their diameters were extremely uniform (Kweon et al., 2004; Fusella et al., 2013).

The third prohibitive observation for the CMPM is the situation when renovation and progression of Golgi cisternae can be blocked. However, under these conditions IGT was observed although it became slower (Dunlop et al., 2017). Also when the opposite membranes of the megacargo domain are connected by protein bridges, its IGT transport was inhibited (Lavieu et al., 2013). Moreover, "the land-locking" of Golgi cisternae does not exclude the possibility that the KARM might also explain these data. Indeed, Golgi cisternae visible within the mitochondria aggregates were relatively close to each other, and could be temporally connected by tubules.

The CMPM has several other problems. The full list of the CMPM problems was presented in our previous review (Mironov et al., 2013). In microsporidia, there are no COPI vesicles for the recycling of resident Golgi proteins (Beznoussenko et al., 2007), although IGT takes place. Also, the study of Patterson et al. (2008) does not support the CMPM, because within the framework of the CMPM, the exit of PCI-GFP from the Golgi zone after bleaching of the whole cell less the Golgi area should not be negatively exponential. This should be composed of two parts to the regression line. The first part should be horizontal, and the second part should be as for linear decay. Another explanation that they presented is the proposal that PCI dots can exit from the GC immediately after their arrival at the cisside of the Golgi stack, without their progression across the stack. Their third explanation suggests that megacargoes might diffuse along the lumen of the united membrane system of the Golgi stack. Finally, the CMPM cannot explain the observation that shows that overexpression of the GDP-mannose transporter in the yeast S. cerevisiae induces the formation of stacked Golgi (Hashimoto et al., 2002). It is not clear how the IGT is organized under these conditions, because the main adaptation of the CMPM for S. cerevisiae is that due to the spatial separation of the different Golgi compartments, there should be a mechanism for the directionality of the delivery of COPI-dependent vesicles. On the other hand, Rambourg et al. (1993) observed that in sec7 mutants maintained at 37◦C in low (0.1%) glucose medium, secretion granules progressively decreased in number, and soon disappeared. Concomitant to this, the networks of Golgi tubules increased in size and complexity, lost their distensions, and then transformed into flattened saccules that formed stacks of up to seven or eight saccules that were similar to the Golgi stacks seen in mammalian cells. Indeed, if we remember that in S. cerevisiae, the different Golgi compartments should have contact with each other to fulfill the transport of cargo from one compartment to another.

#### The Kiss-and-Run Model

To be efficient, the KARM of IGT should be based on several prerequisites (**Figure 8**; see above). Some of them were already observed. Indeed, cisternal distensions containing megacargoes already represent cargo domains. The VSVG domains that do not exchange this cargo with each other were described by us (Mironov et al., 2001). We demonstrated how the KARM explains IGT of soluble cargoes and Golgi glycosylation enzymes (Mironov and Beznoussenko, 2012; Mironov et al., 2013). On cryosections, cisternal pores are not usually particularly visible. However, even in Figure 1 of Bonfanti et al. (1998) the pores were visible. In Figure 2 of Bonfanti et al. (1998), the pores are not particularly visible due to the diffusion of the DAB. The excessive diffusion of the DAB precipitate hid these pores. In Figures 6E,F of Mironov et al. (2017), pores can be seen between the PCI-containing distensions and the rest of the cisternae. In Figure 3 of Mironov and Pavelka (2008), pores are visible as well. Cisternal distensions in the GC of acinar pancreatic cells are separated from the rest of cisternae by rows of pores. This was visible in the studies of Claude (1970), Sabesin and Frase (1977), and Sesso et al. (1994). Initially, albumin is present in cisternal distensions that are filled with VLDL (Figure 6e,f of Dahan et al., 1994). However, these pores appear to be very important for IGT (**Figure 8**).

Pores that separate cisternal distensions from the rest of the Golgi cisternae were shown by Claude (1970) in hepatocytes; by Sabesin and Frase (1977) and by Pavelka and Roth (2005) (their Figure 101A, page 205) in enterocytes, where the GC transports chylomicrons; by Sesso et al. (1994) (their Figures 5, 7a–d, 9, 10) in acinar pancreatic cells; by Ladinsky et al. (2002) after the 20◦C temperature block; by Mironov et al. (2001) (their Figures 4C,E,F) in fibroblasts transporting PCI aggregates; by Pavelka and Roth (2005) (their Figure 98B, p. 199) in hepatocytes during IGT of VLDL. In Figures 6E,F of Mironov et al. (2017), which shows the GC in enterocytes, the pores can be seen between the cisternal distensions filled with chylomicrons and the other parts of the Golgi cisterna.

Thus, restoration of pores in the cisternal rims might be based on this mechanism. The KARM gives the following predictions: (1) if pores inside cisternae are consumed, there should be the need for resting of the Golgi stack; (2) recycling of Ykt6 is improbable. As such, there should be one use of this SNARE, and after consumption of the cytosolic pool of Ykt6, there should be the need for the resting of the whole Golgi complex in the entire cell. Thus, there could be several waves of cargo, and only then would the pores be consumed. Fission and then fusion of COPI vesicles might induce the formation of tubules and restoration of the number of pores along cisternal rims (Park et al., 2015).

Careful analysis of images presented by Ladinsky et al. (2002) and Taylor et al. (1997) revealed that after 2 h of the 20Â◦C temperature block in the presence of inhibitor of protein synthesis, when several waves of cargo protein (VSVG) passed through the GC, the numeric density of pores in Golgi cisternae decreased (Figures 3C–F, 4 by Ladinsky et al., 2002). In contrast, after prolonged inhibition of IGT, this density increased (see Figure 3 by Taylor et al., 1997).

When membranes are transported through the GC, the asymmetric variant of the KARM should be used. According to this, to increase the efficiency of IGT, there should be cargo domains where a cargo is concentrated. These domains should contain a set of SNAREs complementary to those in COPI vesicles (GS27, GOS28) (Fusella et al., 2013), and be somehow separated from the rest of the Golgi cisterna, to facilitate the fission process. Finally, during IGT, all of the cargoes, including large cargo aggregates, should undergo concentrating at the trans-side of the GC. In contrast, according to the CMPM, these first three demands are not necessary, whereas, the fourth is forbidden. Good examples of such cargo domains might be: PCI-containing distensions of Golgi cisternae in collagen-secreting cells (i.e., fibroblasts); chylomicron-containing distensions formed during transcytosis of lipids through enterocytes; and cisternal distensions filled with VLDL particles in hepatocytes. The presence of a row of pores behind the cargo domain during IGT and the concentrating of SNAREs over the cargo domain favor the KARM. Our observation that there is no cargo diffusion between VSVG-GFP domains formed during IGT (Mironov et al., 2001) is in favor of the KARM.

# DISCUSSION

Thus, at the level of ER-Golgi and IGT the VM faces with the problem of the transport of megacargoes. Attempts to modify the VM by addition of so called megabuds and megavesicles were not convincing. We suggest that the megavesicles observed by Gorur et al. (2017) and the large procollagen-positive immobile dots observed by Omari et al. (2018) and McCaughey et al.

(2019) (their Figure 2) represent ER-derived autophagosomes (reticulophagosomes; Fregno and Molinari, 2018; Fregno et al., 2018; Omari et al., 2018; Forrester et al., 2019). Autophagosomes derived from the protrusions of the ER filled with protein aggregates were first described by Omari et al. (2018), (Fregno and Molinari, 2018 see also Fregno and Molinari, 2018; Forrester et al., 2019). In a study of Fregno et al. (2018), this phenomenon was observed upon overexpression of the secretory heavy chain of immunoglobulin M that lacked some domains; the aggregates of this chain were concentrated in ER protrusions with the diameter of ≥450 nm. These protrusions were not coated with a COPII-like coat. After detachment from the ER, these distensions

were delivered to the GC, and then were secreted or fused with lysosomes (see SM).

CMPM has problems at both steps of intracellular transport, namely, at the level of the exit from the ER, it cannot explain the problem of different concentration of different cargoes whereas at the Golgi level, it cannot explain concentration of megacagoes during IGT. Although when we faced similar discrepancy between the cis-to-trans delivery of albumin and PC we tried to combine different models for different cargoes and used the DM for the explanation of this delivery. However, the DM cannot explain the augmentation of albumin concentration at the trans side of the GC. The second main problem of CMPM for IGT is the depletion of several resident proteins in the so called retrograde COPI vesicles. Moreover, the recycling at the level of the trans-most cisterna and the TGN is not possible to explain because COPI vesicles which are considered to be retrograde transport carriers are not formed. The attempt to propose that clathrin-dependent vesicles could execute the recycling of the resident proteins is not successful also due to the absence (or extreme rarity) of COPI-coated buds on the trans-most cisterna (see details in the **Supplementary Materials**).

The DM cannot explain the necessity of SNAREs for intracellular transport and concentration of cargo at different level of the transport. Thus, the VM, DM, and CMPM cannot overcome their prohibitive observations. Also the VM, DM, and CMPM cannot explain the mechanisms of Golgi ribbon formation and the disappearance of the GC in S. cerevisiae, the fragmentation of the Golgi ribbon In contrast, the KARM can do this. The live-cell imaging of RUSH-controlled cargoes shows that different cargoes have different kinetics (Boncompain et al., 2012). One possibility is that they follow different trafficking mechanisms; another one is that rates of different cargo concentration are different. The studies usually considered as the corner stones of the VM, DM, and CMPM could also be easily explained from the point of view of the KARM. Although we have shown here that now the KARM appears to be the most powerful model of IGT, it still has some difficulties. For instance, one of these is the existence of separate and different Golgi compartments in S. cerevisiae. The observation that different Golgi compartments are rarely connected by tubules might provide the explanation, and thus the requirements of the KARM (Beznoussenko et al., 2016). Moreover, Kurokawa et al. (2019) demonstrated that Golgi-resident proteins and a cargo can form two domains within the same Golgi compartment. In S. cerevisiae, the different Golgi compartments are separated. If we assume that in S. cerevisiae IGT occurs according to the CMPM, the mechanism of the vectorial delivery of retrograde COPI vesicles should exist. For instance, in mammalian cells, COPI vesicles are on strings. This prevents the diffusion of COPI vesicles around the GC, and might explain the vectoriality of vesicle movement (Orci et al., 1998). However, proponents of the CMPM have not presented any analysis of this problem in S. cerevisiae. When this review was already submitted, two important studies appeared (Casler et al., 2019; Kurokawa et al., 2019). In both of these, visualization of maturation of the cargo domain was performed in living S. cerevisiae cells. The authors claim that their studies demonstrate the cisterna maturation model. However, their data fit even better to the KARM (**Figure 9**).

The vesicle delivery is a very important problem for both VM and CMPM and especially for Saccharomyces cerevisiae. The diffusion of vesicles through the dense cytosol is very slow, because most of these vesicular carriers have a diameter >50 nm (Luby-Phelps, 1994). In order to solve this, the idea of "vesicles on a string" was proposed (Orci et al., 1998). On the other hand, different Golgi compartments could be getting closer. When these compartments are close by, a burst of COPI vesicle can be formed to provide the transfer of a significant amount of protein from one compartment to another. However, in this case the number of COPI-coated buds on the Golgi compartments should increase. However, this number is much lower than in mammalian cells (Beznoussenko et al., 2016).

The CMPM cannot explain the observation that demonstrates that in S. cerevisiae when cargo exit from the ER is blocked, the GC disappeared (Ayscough and Warren, 1994; Morin-Ganet et al., 2000). The KARM can do this (**Figure 10**). Also, the CMPM cannot explain why after elimination of COPI vesicles using the temperature sensitive mutant of one of the COPI subunits (Matsuura-Tokita et al., 2006), compartment maturation continues, although it becomes slower. In contrast, the KARM posing that cisternal pores are important and are consumed during IGT explains this phenomenon easily. Also we presented the explanation of the corner-stone experiments, which are usually considered in favor of the VM and CMPM. This explanation fits to the KARM.

The important issue, which the KARM should explain, is the role of COPI and COPI-dependent vesicles. Within the framework of the KARM, COPI vesicles are important for: (1) elimination of excessive membrane curvature (Beznoussenko et al., 2015); (2) extraction of Qb SNAREs and slowing down of IGT (Trucco et al., 2004; Fusella et al., 2013; Beznoussenko et al., 2016); and (3) initial retention of Golgi enzymes (Dominguez et al., 1998; Nilsson et al., 2009). Park et al. (2015) showed that COPI can sort anterograde cargoes into COPI-dependent tubules. Importantly, tubulation of the GC accelerates EGT and IGT (Mavillard et al., 2010; Capaci et al., 2019). Membrane coats are necessary for the concentrating of SNAREs (Zeuschner et al., 2006; Pryor et al., 2008; Koo et al., 2011; Fusella et al., 2013). However, it is necessary to test whether the transport of cargoes would be stopped when inhibition of the specific set of SNAREs localized at defined steps of intracellular transport would stop the transport at this specific point. The discovery of such mechanisms, or their absence, might confirm or reject the KARM. The additional role of COPI-dependent vesicles could be their involvement in the uncoating of Golgi membranes from COPI coat.

There are two transport steps where this scheme might be not very obvious; namely, EGT and post-Golgi transport. There, according to the KARM, the tubule from the GC has to move toward the ER exit site (ERES), and this should induce a boluslike delivery of ER-to-Golgi carriers. At the post-Golgi stage, the tubule or endosome per se has to move toward the centrally localized GC. After the fission, the cargo domain should move centrifugally according to the bolus-like mechanism and thin

FIGURE 9 | Scheme of intra-Golgi transport in *S. cerevisiae* within the framework of the KARM. This scheme ensures maturation and progression of cargo domains. According to the main postulate of the KARM, to become separated, it is necessary to have initial fusion. Thus, fusion is the first event, and then fission (at another site) is the second event. The ER is magenta; artificial cargo is blue and indicated with the letter "C"; the cis-Golgi compartment is red; the trans Golgi compartment is green; the post-Golgi compartment is brownish; the exited cargo domain is surrounded with a red line. The situation that is shown here was formed after several rounds of cargo domain detachment from ERES, when most of the Golgi compartments already contain the artificial cargo. (A) The initial situation when most of the Golgi compartments contain cargo domains (blue). (B) Fusion of the cis-Golgi compartment (red) with ERES, and the trans-Golgi compartment (green) with the post-Golgi compartment (brownish). (C) Shift of the COPII coat (yellow) from the cargo domain (blue oval surrounded by a magenta ring). There is fission of the connections between the trans compartment (green) and the cargo domain, which is consumed by the post-Golgi compartment. (D) Detachment of the cis-Golgi compartment connected to two cargo domains from ERES (yellow). (E) Rotation of the cis-Golgi (for the sake of clarity). (F) Fusion between the trans-Golgi compartment and the cis-Golgi compartment connected to cargo domains. (G) Detachment after fission of the cis-Golgi compartment from the trans-Golgi compartment connected to cargo domains. (H) Fusion of the cis-Golgi compartment with ERES, where a new cargo domain covered with Sec31 is prepared. (I) Shift of the Sec31-coat (yellow) from the cargo. (J) Detachment of the cis-Golgi compartment connected with the new cargo domain from ERES, and fusion of the trans-Golgi part connected to cargo domains with the post-Golgi. (K) Detachment of the trans-Golgi compartment from the cargo domains. (L) The final stage when the cargo domain of interest is within the post-Golgi compartment. (Ja–c) The alternative consequence of the final events; namely, the cargo domains connected with the trans-Golgi compartment can be delivered to the post-Golgi one after another.

tubules behind the cargo domain could be observed (Polishchuk et al., 2003). When compartments are separated by significant space the distal compartment moves toward the proximal one, fuses with this cargo domain and traps the cargo domain, exactly as was shown by (Casler et al., 2019 see below). The tubules formed by COPI could fuse with the cargo domain localized within the proximal Golgi domain, as was shown by (Trucco et al., 2004 see Figures 3i–n there). The concentrating of SNAREs at the future fusion sites increases the efficiency of transport, which eliminates stochastic events. Also it is not necessary to have separate retrograde carriers of separate retrograde pathways. During the kiss-and-run process there could be a simultaneous process of anterograde and retrograde exchange between the two compartments. Thus, the KARM has a significant potential for the role of a new paradigm within the transport field. Nevertheless, additional analysis of this issue is necessary.

Thus, the KARM give the following predictions. (1) The cargo should be organized in the domain more or less clearly separated from the domains, which are formed by Golgi-resident

#### REFERENCES


proteins. (2) Initially, there should be lateral contact between the cargo domain and the Golgi domain. Then the cargo has to go together with the Golgi compartment. (3) After its arrival at ERES, the cis-compartment displaces Sec31 from the cargo domain. (4) After arrival of the cis-compartment at ERES, the size of the cargo domain should increase (**C**). (5) The transient phase of maturation when domains of two different Golgi compartments are situated within the same Golgi cisterna should be shorter than other (co-localization) phases when only one Golgi compartments is present within the same Golgi cisternae (This explains why in yeast, intermediate forms of transport carriers are rare). (6) The cargo should be concentrated into two cargo domains. However, if other cargoes (which are invisible to the observer) are transported simultaneously, this multiplication should be slightly lower than 2-fold. (7) During the transitional phase, the initial co-localization would be between cargo and trans and not between cis and trans markers. (8) The cargodomain and the Golgi domain should be in the same membrane disk with pores, but within their own domains. The cargo domains should be separated from the rest of the Golgi cisterna by pores.

#### AUTHOR CONTRIBUTIONS

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

#### FUNDING

FIRC's support of laboratory, INTAS (Project 99-4-1732), Telethon (E.1105), the Italian National Research Council (Convenzione CNR – Consorzio Mario Negri Sud).

#### ACKNOWLEDGMENTS

We thank Drs. C. Berrie, C. Wilson and M. Kreft for discussion, critical suggestions and editing of the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2019. 00146/full#supplementary-material


Bannykh, S. I., Rowe, T., and Balch, W. E. (1996). The organization of endoplasmic reticulum export complexes. J. Cell. Biol. 135, 19–35. doi: 10.1083/jcb.135.1.19


pathway for procollagen VII export from the endoplasmic reticulum. Elife 3:e02784. doi: 10.7554/eLife.02784


**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 Mironov and Beznoussenko. 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.

# The N-Glycosylation Processing Potential of the Mammalian Golgi Apparatus

#### Peter Fisher<sup>1</sup> , Jane Thomas-Oates<sup>2</sup> \*, A. Jamie Wood1,3 \* and Daniel Ungar<sup>1</sup> \*

<sup>1</sup> Department of Biology, University of York, York, United Kingdom, <sup>2</sup> Department of Chemistry and Centre of Excellence in Mass Spectrometry, University of York, York, United Kingdom, <sup>3</sup> Department of Mathematics, University of York, York, United Kingdom

#### Edited by:

Kristian Prydz, University of Oslo, Norway

#### Reviewed by:

Heike Folsch, Northwestern University, United States Johannes Stadlmann, Institute of Molecular Biotechnology (OAW), Austria

#### \*Correspondence:

Jane Thomas-Oates jane.thomas-oates@york.ac.uk A. Jamie Wood jamie.wood@york.ac.uk Daniel Ungar dani.ungar@york.ac.uk

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 28 March 2019 Accepted: 26 July 2019 Published: 13 August 2019

#### Citation:

Fisher P, Thomas-Oates J, Wood AJ and Ungar D (2019) The N-Glycosylation Processing Potential of the Mammalian Golgi Apparatus. Front. Cell Dev. Biol. 7:157. doi: 10.3389/fcell.2019.00157 Heterogeneity is an inherent feature of the glycosylation process. Mammalian cells often produce a variety of glycan structures on separate molecules of the same protein, known as glycoforms. This heterogeneity is not random but is controlled by the organization of the glycosylation machinery in the Golgi cisternae. In this work, we use a computational model of the N-glycosylation process to probe how the organization of the glycosylation machinery into different cisternae drives N-glycan biosynthesis toward differing degrees of heterogeneity. Using this model, we demonstrate the N-glycosylation potential and limits of the mammalian Golgi apparatus, for example how the number of cisternae limits the goal of achieving near homogeneity for N-glycans. The production of specific glycoforms guided by this computational study could pave the way for "glycoform engineering," which will find uses in the functional investigation of glycans, the modulation of glycan-mediated physiological functions, and in biotechnology.

Keywords: computational modeling, Golgi apparatus, glycan biosynthesis, cisternal number, glycan heterogeneity

# INTRODUCTION

N-glycosylation is a process initiated in the endoplasmic reticulum (ER) but specific structural elements such as core fucosylation and branching (**Figure 1A**) are introduced later in the secretory pathway in the Golgi apparatus. N-glycans are modified by a series of sequentially acting glycosidases and glycosyltransferases (**Figure 1B**) that modify glycans in the Golgi apparatus and consequently dictate and ultimately determine the glycan profile of the whole cell. However, the structural modification of N-linked glycans is a complex process that results in numerous different glycan structures. In the absence of a "glycan template," protein glycosylation is inherently heterogenous with a number of factors contributing to the final glycan structure. These variables include the protein structure (Hang et al., 2015; Suga et al., 2018), secretory protein load (Jimenez-Mallebrera et al., 2009), Golgi transport mechanism (Hossler et al., 2007), enzyme protein levels, availability of monosaccharide-nucleotides, and the organization of glycosylation enzymes within the Golgi apparatus (Oka et al., 2004; Zolov and Lupashin, 2005; Fisher and Ungar, 2016). Typically, N-glycan biosynthesis can be characterized as a series of divergent pathways that converge into structural nodes (**Figure 1B**). Following the initial trimming of mannoses by ManI, the number of possible N-glycan structures generated in the Golgi apparatus increases exponentially with each additional monosaccharide until the capping of antennae with sialic acid (Spahn et al., 2016).

N-glycans can play important roles in dictating the properties of glycoproteins. From a biologic's standpoint, controlling the glycans residing on therapeutic antibodies can tune their pharmaceutical properties. For example, glycans lacking core fucose (**Figure 1A**) can increase antibody-dependent cellular cytotoxicity (ADCC) (Shinkawa et al., 2003; Yamane-Ohnuki et al., 2004). Overall it appears that fucosylation and galactosylation are the dominant features of N-glycans that influence Fc-receptor binding (Dekkers et al., 2017), which is linked to ADCC. Furthermore, the presence of sialylated complex type glycans increases the in vivo halflife of therapeutic antibodies compared to oligomannose type glycans (Kanda et al., 2007) further illustrating the attenuating role of N-glycans in IgG properties. Due to the heterogenous nature of glycan biosynthesis the production of homogenous biologics with respect to glycosylation is currently not feasible, giving rise to batch-to-batch variation, and heterogeneity within batches. Glyco-engineering of biologics, for example through the deletion of Mgat1 to eliminate complex and hybrid glycans from CHO cells, reduces glycan heterogeneity, and improves immunological properties (Xu et al., 2016). However, eliminating complex glycans also removes any ability to fine-tune the properties of a therapeutic through presence/absence of other N-glycan features such as those shown in **Figure 1A**. Therefore, devising novel strategies to reduce glycan heterogeneity and/or enrich the relative abundance of a desired glycan is of great value to the pharmaceutical industry.

Glycosylation enzyme levels play a pivotal role in glycan biosynthesis; however, the outcome of enzyme depletion or overexpression can be unexpected. Knockdown of the galactosyltransferase GalT4 alongside overexpression of the branching enzymes Mgat4 and/or 5 in CHO cells substantially increased the abundance of tri- and tetra-antennary N-glycans. Interestingly though, it was the reduction of GalT4 that primarily accounted for the increase in branching (McDonald et al., 2014). Two additional important factors in determining the N-glycans produced by mammalian cells are the distribution of the glycosylation enzymes within the Golgi apparatus (Fisher and Ungar, 2016), and the architecture of the Golgi apparatus itself. For example, disruptions to the conserved oligomeric Golgi (COG) complex, which is involved in tethering and thereby targeting intra-Golgi retrograde vesicles, can lead to glycosylation defects in model cell lines (Shestakova et al., 2006; Bailey Blackburn et al., 2016) and to congenital disorders of glycosylation (reviewed in Wu et al., 2004; Steet and Kornfeld, 2006; Hennet and Cabalzar, 2015). Furthermore, the depletion of the Golgi reassembly stacking proteins (GRASPs) of 55 and 65 kDa results in an acceleration of protein transport, Golgi fragmentation, and impaired glycosylation (Xiang et al., 2013; Bekier et al., 2017). Fragmentation of the Golgi apparatus has also been linked to numerous neurodegenerative disorders (reviewed in Joshi et al., 2015). Alterations to the architecture and the organization of the Golgi apparatus provide a potential route to controlling glycosylation alongside the protein levels of the glycosylation machinery themselves. Other factors such as the availability of the monosaccharide-nucleotide donors are important and worthy of further investigation but are not considered in this work.

In this work we build on previous studies of glycosylation in WT HEK293T cells (Bailey Blackburn et al., 2016) and a stochastic model of N-glycosylation we recently developed and validated (Fisher et al., 2019). We used this model of glycosylation to demonstrate in silico the effects of cisternal number on glycan heterogeneity and complexity. This led us to test computationally what limitations the existing Golgi apparatus architecture places on the degree of glycan homogeneity achievable in mammalian cells. Finally, we used our model to predict strategies to increase the relative abundance of targeted glycan structures.

### RESULTS AND DISCUSSION

It is not the aim of this work to describe in great detail the modeling methodology that has been used (for a more detailed description on the development and validation of the modeling framework used in this work please see Fisher et al., 2019), however, a brief summary of the modeling methodology will give context and a greater understanding of the results of this work. Glycosylation reactions are simulated using a stochastic simulation algorithm (SSA). The SSA incorporates the inherent noise that is present in biological systems and that becomes increasingly relevant when the number of reactants (and enzymes) are small, as is the case for glycosylation in the Golgi. In contrast to models of glycosylation based on ordinary differential equations (ODEs) (Hossler et al., 2007; Krambeck et al., 2009, 2017; Goey et al., 2018), several factors can be included in each parameter, reducing the need for excessive parameterization in our model. As such, we define parameters for each enzyme by using the term "effective enzymatic rate" to encompass the enzyme's protein level, availability of its nucleotidemonosaccharide substrates and the chemical enzymatic rate. In this work we do not wish to consider changes to intrinsic enzymatic properties such as the inherent chemical enzymatic rate constants, therefore any predicted alterations to the effective enzymatic rate parameter are assumed to be equivalent to alterations in the protein level of the respective enzyme.

The effective enzymatic rates, the transit time in the Golgi apparatus, and the composition of glycans entering the Golgi were the parameters that were altered to fit the computed to a target glycan profile. This fitting was done using an approximate Bayesian computation (ABC) algorithm (Marjoram et al., 2003), which samples from parameter distributions to feed the SSA and assess the goodness of fit with a target profile. The shift of the parameter distributions during fitting tells us how the organization (e.g., levels and localizations) of enzymes changes from the starting condition to reach the target glycan profile. We utilize the optimized parameters determined previously for the WT HEK293T cell glycan profile as initial values (Fisher et al., 2019), therefore predicted changes in effective enzymatic rates are

relative to the organization of the glycosylation machinery in WT HEK293T cells.

#### Glycan Heterogeneity and the Golgi Apparatus

We set out to evaluate how the level of glycan heterogeneity is affected by different variables in our model of N-glycosylation. An important variable is the number of cisternal elements, which is known to vary between species, cell types (Mironov et al., 2017), and in pathologies (Joshi et al., 2014). In our earlier work four cisternae were found to be required to optimally model HEK293T cell N-glycans (Fisher et al., 2019). While this number does not necessarily reflect the actual number of cisternae in this cell type, deviations from this number during modeling reflect potential changes in cisternal number. In order to investigate the effect of architecture on the glycan processing potential of the Golgi apparatus, the number of cisternal elements in our stochastic model of the Golgi was varied from the four required for HEK293T cells in previous work (Fisher et al., 2019). The total effective rate for each enzyme was kept constant, as was the total time spent in the Golgi apparatus. As a measure of glycan heterogeneity in these simulations, the number of different glycan structures produced in the simulation was recorded (**Figure 2A**). The trend is of decreasing heterogeneity as the number of cisternal elements is increased from two to ten. In addition to the observed decrease in heterogeneity correlating with increasing cisternal elements, the relative abundance of oligomannose glycans also increases as the trimming of mannose residues becomes less efficient possibly due to the reduced time per cisterna. The reduction in glycan maturity observed in GRASP55/65 double knockout (KO) cells may therefore be a consequence of the fragmented Golgi apparatus (Bekier et al., 2017). Interestingly, by changing the number of cisternal elements the most abundant complex glycans simulated by our model also differed (**Figure 2C**). For example, and in support of the hypothesis that the organization of a fragmented Golgi apparatus may explain the reduction in glycan maturity in GRASP55/65 KO cells, simulations with more than four cisternal elements generated immature complex glycans in comparison to simulations involving one and two cisternal elements, which generated higher relative abundances of more elaborate glycans containing more antennae, and a higher proportion of sialic acids (**Figure 2C**). This is possibly a knock-on effect of the reduced efficiency of oligomannose processing as mentioned above, and could possibly be explained with the shorter processing time in each individual cisterna.

In the previous simulation, the total time taken to traverse the Golgi apparatus was kept constant while the number of cisternal elements was altered. Next, we were interested in what effect varying the transit time per cisterna had on glycan heterogeneity and complexity while keeping the number of cisternae at four. Indeed, altering the time taken to traverse the Golgi apparatus had a different effect than varying the number of cisternal elements (**Figure 2B**). The number of different glycan structures predicted when the time was reduced to 10% of the WT time decreased dramatically (**Figure 2B**). Following a gradual increase and plateauing of glycan heterogeneity with increasing time in the Golgi apparatus, the number of glycan structures begins to decrease again at 250% of WT time. This is due to glycan pathways converging

on structural nodes with terminating sialic acids. For example, the abundance of the fully sialylated bi-antennary glycan, Fuc1GlcNAc4Man3Gal2NeuAc2, increases from 4.5% to close to 10% of the total when the transit time is increased 3.3 fold of that of WT. In contrast, this same glycan is absent from the simulated profile when the transit time is reduced 10-fold compared to WT.

Interestingly, GRASP55/65 depletion accelerates protein trafficking through the Golgi apparatus (Xiang et al., 2013; Bekier et al., 2017), an effect that our model would predict to decrease the relative abundance of complex N-glycans (**Figure 2B**). Indeed, published experimental evidence shows less processing for both cell surface and intracellular glycans in these cells (Bekier et al., 2017). In contrast, there was a minor kinetic delay for the delivery of VSV-G to the plasma membrane in Cog3 knockdown HeLa cells (Zolov and Lupashin, 2005), indicating a reduced rate of anterograde transport. This is unlikely to be the direct cause of the glycosylation defect though, as Cog3 depletion results in relocation and partial degradation of Golgi glycosylation enzymes (Shestakova et al., 2006), as opposed to GRASP55/65 depletion that had no effect on the localization and levels of the glycosylation enzymes studied (Xiang et al., 2013). Furthermore, our model suggests slower transport through the Golgi apparatus would reduce the oligomannose content, a feature that is not observed in COG subunit depleted cell lines (Bailey Blackburn et al., 2016).

# Maximizing the Relative Abundance of Target Glycans

The production of single glycoforms is an important goal in the pharmaceutical industry. We therefore used our modeling framework to assess the glycan processing potential of the Golgi apparatus. In this instance, we tested the potential of the Golgi apparatus to produce a single glycoform, within reasonable biological boundaries. Three complex glycans with partial or complete sialylation and two or three antennae (**Table 1**) were chosen to test this property of the Golgi apparatus. The three complex glycans are referred to as bi-Sia1, bi-Sia2, and tri-Sia<sup>1</sup> when they are described as the target for glycan engineering. When the same structures appear in the simulation as byproducts though, they are referred to in conventional glycan notation. Starting with the fitted parameters for the effective

#### TABLE 1 | Target glycans.

fcell-07-00157 August 9, 2019 Time: 16:32 # 5


enzymatic rates for WT HEK293T cells, we used ABC fitting to maximize the relative abundance of each target glycan. For this purpose, a hypothetical glycan profile was fitted in which 100% of the complex glycan was set as the desired N-glycan. The parameters that were fitted for these experiments were: effective enzymatic rates, transit time per cisterna and composition of glycans entering the Golgi. In addition, we investigated the maximization of target glycans in two scenarios: One, in which the distribution of enzymes between Golgi cisternae can be varied (variable localization); and another one, in which the distribution of enzymes in the Golgi apparatus is fixed to that observed in WT cells (fixed localization).

**Table 1** shows the percentage of the target glycan normalized to the total amount of complex glycan observed experimentally in WT HEK293T cell samples (Bailey Blackburn et al., 2016). For all of the target glycans the proportion could be increased above that found in WT HEK293T cells (**Table 1**). Bi-Sia<sup>2</sup> could be produced by the simulation at the highest relative abundance due to the terminating sialic acids on both antennae, which are endpoint structures. In contrast, homogeneous production of partially sialylated glycans is predicted to be much more difficult if not impossible to achieve according to our stochastic model. When comparing the fits to the bi-Sia<sup>1</sup> and tri-Sia<sup>1</sup> targets we can also say that adding a third antenna makes the drive to homogeneity again more difficult, implying that the inclusion of tri- and/or tetra-antennary glycans in products will inadvertently increase heterogeneity, as was observed in glycan-engineered plant expression systems (Nagels et al., 2012).

As expected, the most abundant glycan by-products predicted when maximizing the production of each target glycan were quite different (**Figure 3**). When maximizing the abundance of bi-Sia1, the major by-products were the un-sialylated biantennary glycan Fuc1GlcNAc4Man3Gal<sup>2</sup> and the fully sialylated Fuc1GlcNAc4Man3Gal2NeuAc2. This result suggests a small window in which the glycoprotein must exit the Golgi apparatus following the initial sialylation but prior to the second sialylation reaction to achieve a bi-Sia<sup>1</sup> glycan structure. The result is similar when enriching tri-Sia1, as the dominant predicted by-products are the over-sialylated tri-antennary glycan, Fuc1GlcNAc5Man3Gal3NeuAc2, and the under sialylated tri-antennary glycan, Fuc1GlcNAc5Man3Gal<sup>3</sup> (**Figure 3**). The major by-products for bi-Sia<sup>2</sup> were the fully sialylated triantennary glycan, Fuc1GlcNAc5Man3Gal3NeuAc<sup>3</sup> suggesting that the modeled sialyation activity is not limiting the production of bi-Sia2, rather the presence of branching enzymes is diverting flux away from bi-Sia2. These by-products could of course be eliminated by modeling a cell line in which the branching enzymes are completely removed, but the ABC algorithm has difficulties pushing enzyme levels to complete elimination i.e., zero, as it will always sample from a non-zero distribution.

Having started with the WT HEK293T cell parameters (Fisher et al., 2019), relative changes in enzyme distributions between Golgi cisternae that maximize the target glycans in silico can be predicted when using localization dependent fitting (**Figure 4**). For all target glycans an increase in the Mgat1 effective enzymatic rate within the second cisterna is required, and a similar change is seen for Fut8 (**Figure 4**). Both of these enzymes act early in the generation of complex N-glycans (Calderon et al., 2016; Fisher et al., 2019). The increases in their levels in an early cisterna presumably ensures that glycan processing can proceed more completely toward the desired products without the accumulation of partially processed intermediates. For the two biantennary glycan targets Mgat2 was required to be redistributed toward the trans side of the Golgi apparatus relative to the simulated WT distribution (**Figure 4**). Analysis of the flux map revealed that the shift in Mgat2 toward the trans side of the Golgi apparatus changed the dominant substrates for Mgat2 (**Figure 5A**). For the model of WT N-glycosylation only ∼8% of the Mgat2 substrates were fucosylated. In contrast, more than 60% of the Mgat2 substrates were fucosylated in the model primed to maximize the abundance of bi-Sia1, and a similar shift in substrate preference is seen for bi-Sia<sup>2</sup> (**Figure 5B**). Unsurprisingly, the effective enzymatic rate of the branching enzyme Mgat5 was predicted to be increased when trying to maximize the abundance of tri-Sia1; however, the distribution of Mgat5 was also required to shift more to the trans side of the Golgi apparatus (**Figure 4**). This shift in Mgat5 coincided with an equivalent shift of GalT into the trans-Golgi, a characteristic that aims to spatially separate the two enzymes within the Golgi apparatus (**Figure 4**). This is in line with competition between galactosylation and glycan branching, a feature that was suggested to explain the demonstrated role of GalT in determining and controlling antenna number (McDonald et al., 2014; Fisher et al., 2019).

If we assume that the intrinsic rates of the glycosylation enzymes are not altered, predicted changes to the effective rates that are made by our model can be rationalized as overexpression or knock down of the relevant protein levels. Such an approach to reducing glycan heterogeneity and enriching the target glycan has indeed been previously applied experimentally for the simple case of core fucosylation (Yamane-Ohnuki et al., 2004). **Figure 6** shows the predicted changes in total effective rate for each enzyme necessary to maximize the relevant target glycan. The

FIGURE 3 | Glycan by-products of maximizing target complex glycans. Relative abundance of the oligomannose pool and the three most abundant complex glycans that are predicted as by-products when maximising the abundance of the indicated target glycans with the indicated type (fixed or variable enzyme localization) of fitting strategy.

predicted distribution in WT HEK293T cells (Fisher et al., 2019) is shown for comparison.

differences in total effective enzymatic rates between the fits using variable and fixed enzyme localizations suggest that the distribution of enzymes is an important factor to be considered when designing genetic manipulations of the glycosylation machinery. When the localization of enzymes is fixed, predicted alterations to the total enzymatic rates are much larger than those required in the variable localization scenario (**Figure 6**). For example, the increase required for Mgat5 is more than threefold higher for the fixed localization fitting of tri-Sia<sup>1</sup> compared with the variable localization fitting of the same glycan. While simply altering the levels of enzymes could conceptually be much simpler through knock-down or overexpression, their distribution may also be changed through engineering of the intra-Golgi vesicular sorting pathway. For example, manipulation of the interactions of the COG vesicle tethering complex (Miller et al., 2013; Willett et al., 2013), or adjustments in the cytosolic and transmembrane

FIGURE 5 | Flux analysis of Mgat2 substrates in WT, bi-Sia1, and bi-Sia<sup>2</sup> simulations. (A) Proportion of biosynthetic flux carried by Mgat2 from its three dominant substrates in WT HEK293T cells (red) and the simulated profiles maximizing bi-Sia<sup>1</sup> (blue) and bi-Sia<sup>2</sup> (purple) fitted with variable enzyme localization. (B) The proportion of substrates for Mgat2 that are fucosylated for the simulations described in "A."

FIGURE 6 | Total enzymatic rate changes to maximize relative abundance of target glycans. Percentage changes in total effective enzymatic rates following fitting to maximize the abundance of bi-Sia<sup>1</sup> (blue), bi-Sia<sup>2</sup> (red), and tri-Sia<sup>1</sup> (green) fitted with variable enzyme localization (solid) and fixed enzyme localization (striped) relative to the parameters predicted for the WT HEK293T cell profile. Total effective enzymatic rates obtained from the different fits were adjusted to ensure that the transit time in each case matched that of the WT HEK293T simulation.

domains of particular glycosylation enzymes could both be used to alter enzyme locations (Ferrara et al., 2006; Becker et al., 2018).

#### A Computationally Informed Strategy for Producing Tetra-Antennary Glycans

Our ABC fitting strategy was unable to generate parameter values for the effective enzymatic rates that would simulate the production of significant amounts of tetra-antennary glycans. Initial parameter values for this fitting were those of the WT HEK293T cell line and therefore the huge shifts required in the effective rate of Mgat4 and 5 are almost never accepted within the fitting as they are considered highly unlikely relative to the WT values. However, it was possible to devise strategies to increase the output of tetra-antennary glycans from our model (**Figure 7A**). Manually elevating the effective rates of Mgat4/5 10-fold, while keeping the distribution of the enzymes the same as in WT HEK293T cells, increased the relative abundance of tetra-antennary glycans about sevenfold, to roughly 25% (**Figure 7A**). However, increasing the effective rates of Mgat4/5 further (100-fold) did not increase the relative abundance of tetra-antennary glycans above that of the 10-fold increased rates, but did increase the abundance of bi-antennary glycans (not shown). In order to investigate why further increasing

the effective enzymatic rates of Mgat4/5 did not result in a concomitant increase in tetra-antennary glycans we examined the flux diagrams for the three conditions (**Figure 7B**). Flux analysis revealed that as the effective enzymatic rate of Mgat4 increases it begins to outcompete ManII and Mgat2 in the earlier cisternae resulting in a reduction in flux toward GlcNAc4Man3, which is an important intermediate on the path of tetra-antennary glycan biosynthesis (**Figure 7B**).

The enzymatic competition between ManII/Mgat2 and Mgat4 is predicted to limit the production of tetra-antennary glycans alongside the known effect of GalT on controlling the degree of N-glycan branching (Fisher et al., 2019). Elimination of GalT was suggested as an approach to producing a larger proportion of tetra-antennary glycans (McDonald et al., 2014). Spatially separating Mgat4/5 from GalT in the Golgi apparatus and therefore minimizing competition between the enzymes (**Figure 7C**) could achieve a similar effect for increasing the relative abundance of tetra-antennary glycans (**Figure 7A**). Indeed, the spatial separation of Mgat5 and GalT that is enforced by the model while enriching tri-Sia<sup>1</sup> (**Figure 4**) demonstrates that cisternal separation allows GalT to evade dominating Mgat5. When applied to the task of increasing the proportion of tetra-antennary glycans the effect was not as large as increasing the effective rates of Mgat4/5, however, the two strategies were synergistic (**Figure 7A**). It is important to note that separation of the competing enzymes increased the relative abundance of tetraantennary glycans but these were often not sialylated. Therefore, in this engineered theoretical scenario an additional cisterna to accommodate sialylation reactions may also be required depending on the target glycan.

#### CONCLUSION

In this work our computational model of mammalian N-glycosylation has been used to probe the glycan processing potential of the Golgi apparatus. While this study was based on computation only, our model's predictions generate some testable hypothesis, which will be interesting to follow up on using laboratory based experiments. Two aspects of Golgi biology, which are commonly affected in pathologies such as CDGs and neurological disorders, the architecture of the Golgi apparatus and transit time through the Golgi apparatus have been shown to be key determinants of the cellular glycan profile. Golgi architecture and transit time should be considered as important factors in glycosylation disorders that likely

have an additive effect in disrupting the protein levels of the glycosylation enzymes.

The goal of generating homogeneous glycoforms of biologics from a process that is inherently heterogeneous is a difficult one. Our model suggests several strategies that we predict will enrich particular target glycan structures; however, our model could not achieve complete homogeneity of glycoforms, suggesting that full uniformity may not be achievable. In the cases of bi-Sia<sup>1</sup> and tri-Sia<sup>1</sup> a higher percentage of the target glycan could be achieved if the substrate specificities of the enzymes were treated as a variable and not fixed. This suggests that when partial sialylation (or galactosylation) is required, protein engineering to adjust the substrate specificities of enzymes is a necessary strategy. A combination of engineering glycosylation enzymes' specificities, controlling effective enzymatic rates and organizing the Golgi apparatus may all be required to attain higher glycan homogeneity, but ultimately the number of cisternae in a mammalian Golgi will likely be a key limiting factor for producing large complex glycans close to homogeneity.

### MATERIALS AND METHODS

#### Modeling Framework

The fitted and validated model of HEK293T cells (Fisher et al., 2019) has been used extensively in this work. This model was presented in detail in Fisher et al. (2019); it is generated using a SSA based on the Gillespie algorithm (Doob, 1945; Gillespie, 1976) that was used to implicitly simulate enzyme competition in the Golgi apparatus. Glycans entered the in silico Golgi apparatus one at a time, hence our model does not account for substrate competition or the potential effects of protein load in the secretory pathway. The processing of 10,000 glycans was simulated in order to generate a simulated glycan profile from which the abundance of individual or structural classes of glycans can be obtained. This profile is compared to a real data set utilizing a Bayesian fitting methodology with priors based upon biological knowledge. In the case of this study the priors were based on the parameters obtained by fitting the WT HEK293T glycan profile.

#### Transit Time and Cisternal Element Number

The enzymatic parameters as well as the proportions of Man9GlcNAc2, Man8GlcNAc2, and GlcMan9GlcNAc<sup>2</sup> obtained for the fitted WT HEK293T cell line (Fisher et al., 2019) were used. The relative time taken for a glycan to transit through each cisternal element was varied and the glycosylation reactions simulated. The number of simulated glycans and the relative abundance of oligomannose glycans was calculated from the average of three simulations for each transit time.

For investigating the effect of cisternal element number on the N-glycosylation process the total effective enzymatic rates for each enzyme were kept constant in addition to the total transit time and proportions of Man9GlcNAc2, Man8GlcNAc2, and GlcMan9GlcNAc2. Extra cisternal elements were added by calculating the average effective enzymatic rate between two



adjacent cisternae and then rescaling the effective enzymatic rates to equal the total for the fitted WT HEK293T cells. By using this strategy, we ensure any alterations to the simulated glycan profile are not the result of changes in enzyme levels. The number of simulated glycans and the relative abundance of oligomannose glycans was calculated from the average of three simulations for each cisternal number.

#### Maximizing Target Glycans

To predict alterations to enzyme levels and localization that can maximize a given target glycan we used the ABC fitting methodology to fit a simulated glycan profile to a hypothetical observed glycan profile. The hypothetical glycan profile was constructed by making the relative abundance of the target glycan to 100% of the combined complex and hybrid glycan abundances. The proportion of oligomannose glycans was kept at the level observed in WT HEK293T cells in all target profiles. ABC fitting methodology was then used to predict alterations in the organization of the Golgi machinery to generate the hypothetical glycan profile. The model was not penalized for what type of byproducts were produced, as such it is reasonable to assume that the predicted by-products would be observed in a true glycan profile of an engineered cell line.

#### Maximising Tetra-Antennary Glycans

WT HEK293T fitted parameters were used as a starting point. The effective enzymatic rates of Mgat4 and Mgat5 were increased 10- or 100-fold and the glycan profile simulated. For the spatial separation of the key enzymes involved in determining tetraantennary glycans, the effective enzymatic rates were condensed into specific cisternae as shown in **Table 2**. The relative abundance of all tetra-antennary glycans was calculated from the average of three simulations for each different scenario.

# DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

#### AUTHOR CONTRIBUTIONS

All authors designed the study, planned the experiments, and wrote the manuscript. PF performed the experiments.

#### FUNDING

fcell-07-00157 August 9, 2019 Time: 16:32 # 10

This work was supported by an Impact Accelerator Award to AW, DU, and JT-O in collaboration with GSK (BB/S506795/1), a BBSRC IB Catalyst grant to DU (BB/M018237/1) and the

#### REFERENCES


York Centre of Excellence in Mass Spectrometry, which was created thanks to a major capital investment through Science City York, supported by Yorkshire Forward with funds from the Northern Way Initiative, and subsequent support from EPSRC (EP/K039660/1 and EP/M028127/1).



cellular cytotoxicity. Biotechnol. Bioeng. 87, 614–622. doi: 10.1002/bit. 20151

Zolov, S. N., and Lupashin, V. V. (2005). Cog3p depletion blocks vesicle-mediated golgi retrograde trafficking in HeLa cells. J. Cell Biol. 168, 747–759. doi: 10. 1083/jcb.200412003

**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 Fisher, Thomas-Oates, Wood and Ungar. 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.

# A New Look at the Functional Organization of the Golgi Ribbon

#### Jaakko Saraste<sup>1</sup> \* and Kristian Prydz<sup>2</sup> \*

<sup>1</sup> Department of Biomedicine and Molecular Imaging Center, University of Bergen, Bergen, Norway, <sup>2</sup> Department of Biosciences, University of Oslo, Oslo, Norway

A characteristic feature of vertebrate cells is a Golgi ribbon consisting of multiple cisternal stacks connected into a single-copy organelle next to the centrosome. Despite numerous studies, the mechanisms that link the stacks together and the functional significance of ribbon formation remain poorly understood. Nevertheless, these questions are of considerable interest, since there is increasing evidence that Golgi fragmentation – the unlinking of the stacks in the ribbon – is intimately connected not only to normal physiological processes, such as cell division and migration, but also to pathological states, including neurodegeneration and cancer. Challenging a commonly held view that ribbon architecture involves the formation of homotypic tubular bridges between the Golgi stacks, we present an alternative model, based on direct interaction between the biosynthetic (pre-Golgi) and endocytic (post-Golgi) membrane networks and their connection with the centrosome. We propose that the central domains of these permanent pre- and post-Golgi networks function together in the biogenesis and maintenance of the more transient Golgi stacks, and thereby establish "linker compartments" that dynamically join the stacks together. This model provides insight into the reversible fragmentation of the Golgi ribbon that takes place in dividing and migrating cells and its regulation along a cell surface – Golgi – centrosome axis. Moreover, it helps to understand transport pathways that either traverse or bypass the Golgi stacks and the positioning of the Golgi apparatus in differentiated neuronal, epithelial, and muscle cells.

Keywords: Golgi ribbon, mitosis, cell migration, cell differentiation, Golgi bypass, centrosome, intermediate compartment, recycling endosome

# INTRODUCTION

The Golgi apparatus modifies, sorts and transports proteins, lipids, and complex carbohydrates at the crossroads of the secretory and endocytic pathways. The Golgi is structurally unique, consisting of polarized stacks of flattened cisternae flanked by tubular networks (Mellman and Simons, 1992; Weidman et al., 1993; Mollenhauer and Morré, 1998; Jackson, 2009). Two opposing hypotheses have been put forward to explain the formation of such complex architecture (Glick, 2002). According to a more traditional view, the biogenesis of the Golgi stacks requires a permanent template; however, the nature of such a template has not been unequivocally established (Palade, 1983; Seemann et al., 2000). According to another proposition, the Golgi apparatus is a selforganizing structure, which assembles from dynamic components, exists in a state of equilibrium, and is capable of de novo formation (Misteli, 2001; Altan-Bonnet et al., 2004; Ronchi et al., 2014). In

#### Edited by:

Daniel Ungar, University of York, United Kingdom

#### Reviewed by:

Brian Storrie, University of Arkansas for Medical Sciences, United States Alexandre A. Mironov, Italian Foundation for Cancer Research (FIRC), Italy

#### \*Correspondence:

Jaakko Saraste jaakko.saraste@uib.no Kristian Prydz kristian.prydz@ibv.uio.no

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 10 May 2019 Accepted: 07 August 2019 Published: 21 August 2019

#### Citation:

Saraste J and Prydz K (2019) A New Look at the Functional Organization of the Golgi Ribbon. Front. Cell Dev. Biol. 7:171. doi: 10.3389/fcell.2019.00171

addition, there is data suggesting that the Golgi apparatus is a modular structure, with the joining of cisternal stacks into a ribbon structure representing the highest order of assembly (Nakamura et al., 2012; **Figure 1**). Evidence for structural Golgi modules may be obtained when looking more closely at different cell types or dividing cells. For example, during mitosis the Golgi stacks undergo disassembly, and resident Golgi enzymes temporarily end up in a vesicular Golgi haze (Shorter and Warren, 2002; Marie et al., 2012). The budding yeast Saccharomyces cerevisiae is generally considered to contain separate Golgi cisternae (Suda and Nakano, 2012); however, formation of stacked Golgi-like structures is observed in mutant yeast cells or under certain growth conditions (Rambourg et al., 1993; Hashimoto et al., 2002). Most typically, invertebrates, plants and many fungi contain individual or pairs of Golgi stacks distributed throughout the cytoplasm close to ER exit sites (ERES). Vertebrate cells display the highest level of complexity as they contain a Golgi ribbon, consisting of numerous cisternal stacks (compact zones) connected by tubular networks (noncompact zones) into a single copy organelle (Ladinsky et al., 1999; Kepes et al., 2005).

However, why vertebrate cells build a Golgi ribbon has generally remained an enigma (Wei and Seemann, 2010; Gosavi and Gleeson, 2017). Namely, ribbon organization is not strictly required for secretion, as clearly demonstrated by experiments with nocodazole, a microtubule (MT)-depolymerizing drug, which causes the replacement of the central Golgi ribbon by ERES-associated ministacks (Cole et al., 1996a; Thyberg and Moskalewski, 1999; Fourriere et al., 2016). It has been suggested that ribbon organization, by allowing lateral mobility of Golgi enzymes between the stacks, ensures correct glycosylation of

FIGURE 1 | Building blocks of the Golgi apparatus. A model suggesting modular assembly and disassembly of the Golgi apparatus, based on its organization in various cell types and during different stages of the cell cycle. The prevailing view is that the preformed Golgi stacks in mammalian cells extend tubules that undergo tethering and fusion, thereby giving rise to a continuous Golgi ribbon consisting of compact (stacked) and non-compact (tubular) regions. Here, we argue that the non-compact zones are structurally more complex, being occupied by pleiomorphic "linker compartments", which due to their function in the biogenesis of the Golgi stacks also dynamically join them together.

cargo proteins (Puthenveedu et al., 2006; Xiang et al., 2013). Based on a rim progression Golgi model, lateral connections between neighboring stacks may facilitate anterograde intra-Golgi transport of large-sized cargo proteins (Lavieu et al., 2014), or allow the formation of large aggregates of endothelial von Willebrand factor (Ferraro et al., 2014). This proposal is in accordance with super-resolution light microscopy (LM) of individual Golgi stacks, showing the preferential localization of bulky, but not small cargo proteins to cisternal rims. Moreover, a large number of cargo processing enzymes localize to the central portion of the cisternae, while transport machinery proteins are found at the periphery of the stacks (Tie et al., 2018).

Furthermore, on top of its classical roles in modification, sorting and transport of cargo, the Golgi apparatus has been assigned novel functions that seem to require an intact ribbon structure. For example, there is considerable evidence that it participates actively in cell signaling (Farhan et al., 2010; Chia et al., 2012; Luini and Parashuraman, 2016; Makhoul et al., 2018). The first signaling event, which is coupled to fragmentation of the Golgi ribbon was identified via the demonstration of a "Golgi checkpoint" regulating mitotic entry (Sütterlin et al., 2002; Colanzi et al., 2007). More recently, the coordinated trafficking and signaling functions of the Golgi apparatus have been implicated in complex cellular processes, such as cell migration, metabolism, and autophagy (Millarte and Farhan, 2012; Makhoul et al., 2018). Strikingly, the Golgi collaborates with the centrosome in providing a platform for the nucleation of MTs (Chabin-Brion et al., 2001; Efimov et al., 2007; Rivero et al., 2009), to support ribbon integrity, cell polarization and motility. Indeed, recent studies indicate that directional cell migration, which involves polarized delivery of lipids and proteins to the cell's leading edge, depends on reorientation of both the centrosome and the Golgi ribbon, as well as an asymmetrical array of Golginucleated MTs (Miller et al., 2009; Yadav et al., 2009; Hurtado et al., 2011). In sum, the discovery of these novel organelle functions raises questions regarding the division of labor between the compact and non-compact regions of the Golgi ribbon. In fact, in specific cell types, the non-compact zones amount to up to 50% of the total volume of the ribbon (Noske et al., 2008).

How are the stacks actually joined together? The prevailing view is based on stereoscopic EM analysis of serial sections in a variety of cell types (Rambourg and Clermont, 1997; Kepes et al., 2005) and fluorescence recovery after photobleaching (FRAP) experiments demonstrating the continuity of the Golgi ribbon (Cole et al., 1996b; Puthenveedu et al., 2006). Accordingly, the cisternal stacks are thought to tether and fuse laterally, resulting in the formation of stable or transient tubular connections. Although it is commonly stated that such fusions only give rise to homotypic links between cisternae occupying equivalent positions in adjacent stacks (**Figure 1**), interconnections may be created between cisternae at different levels of neighboring stacks (Kepes et al., 2005). Of note, in many cell types the lateral tubular networks also appear to be in continuity with a tubular system at the cis-side of the Golgi ribbon (Kepes et al., 2005).

The joining of the Golgi stacks into a ribbon involves complex cellular machinery (Wei and Seemann, 2010; Mironov and Beznoussenko, 2011; Bechler et al., 2012;

Rabouille and Linstedt, 2016; Huang and Wang, 2017). Both centrosome- and Golgi-nucleated MTs participate in this process, contributing to the central positioning or lateral linking of the stacks, respectively (Miller et al., 2009; Lowe, 2011; Yadav and Linstedt, 2011; Nakamura et al., 2012). More recently, an actin-based filament system that collaborates with MTs and Golgi-associated proteins, such as Cdc42, Rab1, and GRASP65 (Kodani et al., 2009; Hehnly et al., 2010; Copeland et al., 2016; Russo et al., 2016; Tang et al., 2016; Xing et al., 2016; Kage et al., 2017; Makhoul et al., 2019), has been implicated in ribbon formation (Egea et al., 2013; Gosavi and Gleeson, 2017). In addition, both membrane flow and cargo load influence the structure and function of the Golgi ribbon (Sengupta and Linstedt, 2011). Accordingly, its integrity depends on ongoing pre-, intra-, and post-Golgi membrane traffic (Yang et al., 2011; Climer et al., 2015; Blackburn et al., 2018; Makhoul et al., 2019) and blocking the transport of cargo-containing ER-to-Golgi carriers (Marra et al., 2007), or depletion of cargo receptors (Mitrovic et al., 2008), results in ribbon fragmentation. Finally, besides cell stress and apoptosis (Machamer, 2015) Golgi fragmentation is associated with various pathological conditions, including neurodegenerative disorders – such as amyotrophic lateral sclerosis (ALS), Alzheimer's and Parkinson's disease – and cancer. Notably, while the causative role of Golgi alterations in the progression of these diseases remains open, there are indications that they result from general effects on membrane traffic (Stieber et al., 1996; Fujita et al., 2006; Rendón et al., 2013; Joshi et al., 2015; Makhoul et al., 2018).

The identification of the roles of various transport machinery proteins – such as the GRASPs, golgins and regulatory GTPases – in ribbon formation is largely based on studies showing that their inhibition or depletion leads to Golgi fragmentation (de Figueiredo et al., 1998; Yadav and Linstedt, 2011; Goud et al., 2018). Two types of fragmentation can be distinguished: first, blocking MT- and dynein-dependent centralization of dynamic intermediate compartment (IC) elements and endosomes – as occurs in cells treated with nocodazole – gives rise to Golgi ministacks close to ERES. This situation is exemplified by knockdown of the dynein receptor golgin-160, or GMAP-210, a tethering protein ("golgin") implicated in ER-Golgi trafficking at the level of the IC (Rios et al., 2004; Yadav et al., 2009; Roboti et al., 2015). In the second form of fragmentation, severing the Golgi ribbon – evidently due to local effects – leaves separated cisternal stacks residing at the cell center next to the centrosome.

In conclusion, the complex machineries implicated in the formation and maintenance of the Golgi ribbon, including local and global players, are difficult to reconcile with the currently popular model depicting narrow tubular connections between the Golgi stacks (**Figure 1**). This relatively simple model places the focus on the cisternal stacks as the basic structural and functional units of the ribbon, but does not adequately take into account the extensive tubular networks that – as mentioned earlier – represent an additional key feature of this organelle (Mollenhauer and Morré, 1998; Kepes et al., 2005). Moreover, several studies indicate that the ultrastructural organization of the non-compact zones within the ribbon is more complex than presented by the prevailing "tubular bridge model" (Thorne-Tjomsland et al., 1998; Ladinsky et al., 1999; Martínez-Martínez et al., 2017).

Therefore, based on the recently discovered spatial and functional connections between the membrane networks operating in ER-Golgi and endocytic trafficking (Marie et al., 2009, 2012; Bowen et al., 2017), we propose an alternative model for Golgi organization in vertebrate cells. According to this model these networks, which co-exist at the cell periphery and around the centrosome, also meet at the level of the Golgi ribbon, representing a permanent template that generates the transient Golgi stacks and simultaneously links them into a continuous structure. Unlike the "tubular bridge model", this "linker compartments model" can clarify the tight coordination of the repositioning of the Golgi ribbon and the centrosome as a prerequisite for cell division and directed cell migration. It is also relevant for understanding the development of endomembranes and their rearrangements during cell differentiation. Furthermore, we discuss the implications of this model for enigmatic processes that take place at opposite sides of the Golgi stacks – such as MT nucleation and autophagy – as well as transport routes that pass through or circumvent the Golgi stacks.

But first, an introduction to the present terminology: In the following we refer to the two interconnected membrane systems defined by Rab1 and Rab11 as biosynthetic and endocytic networks, respectively, and their individual dynamic components as IC elements and recycling endosomes (REs). Based on their accumulation around the centrosome, the central domains of these networks have been previously designated as biosynthetic (BRC) and endocytic recycling compartments (ERC) (Maxfield and McGraw, 2004; Saraste and Goud, 2007). For simplicity, the IC elements and REs at non-compact zones of the Golgi ribbon have been dubbed here as "linker compartments".

## UNLINKING OF THE GOLGI RIBBON DURING CELL DIVISION AND MIGRATION: TWO ALTERNATIVE VIEWS

Despite its complex organization the Golgi apparatus is capable of rapidly changing its shape and cellular location under different physiological conditions. Typically, such dynamic alterations coincide with the repositioning of the centrosome and the unlinking of the Golgi ribbon (Rios and Bornens, 2003; Sütterlin and Colanzi, 2010). These events are necessary, for instance, for equal partitioning of this single-copy organelle during cell division and its reorientation toward the lamellipodium during cell migration. In the following, we discuss these two cellular processes in light of the commonly accepted "tubular bridge model" of the Golgi ribbon (**Figure 1**) and the "linker compartments model" proposed here (see **Figure 2**). Golgi rearrangements are also an integral part of cell differentiation, taking place, for example, during the formation of neuronal extensions and the polarization of epithelial cells (see below; **Figure 5**).

The best-characterized process of physiological Golgi fragmentation takes place as cells prepare for mitosis. At the late G2 stage of the cell cycle the mammalian Golgi ribbon

breaks down into individual stacks due to activation of the membrane fission protein CtBP1/BARS (Hidalgo-Carcedo et al., 2004; Colanzi et al., 2007), and phosphorylation of the two tethering proteins GRASP65 and GRASP55 (Sütterlin et al., 2002; Yoshimura et al., 2005; Feinstein and Linstedt, 2007, 2008; Duran et al., 2008; Cervigni et al., 2015). However, the precise roles of these two factors in this process remain incompletely understood (Ayala and Colanzi, 2017). For example, whereas the function of the GRASPs in homotypic tethering of membranes (via trans-oligomerization) has been extensively characterized (Rabouille and Linstedt, 2016), the mechanism of CtBP1/BARS activation remains unknown. Nevertheless, evidently as a consequence of the joint action of the GRASPs and CtBP1/BARS, the initially asymmetric juxtanuclear Golgi stacks end up circling the nucleus as the cells reach prophase, coinciding with the separation of the centrosomes and initiation of formation of the mitotic spindle (Shorter and Warren, 2002; Wei and Seemann, 2017). If Golgi fragmentation is blocked – for example, by inhibiting CtBP1/BARS activation or the phosphorylation of one of the GRASPs – the progression of cells from G2 to prophase is delayed. This regulatory event of the cell cycle is referred to as the Golgi checkpoint (Sütterlin et al., 2002; Colanzi et al., 2007). A similar controlled unlinking process occurs during mitotic entry in Drosophila S2 cells, despite the fact that the fly Golgi is not a ribbon, but exists as pairs of stacks. Notably, however, in this case the linking or unlinking of the stacks does not involve the single Drosophila GRASP homolog (dGRASP), but is mediated by the Golgi-associated actin cytoskeleton (Kondylis et al., 2007).

Directed migration of fibroblasts is also accompanied by unlinking of the Golgi ribbon, followed by its subsequent relocation to the side of the nucleus facing the leading edge (Kupfer et al., 1982). This process ensures polarized delivery of membrane constituents – lipids and specific proteins, such as integrins – to the leading edge, thereby supporting cell polarization and directed motility (Bisel et al., 2008; Millarte and Farhan, 2012). Besides contributing to linking of the stacks, Golgi-nucleated MTs establish an asymmetric array of filaments, providing tracks for polarized trafficking to the lamellipodium (Efimov et al., 2007; Miller et al., 2009; Rivero et al., 2009). While Golgi and the centrosome are thought to part company as cells enter mitosis (Champion et al., 2017), Golgi relocation during cell migration is intimately coupled to repositioning of the centrosome (Sütterlin and Colanzi, 2010; Hurtado et al., 2011).

In fact, unlinking of the Golgi stacks appears to be a prerequisite for repositioning of the centrosome (Preisinger et al., 2004; Bisel et al., 2008; Millarte and Farhan, 2012). Namely, as in mitosis, this process depends on phosphorylation of GRASP65, and inhibition of this modification – for example, using non-phosphorylatable mutants – blocks centrosome positioning and cell polarization (Bisel et al., 2008). Cell migration is also regulated by GM130, which associates with IC/cis-Golgi membranes via GRASP65 (Preisinger et al., 2004; Saraste and Marie, 2018). GM130 could affect cell polarization and migration via multiple mechanisms (Sütterlin and Colanzi, 2010). One could involve interaction with the Rho family GTPase Cdc42, a key regulator of cell polarization (Etienne-Manneville, 2004; Kodani et al., 2009; Baschieri et al., 2014; see below). Another possible role of GM130 in cell migration could depend on its function in Golgi nucleation of MTs, which provide tracks for transport to the lamellipodium (Rivero et al., 2009). Furthermore, GM130 provides a scaffold for the activation of kinases (YSK1 and MST4) that regulate cell migration (Preisinger et al., 2004).

In summary, the two types of events leading to Golgi fragmentation, taking place at G2/M transition or during cell migration, are at least partly regulated by different signaling pathways (Millarte and Farhan, 2012; Ayala and Colanzi, 2017). Also, the extent of Golgi disassembly differs in these two cases. During mitosis the Golgi undergoes a multi-step disassembly process, which results in the appearance of two components: tubulovesicular membrane clusters concentrating at the spindle poles and a vesicular Golgi haze (Marie et al., 2012; Wei and Seemann, 2017). By contrast, Golgi reorganization during cell migration seems to be less dramatic, possibly limited to unlinking of the ribbon and partial breakdown of the Golgi stacks (Bisel et al., 2008). Interestingly, GRASP65 is phosphorylated at the same site (Ser 277) by ERK or JNK2 during cell migration and mitotic entry, respectively, indicating that Golgi fragmentation during these cellular events shares similar mechanisms. Based on the "tubular bridge model", a commonly held view is that the molecular changes in both cases initially trigger the severing of the tubular connections between the relatively stable cisternal stacks, resulting in the unlinking of the Golgi ribbon. As a consequence, the individual Golgi stacks are thought to be released and even become mobile, allowing their repositioning.

#### A New View of the Golgi Ribbon

Our new model regarding the functional organization of the Golgi ribbon and its behavior at the onset of mitosis and during cell motility (**Figures 2**, **4**) embodies the idea that the noncompact regions are structurally and functionally more complex than proposed by the "tubular bridge model". It is based on the discovery of permanent connections between the IC and the endocytic recycling system and the anchoring of the two networks at the centrosome (Marie et al., 2009, 2012; Bowen et al., 2017; Saraste and Marie, 2018). Indeed, a direct link between the pericentrosomal IC elements and recycling endosomes (REs) – defined by the GTPases Rab1 and Rab11, respectively – persists when the Golgi stacks are disassembled by Brefeldin A (BFA; Marie et al., 2009), a reversible inhibitor that dissociates specific protein coats (COPI, clathrin) from membranes and has been extensively used to study endomembrane organization and protein transport in different cell types (Klausner et al., 1992; Prydz et al., 1992; Marie et al., 2008; Robinson et al., 2015). Here, we propose that – in addition to meeting at the cell periphery and around the centrosome – the central IC elements and REs also co-exist at the non-compact regions of the ribbon (**Figure 2**). Here they co-operate in the biogenesis of Golgi cisternae and consequently act as "linker compartments" that connect the stacks (**Figure 4**) in a process which is expected to be more dynamic than the one depicted in the "tubular bridge model".

The alternative model is supported by EM tomographic studies of both cultured cells and tissues, providing highresolution data on the non-compact regions of the Golgi ribbon (Ladinsky et al., 1999; Marsh et al., 2001; Martínez-Martínez et al., 2017). Ultrastructural analysis shows that these linker regions – besides displaying apparently stable tubular or saccular connections between the neighboring Golgi stacks – are characterized by large openings. Notably, these "wells" are filled with pleiomorphic structures resembling IC elements and endosomes, as well as tubules and coated or noncoated vesicles (Ladinsky et al., 1999). In pancreatic β-cells, where MTs are predominantly nucleated at the Golgi (Zhu et al., 2015), these filaments typically associate with cis-Golgi cisternae and endo-lysosomal compartments in the vicinity of the Golgi ribbon (Marsh et al., 2001). In addition, MTs can be seen passing through the non-compact zones (Marsh et al., 2001; Martínez-Martínez et al., 2017). Collectively, the above features support the conclusion that the non-compact regions represent structurally complex sites for dynamic transport events, rather than consisting solely of narrow tubular connections between the stacks.

Furthermore, it has been recognized for quite some time that a typical feature of the Golgi apparatus in many cell types is the presence of extensive tubular networks (Mollenhauer and Morré, 1998). Indeed, such networks represent a conserved aspect of Golgi structure, being present in animals, plants, and fungi, and – corresponding to roughly half of the total Golgi membrane – can also be expected to play an important role in Golgi function. Importantly, besides the cis- and transaspects of the stacks, they are also found on their lateral sides, contributing in vertebrate cells to the establishment of the non-compact regions of the ribbon. These regions also include saccular elements and display continuity with forming secretory granules. Notably, Mollenhauer and Morré proposed that the tubular networks represent the permanent components of the Golgi ribbon, whereas the Golgi stacks – based on the cisternal progression model – were expected to undergo continuous turnover (Mollenhauer and Morré, 1998).

**Figure 2** shows the application of our alternative "linker compartments model" in the context of Golgi rearrangements taking place during cell division and motility. Experimental support for the mitosis model (**Figure 2A**) was obtained by live cell imaging of cells expressing the IC marker GFP-Rab1. At late G2, jointly with the movement of the duplicated centrosome to the cell center, a pool of IC membranes detaches from the Golgi ribbon. At prophase, this compartment (designated as BRC) – together with the Rab11-positive ERC – first grows and then divides as the centrosomes separate, and finally moves together with the latter to the forming spindle poles (Marie et al., 2012;

**Figure 2A**). Since the separation and expansion of these pericentrosomal compartments coincide with the unlinking of the Golgi ribbon, we proposed that they are derived from its non-compact regions (Marie et al., 2012). How do the "linker compartments" pile up around the centrosome? A simple scenario is that as a consequence of the unlinking of the Golgi stacks and the release of these compartments from the ribbon – for example, due to membrane untethering and/or cytoskeletal rearrangements – they are free to move toward the centrosome in a dynein-dependent fashion, using the radiating centrosomal MTs as tracks (**Figure 2A**).

Besides Rab11 (Marie et al., 2012; Hehnly and Doxsey, 2014), the pericentrosomal ERC at the spindle poles can be visualized via endocytosed transferrin, or antibodies against its receptor (Takatsu et al., 2013; **Figure 3**). The BRC also contains the Rab1 effectors GM130 and p115 (Seemann et al., 2002; Radulescu et al., 2011), as well as GRASP65, which provides a membrane anchor for GM130 (Marie et al., 2012). Based on their proposed function in the biogenesis and maintenance of the Golgi stacks (Saraste and Marie, 2018), the linker compartments dynamically interact with the stacks during interface, as well as with the vesicular Golgi haze during mitosis (Marie et al., 2012). Therefore, Golgi enzymes may also be found at the spindle poles, as a consequence of their missorting due to overexpression and/or tagging.

The present model suggests that the Golgi ribbon consists of two main domains with distinct properties (**Figures 2**, **4**). The non-compact linker regions are considered as the permanent part of the ribbon, which function in the formation of the transient Golgi stacks. As discussed above, the linker compartments are expected to actively communicate with the stacks via vesicular or tubular trafficking. In addition, they can establish more stable connections, allowing communication between neighboring Golgi stacks. This two-component model is in accordance with results suggesting that different parts of the Golgi employ different inheritance strategies (Wei and Seemann, 2009). Thus, the vesicular Golgi haze – evidently together with linker compartments at the cell periphery – can generate transportcompetent Golgi stacks, while a spindle-associated component is required for post-mitotic ribbon formation. Detailed studies of the Golgi reassembly process during mitotic exit can address the validity of this two-domain model. Interestingly, during cytokinesis the reforming Golgi elements in the daughter cells first organize into two unequal membrane clusters at the two sides of the nuclei. The smaller Golgi cluster ("twin Golgi"), localized near the intercellular bridge, then moves to the opposite side of the nucleus to join the larger pericentrosomal Golgi cluster during reformation of the interphase Golgi ribbon (Gaietta et al., 2006; Marie et al., 2012).

The model regarding the role of the pericentrosomal compartments in Golgi repositioning during cell migration (**Figure 2B**) is also based on live imaging of GFP-Rab1 (Marie et al., 2009). Similarly as during G2/M transition, the IC membranes are relocated with the centrosome to the cell center as the cell starts to move. Subsequently, the Rab1-containing IC/cis-Golgi membranes are transferred to the opposite side of the nucleus, apparently utilizing the centralized pericentrosomal compartment as a way station. Finally, due to the function of the linker compartments in reformation of the Golgi stacks, the reoriented Golgi ribbon – simply based on spatial constraints – is positioned at a distance from the centrosome to face the leading edge (**Figure 2B**).

As mentioned earlier, treatment of cells with BFA results in breakdown of the Golgi stacks and accumulation of the linker compartments around the centrosome (Marie et al., 2009), creating a situation very similar to that seen during mitotic onset and cell motility (**Figure 2**). Therefore, it does not come as a total surprise that BFA can "rescue" both an experimentally induced block in mitotic entry (Sütterlin et al., 2002; Feinstein and Linstedt, 2007; Cervigni et al., 2015), and centrosome reorientation in motile cells where ribbon fragmentation has been experimentally inhibited (Bisel et al., 2008). Of note, the linker compartments maintain their close connection during mitosis (Marie et al., 2012; Takatsu et al., 2013; Hehnly and Doxsey, 2014; **Figure 3**), as well as during cell migration (Dale et al., in preparation). Moreover, wound-healing assays reveal that cell motility is not inhibited, but rather enhanced, during the first hours of BFA treatment (Dale et al., in preparation). Furthermore, the ability of BFA to rescue centrosome positioning as a prerequisite to cell migration revealed how critically dependent this process is on the unlinking of the Golgi ribbon (Bisel et al., 2008). Based on the "tubular bridge model" it looked as if the extensive Golgi ribbon would somehow be able to mechanically or sterically inhibit centrosome motility. Simultaneous repositioning of the linker compartments and the centrosome (**Figure 2**) may solve this puzzle and explain the tight coordination of these processes, which may both involve the master of cell polarization, the GTPase Cdc42 (see below).

Finally, these considerations set the stage for a new view of Golgi positioning. According to one popular view the Golgi ribbon is first fragmented, whereafter the individual stacks are free to move across the cytoplasm to find their new location. Alternatively, resident Golgi enzymes may redistribute to the ER and organelle repositioning involves de novo assembly of Golgi stacks at ERES, resulting in ribbon formation at a new site. As a trade-off, the new model emphasizes a novel role of the linker compartments in defining Golgi repositioning, either at the distal side of the daughter nuclei at telophase (**Figure 2A**), or facing the leading edge of a motile cell (**Figure 2B**). We propose that in both situations the cis/medial- and trans-Golgi residents are redistributed to the permanent IC and endosomal networks, respectively, and relocate together with these dynamic elements, resulting in reformation of the Golgi stacks at new locations. The difference is that during mitosis Golgi enzymes are further distributed to the vesicular Golgi haze which, however, still communicates with the compartments at the spindle poles (**Figure 2A**; Marie et al., 2012).

#### SPATIAL ASPECTS OF TRAFFICKING AND SIGNALING

The localization of endosomes and IC elements at the cell center is based on their dynein-dependent movements along MT tracks (Burkhardt et al., 1997; Presley et al., 1997; Horgan et al., 2010;

Granger et al., 2014). The positioning of these compartments at the non-compact zones of the Golgi ribbon, at a distance from the centrosome (**Figure 4**), could be based on simple spatial constraints, created by their centralization and function in the formation of the sizeable Golgi stacks. Alternatively, it could be influenced by their association with actin filaments, mutual adhesion – for example, the establishment of membrane contact sites – or the opposing forces generated by MT motors. Indeed, both the IC elements and REs (containing Rab1 and Rab11, respectively) are capable of moving bidirectionally along MTs. As a consequence, they are also found at the cell periphery (**Figure 4**); for example, in the protrusions or lamellipodia of migrating fibroblasts, and neuronal growth cones (Hattula et al., 2006; Sannerud et al., 2006; Eva et al., 2010; Matsuzaki et al., 2011; Takahashi et al., 2012). Unexpectedly, the well-established IC/cis-Golgi proteins Rab1, Arf1 and GBF1 – the GTP exchange factor of the latter – have been shown to act in endocytic trafficking (Gupta et al., 2009; Mukhopadhyay et al., 2011; Kaczmarek et al., 2017). Arf1 regulates a constitutive clathrin-independent endocytic pathway, which is also mediated by Cdc42, and plays a major role in membrane turnover at the leading edge of migrating cells. Strikingly, the protein profile of the clathrin-independent carriers (CLICs) operating in this pathway includes also the IC proteins Rab1, Sec22b and p58/ERGIC-53 (Howes et al., 2010).

(see also Marie et al., 2012; Takatsu et al., 2013).

Together with their enrollment as linker compartments in the Golgi ribbon these considerations provide a new view on the spatial organization of the MT-based early biosynthetic (IC) and endocytic membrane networks. The emerging cell surface – Golgi ribbon – centrosome axis (**Figure 4**) can provide an explanation for the striking operation of the same transport machineries both at the ER-Golgi boundary and the cell periphery. In the following, we also address the implications of this novel axis for signaling events that regulate the onset of mitosis or cell migration. Furthermore, by shifting the main focus away from the cisternal Golgi stacks, the model is relevant for considering the localization and function of machinery proteins implicated in ribbon formation, such as the GRASPs, as well as Golgiindependent pathways of protein and lipid trafficking.

## Signaling at the Golgi Checkpoint

As discussed above, the fragmentation of the Golgi ribbon at G2 is linked to cell cycle control mechanisms, coinciding with the "Golgi checkpoint" that regulates mitotic entry (Sütterlin et al., 2002; Hidalgo-Carcedo et al., 2004; Yoshimura et al., 2005; Colanzi et al., 2007). Based on the prevailing model of the ribbon (**Figure 1**), the general idea is that this control station monitors the successful splitting of the continuous Golgi ribbon into individual stacks. Thus, despite the fact that severing the tubular connections between the stacks marks only the beginning of a multi-step Golgi disassembly process, the consensus is that the checkpoint oversees organelle inheritance (Wang and Seemann, 2011; Ayala and Colanzi, 2017). Some of the signaling events that link Golgi integrity to mitotic entry have recently been identified. Accordingly, ribbon fragmentation at late G2

at ERES. For simplicity, a structure consisting of five stacks displaying uniform cis-trans polarity is shown, while in reality the Golgi ribbon is a twisted, basket-shaped structure in the perinuclear area of a fibroblastic cell. The blow-up illustrates a non-compact region of the Golgi ribbon. The linker compartments derived from the central domains of biosynthetic (IC) and endocytic (EN) networks are schematically depicted as separate structures, although they are expected to establish tubular and saccular continuities between the neighboring stacks (blue). The centrosomal and non-centrosomal (Golgi-nucleated) MTs with plus-minus polarity are indicated in orange and brown color, respectively.

leads to the activation of a Golgi-localized Src kinase, which phosphorylates another key kinase, Aurora A, resulting in its activation and recruitment to the centrosome (Persico et al., 2010; Barretta et al., 2016). This event is a prerequisite for centrosome maturation, including expansion of the pericentrosomal material, which affects MT nucleation and formation of the mitotic spindle (Wei et al., 2015; Barretta et al., 2016). Importantly, the recruitment of activated Aurora A to the centrosome culminates in the activation of Cdk1, the master kinase that sets mitosis in motion (Champion et al., 2017).

What is the mechanism that couples Golgi unlinking to centrosome maturation? How does the apparently trans-Golgi/TGN-localized Src kinase come in contact with Aurora A at the centrosome? The proposed behavior of the linker compartments at the onset of mitosis (Marie et al., 2012; **Figure 2A**) may provide an answer. Namely, their detachment from the Golgi ribbon at late G2 and movement to the pericentrosomal region may constitute the pathway that mediates the interaction of the two kinases and the recruitment of activated Aurora A to the centrosome (Barretta et al., 2016). Indeed, similar relocation of TGN proteins to the pericentrosomal area takes place when the Golgi stacks are disassembled by BFA (Reaves and Banting, 1992; Molloy et al., 1994). This Golgi ribbon-centrosome axis could also act in the transfer of other key proteins that regulate mitotic entry, such as cyclin B2, the partner of Cdk1 (Jackman et al., 1995) and the phosphatase Cdc25C, an activator of the cyclin B2/Cdk1 complex (Noll et al., 2006). In general, the pericentrosomal accumulation of the IC elements and REs (Marie et al., 2012) could play an important role in the maturation (at G2) and separation (at prophase) of centrosomes, as well as formation of the MT-based mitotic spindle (Hehnly and Doxsey, 2014; Wei et al., 2015; Ibar and Glavic, 2017).

The models in **Figures 2A**, **4** also provide a new perspective to consider the nature of the Golgi checkpoint operating at the G2/M transition. Instead of overseeing the unlinking of the presumably transient Golgi stacks, this control station could monitor the state of the two permanent membrane systems – the biosynthetic (IC) and endocytic networks – meeting at

the non-compact zones of the Golgi ribbon. In case they are found ready for accurate partitioning (Marie et al., 2012), and competent to carry out their mitotic roles, the linker compartments detach from the Golgi ribbon, and relocate to the centrosome. However, if damage is detected (or something is missing), their separation is arrested, and entry into mitosis is delayed. Accordingly, the check-point can control both cell cycle progression and organelle inheritance. This scenario is also compatible with the striking finding that the progression of cells through mitosis is not affected by the presence of BFA (Seemann et al., 2002; Nizak et al., 2004; Marie et al., 2012; **Figure 3**). Although BFA disassembles the Golgi stacks, it allows the linker compartments to detach, partition properly in parallel with centrosome separation, and evidently also support basic trafficking and signaling events that take place during mitosis. Thus, besides their initial unlinking, the subsequent mitotic fate of the cisternal Golgi stacks is a secondary issue. Recently, experimental filling of the Golgi lumen with DAB precipitate was shown to allow mitotic entry, but inhibit the disassembly of the Golgi stacks, resulting in mitotic arrest at the spindle assembly checkpoint (SAC; Guizzunti and Seemann, 2016). An alternative explanation is that the function of the linker compartments is also affected by this treatment, as indicated by the inability of the centrosomes to separate properly. Thus, the ensuing damage to these compartments, rather than that of the Golgi stacks, is the reason for SAC activation.

As cells prepare for division, they change both their shape and internal architecture. At late G2, based on the initial disassembly of integrin-based focal adhesions (FAs) they begin to round up. This dramatic alteration in cell shape is transmitted via the cortical actin meshwork and the radiating MT system to the cell center, resulting in the positioning of the centrosome to the geometric center of the cell and redistribution organelles, such as the Golgi apparatus (Champion et al., 2017). Notably, cell cycle progression is also controlled from the distance, as specific FA components move from the cell surface to the centrosome, where they interact with Aurora A, thereby influencing centrosome maturation and mitotic entry (Pugacheva and Golemis, 2006). Such a complex control of Aurora A activation may involve the trafficking and signaling pathways proposed in **Figure 4**, which not only connect the centrosome with the Golgi ribbon, but also with FAs at the cell periphery. The localization of GBF1 and Arf1 to adhesion sites at the leading edge (Mazaki et al., 2012; Schlienger et al., 2015; Busby et al., 2017) and the proposed roles of Rab1 and Rab11 in integrin trafficking and cell adhesion (Wang et al., 2010; Paul et al., 2015) are in accordance with this possibility.

### Coordinating Golgi and Centrosome Positioning

Another possible example of cross-talk between the cell surface, non-compact zones of the Golgi ribbon and the centrosome (**Figure 4**) is provided by the function of the master regulator of cell polarity – the GTPase Cdc42 of the Rho family – during cell migration (Etienne-Manneville, 2004). Activation of Cdc42 at the leading edge of migrating cells triggers actin polymerization, promoting the formation of cellular protrusions and stabilization and anchoring of the plus-ends of MTs at the actin-based cortical filament meshwork of the lamellipodium. Accordingly, a plasma membrane-associated pool of Cdc42 directs the relocation of the centrosome between the nucleus and the leading edge in a MTand dynein-dependent process that is intimately coupled to Golgi repositioning. Due to reorientation of the MT network, post-Golgi and RE carriers are directed to the lamellipodium, setting the stage for cell polarization and migration.

Another pool of Cdc42 is present in the Golgi region where it interacts with COPI coats and GM130, suggesting that it associates – at least partly – with IC/cis-Golgi membranes (Erickson et al., 1996; Wu et al., 2000; Kodani et al., 2009; Baschieri et al., 2014). In addition, EM has shown the predominant localization of Cdc42 to tubulovesicular membranes at the lateral sides of the Golgi stacks (Luna et al., 2002), in line with the possibility that it also associates with the linker compartments at the non-compact zones of the Golgi ribbon. Indeed, Cdc42 can be recruited from this central pool to the cell surface in an MT- and Arf6-dependent manner, indicating its presence in the REs (Osmani et al., 2010; Baschieri et al., 2014; Farhan and Hsu, 2016). Notably, it also functions in dynein-dependent endosome-to-Golgi trafficking (Hehnly et al., 2009), as well as dynein recruitment to COPI-coated ER-to-Golgi (or intra-Golgi) carriers, indicating a role in Golgi positioning (Hehnly et al., 2010). Therefore, it is tempting to speculate that Cdc42 also regulates the dynein-based pericentrosomal accumulation of the linker compartments during cell migration (**Figure 2B**). Furthermore, the concerted actions of the peripheral and central pools of Cdc42 could explain the tight coupling of centrosome and Golgi re-positioning during this process. Additional effects of Cdc42 on actin dynamics (Luna et al., 2002; Hehnly et al., 2010), MT nucleation, bidirectional trafficking and/or the kinetics of anterograde transport at the Golgi ribbon (Park et al., 2015) could also contribute to Golgi repositioning and polarized delivery of membrane to the leading edge of migrating cells (Farhan and Hsu, 2016).

Is its possible that Cdc42 cycles between the cell periphery and the non-compact zones of the Golgi ribbon (**Figure 4**)? Namely, other GTPases of the Rho-family have been suggested to be transferred from the PM to the ERC (Bouchet et al., 2018). Cdc42 could employ the clathrin-independent endocytic pathway regulated by GBF1 and Arf1. Whatever the precise route, such cycling could explain how the PM pool of Cdc42 can regulate centrosome organization (Kodani and Sütterlin, 2008; Kodani et al., 2009; Herrington et al., 2017).

#### A New Role for GRASPs?

In addition to tethering Golgi cisternae into stacks via their ability to trans-oligomerize, the two mammalian GRASP proteins, GRASP55 and GRASP65, have been implicated as key players in the process that links the Golgi stacks into a ribbon (Puthenveedu et al., 2006; Feinstein and Linstedt, 2008; Jarvela and Linstedt, 2014; Veenendaal et al., 2014; Rabouille and Linstedt, 2016; Bekier et al., 2017; Huang and Wang, 2017). As discussed above, the strongest evidence for the latter role comes from the demonstration that phosphorylation of the GRASPs is

required for Golgi fragmentation and entry of cells into mitosis (Ayala and Colanzi, 2017).

Although GRASP65 and GRASP55 are generally referred to as cis- and medial/trans-Golgi proteins, respectively, and there is evidence suggesting that they function separately to link cisternae at the cis- and trans-sides of the stacks (Jarvela and Linstedt, 2014; Rabouille and Linstedt, 2016), their ultrastructural localizations within the Golgi ribbon have not been firmly established. Interestingly, however, besides being found predominantly at the cis- and lateral sides of the Golgi stacks, the single GRASP (dGRASP) in Drosophila S2 cells has been localized by EM to pleiomorphic tubulo-vesicular elements at the ER-Golgi boundary (Kondylis et al., 2005). More recently, super-resolution microscopy placed both mammalian GRASPs to the cis-side of Golgi ministacks, displaying a localization similar to that of Rab1 (Tie et al., 2018). Moreover, both GRASPs co-localize extensively with Rab1 (Marie et al., 2012; own unpublished data), and there is also previous evidence suggesting IC localization of GRASP65 (Marra et al., 2001).

Notably, at metaphase – following disassembly of Golgi stacks – GRASP65 is found at the spindle poles (Marie et al., 2012). Therefore, the model of Golgi fragmentation during G2/M transition (**Figure 2B**) opens the possibility that GRASP65 – like Rab1 – is present at the linker regions of the Golgi ribbon. This localization would be compatible with its proposed dual role in assembling the cisternal stacks and linking the Golgi ribbon (Rabouille and Linstedt, 2016; Huang and Wang, 2017), and its function in anterograde trafficking (D'Angelo et al., 2009). It might also explain why GRASP65 depletion accelerates cell surface delivery of certain proteins (such as APP, integrin and CD8), affects protein glycosylation and results in missorting of cathepsin D (Xiang et al., 2013; Bekier et al., 2017; Huang and Wang, 2017). Furthermore, one of the GRASPs might even participate in the tethering of the linker compartments at the non-compact zones. Namely, despite lacking Golgi stacks the yeast S. cerevisiae contains a GRASP ortholog Grh1 (Behnia et al., 2007). Moreover, it seems likely that the tubulovesicular networks that constitute the yeast secretory pathway correspond to the biosynthetic and endocytic networks that meet at the noncompact regions of the mammalian Golgi ribbon (Marie et al., 2008; Jackson, 2009; Saraste and Marie, 2018).

Finally, the new model on the spatial organization and dynamics of the biosynthetic (IC) and endocytic networks proposed in **Figure 4** could also explain the findings showing that the mammalian and fly GRASPs can exert their functions not only at the Golgi, but also close to ERES (Kondylis et al., 2005; Kim et al., 2016), or even at the cell periphery (Schotman et al., 2008).

#### Bypassing the Golgi Stacks

Another argument for placing the GRASPs at the non-compact zones of the Golgi ribbon is their participation in Golgiindependent secretory pathways that an increasing number of proteins employ during their intracellular trafficking (Prydz et al., 2013; Gee et al., 2018). Generally, a set of transmembrane proteins – including receptors, ion channels and adhesion proteins – entering the secretory pathway at the ER can reach the cell surface without passing through the Golgi stacks (Martin et al., 2001; Baldwin and Ostergaard, 2002; Yoo et al., 2002; Marie et al., 2008; Hanus et al., 2016). In addition, certain cytoplasmic proteins lacking a signal sequence for ER translocation can reach the extracellular space by crossing the cell membrane directly, or after inclusion into membrane-bound organelles, which fuse with the PM (Dimou and Nickel, 2018). In a variety of organisms, both types of unconventional secretion may involve GRASPs, either for the departure of proteins from the classical secretory route, or their direct inclusion into transport carriers that are related to autophagosomes (Kinseth et al., 2007; Schotman et al., 2008; Dupont et al., 2011; Manjithaya and Subramani, 2011; Kortvely et al., 2016; Zhao et al., 2019). The diversion of the cystic fibrosis-related chloride channel (CFTR) into a Golgi bypass route takes advantage of components of the autophagic machinery and is stimulated by ER stress (Gee et al., 2011, 2018; Noh et al., 2018). Starvation of yeast cells triggers secretion via CUPS (compartment for unconventional protein secretion), a membrane structure formed by ER-Golgi system with contributions from the endosomal pathway (Cruz-Garcia et al., 2018). Notably, there is a close relationship between GRASP-dependent autophagy and GRASP-dependent unconventional secretion (Subramani and Malhotra, 2013). Some unconventional proteins pass via REs, where Rab11A regulates the secretion of e.g., α-synuclein (Liu et al., 2009; Chutna et al., 2014).

Also concerning anterograde trafficking, proteoglycan protein cores that bypass the Golgi apparatus when Golgi passage is inhibited will appear at the PM without polymerized glycosaminoglycan chains. When the proteoglycan serglycin was expressed in epithelial MDCK cells and Golgi bypass was induced by BFA, apical targeting was maintained, indicating that polarized sorting had already taken place prior to Golgi entry, presumably in the IC (Tveit et al., 2009).

Furthermore, the model in **Figure 4** could help to understand the enigmatic itineraries taken by protein toxins during their retrograde transport from the cell surface to the ER. However, while different ER-destined toxins – such as ricin, Shiga, pertussis, and cholera toxins – all engage endogenous cell components to move retrogradely, they do not depend entirely on the same mechanisms at every step of the way (Sandvig et al., 2013). For example, while Shiga and cholera toxins (B subunit) move via REs to the Golgi apparatus in a retromer- or clathrin/AP-1-dependent fashion, respectively, ricin moves from early endosomes to the Golgi apparatus independently of Rab11 and clathrin (Mallard et al., 1998; Iversen et al., 2001; Matsudaira et al., 2015). Ricin may even reach the ER without encountering Golgi enzymes, most likely by an alternative retrograde Golgi bypass mechanism (Llorente et al., 2003). Since the focus on toxin entry to the Golgi apparatus has been at the trans-side, a number of studies have employed recombinant toxins with a sulfation site to monitor retrograde Golgi passage. Based on its permanent character (Saraste and Marie, 2018), it would be logical to assume that the IC is an obligate way station between the Golgi cisternae and the ER; however, this compartment has generally not been addressed in such studies. Nevertheless, the view that toxins destined for the ER pass through the IC is strengthened, since deletion of Rab1A

or Rab1B, or expression of Rab1 mutants, impairs ricin toxicity (Simpson et al., 1995; Bassik et al., 2013). Also, ricin intoxicates mutant CHO cells (END4), where a typical Golgi apparatus disappears, but the IC seems to remain intact (Bau and Draper, 1993). Knockdown of intracellular phospholipase A1 γ (iPLA1γ), which is localized to the IC/cis-Golgi, blocked the ER delivery of cholera toxin B subunit, but not of Shiga toxin (Morikawa et al., 2009). An intact Golgi ribbon is not required for retrograde toxin trafficking, since Drosophila cells, where the dispersed Golgi apparatus existing as individual or pairs of stacks (Kondylis et al., 2007) are also sensitive to ricin (Pawar et al., 2011).

A further indication that molecules pass from RE to IC during their recycling is provided by studies of the cell surface heparan sulfate (HS) proteoglycan Glypican-1, which enters the endocytic pathway where the HS chains are trimmed down by glycosidases, before the protein scaffold recycles to the early secretory pathway to obtain novel HS chains (Mani et al., 2000).

### GOLGI REMODELING BY DIFFERENTIATED CELLS

Cell differentiation requires major changes in the organization of endomembranes and cytoskeletal filaments, which frequently involve repositioning of the centrosome and the Golgi apparatus, as well as fragmentation of the continuous Golgi ribbon (Yadav and Linstedt, 2011; Sanchez and Feldman, 2017; Wei and Seemann, 2017). Another common denominator of neurons, epithelial and muscle cells is that in the course of their differentiation the centrosome loses its role as the major site of MT nucleation. While in some situations this function is taken over by the Golgi apparatus, in other cases the non-centrosomal sites of MT nucleation remain enigmatic (Nishita et al., 2017). In the following we address some of the subcellular rearrangements that accompany the differentiation of these three cell types, with special focus on the IC and endosomal networks and their proposed role in defining Golgi positioning (**Figure 5**).

#### Neurons

The repositioning of the centrosome and the Golgi ribbon within the cell body plays a key role in the early stages of neuronal differentiation. Initially, this process is important for axon specification by ensuring polarized trafficking to the developing axon (de Anda et al., 2005), but it is also required for the formation of dendrites (Horton and Ehlers, 2004). In many neuronal cells the somatic Golgi ribbon faces the primary dendrite and even enters its proximal portion. However, the formation of Golgi outposts (GOPs) at the first branchpoints of the primary dendrite may not be due to dispersal of Golgi ministacks from the cell body, but rather to the movement of the IC elements and REs toward the cell periphery, in this case the growth cones of developing axons and dendrites (Sannerud et al., 2006; Eva et al., 2010; Matsuzaki et al., 2011; **Figure 5A**). This idea is supported by live imaging studies showing that thick tubules containing cis- or trans-Golgi markers move from the cell body to the dendrite, evidently having the capacity to form the GOPs (Quassollo et al., 2015). Of note, the latter have also been implicated in the formation of the dendritic MT network consisting of filaments of variable polarity (Ori-McKenney et al., 2012; **Figure 5A**).

In contrast to GOPs, the IC elements and REs are present throughout the dendritic tree (Hanus et al., 2014; Bourke et al., 2018; **Figure 5A**). At the level of the synapses these elements associate with ERES to establish local secretory units, also referred to as "secretory satellites" (Gardiol et al., 1999; Pierce et al., 2001; Hanus and Ehlers, 2016; Bowen et al., 2017; **Figure 5A**, inset). Strikingly, it turns out that hundreds of locally synthesized transmembrane glycoproteins – including neurotransmitter receptors, ion channels and neuronal adhesion proteins – can reach the synaptic PM with their glycans in the high-mannose form. These proteins employ a BFA-resistant Golgi bypass route across the ERES-IC-RE units (Hanus et al., 2016; Bowen et al., 2017), which has also been suggested to include Golgilike components (Mikhaylova et al., 2016). Nonetheless, these findings open the possibility that Golgi bypass is not limited to the transport of selected proteins under special circumstances, but represents a basic mechanism for cell surface delivery of proteins and lipids (Prydz et al., 2013). In supporting dendritic compartmentalization and synaptic function, the IC elements and REs could also act as sites of MT nucleation (**Figure 5A**), since γ-tubulin is found throughout the dendritic tree, while GOPs are restricted to its proximal parts (Nguyen et al., 2014).

Interestingly, in Drosophila neurons mutations in genes encoding key transport machinery proteins – Sec23 (COPII), Sar1 and Rab1 – cause defects in dendritic rather than axonal morphology, showing that the growing dendrites preferentially depend on a functional early secretory pathway (Ye et al., 2007). Indeed, axons do not contain Golgi outposts (González et al., 2018), pointing to the possibility that the early secretory compartments (ERES and IC) in axons and dendrites differ in their overall organization or activity. Similarly, overexpression of GRASP65 exerts a preferential effect on the outgrowth of dendrites (Horton et al., 2005). Since GRASP65 most likely localizes to the IC and could even mediate the connection between the IC and endosomal networks (see above), its overexpression not only causes fragmentation of the Golgi ribbon in the cell body, but may lead to dysfunction of the IC elements at the neuronal periphery.

#### Skeleletal Muscle Cells

During myogenesis, as mononuclear myoblasts differentiate into multinuclear myofibers, the centrosome undergoes dramatic reorganization as pericentriolar material – including the centrosomal proteins γ-tubulin and pericentrin – first redistributes to the periphery of the nuclei and then to a multitude of sites throughout the cell body (Tassin et al., 1985a; Zaal et al., 2011; Oddoux et al., 2013). The ensuing change in the pattern of MT nucleation is accompanied by fragmentation of the Golgi ribbon and a major reorganization of the Golgi stacks. While myoblasts contain a typical juxtanuclear Golgi next to the centrosome, during myogenesis Golgi elements first circle the nuclei and then are found as dispersed small cisternal stacks throughout the cell body (Ralston, 1993; Tassin et al., 1985b). However, like all membrane compartments in the skeletal muscle

FIGURE 5 | Role of the biosynthetic and endocytic networks in the organization of endomembranes and Golgi positioning in differentiated cell types. In all cells the centrosomal (or centrosome-derived) and non-centrosomal (Golgi-nucleated) MTs with plus-minus polarity are indicated by orange and brown color, respectively. The IC elements (dark green) and REs (light green) are also depicted by different colors. (A) Highly schematic model of a neuron with its cell body, axon and dendritic tree. Golgi stacks (blue) are present in the cell body (Golgi ribbon) and in the proximal branchpoints of the dendritic tree (Golgi outposts), but are lacking from axons. By contrast, in addition to being present in the cell body, IC elements and REs are found throughout the neuronal periphery. The blow-up highlights synapses with local secretory ERES-IC-RE units. Whether axons contain similar structures is presently unclear. (B) Schematic diagram of a small portion of a long multinucleated skeletal muscle cell. In a terminally differentiated myofiber small Golgi outposts are found in the nuclear periphery and – together with ERES – at specific sites within the myofibrillar system. These sites, which function in the nucleation of longitudinal and vertical bundles of non-centrosomal (Golgi-nucleated) MTs most likely contain also IC elements and REs. (C) In a polarized epithelial cell the tight junctions divide the PM into apical and basolateral domains. The apical PM further consists of ciliary and non-ciliary subdomains. In epithelial cells, as in neurons and muscle cells, the centrosome loses its major role as MT-organizing center and (in this case) forms a basal body at the base of the primary cilium. This function is taken over by sub-apical nucleation sites which, however, remain enigmatic. These sites generate a vertical array of MTs typical for the polarized epithelial cell. The apical region may also contain a lateral array of MTs of mixed polarity (not shown). Moreover, a sub-population of non-centrosomal MTs are nucleated by the Golgi apparatus and grow apically. Rab11-containing apical recycling endosomes (AREs) pile-up at the minus ends of the vertical MTs. Similarly as in the pericentrosomal region of a fibroblastic cell (see Figure 4), a pool of IC elements are proposed join the REs at this location. This conclusion is supported by the existence of a circular membrane compartment at the base of the primary cilium, which is known to contain Rab11 and the IC/cis-Golgi protein GM130.

cells, the Golgi elements are also precisely positioned within the myofibrillar network, residing at the intersections of the longitudinal and vertical MT bundles that run across the cells (**Figure 5B**). Indeed, the Golgi elements have been implicated in the nucleation of the filaments that form of the stationery MT lattice typical for myofibers (Oddoux et al., 2013).

Currently, two alternatives have been considered regarding the nature of the ERES-Golgi units of myofibers. Either the Golgi ministacks emerging at these sites are formed de novo, due to recycling of Golgi components via the ER, or correspond to pre-existing elements derived from the fragmented Golgi ribbon that redistribute throughout the muscle cells (Lu et al., 2001; Zaal et al., 2011; Giacomello et al., 2019). As a compromise, we propose that repositioning of the permanent IC and endosomal networks provides the driving force for the rearrangement of the endomembrane system in muscle cells, including formation of the small Golgi stacks. First, there is evidence that both IC elements and endosomes are present at the same MT crossroads sites, where the Golgi elements reside (Rahkila et al., 1997; Kaisto et al., 1999; **Figure 5B**). Second, the IC/cis-Golgi proteins p115 and GM130 – both Rab1 effectors – have been shown to act as master regulators in the organization of early secretory compartments during myogenesis (Giacomello et al., 2019). Third, the nucleation of MTs in mature myofibers is not affected by BFA (Oddoux et al., 2013), raising the possibility that it is accomplished by the drug-resistant IC elements and REs.

#### Epithelial Cells

Epithelial Madin-Darby canine kidney (MDCK) cells grown on filters initially show subcellular organization similar to that of fibroblasts, where the Golgi apparatus and the centrosome localize to one side of the nucleus, and the centrosome-nucleated MTs make up a radial array (**Figure 4**). Early 3-D studies employing confocal microscopy and EM revealed that as a tight and polarized epithelial monolayer is established, both organelles relocate to underneath the apical membrane (Bacallao et al., 1989; Buendia et al., 1990). At the same time, MTs reorganize into noncentrosomal, vertical arrays with their minus ends anchored in the apical region and plus ends pointing toward the basal part of the cell (Bacallao et al., 1989; Toya et al., 2016; **Figure 5C**). In addition, a subset of vertically oriented MTs nucleate at the Golgi membranes (Perez Bay et al., 2013; **Figure 5C**). Generally, the MT organization in different epithelial tissues varies considerably,

involving a number of filament-associated proteins, such as CAMSAP3 (Toya and Takeichi, 2016).

Regarding endomembranes, while the outcome of epithelial differentiation on the endosomal system has been well characterized, its effects on early secretory compartments remain enigmatic. For example, it is unclear how the Golgi ribbon of polarized MDCK cells develops resistance to BFA. In the polarized state, separate pools of early endosomes operate in endocytic uptake at the apical and basolateral plasma membrane domains and the two routes meet in a common pool of late endosomes, localized – like lysosomes – on the apical side of the nucleus, with subsequent exchange of internalized cargo taking place within endosomes all around the nucleus (Bomsel et al., 1989). The apical endocytic machinery displays a relatively speaking much higher capacity of recycling and transcytosis of both fluid and membrane than its basolateral counterpart (Bomsel et al., 1989; Prydz et al., 1992). More recently, apical recycling endosomes (ARE) and common recycling endosomes (CRE) have been added to the picture, both positioned in the apical region on top of the nucleus (Leung et al., 2000). The latter most likely represents the compartment where fluid phase cargo endocytosed from the apical or basolateral surfaces first meet underway to late endosomes and lysosomes (Bomsel et al., 1989; Parton et al., 1989; Wang et al., 2000).

Thus, during differentiation of epithelial cells the endocytic apparatus splits into two systems serving the apical and basolateral domains of the cell. Accordingly, while in nonpolarized MDCK cells different cell surface receptors follow the same recycling route via the Rab11-positive peri-centrosomal ERC, cell polarization involves the development of two compartments specialized into apical and basolateral recycling. Apparently based on its association with the centrosome, one of these compartments – the Rab11-positive ARE – moves to the sub-apical region, while the other – the Rab8-positive CRE, sharing compositional and functional similarity with the trans-Golgi/TGN – remains in the vicinity of the Golgi ribbon (Perez Bay et al., 2016). The permanent connections between the biosynthetic and endocytic networks observed in other cell types raise the possibility that the specialization of the two endocytic recycling circuits of epithelial cells is accompanied by a parallel "duplication" of the IC mediating membrane recycling at the ER-Golgi boundary (**Figure 5C**).

In polarized epithelial cells a primary cilium protruding from the apical membrane is anchored at the basal body, a structure consisting of the mother and daughter centrioles of the centrosome that during cell polarization relocated to the apical membrane, losing most of its pericentriolar material and ability to nucleate MTs (**Figure 5C**). Notably, newly synthesized proteins that are delivered to the ciliary membrane in a Rab11-, Rab8- and exocyst-dependent manner frequently follow pathways that bypass the Golgi stacks (Tian et al., 2014; Bernabé-Rubio and Alonso, 2017; Gilder et al., 2018; Witzgall, 2018). This transport is likely to involve a circular membrane compartment surrounding the base of the cilium that contains Rab11 and the IC/cis-Golgi protein GM130 (Kim et al., 2010; He et al., 2012; Stoops et al., 2015). Looking down from the apical side, the Rab11-positive AREs are normally found throughout the sub-apical region. Notably, however, upon knockout of CAMSAP3 – a protein with a key role in minus-end stabilization and anchoring of the vertical MTs – they accumulate around the basal body, which appears to regain the ability to nucleate a radial array of MTs (Noordstra et al., 2016; Toya et al., 2016). Finally, the periciliary compartment also contains Cdc42, which besides guiding centrosome repositioning during cell migration (see above) and epithelial polarization, is also required for ciliogenesis and ciliary protein trafficking (Wang et al., 2009; Bernabé-Rubio and Alonso, 2017).

#### SUMMARY AND PERSPECTIVES

The present discussion focuses on two interconnected membrane systems, referred to as the biosynthetic and endocytic networks, which play key roles in membrane recycling in eukaryotic cells. Due to their ability to move bidirectionally along MT tracks, these membrane structures can assume wide cellular distributions and provide an essential link between the cell periphery and center in metazoans. Besides operating as a template for the biogenesis of the Golgi stacks, these networks may constitute a basic membrane system that plays an important role in trafficking and signaling events during different phases of the cell cycle; for example during cell division, when many trafficking events mediated by the classical protein coats (clathrin, COPI, COPII) appear to be compromised. Previously, the membranes meeting at the non-compact zones of the mammalian Golgi ribbon have been compared with the secretory system of the yeast S. cerevisiae (Marie et al., 2008; Jackson, 2009; Saraste and Marie, 2018).

The proposed model of the Golgi ribbon (**Figure 4**) is in accordance with the autonomous nature of the Golgi apparatus, as well as the ability of the cisternal stacks to form de novo (Emr et al., 2009). The linker compartments could represent a conserved, permanent aspect of the organelle and establish the basis for its autonomy. Moreover, the present model opens for communication across the Golgi stacks, possibly explaining events that occur on both sides of the organelle. One example is provided by Golgi-nucleation of MTs, which involves the formation of filaments at the cis-side of the stacks, and their stabilization at the trans-side (Efimov et al., 2007; Rivero et al., 2009). An interesting possibility is that the "hot-spots" observed in the nucleation of MTs (Sanders et al., 2017), as well as actinand MT-dependent Golgi exit (Miserey-Lenkei et al., 2017), correspond to the non-compact zones of the Golgi ribbon. Another interesting case deals with autophagy, a complex process where both IC elements and REs have been implicated (Longatti and Tooze, 2012; Ge et al., 2013; Mochizuki et al., 2013; Puri et al., 2013). Again, it is possible that the linker compartments provide the "openings" that allow, for example, the transfer of key transmembrane protein Atg9 from the trans-Golgi/RE system via the IC to the site of autophagosome formation (Lamb et al., 2015; Imai et al., 2016; Mattera et al., 2017; Davies et al., 2018). Interestingly, the induction of autophagy results in the unlinking of the Golgi ribbon (Takahashi et al., 2011; Gosavi et al., 2018).

Regarding the various models on intra-Golgi trafficking, the present view on the organization of the Golgi ribbon (**Figure 4**)

appears to be most compatible with the dynamic cisternal progression or maturation models, rather than the ones assuming vesicular transport between stable Golgi compartments (Glick and Luini, 2011). However, by providing a pathway for the passage of large-sized cargo molecules across the Golgi ribbon the present model of linker compartments would also be in line with the rim progression model of Rothman and coworkers, according to which the dilated rims and central portions of the stacks differ in their dynamics (Lavieu et al., 2013; 2014). Moreover, the proposed existence of permanent biosynthetic (pre-Golgi) and endocytic (post-Golgi) compartments at the two sides of the transient Golgi stacks could clarify a major discrepancy between the cisternal maturation model and the rapid-partitioning model based on the observation that the exit of various cargo from the Golgi typically follows exponential kinetics (Patterson et al., 2008). Thus, the permanent pre- and post-Golgi compartments could represent sites for the formation and fusion of the Golgi cisternae, respectively (Saraste and Kuismanen, 1992). Finally, according to the rapid-partitioning model the Golgi stacks are divided into processing and exit domains, a functional scenario that was previously incorporated in a two-domain structural model of the Golgi ribbon (Jackson, 2009), which has been elaborated further here.

Based on their permanent and dynamic nature, we propose here a dominant role for the biosynthetic and endocytic networks in Golgi positioning during cell division, migration and differentiation. In addition, this perspective is relevant for considering the events that take place in cells treated with MT-disruptive drugs, such as nocodazole, where the MT-based mobility of membranes is blocked and the Golgi ribbon is replaced by dispersed ministacks (Thyberg and Moskalewski, 1999). Two alternatives have been put forward to explain what happens during the drug treatment (when Golgi stacks are relocated to ERES) and wash-out (when a central Golgi ribbon is rapidly re-established). According to one the Golgi enzymes are temporarily redistributed to the ER (Cole et al., 1996a), whereas the other maintains that they remain within the Golgi proper (Pecot and Malhotra, 2006). Common to these models is that they both regard the Golgi stacks as mobile entities, which during drug wash-out can move along MTs to the cell center (Miller et al., 2009). The present considerations open a third possibility, namely that Golgi residents in both situations are redistributed to the biosynthetic (IC) and endosomal networks. How these networks receive them and also maintain their dynamics in the absence of MTs – possibly with the help of an actin-based system? – remain topics of future studies employing e.g., live imaging, super-resolution microscopy or correlative LM-EM.

Finally, the proposed model of the Golgi ribbon (**Figure 4**) prompts new thoughts regarding the development of

#### REFERENCES

endomembranes. Could it be that during evolution, to speed up trafficking and signaling between the periphery and center of the large-sized metazoan cells, their endomembrane system gradually established a connection with the newly arrived MT-system radiating from the centrosome. The centripetal movements of the primordial biosynthetic and endocytic compartments toward the centrosome, and their subsequently acquired capabilities to form cisternal stacks – representing an "annex" for efficient protein modification – could then explain the development of the ribbon-like organization of the Golgi in these cells. The dynamic connection between the Golgi ribbon and the centrosome also allowed the former to develop MT-nucleating activities over time. Consequently, the combined and variable roles of the centrosome and the Golgi in MT-nucleation could account for the diversity of Golgi structures seen in different metazoan cells. For example, lymphoid cells with a predominantly radial array of MTs form a circular Golgi around the centrosome, whereas fibroblasts displaying more active Golgi-based MT-nucleation build a more expanded and irregular perinuclear Golgi ribbon (Rios and Bornens, 2003). Highly differentiated cells again can employ and develop this toolbox to build amazing architectures to fulfill their functional needs (**Figure 5**).

Could the phylogeny of the Golgi ribbon be revealed when looking at the cells of early embryos? Strikingly, during early embryonic development in zebrafish, Golgi markers initially display a dispersed and punctate pattern. Around midgastrulation, the Golgi apparatus condenses in some cell types – like in the most superficial epithelial cells of the gastrula – evidently forming a ribbon, while other cell types maintain a dispersed pattern until later in development (Sepich and Solnica-Krezel, 2016). The morphological variation observed shows a potential correlation between more dispersed Golgi elements and a shorter cell cycle.

#### DATA AVAILABILITY

All data analyzed for this study are included in the manuscript and the supplementary files.

#### AUTHOR CONTRIBUTIONS

Both authors confirm being the sole contributors of this work and have approved it for publication.

#### ACKNOWLEDGMENTS

We thank The Fridtjof Nansen Fund for economical support.

Bacallao, R., Antony, C., Dotti, C., Karsenti, E., Stelzer, E. H., and Simons, K. (1989). The subcellular organization of Madin-Darby canine kidney cells during

Altan-Bonnet, N., Sougrat, R., and Lippincott-Schwartz, J. (2004). Molecular basis for Golgi maintenance and biogenesis. Curr. Opin. Cell Biol. 16, 364–372. doi: 10.1016/j.ceb.2004.06.011

Ayala, I., and Colanzi, A. (2017). Mitotic inheritance of the Golgi complex and its role in cell division. Biol. Cell 109, 364–374. doi: 10.1111/boc.2017 00032

the formation of a polarized epithelium. J. Cell Biol. 109, 2817–2832. doi: 10.1083/jcb.109.6.2817


transition through phosphorylation of GRASP65. J. Cell Sci. 128, 2249–2260. doi: 10.1242/jcs.164871








**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 Saraste and Prydz. 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.

# The Golgin Protein Giantin Regulates Interconnections Between Golgi Stacks

Ayano Satoh<sup>1</sup> \* † , Mitsuko Hayashi-Nishino<sup>2</sup>† , Takuto Shakuno<sup>3</sup> , Junko Masuda<sup>1</sup> , Mayuko Koreishi<sup>1</sup> , Runa Murakami<sup>1</sup> , Yoshimasa Nakamura<sup>4</sup> , Toshiyuki Nakamura<sup>4</sup> , Naomi Abe-Kanoh4,5, Yasuko Honjo<sup>6</sup> , Joerg Malsam<sup>7</sup> , Sidney Yu<sup>8</sup> and Kunihiko Nishino<sup>2</sup>

<sup>1</sup> Graduate School of Interdisciplinary Science and Engineering in Health Systems, Okayama University, Okayama, Japan, 2 Institute of Scientific and Industrial Research, Osaka University, Osaka, Japan, <sup>3</sup> Graduate School of Natural Science and Technology, Okayama University, Okayama, Japan, <sup>4</sup> Graduate School of Environmental and Life Science, Okayama University, Okayama, Japan, <sup>5</sup> Department of Public Health and Applied Nutrition, Institute of Biomedical Sciences, Graduate School Tokushima University, Tokushima, Japan, <sup>6</sup> Research Institute for Radiation Biology and Medicine, Hiroshima University, Hiroshima, Japan, <sup>7</sup> Center for Biochemistry (BZH), Heidelberg University, Heidelberg, Germany, <sup>8</sup> School of Biomedical Sciences, The Chinese University of Hong Kong, Hong Kong, Hong Kong

#### Edited by:

Yanzhuang Wang, University of Michigan, United States

#### Reviewed by:

Martin Lowe, The University of Manchester, United Kingdom Suzanne Pfeffer, Stanford University, United States Nobuhiro Nakamura, Kyoto Sangyo University, Japan

\*Correspondence:

Ayano Satoh ayano113@cc.okayama-u.ac.jp †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

> Received: 18 March 2019 Accepted: 29 July 2019 Published: 27 August 2019

#### Citation:

Satoh A, Hayashi-Nishino M, Shakuno T, Masuda J, Koreishi M, Murakami R, Nakamura Y, Nakamura T, Abe-Kanoh N, Honjo Y, Malsam J, Yu S and Nishino K (2019) The Golgin Protein Giantin Regulates Interconnections Between Golgi Stacks. Front. Cell Dev. Biol. 7:160. doi: 10.3389/fcell.2019.00160 Golgins are a family of Golgi-localized long coiled-coil proteins. The major golgin function is thought to be the tethering of vesicles, membranes, and cytoskeletal elements to the Golgi. We previously showed that knockdown of one of the longest golgins, Giantin, altered the glycosylation patterns of cell surfaces and the kinetics of cargo transport, suggesting that Giantin maintains correct glycosylation through slowing down transport within the Golgi. Giantin knockdown also altered the sizes and numbers of mini Golgi stacks generated by microtubule de-polymerization, suggesting that it maintains the independence of individual Golgi stacks. Therefore, it is presumed that Golgi stacks lose their independence following Giantin knockdown, allowing easier and possibly increased transport among stacks and abnormal glycosylation. To gain structural insights into the independence of Golgi stacks, we herein performed electron tomography and 3D modeling of Golgi stacks in Giantin knockdown cells. Compared with control cells, Giantin-knockdown cells had fewer and smaller fenestrae within each cisterna. This was supported by data showing that the diffusion rate of Golgi membrane proteins is faster in Giantin-knockdown Golgi, indicating that Giantin knockdown structurally and functionally increases connectivity among Golgi cisternae and stacks. This increased connectivity suggests that contrary to the cis-golgin tether model, Giantin instead inhibits the tether and fusion of nearby Golgi cisternae and stacks, resulting in transport difficulties between stacks that may enable the correct glycosylation of proteins and lipids passing through the Golgi.

Keywords: Golgi, golgins, glycosylation, endoplasmic reticulum, electron tomography

# INTRODUCTION

Eukaryotic cells have various forms of glycans on their cell surfaces that are important for cell–cell communications, development, differentiation, infection, and signaling. Most of these glycans are attached to lipids and proteins initially in the endoplasmic reticulum (ER) and are further extended and trimmed in the Golgi apparatus.

Glycan structures vary depending on cell types and species. One of the reasons for such structural variation is differences in the expression patterns of glycosyltransferases and glycosidases responsible for glycan biosynthesis, although it may also depend on the structure of the Golgi (Koreishi et al., 2013a; Xiang et al., 2013). The Golgi apparatus is usually a pancake-like structure consisting of a stack of several flat membrane cisternae, which is further linked laterally and forms a ribbon-like structure [Golgi ribbon (Wei and Seemann, 2010; Gosavi and Gleeson, 2017)]. The stacking of Golgi cisternae is secured by Golgi stacking proteins GRASP55/65, and their loss disrupts Golgi stacking and accelerates cargo transport without affecting the lateral linking (ribbon formation) of the stacks (Xiang et al., 2013). Glycan analyses showed that the loss of GRASP55/65 decreased the cell surface expression of high-mannose- and complex-type glycans, which are formed in the Golgi (Xiang et al., 2013). The authors of this study proposed that GRASP55/65 may slow cargo transport down to ensure correct glycosylation occurs through the Golgi by maintaining Golgi stacks in normal cells (Xiang et al., 2013).

Similar to this study, we previously reported that loss of Giantin, the longest golgin which is a family of Golgi-localized coiled-coil proteins, also accelerates cargo transport by affecting lateral linking of the stacks (Koreishi et al., 2013a). Lectin staining showed that the loss of Giantin increased the proportion of highly branched glycans with sialic acids, representing complex-type glycans (Koreishi et al., 2013a). The key difference between our finding and that of Xiang et al. with regard to glycan structures is whether lateral stack linking is also affected.

Giantin, also known as GCP364, was originally identified as a C-terminally anchored Golgi membrane protein (Linstedt and Hauri, 1993; Linstedt et al., 1995; Toki et al., 1997). It is also recognized as an autoantigen in autoimmune diseases (reviews Stinton et al., 2004; Saraste, 2016). Golgins share predicted coiledcoil structures known to form long rod-like structures, which function in tethering vesicles, membranes, and other cytoskeletal factors. Giantin is thought to tether coatomer protein I (COPI) coated vesicles to cis-Golgi membranes (Sönnichsen et al., 1998; Lesa et al., 2000; Linstedt et al., 2000; Shorter and Warren, 2002, the cis-golgin tether model). In this model, Giantin on COPI vesicles and golgin GM130 in cis-Golgi membranes are linked by a soluble protein, p115; therefore, COPI vesicles are tethered to cis-Golgi. Although Giantin localizes to both the Golgi and COPI vesicles, it is distributed asymmetrically, with more seen in COPI vesicles (Sönnichsen et al., 1998).

To understand the cis-golgin tether model, we previously performed RNA interference (RNAi) of Giantin. Without the tether, it was presumed that cells would be filled with untethered vesicles that may cause incorrect vesicle transport. Anterograde transport was accelerated and sialic acid-bearing cell surface glycans were increased by the loss of Giantin. However, we observed surprisingly few vesicles, which were likely untethered vesicles. This suggested that Giantin has other functions than the COPI vesicle tether. Indeed, we detected an alteration of lateral linking of the stacks following the loss of Giantin (Koreishi et al., 2013a). To gain insights into the structural alteration, which may have caused transport alteration, we herein performed 3D modeling of Golgi cisternae and stacks, and found them to be tightly linked laterally by Giantin depletion. This finding is supported by the observed increased lateral diffusion of Golgi membrane proteins.

#### MATERIALS AND METHODS

### Cell Lines, Culture, Small Interfering (si)RNA Transfection, and Fluorescence Recovery After Photobleaching (FRAP) Experiments

HeLa cells (CCL-2, ATCC, Manassas, VA) were maintained and transfected with Giantin siRNA as described previously (Koreishi et al., 2013a). HeLa cells stably expressing murine Golgi mannosidase II (ManII)-GFP (a gift from J. White, European Molecular Biology Laboratory, Heidelberg, Germany) were established as previously described (Maday et al., 2008; Koreishi et al., 2013b). FRAP experiments were performed as described in earlier studies (Rutz et al., 2009; Koreishi et al., 2013b). The relative diffusion rate and % maximum recovery were obtained by fitting the FRAP data to Ellenberg's diffusion equation (Ellenberg et al., 1997) shown as below.

%FRAP = (%maximum recovery) <sup>∗</sup> (1− √ ((diffusion rate)/ ((diffusion rate) + π ∗ (time after recovery))))

#### Conventional Electron Microscopy

Conventional electron microscopy was performed as previously described (Hayashi-Nishino et al., 2009) with slight modifications. In brief, cells were fixed in 2.5% glutaraldehyde in 0.1 M sodium phosphate buffer, pH 7.4 (PB) for 1 h. Cells were washed in PB, then scraped and collected as pellets. These were post-fixed in buffer containing 1% OsO<sup>4</sup> and 0.5% potassium ferrocyanide, dehydrated in a series of graded ethanol solutions, followed by propylene oxide, and embedded in epoxy resin.

#### Electron Tomography

Cell specimens prepared as above were cut into 300-nm thick sections, collected on formvar/carbon-coated grids, stained with uranyl acetate and lead citrate, and examined using a JEM-2100 transmission electron microscope (The Japan Electron Optics Laboratory Co., Ltd. (JEOL; Tokyo, Japan) at accelerating voltages of 200 kV. Tilt series data for each section were recorded around two orthogonal axes (1◦ interval over ± 60◦ for each axis) using a 2k × 2k CCD camera (Gatan US1000, Gatan Inc., Warrendale, PA). Tomograms were computed and joined to form a dual-axis tomogram using the IMOD software package. Subcellular structures within the 3D volume were segmented, and their surfaces were modeled with IMOD (Kremer et al., 1996; Mastronarde, 1997). Contours of membranes of the Golgi apparatus and other membrane structures visible at the Golgi regions were traced manually to generate the final models. Vesicles were represented by spheres.

#### Cell Cycle Analysis

Cell cycle analysis was performed as described (Liu et al., 2017). Briefly, HeLa cells with or without Giantin RNAi were treated

with trypsin-EDTA and fixed with ice-cold 70% ethanol overnight at –20◦C. After washing with phosphate-buffered saline, cells were stained with 0.1 mg/ml propidium iodide (Thermo Fisher Scientific, Waltham, MA, United States) containing 0.1 mg/ml RNase A for 30 min. DNA contents in stained cells were measured by a Tali image-based cytometer (Thermo Fisher Scientific) or a BD Accuri C6 flow cytometer (Becton, Dickinson and Company, Franklin Lakes, NJ, United States).

#### RNA Isolation and Quantitative (q)PCR

Total RNA was isolated using the Total RNA Extraction Kit (Viogene, New Taipei City, Taiwan) and reverse-transcribed using the SuperScript VILO cDNA Synthesis Kit (Thermo Fisher Scientific). Alternatively, total RNA extraction and reverse transcription were performed using the SuperPrep <sup>R</sup> Cell Lysis & RT Kit for qPCR (Toyobo, Tokyo, Japan). The relative expression of mRNAs was quantified using the LightCycler <sup>R</sup> 480 with SYBR Green I Master (Indianapolis, IN). Primer sets used were as follows: human beta-actin fwd: 5 0 -CCAACCGCGAGAAGATGA-3<sup>0</sup> and human beta-actin rev: 5 0 -TCCATCACGATGCCAGTG-3<sup>0</sup> ; human GALNT5 fwd: 5<sup>0</sup> - AGAGCCATTGAAGACACCAGA-3<sup>0</sup> and human GALNT5rev: 5 0 -CACTGGTGGTTGGGAGGTTA-3<sup>0</sup> ; human GALNT5 fwd2: 5<sup>0</sup> -CCAGTGGATAGAGCCATTGAA-3<sup>0</sup> and human GALNT5 rev2: 5<sup>0</sup> -TGGTTGGGAGGTTATTGTGAA-3<sup>0</sup> ; human ST6GALNAC3 fwd: 5<sup>0</sup> -CACAGAGAAGCGCATGAGTTA-3<sup>0</sup> and human STGALNAC3 rev: 5<sup>0</sup> -TCTTCAAGGCGTGAACA AAA-3<sup>0</sup> ; human EXTL1 fwd: 5<sup>0</sup> -TGTGAGCAAGACCCTGGAC-3 0 and human EXTL1 rev: 5<sup>0</sup> -CCAGAGATGAGGCAGAAGGT-3 0 ; and human ST6GAL2 fwd: 5<sup>0</sup> -TCATCCTAAATTTATA TGGCAGCTC-3<sup>0</sup> and human ST6GAL2 rev: 5<sup>0</sup> -TGAGGA TGCCCAAGCAGT-3<sup>0</sup> .

#### Statistical Analysis

All statistical comparisons in this study were performed using the Student's t-test for independent samples or a non-parametric test, Mann-Whitney U Test calculated by an online calculator: https: //www.socscistatistics.com/tests/mannwhitney.

#### RESULTS

#### Loss of Giantin Elongates Golgi Cisternae

In many mammalian cells, Golgi stacks are connected as ribbon-like structures that are dependent on the integrity of microtubules, the depolymerization of which by Nocodazole disperses the ribbon-like Golgi into mini Golgi stacks that can be observed by light microscopy. Our previous work using Nocodazole suggested that the connectivity between Golgi stacks may be altered by the loss of Giantin (Koreishi et al., 2013a). To gain insights into these changes associated with the loss of Giantin, we performed conventional electron microscopy of Giantin siRNA-treated and control cells (**Figure 1A**). All the efficiencies of Giantin knockdown shown in this study were

>90% as estimated by immunofluorescence (**Supplementary Figure S1**). Compared with control cells, Giantin siRNA-treated cells had Golgi cisternae that were around 1.5 times longer (**Figure 1B**). **Supplementary Figure S2** depicts the distribution of the cisternal lengths as a histogram. Because there was no significant change in the numbers of Golgi cisternae per stack following Giantin RNAi (**Supplementary Figure S3**), the observed elongation is likely the result of lateral fusions of Golgi cisternae rather than increased membranes in Golgi areas. Importantly, other siRNA to Giantin (siRNA5) also had the similar effect on cisternal lengths, and exogenous expression of rat Giantin, which is resistant to siRNAs used, reduced the effect partly (**Supplementary Figure S4**).

#### Loss of Giantin Connects or Fuses Golgi Cisternae and Stacks

To investigate the lateral fusion of Golgi cisternae in living cells, we performed FRAP of the Golgi membrane protein ManII-GFP (Storrie et al., 1998; Ward et al., 2001). As shown in **Figure 2**, FRAP of ManII-GFP in Giantin siRNA-treated cells was much faster than that in control cells. The FRAP data were curvefitted to Ellenberg's diffusion equation revealing that the relative diffusion rate of ManII-GFP in Giantin-siRNA-treated Golgi was 2.3-fold compared to the control. These data suggested that Golgi proteins in Giantin siRNA-treated Golgi move faster and more readily than in control cells, indicating that the connectivity between Golgi cisternae and stacks was functionally increased by the loss of Giantin. Importantly, other siRNA to Giantin (siRNA5) also had the similar effect on FRAP, and exogenous

expression of rat Giantin, which is resistant to siRNAs used, reduced the effect partly (**Supplementary Table S1**).

For further structural analysis, we carried out 3D modeling of electron tomograms of Golgi cisternae/stacks following the loss of Giantin. Approximately 10 tomograms were used for 3D modeling by IMOD software for 3D reconstruction of EM serial sections. Compared with Golgi cisternae of control cells, those of Giantin siRNA-treated cells were much smoother with fewer and smaller wells [Rambourg and Clermont (1990) and fenestrae which were previously defined as "non-compact regions" (Ladinsky et al., 1999; **Figure 3A**)]. Of note, one of the tomographic slices used in 3D modeling shown in the top panels of **Figure 3A** reveals longer cisternae in Giantin siRNA-treated cells than those in control cells. This is in a good agreement with **Figure 1A**. Examples of tomograms and modeling are shown in **Supplementary Movies** and **Supplementary Figure S5**.

The smoothness of the cisternae was then quantified by comparing the volumes of 3D models without fenestrae to with fenestrae (**Figure 3B**). In this quantification, the difference between cisternal volumes of the actual model and those of the no-fenestra model indicates the volumes of fenestrae. An example of the no-fenestra model is shown in **Supplementary Figure S6**. The volume differences, i.e., the volume of the

siRNA-treated cells. Bar, SD (n = 4, <sup>∗</sup>P < 0.005).

fenestrae, in control cells depended on the positions of the cisternae (C2–C3), but were approximately 20∼30%. In contrast, the differences in Giantin siRNA-treated cells were less than 10%, which is indicative of smoother cisternae. We found no significant difference in C1 and C4 cisternae because of their

heterogeneity. We also tried to quantify the smoothness by comparing surface areas to volumes. If the structure has a rough surface, the ratio between the surface area and volume would be higher than if the structure was smooth. However, because cisternae appeared to be thicker near the fenestrae (**Supplementary Figure S7**), we concluded that quantification of smoothness in this way is not appropriate (data not shown). The total volumes (including those of un-modeled fenestrae) of C2 cisternae are 0.011 ± 0.006 µm<sup>3</sup> and 0.0097 ± 0.003 µm<sup>3</sup> for control and RNAi cells, respectively. Importantly, the loss of Giantin did not change cell cycle progression (data not shown).

#### DISCUSSION

Our previous data suggested that the Golgi stack loses its independence of individual Golgi stacks following Giantin knockdown, allowing easier transport among Golgi stacks that may be responsible for increased transport and abnormal glycosylation (Koreishi et al., 2013a). In this study, we confirmed by 3D modeling of electron tomograms that Golgi stacks and cisternae have less independence after the loss of Giantin.

Giantin has a proposed function as a cis-Golgin tether which tethers COPI vesicles to cis-Golgi membranes (Sönnichsen et al., 1998; Linstedt et al., 2000; Alvarez et al., 2001). In the present study, we modeled vesicles around Golgi stacks but were unable to see more untethered COPI vesicles after the loss of Giantin (**Supplementary Figure S8**). To test whether the loss of Giantin increases untethered COPI vesicles, we may need to use larger specimens for future modeling.

#### Golgi 3D Structures

3D Golgi modeling has previously revealed novel Golgi structures including connected Golgi cisternae, branched Golgi cisternae, and autophagosome-like Golgi (reviews Mogelsvang et al., 2004; Marsh, 2005; Martínez-Alonso et al., 2013). Among these works, the first and most crucial examination was 3D modeling of the Golgi in normal rat kidney cells (Ladinsky et al., 1999). Importantly, this work quantitatively described non-compact regions that include unconnected cisternae, some vesicles, and fenestrae. We also identified this "non-compact region" between Golgi stacks in HeLa cells in the present study, although few vesicles were observed in this region.

Our comparison of cisternal volumes between actual and nofenestra models revealed a relative volume of the non-compact region in control HeLa cells of around 20% (**Figure 3**). This volume was reduced by the loss of Giantin, suggesting that Giantin plays an essential role in the generation of the noncompact region although the function of this region remains unclear. Extrapolating from our previous work showing that the loss of Giantin increases cargo transport and affects cell surface glycans, we speculate that the non-compact region organizes transport through the Golgi. Further, our previous work by 2D immuno-EM showed that Giantin localizes to the rims of Golgi cisternae that may be part of the fenestrae/non-compact regions. It is noteworthy that depletion of the Golgi-localized small GTPase Rab6 increased the lateral continuity of the Golgi (Storrie et al., 2012), and Rab6 was also shown to interact with Giantin (Rosing et al., 2007). Taken together, Giantin may function together with Rab6 in the maintenance of the noncompact region. Another possibility is that the loss of Giantin reduces fission of the Golgi cisternae or changes in the flux of vesicle transport within the Golgi. Although we are unable to test this possibility, the loss of Giantin did not increase the number of vesicles in the areas observed (**Supplementary Figure S8**), suggesting that these may be comparable with or without Giantin.

Our previous work showed that the loss of Giantin altered lateral connectivity of Golgi stacks using microtubule depolymerizer, Nocodazole. Although this appeared to contradict the present study, the previous study, in particular, the EM study was done in the presence of Nocodazole suggesting that microtubules may contribute to the Giantin-regulated lateral connectivity. From this viewpoint, the Golgi-microtubule associate proteins (GMAPs) together with Giantin may also play a key role in the independence of Golgi cisternae, but we have not identified such GMAPs, yet. Our updated working model is shown in **Supplementary Figure S10**.

#### Giantin and Glycosylation

Our previous data showed that the loss of Giantin altered the glycosylation patterns of cell surface proteins (Koreishi et al., 2013a). Golgins, including Giantin, have been shown to interact with some glycosyltransferases and are required for their targeting to functional sites (Petrosyan et al., 2012). These golgin-interacting glycosyltransferases are responsible for the biosynthesis of mucin-type glycoproteins. We did not test whether the localization of these enzymes or the surface expression of mucin-type glycoproteins is changed after the loss of Giantin, but the observed alterations in glycosylation patterns might reflect the targeting failure of glycosyltransferases.

Another study showed that loss of Giantin by clustered regularly interspaced short palindromic repeats (CRISPR)/CRISPR-associated protein 9 affected glycosyltransferase expression levels in the RPE-1 cell line derived from human retina (Stevenson et al., 2017). We also tested those glycosyltransferase expression levels in HeLa cells after Giantin RNAi, but found no significant changes (**Supplementary Figure S9**). The authors also claimed that the knock-out of Giantin by CRISPR in RPE1 cells did not change the cisternal lengths and cargo transport. Therefore, the phenotypes in cisternal lengths, glycosylation, and transport that we observed by the knock-down of Giantin with siRNAs can be specific to HeLa cells.

#### Cell-Type Specific Functions of Giantin

Giantin is expressed ubiquitously (Fagerberg et al., 2014), and is widely used as a Golgi marker protein. Giantin knock-out or mutant animal models have revealed the cell-type specific function of Giantin (Katayama et al., 2011, 2018; Lan et al., 2016; Bergen et al., 2017; McGee et al., 2017; Stevenson et al., 2017). Moreover, Giantin gene mutations and pathogenic changes in Giantin levels have been reported in various diseases, including sickle cell disease (Alsultan et al., 2018; Amlie-Lefond et al., 2018), bipolar disorder (Kerner et al., 2013), hepatocellular carcinoma

(Choi et al., 2017), leukemia (Troadec et al., 2017), pulmonary disease (Raparelli et al., 2018), and post-alcohol recovery (Casey et al., 2018). These various phenotypes might be caused by the expression of Giantin binding partners, including GCP60 (Sohda et al., 2001), RCAN2 (Stevenson et al., 2018), PRMT5 (Sohail et al., 2015), PLK3 (Ruan et al., 2004), Rab6, p115, and Rab1 (Beard et al., 2005). In the present study, we noticed that immunofluorescence of GCP60 around the Golgi was less distinct after the loss of Giantin (data not shown). Therefore, GCP60 may function together with Giantin in the independence of Golgi stacks.

#### CONCLUSION AND PERSPECTIVES

Our study found that Golgi stacks and cisternae changed their structure following the loss of Giantin, which may correlate with our previous finding of the alteration of glycosylation patterns of cell surface proteins. This is reasonable considering that glycosylation is a series of chemical reactions and that Golgi stacks and cisternae are reaction vessels. In other words, structural changes to reaction vessels by the lateral linking of Golgi cisternae may alter temporal and local concentrations of enzymes and substrates for glycosylation reactions, which may result in changes to surface glycosylation patterns and secretion.

The correlation between structural changes of the Golgi and glycosylation patterns was previously explained as a result of 'sorting zones' in the Golgi (Yano et al., 2005) in which each zone has a unique role in providing glycosylation and sorting. For example, one zone is responsible for making glycosaminoglycans, and the others are responsible for protein N-glycosylation or GPI-anchoring. It could be speculated that these sorting zones are lost following Giantin knockdown, leading to changes in surface glycosylation patterns. Future studies using recent technologies including superresolution microscopy (Tie et al., 2018), cryo-focused ion beam scanning electron microscopy (FIB-SEM) with membrane segmentation by deep learning, and other new methods (Ranftler et al., 2017; Gorelick et al., 2019) will help dissect the structural changes associated with Golgi stacks and cisternae and sorting zones.

#### REFERENCES


# DATA AVAILABILITY

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

#### AUTHOR CONTRIBUTIONS

MH-N, JoM, and AS: electron microscopy. TS, MH-N, KN, and AS: electron tomography and 3D modeling. MK, RM, YH, SY, and AS: giantin RNAi. YN, TN, NA-K, JuM, and AS: cell cycle analysis. JuM and AS: plasmid construction.

# FUNDING

This work was supported in part by MEXT/JSPS KAKENHI grant numbers 23570167 (AS), 26440055 (AS), 17H01214 (AS), 17K08827 (MH-N), 17H06422 (MH-N), 17H03818 (YN), and 18K06133 (AS), the Network Joint Research Center for Materials and Devices (AS, MH-N, and KN), the Naito Foundation (MH-N and KN), and The Futaba Foundation (AS).

# ACKNOWLEDGMENTS

We thank Risa Matsumoto, Chiu Tzu Hsuan, Saki Okamoto, and Yui Shigechika for their technical help. We are grateful to the technical staffs, especially, Iwao Nishimura, of the Comprehensive Analysis Center of the Institute of Scientific and Industrial Research (Osaka University) and Organelle Lab (Okayama University) for their assistance. We also thank Sarah Williams, Ph.D., from Edanz Group (www.edanzediting.com) for editing a draft of this manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2019.00160/ full#supplementary-material


targeting of an inner nuclear membrane protein in interphase and mitosis. J. Cell Biol. 138, 1193–1206. doi: 10.1083/jcb.138.6.1193


cisternal number: linkage between Rab6 and Golgi-coated vesicles/cisternae. Traffic 13, 727–744. doi: 10.1111/j.1600-0854.2012.01343.x


**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 Satoh, Hayashi-Nishino, Shakuno, Masuda, Koreishi, Murakami, Nakamura, Nakamura, Abe-Kanoh, Honjo, Malsam, Yu and Nishino. 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.

# Visualization of Protein Sorting at the Trans-Golgi Network and Endosomes Through Super-Resolution Imaging

Yan Huang1†, Tianji Ma1†, Pik Ki Lau1†, Jinhui Wang<sup>1</sup> , Teng Zhao<sup>2</sup> , Shengwang Du3,4 , Michael M. T. Loy <sup>4</sup> and Yusong Guo1,5 \*

*<sup>1</sup> Division of Life Science, Hong Kong University of Science and Technology, Hong Kong, China, <sup>2</sup> Light Innovation Technology Limited, Hong Kong, China, <sup>3</sup> Department of Chemical and Biological Engineering, Hong Kong University of Science and Technology, Hong Kong, China, <sup>4</sup> Department of Physics, Hong Kong University of Science and Technology, Hong Kong, China, <sup>5</sup> Hong Kong University of Science and Technology Shenzhen Research Institute, Shenzhen, China*

#### Edited by:

*Yanzhuang Wang, University of Michigan, United States*

#### Reviewed by:

*Heike Folsch, Northwestern University, United States Juan S. Bonifacino, Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), United States*

> \*Correspondence: *Yusong Guo guoyusong@ust.hk*

*†These authors have contributed equally to this work*

#### Specialty section:

*This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology*

Received: *16 April 2019* Accepted: *19 August 2019* Published: *03 September 2019*

#### Citation:

*Huang Y, Ma T, Lau PK, Wang J, Zhao T, Du S, Loy MMT and Guo Y (2019) Visualization of Protein Sorting at the Trans-Golgi Network and Endosomes Through Super-Resolution Imaging. Front. Cell Dev. Biol. 7:181. doi: 10.3389/fcell.2019.00181* The *trans*-Golgi network (TGN) and endosomes are essential protein sorting stations in the secretory transport pathway. Protein sorting is fundamentally a process of spatial segregation, but the spatial relationships among the proteins that constitute the sorting machinery have not been systematically analyzed at high resolution in mammalian cells. Here, using two-color STORM imaging, we show that the TGN/endosome-localized cargo adaptors, AP-1, GGA2 and epsinR, form elongated structures of over 250 nm in length at the juxta-nuclear Golgi area. Many of these structures are associated with clathrin. We found that AP-1 is spatially segregated from AP-3 and GGA2, whereas a fraction of AP-1 and GGA2 punctae are associated with epsinR. Moreover, we observed that the planar cell polarity cargo proteins, Vangl2 and Frizzled6 associate with different cargo adaptors—AP-1 and GGA2 or epsinR, respectively—when exiting the TGN. Knockdown analysis confirms the functional significance of this segregation. Our data indicates that TGN/endosome-localized cargo adaptors have distinct spatial relationships. The spatially segregated cargo adaptors GGA2 and AP-1 regulate sorting of Frizzled6 and Vangl2, respectively and spatially associated cargo adaptors can cooperatively regulate a specific sorting process.

#### Keywords: trans-Golgi network, cargo adaptor, sorting, clathrin, storm

#### INTRODUCTION

The trans-Golgi network (TGN) and endosomes are important transport hubs in the secretory transport pathway. To ensure the fidelity of protein transport, elaborate protein sorting machineries are employed to accurately package cargo proteins into specific transport vesicles that are delivered to various downstream compartments. Defects in the cargo sorting process cause protein mis-targeting and give rise to defects in cell polarity, immunity, and regulated secretion (Guo et al., 2014).

The key players that mediate protein sorting at the TGN and endosomes include various cytosolic cargo adaptors. Once recruited onto the membranes, these cargo adaptors recognize sorting motifs in the cytosolic domains of the transmembrane cargo proteins. Some cargo adaptors subsequently recruit clathrin. Polymerized clathrin along with their associated clathrin adaptors form characteristic electron dense membrane coat structures and this process leads to a sequestration of the associated cargo proteins into coated membrane patches (Guo et al., 2014). The assembly of vesicle coat structures also promotes deformations of the lipid bilayer leading to vesicle budding.

Protein sorting is fundamentally a process of spatial segregation: cargo sorting machineries concentrate specific cargo proteins in specific membrane microdomains from which other proteins, such as resident proteins are excluded. Despite its importance, the spatial distributions of the key players that participate in the protein sorting process at the TGN and endosomes have not been systematically analyzed at high resolution in mammalian cells. Moreover, the key step in the protein sorting process, namely the association of a specific cargo molecule with a specific cargo adaptor, has not been intensively studied using high resolution fluorescence microscopy. Addressing these questions will significantly advance our understanding of how cargo sorting machineries function to package specific cargo proteins into transport vesicles at the TGN and endosomes.

Conventional fluorescence microcopy cannot achieve the resolution necessary to distinguish the spatial relationships among Golgi-localized proteins. Immunogold electron microscopy (EM) has demonstrated that the majority of the cargo adaptors, the adaptor protein complex-1 (AP-1) and GGA2, are present in separate clathrin-coated budding profiles in Drosophila Dmel2 cells (Hirst et al., 2009). Immunogold labeling also revealed that the adaptor protein complex-3 (AP-3) and AP-1 are found on distinct buds arising from endosome-associated tubules in HepG2 cells (Peden et al., 2004). This suggests that AP-1 is not present in the same vesicles as GGAs or AP-3. Although immunogold EM analysis provides details of protein localizations at fine resolution, it requires a long procedure of sample preparation. Moreover, the efficiency of immunogold labeling relies on both the antibody quality and the target protein condition in the ultra-thin section, which in practice is usually affected by intricate sample preparation (Griffiths and Hoppeler, 1986; Howell et al., 1987; D'Amico and Skarmoutsou, 2008). Therefore, immunogold labeling for transmission EM generally displays a lower efficiency than that immunostaining for confocal microscopy does in regard to the labeling efficiency. For this reason, immunogold labeling has been used for detailed subcellular localization of the target protein instead of providing a comprehensive view of the spatial distributions of adaptor and cargo protein on the whole organelle.

Here, we utilized two-color stochastic optical reconstruction microscopy (STORM) to provide a wider view of the spatial distributions of proteins that participate in the TGN/endosome sorting process at fine resolution. Two-color STORM imaging simultaneously captures both fluorescent dyes in the same camera frame to achieve a spatial resolution of 20 nm (Zhao et al., 2015). Our set up was equipped with an active sample locking mechanism that effectively abolishes drift into the x-y plane and the z-axis. This allowed the position of the sample to be stabilized to within 1 nm (Zhao et al., 2015). We found that the TGNand endosome- localized cargo adaptors can be assembled into elongated punctate structures of over 250 nm in length at the juxta-nuclear Golgi area. We revealed that the cargo adaptors, AP-1 and GGA2, are spatially segregated and they function to mediate sorting of planar cell polarity proteins, Vangl2 and Frizzled6, respectively. These analyses indicate that spatial segregation of cargo adaptors may contribute to the protein sorting process. In contrast, another cargo adaptor, epsinR, displayed a GGA2 and an AP-1 associated pattern. Both GGA2 and epsinR regulate sorting of Frizzled6, suggesting that spatially associated cargo adaptors can cooperatively regulate sorting of a specific cargo molecule. Our super-resolution imaging studies revealed that Frizzled6 and Vangl2 are spatially associated with different clathrin adaptors, providing a direct observation of the differential cargo sorting process.

# RESULTS

# Analysis of the Spatial Relationships Between Clathrin and

#### TGN/endosome-localized Cargo Adaptors

To test whether two-color STORM can be used to compare the spatial relationships between different Golgi-localized proteins, we analyzed the localization patterns of a cis-Golgi marker, GM130, and a trans-Golgi marker, Golgin-97. Using conventional light microscopy, the localization pattern of GM130 was not clearly distinguishable from that of Golgin-97 (**Figure 1A**). By contrast, with STORM, we observed a clear separation of GM130 from Golgin-97 (**Figures 1B,C**). To ensure that the separation of GM130 from Golgin-97 in the superresolution images was not due to sample drift, we analyzed the localization patterns of the same protein, GM130, labeled with two different dyes, Alexa 750 and Alexa 647. STORM images of Alexa 750 dye-labeled GM130 (GM130 <sup>750</sup>) and Alexa 647 dye-labeled GM130 (GM130 <sup>647</sup>) overlapped almost perfectly (**Figures 1D,E**). Quantification analysis indicated that the percentage of overlapping between GM130 <sup>750</sup> and GM130 <sup>647</sup> was significantly higher than that of GM130 and Golgin-97 (**Figure 1F**). We noticed that GM130 <sup>750</sup> and GM130 <sup>647</sup> were not completely overlapped. A possible explanation is that not all the GM130 proteins are labeled with both Alexa 647 dye and Alexa 750 dye. To ensure that this is not due to a systematic error, we performed experiments using imaging beads coated with both Alexa 750 and Alexa 647. Our results indicate that Alexa 750 and Alexa 647 showed a nearly perfectly overlapped pattern (**Figures S1A–D**). Thus, two-color STORM is a feasible tool to reveal the spatial relationship between different Golgilocalized proteins.

Using two-color STORM, we sought to analyze the spatial relationships between clathrin and cargo adaptors in COS7 cells. COS7 cells are flat so that the Golgi is close to the coverslips for STORM imaging. When antibodies are commercially available, we determined the localization of the endogenous proteins, including clathrin heavy chain (CHC), GGA2, the γ subunit of the adaptor complex-1 (γ1) and the δ subunit of the adaptor complex 3 (AP3δ1). We generated a FLAG-tagged version of epsinR and used antibodies against FLAG to label epsinR after transient transfection. In COS7 cells, CHC, γ1, GGA2, and FLAG are highly clustered at the juxtanuclear area where

the Golgi is localized and some punctate structures in the cell periphery, presumably on endosomes (**Figures S2A–L**,**P–R**). In comparison, the structures positive for AP3δ1 were more spread out (**Figures S2M–O**). Similar patterns labeled were observed in HeLa cells (**Figures S3A,C,E,G,I**). In addition, the fluorescent intensities of the juxtanuclear structures labeled for the cargo adaptor were greatly diminished when cells were transfected with siRNAs against the corresponding protein (**Figures S3B,D,F,H,J**). These siRNAs effectively reduced the expression level of their target proteins (**Figures S3K–M** and **Figure 4Q**) (Ma et al., 2018). This result demonstrates that the signals we detected were specific. We selected the whole juxtanuclear Golgi area where the majority of the cargo adaptors were localized for the following super-resolution imaging analyses.

AP-1 is a major cargo sorting complex at the TGN and it recruits clathrin to membranes during vesicle formation. Super-resolution images revealed that both AP-1, indicated by its γ1 subunit, and clathrin, marked by clathrin heavy chain (CHC), colocalized in many of the punctate structures in the juxtanuclear area (**Figure 2A**, and magnified view in **Figures 2B,C**). In addition to the AP-1 positive structures, clathrin exhibited additional locations, presumably associated with other clathrin adaptors.

We then performed 3-D super-resolution imaging analysis to analyze the spatial relationships between clathrin and AP-1 (**Figures 2D–G'**, **Videos S1–S3**). 3-D super-resolution imaging analysis revealed that AP-1 can form elongated structures with over 250 nm in length (**Figures 2E–F'**, **Videos S1, S2**). Quantification analysis of the super-resolution images indicates that around 46% of AP-1 structures were associated with clathrin (**Figure 2S**). Many of the AP-1 structures that were not associated with clathrin were small structures.

We also performed 3-D super-resolution imaging analysis of the spatial relationships between clathrin and two other cargo adaptors, epsinR and GGA2. Both epsinR and GGA2 also exhibited punctate localization patterns and many of these punctae were associated with clathrin (**Figures 2H–O'**, **Videos S4–S9**). Quantification analysis indicates that around 46% of epsinR and around 44% of GGA2 structures were clathrin-associated (**Figure 2S**). Similar to AP-1, some epsinR and GGA2 punctae were elongated with over 250 nm in length

respectively. (H–K'). COS7 cells were transfected with epsinR-FLAG. Day 1 after transfection, cells were co-stained with antibodies against γ1-adaptin and FLAG tag. 3- Dimensional localization patterns of epsinR-FLAG and clathrin heavy chain in the same COS7 cell were analyzed through two-color STORM. Magnified views of the indicated area in (H) were shown in (I–K'). (L–O') COS7 cells were co-stained with antibodies against clathrin heavy chain and GGA2. Localization patterns of GGA2 and clathrin heavy chain in the same COS7 cell were analyzed by two-color STORM in 3-dimension. Magnified views of the indicated area in (L) were shown in (M–O'). (P–R) COS7 cells were co-stained with antibodies against clathrin heavy chain and AP3δ1. Localization patterns of AP-3 marked by AP3δ1 and clathrin heavy chain in the same COS7 cell were analyzed by two-color STORM in 2-dimension. Magnified views of the indicated area in panel P were shown in (Q,R). (S) Quantification of fractions of AP-1, GGA2, AP-3, or epsinR punctae that were associated with clathrin based on super-resolution images acquired from three independent repetitions in each experimental group (mean ± SD, based on ≥9 super-resolution images). \*\*\**p* < 0.001 by two-tailed Student's *t*-test. Scale bar in each panel was indicated. The super-resolution images showed the localization patterns of the indicated protein in the whole juxtanuclear area where the majority of the cargo adaptors were localized. Each dot in the graphs represents data of the whole juxtanuclear Golgi area of each individual cell.

(**Figures 2I,N**, **Videos S4, S8**). In contrast, the percentage of AP-3 structures, labeled by AP3δ1, showed a significantly reduced association with clathrin at the juxtanulear area (**Figures 2P–R**, and quantification in **Figure 2S**).

If the cargo adaptors and clathrin are assembled on vesicle membranes, they are expected to be localized on ring structures. However, we did not detect clear ring structures labeled by antibodies against cargo adaptors and clathrin in any of the super-resolution images. As a control for our analysis, we performed super-resolution imaging analysis to analyze the localization pattern of clathrin on the plasma membrane of COS7 cells. To induce endocytosis, we treated COS7 cells with epidermal growth factor (EGF) and analyzed the localization of clathrin heavy chain (CHC) 5 min after EGF treatment. We detected ring structures of CHC with the diameter around 100 nm (**Figure S4A** and magnified views shown in **Figures S4B–D**). This analysis indicates that our STORM can detect ring-localization of clathrin around the vesicle coats. A possible explanation for the failure to detect clear ring structures of clathrin at the juxtanuclear TGN area is that the area of plasma membrane that was captured in STORM was flat and attached to the coverslip in 2D, whereas the TGN has 3D shapes. Thus, it is relatively easier to capture the clathrin-coated pits (CCPs) on the plasma membrane than to capture the CCPs at the TGN. Another possibility is that the vesicle coats are unstable and are disassembled after vesicles are formed at the TGN or endosomes. It is also possible that the antigen might be inaccessible in the context of the coat or other crowded environments at the TGN or endosomes. We propose that the punctate structures detected in our super-resolution imaging analyses represent cargo adaptors and clathrin that are assembled on the microdomains on the TGN or the endosomal membranes but not on the vesicle membranes.

## Analysis of the Spatial Relationships Among the TGN/endosome-localized Cargo Adaptors

Next, we analyzed the spatial relationships among the TGNand endosome- localized cargo adaptors at fine resolution. We first analyzed the spatial relationship between AP-1 and AP-3, which are marked by their γ1 and δ1 subunits, respectively. Super-resolution images of AP-1 and AP-3 showed that AP-1 and AP-3 remained largely separated at the juxtanuclear area (**Figures 3A–D** and quantification in **Figure 3I**) suggesting that they may mediate distinct steps in cargo sorting. We then analyzed the spatial relationship between AP-1 and epsinR. EpsinR directly interacts with the appendage domain of AP-1 (Owen et al., 2004) and colocalized with AP-1 when viewed under conventional microscope (Hirst et al., 2003). Superresolution imaging analysis of AP-1 and epsinR showed that some of the epsinR structures were associated with AP-1 structures (**Figures 3E–H**, and quantification in **Figure 3I**) and some of the AP-1 and epsinR structures were separated from one another (**Figures 3E,G**). EpsinR also interacts with the GAE domain of

FIGURE 3 | Spatial relationship analysis between clathrin adaptors at the juxtanuclear area. (A–D,N–Q) COS7 cells were co-stained with antibodies against γ1-adaptin and δ3-adaptin or co-stained with antibodies against γ1-adaptin and GGA2. Two-color STORM was then utilized to visualize localizations of AP-1 and AP-3 (A–D) and the localization of AP-1 and GGA2 (N–Q). (E–H,J–M). COS7 cells were transfected with epsinR-FLAG. Day 1 after transfection, cells were co-stained with antibodies against γ1-adaptin and FLAG tag (E–H) or with antibodies against GGA2 and FLAG tag (J–M). Two-color STORM was then utilized to visualize localizations of γ1 and epsinR-FLAG and the localizations of GGA2 and epsinR-FLAG. The Magnified views of the indicated area in (A,E,J,N) were indicated in (B–D,F–H,K–M,O–Q). The scale bar of each panel was indicated. (I,R) The percentage of AP-1 punctae that were associated with epsinR or AP-3 was quantified (I) and the percentage of GGA2 punctae that were associated with epsinR or AP-1 was quantified (R) (mean ± SD, based on ≥7 super-resolution images). \*\*\**p* < 0.001 by two-tailed Student's *t*-test. AP-1 and AP-3 were labeled by antibodies against γ1 and δ3, respectively. The super-resolution images showed the localization patterns of the indicated protein in the whole juxtanuclear area where the majority of the cargo adaptors were localized. Each dot in the graphs represents data of the whole juxtanuclear Golgi area of each individual cell. Quantification analyses were based on super-resolution images acquired from three independent repetitions for each experimental group.

GGA2. Similar to the spatial relationship between AP-1 and epsinR, we found that some GGA2 punctae were associated with epsinR punctae (**Figures 3J–M**). Quantification analysis indicates that around 41% of the AP-1 structures and 47% of the GGA2 structures were spatially associated with epsinR (**Figures 3I,R**).

We then analyzed the spatial relationships between AP-1 and GGA2. In yeast, these two cargo adaptors are adjacent to, but distinct from one another, and Gga proteins and AP-1 are sequentially assembled at the TGN (Daboussi et al., 2012). In yeast, a key GGA protein, Gga2p, directly binds phosphatidylinositol (PtdIns) 4-kinase (PI4K), Pik1 (Daboussi et al., 2017), which recruit Pik1 to the TGN to produce PtdIns4P and induce a second wave of AP-1 assembly (Daboussi et al., 2012, 2017). The sequential assembly process allows assembly of GGA proteins and AP-1 to be separated in time and space. Mammalian GGA2 also directly interacts with the Pik1 homolog, PI4KIIIβ, and regulates its membrane association (Daboussi et al., 2017). A poor colocalization between GGA1 and AP-1 was detected in mammalian cells (Mardones et al., 2007). Here, we found that around 8% of the GGA2 punctae were spatially

each experiment). \*\**p* < 0.01 by two-tailed Student's *t*-test.

associated with the AP-1 punctae and the majority of AP-1 and GGA2 punctae were separated in COS7 cells (**Figures 3N–Q** and quantification in **Figure 3R**).

## Visualization of Sorting of Planar Cell Polarity Proteins, Vangl2 and Frizzled6, Upon TGN Exit Through Super-Resolution Imaging Analysis

Our analyses indicate that GGA2 and AP-1 are largely spatially segregated from each other suggesting that GGA2 and AP-1 can function to mediate sorting of distinct cargo proteins at the TGN and endosomes. AP-1 has been shown to regulate sorting of a planar cell polarity protein, Vangl2, at the TGN (Guo et al., 2013). TGN sorting of another PCP protein, Frizzle6, which is localized on the cellular boundaries opposing to Vangl2, is independent of AP-1 but dependent on epsinR (Ma et al., 2018). Our super-resolution imaging analyses indicate that epsinR showed an AP-1-associated pattern and a GGA2-associated pattern. This observation prompted us to test whether it is the GGA2-associated epsinR that regulates sorting of Frizzled6 at the TGN. To test this, we first analyzed whether GGA2 is important for TGN export of Frizzled6. Cells treated with siRNA against GGA2 did not show detectable defects in HA-Frizzled6 localization at the steady state, possibly due to the presence of functional redundancy. To test whether knockdown of GGA2 causes a delay in the kinetics of TGN export of Frizzled6, we incubated HeLa cells at 20◦C in the presence of cycloheximide to accumulate newly-synthesized HA-Frizzled6 in the TGN. As the HA tag is exposed at the extracellular domain of Frizzled6, we performed a surface labeling experiment to detect the surface-localized HA-Frizzled6. After a 20◦C incubation, HA-Frizzled6 had accumulated at the juxtanuclear area, colocalized with TGN46 and the majority of cells showed no detectable surface-localized Frizzled6 (**Figures 4A–H**, and quantification in **Figure 4R**). When cells were shifted to 32◦C, Frizzled6 in most control siRNA treated cells showed a detectable surface-localized pattern (**Figures 4I–L**, and quantification in **Figure 4R**). Through this method, we found that cells treated with siRNA against GGA2, which efficiently reduced the expression of GGA2 (**Figure 4Q**), showed a significant reduction of the percentage of cells showing surface-localized Frizzled6 (**Figures 4M–P**, and quantification in **Figure 4R**). In contrast, knockdown of GGA2 did not cause a defect in the TGN export of HA-Vangl2, whereas the majority of cells treated with siRNA against CHC showed strong accumulations of HA-Vangl2 at the juxtanuclear area (**Figure S5**). Using a similar temperature shift approach, we detected a significant reduction of the percentage of cells showing surface-localized Frizzled6 in GGA3 knockdown cells but not GGA1 (**Figure S6**), indicating that GGA3 and GGA2 may redundantly regulate sorting of Frizzled6 at the TGN.

Next we sought to visualize these sorting events using twocolor STORM. We first investigated whether Vangl2 was enriched in membrane patches coated by AP-1 upon TGN exit. COS7 cells expressing HA-Vangl2 were incubated at 20◦C to accumulate newly-synthesized Vangl2 at the TGN. We analyzed the spatial relationship between Vangl2 and AP-1 after the cells were incubated at 32◦C for 5 min to restore the protein sorting process. In this incubation, the extent of exit from the TGN was minimal and the majority of cargo proteins remained localized at or around the juxtanuclear area (**Figure S7**). We selected the juxtanuclear area where the majority of Vangl2 was localized for the super-resolution imaging analysis. STORM images of Vangl2 and AP-1 in 2- and in 3-dimensions showed that many Vangl2 structures are associated with AP-1 (**Figure 5A** and the magnified view in **Figures 5B–E** and the magnified view in **Figures 5F–H'**, **Videos S10–S12**). A broad range in the degree of co-localization was apparent from a quantification of 7 super-resolution images from 7 cells (**Figure 5U**), possibly due to an asynchronous budding of vesicles from the TGN followed by coat protein release from the membrane. As a control, we analyzed the spatial relationship between Vangl2 and AP-3. The majority of Vangl2 structures were not associated with the AP-3 (**Figures 5I,U**). As another control, we analyzed the spatial relationship of AP-1 and Vangl2 bearing mutations in the tyrosine sorting motif (Vangl2 Y279A,Y280A), which we previously implicated in the interaction with AP-1 (Guo et al., 2013). STORM imaging analysis showed that the majority of Vangl2 Y279A,Y280A punctate structures were not adjacent to AP-1 (**Figures 5J,U**).

To test whether Frizzled6 is associated with AP-1 upon TGN exit, we analyzed the spatial relationship between HAtagged Frizzled6 and AP-1 under the same conditions of temperature shift. Super-resolution imaging analysis indicated that the majority of Frizzled6 punctae were not associated with AP-1 (**Figures 5K,U**) consistent with our previous report that sorting of Frizzled6 is not mediated by AP-1. Moreover, we found that many Frizzled6 structures but not Vangl2 structures were associated with GGA2 under the temperature shift conditions (**Figures 5L–P**). Quantification of 6 superresolution images from 6 cells of each group indicate that the percentage of Frizzled6 punctae that were associated with GGA2 is significantly higher than the percentage of Vangl2 punctae that were associated with GGA2 (**Figure 5U**). These analyses indicate that GGA2 regulates TGN export of Frizzled6 but not Vangl2 and provide direct evidence demonstrating sorting of specific cargo proteins by specific cargo adaptors at the TGN.

We recently showed that TGN export of Frizzled6 is also regulated by epsinR. We found that many Frizzled6 structures were associated with epsinR under the temperature shift conditions (**Figures 5Q–T**). Quantification analysis indicates that the percentage of Frizzled6 punctae that were associated with epsinR is significantly higher than the percentage of Frizzled6 punctae that were associated with AP-1 (**Figure 5U**). These analyses indicate that sorting of Frizzled6 at the TGN is regulated by the GGA2-associated epsinR but not the AP-1 associated epsinR.

Next, we analyzed the spatial relationships between Vangl2 and Frizzled6 upon TGN exit in COS7 cells co-transfected with HA-Frizzled6 and Myc-Vangl2. After an incubation at 20◦C, we observed that many Frizzled6 and Vangl2 structures overlapped at the juxtanuclear area (**Figures S8A–C**), suggesting that they were intermixed with each other on the TGN membranes in a condition of blocked exit from the TGN. After the temperature was released at 32◦C for 5 min, we found the majority of

Vangl2 (wt) and the γ subunit of AP-1 at 2-dimension (A–D) and 3-dimension (E–H'). Magnified views of the indicated areas in (A) were shown in (B–D). Magnified views of the indicated areas in (E) were shown in (F–H). 180◦ rotated views of the (F–H) were shown in (F'–H'). The scale bar of each image was indicated. (I–K) COS7 cells were transfected with HA-Vangl2 (wt) or HA-Vangl2 Y279A,Y280A or HA-Frizzled6. Day 1 after transfection, cells were incubated at 20◦C for 2 h and then at 32◦C for 5 min. After incubation, two-color STORM was utilized to visualize localizations of Vangl2 (wt) and the δ subunit of AP-3 (I), Vangl2 Y279A,Y280A and the γ subunit of AP-1 (J), Frizzled6 and the γ subunit of AP-1 (K). Scale bar, 500 nm. (L–Q) COS7 cells were transfected with HA-Vangl2 (L) or HA-Frizzled6 (M–P) or co-transfected with epsinR-FLAG and HA-Frizzled6 (Q–T). Day 1 after transfection, cells were incubated at 20◦C for 2 h and then at 32◦C for 5 min. After incubation, two-color STORM was utilized to visualize localizations of Vangl2 and GGA2 (L), Frizzled6 and GGA2 (M–P) or epsinR-FLAG and Frizzled6 (Q). Magnified views of the indicated areas in (M,Q) are shown in (N–P,R–T). (U) Quantifications of the percentage Vangl2 or Frizzled6 punctae that are associated with the indicated cargo *(Continued)*

FIGURE 5 | adaptor (mean ± SD, based on ≥6 super-resolution images in each experimental group). \*\**p* < 0.01, \*\*\**p* < 0.001 by two-tailed Student's *t*-test. The super-resolution images showed the localization patterns of the indicated protein in the whole juxtanuclear area where the majority of the cargo adaptors were localized. Each dot in the graphs represents data of the whole juxtanuclear Golgi area of each individual cell. Quantification analyses were based on super-resolution images acquired from three independent repetitions for each experimental group. (V) The proposed model showing that AP-1, AP-3, epsinR, and GGA2 are assembled into elongated structures on distinct microdomains on TGN and endosomal membranes with distinct spatial relationships: the spatial segregated GGA2 and AP-1 regulates sorting of Frizzled6 and Vangl2, respectively; the spatially associated cargo adaptor, GGA2 and epsinR, cooperatively regulate sorting of Frizzled6.

Vangl2 and Frizzled6 were not overlapped (**Figures S8D–F**, and quantification in **Figure S8G**), suggesting that Vangl2 and Frizzled6 spatially segregate on exit from the TGN.

#### DISCUSSION

Protein sorting at the TGN and endosomes play important roles in targeting various cargo proteins to their specific destinations. A number of cargo adaptors have been identified to mediate this essential cellular process but their spatial relationships at high resolution have not been systematically analyzed. Using two-color STORM imaging, we demonstrated that the TGNand endosome-localized cargo adaptors, AP-1, AP-3, GGA2, and epsinR, are enriched in distinct microdomains (**Figure 5V**). These cargo adaptors can form elongated structures of over 250 nm in length. Many of the elongated structures formed by GGA2, AP-1, and epsinR are associated with clathrin. In contrast, AP-3 showed a significantly reduced association with clathrin compared to the other three cargo adaptors. In addition, we showed that the majority of AP-3 structures are segregated from AP-1 structures (**Figure 5V**). Similar observations of these spatial relationship have been detected on the buds arising from endosome-associated tubules by immunogold electron microscopy (Peden et al., 2004).

Our high resolution imaging analysis also revealed that AP-1 structures are largely segregated from GGA2 structures and there are at least two populations of epsinR, one associated with AP-1 and the other associated with GGA2 (**Figure 5V**). In yeast, Ent3 and Ent5, the yeast homologs of epsinR, were shown by structured illumination microscopy (SIM) to be predominantly colocalized with Gga2p and AP-1, respectively. Gga2p and AP-1 are adjacent to but distinct from each other (Daboussi et al., 2012). These observations suggest that the spatial relationships among the cargo adaptors are conserved from yeast to mammalian cells.

The resolution of a STORM image can be indicated by the mean localization error of single molecular fluorescence. We examined the STORM resolution by analyzing the localization error distribution of single molecular fluorescence from the localization table of the raw STORM data of **Figure 5A** (**Figure S9**) and find the mean localization error of 11.2 nm. The resolution of STORM images is also affected by low localization density, but we do not have any structure that is too sparse to affect the conclusions. Besides, we have a robust protocol for labeling and tested different structures and no such issues were found (Huang et al., 2017; Liu et al., 2017; Cheng et al., 2018; Mao et al., 2019). Moreover, we measured the density of localization in **Figures 3A,E**, **5A,I** to be >18 counts/nm<sup>2</sup> at the least dense region, which means the average distance between adjacent localized sites is below 20 nm. These evidence suggests that our STORM system can achieve 20 nm resolution.

Both epsinR and GGA2 regulate TGN export of Frizzled6 but not Vangl2, suggesting that these two spatially associated cargo adaptors can cooperatively regulate a specific protein sorting process. This co-operativity has also been detected in yeast. Ent3p, which spatially associates with Gga2p, functions primarily in GGA-dependent transport (Costaguta et al., 2006). Ent3p, but not Ent5p, facilitates binding of Gga2p to the endosomal syntaxin Pep12p (Copic et al., 2007). In contrast, Ent5p, which spatially associates with AP-1, is more critical for AP-1-mediated transport (Costaguta et al., 2006).

EpsinR contains an N-terminal ENTH domain and a Cterminal unfolded region. The C-terminal unfolded region of epsinR directly interacts with the appendage domain of AP-1 (Owen et al., 2004) indicating that this direct interaction may induce their spatial association. Interestingly, we recently found that Frizzled6 also interacts with the C-terminal unfolded region of epsinR (Ma et al., 2018). The interaction between Frizzled6 and epsinR causes dissociation of epsinR from AP-1 (Ma et al., 2018). This result indicates that the presence of Frizzled6 can induce the separation of epsinR from AP-1 to allow these two cargo adaptors to perform distinct cargo sorting functions (Ma et al., 2018). In mammalian cells, knockdown of either AP-1 or epsinR causes reduction of the other in clathrin coated vesicles (Hirst et al., 2004). Acute inactivation of epsinR blocked the production of the entire population of clathrin-coated vesicles, suggesting a more global function of epsinR (Hirst et al., 2015). The functional role of epsinR remains to be elucidated in the AP-1-mediated vesicular trafficking process.

Spatially segregated assembly of cargo adaptors, such as GGA2 and AP-1, may provide a mechanism to allow these cargo adaptors to package distinct cargo molecules into transport vesicles. We previously showed that TGN export of Vangl2 but not Frizzled6 depends on AP-1 (Ma et al., 2018). Here, we have shown that GGA2 regulates TGN export of Frizzled6 but not Vangl2 suggesting that sorting of Vangl2 and Frizzled6 is regulated by the spatially segregated cargo adaptors, AP-1 and GGA2, respectively. Moreover, through super-resolution imaging analysis, we observed specific associations of Vangl2 and Frizzled6 with AP-1 and GGA2, respectively, upon exiting the TGN, confirming this differential sorting process.

How can cargo adaptors be recruited onto spatially segregated microdomains on the TGN or endosomes? One possible explanation is that the spatial separation is caused by a sequential assembly process as exemplified by GGAs and AP-1. In this scenario, assembly of GGAs on TGN membranes recruits PI4 kinase which in turn generates PI4P that induces a second wave of AP-1 assembly (Daboussi et al., 2012, 2017). Thus a peak of GGA assembly at a specific site will be followed by a peak of AP-1 assembly (Daboussi et al., 2012). Such a sequential assembly process correlates with sequential cargo sorting events (McDonold and Fromme, 2014). This process may prevent the assembly of GGA and AP-1 at the same time and space. Another possible explanation is that binding of cargo adaptors to specific cargo molecules may cause polymerization of the cargo adaptors. This polymerization process will enrich the cargo adaptors as well as their associated cargo molecules into specific membrane domains. This hypothesis is supported by evidence showing that binding of cargo to AP-1 promotes polymerization of AP-1 (Meyer et al., 2005; Lee et al., 2008). In addition to inducing polymerization of cargo adaptors, cargo molecules can stabilize the membrane association of cargo adaptors to allow sufficient time for vesicle coat assembly (Crottet et al., 2002; Lee et al., 2008; Caster et al., 2013; Guo et al., 2013, 2014; Ren et al., 2013). Thus, cargo molecule and cargo adaptors can form a positive feedback loop which will induce assembly of specific cargo sorting machinery on distinct membrane microdomains.

Cargo adaptors that mediate protein sorting at the TGN and endosome are present in a sophisticated interaction network and defects in these cargo adaptors can directly or indirectly influence the functions of other cargo adaptors. Even the spatially segregated cargo adaptors are not functionally independent of each other. Evidence suggests that acute inactivation of AP-1 in mammalian cells using the "knocksideways" system depletes GGA2 from clathrin coated vesicles and causes stronger defects in cargo sorting than acute inactivation of GGA2 does (Hirst et al., 2012). However, knockdown analysis indicates that incorporation of GGA2 into clathrin coated vesicles is not affected by AP-1 depletion and vice versa (Hirst et al., 2009), suggesting that knock down cargo adaptors produces a weaker phenotype than knocking the cargo adaptors sideways. Deletion AP-1 but not GGA2 by siRNA knockdown regulates HIV protein Nef-mediated downregulation of histocompatibility complex (MHC) class I (Hirst et al., 2009), which is consistent with our study indicating that these two cargo adaptors can function to regulate sorting of distinct cargo molecules.

We used COS7 cells for the spatial analysis. The spatial relationships of the cargo adaptors are consistent with the spatial relationships of cargo adaptors observed in yeast and human fibroblasts (Hirst et al., 2003; Mardones et al., 2007). An important future step of our study is to measure the extent of co-localization using polarized cells or when cells are in different differentiation states. Because of the limitation of two color, we cannot distinguish whether the signal we detected was located at the Golgi or endosomes. Another important future step of our study is to build a three-color super-resolution imaging system to demonstrate where the segregation took place. As the majority of the signal of AP-1, GGA2, and epsinR were located at the juxtanuclear Golgi area and the superresolution images showed the localization patterns of these cargo adaptors in the whole juxtanuclear Golgi area, we propose that the majority the signal detected in this study was from the Golgi.

Altogether, our high-resolution imaging analysis indicates that cargo adaptors, AP-1, AP-3, and GGA2, are assembled into large elongated structures on distinct exit domains at the TGN or endosomes to mediate sorting of specific cargo molecules (**Figure 5V**). The spatially segregated GGA2 and AP-1 regulates sorting of Frizzled6 and Vangl2, respectively. And the spatially associated cargo adaptor, GGA2 and epsinR, cooperatively regulate sorting of Frizzled6.

# MATERIALS AND METHODS

# Cell Lines, Antibodies, and Plasmids

HeLa cells were kindly provided by the University of California-Berkeley Cell Culture Facility and were confirmed by short tandem repeat profiling. COS-7 cells were obtained from ATCC (Cat#ATCC CRL-1651, RRID: CVCL\_0224). The cells were tested negative for mycoplasma contamination. COS7 and HeLa cells were maintained in GIBCO Dulbecco's Modified Eagle Medium containing 10% Fetal Bovine Serum (FBS), 10 mU/ml of penicillin and 0.1 mg/ml of streptomycin. Transfection of DNA constructs into COS7 cells and immunofluorescence were performed as described (Guo et al., 2013). Temperature shift experiment was performed as described (Guo et al., 2013).

The commercial antibodies were: mouse anti-golgin-97 (Invitrogen #A21270, RRID: AB\_221447), mouse anti-GM130 (BD Biosciences #610823, RRID: AB\_398142), mouse anti-γ1 adaptin (BD Bio #610385, RRID: AB\_397768), rabbit anti-HA (Cell Signaling #3724, RRID: AB\_1549585), rabbit anti-clathrin heavy chain (abcam #ab21679, RRID: AB\_2083165), mouse anti-δ subunit of AP-3 (Developmental Studies Hybridoma Bank, number anti-delta SA4, RRID: AB\_2056641), sheep anti-TGN46 (AbD Serotec, number AHP500G, RRID: AB\_323104), mouse anti-FLAG (Sigma, number F3165, RRID: AB\_259529), rabbit anti-FLAG (Sigma, number F7425, RRID: AB\_439687), rabbit anti-γ1-adaptin (Proteintech, number 13258-1-AP, RRID: AB\_2058209), mouse anti-GGA2 (BD Bioscience, number 612612, RRID: AB\_399892), mouse anti-HA (BioLegend, number 901501, RRID: AB\_2565006), rabbit anti-GGA1 (Thermo Fisher Scientific, number PA5-12130, RRID: AB\_2232367), rabbit anti-GGA3 antibody (Cell Signaling Technology, number 4167s, RRID: AB\_1903987).

The plasmid encoding HA-tagged mouse Vangl2 was cloned in pCS2 (Merte et al., 2010). The plasmid encoding HA-Frizzled6 was generated by cloning full length mouse Frizzled6 amplified from Frizzled6-Myc into pcDNA4/TO and an HA tag was inserted by PCR between F22 and T23 (Ma et al., 2018). The plasmid encoding FLAG-tagged full length epsinR was generated by cloning full length epsinR from human epsinR cDNA (OriGene) into p3XFLAG-CMV-14 (Sigma Aldrich).

The target sequence of the siRNAs against clathrin heavy chain, AP1γ1-adaptin and AP3δ1-adaptin were described previously (Ma et al., 2018). The target sequence of siRNA against GGA2 was: CCGGAAGACATCAAGATTCGA. The target sequence of siRNA against GGA1 was GCCGAAGAATGTGATCTTT, the target sequence against GGA3 was GTGAGATGCTGCTTCATTA.

# Transfection and Immunofluorescence Staining

Transfection of siRNA or DNA constructs into HeLa cells or COS7 cells and immunofluorescence were performed as described (Guo et al., 2013). Cells growing on coverslips were fixed in 4% PFA for 20 min then washed with PBS and incubated with permeabilization buffer (2.5% FBS, 0.1% TX-100, 0.2 M Glycine in PBS) at RT for 30 min. Then cells were sequentially incubated with primary antibody and secondary antibody diluted in permeabilization buffer for 30 min. Each antibody incubation was following by five times wash with PBS. After staining, cells were fixed again in 4% PFA for 20 min and then washed with PBS. Surface labeling procedure was described (Ma et al., 2018). Briefly, to label surface-localized Frizzled6 containing an HA tag in its extracellular domain, HeLa cells expressing HA-Frizzled6 were incubated with the mouse anti-HA antibody (1:200) in PBS containing FBS (250 µl in 10 ml) for 40 min on ice. After several washes with PBS, cells were fixed for 15 min with 4% paraformaldehyde in PBS and then a normal immunofluorescence procedure was performed. Fluorescent images were acquired with a Zeiss Axioobserver Z1 microscope system.

# Quantitative Real-Time PCR

HeLa cells transfected with control siRNA or AP3δ1 siRNA. Seventy two hours after transfection, HeLa cells were harvested for RNA extraction. Total RNA was extracted using TRIzol reagent (Invitrogen, Catalog number: 15596026). Subsequently, RT-qPCR were performed as described previously (Amatori et al., 2017). RT-qPCR experiment was conducted in technical triplicates. 2-11CT method was used to detect the gene expression variation between the control group and the knockdown group.

#### Super-Resolution Imaging Analysis

The two-color STORM was built as described (Wang et al., 2016). Briefly, this STORM was built using a Nikon Ti-E inverted microscope with perfect focus system. A 656.5 nm DPSS laser and a 750 nm diode laser were used to provide laser excitations for Alexa 647 channel and Alexa 750 channel, respectively. A 100× TIRF objective (Nikon) was used to collect the fluorescent signals of both channels. Subsequently, the fluorescent signals passed through a home-built channel splitter before simultaneously forming images on an EMCCD (Andor, IXon-Ultra). Emission filters (Chroma ET700/50 m and ET810/90 m) were placed in each path of the channel to block the excitation laser.

The imaging buffer for two-color STORM was designed for the two dye combination Alexa 647 and Alexa 750 (Zhao et al., 2015). The buffer contained 10% (w/v) glucose, 25 mM Tris(2-carboxyethyl)phosphine hydrochloride solution (TCEP, Sigma-Aldrich 646547), 2 mM cyclooctatetraene (Sigma-Aldrich 138924), 560µg/ml glucose oxidase, 40µg/ml catalase, 50 mM Tris-Cl pH 8.0, 1 mM ascorbic acid and 1 mM methyl viologen. The composition of the imaging buffer provided matched and balanced switching characteristics for both dyes (Zhao et al., 2015). The sample was mounted on a customized glass-bottom chamber filled with imaging buffer. The desired position is identified using conventional fluorescence image with relatively low laser excitation power, typically 60 W/cm<sup>2</sup> for a 656.5 nm laser and 80 W/cm<sup>2</sup> for a 750 nm laser. After the region of interest was identified, the laser power was increased to 4 kW/cm<sup>2</sup> in both channels enabling rapid "blinking" of dye molecules for single molecule detection and localization. The blinking was recorded by an EMCCD with 100× EM gain at 30 Hz for 15,000 to 20,000 frames based on the abundance of proteins. When each frame was captured, the peak finding algorithm recognized the sites of blinking, followed by the fitting algorithm that determined the centroid of each blinking with nanometer accuracy. These centroids were registered to the final super-resolution image. In addition, active sample locking was applied to stabilize the sample with nanometer accuracy in the x-y plane and z-axis during acquisition. Optical astigmatism was used to achieve 3 dimensional STORM (Huang et al., 2008). The data acquisition and single-molecule localization are carried out with software called Rohdea developed by NanoBioImaging Ltd, while the rendering is done with QuickPALM (Henriques et al., 2010) from Fiji (win64 version, National Institutes of Health, USA). Gaussian rendering is used to simulate sub-diffraction spot. Fiji plug-in 3D viewer was used to record 3D rotation films.

Fiji was used for calculating the percentage of overlapped pixels based on a modified quantification method (Guo et al., 2008), utilizing the following procedures: (1) a threshold equaling to 25% of the maximal pixel value of each marker was chosen; (2) the average pixel intensity for the two thresholded images was equalized using the divide function; (3) the number of above-threshold pixels for each marker was determined per cell; (4) the number of above-threshold overlapped pixels was determined using the colocalization highlight function in MBF plugin collection (ImageJ) with a fixed ratio of 0.6; (5) for each marker, the number of overlapped pixels was divided by the number of above-threshold pixels to yield the percentage of a given maker's area in a cell that overlapped with the other marker; (6) the average of the two values, each representing the percentage of overlapped pixels for each marker, was used as final value indicating the percentage of overlapped pixels.

To measure the percentage of the punctate structures in one channel (channel A) that were associated with the punctate structures in the other channel (channel B), the number of punctates in channel A that were associated with the punctate structures in channel B was divided by the total number of the punctates in channel A in each super-resolution image. The number of punctate structures in each super-resolution image was counted manually.

# DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

# AUTHOR CONTRIBUTIONS

YH, TM, PL, JW, TZ, and YG: data curation, formal analysis, validation, and investigation. SD, ML, and YG: resources and supervision. YG: conceptualization, writing-review and editing, writing original draft, and project administration.

#### FUNDING

This work was supported by the Hong Kong Research Grants Council Grants 26100315, 16101116, 16102218, AoE/M-05/12- 3, C4002-17G, C4012-16E, and HKUST12/CRF/13G to YG. This work was also supported by a grant from National Natural Science Foundation of China (NSFC 31871421) and a free explore project from Shenzhen Science and Technology Innovation Committee (JCYJ20180306174847511) to YG. In addition, the work was supported by a grant from the Offices of Provost, VPRG, and Dean of Science, HKUST (HKUST VPRGO12SC02) to SD.

#### ACKNOWLEDGMENTS

We thank Daiying Xu (NanoBioImaging LTD, Hong Kong) for their technical support for the STORM imaging analysis.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2019. 00181/full#supplementary-material

Figure S1 | Localization patterns of imaging beads coated with both Alexa 750 and Alexa 647 revealed through two-color STORM. (A) Alexa 750 and Alexa 647 showed a perfectly overlapped pattern. Scale bar, 500 nm. (B–D) Magnified views of the area indicated by arrows in (A). Scale Bar, 50 nm.

Figure S2 | The localization patterns of clathrin and cargo adaptors in COS7 cells. COS7 cells were untransfected (A–O) or transfected with plasmids encoding FLAG-tagged epsinR (P–R). Day 1 after transfection, cells were analyzed by immunofluorescence utilizing the indicated antibodies. Representative fluorescent images were shown. Scale bar, 10µm.

Figure S3 | The localization patterns of clathrin and the indicated cargo adaptors in HeLa cells. HeLa cells were transfected with control siRNA (A,C,E,G,I) or transfected with siRNA against the indicated cargo adaptors (B,D,F,H) or clathrin heavy chain (J). Day 3 after transfection, cells were analyzed by immunofluorescence utilizing the indicated antibodies. Representative fluorescent images were shown. Scale bar, 10µm. (K–M) HeLa cells were mock transfected or transfected with siRNA against AP1γ1, clathrin heavy chain or AP3δ1. On day 3 after transfection, cells were analyzed by immunoblot (K,L) or RT-PCR (M).

Figure S4 | The localization patterns of clathrin heavy chain on the plasma membrane of COS7 cells after stimulating with epidermal growth factor (EGF). (A) COS7 cells were treated with 5 ng/ml EGF for 5 min and then stained with antibodies against clathrin heavy chain. Localization patterns of clathrin heavy

#### REFERENCES


chain were analyzed by STORM. Scale Bar, 500 nm. (B–D) Magnified views of the structures highlighted by arrows in (A). Scale Bar, 100 nm.

Figure S5 | TGN export of Vangl2 is independent of GGA2. (A–I) HeLa cells were mock transfected (A–C) or transfected with siRNA against clathrin heavy chain (CHC, D–F) or transfected with siRNA against GGA2 (G–I) and re-transfected after 48 h with plasmids encoding HA-Vangl2 (A–I). On day 3 after knockdown, cells were incubated at 20◦C for 2 h then shifted to 32◦C for 50 min in the presence of cycloheximide. After incubation, cells were analyzed by immunofluorescence. Scale bar, 10µm. (J) Quantification of the percentage of cells showing TGN-accumulated Vangl2 in cells treated with control siRNA or siRNA against CHC or GGA2 after incubation at 32◦C (mean ± SD; *N* = 3; >150 cells counted for each experiment).

Figure S6 | Knockdown of GGA3 but not GGA1 causes defects in surface delivery of Frizzled6. (A–L) HeLa cells were mock transfected (A–D) or transfected with siRNA against GGA1 (E–H) or siRNA against GGA3 (I–L) and re-transfected after 48 h with plasmids encoding HA-Frizzled6. On day 3 after knockdown, cells were incubated at 20◦C for 2 h then shifted to 32◦C for 50 min in the presence of cycloheximide. After incubation, cells were analyzed by immunofluorescence. The surface-localized HA-Frizzled6 and the total HA-Frizzled6 were stained by mouse and rabbit anti-HA antibodies, respectively. Scale bar, 10µm. (M) HeLa cells were mock transfected or transfected with siRNA against GGA1 or GGA3. On day 3 after transfection, cells were analyzed by immunoblot. (N) Quantification of the percentage of cells showing detectable surface localized Frizzled6 in cells treated with control siRNA or siRNA against GGA1 or GGA3 after incubation at 32◦C (mean ± SD; *N* = 3; >100 cells counted for each experiment). ∗∗*p* < 0.01 by two-tailed Student's *t*-test.

Figure S7 | Localizations of Vangl2 and Frizzled6 after temperature shit. (A–F) COS7 cells were transfected with HA-Vangl2 (A–C) or HA-Frizzled6 (D–F). Day 1 after transfection, cells were incubated at 20◦C for 2 h and then at 32◦C for 5 min. The localizations of Vangl2, Frizzled6, and TGN46 was then analyzed by immunofluorescence. Scale bar, 10µm.

Figure S8 | Analysis of the spatial relationships between Vangl2 and Frizzled6 upon exiting the TGN. (A–F) COS7 cells were co-transfected with HA-Frizzled6 and Myc-Vangl2. Day 1 after transfection, cells were incubated at 20◦C for 2 h (A–C) or incubated at 20◦C for 2 h and then at 32◦C for 5 min (D–F). After incubation, cells were stained with antibodies against HA tag and Myc tag. Two-color STORM was then utilized to visualize localizations of Vangl2 and Frizzled6. (G) Quantification of the percentage of overlapped pixels between Vangl2 and Frizzled6 in cells incubated at 20◦C for 2 h or in cells incubated at 20◦C for 2 h and then at 32◦C for 5 min (mean ± SD, based on three super-resolution images each). ∗∗∗*p* < 0.001 by two-tailed Student's *t*-test.

Figure S9 | The histogram of the localization error of Figure 5A. The software rejects any fitting with error >20 nm.

Videos S1–S3 | 360◦ rotated views of AP-1 (red) and clathrin (green) of the indicated areas in Figure 2D.

Videos S4–S6 | 360◦ rotated views of epsinR (red) and clathrin (green) of the indicated areas in Figure 2H.

Videos S7–S9 | 360◦ rotated views of GGA2 (red) and clathrin (green) of the indicated areas in Figure 2L.

Videos S10–S12 | 360◦ rotated views of AP-1 (red) and Vangl2 (green) of the indicated areas in Figure 5E.


**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 Huang, Ma, Lau, Wang, Zhao, Du, Loy and Guo. 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.

# Directing Traffic: Regulation of COPI Transport by Post-translational Modifications

Peter M. Luo and Michael Boyce\*

Department of Biochemistry, Duke University School of Medicine, Durham, NC, United States

The coat protein complex I (COPI) is an essential, highly conserved pathway that traffics proteins and lipids between the endoplasmic reticulum (ER) and the Golgi. Many aspects of the COPI machinery are well understood at the structural, biochemical and genetic levels. However, we know much less about how cells dynamically modulate COPI trafficking in response to changing signals, metabolic state, stress or other stimuli. Recently, post-translational modifications (PTMs) have emerged as one common theme in the regulation of the COPI pathway. Here, we review a range of modifications and mechanisms that govern COPI activity in interphase cells and suggest potential future directions to address as-yet unanswered questions.

#### Edited by:

Yanzhuang Wang, University of Michigan, United States

#### Reviewed by: Alberto Luini,

Italian National Research Council (CNR), Italy Elizabeth Sztul, The University of Alabama at Birmingham, United States Blanche Schwappach, University of Göttingen, Germany

> \*Correspondence: Michael Boyce michael.boyce@duke.edu

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

Received: 15 April 2019 Accepted: 23 August 2019 Published: 11 September 2019

#### Citation:

Luo PM and Boyce M (2019) Directing Traffic: Regulation of COPI Transport by Post-translational Modifications. Front. Cell Dev. Biol. 7:190. doi: 10.3389/fcell.2019.00190 Keywords: COPI vesicle trafficking, coatomer, interphase, post-translational modifications, phosphorylation, ubiquitination, glycosylation, myristoylation

# INTRODUCTION

The coat protein complex I (COPI) mediates multiple lipid and protein trafficking paths between the endoplasmic reticulum (ER) and the Golgi. Mammalian COPI carries out this essential function through the combined action of seven core coatomer subunits: α-COP, β-COP, β 0 -COP, γ-COP, δ-COP, ε-COP, and ζ -COP (Popoff et al., 2011; Jackson, 2014; Rout and Field, 2017; Arakel and Schwappach, 2018; Bethune and Wieland, 2018). The genetic, biochemical, and structural details of the COPI pathway have been studied extensively. Here, we first provide a succinct overview of COPI mechanism and function, and we refer the interested reader to several excellent review articles for more comprehensive detail (Popoff et al., 2011; Jackson, 2014; Rout and Field, 2017; Arakel and Schwappach, 2018; Bethune and Wieland, 2018).

Briefly, COPI carrier formation is initiated when the small GTPase Arf1 exchanges GDP for GTP with the assistance of its guanine nucleotide exchange factor (GEF), GBF1 (Popoff et al., 2011; Jackson, 2014; Rout and Field, 2017; Arakel and Schwappach, 2018; Bethune and Wieland, 2018). Arf1-GTP inserts a myristoylated α-helix into the originating membrane (e.g., the cis-Golgi) and recruits the heptameric coatomer complex en bloc from cytoplasmic pools to the membrane (Popoff et al., 2011; Jackson, 2014; Rout and Field, 2017; Arakel and Schwappach, 2018; Bethune and Wieland, 2018). Stably bound COPI complexes assemble into an ordered lattice on the lipid bilayer, interacting with cargo through N-terminal domains of the α and β 0 subunits, and promoting membrane curvature (Popoff et al., 2011; Jackson, 2014; Rout and Field, 2017; Arakel and Schwappach, 2018; Bethune and Wieland, 2018). Bilayer curvature triggers the recruitment of GTPase-activating proteins (GAPs) – ArfGAPs1-3 in mammals – which stimulate Arf1's GTPase activity (Popoff et al., 2011; Jackson, 2014; Rout and Field, 2017; Arakel and Schwappach, 2018; Bethune and Wieland, 2018). GTP hydrolysis by Arf1 is required for cargo sorting, suggesting

that COPI undergoes GTPase-driven conformational changes, though these steps are incompletely understood (Popoff et al., 2011; Jackson, 2014; Rout and Field, 2017; Arakel and Schwappach, 2018; Bethune and Wieland, 2018). Near the end of COPI assembly, extreme curvature promotes vesicle scission, also an Arf1-dependent event (Popoff et al., 2011; Jackson, 2014; Rout and Field, 2017; Arakel and Schwappach, 2018; Bethune and Wieland, 2018). Most COPI uncoats from the resulting, fully formed vesicle, likely coordinated by final rounds of GTP hydrolysis and the dissociation of Arf1 from the membrane (Popoff et al., 2011; Jackson, 2014; Rout and Field, 2017; Arakel and Schwappach, 2018; Bethune and Wieland, 2018). Finally, COPI vesicles dock to their destination membranes on the ER or Golgi. This step is aided by interactions among accessory proteins and lingering coatomer complexes on the vesicles, and by resident tethering and fusion components on the target organelle (Popoff et al., 2011; Jackson, 2014; Rout and Field, 2017; Arakel and Schwappach, 2018; Bethune and Wieland, 2018).

COPI is generally agreed to mediate retrograde protein and lipid trafficking within and from the Golgi to the ER, and, in animals, probably mediates anterograde trafficking from the ER-Golgi intermediate compartment (ERGIC) to the Golgi proper (Emr et al., 2009; Popoff et al., 2011; Jackson, 2014; Rout and Field, 2017; Arakel and Schwappach, 2018; Bethune and Wieland, 2018). The role of the COPI system in trafficking among the Golgi stacks is more controversial. One model invokes the forward movement of vesicles from cis to medial to trans Golgi cisternae, with COPI participating (Emr et al., 2009; Nakano and Luini, 2010; Glick and Luini, 2011; Papanikou and Glick, 2014). However, a newer and more broadly accepted model suggests that individual Golgi cisternae progressively acquire components and characteristics of cis, then medial, then trans Golgi character (Emr et al., 2009; Nakano and Luini, 2010; Glick and Luini, 2011; Papanikou and Glick, 2014). The role or requirement for COPI in this cisternal maturation process may be minor or nonexistent. Elements of both vesicular trafficking and cisternal maturation mechanisms could, in principle, co-occur within a single organism or cell type, and further research will be needed to clarify these ambiguities (Emr et al., 2009; Nakano and Luini, 2010; Glick and Luini, 2011; Papanikou and Glick, 2014). Nevertheless, COPI is a fundamental and essential pathway in the endomembrane system of all eukaryotic cells, regardless of its precise role in intra-Golgi trafficking.

The COPI pathway has been studied for decades through a range of elegant genetic, cellular, biochemical and biophysical studies, reviewed elsewhere (Popoff et al., 2011; Jackson, 2014; Rout and Field, 2017; Arakel and Schwappach, 2018; Bethune and Wieland, 2018). However, major aspects of COPI trafficking remain poorly understood. For example, little is known about how cells and tissues tune the activity of COPI trafficking in response to metabolic changes, physiological signals or stresses, such as the unfolded protein response or infection. In all of these contexts, COPI flux must adapt to the changing amount and type of client cargoes produced, but the nature of this rapid regulation is largely unclear. One key clue may come from related studies of Golgi dynamics during mitosis. The Golgi disperses in a regulated fashion during cell division for redistribution to the daughter cells, and numerous studies have dissected the mechanisms regulating this COPIdependent process (Cancino and Luini, 2013; Huang and Wang, 2017). Several lines of evidence have defined a clear role for post-translational modifications (PTMs) in these cell cycledependent changes in Golgi structure (Cancino and Luini, 2013; Huang and Wang, 2017). Whether analogous mechanisms are used to tune COPI traffic in interphase cells is less well understood, but a variety of PTMs has been implicated in modulating interphase COPI activity, indicating the existence of robust and complex systems of regulation. Here, we focus specifically on major examples of interphase COPI regulation by PTMs and assess the future directions of this emerging area of cell biology.

# PHOSPHORYLATION

Protein phosphorylation has long been implicated in COPI regulation. For example, early studies found that inhibiting global dephosphorylation in hepatocytes disrupts the localization of β-COP and the architecture and function of the Golgi (Reaven et al., 1993), and demonstrated that β- and δ-COP purified from rat liver cytosol are themselves phosphorylated (Sheff et al., 1996). In the years since, several major kinase pathways were discovered to influence COPI function directly. Below we provide three examples of these links between phosphorylation and the COPI pathway. For a compendium of specific PTMs discussed throughout this review, see **Table 1**. For a continually updated and annotated database of all observed, COPI-relevant phosphosites and other PTMs, we direct the reader to the public PhosphoSitePlus resource<sup>1</sup> (Hornbeck et al., 2015).

Cyclic AMP (cAMP)-dependent protein kinase A (PKA) is one of the best-documented regulators of interphase COPI trafficking (**Figure 1**). Early studies by the Velasco lab revealed that a small molecule inhibitor of PKA retarded the trafficking of the vesicular stomatitis virus G glycoprotein (VSVG) through the Golgi (Muniz et al., 1996). Conversely, PKA activators, such as forskolin, which stimulates cAMP synthesis, accelerated VSVG transport and altered Golgi cisternal structure, suggesting a possible impact on COPI (Muniz et al., 1996). Using in vitro reconstitution assays, the same group demonstrated that PKA or cAMP increased the binding of Arf1 to salt-washed Golgi membranes, whereas PKA inhibition or depletion, or dephosphorylation of Golgi membranes, had the opposite effect (Martin et al., 2000). The authors also showed that increases in cAMP levels in live cells triggered the redistribution of Arf1 from the cytoplasm to the Golgi, consistent with a regulatory role for PKA in COPI initiation (Martin et al., 2000). Subsequent work has supported this model, demonstrating that Golgi-associated PKA is rapidly activated by cAMP, inducing changes in the morphology of early Golgi stacks and accelerating trafficking (Mavillard et al., 2010). These results pointed to a possible role for Golgi-localized PKA in regulating the COPI pathway in response to extracellular signals.

<sup>1</sup>https://www.phosphosite.org

TABLE 1 | Compendium of specific PTMs discussed in this review. Please see text for details.


The relevant PKA substrates and mechanism(s) of action in the COPI pathway remain incompletely understood, but current evidence points to several candidates. In an early clue that PKA signaling may influence membrane trafficking through Arf proteins in particular, the Vaughan lab showed that PKA binds and phosphorylates ArfGEF1 and -2, and both GEFs translocate

from the cytosol to endomembranes upon forskolin-mediated cAMP upregulation (Li et al., 2003; Kuroda et al., 2007). Indeed, genetic knockdown of the PDE3A phosphodiesterase, which terminates cAMP signaling, decreased PKA-dependent ArfGEF membrane association and Arf1-GTP (Puxeddu et al., 2009). These results suggested a direct functional link between the PKA and Arf pathways, but the implications for COPI in particular remained uncertain, since ArfGEF1 and -2 are not major regulators of COPI trafficking. In subsequent work, both cytosolic and recombinant-purified PKA was also shown to phosphorylate the KDEL receptor (KDELR) itself at Ser209, promoting its interaction with COPI coatomer and ArfGAP (Cabrera et al., 2003) (**Figure 1A**). Although coatomer and ArfGAP1 can bind directly to unphosphorylated KDELR in vitro (Yang et al., 2002), PKA phosphorylation at Ser209 may play a regulatory role in vivo, as PKA inhibition prevented COPI-mediated retrograde transport of wild type, but not phosphomimetic Ser209Asp mutant, KDELR (Cabrera et al., 2003). In later work, anterograde ER-to-Golgi traffic was shown to trigger the G<sup>s</sup> subunit of the heterotrimeric G protein complex and adenylyl cyclase, leading to the phosphorylation of α-, δ-, ε-, and ζ -COP and actin cytoskeletal regulators by PKA, ultimately stimulating retrograde trafficking to return KDELR and other COPI cargoes to the ER (Cancino et al., 2014). Taken together, these studies suggest a key role for PKA in balancing the anterograde and retrograde (i.e., COPI-dependent) pathways of Golgi trafficking to maintain organellar homeostasis. On the other hand, PKA is reported to have disparate effects on COPI trafficking, from promoting Arf1 membrane binding and transport (Martin et al., 2000; Cabrera et al., 2003; Mavillard et al., 2010; Cancino et al., 2014) to decreasing ArfGEF activity and Arf1-GTP levels (Li et al., 2003; Kuroda et al., 2007). It may be that spatially and temporally restricted activation (e.g., by cAMP and/or enzyme localization) allows PKA to exert divergent effects on trafficking in different contexts.

The non-receptor tyrosine kinase Src has also been implicated in COPI trafficking in several contexts. Early work by Brown et al. (1998) found that the ArfGAP ASAP1 interacts biochemically with the SH3 domain of Src and is tyrosine-phosphorylated, suggesting that ASAP1 might link Src signaling to membrane trafficking (**Figure 2A**). Although ASAP1 is thought to function primarily in regulating cytoskeletal rearrangements at the plasma membrane, recent work suggests that the Src/ASAP1 axis may impact on COPI as well. The Lee lab demonstrated that ER stress causes the integral ER membrane protein and stress sensor Ire1α to bind Src, promoting ASAP1 phosphorylation and recruitment to the Golgi membrane (Tsai et al., 2018) (**Figure 2A**). Relocalized ASAP1 bound to GBF1, increasing its GEF activity to facilitate ER stress-induced Arf1-GTP recruitment to the cis-Golgi (Tsai et al., 2018) (**Figure 2A**). These events likely impact on COPI, because ER stress-induced Src signaling and ASAP1 phosphorylation led to KDELR1 dispersion from the Golgi and suppression of retrograde trafficking (Tsai et al., 2018). In other work supporting the Src/COPI connection, Bard and coworkers observed collapsed Golgi stacks and distended cisternae in cells lacking the three ubiquitous Src-like kinases Src, Yes and Fyn (Bard et al., 2003). Expression of active Src induced KDELR dispersion from the Golgi, whereas transport of Pseudomonas exotoxin, a COPI-dependent KDELR client, was accelerated by chemical or genetic Src inhibition (Bard et al., 2003). Src also functionally interacts with the Rab family of GTPases, themselves important regulators of membrane trafficking (Hutagalung and Novick, 2011). Tisdale and Artalejo (2006) demonstrated that Rab2, which is required for transport between the ER and Golgi, promotes Src membrane association and its phosphorylation of the atypical kinase aPKCι/λ (see also below) on the ERGIC, leading to β-COP recruitment and retrograde trafficking (**Figure 2B**).

Other studies revealed that Src may also play a role analogous to PKA in regulating Golgi homeostasis by balancing forward and reverse transport. The Luini lab demonstrated that the anterograde trafficking of ER chaperones to the Golgi initiates a KDELR-dependent signaling cascade (Pulvirenti et al., 2008). Golgi-localized Src interacts with the KDELR directly, leading to Src activation and the upregulation of forward intra-Golgi transport through tyrosine phosphorylation (Pulvirenti et al., 2008). As with PKA and retrograde trafficking,

then, Src activity may serve to balance anterograde transport to and within the Golgi to maintain organelle homeostasis.

More recently, the Bard lab reported a role for Src in growth factor-triggered redistribution of polypeptide N-acetylgalactosaminyltransferases (ppGalNAcTs), enzymes that initiate mucin-type O-glycosylation, from the Golgi to the ER (Gill et al., 2010). The authors reported that stimulation with epidermal growth factor or platelet-derived growth factor induces Src-driven redistribution of coatomer and ppGalNAcTs in an Arf1-dependent manner, implicating COPI trafficking in the reduction of mucin-type glycan biosynthesis through retrograde transport of glycosyltransferases (Gill et al., 2010; Chia et al., 2014). These results have been disputed, raising questions about the generality of the observations (Bard and Chia, 2017; Herbomel et al., 2017). Interestingly, however, the Bard lab has found evidence of O-glycosylation machinery relocalization to the ER in multiple human cancers, suggesting that activation of oncogenes like Src may dysregulate COPI or other trafficking pathways, resulting in aberrant glycoconjugates that might promote tumorigenesis (Gill et al., 2013; Nguyen et al., 2017). Testing this hypothesis will be an important goal of future work.

The protein kinase C (PKC) family is a final example of a welldocumented COPI regulatory kinase. At the biochemical level, an early study showed that PKC promotes the binding of β-COP and Arf1 to Golgi membranes in a GTP-dependent manner, both in rodent cells and in vitro (De Matteis et al., 1993). Similarly, the Mochly-Rosen group showed that β 0 -COP interacts directly with PKCε in vitro, and the two proteins colocalize to the Golgi membrane of cardiomyocytes (Csukai et al., 1997). These results suggested that PKC may modulate COPI activity in response to upstream regulatory signals. Consistent with this notion, Westermann et al. (1996) showed that activation of membranebound PKC by phorbol 12-myristate 13-acetate increased the secretion of heparan sulfate proteoglycans from the trans-Golgi in a hepatocellular carcinoma cell line, though a specific role for COPI was not proven.

In another context, osmotically induced cell volume changes were shown to perturb COPI trafficking, as judged by an assay of brefeldin A-induced β-COP Golgi membrane dissociation (Lee and Linstedt, 1999). Interestingly, following several hours of osmotic stress, Golgi resident proteins returned to their proper location in a PKC-dependent (but protein synthesisindependent) manner, providing additional evidence for dynamic post-translational regulation of cargo trafficking by PKC (Lee and Linstedt, 1999).

Finally, as noted above, the Tisdale group identified a functional interaction between β-COP and PKC, focusing on the ι/λ isoform (Tisdale and Jackson, 1998; Tisdale, 2000) (**Figure 2B**). The authors showed that Rab2 requires PKCι/λ in order to promote the recruitment of β-COP to ERGIC membranes (Tisdale and Jackson, 1998; Tisdale, 2000) (**Figure 2B**). Perhaps surprisingly, β-COP recruitment did not require PKCι/λ kinase activity, whereas Rab2-mediated vesicle budding did (Tisdale, 2000). Though the mechanism remains to be fully elucidated, these results suggest that ERGIC-localized PKCι/λ may coordinate the Rab and coatomer families to regulate retrograde trafficking.

PKA, Src, and PKC represent perhaps the bestdocumented examples of interphase COPI regulation through phosphorylation. However, it is clear that more discoveries await us. Functional interactions have been reported between

adenosine monophosphate-activated protein kinase (AMPK) and GBF1 (Miyamoto et al., 2008; Mao et al., 2013), casein kinase (CK) I and ArfGAP1 (Yu and Roth, 2002), and CKII and p115, a Golgi-resident COPI vesicle tethering protein (Dirac-Svejstrup et al., 2000) (**Table 1**). More recently, evidence has emerged that ArfGAP1 is regulated by the leucine-rich repeat kinase 2 (LRRK2), which is dysregulated in both inherited and sporadic forms of Parkinson's disease (PD) (Stafa et al., 2012). The authors demonstrated that LRRK2 and ArfGAP1 interact in vitro and in brain tissue, and PD-associated mutations in LRRK2 alter this association (Stafa et al., 2012). Interestingly, ArfGAP1 promotes both the kinase and GTPase activities of LRRK2, and is phosphorylated directly by LRRK2 (Stafa et al., 2012). These effects may be functionally important in PD, because silencing ArfGAP1 expression in primary cortical neurons rescued the neurite shortening phenotype caused by overexpression of a disease-associated mutant LRRK2, whereas co-expression of ArfGAP1 and LRRK2 synergistically promoted neurite shortening (Stafa et al., 2012). In future work, it will be important to determine the role (if any) for COPI trafficking in LRRK2/ArfGAP1 signaling in the nervous system.

These and other results suggest that COPI trafficking is subject to a wide range of combinatorial inputs from diverse phosphorylation cascades. Supporting this general notion, genetic screens based on high-content imaging indicate that many additional kinases and phosphatases regulate secretion in general and COPI in particular, calling for further investigation (Farhan et al., 2010; Chia et al., 2012). Similarly, sophisticated mass spectrometry (MS)-based phosphoproteome profiling studies have demonstrated intriguing changes in the phosphorylation of coatomer proteins in response to such stimuli as glucose or insulin signaling (Sacco et al., 2016) and circadian oscillation (Robles et al., 2017). The responsible kinases remain to be identified experimentally, but computational analyses of these MS datasets suggest that PKA, PKC and CKII may account for many of the observed phosphosite changes, perhaps including those on COPI proteins (Sacco et al., 2016). We anticipate that efforts combining classical genetics and biochemistry with state-of-the-art imaging and MS approaches will reveal fascinating new mechanisms and functions of COPI regulation by phosphorylation.

# O-GLCNACYLATION

O-linked β-N-acetylglucosamine (O-GlcNAc) is an abundant, single-sugar modification of serines and threonines on nuclear and cytoplasmic proteins (Hart, 2014; Bond and Hanover, 2015; Yang and Qian, 2017). Like phosphorylation, O-GlcNAc can cycle on and off substrates rapidly, sometimes on the timescale of minutes, thanks to the action of dedicated enzymes that add (O-GlcNAc transferase, OGT) and remove (O-GlcNAcase, OGA) the PTM. O-GlcNAc signaling governs myriad cellular processes and is dysregulated in many diseases, including cancer, diabetes and neurodegeneration (Hart, 2014; Bond and Hanover, 2015; Yang and Qian, 2017). Recent evidence also indicates that O-GlcNAcylation influences COPI trafficking (**Figure 2**). In a first report, Deng et al. (2014) used a protein microarray assay to discover direct biochemical interactions between OGT and several components of Golgi trafficking pathways, including ε-COP and several Rabs. These results hinted at a potential role for O-GlcNAcylation in COPI trafficking, but the relevant OGT substrate(s) remained unclear. Subsequently, our lab used a quantitative glycoproteomics approach and MS site-mapping to discover that γ1-COP (one of two mammalian γ isoforms) is O-GlcNAc-modified on at least 11 residues (Cox et al., 2018) (**Table 1**). Brefeldin A treatment reduced the O-GlcNAcylation of endogenous γ1- COP in human cells, suggesting a potential connection between coatomer glycosylation and pathway regulation (Cox et al., 2018). The functional impact of γ1-COP O-GlcNAcylation remains to be determined. However, because O-GlcNAc is a nutrientsensitive PTM with ubiquitous roles in metabolic signaling (Ong et al., 2018; Hart, 2019), it is tempting to speculate that O-GlcNAcylation of γ1-COP or other coatomer components regulates COPI activity based on metabolic state. In addition, phosphorylation and O-GlcNAc are both O-linked modifications and can compete for identical or nearby residues on substrate proteins, giving rise to complex functional interplay between these PTMs (Hart et al., 2011) (see also below). Of note, five of the O-GlcNAc sites we identified on γ1-COP are also reported phosphosites, suggesting that coatomer may be regulated by crosstalk between these two O-linked PTMs (Cox et al., 2018). Experiments to test this possibility are underway.

# MYRISTOYLATION

Perhaps the first evidence of PTMs in the COPI pathway came from the discovery that Arf1 is myristoylated at its N-terminus (Kahn et al., 1988) (**Table 1**). This form of acylation, previously known to occur on other mammalian and viral GTPases, proved to be essential for Arf1's membrane trafficking function (Franco et al., 1996; Kaczmarek et al., 2017). Mechanistically, myristoylation is required for the ability of GTP-bound Arf1 to bind target membranes and insert its α-helix into the bilayer (Antonny et al., 1997; Goldberg, 1998; Mossessova et al., 1998; Liu et al., 2009, 2010). To date, nearly all studies on Arf1 myristoylation have focused on its role in constitutive – as opposed to regulated – COPI trafficking. It will be interesting to learn from future work whether Arf1 myristoylation serves as a control point in any physiological contexts (e.g., if cells adjust the stoichiometry of available Arf1 via regulation of its acylation or the proteolytic removal of its myristoylated N-terminus). Several reports on the manipulation of Arf1 myristoylation by bacterial pathogens (see below) demonstrate the plausibility of this notion.

# UBIQUITINATION

Many COPI-relevant proteins are ubiquitinated in both yeast and mammals, including multiple coatomer components (Cohen et al., 2003; Hitchcock et al., 2003; Peng et al., 2003). However, little is known about the biochemical nature or functional impact

that ubiquitination has on COPI activity. In one investigation of this question, Maruyama et al. (2008) examined mammalian ε-COP ubiquitination by PIRH2, an E3 ligase involved in androgen receptor (AR) signaling (**Figure 3A**). PIRH2 binds ε-COP and directly ubiquitinates it, promoting its proteasomemediated degradation (Maruyama et al., 2008) (**Figure 3A**). In prostate cancer cells, dihydrotestosterone stimulation induced a PIRH2-AR interaction and concomitant ε-COP ubiquitination and degradation (Maruyama et al., 2008) (**Figure 3A**). In the same system, the authors demonstrated that overexpression of PIRH2 inhibits the secretion of prostate-specific antigen, suggesting a functional role for ubiquitination in linking hormone signaling to secretion (Maruyama et al., 2008).

In another example in yeast, the Kaminska lab built on prior proteomic studies (Hitchcock et al., 2003; Peng et al., 2003) to genetically dissect the role of ubiquitination in regulating actin remodeling and COPI trafficking (Jarmoszewicz et al., 2012). The authors demonstrated that a mutation in the rsp5 ubiquitin ligase gene inhibits trafficking in loss-of-function genetic backgrounds for ret1 or sec28, which encode α- and ε-COP, respectively (Jarmoszewicz et al., 2012). Rsp5/ret1 double mutants exhibited reduced retrograde trafficking of the ER chaperone Kar2p and a hypersensitivity to neomycin, a drug that competes with coatomer for binding to dibasic motifs on COPI cargoes (Jarmoszewicz et al., 2012). These results indicate a functional impact of Rsp5p-mediated ubiquitination on Golgi-to-ER transport. Interestingly, the authors also identified a role for the actin cytoskeleton in these trafficking effects, as cells mutated in arp2 (encoding an actin nucleating complex factor) or sla1 (encoding a component of the PNA1 actin regulatory complex) also exhibit trafficking defects in a rsp5 mutant background (Jarmoszewicz et al., 2012). It will be interesting to learn the direct ubiquitination targets of Rsp5p that mediate these effects on COPI trafficking and to determine whether its mammalian ortholog, NEDD4, has similar functions. NEDD4 is known to participate in other membrane trafficking events (Boase and Kumar, 2015), suggesting a potentially evolutionarily conserved impact on COPI as well.

# TRAFFICKING REGULATION THROUGH CARGO PTMs

The above studies focused on COPI regulation through modification of the trafficking machinery itself. However, several lines of evidence indicate that PTMs on vectorially transported COPI cargoes can also influence their trafficking during interphase, providing another layer of regulatory control. Perhaps the best-studied example of this regulation is on potassium channel subfamily K (KCNK, also known as TASK-1) proteins (**Figure 1B**). In initial work, the Goldstein lab demonstrated that KCNK3 comprises two motifs that regulate its trafficking: a canonical N-terminal dibasic ER retrieval motif, which binds β-COP, and a C-terminal motif, which binds 14-3-3β in a phosphorylation-dependent manner (O'Kelly et al., 2002) (**Figure 1B**). The authors reported that β-COP binding and retrograde trafficking keep KCNK3 predominantly in the ER, whereas phosphorylation of the C-terminal motif on Ser393 by PKA recruits 14-3-3β, displaces β-COP and allows anterograde KCNK3 trafficking to the plasma membrane for active channel function (O'Kelly et al., 2002) (**Figure 1B**). Follow-up studies by the same group further clarified the mechanism of this regulation, demonstrating that 14-3-3 binding to KCNK3 promotes the subsequent binding of the p11 annexin in some tissues to promote forward transport of the channel (O'Kelly and Goldstein, 2008).

In complementary work, the Schwappach lab showed that phosphorylation of the C-terminal motif prevented β-COP binding to KCNK3 and the paralogous channel KCNK9 (also called TASK-3), even in the absence of 14-3-3 proteins (Kilisch et al., 2016). Interestingly, the authors identified another nearby PKA phosphorylation site, Ser373, on KCNK3 alone, which,

when modified, inhibits COPI binding to permit anterograde transport (Kilisch et al., 2016) (**Table 1**). These results suggested that stimulus-induced PKA activity might modulate potassium signaling through regulated transport of ion channels. Supporting this model, the Schwappach lab reported that ATPsensitive potassium channels, comprising Kir6.2 and SUR1 subunits, are retained in the Golgi in ventricular cardiomyocytes (Arakel et al., 2014). β-adrenergic stimulation, as would occur in vivo through sympathetic nervous system action, activates PKA, which phosphorylates the C-terminal domain of Kir6.2, displacing coatomer proteins and facilitating trafficking to the cell surface (Arakel et al., 2014) (**Figure 1B**). Beyond potassium channels, similar phosphorylation events have been reported to regulate the COPI association and forward trafficking of other cargoes, including major histocompatibility complex proteins and nicotinic acetylcholine receptors, suggesting that this mode of regulation may be widespread (O'Kelly et al., 2002; Khalil et al., 2005; Sand et al., 2014).

Cargo ubiquitination is also known to affect COPI transport. For example, Xu et al. (2017) showed that a subset of COPI coats localizes to the early endosome in yeast, and that the N-terminal WD40 propeller domains of both β 0 - and α-COP bind to Lys63 linked polyubiquitin on proteins, such as the v-SNARE Snc1 (**Figure 3B**). Deletion of the β 0 -COP propeller (but not mutation of the dibasic binding site alone) trapped Snc1 at the early endosome, whereas replacing the WD40 domain with unrelated ubiquitin-binding domains restored the recycling of Snc1 to the plasma membrane (Xu et al., 2017) (**Figure 3B**). These results suggest a role for yeast COPI in plasma membrane-endosome recycling that is mediated by recognition of polyubiquitin, rather than a dibasic motifs, on cargoes (Xu et al., 2017) (**Figure 3B**).

In another example, the Aniento group reported that the N-glycosylation of a member of the p24 protein family affects the retrograde transport of the K/HDEL receptor ERD2 in plants (Pastor-Cantizano et al., 2017) (**Figure 4**). p24 proteins associate with COPI vesicles and likely participate in coatomer recruitment and cargo selection, particularly for glycosylphosphatidylinositol-anchored proteins, but they remain incompletely understood (Pastor-Cantizano et al., 2016). Aniento and coworkers demonstrated that N-glycosylation of Arabidopsis p24δ5 is required for its binding to ERD2 and for the COPIdependent trafficking of ERD2 from the Golgi to the ER (Pastor-Cantizano et al., 2017) (**Figure 4**). p24δ5 glycosylation does not affect its binding to the coatomer itself, and so the precise biochemical mechanism of this regulation remains to be determined (Pastor-Cantizano et al., 2017). However, several fungal and mammalian p24 proteins are glycosylated as well, suggesting a possible evolutionarily conserved mode of retrograde trafficking regulation through glycosylation, an interesting topic for future work.

As these examples show, a range of cargo PTMs can influence COPI transport (**Table 1**), raising the possibility of combinatorial control of COPI trafficking through multiple PTMs on individual cargoes. A recent report on hyaluronan synthase 2 (HAS2) by the Deen group illustrates this possibility (Melero-Fernandez de Mera et al., 2018) (**Table 1**). The authors showed that phosphorylation of HAS2 at Thr110 is required for its transit

from the ER to the Golgi and subsequently to the plasma membrane (Melero-Fernandez de Mera et al., 2018). In addition, HAS2 Ser221 is alternatively O-GlcNAcylated or phosphorylated, affecting HAS2 by controlling both its rate of anterograde trafficking through the Golgi and its rate of endocytosis from the plasma membrane to endolysosomes, where it is degraded (Melero-Fernandez de Mera et al., 2018). Finally, the authors show that HAS2 ubiquitination on Lys190 is essential for its activity and may play a role in its transport from the Golgi to the plasma membrane (Melero-Fernandez de Mera et al., 2018). At present, little is known about the upstream factors controlling HAS2 PTMs or whether they impinge directly on COPI transport and/or other trafficking pathways. Nevertheless, these results highlight the possibility that dynamic stimuli might signal through combinations of diverse PTMs to control cargo movement through COPI and other transport systems.

### PATHOGEN SUBVERSION OF COPI TRAFFICKING THROUGH PTMs

Like many fundamental cell biological processes, COPI trafficking can be subverted by pathogens to promote their own survival and growth, and PTMs have been implicated in this phenomenon (**Figure 5** and **Table 1**). For example, the Alto lab demonstrated that the IpaJ type III effector protein of Shigella flexneri is a cysteine protease specific for the myristoylated N-terminus of Arf proteins (Burnaevskiy et al., 2013) (**Figure 5A**). Shigella infection had been reported previously to disrupt Golgi morphology, but the mechanism remained uncertain. As the authors showed, IpaJ cleaves the peptide bond between the myristoylated-Gly2 and Arg3 residues of Arf1, effectively deacylating the protein and presumably directly causing the dramatic impact on COPI trafficking and Golgi homeostasis observed during Shigella infection (Burnaevskiy et al., 2013). Interestingly, IpaJ can cleave a range of N-myristoylated host proteins in vitro but exhibits remarkable specificity for Golgi GTPases in infected cells (Burnaevskiy et al., 2015). The specific advantages that Shigella derives from sabotaging COPI transport are not entirely clear, but these results raise the intriguing possibility that other pathogens – and

immunity-related GTPase M.

perhaps even endogenous cellular machinery – may manipulate trafficking through Arf1 deacylation.

In a second, recent example, Hansen et al. (2017) demonstrated that the human immunity-related GTPase M (IRGM) influences Golgi trafficking during hepatitis C virus (HCV) infection (**Figure 5B**). The authors show that IRGM localizes to the Golgi and participates in HCV-directed Golgi fragmentation (Hansen et al., 2017). Specifically, HCV infection triggers the IRGM-mediated phosphorylation of GBF1, probably by AMPK (Hansen et al., 2017). This PTM likely prolongs Arf1 activation, disrupting the COPI pathway and fragmenting the Golgi to promote viral replication (Hansen et al., 2017). It will be interesting to dissect the mechanism and effects of HCV's manipulation of IRGM during infection, and to determine the normal role that IRGM plays, if any, in the COPI pathway of healthy cells.

Rab1 is a small GTPase known to regulate several membrane trafficking pathways, including COPI (Hutagalung and Novick, 2011; Yang et al., 2016; Saraste and Marie, 2018). Recently, Rab1 has also emerged as a major target of several pathogen-directed PTMs, particularly by the bacterium Legionella pneumophila, the causative agent of Legionnaire's disease (Machner and Chen, 2011; Goody and Itzen, 2013; Misch, 2016). Legionella induces the reversible adenylylation of Rab1, with the secreted bacterial effector proteins SidM/DrrA and SidD covalently adding and removing adenosine monophosphate (AMP) to Rab1, respectively, at different time-points during infection (Muller et al., 2010, 2012; Neunuebel et al., 2011; Tan and Luo, 2011; Chen et al., 2013; Luitz et al., 2016) (**Table 1**). Rab1 AMPylation locks it into its active (GTP-bound) conformation, presumably to manipulate host membrane transport pathways, though whether COPI trafficking or other processes regulated by Rab1 (e.g., endocytosis, autophagy) are the key targets for Legionella remains unclear (Derre and Isberg, 2004; Kagan et al., 2004; Joshi and Swanson, 2011; Mukherjee et al., 2011). Complicating the picture, the coopting of Rab1 by SidM/DrrA binding also occurs through PTM-independent mechanisms, including alteration of the Rab1 GTPase activity and blocking the association of Rab1 with host GDP-dissociation inhibitor (GDI) proteins (Machner and Isberg, 2006, 2007; Murata et al., 2006; Ingmundson et al., 2007). More work will be needed to understand the mechanisms and downstream functional effects of SidM/DrrA and SidD during Legionella infection.

Remarkably, Rab1 is subject to other PTMs by Legionella effectors as well. For example, the bacterial protein AnkX/LegA8 attaches a phosphocholine lipid moiety to Rab1, a PTM that can be removed by pathogen-encoded Lem3/lpg0696 (Mukherjee et al., 2011; Tan et al., 2011; Goody et al., 2012; Allgood et al., 2017) (**Table 1**). Phosphocholination inhibits Rab1 function and is required for membrane remodeling during Legionella infection (Mukherjee et al., 2011; Tan et al., 2011; Goody et al., 2012; Allgood et al., 2017). More recently, the Legionella effector SetA was shown to glucosylate Rab1, inhibiting its GTPase activity and its interaction with GDI proteins (Wang et al., 2018) (**Table 1**). Taken together, these studies demonstrate that Legionella invests significant resources in hijacking Rab1, suggesting an important role for the remodeling of endomembrane traffic during infection. It will be important to learn the particular pathways controlled by Rab1 that are most critical during Legionella pathogenesis, and whether these are differentially affected by discrete pathogen-directed PTMs (Mukherjee et al., 2011; Tan et al., 2011; Goody et al., 2012; Allgood et al., 2017; Wang et al., 2018).

Recently, research on pathogen/Rab1 interactions has facilitated separate discoveries on the native regulation of Rab1 by endogenous PTMs. Mammalian TGF-β-activated kinase 1 (TAK1), which regulates the AP-1 and NF-κB transcription pathways during innate immune signaling, was found to phosphorylate Rab1 at a site on its switch II region, near the hotspot modified by Legionella effectors (Levin et al., 2016) (**Table 1**). Phosphorylation of Rab1 by TAK1 disrupts its interaction with GDI proteins (but not GEFs or GAPs), and

is essential for Rab1 function (Levin et al., 2016). Notably, Legionella infection reduces Rab1 phosphorylation by TAK1, suggesting that the pathogen may subvert normal cellular Rab1 PTMs as a way of interfering with TAK1-mediated innate immune signaling (Levin et al., 2016). This work provides an excellent example of how discoveries from the eukaryotic trafficking and microbial pathogenesis fields can illuminate each other.

#### CONCLUSION

Partitioning proteins and lipids into specialized compartments is a signature feature of all eukaryotes. To create and maintain this organellar homeostasis, cells rely on a variety of essential and highly regulated trafficking pathways, including COPI. Since its discovery, great strides have been made in understanding the structure and function of COPI in constitutive vesicle formation and trafficking. Despite these achievements, however, we know relatively little about the dynamic regulation of COPI transport in interphase cells experiencing fluctuating signals, nutrients, developmental cues or stresses. PTMs clearly provide one – though not the only – broad mechanism that eukaryotes use to tune COPI activity. The studies reviewed here highlight the role that diverse PTMs play in interphase COPI trafficking. We anticipate that recent advances in high-content imaging, MS, genome engineering, structural biology and computational modeling will reveal new examples as well, eventually leading to an integrated model for how cells modulate COPI transport in real time. To this end, one important future goal will

#### REFERENCES


be to comprehensively site-map functionally important PTMs on coatomer proteins, Arfs and ArfGEFs, and integrate this information with improved structural data on the COPI coat, in order to gain biophysical insight into the mechanism of trafficking regulation by COPI PTMs. Knowledge from such efforts may also shed light on other, longstanding questions in the COPI field, such as why mammals encode multiple similar isoforms of some critical COPI components (e.g., γ- and ζ -COP, GBF1 isoforms) and how these may be differentially regulated by PTMs, how under-studied PTMs govern other components of the COPI pathway (e.g., ArfGAPs), how COPI transport influences the distribution of lipids and glycans, which PTMs are required in intact tissues and organisms to maintain COPI homeostasis, and how these processes might be dysregulated in disease. We expect that exciting new answers to these questions will be found in the coming years, with PTMs playing a central role.

# AUTHOR CONTRIBUTIONS

PL and MB wrote the manuscript, performed revisions, and read and approved the submitted version.

# FUNDING

Work in the Boyce Lab related to this study was supported by the National Institute of General Medical Sciences grant R01GM117473 to MB.





**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 Luo and Boyce. 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.

# Misplaced Golgi Elements Produce Randomly Oriented Microtubules and Aberrant Cortical Arrays of Microtubules in Dystrophic Skeletal Muscle Fibers

Sarah Oddoux † , Davide Randazzo† , Aster Kenea, Bruno Alonso, Kristien J. M. Zaal and Evelyn Ralston\*

*Light Imaging Section, Office of Science and Technology, National Institute of Arthritis and Musculoskeletal and Skin Diseases, National Institutes of Health, Bethesda, MD, United States*

#### Edited by:

*Jaakko Saraste, University of Bergen, Norway*

#### Reviewed by:

*Martin Lowe, University of Manchester, United Kingdom Antonino Colanzi, Institute of Biochemistry and Cell Biology (CNR), Italy*

#### \*Correspondence:

*Evelyn Ralston evelyn.ralston@nih.gov*

*†These authors have contributed equally to this work*

#### Specialty section:

*This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology*

Received: *13 May 2019* Accepted: *13 August 2019* Published: *18 September 2019*

#### Citation:

*Oddoux S, Randazzo D, Kenea A, Alonso B, Zaal KJM and Ralston E (2019) Misplaced Golgi Elements Produce Randomly Oriented Microtubules and Aberrant Cortical Arrays of Microtubules in Dystrophic Skeletal Muscle Fibers. Front. Cell Dev. Biol. 7:176. doi: 10.3389/fcell.2019.00176* Differentiated mammalian cells and tissues, such as skeletal muscle fibers, acquire an organization of Golgi complex and microtubules profoundly different from that in proliferating cells and still poorly understood. In adult rodent skeletal muscle, the multinucleated muscle fibers have hundreds of Golgi elements (GE), small stacks of cisternae that serve as microtubule-organizing centers. We are interested in the role of the GE in organizing a peculiar grid of microtubules located in the fiber cortex, against the sarcolemma. Modifications of this grid in the *mdx* mouse model of Duchenne muscular dystrophy have led to identifying dystrophin, the protein missing in both human disease and mouse model, as a microtubule guide. Compared to wild-type (WT), *mdx* microtubules are disordered and more dense and they have been linked to the dystrophic pathology. GE themselves are disordered in *mdx*. Here, to identify the causes of GE and microtubule alterations in the *mdx* muscle, we follow GFP-tagged microtubule markers in live *mdx* fibers and investigate the recovery of GE and microtubules after treatment with nocodazole. We find that *mdx* microtubules grow 10% faster but in 30% shorter bouts and that they begin to form a tangled network, rather than an orthogonal grid, right after nucleation from GE. Strikingly, a large fraction of microtubules in *mdx* muscle fibers seem to dissociate from GE after nucleation. Moreover, we report that *mdx* GE are mispositioned and increased in number and size. These results were replicated in WT fibers overexpressing the beta-tubulin tubb6, which is elevated in Duchenne muscular dystrophy, in *mdx* and in regenerating muscle. Finally, we examine the association of GE with ER exit sites and ER-to-Golgi intermediate compartment, which starts during muscle differentiation, and find it persisting in *mdx* and tubb6 overexpressing fibers. We conclude that GE are full, small, Golgi complexes anchored, and positioned through ER Exit Sites. We propose a model in which GE mispositioning, together with the absence of microtubule guidance due to the lack of dystrophin, determines the differences in GE and microtubule organization between WT and *mdx* muscle fibers.

Keywords: muscle, mdx, Golgi, microtubules, ERES, ERGIC, MTOC, tubb6

#### INTRODUCTION

It has long been known that Golgi complex (GC) and microtubules are tightly interwoven (Burkhardt, 1998). However, there is a large gap between their organization in proliferating cells and in differentiated tissues (Yadav and Linstedt, 2011). In proliferating cells, the main features of GC and microtubule organization are now understood at a molecular and mechanistic level (Wei and Seemann, 2017). Briefly, microtubules form a single radial array growing from the centrosome (or microtubule-organizing center, MTOC) to the cell membrane. A single GC is dynamically kept near the centrosome through the action of minus-end directed microtubule motors. This organization is well-suited to the needs of cell division in which microtubules play an essential role. Some proliferating cells, however, show two distinct sets of microtubules: one radial array nucleating from the centrosome and one directional array originating from the GC itself (Efimov et al., 2007; Rios, 2014) and involved in functions such as cell migration (Miller et al., 2009).

When cells leave the mitotic cycle and differentiate, microtubule, and GC organization are profoundly changed to support the architecture and function of the developing tissue. Different cell types such as those found in muscle, plants, and brain, adopt very different, complex, subcellular organizations (Oddoux et al., 2013; Chen et al., 2016; Mikhaylova et al., 2016). Few of these organizations are understood at a mechanistic level and even at a descriptive level there remains a lot to understand. This is the case in myogenesis (skeletal muscle development). Extensive reorganizations take place during differentiation, which can be studied in cell culture. However, muscle cell cultures do not replicate the subsequent changes that take place during muscle maturation and regeneration and are key to the role of GC and microtubules in muscle diseases.

Undifferentiated muscle cells (myoblasts) have the typical organization of proliferating cells described above. As musclespecific transcription factors become expressed, myoblasts stop dividing. Their MTOC redistributes to the nuclear membrane (Tassin et al., 1985a). Microtubule nucleation and GC follow (Tassin et al., 1985b). The single GC is replaced by numerous smaller Golgi elements (GE) and the radial microtubule array replaced by mostly parallel microtubules, which contribute to elongation of the cell shape (Zaal et al., 2011). GE are small stacks of cisternae. They are positioned next to ER exit sites (ERES), as are GC fragments of proliferating cells treated with drugs that depolymerize microtubules (Cole et al., 1996; Lu et al., 2001). GE and ERES, from this stage on, are not distinct organelles positioned close by but are single secretory units formed of two parts. The next important change in cellular architecture is the fusion of the differentiated myoblasts (also known as myocytes) to form multinucleated myotubes. Myotubes in muscle cell cultures can reach hundreds of microns in length, with tens of nuclei distributed from end to end in a seamless cytoplasm.

In vivo, myotubes mature into muscle fibers, arguably the largest mammalian cells (with radii up to 100µm). Innervation is an essential step in the maturation of myotubes into muscle fibers. Once innervated, muscle fibers are exposed to stimulation by patterned activity, which regulates fiber contraction, speed, metabolism (Schiaffino et al., 2015), and subcellular organization, including that of microtubules and GE (Ralston et al., 2001). Muscle fibers are divided into different types, related to their activity (fast, slow, and intermediate) but they are plastic: changes in contractile activity, experimental or due to disease, remodel muscle fibers (Dowling et al., 2016).

When considering the radial organization of muscle fibers, two domains can be distinguished: core and cortical areas. Their main distinguishing feature is the concentration, in the core, of the contractile myofibrils and of the organelles involved in excitation-contraction; the cortex is a thin lining that resembles the cytoplasm of non-muscle cells. The fiber cortex houses the nuclei which are distributed evenly, radially and longitudinally. Mitochondria, microtubules and GE are found in both cortex and core, but their organization is different in each domain and depends on fiber type (Ralston et al., 1999). In particular, GE and microtubules form a unique orthogonal grid in the fiber cortex of all except slowtwitch fibers, i.e., in most muscles of the mouse. This grid is composed of single and bundled microtubules, with GE positioned at microtubule intersections. The microtubule grid periodicity matches that of the contractile proteins; longitudinal microtubules are aligned with the axis of the muscle fiber and transverse microtubules are aligned with Z-disks (Kaisto and Metsikkö, 2003).

To find out whether this microtubule grid is dynamic and to resolve the "chicken-egg" question of whether microtubules position GE or GE position microtubules, we investigated them both in live muscle fibers expressing fluorescently tagged markers (Oddoux et al., 2013). These experiments were done on collagenase-dissociated fibers of the mouse flexor digitorum brevis (FDB) muscle and were further validated by intravital recordings (Oddoux et al., 2013). We found out that muscle GE are static and nucleate dynamic microtubules that grow along stable microtubules, all together forming a permanent grid superimposed on the orthogonal network formed by dystrophin (Prins et al., 2009).

Dystrophin is the protein that is absent in Duchenne muscular dystrophy (DMD), an X-linked fatal muscle disease, and in the mdx mouse, a DMD animal model (Bulfield et al., 1984; Hoffman et al., 1987). Several observations led to the proposal that microtubule-dystrophin interactions play a role not only in muscle fiber organization but also in the pathology of DMD. First, microtubules and GE of mdx muscles were found to be disorganized and rescued by expression of a micro-dystrophin construct (Percival et al., 2007). Importantly, dystrophin was shown to bind microtubules (Prins et al., 2009; Belanto et al., 2014) and to be positioned along them (Prins et al., 2009).

**Abbreviations:** DMD, Duchenne muscular dystrophy; EDL, extensor digitorum longus; ERES, ER exit site; ERGIC, ER-Golgi intermediate compartment; FDB, flexor digitorum brevis; fps, frame per second; GE, Golgi Element; GFP, Green fluorescent protein; MTOC, Microtubule organizing center; SR, sarcoplasmic reticulum; TGN, trans-Golgi network; WT, wild-type.

Microtubules have also been identified as part of the dystrophin complexome (Murphy and Ohlendieck, 2016). Dystrophin is a constituent of the costameres (Ervasti, 2003), protein assemblies of the muscle cortex that form periodic links between sarcolemma and myofibrils at the level of the Zdisks, thus perfectly positioned to interact with cortical microtubules. Finally, microtubules were implicated in dystrophinopathy: the production of reactive oxygen species (ROS) and other hallmarks of DMD, upon experimental stretching of mdx muscle fibers ex vivo, could be prevented by depolymerizing microtubules (Khairallah et al., 2012; Kerr et al., 2015).

However, recent work has made matters more complicated. First, a grid-like microtubule organization and normal physiology was restored in a transgenic mdx mouse line expressing a dystrophin construct lacking the microtubulebinding domain (Belanto et al., 2016). Then, microtubule organization was rescued in an mdx mouse line following elimination of ROS production without restoration of dystrophin expression (Loehr et al., 2018). The same work showed that fibrosis plays a larger role than microtubules in mdx pathology. Finally, we found out that experimentally overexpressing the β-tubulin tubb6, which is highly elevated in DMD and mdx muscle, destroys the microtubule grid of wild-type (WT) mouse muscles (Randazzo et al., 2019). Conversely, knocking down tubb6 in mdx muscle fibers improved their microtubule organization (Randazzo et al., 2019). Thus, dystrophin seemed neither necessary nor sufficient to organize muscle microtubules. Additionally, it is known that the mere presence of dystrophin in a muscle fiber is not sufficient to ensure its positioning in the costameres. Instead, dystrophin binding to the costameres depends on a cascade of proteins including ankyrin-B, spectrin, dynactin-4, and, amazingly, microtubules (Ayalon et al., 2008, 2011). Furthermore, dystrophin positioning, like that of microtubules and GE, depends on muscle activity (Bezakova and Lømo, 2001).

Our previous work with WT muscle fibers led to a model, schematically represented in **Figure 1**. Several features were important: first, GE themselves formed a grid and were anchored along the transverse and longitudinal components of dystrophin; second, microtubules grew from the GE in a limited range of directions, mostly transverse and longitudinal; third, dystrophin was available for guidance. Here, we investigate the consequences of the absence of guidance by dystrophin in mdx fibers. We follow microtubule dynamics in live and fixed FDB fibers of the mdx mouse and follow recovery of the microtubule network from nocodazole depolymerization. We find that the differences in microtubule directionality and density can be observed at the earliest stage of microtubule growth but that the main features of the WT network, i.e., a dynamic though permanent microtubule network are maintained in mdx fibers. Furthermore, GE are associated with ERES in mdx as they are in WT fibers. We conclude that mispositioning of muscle GE, which are full, though small, Golgi complexes, and the random orientation of growing microtubules cause microtubule disorder in mdx muscle.

FIGURE 1 | Model of GE and microtubule organization in the cortex of WT muscle fibers. GE and microtubules both form an orthogonal grid that is aligned with dystrophin. Dystrophin is part of the costameres (muscle "ribs") in the transverse direction and also forms longitudinal lines. Dynamic microtubules, with the protein EB3 marking their growing plus-end, are nucleated on the GE and grow bidirectionally mostly along dystrophin tracks. GE are static and the whole grid is stable over time.

# RESULTS

## GFP-Tagged Microtubule Markers Visualized in Live WT and mdx Muscle Fibers Highlight Directionality and Density Differences

To compare the dynamic features of mdx GE and microtubules to those of WT, we followed fluorescent markers of αtubulin (GFP-tubulin) and of the plus-end microtubule protein EB3 (EB3-GFP) in live FDB fibers of the two genotypes (**Figure 2**) as previously done for WT muscle fibers only (Oddoux et al., 2013). The video recordings (found in the **Supplementary Material**) are presented here as color-coded projections in which color serves as an indication of time (**Figures 2A,C,F,G** represent **Supplementary Videos 1, 3, 6, 7**). Panels B, D, and E present single images of GFP-tubulin recordings (**Supplementary Videos 2, 4, 5**).

An even superficial examination of the panels reveals mdx vs. WT differences in microtubule directionality: WT panels show numerous transverse (vertical) microtubules lacking from mdx panels. These differences were confirmed by analysis with the TeDT microtubule directionality software (Liu and Ralston, 2014; **Supplementary Figure 2A**).

FIGURE 2 | While *mdx* and WT muscle microtubules differ in their pattern and density, both have dynamic but stable microtubule networks nucleated from stationary MTOC. Muscle fibers from the *flexor digitorum brevis* muscle expressing GFP-tubulin or EB3-GFP, as indicated on the left side, were imaged at the rate of 93 f/min with a 63x N.A. 1.4 objective on a Leica SP5 confocal microscope fitted with a heating stage and objective heater at 37 deg. (A,C,F,G) show projections of 2 min recordings. These recordings are in the Supplementary Videos 1–7 and correspond to (A–G). The long axes of the muscle fibers are parallel with the long side of the images. The projections were color-coded to indicate time, as shown by the bar above the images. The main differences between WT and *mdx* microtubules are their directionality and density. (A,F) show many transverse microtubules while panels (C,G) show very few, if any. Because microtubules form bundles and grow along one another, GFP-tubulin recordings show little color, unless a new track is formed or a microtubule shrinks. In EB3 recordings, only growing microtubules are seen since EB3 detaches from shrinking microtubules. White tracks (merging of rainbow colors) are those present through the recordings. Arrowhead pairs in images (B,D,E) bookend lines selected to draw kymographs (1–3) of GFP-tubulin sequences. A kymograph is a time-distance graph obtained by repetition of the same line for each time-point. Kymograph 1 reflects successive production of microtubules that grow along the line at steady velocity; kymographs 2–3 reflect the presence of unstable microtubules both along and across the kymograph line. For EB3-GFP recordings, the color coding in panels H and I represents the relative instantaneous velocity from blue for the slowest to red for the fastest. Kymographs of selected tracks (boxes 1–6) show that EB3 comets do not progress at a constant speed but speed up and slow down abruptly but EB3-GFP growth velocity is independent of the growth direction. These projections were thresholded and binarized to suppress the background striations visible in F, G and inherent to EB3-GFP expression. Bars: (A–H): 10µm; (I): 2µm. In the kymographs, vertical bars: 30 s, GFP-tubulin horizontal bars: 5µm; EB3-GFP horizontal bars: 2µm.

The color coding of the projections also gives a snapshot of microtubule dynamics: stationary microtubules appear white or lightly colored (**Figures 2A,C**), while projection of the growing plus-ends forms tracks of successive colors (**Figures 2F,G**). In this respect, mdx recordings look similar to WT ones, indicating that, in both, dynamic microtubules grow along stable microtubules. Kymographs (boxes 1–3) give a different visual presentation of the complexity of microtubule dynamics in mdx. To obtain these, a line was drawn along a microtubule track (between two arrowheads in **Figures 2B,D,E**) and plotted as a stack with the line repeated at each time-point (see methods). The WT kymograph, with successive triangles of similar slopes, indicates repetitive growth of microtubules along the kymograph line, from a common point of origin and with constant velocity. The mdx kymographs, in contrast, reflect longer or shorter-lived and random-oriented microtubules that do not precisely follow the central line. In addition, the density of mdx microtubules is higher than that of WT.

The interpretation of the visual differences in the kymographs is supported by quantitation of microtubule growth parameters (**Table 1**). Two parameters differ significantly: mdx microtubules grow significantly faster than WT ones by ∼10% and their growth duration is shorter by ∼30%.

In all projections, MTOC (arrows in **Figures 2A–G**) appear as lighter colored points shown in videos and kymographs to be static. Watching them in loops, we started noticing differences between WT and mdx recordings. WT MTOC are distant enough from one another to not interfere. Microtubules grow periodically in only a few directions, mainly along or at 90 degrees off the fiber axis. In mdx recordings, in contrast, new microtubule growth appears stochastic in both direction and timing. Microtubules originating elsewhere often cross the area under observation. Examples of one WT and one mdx MTOC and surroundings are shown in projection in **Supplementary Figure 1** and in **Supplementary Video 8** (WT) and **Supplementary Video 9** (mdx), respectively.

Thus, the presence of dystrophin seems to be necessary for spatial and temporal organization of microtubules.

### Differences Between mdx and WT Golgi-Microtubule Patterns Manifest Early on During Microtubule Recovery From Nocodazole-Induced Depolymerization

To verify that GE are the MTOC of mdx as they are those of WT microtubules, and to find out more about microtubule orientation at early growth stages, we decided to disassemble the muscle fiber microtubule network and to watch GE and microtubules during recovery. After treatment for 4 h with the drug nocodazole to depolymerize microtubules (see methods), microtubules were rare, with only stable, nocodazole-resistant microtubules left (not shown). Nocodazole was washed out and fibers were left to recover for 2, 5, and 15 min. Untreated and recovered samples were then stained for the cis-Golgi marker GM130 and for tyrosinated α-tubulin, a marker of dynamic microtubules. Representative images from three independent experiments as well as directionality measurements are shown in **Figure 3**. After 2 min of recovery, GE were surrounded by short microtubules in both genotypes but were positioned differently. WT GE often formed doublets (arrows) aligned with the axis of the fibers, while mdx GE formed fewer doublets and were unaligned. An orthogonal grid could be superimposed on an image of WT but not of mdx GE. Most microtubules were shorter than 1µm and surrounded by GE but some appeared to be detached from GE. At this stage TABLE 1 | Microtubule dynamic parameters (α-tubulin-GFP), *mdx* vs. WT.


*Movement analysis was done in ImageJ as described in Methods. Image sequences were taken from 3 WT and 3 mdx FDB muscles, each one from a different animal. For each one, 15 fibers were imaged for 2 min at 0.65 frames/ms. For WT fibers, a total of 42 microtubules were included in the calculations; for mdx fibers, a total of 47 microtubules. The stars (*\**) tag the p values that indicate statistical significance.*

there were no significant differences in the directionality plots which did not show any preferential directionality. Between 2 and 5 min, microtubules grew and established significant differences in density and directionality between mdx and WT samples. A majority of WT microtubules remained connected to GE but mdx microtubules filled out the space. Microtubule clusters apparently without GE, some shaped as asters, were present (**Figure 3** arrowheads and **Supplementary Figure 2A**). We counted such microtubule clusters in confocal images from three independent experiments. In WT fibers, 28% (n = 126) had no GM130 staining. In mdx fibers this proportion increased to 55% (n = 162). The difference between mdx and WT was highly significant (p < 0.001). At 15 min recovery, a full network was reorganized that appeared to link all microtubules. The muscle fiber surface occupied by microtubules and by nucleating GE differed between mdx and WT as did the ratio of microtubules to GE number (**Supplementary Figure 2B**). Thus, GE remain the main MTOC in mdx muscle fibers (besides nuclei) but they are positioned in a disordered way. The long treatment in nocodazole did not cause further fragmentation of GE as would be the case for a classical GC. It did not cause their fusion either. Thus, GE remain anchored in the absence of dystrophin but the anchoring points are not positioned along a grid. Anchoring and positioning of GE may be regulated by distinct mechanisms. Furthermore, the presence in mdx fibers of an increased proportion of microtubules not clearly associated with GE likely contributes to the higher density and the disorder of the mdx microtubule network.

#### Nuclei-Originating Microtubules Are Stabilized in mdx Fibers

In addition to GE-nucleated microtubules, the cortical area of muscle fibers has microtubules originating from nuclear membranes. In WT fibers, such microtubules join the grid and do not stand out (Oddoux et al., 2013).

nocodazole washout. Impressively, the original network organization is almost recovered in 15 min. At 2 min, GE appear as mostly independent seeds with short microtubules. After 5 min, microtubules form asters (groups of microtubules with a common point of origin) and start to connect. Some of the asters, especially in *mdx* samples (arrowheads) do not seem to be associated with a GE (see text). In WT samples, control fibers show groups of GE along microtubule bundles (arrow). After 2 and 5 min recovery, only pairs of GE (arrows) are seen. At 15 min they appear to be connecting. The directionality of the MTs at each stage was analyzed with the TeDT software (right side column). The curves represent the normalized frequency of the angles that microtubules form with the fiber axis, with 0◦ /180◦ corresponding to the longitudinal fiber axis. In the control, significant differences exist along the whole range of angles: WT microtubules have a peak at 90◦ that is absent in *mdx (Continued)*

FIGURE 3 | microtubules and are also more polarized into longitudinal and transverse microtubules while *mdx* microtubules are found in more orientations. After 2 min of nocodazole washout, the curves are very close but 5 min suffice to make them significantly different. After 15 min, the distributions and density of WT and *mdx* MTs are similar to those in the steady state (\**p* < 0.0001). Directionality measurements were done on two areas per fiber with 18 *mdx* and 17 WT steady-state fibers, 16 *mdx* and 14 WT fibers for 2 min, 15 *mdx* and 15 WT fibers for 5 min and 20 *mdx* and 20 WT fibers for 15 min time points. The fibers were representative of three independent experiments. Bar: 8µm.

However, in mdx FDB fibers stained for dynamic and stable microtubules we noticed long microtubule stretches radiating from nuclear membranes (**Figure 4A**, red) and positive for detyrosinated tubulin (**Figure 4A**, green), which marks stable microtubules. Similar patterns were also found in GFP-tubulin recordings of mdx fibers (**Figure 4B**). Thus, microtubules originating from nuclear membranes are stabilized and reach longer distances in mdx than WT fibers, forming a distinct set of microtubules, often linking neighboring nuclei. Nuclear dynamics are known to be altered in mdx fibers (Iyer et al., 2017).

detyrosinated tubulin (green) and nuclei (blue); (B) Single frames of GFP-tubulin recordings. Bars: (A) 5µm; (B) 7.5µm.

# GE Positioning and Size Distribution Are Altered in Both mdx and tubb6 OE Fibers

If GE mispositioning is part of the mdx microtubule disorder, we expect other conditions that cause a similar microtubule disorder to involve GE alterations as well. We thus tested the effect on GE of overexpression of the β-tubulin tubb6 in WT fibers, a condition that mimics several aspects of mdx dystrophinopathy (Randazzo et al., 2019). After staining WT, mdx, and tubb6 OE fibers for the cis-Golgi protein GM130 and for microtubules we noticed differences in GE positioning (**Figure 5A**). In WT fibers, most GE (79%) are found in pairs

or small groups, along microtubule bundles (see **Figure 5A** and enlarged details in **Figure 5B**) that are parallel to the fiber axis. Such groups are less frequent in mdx (28%) whose GE are found along bundled or single microtubules that have various orientations at an angle with the fiber axis. Tubb6 OE fibers showed similarly dense and disordered microtubules with an intermediate proportion of grouped GE (50%) and occasional longitudinal bundles. Quantitation is presented in **Table 2**. We also noted that the size of GE is more variable in mdx and tubb6 OE WT fibers than in WT fibers, with some of them quite larger. This is not related to a change in the whole fiber: the proportion of fiber surface occupied by GE in muscle fibers is significantly increased for mdx and tubb6 OE fibers (**Figure 5C**). Immunoblots of WT and mdx gastrocnemius muscle extracts for GM130 also showed a significant increase of the GE marker (**Figures 5D,E**). Golgi distribution is thus disordered in tubb6 OE fibers as in mdx fibers and the surface of GE in both mdx and tubb6 OE fibers is significantly increased compared to WT.

### WT GE Aligned in Rows Do Not Share a Homogenous cis-trans Polarity

In the cartoon of **Figure 1**, GE are represented as small blobs, as we view them in light microscopy with a single marker. Staining of muscle fibers with markers of both cis- and trans-Golgi cisternae, however, resolves cis-trans polarity of the GE (Ploug et al., 1998). Around nuclei, GE have a uniform polarity with the cis-trans axis pointing away from the nuclear membrane. The orientation of the GE along costameres could impose a directionality to the microtubules they nucleate. If this were the case, we would expect a row of GE along WT costameres to have a uniform cis-trans polarity, but we would expect mdx

increase in GE components. Bars: 10µm.



*Confocal images of GM130 and* α*-tubulin staining were counted in ImageJ for the total number of GE and manually for the number of those GE associated with microtubule bundles, with nuclei, and with transverse microtubules. Remaining GE are single and associated with unbundled microtubules.*

TABLE 3 | GM130 puncta association with markers of ERES, ERGIC, and TGN.


*To examine whether muscle GE all represent full Golgi, single confocal images from WT, mdx, and GFP-tubb6-overexpressing FDB fibers were stained for GM130 and Sec31, GLUT4, or p58. Confocal images were collected. They were opened in Photoshop and GM130 puncta were annotated as associated with the 2nd marker or not. The table shows the numbers of puncta counted.*

and tubb6 OE GE to be randomly oriented. To evaluate this hypothesis, we needed a trans-Golgi marker but none of the trans- specific antibodies that we tested gave an acceptable staining of mouse muscle fibers. We then used the P-1 antibody against the muscle and fat cell-specific glucose transporter GLUT4, which accumulates in the TGN and colocalizes with TGN38 (Ploug et al., 1998). We carried out imaging in the super-resolution Lightning mode of the Leica SP8 confocal and displayed the results in 3D (**Figure 6**, see methods). GLUT4 is not only in the TGN, hence small puncta show up in the staining. In WT fibers, however (six images with a total of 159 GE), we found no uniform polarity along GE rows (**Figure 6A**), although the expected cis-trans polarity was present in the perinuclear GE (**Figure 6B**). Occasional close pairs of GE had common polarity (**Figure 6C**, black underlining), others did not (**Figure 6C**, white underlining). Mdx and tubb6 OE fibers showed equally uncoordinated GE orientation (data not shown). We therefore rule out alignment of GE with all cis-trans axes in the same direction as a source of microtubule alignment in WT muscle fibers.

### GE, ERES, and ER to Golgi Intermediate Compartment (ERGIC) Remain Juxtaposed in mdx and tubb6 OE Muscle Fibers and Undergo Correlated Volume Changes

The concerted reorganization that brings together Golgi and ERES during muscle differentiation (Lu et al., 2001) turns GE into dynamically tethered extensions of the ER and is thus important for their localization. We asked whether this organization is maintained in mdx and tubb6 OE fibers and whether ERGIC are reorganized simultaneously to ERES (Saraste and Marie, 2018). Immunostaining of differentiating C2 muscle cells for GM130 and the ERES protein Sec31 (Shugrue et al., 1999) or the ERGIC protein p58 (Lahtinen et al., 1996) shows a similar reorganization of GE and ERGIC (**Supplementary Figure 3**). Staining of FDB fibers (**Figure 7A** and Rahkila et al., 1997) reveals the tight assemblies of GE-ERES and GE-ERGIC. In mdx and tubb6 OE fibers they remain but are larger. ERES in each of the three samples surround GE, while p58 appears colocalized with the cis-Golgi marker, as expected. Immunoblots show both Sec31 and p58 significantly increased in mdx fibers (**Figures 7B–D**) and volume measurements in the software Imaris show different distributions and larger sizes in mdx and tubb6 OE fibers (**Figure 7E**). Finally intensity measurements show that GE and ERES staining correlate (**Figure 7F**). It is generally accepted that reciprocal interactions between GC and ERES determine their size, which is linked to the demand for protein transport (Hammond and Glick, 2000; Stephens, 2003; Sengupta and Linstedt, 2011; Glick, 2014).

We then examined GE and ERES association in Lightning super-resolution and in 3D. The images revealed, unexpectedly, that GE-ERES in mdx and WT are polarized, with ERES on the sarcolemmal side (**Figure 8**), suggesting that cortical muscle GE are associated with an ER-like subsarcolemmal tubular system described by Jayasinghe et al. (2013). The association between GE and ERES is thus maintained in mdx and tubb6 OE fibers despite changes in their respective sizes, possibly indicating increased protein trafficking.

Finally, counting GM130's pairwise association with ERES, ERGIC and TGN markers, in WT, mdx, and GFP-tubb6 OE fibers (**Table 3**) indicated that muscle GE are all small but fully functional Golgi complexes, unlike neuronal Golgi which include Golgi outposts and an incomplete Golgi-related satellite organelle (Mikhaylova et al., 2016).

### Parallels Between Fiber Cortex Microtubules and Fiber Core SR

The cortical microtubule grid differs in its organization from the core microtubules, generally longitudinal without a transverse

component. However, we often encounter cortical microtubule bundles, in mdx fibers, that form a 20–40 deg angle with the longitudinal fiber axis (see for example box 2 in **Figure 5**). We were curious about this recurrent orientation and expected to observe it in Second Harmonic Generation (SHG) images of muscle fibers. SHG visualizes muscle myosin heavy chain without need for staining and since SHG is visualized on a confocal microscope it would be simple to determine at what depth such angles form. However, this was not the case. We found an explanation in electron micrographs of muscle fibers of the extensor digitorum longus (EDL) mdx mouse muscle (**Figures 9A–C**). Although the Z disks, I and A bands are reasonably well-aligned in these fibers, they are interrupted by cytoplasmic incursions at an angle of ∼20 deg. that mostly contain longitudinal SR tubules (**Figure 9C**, arrow). In fibers that are richer in mitochondria these are also present and we encountered a GE at the origin of such channels (**Figure 9B**). Similar angular incursions can be seen in autofluorescence images of mdx (**Figure 9D**) but not of WT fibers (**Figure 9E**). Finally, a microtubule image (**Figure 9F**) collected at the corecortex contact of a fiber stained for calsequestrin, an SR marker (**Figure 9G**), show similar interruptions of the normal pattern.

FIGURE 7 | GE-ERES and GE-ERGIC assemblies in *mdx* and tubb6 OE fibers are larger than in WT fibers. (A) In mature FDB muscle fibers stained for GM130 (red) and Sec31 or p58 (green), GE are more numerous than in muscle cultures (Supplementary Figure 3) but smaller and tightly associated with ERES or ERGIC. The three subcellular compartments differ in number and size between WT, *mdx*, and tubb6 OE fibers. The fixation procedure for p58 differed from that for Sec31 (see methods), possibly contributing to a higher background for p58. (B) Immunoblots of *gastrocnemius* muscle show that there is a 3-fold significant increase of Sec31 (\*\**p* < 0.01) and a 2-fold significant increase of p58 (\**p* < 0.05) in *mdx* compared to WT extracts (C,D). (E) To quantitate in 3D we used the software Imaris. We first calculated GE+ERES volumes and plotted them as a histogram showing proportion of each size to the total. (F) We also plotted Sec31 vs. GM130 staining intensity (in arbitrary units) measured for each GE, also in Imaris, and plotted one against the other. The number of GE included in calculations E and F was 1,052 for WT, 957 for *mdx*, and 1,013 for tubb6 OE, with *n* = 3 animals per group. Quantitation thus confirms the increase in size from WT to *mdx* and *mdx* to tubb6 OE; it also shows that the volume of GE and ERES stainings are correlated, regardless of genotype or condition. Bar: 10µm.

There is therefore a link between microtubules, GE, and SR positioning that extends from cortex to core of the fibers and is affected by the absence of dystrophin, possibly contributing to the weakness of the dystrophic muscle and to structural changes throughout the fiber (Friedrich et al., 2010).

## DISCUSSION

Until we observed the dynamics of WT microtubule and Golgi markers in live muscle fibers (Oddoux et al., 2013), we had no comprehension of their organization. Was this network stable or dynamic? How did GE happen to be positioned at the intersections of a microtubule grid? We answered several basic questions and, as is generally the case, came up with new questions. Here we start investigating the disorganization of the GE-microtubule network in the muscle of the mdx mouse in search of answers.

The main new question was: is it possible that dystrophin guides microtubules and anchors GE at the same time? The results presented here show that dystrophin is not necessary to anchor GE but that GE positioning is abnormal in the absence of dystrophin. Thus, dystrophin may play a role, possibly indirect, in positioning GE. To the best of our knowledge, there is no known direct interaction between GE and dystrophin although costameres contain numerous other proteins (Ervasti, 2003). We propose that GE are anchored by ERES, which are membrane domains of the ER. Muscle fibers have a core ER, the sarcoplamic reticulum (SR). The SR is compartmentalized into domains involved in calcium handling and muscle contraction and into domains involved in protein secretion. ERES are part of the latter (Kaisto and Metsikkö, 2003). Interestingly, a subsarcolemmal tubular system shaped like the ER was observed after labeling of skinned muscle fibers with fluorescent dyes (Jayasinghe et al., 2013). We do not know whether the periodic organization of the SR is perturbed in mdx fibers but electron microscopy of mdx muscle fibers reveals channels that contain SR and cross myofibrils at an abnormal angle of similar magnitude to that often seen in mdx microtubules (**Figure 9**).

We do not know why GE and ERES are increased in number and volume in mdx and tubb6 OE fibers. These changes suggest variations in protein secretion. Autophagy defects have been noticed in DMD and mdx mice and proposed as a target for DMD treatment (De Palma et al., 2012), so far without success. It is conceivable that Golgi membranes, which serve as source of membrane for autophagosomes (Geng and Klionsky, 2010) are upregulated in an attempt by muscle fibers to compensate for deficient autophagy.

Other organelles may be involved in muscle GE positioning. In WT muscle fibers, GE are often found in pairs (Percival et al., 2010). Interestingly, we found out that these pairs bookend or are bookended by lysosomes (Fukuda et al., 2006). Lysosomes are held by stable, detyrosinated microtubules (Mohan et al.,

FIGURE 9 | Link between microtubule orientation in the fiber cortex and sarcoplasmic reticulum (SR) orientation in the fiber core. The grid organization is typical of cortical microtubules and GE, likely because of their links to the costameres which only extend through the cortical layer. In the fiber core, longitudinal, and transverse microtubules are in different planes (not shown). However, the angular distribution of microtubules in the *mdx* fiber cortex mirrors that of the SR in the fiber core. On the left side we show the SR organization in electron micrographs of 5 mo-old *mdx extensor digitorum longus* (EDL) muscle. The myofibrillar organization is relatively normal (see thin dark band of Z-disks), despite interruptions by strands of longitudinal SR (best shown in C, white arrow) that often form V shapes (red dotted lines in A,B). Similar V shapes are found in autofluorescence images of *mdx* FDB fibers, but not of WT ones (D,E). They are also found in tubulin-stained *mdx* FDB fibers (F) that were also stained for calsequestrin, an SR marker (G). Bars: (A) 500 nm; (B) 1µm; (C) 500 nm; (D–G) 10µm.

2019) but this work concerns epithelial cells, not muscle. The role of microtubule stabilization in mdx and DMD pathologies is not clear. An increase in microtubule detyrosination has been proposed to make mdx microtubules more rigid (Kerr et al., 2015) but there is no consensus that there is a specific increase in detyrosination in mdx since its detyrosination level is increased proportionately to an increase in α-tubulin expression (Belanto et al., 2016; Loehr et al., 2018; Randazzo et al., 2019). We only find a minority of muscle microtubules to be detyrosinated (Randazzo et al., 2019 and **Figure 4A**) but a redistribution of stable microtubules that could affect various organelles cannot be ruled out.

GE pairs and rows are associated with microtubule bundles (**Figure 5**). This makes sense since muscle microtubules grow bidirectionally along pre-existing microtubules (Oddoux et al., 2013). It is likely that the more GE are positioned in a row, the higher the chance that their microtubules will interact and bundle, perhaps as a reinforcement of microtubule strength and for their protection against severing enzymes (Burkart and Dixit, 2019; Kuo et al., 2019).

It will be interesting to find out more about muscle GC nucleation; so far it has been studied mostly in non-muscle cells. Hot spots and periodic inactivation have been reported (Sanders and Kaverina, 2015) that could contribute to the stochastic character of nucleation from GE. Muscle GE are small however, barely the size of the GC hot spots. Nucleating proteins at non-muscle GC are the same as those involved in centrosomal nucleation (γ-tubulin and the γ-TuRC protein complexes) but with different scaffolding proteins to link them to the GC (Sanders et al., 2017). We have been unable to detect γ-tubulin at muscle GE (Oddoux et al., 2013) but γtubulin is even difficult to detect in myotubes because its level is low (Bugnard et al., 2005). The current consensus implicates AKAP450, CDK5Rap2, myomegalin, pericentrin, and centrosomin in the scaffolding of the nucleating machinery at the GC (Sanders and Kaverina, 2015). Pericentrin is the only one of these proteins that we identified in muscle GE; we could not detect AKAP450 (Oddoux et al., 2013). Finding good antibodies has been limiting. In contrast there has been recent progress in understanding microtubule nucleation from muscle nuclei, which can be approached in muscle cultures. Microtubules play a role in nuclear positioning and are important for muscle function (Metzger et al., 2012). The long, detyrosinated microtubules linking nuclei in mdx fibers (**Figure 4**) suggest that nucleiand GE-based muscle microtubules are distinctly regulated. Microtubule stabilization has been considered a source of pathology in mdx (Kerr et al., 2015) but may be a response to the loss of the microtubule grid that normally integrates the nuclei. The central scaffolding protein for nuclei-linked microtubules has been identified as nesprin-1 (Espigat-Georger et al., 2016; Gimpel et al., 2017).

Dystrophin is not the only muscle protein that has been implicated in microtubule guidance; some of these, such as CLIMP-63 (Osseni et al., 2016), a component of the triads, could direct core microtubules and complement dystrophin, CLIMP-63 working in the fiber core, dystrophin in the fiber cortex. Finally, a nNOS splice variant located in muscle GE has also been implicated in microtubule maintenance (Percival et al., 2010). Interestingly, GE themselves are abnormal in these muscle fibers, supporting the importance of Golgi integrity.

In summary, our work in live fibers of mdx and WT mouse muscle has revealed both common features and distinguishing ones that we consider essential. The common features are static GE nucleation and dynamic microtubules coursing along stable microtubules, while essential differences are mispositioning of GE and randomness of newly nucleated microtubules. We hope that the new model (**Figure 10**), derived from these experiments, will help ask further questions about this fascinating organization and understand it at a molecular and mechanistic level. Microtubules play important roles in in all mammalian cells. It is regrettable that we still understand so little of their organization in the largest of human cells, especially given the likely implication of microtubules in DMD, one of muscle's worst diseases.

#### MATERIALS AND METHODS

#### Antibodies

Several antibodies were generous gifts: rabbit anti-sec31/Vp137 from Fred Gorelick and Chris Shugrue (Yale University, New Haven, CT); rabbit anti-p58 ERGIC serum and monoclonal from Jaakko Saraste (University of Bergen, Norway); rabbit anti-GLUT4 (P-1; Ploug et al., 1990) from Thorkil Ploug (University of Copenhagen, Denmark) and rabbit anti-tubb6 (Randazzo et al., 2019) from Jim Ervasti (University of Minnesota-Twin Cities, Minneapolis, MN). Mouse anti-GM130 was purchased from BD Biosciences (Franklin Lake, NJ), rabbit anti-GM130,

anti-calsequestrin, and anti-α-tubulin from Abcam (Cambridge, MA), mouse anti-α-tubulin DM1A and rabbit anti-p58 from Sigma-Aldrich (St. Louis, MO), rat anti-α-tubulin from Novus Biologicals (Centennial, CO).

GE are static and anchored by ERES. Microtubule bundles form between

close GE by microtubule-microtubule interactions.

#### Plasmids

Tubulin and EB3 plasmids are based on pEGFP-N1 or pEGFP-C1 vectors (Takara Bio Inc., formerly Clontech). p-EB3-GFP-N1 cDNA was a gift from A. Akhmanova (Stepanova et al., 2003). The tubb6 plasmid was generated in the laboratory of Jim Ervasti at the University of Minnesota (Minneapolis, MN). It expresses mouse tubb6 and was built on a pDEST40 vector backbone with N-terminal or C-terminal EGFP tag as fully described in Randazzo et al. (2019).

#### cDNA Injection and Electroporation Into Mouse Muscles

All animal protocols were reviewed and approved by the Animal Care and Use Committee of the National Institute of Arthritis and Musculoskeletal and Skin Diseases. Mice were obtained from the Jackson Laboratory (Bar Harbor, ME) and housed and cared for by the personnel of the building 50 Animal Care Facility. Strains used were C57BL/6 or C57BL/10ScSnJ (both WT) and C57BL/10ScSn-DMDmdx /J (mdx), 6–8 week old unless otherwise mentioned. Since Duchenne Muscular Dystrophy affects males only, all animals were male. Anesthesia with 4% isoflurane and anti-pain injection of 0.05 mg buprenorphine-HCl as well as subsequent steps were as described in Oddoux et al. (2013) and Randazzo et al. (2019). Briefly, hyaluronidase (10 µl of 0.36 mg/ml) was injected through the skin of the heel and left for 1 h. Endotoxin-free plasmid (Genewiz, South Plainfield, NJ) was then injected at 5 mg/ml in sterile Dulbecco's PBS. Electroporation followed, with acupuncture needles as electrodes connected to an ECM 830 BTX electroporator (BTX Harvard Apparatus). Six pulses of 20 ms each at 1 Hz were applied to yield an electric field of 75 V/cm. After 6 days the animals were euthanized, and muscles collected.

#### Muscle Fiber Preparation

Briefly, FDB muscles were dissected in sterile PBS, rinsed in sterile DMEM, and incubated for 3 h at 37◦C in 1.5 mg/ml type I collagenase from Clostridium histolyticum (Sigma-Aldrich) and 1 mg/ml bovine serum albumin in DMEM. Fibers were triturated with 1 ml glass pipets and plated on 22 × 22 mm #1.5 glass coverslips that had been coated for 1 h with a 1:10 dilution of Matrigel (BD Bioscience). Two hours after plating in 0.1 ml of fusion media (DMEM supplemented with 4% horse serum) fibers were fed with the same media supplemented with penicillinstreptomycin (50 U/ml). They were fixed within 24 h to prevent denervation-related reorganization of microtubules and other cytoskeletal components (Ralston et al., 2001, 2006).

# Microtubule Recovery After Nocodazole Treatment

FDB fibers were treated with 4µg/ml nocodazole for 4 h at 37◦C, rinsed three times with fusion medium and left to recover at 37◦C for 2, 5, or 15 min. They were then fixed at room temperature with para-formaldehyde (Electron Microscopy Sciences, Atfield PA) diluted to 4% in PBS.

#### Fiber Immunofluorescence

Fixed FDB fibers were blocked for 2 h at room temperature in PBS containing 5% bovine serum albumin, 1% normal goat serum, and 0.04% saponin, except for staining with p58 antibodies which needed 0.1% Triton-X100 instead of saponin for permeabilization or −20◦C methanol for 20 min for fixation and permeabilization. Fibers were incubated with primary antibodies for 2 h at room temperature (or overnight at 4◦C) and with secondary antibodies for 2 h at room temperature, counterstained with Hoechst 33342, and mounted in Vectashield (Vector Laboratories, Burlingame, CA). All washes were 5– 15 min at room temperature with PBS containing 0.04% saponin or with PBS alone for samples permeabilized with Triton X-100.

#### Light Microscopy Imaging

Images from fixed fibers were acquired on a Leica TCS SP5 with LASAF software or on a Leica TCS SP8 X with LASX software and Lightning, as specified, with 63x N.A. 1.4 Plan Apo objectives. Images were 1,024 or 2,048 pixels wide; pinhole was set to 1.0 Airy Units and channels were scanned sequentially line by line when possible. For Lightning, images were acquired with the same objective, with a pinhole decreased to 0.7–0.5 AU depending on the degree of resolution improvement demanded. Image size was increased in width as set by the LASX. For Z-stacks, system optimized parameters were adopted i.e., optical sections were 0.3 µm-thick for pinhole 1.0, 0.17 µm-thick for pinhole 0.7 and 0.13 µm-thick for pinhole 0.5. 3D image rendition was obtained in the Leica LASX software.

2-Photon excited fluorescence images (2PEF) were obtained with excitation at 920 nm by a 3W MaiTai HP Ti:sapphire photon pulsed laser (Spectra-Physics, Santa Clara, CA) on the Leica SP8 X. Images were collected in a non-descanned detector with a bandpass filter 460/50 nm.

Live fibers were imaged at 0.65 ms/frame in phenol red–free HEPES-supplemented medium (20 mM). Temperature was kept at 37◦C, using a Tokai Hit heated stage insert and objective heater. The confocal pinhole was increased up to three Airy Units to increase depth of field and signal intensity while avoiding light damage to the live fibers.

All images were exported as.tif files with lossless compression and assembled into images in Photoshop CC2019 on iMacs working in OS 10.13.6 (High Sierra). Contrast and brightness were adjusted linearly if needed using Levels.

#### Image Analysis

As described in Oddoux et al. (2013) EB3-GFP speed was analyzed with PlusTipTracker and PlusTip GetTracks. GFPtubulin MT growth rate was analyzed with ImageJ (freely available at rsb.info.nih.gov/ij). Kymographs were generated on single or bundled microtubules in ImageJ or FiJi. A rectangle was drawn whose length represents the duration of an event and width the distance covered. Speed was obtained as distance divided by duration, pixels converted into time or length units. Pauses: only the time can be measured. All movies were recorded for 2 mins at 0.65 ms/frame.

Microtubule directionality was measured in the Matlab software TeDT (Liu and Ralston, 2014). The nuclear areas were masked as well as stable MTs remaining after 2 min of nocodazole washout. All the images collected were analyzed. The average directionality curves are calculated from the averages for each fiber, to avoid increasing the weight of better fibers for which more images were obtained than for poorer quality fibers that gave no more than 1 or 2 images suitable for analysis. Some fibers, poorly attached and in a bad shape, were excluded from the study.

For the nocodazole washout experiment, ImageJ was used for all image treatment. For microtubule density quantification, nuclear areas were cropped from the images. For early stage of nocodazole washout (2 min), long nocodazole-resistant microtubules were also cropped from the images to avoid a signal confusing microtubule dynamic recovery. All images were thresholded to make them binary and the number of pixels of microtubules on each image was measured. The total fiber area, minus nuclei, was measured in pixels. For Golgi density quantification, all 415 images were binarized and analyzed in bulk. The ImageJ function "analyze particle" was thresholded at five square pixels to avoid counting the background signal as GE. The size distribution of GE was plotted by binning GE surface by 0.1 µm<sup>2</sup> steps. The relative frequencies are in % of the total.

Statistical analyses were performed with Prism 8. T-tests with Welch's correction were used as most of the data compared did not have the same standard deviation.

3D analysis of confocal Z-stack acquisitions of FDB fibers was performed on Imaris software (version 9.3.1, Bitplane AG, Zurich, Switzerland). 3D reconstruction and surface modeling was used to measure the intensity of Sec31 and GM130 and calculate the volume of GE. Data were exported from Imaris as.xls files and processed on Prism 8 to generate graphs.

#### Immunoblots

Immunoblotting was performed as described in Randazzo et al. (2019). Briefly, FDB fibers were plated on Matrigelcoated coverslips. One day after plating, fibers were rinsed once in calcium-free PBS, collected with a rubber spatula and spun down. The pellet was resuspended in 1X sample buffer (National Diagnostic), boiled for 5 min and incubated for 30 min at r.t. Lysates from Gastrocnemius were obtained from liquid nitrogen flash-frozen muscles, grounded with mortar and pestle and the powder resuspended in 1X sample buffer (National Diagnostic). Extracts were then boiled for 5 min, incubated at room temperature for 30 min and centrifuged at 14,000 rpm for 10 min. Proteins were separated on 10% NOVEX NuPAGE MES (2-(N-morpholino) ethanesulfonic acid) Bis-Tris (Bis(2-hydroxyethyl)aminotris(hydroxymethyl)methane) Bis-Tris (Bis(2 hydroxyethyl)amino-tris(hydroxymethyl)methane)1.0 mm gels (Invitrogen) using 1X NuPAGE MES Sodium Dodecyl Sulfate running buffer. The loading amount of each extract was established by first running a gel with the same amount for all extracts and staining the gel with a colloidal blue staining kit (Invitrogen). A 600 dpi image of the gel was obtained with a scanner and the relative loading amount of each extract was calculated by analyzing the intensity of the corresponding lane in ImageJ. Based on the quantitation, a new gel was run under the same conditions. After the run, proteins were transferred to NOVEX nitrocellulose membrane (0.45µm) in 1X NOVEX NuPAGE transfer buffer (Invitrogen) with 10% methanol for 1 h at 100 V. Blocking of the membranes was performed for 1 h using 5% no-fat milk in Tris-buffered saline solution in presence of Tween-20 (TBS-T: 25 mM Tris, 136 mM NaCl, 3 mM KCl, 0.1% Tween-20, pH 7.4). Primary and horseradish peroxidase-conjugated secondary antibodies were diluted in blocking buffer and incubated for 1 h at room temperature. Washes were performed in TBS-T for three times, 10 min

each, after incubations with primary and secondary antibodies. Enhanced Chemiluminescence was performed with SuperSignal West Pico chemiluminescence or SuperSignal Femto Maximum Sensitivity substrates (Thermo Fisher Scientific, Rockford, IL) and peroxidase activity was detected using an XRS<sup>+</sup> Chemidoc (Bio-Rad, Hercules, CA). Different exposures were captured for each blot and protein bands quantitation was performed using the Image Lab Software v.5.2.1 (Bio-Rad) avoiding pixel saturation on the bands of interest.

#### Electron Microscopy

Muscles were fixed in 2% p-formaldehyde/2% glutaraldehyde in 0.1 M sodium cacodylate buffer for 2 h at room temperature and stored at 4◦C. Fixed samples were washed in buffer and treated with 1% osmium tetroxide in 0.1 M cacodylate buffer at pH 7.4 for 1 h on ice, washed and en bloc stained with 1% uranyl acetate in 0.1 N acetate buffer at pH 5.0 overnight at 4 ◦C, dehydrated with a series of graded ethanol, and embedded in epoxy resin. Sections 70–90 nm thick were counterstained with uranyl acetate and lead citrate, examined under a JEOL 1200EX transmission electron microscope, and photographed with a bottom-mounted digital CCD camera (AMT XR-100, Danvers, MA, USA). Electron microscopy was carried out at the EM Facility of the National Institute for Neurological Diseases and Stroke.

# DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the NIAMS Animal Care and Use Committee. The protocol was approved by the NIAMS Animal Care and Use Committee.

# AUTHOR CONTRIBUTIONS

This work was initiated by SO and KZ and continued and completed by DR, AK, and BA. Each of them performed, to different degrees, the experimental work, collected confocal images, and carried out image analysis. SO also drew the cartoon figures. ER coordinated and led the work, collected images, and drafted the manuscript. All co-authors read and contributed to the manuscript.

# FUNDING

This research was funded by the Intramural Research Program of the National Institute of Arthritis and Musculoskeletal and Skin Diseases.

# ACKNOWLEDGMENTS

We thank the several colleagues, mentioned in the Methods section, who gave us antibodies and plasmids and former laboratory members Victoria Tate and Shuktika Nandkeolyar who participated in the early stages of this work. We also thank Juraj Kabat (Biological Imaging Section, NIAID) for generous help with Imaris and Dan Sackett (NICHD) for critical reading of the manuscript.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2019. 00176/full#supplementary-material

#### REFERENCES


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

This work is authored by Oddoux, Randazzo, Kenea, Alonso, Zaal and Ralston on behalf of the U.S. Government and, as regards Dr. Oddoux, Dr. Randazzo, Dr. Kenea, Dr. Alonso, Dr. J Zaal, and Dr. Ralston and the U.S. Government, is not subject to copyright protection in the United States. Foreign and other copyrights may apply. 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.

# BML-265 and Tyrphostin AG1478 Disperse the Golgi Apparatus and Abolish Protein Transport in Human Cells

Gaelle Boncompain<sup>1</sup> , Nelly Gareil<sup>1</sup> , Sarah Tessier<sup>2</sup> , Aurianne Lescure<sup>2</sup> , Thouis R. Jones<sup>2</sup> , Oliver Kepp3,4, Guido Kroemer3,4,5,6,7, Elaine Del Nery<sup>2</sup> and Franck Perez<sup>1</sup> \*

#### Edited by:

Vladimir Lupashin, University of Arkansas for Medical Sciences, United States

#### Reviewed by:

Elizabeth Sztul, University of Alabama at Birmingham, United States Nobuhiro Nakamura, Kyoto Sangyo University, Japan Martin Lowe, The University of Manchester, United Kingdom

\*Correspondence:

Franck Perez franck.perez@curie.fr

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

Received: 11 July 2019 Accepted: 27 September 2019 Published: 11 October 2019

#### Citation:

Boncompain G, Gareil N, Tessier S, Lescure A, Jones TR, Kepp O, Kroemer G, Del Nery E and Perez F (2019) BML-265 and Tyrphostin AG1478 Disperse the Golgi Apparatus and Abolish Protein Transport in Human Cells. Front. Cell Dev. Biol. 7:232. doi: 10.3389/fcell.2019.00232 <sup>1</sup> Dynamics of Intracellular Organization Laboratory, Institut Curie, PSL Research University, Sorbonne Université, Centre National de la Recherche Scientifique, UMR 144, Paris, France, <sup>2</sup> BioPhenics High-Content Screening Laboratory, Cell and Tissue Imaging Facility (PICT-IBiSA), Institut Curie, PSL Research University, Translational Research Department, Paris, France, <sup>3</sup> Equipe Labellisée par la Ligue Contre le Cancer, Université de Paris, Sorbonne Université, INSERM U1138, Centre de Recherche des Cordeliers, Paris, France, <sup>4</sup> Metabolomics and Cell Biology Platforms, Gustave Roussy, Villejuif, France, <sup>5</sup> Suzhou Institute for Systems Medicine, Chinese Academy of Medical Sciences, Suzhou, China, <sup>6</sup> Pôle de Biologie, Hôpital Européen Georges Pompidou, AP-HP, Paris, France, <sup>7</sup> Department of Women's and Children's Health, Karolinska University Hospital, Karolinska Institute, Stockholm, Sweden

The steady-state localization of Golgi-resident glycosylation enzymes in the Golgi apparatus depends on a balance between anterograde and retrograde transport. Using the Retention Using Selective Hooks (RUSH) assay and high-content screening, we identified small molecules that perturb the localization of Mannosidase II (ManII) used as a model cargo for Golgi resident enzymes. In particular, we found that two compounds known as EGFR tyrosine kinase inhibitors, namely BML-265 and Tyrphostin AG1478 disrupt Golgi integrity and abolish secretory protein transport of diverse cargos, thus inducing brefeldin A-like effects. Interestingly, BML-265 and Tyrphostin AG1478 affect Golgi integrity and transport in human cells but not in rodent cells. The effects of BML-265 are reversible since Golgi integrity and protein transport are quickly restored upon washout of the compounds. BML-265 and Tyrphostin AG1478 do not lead to endosomal tubulation suggesting that, contrary to brefeldin A, they do not target the trans-Golgi ARF GEF BIG1 and BIG2. They quickly induce COPI dissociation from Golgi membranes suggesting that, in addition to EGFR kinase, the cis-Golgi ARF GEF GBF1 might also be a target of these molecules. Accordingly, overexpression of GBF1 prevents the effects of BML-265 and Tyrphostin AG1478 on Golgi integrity.

Keywords: golgi, membrane trafficking, high-content screening, EGFR kinase inhibitor, GBF1

# INTRODUCTION

The Golgi apparatus lies at the center of the secretory pathway of eukaryotic cells. This organelle receives neo-synthesized secretory proteins from the ER and sorts them to their destination compartments, such as the endosomes or the plasma membrane (Boncompain and Weigel, 2018). In the Golgi complex, cargos are often processed and modified through proteolysis and

glycosylation. Defects in the function and/or localization of Golgi associated proteins regulating trafficking and especially of glycosylation enzymes could lead to diverse diseases (Zappa et al., 2018). Golgi-resident glycosylation enzymes are type II transmembrane proteins transported from the ER to the Golgi in a COPII-dependent manner (Ward et al., 2001). The Golgi apparatus is a dynamic polarized organelle the composition of which is controlled by a balance of anterograde and retrograde fluxes of proteins. In particular, the steady-state intra-Golgi localization of Golgi-resident glycosylation enzymes is ensured by constant recycling from cisterna to cisterna up to the ER (Storrie et al., 1998). Retrograde transport of glycosylation enzymes occurs via COPI vesicles either by direct binding to coatomer (Liu et al., 2018) or through interaction with GOLPH3 (Tu et al., 2008; Ali et al., 2012). COPI coat recruitment is modulated by ADP-ribosylation factors (ARF), which cycle from an inactive state (GDP-bound) to an active state (GTP-bound) (Serafini et al., 1991; Donaldson et al., 1992a). ARFs are activated by guanine nucleotide exchange factors (GEF). At the cis-Golgi, the Golgi brefeldin A-resistant factor 1 (GBF1) plays a pivotal role in regulating organelle structure and vesicle trafficking by catalyzing the activation of ARF1 leading to COPI assembly and recruitment to Golgi membranes (Presley et al., 2002).

Because anterograde transport of cargos, especially of Golgiresident glycosylation enzymes, is constantly counterbalanced by retrograde transport (Cole et al., 1996; Storrie et al., 1998; Sengupta et al., 2015), it is difficult to distinguish between defects in the anterograde or the retrograde transport when a Golgi enzyme localization is perturbed. The development of the Retention Using Selective Hooks (RUSH) assay enables to overcome this difficulty through synchronization of fluxes, enabling quantitative analysis of anterograde transport of cargos (Boncompain et al., 2012). The RUSH assay uses physiological conditions for mammalian cells and is induced by simple addition of the non-toxic vitamin biotin. We have previously shown that the RUSH assay is amenable to high-content phenotypic screening (Boncompain et al., 2012, 2019; Liu et al., 2017; Zhao et al., 2018). To identify small molecules that regulate the trafficking of Golgi-resident glycosylation enzymes, we combined the RUSH assay to high content screening of chemical libraries using Mannosidase II (ManII) as a model Golgi enzyme. We identified several molecules either inhibiting or accelerating ManII transport from ER to Golgi. Supervised classification enabled the discovery of two small molecules displaying BFA-like activity on which we focused the present study. We further showed that these molecules, namely BML-265 and Tyrphostin AG1478, described as EGFR kinase inhibitors have additional intracellular biological effects. Both compounds induced reversible Golgi disruption and inhibition of the secretory transport in human cells, but not in rodent cells. The analysis of their effects on endosomal tubulation and dissociation of COPI from Golgi membranes suggested the cis-Golgi ARF GEF GBF1 to be the target of BML-265. This was further supported by overexpression of human GBF1, which prevents the effects of BML-265 and Tyrphostin AG1478.

# MATERIALS AND METHODS

# Screening Procedure

#### Cell Seeding

HeLa cells stably expressing Streptavidin-KDEL\_ManII-SBP-EGFP (Boncompain et al., 2012) were counted using T4 Cellometer (Nexcelom) and seeded in 384-well plates (ViewPlate-384 Black Perkin Elmer, catalog number 6007460) at 2,000 cells/well using a Multidrop Combi (Thermo Fisher Scientific) in 40 µl of cell media. The screen was performed at same early cell passages in two replicate experiments.

#### Compound Libraries

A FDA-approved drug library, comprised of 640 compounds diluted in Dimethyl Sulfoxide (DMSO) was purchased from Enzo Life Sciences (BML-2842) together with 80 known kinase inhibitors of well-defined activity (Screen-Well Kinase Inhibitor Library, catalog number BML-2832), 33 phosphatase inhibitors (Phosphatase Inhibitor library, catalog number BML-2834) and 53 protease inhibitors (Screen-WellTM Protease Inhibitor Library, catalog number BML-2833). 24 h after seeding, 10 µL of compounds was transferred to 384-well plates using the MultiChannel ArmTM 384 (MCA 384) (TECAN) to the cells, to a final concentration of 10 µM and 0.5% of DMSO. Nocodazole (2.5 µM final dilution) and brefeldin A (10 µM final dilution) were added as biologically relevant phenotypic controls in a single control plate. All chemical compounds were diluted in dimethyl sulfoxide (DMSO) as 10 mM stock solution and robotically reformatted in-house into 384-well source plates.

#### Compound Treatment and Biotin Pulse

After 90 min of incubation with the compounds at 37◦C and 5% CO2, cells were treated with 40 µM of biotin for 30 min at 37◦C. The screening was performed in three biological replicates. After biotin treatment, cells were processed for immunofluorescence performed as follows: cells were fixed with a fresh solution of 3% paraformaldehyde for 15 min using the MultiChannel ArmTM 384 (MCA 384) (TECAN). Cells were quenched with 50 mM NH4Cl in phosphate buffered saline (PBS) solution for 5 min, and then incubated for 1 h at room temperature (RT) with a primary mouse anti-GM130 antibody (1:500, BD Biosciences, catalog number 610822) diluted in 1% BSA-0.2% saponin. Cells were further washed with PBS and co-incubated for 1h RT with a secondary anti-mouse Alexa 647 antibody (1:400, Thermo Fisher Scientific, catalog number A-31571) and Hoechst 33242 (DNA) (1:500, Sigma, catalog number 14533).

For dose-response experiments, cells were treated 24 h after seeding with compounds (20 µL of media/DMSO solution) and vehicle DMSO 0.5%, titrated in a 8- point, three-fold dilution starting at a concentration of 10 µM. The assay plates were then incubated 24 h at 37◦C at 5% CO<sup>2</sup> and processed as described above.

#### Image Acquisition and Analysis

Image acquisition was performed using the INCell 2000 automated widefield system (GE Healthcare, United States) at a 20X magnification (Nikon 20X/0.45, Plan Apo, CFI/60), using the

same exposure time for all plates in the experiment and across replicate experiments. Plates were loaded onto the microscope system with a Kinedx robotic arm (PAA, United Kingdom). Images of four different positions in each well were acquired, each containing channels for Hoechst 33242, ManII-SBP-EGFP and GM130. The total number of cells measured in a well was typically around 300. Computational image processing operations were performed using the dual area object analysis in the INCell Analyzer Workstation 3.7 software (GE Healthcare). Nuclei were defined based on DNA staining and cell region was segmented using top-hat and collar segmentation, respectively. ManII-SBP-EGFP in the Golgi area was identified as inclusions in the collar cell area using multiscale top-hat segmentation, and quantified by the average intensity of pixels within the defined inclusion region. Individual cells were then gated to single phenotypes using CellProfiler Analyst software (Jones et al., 2008). Briefly, cell samples from all replicate experiments were sorted into four-defined classes (1. Golgi-disrupted, 2. ER + Golgi, 3. ER-retained and 4. Golgi) using DMSO, brefeldin A and nocodazole-treated wells as a training set for the classifier active learning module. Obtained data was normalized using the robust Z-score method under the assumption that most compounds are inactive and can serve as controls (Malo et al., 2006; Birmingham et al., 2009). Plate positional effects were corrected using median polishing (Mosteller and Tukey, 1977; Birmingham et al., 2009) applied to each phenotypic class. Hits for each compound were identified as follows: sample median and median absolute deviation (MAD) were calculated for each replicate from the population of screening data points (named as sample) and used to compute Robust Z-scores [RZ-scores (Iglewicz and Hoaglin, 1993)] according to the formula: RZscore = (Perturbator value-median (reference sample))/(1.4826 × MAD). A compound was identified as a 'hit', if the RZ-score was <−2 or >2.

#### Cells, Plasmids, Transfection

Hela wildtype and stably expressing Str-KDEL\_ManII-SBP-EGFP (Boncompain et al., 2012) cells, mouse embryonic fibroblasts (MEF) and normal rat kidney (NRK) cells were cultured in Dubelcco's modified Eagle medium (DMEM) (Thermo Fisher Scientific) supplemented with 10% Fetal Calf Serum (FCS, GE Healthcare), 1 mM sodium pyruvate and 100 µg/ml penicillin and streptomycin (Thermo Fisher Scientific).

RUSH plasmids were previously described TNF-SBP-EGFP, SBP-EGP-GPI (Boncompain et al., 2012), SBP-EGFP-EGFR (Scharaw et al., 2016).

The plasmid coding for Venus-GBF1 was a kind gift from C. Jackson (Institut Jacques Monod, Paris, France).

HeLa cells were transfected using calcium phosphate as described previously (Jordan et al., 1996).

#### Chemicals

BML-265 was purchased from Enzo Life Sciences. Tyrphostin AG1478 and Erlotinib were ordered from Cayman chemical. DMSO, brefeldin A, Golgicide A and D-biotin were purchased Sigma-Aldrich.

#### Antibodies and Immunofluorescence

Monoclonal mouse anti-GM130 was purchased from BD Biosciences (catalog number 610823, used at 1:1000) and rabbit polyclonal anti-GM130 from Abcam (catalog number ab52649, dilution 1:2000). Anti-GFP was purchased from Merck (catalog number 11814460001, use at 1:1000). Anti giantin (TA10) was obtained from the recombinant antibody platform of the Institut Curie (dilution 1:100). Anti betaCOP (mAD) and anti-transferrin receptor (OKT9) were a gift from T. Kreis, University of Geneva, dilution 1:200 and 1:1000 respectively. The anti-betaCOP antibody requires cell fixation using methanol. The antibody directed to GalT (B4GalT1) used in **Supplementary Figure S2** was purchased from Abnova (catalog number PAB20512, dilution 1:1000) and the one used in **Supplementary Figure S3** was obtained from CellMAb (catalog number CB02, dilution 1:100). The anti-TGN46 was purchased from BioRad (catalog number AHP500G, dilution 1:2000). The antibody directed to Sec24C was a kind gift from David Stephens (University of Bristol, United Kingdom) (dilution 1:500). The anti-EEA1 antibody was purchased from BD biosciences (catalog number 610456, dilution 1:2000). The anti-LAMP1 was purchased from Merck (catalog number L1418, dilution 1:3000).

Alexa488, Cy3 and Cy5 conjugated secondary antibodies were purchased from Jackson Immunoresearch and used at 1:400.

Immunostaining of intracellular targets was performed after fixation with paraformaldehyde 3% for 15 min at room temperature. After 3 washes with PBS, cells were permeabilized with PBS supplemented with BSA 2 g/l and saponin 0.5 g/l for 5 min at room temperature. Antibodies were diluted in PBS supplemented with BSA 2 g/l and saponin 0.5 g/l and incubated with cells for 45 min. Coverslips were mounted in Mowiol containing DAPI.

Immunostaining to monitor the presence of EGFP-tagged cargos at the cell surface was performed on living cells seeded on glass coverslips. Cells were washed with ice cold PBS and incubated on ice with the anti-GFP antibody diluted in PBS for 45 min. Cells were then washed 3 times with ice cold PBS and fixed with paraformaldehyde 2% for 15 min at room temperature. After 3 washes with PBS, the cells were incubated with the secondary antibodies for 40 min at room temp.

After 3 washes with PBS, coverslips were in Mowiol. Observation and acquisition of pictures of fixed samples were performed using an epifluorescence microscope (Leica) equipped with a Coolsnap camera (Roper Scientific) using the software Metamorph (Molecular Devices).

#### EGFR Phosphorylation and Immunoblot

Cells were serum starved overnight. Cells were then pre-treated with the indicated molecules at 10 µM for 1h. They were then stimulated with human EGF at 50 ng/ml final for 10 min.

Cells were lysed in Laemmli 2,5 × buffer containing betamercaptoethanol. Samples were then loaded on acrylamide gels (Criterion TGX, BioRad) and transferred on nitrocellulose membrane (GE Healthcare) using Power blotter Semi dry from Thermo Scientific. Blocking and incubation of antibodies were performed in PBS supplemented with 0.1% Tween and 5% milk.

Detection was performed using SuperSignal West Pico Substrate from Thermo and Chemidoc machine (BioRad).

Human EGF was purchased from Sigma-Aldrich (catalog number E9644).

Antibodies to detect phosphorylated EGFR and total EGFR were purchased from Cell Signaling Technology (PhosphoPlus EGFR (Tyr1068) Antibody Duet, catalog number 11862S, dilution 1:1000). Anti-actin (clone AC40) used as loading control was purchased from Sigma Aldrich (catalog number A3853, dilution 1:1000). Secondary anti-mouse and anti-rabbit poly-HRP antibodies were purchased from Thermo Scientific (dilution 1:10000).

#### Real Time Imaging

HeLa cells stably expressing Streptavidin-KDEL\_ManII-SBP-EGFP were grown on 25 mm glass coverslips. Cells were maintained in presence of biotin at 40 µM to allow stable localization of ManII-SBP-EGFP in the Golgi apparatus. Coverslips were transferred to an L-shape tubing equipped Chamlide chamber (Live Cell Instrument). Pre-warmed Leibovitz medium (Life Technologies) supplemented with 40 µM of biotin was used. BML-265 diluted at 10 µM in Leibovitz supplemented with biotin was added at time 0. After the indicated time, BML-265 was washed thanks to several washes with pre-warmed Leibovitz. For the recovery period, medium was replaced by pre-warmed Leibovitz containing biotin. Imaging was performed at 37◦C in a thermostat controlled chamber using an Eclipse 80i microscope (Nikon) equipped with a spinning disk confocal head (Perkin) and an Ultra897 iXon camera (Andor). Image acquisition was performed using MetaMorph software (Molecular Devices). Maximum intensity projections of several Z-slices are shown.

# RESULTS

### High-Content Phenotypic Screen Led to Identification of Molecules Regulating the Trafficking of ManII

Using the previously described HeLa cell line stably expressing Streptavidin-KDEL\_ManII-SBP-EGFP (Boncompain et al., 2012), we designed a phenotypic screen to identify small molecules modulating the trafficking of the Golgi enzyme Mannosidase II (ManII) (**Figure 1A**). The cells co-express a non-fluorescent ER hook (Str-KDEL) with ManII fused to a Streptavidin Binding Peptide (SBP) and an EGFP (ManII-SBP-EGFP) enabling RUSH control of ManII trafficking. ManII-SBP-EGFP is retained in the ER in the absence of biotin and is transported to the Golgi apparatus after incubation with biotin (**Figure 1B**). This set-up was used to screen small molecules from a library of FDA-approved molecules and inhibitors of kinases, phosphatases and proteases (see Materials and Methods section for details). Cells seeded in 384-well plates were incubated with small molecules, diluted at a final concentration of 10 µM in DMSO, for 90 min. Biotin was then added for 30 min to induce the transport of ManII-SBP-EGFP from ER to Golgi. Cells were then fixed and nuclei were stained using DAPI. The Golgi disrupting agent, brefeldin A (BFA) and the microtubule-depolymerizing drug, nocodazole (Noco) were used as positive phenotypic controls for ManII ER-retention and Golgi disruption, respectively. In addition, wells not incubated with biotin were included in order to prevent ManII-SBP-EGFP exit from the ER. 0.5% of DMSO was used as solvent control. After image acquisition and segmentation, several features were measured (see Materials and Methods section for details). Samples were classified into four phenotypical classes (1. ERretained, 2. ER + Golgi, 3. Golgi disrupted and 4. Golgi) using above-mentioned controls as training set using Cell Profiler Analyst (**Figure 1C**). A cell count was also used to evaluate toxic effects of the compounds.

Robust z-score for each phenotypic class was calculated (see Materials and Methods) allowing identification of outliers. For instance, the ER-retained robust Z-score of the condition "DMSO" is very different (20.46) from the one of the condition "DMSO + biotin" (−2.19) (**Figure 1D**). "DMSO + biotin" corresponds to the control condition, for which ER to Golgi transport of ManII occurred normally while "DMSO" (i.e., absence of biotin) is exemplifying the robust Z-score obtained by a hit preventing normal transport and inducing ER-retention. Please note that the score for Golgi-disruption of cells incubated with nocodazole and biotin ('Noco + biotin') is higher than the one of cells treated with nocodazole only ('Noco') because our analysis uses ManII-SBP-EGFP signal to detect Golgi elements and not an independent Golgi marker. In consequence, the accumulation of ManII-SBP-EGFP in Golgi mini-stacks due to biotin addition leads to a better detection of Golgidisruption (**Figure 1D**).

As we were primarily looking for molecules inhibiting ER to Golgi transport of ManII, the results were sorted using ER-retained score (**Figures 1D,E**). The two top hit molecules identified as inhibitors of ManII ER to Golgi transport were BML-265 and Tyrphostin AG1478. Their score for the 4 phenotypic classes was similar to the controls DMSO [No biotin], BFA and BFA + biotin confirming that they inhibit ManII transport to the Golgi (**Figures 1D,E**).

We then performed a dose-response analysis, in triplicates, using serial dilutions (10 doses from 30 µM to 1.52 nM) of 50 selected molecules either inhibiting or accelerating ER to Golgi transport. The same quantitative analysis that was done in the primary screen was carried out here. Four families of compounds were identified based on the dose-response profile of 'ER-retained' and 'Golgi disrupted' scores (**Supplementary Figure S1**). Some molecules inhibited ER to Golgi transport at increasing doses without affecting the integrity of the Golgi apparatus. These molecules are suspected to affect cell homeostasis or to be energy poisons as this family includes ouabain (Na+, K<sup>+</sup> ATPase inhibitor). Based on the doseresponse profiles, we identified two groups of molecules affecting microtubules. A first group contained depolymerizing agents such as albendazole, while a second one was composed of destabilizing agents such as vinca alkaloides (e.g., vinblastine). The dose-response profiles of 'ER retained' and 'Golgi disrupted' scores showed that BML-265 and Tyrphostin AG1478 displayed

FIGURE 1 | High-content screening for molecules regulating ER to Golgi transport of ManII. (A) Scheme of the screening processes. Cells stably expressing Str-KDEL\_ManII-SBP-EGFP were incubated with each small molecule for 90 min. Synchronized transport of ManII was induced by addition of biotin. After 30 min, cells were then fixed and nuclei were stained with DAPI. (B) Pictures from screening plates depicting the controls DMSO (no traffic) and DMSO + biotin (transport to the Golgi). Scale bar: 10 µm. (C) Scheme of the four classes used for the classification of quantitative analysis of the pictures. (D,E) Scores obtained for the parameters 'ER-retained,' 'ER + Golgi,' 'Golgi,' 'Golgi-disrupted,' and cell count for each control (D) or small molecule compound (E).

similar effects (BFA-like). They inhibited the trafficking of ManII to the Golgi starting from 123 – 370 nM and induced Golgi disruption on the same concentration range. Note that, surprisingly, no Golgi disruption was detected at high doses of BML-265 and Tyrphostin AG1478. This is due to the way we quantified Golgi organization. Indeed, at high doses, the Golgi apparatus was completely disassembled and no visible structures remained. In consequence, our analysis for Golgi disruption did not detect any Golgi structures and scored it as "no Golgi disruption" (**Supplementary Figure S1**).

#### The Effects of BML-265 and Tyrphostin AG1478 on Golgi Integrity and Trafficking Indicate BFA-Like Activity

The above-mentioned effects of BML-265 and Tyrphostin AG1478 were reminiscent of the ones obtained with BFA used as a control in our experiments. In addition, whereas BML-265 and Tyrphostin AG1478 are both annotated as EGFR kinase inhibitors, Tyrphostin AG1478 was described to target the cis-Golgi ADP ribosylation factor guanine nucleotide exchange factor (ARF GEF) named GBF1 (Pan et al., 2008). We thus decided to compare the efficacy of BML-265, Tyrphostin AG1478 and BFA on the trafficking of ManII to the Golgi apparatus. HeLa stably expressing Str-KDEL\_ManII-SBP-EGFP were incubated with serial dilutions of the molecules (from 0.019 nM to 30 µM) for 90 min and biotin was then added to induce the transport of ManII-SBP-EGFP. Our results show that BFA is more potent than BML-265 which is itself more potent than Tyrphostin AG1478 (**Figure 2A**). Whereas IC50 of BFA is about 2 nM, IC50 of BML-265 is about 200 nM and of Tyrphostin AG1478 about 1 µM. Strikingly, the chemical structures of BML-265 and Tyrphostin AG1478 are closely related while they are different from the one of BFA (**Figure 2B**).

We then assessed the effects of BML-265 and Tyrphostin AG1478 on the trafficking of secretory cargos addressed to the cell surface. For this purpose, we used TNF (type II transmembrane protein), GPI (GPI anchor) and EGFR (type I transmembrane protein) as RUSH cargos (Boncompain et al., 2012; Fourriere et al., 2016; Scharaw et al., 2016). In these fusion constructs, the EGFP is exposed to the extracellular space when the cargo reaches the plasma membrane. The effects of the molecules on the anterograde transport of TNF, EGFP-GPI and EGFR were thus assessed by immunofluorescence on non-permeabilized cells using an anti-GFP antibody. The Golgi apparatus was stained using an anti-GM130 antibody. As observed with BFA, pretreatment of the cells with BML-265 and Tyrphostin AG1478 for 60 min prior to addition of biotin prevented transport of the cargos to the cell surface (**Figure 3**). The EGFP-tagged cargos are detected inside the cells, probably in the ER. Consistently with the results obtained during the screening and IC50 experiments, the Golgi apparatus is disrupted upon incubation with BML-265 and Tyrphostin AG1478.

Altogether, our results indicate that BML-265 and Tyrphostin AG1478 display BFA-like effects on trafficking and Golgi integrity.

# BML-265 Disperses the Golgi Apparatus in Human Cells but Not in Rodent Cells

Tyrphostin AG1478 was previously reported to affect Golgi integrity in human but not in rodent cells. We thus analyzed the effects of BML-265 on the integrity of the Golgi complex by immunolabeling using an anti-GM130 antibody using a human cell line and two rodent cell lines. As observed with BFA, incubation of the human epithelial cell line HeLa with BML-265 and Tyrphostin AG1478 led to redistribution of GM130, indicating Golgi disruption (**Figure 4A**). Similar results were obtained when staining for other Golgi markers, namely TGN46, GalT and giantin. As expected, the intracellular distribution of ER exit sites, stained with an anti-Sec24 antibody, was also affected. In contrast, early endosomes and late endosomes/lysosomes organization was not perturbed after incubation with the molecules compared to DMSO control (**Supplementary Figure S2**). However, these molecules did not affect Golgi localization and morphology of mouse fibroblasts (MEF) and rat epithelial cells (NRK). In contrast, BFA disrupted the Golgi complex of these two rodent cell models (**Figures 4B,C**).

BML-265 and Tyrphostin AG1478 are both annotated in the library as analogs of Erlotinib (Tarceva trade mark; OSI Pharmaceuticals, Genentech and Roche), one of the several EGFR tyrosine kinase inhibitors, which has been largely studied in clinical trials, with proven efficacy in humans. BML-265, Tyrphostin AG1478 and Erlotinib have different structures but share a quinazoline group (**Figure 2B** and **Supplementary Figure S3**). We verified that BML-265 was indeed able to prevent EGFR phosphorylation upon stimulation with EGF. Importantly, Erlotinib did not induce Golgi disruption even at high doses (**Supplementary Figure S3**). These results suggest that effects of BML-265 and Tyrphostin AG1478 on Golgi integrity and function are independent of their capability to inhibit EGFR phosphorylation.

## BML-265 Has Reversible Effects on Golgi Integrity

Treatment of HeLa cells for 1h with BML-265 or Tyrphostin AG1478 leads to a complete redistribution of Golgi complex proteins throughout the cytoplasm (**Figure 5A**). However, the effects of both compounds are partially reversible. 1 h after addition of the molecules, the medium was removed and cells were washed before being incubated in normal medium. Cells were then fixed and the Golgi complex immunolabeled using an anti-GM130 antibody. Forty-five min after washout, cells recovered a normal Golgi organization and localization (**Figure 5A**). We next assessed by real-time imaging the reformation of the Golgi complex after washout of BML-265. BML-265 was added to HeLa stably expressing ManII-SBP-EGFP. Quickly after addition of BML-265, transient ManII-SBP-EGFP positive tubules were observed and ManII-SBP-EGFP signal intensity at the Golgi decreases while increasing in the whole cell corresponding to ER relocation of ManII-SBP-EGFP (**Figure 5B** and **Supplementary Movie S1**). Washout of BML-265 was then performed and cells imaged in real-time. The

Golgi complex visualized by ManII-SBP-EGFP redistributed to the perinuclear area. Signal intensity of ManII-SBP-EGFP at the Golgi increased while ER signal decreased showing that ER to Golgi transport was restored.

#### BML-265 Might Target GBF1

BFA inhibits Arf1 activation (Donaldson et al., 1992b; Helms and Rothman, 1992) by targeting several ARF GEF: GBF1 at the cis-Golgi and BIG1 and BIG2 at the TGN (Claude et al., 1999; Yamaji et al., 2000). Due to its effects on the TGN ARF GEF, BFA induces tubule formation at the TGN and on endosomes (Lippincott-Schwartz et al., 1991; Wood et al., 1991) as confirmed by immunostaining using an anti-Transferrin receptor (TfR) antibody. In contrast, BML-265 and Tyrphostin AG1478 did not induce tubulation of TfR-positive compartments, suggesting that they do not affect the TGN ARF GEF (**Figure 6A**). GBF1 is an ARF GEF present at the cis-Golgi involved in the recruitment of the COPI coat on Golgi membranes. We next assessed the distribution of COPI after treatment with BML-265. In HeLa cells treated with the molecules for 5 min, the distribution of the COPI coat was monitored using immunolabeling with an antibetaCOP antibody and the Golgi complex was detected using an anti-Giantin antibody. Whereas in non-treated cells, betaCOP is found in the Golgi area, probably associated to the Golgi complex and to vesicles, betaCOP is absent from the Golgi complex in BML-265 treated cells even in cells still displaying a perinuclear Golgi complex after this short treatment time (**Figure 6B**). As expected, BFA also induced rapid dissociation of COPI coat from Golgi membranes. We also detected dissociation of betaCOP from Golgi membranes for two molecules known as inhibitors of GBF1, Tyrphostin AG1478 and Golgicide A (GCA) (Pan et al., 2008; Saenz et al., 2009) (**Figure 6B**). Overexpression of human wild-type GBF1 prevented Golgi dispersal induced by incubation with BML-265 or Tyrphostin AG1478 (**Figure 6C**).

Altogether these results suggest that BML-265 targets GBF1 causing Golgi dispersal and inhibiting secretory protein trafficking.

#### DISCUSSION

In the past years, several image-based and high-throughput screens led to the identification of small molecules which affect membrane trafficking (Mishev et al., 2013). Among those, AMF-26/M-COPA, Golgicide A, Exo2, LG-186 and Tyrphostin AG1478 were identified as molecules displaying effects similar

to BFA, but being more specific because they specifically target the ARF GEF GBF1 (Pan et al., 2008; Spooner et al., 2008; Saenz et al., 2009; Boal et al., 2010; Ohashi et al., 2012). In the present study, we initially intended to search for compounds that regulate the ER to Golgi transport of the Golgi-resident glycosylation enzyme ManII. Our study using a compound library composed of FDA-approved molecules as well as inhibitors of proteases, phosphatases and kinases revealed several compounds able to modulate ManII trafficking, either by inhibiting or accelerating ER to Golgi transport. Further studies will be necessary to clarify the biological activity of these molecules on ER to Golgi transport and more largely on secretory protein trafficking. We

FIGURE 4 | BML-265 affects Golgi integrity in human cells but not in rodent cells. HeLa cells (A), mouse embryonic fibroblasts (MEF) (B) or normal rat kidney (NRK) cells (C) were incubated with the indicated molecules at 10 µM final for 1h30. Cells were then fixed and the Golgi apparatus was stained using an anti-GM130 antibody (green on merge). Nuclei were stained using DAPI (blue on merge). Scale bar: 10 µm.

focused our attention on the previously studied Tyrphostin AG1478 as well as BML-265, which both prevent ManII transport to the Golgi, induce Golgi disassembly and prevent secretory protein transport. BML-265 and Tyrphostin AG1478 display a very close structure and are both annotated as Erlotinib analogs, being inhibitors of EGFR kinase activity. They share a

quinazoline moiety with Erlotinib and other described inhibitors of receptor kinases (Fry et al., 1994; Ward et al., 1994; Stamos et al., 2002). Erlotinib as well as Gefitinib, Lapatinib and several Tyrphostins were present in our library but were not scored as molecules inhibiting the transport of ManII and/or affecting Golgi integrity. The similarity between BML-265 and Tyrphostin AG1478 in terms of structure and effects, and the absence of effects of other Erlotinib analogs, suggest that they both act on Golgi function independently of their ability to inhibit EGFR phosphorylation. Interestingly, other molecules known to target GBF1, such as Golgicide A, do not bear a quinazoline moiety (Saenz et al., 2009) and were not reported to inhibit EGFR phosphorylation.

Our results for BML-265 and Tyrphostin AG1478 as well as previous published work on Tyrphostin AG1478 (Pan et al., 2008) show that these small molecules induce Golgi disruption in human cells, but not in rodent cells. The differential effects of BML-265 and Tyrphostin AG1478 on human versus rodent cells are interesting in the point of view of their clinical use. EGFR is activated due to mutation and/or overexpression in diverse epithelial tumors and is associated with poor prognosis (Nicholson et al., 2001). Consequently, inactivation of EGFR signaling pathway is a target for cancer treatment. Erlotinib is given to patients sometimes in combination with cetuximab for the treatments of some cancers. Considering that mouse cells are sensitive only to the EGFR kinase inhibition activity of BML-265 and Tyrphostin AG1478, in vivo tests in mouse models would underestimate their toxic effects and fail to early capture clinically significant secondary effects that might arise in humans. The suggested target for these compounds is the cis-Golgi ARF GEF GBF1 since they exert BFA-like effects, but do not induce endosome tubulation. Nucleotide exchange activity of GBF1 is mediated by its catalytic Sec7 domain (Renault et al., 2003). The Sec7 domains of human and mouse GBF1 display 98% of similarity in their amino acid sequence. Even though the catalytic domain of GBF1 is highly conserved in human and mouse cells, this difference might be sufficient to modify the putative binding sites of BML-265 and Tyrphostin AG1478. In the present study, we showed that overexpression of human GBF1 prevents the effects of BML-265 on Golgi integrity. This result strongly suggests that GBF1 might be a target of BML-265. However, we cannot exclude indirect effects of the molecules on GBF1 neither the existence of other cellular upstream targets. The mode of interaction of these molecules with human GBF1 as well as the mechanisms of inactivation of GBF1 require further investigation.

FIGURE 6 | BML-265 exerts its effects on Golgi integrity and trafficking through targeting GBF1. (A) HeLa cells were incubated with the indicated molecules at 10 µM for 1 h. After fixation, endosomes were stained using an anti-transferrin receptor (TfR) antibody. Scale bar: 10 µm. The inset displayed an enlarged view of the boxed area. (B) HeLa cells were incubated with the indicated molecules at 10 µM for 5 min and were then fixed using methanol. COPI coat was stained using an anti-betaCOP antibody, the Golgi apparatus using anti-giantin antibody and nuclei using DAPI. Scale bar: 10 µm. (C) HeLa cells transiently expressing human GBF1 tagged with Venus (Venus-hGBF1) or cytoplasmic EGFP as a transfection control were incubated with the indicated molecules at 10 µM for 1 h. After fixation, the Golgi apparatus was immunolabeled using an anti-giantin antibody. Scale bar: 10 µm.

# DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the manuscript/**Supplementary Files**.

# AUTHOR CONTRIBUTIONS

GB carried out the experiments and analyzed the data. NG carried out the experiments. ST and AL carried out high content screens. TJ and ED analyzed the high-content screening data. ED supervised the high content-screening. OK and GK helped in the analysis of the high content screening data. GB, FP, and ED wrote the manuscript. GB and FP designed the study.

# FUNDING

Work performed in the F. Perez laboratory was funded by Centre Nationale de la Recherche Scientifique, the Fondation pour la Recherche Medicale (FRM DEQ20120323723), the Labex CellTisPhyBio, and the Agence Nationale de la Recherche (ANR-12-BSV2-0003-01). The lab of F. Perez is part of Labex CelTisPhyBio (11-LBX-0038) and Idex Paris Sciences et Lettres (ANR-10-IDEX-0001-02 PSL). GK is supported by the Ligue contre le Cancer (équipe labellisée); Agence Nationale de la Recherche (ANR) – Projets blancs; ANR under the frame of E-Rare-2, the ERA-Net for Research on Rare Diseases; Association pour la recherche sur le cancer (ARC); Cancéropôle Ile-de-France; Chancellerie des Universités de Paris

(Legs Poix), Fondation pour la Recherche Médicale (FRM); a donation by Elior; European Research Area Network on Cardiovascular Diseases (ERA-CVD, MINOTAUR); Gustave Roussy Odyssea, the European Union Horizon 2020 Project Oncobiome; Fondation Carrefour; High-end Foreign Expert Program in China (GDW20171100085 and GDW20181100051), Institut National du Cancer (INCa); Inserm (HTE); Institut Universitaire de France; LeDucq Foundation; the LabEx Immuno-Oncology; the RHU Torino Lumière; the Seerave Foundation; the SIRIC Stratified Oncology Cell DNA Repair and Tumor Immune Elimination (SOCRATE); and the SIRIC Cancer Research and Personalized Medicine (CARPEM).

#### ACKNOWLEDGMENTS

The authors acknowledge the Cell and Tissue Imaging Facility (PICT-IBiSA), Institut Curie, a member of the French National Research Infrastructure, and France-BioImaging (ANR10-INBS-04). The authors also thank the recombinant antibody platform of the Institut Curie.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2019.00232/ full#supplementary-material

# REFERENCES


FIGURE S1 | A dose-response analysis of 50 molecules led to the identification of 4 classes of molecules. HeLa cells stably expressing Str-KDEL\_ManII-SBP-EGFP were incubated with serial dilutions of 50 selected molecules. Trafficking was induced by addition of biotin and phenotypic analysis was carried out using the approach developed for the screen. 'ER-retained' (A) and 'Golgi-disrupted' (B) scores were calculated and displayed on the graphs. Three independent replicates were performed and reported here (colored curves).

FIGURE S2 | Effects of BML-265, AG1478 on different markers of intracellular compartments. (A) HeLa cells were incubated with the indicated molecules at 10 µM final for 1 h. After fixation, immunostaining using anti-TGN46 (left), anti-Sec24 (middle) or anti-GalT (right) antibodies was performed. Scale bar: 10 µm. (B) HeLa cells were incubated with the indicated molecules at 10 µM final for 1 h. Immunolabeling of early endosomes using an anti-EEA1 antibody, of the Golgi apparatus using an anti-Giantin antibody and of late/endosomes/lysosomes using an anti-LAMP1 antibody was performed. Scale bar: 10 µm.

FIGURE S3 | BML-265 behaves as Erlotinib analog and inhibits EGFR phosphorylation but Erlotinib does not affect Golgi integrity. (A) HeLa cells were serum starved overnight and were treated with the indicated molecules at 10 µM for 90 min. Cells were then incubated with EGF at 50 ng/ml for 10 min. Phosphorylated EGFR (P-EGFR), total EGFR and actin (used as a loading control) were detected by immunoblot. (B) Scheme of the Erlotinib molecule. (C) HeLa cells were incubated with Erlotinib for 90 min at the indicated concentrations. The Golgi apparatus was stained using 3 different antibodies: anti-GalT (green), anti-GM130 (rouge), and anti-giantin (bleu). Scale bar: 10 µm.

MOVIE S1 | The effects of BML-265 are reversible. HeLa cells stably expressing Str-KDEL\_ManII-SBP-EGFP maintained in presence of biotin 40 µM to allow stable Golgi localization of ManII-SBP-EGFP. BML-265 at 10 µM was added at time 0. Time is indicated in min:sec. After 50 min, BML-265 was washed. Pictures were acquired every 30 s. Maximum projection of 11 z-slices is shown.

coatomer protein beta-COP to Golgi membranes. Proc. Natl. Acad. Sci. U.S.A. 89, 6408–6412. doi: 10.1073/pnas.89.14.6408



**Conflict of Interest:** OK and GK are cofounders of Samsara Therapeutics.

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 Boncompain, Gareil, Tessier, Lescure, Jones, Kepp, Kroemer, Del Nery and Perez. 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.

# Protein Amphipathic Helix Insertion: A Mechanism to Induce Membrane Fission

Mikhail A. Zhukovsky\*, Angela Filograna, Alberto Luini, Daniela Corda\* and Carmen Valente\*

Institute of Biochemistry and Cell Biology, National Research Council, Naples, Italy

#### Edited by:

Vladimir Lupashin, University of Arkansas for Medical Sciences, United States

#### Reviewed by:

Bruno Goud, Centre National de la Recherche Scientifique (CNRS), France Patricia Bassereau, Institut Curie, France

#### \*Correspondence:

Mikhail A. Zhukovsky mikhail.zhukovsky@ibbc.cnr.it Daniela Corda daniela.corda@cnr.it Carmen Valente carmen.valente@ibbc.cnr.it

#### Specialty section:

This article was submitted to Membrane Traffic, a section of the journal Frontiers in Cell and Developmental Biology

Received: 16 May 2019 Accepted: 06 November 2019 Published: 10 December 2019

#### Citation:

Zhukovsky MA, Filograna A, Luini A, Corda D and Valente C (2019) Protein Amphipathic Helix Insertion: A Mechanism to Induce Membrane Fission. Front. Cell Dev. Biol. 7:291. doi: 10.3389/fcell.2019.00291 One of the fundamental features of biomembranes is the ability to fuse or to separate. These processes called respectively membrane fusion and fission are central in the homeostasis of events such as those related to intracellular membrane traffic. Proteins that contain amphipathic helices (AHs) were suggested to mediate membrane fission via shallow insertion of these helices into the lipid bilayer. Here we analyze the AH-containing proteins that have been identified as essential for membrane fission and categorize them in few subfamilies, including small GTPases, Atg proteins, and proteins containing either the ENTH/ANTH- or the BAR-domain. AH-containing fission-inducing proteins may require cofactors such as additional proteins (e.g., lipid-modifying enzymes), or lipids (e.g., phosphatidylinositol 4,5-bisphosphate [PtdIns(4,5)P2], phosphatidic acid [PA], or cardiolipin). Both PA and cardiolipin possess a cone shape and a negative charge (−2) that favor the recruitment of the AHs of fission-inducing proteins. Instead, PtdIns(4,5)P<sup>2</sup> is characterized by an high negative charge able to recruit basic residues of the AHs of fission-inducing proteins. Here we propose that the AHs of fission-inducing proteins contain sequence motifs that bind lipid cofactors; accordingly (K/R/H)(K/R/H)xx(K/R/H) is a PtdIns(4,5)P2-binding motif, (K/R)x6(F/Y) is a cardiolipin-binding motif, whereas KxK is a PA-binding motif. Following our analysis, we show that the AHs of many fission-inducing proteins possess five properties: (a) at least three basic residues on the hydrophilic side, (b) ability to oligomerize, (c) optimal (shallow) depth of insertion into the membrane, (d) positive cooperativity in membrane curvature generation, and (e) specific interaction with one of the lipids mentioned above. These lipid cofactors favor correct conformation, oligomeric state and optimal insertion depth. The most abundant lipid in a given organelle possessing high negative charge (more negative than −1) is usually the lipid cofactor in the fission event. Interestingly, naturally occurring mutations have been reported in AH-containing fission-inducing proteins and related to diseases such as centronuclear myopathy (amphiphysin 2), Charcot-Marie-Tooth disease (GDAP1), Parkinson's disease (α-synuclein). These findings add to the interest of the membrane fission process whose complete understanding will be instrumental for the elucidation of the pathogenesis of diseases involving mutations in the protein AHs.

Keywords: membrane fission, membrane scission, fission-inducing protein, amphipathic helix, shallow insertion, lipid cofactor, lipid-binding site, neck-hemifission model

# INTRODUCTION

fcell-07-00291 December 6, 2019 Time: 15:40 # 2

Membrane fission (or scission), the process opposite to membrane fusion, consists in the splitting of one membrane into two separate membranes; instead, membrane fusion is a process by which two biological membranes converge into one membrane (Falanga et al., 2009; Kozlov et al., 2010; Scott and Youle, 2010; Knorr et al., 2017). Several examples of membrane fission processes can be found in the literature: cell division (Bohuszewicz et al., 2016; Caspi and Dekker, 2018; Stoten and Carlton, 2018), endocytosis (Renard et al., 2015; Ferreira and Boucrot, 2018; Kaksonen and Roux, 2018; Mettlen et al., 2018; Sandvig et al., 2018; Genet et al., 2019), caveolae biogenesis (Parton et al., 2006; Kirkham et al., 2008; Ariotti et al., 2015), budding of vesicles from endomembranes (Lee et al., 2005; Krauss et al., 2008; Hagen et al., 2015; Park et al., 2019), nuclear envelope repair (Isermann and Lammerding, 2017), neuron pruning (Loncle et al., 2015), mitochondrial division (Adachi et al., 2016; Huber et al., 2016; Pagliuso et al., 2018; Tandler et al., 2018; Tilokani et al., 2018; Agrawal and Ramachandran, 2019; Irajizad et al., 2019; Yoshida and Mogi, 2019), plastid division (Osteryoung and Pyke, 2014; Yoshida, 2018; Yoshida and Mogi, 2019), peroxisome fission (Huber et al., 2016; Schrader et al., 2016; Su et al., 2018), endosome fission (Daly and Cullen, 2018; Hoyer et al., 2018; Kitamata et al., 2019), vacuole fission (Gopaldass et al., 2017), Golgi membrane fission (Weigert et al., 1999; Hidalgo Carcedo et al., 2004; Bonazzi et al., 2005; Yang et al., 2005, 2008, 2011; Corda et al., 2006; Colanzi et al., 2007; Valente et al., 2013), macropinocytosis (Liberali et al., 2008), macroautophagy (Knorr et al., 2015; Yu and Melia, 2017; Osawa et al., 2019), microautophagy (Uttenweiler and Mayer, 2008; Li et al., 2012), thylakoid membrane remodeling (Chuartzman et al., 2008), formation of multivesicular bodies (Piper and Katzmann, 2007), sporulation in bacteria (Doan et al., 2013; Gifford and Meyer, 2015); formation of bacterial membrane vesicles (Eriksson et al., 2009; Schwechheimer and Kuehn, 2015; Bitto and Kaparakis-Liaskos, 2017), bacterial chromatophores (Bohuszewicz et al., 2016) and bacterial magnetosomes (Uebe and Schüler, 2016); formation of membrane vesicles in archaea (Ellen et al., 2010), virus budding from the cell (Rossman and Lamb, 2013; Adu-Gyamfi et al., 2015; Bigalke and Heldwein, 2015; Herneisen et al., 2017). Most of these membrane fission reactions are mediated by specialized proteins (Campelo and Malhotra, 2012; Frolov et al., 2015; Renard et al., 2015) that include the AHcontaining proteins (see, e.g., Antonny et al., 1997; Farsad et al., 2001; Boucrot et al., 2012; Martyna et al., 2017).

Based on the many examples of membrane fission driven by AH insertion into the membrane bilayer, we have summarized in this review the known AH-containing proteins and related mechanisms linked to the fission event. Following a brief outline on the hemifission intermediate, we here propose to divide the membrane fission processes in four classes based on direct/indirect energy consumption and on the budding toward the cytosol or away from the cytosol. We then discuss the fission events mediated by the AH-containing proteins and underline the role of specific lipid cofactors and of the oligomeric state. We also report the naturally occurring mutations of these proteins.

#### PROTEIN-MEDIATED MEMBRANE FISSION

#### Stages of Fission Process: "Neck-Hemifission" Model

Membrane fission proceeds through the following steps: membrane neck intermediate, hemifission intermediate, and two separate membrane formation (Corda et al., 2002; Kozlovsky and Kozlov, 2003; Bashkirov et al., 2008; Campelo and Malhotra, 2012; Frolov et al., 2015; Antonny et al., 2016; McDargh and Deserno, 2018; Pannuzzo et al., 2018). Hemifission is an intermediate state where the proximal monolayer (also called contacting monolayer) of the bilayer forming the neck coalesces, thus breaking the inner volume into two parts, while the distal monolayer remains continuous. Separation of distal monolayers completes the fission process. The existence of hemifission

**Abbreviations:** AH, amphipathic helix; ANKHD1, ankyrin repeat and KH domain-containing protein 1; Arf, ADP ribosylation factor; ArfGAP1, ADPribosylation factor GTPase-activating protein 1; Atg, autophagy-related protein; ATP, adenosine triphosphate; AtPmtA, Agrobacterium tumefaciens phospholipid N-methyltransferase; BAR, Bin/Amphiphysin/Rvs; BDLP1, bacterial dynaminlike protein 1; Bif-1, Bax-interacting factor 1; BIN1, bridging integrator-1; CALM, Clathrin Assembly Lymphoid-Myeloid leukemia protein; CCT, cytidine triphosphate:phosphocholine cytidylyltransferase; CCVs, clathrin-coated vesicles; CdvC, Cell division protein C; CL, cardiolipin; CME, clathrin-mediated endocytosis; CNM, centronuclear myopathy; COP, coat protein; CtBP1-S/BARS, C-terminal binding protein 1 shorter isoform/brefeldin A ADP-ribosylated substrate; DGS, diglucosyldiacylglycerol synthase; DHN1, dehydrin 1; Dnm2, dynamin2; DRP, dynamin-related protein; EcMurG, Escherichia coli MurG; EHD1, Eps15-homology domain-containing protein 1; ENTH/ANTH, Epsin N-Terminal Homology/AP180 N-Terminal Homology; EPR, electron paramagnetic resonance; ER, endoplasmic reticulum; ESCRT, endosomal sorting complexes required for transport; F-BAR, FCH-BAR; FEME, fast endophilin-mediated endocytosis; FisB, fission protein B; GDAP1, Ganglioside-induced Differentiation Associated Protein 1; GDP, guanosine diphosphate; GMAP-210, Golgi Microtubule-Associated Protein 210; GTP, guanosine triphosphate; GUVs, giant unilamellar vesicles; I-BAR, Inverse BAR; IMM, inner mitochondrial membrane; INF2, inverted formin-2; LPAAT, lysophosphatidic acid acyltransferase; MFF, mitochondrial fission factor; MGS, monoglucosyldiacylglycerol synthase; MiD49 and MiD51, mitochondrial dynamics proteins 49 and 51; MinD, septum site-determining protein MinD; MMPE, monomethyl-phosphatidylethanolamine; MurG, N-acetylmuramyl-(pentapeptide) pyrophosphoryl-undecaprenol N-acetylglucosamine transferase; N-BAR, N-terminal amphipathic helixcontaining BAR; OMM, outer mitochondrial membrane; Opi1, overproduction of inositol 1; PA, phosphatidic acid; PBP, phosphatidylinositol bisphosphate; PC, phosphatidylcholine; Pcyt1a, cytidine triphosphate:phosphocholine cytidylyltransferase alpha; Pex11, peroxisomal membrane protein 11; PG, phosphatidylglycerol; PICK1, protein interacting with C kinase 1; PM, plasma membrane; PmtA, phospholipid N-methyltransferase; POPG, 1-palmitoyl-2 oleoyl-sn-glycero-3-phosphatidylglycerol; PROPPIN, β-propellers that bind polyphosphoinositides; PS, phosphatidylserine; PtdIns, phosphatidylinositol; PtdIns3P, phosphatidylinositol 3-phosphate; PtdIns4P, phosphatidylinositol 4-phosphate; PtdIns5P, phosphatidylinositol 5-phosphate; PtdIns(3,4)P2, phosphatidylinositol 3,4-bisphosphate; PtdIns(3,5)P2, phosphatidylinositol 3,5-bisphosphate; PtdIns(3,4,5)P2, phosphatidylinositol 3,4,5-trisphosphate; PtdIns(4,5)P2, phosphatidylinositol 4,5-bisphosphate; Sar, secretion-associated Ras-related protein; SH3YL1, SRC homology 3 domain containing Ysc84-like 1; SNCA, α-synuclein; Snf7, sucrose non-fermenting protein 7; Spo20, sporulationspecific protein 20; STIM2, stromal interaction molecule 2; Tam41, translocator assembly and maintenance protein 41; VHS, Vps27, Hrs, and STAM; VP40, viral protein 40; Vps4, Vacuolar protein sorting-associated protein 4A; wt, wild type.

intermediate is confirmed by experimental observation that membrane fission is non-leaky (Bashkirov et al., 2008).

Neck and hemifission intermediates of the membrane fission process resemble pore and hemifusion intermediates of the opposite membrane fusion process. Membrane fusion proceeds via an intermediate named stalk, a lipidic hourglass-shaped connection between the contacting membrane leaflets. As a result of the radial expansion of the stalk, a hemifusion diaphragm forms. This intermediate is a bilayer formed by two distal leaflets of the fusing membranes. Finally, a fusion pore forms and establishes continuity between aqueous spaces of fusing membrane-enclosed compartments. The difference between membrane neck (sometimes referred to as "fission pore") and fusion pore is that neck narrows, while fusion pore expands. By analogy with the "stalk-pore model" of membrane fusion (Chernomordik and Kozlov, 2008; Martens and McMahon, 2008; Kielian, 2014; Podbilewicz, 2014; Zhukovsky et al., 2019a; and references therein), we will call this membrane fission pathway as "neck-hemifission model."

#### Classes of Membrane Fission Reactions

Renard et al. (2018) suggested to classify membrane fission mechanisms into two main categories: active fission (with the direct consumption of cellular energy by nucleoside triphosphate hydrolysis) and passive fission (without the direct use of energy). Many membrane fission processes are dependent on the interaction of fission-inducing proteins with specific lipid molecules (see below). In some cases, passive fission might be energized indirectly, via the energy used in the synthesis of these lipid cofactors (Gopaldass et al., 2017).

Moreover, membrane fission processes can be classified into two types: "normal topology fission" (membrane-enclosed compartments bud toward the cytosol) and "reverse topology fission" (membrane-enclosed compartments bud away from the cytosol) (Votteler and Sundquist, 2013; Frolov et al., 2015; Schöneberg et al., 2017; Caspi and Dekker, 2018; Snead and Stachowiak, 2018). During normal topology fission, distal monolayer of a membrane-enclosed compartment is cytoplasmic, whereas proximal monolayer is exoplasmic (**Figure 1**). During reverse topology fission, distal monolayer is exoplasmic, and proximal monolayer is cytoplasmic (**Figure 1**). We would like to point out that after "normal topology fission," disjoint union of two membrane-enclosed compartments (as in the case of formation of post-Golgi carriers) or joint union (as in the case of endocytosis) can form. Similarly, after "reverse topology fission," disjoint union (as in the case of cell division) or joint union [as in the case of herpesvirus primary envelopment (Bailer, 2017; Klupp et al., 2018) or formation of multivesicular bodies (Piper and Katzmann, 2007)] can form (**Figure 1**). Based on these considerations, we propose to divide membrane fission processes into four classes (**Figure 2**):

Class I: Active mechanism, normal topology. Examples: fission mediated by the large GTPase dynamin and other members of the dynamin superfamily (Daumke and Praefcke, 2016; Ramachandran and Schmid, 2018; Ford and Chappie, 2019; Jimah and Hinshaw, 2019) and by the small AH-containing GTPases Arf1 (Krauss et al., 2008; Beck et al., 2011; Bottanelli et al., 2017; Dodonova et al., 2017) and Sar1 (Bielli et al., 2005; Lee et al., 2005; Bacia et al., 2011; Hariri et al., 2014; Hanna et al., 2016; Melero et al., 2018).

Class II: Passive mechanism, normal topology. Examples: fission mediated, in the absence of nucleoside triphosphate hydrolysis, by Bacillus subtilis FisB (Doan et al., 2013), C-terminal Binding Protein 1 Short form/Brefeldin A ADP-Ribosylation substrate (CtBP1-S/BARS) (Spanò et al., 1999; Weigert et al., 1999; Hidalgo Carcedo et al., 2004; Valente et al., 2005, 2012; Liberali et al., 2008; Pagliuso et al., 2016; Zhukovsky et al., 2019b), and by numerous AH-containing proteins, such as endophilins (Rostovtseva et al., 2009; Ambroso et al., 2014; Genet et al., 2019), amphiphysins (Wu and Baumgart, 2014; Snead et al., 2019), epsins (Ford et al., 2002; Boucrot et al., 2012; Brooks et al., 2015), α-synuclein (Nakamura et al., 2011; Braun et al., 2017; Pozo Devoto and Falzone, 2017; Fakhree et al., 2019), GDAP1 (Huber et al., 2016), ankyrin repeats and KH domain-containing protein 1 (ANKHD1) (Kitamata et al., 2019), Saccharomyces cerevisiae Atg18 (Gopaldass et al., 2017), Agrobacterium tumefaciens PmtA (Danne et al., 2017b), EcMurG (van den Brink-van der Laan et al., 2003), Acholeplasma laidlawii MGS (Eriksson et al., 2009; Ariöz et al., 2014; Ge et al., 2014) and DGS (Eriksson et al., 2009).

Class III: Active mechanism, reverse topology. Examples: fission mediated by the ESCRT machinery (Chiaruttini and Roux, 2017; Schöneberg et al., 2017; Stoten and Carlton, 2018; Ahmed et al., 2019; Gatta and Carlton, 2019), bacterial cell division (Bohuszewicz et al., 2016; Vedyaykin et al., 2019) and archaeal cell division (Caspi and Dekker, 2018). Of note, Vps4 ATPase (Monroe et al., 2014; Caspi and Dekker, 2018), BDLP1 containing N-terminal GTPase domain (Bohuszewicz et al., 2016), and CdvC containing ATPase subunit (Caspi and Dekker, 2018) proteins provide energy for the ESCRT pathway, bacterial cell division, and archaeal cell division, respectively. Interestingly, archaeal CdvC is the homolog of eukaryotic Vps4 (Caspi and Dekker, 2018).

Class IV: Passive mechanism, reverse topology. Examples: fission mediated by the herpesvirus fission machine (Bigalke and Heldwein, 2015), influenza A virus AH-containing protein M2 (Herneisen et al., 2017; Martyna et al., 2017; Madsen et al., 2018), Ebola virus VP40 (Soni and Stahelin, 2014; Adu-Gyamfi et al., 2015).

Although in Renard et al. (2018) CtBP1-S/BARS-mediated fission is classified as an active fission mechanism, we propose that this process should be considered as passive, because no direct consumption of energy by nucleoside triphosphate hydrolysis takes place during this process.

Any membrane fission reaction belongs to the category of active or passive fission and almost all fission processes can be classified as normal topology or reverse topology fission. Moreover, only very few fission processes (such as division of IMMs) are mediated by proteins not residing in the cytosol.

(yellow circles) are considered. When membrane fission takes place, separation of proximal leaflets is followed by separation of distal leaflets. During normal topology fission (membrane-enclosed compartment buds toward the cytoplasm), distal leaflet is cytoplasmic, whereas proximal leaflet is exoplasmic. During reverse topology fission (membrane-enclosed compartment buds away from the cytoplasm), distal leaflet is exoplasmic, and proximal leaflet is cytoplasmic. During both normal and reverse topology fission reactions, disjoint or joint union of membrane-enclosed compartments can form. If disjoint union forms, outer leaflet is distal, and inner leaflet is proximal. When joint union forms, outer leaflet is proximal, whereas inner leaflet is distal. Fission-inducing proteins are usually present on the cytoplasmic side of the membrane.

Therefore, with very few exceptions, all fission reactions belong to one of the four classes described above. Moreover, many fission reactions mediated by AH-containing proteins are very diverse. However, assigning of these reactions to one of four classes I–IV will allow to find similarities in fission processes which seem to be completely different.

#### FISSION DRIVEN BY AMPHIPATHIC HELIX-CONTAINING PROTEINS

Amphipathicity is the segregation of hydrophobic and hydrophilic amino acid residues between the two opposite faces of the protein α-helix, a distribution well suited for membrane binding (Drin and Antonny, 2010; Giménez-Andrés et al., 2018). A hydrophobic moment was introduced (Eisenberg et al., 1982, 1984) to estimate whether a protein sequence, when considered as helical, exhibits one polar and one hydrophobic face (Drin and Antonny, 2010). A large value of hydrophobic moment suggests that the protein sequence can be folded as AH perpendicularly to its axis (Eisenberg et al., 1982).

Various fission reactions are known to be driven by the large GTPase dynamin and other mechanoenzymes belonging to the dynamin superfamily (Daumke and Praefcke, 2016; Ramachandran and Schmid, 2018; Ford and Chappie, 2019; Jimah and Hinshaw, 2019; and references therein). In other fission processes that are not dependent on mechanoenzymes, proteins that contain AHs often play an important role (see, e.g., Antonny et al., 1997; Farsad et al., 2001; Boucrot et al., 2012; Martyna et al., 2017; and references therein).

Amphipathic helices are involved in many biological processes such as protein–protein interactions or interaction with biomembranes (Segrest et al., 1990, 1992). When an AH binds to membrane, it orients itself parallel to the membrane

plane. The hydrophobic side of this helix inserts into the interior of the bilayer, while residues of the hydrophilic side interact with the lipid headgroups (Drin and Antonny, 2010; Roberts et al., 2013). The insertion of AHs into the bilayer promotes membrane curvature thus supporting the fission reaction (Campelo et al., 2008; Martyna et al., 2017).

Moreover, some protein segments (named "AH motifs") of membrane-binding proteins are not completely helical in solution, but become AHs upon interaction with the membrane (Cornell and Taneva, 2006; Drin and Antonny, 2010; Ambroso et al., 2014; Chong et al., 2014; Horchani et al., 2014; Barneda et al., 2015; Frolov et al., 2015; and references therein). The presence of anionic lipids in the membrane often promotes the AH folding (see, for example, Davidson et al., 1998; Kweon et al., 2006; Fernandes et al., 2008; Horchani et al., 2014; Brady et al., 2015; Mizuno et al., 2017; Ryan et al., 2018).

The ability to sense membrane curvature by some proteins favors their interaction with curved membranes. This process not always leads to curvature generation. However, other AHs are able to induce curvature (Drin and Antonny, 2010; Wilz et al., 2011), an event required for proteins mediating membrane fission. The basic principles governing membrane curvature generation remain to be clarified (Khattree et al., 2013).

Amphipathic helices characterized by the presence of basic residues at the polar-non-polar interface are well suited for interaction with membranes (Mishra and Palgunachari, 1996; Davidson et al., 1998; Polozov et al., 1998; Mozsolits et al., 2004), especially with those containing negatively charged phospholipids by electrostatic attraction (Polozov et al., 1998). It has been hypothesized that interfacial positively charged residues help to anchor the AH in the lipid bilayer (Mishra and Palgunachari, 1996). We would like to emphasize that the AHs of many fission-inducing proteins contain at least two basic residues (Arg and/or Lys) each, at the polar-non-polar interface. Examples are: H0 helix of endophilin A1 contains Lys7 and Lys16 (Ambroso et al., 2014); AH of PmtA contains Lys6, Arg8, Lys12 (Danne et al., 2017b); AH of Atg18 contains Arg372 and Arg377 (Gopaldass et al., 2017). In PmtA, Arg8 and Lys12 are involved in protein attachment to anionic lipids (Danne et al., 2017b). Overall, the AHs of many fission proteins contain at least three positively charged residues. On the contrary, the AHs of proteins that sense but do not generate membrane curvature, such as ArfGAP1, GMAP-210, Kes1p, often contain not more than one basic residue (Drin and Antonny, 2010).

Campelo et al. (2008), similarly to Zemel et al. (2008), predicted that shallow (penetrating ca. 40% of the monolayer thickness) insertion of AH is particularly effective in the generation of membrane curvature. Since AHs and AHcontaining protein segments are particularly suitable as membrane curvature-generating inclusions, their shallow insertions were suggested to be sufficient to mediate membrane fission (Boucrot et al., 2012).

The insertion depth of the AHs of fission-inducing proteins was studied experimentally. Using two hydrophobic quenchers of tryptophan fluorescence (shallow quencher and deep quencher), Hanna et al. (2016) detected shallow penetration of the Sar1B AH into the membrane, just below the hydrophilic headgroups. EPR measurements demonstrated shallow insertion of the AHs of endophilin A1 (Gallop et al., 2006), epsin (Lai et al., 2012), and α-synuclein (Jao et al., 2008) that embed at the level of lipid phosphate groups (Gallop et al., 2006; Lai et al., 2012) or just below this level (Jao et al., 2008). According to molecular dynamics simulations, the insertion depth of AHs from epsin and Sar1p is lower than that of Arf1, and these helices from epsin and Sar1p generate higher membrane curvature than the AH from Arf1 (Li, 2018). Interaction of the AHs of fission-inducing proteins with specific lipids might ensure the optimal insertion depth for the generation of membrane curvature and for the interaction with other fission-inducing proteins, as discussed below.

An additional mechanism foresees a cooperativity among proteins where the first insertion of an AH from a given protein able to sense membrane curvature (i.e., to bind selectively to curved membranes) favors the insertion of additional AHs (Gallop et al., 2006; Lundmark et al., 2008; Miller et al., 2015; Hanna et al., 2016; Martyna et al., 2017). In turn, these AHs induce more membrane curvature facilitating the insertion of additional AHs. Due to this positive cooperativity in membrane curvature generation, the runaway process leading to membrane fission might take place once a critical concentration of AH-containing fission-inducing proteins is reached (Miller et al., 2015).

Amphipathic helix-containing fission-inducing proteins are recognized as a separate superfamily of scission factors (Gallop et al., 2006; Boucrot et al., 2012; Martyna et al., 2017). Among these proteins there are: Arf1 (Krauss et al., 2008; Beck et al., 2011; Bottanelli et al., 2017; Dodonova et al., 2017), Sar1 (Bielli et al., 2005; Lee et al., 2005; Bacia et al., 2011; Hariri et al., 2014; Hanna et al., 2016; Melero et al., 2018), epsins (Ford et al., 2002; Boucrot et al., 2012; Brooks et al., 2015), CALM (Miller et al., 2015), endophilin A1 (Gallop et al., 2006; Ambroso et al., 2014), endophilin A2 (Boucrot et al., 2015; Renard et al., 2015; Simunovic et al., 2017; Genet et al., 2019), endophilin A3 (Boucrot et al., 2012), endophilin B1 (Rostovtseva et al., 2009; Takahashi et al., 2016), mammalian amphiphysin 1 (Snead et al., 2019) and amphiphysin 2 (Wu and Baumgart, 2014), Drosophila amphiphysin (Isas et al., 2015), PICK1 (Karlsen et al., 2015), α-synuclein (Nakamura et al., 2011; Braun et al., 2017; Pozo Devoto and Falzone, 2017; Fakhree et al., 2019), GDAP1 (Huber et al., 2016), caveolin-1 (Parton et al., 2006; Kirkham et al., 2008; Ariotti et al., 2015), ANKHD1 (Kitamata et al., 2019), peroxins Pex11B (Yoshida et al., 2015) and Pex11p (Opalinski et al., 2011 ´ ; Su et al., 2018), yeast PROPPIN Atg18 (Gopaldass et al., 2017), Snf7 belonging to the ESCRT complex (Buchkovich et al., 2013), influenza A virus M2 (Rossman et al., 2010; Roberts et al., 2013; Herneisen et al., 2017; Madsen et al., 2018), and bacterial enzymes PmtA (Danne et al., 2017b), MurG (van den Brink-van der Laan et al., 2003; Lind et al., 2007; Albesa-Jové et al., 2014), MGS (Lind et al., 2007; Eriksson et al., 2009; Ge et al., 2014), and DGS (Eriksson et al., 2009). Unlike wt proteins, mutants of Arf1 (Beck et al., 2011), Sar1 (Lee et al., 2005), endophilin A1 (Farsad et al., 2001; Masuda et al., 2006; Mim et al., 2012), Drosophila amphiphysin (Yoon et al., 2012), mammalian amphiphysin 1 (Farsad et al., 2001), Atg18 (Gopaldass et al., 2017) lacking AHs were unable to generate membrane curvature.

Increasing of the hydrophobic moment favors folding of AH motif into an α-helix and binding to membrane (Drin and Antonny, 2010). Thus, it is not surprizing that mutations of hydrophobic residues, within AHs, to Ala or to hydrophilic residues (substitutions that decrease hydrophobic moment) inhibit the ability of fission proteins such as Arf1 (Krauss et al., 2008), Sar1p (Lee et al., 2005), endophilin A1 (Farsad et al., 2001; Gallop et al., 2006; Masuda et al., 2006; Suresh and Edwardson, 2010), epsin (Ford et al., 2002; Boucrot et al., 2012), Pex11p (Opalinski et al., 2011 ´ ), Pex11B (Yoshida et al., 2015), yeast Atg2 (Kotani et al., 2018), mammalian Atg2A (Tamura et al., 2017), caveolin-1 (Parton et al., 2006; Kirkham et al., 2008; Ariotti et al., 2015), M2 (Rossman et al., 2010; Roberts et al., 2013), PmtA (Danne et al., 2017b), to generate membrane curvature. On the contrary, mutations of hydrophobic residues to more bulky hydrophobic Trp (substitutions that increase hydrophobic moment) promoted the ability of Arf1 (Krauss et al., 2008), epsin (Ford et al., 2002; Boucrot et al., 2012), and Pex11p (Opalinski ´ et al., 2011) to induce membrane curvature, see **Table 1**.

Substitution of positively charged residues of the AH hydrophilic face for negatively charged residues leads to repulsion from similarly charged membrane and, thus, to the reduced ability of AH to induce membrane curvature. This conclusion was confirmed experimentally. Mutations of basic Lys and Arg to negatively charged Glu within AH of endophilin A1 (Gallop et al., 2006) inhibited the ability of this fission-inducing protein to generate membrane curvature.

The influence of the mutations of AH residues on the ability of AH-containing proteins to generate membrane curvature is summarized in **Table 1**.

Some of AH-containing fission-inducing proteins, such as the small GTPases Arf1 and Sar1, mediate class I fission reactions. Other proteins, such as yeast Atg18 and bacterial enzymes PmtA, MurG, MGS, DGS, drive class II fission processes. Snf7 protein is involved in class III membrane fission reactions of the ESCRT pathway. Influenza virus M2 mediates class IV fission reaction. However, some AH-containing proteins, such as endophilin B1 and amphiphysin 1, could be involved, in complex with mechanoenzymes belonging to the dynamin superfamily, in class I fission reactions (Takahashi et al., 2016; Takeda et al., 2018), or, in the absence of mechanoenzymes, in class II fission reactions (Rostovtseva et al., 2009; Snead et al., 2019). Hence, belonging to a certain class is a property of the process, not a property of the protein molecule.

We should keep in mind that some membrane fissioninducing protein complexes, such as herpesvirus fission machine (Bigalke and Heldwein, 2015; Lorenz et al., 2015; Klupp et al., 2018), Ebola virus fission protein VP40 (Soni and Stahelin, 2014;


TABLE 1 | Mutations in amphipathic helices of fission-inducing proteins that inhibit or promote the ability of these proteins to generate membrane curvature.

Adu-Gyamfi et al., 2015), and FisB that mediates fission during sporulation in B. subtilis (Doan et al., 2013), contain neither proteins belonging to the dynamin superfamily nor AH-containing proteins.

#### SUBFAMILIES OF AMPHIPATHIC HELIX-CONTAINING FISSION-INDUCING PROTEINS

At least four subfamilies of proteins containing AHs able to drive membrane fission can be identified so far: small GTPases, ENTH/ANTH domain-containing proteins, BAR domain-containing proteins, and Atg proteins. Moreover, we suppose that synuclein family of proteins can be also considered as a subfamily of AH-containing fission-inducing proteins.

#### Small GTPases

The small GTPases Sar1 (Bielli et al., 2005; Lee et al., 2005; Bacia et al., 2011; Hariri et al., 2014; Hanna et al., 2016; Melero et al., 2018) and Arf1 (Krauss et al., 2008; Beck et al., 2011; Bottanelli et al., 2017; Dodonova et al., 2017) are involved in the intracellular trafficking of proteins and lipids in COP-coated vesicles (Graham and Kozlov, 2010; Adolf et al., 2013; Cevher-Keskin, 2013; Yorimitsu et al., 2014; and references therein). Sar1 plays a role in the COPII-mediated anterograde trafficking from the ER to the Golgi apparatus, whereas Arf1 is involved in COPI-mediated retrograde trafficking from the Golgi to the ER, as well as in the intra-Golgi transport. Sar1 and Arf1, as well as other small GTPase family proteins, share structural similarities. These two proteins are highly conserved in evolution. Yeast S. cerevisiae has one Sar1, whereas mammals have two Sar1 paralogs, denoted Sar1A and Sar1B in humans (Loftus et al., 2012). Sar1 and Arf1 function as molecular switches. They are cytosolic and inactive when bound to GDP. Upon exchange of GDP to GTP, these small GTPases undergo a conformational change into their active form. In this way, the N-terminal AH becomes exposed. This helix has one hydrophobic face, which requires engagement in a hydrophobic environment and thus it inserts into the cytoplasmic-membrane leaflet (Graham and Kozlov, 2010; Adolf et al., 2013). The AH of Arf1 is myristoylated, unlike the AH of Sar1. Membrane-bound Sar1/Arf1 recruit coat proteins and initiate the formation of COP-coated vesicles. Insertion of the AHs of Sar1 and Arf1 into the membrane drives fission and, in turn, the release of these COP-coated vesicles (Bielli et al., 2005; Lee et al., 2005; Krauss et al., 2008; Hanna et al., 2016).

#### ENTH/ANTH Domain-Containing Proteins

Epsins are highly conserved fission-inducing proteins that contain the N-terminal ENTH domain (Sen et al., 2012). Epsins play a role in the fission of CCVs during CME (Ford et al., 2002; Boucrot et al., 2012; Brooks et al., 2015). Four epsin paralogs are known in mammals: epsin 1 to 3 and epsin-related (EpsinR) (Takatori and Tomita, 2018). The role of epsin 1 in membrane curvature generation is well known (Ford et al., 2002; Kweon et al., 2006), although epsin 2 and epsin 3 are also involved in CME (Boucrot et al., 2012). The ENTH domain is approximately 140 residues long and has a compact globular structure of seven α-helices followed by an eighth helix that is not aligned with the other seven (Sen et al., 2012). Upon

interaction with PtdIns(4,5)P2-containing membrane, ENTH domain of epsin 1 undergoes a conformational change that leads to the formation of an additional N-terminal α-helix denoted helix 0 (H0) (Ford et al., 2002) consisting of residues Met1-Tyr17 (Martyna et al., 2017), or residues Ser4–Val14, as reported by Lai et al. (2012). This is the helix inserting into the membrane. Primary sequences of the N-termini (residues 1–17) of human epsins 1, 2, 3 are very similar. It has been proposed that insertion of H0 helix into the bilayer induces membrane curvature required for fission (Ford et al., 2002; Boucrot et al., 2012), although recent study raised doubts on this mechanism and suggested that, instead, crowding of intrinsically disordered protein domains is able to drive fission (Snead et al., 2017).

Dynamin plays an important role in the fission of CCVs. Nevertheless, epsin is able to mediate CCV fission in dynamindepleted cells. Boucrot et al. (2012) suggested that epsin molecules present at the neck of a budding vesicle might provide the force that destabilizes the neck of budding vesicle, thus leading to fission.

It should be emphasized that the role of epsins in the formation and fission of CCVs is somewhat similar to the role of Sar1 and Arf1 in the formation and fission of COP vesicles (Boucrot et al., 2012; Adolf et al., 2013). In both cases, an external trigger leads to the exposure of an AH that inserts into the cytosolic leaflet of the membrane and is assumed to drive the fission of vesicles. In case of small GTPases, exchange of GDP for GTP plays a role of such a trigger, whereas in case of epsin, AH exposure is triggered by the interaction with PtdIns(4,5)P2 containing membrane. Epsin, as well as small GTPases, might create a high-energy state at the neck of budding vesicle, and this state is relaxed by fission of this vesicle from the donor membrane (Boucrot et al., 2012; Adolf et al., 2013). Moreover, small GTPases are responsible for recruitment of COP proteins, whereas epsin is involved in clathrin coat assembly.

The CALM protein, also known as PICALM, is also involved in fission of CCVs during CME (Miller et al., 2015). CALM contains a N-terminal ANTH domain that is approximately twice as big as the ENTH domain, but has the same core of α-helices (Duncan and Payne, 2003; Legendre-Guillemin et al., 2004; Takatori and Tomita, 2018; Kaneda et al., 2019). ENTH and ANTH domains are so similar that several ANTH domains were originally designated ENTH domains, although later these two domains were subclassified as two distinct domains, ENTH and ANTH (Ford et al., 2002). Structurally, ENTH and ANTH domains are similar to the VHS domain (De Craene et al., 2012). Most ENTH and ANTH domains are lipid-binding domains, they interact specifically with phosphoinositides present in the membrane. However, the mechanisms of the interaction of ENTH and ANTH domains with PtdIns(4,5)P<sup>2</sup> are quite different (Kaneda et al., 2019; and references therein).

ANTH domain-containing protein CALM, like ENTH domain-containing protein epsin 1, are characterized by an N-terminal AH that is supposed to insert into the membrane and to play a role in membrane fission (Miller et al., 2015). Proteins containing ENTH/ANTH domains (epsins 1, 2, 3 and CALM) are somewhat similar and can be classified as separate subfamily of AH-containing fission-inducing proteins. Possibly, the plant ANTH domain-containing protein AP180 also belongs to this subfamily (Kaneda et al., 2019). Similarly, new fission proteins belonging to this subfamily might be discovered in the future, based on the identification of AHs in their structure.

# BAR Domain-Containing Proteins

The BAR domain is found in many proteins implicated in various cellular processes, most of which are related to membrane remodeling (Stanishneva-Konovalova et al., 2016; Carman and Dominguez, 2018; Nishimura et al., 2018; Simunovic et al., 2019). BAR domains typically form banana-shaped homodimers, where each protomer consists of three-helix antiparallel coiledcoil structure. Because of its curved shape, the BAR domain is able to sense membrane curvature, i.e., to bind preferentially to curved membranes. Thus, the concave shape of the BAR-domain dimer is positively charged and interacts with the negatively charged cytoplasmic monolayer of biomembranes. Most BAR domain proteins also contain auxiliary domains involved in the interactions with other proteins or with membranes (Mim et al., 2012; Carman and Dominguez, 2018). Based on their structural properties, BAR domain proteins can be classified into three main groups: a classical BAR (including N-BAR), F-BAR, and I-BAR (Ahmed et al., 2010; Stanishneva-Konovalova et al., 2016; Nishimura et al., 2018). Proteins belonging to the N-BAR subgroup contain an N-terminal sequence H0 that folds into an AH upon membrane binding. Endophilins (Kjaerulff et al., 2011) and amphiphysins (Wu et al., 2009; Prokic et al., 2014) belong to the N-BAR subgroup and thus contain these H0-AH motifs. Moreover, endophilins contain an additional AH, known as H1I (helix 1 insert). In mammals endophilins are encoded by five genes: endophilins A1, A2, A3, B1, B2, whereas amphiphysins are encoded by two genes: amphiphysin 1 and amphiphysin 2 (also known as BIN1) (Kjaerulff et al., 2011; Prokic et al., 2014).

Bin/Amphiphysin/Rvs domain protein PICK1 contains an internal AH (Karlsen et al., 2015; Herlo et al., 2018). Mammalian endophilins A1 (Gallop et al., 2006; Ambroso et al., 2014), A2 (Boucrot et al., 2015; Renard et al., 2015; Simunovic et al., 2017; Genet et al., 2019), A3 (Boucrot et al., 2012) and B1 (Rostovtseva et al., 2009; Takahashi et al., 2016), mammalian amphiphysin 1 (Snead et al., 2019) and amphiphysin 2 (Wu and Baumgart, 2014), Drosophila amphiphysin (Isas et al., 2015), as well as mammalian PICK1 (Karlsen et al., 2015) are all involved in membrane fission. The AHs were suggested to generate membrane curvature and to play a role in membrane fission mediated by endophilins (Gallop et al., 2006; Masuda et al., 2006; Boucrot et al., 2012; Yoon et al., 2012) and amphiphysins (Peter et al., 2004; Boucrot et al., 2012; Yoon et al., 2012; Wu and Baumgart, 2014). Moreover, a higher number of AHs in endophilins compared with amphiphysins was suggested to correlate with increased ability to penetrate the membrane and to drive membrane fission (Boucrot et al., 2012; Yoon et al., 2012). However, other studies question the role of these AHs in membrane curvature generation (e.g., see Chen et al., 2016). The BAR domains were suggested to restrict (Boucrot et al., 2012) or to promote membrane fission (Snead et al., 2019). Finally, Snead et al. (2019) reported that the steric

pressure among disordered domains of amphiphysin 1 plays a role in fission. Based on these reports, the molecular mechanism taking place in endophilin- and amphiphysin-induced fission remains to be fully elucidated.

Like AH motif of epsin (Ford et al., 2002), AH motifs of BAR domain-containing fission-inducing proteins endophilins and amphiphysins are also disordered in solution, but become α-helical upon interaction with the membrane (Löw et al., 2008; Jao et al., 2010). In this they differ from the fission-inducing small GTPases Arf1 and Sar1, whose AHs are already formed but are buried in the GDP-bound conformation, and only become exposed upon exchange of GDP to GTP (see above).

Often, endophilins mediate or inhibit fission cooperatively with the proteins belonging to the dynamin superfamily. Endophilin A2 and dynamin contribute to the fission of Shigatoxin-induced tubules (Renard et al., 2015). In a dynamindependent manner, endophilins A2 and A1 mediate clathrinindependent internalization route named FEME (Boucrot et al., 2015). Endophilin A1 inhibits dynamin-mediated membrane fission via intercalation between turns of the dynamin helix (Hohendahl et al., 2017). Endophilin B1 and dynamin 2 cooperatively induce the fission of Golgi membranes during autophagy (Takahashi et al., 2016), while the dynamic clustering of dynamin-amphiphysin 1 rings was suggested to regulate fission of large unilamellar vesicles (Takeda et al., 2018).

Interestingly, predicted curved structure of the ten C-terminal ankyrin repeats of AH-containing fission-inducing protein ANKHD1 is similar to the lipid-binding surface of BAR domaincontaining proteins (Kitamata et al., 2019).

#### Atg Proteins

Two Atg proteins, Atg2 and Atg18, contain AHs and mediate membrane fission (Gopaldass et al., 2017; Tamura et al., 2017; Kotani et al., 2018). Macroautophagy (hereinafter called autophagy) is a degradation process highly conserved from yeast to mammals (Feng et al., 2014; Knorr et al., 2015; Yu and Melia, 2017; Osawa et al., 2019; and references therein). During autophagy, a bowl-shaped membrane vesicle called phagophore appears, engulfs a part of the cytoplasm (including cytotoxins, damaged organelles, and invasive microbes), and expands. Finally, the open end of phagophore closes, and thus, the inner and outer bilayers become separate entities, and a double-membrane vesicle called autophagosome forms. Closure of the phagophore, leading to the autophagosome formation, is a membrane fission event (Knorr et al., 2015; Yu and Melia, 2017); and, moreover, is a reverse-topology fission event (Schöneberg et al., 2017). Finally, the outer membrane of the autophagosome fuses with a lysosome or vacuole, and captured cytoplasmic contents, together with the inner membrane of the autophagosome, is degraded by hydrolases. The resulting materials, such as amino acids, are released to the cytosol and recycled.

Many Atg proteins are required for the formation of autophagosome. Atg2 is one of these proteins. Two mammalian homologs of yeast Atg2 were identified: Atg2A and Atg2B (Velikkakath et al., 2012). Yeast Atg2 (Kotani et al., 2018) as well as mammalian Atg2 proteins (Velikkakath et al., 2012; Tamura et al., 2017; Tang et al., 2017) play a role in autophagosome formation. Two mammalian Atg2 homologs, Atg2A and Atg2B, have redundant functions in autophagy (Tamura et al., 2017). In the absence of mammalian Atg2A and Atg2B, the closure of autophagosomes is impaired (Velikkakath et al., 2012; Tang et al., 2017), indicating that these proteins are involved in membrane fission event, although other molecules might also be required for this process. Both yeast Atg2 (Kotani et al., 2018) and mammalian Atg2A (Tamura et al., 2017; Chowdhury et al., 2018) contain AHs, and the AH of yeast Atg2 corresponds to the AH of mammalian Atg2A (Kotani et al., 2018). These AHs are required for autophagosome formation (Tamura et al., 2017; Kotani et al., 2018). Mutations of hydrophobic residues of AH from human Atg2A to negatively charged residues abolish membrane-binding capability (Chowdhury et al., 2018) and causes defects in autophagy (Tamura et al., 2017). Similarly, mutations of hydrophobic residues of yeast Atg2 to negatively charged Asp inhibit autophagy (Kotani et al., 2018). From these observations, it derives that yeast and mammalian homologs of Atg2 belong to the family of AH-containing membrane fissioninducing proteins.

Autophagy-related proteins are classified into six functional groups (Velikkakath et al., 2012; Osawa et al., 2019). Atg2 and Atg18 belong to the same group. Atg18 forms a complex with Atg2 and also plays a role in autophagosome formation (Kotani et al., 2018). On top of its role in autophagy, Atg18 also mediates vacuole fission (Gopaldass et al., 2017). Amphipathic α-helical character of the Atg18 segment (residues ∼Pro366- Ser383) is required for Atg18-driven vacuole fission (Gopaldass et al., 2017). Vacuole fission is a normal topology fission event, and, hence, Atg18 is an interesting kind of protein that is involved in fission processes of two opposite topologies: formation of autophagosome (reverse topology) and vacuole fission (normal topology). It is noteworthy that Atg18 displays different requirements in these two fission processes. In vacuole fission, AH is crucial, whereas its binding partner Atg2 is dispensable. In autophagy, interaction of Atg18 with Atg2 is necessary, whereas the Atg18 AH is of moderate importance (Gopaldass et al., 2017). However, the ability of Atg18 to bind phosphatidylinositol 3-phosphate (PtdIns3P) is required for both processes: autophagy (Kotani et al., 2018) and vacuole fission (Gopaldass et al., 2017).

Interestingly, the AH motif of Atg18, similar to the AH motifs of epsins, endophilins and amphiphysins, is unstructured in solution, but transforms into an AH upon membrane binding (Gopaldass et al., 2017). Whereas small GTPases Arf1 and Sar1, ENTH/ANTH domain fission-inducing proteins, endophilins and amphiphysins contain N-terminal AH (see above), AHs of Atg proteins, Atg2 and Atg18, similar to the AH of PICK1, are internal and far, in primary sequence, from the N-terminus.

#### Synucleins

Mitochondria are double-membrane-bound organelles that constantly undergo fusion and fission events, referred to as mitochondrial dynamics (Tilokani et al., 2018). Fission of OMM occurs at the ER contact sites. Protein adaptors, such as MFF and MiD49 and MiD51, recruit Drp1, a soluble protein belonging

to the dynamin superfamily. Drp1 mediates OMM constriction, whereas final step, namely scission, is mediated by another mechanoenzyme, dynamin2 (Dnm2). Other proteins, such as INF2, Spire1C, actin, and Myosin IIA play a role in this fission process (Pagliuso et al., 2018; Tilokani et al., 2018; and references therein). Simplified model for OMM fission is shown in Figure 4 of Tilokani et al. (2018). Fission of IMM occurs independently of OMM fission (Agrawal and Ramachandran, 2019). Mechanisms regulating constriction of IMM are poorly understood.

Proteins belonging to the synuclein family (Lavedan, 1998), namely α-synuclein (Kamp et al., 2010; Nakamura et al., 2011; O'Donnell et al., 2014; Martinez et al., 2018) and β-synuclein (Nakamura et al., 2011; Taschenberger et al., 2013), are involved in mitochondrial fragmentation. α-Synuclein contains a uniquely long α-11/3 AH that forms upon binding to a lipid bilayer (George et al., 1995; Davidson et al., 1998; Drin and Antonny, 2010; Braun et al., 2017). Similar AHs are present also in β-synuclein and γ-synuclein (Ducas and Rhoades, 2012). β-Synuclein mediates less fragmentation than α-synuclein, whereas γ-synuclein causes very little if any fragmentation (Nakamura et al., 2011). Therefore, in this section we mostly discuss α-synuclein.

α-Synuclein is closely associated with the development of Parkinson's disease. The function of this protein is not well understood. As already mentioned, α-synuclein drives fragmentation of mitochondria (Kamp et al., 2010; Nakamura et al., 2011; O'Donnell et al., 2014; Martinez et al., 2018). Uncontrolled mitochondrial fragmentation might contribute to the development of Parkinson's disease (Varkey et al., 2010; Panchal and Tiwari, 2019) and, importantly, duplication and triplication of SNCA (α-synuclein) gene cause a severe form of this disease (Olgiati et al., 2015; Konno et al., 2016). The authors of Kamp et al. (2010) and Nakamura et al. (2011) hypothesized that fragmentation of mitochondria might be independent of Drp1, whereas Martinez et al. (2018) reported that Drp1 is required for α-synuclein-mediated mitochondrial fragmentation. α-Synuclein might drive fragmentation of mitochondria by decreasing the rate of fusion or, alternatively, by promoting fission (Nakamura, 2013). Kamp et al. (2010) hypothesized that the influence of this protein on mitochondrial dynamics is based on inhibition of fusion, whereas Nakamura et al. (2011) suppose that, quite opposite, α-synuclein acts by promoting membrane fission. The latter hypothesis is supported by the fact that α-synuclein is able to convert large vesicles (liposomes) into smaller vesicles, i.e., mediate membrane fission (Varkey et al., 2010; Nakamura et al., 2011; Fakhree et al., 2019). AH of α-synuclein contributes to the generation of membrane curvature (Braun et al., 2014, 2017; Fakhree et al., 2019). Thus, it is reasonable to hypothesize that α-synuclein and, possibly, also β-synuclein belong to the large family of AH-containing fissioninducing proteins.

In summary, in most AH-containing fission-inducing proteins AHs were shown or suggested to play an important role in various fission events. As indicated, the AH penetration in the lipid bilayer can be the starting point of the fission process, followed by the recruitment of other proteins that mediate the final fission event (Sundborger et al., 2011; Meinecke et al., 2013). Shallow insertion of AHs into the membrane was suggested to drive membrane fission (Boucrot et al., 2012). However, few recent publications of Jeanne Stachowiak's group consider also an alternative mechanism foreseeing that the steric pressure generated by bulky disordered domains can drive membrane fission (Snead et al., 2017, 2019; Snead and Stachowiak, 2018), whereas the role of insertions, such as AHs, is to anchor proteins to the membrane (Snead et al., 2017). Whether the different mechanisms coexist or one prevails over the others remains to be defined.

# THE ROLE OF LIPID COFACTORS IN MEMBRANE FISSION DRIVEN BY AMPHIPATHIC HELIX-CONTAINING PROTEINS

## Lipid Cofactors: An Overview

Many AH-containing fission-inducing proteins require one or two specific lipids to generate membrane curvature and to complete the fission process (**Figure 3**). Most of these lipids are anionic. Such lipids could be named lipid cofactors (Ramachandran, 2018), or lipid factors (Danne et al., 2017b), or lipid ligands (Gopaldass et al., 2017). Examples of lipid cofactors of AH-containing fission proteins are: (1) PtdIns(4,5)P<sup>2</sup> for epsin (Kweon et al., 2006; Boucrot et al., 2012; Lai et al., 2012), mammalian amphiphysin 2 (Lee et al., 2002; Wu and Baumgart, 2014), Drosophila amphiphysin (Yoon et al., 2012), and mammalian endophilin A1 (Yoon et al., 2012); (2) cardiolipin and MMPE for PmtA (Danne et al., 2017b); (3) PtdIns3P and PtdIns(3,5)P<sup>2</sup> for Atg18 (Gopaldass et al., 2017; Scacioc et al., 2017); (4) cholesterol for M2 protein from influenza A virus (Ekanayake et al., 2016; Elkins et al., 2017, 2018; Herneisen et al., 2017; Pan et al., 2019; and references therein), see **Table 2**.

The lipid cofactor-binding sites were found in the AHs of few fission-inducing proteins: PtdIns(4,5)P2-binding site in epsin (Kweon et al., 2006; Lai et al., 2012), site that binds PtdIns(3,5)P<sup>2</sup> and PtdIns3P in Atg18 (Gopaldass et al., 2017), and cholesterol-binding site in M2 (Elkins et al., 2017, 2018). It is not surprizing that anionic lipid-binding sites contain basic residues: Arg7, Arg8, Lys11 of epsin belong to the PtdIns(4,5)P2 binding site, whereas Arg285 and Arg286 of Atg18 are involved in the interaction with PtdIns(3,5)P<sup>2</sup> and PtdIns3P. Accordingly, other fission-inducing proteins may contain few basic residues as anionic lipid cofactor-binding sites that are located close to each other in primary sequence.

In some membrane fission reactions involving AH-containing proteins, the role of specific lipids is not yet fully established. Possibly, cardiolipin is a cofactor for membrane fission mediated by Pex11 [peroxisome fission (Opalinski et al., 2011 ´ ; Su et al., 2018)]; bacterial MGS (Eriksson et al., 2009; Ariöz et al., 2013), DGS (Edman et al., 2003; Eriksson et al., 2009), and MurG (van den Brink-van der Laan et al., 2003; Lind et al., 2007) proteins; and by MinD [bacterial division (Zhou and Lutkenhaus, 2003; Renner and Weibel, 2012)]. PA might be a lipid cofactor for endophilin B1 (also known as Bif-1) (Zhang et al., 2011)

lipid composition. Of note, proteins are not able to induce membrane fission in the absence of these lipids (left panel; purple dashed line). Conversely, in the presence of such lipid cofactors, fission occurs through the following steps: membrane neck, hemifission and formation of two separate membranes (right panel; green dashed line). Lipid cofactors include: PtdIns(4,5)P2, PA, cardiolipin, MMPE, PtdIns3P, PtdIns(3,5)P<sup>2</sup> and cholesterol. See text for details.

TABLE 2 | Examples of lipid cofactors of fission-inducing proteins containing amphipathic helices.


The indicated lipids are specifically required for membrane fission reactions mediated by these proteins. Other lipids cannot replace these lipid cofactors to generate membrane curvature (i.e., tubulation and vesiculation). The topology of fission reaction for each protein is also shown.

that contains two AHs (Kjaerulff et al., 2011) and plays a role in membrane fission (Rostovtseva et al., 2009; Takahashi et al., 2011, 2016). Possibly, PtdIns(4,5)P<sup>2</sup> is a lipid cofactor in membrane vesiculation mediated by PICK1 (Pan et al., 2007; Karlsen et al., 2015). PS might be a lipid cofactor in the fission reaction mediated by ANKHD1 (Kitamata et al., 2019). Arf1 binds to PtdIns(4,5)P<sup>2</sup> and PA (Randazzo, 1997; Manifava et al., 2001; Krauss et al., 2008), but the role of these interactions in Arf1-mediated membrane fission is not completely clear.

As we already mentioned, an AH-containing protein α-synuclein induces fragmentation of mitochondria (Kamp et al., 2010; Nakamura et al., 2011; O'Donnell et al., 2014; Martinez et al., 2018) and converts large vesicles (liposomes) into smaller vesicles (Varkey et al., 2010; Nakamura et al., 2011; Fakhree et al., 2019). Moreover, AH plays a role in the α-synuclein-driven generation of membrane curvature (Braun et al., 2014, 2017; Fakhree et al., 2019). Hence, we can consider α-synuclein as an AH-containing fission-inducing protein. Interestingly, α-synuclein binds to cardiolipin (Nakamura et al., 2008; Robotta et al., 2014; Ryan et al., 2018) and PA (Perrin et al., 2000; Nakamura et al., 2011; Mizuno et al., 2017) with high affinity. Moreover, this protein prefers negatively charged lipids cardiolipin (Nakamura et al., 2008) and PA (Nakamura et al., 2008; Mizuno et al., 2017) to another negatively charged lipid PS. The cone shape of cardiolipin, whose tail is wider than its headgroup (see below), might contribute to the affinity of α-synuclein to cardiolipin-containing bilayers (Ghio et al., 2019; and references therein). PA (Mizuno et al., 2017) and cardiolipin (Ryan et al., 2018) promote the formation of the AH of α-synuclein. Importantly, α-synuclein mediates fission of liposomes that contain both PC and cardiolipin, but not liposomes that contain only PC (Nakamura et al., 2011). We propose that cardiolipin and, possibly, also PA are cofactors in membrane fission mediated by AH-containing protein α-synuclein.

To the best of our knowledge, membrane fission mediated by small GTPase Sar1 does not need any lipid cofactors. It has been stressed that Arf1 and Sar1 use the energy of GTP hydrolysis, whereas no enzymatic activity has been found or predicted for Atg18 (Gopaldass et al., 2017). However, Atg18 membrane recruitment and oligomerization needs PtdIns3P and PtdIns(3,5)P<sup>2</sup> as lipid cofactors (see **Table 2**), and synthesis of these lipids is energy-consuming. Hence, this fission process can be energized indirectly, via the energy invested into the synthesis of lipid cofactors (Gopaldass et al., 2017). Possibly, such indirect driving force of the reaction can also be considered for other AH-containing fission proteins.

Anionic lipid cofactors cannot be replaced with other anionic lipids: see **Table 2**. Tubulation mediated by endophilin A1 requires PtdIns(4,5)P<sup>2</sup> but not PS (Yoon et al., 2012). PmtA tubulates liposomes containing cardiolipin, but is less efficient in tubulating liposomes containing PG, PtdIns4P or PA (Danne et al., 2017b). The AH-containing fission proteins epsin (Boucrot et al., 2012), PmtA (Danne et al., 2017b), Atg18 (Gopaldass et al., 2017) mediate fission of membranes containing their lipid cofactors, but they are inefficient with membranes containing other acidic lipids.

Electrostatic interaction between positively charged residues belonging to AH and negatively charged phospholipids was proposed to overcome the energetic cost of spreading lipids apart and thus to play an important role in the generation of membrane curvature needed for the fission event (Khattree et al., 2013; Roberts et al., 2013; Brady et al., 2015). Most probably, anionic lipid cofactors are required for the recruitment of AHcontaining fission proteins. Furthermore, as pointed out above, the shallow penetration of some AHs leads to the generation of local membrane curvature through the pushing of the lipid headgroups apart (the "wedge" mechanism) (McMahon and Gallop, 2005; Campelo et al., 2008; Drin and Antonny, 2010; Baumgart et al., 2011; Lai et al., 2012; McMahon and Boucrot, 2015; and references therein). At least in some cases, the depth of the AH insertion into the membrane depends on the presence of negatively charged lipids in the membrane (Sani et al., 2012; Palomares-Jerez et al., 2013). Indeed, the interaction of curvature-inducing AHs with anionic lipids might contribute to the optimal depth of AH insertion.

As pointed out above, the presence of anionic lipids in the membrane often promotes the folding of AH (Davidson et al., 1998; Kweon et al., 2006; Fernandes et al., 2008; Horchani et al., 2014; Brady et al., 2015; Mizuno et al., 2017; Ryan et al., 2018). Thus, another mode for the anionic lipid to mediate membrane curvature generation and promote membrane fission is facilitating the folding of AH motif, once the protein segment comes to close contact with the membrane.

Simple electrostatic attraction between acidic lipids and basic residues of AHs does not explain the specificity of the action of lipid cofactors on fission proteins. AH-containing fission proteins epsin (Kweon et al., 2006), endophilin A1 (Yoon et al., 2012), amphiphysin 2 (Lee et al., 2002), PmtA (Danne et al., 2017b), Atg18 (Gopaldass et al., 2017) interact specifically with the membranes containing their lipid cofactors, but poorly with the membranes containing other anionic lipids. For example, PmtA binds cardiolipin-containing liposomes, but poorly binds PA-containing liposomes (Danne et al., 2017b). Most probably, a specific interaction of fission proteins with their lipid cofactors is necessary for the correct conformation and oligomeric structure of these proteins.

Reverse-topology membrane fission-inducing protein M2 from influenza A virus contains an AH, but uses neutral cholesterol as a cofactor in the membrane fission process (Elkins et al., 2017; Herneisen et al., 2017; Pan et al., 2019; and references therein), see **Table 2**. However, we should keep in mind that in the case of reverse topology fission, AHs insert into the proximal leaflet of the budding membrane-enclosed compartment, and not into the distal leaflet, as in the case of the normal-topology fission. Hence, the process of membrane remodeling is fundamentally different, and the requirements for the lipid cofactor might also be different.

Cholesterol may control the insertion depth of the M2 AH in the bilayer (Kim et al., 2015; Herneisen et al., 2017), thus, contributing to the optimal depth of the AH insertion, as we have discussed above for the interaction of the AHs of other fission-inducing proteins with their anionic lipid cofactors. Moreover, cholesterol was suggested to promote a more compact

conformation of the M2 AH (Herneisen et al., 2017). This protein was shown also to cluster at the phase boundary of cholesterolrich liquid-ordered phase and cholesterol-poor liquid-disordered phase (Roberts et al., 2013). Such a localization of a fissioninducing protein at the neck of the budding virus can generate the membrane curvature required for membrane fission (Elkins et al., 2017; Martyna et al., 2017). A similar mechanism might be applicable for few other membrane fission processes mediated by AH-containing proteins interacting with specific lipid cofactors.

Some fission-inducing proteins, in which no definitely established AHs have yet been identified, also require lipid cofactors for their activity. For example, PA is a lipid cofactor for CtBP1-S/BARS (Weigert et al., 1999; Pagliuso et al., 2016); PS is a lipid cofactor for Ebola virus VP40 (Soni and Stahelin, 2014; Adu-Gyamfi et al., 2015; Del Vecchio et al., 2018); PtdIns(4,5)P<sup>2</sup> is a putative lipid cofactor for dynamin (Jost et al., 1998); cardiolipin is a probable lipid cofactor for FisB (Doan et al., 2013) and, moreover, activates mechanoenzyme Drp1 that plays a role in mitochondrial fission (Francy et al., 2017).

Lipid cofactors should not be confused with lipids that just recruit fission-inducing proteins to the membrane, but are not required for the fission reaction per se. Unlike lipids that just recruit proteins, lipid cofactors are indispensable for fission: membrane curvature generation cannot proceed efficiently without them, see **Table 2**. For example, PtdIns4P was suggested to have a direct role in the anchoring of the protein complex-bound CtBP1-S/BARS (Valente et al., 2013). However, only PA, but not PtdIns4P, allowed CtBP1-S/BARS to induce liposome tubulation (Yang et al., 2008). Hence, PA, but not PtdIns4P, can be considered a lipid cofactor in CtBP1-S/BARSmediated fission. However, often the information concerning particular lipid is incomplete. In this case, it is difficult to figure out if this lipid a true cofactor or protein-recruiting lipid.

Interestingly, two fission-inducing proteins, A. tumefaciens PmtA and mammalian CtBP1-S/BARS, are known to play a role in the synthesis of their own lipid cofactors. The difference between PmtA and CtBP1-S/BARS is that PmtA is itself an enzyme that produces its own lipid cofactor MMPE (Danne et al., 2017b), whereas CtBP1-S/BARS binds to and activates an enzyme, LPAATδ that catalyzes the synthesis of PA required for CtBP1-S/BARS-mediated membrane fission (Pagliuso et al., 2016; Zhukovsky et al., 2019b). Such production of lipid cofactors promoted by the very fission-inducing proteins and then using them guarantees availability of these lipids where and when they are needed.

### Cardiolipin, PA, and PtdIns(4,5)P<sup>2</sup> Act as Lipid Cofactors in Several Membrane Fission Reactions

We noted that the three negatively charged lipids PtdIns(4,5)P2, cardiolipin, and PA might play a role of lipid cofactors in several fission reactions mediated by different proteins. Cardiolipin is an unusual phospholipid that consists of two PA molecules connected with a glycerol backbone and thus contains two phosphates and four fatty acid tails. Cardiolipin is present in the membranes of most bacteria, whereas in mammalian and plant cells it is found mostly in the IMM and, to a lesser extent, in the OMM and in peroxisomes (Wriessnegger et al., 2007; Musatov and Sedlak, 2017; Pagliuso et al., 2018). PA containing a unique phosphomonoester headgroup is a minor phospholipid present in various organelles (Zhukovsky et al., 2019a). PtdIns(4,5)P<sup>2</sup> is enriched in the PM of eukaryotic cells (Olivença et al., 2018; Dickson and Hille, 2019).

The role that cardiolipin and PA play in membrane fission reactions might be determined by their structure, namely combination of charge and shape (Danne et al., 2017b; Agrawal and Ramachandran, 2019; Zhukovsky et al., 2019a). Packing defects are locations on the membrane surface where the hydrophobic interior of the membrane is exposed to the solvent. AHs have a tendency to bind such defects induced by coneshaped lipids, i.e., lipids whose tails are wider than their headgroups (Ouberai et al., 2013; Vanni et al., 2013; Lauwers et al., 2016). At physiological conditions, cardiolipin carries a charge −2 (Arnarez et al., 2016; Sathappa and Alder, 2016; Boyd et al., 2018; and references therein) and, moreover, has a cone shape (Arnarez et al., 2016; Carranza et al., 2017; Basu Ball et al., 2018; Su et al., 2018; and references therein). PA also has a cone shape (Kooijman et al., 2005). According to the electrostatic/hydrogen bond switch mechanism, the PA charge can change from −1 to −2 upon interaction with Lys or Arg (Kooijman et al., 2007). Hence, cone-shaped lipid PA, as cone-shaped lipid cardiolipin, is able to have charge −2. Combination of double negative charge and cone shape of cardiolipin and PA allows positively charged protein segments, such as AHs with basic residues on the hydrophilic face, to interact with the membranes containing these lipids. The combination of net charge and cone shape of cardiolipin seems to be responsible for the membraneremodeling activity of PmtA (Danne et al., 2017b).

PtdIns(4,5)P2, like cardiolipin and PA, is a lipid cofactor in membrane fission reactions mediated by various proteins, such as epsin (Kweon et al., 2006; Boucrot et al., 2012; Lai et al., 2012), mammalian amphiphysin 2 (Lee et al., 2002; Wu and Baumgart, 2014), Drosophila amphiphysin (Yoon et al., 2012) and mammalian endophilin A1 (Yoon et al., 2012). Like cardiolipin and PA, PtdIns(4,5)P<sup>2</sup> possesses a high negative charge. The charge of PtdIns(4,5)P<sup>2</sup> depends on various factors, but at the physiological conditions, the charge of this lipid is approximately −4 (McLaughlin et al., 2002; Heo et al., 2006; Kooijman et al., 2009; and references therein). However, unlike cardiolipin and PA that have a cone shape, PtdIns(4,5)P<sup>2</sup> and other phosphoinositides have an inverted-cone shape, i.e., their headgroups are wider than their tails (Martin, 2012; Suetsugu et al., 2014; McMahon and Boucrot, 2015; and references therein). Besides PtdIns(4,5)P2, other phosphoinositides also function as lipid cofactors in membrane fission processes, e.g., PtdIns3P and PtdIns(3,5)P<sup>2</sup> for Atg18 (Gopaldass et al., 2017). We hypothesize that, despite of their shape, PtdIns(4,5)P<sup>2</sup> and PtdIns(3,5)P<sup>2</sup> are able to be lipid cofactors required for membrane fission reactions because of extremely high negative charge that recruits basic residues of the AHs of fission-inducing proteins to PBP-containing membranes due to the electrostatic attractions. If a PBP-containing membrane contains also coneshaped lipids, membrane packing defects are present in such

membrane, and some of these defects are present in the vicinity of PBP molecules. If a membrane contains cone-shaped lipids, membrane packing defects should be evenly distributed on the membrane surface (Vamparys et al., 2013). Co-localization of some of these defects with highly charged PBPs creates suitable condition for the recruitment of AHs belonging to the fission-inducing proteins.

Alternatively, we hypothesize that, while AH-containing fission-inducing proteins interacting with cardiolipin and PA play a major role in fission, AH-containing proteins that bind PtdIns(4,5)P<sup>2</sup> often mediate membrane fission cooperatively with the proteins belonging to the dynamin superfamily (Boucrot et al., 2012, 2015; Meinecke et al., 2013). Possibly, interaction with cone-shaped lipids, such as cardiolipin and PA, is required for AH-containing proteins that mediate fission in the absence of proteins belonging to the dynamin superfamily, but is not required for AH-containing proteins that mediate fission cooperatively with the proteins belonging to this superfamily.

Of note, PS, phosphatidylinositol (PtdIns) and PG that possess net charge −1 at neutral pH (Li et al., 2015) are less frequently used as lipid cofactors in membrane fission processes, compared with PA, cardiolipin and PtdIns(4,5)P<sup>2</sup> that possess higher negative charge.

# Putative Sequence Motifs That Specifically Recognize Cardiolipin, PA, and PtdIns(4,5)P<sup>2</sup>

It is reasonable to hypothesize that AHs of proteins interacting specifically with each of three lipids cardiolipin, PA, PtdIns(4,5)P2, might contain sequence motifs that recognize these lipids. Primary sequences of the AHs of fission-inducing proteins that were shown or suggested to use cardiolipin and PtdIns(4,5)P<sup>2</sup> as cofactors are presented in **Tables 3**, **4**. We found that (K/R)x6(F/Y) motif is present in all but one AHs of cardiolipin-binding fission-inducing proteins we know, whereas (K/R/H)(K/R/H)xx(K/R/H) motif is present in most AHs of PtdIns(4,5)P2-binding fission-inducing proteins we know. Here K is Lys, R is Arg, F is Phe, Y is Tyr, H is His, and x is any residue. Although it is quite possible that the (K/R)x6(F/Y) and (K/R/H)(K/R/H)xx(K/R/H) motifs in some of AHs of fissioninducing proteins are not involved in the specific interaction with lipids, we suppose that in most cases these two motifs are functionally essential.

#### Cardiolipin

Among proteins that are involved in membrane fission and are listed in the **Tables 3**, **4**, the (K/R)x6(F/Y) motif is present in AHs of six out of seven cardiolipin-binding proteins, but absent in AHs of all seven PtdIns(4,5)P2-binding proteins. In general, (K/R)x6(F/Y) motifs in the AHs of various proteins are rare. Among 16 AHs shown in the Table 1 of Drin and Antonny (2010), this motif is present only in α-synuclein, a protein that binds cardiolipin (Nakamura et al., 2008; Robotta et al., 2014; Ryan et al., 2018). However, except proteins listed in the **Table 3**, this motif is present in the AHs of at least two proteins that interact with cardiolipin but, to the best of our knowledge, are not known to be involved in membrane fission: PmtA from Bradyrhizobium japonicum (Danne et al., 2017a) and ATG3 (Nath et al., 2014; Hervás et al., 2017). Arg14 and Phe21 belong to



Sequences are retrieved from indicated references. Residues belonging to the (K/R)x6(F/Y) motif are shown in red. Gly are underlined.

TABLE 4 | Amphipathic helices of fission-inducing proteins that use PtdIns(4,5)P<sup>2</sup> as a lipid cofactor.


Sequences are retrieved from indicated references. Residues belonging to the (K/R/H)(K/R/H)xx(K/R/H) motif are shown in red.

the (K/R)x6(F/Y) motif in the AH of PmtA from B. japonicum (Danne et al., 2017a), whereas Lys11 and Tyr18 belong to the (K/R)x6(F/Y) motif in the AH of Atg3 (Hervás et al., 2017). In the AH of cardiolipin-binding fission-inducing protein PmtA from A. tumefaciens (Danne et al., 2017b), both Lys12 and Phe19 belonging to the Kx6F motif are critical for membrane binding (Danne et al., 2015). Based on these observations, we make a conclusion that, most probably, many (K/R)x6(F/Y) pairs in the AHs of cardiolipin-binding proteins are functional cardiolipinbinding motifs.

In the protein α-helix, each amino acid residue corresponds to a 100◦ turn in the helix. (K/R)x6(F/Y) motif contains one basic residue (Arg or Lys) and one aromatic residue (Phe or Tyr) spaced roughly 700◦ in the circle, very close to 720◦ (two turns). Hence, basic residue and aromatic residue of the (K/R)x6(F/Y) motif are on the same face of the helix, two turns apart, see **Figure 4**. Such localization might be suitable for two residues to interact with the same cardiolipin molecule.

Kx6F motifs in MGS and MinD contain positively charged Lys followed by Gly. Interestingly, a positively charged residue preceding flexibility-conferring Gly may be considered a footprint for cardiolipin binding (Planas-Iglesias et al., 2015).

In (K/R)x6(F/Y) motifs within AHs of cardiolipin-binding proteins, Gly is often present close to basic (Lys or Arg) and aromatic (Phe or Tyr) residues, see **Table 3**. Hence, in most (K/R)x6(F/Y) motifs, three residues are present: positively charged residue (Lys or Arg), aromatic residue (Phe or Tyr), and Gly. Such a combination of these three kinds of

the (K/R)x6(F/Y) motifs are highlighted. Color coding for residues: yellow for hydrophobic, purple for Ser (S) and Thr (T), blue for Lys (K) and Arg (R), red for acidic, pink for Asn (N) and Gln (Q), gray for small residues (Ala, A and Gly, G), green for Pro (P), and light blue for His (H). The arrow in helical wheels corresponds to the hydrophobic moment. Species names and UniProt accession numbers: Acholeplasma laidlawii MGS, Q93P60; Escherichia coli MinD, P0AEZ3; A. laidlawii DGS, Q8KQL6; Penicillium chrysogenum Pex11p, B6GZG8; Agrobacterium tumefaciens AtPmtA, A0A2L2L7Q9.

residues is somewhat similar to the cardiolipin-binding motif (Y/W/F)(K/R)G (Ruprecht et al., 2014; Kunji et al., 2016; Duncan et al., 2018), where W is Trp and G is Gly. We suppose that, as suggested for the (Y/W/F)(K/R)G motif, aromatic residues (Phe or Tyr) in (K/R)x6(F/Y) motif might be involved in the interaction with acyl chain (Ruprecht et al., 2014), whereas neighboring Gly might be involved in the interaction with phosphate groups (Kunji et al., 2016; Duncan et al., 2018).

#### PtdIns(4,5)P<sup>2</sup>

Lys, Arg, and His are basic amino acid residues. Hence, all three residues belonging to the putative PtdIns(4,5)P2 binding (K/R/H)(K/R/H)xx(K/R/H) motif are positively charged. This motif is present in five out of seven AHs of fissioninducing PtdIns(4,5)P2-binding proteins we know, but in AHs of only two out of seven cardiolipin-binding fission-inducing proteins we know, see **Tables 3**, **4**. AH of Arf1 does not contain (K/R/H)(K/R/H)xx(K/R/H) motif: see **Table 4**. However, if we consider Arg19 adjacent to this AH, we find that Arf1 contains a K15K <sup>16</sup>xxR<sup>19</sup> sequence that conforms to the (K/R/H)(K/R/H)xx(K/R/H) motif. Double mutation of two Lys belonging to this K15K <sup>16</sup>xxR<sup>19</sup> motif to Leu affected Arf1 binding to PtdIns(4,5)P<sup>2</sup> but not to PA (Randazzo, 1997). Moreover, (K/R/H)(K/R/H)xx(K/R/H) motifs are present in the AHs of at least three PtdIns(4,5)P2-binding proteins that, to the best of our knowledge, are not known to be involved in membrane fission: Spo20 (Horchani et al., 2014), STIM2 (Bhardwaj et al., 2013), and SH3YL1 (Hasegawa et al., 2011). His75, Lys76, and His79 belong to the (K/R/H)(K/R/H)xx(K/R/H) motif in the AH of Spo20 (Horchani et al., 2014); K732K <sup>733</sup>PSK<sup>736</sup> and K <sup>742</sup>K <sup>743</sup>KSK<sup>746</sup> are two (K/R/H)(K/R/H)xx(K/R/H) motifs in the AH of STIM2 (Bhardwaj et al., 2013), whereas Lys14, Lys15, and Lys18 belong to the (K/R/H)(K/R/H)xx(K/R/H) motif in the AH of SH3YL1 (Hasegawa et al., 2011). Residues belonging to the (K/R/H)(K/R/H)xx(K/R/H) motif within AHs play a role in the interaction with PtdIns(4,5)P2: Arg7, Arg8, Lys11 in epsin 1 (Ford et al., 2002; Lai et al., 2012) and Lys14, Lys15, Lys18 in SH3YL1 (Hasegawa et al., 2011). We make a conclusion that, most probably, many (K/R/H)(K/R/H)xx(K/R/H) motifs are functional PtdIns(4,5)P2-binding motifs in the AHs of various proteins that interact with this lipid.

Taking into account that each residue corresponds to a 100◦ turn in the protein α-helix, all three basic residues belonging to the (K/R/H)(K/R/H)xx(K/R/H) motif are located on the same face of the helix: see **Figure 5**. Such localization might be favorable for the interaction of these residues with the same PtdIns(4,5)P<sup>2</sup> molecule. Possibly, in the (K/R/H)(K/R/H)xx(K/R/H) motifs of various proteins, two adjacent positively charged residues interact with 1- and 4-phosphate groups of PtdIns(4,5)P2, similar to the two Arg belonging to the R7R 8 xxK<sup>11</sup> motif in epsin 1 (Lai et al., 2012).

According to Nishimura et al. (2018), the BAR domain-containing proteins do not have any specific phosphoinositide-binding pockets. However, we hypothesize that (K/R/H)(K/R/H)xx(K/R/H) motif is a specific PtdIns(4,5)P2 binding motif in AHs (and, perhaps, other segments) of various proteins, including BAR domain-containing proteins. We predict that residues Lys7, Lys8, His11 in mammalian endophilin A1, residues Lys20, Lys21, Arg24 in mammalian amphiphysin 2, residues Lys15, His16, Arg19 in Drosophila amphiphysin are involved in the interaction with PtdIns(4,5)P2. We expect that mutations involving these residues will inhibit such interaction as well as membrane curvature generation and membrane fission mediated by these proteins.

#### Phosphatidic Acid

At least eight AHs of PA-binding proteins contain KxK (Lys – any residue – Lys) motifs, see **Table 5**. One of these proteins, α-synuclein, might mediate membrane fission (see above), whereas other seven proteins, Atg3 (Nath et al., 2014; Hervás et al., 2017), CDeT11-24 (Petersen et al., 2012), Opi1 (Loewen et al., 2004; Hofbauer et al., 2018), Spo20 (Horchani et al., 2014), DHN1 (Koag et al., 2003, 2009), CCT also known as Pcyt1a (Cornell, 2016), and Tam41 (Jiao et al., 2019), are not known to be involved in fission (to the best of our knowledge). In Atg3, residues Lys9 and Lys11, belonging to the KxK motif, are essential for membrane interaction (Hervás et al., 2017). In Opi1, mutation of five Lys (including Lys119 and Lys121 belonging to the KxK motif) to Arg leads to the loss of the preference for PA-containing over PS-containing liposomes (Hofbauer et al., 2018).

Amphipathic helices of few PA-binding proteins, including fission-inducing proteins endophilin B1 (Rostovtseva et al., 2009; Kjaerulff et al., 2011; Zhang et al., 2011) and Arf1 (Krauss et al., 2008; Beck et al., 2011; Martyna et al., 2017) do not contain KxK motifs. Moreover, the authors of Putta et al. (2016), Kassas et al. (2017), Zegarlinska et al. (2018) ´ , and Tanguy et al. (2019) stress that there is no known well-defined PA-binding motif or domain. However, here we hypothesize that KxK motif in AHs is involved in specific interaction with PA. It is possible that other sequence motifs recognizing PA will be discovered in the future.

If KxK motif is present in the protein α-helix (each amino acid residue corresponds to a 100◦ turn in the helix), two Lys belonging to the KxK motif are on two opposite faces of the helix, see **Figure 6**. Hence, in the AH, KxK motif usually places two Lys at the polar/non-polar interface. Such a position of basic residues is typical for AHs that interact with membranes (Mishra and Palgunachari, 1996; Davidson et al., 1998; Polozov et al., 1998; Mozsolits et al., 2004), especially with membranes that contain acidic phospholipids (Polozov et al., 1998). Most Lys residues in the AH of α-synuclein belong to the KxK motifs, and, interestingly, Perrin et al. (2000) stress that such localization of Lys along the polar/non-polar interface might be important for α-synuclein interaction with PA-containing membranes. At first sight, such localization of two Lys, at the opposite sides of an α-helix, is not optimal for the interaction with the same lipid molecule. However, Hofbauer et al. (2018) underline that KRK motif in Opi1 interacts specifically with PA. Moreover, this motif forms a "three-finger grip" that tightly binds the PA phosphate headgroup but fails to accommodate the larger PS headgroup (Hofbauer et al., 2018). We suppose that, similarly, KxK motif in the AH of other proteins might form a "two-finger grip" that binds specifically the PA phosphate headgroup, although with somewhat lower affinity compared with KRK motif, if the residue between two Lys is not positively charged.

As we pointed out, the penetration of AHs into ca. 40% of monolayer thickness is optimal for generation of membrane curvature and driving membrane fission (Campelo et al., 2008; Zemel et al., 2008; Boucrot et al., 2012; Hanna et al., 2016). We hypothesize that the specific interaction of lipid cofactor with sequence motif ensures optimal depth of the insertion of AH into lipid bilayer. Insertion of AH approximately at the level of membrane hydrophilic-hydrophobic interface is optimal for generating membrane curvature (Li, 2018). The KxK motifs within PA-associated AH places two Lys close to this interface (see **Figure 6**) and, hence, is able to ensure shallow insertion of this AH into the membrane and generation of the high membrane curvature, required for driving membrane fission. Moreover, in few AHs of cardiolipin-binding proteins, such as AHs from MGS (Eriksson et al., 2009), AtPmtA (Danne et al., 2017b), Pex11p (Opalinski et al., 2011 ´ ), the orientation of the (K/R)x6(F/Y) motif ensures the position of these two residues (basic Lys or Arg and aromatic Phe or Tyr) close to the hydrophilichydrophobic interface, so that basic residue is somewhat closer to the hydrophobic face of the helix, and aromatic residue is somewhat closer to the hydrophilic face of the helix, see **Figure 4**. We suggest that such orientation also ensures shallow insertion of the AH, that is optimal for driving fission.

Interestingly, α-synuclein that might use cardiolipin and, possibly, also PA as lipid cofactors in fission reaction (see above) contains, in its AH, putative cardiolipin-binding Kx6Y motif


Only one protein in this Table (α-synuclein) was suggested to mediate membrane fission. Sequences are retrieved from indicated references. Residues belonging to the KxK motif are shown in red.

as well as five copies of putative PA-binding KxK motifs, see **Tables 3**, **5**. Moreover, Atg3 that is not known to be involved in any membrane fission process, interacts with both cardiolipin and PA (Nath et al., 2014; Hervás et al., 2017), and contains, in its AH, putative cardiolipin-binding K11x6Y <sup>18</sup> motif overlapping with the putative PA-binding K<sup>9</sup> xK<sup>11</sup> motif.

We predict that mutations of residues belonging to the KxK motifs in AHs of PA-binding proteins as well as (K/R)x6(F/Y) motifs in AHs of cardiolipin-binding proteins will inhibit the binding of these proteins to the membranes containing these lipids. For proteins that possess an ability to generate membrane curvature or to mediate membrane fission, these processes will also be inhibitied in the mutants.

Interestingly, Lys in cardiolipin-binding and PA-binding AHs are much more frequent than Arg, see **Tables 3**, **5**. Mutation of five Lys to Arg in Opi1 leads to the loss of the preference for PAcontaining over PS-containing liposomes (Hofbauer et al., 2018). Due to its longer aliphatic side chain, Lys penetrate more deeply into lipid bilayer, and thus interaction of lipids with Lys might be more stable than interaction of lipids with Arg (Davidson et al., 1998). Furthermore, the increase in PA charge induced by Lys is higher than that induced by Arg (Kooijman et al., 2007). For these reasons, the presence of Lys in sequence motifs that recognize cardiolipin and PA is preferable to the presence of Arg.

Amphipathic helices of some of the fission-inducing proteins that use cardiolipin, PtdIns(4,5)P<sup>2</sup> and PA as cofactors do not contain, correspondingly, (K/R)x6(F/Y) motif, (K/R/H)(K/R/H)xx(K/R/H) motif, and KxK motif. Other residues yet to be identified are therefore involved in the interaction with these lipids.

## Characteristics of Lipid Cofactors Involved in Membrane Fission Processes in Different Organelles

We noted that in many membrane fission processes mediated by AH-containing proteins in a particular organelle or bacterium, the most abundant lipids in this organelle (or bacterium) possessing high negative charge (more negative than −1) play a role of lipid cofactors. Moreover, fission-inducing proteins often contain in their AHs sequence motifs that recognize these lipids. The following highly charged lipids are over-represented in specific organelles and hence might play the role of lipid cofactors with various fission-inducing proteins: cardiolipin in bacteria, mitochondria (Musatov and Sedlak, 2017; Pagliuso et al., 2018) and peroxisomes (Wriessnegger et al., 2007); PtdIns3P and PtdIns(3,4)P<sup>2</sup> in the endosomes; PtdIns4P in the Golgi apparatus; PtdIns5P in the nuclear envelope; PtdIns(4,5)P<sup>2</sup> in the PM; PtdIns(3,5)P<sup>2</sup> in endosomes, lysosomes and vacuoles (Takatori et al., 2016; Olivença et al., 2018; Dickson and Hille, 2019).

Knowing the primary sequence of the AH of a fissioninducing proteins allows to anticipate which lipid is used by this protein as a cofactor in fission reactions, and which organelle this protein functions in, see **Table 6**. For example, eukaryotic fission protein containing (K/R)x6(F/Y) motif in its AH might be expected to use cardiolipin as cofactor and, hence, to mediate fission of mitochondria or peroxisomes. On the other hand, a fission-inducing protein that does not contain (K/R)x6(F/Y) motif but contains a (K/R/H)(K/R/H)xx(K/R/H) motif in its AH might use PtdIns(4,5)P<sup>2</sup> as cofactor and, hence, might be involved in endocytosis.

As pointed out above, we hypothesize that the most abundant organelle-specific lipid possessing high negative charge might play a role of a lipid cofactor in membrane fission processes.

We have already mentioned AH-containing proteins (such as α-synuclein, Pex11, PmtA, MGS, DGS, epsin, endophilin A1) that mediate fission in different organelles and follow this rule, see **Table 6**. Most probably, a similar pattern also applies to fission-inducing proteins in general. For example, cardiolipin is a putative lipid cofactor for FisB that mediates fission during sporulation in a bacterium B. subtilis (Doan et al., 2013). FisB is not known to contain any AH. Moreover, cardiolipin activates the mechanoenzyme Drp1 that plays an important role in mitochondrial fission (Francy et al., 2017). Besides being cofactor for fission mediated by AH-containing proteins such as epsin (Kweon et al., 2006; Boucrot et al., 2012; Lai et al., 2012) and endophilin A1 (Yoon et al., 2012), PtdIns(4,5)P<sup>2</sup> is a putative cofactor for the mechanoenzyme dynamin (Jost et al., 1998) that is responsible for endocytosis, i.e., fission at PM. PtdIns4P, whose charge is between −2 and −3 at physiological pH (van Paridon et al., 1986), was suggested to play a direct role in membrane fission of Golgi membranes mediated by

Schizosaccharomyces pombe Tam41, O74339.

TABLE 6 | Typical lipid cofactors that are used by fission-inducing proteins in different organelles and in bacterial cells, and putative sequence motifs within amphipathic helices of these proteins that specifically bind their lipid cofactors.


CtBP1-S/BARS (Yang et al., 2008; Valente et al., 2013). Moreover, besides being cofactor for vacuole fission mediated by AHcontaining fission-inducing protein Atg18 (Gopaldass et al., 2017), PtdIns(3,5)P<sup>2</sup> plays an important role in various fission processes in vacuoles and endo-lysosomes (Dove et al., 2009; Miner et al., 2019; and references therein). We are not aware of any role that PtdIns5P plays in any fission process. However, this phenomenon might be expected in an organelle such as the nuclear envelope, in which the content of this lipid is high, compared with other highly charged lipids. This type of reasoning can be applied to other phosphoinositides enriched in different organelles.

### OLIGOMERIZATION OF AMPHIPATHIC HELIX-CONTAINING FISSION-INDUCING PROTEINS

An oligomeric structure is required for membrane fission and membrane curvature generation mediated by various AHcontaining fission-inducing proteins. For example, Arf1 (Beck et al., 2008, 2011), Sar1 (Hariri et al., 2014), epsin 1 (Lai et al., 2012), endophilin A1 (Gallop et al., 2006; Mim et al., 2012; Yoon et al., 2012; Capraro et al., 2013), endophilin A2 (Gortat et al., 2012), Drosophila amphiphysin (Peter et al., 2004; Yoon et al., 2012), ANKHD1 (Kitamata et al., 2019), GDAP1 (Huber et al., 2016) form dimers. Endophilin B1 (Rostovtseva et al., 2009), PICK1 (Karlsen et al., 2015) and influenza virus M2 (Rossman et al., 2010) form tetramers (dimers of dimers). In some cases, dimers may direct the formation of higher order oligomers, such as, for example, in the case of epsin 1 (Yoon et al., 2010; Lai et al., 2012) and endophilin A1 (Gallop et al., 2006; Yoon et al., 2012). Moreover, dimers of Sar1 (Hariri et al., 2014), epsin 1 (Lai et al., 2012), and endophilin A1 (Mim et al., 2012; Capraro et al., 2013) were reported to form ordered lattices. Human Pex11B (Yoshida et al., 2015), fungal Pex11p (Su et al., 2018), Snf7 (Saksena et al., 2009), caveolin-1 (Fernandez et al., 2002), and α-synuclein (Nakamura et al., 2011) form oligomers. The protein mutations that disrupt their oligomerization inhibit the ability of Arf1 (Beck et al., 2008, 2011), epsin 1 (Yoon et al., 2010), and endophilin A2 (Gortat et al., 2012) to deform or vesiculate membranes.

Sometimes, lipid cofactors required for membrane fission events promote oligomerization of AH-containing fissioninducing proteins. Thus, PtdIns(3,4)P<sup>2</sup> induces Atg18 oligomerization (Gopaldass et al., 2017), whereas cholesterol prevents disruption of influenza virus M2 tetramers (Ekanayake et al., 2016) and shifts these tetramers toward a more compact conformation (Herneisen et al., 2017). In general, lipid cofactors such as PtdIns(4,5)P<sup>2</sup> for endophilin A1 and Drosophila amphiphysin (Yoon et al., 2012), can promote membrane penetration of the AHs and thus induce protein oligomerization upon membrane binding.

Formation of the oligomers of some AH-containing fissioninducing proteins, such as epsin 1 (Yoon et al., 2010), endophilin A1 (Mim et al., 2012), peroxins Pex11B (Yoshida et al., 2015) and Pex11p (Su et al., 2018), influenza virus M2 (Ekanayake et al., 2016), is driven, at least partially, by direct interaction of their AHs. Membrane curvature generated by the insertion of AH of fission-inducing protein into the membrane is amplified upon protein oligomerization due to the concerted penetration of multiple AHs, ultimately causing membrane tubulation or fission (Campelo et al., 2008; Yoon et al., 2010; Lai et al., 2012; Hariri et al., 2014; Pan et al., 2019).

Interestingly, oligomerization of endophilin B1 is induced by chaperone activity of the apoptosis regulator Bax (Rostovtseva et al., 2009). Bax does not stably associate with endophilin B1 oligomers. Despite of this, even when the concentration of Bax is 10 times lower than the concentration of endophilin B1, Bax induces endophilin B1 oligomerization, a structure required for membrane fission. As result, endophilin B1 with Bax induces massive vesiculation of liposomes, whereas endophilin B1 alone and Bax alone do not have such an effect (Rostovtseva et al., 2009). Interestingly, in normal cells, endophilin B1 and Bax do not interact. However, during apoptosis they bind transiently and, due to the membrane rearrangements mediated by oligomeric endophilin B1, they promote release of cytochrome C from mitochondria (Rostovtseva et al., 2009).

### DISEASES ASSOCIATED WITH MUTATIONS IN THE AMPHIPATHIC HELICES OF FISSION-INDUCING PROTEINS

The role that AHs play in the function of fission-inducing proteins is emphasized by the fact that few naturally occurring mutations of residues belonging to the AHs are associated with human diseases. Some of these mutants are defective in inducing membrane curvature, whereas other mutants generate membrane curvature more vigorously than wt proteins. Both kinds of mutations might contribute to the development of diseases.

Amphiphysin 2 (also known as BIN1) contains an AH (Nicot et al., 2007; Böhm et al., 2014; Hohendahl et al., 2016) and induces membrane vesiculation (Wu and Baumgart, 2014). Deletion mutation 1Lys21 and substitution mutations Arg24Cys and Lys35Asn (Nicot et al., 2007; Böhm et al., 2014) within the AH Val18-Ala36 (Hohendahl et al., 2016) of amphiphysin 2 cause CNM. Unlike wt amphiphysin 2, none of these three mutants induce membrane tubulation (Nicot et al., 2007; Böhm et al., 2014). Nicot et al. (2007) and Hohendahl et al. (2016) underlined that these mutations change the charge of the hydrophilic face of the AH, and such change might disrupt the interaction with negatively charged lipids of the membrane and, in turn, may lead to a defect in the ability to generate membrane curvature.

Expression of GDAP1 induces fission of peroxisomes and mitochondria, and this fission is critically dependent on the AH Val292-Phe309 (Huber et al., 2016). The Asn297Lys mutation into the AH is associated with Charcot-Marie-Tooth disease (Moroni et al., 2009).

α-Synuclein induces mitochondrial fragmentation (Kamp et al., 2010; Nakamura et al., 2011; O'Donnell et al., 2014; Martinez et al., 2018) possibly explained by its role in membrane fission (Varkey et al., 2010; Nakamura et al., 2011;

Zhukovsky et al. Amphipathic Helix-Containing Fission Proteins

Fakhree et al., 2019). α-Synuclein contains a long AH that forms on membrane binding (George et al., 1995; Davidson et al., 1998; Drin and Antonny, 2010; Braun et al., 2017), and the insertion of this AH into membranes contributes to membrane curvature generation (Braun et al., 2014, 2017; Fakhree et al., 2019). The role of α-synuclein in mitochondrial fission was suggested to be related to the pathogenesis of Parkinson's disease (Nakamura et al., 2011; Pozo Devoto and Falzone, 2017). Indeed, a few naturally occurring mutations of residues belonging to the α-synuclein AH (Pozo Devoto et al., 2017) are associated with Parkinson's disease: Ala30Pro (Krüger et al., 1998), Glu46Lys (Zarranz et al., 2004), His50Gln (Appel-Cresswell et al., 2013), Gly51Asp (Kiely et al., 2013; Lesage et al., 2013), Ala53Thr (Polymeropoulos et al., 1997), and Ala53Glu (Pasanen et al., 2014). Overexpression of the Ala53Thr α-synuclein mutant led to more significant reductions of mitochondrial size than the overexpression of wt α-synuclein (Pozo Devoto et al., 2017). Uncontrolled mitochondrial fragmentation might contribute to the development of Parkinson's disease (Varkey et al., 2010; Panchal and Tiwari, 2019). Hence, the role of Ala53Thr mutation in the development of Parkinson's disease might be, at least partially, explained by the extensive mitochondrial fragmentation mediated by this mutant involving residue belonging to the AH.

In the future, the discovery of new naturally occurring mutations of residues within the AHs of fission-inducing proteins will facilitate the ultimate definition of the role that the interaction of AHs with the membranes play in the pathogenesis of the related diseases.

#### CONCLUSION

In this review, we have considered fission-inducing proteins that contain AHs. Other fission-inducing proteins are known (Campelo and Malhotra, 2012; Frolov et al., 2015; Renard et al., 2015) indicating that the molecular events finally causing a bilayer to divide may follow different strategies. This would parallel the case of membrane fusion, where the structures of fusion-inducing proteins belonging to different classes are very different from each other and mechanisms of membrane

#### REFERENCES


rearrangements mediated by these proteins are also dissimilar (Kielian, 2014; Podbilewicz, 2014). The aspects that remain to be elucidated in both membrane fusion and fission are indeed the very last steps of these processes. While it is now relatively clear how the hemifission (or hemifusion) intermediate is obtained by the fission- or fusion-inducing proteins (see above and Chernomordik and Kozlov, 2008; Podbilewicz, 2014), the disruption of this structure and formation of the two bilayers (or their fusion) is an energy-demanding event not yet fully clarified. Protein rearrangements, lipid modifications or synthesis, GTP/GDP can all potentially contribute to this energy requirement. How they function, or if other molecules or sources of energy intervene, is still matter of debate. These are fascinating aspects in the membrane dynamics field that we are confident will be solved in the near future thanks to the continuously improved knowledge of the systems, together with the new morphological and instrumental tools being developed.

#### AUTHOR CONTRIBUTIONS

MZ, AL, DC, and CV wrote the manuscript. AF and CV made the figures. MZ and AF made the tables.

#### FUNDING

The work in the authors' laboratories was supported by the Italian Association for Cancer Research (AIRC) (DC: IG10341 and IG18776; AL: IG20786 and IG15767), the AIRC-Fondazione Cariplo TRansforming IDEas in Oncological research project (TRIDEO) (CV: IG17524), the PRONAT project, the PRIN project no. 20177XJCHX, the SATIN POR project 2014–2020, and the Italian-MIUR Cluster project Medintech (CNT01\_00177\_962865).

#### ACKNOWLEDGMENTS

We thank the funding agencies for supporting our research and Dr. R. Le Donne for some help in figure preparation.





and function of the BAR domain protein PICK1. Cell Rep. 23, 2056–2069. doi: 10.1016/j.celrep.2018.04.074



in lipid micelles and bilayers. Biophys. J. 95, 4315–4323. doi: 10.1529/biophysj. 108.134155




localization to autophagic membranes and lipid droplets. FEBS Lett. 591, 3819–3830. doi: 10.1002/1873-3468.12901



**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 Zhukovsky, Filograna, Luini, Corda and Valente. 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.