# ZEBRAFISH IN DEVELOPMENT AND DISEASE

EDITED BY : Gokhan Dalgin, Rebecca Ann Wingert and Ryan M. Anderson PUBLISHED IN : Frontiers in Cell and Developmental Biology and Frontiers in Genetics

#### 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 978-2-88963-248-0 DOI 10.3389/978-2-88963-248-0

#### About Frontiers

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

#### Frontiers Journal Series

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

#### Dedication to Quality

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

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

#### What are Frontiers Research Topics?

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

# ZEBRAFISH IN DEVELOPMENT AND DISEASE

Topic Editors: Gokhan Dalgin, University of Chicago, United States Rebecca Ann Wingert, University of Notre Dame, United States Ryan M. Anderson, Indiana University, Purdue University Indianapolis, United States

There are only a few vertebrate systems that can be used to model human diseases for biomedical discovery. The zebrafish model provides key advantages over existing models. Their externally developing embryos provide high-throughput non-invasive imaging, chemical screening, forward and reverse genetics, and their regeneration capacity make zebrafish a valuable system for novel discovery.

Developmental studies using zebrafish has influenced discoveries in many human health-related conditions. This Research Topic covers all aspects of zebrafish studies, providing developmental mechanisms to human health conditions. The aim of the Research Topic was to foster a platform to bring all levels of zebrafish research including but not limited to development, disease, regeneration, drug screening, bioinformatics and Omics studies.

Citation: Dalgin, G., Wingert, R. A., Anderson, R. M., eds. (2019). Zebrafish in Development and Disease. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-248-0

# Table of Contents

*04 Problems in Fish-to-Tetrapod Transition: Genetic Expeditions Into old Specimens*

Thomas W. P. Wood and Tetsuya Nakamura

*21 Using Zebrafish to Study Collective Cell Migration in Development and Disease*

Hannah M. Olson and Alex V. Nechiporuk


Amrita Mandal, Katherine Pinter and Catherine M. Drerup

*141 Expanding the CRISPR Toolbox in Zebrafish for Studying Development and Disease*

Kaili Liu, Cassidy Petree, Teresa Requena, Pratishtha Varshney and Gaurav K. Varshney

# Problems in Fish-to-Tetrapod Transition: Genetic Expeditions Into Old Specimens

Thomas W. P. Wood and Tetsuya Nakamura\*

*Department of Genetics, Rutgers, The State University of New Jersey, Piscataway, NJ, United States*

The fish-to-tetrapod transition is one of the fundamental problems in evolutionary biology. A significant amount of paleontological data has revealed the morphological trajectories of skeletons, such as those of the skull, vertebrae, and appendages in vertebrate history. Shifts in bone differentiation, from dermal to endochondral bones, are key to explaining skeletal transformations during the transition from water to land. However, the genetic underpinnings underlying the evolution of dermal and endochondral bones are largely missing. Recent genetic approaches utilizing model organisms—zebrafish, frogs, chickens, and mice—reveal the molecular mechanisms underlying vertebrate skeletal development and provide new insights for how the skeletal system has evolved. Currently, our experimental horizons to test evolutionary hypotheses are being expanded to non-model organisms with state-of-the-art techniques in molecular biology and imaging. An integration of functional genomics, developmental genetics, and high-resolution CT scanning into evolutionary inquiries allows us to reevaluate our understanding of old specimens. Here, we summarize the current perspectives in genetic programs underlying the development and evolution of the dermal skull roof, shoulder girdle, and appendages. The ratio shifts of dermal and endochondral bones, and its underlying mechanisms, during the fish-to-tetrapod transition are particularly emphasized. Recent studies have suggested the novel cell origins of dermal bones, and the interchangeability between dermal and endochondral bones, obscuring the ontogenetic distinction of these two types of bones. Assimilation of ontogenetic knowledge of dermal and endochondral bones from different structures demands revisions of the prevalent consensus in the evolutionary mechanisms of vertebrate skeletal shifts.

Keywords: fish, tetrapods, dermal bones, endochondral bones, skull, girdle, fin

### INTRODUCTION

The fish-to-tetrapod transition during the Devonian is one of the prominent events in vertebrate evolution. The invasion of the land from the water necessitated the evolution of the novel structures in skeletons, musculatures, innervations, visceral organs, and respiratory systems in order to adapt to a terrestrial life (Romer, 1949; Clack, 2012; Dial et al., 2015). Robust appendage skeletons with associated musculatures evolved to support body weight against gravitational force and to aid with movement on land (Coates, 1996; Shubin et al., 2006). To facilitate respiration in air, gill breathing had transformed to lung breathing through morphological innovations (Zheng et al., 2011; Sagai et al., 2017).

#### Edited by:

*Gokhan Dalgin, University of Chicago, United States*

#### Reviewed by:

*Raman Chandrasekar, Kansas State University, United States Randall Dahn, University of Wisconsin-Madison, United States*

> \*Correspondence: *Tetsuya Nakamura nakamura@dls.rutgers.edu*

#### Specialty section:

*This article was submitted to Molecular Medicine, a section of the journal Frontiers in Cell and Developmental Biology* Received: *30 March 2018*

Accepted: *15 June 2018* Published: *16 July 2018*

#### Citation:

*Wood TWP and Nakamura T (2018) Problems in Fish-to-Tetrapod Transition: Genetic Expeditions Into Old Specimens. Front. Cell Dev. Biol. 6:70. doi: 10.3389/fcell.2018.00070* Sensory systems have been dramatically reconstructed, and the lateral line that sensed physical movement of water was simply lost due to a lack of necessity on land (Piotrowski and Baker, 2014). The function of the kidney shifted from osmoregulation to the excretion of nitrogen in the form of urine or uric acid paste (Mahasen, 2016).

Among the many evolutionary novelties associated with terrestriality, the evolution of the skeleton has attracted much interest from evolutionary biologists. This is mainly because of the ability to trace its evolutionary history through relatively well-preserved structures in the fossil record. Skeletal shifts in the fish-to-tetrapod transition have been understood through the so-called dermal-to-endochondral transition (Westoll et al., 1977; Shubin et al., 2006; Hirasawa and Kuratani, 2015). A fish possesses a significant amount of dermal bones: the dermatocranium, shoulder girdle, jaw, teeth, scales, and fin rays. As lobe-finned fish (sarcopterygians) invaded land, the necessities of these dermal bones for survival fitness largely changed and dermal bones had been reorganized. The dermatocranium of fish skulls, that consists of multiple dermal plates, has experienced major reorganizations in its morphology and composition during the evolution into tetrapods (Coates, 1996; Daeschler et al., 2006). The cranial dermal bones articulate with the shoulder girdle at its posterior end, but, the tetrapod skull has lost several posterior bones, resulting in the dissociation of the skull from the shoulder girdle—the origin of the neck. Dorsal and anal fins are indispensable to stabilize the fish-body for fast swimming in water. However, they are a hindrance on land and have been lost in the tetrapod lineage. Scales that serve as osmotic controls and as protection of the body in an aquatic habitat were also lost during landing. With this concomitant loss of dermal bones, the relative ratio of endochondral bones in vertebrae, girdle, and fins increased in early tetrapods. The endochondral bones in the pectoral girdle enlarged and evolved into the scapula of modern tetrapods (Shubin et al., 2006). In accompaniment with the loss of dermal fin rays in paired fins that propel and balance their bodies, early tetrapods evolved endochondral digits and wrists/ankles (Clack, 2009a; Schneider and Shubin, 2013; Pieretti et al., 2015), helping them to acquire diverse locomotory abilities and extensive maneuverability.

Dermal and endochondral bones are histologically and ontogenetically distinct (Hall, 2005). Dermal bones develop directly from the mesenchyme without a pre-formed cartilaginous model (intramembranous ossification), whereas endochondral bones develop by converting cartilage to bone. The dermoskeleton appears to have originated in the teeth of agnathans such as conodonts and diversified into the dermal skull roof, scales, or fin rays during fish evolution, though evolutionary trajectories of dermal bones are under debate (Donoghue and Sansom, 2002; Sire and Huysseune, 2003; Hirasawa and Kuratani, 2015). Endochondral ossification is hypothesized to have arose much later, in Osteoichthyes (Hirasawa and Kuratani, 2015). Even though a large number of dermal bones have been lost in the transition from fish, tetrapods still possess comparable developmental pathways of dermal bones, which grow calvaria and clavicle bones. The vestiges of the dermal-to-endochondral transition in the developmental programs of vertebrate skeleton cause multiple types of congenital and postnatal skeletal diseases in humans (Gilbert, 2000; Hall, 2005; Wagner and Aspenberg, 2011), such as Progressive Osseous Heteroplasia (POH) or Albright Hereditary Osteodystrophy (AHO) (Regard et al., 2013; Pignolo et al., 2015), which both develop heterotopic dermal bones.

Regardless of the central roles of dermal and endochondral bones in skeletal evolution and clinical cases, scrutiny of the genetic mechanisms that differentiate these distinct types of bones is not enough. With the advent of new technology in molecular biology and paleontology, these problems are now becoming more amenable. The invention of novel sequencing methods promotes comparative genomics with high-throughput output across a diverse range of species (Braasch et al., 2016; Smith et al., 2018). Rapid advancements of genetic manipulation techniques opens a new window to observe functioning of target genes in model and non-model organisms that hold prominent positions in vertebrate evolution (Parker et al., 2014). Furthermore, high-resolution CT scanning is revealing fine details of fossil records and living taxa (Giles et al., 2017; Pardo et al., 2017). These advancements of technology fill in the gaps of understanding evolutionary mechanisms of vertebrates by involving model and non-model organisms into lab-level experiments.

In this article, we summarize the current understanding and problems in the developmental processes and evolutionary shifts of dermal and endochondral bones. First, we review the evolution of cranial dermal bones and the underlying developmental mechanisms. Then, we discuss the skeletal shift from dermal to endochondral bones in the shoulder girdle and current perspectives of the underlying mechanisms. Third, we highlight the approaches that integrate developmental mechanisms into comparative anatomy to answer the fin-tolimb conundrum. Assimilating current knowledge about the molecular mechanisms underlying skeletal shifts of distinct structures would take us one step closer to elucidating the fishto-tetrapod transition.

### REARRANGEMENTS OF SKULL DERMAL BONES

The dermal skull roof is one of the remarkable exemplars for continuous modifications of skeletons during the fish-totetrapod transition. The skull roof of sarcopterygians that led to the tetrapod lineage consists of multiple dermal units: such as nasal, parietal (or frontal), temporal, intratemporal, or opercular (**Figure 1A**). These dermal bones encase endochondral cranial components and protect the primary operative unit of the central nervous system—the brain. The comparative studies of skull morphology between fish and tetrapods highlight apparent distinctions of their broad proportions: the snout is relatively longer, the orbits are located more dorsally, and the skull is flatter in tetrapods. The functional reasons for these skeletal modifications are obscure, but they are likely to be linked with sensory and feeding requirements, and also size of the otic capsules (Clack, 1989). The proportional shifts of skulls from

(2012), Jarvik (1980), and White et al. (2016).

fish to tetrapods are direct consequences of the remodeling of dermal and endochondral bone morphologies, and also simple reductions of dermal bones. Eusthenopteron is one of the rhipidistians that lived in shallow freshwater during the late Paleozoic (Andrews and Westoll, 1970). The dermal skull roof of Eusthenopteron consists of remarkably large parietals (referred to as "frontals" in actinopterygians) and postparietals (Moy-Thomas and Miles, 1971). Orbits are located at more front of the skull compared with that of tetrapods (**Figure 1**). Panderichthys shares multiple unique features with tetrapods: flat skull, medially located eyes, and no dorsal and anal fins (Schultze and Arsenault, 1985; Boisvert et al., 2008). However, the shape of the four median pairs of cranial dermal bones—nasals, frontals, parietals, and postparietal bones—are rather similar to those of fish. The skull of Tiktaalik is also flat with large frontals and postfrontals, which bridges the gap of skull morphology between fish and tetrapods (Daeschler et al., 2006). Acanthostega, which has limbs with digits, further fills the gap in our understanding of the difference in the skull roof morphology between fish and tetrapods with its intermediate features—large nasals and frontals, and also the relatively small parietal and postparietal bones (Coates, 1996). In the early tetrapod Ichtyostega, nasals and frontals are long, and parietals and postparietals are rather short compared to its early fossil relatives (Jarvik, 1980). Due to these remodeling of dermal bones, the orbits have shifted to a further posterodorsal position—one of the shared features with early tetrapods.

On the lateral side of the skull, gills of osteichthyan fish are covered externally by an opercular series that consists of four dermal bones: the preopercular, opercular, subopercular, and the interopercular. A group of sarcopterygians had adapted to terrestrial life and evolved lungs instead of gills with the gradual loss of opercular series from the skull (**Figure 1**). Rhipidistians (including Eusthenopteron) and Panderichthys have operculars and suboperculars on the back of their cheek bones (**Figure 1A**), whereas Acanthostega and Tiktaalik barely possess the opercular bone series, but have a rather small preopercular (Coates, 1996; Daeschler et al., 2006; **Figure 1B**). The skull of Ichthyostega also shows a vestige of a preopercular at the back of the squamosals and qudratojugal without any sign of other opercular bone series (Jarvik, 1980; **Figure 1C**). It remains elusive how much of the function of these preoperculars is for gill breathing or whether they are necessary for other physiological activities in these vertebrates.

Extrascapular series connect the skull to the shoulder bones at the most posterior parts of the skull, and they are another example for the loss of dermal bones in the water-to-land transition (**Figure 1**). Rhipidistians retain extrascapular series medial extrascapular and lateral extrascapular—at the posterior of postparietal and tarbular (Andrews and Westoll, 1970). Supracleithrum (girdle bone) articulates to these extrascapular bones via the post-temporal from the posterior side of the skull (Andrews and Westoll, 1970; **Figure 1A**). However, Acanthostega and Tiktaalik have lost extrascapular series, resulting in a dissociation of the skull and girdle which led to terrestrial locomotion (Coates, 1996; Daeschler et al., 2006; **Figure 1B**).

Although completing the details of the skeletal shifts from fish to tetrapods is formidable, the remodeling of the dermal skull roof, and loss of extrascapular and opercular series are general trends with alteration of their habitats across vertebrate phylogeny (Clack, 2012). Of particular note, actinopterygians, including zebrafish which is one of the well-established model organisms, possess comparable opercular and girdle bones to those of the rhipidistians, though genetic programs in teleosts underlying development of these bones might be derived compared with rhipidistians.

#### DEVELOPMENTAL BASIS FOR CRANIAL DERMAL BONES

Along with the extraordinary discoveries of the transitional specimens from water to land, the molecular mechanisms underlying development of cranial dermal bones have been vigorously studied in teeth, operculum, calvaria, mandibule, and other bones (Tucker and Sharpe, 2004; Huycke et al., 2012; Rasch et al., 2016). The dawn of the description of cranial dermal bone development in fish dates back to the early nineteenth century (Pehrson, 1940, 1944, 1958; Aumonier, 1941; Kapoor, 1970; De Beer, 1985; Cubbage and Mabee, 1996). High-resolution histological approaches, including Scanning Electron Microscopy and Transmission Electron Microscopy, have highlighted the developmental process of dermal bones and disparities of structures between dermal and endochondral bones of extinct and living taxa (Verreijdt et al., 2002; Sire and Huysseune, 2003; Witzmann, 2009; Keating and Donoghue, 2016). However, in spite of the continuous efforts (Sire and Huysseune, 2003), the understanding of the development and the evolution of dermal bones still remains poor. Since the changes of cell origins in cranial dermal bone development across species are summarized in previous studies (Piekarski et al., 2014; Hirasawa and Kuratani, 2015; Maddin et al., 2016), we review current perspectives about molecular pathways that are responsible for cranial dermal bone development with the significant emphasis on interchangeability of dermal and endochondral bones.

Dermal bones retain three layers: for example, dentine tubercles are at the outermost surface, middle spongy layer, and basal laminated layer in agnathans (Donoghue and Sansom, 2002). The middle spongy layer is highly vascularized and transports oxygen, proteins, and hormones to support osteogenesis and homeostasis (Percival and Richtsmeier, 2013). Although the details of structures have been modified in different species, the three-layer structure is conserved in vertebrate evolution. The development of dermal bones begins with mesenchymal cell condensations under epithelial layers at their early stages. The aggregated cells directly differentiate into osteoblasts that express osteoblast markers, such as Runx2 (Abzhanov et al., 2007). In many cases, the epithelialmesenchyme interaction plays a central role in the initiation of dermal ossification (Sire and Huysseune, 2003). Due to this specific developmental process and the thin structures of dermal bones, surface bones of the fish body consist of dermal bones, not endochondral bones such as scales or spines. Osteoblasts, in turn, induce mineralization around them and differentiate into osteocytes. Distinct from dermal bones, endochondral bones develop from a cartilage template (Long and Ornitz, 2013). The first step of endochondral development is also forming mesenchymal cell condensations, but aggregated mesenchymal cells subsequently differentiate into chondrocytes that express a cartilage marker, Sox9. Chondrocytes further differentiate into hypertrophic chondrocytes that induce invasions of blood vessels into developing bones. Chondrocytes are replaced by osteoblasts at the center of long bones, but they do not differentiate into osteoblasts at peripheral regions but rather continue to stimulate cell proliferation and bone growth by Indian Hedgehog (Kronenberg, 2003).

Due to the entrenched idea that dermal bones develop from neural crest cells whereas endochondral bones originate from mesenchymal cells (Smith and Hall, 1990; Sire and Akimenko, 2004), their developmental processes have been hypothesized to be completely distinct. To gain a deeper understanding into the developmental process of dermal bones, Abzhanov and colleagues extensively investigated the gene expression profiles in developing dermal dentary bones of chicken and mouse embryos (Abzhanov et al., 2007). Expanding the previous knowledge of the similarity in some gene expression pattern in dermal and endochondral development including Runx2, they discovered that dermal dentary bone expresses some endochondral genes, which had been regarded as specific markers for chondrocytes, Collagen type 2 and 9. Moreover, Abzhanov et al. performed fluorescent double in situ hybridization that stained different combinations of osteoblast markers and identified four distinct stages in dermal-bone development, including "chondrocytelike" osteoblasts in the developmental process of dermal dentary bone. These data suggest that the developmental programs of dermal and endochondral bones are in part similar at least in terms of gene expression profiles.

The developmental process of the dermal skull roof (calvaria) has attracted a large amount of attention due to the significance not only in the evolutionary history of vertebrates (Hanken and Hall, 1993; Janvier, 1996), but also in clinical aspects (Tubbs et al., 2012), resulting in rapid advancements of understanding in the developmental mechanisms of cranial dermal bones. Mouse research has considerably contributed to the elucidation of molecular pathways that regulate calvaria development, such as Bone Morphogenetic Protein (BMP), Fibroblast Growth Factor (FGF), Wingless type MMTV integration site family (WNT), Growth and Differentiation Factor (GDF), TWIST, Engrailed 1, Foxc1, and others (Ishii et al., 2015). Bmp2, 4, and 7 are expressed in cranial neural crest cells and are one of main factors that develop the calvaria. Conditional knockout of Bmp2, Bmp4, and Bmp7 in cranial neural crest cells during mouse embryonic development by using the Wnt-Cre transgenic mice line resulted in the enlarged frontal fontanelle (Bonilla-Claudio et al., 2012). The conditional double knockout mouse of Msx1 and Msx2, downstream targets of the BMP signal pathway, in cranial neural crest cells, also exhibited severe reduction of frontals, and unexpectedly, newly synthesized cartilage compensated the lack of dermal frontals in the mutant mouse (Roybal et al., 2010). Roybal and colleagues identified the cell source of this ectopic cartilage, which indeed developed from a part of the neural crest cells that do not contribute to dermal bones under normal conditions (Roybal et al., 2010). This data implies a dual role of MSXs in calvaria development—an inducer of dermal bones and suppressor of cartilage bones (**Figure 1D**). Thus, the level of BMP signaling is likely to be one of main factors that controls the ratio of dermal and endochondral bones through SMAD1/5/8 and MSX which, in turn, induce downstream target genes in cranial bones (Bonilla-Claudio et al., 2012). Abzhanov and colleagues showed that mis-expression of Bmp4 by virus infection replaces dermal bones by cartilage in chicken frontal bone (Abzhanov et al., 2007), though the result is intuitively opposite to that of the Msx conditional knockout mouse. In either scenario, BMP signaling is likely to regulate dermal and endochondral bone differentiation in the concentration-dependent manner. We do not possess any evidence to substantiate the necessities of the exchange program between dermal and endochondral differentiation in mutant mice. The interchangeability between dermal and endochondral development in cranial bones could be the compensatory mechanism to ensure the development of skull bones, yet their contribution to development and evolution is unknown.

The WNT signaling pathway is another major regulator that develops cranial dermal bones and is a balancer for the ratio of dermal and endochondral bones as well (**Figure 1D**). The conditional knockout mice for β-catenin, an intracellular signal transducer of the WNT signal, in cranial neural crest cells and paraxial mesoderm (PAM) by Engrailed-Cre transgenic mouse (Goodnough et al., 2012) did not form calvaria. This conditional knockout mouse rather grew cartilage in the original position of calvaria. The phenotype in which dermal bones are replaced by endochondral bones is reminiscent of the conditional double knockout of Msx1 and 2 and implies that the BMP and WNT pathways crosstalk in calvaria development. Goodnough and colleagues further discovered that β-catenin functions to switch differentiation from chondrocytes to osteoblasts via TWIST1, which binds to the 3′ untranslated region of Sox9 mRNA (Goodnough et al., 2012; **Figure 1D**).

The new study added another key player into the development of cranial bones. Barske et al. identified Nuclear Receptor 2f genes (Nr2fs) as repressor genes of cartilage development in dermal maxilla in zebrafish (Barske et al., 2018). Expression of Nr2f is repressed by Endothelin-1 in the mandible, resulting in endochondral ossification. Though the function of NRF2 in the limb bud is likely to be different from that of maxilla (Barske et al., 2018), it is fascinating to test NRF2s function in calvaria development. One of the subsequent challenges is elucidating how these distinct signal pathways synergistically or redundantly regulate dermal bone development and how they switch the differentiation between dermal and endochondral bones.

Classical surgical experiments suggested that epithelialmesenchymal interaction initiates dermal bone development in the mesenchyme, yet the molecular mechanisms had remained elusive for a quarter of a century (Hall, 2005). Goodnough and colleagues deployed a conditional knockout system to answer how the epithelial layer promotes development of dermal bones in the cranial mesenchyme (Goodnough et al., 2014); the conditional deletion of Wls, the trafficking regulator of the WNT ligand, by Crect (ectodermal Cre) mice showed the entire loss of the calvaria bones. However, the conditional knockout of Wls in the underlying mesenchyme by Dermo1Cre (Cre expression in cranial mesenchyme and meningeal progenitor cells) did not show a comparable dermal phenotype (Goodnough et al., 2014). These data suggest that the WNT signal from the ectoderm, presumably through diffusion of proteins, initiates cranial dermal bone development in the underlying mesenchyme. Because multiple WNT ligands are expressed in the epithelial layer and cranial mesenchyme, it is timeintensive to identify the major WNT ligand that transmits signal from the ectoderm to the mesenchyme (Goodnough et al., 2014). While many genes have been discovered to be responsible for cranial dermal bone development, understanding the initial mechanisms that initiate dermal differentiation—the upstream of the WNT signal in the ectoderm—would be one of next challenges. Further experiments are expected to identify molecules or any other cues that trigger the differentiation of cranial dermal bones at the upstream of the WNT signal pathway.

Along with the discoveries of many genes that develop cranial dermal bones, actual genetic loci that have contributed to vertebrate cranial evolution is still obscure. The BMP Wood and Nakamura Skeletal Evolution From Fish to Tetrapods

and WNT signals, which are necessary for the development of the mouse calvaria, are also involved in developing and diversifying skull shapes in fish (Albertson et al., 2005). The inhibition of WNT signaling by a chemical agonist or antagonist in cichlids created the diversity of preorbital morphologies, which phenocopy cichlid facial diversity (Parsons et al., 2014). The detection of single nucleotide polymorphism (SNP) in β-catenin of cichlids also supports the involvement of the WNT signal in facial diversity (Loh et al., 2008). Further assessments to test the functional roles of the BMP and WNT signal pathways in development of the fish skull would enrich developmental and evolutionary basis of fish skull diversity. During the fish-to-tetrapod transition, many cranial dermal bones are lost or reorganized (Clack, 1989; Daeschler et al., 2006). It is likely that different genetic loci contributed to the modifications of different bones, and/or multiple genetic loci contributed to a rearrangement of a single bone, as we have observed in complex evolutionary traits of other vertebrates (Linnen et al., 2013). To gain complete pictures of the molecular mechanisms underlying the evolution of cranial dermal bones, strategies that comprehensively identify genetic loci which contribute to specific phenotypes are ongoing and are successfully capturing responsible regions (Jones et al., 2012; Parsons and Albertson, 2013; Miller et al., 2014). The contribution of genetic loci, which are identified by these QTL studies in closely related species, to a more long-time scale evolution of vertebrates (i.e., the fish-to-tetrapod transition) remains elusive.

### DISCONNECTION OF THE SHOULDER GIRDLE FROM THE SKULL

The pectoral girdle morphologies of fish and tetrapods are spectacularly diverse. Since the girdle links appendicular bones to the body trunk and serves as the base of attachment for muscles of the neck and pectoral appendages, the girdle holds a critical position in the evolution of vertebrate locomotion. Comparison of pectoral girdle morphology across vertebrates highlights the general trend in girdle evolution—the reduction of dermal bones and enlargement of endochondral bones. In fish, the series of pectoral girdle bones mainly consists of dermal bony plates: the supracleithrum, postcleithrum, cleithrum, and the clavicle along the dorsoventral axis (Andrews and Westoll, 1970; **Figure 2A**). The supracleithrum articulates the pectoral girdle series to the post-temporal (**Figure 1A**) and orchestrates movements of the head and paired pectoral appendages. Also, the fish pectoral girdle has the scaupulocoracoid, a relatively small endochondral bone that connects the humerus to the pectoral girdle via the glenoid fossa. To invade land, early vertebrates required a robust skeleton to support their body weight without buoyancy. During the fish-to-tetrapod transition, dermal bones had become reduced and simultaneously the endochondral scapulocoracoid had enlarged (**Figures 2A–C**). The scapulocoracoid had split into the scapula and procoracoid bones, which both develop from different developmental centers. In amniotes, the shoulder girdle has been further modified and it has become three bones: the scapula, the procoracoid, and the coracoid.

The developmental processes of girdle bones (Grandel and Schulte-Merker, 1998; Davis et al., 2004; Sears, 2004; Pomikal and Streicher, 2009; Boisvert et al., 2013; Sears et al., 2013; Dillman and Hilton, 2015; Warth et al., 2017) and the associated muscles (Ericsson et al., 2013; Diogo et al., 2014; Masyuk et al., 2014; Pu et al., 2016) have been described in many taxa, yet the problematic evolutionary history of girdle bones has hampered us from understanding the cellular origins of girdle bones. Kague and colleagues tested whether neural crest cells contribute to girdle bones in zebrafish by using Wnt1-Cre transgenic fish and confirmed that neural crest cells do not migrate into dermal girdle bones including the supracleithrum, postcleithrum, cleithrum, and the endochondral scapulocoracoid. This finding raises a possibility that the zebrafish girdle bones originate from the lateral plate mesoderm (LPM) or the PAM, yet these hypotheses have not been tested. The cell origins of girdle bones, particularly the scapula, have been more intensively investigated in tetrapods: salamander, turtle, chicken, and mouse embryos (Burke, 1991; Huang et al., 2000; Malashichev et al., 2008; Valasek et al., 2010; Kague et al., 2012; Nagashima et al., 2016). All of these studies support the ontogenetically conserved pattern across tetrapods. The tetrapod scapula has, at the very least, a dual origin—the LPM and somites. Piekarski and Olsson transplanted GFP-labeled somites to wild-type axolotl and demonstrated the somitic contribution to the suprascapular (Piekarski and Olsson, 2011). Burke performed the removal of somites adjacent to the fore limb region of turtle embryos, which resulted in scapular defects (Burke, 1991). Huang and colleagues performed chick-quail chimeric analyses and revealed that the head and neck of the scapula originates from the LPM, but the scapula blade develops from the dermomyotome of somite 17–24 (Huang et al., 2000). Intriguingly, the cells maintain their topology of original somites in the developing scapular such that muscles attach to a specific position of the scapula blade that originate from same somites. The developmental logic underlying the dual origin of scapular cells is explained by Piekarski's non-canonical "position-dependent" hypothesis that proposes that the scapula develops from its most adjacent tissue, either somite or LPM (Piekarski and Olsson, 2011). Mouse studies also showed that the mouse scapula originates from somites by using Pax3-Cre transgenic mice (Valasek et al., 2010). In addition, Matsuoka et al. showed that neural crest cells contribute to the scapular spine, coracoid, and acromion by using Wnt1-Cre lineage trace system in mice (Matsuoka et al., 2005). Further research could test the contribution of neural crest cells into the girdle bones of primitive actinopterygian or elasmobranchs; whether non-neural crest origin of zebrafish girdle is an evolutionary conserved pattern in fish lineage.

Knowledge about the developmental programs of fish girdle bones is still fragmented. Syu homozygous zebrafish that have a mutation in the Sonic hedgehog a coding sequence showed a severe defect in cleithrum and pectoral fin development (Neumann et al., 1999). The morpholino knockout of T-box gene, Tbx5, also affected the pectoral fin module; with the loss

bone development is still poor (see text). acl, anocleithrum; acr, acromion; cl, cleithrum; cla, clavicle; g, glenoid fossa; ic, interclavicle; sc, scapula; sca, scapulocoracoid; scl, supracleithrum; sp, spine; Illustrations in (A–C) are redrawn with permissions form Andrews and Westoll (1970), Jarvik (1980), and Gregory (1951). The illustration of the mouse scapula is adapted with permission from Kuijper et al. (2005).

of pectoral fins, the scapulocoracoid, postcoracoid process, and the cleithrum were also severely affected (Ahn et al., 2002). Further, Dlx5a and Dlx6a morpholino-mediated knockdown in zebrafish disrupted not only the pectoral fin, but also girdle bone development (Heude et al., 2014). These findings indicated that the development of fins and girdles, either dermal or endochondral, are interlinked by the same genes and cannot be simply separated because of their close topology in the developing body.

In tetrapods, especially in chickens and mice, more genes have been uncovered in scapula and clavicle development (Huang et al., 2006; **Figure 2D**). The cranial part of the pectoral girdle, including the acromion, coracoid process, and the glenoid fossa, develops from the LPM. The developmental programs of the anterior scapular bone shares the same set of genes with the limb bud as we have seen in the development of the zebrafish pectoral girdle; Dlx (Heude et al., 2014), Islet1 (Itou et al., 2012), Tbx5 (Valasek et al., 2011), and Twist1 (Krawchuk et al., 2010; Loebel et al., 2012) affect both girdle and limb development. Contrary to the canonical hypothesis, the scapular blade develops from dermomyotome, not sclerotome, which goes through the epithelial-mesenchymal transition (EMT) (**Figure 2D**). Several transplantation studies in chicken embryos showed that signals from the ectoderm to dermomyotome are necessary for this EMT (Malashichev et al., 2005). The molecules that induce EMT in the dermomyotome from the ectoderm have not been identified, yet the attenuation of the WNT signal is likely to be involved. Moeller and colleagues ectopically expressed Carboxypeptidase Z (CPZ), which possesses a WNT binding domain, in somites of chicken embryos and discovered that the WNT signal downregulates Pax1 expression that is necessary for scapular development and promotes Pax3/7 expression that is necessary for limb muscle development (Moeller et al., 2003). WNT6 in the ectoderm was suggested as a primary diffusible ligand to maintain Pax3/7 expression, nevertheless further experiments to verify its function in the EMT are imperative (Schmidt et al., 2004). Once the EMT produces prospective scapular cells with Pax1 expression, Pbx family genes become key players to regulate and pattern scapular development (**Figure 2D**). PBX1/2/3 are expressed in the proximal limb bud and promote cartilaginous condensation through binding with EMX2 (Capellini et al., 2010, 2011). In parallel with developing cell condensation, PBX and BMP (Hofmann et al., 1998; Capdevila et al., 1999) regulate Hoxa5 and Pax1 (Timmons et al., 1994; Hofmann et al., 1998; Aubin et al., 2002) to pattern the acromion and the scapular head. The patterning of posterior scapula is established by Alx1 (Capellini et al., 2010), Tbx15, and Gli3 (Kuijper et al., 2005) that are also downstream of PBX1.

Despite of the discovery of a number of genes for scapular development, little is known about the molecular mechanisms underlying the development of the dermal clavicle in the shoulder girdle. Kuijper and colleagues investigated the girdle phenotype of triple knockout mice of Alx4, Cart1, and Tbx15 and discovered that the clavicle showed a severe phenotype while the scapula showed a minorly affected morphology in these mutant mice (Kuijper et al., 2005). These data suggest that these genes may regulate more or less specifically dermal bone development in the shoulder girdle, but the precise mechanisms are unknown.

Fish girdle bones are almost all dermal bones, which may utilize epithelial-mesenchymal interaction or ossify by themselves without any input from epithelial tissue. The ectodermal signal in the scapular development of tetrapods is most likely important to differentiate the competent cell population for the prospective shoulder girdle, not to trigger the bone developmental program itself. As reviewed above, in the shoulder girdle, the developmental programs of dermal and endochondral bones are presumably intermingled; the genes affecting endochondral bone development also affect dermal bone development in most cases. This data implies that the cell sources and developmental programs for these two types of bones in the shoulder girdle are not obviously separated due to their complicated evolutionary history between the skull and body trunk. It is compelling to test how the epithelial-dermomyotome interaction is conserved in the development of the scapulocoracoid, and dermal bones of fish.

### FINS INTO LIMBS

The evolution of tetrapod limbs from fish fins is one of the most remarkable transitions in vertebrate history (Clack, 2009a,b). Whereas fish fins consist of endochondral bones in a proximal domain and dermal fin rays in a distal domain, tetrapod limbs are exclusively composed of endochondral bones. In Eusthenopteron, the scapulocoracoid articulates with the humerus that further connects to the ulna and the radius (Andrews and Westoll, 1970; **Figure 3A**). The distal ends of the ulna and the radius attach to the preaxial radials, which are followed by the lepidotrichia. The pectoral fin of Tiktaalik presents an intermediate structure between fish and tetrapods with regards to its morphology and function. The Tiktaalik fin possesses elaborated distal endochondral bones; their morphology and mobility is reminiscent of distal appendages of tetrapods (Shubin et al., 2006; **Figure 3B**). In contrast to the extension of endochondral domain toward the distal direction, the dermal fin rays of Tiktaalik are much reduced compared with that of fish. A further transition from fins into limbs is observed in Acanthostega. The paired appendages of Acanthostega retain comparable digits to those of tetrapods with a stout humerus, ulna, and radius, but without any evidence of fin rays (Coates, 1996; **Figure 3C**). The number of digits is eight in the fore limb of Acanthostega, but Ichthyostega shows a reduction of digit number toward five, which is the shared feature with later tetrapods—pentadactylism (Jarvik, 1980). The reduction of dermal bones and the increase of endochondral bones in the evolution of paired appendages is a similar trend such as when shoulder girdle bones evolved and this trend is significantly associated with the functional importance of the endochondral appendage in terrestrial life. However, it is not understood whether underlying molecular mechanisms of shoulder and limb evolution are common, or if they employ distinct mechanisms.

Since fish lepidotrichia, one of the major components of fin rays, directly ossifies without a cartilage model, the cell origin of fin rays had been assumed to be neural crest cells like other dermal bones, such as scales. However, the recent genetic labeling of the specific cell population in zebrafish revealed the other cell origins of fin rays. Lee and colleagues used Tbx6 promoter-Cre transgenic fish to trace the derivatives of PAM and discovered that the PAM solely contributes to the development of fin rays in the caudal fin (Lee et al., 2013). Another line of evidence from medaka also supported this conclusion (Shimada et al., 2013). Shimada and colleagues transplanted somite cells of transgenic fish that ubiquitously expresses DsRed into that of wild-type fish and confirmed that these red fluorescent cells contribute to median fin rays. In parallel with this transplantation experiment, they proved the contribution of the PAM to median fin rays by "IR-LEGO" in which Cre-loxP system irreversibly marks target cells with heat shock stimulus. Deploying the same labeling system, they discovered that the LPM, not neural crest cells, contributes to lepidotrichia formation in the pectoral fin (Yano et al., 2012; Shimada et al., 2013). Therefore, model organisms, zebrafish and medaka, provide us the ability to perform genetic experiments to test our hypotheses, yet, another compelling experiment is to test the evolutionary synapomorphy with primitive actinopterygians. Tulenko and colleagues injected DiI into the LPM shortly after gastrulation and confirmed that DiI-labeled LPM cells contribute to pectoral fin fold (Tulenko et al., 2017). The distinct cell origins of lepidotrichia (PAM or LPM) in unpaired and paired fins refuels the discussion of the origin of paired fins, which share same Shh cis module with unpaired fins (Freitas et al., 2006; Letelier et al., 2018). Mice and chicken studies have provided us with comparable knowledge of the cellular origins of the limb bud to that of fish. Gros and Tabin

FIGURE 3 | *hox13* genes are expressed in a distal domain of the endochondral disk. The cells that experienced the late phase *hox* activity migrate out from the distal endochondral disk into the fin fold, that differentiate into the lepidotrichia. In mice, *Hox13* genes are expressed in the autopod, which develop digits at a later stage. HOXA13 is suggested to bind the regulatory region of *Bmp2* and *4* by ChIP experiments (Knosp et al., 2004). While many of the downstream genes of HOX13 have been explored in mice, genes regulated by HOX13 in fish have not been identified. Note that the ossification of lepidotrichia and digits takes place at a later stage than the expression of *Hox13* genes. (E) Hypotheses for the fin-to-limb transition. The fin rays (cells shaded by light orange) have degenerated and the endochondral domain (cells shaded by light gray) expanded during the appendage evolution (top). This hypothesis supports that digits and wrists are novel domains that have been acquired as fish have evolved to tetrapods. Another hypothesis claims that the fish fin has an antecedent of digits and wrists (bottom). The cell histories between fish fin rays and tetrapod digits are comparable in terms of their *hox* expression pattern during the embryonic development (cells shaded by orange). In this hypothesis, the cell differentiation program of fin rays might have changed to become endochondral bones, resulting in the acquisition of digits and the wrist. dr, distal radials; hu, humerus; int, intermedium; ir, intermediate radials, le, lepidotrichia; po, postaxial process, pr, preaxial radials; pro, proximal radials; ra, radius; ul,ulna, uln, ulnae; Illustrations of (A–C) are redrawn with permissions from Andrews and Westoll (1970), Shubin et al. (2006), and Coates (1996).

demonstrated that the epithelial cells of the somatopleural LPM contribute to the limb bud mesenchyme through EMT (Gros and Tabin, 2014). Further, the genetically labeled mesenchymal cells which expressed HoxA13 exclusively contribute to digits and wrist bones (Scotti et al., 2015). This evidence collectively suggests that LPM-derived cells express HoxA13 and contribute to endochondral bones, yet no scrutiny of cell origins of HoxA13 cells have been conducted to date.

The genetic underpinnings of the development of paired appendages have been profoundly investigated in the embryonic limbs of chickens and mice (Zeller et al., 2009). Hox genes— Homeodomain-containing transcription factors—play pivotal roles in body patterning during embryonic development and are expressed in the limb bud (Zakany and Duboule, 2007). In early development of the limb bud, the genes in HoxA and HoxD clusters are expressed in a nested manner from the posterior to anterior limb bud; expressions from "anterior" genes (3′ side genes of Hox clusters. e.g., Hoxa1 or Hoxa2) of HoxA and HoxD clusters are relatively broader than that of "posterior" genes (5′ side genes of Hox clusters. e.g., Hoxa13 or Hoxd12) to the anterior domains of the limb bud. According to this nested expression pattern, Hox genes produce positional information along the anteroposterior axis inside the limb bud. Sonic Hedgehog is expressed at the Zone of Polarizing Activity (ZPA) and constitutes a positive feedback loop with Hox genes, resulting in cell proliferation and also providing positional information with Hox genes (Zeller et al., 2009) in the mesenchyme. At later developmental stages, anterior HoxA and D expression is limited in a proximal domain of the limb bud, while posterior HoxA and D genes gain a new expression domain in the autopod (**Figure 3D**). The functional roles of these site-specific Hox expressions were tested in knockout mice. The Hoxa13 and d13 combinatorial knockout mice lost the entire autopod (Fromental-Ramain et al., 1996b), while the Hoxa9 and d9 knockout mice eliminated the humerus without affecting autopod development (Fromental-Ramain et al., 1996a). Recent studies revealed genomic underpinnings of the two-phase Hox expression; the long-range contacts of the regulatory regions to HoxA and D clusters from centromeric side or telomeric side establish 3D chromatin loop structures (Topologically Associating Domain, TAD) (Montavon et al., 2011; Andrey et al., 2013) and assure physical proximity of the enhancers to Hox genes. These large chromatin structures regulate a group of target gene expression in a time- and site- specific manner and develop two distinct domains in the limb bud—the proximal limb (the humerus, ulna and the radius) and the distal limb (digits).

Hox genes are expressed in a nested manner in the endochondral disk of pectoral fins of fish as well (Ahn and Ho, 2008). At early stages, expression of "posterior" hoxa and d genes are more restricted to a posterior domain of the fin bud compared with that of "anterior" genes. At later stages, expression from posterior hoxa11-13 and d11-13 is restricted to a distal domain of the fin bud as "posterior genes" of mice in the limb bud (**Figure 3D**). Woltering et al. and Gehrke et al. revealed that the chromosomal topologies underlying these twophase hox expression patterns are the shared feature with that of mouse by 4C-sequencing, suggesting that fish fins already retain a dual TAD system in paired fins before evolving digit and wrist (Woltering et al., 2014; Gehrke et al., 2015). Other major genetic networks for the development of paired appendages are also peculiarly conserved between the endochondral disk of fish and limb bud of tetrapods: Tbx5 (Ahn et al., 2002; Adachi et al., 2016), Shh (Neumann et al., 1999), Fgf8 (Jovelin et al., 2007), Bmp2 (Laforest et al., 1998), and others. During the fin fold development, the LPM migrates from the proximal fin bud into the distal fin fold, and the cell configuration becomes flat and thin in the distal domain (Thorogood, 1991; Yano et al., 2012). Concomitant with physical cell migration, hoxa13a expression also migrates out from the endochondral disk to the proximal fin fold (Ahn and Ho, 2008; Nakamura et al., 2016; Tulenko et al., 2016). Surprisingly, during the late development of fin fold, the gene expression profile of the fin fold in paddlefish and of the autopod in mice is rather similar despite their distinct histological structures (Tulenko et al., 2017). Shh is expressed at the posterior edge of the pectoral fin fold, and Fgf10 and hoxa13 is in the broad domain of the fin fold, all of which are reminiscent of gene expression patterns of the mesenchyme in the mouse limb bud (Tulenko et al., 2017).

The shared developmental programs between the fin fold and the autopod inevitably drive further questions. For example, how do these two appendage primordia develop into distinct dermal or endochondral bones from the conserved gene expression patterns? The signaling pathways underlying the development of dermal fin rays has been vigorously studied in normal development and also in the regeneration process after amputations of fins (Wehner and Weidinger, 2015). As the fin fold develops, actinotrichia form at a distal tip and leads lepidotrichia development at its proximal part (Wood and Thorogood, 1984; Durán et al., 2011). Actinodin (And) 1 and 2 serve as non-collagenous components of actinotrichia in fin development, but they had been lost from the genome of the tetrapod lineage (Zhang et al., 2010; **Figure 3D**). Zhang and colleagues tested their function in zebrafish by morpholinomediated knockdown, resulting in the loss of fin rays from the pectoral fin. Given that these genes were lost from the tetrapod genome, they posited that the loss of And genes from the fish genome is likely to have promoted the fin-to-limb transition through the loss of lepidotrichia (Zhang et al., 2010; Lalonde et al., 2016). Currently, the analysis of regulatory mechanisms of And1 expression is ongoing (Lalonde et al., 2016). With the developmental pathways of fin rays, the mechanisms that initiate dermal fin ray differentiation in the fin fold remain enigmatic. The WNT and SHH signal from the epithelial cells stimulates cell differentiation during the regeneration of fin rays, but the genetic mechanisms that induce the differentiation of dermal bones in normal fin rays remain unknown (Quint et al., 2002; Wehner and Weidinger, 2015). Harris et al. screened eda (ectodysplasin) mutant fish, which show the drastic loss of dermal fin rays from paired and unpaired fins as well as the phenotypes of other dermal bones such as the loss of scales (Harris et al., 2008). Since eda is expressed in the epidermal placode and edar (ectodysplasin receptor) is in the basal cells of the forming placode in scale development, the EDA signaling pathway is likely to also be involved in fin ray development through epithelialmesenchymal interaction. Though it is a widely shared consensus that the apical fold [AF-the epithelial structure that forms after Apical Ectodermal Ridge (AER)] plays a critical role in fin fold development (Yano et al., 2012), little is known about how the AF interacts with the underlying mesenchyme and induces dermal ossification.

Due to the apparent loss of the autopod domain in Hoxa13 and Hoxd13 double knockout mice (Fromental-Ramain et al., 1996b), the function of HOX13 has been one of main foci in understanding the endochondral development of the limb bud (**Figure 3D**). The overexpression of HOXA13 in the limb bud alternated cell-cell adhesion and affected the size of cartilage condensation (Yokouchi et al., 1995). Knosp et al. identified Bmp2 and Bmp7 as direct targets of HOXA13 by using a ChIP assay (Knosp et al., 2004; **Figure 3D**). BMP signaling is well known to play key roles in cartilage and bone development, identities of digits, and interdigital apoptosis during limb development (Suzuki, 2013). Thus, they are likely to be directly involved in the differentiation process of endochondral bones at the downstream of HOXA13. HOXD13 is also suggested to bind the regulatory regions of Sfrp1, Barx1, and Fbn1, all of which are indispensable for normal skeletogenesis (Salsi et al., 2008). Recent advancements of technology in genomics have further accelerated extensive identification of the downstream genes of HOX13 groups. Whole genome tiling arrays and RNAsequencing explored whole gene expression profile including non-coding RNA in wild-type and Hox9-11 mutant mice (Gyurján et al., 2011; Raines et al., 2015). These experiments successfully identified a number of Hox downstream genes, including bone developmental factors such as Runx, in the limb development. ChIP-sequencing of HOXA13 and HOXD13 in cultured cells of chicken and mouse limb buds suggested that HOX13 and CTCF co-bind to a number of genomic loci (Beccari et al., 2016; Jerkovic et al., ´ 2017). Given that the CTCF transcription factor regulates 3D chromatin structures, this finding implies that HOX13 transcription factors not only bind directly to their target sites, but rather regulate chromatin structures broadly with CTCF, which consequently shifts the broad gene expression profile.

In parallel with the elucidation of the developmental programs of dermal fin rays and endochondral bones in fish, the evolutionary mechanisms from fins into limbs have been closely investigated (Sordino et al., 1995; Woltering and Duboule, 2010; Schneider and Shubin, 2013; Yano and Tamura, 2013; Woltering et al., 2014; Onimaru et al., 2015; Tanaka, 2018). Sordino et al. suggested that digits and wrists are novel domains of tetrapods due to a striking difference in the expression pattern of Hoxa-11 between fish and tetrapods (Sordino et al., 1995). Woltering et al. injected a tetraodon BAC vector containing hoxa13b region into mice and observed the expression pattern of hoxa13b, which is regulated by fish regulatory domains in mouse limbs. They observed that the expression domain of the fish hoxa13b is confined to a proximal domain of the limb bud of mice, not in a distal domain like mammal Hoxa13 (Woltering et al., 2014). Woltering et al. also state that while fish have most of the necessary genes and regulatory architecture indispensable to form digits, 5′ regulatory landscapes in fish cannot specify a distinct digit territory (Woltering et al., 2014). This suggests that distal pectoral fins and distal limb buds are not comparable in a classical sense of homology as there is a lack of a common ancestral structure. Freitas et al. overexpressed hoxd13a in order to investigate the function of 5′ hoxd expression in fin development and discovered that the overexpression of hoxd13a results in the proliferative expansion of chondrogenic tissue distally that is akin to autopod development (Freitas et al., 2012). Leite-Castro et al. also propose possible mechanisms of the fin-to-limb transition, a consequence of various modifications in HoxA genes, such as: expansion of polyalanine repeats within the HOXA11 and HOXA13 proteins, an acquisition of novel ncRNA with an inhibitory function of HOXA11 or cis-regulatory evolution of hoxa13 (Leite-Castro et al., 2016).

Contrary to the entrenched idea that dermal bones and endochondral bones are ontogenetically and histologically distinct, recently new evidence has implied a possible ontogenetic interchange between these two types of bones in appendage evolution. First, the development of dermal fin rays is unique; the gene expression profile during their development is at an intermediate state between endochondral and dermal bones in appendage evolution. Col2a1 and Col10a1, which are regarded as specific markers of endochondral bones, are expressed despite the absence of cells stained by cartilage staining in fin rays (Smith et al., 2006). Second, functional analyses of hox13 genes provide a new insight for development of the fin fold. Lalonde and Akimenko deleted hoxa13a/hoxd13a expressing cells and observed the defects in the formation of fin rays (Lalonde and Akimenko, 2018). Double knockout zebrafish of hoxa13a and a13b, and triple knockout fish of hoxa13a, a13b, and d13a lost dermal fin rays (Nakamura et al., 2016). These data demonstrated that hox13 genes are indispensable for fin ray development, whereas hoxa13 and d13 are necessary to develop endochondral digits in mice. Third, double knockout zebrafish of hoxa13a and a13b, and triple knockout fish of hoxa13a, a13b, and d13a increased the number of distal endochondral bones along the proximodistal axis with the loss of dermal fin rays (Nakamura et al., 2016). Also, the repetitive excisions of the AF, which is critical to develop fin rays, from the developing pectoral fin, extended the size of the endochondral disk distally (Yano et al., 2012). Summarizing all data leads us to a novel hypothesis the developmental program between dermal and endochondral programs are interchangeable and the dermal genetic network has been replaced by the endochondral network in appendage evolution (Nakamura et al., 2016; Tulenko et al., 2016, 2017; Paço and Freitas, 2018; **Figure 3E**). Given that LPM cells contribute to lepidotrichia, LPM cells that express hoxa13 genes are most likely to differentiate into dermal fin rays in fish, whereas LPM cells that experience hoxa13 and d13 develop endochondral digits in tetrapods. Further dissection of the cell origins and fate mapping of Hoxa13 cells in the fish fins and tetrapod limbs would provide us with more insights for the mechanisms of the fin-to-limb transition.

Though the genetic mechanisms underlying for interchanges between dermal and endochondral bones remain elusive, gradual losses of gene expression that are indispensable for the development of fin rays such as And1, are likely to play roles (Zhang et al., 2010). Masselink and colleagues' research would illuminate a path to approach the underlying mechanisms. They discovered that whereas ectodermal cells develop an AER which promotes tissue growth via Fgf8 and Wnt3 in tetrapods (Lewandoski et al., 2000; Barrow, 2003; Boulet et al., 2004), somitic cells contribute to the AF development (Masselink et al., 2016). They removed somite-derived cells by a genetically targeted cell ablation system before the AF induction, resulting in the severe disruption of AF development, as well as a lack of actinotrichia deposition. Furthermore, they marked somitederived cells via the introduction of the photoactive protein Kaede, and induced apoptosis by laser illumination. The removal of the somite-derived cells from the AF significantly decreased the size of the actinotrichia as well as a reduction by 30% in the length of the fin fold. They concluded that the swap of the cell contributions from somitic cells to ectodermal cells in the AF was likely to drive the evolutionary shift from the AF to the AER and, consequently, lead to the fin-to-limb transition. Extensive investigations and comparative studies of the cell lineages that contribute to the AF, AER, fin fold, and the endochondral disk would shed light on the evolutionary trajectories of appendages.

#### GENETICS AND GENOMICS INTO OLD QUESTIONS

Endless discoveries in paleontology have led to continuing scientific questions in vertebrate evolution. The skeletal shifts between dermal and endochondral bones are examples of major vertebrate transitions from water to land, yet revealing the genetic mechanisms underlying their evolution is a long endeavor in evolutionary biology. Integration of novel techniques in both molecular biology and paleontology would accelerate the understanding in developmental and evolutionary mechanisms of cranial dermal bones, shoulder girdles and appendages during the fish-to-tetrapod transition.

In contrast to the canonical hypotheses, new studies demonstrated that the cell origins of some dermal bones are obviously from the LPM and the PAM, not neural crest cells. Furthermore, developmental programs for dermal bones and endochondral bones are ontogenetically interchangeable in skull development by deploying distinct cell origins. They may be interchangeable even in a single cell population of appendage development, though further assessment is necessary. These new discoveries make our understanding of the border line between dermal and endochondral bones obscure; the two types of bones are more similar in terms of their ontogenetic history and characters than we expected. However, the current data has not been so abundant as to be conclusive. For example, the information of cell origins of dermal and endochondral bones in pectoral fin development are still fragmentary. Complete understanding of the cell lineages with genetic labeling in each ossification in wild-type and genetic mutants would provide us an opportunity to propose a more reliable model for the fin-to-limb transition. The most recent studies have elegantly showed that we are able to track the entire cell lineages of early vertebrate development (Briggs et al., 2018; Farrell et al., 2018; Wagner et al., 2018). Comparative analysis of the cell lineages of appendages between fish and tetrapods by deploying these state-of-the-art techniques would help us better our understanding of cell history in appendage development. Also, explicit understanding of genetic basis of dermal and endochondral interchanges needs a significant amount of future work. Previous studies have revealed molecular mechanisms of dermal and endochondral ossifications, yet, the genetic switches to determine dermal and endochondral is still obscure—one of the most critical questions in this field. The integrative approach of high-throughput comparative sequencing, such as RNA-sequencing or ChIP-sequencing, and subsequent functional tests would be a powerful means to discover the genetic loci that have been responsible for dermal-to-endochondral transitions.

Assimilating knowledge from the different structures would lead us one step closer to understanding the whole picture of the vertebrate skeletal shifts. Though the details of genetic mechanisms that regulate dermal and endochondral bones are disparate in each structure, the principal components in bone differentiation, such as Bmp, Sox9, Runx2 are clearly conserved. The distinct developmental programs in the skull, the shoulder, and fins are likely to be explained by genetic modifications or deployment of different molecules through a long journey of ancestral vertebrates. Future elaboration of molecular mechanisms in distinct types of ossification in multiple structures will shed light on the common and derived genetic mechanisms of dermal and endochondral development.

Embedding the knowledge about the developmental programs of bones into the background of paleontology with newly emerging tools, shifts our understanding of vertebrate evolution into a new era. Reevaluation of the mechanisms underlying the major skeletal shifts in vertebrates with genomics, genetics, and imaging techniques will cast a new light on the deep history of our ancestors.

#### AUTHOR CONTRIBUTIONS

TN served as the main author of the body of the manuscript. He also wrote the figure captions as well as completing parts the figure illustrations. TW drafted and made significant edits to the

#### REFERENCES


main content of the text, as well as editing the figure captions. He also contributed to the illustration of the figures.

#### ACKNOWLEDGMENTS

We would like to thank Gayani Senevirathne and Thomas A. Stewart for constructive comments which improved this manuscript substantially. This work was performed with the institutional support provided by the Rutgers University School of Arts and Sciences and the Human Genetics Institute of New Jersey and with 2016 Grant for Basic Science Research Projects (160335) provided by Sumitomo Foundation.

program to control facial skeletal development. Development 139, 709–719. doi: 10.1242/dev.073197


and changes cell adhesiveness in chick limb buds. Genes Dev. 9, 2509–2522. doi: 10.1101/gad.9.20.2509


homology of zebrafish swimbladder and mammalian lung. PLoS ONE 6:e24019. doi: 10.1371/journal.pone.0024019

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

Copyright © 2018 Wood and Nakamura. 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.

## Using Zebrafish to Study Collective Cell Migration in Development and Disease

#### Hannah M. Olson1,2 and Alex V. Nechiporuk <sup>1</sup> \*

*<sup>1</sup> Department Cell, Developmental & Cancer Biology, The Knight Cancer Institute, Oregon Health & Science University, Portland, OR, United States, <sup>2</sup> Neuroscience Graduate Program, Oregon Health & Science University, Portland, OR, United States*

Cellular migration is necessary for proper embryonic development as well as maintenance of adult health. Cells can migrate individually or in groups in a process known as collective cell migration. Collectively migrating cohorts maintain cell-cell contacts, group polarization, and exhibit coordinated behavior. This mode of migration is important during numerous developmental processes including tracheal branching, blood vessel sprouting, neural crest cell migration and others. In the adult, collective cell migration is important for proper wound healing and is often misappropriated during cancer cell invasion. A variety of genetic model systems are used to examine and define the cellular and molecular mechanisms behind collective cell migration including border cell migration and tracheal branching in *Drosophila melanogaster,* neural crest cell migration in chick and *Xenopus* embryos, and posterior lateral line primordium (pLLP) migration in zebrafish. The pLLP is a group of about 100 cells that begins migrating around 22 hours post-fertilization along the lateral aspect of the trunk of the developing embryo. During migration, clusters of cells are deposited from the trailing end of the pLLP; these ultimately differentiate into mechanosensory organs of the lateral line system. As zebrafish embryos are transparent during early development and the pLLP migrates close to the surface of the skin, this system can be easily visualized and manipulated *in vivo*. These advantages together with the amenity to advance genetic methods make the zebrafish pLLP one of the premier model systems for studying collective cell migration. This review will describe the cellular behaviors and signaling mechanisms of the pLLP and compare the pLLP to collective cell migration in other popular model systems. In addition, we will examine how this type of migration is hijacked by collectively invading cancer cells.

Keywords: collective cell migration, posterior lateral line, posterior lateral line primordium, collective cell invasion, cancer

### INTRODUCTION

Cellular migration is necessary both during development and adulthood and has been widely studied in populations of cells that migrate independently. However, cells can also migrate in groups in a process known as collective cell migration. During collective cell migration, cells maintain cell-cell contacts, exhibit both morphological and behavioral polarization and interact with neighboring cells within the collective to affect each others behavior. This process is important

#### Edited by:

*Gokhan Dalgin, University of Chicago, United States*

#### Reviewed by:

*Isaac Skromne, University of Richmond, United States Kandice Tanner, National Institutes of Health (NIH), United States*

> \*Correspondence: *Alex V. Nechiporuk nechipor@ohsu.edu*

#### Specialty section:

*This article was submitted to Molecular Medicine, a section of the journal Frontiers in Cell and Developmental Biology*

> Received: *01 May 2018* Accepted: *16 July 2018* Published: *17 August 2018*

#### Citation:

*Olson HM and Nechiporuk AV (2018) Using Zebrafish to Study Collective Cell Migration in Development and Disease. Front. Cell Dev. Biol. 6:83. doi: 10.3389/fcell.2018.00083* during the morphogenesis of multiple organ systems, as well as during wound healing in adults. In addition, invading cancer cells exhibit many hallmarks of collective cell migration.

Collectives can be organized in a variety of different forms, including loose chains or strands, tight clusters, tubes, or epithelial sheets (**Figure 1**). Neural crest cell migration is an example of chain migration (**Figure 1A**; Rupp and Kulesa, 2007). During migration, neural crest cells migrate out the neural tube in a chain like manner with the ultimate goal of reaching distant sites and differentiating into numerous cell types (Theveneau and Mayor, 2012). Throughout migration, neural crest cells maintain transient adherens junctions when briefly in contact with each other. Specifically, when two migrating neural crest cells touch, they induce contact inhibition of locomotion (Mayor and Carmona-Fontaine, 2010). This causes the two cells to retract cellular extensions at the site of contact and initiate new extensions on the opposing side of contact of the cell. This behavior restricts protrusions within the interior portion of the chain of migrating cells and promotes protrusive behavior along the edges of the chain, specifically the leading edge (Carmona-Fontaine et al., 2008). This behavior is thought to be important for directional migration and self-organization, and loss of this behavior has been shown in invasive cancer cells (Carmona-Fontaine et al., 2008; Astin et al., 2010).

Collectives can also migrate in a much more cohesive group, often referred to as cluster cell migration (**Figures 1B**, **2**). During migration of this type, cells maintain adherens junctions while migrating, thus remaining tightly connected. Examples of cluster cell migration include pLLP migration in zebrafish (**Figures 2**, **3**), border cell migration in D.melanogaster (**Figure 1B**), Kupffer vesicle organogenesis in zebrafish, and movement of invasive clusters of tumor cells (**Figure 6**). During border cell migration a group of cells delaminates from the follicular epithelium of the D.melanogaster egg chamber and migrates across the chamber toward the developing oocyte (Montell et al., 1992). During this migratory process, these cells maintain adherens junctions while migrating. Similarly, during Kupffer vesicle organogenesis, a group of around 20-30 cells cluster together and migrate cohesively. Finally, invasive groups of tumor cells often migrate as clusters during the invasion of many epithelial-based tumors (**Figure 6**; Freidl et al., 2004; Christiansen and Rajasekaran, 2006; Alexander et al., 2008). This type of collective invasion is discussed in more detail below.

Collective cell migration also contributes to the process of branching morphogenesis. This cellular behavior drives the formation of highly branched tubular structures including mammalian kidneys, lung, prostate and mammary gland as well as D.melanogaster trachea (**Figure 1C**; Sutherland et al., 1996; Ewald et al., 2008; Metzger et al., 2008). During branching morphogenesis, epithelial sheets reorganize into tubelike structures through multiple cellular behaviors, one of which is a specialized type of collective cell migration called invasive branching. During this process, extension of new branches is guided through invasive migratory behavior of a tip cell, which exhibits dynamic protrusive behavior. The cells that lag behind the tip cell are referred to as "stalk cells" and maintain cadherinmediated adhesion to each other as well as to the tip cell.

In contrast to branching morphogenesis, epithelial sheet migration involves collective movement of a leading cell front, rather than individual cells (**Figure 1D**). Epithelial sheet migration mediates wound healing in the adult, dorsal closure in D. melanogaster, and migration of germ layers in some animals (Martin and Parkhurst, 2004; Solnica-Krezel, 2005). This type of collective cell migration is also observed in vitro after a scratch wound assay of endothelial and epithelial cells **Figure 1D**. In this assay, a scratch is made across a confluent sheet of cells and sheet migration is observed to "heal" or repair the "wound," as cells migrate as a cohesive front to fill in the space that was created by the scratch. On a cellular level, sheet migration is characterized as a monolayer of migrating cells that maintain strong adherens and tight junctions as well as apico-basal polarity while migrating (Bahri et al., 2010). These junctions restrict movement within the sheet and limit cellular rearrangement (Zallen and Blankenship, 2008). Cells at the leading or "free edge" take on leader cell positions and extend actin-based cellular protrusions (**Figure 1D**; Poujade et al., 2007; Vitorino and Meyer, 2008; Omelchenko et al., 2014). Follower cells also exhibit protrusive behavior through the extension of cryptic lamellipodia, as seen during wound closure in MDCK cells in vitro (**Figure 1D**; Fenteany et al., 2000; Farooqui and Fenteany, 2005). These cryptic lamellipodia form at the basal side of the follower cells and protrude under cells in front. These lamellipodia are necessary for generating traction against the basal lamina (Fenteany et al., 2000; Farooqui and Fenteany, 2005).

From these examples, it is clear that collective cell migration is employed during the morphogenesis of multiple organ systems. Despite this diversity, collectives often employ conserved cellular strategies during the migratory process. Our understanding of these mechanisms comes from studying various models including pLLP migration. Here, we will first review the cellular and molecular mechanisms that drive pLLP migration. Then we will briefly review a few other examples of collective cell migration and compare the pLLP to these other forms of collective cell migration. Finally, we will discuss how mechanisms of collective cell migration can be misappropriated during cancer cell invasion.

### Zebrafish Posterior Lateral Line Primordium as a Model System to Understand Collective Cell Migration

Over the years a variety of model systems have been used to understand cellular and molecular mechanisms of collective cell migration. These range from the slime mold Dictyostelium discoideum to study movement of cell aggregates, to the mouse retina to investigate blood vessel branching. With zebrafish emerging as a genetic model system in the 1980s and 1990s, collective cell migration of the lateral line primordium during embryogenesis became one of the premier model systems to dissect collective cell migration. Lateral line research has a rich

**Abbreviations:** pLL, posterior Lateral Line; pLLP, posterior Lateral Line Primordium; NM, Neuromast.

delaminate away and migrate as chains toward the Cxcl12a source. Cells display transient adherens junctions. (B) Cluster cell migration of border cells in *Drosophila melanogaster.* Cells maintain tight adherens junctions while migrating with the leading cell in front exhibiting extensive protrusive behavior. These cells migrate toward the EGF/PVF-1 source. (C) Branching morphogenesis of *Drosophila melanogaster* trachea. While the leading cell migrates toward the source of Fgf, trailing cells form tube like structures. (D) Epithelial sheet migration-wound healing. Leading cells on either side of the wound migrate toward each other to close the wound. Leading cells extend filopodial protrusions toward each other. Follower cells extend cryptic lamellipodia underneath cells in front of them. Adherens junctions are maintained during migration.

history, as studies examining this system date back to the 1600s, when scholars first discovered that various fish species contain a row of small pores that stretch along the trunk of the fish from the head to the tail. At that time it was believed that this was a glandular system required for the secretion of mucous that covered the fish (Monro, 1785). However, in the 1800's that view was challenged when it discovered that this system is in fact a mechanosensory system, similar to the touch sensory system within the skin (Knox, 1825). Around 1850, Leydig discovered that within the row itself there were actually small sensory organs and in 1870, Schulze postulated that these sensory organs were actually similar to those within the inner ear and that movement of water stimulated their sensory capabilities (Leydig, 1850; Schulze, 1861, 1870). Up until this point however, the majority of research had focused on observing the system but not perturbing it. Fuchs (1894) was one of the first scholars to experimentally test the system and discovered that the lateral line in Torpedo only responded to changes in tactile stimulation but not in regards to changes in temperature or chemical stimulation (Fuchs, 1894). Finally, in 1904 Parker took a systematic approach to examine factors that stimulated lateral line sensory organs in eight different species of fish (Parker, 1904). He tested light, heat, salinity, food, oxygen, carbon dioxide, foulness of water, water current, water pressure, high frequency vibrations (hearing), and low frequency vibrations. Interestingly, he found that only vibrations of low frequency were sensed by the lateral line. He

FIGURE 2 | Posterior Lateral Line formation (pLL) and posterior Lateral Line Primordium (pLLP) migration. (A) pLLP begins migrating around 20 hours post-fertilization (hpf). (B) At 30 hpf the pLLP has migrated about half way down the trunk and deposited 3 neuromasts (NMs). (C) pLLP migration is complete at 48 hpf with the deposition of the terminal cluster of NMs.

concluded that the lateral line system was a mix of the touch (skin) sensory system and the hearing (ear) sensory system. Further studies confirmed these findings and expanded on them culminating in the conclusion that the lateral line in aquatic vertebrates is a mechanosensory system that detects changes in water current and is necessary for behaviors such as feeding and swimming (Montgomery et al., 2000).

Harrison (1904) was the first to determine that this mechanosensory system develops through the migration of a group of cells that deposits smaller clusters of cells while

migrating (Harrison, 1904). Specifically, Harrison used chimeric frog embryos to study posterior lateral line (pLL) development. In this study Harrison fused the head of a black tadpole to the tail of brown tadpole and observed pLL development. Surprisingly, he witnessed a dark streak (black cells) appear on the brown tadpoles tail and this streak separated into pigmented dots. From this he concluded that the lateral line develops by the concerted migration of a group of cells down the trunk of fish and amphibians. Further research in the 1920s and 1930s identified the origin of this group of cells. Stone (1933) discovered that this group of cells originates from the post auditory placode (Stone, 1933). Specifically, Stone stained salamander embryos with Nile blue sulfate and then grafted the post auditory placode from these stained donor embryos to unstained host embryos. Following transplantation he observed a group of blue stained cells migrating along the trunk of the embryo that deposited small clusters of blue stained cells (presumptive mechanosensory organs—neuromasts). These clusters then differentiated into the sensory organs that form the lateral line. These experiments identified the post auditory placode as the group of cells giving rise to the pLL. As evident from these classical experiments, the close proximity to the surface of the skin makes this an easily tractable system to study mechanisms of mechanosensation and collective cell migration.

### Posterior Lateral Line Development and Posterior Lateral Line Primordium Migration

Since the pLLP discovery in the early twentieth century, the migratory behavior of this system has been actively investigated in various aquatic species including zebrafish. In zebrafish, the posterior Lateral Line Primordium (pLLP) is a group of around 100 cells that migrates along the lateral aspect of the trunk of the embryo during embryogenesis (**Figure 2**). The pLLP and sensory neurons of the pLL ganglion are both derived from the pLL placode, a transient thickening of the embryonic ectoderm positioned caudally to the developing otic vesicle (**Figure 2**; Mizoguchi et al., 2011). Surprisingly, little is known about molecular pathways that regulate pLL placode induction and differentiation and this topic has been discussed elsewhere (Sarrazin et al., 2010; McCarroll et al., 2012; Piotrowski and Baker, 2014; Nikaido et al., 2017). At about 22 hours post fertilization (hpf), the distal portion of the pLL placode begins migrating along the lateral aspect of the trunk, whereas the proximal portion, comprised of sensory neurons, stays behind (**Figure 2A**). The pLLP continues migrating along the trunk of the zebrafish until it reaches the tip of the tail at 48 hpf (**Figure 2C**). As the pLLP migrates, it deposits clusters of about 20–30 cells from its trailing (caudal) end; these clusters will differentiate into mechanosensory neuromasts (NMs) (**Figures 2**, **3**). The pLLP also lays down a single line of inter-neuromast cells (**Figure 2**; Metcalfe et al., 1985), which are latent precursors that will differentiate into additional NMs during larval development. Migration of the pLLP is complete with the deposition of the terminal cluster, a group of two to three NMs that are located at the distal region of the trunk (**Figure 2C**). Rapid embryonic development, optical translucence, and genetic tractability make the zebrafish a particularly suitable model system to define cellular and molecular mechanisms of collective cell migration.

### How Is the Posterior Lateral Line Primordium Organized?

Cells within the pLLP display differential morphology and different states of differentiation depending on their location and can be generally designated as leader or follower cells. The leading third of the pLLP is comprised of 2–3 tip cells of mesenchymal character and less differentiated epithelial cells (**Figure 3**, Blue, Green). Follower cells within the trailing two-thirds of the pLLP (**Figure 3**, Yellow) form polarized rosettes. Tip cells exhibit flat mesenchymal morphology (**Figure 3**), display active protrusive behavior in their leading edge, and respond to guidance cues that steer the collective. Cells proximal to tip cells in the leading region, exhibit columnar epithelial morphology (**Figure 3**). Cells within the trailing region (last 2/3 of the pLLP) apically constrict to form epithelial rosette structures of the proto-NM (**Figure 3**; Lecaudey et al., 2008). The remaining cells in the trailing region contribute to the inter-neuromast cells and are deposited between the NMs. These cells are positioned on the periphery of the pLLP, surrounding the cells that have formed rosettes (Dalle Nogare et al., 2017).

In addition to differences in morphology, leader and follower cells within the pLLP show differences in their fate. As the pLLP migrates it deposits proto-NMs every 5–7 somites. At the onset of migration the pLLP contains 2 to 3 proto-NMs; however, by the end of migration the pLL consists of 5–6 NMs and the terminal cluster of NMs (**Figure 2**). Thus, new proto-NMs must be generated during pLLP migration. These new proto-NMs are generated by cell proliferation throughout the pLLP (Dalle Nogare et al., 2017). Newly generated cells within the leading region of the pLLP differentiate last and ultimately contribute to the terminal cluster of NMs, while those in the trailing region begin differentiating into cells that will contribute to more proximal NMs and inter-neuromast cells (Dalle Nogare et al., 2017). As a new proto-NM begins differentiating, cells undergo apical constriction to form a rosette-like structure that constitutes the proto-NM (**Figure 3**).

### Chemokine Signaling During Posterior Lateral Line Primordium Migration

Similarly to neural crest cell migration, the pLLP uses Cxcl12a as a chemotactic cue. The pLLP migrates along the myoseptum of the zebrafish embryo where cxcl12a is uniformly expressed (**Figure 4B**; David et al., 2002; Li et al., 2004; Haas and Gilmour, 2006; Dambly-Chaudiere et al., 2007; Valentin et al., 2007). Loss of Cxcl12a leads to a failure of migration (David et al., 2002; Valentin et al., 2007), whereas ectopic expression of cxcl12a results in a redirection of pLLP toward the "new" Cxcl12a source (Li et al., 2004).

While pLLP can migrate toward an ectopic source of the ligand, two recent studies demonstrated that the Cxcl12a does not present as a gradient along the trunk (Donà et al., 2013; Venkiteswaran et al., 2013). Instead, the pLLP produces an

internal gradient of Cxcl12a through differential expression of two chemokine receptors, cxcr4b and cxcr7b. cxcr4b is expressed within the leading region whereas cxcr7b is expressed within the trailing region (**Figure 4B**; Haas and Gilmour, 2006; Dambly-Chaudiere et al., 2007; Valentin et al., 2007). Loss of either Cxcr4b or Cxcr7b leads to a failure in migration (Haas and Gilmour, 2006; Valentin et al., 2007) indicating the necessity of both receptors to ensure proper pLLP migration. Additionally, wildtype cells transplanted to the leading region of cxcr4b mutant pLLP can rescue migratory defects in these mutants (Haas and Gilmour, 2006). This is also true for wild-type cells transplanted to the trailing region of cxcr7b mutants (Valentin et al., 2007). However, when wild-type cells are transplanted to the leading region in cxcr4b mutants, impaired migration is not rescued. Finally, when cxcr7b mutant cells are transplanted into the leading region of cxcr4b mutants, migration is rescued (Valentin et al., 2007). Altogether, these transplantation experiments underscore the necessity for region-specific distribution of Cxcr4b and Cxcr7b during pLLP migration. Two recent studies used live imaging to visualize chemokine-receptor internalization to demonstrate that Cxcl12a binds to Cxcr7b and then is internalized with the Cxcr7b receptor (Donà et al., 2013; Venkiteswaran et al., 2013). In doing so, this creates an internal gradient of Cxcl12a, with low levels of Cxcl12a in the trailing region and high levels of Cxcl12a in the leading region. Previous reports support this model as Cxcr7b acts as a ligand sink in other contexts (Dambly-Chaudiere et al., 2007; Boldajipour et al., 2008; Naumann et al., 2010; Mahabaleshwar et al., 2012).

At this point, it is unclear how this region specific expression of cxcr4b and cxcr7b in the leading and trailing region arises. Aman and Piotrowski (2008) argued that cxcr7b expression is downstream of two signaling pathways active within the pLLP, Wnt (leading region) and Fgf (trailing region) signaling (**Figure 4A**). When Wnt was constitutively active or Fgf was inhibited there was a reduction in cxcr7b expression. Additionally, while inhibition of Wnt signaling had no effect of cxcr4b expression it did lead to an expansion of cxcr7b expression into the leading region. However, a separate study did not report any effect on cxcr4b or cxcr7b expression in the absence of Wnt signaling (Valdivia et al., 2011). It should be also noted that expression of chemokine receptors does not mirror Wnt and Fgf signaling domains, suggesting that these receptors are not directly regulated by these signaling pathways (**Figure 4**). Further experiments are needed to determine how chemokine receptor expression is regulated during pLLP migration.

### Signaling Pathways Within the Posterior Lateral Line Primordium

Within the pLLP there are a number of signaling pathways that regulate patterning, maintain migratory behavior, and initiate proto-NM differentiation. Among the main pathways are the canonical Wnt signaling pathway (**Figure 4**; Aman and Piotrowski, 2008), active in the leading region of the pLLP, the Fgf pathway (**Figure 4**; Lecaudey et al., 2008; Nechiporuk and Raible, 2008), active in the trailing region of the pLLP, and the Notch-Delta pathway active in the forming proto-NM in the trailing region (**Figure 5**; Matsuda and Chitnis, 2010). Canonical Wnt signaling maintains proliferation, patterning, and migration of the pLLP. Additionally, canonical Wnt signaling is necessary for initiating expression of fgf3 and 10a ligands in the leading cells (Aman and Piotrowski, 2008; Matsuda et al., 2013). These ligands activate Fgf signaling in the trailing region and initiate proto-NM differentiation. Wnt signaling also initiates the expression of dusp6 and sef, which are inhibitors of Fgf signaling, allowing for restriction of the Wnt signaling domain to the leading region (Aman and Piotrowski, 2008). Furthermore, Fgf signaling also induces expression of dkk1 and dkk2 (**Figure 3**; Aman and Piotrowski, 2008; McGraw et al.,

initiates apical constriction of cells through reorganization of the actin cytoskeleton. (B) proto-NM maturation signaling. Fgf luminal signaling initiates *atoh1a* and *notch3* expression. Atoh1a induces expression of *deltaA* and *fgf10a.* DeltaA interaction with Notch3 initiates lateral inhibition allowing for *atoh1a* expression to be localized to the central cell (hair cell precursor) and surrounding cells to remain as supporting cells.

2014), both of which are inhibitors of Wnt signaling to restrict Fgf signaling to the trailing region. Thus, both Wnt and Fgf mutually inhibit each other to generate region specific signaling domains (**Figure 4A**).

#### Canonical Wnt Signaling Is Necessary for Posterior Lateral Line Primordium Migration and Proto-NM Formation

As mentioned above, canonical Wnt signaling is active in the leading third of the pLLP and is necessary for pLLP migration and patterning. Global inactivation of Wnt signaling during pLLP migration causes a loss of proliferation, cell death, and a loss of patterning within the pLLP (Aman and Piotrowski, 2008; McGraw et al., 2011). Additionally, overexpression of Wnt signaling leads to overproliferation and premature termination of the pLLP (Aman and Piotrowski, 2008). Interestingly, exclusive loss of Lef1, a downstream effector of canonical Wnt signaling, causes defects in migration and cellular behavior within the leading region with no effects on cell death or proliferation (McGraw et al., 2011). Specifically, cells in the leading region of the pLLP are prematurely incorporated into NMs and deposited early. This gradual depletion of cells in the leading region ultimately results in a dispersal of the pLLP and an absence of the terminal cluster of NMs (McGraw et al., 2011; Valdivia et al., 2011). In summary, results from many studies highlight the significance of canonical Wnt signaling in regulating multiple cellular behaviors within the pLLP; however, how these behaviors are executed downstream of Wnt is not well understood.

## Fgf Signaling Is Necessary for Neuromast Formation and Differentiation

The Fgf signaling pathway functions downstream of Wnt and is necessary for proto-NM formation (**Figure 4A**; Lecaudey et al., 2008; Nechiporuk and Raible, 2008; Chitnis et al., 2012). Blocking or reducing Fgf activity inhibits the formation of rosettes and ultimately halts pLLP migration (Lecaudey et al., 2008; Nechiporuk and Raible, 2008; Chitnis et al., 2012;), whereas ectopic expression of Fgf leads to the formation of additional rosettes (Lecaudey et al., 2008).

On a cellular level, Fgf signaling promotes the shape change of epithelial cells from a columnar to an apically constricted morphology during the formation of rosettes or proto-NMs (**Figures 3**, **4A**; Lecaudey et al., 2008; Nechiporuk and Raible, 2008; Chitnis et al., 2012). Two studies published in the same year presented complimentary findings related to the intracellular pathway that drives apical constriction of pLLP cells. In the first study, researchers found that Fgf signaling activates Ras-MAPK signaling which induces Rock2a localization to the apical portion of cells where it activates myosin regulatory light chain and induces apical constriction (**Figure 5A**; Harding and Nechiporuk, 2012). The second study demonstrated that Fgf signaling transcriptionally regulates Schroom3, a scaffold protein that binds to Rock and has been shown to activate apical constriction in other contexts (**Figure 5A**; Ernst et al., 2012).

#### Neuromast Maturation

Neuromasts are the sensory organs that comprise the pLL. These sensory organs are composed of hair cells and supporting cells. Hair cells lie within the middle of the neuromast with

clusters are not known (question marks). It has been speculated that behaviors of such clusters exhibit many similarities to leading front cells present in collective cell migration during embryogenesis. Adherens junctions are maintained throughout invasion process.

support cells surrounding the hair cells. When hair cell bundles are deflected by changes in water current, this information is mechanotransduced through the hair cell and then transmitted back to the brain where it is further processed.

Differentiation into hair and supporting cell precursors occurs during pLLP migration and is driven by Fgf signaling. In addition to its role in apical constriction, Fgf signaling also initiates expression of atoh1a (**Figure 5B**), a transcription factor that is a master regulator of hair cell fate and thus its activation initiates a hair cell program in a small subset (1 to 2 cells) of cells within a forming proto-NM (Sarrazin et al., 2006; Nechiporuk and Raible, 2008). atoh1a expression and action is limited to a single focus through Notch-Delta lateral inhibition. atoh1a expression induces expression of deltaA, whereas expression of its receptor notch3 is induced by Fgf signaling within the forming proto-NM. Therefore, the DeltaA ligand (driven through the Atoh1a transcriptional program) interacts with the Notch3 receptor (driven by Fgf signaling) on neighboring cells within the proto-NM and inhibits expression of atoh1a, in neighboring cells therefore specifying them as supporting cells (**Figure 5B**; Itoh and Chitnis, 2001; Matsuda and Chitnis, 2010). If Notch3 is blocked, proto-NMs generate more hair cells at the expense of supporting cells (Matsuda and Chitnis, 2010). Through this mechanism, atoh1a expression is restricted to the central cell, inducing hair cell progenitor fate in that cell.

Atoh1a also induces expression of both the fgf10a ligand and atoh1b (Millimaki et al., 2007; Matsuda and Chitnis, 2010) within the same hair cell precursor. Expression of fgf10a from the central hair cell progenitor initiates a new Fgf signaling center within the trailing region that promotes maturation of proto-NMs (**Figure 5B**; Matsuda and Chitnis, 2010). Atoh1b maintains atoh1a expression within the central proto-NM and inhibition of Atoh1b results in a reduction in atoh1a expression (Millimaki et al., 2007; Matsuda and Chitnis, 2010). Notably, Fgf ligands accumulate in a microlumen at the apical center of the rosette (Durdu et al., 2014). Inhibiting the formation of the microlumen by knockout of Schroom3 results in a reduction of Fgf response in cells that comprise rosettes. This suggests that the microlumen acts to coordinate Fgf signaling among cells of the rosette during migration. In summary, Fgf signaling plays a key role in both the establishment of hair cell precursors as well as maturation of proto-NMs.

### Cell-Cell Adhesion and Cytoskeletal Regulation During Posterior Lateral Line Primordium Migration

During pLLP migration cells remain in a close contact as they cohesively migrate along the trunk of the zebrafish. Cells within the cluster are connected by cadherin mediated adherens junctions. Specifically, E-cadherin and N-cadherin are both expressed in the pLLP but show specific localization within the proto-NM. In the proto-NM, N-cadherin is expressed in both the hair cell progenitor cell (central cell of the proto-NM) and supporting cells, whereas E-cadherin is only expressed in the hair cell progenitor cell (Matsuda and Chitnis, 2010). However, Revenu et al., reported in 2014 that E-cadherin and Ncadherin are expressed in all cells within the pLLP as evidenced by antibody staining against both cadherins (Revenu et al., 2014). Revenu et al. also examined the maturation of adherens junctions using a BAC fluorescent reporter of N-cadherin as N-cadherin shows enhanced localization at apical junctions. Specifically, they reported a role for N-cadherin in initiating the change in morphology from mesenchymal in the leading cells to columnar epithelial in more trailing cells. The authors showed that N-cadherin clusters first and then epithelial columnar reorganization follows. Finally, using tandem fluorescent protein timers, a readout of protein turnover, the authors determined that the N-cadherin localized at apical junctions is more stable than N-cadherin localized at the basolateral membrane. Moreover, the apical junctions become progressively more stable from the leading to the trailing region.

Adherens junctions between cells often trigger activation of intracellular signaling pathways via various binding partners such as catenins. Recently, one of the catenins expressed in the pLLP, Catenin Delta 1, was shown to regulate Rac1 signaling in cultured cells (Mizoguchi et al., 2017). Specifically, mutation of Mib1, an E3 ubiquitin ligase, caused an accumulation of Catenin Delta 1 and hyperactivation of Rac1, which in turn induced ectopic, random non-persistent protrusions and ultimately impaired migration of cultured cells (Mizoguchi et al., 2017). The authors showed that Mib1 is required for pLLP migration and normal protrusive behaviors of pLLP cells. However, it is not clear if Mib1 also regulates Rac1 activity in the pLLP, similar to the in vitro model. In fact, not much is known about how Rac1 and other modulators of protrusive activity are regulated during pLLP migration.

Below we will compare mechanisms active during pLLP migration to other examples of collective cell migration and then focus on how collectively invading cancer cells subvert these mechanisms to invade surrounding tissues.

### Distribution of Leaders and Followers Within Collectives

Similar to the pLLP, most collectives show division into two different populations of cells, leaders and followers (**Figure 1**). Leaders are the cells that detect and sense chemotactic cues, exhibit active protrusive behavior, and produce molecular or mechanical cues to guide the trailing population to the proper destination. While leader cells exhibit common behaviors, follower cells have diverse functions and fates depending on the context. For example, follower cells contribute to trachea bronchi or blood vessels during branching morphogenesis whereas during pLLP migration follower cells ultimately contribute the sensory organs of the lateral line system. Leaders and followers show differences in morphology with leaders displaying mesenchymal morphology and in some contexts follower cells displaying polarization. Interestingly, the assignment as leader or follower is not always permanent during this migratory process. In border cell migration for example, the leader cell rotates as the cell with the highest levels of RTK/MAPK signaling acquires the leader cell position (Bianco et al., 2007). Another example in which leader cell identity is not maintained throughout migration is during neural crest cell migration (Kuriyama et al., 2014). Whereas it is not exactly clear what determines neural crest cell leader position, leader cells maintain higher levels of RAC1 activity and exhibit greater protrusive behavior than follower cells (Theveneau et al., 2010). Further, during branching morphogenesis of the trachea in D. melanogaster, cells that receive the highest level of the breathless (Fgf) signal take on the role of the leader cell (Caussinus et al., 2008; Lebreton and Casanova, 2014).

Although in many examples of collective cell migration, it is not uncommon for cells to switch positions during migration, this is not observed in wild-type pLLPs. In some instances cells within the pLLP can be forced to move into new positions as a result of experimental manipulation. Haas and Gilmour (2006) found that when wild-type cells were transplanted into cxcr4b mutant embryos, they often ended up at the leading edge of the pLLP. Live imaging of chimeric pLLP revealed that this resulted from the tumbling behavior exhibited by wild-type cells in cxcr4b mutant pLLPs. They suggested that the tumbling behavior exposes wild-type cells to the Cxcl12a signal and "captures" these cells at the leading edge of the pLLP. Once wild-type cells are captured in the leading region, the pLLP commences normal migration and proper deposition of NMs. A similar tumbling behavior was observed in chimeric primordia that contained lef1 mutant cells (McGraw et al., 2011). When lef1 mutant cells were too close to the leading edge (a couple of cell diameters), the chimeric pLLP stopped and tumbled until mutant cells were "pushed back" and excluded from this leading region (McGraw et al., 2011). Once this occurred, the pLLP resumed its migration. The reason for this tumbling behavior is not clear and it is unlikely that this results from lef1 mutant cells being unable to sense the chemokine, as lef1 mutants still express normal levels of the cxcr4b receptor. Although collectives determine leading and following positions differently, the division of collectives into these two populations allows for cohesive migration of these cohorts of cells.

### Chemotactic Cues Guide Collectives During Migration

In order for collectives to migrate in a directional manner they must respond to specific chemotactic cues. Usually, these cues appear as a gradient, with the collective migrating to the highest levels of chemokine or secreted ligand. Leading cells sense these cues and in turn change their behavior, morphology, and protrusive behavior to respond to the cue appropriately (Sutherland et al., 1996; Haas and Gilmour, 2006; Prasad and Montell, 2007). This is often achieved by regulating proteins that remodel the cytoskeleton. For example, RAC1 activation in leader cells in response to guidance cues has been observed in numerous examples of collective cell migration including border cell migration, D.melanogaster trachea formation, and neural crest cell migration (Murphy and Montell, 1996; Chihara et al., 2003; Theveneau et al., 2010; Scarpa et al., 2015).

Despite differences in organization of collectives and environmental contexts through which they migrate, some collectives utilize common guidance cues during migration. For example both neural crest cells in Xenopus and the cells within pLLP respond to Cxcl12a during migration (**Figures 1A**, **4B**; David et al., 2002; Li et al., 2004; Theveneau et al., 2010). Overexpression of Cxcl12a during neural crest cell migration leads to aberrant migration whereas loss of Cxcl12a during pLLP migration results in inhibition of migration (Valentin et al., 2007; Olesnicky Killian et al., 2009). Additionally, in both of these examples Cxcl12a is uniformly expressed and self-generated gradients are produced by the collectives. During neural crest cell migration, neural crest cells migrate toward the epibranchial placodes, the source of the Cxcl12a (Theveneau and Mayor, 2013). Neural crest cells use contact inhibition of locomotion to facilitate proper migratory behavior. Specifically, neural crest cells extend protrusions that interact with the placodes during the onsent of migration. This induces a repulsive response by placodal cells in which focal adhesions are dissassembled and placodal cells migrate away. Neural crest cells then migrate toward the placodal cells again and migration occurs in a "chase and run" manner (Theveneau et al., 2013).

In the case of pLLP migration, Cxcl12a is expressed uniformly throughout the myoseptum of the zebrafish (**Figure 4B**; David et al., 2002; Li et al., 2004). In this case, domain specific expression of Cxcl12a receptors, cxcr4b and cxcr7b, produces a local gradient within the pLLP. cxcr4b is expressed in the leading region, whereas cxcr7b is expressed in the trailing region. Cxcr4b acts as the chemoreceptor initiating a G-protein signaling cascade within the leading cells whereas expression of cxcr7b in the trailing region acts as a ligand sink creating a gradient within the pLLP itself (Haas and Gilmour, 2006; Dambly-Chaudiere et al., 2007; Valentin et al., 2007). These differences in the ligand binding and ligand-receptor turnover lead to a gradient of Cxcl12a response within the pLLP.

Other examples of different guidance cues used during the migration of collectives include EGF and PVF-1, two molecules within the developing D. melanogaster oocyte that are necessary for proper border cell migration (**Figure 1C**). Mutation of either EGFR or PVR results in uncontrolled protrusive behavior and defects in migration (Prasad and Montell, 2007). Finally, Fibroblast Growth Factor (FGF) is used as a chemotactic cue during branching morphogenesis in the trachea, mammary gland, and lung (**Figure 1D**; Sutherland et al., 1996; Ewald et al., 2008; Metzger et al., 2008) as well as in other examples of collective cell migration such as nephric duct migration, wound healing, and endothelial cell migration (Werner et al., 1992; Vitorino and Meyer, 2008; Attia et al., 2015). In summary, collectives recognize a variety of guidance cues and employ diverse strategies as to interpret these cues and maintain their migratory behaviors.

### Cell-Cell Junctions and Cell-ECM Interactions During Collective Cell Migration

In order for cells to migrate cohesively as a group during collective cell migration, cells must communicate often through stable or transient cell-cell junctions. These junctions usually consist of cadherins, desmosomes, and tight junction proteins; loss of these structures often leads to improper or failed migration. Cadherin junctions are the most prevalent junctions observed during collective cell migration. For example, cellcell junctions between migrating border cells and nurse cells are mediated by the transient presence of E-cadherin and loss of E-cadherin results in decreased protrusion formation at the front of the cluster and ultimately impaired migration (Niewiadomska et al., 1999; Geisbrecht and Montell, 2002; Cai et al., 2014). Additionally, during neural crest cell migration in Xenopus, Cadherin-11 is necessary for contact inhibition of locomotion, which allows for processive and directed migration of neural crest cells as described above (Becker et al., 2014). Loss of Cadherin-11 results in non-directional migration and impaired adhesive ability (Becker et al., 2014). In addition to cadherin-based cell-cell junctions, desmosomes and tight junctions are also observed during collective cell migration. In wound healing, both desomosomal-junctions and tight junctions are necessary for proper healing (Danjo and Gipson, 1998; Shaw and Martin, 2009). Knockdown of either desomosomal or tight junction proteins leads to decreases in cell migration velocity and ultimately impairment of migration to close the wound (Bazellières et al., 2015). Desomosomal and tight junctions are also observed during mammary gland morphogenesis. Interestingly, these two junctional complexes show differences in cellular localization within the developing mammary gland (Shamir and Ewald, 2015). Tight junctions are only seen at the apical portion of cells that face lumens, whereas desomosomal junctions connect interior portions of cells in the mammary duct (Shamir and Ewald, 2015).

During migration, cells need to generate force to processively migrate toward their destination. To achieve this, cells adhere to the extracellular matrix as well as use supracellular organization to generate force between the leaders and the followers. To connect with the extracellular matrix cells utilize integrins, which link the intracellular cytoskeleton to the extracellular matrix. This allows transduction of mechanical signals as well as force generation through the recruitment of cytoskeletal adaptor proteins, which couple the integrins and their extracellular binding partners (Nobes and Hall, 1999; Zaidel-Bar et al., 2007). During wound healing, integrin-mediated signaling induces cytoskeletal rearrangements that initiate leader cell properties at the wound edge (Etienne-Manneville and Hall, 2001). These leader cells use the integrins α2β1, α5β1, and αvβ3 to generate force on a collagen substrate and initiate movement to close in the wound (Grose et al., 2002). Similarly, β1 integrins are used during angiogenesis to couple the extracellular matrix to the cytoskeleton within the endothelial collective. β1 integrins activate guanine nucleotide exchange factors (GEFs) for Cdc42 and Rac1 as well as kinases such as Src and FAK (Lamalice et al., 2007; Osmani et al., 2010) to promote protrusive behavior at the leading edge of tip cells, which is necessary for appropriate migratory behavior (Scales and Parsons, 2011; Lawson and Burridge, 2014).

In addition to integrin-mediated force generation, collectives also maintain collective movement and force generation through supracellular cytoskeletal organization. Specifically, focal adhesions at the leading front edge associate with actin-myosin cables, which initiate contraction and force generation. These actin-myosin cables ultimately extend through multiple layers of follower cells allowing for force generation throughout the entire collective instead of the first row of cells (Li et al., 2012; Reffay et al., 2014). However, during collective cell migration the largest forces are generated at the leading edge of collectives with a decrease in force strength in back of the collective (du Roure et al., 2005; Trepat et al., 2009; Tambe et al., 2011; Anon et al., 2012; Cai et al., 2014). For example, during border cell migration, tension decreases at the back of the cluster (Cai et al., 2014). A similar distribution of force is observed in wound healing, where the greatest traction forces are exhibited at the leading edge of the monolayer but forces are maintained among follower cells in the collective (du Roure et al., 2005; Trepat et al., 2009; Tambe et al., 2011; Anon et al., 2012).

In addition to adhering to the extracellular matrix for force generation, cells in collectives also remodel the extracellular matrix while migrating. Migrating and/or surrounding cells deposit new basement membrane to form migrating tracks. The deposition of a new basement membrane forms a smooth surface to promote migration. During neural crest cell migration in both Xenopus and chick embryos, neighboring cells deposit fibronectin along neural crest migrating streams to facilitate migration (Alfandari et al., 2003). Similarly, astrocytes underlying the migrating endothelial cells secrete fibronectin during angiogenesis (Stenzel et al., 2011). Fibronectin then induces tip cell filopodia promoting migratory behavior (Stenzel et al., 2011). When fibronectin is specifically deleted from the astrocytes, endothelial cells show defects in migration (Stenzel et al., 2011). In addition, the basement membrane that is produced by endothelial cells and pericytes during angiogenesis helps stabilize the migrating blood vessels (Eming et al., 2007). The deposition of new basement membrane allows for migrating collectives to migrate through the path of least resistance.

### How Does Collective Cell Migration Influence Our Understanding of Invasive Cancer?

The dogma surrounding cancer cell invasion for many years was that single cells would detach from cancerous tumors, enter the blood stream and metastasize in other tissues. However, in the last 50 years that viewpoint has been gradually expanded and it is now widely recognized that in many cases clusters of cells can also detach from the primary tumor to initiate metastasis (**Figure 6**).

It was first reported more than half a century ago that cancer cell metastasis could be associated with both single and clusters of tumor cells (presumably detached from the primary tumor) found in blood samples of patients (**Figure 6**) (Zeidman and Buss, 1952). Subsequent research indicated tumor cell clusters were better at initiating metastasis than single cells when intravenously injected into mice (Liotta et al., 1976). Further studies confirmed this observation and revealed that tumor cell clusters are actually 20–30-fold better at initiating metastasis in vitro (Cheung et al., 2016) and in vivo (Hou et al., 2012; Maddipati and Stanger, 2015; Cheung et al., 2016).

Research over the last two decades strongly suggests that collective cell invasion mediates metastasis in numerous epithelial cancers including prostate, pancreatic, lung, colorectal and breast cancer (Friedl et al., 1995; Nabeshima et al., 2000; Hegerfeldt et al., 2002; Aceto et al., 2014; Gundem et al., 2015; Maddipati and Stanger, 2015; Cheung et al., 2016). However, only recently studies provided most rigorous evidence for collective cell invasion mechanisms in both humans and mouse models. Recent advances in microfluidics and next generation nucleic acid sequencing allowed for isolation and interrogation of a small number of circulating tumor microclusters. In humans, RNA sequencing of circulating cancer cell clusters in comparison to circulating single cancer cells (Aceto et al., 2014) revealed a small subset of differentially expressed genes. One of the transcripts enriched in microclusters was Plakoglobin, a component of desmosomes and adherens junctions. This study revealed that Plakoglobin plays a role in maintaining adherens junctions in cancer cell clusters and thus enhances their metastatic potential. Interestingly, Plakoglobin has been shown to be important in focal adhesions during collective cell migration of mesendoderm in Xenopus (Bjerke et al., 2014). Another study investigating breast cancer invasion in a mouse model, found that Keratin14 is upregulated in circulating tumor cell clusters and lung metastasis (Cheung et al., 2016). When Keratin14 was depleted, it disrupted the metastases markers Tenascin C, Jagged 1, and Epiregulin. As Keratin14 is also enriched in desmosomes, this study further emphasizes a critical role for cell-cell adhesions during tumor cluster invasion.

Interestingly, these invasive fronts share molecular similarities to collective cell migration observed during embryonic development. For example, the leading edge of invasive cancer clusters share common molecular properties observed at the leading edges of collectives including cell-cell junctions and cell adhesion receptors (**Figure 6**) (Freidl et al., 2004; Christiansen and Rajasekaran, 2006; Alexander et al., 2008). In addition, collectively invading cancer cells use self-generated gradients to promote migration. Similar to self-generated gradients by border cells and the pLLP, a study examining invasion of melanoma cells identified a mechanism by which tumor cells generated a gradient of lysophosphatidic acid (LPA) (Muinonen-Martin et al., 2014) and used this gradient to promote cancer cell invasion. In this study, tumor cells acted as a ligand sink by breaking down LPA into byproducts. This created a gradient in which LPA was high in surrounding tissues but low within the tumor itself. Melanoma tumors then used this gradient to migrate to higher sources of LPA in the surrounding tissue, initiating metastasis.

Common signaling pathways used during development as well as collective cell migration are often reactivated during collective invasion (Korc and Friesel, 2009; Katoh, 2017; Bach et al., 2018). For example, many downstream components of the canonical Wnt signaling pathway are misregulated in a variety of cancers (hepatocellular, colorectal, orapharyngeal squamous cell carcinoma) and their associated metastasis including Axins, βcatenin, and TCF/Lef1 transcription factors (**Figure 6**) (Lammi et al., 2004; Salahshor and Woodgett, 2005; Marvin et al., 2011; Papagerakis et al., 2012). Notably, Lef1 is active at the leading front of invasive lung and colorectal cancers, similar to canonical Wnt signaling being active within the leading region of the pLLP (Nguyen et al., 2009; Wang et al., 2013). In addition to mutations in canonical Wnt signaling, other signaling pathways active during pLLP migration are also implicated in certain types of metastatic cancer. Mutations in the Notch/Delta pathway have been associated with poor prognosis in colorectal and breast cancer as activation of this pathway is associated with metastasis. Although it is not clear whether this pathway is involved in collective cell invasion (Leong et al., 2007; Wang et al., 2013).

Cancer cells also make use of chemotactic signals during metastases. In particular, numerous invasive cancers show abnormal expression of the chemokine receptors CXCR4 and CXCR7 as well as the ligand CXCL12 both within the tumors and at potential sites of metastases. For example in breast cancer, tumor cells express high levels of CXCR4 and metastatic target tissues (lung, liver, bone) express high levels of the ligand CXCL12 (**Figure 6**) (Wang et al., 2013; Wu et al., 2015). Further, high levels of CXCR4 and CXCR7 are associated with shorter survival times than those with low levels (Wu et al., 2015). Although, this association does not seem to hold true for other types of cancers. In a pancreatic cancer in vivo mouse model, cells producing CXCL12 showed deficits in migration and poor metastatic potential in comparison to control cells producing no CXCL12 (Roy et al., 2014). Based on this evidence it is possible that the CXCL12 chemokine may act differently in various cancer contexts. Despite the known prevalence of mutations within these signaling pathways, the mechanisms by which these mutations induce and promote or inhibit collective invasion and metastases remain unknown. Understanding how these signaling pathways regulate collective cell migration of the pLLP may provide clues as to how these pathways are hijacked during cancer invasion.

Based on the similarities between collective invasion and collective cell migration, we can use models of collective cell migration during development to discern mechanisms used by tumor clusters during metastasis. For example, as Lef1 is upregulated at the invasive fronts of both lung and colorectal invasive cancers and canonical Wnt signaling via Lef1 is active in the leading region of the pLLP we can use the leading region of the pLLP as a model for collective cancer invasion. We can study cellular adhesion, protrusive behavior, and cell-ECM interactions using the pLLP model to identify cellular mechanisms that promote cancer front migration and metastasis. Identification of cellular pathways that act downstream of Lef1 in the pLLP may provide clues as to how these factors are misregulated during invasive cancers that show increased Lef1 expression at their

#### REFERENCES


leading edge. Thus, further insights gained through studies of pLLP leading edge behavior could provide insight into how these invasive clusters promote metastasis.

#### CONCLUSIONS

Collective cell migration is a widely used developmental process that initiates and promotes morphogenesis of many different organ systems. While collectives are organized into a variety of different forms, they often share similar cellular strategies. Collectives are guided by leading cells that sense and respond to the extracellular environment, specifically chemotactic cues. These chemotactic cues are then transmitted through specific signaling pathways to initiate molecular changes that guide migration as well as differentiation. Insights gained from studying mechanisms of collective cell migration can be used to identify mechanisms by which invasive cancers hijack developmental machinery to promote metastasis.

#### AUTHOR CONTRIBUTIONS

HO wrote the manuscript. AN edited the manuscript.

#### FUNDING

Research in the AN Laboratory is supported by funds from the National Institutes of Health and Presidential Bridge Funding from Oregon Health and Science University.

pak is required for recruitment of the Scribble complex and septate junction formation. Development 137, 2023–2032. doi: 10.1242/dev.045088


Cancer Res. 12, 731–733.

be construed as a potential conflict of interest.

Copyright © 2018 Olson and Nechiporuk. 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.

# Zebrafish as a Model for Obesity and Diabetes

#### Liqing Zang1,2, Lisette A. Maddison<sup>3</sup> and Wenbiao Chen<sup>1</sup> \*

<sup>1</sup> Department of Molecular Physiology and Biophysics, Vanderbilt University School of Medicine, Nashville, TN, United States, <sup>2</sup> Graduate School of Regional Innovation Studies, Mie University, Tsu, Japan, <sup>3</sup> Center for Reproductive Biology, Washington State University, Pullman, WA, United States

Obesity and diabetes now considered global epidemics. The prevalence rates of diabetes are increasing in parallel with the rates of obesity and the strong connection between these two diseases has been coined as "diabesity." The health risks of overweight or obesity include Type 2 diabetes mellitus (T2DM), coronary heart disease and cancer of numerous organs. Both obesity and diabetes are complex diseases that involve the interaction of genetics and environmental factors. The underlying pathogenesis of obesity and diabetes are not well understood and further research is needed for pharmacological and surgical management. Consequently, the use of animal models of obesity and/or diabetes is important for both improving the understanding of these diseases and to identify and develop effective treatments. Zebrafish is an attractive model system for studying metabolic diseases because of the functional conservation in lipid metabolism, adipose biology, pancreas structure, and glucose homeostasis. It is also suited for identification of novel targets associated with the risk and treatment of obesity and diabetes in humans. In this review, we highlight studies using zebrafish to model metabolic diseases, and discuss the advantages and disadvantages of studying pathologies associated with obesity and diabetes in zebrafish.

Keywords: zebrafish, obesity, diabetes, transgenic models, disease models, animal

## INTRODUCTION

The prevalence of overweight and obesity has steadily increased worldwide in the past several decades. In 2016, more than 1.9 billion adults were overweight, and of these over 650 million were obese (WHO, 2017). This is primarily due to excess food consumption (Vandevijvere et al., 2015). Overweight and obesity are major risk factors for numerous chronic diseases, including cardiovascular diseases, diabetes, and certain types of cancer (Haslam and James, 2005). In the United States, class III obese individuals (BMI ≥ 40 kg/m<sup>2</sup> ) have a six-fold increase in diabetes risk over normal-weight individuals (Leung et al., 2017) and more than 90% of people with type 2 diabetes mellitus (T2DM) are overweight or obese. The global increase of overweight and obesity largely explains the incidence and prevalence of type 2 diabetes over the past 20 years. Obesity and T2DM can substantially decrease life expectancy, diminish quality of life, and impose a large economic burden to society (Leung et al., 2017).

Both obesity and T2DM have high heritability (Poulsen et al., 2001; Willemsen et al., 2015). Recent genome wide association studies and whole exome sequencing studies have identified a large number of genetic variants that are associated with overweight/obesity and/or T2DM (Lawlor et al., 2017; Loos, 2018). In most cases, however, the causative genes for these linked variants are

#### Edited by:

Ryan M. Anderson, Indiana University, Purdue University Indianapolis, United States

#### Reviewed by:

Marta Letizia Hribal, Università Degli Studi Magna Græcia di Catanzaro, Italy Anne-Francoise Burnol, Institut National de la Santé et de la Recherche Médicale (INSERM), France

\*Correspondence:

Wenbiao Chen wenbiao.chen@Vanderbilt.Edu

#### Specialty section:

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

> Received: 02 April 2018 Accepted: 25 July 2018 Published: 20 August 2018

#### Citation:

Zang L, Maddison LA and Chen W (2018) Zebrafish as a Model for Obesity and Diabetes. Front. Cell Dev. Biol. 6:91. doi: 10.3389/fcell.2018.00091 uncertain and the mechanism by which these variants contribute to the disease phenotypes is unclear (Loos, 2018). Furthermore, the aggregate effect of all the variants only account for a small fraction of the heritability of these conditions (Fuchsberger et al., 2016). It is likely that more alleles are yet to be discovered to play a role in obesity and T2DM susceptibility.

The wealth of human genetic and epidemiological data on obesity and T2DM provides ample opportunity for mechanistic investigations in animal models. Zebrafish is a well-established model system for developmental biology, human genetics, and human diseases (Dooley and Zon, 2000; Gibert et al., 2013; Freifeld et al., 2017). Several features have propelled zebrafish to its current prominence in developmental biology and disease modeling. It is a vertebrate, having high degree of genetic, anatomical, and physiological similarities to humans. It is fecund, easy to maintain in large number and has a relative short generation time, allowing facile genetic, and chemical genetic screens (Kimmel et al., 1995; MacRae and Peterson, 2015). Its external development affords easy accessibility to embryonic and genetic manipulations (Kimura et al., 2014; Hoshijima et al., 2016; Yin et al., 2016). The optical transparency of its embryos permits time lapse live imaging (Hall et al., 2009; Herrgen et al., 2009; Feierstein et al., 2015). Although traditionally used for developmental biology, zebrafish has recently been used to investigate metabolic diseases. Here, we will review some of the recent studies using zebrafish to model human metabolic diseases, with an emphasis on obesity, and diabetes. We will discuss the advantages and disadvantages of studying pathologies associated with obesity and diabetes in zebrafish.

## ZEBRAFISH OBESITY MODELS

### Lipid Metabolism and Adipose Biology in Zebrafish

Obesity is a consequence of positive energy balance. Regulation of energy intake and expenditure involves many organ systems including the brain, intestines, skeletal muscle, and adipose tissue (Cai, 2013; Dailey, 2014; Periasamy et al., 2017). Therefore, whole animal models are essential for better understanding of the development and progression of metabolic dysfunction. Zebrafish is an excellent model in which to study metabolic dysfunction because they have the key organs that are important for regulation of energy homeostasis and metabolism in mammals, including digestive organs, adipose tissues, and skeletal muscle (Lieschke and Currie, 2007; Schlegel and Stainier, 2007). The key functions such as appetite regulation, insulin regulation and lipid storage are also well conserved (Elo et al., 2007; Flynn et al., 2009; Nishio et al., 2012). Similar to mammals, excess nutrients in zebrafish cause increased plasma triglyceride levels and hepatic steatosis (Oka et al., 2010). Obese zebrafish also exhibit dysregulation of pathways that control lipid metabolism, including SREBF1, PPARs, NR1H3, and LEP (Oka et al., 2010). The conservation of these metabolic pathways that play key roles in adipocyte differentiation, energy homeostasis (Den Broeder et al., 2015), and cholesterol metabolism (Schlombs et al., 2003) demonstrates zebrafish as a suitable model for human lipid metabolism. However, zebrafish is an ectotherm species and its metabolic rate is not regulated by environmental temperature. Consistent with this, zebrafish does not have brown adipocyte tissues (BAT).

A primary characteristic of obesity is adipose hypertrophy and hyperplasia. Zebrafish have multiple adipose tissue depots and their development has been characterized (Flynn et al., 2009). Neutral lipid droplets first appear in visceral adipocytes and accumulate as zebrafish grow. Similar to mammalian white adipose tissue (WAT), early-stage zebrafish adipocytes contain multiple small lipid droplets while mature zebrafish adipocytes have a single large lipid droplet. As occurs in mammals, the adipocyte lineage expresses pparg, and fabp4 (Flynn et al., 2009). Visceral adiposity is a critical risk factor for T2DM and other metabolic diseases (Ahima and Lazar, 2013). In zebrafish, like in mammals, lipids are stored in visceral, intramuscular and subcutaneous adipocyte depots (Song and Cone, 2007), providing the opportunity to understand the regulation of body fat distribution. The high degree of conservation in distribution and formation of adipose tissue in the zebrafish compared to mammals makes it an appropriate model to study obesity.

### Methods to Quantitate Adiposity in Zebrafish

Quantitative measures of adiposity are important to assess the degree of obesity-related metabolic derangements. Body mass index (BMI) and quantitative computed tomography (CT) are widely used measurements of adiposity in humans but are more difficult to apply in zebrafish. Commonly used lipophilic dyes for visualizing lipids in histological sections and cultured cells, including Nile red, Oil red O, and Sudan black B, have been utilized to detect lipids in adult zebrafish sections and fixed zebrafish larvae (Marza et al., 2005; Schlegel and Stainier, 2006). With the optical transparency of zebrafish larvae, live-imaging, and fluorescence based screens have been developed for the study of digestive physiology or lipid metabolism. In particular, Nile red has been used for live imaging and quantification of intracellular neutral lipid droplets (Greenspan et al., 1985) as well as for purification of adipocyte tissues (Jones et al., 2008; Flynn et al., 2009; Oka et al., 2010). In addition, a variety of fluorescent lipid analogs and tracers are available, including BODIPY Fatty Acid Analogs, BODIPY-cholesterol analogs and fluorescence reports like PED6, for tracking the metabolism and distribution of exogenous lipids in live zebrafish (Hölttä-Vuori et al., 2010; Anderson et al., 2011). 3D micro-CT is also available for this small animal and allows volume measurement of total adipocyte tissue as well as different fat depots (Hasumura et al., 2012; **Figure 1**). Recently, Landgraf et al. compared the methodology of quantify zebrafish body fat mass using MR images (MRI) and EchoMRI 4in1 (EchoMRITM; Landgraf et al., 2017). The body fat mass of 8 adult male zebrafish was measured using the two methods and the two techniques showed high correlation. Overall, these methods provide accurate measurements of zebrafish adiposity and provide means for longitudinal monitoring.

A number of obesity models have been developed in zebrafish using diet and genetic manipulations (**Table 1**). The next section describes the details of these models.

### High-Fat Diet/Over-Nutrition Induced Obesity

A common approach to induce obesity is excess fat intake. Obese zebrafish can be conveniently produced by overfeeding starting from the onset of feeding at 5 dpf (day post fertilization). This is advantageous over rodent models since diet can only be manipulated after weaning, which is at least 3 weeks after birth. Although early larvae have no WAT (Imrie and Sadler, 2010), lipid droplets accumulate in the blood stream and other measures such as whole-larval triacylglycerol level may be used as an indicator for quantifying obesity progression (Schlegel and Stainier, 2006; Tingaud-Sequeira et al., 2011). Although heavy cream has been used (Schlegel and Stainier, 2006), chicken egg yolk solution is the most used high-fat diet for zebrafish larvae and juveniles (Tingaud-Sequeira et al., 2011; Zhou et al., 2015; Kopp et al., 2016; den Broeder et al., 2017). The high fat diet rapidly increases zebrafish adiposity.

Adult zebrafish have also been used as an obesity model. The first diet-induced obese (DIO) zebrafish model was reported in 2010 (Oka et al., 2010) where adult fish (3.5 months of age) were fed with 60 mg or 5 mg of freshly hatched Artemia per day for 8 weeks (150 calories vs. 20 calories). The overfed zebrafish exhibited increased BMI, hypertriglyceridemia and hepatosteatosis compared to the normal fed zebrafish. Male and female zebrafish showed similar responses to over-nutrition treatment. Furthermore, comparative transcriptome analysis of visceral adipose tissue among zebrafish, mouse, rat and human revealed that lipid metabolism networks of zebrafish are similar to those of mammals.

Besides overfeeding with Artemia, several other methods have been used to generate obese zebrafish. Meguro et al. developed custom high-fat zebrafish diets containing 20% corn oil or lard. They demonstrated that these high-fat diets make zebrafish obese (Meguro et al., 2015). Similarly, overfeeding the combination of a commercial tropical fish flakes (Tetramin) and 20% crude vegetable oil to zebrafish for 256 days triggers cardiovascular overload (Vargas and Vásquez, 2017).

It is interesting to note that obesity induced by overfeeding normal fat diet differs from that induced by high fat diet. Landgraf et al. compared the metabolic phenotype of obesity induced by overfeeding of a normal fat diet (NFD; Artemia cysts, 22% fat) to that by high fat diet (HFD; egg yolk powder, 59% fat). Although both increase adiposity, fish with NFD-induced obesity are metabolically healthy. In contrast, fish with HFD-induced obesity are metabolically unhealthy, with glucose intolerance, fatty liver, and preferential increase of visceral fat (Landgraf et al., 2017). This is consistent with the "obesity paradox" observed in humans. Overweight and obesity are not always associated with insulin resistance, the major driver and precursor of T2DM. In fact, overweight has also been paradoxically associated with lower mortality (Kokkinos et al., 2009; Flegal et al., 2013). This is thought to be due to the difference in the fat distribution pattern. In general, the visceral fat mass is a better predictor for insulin resistance and T2DM than BMI (Lebovitz and Banerji, 2005; InterAct et al., 2012). Therefore, the pathophysiological consequences of different fat depots are also conserved in zebrafish.

#### Genetic Models of Obesity

Obesity is a complex disease that results of an interaction between genetic and environmental factors. Genetically-modified animal models that reflect human obesity pathology are needed for understanding the physiological and genetic basis of obesity and for the development of pharmaceuticals to treat obesity. Genetic models of obesity have been characterized in zebrafish including transgenic lines expressing obesogenic genes or mutants from targeted mutagenesis and genetic screens.

#### TABLE 1 | Zebrafish obesity models.


Underscoring the conservation of metabolic regulation, genetic manipulation of pathways that control body weight in mammalian systems also causes obesity in zebrafish. Transgenic zebrafish obesity models are often generated by mimicking existing mammalian models. The central melanocortin system (CMS), including peptides derived from proopiomelanocortin (POMC), their receptors (MC3R and MC4R), and Agoutirelated peptide (AgRP), regulates energy homeostasis in zebrafish as it does in mammals (Ringholm et al., 2002; Hansen et al., 2003). Suppression of central melanocortin receptors by ectopic expression of the hair follicle restricted Agouti due to chromosomal translocations underlies one of the classical obese mouse models, Agouti Yellow (Bultman et al., 1992; Lu et al., 1994). This led to the identification of the endogenous melanocortin antagonist AgRP, whose transgenic overexpression in brain also causes obesity (Ollmann et al., 1997). In the zebrafish a genetic model of obesity has been developed by overexpressing AgRP [Tg(b-actin:AgRP)] (Song and Cone, 2007). These transgenic zebrafish exhibit an increase in body weight, linear growth, visceral adipose accumulation, and total triglycerides in all stages. The increased adiposity results from both hypertrophic and hyperplastic growth of adipocytes (Song and Cone, 2007). This transgenic zebrafish model demonstrates that central regulation of metabolism is conserved. The microRNA miR-27b has been suggested to be a regulatory hub for lipid metabolism by inhibiting the translation of a number of key lipid-metabolism genes (Vickers et al., 2013). Although several lines of evidence from cell culture support a role of miR-27b in lipid metabolism (Karbiener et al., 2009; Kang et al., 2013), there was a lack of in vivo supporting evidence. Recently, Hsu et al. generated transgenic zebrafish lines to deplete miR-27b by expressing a miR-27b sponge (C27bSP) driven by the ubiquitous betaactin promoter (bC27bSP) or the hepatocyte-specific fabp10 promoter (hC27bSP). They demonstrated that the transgenic fish display hyperlipidemia, hepatic steatosis and increased white adipose tissue mass (Hsu et al., 2017), supporting a role of miR-27b in lipid metabolism in vivo. Another obesogenic model, Tg(krt4Hsa.myrAkt1)cy<sup>18</sup> , was initially generated to study skin cancer. However, the transgenic adults were found to be obese, with increased BMI, adipocyte hyperplasia, abnormal fat deposition, and glucose intolerance. These phenotypes likely result from ectopic expression of the constitutively active human AKT1 in several mesenchymal derived tissues (Chu et al., 2012). Tg(krt4Hsa.myrAkt1)cy<sup>18</sup> fish differ from Tg(b-actin:AgRP) in two aspects. First, Tg(krt4Hsa.myrAkt1)cy<sup>18</sup> fish do not display adipocyte hypertrophy. Second, Tg(krt4Hsa.myrAkt1)cy<sup>18</sup> fish exhibit ectopic lipoma-like adipose tissue in dorsal muscle, gill arches, and tail bone tissues, whereas Tg(b-actin:AgRP) fish show a normal distribution of adipocyte tissues.

Multiple mutant lines have been identified that reveal genes and pathways contributing to lipid metabolism and adipose tissue regulation. These mutants are identified often because they have fatty liver at larval stages. Still other mutants are identified as adults due to increased adiposity. Although these mutants may share a common phenotype, they are due to the disruption of a diverse number of processes. Many mutants have larval hepatic steatosis, or fatty liver due to ER stress. In a "shelf screen" for liver size, Foie gras, and cdipthi559Tg/<sup>+</sup> were identified because they displayed fatty liver by 5 days of age (Sadler et al., 2005; Thakur et al., 2011). The affected gene product in cdipthi559Tg/+, Cdipt, is necessary for phosphatidylinositol synthesis and lack of phosphatidylinositol causes ER stress and lipid accumulation in hepatocytes (Thakur et al., 2011). The Foie gras mutant results from mutation in transport protein particle 11 (trappc 11) that encodes a protein critical for ER to Golgi vesicular transport. As a result, the mutation causes pathogenic ER stress in hepatocytes, leading to fatty liver (DeRossi et al., 2016).

Beyond ER stress, other pathways leading to fatty liver have been identified using staining of never-fed mutant larvae with lipophilic dyes (Schlegel and Stainier, 2006; Kim et al., 2015; Hugo and Schlegel, 2017). One of these mutants, harvest moon (hmn), results from a mutation in glutamine-fructose-6 phosphate transamidase (gfpt1) gene (Hugo and Schlegel, 2017), while another mutant, 7466mu110, is caused by a mutation in vacuole membrane protein 1 (vmp1; Kim et al., 2015). These mutants feature lipid accumulation in hepatocytes and increased whole body adiposity and may provide clues to pathogenesis of fatty liver. Another mutant with fatty liver, ducttrip (dtp), was identified in a screen for mutations affecting the development of exocrine pancreas. The dtp mutant stems from a mutation in the gene encoding S-adenosylhomocysteine hydrolase and larvae exhibit mitochondrial dysfunction and liver degeneration in addition to hepatic steatosis and disrupted exocrine pancreas (Yee et al., 2005). While most fatty liver mutants do not survive to adulthood there are exceptions such as red moon (rmn) where both larvae and adults exhibit increased liver neutral lipids. The mutant is due to a loss-of-function of β-hydroxybutyrate transporter (slc16a6a; Hugo et al., 2012) and fatty liver results from the diversion of entrapped ketogenic precursor into lipids. Furthermore, the mutants are less tolerant of starvation. This mutant thus reveals a role of ketone body export in fasting energy homeostasis (Hugo et al., 2012). This diverse group of mutants highlights the complex regulation of lipid metabolism and how disruption at one node can lead to a phenotype of hepatic steatosis.

Other zebrafish mutants have alterations in adipose tissue. The zebrafish vizzini mutant exhibits decreased somatic growth and increased subcutaneous and visceral adipose tissues relative to body size. In vizzini, the subcutaneous adipose tissue (SAT) lipid droplets are extremely large although the number of lipid droplets in adipocytes is unchanged (McMenamin et al., 2013). This is due to a mutation in growth hormone 1 gene (gh1) leading to a premature stop codon. The phenotype is consistent with GH-deficient mice and humans that develop enlarged volume of SAT (Li et al., 1990; Wabitsch et al., 1995). Mutations in cyp2r1 gene (Peng et al., 2017) also results in growth retardation and increased adiposity. These cyp2r1 mutants are deficient in 1α,25(OH)2D3, the principal active form of vitamin D3, and 25(OH)D3 treatment rescues the growth and adiposity defects. In mammals, genetic and epidemiological data suggest a role of vitamin D deficiency in obesity, but vitamin D supplement fails to reduce the risk of obesity and associated pathologies (Rosen et al., 2012). These zebrafish mutants support a role of vitamin D in lipid metabolism and distribution. Mechanisms underlying distribution of adipose are critical as visceral fat is a better predictor than BMI of risk for cardiovascular diseases, insulin resistance and T2DM (Lebovitz and Banerji, 2005; InterAct et al., 2012). The loss of plexin d1 function in zebrafish specifically impacts visceral fat (Minchin et al., 2015). PLEXIN D1 is one of the 67 genes identified in GWAS analyses in humans to be associated with visceral fat mass (Shungin et al., 2015), but the function of PLEXIN D1 in the distribution of fat mass was unknown. In zebrafish plxnd1 mutants, visceral fat is reduced due to a decrease of lipid droplet size and adipocyte hyperplasia. Consequently, with high fat diet, the mutants preferentially store lipid in subcutaneous adipose tissue and are protected from developing insulin resistance (Minchin et al., 2015).

Taken together, these transgenic and mutant zebrafish lines further demonstrate conserved regulation of lipid metabolism. They also provide models in which to address mechanistic understanding of the underlying phenotypes.

#### Utilities of Zebrafish Obesity Models

One advantage of zebrafish models is the amenability for quick identification of chemical and genetic modifiers of the phenotype. The diet-induced obesity models have been used to test the effects of some dietary supplements on body fat accumulation. Green tea extract inhibited lipid accumulation (Tainaka et al., 2011; Meguro et al., 2015; Meguro and Hasumura, 2018), by decreasing the visceral adipose tissue volume and altering the expression of lipid catabolism genes (Hasumura et al., 2012). Eriocitrin, an antioxidative flavonoid in lemon, showed lipid-lowering effects in DIO zebrafish similar to that reported in a high-fat diet in rats (Hiramitsu et al., 2014). Oral administration of Yuzu (Citrus junos Siebold ex Tanaka) peel to DIO zebrafish exhibited antiobesity effects by activating hepatic PPARα and adipocyte PPARγ pathways (Zang et al., 2014). Rhamnan sulfate, a sulphated polysaccharide from a marine green alga (Monostroma nitidum), also attenuated hepatic steatosis by suppressing lipogenesis (Zang et al., 2015a). Recently, a natural polyphenol, resveratrol, was reported to have anti-obesity effects via regulating lipid metabolism (Ran et al., 2017). Overall, DIO zebrafish is an attractive model system to evaluate the effects of functional foods and compounds on obesity development and treatment.

The diet-induced obesity models have also been used in drug testing. Tingaud-Sequeira et al. assessed the effect of small molecules on the whole-body adiposity after 24-h fasting in larvae previously overfed with egg yolk power (Tingaud-Sequeira et al., 2011). They found that two PPARγ agonists, rosiglitazone and TBT, a biocide found in antifouling paints, increases adiposity by inducing adipocyte hypertrophy and are thus obesogenic. In contrast, a PPARγ antagonist and an α1 adrenergic receptor agonist, known to promote lipolysis, are antiobesogenic. Zhou et al. also performed proof of principle drug testing experiments using a similar model and found that all the 5 human hypolipidemic drugs exhibit significant hypolipidemic effect in zebrafish as they do in humans (Zhou et al., 2015). These results demonstrate the value of zebrafish obesity model on drug screening.

Genetic zebrafish models of obesity also provide mechanistic insights into the underlying causes. For instance, studies in the cyp2r1 mutants that show increased adiposity and growth retardation identified pgc1a as a direct target for vitamin D receptor. As Pgc1a is a known master regulator of mitochondrial biogenesis, the study further showed that the increased adiposity results from impaired mitochondrial function (Peng et al., 2017). Similarly, using the plxnd1 mutants, Minchin et al. investigated transcriptional changes in extracellular matrix genes (Minchin et al., 2015). They found that the mRNA and protein product of col5a1 was increased and the visceral fat in the plxnd1 mutants had more pronounced fibrillogenesis. Knocking down col5a1 normalized the defects in visceral fat.

One cautionary note is that not all of the lipid metabolism genes are highly conserved in sequence and function in zebrafish. For example, the leptin protein of zebrafish is only 19% identical to the human protein. In mice and humans, leptin is an adipostatic hormone that regulates adipose mass, and failure of leptin signaling results in hyperphagia and obesity (Myers et al., 2010). Unlike mammals, leptin, and leptin receptor are not expressed in adipose tissue in zebrafish. Leptin receptor-deficient zebrafish primarily have disrupted glucose homeostasis (Michel et al., 2016), which is different from phenotypes observed in mouse models such as severe hyperphagia, hyperlipidemia and morbid obesity (Yen et al., 1976).

These examples illustrate the utility of zebrafish models for mechanistic investigations, drug testing and drug discovery in obesity and lipid metabolism. Thus far, the power these models remain largely untapped. It is anticipated that more mechanistic discoveries will be made from these and other zebrafish obesity models.

### ZEBRAFISH DIABETES MODELS

#### Pancreas Development and Glucose Homeostasis in Zebrafish

The morphogenesis and basic cellular architecture of zebrafish pancreas is similar to mammalian pancreas (Tehrani and Lin, 2011) with both exocrine and endocrine compartments. The endocrine compartment consists of glucagon-secreting α-cells, insulin-producing β-cells, somatostatin-producing δ-cells, ghrelin-producing ε-cells and pancreatic polypeptide producing PP-cells. These cells are arranged in a manner similar to mouse islets (Argenton et al., 1999; Biemar et al., 2001). The signaling pathways and mechanisms of zebrafish endocrine pancreas development are highly homologous to those of mammals (Kinkel and Prince, 2009). In addition to the pancreas, development and function of other organ systems involved in glucose homeostasis, including brain, liver, adipocyte tissue and skeletal muscle, are also conserved (Maddison and Chen, 2017). The conservation of the pancreas structure and glucose homeostasis system make zebrafish useful to identify novel targets in pancreas related diseases such as diabetes.

### Tools to Studies Glucose Homeostasis in Zebrafish

Numerous transgenic zebrafish strains with a fluorescent protein expression have been widely used to study pancreas development and glucose homeostasis in a whole living vertebrate (Kinkel and Prince, 2009; Tiso et al., 2009; Prince et al., 2017). For example, Tg(-1.2ins:EGFP) transgenic lines, where GFP expression is driven by the zebrafish preproinsulin promoter, provide a convenient fluorescent marker of β-cells (Xu et al., 2010) and insulin-expressing cells of the pancreatic islets can be visualized under a fluorescent microscope. Additionally, a transgenic line, Tg(gcga:GFP), where GFP is driven by zebrafish preproglucagon promotor, marks pancreatic α-cell (Zecchin et al., 2007). Using these cell-specific transgenic lines, the β-cell and α-cell area and total numbers are easily measured to evaluate alterations in cell mass and number, which is a predictor for glucose clearance (Li et al., 2015; Maddison et al., 2015).

Methods for zebrafish pancreas function have been established, including fasting and postprandial glucose measurement, and intraperitoneal glucose tolerance tests as well as techniques for pancreas dissection and islet cell culture (Eames et al., 2010; Eames Nalle et al., 2017). In larvae, blood collection for glucose measurement is not a viable methodology but free glucose in whole larvae can be measured by a coupledenzyme fluorescent assay (Jurczyk et al., 2011). For adult zebrafish, the small size (3–4 cm) makes blood collection challenging. Nevertheless, several protocols for blood collection have been developed, such as lateral incision in the region of the dorsal aorta, decapitation and tail ablation although these methods require sacrifice of the animal (Jagadeeswaran et al., 1999; Eames et al., 2010; Velasco-Santamaría et al., 2011). However, a method for repeated blood collection in the same individual adult zebrafish has been developed (Zang et al., 2013, 2015b). Blood glucose can be measured by hand-held glucose-meters (Eames et al., 2010; Zang et al., 2015b). Protocols for glucose tolerance test (GTT) have also been developed in zebrafish (Kinkel et al., 2010; Matsuda et al., 2017; Zang et al., 2017), which is the most used approach to diagnose diabetes mellitus or glucose intolerance in humans.

Measuring insulin and insulin function has presented more of a challenge. As surrogate indicators, insulin mRNA levels can be determined directly by qPCR (Michel et al., 2016) and insulin promoter activity may be determined indirectly by measuring EGFP signal intensity in Tg(−1.0ins:EGFP)sc<sup>1</sup> zebrafish (Zang et al., 2017). An insulin antibody for immunostaining both in whole larvae or adult zebrafish histologic sections is also available (Kimmel et al., 2015). Semi-quantitative dot-blot has been used to compare insulin levels in different fish simultaneously (Olsen et al., 2012). But insulin release has yet to be reliably measured in zebrafish. GFP has been used to replace the C-peptide of proinsulin in a transgenic line as one method to measure insulin release (Eames et al., 2013). Phosphorylation of Akt has been used to assess insulin function as a method to investigate early stage insulin resistance (Maddison et al., 2015; Landgraf et al., 2017). Insulin sensitivity can also be assessed by intraperitoneal injection of insulin in hyperglycemic zebrafish (Capiotti et al., 2014; Maddison et al., 2015).

Much of the biology in glucose homeostasis, from genes to organs, is conserved from zebrafish to humans. The application of powerful live imaging in zebrafish, coupled with genetic, and chemical genetic manipulations, will likely yield insights to many outstanding questions in diabetes.

#### Zebrafish Diabetes Models

**Table 2** summaries zebrafish diabetes models developed by diet and genetic manipulations.

#### Type 1 Diabetes Mellitus Models

Type 1 diabetes mellitus (T1DM) is primarily an autoimmune disease caused by destruction of insulin producing pancreatic β-cells. Although an autoimmune derived T1DM model is lacking in zebrafish, several models have been developed using targeted β-cell damage. Three methods of β-cell destruction have been applied: surgical removal, chemical-dependent ablation, and genetic ablation. Pancreatectomy is feasible under the microscope in transgenic zebrafish with islet specific expression of GFP (Moss et al., 2009; Delaspre et al., 2015). However, this method is technically difficult and is not commonly used in zebrafish. Chemical-induced diabetes is widely used in rodents and also in zebrafish. Intraperitoneal injection of streptozotocin (STZ) is effective at β-cell ablation in adult zebrafish and eventually causes elevated fasting blood glucose and reduced insulin levels (Moss et al., 2009; Olsen et al., 2010; Intine et al., 2013). A total of 6 administrations of STZ within 4 weeks induces stable hyperglycemia and diabetic complications including retinopathy, nephropathy, and impaired fin regeneration. Alloxan can also selectively kill β-cells in zebrafish larvae (Nam et al., 2015; Castañeda et al., 2017). However, these compounds also exhibit other toxicity. Multiple genetic model of T1DM have been reported. Although stable expression of cell-lethal diphtheria toxin A chain (DTA) can eliminate all β-cells, these fish have growth retardation and fail to thrive (Ninov et al., 2013). Therefore, inducible β-cell ablation has been the preferred method for modeling T1DM. Two approaches of inducible β-cell ablation have been reported. In one approach, transgenic zebrafish lines with β-cell specific expression of the bacterial nitroreductase (NTR) enzyme, are exposed to the prodrug metronidazole (MTZ), the NTR substrate, which is converted into a cytotoxic compound that rapidly induces β-cell apoptosis (Curado et al., 2007, 2008; Pisharath et al., 2007; Ye et al., 2015). This NTR/MTZ ablation system is used for β-cell regeneration research as the elimination of β-cell occurs in 18–24 h after MTZ administration and recovers within 3–4 days after MTZ washout. A different approach has been to use a combinatorial, inducible transgene where the insulin promoter drives the expression of a doxycycline/ecdysone-dependent transcription factor and the TetOR-based promoter to express activated human Bid that triggers apoptosis (Li et al., 2014). Ablation models all face the same hurdle in that zebrafish have a remarkable regenerative capacity and β-cell mass is restored once the ablation mechanism is removed.

#### Type 2 Diabetes Mellitus Models

T2DM is characterized by insulin resistance and β-cell dysfunction. β-cell death may also occur in long standing T2DM. Both nutritional and genetic approaches have been used to generate T2DM models in zebrafish. Immersion of zebrafish in glucose solution is a widely-used method because of its convenience. Immersing adult zebrafish into alternating concentrations of 0 and 2% glucose every other day for 28–30 days, or chronic exposure to 2% glucose solution for 14 days, induces diabetic phenotypes, including elevated blood glucose levels and impaired response to exogenous insulin (Gleeson et al., 2007; Alvarez et al., 2010; Capiotti et al., 2014), similar to mice following 6 weeks of high-galactose diet (Joussen et al., 2009). Young zebrafish (4–11 months) acclimate to glucose exposure better than older zebrafish (1–3 years), but persistent hyperglycemia, can be achieved even in young zebrafish by gradually increasing the glucose concentration (Connaughton et al., 2016).

Obesity is the major risk factor for T2DM. High-fat diet causes both obesity and T2DM in rodent models (Winzell and Ahrén, 2004). In zebrafish, overfeeding with a commercial food quickly caused insulin resistance, elevated fasting blood glucose, and impaired glucose tolerance (Zang et al., 2017). Calorie restriction


TABLE

2


diabetic

models.

and anti-diabetic drugs (metformin and glimepiride) ameliorated the hyperglycemia in the overfed zebrafish. These drugs are both frequently prescribed treatments for T2DM and their effectiveness in the zebrafish model demonstrates conservation in glucose homeostasis pathways.

Insulin resistance is a major driver of T2DM. Our lab has developed two transgenic models of insulin resistance. In one model, skeletal muscle insulin resistance is achieved by transgenic expression of a dominant-negative IGF-I receptor (IGF-IR) in skeletal muscle. The transgenic fish showed impaired Akt phosphorylation postprandially or after insulin administration (Maddison et al., 2015). These fish had significantly increased fasting blood glucose as early as 3-month old compared to wild-type fish and is exacerbated by overfeeding (unpublished data). In the other model, insulin resistance is achieved through liver specific knockdown of the insulin receptors using CRISPR/Cas9 (Yin et al., 2015). Similar to mice and humans, liver insulin resistance causes fasting hypoglycemia and postprandial hyperglycemia. Since muscle and liver insulin resistance are thought to be the major drivers of T2DM, these models will be useful to dissect the progression of T2DM.

Another type of diabetes, MODY (maturity-onset diabetes of the young), is a rare, autosomal dominant, noninsulin-dependent and monogenic form of diabetes resulting from pancreatic βcell dysfunction with an onset before 25 years of age. Since this disease is caused by mutation in a single gene, with different genes leading to different forms, MODY models can be developed by targeted gene ablation. However, as in mice, the mode of inheritance in MODY gene mutations is usually recessive, not autosomal dominant. MODY5 stems from mutations in hepatocyte nuclear factor 1β (HNF1β). A zebrafish hnf1ba mutant line (hnf1bas430) was identified from a zebrafish ENU mutagenesis screen (Lancman et al., 2013). The homozygous mutants exhibit pancreas hypoplasia and reduced β-cell numbers similar to MODY5. MODY6 results from mutations in NEUROD1 (Malecki et al., 1999). In mice, disruption of NeuroD1 leads to diabetes and premature death (Naya et al., 1997). In zebrafish neurod1 deficiency led to failed endocrine cell differentiation and increased free glucose levels in larvae (Dalgin and Prince, 2015). MODY4 is a result of PDX1 mutation (Stoffers et al., 1997) and a pdx1 mutant line exhibited reduced β-cell numbers, disrupted glucose homeostasis, sensitivity to over-nutrition and is responsive to anti-diabetic drug treatment

#### REFERENCES


(Kimmel et al., 2015). The adult pdx1 mutant zebrafish have small body size and decreased viability. MODY10 results from mutations in INS gene (Meur et al., 2010). A transgenic line expressing a mutated preproinsulin protein (C43G) has been developed (Eames et al., 2013). Interestingly, glucose homeostasis and β-cell mass were not altered in these fish, even though excess proinsulin accumulates in endoplasmic reticulum (ER). This could be due to the regenerative capacity of the zebrafish leading to turnover of the dysfunctional β-cells. However, this provides an opportunity to investigate misfolded proinsulin and ER stress in a non-diabetic in vivo system. Together, these MODY models develop phenotypes observed in patients, further supporting the utility of zebrafish as a diabetes model.

Although an appropriate model is still lacking for studying the long-term effect of diabetes, there have been approaches to study diabetic complications. For example, long immersion of larval or adult zebrafish in glucose solution has been used to model chronic hyperglycemia (Capiotti et al., 2014; Connaughton et al., 2016). This approach has been used to study diabetic retinopathy (Gleeson et al., 2007; Jung et al., 2016) as well as changes in bone metabolism (Carnovali et al., 2016). Inducing hyperglycemia through repeated STZ treatment in adult fish can impair wound healing (Olsen et al., 2010) and can cause heritable epigenetic changes after normalization of glycemia (Olsen et al., 2012). These studies underscore the lasting consequences of disrupting glucose control in zebrafish.

Overall, zebrafish offers particular advantages to the study of metabolic diseases. Models for studying obesity, pancreas regeneration, hyperglycemia, and diabetic complications have been established and will promote the understanding of the disease mechanisms, and provide new targets for disease therapy.

### AUTHOR CONTRIBUTIONS

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

### FUNDING

Supported by ADA 1-13-BS-27 and a Vanderbilt DRTC Pilot and Feasibility Grant (WC), R01 DK109407 (WC), and by JSPS KAKENHI Grant Number 15KK0305 (LZ).

of dietary fat and cholesterol. Methods Cell Biol. 101, 111–141. doi: 10.1016/B978-0-12-387036-0.00005-0


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

Copyright © 2018 Zang, Maddison and Chen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) 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.

# Rare Genetic Blood Disease Modeling in Zebrafish

Alberto Rissone and Shawn M. Burgess\*

Translational and Functional Genomics Branch, National Human Genome Research Institute, National Institutes of Health, Bethesda, MD, United States

Hematopoiesis results in the correct formation of all the different blood cell types. In mammals, it starts from specific hematopoietic stem and precursor cells residing in the bone marrow. Mature blood cells are responsible for supplying oxygen to every cell of the organism and for the protection against pathogens. Therefore, inherited or de novo genetic mutations affecting blood cell formation or the regulation of their activity are responsible for numerous diseases including anemia, immunodeficiency, autoimmunity, hyper- or hypo-inflammation, and cancer. By definition, an animal disease model is an analogous version of a specific clinical condition developed by researchers to gain information about its pathophysiology. Among all the model species used in comparative medicine, mice continue to be the most common and accepted model for biomedical research. However, because of the complexity of human diseases and the intrinsic differences between humans and other species, the use of several models (possibly in distinct species) can often be more helpful and informative than the use of a single model. In recent decades, the zebrafish (Danio rerio) has become increasingly popular among researchers, because it represents an inexpensive alternative compared to mammalian models, such as mice. Numerous advantages make it an excellent animal model to be used in genetic studies and in particular in modeling human blood diseases. Comparing zebrafish hematopoiesis to mammals, it is highly conserved with few, significant differences. In addition, the zebrafish model has a high-quality, complete genomic sequence available that shows a high level of evolutionary conservation with the human genome, empowering genetic and genomic approaches. Moreover, the external fertilization, the high fecundity and the transparency of their embryos facilitate rapid, in vivo analysis of phenotypes. In addition, the ability to manipulate its genome using the last genome editing technologies, provides powerful tools for developing new disease models and understanding the pathophysiology of human disorders. This review provides an overview of the different approaches and techniques that can be used to model genetic diseases in zebrafish, discussing how this animal model has contributed to the understanding of genetic diseases, with a specific focus on the blood disorders.

Keywords: zebrafish, hematopoiesis, modeling human diseases, SCID, blood, genome editing

#### Edited by:

Gokhan Dalgin, The University of Chicago, United States

#### Reviewed by:

Eirini Trompouki, Max-Planck-Institut für Immunobiologie und Epigenetik, Germany Julien Y. Bertrand, Université de Genève, Switzerland

> \*Correspondence: Shawn M. Burgess burgess@mail.nih.gov

#### Specialty section:

This article was submitted to Stem Cell Research, a section of the journal Frontiers in Genetics

Received: 31 May 2018 Accepted: 09 August 2018 Published: 31 August 2018

#### Citation:

Rissone A and Burgess SM (2018) Rare Genetic Blood Disease Modeling in Zebrafish. Front. Genet. 9:348. doi: 10.3389/fgene.2018.00348

## INTRODUCTION

fgene-09-00348 August 30, 2018 Time: 10:38 # 2

Genetic diseases can be both inherited and acquired. In particular, most of the inherited diseases belong to the category called "orphan diseases" (Strynatka et al., 2018). The term orphan disease can refer to two different types: common diseases neglected by doctors or rare diseases with a variable incidence in the population (Aronson, 2006). Although for different reasons, both have in common minimal scientific research about their genetic causes and molecular mechanisms and a lack of investments by pharmaceutical sector to develop new treatments (Strynatka et al., 2018). The definition of rare disease in not universal and depends on the country. In the United States, for example, a disease is considered rare when affecting fewer than 1 person in 200,000, but in Japan and Australia the numbers are very different: 1/50,000 and 1/2,000, respectively (Lavandeira, 2002).

With the advent of next-generation sequencing technologies and a progressive reduction in sequencing costs, we will begin to see a dramatic increase in the identification of the genes responsible for human genetic disorders. Model organisms played a pivotal role in genotype-to-phenotype studies, in particular when the association is unclear (Strynatka et al., 2018). However, with an estimated total number of Mendelian genetic diseases between 7,000 and 15,000 (Boycott et al., 2013) and only ∼1,500 drugs approved by FDA, most of the genetic diseases still have no effective treatment, indicating a constant need for new experimental animal models.

#### Animal Models to Study Genetic Diseases

Animal models are fundamental tools in biomedical research because they can fill the gap between basic science and the treatment of human diseases (Zon, 2016). Several different animal models can be used to study the gene function providing new insight into pathophysiology of human disorders (Bier and McGinnis, 2004). Simple models such as Saccharomyces cerevisiae (Foury, 1997) and Dictyostelium discoideum (Firtel and Chung, 2000; Chung et al., 2001) proved to be very helpful in elucidating the basic mechanisms of eukaryotic cell function, such as the regulation of the cell cycle, the mechanisms of DNA damage and repair, metabolism, and cell signaling. Similarly, invertebrates like Caenorhabditis elegans (Aboobaker and Blaxter, 2000; Culetto and Sattelle, 2000) and Drosophila melanogaster (Bernards and Hariharan, 2001; Reiter et al., 2001; Chien et al., 2002) represent outstanding models to study genes involved in more complex body plans (Bier and McGinnis, 2004). However, their very high evolutionary distance with a low rate of sequence conservation compared to vertebrates and the huge difference in their anatomy and physiology, limit their use in studying vertebratespecific embryonic development and in directly modeling human diseases.

Traditionally, among mammals Mus musculus (mouse) and Rattus norvegicus (rat) are the species most commonly used as a vertebrate model organisms. Particularly the mouse with its small size, genomic resources, genetic tractability, and anatomic and physiologic conservation with humans, elected it as the favored species to model human genetic disorders. Although in the past, mouse models were usually generated using homologous recombination methods in embryonic stem cells (ESCs) it was a laborious, time consuming and not so efficient approach. With the advent of the new genome-editing techniques the overall process has been speed-up and today the generation of new mouse models require just few weeks, instead of the previous 1–2 years (Ott de Bruin et al., 2015). However, the maintenance of large mouse colonies is still expensive reducing its use in large-scale genetic screens and phenotyping studies. In addition, because of the complexity of human diseases and the intrinsic differences between humans and other species, it is often the case that some aspects of the model organisms physiology makes it a poor model for a specific disease, and so multiple model organisms are needed. Based on several features described in detail below, zebrafish represents a good compromise for modeling human diseases, filling the gap between the invertebrate and mammalian model systems.

### Zebrafish as an Animal Model

The zebrafish (Danio rerio) is a tropical freshwater fish from South-East Asia which has in recent decades gained popularity in the research community. Zebrafish popularity began at the end of the last century (1970s–1980s), when they became a new genetic model for developmental biologists. However, because of the numerous advantages that zebrafish offer, it has rapidly become popular in the study of human disease.

Zebrafish belong to the teleost clade, also known as "bony" fish. The eggs are externally fertilized which allows for simple experimental manipulation of the embryos, and each mating produces a high number (usually >100) of embryos. The embryo development is very fast compared to other vertebrate models such as mice, and a few days after the hatching (48– 72 h post fertilization) zebrafish embryos already show all the major organs of the adult animals. Notably, the anatomy and physiology of most of the zebrafish organs are very similar to those of mammals and in terms of hematopoiesis, Teleosts have all the different hematopoietic cellular elements found in mammals (i.e., erythroid, myeloid and lymphoid lineages).

Although under normal conditions zebrafish embryos are not completely transparent, they may be treated with 1-phenyl 2-thiourea (PTU) at ∼24 h post fertilization (hpf) which will inhibit melanin formation resulting in almost transparent embryos that will continue to remain in this condition as long as the PTU treatment is continued (Karlsson et al., 2001). Alternatively, numerous genetic pigmentation mutants with different levels of transparency are available and can permit in vivo imaging from the embryo phases to adulthood (White et al., 2008).

The zebrafish is well suited for molecular and genetic analysis of temporal and spatial gene expression using whole mount in situ hybridization (WISH) (Thisse and Thisse, 2008); moreover, a very long list of transgenic lines (including inducible models)

are publicly available that allows study of tissue and organ development in vivo and in real-time during all the phases of embryo development (Kondrychyn et al., 2011; Ruzicka et al., 2015). For a comprehensive list of transgenic lines helpful in studying zebrafish hematopoiesis see Gore et al. (2018).

Thousands of mutations obtained using large scale mutagenesis screens are available and moreover new mutations can be easily introduced in zebrafish genome using the most recent techniques of site-specific genome editing such as the Clustered Regularly Interspaced Short Palindromic Repeat/CRISPR associated protein 9 (CRISPR/Cas9). The zebrafish genome has been fully sequenced and high-quality assemblies are publically available (Howe et al., 2013). Genomic analysis shows that there is a high degree of sequence conservation and synteny between the zebrafish and human genomes. Zebrafish, especially during its embryonic stages, proved to be very suitable for medium- to large-throughput drug screening, because it is possible to add the different compounds directly into the embryo medium. Moreover, usually zebrafish bioassays are cheaper and faster than the comparable mouse assays. Finally, maintenance costs of zebrafish model are lower than those for mammals. While this review focuses on mutational analysis in early embryos, adult zebrafish are increasingly being used to study some blood diseases as well, particularly blood cancers (Langenau et al., 2003; Alghisi et al., 2013).

Like any other animal model and despite its numerous advantages and unique features, the zebrafish model system is not devoid of disadvantages and/or limitations. One of major limitations is the teleost-specific genome duplication. This event occurred ∼400 millions of years ago and corresponded to a complete duplication of teleost genome (Meyer and Van de Peer, 2005). After the duplication event, the majority of the duplicated genes were lost or became pseudogenes. However, roughly 20% of the genes maintained two functional copies in the genome. As a results, zebrafish and other teleost species have a higher number of protein coding genes (∼26,000) compared to other vertebrates like human, mouse or chicken (∼21,000) (Howe et al., 2013). This information must be taken in consideration during reverse genetic analysis of the duplicated genes as a result of compensatory effects, most of the cases knocking out one copy of the gene is not enough to mimic the effects of a null allele. Another important aspect to take into consideration is that, after gene duplication, each duplicate can functionally diverge from each other through subfunctionalization and/or neo-functionalization events (Ohno, 1970; Postlethwait et al., 2004; Rastogi and Liberles, 2005). Although the modern genome-editing technologies allow the targeting of multiple genes at the same time, somewhat overcoming the problem duplicate genes, neo-functionalization events could have partially changed the gene's function. These phenomena could be responsible for discrepant functional outcomes among models in different species and potentially could limit the use of zebrafish in modeling human diseases. Unfortunately, it is currently not possible to determine a priori whether this would be an issue for any given gene duplication.

## Zebrafish as a Tool to Study Human Diseases

The recent advent in the zebrafish field of targeted genome editing techniques, such as ZFN, TALEN, and in particular CRISPR/Cas9, has opened up the model to reproduce human pathological conditions of known disease-related genes and to study their effects in vivo, with the ultimate goal of identifying new therapeutic targets (Detrich et al., 1999; Langheinrich, 2003; Santoriello and Zon, 2012). Although historically the first successful attempts to use zebrafish for genetic studies were represented by forward genetic approaches using chemical or insertional mutagenesis techniques (Haffter and Nusslein-Volhard, 1996; Golling et al., 2002; Varshney et al., 2013), later on, thanks to the development of knockdown and targeted genome editing techniques, this model system proved that it could be efficiently used in reverse genetic approaches as well.

#### Knockdown Approach to Study Gene Function

In zebrafish, with the exception of limited cases where RNAi has been used to knockdown specific targets (Oates et al., 2000), the knockdown approach has been performed through the use of morpholinos (MOs) (Nasevicius and Ekker, 2000). Because of their ease of use, MOs represented the first and they are still today one of the most popular approaches to perform reverse genetic analysis in zebrafish (Eisen and Smith, 2008; Bill et al., 2009; Timme-Laragy et al., 2012; Stainier et al., 2017). MOs are modified antisense oligonucleotides (ASOs) that are manually microinjected in the embryos at the first stages of development (1–4 cells maximum). MO oligonucleotides are very stable because they are not targeted by nuclease enzymes and they do not act through an RNaseH mechanism, as in the case of RNA interference (RNAi) technology (Eisen and Smith, 2008). Instead, through the binding to their RNA targets (pre-mRNA or mRNA), MOs induce a transient dose-dependent knockdown effect in the injected embryos (morphants). During embryonic development, MOs concentration is gradually reduced in the cells due to cell divisions and therefore they do not typically stay effective beyond 3–5 days post fertilization (dpf) (Timme-Laragy et al., 2012).

Usually, low doses of MOs are well tolerated by zebrafish embryos, allowing the targeting of more than a single transcript at the same time to study synergistic effects of multiple knockdowns (Rissone et al., 2012). While the report of potential off-target effects mediated by the tp53 activation following MOs injection (Robu et al., 2007) raised several criticisms to the validity of some results obtained with them (Blum et al., 2015; Kok et al., 2015; Stainier et al., 2015), they have been extensively used to study gene function during embryo development and to confirm the role of candidate genes involved in human diseases. Usually, if used with full knowledge of the potential risks and limitations and with all the essential controls (Eisen and Smith, 2008; Stainier et al., 2015, 2017), MOs represents a good starting point to infer gene function in a fast and inexpensive way and, at least in one case, it has been shown that MOs action can prevent the genetic compensatory effects induced in some mutant animals (Rossi et al., 2015).

Recently, an alternative knockdown approach requiring the injection of RNA–DNA hybrid ASOs (also known as gapmers) has been used to overcome some of the limitations of MOs (Pauli et al., 2015): first, each molecule of a steric-blocking MO will only bind a single target RNA transcript; second, translational MOs, designed to block the ATG codon of mRNAs, do not induce the degradation of the target transcripts impeding the quantification of their knockdown efficiency. In contrast, gapmer ASOs contains a central DNA region, which triggers an RNAse H-mediated degradation of the target RNAs. The molecules also have flanking 2<sup>0</sup> -modified nucleosides at both ends to protect them from exonucleases activity and to increase the affinity for the targets (Evers et al., 2015). While they have already been used in cell culture (Dimitrova et al., 2014) and other species (Heasman et al., 1994; Zhang et al., 1998; Wheeler et al., 2012), gapmers-mediated knockown represents a relatively unexplored approach in zebrafish. Pauli et al. (2015) tested the feasibility of this approach in zebrafish targeting ∼20 protein-coding and non-coding transcripts with known embryonic loss-of-functions phenotypes and showing that gapmers can represent an effective RNA knockdown alternative. Although representing a promising tool, more studies are required to further confirm their potential utility.

Finally, given the continuous reduction of the costs required to create mutant alleles in zebrafish (Varshney et al., 2015b), in the future the studies involving the use of MOs or other knockown techniques could and should include a comparison of the phenotypes observed in bona fide genetic mutants as a control (Stainier et al., 2017).

#### Zebrafish Genome Editing Tools

In zebrafish, large-scale genetic screens using random mutagenesis were successfully introduced at the end of the 1990s (Haffter and Nusslein-Volhard, 1996; Haffter et al., 1996). These forward genetic techniques proved to be very helpful in identifying mutants presenting phenotypes typical of several human disorders (North and Zon, 2003; Amsterdam and Hopkins, 2006; Bradford et al., 2017; Howe et al., 2017). However, these approaches were not devoid of limitations; especially in the amount of efforts required to isolate each mutation by positional cloning and they were gradually replaced by the use of reverse genetic approaches. In particular, the advent of the TILLING (or Targeting Induced Local Lesions in Genome) system in early 2000s (Wienholds et al., 2002) allowed the researchers to screen for mutations in specific genes of interest. Later, new and more efficient tools for targeted mutagenesis were developed and they were quickly adopted by the zebrafish community (Doyon et al., 2008; Meng et al., 2008; Huang et al., 2011; Bedell et al., 2012). Briefly, all the major genome editing techniques are based on the coupling of DNA-binding domains or guide RNA molecules to proteins with nuclease activity used to induce double-stranded breaks (DSB) in the target genomic regions. Then inductions of DSBs prompts the activity of two different cellular DNA repair mechanisms: (a) the most common, but highly error–prone nonhomologous end joining (NHEJ) and (b) the homology-directed repair (HDR) which is rarer in vivo and requires a template DNA to repair the DSB (Symington and Gautier, 2011). The HDR mechanism seems to be particularly challenging in zebrafish and so far, despite numerous efforts of the community to optimize the mutagenic protocol in order to increase its frequency in zebrafish, very few successful cases are described in literature (Hruscha et al., 2013; Hwang et al., 2013a; Auer et al., 2014; Irion et al., 2014; Kimura et al., 2014; Hisano et al., 2015; Li et al., 2015; Armstrong et al., 2016; Hoshijima et al., 2016; Zhang et al., 2016; Moreno-Mateos et al., 2017; Zhang et al., 2018). In contrast, in zebrafish the NHEJ repair mechanism works very efficiently and usually it is associated with loss/gain of small fragments of genomic DNA in the range of 1 bp to ∼40 bps. Selecting frame-shift mutations introduced by NHEJ potentially impairs the structure and/or the functionality of the targeted protein.

The first examples of zebrafish mutants obtained with Zinc Finger Nucleases (ZFNs) and Transcription Activator-Like Effector Nucleases (TALENs) were published in the late 2000s and 2011, respectively (Doyon et al., 2008; Meng et al., 2008; Huang et al., 2011; Sander et al., 2011). They both allowed an easy targeting and recovery of the different mutations introduced in the specific genomic regions, although their major limitations resided in their still high costs and in the efforts necessary to develop the modular DNA-binding motifs responsible for the sequence specificity.

The recent advent of CRISPR/Cas9 system moved the versatility and affordability of the genome editing in zebrafish to a new level, allowing the targeting of multiple regions at the same time (multiplexing) with a consistent reduction of the costs (Hwang et al., 2013b; Jao et al., 2013; Varshney et al., 2015a). In the CRISPR/Cas9 system, the sequence specificity is obtained using an ∼20 base pairs long guide RNA (gRNA), while the DNA double-strand cleavage activity is attained using the Cas9 endonuclease activity. Compared to ZFNs and TALENs, CRISPR/Cas9 system has a similar or better efficiency in targeting genomic DNA, with a higher versatility and simplicity of design. The only limitation in the target design consist in the presence of a protospacer adjacent motif (PAM) directly upstream the target region bound by the gRNA. The PAM sequence depends on the Cas9 protein used, but in most of the cases is the nucleotide sequence NGG (Varshney et al., 2015b). Based on the biological target system used (cells or animal embryos) the gRNA sequence is delivered by transfection of specific gRNA-containing vectors or direct injection of gRNAs generated by in vitro transcription. In zebrafish, its intrinsic features such as the external fertilization and the easy manipulation of the embryos, allow the direct coinjection of Cas9 mRNA (or Cas9 protein) and the gRNAs into the embryos during the earliest stages of development (similarly to MOs). One further advantage of the zebrafish embryos is that they can easily tolerate multiple gRNAs at the same time, making possible a multiplexing approach that dramatically increases its versatility. It becomes possible to target multiple genes at the same time or to use specific gRNA directed against different regions of a single gene to maximize the number of mutated allele or to induce the deletion of a large genomic region based on the specific needs of the researchers. Another benefit of a multiplexing approach in zebrafish consists in overcoming the potential compensatory effects due to the duplicated genes, which clearly represents a problem for genetic analysis in this

animal model. In contrast, an obvious downside of a multiplexing approach is represented by the increased probability of offtarget activity, which seems to be relatively rare, but detectable (Varshney et al., 2015a). Recently, in order to maximize the specificity of the CRISPR/Cas9 system, different variants of the Cas9 enzyme (Cas9n and Cas9/FokI) were successfully developed and proved to work very well in limiting nuclease activity to specific genomic sites, although reducing at the same time the overall multiplex potential of the system (Ran et al., 2013; Tsai et al., 2014). Many human diseases are due to mutations predicted to cause single or multiple amino acid substitutions that partially inhibit gene activity, instead of completely impairing protein function or mRNA stability. Therefore, in order to better mimicking the mutations found in human patients, the ability to introduce in an animal substitution genetic mutations (in contrast to null alleles) represents one of the current challenges of the genomic editing era. Recently, a new genome editing technique called "base editing" has been developed and tested in mammalian cells and different model species (Komor et al., 2016; Kim et al., 2017). Through the fusion of a cytidine deaminase enzyme to the N-terminal region of a Cas9 nickase (nCas9) protein, this new technology allows direct conversion of one single base in a programmable way bypassing the DSBs. Recently, Zhang et al. (2017) adapted a similar approach to work efficiently in zebrafish, further increasing its versatility as animal model in modeling human diseases.

#### Zebrafish Hematopoiesis

Usually, during both zebrafish and mammal embryo development, hematopoiesis is obtained in three distinctive, but partially overlapping, processes termed "hematopoietic waves" (Ciau-Uitz et al., 2014). For correct hematopoietic development, both their timing and embryonic localization need to be strictly regulated (for a comprehensive list of the genes expressed and involved in the different phases of zebrafish hematopoiesis see Gore et al., 2018 ). Based on the type of blood cells originated, the three major hematopoietic waves are distinguished in primitive, prodefinitive (or intermediate) and definitive. In mammals, during the first two hematopoietic waves, red blood cells and macrophages, and erythroid-myeloid precursors are generated extra-embryonically in the yolk sac blood island (Ciau-Uitz et al., 2014). In contrast, the definitive hematopoiesis produces all the major hematopoietic cell types (erythroid, myeloid, and lymphoid) through the creation of hematopoietic stem and precursor cells (HSPCs).

Over the years, several reports pointed out the utility of the zebrafish model to study the different aspects of vertebrate hematopoiesis (Gore et al., 2018). Despite the >400 million years of evolutionary distance (Postlethwait et al., 1999) and the different embryonic territories involved (Ciau-Uitz et al., 2014), zebrafish and mammal species share the same key genetic regulation and mechanisms (Sood and Liu, 2012; Avagyan and Zon, 2016; Gore et al., 2018).

#### Zebrafish Primitive Hematopoiesis

In zebrafish, primitive hematopoiesis starts around 11 h post fertilization (hpf) during somitogenesis (Davidson and Zon, 2004). Specific cells inside the anterior and posterior lateral mesoderm (ALM and PLM, respectively) start expressing endothelial and hematopoietic markers generating different populations of vascular and hematopoietic precursors. Later, the ALM gives rise to the rostral blood island (RBI) region. The cells in the ALM/RBI region generate primitive myeloid precursors which eventually differentiate into macrophages and neutrophils (Herbomel et al., 1999). Specifically, primitive myeloid precursors start to express the transcription factor pu.1 and then they leave the RBI spreading on the yolk sac (Galloway et al., 2005; Rhodes et al., 2005; Monteiro et al., 2011). Following pu.1 activation, different myeloid markers start to be expressed in these migrating cells and finally, with the expression of irf8, cebpa, and cebp1 genes, these precursors begin to assume a more restricted myeloid identity (Li et al., 2011; Jin et al., 2012, 2016; Mommaerts et al., 2014; Dai et al., 2016). Precursor cells from RBI are also responsible for the generation of mast cells (Dobson et al., 2008) and microglia, which will eventually colonize the brain (Xu et al., 2015, 2016). In contrast, the future hematopoietic precursors that reside in the PLM migrate, converging to the midline of the embryo body and forming the intermediate cell mass (ICM). The ICM occupies the space between the yolk extension and the notochord and it extends throughout the trunk of the embryos to the end of the yolk extension (Ciau-Uitz et al., 2014). Although the hematopoietic precursors forming the ICM region during development express both erythroid and myeloid markers (such as gata1a, and spi1b or mpx, respectively) this posterior first hematopoietic wave seems to generate mostly primitive erythrocytes that enter the circulation at ∼24 hpf, when the zebrafish heart starts beating (Berman et al., 2003). Primitive erythroid and myeloid cells are then gradually replaced by blood cells produced during the following hematopoietic waves.

#### Zebrafish Prodefinitive Hematopoiesis

Around 24 hpf, the zebrafish embryos switch from the primitive to the definitive hematopoiesis. This transition occurs concurrently in two different regions of the embryo: the posterior blood island (PBI) and the ventral wall of the dorsal aorta (VDA). From the VDA originates the precursors of HSPCs; while in the PBI, which is the most caudal region of the ICM right after the end of the yolk extension, only erythroid-myeloid progenitors (EMP) begin to differentiate (Bertrand et al., 2007). Because they do not derive from HSPCs, the hematopoietic potential of EMPs is limited to the erythroid and myeloid lineages (Gore et al., 2018). However, a very recent temporally–spatially resolved fate-mapping analysis (Tian et al., 2017) showed that the ventral endothelium in the VDA and PBI regions, gives rise to a transient wave of T lymphopoiesis which does not require HSPCs. Notably, the generated T-cells are mostly CD4 Tαβ cells and are temporally limited to the larval stages of development (Tian et al., 2017).

#### Zebrafish Definitive and Adult Hematopoiesis

The formation of HSPCs is the characteristic step of definitive hematopoiesis. Compared to previous hematopoietic progenitors generated during primitive hematopoiesis, HSPCs are selfrenewable pluripotent stem cells responsible for the formation of

all the hematopoietic lineages during the zebrafish development and the adult phase. In zebrafish the definitive hematopoiesis starts around ∼24 hpf with the formation of the first runx1 and then cmyb positive cells in a specific region of the trunk of the embryos, the VDA which functionally corresponds to the aorta-gonad-mesonephros (AGM) region in mammals. In the VDA, a subset of endothelial cells starts to differentiate in HSPCs and they eventually bud from the aorta and colonize the region between the aorta and the posterior cardinal vein (PCV) (Bertrand et al., 2010; Kissa and Herbomel, 2010). This process is usually indicated as endothelial-hematopoietic transition (EHT) (Kissa and Herbomel, 2010; Bresciani et al., 2014). Then, HSPCs enter the circulation through the PCV and move caudally to colonize by 2 dpf the so called caudal hematopoietic tissue (CHT) which consists of the region where the vascular plexus connecting the dorsal aorta and the cardinal vein resides (Murayama et al., 2006; Bertrand et al., 2010; Kissa and Herbomel, 2010; Sood et al., 2010; Bresciani et al., 2014; Gore et al., 2016; Tian et al., 2017). This region is also indicated as PBI. Due to the reduced blood flow and the presence of specific signals on the surface of HSPCs and endothelial cells of the caudal vein plexus (CVP), HSCPs leave the circulation and extravasate in the PBI region, where they are surrounded by endothelial and perivascular mesenchymal stromal cells that modulates their subsequent proliferative and/or differentiation fate (Murayama et al., 2006; Jin et al., 2007, 2009; Tamplin et al., 2015). The CHT region is the teleost homologous of the mammalian fetal liver and there, part of the HSCPs proliferates and gives rise to erythroid and myeloid cells, and some of them through circulation, move anteriorly to the thymus and to the anterior part of the kidney by 3 and 4 dpf, respectively (Kissa et al., 2008). Like in mammals, the zebrafish thymus is the site of differentiation and maturation of T-lymphocytes (Hess and Boehm, 2012), while the kidney marrow, where HSPCs reside through adulthood, is analogous to mammalian bone marrow (Bertrand and Traver, 2009). Both organs represent the sites where the adult hematopoiesis reside (de Jong and Zon, 2005).

### ZEBRAFISH MODELS OF BLOOD DISEASE

Historically, the zebrafish mutant sauternes (sau), represented the first example of a growing list of zebrafish contributions to the study of human diseases. Isolated during a large zebrafish forward genetic screening, the mutated gene encodes δ-aminolevulinate synthase (ALAS-2), an enzyme involved in the first step of heme biosynthesis (Brownlie et al., 1998). Notably, because mutations in ALAS2 gene are responsible for the congenital sideroblastic anemia (CSA) in humans, zebrafish sauternes mutants represented also the first animal model for the disease (Brownlie et al., 1998). Since then, through both forward and reverse genetic analysis, the zebrafish have proven to be instrumental in the study of human blood disorders (Berman et al., 2003; North and Zon, 2003; Forrester et al., 2012; Moore and Langenau, 2012; Santoriello and Zon, 2012; Zhang and Yeh, 2012; Avagyan and Zon, 2016; Robertson et al., 2016; Gore et al., 2018). In the following sections we will discuss a few of these disease models.

## Zebrafish Erythroid and Myeloid Models of Disease

#### Diamond–Blackfan Anemia

Diamond–Blackfan anemia (DBA) is a congenital bone marrow failure syndrome characterized by a complex array of hematopoietic and non-immunological defects. Patients with DBA are generally diagnosed during infancy or early childhood and present erythrocyte aplasia with anemia, macrocytosis, reticulocytopenia, and a paucity of red blood cell precursor cells within a normocellular marrow, associated with growth retardation, and limb, cardiac, and/or craniofacial malformations and have a predisposition to cancer (O'Brien et al., 2017). Other specific features of DBA are an elevated erythrocyte adenosine deaminase (ADA) activity and an elevated fetal hemoglobin concentration (McGowan and Mason, 2011). The primary treatment of DBA is corticosteroids, but ∼40% of patients lose steroid responsiveness, requiring chronic red cell transfusions (O'Brien et al., 2017). The only definitive treatment for DBA consists of hematopoietic stem cell transplantation (HSCT). In the 65% of patients DBA is caused by heterozygous mutations in ribosomal genes; while in the remaining ∼35% the genetics causes are still unknown (O'Brien et al., 2017). While, the most commonly mutated gene in DBA patients is RPS19 (Draptchinskaia et al., 1999; Campagnoli et al., 2008), germ-line mutations in genes encoding both small and large components of the ribosomal subunits (RPS24, RPS17, RPS7, RPS10, RPS26 and RPL35A, RPL5, RPL11, RPL26, respectively) have also been described (Farrar et al., 2011; Gazda et al., 2012).

In mice, homozygous knockout mutants for the rps19 gene presented early embryonic lethality (Matsson et al., 2004). Heterozygous mice did not present hematologic or developmental phenotypes, because of a genetic compensation from the wild-type rps19 locus (Matsson et al., 2006). Another mouse model with a ethylnitrosourea (ENU)-induced missense point mutation in rps19 gene showed a hematopoietic defect that was rescued by p53 knockdown (McGowan et al., 2008; McGowan and Mason, 2011).

To date two independent models of DBA were developed in zebrafish using MOs knockdown approaches against the ribosomal protein S19 (rps19) transcripts (Danilova et al., 2008; Uechi et al., 2008). In both cases, rps19 knockdown recapitulated the phenotypes observed in DBA patients such as the defective erythropoiesis and morphologic abnormalities. A subsequent zebrafish knockdown model of rpl11, another ribosomal protein found mutated in DBA patients (Chakraborty et al., 2009), confirmed the central role for p53 activation in the pathophysiology of DBA (Ball, 2011). More recently, the genetic models of rps19 and rpl11 deficiency were developed (Zhang et al., 2014). Both models presented a reduction of protein production and in particular of globin proteins in red blood cells, suggesting that the protein reduction could be a key contributing factor to erythroid defects observed in DBA (Zhang

et al., 2014). Finally, Danilova et al. (2018) investigated the role of immune system in DBA. Using rpl11 mutants and rps19 morphants, they showed an increased level of interferon network, inflammatory pathways and complement system suggesting that the activation of the innate immune system could contribute to the physiopathology of DBA (Danilova et al., 2018).

Recently, another gene of the rps family, rps29, has been associated with DBA using whole-exome sequencing and functional studies in a zebrafish model of rps29 deficiency (Taylor et al., 2012; Mirabello et al., 2014).

Finally, DBA patients present a high incidence of cancer, with particularly high risks of leukemia, osteosarcoma, myelodysplastic syndrome and colon adenocarcinoma (Vlachos et al., 2012); notably, rps and rpl heterozygous mutations also cause tumors in zebrafish (Amsterdam et al., 2004; Lai et al., 2009).

#### Erythropoietic Protoporphyria

The different inherited porphyrias are genetic diseases affecting heme biosynthesis, caused by mutations in specific enzymes of the heme biosynthetic pathway. As a result of these enzymatic deficiencies, the intermediates of the heme biosynthetic pathway (porphyrinogens, porphyrins and their precursors) are produced in excess and accumulate in tissues resulting in neurological, photo-cutaneous symptoms, and hematological disturbances (Richard et al., 2008). Based on which tissue accumulates porphyrin, this group of diseases can be divide into erythropoietic or hepatic. Three different erythropoietic porphyrias have been described: erythropoietic protoporphyria (EPP), which is the most frequent, congenital erythropoietic porphyria (CEP), and the very rare hepatoerythropoietic porphyria (HEP) (Richard et al., 2008). In humans, EPP is associated with inherited defects in the ferrochelatase (FECH) gene which catalyzes the insertion of a ferrous iron into protoporphyrin IX (PPIX) to form heme (Puy et al., 2010). The main clinical manifestation is painful skin inflammation after short exposure to sunlight. However, because the heme formation mainly occurs in the bone marrow and liver, mutations affecting FECH activity lead to PPIX accumulation in the bone marrow, erythrocytes, plasma, and liver and it has been estimated that up to 20% of the EPP patients have liver injury and approximately 2–5% develop serious liver damage or even liver failure (Wang et al., 2018).

After chemical mutagenesis using ENU, a viable autosomal recessive mutation in mouse fech gene was isolated and characterized in the early 1990s (Tutois et al., 1991; Boulechfar et al., 1993). Homozygous null mice present a severe reduction of FECH enzymatic activity and they exhibit jaundice, photosensitivity and dramatic hepatic dysfunction (Tutois et al., 1991).

A zebrafish genetic model of EPP, with mutations in the ferrochelatase (fech) gene, was obtained from a large-scale genetic screen (Childs et al., 2000). Zebrafish ferrochelatase mutants (Dracula) present a light-dependent lysis of red blood cells and liver disease (Childs et al., 2000).

A zebrafish mutant (ypetp61), which represented the first genetic model of HEP, was obtained in a forward genetic screen (Wang et al., 1998). The mutant presents a mutation in the uroporphyrinogen decarboxylase (urod) gene. Homozygous embryos die due to photo-ablation of their auto-fluorescent blood cells upon light exposure (Wang et al., 1998) and present clinical similarities to the defects observed in HEP patients.

#### Systemic Mastocytosis

Mastocytosis refers to a group of hematological disorders characterized by an increase in mast cell production, as well as abnormal morphology with aberrant surface receptor expression of tissue mast cells. These disorders are usually the result of various gain-of-function mutations affecting the tyrosine kinase KIT receptor, leading to increased accumulation and survival of tissue mast cells (Klaiber et al., 2017). Mastocytosis is divided in different subtypes: mastocytoma, urticaria pigmentosa, and systemic mastocytosis (SM), which represents the most severe subtype because mast cells accumulate in multiple organs. In the case of SM, a c-Kit D816V mutation is the most common cause of the disease and it gives rise to a constitutively active form of the protein that activates PI3K, Jak-STAT, and MAPK pathways (Lennartsson and Ronnstrand, 2012). Mouse models of SM using D816V mutation of human (Zappulla et al., 2005) or mouse (Gerbaulet et al., 2011) c-Kit gene have been developed. Similarly, a zebrafish transgenic model ubiquitously expressing the human KIT-D816V mutation has been developed (Balci et al., 2014). Adult transgenic fish demonstrate a myeloproliferative disease phenotype with a strong accumulation of mast cells in the kidney marrow and high expression levels of endopeptidases, consistent with SM defects observed in patients. Moreover, the zebrafish model showed a higher incidence of disease than in the transgenic mice overexpressing the same human mutant gene (Balci et al., 2014).

#### Zebrafish Models of Primary Immunodeficiencies

Primary immunodeficiencies (PIDs) represent a heterogeneous group of genetic disorders characterized by the partial or complete absence of the immune system or its improper activity (Al-Herz et al., 2011). So far, more than 230 PID-causing genes have been identified and novel gene defects continue to be discovered (Al-Herz et al., 2011). Among PIDs, severe combined immunodeficiencies (SCIDs) are the most severe forms, resulting in a block of the development of T, B and/or NK cells and, consequently, in a high susceptibility to any kind of infection. For the most severe forms of PIDs the HSCT represents the current treatment of choice and, when a histocompatibility leukocyte antigen (HLA)-matched donor is not available, conditioning chemotherapy may be needed to facilitate robust and sustained engraftment of donor cells and improve immune reconstitution (Pai et al., 2014; Ott de Bruin et al., 2015). Some SCIDs have also been successfully treated with gene therapy (Fischer et al., 2013; Mukherjee and Thrasher, 2013). Unfortunately, for many cases of PIDs the genetic causes are still unknown or poorly understood (Shearer et al., 2014). Although current advances in analyzing the genome or exome sequences of patients and their relatives uncover many

sequence polymorphisms (SNPs) possibly affecting the blood disorders, in vivo analysis still represent the golden standard to functionally confirm their effects and, from this point of view, the zebrafish can provide a good platform to test the functional consequences of different genetic variants (Iwanami, 2014).

#### Reticular Dysgenesis

Reticular dysgenesis is one of the most rare and severe forms of SCIDs. The disease is clinically characterized by congenital lymphopenia, lymphoid and thymic hypoplasia with agranulocytosis and sensorineural deafness (Hoenig et al., 2018) and is caused by mutations in adenylate kinase 2 (ak2) gene (Lagresle-Peyrou et al., 2009; Pannicke et al., 2009; Six et al., 2015). Ak2 protein is mostly located in the mitochondrial membrane space where it catalyzes the conversion of 1ATP + 1AMP = 2ADP sustaining the mitochondrial oxidative phosphorylation (Dzeja and Terzic, 2003). In mouse ak2 mutations are embryonically lethal, therefore zebrafish represented an alternative to try to model the disease. The first attempts to model RD in zebrafish consisted of embryonic knockdown with a splicing-MO mimicking one of the mutations found in a patient (Pannicke et al., 2009). Overall, larvae showed a reduction of the ikaros signal in the thymus indicating an impairment of leukocyte development during definitive hematopoiesis (Pannicke et al., 2009). Recently, the first knockdown results were independently confirmed using two other different MOs and, more importantly, by two distinct zebrafish genetic models (with a missense point mutation and a frame shift mutation, respectively) and a patient-derived iPSCs model of RD (Rissone et al., 2015). As previously shown in vitro in fibroblast of RD patients, ak2 deficiency in zebrafish induces an increased level of oxidative stress resulting in increased apoptosis and cell death of the HSPC population. In vitro differentiated iPSCs recapitulate the promyeloid block of their differentiation that has been described in the bone marrow of RD patients (Lagresle-Peyrou et al., 2009; Hoenig et al., 2017). Notably, in zebrafish antioxidant treatments with N-acetyl cysteine or Glutathione (GSH) were able to reduce the cellular oxidative stress in vivo rescuing the hematopoietic phenotypes; moreover similar results were obtained in the RD-patient derived iPSCs model, where the GSH, but not the all-trans retinoic acid (ATRA), treatment was able to significantly increase the differentiation of AK2 deficient iPSCs into mature granulocytes (Rissone et al., 2015). Interestingly, a recent report showed that in ak2-deficiencient hematopoietic progenitors obtained from a different RDpatient derived iPSCs, the intracellular ATP redistribution is impaired with a strong ATP depletion in the nucleus and an altered global transcriptional profile (Oshima et al., 2018).

#### Wiskott–Aldrich Syndrome

The Wiskott–Aldrich syndrome (WAS) is a rare X-linked recessive disease (with an estimated incidence of less than 1 in 100,000 births) characterized by eczema, bleeding diathesis, and recurrent infections that occurs in boys (Ochs and Thrasher, 2006; Puck and Candotti, 2006). The disease is associated with mutations in a gene on the short arm of the X chromosome (Xp11.23) that was originally termed the WAS gene (Derry et al., 1994). The protein encoded by the WAS gene (WASp) is a major regulator of actin polymerization and it plays a role in the remodeling of the cytoskeleton during the formation of the immunological synapse between T cells and the and antigen-presenting cells. Mutations in WASp can prevent the formation of the immunologic synapse, impairing T-cell function and compromise the locomotion and the adhesion of other immunological cells such as B cells, macrophages, dendritic cells, etc. (Ochs and Thrasher, 2006; Puck and Candotti, 2006). Moreover, constitutively activating mutations of WASp are responsible for the X-linked severe congenital neutropenia (XLN) disorder (Devriendt et al., 2001).

Different knock-out and knock-in murine models of WAS were developed (Leon et al., 2016). The complete inactivation of WASp mimicked the thrombocytopenia although failed to reproduce the microcytosis observed in human patients (Sabri et al., 2006; Marathe et al., 2009). Notably, murine models have been successfully used to conduct preclinical trials evaluating somatic gene therapy as an alternative to transplantation (Dupre et al., 2006; Bosticardo et al., 2011, 2014; Uchiyama et al., 2012).

In zebrafish the was gene is duplicated and both present a very similar expression pattern (Cvejic et al., 2008). Morpholino analysis targeting wasa or wasb showed that they exhibit different levels of disruption to the wound inflammatory response. In particular, wasa morphants showed the strongest phenotypes, which consisted of impaired migration of neutrophils and macrophages in a tail wound assays and a thrombosis and/or bleeding phenotype that mirrored the human syndrome (Cvejic et al., 2008). Morpholino studies were then confirmed by two different mutant alleles (Cvejic et al., 2008). More recently, a zebrafish wasa null mutant allele modeling WAS and XLN disorders was characterized (Jones et al., 2013). The null mutant showed defects in the wound-induced inflammatory response, due to inefficiency in forming and maintaining new leading pseudopods, and also defects in immune-cell-mediated resistance to bacterial infection, as observed in WAS patients (Jones et al., 2013).

#### WHIM Syndrome

The warts, hypogammaglobulinemia, infections, and myelokathexis (WHIM) syndrome is caused by dominant mutations in chemokine receptor CXCR4 that induce the truncation of its carboxy-terminal domain. This leads to a defect in the internalization of the CXCR4 receptor after the binding to the sdf1 ligand (which is encoded in humans by CXCL12 gene) and it induces an increased signaling and enhanced migration after stimulation by chemokine (Hernandez et al., 2003). The WHIM syndrome is an inherited immunodeficiency that presents a range of symptoms, including human papillomavirus (HPV)-induced warts, reduced longterm immunoglobulin G (IgG) titers, recurring infections, retention of neutrophils in the bone marrow (myelokathexis), and leukopenia (Kallikourdis et al., 2015). In a mouse model

of the WHIM syndrome, which recapitulates the defects observed in human patients, the expression of the mutant forms of CXCR4 in hematopoietic stem cells blocks the release of neutrophils from the bone marrow, inducing apoptosis in neutrophils and eventually neutropenia (Kawai et al., 2007). A stable transgenic line specifically expressing in neutrophils the homologous CXCR4 receptor truncation mutations found in WHIM patients was generated in zebrafish to model the disorder. As observed in the mouse model and in human patients, the zebrafish model showed neutrophil retention in hematopoietic tissue and an impairment of neutrophil motility and wound recruitment. The neutrophil retention is SDF1 dependent, because depletion of SDF1 using MOs restores neutrophil chemotaxis to wounds (Walters et al., 2010).

#### Chronic Granulomatous Disease

Chronic granulomatous disease (CGD) is an inherited PID caused by functional impairment of the NADPH oxidase complex in neutrophilic granulocytes and monocytes compromising their ability to produce ROS that are highly toxic to phagocytosed microorganisms. CGD is characterized by recurrent and severe infections, dysregulated inflammation, and autoimmunity, and patients are at increased risk of lifethreatening infections with catalase-positive bacteria and fungi and inflammatory complications such as CGD colitis (Arnold and Heimall, 2017). Mutations in any of the five structural subunits of the NADPH oxidase complex result in defective ROS production that are highly toxic to phagocytosed microorganisms.

Although only mouse genetic models of CGD are currently available (Schaffer and Klein, 2013), studies in zebrafish using MOs targeting different components of the PHOX complex successfully demonstrated that phagocyte-mediated killing of Candida albicans (Brothers et al., 2011) and Mycobacterium marinum (Yang et al., 2012) are dependent on their ability to generate an oxidative burst (Harvie and Huttenlocher, 2015).

#### Leukocyte Adhesion Deficiency

Leukocyte Adhesion Deficiency (LAD) is a group of disorders characterized by devastating bacterial infections associated with an increased number of circulating neutrophils (neutrophilia) (Burns et al., 2017). So far, four different types of LAD have been described (Burns et al., 2017). The LADs are usually distinguished by Roman numerals, I, II, III, and IV. There are mouse models for each of the four diseases, and additional nonmurine animal models for two of them (Hanna and Etzioni, 2012). In particular, LAD IV (also indicated as Rac2-deficiency), is a very rare autosomal recessive disorder in which loss of function Rac2 mutations cause defects of neutrophil F-actin assembly, adhesion and migration (Schaffer and Klein, 2013). Although, because of the additional role of RAC2 in the NADPH complex, the phenotypes of RAC2-deficiency overlap those observed in CGD. Due to RAC2's role in cell adhesion and migration, and other pathways, the phenotypes are more severe than in CGD. Recently, two models of RAC2-deficiency have been developed in zebrafish (Deng et al., 2011) using MOs and expressing in zebrafish neutrophils the human dominant inhibitory Rac2D57N mutation found in patients, respectively. Both models present a failure in wound healing due to impaired neutrophil chemotaxis and in both, the neutrophils fail to respond to a Pseudomonas aeruginosa infection (Deng et al., 2011). As observed in patients, LAD fish exhibit neutrophilia from hematopoietic tissue without increased production of neutrophils and a defect in leaving the vasculature to reach the sites of tissue damage. Notably, neutrophil retention in the CHT of WHIM fish is reduced by the expression of Rac2D57N in neutrophils, suggesting that Rac2 signaling is also necessary in CXCR4-mediated neutrophil retention in hematopoietic tissues (Deng et al., 2011).

#### ZAP70 Deficiency

Zeta-chain (TCR) associated protein kinase, 70 kDa (ZAP70) deficiency is a rare form of SCID characterized by a deficit of mature CD8+ T cells along with a regular number of nonfunctional circulating CD4+ T cells unable to mount an effective T cell response (Arpaia et al., 1994; Chan et al., 1994; Elder et al., 1994; Hivroz and Fischer, 1994). ZAP70 is an important mediator of T cell activation, proliferation, and differentiation (Wang et al., 2010). Mouse models of the disease present an even more severe block in T cell differentiation phenotypes with a lack of mature T cells (Negishi et al., 1995; Kadlecek et al., 1998), that usually is explained by a compensatory effect induced in humans by spleen tyrosine kinase (SYK) protein (Kadlecek et al., 1998). Although defects in lymphatic or blood endothelial specification have not been reported for ZAP70-deficient mice or humans, a first model in zebrafish using knockdown approaches indicated a role for both syk and zap70 in vascular embryonic development (Christie et al., 2010). However, a more recent zebrafish genetic model, where a frame-shift mutation was introduced using a TALEN targeting exon 2 of the zap70 gene, failed to show any vascular and/or lymphatic defects (Moore et al., 2016). However, zebrafish zap70 null mutants presented a reduction of developing thymocytes and mature T cells during embryo development and later they develop a T cell-specific immunodeficiency that cannot be compensated by syk protein, fully confirming the data obtained in mouse models (Moore et al., 2016).

### CONCLUSION AND FUTURE PERSPECTIVES

Thanks to the next-generation sequencing techniques, the ability to identify gene defects in small populations or even in single patients with inherited diseases is increased rapidly. More importantly, the overall costs of whole-exome sequencing and whole genome-wide association studies are steadily dropping and, based on all the predictions, they will continue to decrease in the coming years. In this scenario, the use of the latest genomeediting techniques such as CRIPR/Cas9 in association with the numerous unique features of the zebrafish model, will represents a huge boost in the modeling and in the understanding of the physiopathology of human diseases.

## AUTHOR CONTRIBUTIONS

fgene-09-00348 August 30, 2018 Time: 10:38 # 10

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

### REFERENCES


### FUNDING

This research was funded by the Intramural Research Program of the National Human Genome Research Institute; National Institutes of Health (SB: 1ZIAHG000183).


mutation provides a model for erythropoietic protoporphyria. Curr. Biol. 10, 1001–1004. doi: 10.1016/S0960-9822(00)00653-9



erythropoietin response and biochemical markers for Diamond-Blackfan anemia. Blood Cells Mol. Dis. 36, 259–264. doi: 10.1016/j.bcmd.2005.12.002


Primary Immune Deficiency Treatment Consortium experience. J. Allergy Clin. Immunol. 133, 1092–1098. doi: 10.1016/j.jaci.2013.09.044


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

Copyright © 2018 Rissone and Burgess. 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.

# Transmission Disrupted: Modeling Auditory Synaptopathy in Zebrafish

Katie S. Kindt<sup>1</sup> \* and Lavinia Sheets<sup>2</sup> \*

<sup>1</sup> Section on Sensory Cell Development and Function, NIDCD/National Institutes of Health, Bethesda, MD, United States, <sup>2</sup> Department of Otolaryngology, Washington University School of Medicine, St. Louis, MO, United States

Sensorineural hearing loss is the most common form of hearing loss in humans, and results from either dysfunction in hair cells, the sensory receptors of sound, or the neurons that innervate hair cells. A specific type of sensorineural hearing loss, referred to as auditory synaptopathy, occurs when hair cells are able to detect sound but fail to transmit sound stimuli at the hair-cell synapse. Auditory synaptopathy can originate from genetic alterations that specifically disrupt hair-cell synapse function. Additionally, environmental factors such as noise exposure can leave hair cells intact but result in loss of hair-cell synapses, and represent an acquired form of auditory synaptopathy. The zebrafish model has emerged as a valuable system for studies of hair-cell function, and specifically hair-cell synaptopathy. In this review, we describe the experimental tools that have been developed to study hair-cell synapses in zebrafish. We discuss how zebrafish genetics has helped identify and define the roles of hair-cell synaptic proteins crucial for hearing in humans, and highlight how studies in zebrafish have contributed to our understanding of hair-cell synapse formation and function. In addition, we also discuss work that has used noise exposure or pharmacological mimic of noise-induced excitotoxicity in zebrafish to define cellular mechanisms underlying noise-induced haircell damage and synapse loss. Lastly, we highlight how future studies in zebrafish could enhance our understanding of the pathological processes underlying synapse loss in both genetic and acquired auditory synaptopathy. This knowledge is critical in order to develop therapies that protect or repair auditory synaptic contacts.

Keywords: zebrafish model system, hair cells (HCs), ribbon synapse, deafness/hearing loss, synaptic transmission

#### INTRODUCTION

Sensory hair cells in our inner ear must both reliably transduce and transmit auditory and vestibular stimuli (Harris et al., 1970; Eatock and Fay, 2006; **Figures 1A,B**). Hair cells transduce stimuli when apically localized mechanosensitive channels are activated, leading to graded depolarization of the hair-cell membrane (Gillespie and Walker, 2001; **Figure 1C**). Hair-cells transmit stimuli at the hair-cell synapse. At the synapse, hair-cell depolarization opens voltagegated calcium channels; calcium influx through these channels drives synaptic vesicle fusion and glutamate release onto innervating afferent nerves (Fuchs, 2005; Moser et al., 2006; **Figure 1C**). If either the hair cells or downstream afferent nerves are damaged or dysfunctional, the pathological consequence is sensorineural hearing loss. Sensorineural hearing loss can be caused by genetic factors, infections, toxins, age and excessive noise, and is the most common form of hearing loss in humans (90% of cases) (Eggermont, 2017; World Health Organization, 2018).

#### Edited by:

Rebecca Ann Wingert, University of Notre Dame, United States

#### Reviewed by:

Paola Rizzo, University of Ferrara, Italy Joachim Berger, Monash University, Australia

#### \*Correspondence:

Katie S. Kindt katie.kindt@nih.gov Lavinia Sheets sheetsl@wustl.edu

#### Specialty section:

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

Received: 30 April 2018 Accepted: 23 August 2018 Published: 11 September 2018

#### Citation:

Kindt KS and Sheets L (2018) Transmission Disrupted: Modeling Auditory Synaptopathy in Zebrafish. Front. Cell Dev. Biol. 6:114. doi: 10.3389/fcell.2018.00114

One form of sensorineural hearing loss—auditory synaptopathy—results when hair cell transduction is intact, yet synaptic transmission of sound stimuli from hair cells to downstream afferent nerves is disrupted. Auditory synaptopathy can result from genetic alterations that disrupt molecules requires for synapse function. Additionally, auditory synaptopathy and associated hearing loss can also be acquired though noise exposure (Bharadwaj et al., 2014). Both genetic and acquired auditory synaptopathy stem from dysfunction of specialized synapses in hair cells called ribbon synapses. Ribbon synapses have a unique presynaptic specialization called a synaptic ribbon or ribbon body which tethers synaptic vesicles near calcium channels at the presynaptic active zone (**Figure 1C**). Ribbon synapses are required for transmitting stimuli in a fast and sustained manner needed for precise sensory encoding, and are structurally and functionally unique from classic neuronal synapses (Fuchs, 2005; Moser et al., 2006). Understanding the unique properties of hair-cells synapses is an active area of study, and continued research is necessary in order to define the pathologies underlying auditory synaptopathy.

Our current understanding of how hair-cell synapses function, and the underlying causes of auditory synaptopathy, has been built by genetic studies in several model systems. In particular, the zebrafish model has been used to help identify and define the functions of genes important for hair-cell synapse function. Zebrafish hair cells are remarkably similar to mammalian hair cells at the molecular and cellular level (Coffin et al., 2004). Genetic studies have demonstrated that numerous genes required for hearing and balance in zebrafish are also required in mice and humans (Coffin et al., 2004; Nicolson, 2005; Varshney et al., 2016). In addition to genetic conservation, one significant advantage of studying hair cells and their synapses in zebrafish is the ability to study hair cells in vivo. In mammals, the inner ear is encased in bone, making it impracticable to study these sensory cells in their native environment. In contrast, larval zebrafish are transparent, and hair cells are optically accessible in whole larvae. In zebrafish larvae, hair cells are present both within the inner ear and the lateral-line system—a sensory organ used to detect the movement of water. The lateral line is made up of clusters of hair cells called neuromasts that are arranged in series along the fish body and head (**Figures 1A,B**). Neuromast

bodies of neurons in these ganglia project to and innervate hair cells in the lateral line. (B) An overview of the anatomy of a single patch of hair cells in the lateral line, referred to as a neuromast. Hair cells (pink) are surrounded by supporting cells (internal, blue and peripheral, orange) and innervated by both afferent (green) and efferent neurons (yellow). Mechanosensory hair bundles (purple) at the apex of hair cells project out into the water to detect local water flow. (C) Diagram of a single hair cell. Hair cells are activated when hair bundles are deflected, for example by local water flow. This apical deflection opens mechanosensitive channels allowing in cations including potassium and calcium. This apical activity depolarizes the hair cell, resulting in presynaptic calcium influx and release of glutamate onto the afferent neuron. Inset: magnified view of a hair-cell ribbon synapse. Shown are key evolutionarily conserved synaptic proteins discussed in this review. In hair cells, a presynaptic density called a ribbon (red) helps to recruit synaptic vesicles (white circles) to the synapse near clusters of calcium channels (CaV1.3). The ribbon is made up primarily of the protein Ribeye. Slc17A8 (Vglut3) and DMXL2 (Rbc3a) colocalize in or near synaptic vesicles. Synaptojanin and Otoferlin are also critical for ribbon-synapse function although their precise localization has not been definitively shown.

hair cells are particularly advantageous for hair-cell assessment because these cells are located superficially just beneath the fish skin, with their apical hair bundles protruding into the aqueous environment. This access makes it relatively straightforward to apply pharmacological agents, to stimulate the hair cells with fluid-flow, and assess hair-cell structure and function in vivo. Moreover, relative to mammals where hair cells mature over several weeks (Kraus and Aulbach-Kraus, 1981; Romand and Varela-Nieto, 2003), the hair cells in zebrafish mature rapidly (<24 h; Kindt et al., 2012; Dow et al., 2015), making it possible to study the entirety of hair-cell development in a single imaging session.

In addition to hair-cell accessibility and rapid development, the zebrafish model is valuable for hearing and balance research because, similar to mice, it is genetically tractable model system. Zebrafish are amenable to rapid genetic modification, including transgenic modification to express tissue specific transgenes encoding fluorescent markers or gene products (Kwan et al., 2007). The use of fluorescent markers is especially useful in the transparent larvae where hair-cell structures can easily be visualized in vivo and dynamic cellular processes can be imaged in a live, intact preparation.

In this review, we provide an overview of tools and techniques developed in the zebrafish model to examine haircell synapse structure and function. We also describe genetic studies in zebrafish that have helped define the roles of key hair-cell synaptic proteins. Given the recent advances in gene-editing technology, we highlight how zebrafish genetics could be applied to further our understanding of the genetic causes of auditory synaptopathy. Lastly, we outline preliminary studies that have explored the potential for using zebrafish to model noise-exposure and its associated excitotoxicity. We conclude with a discussion on how noise exposure studies in zebrafish could be expanded to further our understanding of the specific pathological changes that lead to acquired, noise-induced auditory synaptopathy.

#### TOOLKIT TO ASSESS HAIR-CELL SYNAPSE FUNCTION AND MORPHOLOGY IN ZEBRAFISH

Over the years, experimental techniques have been developed to study hair cells and hair-cell synapses in zebrafish. These techniques include: optical and ultrastructural analyses to visualize hair-cell synapse morphology, and functional assays to examine how hair cells transduce and transmit sensory stimuli. In the section below, we outline these methods and tools.

### Morphological Analysis of Hair-Cell Synapses in Zebrafish

Genetic mutations or environmental insults such as noise exposure can specifically affect the spatial organization of haircell synaptic structures (Paquette et al., 2016; Ryan et al., 2016; Song et al., 2016). In the mammalian inner ear, hair-cell synapses are commonly characterized ultrastructurally using transmission electron microscopy (TEM) to examine synapses in either single or serial-sections. In addition, these synapses can be examined using confocal microscopy to visualize immunolabel of haircell synaptic proteins (Liberman et al., 2011; Valero et al., 2017; Becker et al., 2018; Jean et al., 2018).

Similar to work in mammals, precise ultrastructural measurements can be obtained from zebrafish hair-cell synapses using TEM (**Figure 2A**). For example, in zebrafish, the synaptic ribbon can be seen clearly in TEM as an electron-dense region that is adjacent to the postsynaptic density on the innervating afferent neuron (**Figure 2A**, ribbon and PSD). TEM is the most accurate way to determine the size of the synaptic ribbon. TEM can also be used to visualize the synaptic vesicles tethered to the synaptic ribbon and near the active zone (**Figure 2A**, SVs). Currently TEM is the only method able to quantify the number and distribution of these synaptic vesicles populations. While these ultrastructural measurements are valuable, preparing, sectioning, imaging and analyzing TEM samples requires considerable time and effort. Moreover, in most cases, TEM is only able to capture a subset of synapses within each hair-cell organ.

In contrast to electron micrographs, quantitative analysis of immunolabeled epithelial whole mounts provide the advantage of being able to examine synaptic features of large numbers of hair cells. This advantage has been used by a number of groups to characterize relative variances in size and morphologies of pre- and postsynaptic components in mammalian and zebrafish hair-cell organs (Wong et al., 2014; Paquette et al., 2016; Suli et al., 2016; Becker et al., 2018; Jean et al., 2018). Further benefits of using the larval zebrafish lateral-line system for quantitative imaging of immunolabeled structures are twofold. First, the relative simplicity; each neuromast contains 10–16 hair cells with ∼3 synapses per cell of similar morphology, making comparative analyses of synapses in numerous lateralline organs straightforward (Sheets et al., 2012; Suli et al., 2016). This in contrast to mammalian auditory hair cells where the number and size of synapses can vary among individual hair cells depending on location within the Organ of Corti—the sensory organ for hearing (Meyer et al., 2009). Second, the synaptic ribbons in the zebrafish lateral line (TEM, circular 300 nm diameter; Suli et al., 2016) are on average larger compared to those in mammalian auditory hair cells (mouse TEM length (longest axis of the ribbon) ∼120 nm prehearing and ∼170 nm hearing; Wong et al., 2014)). These larger synaptic ribbons are near the resolution of light microscopy (∼200–270 nm) making it relatively straightforward to resolve these structures.

A previous challenge using the zebrafish lateral-line system for analysis of immunolabeled synaptic structures was a scarcity of antibodies that interact with zebrafish hair-cell synaptic proteins. In recent years a number of commercially available antibodies have been identified that label zebrafish synaptic components including synaptic vesicles (**Figure 2F**, CSP), synaptic ribbons (**Figure 2E**, CtBP), postsynaptic densities (**Figure 2E**, MAGUK), afferent (**Figure 2B**, ZN-12) and efferent (**Figure 2C**, Synaptophysin and **Figure 2D**, Vamp2 and TH) lateral-line neurons, and glutamate receptor subunits (Summary of Abs: **Table 1**). Immunolabels can be used in combination with

(pink) to label both afferent fibers and all efferent synapses respectively. (D) Efferent synapses, which can also be labeled with a Vamp2 antibody (pink), can be further sub-classified by a co-label such a tyrosine hydroxylase (TH, green; white overlap indicates dopaminergic synapses). (E) Pre- and post-synaptic densities can be labeled with CtBP (pink) and pan-MAGUK (green) antibodies respectively. (F) Synaptic vesicles, labeled with cysteine string protein (CSP, pink) are enriched at the basolateral membrane of hair cells near synaptic ribbons labeled with Ribeye antibody (green). Scale bar = 100 nm in (A) and 5 µm in (E) (for B–E) and (F).

transgenic lines that specifically mark cell types of interest such as hair cells, glia-like supporting cells, as well as the innervating neurons (**Figures 2B,C**, neurod:GFP; Obholzer et al., 2008; Behra et al., 2012, p. 1; Tabor et al., 2014; Toro et al., 2015).

Transmission electron microscopy and immunostaining provide important information regarding synaptic structure and the localization of proteins at hair-cell synapses. Unfortunately, there are not antibodies for all synaptic proteins, and these approaches do not provide temporal information regarding the dynamics of these proteins within hair cells. Therefore, transgenic fish lines expressing fluorescently tagged synaptic proteins provide a powerful way to determine the localization of these molecules in vivo (Trapani et al., 2009; Sheets et al., 2011; Einhorn et al., 2012). For example, transgenic fish expressing fluorescently tagged Ribeye, one of the main structural components of synaptic ribbons, have been used to identify the location of ribbons in developmental analyses and functional imaging experiments (Tg[myo6b:ribeye-mcherry]; **Figures 3F,G**; Pujol-Martí et al., 2014; Zhang et al., 2018). More recent work has used tagged proteins to investigate the structural dynamics of synaptic ribbons. For example, fish expressing fluorescently tagged Ribeye have been used along with fluorescence recovery after photobleaching (FRAP) to determine the stability and turnover of Ribeye within synaptic ribbons, and the exchange of Ribeye between synaptic ribbons (Graydon et al., 2017; Chen et al., 2018). In the future, the creation of additional zebrafish transgenic lines will provide a valuable resource in this in vivo model to study the localization and dynamics of hair-cell synaptic proteins.

### Assays for Hair-Cell Mechanotransduction in Zebrafish

Hair-cell function occurs primarily within two main structural domain, the apical hair bundle and the basal ribbon synapse. These structural domains are required for mechanotransduction and neurotransmission respectively. In response to sensory stimuli, apical hair bundles are deflected; this deflection opens mechanosensitive ion channels. This apical activity is essential to initiate hair-cell depolarization, and opening of calcium channels at the synapse, leading to sensoryevoked neurotransmission. Therefore, these two domains are functionally linked and, in order for proper hair-cell synapse function, apical hair bundles must be intact and functional. To



assay for normal mechanotransduction in zebrafish, microphonic potential measurements and FM 1-43 dye labeling can be used (**Figure 3A**; Nicolson et al., 1998; Seiler et al., 2004). While these measurements are most straightforward in lateral-line hair cells, they have also been adapted to examine hair cells in the zebrafish inner ear (Tanimoto et al., 2011). In the lateral line, microphonic potentials are recorded by placing an extracellular electrode near the apical hair bundles of an individual neuromast. Hair bundles are then deflected with a fluid jet and the flow of current into mechanosensitive ion channels in the bundles of all hair cells within a neuromast can be measured (Trapani and Nicolson, 2010; Olt et al., 2016b). For FM 1-43 analysis, larvae are briefly immersed in the vital dye FM 1-43. This dye rapidly enters hair cells when mechanosensitive ion channels are functional (**Figure 3A**; Seiler and Nicolson, 1999). These combined methods provide a way to ensure that hair-cell mechanotransduction is intact, and represent important assays in the characterization of zebrafish auditory and vestibular mutants. For example, mutants without microphonic potentials or FM 1- 43 label can be classified as mechanotransduction mutants. In contrast, mutants with intact microphonic potentials and FM 1-43 label indicate that there is a disruption downstream of mechanotransduction. Importantly, a subset of zebrafish mutants with normal mechanotransduction have been shown to be required for hair-cell synapse function. (See the genetics section below; Nicolson et al., 1998; Sidi et al., 2004; Obholzer et al., 2008; Trapani et al., 2009; Einhorn et al., 2012). Overall, these assays are useful in determining whether specific mutations disrupt components of the transduction apparatus, or potentially affect a process downstream, including hair-cell transmission.

#### Electrophysiological Approaches to Study Hair-Cell Synapses in Zebrafish

In addition to microphonics and FM 1-43, several additional methods have been established to assay hair-cell function, and specifically hair-cell synapse function in zebrafish. These methods utilize either electrophysiology or imaging-based approaches (Olt et al., 2016b; Zhang et al., 2016). In general, imaging approaches offer superior spatial resolution, while electrophysiological recordings offer greater sensitivity. Both of these methods have proved invaluable in the analysis of molecules required for haircell synapse function in zebrafish.

To study presynaptic function, the gold standard in the hair-cell field is whole-cell patch-clamp recordings. Using this approach, stimulatory voltage steps can be applied to electrically isolated individual hair cells in order to obtain important information on synapse function; for example, calcium currents

reliant on presynaptic calcium channels can be isolated and characterized (Brandt et al., 2005; Ricci et al., 2013; Olt et al., 2014). In addition, increases in cell-membrane capacitance that are associated with vesicle fusion can also be measured. These hair-cell recording have been pioneered in zebrafish by several groups (Ricci et al., 2013; Olt et al., 2014). Using this method, both presynaptic calcium currents and capacitance changes have been recorded in larval and juvenile zebrafish lateral-line hair cells (Olt et al., 2014, 2016a).

In addition to hair-cell electrophysiological measurements, recordings from the afferent posterior lateral-line ganglia have also been used to characterize hair-cell synapse function (Trapani and Nicolson, 2010). Within each neuromast there are two populations of hair cells that respond to water flow in two directions along a single axis. Afferent neurons that innervate neuromasts form bouton-like synapses on multiple hair cells that respond to one of these two directions (Nagiel et al., 2008; Faucherre et al., 2009; Sheets et al., 2011; Pujol-Martí et al., 2014). Therefore, electrophysiological recording from afferents represent the summed output of several hair cells within a neuromast.

Afferent recordings in zebrafish have primarily measured extracellular currents from afferent cell bodies in a loosepatch configuration. Initially, a successful afferent recording is identified by detection of spontaneous spiking that results from hair-cell neurotransmission in the absence of stimuli (**Figure 3D**; Trapani and Nicolson, 2011). Spontaneous spike rate on its own has been shown to be an important feature of synapticrelease properties in both zebrafish and mammals (Pfeiffer and Kiang, 1965; Merchan-Perez and Liberman, 1996; Furman et al., 2013; Sheets et al., 2017). After identifying a spiking afferent neuron, a fluid-jet can be used to deflect the hair bundles of each neuromast along its axis of sensitivity in order to identify an afferent neuron and neuromast that are synaptically paired (**Figures 3B,C**). Each afferent will respond to a single direction along the neuromast's axis of sensitivity (**Figures 3B,C**). Once this pairing is identified, many features relevant to hair-cell synapse function can be quantified using these recordings. For example, during fluid-jet stimulation, the timing and number of afferent spikes can be recorded. While spike number can provide important information on vesicle release and replenishment (Trapani et al., 2009; Einhorn et al., 2012; Sheets et al., 2017),

measurements such as first spike latency (time from stimulus to the first spike) can provide important information on the timing of vesicle release (Sheets et al., 2017). Additionally, during a sinusoidal stimulus where the fluid-jet is used to stimulate both populations of hair cells within a neuromast along its axis of sensitivity, phase-locking can be measured (**Figure 3C**). Phaselocking assesses how consistently the afferent neuron spikes during a particular phase of the sine stimulus and reflects the fidelity of neurotransmission. Overall, afferent recordings have been able to resolve subtle yet important differences in zebrafish mutants that alter hair-cell synapse function (Obholzer et al., 2008; Trapani et al., 2009; Einhorn et al., 2012).

### Using Functional Imaging to Study Hair-Cell Synapses in Zebrafish

Because the zebrafish model system is genetically tractable, work in the last decade has expanded toward using genetically encoded optical indicators to examine hair-cell function in zebrafish (Kindt et al., 2012; Esterberg et al., 2014; Zhang et al., 2016, 2018; Sheets et al., 2017). These studies have utilized transgenic zebrafish that express functional indicators in either hair cells or afferent neurons using cell-type specific promoters (**Figures 3E–G**). An important advantage of functional imaging over electrophysiogical recordings is the ability to resolve activity spatially among multiple cells and subcellularly within individual cells. In contrast, microphonics or extracellular afferent recordings are the readout of many hair cells within a neuromast, and whole-cell recordings examine activity from one hair cell at a time.

The majority of functional imaging studies have used genetically encoded indicators of calcium because calcium influx is an important feature of both apical mechanotransduction and basal neurotransmission in hair cells (**Figure 1C**). Initial work primarily used genetically encoded calcium indicators localized to the cytosol of lateral-line hair cells. During fluidjet stimulation, it was shown that calcium signals could be reliably measured in the cytosol (Kindt et al., 2012; Sheets et al., 2012, 2017; Sebe et al., 2017). Disrupting hair-cell mechanotransduction abolished these cytosolic signals indicating that they are mechanically evoked calcium signals. In addition, it was possible to determine the contribution of presynaptic calcium channels to these cytosolic calcium responses by using pharmacology or mutant analysis (Kindt et al., 2012; Sheets et al., 2012, see the genetics section below).

Unfortunately, it is challenging to use a cytosolic calcium indicator to understand the physiological properties of channel activity within hair cells. Therefore, more recent studies have used localized calcium indicators to examine subcellular activity within hair cells (Sheets et al., 2017; Zhang et al., 2018). In particular, membrane-localized calcium indicators have proved advantageous to assess localized calcium influx into the hair-cell bundle and at the hair-cell synapse (Zhang et al., 2018). This approach is particularly useful because both of these measurements can be made within the same cell by using either an apical or basal imaging plane respectively (**Figures 3E,E**'). To determine the location of synaptic ribbons and measure presynaptic calcium signals at sites of release, calcium indicator lines can be combined with an additional transgenic line that marks synaptic ribbons via a Ribeye-mCherry fusion protein (Tg[myo6b:ribeye-mcherry]; **Figures 3E–G**; Sheets, 2017; Zhang et al., 2018). Similar to work in hair cells, membrane-localized calcium indicators have also been used to assay postsynaptic activity in the afferent process beneath neuromast hair cells (**Figure 3G**). In response to fluid-jet stimulus, afferent calcium signals can be detected at sites adjacent to synaptic ribbons that are marked using a Ribeye-mCherry transgenic line (Zhang et al., 2018).

While calcium indicators have proved useful, other indicators of activity have been utilized to examine hair-cell synapse function in zebrafish. For example, a genetically encoded indictor called SypHy has been used as a readout of vesicle fusion at synaptic ribbons (**Figure 3F**; Zhang et al., 2018). In the future, SypHy may provide valuable information regarding the spatial properties of vesicle fusion at hair-cell synapses. At the postsynapse, instead of a calcium indicator, recent work has used a genetically encoded glutamate sensor to measure postsynaptic activity (Pichler and Lagnado, 2018). Overall, these functional studies highlight that the zebrafish is an excellent model to test the efficacy of using genetically encoded optical indicators, with the hope that they can ultimately be used in both zebrafish and mammalian models to assess hair-cell synapse function. With the advance of functional imaging, it also will be informative to combine electrophysiology and functional imaging in zebrafish to gain a more comprehensive understanding of the temporal and spatial properties underlying hair-cell synapse function.

### THE GENETICS OF HEARING LOSS AFFECTING HAIR-CELL SYNAPSES IN MAMMALS AND ZEBRAFISH

To date, numerous genetic studies in zebrafish, mice and humans have uncovered molecules required for hearing and balance (Nicolson, 2005; Safieddine et al., 2012; Pan and Holt, 2015). In zebrafish, mutants with hearing and balance defects were first identified behaviorally in large-scale forward genetic screens (Nicolson et al., 1998; Lin et al., 2016). These fish were initially identified as motility mutants with circling behavior, an indicator of vestibular dysfunction; additional screening showed these fish also lacked acoustic-vibrational startle responses, an indicator of deafness (Nicolson et al., 1998; Tubingen 2000 Screening Consortium). Over the last 20 years considerable work has focused on identifying the lesions underlying deafness in these zebrafish mutants. Concurrent work in mice and humans has revealed, perhaps not surprisingly, that orthologous genes, when mutated, cause deafness in zebrafish and mammals. This body of work supports functional conservation of deafness genes among vertebrates.

During characterization of these genes, zebrafish deafness mutants were classified based on morphological and functional assays. For example, one class of mutations disrupted overall hair-cell morphology while another class specifically affected hair-bundle integrity. In most zebrafish mutants with disrupted

hair-bundle integrity, mechanotransduction was also affected (Ernest et al., 2000; Seiler et al., 2004; Söllner et al., 2004; Gleason et al., 2009). Importantly, a distinct class of zebrafish mutants had normal hair-cell and hair-bundle morphology and intact mechanotransduction, indicating that that the affected genes altered function downstream of hair-bundle function. Further characterization of 6 of these zebrafish mutants revealed molecules that are specifically required for proper hair-cell neurotransmission: CaV1.3, Vglut3, Nsf, Rabconnectin 3α, Synaptojanin and Wrb (Sidi et al., 2004; Obholzer et al., 2008; Trapani et al., 2009; Mo and Nicolson, 2011; Einhorn et al., 2012; Lin et al., 2016).

In the section below, we outline how studies characterizing zebrafish auditory and vestibular mutants revealed the function of these molecules in hair-cell neurotransmission. Further, we discuss how these mutants have expanded our understanding of each molecule's contribution to hair-cell synaptic development, maintenance and function. To date, mutations in four of the synapse-associated genes identified above have also been associated with human hearing loss (Rodríguez-Ballesteros et al., 2008; Ruel et al., 2008; Baig et al., 2011). This genetic conservation between zebrafish and humans suggests that thereis alsofunctional conservation at hair-cell synapses, and further supports that the zebrafish model is useful for studying auditory synaptopathy.

#### CaV1.3; Sinoatrial Node Dysfunction and Deafness (SANDD) Syndrome

In humans, loss of function mutations in CaV1.3 (CACNA1D) results in Sinoatrial Node Dysfunction and Deafness (SANDD), a disorder whereby affected individuals have abnormal heart rhythms and severe deafness (Platzer et al., 2000; Baig et al., 2011). CaV1.3 channels are the presynaptic calcium channel required for neurotransmission at hair-cell synapses. The requirement for CaV1.3 channels in hearing is highly conserved; mutations in caV1.3 result in profound deafness in human, mice and zebrafish (Sidi et al., 2004; Brandt et al., 2005; Baig et al., 2011).

CaV1.3 channels are part of the L-type calcium channel family—they are uniquely sensitive to dihydropyridines and have large single channel conductance (Bean, 1989). CaV1.3 channels also activate with rapid kinetics at low voltages relative to other CaV1 channels (Lipscombe et al., 2004) and inactivate slowly in hair cells. These properties make CaV1.3 channels ideal for mediating rapid and continuous exocytosis. In zebrafish and mice, CaV1.3 channels cluster tightly at synaptic ribbons (Frank et al., 2009, 2010; Sheets et al., 2011; Wong et al., 2014). Presynaptic clustering of CaV1.3 channels is thought to be important to tightly couple calcium influx and vesicle release. Precise control of vesicle release is an important feature for reliable sensory encoding in hair cells (Brandt et al., 2003, 2005; Wong et al., 2014).

In addition to its role in hair-cell neurotransmission, CaV1.3 channels are essential for hair-cell development and synapse maintenance. While caV1.3 knockout mice initially form synapses, they progressively lose hair cells and postsynaptic afferents that innervate the remaining hair cells degenerate (Glueckert et al., 2003). In zebrafish caV1.3a mutants, lateral-line hair cells were also shown to initially form synapses, a phenotype that can be assessed in developing hair cells when larvae are 3-days-old, 2 days prior to the onset of lateral-line function (Trapani and Nicolson, 2010; Suli et al., 2016). Nevertheless, in the zebrafish, a progressive loss of juxtaposition between hair-cell pre- and postsynaptic components was observed in functionally mature hair cells of caV1.3a mutants just 2 days later (Sheets et al., 2012). Additionally, this work found hair-cell ribbons were significantly enlarged in both 3- and 5-day-old caV1.3a mutants. The rapid formation (by day 3) and subsequent loss of synaptic juxtaposition (at day 5) in zebrafish caV1.3a mutants is one example of how quickly phenotypic differences in synapse development and maintenance can be assayed in zebrafish. By taking advantage of these distinct developmental time points, this study also found that transiently (1 h.) treating the developing hair cells of 3 day-old zebrafish with dihydropyridine agonists and antagonists was able to rapidly decrease or increase presynaptic-ribbon size respectively (Sheets et al., 2012). In 5 dayold zebrafish, these compounds affected synaptic-ribbon size to a far lesser degree, revealing that calcium influx through CaV1.3 channels could dramatically influence presynaptic morphology during a critical window of development.

Cumulatively, these results support that functional CaV1.3 channels are necessary to properly form synaptic ribbons and to maintain pre- and postsynaptic juxtaposition in zebrafish hair cells. Although is it clear that CaV1.3-dependent presynaptic calcium influx regulates presynaptic size during zebrafish haircell development, it is less clear what role these channels play in synapse maintenance. It has been proposed that synapse maintenance could require release of synaptic vesicle contents (Mo and Nicolson, 2011). Consistent with this idea, zebrafish caV1.3a mutants lack evoked and spontaneous synaptic vesicle release (Trapani and Nicolson, 2011). Because formation and maintenance of hair-cell synapses in mutants lacking synaptic glutamate release appear relatively normal (vglut3−/−; see next section), it possible that the release of trophic factors from the synapse is required for maintenance of hair-cell synaptic connections (Fritzsch et al., 1997, 2004; Mo and Nicolson, 2011; Kersigo and Fritzsch, 2015). In the future, it may be informative to use zebrafish as a platform to screen for compounds that maintain synaptic juxtaposition in caV1.3a mutants in order to identify factors necessary for hair-cell synapse maintenance.

#### SLC17A8/vglut3; Autosomal Dominant Deafness-25 (DFNA25)

In human patients, autosomal dominant deafness-25 (DFNA25) is a progressive, high frequency non-syndromic hearing loss caused by a heterozygous mutation in the SLC17A8 gene encoding Vesicular Glutamate Transporter-3 (VGLUT3) (Ruel et al., 2008). Vglut3—a transporter that packages glutamate into synaptic vesicles—was identified independently in both zebrafish loss-of-function mutants and mouse knockouts as a critical component for hearing and hair-cell neurotransmission (Obholzer et al., 2008; Seal et al., 2008). These studies found that both zebrafish hair cells and mammalian auditory hair cells express vglut3. In zebrafish and mice, vglut3 mutant hair cells

have normal microphonic potentials, suggesting that Vglut3 is not required for mechanotransduction (Obholzer et al., 2008; Ruel et al., 2008). Additionally, vglut3 mutant mice and zebrafish show normal calcium responses, and in mice exocytosis is not altered. In zebrafish, despite normal mechanotransduction and evoked calcium responses in hair cells, no postsynaptic spikes are detected in the innervating afferent neurons (Obholzer et al., 2008; Sheets et al., 2012). Similarly, in vglut3 knockout mice, auditory nerves lacked responses to auditory stimuli, despite normal hair-cell calcium currents and exocytosis (Ruel et al., 2008).

In vglut3 zebrafish mutants, afferent innervation appears relatively normal (Sheets et al., 2012). There does not appear to be any major structural changes in presynaptic ribbon morphology beyond the normal variances in mice and zebrafish (Obholzer et al., 2008; Ruel et al., 2008). Overall, these mild synaptic morphology phenotypes in vglut3 mutants are quite different compared to caV1.3 mutants where the synapses ultimately degenerate. Synapses may be preserved due to the presence of normal calcium currents and hair-cell exocytosis despite an absence of glutamate release (Ruel et al., 2008). This supports an important role for exocytosis in the release of other trophic factors in synaptic maintenance. Relatively normal development and maintenance of hair cells in vglut3 knockout mice have made this deafness model a promising target of virally-mediated gene therapy (Akil et al., 2012).

Notably, there are a few morphological differences observed between Vglut3-deficient zebrafish lateral-line hair cells and Vglut3-deficient mouse auditory hair cells. Zebrafish vglut3 mutant synaptic ribbons have a reduced number of ribbonassociated vesicles, which is not observed in mouse auditory hair cells. Additionally, there is evidence that the glutamate transporter Vglut1 is also expressed in zebrafish hair cells, yet Vglut1 appears unable to compensate for neurotransmission in the vglut3 mutant (Obholzer et al., 2008). Further studies in zebrafish are needed to confirm the presence of Vglut1 in hair cells and to define its function in order to understand the unique functional role of Vglut3 in hair-cell neurotransmission.

#### DMXL2/Rabconnectin 3α; Autosomal Dominant Deafness-71 (DFNA71)

A recent study identified a heterozygous missense variant of DMXL2 that is associated with dominant, non-syndromic hearing loss in humans (Chen et al., 2017). Notably, this gene had been previously identified in a zebrafish hearing and balance mutant and represents an example of hereditary deafness gene identified in zebrafish prior to mice or humans. DMXL2 encodes Rabconnectin 3α—the α-subunit of the Rabconnectin protein complex. Rbc3α zebrafish mutants, when compared to vglut3 and caV1.3a zebrafish mutants, have relatively mild to moderate auditory and vestibular deficits (Einhorn et al., 2012). Phenotypically, similar to vglut3 mutants, both pre- and post-synaptic morphology appeared normal in rbc3α mutants. Subsequent analysis of Rbc3α localization revealed it was enriched basolaterally and overlapped with Vglut3 in hair cells, suggesting that Rbc3α is localized to synaptic vesicles.

This study found that zebrafish hair cells deficient in Rbc3α impacted Vacuolar-type H+-ATPase (V-ATPase) localization at the base of hair cells (Einhorn et al., 2012). V-ATPase generates a proton gradient and acidifies subcellular compartments, including synaptic vesicles. This gradient is important for the accumulation of glutamate into synaptic vesicles. These results suggested that rbc3α mutants could have deficient synapticvesicle acidification. To determine whether vesicles properly acidified, this study pioneered the use of lysotracker in hair cells. Lysotracker is a membrane-permeable vital dye that labels acidic organelles. In combination with a live presynaptic ribbon label (Ribeye-GFP), lysotracker brightly labeled rings around synaptic ribbons that likely correspond to ribbon-associated vesicles. The intensity of lysotracker labeling was dramatically reduced in rbc3α mutants, indicating reduced acidification of organelles and vesicles surrounding synaptic ribbons.

To better understand the relationship between Rbc3α and the V-ATPase, this work used the genetic tractability of the zebrafish model to rapidly express tagged proteins in hair cells. Transient expression of either the cytosolic (V1) or the membrane (V0) subunits of the V-ATPase in wild-type and Rbc3α deficient hair cells revealed that Rbc3α is required to traffic or assemble the V1 subunit at the base of hair cells (Einhorn et al., 2012). Therefore, Rbc3α is required for proper V-ATPase localization and ultimately synaptic vesicle acidification. Because synaptic vesicle acidification contributes to vesicular glutamate accumulation, this observation suggests that rbc3α mutant synaptic vesicles contain less glutamate. Accordingly, afferent recordings revealed reduced glutamate-dependent evoked release from rbc3α mutant hair cells. In addition, rbc3α mutants showed loss of fidelity of phase-locked spiking at higher frequencies (20 vs. 60 Hz) (Einhorn et al., 2012). Currently there is no other established model to study DMXL2-related human hearing loss. In the future, it will be interesting to test whether expression of the dominant human mutation in zebrafish hair cells also disrupts hearing and balance, which could provide further insight into the pathology underlying this genetic lesion.

#### Synaptojanin

Proteins in the Synaptojanin family are lipid phosphatases that play an important role in endocytosis and vesicle recovery at synapses (Harris et al., 2000; Song and Zinsmaier, 2003). Although no members of the Synaptojanin family have been associated with hearing loss in humans, mutations that abolish the lipid phosphatase activity of Synaptojanin 2 result in progressive age-related hearing loss in mice without any other accompanying phenotype (Manji et al., 2011). Synaptojanin 2 is expressed in auditory hair cells, and the progressive hearing loss observed in Synj2 mutants appears to be due to degeneration or loss of hair bundles, and sunken appearance of cell bodies, followed by hair-cell death. These observations support that Synaptojanin 2 plays an important role in hair-cell survival, but the process by which it contributes to cell survival is not understood.

While mice express Synaptojanin 2 in hair cells, zebrafish express Synaptojanin 1 (McDermott et al., 2007; Trapani et al., 2009). Synj1 zebrafish mutants, similar to rbc3α

zebrafish mutants, have moderate auditory and vestibular defects. These behavioral defects were accompanied by morphological disruptions in Synj1-defficient hair cells; specifically, basal membrane protrusions, or blebbing. These protrusions were dependent on functional Cav1.3 channels and were observed in 1/3 of synj1 mutant hair cells. In contrast to mice Synj2 mutants, other aspects of hair-cell morphology in zebrafish synj1 mutants appeared normal. Transmitted electron micrographs revealed fewer vesicles at synj1 ribbons, indicating vesicle recycling was impaired. Reduced vesicles and basal membrane protrusions in mutant synj1 hair cells led to deficits in synapse function; afferent recordings in synj1 mutants revealed a delay in afferent spike timing and impaired phase-locking in response to highfrequency stimuli (Trapani et al., 2009). Speculatively, impaired hair-cell membrane recycling could contribute to a progressive degeneration of hair cells. It will be informative in follow-up studies to examine whether disrupting Synj1 function impacts zebrafish hair-cell maintenance and survival in older larvae.

### Otoferlin and WRB (pwi); Autosomal Recessive Deafness-9 (DFNB9)

In human patients, mutations in the Otoferlin gene gives rise to neurosensory non-syndromic recessive deafness DFNB9 (Yasunaga et al., 1999), and is the cause of ∼1.4–5% of the cases of autosomal recessive hearing loss (Santarelli et al., 2015). Otoferlin exists in a long form containing six C2 domains (C2A-F) and a short form containing three C2-domains (Yasunaga et al., 2000). C2 domains are important for membrane localization and bind calcium (Lek et al., 2012). Mutations in nearly any of the C2 domains in the long form (C2B, C, D, E, or F) are linked to deafness in humans and mice (Yasunaga et al., 2000; Ramakrishnan et al., 2009), indicating that the presence of the long form is important for hearing. Knockdown of Otoferlin in zebrafish results in deafness (Chatterjee et al., 2015). Otoferlin is proposed to be an essential regulator of hair-cell neurotransmission, functioning to both couple calcium signaling with vesicle fusion and to regulate vesicle replenishment (Roux et al., 2006; Pangršic et al., 2015 ˇ ; Vogl et al., 2016; Michalski et al., 2017). Otoferlin's sequence identity and protein localization are highly conserved between divergent species (Goodyear et al., 2010). Additionally, Otoferlin's function also appears to be conserved; acoustic startle responses can restored in zebrafish otof knockdowns using exogenous mouse Otoferlin (Chatterjee et al., 2015).

A highly conserved role for Otoferlin in hair-cell synapse function is further supported by the identification of a gene important for hearing and vision in a large-scale mutagenesis screen. A null mutation the gene pinball wizard (pwi) resulted in zebrafish with impaired acoustic startle response, vestibular abnormalities and defective optokinetic response (Lin et al., 2016). Pwi encodes tryptophan-rich basic (WRB) protein, a small transmembrane protein found in the endoplasmic reticulum that is a receptor for insertion of tail-anchored (TA) proteins. Disruption of TA-protein membrane insertion would likely result in disruption of TA-protein trafficking, and numerous hair-cell vesicular proteins are TA, including Otoferlin. A subsequent report further examined zebrafish pwi mutants, and verified that Wrb is necessary for normal Otoferlin protein levels in hair cells and hearing in zebrafish (Vogl et al., 2016). Additionally, this study showed mutating wrb in mice disrupted ER-insertion of Otoferlin into vesicles, which greatly reduced Otoferlin levels in auditory hair cells. A reduction in Otoferlin levels contributed to impaired sustained exocytosis at Wrb-deficient hair-cell synapses and disruptions in sound encoding. These observations further support the functional conservation of hair-cell synaptic proteins between zebrafish and mammals. This works also demonstrates the effectiveness of using the zebrafish model to identify novel proteins involved in auditory synaptopathy and to define the molecular functions of these proteins in hair cells.

#### Ribeye

Ribeye is the main component of synaptic ribbons and is a presynaptic protein that is unique to ribbon synapses (Schmitz et al., 2000). Ribeye is a splice isoform of the transcriptional co-repressor CtBP2. As CtBP2 is a protein that regulates a number of diverse transcriptional targets, knockouts of CtBP2 are embryonic lethal (Hildebrand and Soriano, 2002). Biochemical studies have demonstrated that individual Ribeye subunits selfassociate, and this self-association may form synaptic ribbons (Magupalli et al., 2008). There is currently no known mutation in CtBP2/Ribeye that contributes to hearing loss in humans. Nevertheless, studies in zebrafish and mouse models have depleted or knocked out Ribeye expression (while leaving the transcriptional co-repressor CtBP2's function intact) to understand the role of the synaptic ribbon in hair-cells. These genetic studies motivated discussion of Ribeye in this section.

In zebrafish, there are two paralogs of Ribeye and both are found in hair cells. Two main studies have examined the role of these Ribeye paralogs in zebrafish hair cells; one study transiently knocked down Ribeye and the other study created a genetic mutant that permanently eliminated nearly all Ribeye in hair cells. Transient knockdown of both Ribeye transcripts during development resulted in reduced number of hair-cell synaptic ribbons which correlated with reduced afferent innervation and reduced afferent firing (Sheets et al., 2011). By contrast, while ribeye mutant zebrafish also eliminated synaptic ribbons, these genetic mutants did not appear to affect afferent innervation nor significantly disrupt lateral-line afferent firing properties (Lv et al., 2016). This latter work suggests that compensatory mechanisms may be engaged when Ribeye is severely and permanently depleted. One additional similarity observed with either transient Ribeye knockdown and ribeye mutants was that CaV1.3a channels failed to localize and cluster at the synapse. Despite this clustering defect, an enhancement of CaV1.3a channel currents was observed in ribeye mutants (Lv et al., 2016).

In the mouse knockout of Ribeye, the entire ribbon structure was shown to be absent in hair cells (Maxeiner et al., 2016; Becker et al., 2018; Jean et al., 2018). Yet the absence of synaptic ribbons in knockout mice did not disrupt CaV1.3 localization at the hair-cell synapse. Instead, Ribeye was shown to be important for presynaptic CaV1.3 calcium channel organization; without Ribeye there was a preponderance of small CaV1.3 clusters at each synapse instead of a single organized structure (Jean

et al., 2018). Functionally, Ribeye knockout mice showed minor auditory deficits despite the absence of synaptic ribbons in hair cells (Becker et al., 2018; Jean et al., 2018). Both zebrafish and mouse studies cumulatively revealed that loss of synaptic ribbons via mutation or knockout of Ribeye leads to surprisingly minor deficits hair-cell synapse function and support the idea that compensatory mechanisms exist in both model systems.

In addition to these loss of function models, work in zebrafish has also demonstrated that overexpression of exogenous Ribeye in hair cells can enlarge synaptic ribbons and influence synaptic activity (Sheets et al., 2017). Synaptic ribbons in hair cells overexpressing Ribeye were ∼2 fold larger and transmission electron micrographs showed that these synaptic ribbons had a greater number of synaptic vesicles relative to wild-type siblings. Hair cells containing enlarged synaptic ribbons had less tightly clustered CaV1.3a channels yet showed increased CaV1.3a channel currents and correspondingly larger ribbon-localized calcium signals. Despite larger calcium signals, there was no change in exocytosis or afferent spike number in response to strong stimulus. Importantly, enlarged synaptic ribbons resulted in a significant reduction in spontaneous afferent activity, and disrupted evoked release at the onset of stimuli. These results indicate that enlarging the synaptic ribbon can influence the activity of innervating afferent neurons and degrade sensory encoding. These observations may have clinical significance; in noise-exposed guinea pig, the synaptic-ribbon size gradient found in auditory hair cells is disrupted (Furman et al., 2013) and synaptic ribbon volume is increased (See section on noiseexposure below, Furman et al., 2013; Song et al., 2016). This increase in synaptic ribbon volume is accompanied by deficits in intensity and temporal coding by auditory nerve fibers (Song et al., 2016). An interesting prospect for future zebrafish work is to determine whether noise can also induce changes in synaptic ribbon size and whether these changes can influence afferent neuron function, and if so by what mechanisms.

### THE FUTURE OF USING ZEBRAFISH GENETICS TO STUDY HAIR-CELL SYNAPSES

Action potentials do not drive neurotransmitter release at haircell synapses. Instead, in order to convey information about timing and intensity of stimuli, hair-cell neurotransmission is driven by graded depolarizations (Glowatzki and Fuchs, 2002; Trussell, 2002). While hair-cell synapses contain many of the same molecular components as conventional synapses, such as presynaptic calcium channels and postsynaptic glutamate receptors, their specialized function may require synaptic proteins that are unique to hair cells. Indeed, a number of molecules that are required at neuronal synapses are not present in mammalian hair cells, including Munc-13 and CAPS (two important proteins for synaptic vesicle tethering and priming), Synaptotagmins 1 and 2 (calcium sensors for vesicle fusion), Complexins (which regulate vesicle fusion), and Synaptophysins (Safieddine and Wenthold, 1999; Strenzke et al., 2009; Johnson et al., 2010; Uthaiah and Hudspeth, 2010; Vogl et al., 2015). These important synaptic functions are instead thought to be accomplished by specialized hair-cell synaptic proteins. One notable example is Otoferlin which, as highlighted in this review, appears to act in place of Synaptotagmins and functions as a calcium sensor for vesicle fusion (Johnson and Chapman, 2010; Michalski et al., 2017).

Identifying the unique molecular players at ribbon synapses in hair cells has been hampered by the difficultly in acquiring a sufficient amount of material for biochemical and proteomic approaches (Uthaiah and Hudspeth, 2010; Kantardzhieva et al., 2012). Moreover, it is possible that proteins found at both conventional and hair-cell synapses may be present but not be functioning in the same way. For example, SNARE proteins that are required for vesicle fusion at conventional synapses may also be present in hair cells (Uthaiah and Hudspeth, 2010), but do not appear to be required in mouse hair cells for synapticvesicle fusion (Nouvian et al., 2011). In future studies, with the advent of CRISPR technology, zebrafish could be used as platform to rapidly and inexpensively identify what synaptic proteins are present in hair cells. Using this approach, it may be possible to identify the molecular equivalents of neuronal molecules that are not present in hair cells, and determine whether neuronal synaptic proteins have specialized functions when they are present in hair cells. In support of this idea, a recent study in zebrafish demonstrated that targeted mutagenesis of protein-coding genes using CRISPR-Cas9 is a powerful and high-throughput way to assess the role of candidate deafness genes identified in humans (Varshney et al., 2015). This work also highlights that the zebrafish model is a useful platform to not only rapidly evaluate the role of both known human deafness genes but also probe for yet unknown molecules that may be required at hair-cell synapses. CRISPR technology, combined with the functional and morphological toolkit outlined in this review, make zebrafish a favorable model to use toward determining the complete molecular composition of hair-cell synapses.

### NOISE EXPOSURE, EXCITOXICITY AND ACQUIRED HEARING LOSS IN ZEBRAFISH

In addition to gene mutations that cause hereditary forms of hearing loss, environmental factors such as intense or prolonged noise exposure can result in an acquired form of hearing loss. In humans, intense noise exposure can rapidly lead to profound hearing loss. In other cases following noise exposure, hearing loss is not profound but rather hearing sensitivity is diminished and higher sound pressure levels are needed to perceive a given stimulus (Mills et al., 1979). This diminished hearing sensitivity for a given stimulus is referred to as an elevated shift in hearing threshold.

After noise exposure, hearing loss can either be permanent or temporary depending on the intensity, duration and repetition of the exposure. Cumulatively, studies in mammals have demonstrated that noise exposures can result in damage or loss of hair cells, hair-cell synapses or the innervating afferent neurons (Bohne, 1976; Dinh et al., 2015; Liberman and Kujawa, 2017).

Intense noise exposures that result in permanent hearing loss are accompanied by progressive hair-cell death and loss of afferent neurons (Ryan et al., 2016). Notably in mice, intense impulse noise (i.e., blast) also appeared to result in a significant decrease in the number of hair-cell synapses in surviving hair cells (Cho et al., 2013). This synaptic pathology supports the hypothesis that intense noise exposures contributes to both hair-cell damage, synapse loss and ultimately a permanent, acquired hearing loss.

By contrast, moderate noise exposures are initially accompanied by elevated shifts in hearing threshold, but the thresholds eventually return to normal (Ryan et al., 2016). These noise exposures leave auditory hair cells intact, but contribute to afferent terminal swelling and a subsequent reduction in synaptic contacts, followed by progressive loss of auditory nerves (Kujawa and Liberman, 2009; Lin et al., 2011; Jensen et al., 2015). Currently it is hypothesized that, although clinical hearing thresholds return to normal, there may be subclinical hearing deficits associated with hair-cell synapse loss and afferent nerve degeneration (Bharadwaj et al., 2014). These deficits include difficultly resolving sounds in challenging listening environments such as discerning speech in a noisy room (Moser and Starr, 2016). Afferent-terminal swelling and synapse loss are thought to be a consequence of excess glutamate accumulation in the synaptic cleft during noise over-exposure, resulting in glutamate excitotoxicity (Puel et al., 1994, 1998; Hakuba et al., 2000; Ruel et al., 2005). Similar to genetic lesions that impair hair-cell synapse function, acquired noise-induced hearing loss resulting from a reduction in hair-cell synapses is a form of auditory synaptopathy (Moser and Starr, 2016).

#### Intense Noise Exposure in Zebrafish

While zebrafish have been used extensively to understand how ototoxic agents, such as aminoglycoside antibiotics and platinum-based cancer therapeutics, damage hair-cell organs (Coffin et al., 2010; Namdaran et al., 2012; Ou et al., 2012), less work has been done to model the toxic effects of noise damage on these organs. Currently there are only a few published studies exploring noise exposure paradigms in zebrafish (Schuck and Smith, 2009; Sun et al., 2011; Uribe et al., 2018). In studies examining auditory over-stimulation, adult zebrafish were exposed to a 100 Hz pure tone at 179 dB for 36 h. After this intense exposure, hair cells in the saccular epithelia (a hair-cell organ in the zebrafish inner ear used to detect sound (Schuck and Smith, 2009)) showed damage or loss of apical mechanosensory structures immediately following noise exposure. The damage was most apparent in the caudal region of the saccule which corresponds to a region sensitive to low frequency tones. Overall, the cellular damage following intense noise exposure was similar to damage that has been observed in mammalian models (Wang et al., 2002). Similar to the regenerative capability that has been demonstrated in the lateral-line system (Coffin et al., 2010; Namdaran et al., 2012; Ou et al., 2012), there was evidence of haircell proliferation in the adult zebrafish inner ear just days after noise exposure. Follow up work from this study later revealed that Growth Hormone (GH) may be important for hair-cell proliferation after this level of trauma (Schuck and Smith, 2009; Sun et al., 2011).

The role of GH is intriguing because work in mammals indicates that other hormones and neurotrophins including Thyroid hormone (TH), Neurotrophin-3 (NT-3), Brain-derived neurotrophic factor (BDNF) may be important for the survival and recovery of afferent terminals and synapses (Wan and Corfas, 2015). More specifically, studies in noise-exposed mice suggest NT-3/TrkC signaling promotes synaptic repair and regeneration in auditory hair cells (Wan et al., 2014; Suzuki et al., 2016), while BDNF/TrkB signaling regulates time-dependent noise sensitivity and protects against synapse loss during periods of wakefulness (Meltser et al., 2014). In zebrafish, it has been demonstrated that hair-cell synapse stabilization during normal development requires the protein N-ethylmaleimide-sensitive factor (Nsf) in order to release neurotrophic factors including BDNF (Mo and Nicolson, 2011). Future studies in the zebrafish model could provide mechanistic information toward how these trophic factors provide protection or promote repair of hair-cell synapses following noise exposure.

### Modeling Noise-Induced Excitotoxicity in Zebrafish

In addition to the few noise exposure studies in zebrafish, recent work has used pharmacology to model glutamate excitotoxicity associated with noise exposure (Sebe et al., 2017; Sheets, 2017). These models are based on the premise that, during noise exposure, excess glutamate accumulates in the synaptic cleft leading to over activation of ionotropic glutamate receptors (iGluR) and subsequent excitotoxic damage. Application of the iGluR agonists α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA), Kainic acid (KA) and N-methyl-D-aspartate (NMDA) to mammalian inner ears or hair-cell explants has also been used to mimic glutamate excitotoxicity associated with noise exposure. In mammals, application of AMPA or KA results in overactivation of the iGluR receptors mediating neurotransmission on postsynaptic afferent terminals (Puel et al., 1994; Zheng et al., 1997; Sun et al., 2001; Sasaki et al., 2012). Subsequently, the afferent terminals swell and, in the case of KA application on auditory hair-cell explants, neurites retract (Wang and Green, 2011). In cases of AMPA administration to inner ears, a small but significant percentage of inner hair cells are also lost 7 days following exposure (Hakuba et al., 2003; Hyodo et al., 2009). Two recent pharmacological studies have used iGluR agonists to mimic noise exposure in zebrafish larvae. These studies took advantage of the accessibility of the larval lateral-line organ to apply iGluR agonists externally, and image hair cells during and after drug exposure (Sebe et al., 2017; Sheets, 2017).

In one study, a single, short application of AMPA (100 or 300 µM AMPA, 15 min) was examined (Sebe et al., 2017). This treatment resulted in swelling of lateral-line afferent nerve terminals. This swelling was similar to what has been reported in mammal auditory system after AMPA exposure (Puel et al., 1994; Hyodo et al., 2009). Functional calcium imaging revealed a loss of activity in afferent nerve terminals after this treatment. Lateral-line hair cells, on the other hand, were not morphologically affected by these treatments and showed normal

mechanotransduction, suggesting glutamate excitotoxicity underlies afferent terminal damage and synaptopathy (Sebe et al., 2017). Importantly, this work demonstrated that the excitotoxic effects of AMPA occurred through calcium-permeable AMPA receptors (CP-AMPARs), as blocking CP-AMPARs prevented postsynaptic swelling and loss-of-function. Furthermore, this study used immunohistochemistry, electrophysiology and pharmacology to demonstrate that CP-AMPARs are present and mediate synaptic currents not only within the postsynapses of lateral-line afferents of the zebrafish lateral line, but also within rat and bullfrog auditory synapses (Sebe et al., 2017). The morphological and functional conservation of CP-AMPARs among species indicates the mechanism underlying glutamate excitotoxicity at hair-cell synapses may also be conserved.

A second study used similar methodology but applied iGluR agonists for longer durations (Sheets, 2017). Here the iGluR agonist NMDA or non-sensitizing AMPA/kainite receptor agonist KA were applied over longer time scales (10–600 µm; 1 h). Similar to what has been demonstrated in the mammalian auditory system after NMDA and KA application (Puel et al., 1994), this study found that NMDA did not cause appreciable swelling in the afferent nerve terminals, while KA was extremely potent and caused swelling and even bursting of afferent nerve terminals. In addition, exposure to either NMDA or KA induced apoptotic hair-cell death in a dose-dependent manner (Sheets, 2017). Remarkably, hair-cell death was independent of damage to post-synaptic terminals—loss of hair cells following NMDA and KA application occurred even in the absence of afferent neurons. Further, this work identified AMPA, Kainate and NMDA receptor subunits that appear to be expressed in hair cells, suggesting that presynaptic iGluR receptors may contribute to hair-cell excitotoxic damage.

#### THE FUTURE OF NOISE EXPOSURE AND EXCITOTOXICITY RESEARCH IN ZEBRAFISH

Currently there are no published zebrafish studies in adults or larvae using noise exposures to model acquired auditory synaptopathy in the ear. In the future, it will be useful to create protocols to modulate the intensity and duration of noise exposures in order to define the pathological changes in the zebrafish inner ear that are associated with moderate noise exposure. By modulating the intensity and duration of noise exposures in zebrafish, it will be possible to examine the dynamic progression of damage following moderate noise exposure, including loss of afferent fibers and synapses (Kujawa and Liberman, 2009; Shi et al., 2013; Song et al., 2016). In addition to noise damage paradigms to study pathology in the zebrafish inner ear, it will be experimentally worthwhile to develop approaches to directly mechanically over-stimulate the well characterized lateral-line organs. Here, noxious water flow could be used to over-stimulate lateral-line hair cells. Although a previous report outlined a microfluidic device that could be used to confer damage to the lateral-line system in larval zebrafish, no studies have demonstrated the effectiveness of this design (Kwon et al., 2014).

After establishing both moderate and intense noise exposure methods in zebrafish, it will be informative to apply the same tools and assays that have been used to understand the effects ototoxic drugs on hair-cell pathology. For example, the zebrafish lateral line has been used to screen for compounds that protect hair cells or promote hair-cell regeneration during and after ototoxic insult (Coffin et al., 2010; Ou et al., 2012; Esterberg et al., 2014, 2016; Hailey et al., 2017). Based on ototoxicity studies, it is likely that the zebrafish lateral-line system or inner ear could also be used as a screening platform to identify compounds that are protective during noise exposure or lateral-line over-stimulation. Alternatively, zebrafish could be used to identify compounds that promote synaptogenesis after synapse loss and afferent nerve damage.

These studies will be particularly advantageous in larval zebrafish where transgenic lines (**Figure 3**), can be used to image hair cells in vivo. Using live imaging, it will be possible to examine the morphology of hair cells, ribbon synapses and afferent nerve terminals during noise exposure, as well as during repair and regeneration. Currently, it is not possible to visualize these changes in the hair-cell organs in living mammals. The ability to visualize morphological changes during and after insults in whole animals is an important advantage to using zebrafish for these studies. These live imaging approaches could reveal the specific pathological changes accompanying both moderate and intense noise exposures. While imaging morphological changes accompanying noise exposure or lateral-line over stimulation will be invaluable, it will also be interesting to explore the functional consequences to the synapse during the noxious insult, as well as during recovery. These studies could be accomplished using electrophysiology and imaging-based methods that have been established in the zebrafish lateral line (**Figure 3**).

Finally, in addition to examining the morphological and functional consequences of noise exposure or lateral-line over stimulation in zebrafish, it will also be worthwhile to explore the downstream molecular mechanisms underlying the observed pathologies, as well as the recovery. In recent years, several models and methods have been developed to profile gene expression changes in specific cell types, including zebrafish hair cells (Steiner et al., 2014; Esterberg et al., 2016; Barta et al., 2018; Matern et al., 2018). It will therefore be beneficial and informative to use these approaches to define the pre- and postsynaptic molecular pathways underlying the pathologies during and recovery after noise damage.

### AUTHOR CONTRIBUTIONS

KK and LS wrote the review article and made the figures.

## FUNDING

This work was supported by NIH/NIDCD intramural research funds 1ZIADC000085-01 (KK), and NIH/NIDCD Grant R01- DC-016066 (LS).

### REFERENCES

fcell-06-00114 September 8, 2018 Time: 13:42 # 14


J. Neurosci. Off. J. Soc. Neurosci. 32, 11144–11156. doi: 10.1523/JNEUROSCI. 1705-12.2012


in the lateral line of zebrafish (Danio rerio). J. Assoc. Res. Otolaryngol. 4, 219–234. doi: 10.1007/s10162-002-3022-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 © 2018 Kindt and Sheets. 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.

# Leveraging Zebrafish to Study Retinal Degenerations

Juan M. Angueyra<sup>1</sup> \* and Katie S. Kindt<sup>2</sup> \*

<sup>1</sup> Retinal Neurophysiology Section, National Eye Institute, National Institutes of Health, Bethesda, MD, United States, <sup>2</sup> Section on Sensory Cell Development and Function, National Institute on Deafness and Other Communication Disorders, National Institutes of Health, Bethesda, MD, United States

Retinal degenerations are a heterogeneous group of diseases characterized by death of photoreceptors and progressive loss of vision. Retinal degenerations are a major cause of blindness in developed countries (Bourne et al., 2017; De Bode, 2017) and currently have no cure. In this review, we will briefly review the latest advances in therapies for retinal degenerations, highlighting the current barriers to study and develop therapies that promote photoreceptor regeneration in mammals. In light of these barriers, we present zebrafish as a powerful model to study photoreceptor regeneration and their integration into retinal circuits after regeneration. We outline why zebrafish is well suited for these analyses and summarize the powerful tools available in zebrafish that could be used to further uncover the mechanisms underlying photoreceptor regeneration and rewiring. In particular, we highlight that it is critical to understand how rewiring occurs after regeneration and how it differs from development. Insights derived from photoreceptor regeneration and rewiring in zebrafish may provide leverage to develop therapeutic targets to treat retinal degenerations.

#### Edited by:

Gokhan Dalgin, The University of Chicago, United States

#### Reviewed by:

Glenn Prazere Lobo, Medical University of South Carolina, United States Deborah Stenkamp, University of Idaho, United States

> \*Correspondence: Juan M. Angueyra angueyra@nih.gov Katie S. Kindt katie.kindt@nih.gov

#### Specialty section:

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

Received: 28 May 2018 Accepted: 20 August 2018 Published: 19 September 2018

#### Citation:

Angueyra JM and Kindt KS (2018) Leveraging Zebrafish to Study Retinal Degenerations. Front. Cell Dev. Biol. 6:110. doi: 10.3389/fcell.2018.00110 Keywords: zebrafish, retinal degeneration, regeneration, photoreceptor cells, Müller glia, developmental biology, rewiring, retinal circuitry and visual pathways

#### RETINA AND PHOTORECEPTORS

Similar to many organs, the eye is structurally well-conserved between zebrafish and mammals. For example, the eyes of zebrafish have the same gross structure as human and other mammalian eyes, and contain a cornea, lens, vitreous, retina, pigment epithelium, choroid and sclera (**Figure 1A**). Furthermore, the development of the eye during embryogenesis is also conserved, and complimentary work in zebrafish, mice and other species has helped to delineate the key developmental events in eye morphogenesis across vertebrates (Bibliowicz et al., 2011; Stenkamp, 2015).

Within the eye, the retina is of particular interest because it is the site of sensory detection and damage to the retina results in vision loss. The vertebrate retina is a highly structured neuronal tissue that lines the back of the eye. It is responsible for both the detection and processing of visual information, before it is relayed to higher-order visual centers. To achieve this, the retina is equipped with a variety of neurons that are arranged into three nuclear layers and project into two synaptic layers (**Figures 1B,C**). Within these layers, retinal neurons assemble into multiple, distinct circuits that encode different aspects of the visual information (Gollisch and Meister, 2010). The encoding of visual information starts when light is detected by the rod and cone photoreceptors. The highly sensitive rods are mainly used during dim-lighting conditions, while the more adaptable but less sensitive cones function from dawn until dusk. The retina of the nocturnal rodents commonly

allowing the identification of the nuclei of M- and L-cones between the rows of UV- and S-cones. (F) Diagram of the vertebrate retina and the retinal cells. Inset highlights the synapse between cones and horizontal and bipolar cells, where the cone synaptic terminal contains synaptic vesicles (white) attached to the synaptic ribbon (black). In close apposition to the ribbon, the dendrites on On-bipolar cells (On-BCs) invaginate into the synaptic terminal and are flanked by two horizontal cell (HC) processes. Off-bipolar cells make more basal contacts in close proximity but not apposed to the synaptic ribbon.

used in research like mice and rats is rod dominated (97% rods and 3% cones) (Carter-Dawson Louvenia and Lavail Matthew, 1979), as is the peripheral human retina. In contrast, zebrafish have a cone-dominated retina (∼40% rods and ∼60% cones) (Fadool, 2003), similar to the central human retina, which provides high-acuity vision, and is essential for most day-today visual tasks. Therefore, the zebrafish retina is uniquely positioned to understand the molecular mechanisms relevant to development and regeneration of the photoreceptors that are most relevant for human vision.

The zebrafish retina contains four different cone photoreceptor subtypes (UV-, S-, M-, and L-cones). Each subtype is defined by specific opsin expression that confers a particular wavelength-sensitivity, and morphology (short or long, and single or paired with another cone type) (**Figures 1D,E**). UV-cones express sws1, an opsin with peak sensitivity (λmax) in the ultraviolet range (λmax = 354 nm), and are short-single cones morphologically. S-cones express sws2, with peak sensitivity at short wavelengths (λmax = 416 nm), and are long-single cones morphologically. M-cones express opsins of the Rh2 class, which have undergone tandem quadruplications (Rh2-1 to Rh2-4), with peak sensitivities at mid wavelengths (λmax = 467 nm, 476 nm, 488 nm, and 505 nm respectively). L-cones express one of two tandemly duplicated opsin genes from the lws class, with peak sensitivities at longer wavelengths (λmax = 548 nm and 558 nm). M- and L-cones are morphologically arranged as a double cone, where the L-cone is the long (or principal) member of the pair and the M-cone is the short (or accessory) member (Raymond et al., 1996; Vihtelic et al., 1999; Chinen et al., 2003).

In both rod and cone photoreceptors, light triggers the activation of opsins followed by the rest of the phototransduction cascade. In zebrafish and in mammals, this cascade ultimately leads to changes in photoreceptor membrane potential, and to modulation of neurotransmitter release in the synaptic terminal. Visual information is then directly transmitted from photoreceptors to several subtypes of horizontal (inhibitory

interneurons that locally modulate photoreceptor synaptic output) and bipolar cells (glutamatergic neurons that transmit light signals into the next processing layer). Information is further processed in the next synaptic layer, where bipolar cells (BC) provide excitation to ganglion cells (glutamatergic spiking neurons), while amacrine cells (local interneurons) provide modulation pre- and/or post-synaptically. The axons of the retinal ganglion cells (or RGCs) form the optic nerve, and relay the pre-processed visual information to central targets in the brain. Additionally, the retina contains microglia (resident immune cells located primarily in the synaptic layers) and two types of true glial cells: Müller cells (a type of radial glia) and astrocytes (associated with axons of RGCs) (**Figure 1F**). The photoreceptors also closely associate with the retinal pigment epithelium (RPE), which provides structural, trophic and metabolic support and is directly involved in the recycling of opsins.

### RETINAL DEGENERATIONS

Within the retina, rods and cones are particularly vulnerable to metabolic, genetic or environmental insults; phototransduction and synaptic signaling are demanding processes that require high metabolic rates. Retinal degenerations (RD) are disorders characterized by photoreceptor death and can affect both rods and cones, or each photoreceptor type individually. RD have multiple causes that can be broadly divided based on whether the cause is primary or secondary. In primary RD, photoreceptors are directly affected, for example due to mutations that affect phototransduction proteins. In secondary RD, other cells types, for example the RPE or a single type of photoreceptor degenerates; this can lead to secondary degeneration of other photoreceptors. The main forms of RD in humans are Age-related Macular Degeneration (AMD), Retinitis Pigmentosa (RP), and Leber Congenital Amaurosis (LCA). AMD is a multi-factorial disease that affects the RPE. Advanced RPE dysfunction leads to secondary rod and cone loss. In industrialized countries, AMD is the leading cause of blindness (De Bode, 2017). AMD is associated with age, smoking, nutritional deficiencies, inflammation and mutations or polymorphisms in more than 30 genes, with many more genes still to uncover (Jager et al., 2008; Warwick and Lotery, 2018). RP encompasses a set of complex hereditary diseases that can be caused by a plethora of mutations mapped to more than 70 human genes (Daiger et al., 2013; Farrar et al., 2017) 1 . Most forms of RP in humans are characterized by an initial death of rods, with subsequent secondary cone death. RP presents clinically first with night-blindness, followed by decreases in peripheral visual acuity (tunnel vision) that eventually progresses toward the central areas of the retina. LCA encompasses a group of early-onset and progressive rod and cone dystrophies. Again, 25 genes have already been identified as causes of LCA, and most are expressed either in photoreceptors or in the RPE (Kumaran et al., 2017).

Early forward genetic screens in zebrafish that evaluated defects in the histology of the photoreceptor layer identified many genes that cause RD (Brockerhoff and Fadool, 2011). Most human homologs of these genes have also been shown to cause RD in humans. Subsequent screens in zebrafish assayed for defects in visual function rather than morphological defects. These screens exploited reliable visual behavioral assays in larvae, including the optokinetic response (OKR) (Brockerhoff et al., 1995) or the escape response to moving dark objects (Li and Dowling, 1997) (see below). Together these screens helped, not only to establish additional genetic models for photoreceptor degenerations, but also to identify additional and in some cases novel genes involved in RD. For example, one zebrafish mutant that lacked OKR was linked to a mutation in pde6c, a novel cone-specific phototransduction gene. Mutations in pde6c cause cone degeneration in zebrafish with secondary rod degeneration (Stearns et al., 2007). The discovery of this mutant was beneficial in two ways. First pde6c mutants were used to confirm the existence of rod-specific progenitors in the zebrafish retina (Morris et al., 2008). Second this mutant was used to identify pde6c as the causative gene in cone-photoreceptor loss of function 1 (cpfl1), a specific type of cone degeneration in mouse and humans (Chang et al., 2009). This is an excellent example of how zebrafish can be used to identify novel genes and pathways involved in RD. More recently, zebrafish models of RD are being leveraged in pharmacological screens, to test or find novel treatments for human RD (Moosajee et al., 2008; Ganzen et al., 2017). Overall these zebrafish genetic screens highlight the conservation of molecules underlying RD between zebrafish and mammals. Given this level of conservation, it is likely similar molecules may be required to regenerate and rewire the zebrafish and mammalian retina after RD.

### THERAPIES FOR RD

In humans, photoreceptor loss in RD is permanent (regardless of their diverse causes and speed of progression) and therefore remain largely untreatable and lead to progressive loss of vision and ultimately blindness. In early stages of RD in humans, current treatments include neuroprotective agents (Trifunovic et al., 2012) and antibody therapies in cases where the underlying mutation is well characterized (Lazic and Gabric, 2007). Unfortunately, these treatments only slow down the progression of disease and have variable outcomes (Pardue and Allen, 2018). In addition to these treatments, gene therapy has been used to improve vision in patients with LCA caused by mutations in RPE65, an RPE-specific protein involved in the recycling of retinoids (Cideciyan, 2010; Jacobson et al., 2012), but the improvement may not be long-lasting (Jacobson et al., 2015).

In late stages of RD in humans, when there is widespread photoreceptor loss, two distinct approaches for treatment exist. The first seeks to bypass the need for photoreceptors. This can be achieved by either making the surviving retinal bipolar or ganglion cells photosensitive using optogenetics (Busskamp et al., 2010; Yue et al., 2016) or synthetic photo-switchable compounds (Polosukhina et al., 2012; Tochitsky et al., 2017). Additionally,

<sup>1</sup>http://www.sph.uth.tmc.edu/retnet

retinal prostheses capable of stimulating RGCs directly have been developed, attempting to encode visual information directly into these output neurons (da Cruz et al., 2016; Lewis et al., 2016). Of note, the use of retinal prostheses for blindness was approved by the European Union in 2011, and by the FDA in 2013. Use of these prostheses has led to some successful reacquisition of very basic visual functions but only for limited periods of time (Mills et al., 2017). The second and more promising approaches aims to replace the lost photoreceptors by transplantation or by stimulating regeneration. These later approaches have received special attention because they have the capability of renewing the native function of the retina, and could provide a real cure for RD. Due to this potential for complete functional recovery, and because of recent and important developments in the field, transplantation and regeneration will be a focus of this review.

#### BARRIERS IN PHOTORECEPTOR TRANSPLANTATION AS A THERAPY FOR RD

Just a decade ago, the prospect of producing photoreceptors from stem cells seemed like an overly daunting task (Adler, 2008). Nevertheless, in the last few years, several laboratories have successfully developed protocols to produce eyecup-like structures from induced-pluripotent stem cells (iPSCs) in the span of weeks. Some of these eyecups are able to acquire a layered structure reminiscent of the retina, with photoreceptorlike cells that contain outer segments, express phototransduction proteins (Wahlin et al., 2017), and have some capacity for light responsiveness (Zhong et al., 2014), and vesicular release (Wahlin et al., 2017). The successful development of these eyecups opens the possibility of harvesting cells from an individual to generate iPSCs, and re-differentiate them into photoreceptors that could be then transplanted back into patients with RD. Based on these prospects, recent work in the retinal field has focused on using mice to explore photoreceptor transplantation as a therapy for RD.

Initial transplantation studies in mice attempted to introduce rod photoreceptors, with the best rates of integration (never surpassing a few percent) achieved by transplanting immature rod precursors (MacLaren et al., 2016). Follow-up studies presented equally promising examples of integration, and in some cases demonstrated functional recovery of vision (Santos-Ferreira et al., 2015; Smiley et al., 2016). However, it has recently been discovered that many of these results are due to the exchange of cytoplasmic material (including RNA and/or proteins) between donor cells and the host retina, and not integration of transplanted photoreceptors (Pearson et al., 2016; Santos-Ferreira et al., 2016; Singh et al., 2016; Ortin-Martinez et al., 2017). In light of this recent discovery, it will be important to carefully interpret how functional recovery of vision occurred after transplantation/cytoplasmic exchange in degenerated or degenerating retinas (Homma et al., 2013; Singh et al., 2013; Santos-Ferreira et al., 2015). Even if cytoplasmic exchange results in recovery photoreceptor function, it occurs at very low rates (a few percent of host positive cells, for transplantations of tens to hundreds of thousands of donor cells). Such low rates casts doubt on the prospect of leveraging this process as a viable therapeutic strategy, especially in advanced cases of degeneration.

Even with viable evidence for successful photoreceptor transplantation (Waldron et al., 2018), there are additional concerns for this type of therapy. For example, subretinal injection of a mass of cells, the most common transplantation method, causes inflammation and scarring, and inhibits the migration of transplanted cells (Barber et al., 2013). Additionally, it is also unclear if transplanted photoreceptors are capable of rewiring properly into the host retina. This problem is further compounded by our incomplete understanding of how photoreceptors normally wire during development. To date, only a handful of genes are known to be involved in synapse formation between photoreceptors and their postsynaptic targets (Simmons et al., 2017; Zhang et al., 2017; Miller et al., 2018; Sarria et al., 2018; Ueno et al., 2018). Despite this work, we still have little insight on the molecules that drive the initial recognition between these cells, or on the processes that promote, inhibit or refine synapse formation. Unveiling genes involved in photoreceptor synapse formation during normal development in zebrafish could provide direct therapeutic targets to promote rewiring of transplanted photoreceptors.

#### USING ZEBRAFISH TO EXPLORE RETINAL REGENERATION AS A THERAPY FOR RD

Cumulatively, work on transplantation therapies has highlighted that alternative therapies, such as photoreceptor regeneration, could be a promising alternative. Unfortunately, in mammals there is no regeneration in the retina after damage or RD. In contrast, the zebrafish retina has the innate capacity for regeneration. This capacity may be due to the continued growth of the zebrafish retina into adulthood, as well as the ability of the zebrafish to maintain a population of multipotent stem cells within the retina.

Larval zebrafish form a functional visual system by 4 days post fertilization (4 dpf), and are able to perform complex visual guided behaviors (like prey capture, see below) by 5 dpf (Patterson et al., 2013). This rapid onset of sensory function is critical to survival of the animal. As larvae progress into adulthood, zebrafish continue to grow in size. This growth requires organs like the eye and retina to grow as well. In the retina this growth occurs in the ciliary marginal zone (CMZ). The CMZ maintains a niche of pluripotent cells at the edge of the retina that continually adds neurons in peripheral concentric rings (Centanin et al., 2011). In addition to this continual growth, zebrafish can also regenerate their retinas after injury. In fact, robust retinal regeneration and rewiring have been demonstrated in genetic models of RD and in other models that incur retinal injury. Overall, given that zebrafish is a genetically tractable model with active retinal regeneration, it is poised to uncover the molecular processes that control retinal regeneration and rewiring.

In teleost fish, retinal regeneration after injury has a rich history. It was first reported in goldfish (Lombardo, 1968) and later in cichlids (Johns and Fernald, 1981) and trout (Julian et al., 1998). In zebrafish retinal regeneration is robust after resection (Cameron, 2000), mechanical damage (Fausett and Goldman, 2006), light damage (Bernardos et al., 2007; Thomas et al., 2012), thermal damage (Raymond et al., 2006), pharmacological damage (Fimbel et al., 2007; Sherpa et al., 2008; Nagashima et al., 2013; Tappeiner et al., 2013; Sherpa et al., 2014) and selective ablation of particular cell-types (Montgomery et al., 2010; D'Orazi et al., 2016; Hagerman et al., 2016; Yoshimatsu et al., 2016; White et al., 2017). In teleosts, regeneration can occur from cells generated in the CMZ (Raymond et al., 2006), a dedicated population of progenitors that are committed to a rod fate (Bernardos et al., 2007; Morris et al., 2008), and the Müller glia (Fausett and Goldman, 2006; Bernardos et al., 2007; Fimbel et al., 2007).

Due to its location at the edge of the retina, the CMZ is only involved in regeneration if the injury involves the peripheral retina. During regeneration, the CMZ is capable of giving rise to all retinal neurons except rod photoreceptors (Stenkamp et al., 2001; Raymond et al., 2006). Instead, rods originate from rodspecific progenitors. These progenitors were first identified in goldfish and cichlids (Johns and Fernald, 1981) and were later found in other teleost fish including trout and zebrafish (Julian et al., 1998). Rod-specific progenitors are thought to be important for maintaining the density of rods as the eye grows, and lineage tracing revealed that these rod progenitors derive from Müller cells that slowly and continuously divide in the normal retina (Otteson et al., 2001; Raymond et al., 2006; Bernardos et al., 2007; Nelson et al., 2008). During regeneration, there is an expansion in the number of photoreceptor progenitors, but these mainly derive from actively dividing Müller glia. In fact, in zebrafish the Müller glia are the primary source of regenerated neurons after injury. During regeneration they can act as multipotent stem cells, dividing and differentiating into any retinal cell type (Ramachandran et al., 2010b). In contrast to zebrafish, in humans and other mammals Müller glia do not remain multipotent and therefore cannot readily replace lost neurons in the retina. Because Müller glia are the primary source of regenerated retinal neurons and can regenerate all retinal neurons, considerable work has been dedicated to understanding the differences between the Müller glia of zebrafish and mammals.

A series of studies that investigated the response of zebrafish Müller glia to retinal injury, have unveiled the key transcription factors in a gene regulatory network that controls retinal repair. Shortly after injury, cytokines and growth factors activate the beta-catenin and stat3 pathways (Kassen et al., 2007; Wan et al., 2014). These pathways upregulate the expression of ascl1 (Fausett et al., 2008), a key transcription factor that (through lin-28) leads to the suppression of let7 microRNA (Ramachandran et al., 2010a). In the uninjured retina, let7 normally represses the expression of many regeneration-induced genes (including ascl1 and lin-28), closing the loop of a system poised to control Müller glia response to injury (Wan and Goldman, 2016) (**Figure 2A**).

In contrast, mammalian Müller glia does not readily divide (Wan et al., 2008) and responds to retinal injury with an inflammatory response known as reactive gliosis, characterized by an increase in size and overproduction of intermediate filaments, and leading to distortion of the architecture of the retina without repair (Dyer and Cepko, 2000; Bringmann et al., 2009; Thomas et al., 2016) (**Figure 2C**). Significant efforts have been devoted to understanding the differences between these species, in the hope of stimulating regeneration in mammals. This work has shown that ascl1 is not upregulated in mice after retinal injury (Karl et al., 2008), but ascl1 overexpression in mammalian Müller cells in vitro is sufficient to induce production of neurons (Pollak et al., 2013). Moreover, induction of expression of ascl1 in Müller cells in vivo, followed by retinal injury, induces division and production of all classes of retinal neurons, but only in young mice (Ueki et al., 2015). Recently, a successful report of regeneration in adult mice has shown that new bipolar- and amacrine-like cells derived from Müller glia can rewire into the retina. For this work, in addition to overexpression of ascl1, inhibition of histone deacetylation was also required (Jorstad et al., 2017). Unfortunately, no other retinal cell types are produced with this protocol (**Figure 2D**). Further insight into retinal regeneration has been gained from studying a related telost, medaka, which shows a restricted capacity for regeneration. In medaka fish, after injury, Müller glia do not readily proliferate, and new retinal progenitors commit almost exclusively to a photoreceptor fate. Comparisons in the Müller glia response to retinal injury in both medaka fish and zebrafish concluded that sustained expression of the transcription factor sox2 in adult Müller cells is key for maintaining multipotency (Lust and Wittbrodt, 2018) (**Figure 2B**). While it is clear that important strides have been taken to attempt retinal regeneration in mammals, before regeneration can be used as a viable therapy, we need a deeper understanding on the mechanisms that maintain cells with a regenerative potential in zebrafish throughout adulthood.

### INSIGHTS FROM ZEBRAFISH ON REWIRING AFTER REGENERATION

Even after successful transplantation or regeneration of photoreceptors, the biggest hurdle in these RD therapies is ensuring that the new photoreceptors rewire into the appropriate retinal circuits so that they are able to restore normal visual function. Once again, zebrafish has offered a unique opportunity to study rewiring after injury and regeneration, in both larvae and adults. Cumulatively, this work has demonstrated that the extent and time course of regeneration and rewiring is determined by lesion-specific differences, in particular the extent of injury and the number of cells that need to be replaced.

Adult teleosts are able to regenerate their retinas even after extensive retinal damage. Early seminal studies in adult goldfish, and later studies in adult zebrafish, showed robust retinal regeneration and rewiring after surgical retinal extirpation (Hitchcock et al., 1992; Cameron, 2000) or pharmacologically induced death of all retinal neurons (Raymond et al., 1988; Sherpa et al., 2008). Under these lesion paradigms, all retinal cell types were regenerated. Importantly, with regards to rewiring, the retinal lamination was reestablished (Raymond et al., 1988;

inhibits the production of let7, releasing the inhibition of this pathway. Expression of the transcription factor Pax6b (normally inhibited by the microRNA mi203), allows the amplification and production of retinal progenitors which are then able to redifferentiate into any retinal neurons or Müller glial cells. (B) In medaka fish, lack of production of Sox2 after the production of retinal progenitors restricts their fate to photoreceptors and does not allow the production of other cell types, including new Müller glia. (C) In mice and other mammals, retinal injury does not lead to expression of ascl1 and the rest of regeneration-related genes. Instead Müller glia activate the production of intermediate filaments and increase their size (reactive gliosis) and do not lead to the production of new retinal cells or injury repair. (D) Retinal regeneration can be stimulated in mice by artificially overexpressing ascl1 and inhibiting epigenetic changes (histone deacetylation in particular). With this treatment, Müller glial cells are able to divide and produce retinal progenitors, but their fate is restricted to bipolar and amacrine cells.

Sherpa et al., 2008), and synaptic connections were reformed between retinal neurons (Hitchcock and Cirenza, 1994) and between RGCs and the brain (Stuermer et al., 1985). Parallel work demonstrated that after extensive retinal damage in adult goldfish and zebrafish, visual function is also recovered (Mensinger and Powers, 1999, 2007; Lindsey and Powers, 2007; Sherpa et al., 2008). Nevertheless, because the Müller glia are the main source of regenerated retinal neurons, complete ocular excision prevents regeneration (Mensinger and Powers, 2007).

In adult teleosts, although there is robust regeneration after extensive retinal injury, the time required for regeneration and functional recovery depends on the extent of injury. For example, the differences in regeneration and rewiring were examined after surgical extirpation of ∼25, 50, and 75% of the adult retina. This work found that after surgical extirpation of 25% of the retina, 17 weeks (120 days) are required for regeneration and reestablishment of lamination and 25 weeks (180 days) for functional recovery. Additional extirpation lengthened the time and extent of recovery for both lamination and functional recovery (Mensinger and Powers, 2007). Similar results were observed in adults after pharmacological damage and death of all retinal neurons (induced by high doses of intraocular ouabain). In this study functional recovery started 5 weeks after the injury with further improvement by weeks 7 – 10, albeit with decreased sensitivity (Mensinger and Powers, 1999; Lindsey and Powers, 2007).

In adult zebrafish, the extent and time course of retinal regeneration and rewiring is also dependent on the injury. Extensive pharmacological damage of all retinal layers with ouabain leads to regeneration, and the newly formed cells are capable to reorganize into the three distinct nuclear layers by week 3 after injury. After 14 weeks, the regenerated retinas are well laminated (clear nuclear and plexiform layers), the optic nerve has regrown, and there is functional recovery (Sherpa et al., 2008). Interestingly, with pharmacological damage (using lower doses of ouabain) that spares photoreceptors and Müller glia but still induces a loss of cells in the INL and GCL, regeneration is faster, with significant recovery of function after only 8 weeks (Sherpa et al., 2014; McGinn et al., 2018). In this lesion paradigm, rewiring of a specific (but heterogeneous) subset of regenerated BCs was closely examined. The regenerated BCs had largely normal morphology, and, as a population, were able to reproduce the diversity of connectivity patterns observed in the surviving photoreceptors, with only a few errors in lamination or abnormal

dendritic or axonal arborization, again emphasizing the robust regeneration of zebrafish (McGinn et al., 2018). Nevertheless, in the context of extensive damage, regeneration in adult teleosts is far from perfect. Several structural defects are common including areas with defects in the formation or absence of plexiform layers, disorganization of nuclear layers, presence of cells in the incorrect layer (e.g., RGCs in INL), failure to reestablish the photoreceptor mosaic, formation of photoreceptor rosettes, overproduction of neurons, and the generation of cell types that were not initially damaged (Raymond et al., 1988; Hitchcock et al., 1992; Cameron, 2000; Vihtelic Thomas and Hyde David, 2000; Stenkamp et al., 2001; Stenkamp and Cameron, 2002; Sherpa et al., 2008; Powell et al., 2016).

More recently, the genetic tractability of zebrafish has enabled researchers to damage specific retinal cell types and study their rewiring after regeneration. This work was accomplished by using the recently developed nitroreductase–metronidazole (NTR–MTZ) system. For this method, transgenic zebrafish are created expressing the NTR gene (nfsb) under the control of a cell-specific promoter. When these transgenic zebrafish are treated with the compound MTZ, the NTR converts MTZ into a cytotoxic compound. Because this compound does not diffuse to neighboring cells, the resulting ablation is restricted to the NTR-expressing cells. Importantly, this process is reversible, and removal of MTZ solution makes it possible to examine regeneration and rewiring. To date, the NTR-MTZ system has been used to selectively ablate rods, and specific subtypes of cones, bipolar cells and glial cells (Zhao et al., 2009; Ariga et al., 2010; Montgomery et al., 2010; Fraser et al., 2013; D'Orazi et al., 2016; Hagerman et al., 2016; White et al., 2017). Importantly, several of these studies demonstrated that other cells in the retina that did not express NTR were not ablated after MTZ application. In addition, these treatments did not affect the surrounding retinal architecture (Zhao et al., 2009). This highlights the specificity and power of the NTR-MTZ system.

After genetic ablation and removal of MTZ, in each instance the targeted cells regenerated after several days, although the exact time-course varied depending on the ablated cell type and the age of zebrafish treated. For example, after using the NTR–MTZ to completely ablate rods in adult zebrafish, newly generated rods were identified within a week after removal of MTZ, and repopulation of rods attained pre-injury levels within 4 weeks (Montgomery et al., 2010) – a very similar time course required for the regeneration of cones in adults (Raymond et al., 2006; Bernardos et al., 2007). In larvae, the regeneration of cells occurs at a faster timescale. For example, after ablation of rods using the NTR-MTZ system in 5 dpf larvae, newly formed rods attained control levels in just 6 days (White et al., 2017). Similarly, cones ablated between 4 and 6 dpf regenerate in 7 – 10 days (Fraser et al., 2013; Yoshimatsu et al., 2016) and BCs ablated at 7 dpf regenerate in 13 days (D'Orazi et al., 2016).

In the majority of these studies, regeneration was confirmed morphologically. In a subset of studies, after regeneration, the analysis was extended to include behavior or rewiring. For example, in one study, either the UV- or S-cones were ablated (in 7 dpf larvae) and the optomotor response (OMR) (see below) was assayed after ablation and following UV- or S-cone regeneration respectively (Hagerman et al., 2016). The OMR was reduced immediately after ablation of either UV- or S-cones. Surprisingly, while the OMR recovery took 4 days for the UV-cone ablation, the OMR recovered in just 1 day following S-cone ablation, before new S-cones were produced. These differences in behavioral recovery suggest that there may be a capacity for plasticity amongst the remaining cells, used to compensate for the ablated cells during the recovery phase. It is possible that this shortterm plasticity relies on activity from other cone subtypes and/or on synaptic remodeling. Evidence for such synaptic remodeling has been reported in a parallel study (Yoshimatsu et al., 2016). In this study, a subtype of horizontal cell (H3) that normally connects preferentially to UV- and S-cones, was able to reconnect to UV-cones after UV-cone specific ablation and regeneration. Nevertheless, if UV-cone regeneration was delayed, the H3 made additional contacts with S-cones and even M- and L-cones, suggesting functional compensation at the level of rewiring. Another recent study examined rewiring after selective loss of a subpopulation of BC using the NTR-MTZ system in 5 dpf larvae (D'Orazi et al., 2016). Thirteen days after the ablation of these BC, the majority of regenerated BC were morphologically normal but the rewiring did not fully recapitulate development, with a relative loss of selectivity for specific cone subtypes. Additionally, BC axons contained significantly more synapses.

As a whole, work in this field proves that retinal regeneration in zebrafish is a robust process, but also suggests that some of the developmental cues required to refine synapse number or proportion of photoreceptor subtypes innervated may not be present during regeneration. In the future, it will be important to further understand what cues are present during development that enable photoreceptors to wire into different retinal circuits. It will be especially important to understand how specific photoreceptor subtypes recognize the different bipolar- and horizontal-cell subtypes, and the factors required for the formation of these synapses. It will also be important to recognize and examine the developmental and environmental differences between larval and adult zebrafish retinas. This knowledge will provide a comprehensive understanding of the differences between development and regeneration, between wiring, rewiring and remodeling, and likely uncover manipulations that could be used to modify or refine rewiring in the context of treatments for RD.

### THE ZEBRAFISH TOOLKIT FOR THE STUDY OF RETINAL DEVELOPMENT AND REGENERATION

There are numerous factors that have established zebrafish as a valuable model organism for the study of human disease including rapid development, large clutch sizes, ease of maintenance, genetic conservation, accessibility to genetic manipulations, and optical transparency of embryos. In addition to these advantages, we have described several examples of how zebrafish has been a useful model to investigate retinal development and regeneration. To aid in these studies, multiple tools have been developed that have direct application for the

study of retina in zebrafish. We have summarized some of these tools here as a convenient reference.

### Imaging Retinal Cells and Their Connectivity

Currently there are established transgenic lines that make it straightforward to visualize each cell subtype within the retina, as well as the lamination and precise wiring of these cells (**Table 1**). For example, taking advantage of the specificity of opsin expression in the different photoreceptor subtypes, promoters from each opsin have been utilized to create transgenic lines that label rods (rods: rho) (Fadool, 2003) and each of the four cone subtypes (UV-cones: opn1sw1; S-Cones: opn1sw2; M-Cones: opn1mw2; L-Cones: opn1lw1) (Takechi et al., 2003, 2008; Tsujimura et al., 2007, 2010) (**Figures 3A,B**). Additionally, multiple lines exist to label horizontal and BC. Within the retina, horizontal cells can be specifically labeled exploiting the promoter for connexin 55.5 (Weber et al., 2014; Klaassen et al., 2016), or using an enhancer-trap line that labels a combination of horizontal and amacrine cells (Torvund et al., 2017). BC represent a more diverse cell class. On-BC can be labeled using the promoter for grm6b, a metabotropic glutamate receptor that is key for the detection of glutamate release by photoreceptors (Glasauer et al., 2016) (**Figure 3C**), and different subtypes of BC have been labeled with enhancer-trap lines (D'Orazi et al., 2016), or using promoters for transcription factors (Vitorino et al., 2009) or for other bipolar-specific proteins (Schroeter et al., 2006). The promoter for gfap (glial fibrillar acidic protein) can be used to label Müller glia (Raymond et al., 2006; Bernardos et al., 2007). Additionally, lines that use the promoter for mpeg1 label all macrophages (Ellett et al., 2011), allow visualization of the retinal microglia and macrophages in both the normal and regenerating retina (Mitchell et al., 2018).

Most of these lines allow high-resolution imaging of not only the lamination but also the synaptic terminals of these distinct cells types (Noel and Allison, 2018) (**Figure 3D**). For example, in some of these transgenic lines, the connections of photoreceptors have been used to accurately track wiring during development (Yoshimatsu et al., 2016) or rewiring after regeneration (D'Orazi et al., 2016). They can also be adapted for live imaging, allowing connections to be dynamically tracked during circuit formation or during the integration of regenerated neurons into functional retinal circuits (Ariga et al., 2010; Duval et al., 2013). Using the same promoters, several transgenic lines have been developed to express the NTR gene (nfsb) and ablate specific subsets of retinal cells (see above, **Table 1**). This approach has enabled the study of regeneration and rewiring after selective ablation (Zhao et al., 2009; D'Orazi et al., 2016; Yoshimatsu et al., 2016; McGinn et al., 2018; Noel and Allison, 2018). Lastly, it is straightforward to perform sparse labeling to capture the fine detail of individual retinal neurons and their contacts during development or regeneration, by using transient expression in larvae (with the same promoters leveraged for transgenic lines) (Klaassen et al., 2016; Yoshimatsu et al., 2016). Alternatively, inorganic fluorescent dyes like DiI can be used in larvae or in adults (**Figures 3E,F**) (Connaughton et al., 2004; Li et al., 2009; Li et al., 2012).

While transgenic lines represent powerful tools to visualize cells and processes within the retina in situ, antibodies against specific markers have been identified that can also be exploited to label subsets of retinal cells (**Table 2**). In addition to antibodies, cones can also easily be labeled using fluorescently-tagged peanut agglutinin (or PNA), a lectin protein that binds to the cone sheath (Hageman and Johnson, 1986; Shi et al., 2017). In combination with the transgenic lines outlined above, these labels can be used to visualize multiple types of cells simultaneously within the retina (**Figure 3C**). In addition to simply marking specific retinal cell types, antibodies against synaptic markers are of particular usefulness to characterize retinal wiring. In photoreceptors most presynaptic markers label components of the photoreceptor ribbon synapse, like Ribeye or Syntaxin, or adjoining structures like the voltage-gated calcium channels or the synaptic vesicles (Huang et al., 2012; Lv et al., 2012; Daniele et al., 2016). Postsynaptically, photoreceptor synapse markers include: components of the postsynaptic density itself (e.g., MAGUK) and glutamate receptors (grm6 for ON-bipolars, gria4 for OFF-bipolars, gria2 for horizontal cells) (Yazulla and Studholme, 2001) (**Figure 3D**). These synaptic markers are extremely important for understanding the correct development of synapses or the correct rewiring of photoreceptors after regeneration. For example, alterations in the synapses between cones and Off-BC caused by mutations in pappaa, a protein recently identified in a behavioral screen, were identified by labeling the photoreceptor synaptic vesicles, but could not be seen by labeling specific cell-types, as retinal lamination was not altered (Miller et al., 2018).

#### Functional and Behavioral Methods to Assess the Zebrafish Retina

One of the most exploited assays for visual function in humans and many animal models are electroretinograms (ERG). ERG measure bulk electrical signals produced by the whole retina in response to light stimulation and has been adapted to both zebrafish larvae (Nelson and Singla, 2009; Chrispell et al., 2015) and adults (Hughes et al., 1998). Through analysis of the different ERG waves the overall activity of photoreceptors and BC can be evaluated. Using well-designed stimuli or pharmacological agents, other properties like the kinetics of photoreceptor adaptation can also be measured (Korenbrot et al., 2013). Additionally, ERG signals have a spectral signature based on the signals generated by specific subsets of photoreceptor and their downstream partners. These signatures can be utilized to isolate the contributions of each element in different conditions (Nelson and Singla, 2009). ERG measurements have been used in the adult to demonstrate that the retina can recover function after damage and subsequent regeneration (McGinn et al., 2018). In the future ERG could be used both in larvae and adults after NTR-MTZ ablation of specific cells, to assess the functional recovery of retinal processing during regeneration.

While the ERG can provide a powerful readout of retinal activity, functional imaging using genetically encoded calcium

TABLE 1 | Toolkit for the study of retinal development and regeneration: transgenic lines.


(Continued)

#### TABLE 1 | Continued


sensors has also been developed to measure response in visual centers, especially in the tectum (Förster et al., 2017). Currently these approaches remain challenging in the retina due to the RPE which creates an optical barrier, making it difficult to image directly through the eye. It is possible to use these indicators to image through the lens in adults (Duval et al., 2013). Also, in larvae some transparency can be achieved using PTU to inhibit melanophore production, but this treatment may alter visual function (Antinucci and Hindges, 2016). Better imaging has been achieved with mutant lines that genetically remove the different classes of pigmented cells (White et al., 2008; Antinucci and Hindges, 2016). In the future, additional functional imaging using calcium sensory in photoreceptors along with newly developed neurotransmitter sensors, like iGluSnfR (Marvin et al., 2013), will be an important in vivo approach to assess pre- and post-synaptic function with in developing and regenerating retinal circuits. (Zhang et al., 2016; MacDonald et al., 2017).

In addition to electrically or optically recording the activity of cells within the retina, there are many well characterized visual behaviors that can be used to evaluate retinal function. The optokinetic response is an extremely robust behavioral assay, where a visual stimulus of moving stripes is tracked by eye movements. This behavior is already present and reliable by 5 dpf and requires minimal equipment to setup (Brockerhoff et al., 1995; Neuhauss et al., 1999; Neuhauss, 2003). In the related optomotor response, tracking of moving stripes is followed by swimming in the same direction as the stimulus (Neuhauss et al., 1999; Neuhauss, 2003). This assay has been used to characterize the overall recovery of vision after photoreceptor ablation (Hagerman et al., 2016). At around 5 dpf, larvae also start hunting for small prey using visual cues, another visual behavior that can be quantified (Borla et al., 2002; Gahtan et al., 2005; McElligott and O'Malley, 2005). Larvae also innately exhibit phototaxis and photoavoidance (Brockerhoff et al., 1995; Orger and Baier, 2005; Burgess et al., 2010), and an escape response in response to sudden decreases in illumination (Burgess and Granato, 2007).

Adult zebrafish also exhibit an escape response to threatening objects, characterized by rapid turning and swimming away from the threat. The escape response can be elicited by placing fish in a clear tank with a central pole that serves for hiding, and an external rotating drum with a single black stripe to act as a threatening stimulus. Use of this assay allowed to measure behavioral rod and cone thresholds and the time course of photoreceptor adaptation and as part of the screening in a forward-genetic screen for visual mutations (Li and Dowling, 1997; Li and Dowling, 2000). Zebrafish and other teleosts determine their body position using a combination of their sense of balance and the source of illumination, which in their natural environment tends to come from above. Thus, they tend to tilt their bodies such that their backs are turned against the source of illumination (dorsal light response or DLR). Tilt can be induced by uneven-illumination between the two eyes (e.g., side illumination) (Silver, 1974; Neuhauss, 2003) or by unilateral

ocular damage, where recovery of a normal tilt is a sign of functional recovery (Mensinger and Powers, 1999; Lindsey and Powers, 2007; Mensinger and Powers, 2007; Sherpa et al., 2014). In zebrafish, information from the vestibular system is capable of overriding visual input, so that the DLR is only apparent when the vestibular system has been damaged (Nicolson et al., 1998) or when fish are placed head down in a tightly fitting tube (Neuhauss, 2003).

In captivity, zebrafish are conditioned to move toward the front of the tank and wait for food whenever a person approaches. This conditioned learning can be exploited to test for visual function (place preference test) (Sherpa et al., 2014). A very similar test has been recently used in cichlids to demonstrate their ability to truly discriminate colors (Escobar-Camacho et al., 2017) (**Table 3**).

Interestingly, some of these behaviors seem to rely on only into small subsets of retinal circuits. For example, the OKR seems to be mainly driven by M- and L-cone signals (Orger and Baier, 2005), while the tracking of small dots, at least in goldfish, depends mainly on M-cone signals (Gehres and Neumeyer, 2007). Similarly, photoavoidance is robustly driven by UV-light, and presumably UV-cone signals (Guggiana-Nilo and Engert, 2016). From previous work that has shown fast functional OKR recovery after S-cone ablation even before regeneration and rewiring occur (Hagerman et al., 2016), it is clear that not enough is known about the function, recovery and plasticity of circuits in the retina. Together, behavioral assays along with NTR-mediated ablation of a given cone subtype, could be used to further expand our understanding of these circuits, and the functional consequences of regeneration and rewiring of the different retinal circuits.

### Forward and Reverse Genetic Approaches

Zebrafish forward genetic screens are extremely powerful, and have been successfully used to uncover novel genes that are involved in photoreceptor function and RD (Brockerhoff and Fadool, 2011). It is likely that any additional screening that relies on alterations in visual behaviors will continue to uncover new genes that affect photoreceptor function or cause RD. As more causes of RD continue to be unveiled, we now understand that there are limited number of converging pathways that eventually lead to photoreceptor degeneration, including: classic apoptosis, oxidation, activation of proteolytic pathways, misbalance in intracellular levels of cGMP and calcium, and epigenetic regulation (Trifunovic et al., 2012). With this knowledge, reversegenetic approaches using CRISPR-mediated gene editing to target previously undescribed components within these pathways are likely to be extremely useful in the future. Recent studies in

#### TABLE 2 | Toolkit for the study of retinal development and regeneration: antibodies and fluorescent labeling.


ganglion cells


TABLE 3 | Toolkit for the study of retinal development and regeneration: electroretinograms and visually guided behaviors.

zebrafish have already demonstrated that zebrafish can be used as a platform to rapidly perform genetic screens using CRISPR (Varshney et al., 2015, 2016; Shah et al., 2016). In addition to genetic screens, similar to what has been done in fin and hair-cell regeneration studies (Mathew et al., 2007; Namdaran et al., 2012), pharmacology-based screening could be used to isolate novel compounds with the ability to promote or prevent photoreceptor regeneration. Similar screens could also be accomplished using behavioral assays, evaluating the recovery of visual function after regeneration. Some of the success of such screens in hair cells of the lateral line stems from the fast regeneration times (∼2 days) and the small number of cell types that have to be regenerated. In paradigms of regeneration after extensive damage (surgical of pharmacological) in adults, the long regeneration times (8–14 weeks) and the diversity of cell types that need to be regenerated might present insurmountable hurdles for screens. Nevertheless, photoreceptor regeneration occurs within 4 weeks in adults and in ∼10 days in larvae after cell-specific ablations, opening up the possibility to carry out such screens.

While the majority of regeneration studies in zebrafish have focused on regenerating damaged cells, for functional recovery after regeneration, it is imperative that new cells integrate appropriately into their specific retinal circuits. Our current knowledge on how retinal circuits in the outer retina form during development and after regenerations is limited. Studies into these processes suggest that rewiring after regeneration is not a complete recapitulation of development. First of all, during development retinal cells are derived from retinal progenitors and retinal circuits assemble properly even in the absence of Müller glia (Williams et al., 2010). In contrast, during regeneration, Müller glia are the principal source of new retinal cells. Second, not all the transcription pathways that are active during development are reactivated during regeneration (Veldman et al., 2007; Sherpa et al., 2014). Third, there seems to be more plasticity and a greater capacity for compensation. It appears that during rewiring, at least in larvae, maintaining inputs and outputs is more important than the absolute selectivity of connections (D'Orazi et al., 2016; Yoshimatsu et al., 2016). In adult zebrafish, after pharmacological ablation of bipolar, amacrine and ganglion cells (but survival of photoreceptors) and their subsequent regeneration, various subtypes of BCs seem to be able to recapitulate the diversity of connectivity that is found in uninjured eyes, but as a population, selectivity for photoreceptors seems to be restored (McGinn et al., 2018). Further investigation is required to explain the disagreement between these studies since there are many differences including ablation technique (NTR-MTZ vs. ouabain), age of ablation (larvae vs. adults), time between ablation and assessment of connectivity (1–2 weeks vs. 8 weeks) and subtypes of cells studied. Yet, this raises interesting questions: is the capacity for compensation only present in larvae and lost in the adult? or, is compensation only present in the initial phase after regeneration and normal selectivity of connections reestablished over time?

It is interesting to note that selectivity is not lost during rewiring, it is just more permissive, and it is very likely that the same molecules that allow recognition between retinal cell types are used in both development and regeneration. Only a handful of cell-adhesion and synaptic molecules are known to be necessary for the formation of synaptic contacts between photoreceptors and downstream retinal cells (Zhang et al., 2017; Miller et al., 2018). Some of these molecules are key across all photoreceptors, while others are specific to rods (Cao et al., 2015; Wang et al., 2017) or to cones (Sarria et al., 2018; Ueno et al., 2018). To date, the molecular mechanisms involved in the recognition between specific photoreceptor subtypes and their synaptic partners (horizontal and bipolar cells) are not known. Any of the genetic screening tools mentioned above could be combined with the NTR-MTZ transgenic lines (**Table 1**), to target specific cell subtypes and elucidate the mechanisms that enable rewiring. Research in this front could have a very

significant impact in phenotyping vision loss in RD, and to develop manipulations that could ultimately enable rewiring of transplanted or regenerated photoreceptors into proper retinal circuits.

#### Gene-Expression Profiling (RNAseq)

During the last decade, advances in the capacity of highthroughput sequencing has allowed to profile the transcriptomes of whole tissues or dissociated single cells. Gene-expression profiling of retinal cells in mice has given great insight in the classification of retinal cells into different (and even novel) subtypes (Macosko et al., 2015), especially for BC, where clear differences in molecules involved in cell-recognition and synapse formation were detected (Shekhar et al., 2016). These techniques can be applied in zebrafish, especially using transgenic lines, as has been recently reported for rods (Sun et al., 2018). RNAseq of zebrafish retinal cells could help unveil the genes that are required for synapse formation between BC and photoreceptors, genes that could be essential to promote rewiring in RD therapies. RNAseq could also be exploited to study the changes in gene expression that occur during degeneration and regeneration. This could be accomplished by profiling single cells in the most relevant time points after photoreceptor death. With a focus on Müller glia, further insights could be gained into the gene networks that allow pluripotent and functional recovery in zebrafish. This knowledge will also be extremely valuable for the treatment of RD and understanding how to initiate regeneration after RD.

### CONCLUSION AND OUTLOOK

Our field is developing a deep understanding on many aspects of RD, including risk factors, underlying genetic causes, molecular pathways that lead to photoreceptor death, and the manipulations that could slow down the progression of the disease. During the last decade, we have made significant progress into revolutionary therapies that could, 1 day, cure blindness.

Despite all of the research on RD and regeneration, there are still gaps in our current knowledge that limit our capacity to understand certain aspects of RD and hinder our ability to develop therapies. The zebrafish is an advantageous model to fill in these gaps, especially at a mechanistic level. As a relevant example, we have discussed how zebrafish has been

#### REFERENCES


used to delineate molecular pathways within Müller glia that allow regeneration of retinal cells even in the adult zebrafish. This knowledge has been directly applied into the mouse retina and successfully used to generate new and functional bipolar and amacrine cells (Jorstad et al., 2017). Although studies have been able to stimulate retinal regeneration in mice using manipulations derived from the study of zebrafish retinal regeneration, but we do not yet fully understand the pathways required to regenerate each retinal cell type, or how these pathways are regulated to regenerate specific subpopulations. Further studies are required to understand the mechanisms that allow zebrafish Müller glia to not only produce any retinal cell, but also to specifically replace the lost population without overtly producing proliferation of undamaged cell types (D'Orazi et al., 2016; Yoshimatsu et al., 2016; McGinn et al., 2018).

In addition to gaps in knowledge on the role of Müller glia and regeneration, we also have an incomplete grasp on the processes that are involved in recognition between photoreceptors and their postsynaptic targets, with only a handful of molecules known to be involved in the correct formation of synapses. This leaves us with little leverage on manipulations that could promote integration of new photoreceptors into the surviving retinal circuits. Solving these issues and finding viable therapeutic options for RD will certainly require diverse approaches. Research in zebrafish is uniquely poised to make additional key contributions into RD, especially on unveiling the molecular mechanisms involved in photoreceptor regeneration and the processes that guide wiring during development and rewiring after regeneration of photoreceptors into retinal circuits.

#### AUTHOR CONTRIBUTIONS

JA conceived and carried out the literature review research, designed the figures and diagrams, acquired the images, and wrote the article. KK conceived and carried out the literature review research and wrote the article.

### FUNDING

This work was supported by NIH/NIDCD intramural research funds 1ZIADC000085-01 (KK), and NIH/NEI intramural research funds ZIAEY000488-10 (JA).


review and meta-analysis. Lancet Glob. Health 5, e888–e897. doi: 10.1016/ S2214-109X(17)30293-0





differences in photoreceptor loss. Exp. Eye Res. 97, 105–116. doi: 10.1016/j.exer. 2012.02.004



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

Copyright © 2018 Angueyra and Kindt. 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.

# Genetic Models of Leukemia in Zebrafish

Jeremy T. Baeten and Jill L. O. de Jong\*

Department of Pediatrics, University of Chicago, Chicago, IL, United States

The zebrafish animal model is gaining increasing popularity as a tool for studying human disease. Over the past 15 years, many models of leukemia and other hematological malignancies have been developed in the zebrafish. These confer some significant advantages over similar models in other animals and systems, representing a powerful resource for investigation of the molecular basis of human leukemia. This review discusses the various zebrafish models of lymphoid and myeloid leukemia available, the major discoveries that have been made possible by them, and opportunities for future exploration.

Keywords: zebrafish, leukemia, animal models, ALL, AML, MDS, MPN

#### Edited by:

Rebecca Ann Wingert, University of Notre Dame, United States

#### Reviewed by:

Lasse Dahl Ejby Jensen, Linköping University, Sweden Eirini Trompouki, Max-Planck-Institut für Immunbiologie und Epigenetik, Germany

> \*Correspondence: Jill L. O. de Jong jdejong@peds.bsd.uchicago.edu

#### Specialty section:

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

Received: 30 April 2018 Accepted: 23 August 2018 Published: 20 September 2018

#### Citation:

Baeten JT and de Jong JLO (2018) Genetic Models of Leukemia in Zebrafish. Front. Cell Dev. Biol. 6:115. doi: 10.3389/fcell.2018.00115

#### Leukemia

INTRODUCTION

Leukemia is a broad designation encompassing hematological malignancies that produce the expansion of blood cells, typically starting in the bone marrow. In 2015, there were over 2.3 million patients suffering from leukemia, resulting in over 350,000 deaths worldwide (GBD 2015 Mortality and causes of death collaborators, 2016). In the United States, an estimated 62,130 new leukemia cases were diagnosed and 24,500 deaths caused by leukemia in 2017, with a 5 years survival rate of ∼63% (NCI SEER Cancer Stat Facts: Leukemia). Although the majority of leukemias affect adults, leukemia is also the most common cancer diagnosis in children.

Leukemias are categorized by two major criteria into four groups. The first criterion relates to the cell of origin: leukemias of lymphoid origin are classified "lymphocytic or lymphoblastic" and those of myeloid origin called "myelogenous or myeloid." The second criterion deems rapidly growing leukemias as "acute" and those with more indolent growth as "chronic." The causes of these different malignancies are varied. Some are directly linked to a chromosomal abnormality, such as the Philadelphia chromosome in chronic myelogenous leukemia (Bartram et al., 1983) or the increased incidence of leukemia in patients with trisomy 21 (down syndrome) (Evans and Steward, 1972). However, the etiology of most leukemias is less straightforward. Some leukemias involve mutations and/or translocations of multiple genes associated with growth, differentiation and survival of blood cells. Others have a normal karyotype and no known genetic mutations, highlighting the need for further studies in animal models to uncover these unknown drivers of hematopoietic malignancy.

#### Zebrafish as a Model Organism

Danio rerio, commonly known as the zebrafish, is a small tropical fish popular in pet stores and aquariums. Since the 1970's when George Streisinger first began using the zebrafish as a model organism (Walker and Streisinger, 1983), more and more labs have begun utilizing this powerful tool for studying development and disease due to numerous advantages over other model systems (**Figure 1**).

Zebrafish fertilization and development occurs externally in optically clear embryos that are easily observed and manipulated. Development is much faster than mammals, with most major organs forming by 2–3 days post-fertilization (dpf). Animals reach sexual maturity by 2–3 months of age (Kimmel et al., 1995) and a single breeding pair produces several hundred embryos weekly. This fecundity coupled with their rapid development makes the zebrafish an excellent model for large-scale screening. Forward and reverse genetic screens as well as toxicity and drug screens in zebrafish have been performed around the world over the past three decades (Ransom et al., 1996; Weinberg et al., 1996; Sood et al., 2006; North et al., 2007; Ridges et al., 2012), including significant work more recently to evaluate therapeutics in zebrafish leukemia models (Mizgirev and Revskoy, 2010; Deveau et al., 2017). Although teleosts (like the zebrafish) and mammals diverged from a common ancestor approximately 340 million years ago, they still share a remarkable amount of their genomes, with a zebrafish ortholog identified for 82% of known disease-causing genes in humans (Howe et al., 2013). Many of these zebrafish genes have already been shown to recapitulate human disease when affected in zebrafish, including several connected to hematopoiesis (Brownlie et al., 1998; Wang et al., 1998) and cancer, as we will discuss in this review.

Several systems have been developed within the zebrafish model to create transgenic and knockout animals. Because the zebrafish embryos are externally developed, it is possible to microinject directly into the single-cell for the first 15–30 min following fertilization. Although, the first transgenic zebrafish were created through injection of naked, linearized DNA (Stuart et al., 1988), more efficient systems of genomic incorporation are now available. The tol2 transposon system creates randomly inserted transgenes that heavily favor single copy insertions (Urasaki et al., 2006) and the I-SceI meganuclease system inserts one or more copies into double-stranded breaks in the genome (Grabher et al., 2004; Ogino et al., 2006). In addition to the ability to integrate transgenes, the advent of CRISPR/Cas9 technology has made it possible to directly edit the zebrafish genome; from creating knockouts to mimicking human mutations to introducing specific SNPs. The use of CRISPR/Cas9 in zebrafish was first described by Keith Joung's lab in 2013 (Hwang et al., 2013), and has since spread throughout the field to become a common tool in many labs' arsenal, just as it has throughout the biomedical community at large (Hruscha et al., 2013; Ablain et al., 2015).

Over the years, the zebrafish community has amassed a large number of inbred, transgenic, knockout, or other specialized lines that have been characterized and maintained for various applications. Important to leukemia models are several lines that allow for transplantation of tumors without the need for pre-transplant immune ablation. Generated by parthenogenesis, the clonal golden lines (CG1 and CG2) allow for syngeneic transplantation within a genetically identical line, similar to transplantation experiments using inbred mice (Mizgireuv and Revskoy, 2006; Smith et al., 2010). The rag2 (E450fs) mutant line has reduced numbers of functional T- and B-cells, and thus is unable to mount a significant immune response against transplanted cells (Tang et al., 2014). The c-mybI181N hypomorphic mutant is another immunocompromised line that has shown promise in xenograft experiments (Hess and Boehm, 2016). These lines allow for immunologically unmatched transplantation from other zebrafish lines as well as xenografts.

A common challenge for many model systems is the ability to visualize and trace the fate of a cancer cell within an animal over time. These issues are often circumvented by euthanizing, sectioning, and staining multiple animals at different timepoints, however, this increases the number of animals required, increases time commitment, and may blur inter-individual variability. In the zebrafish, fluorescently tagged proteins or cells can be clearly imaged from embryo to adulthood in live animals by confocal or lightsheet microscopy (Kaufmann et al., 2012) and at even

greater resolutions in the pigment-less Casper line (White et al., 2008). An excellent example of this utility was described by Kaufman et al. (2016), when they used a crestin-EGFP line to show melanoma initiation and progression from a single cell. Also, with the macroscope developed by the Langeneau lab, high-throughput imaging of adult fish is possible for transgenic lines or screening for tumor engraftment in transplantation models (Blackburn et al., 2011).

#### Hematopoiesis: Zebrafish and Human

Many of the transcription factors and major signaling pathways controlling hematopoietic differentiation are mutated or dysregulated in the transformation and progression of leukemia. Therefore, in addition to the general advantages described above, the zebrafish is an appealing model for studying leukemia because of the close parallels to mammalian hematopoiesis (de Jong and Zon, 2005). Though the locations of hematopoiesis are not perfectly shared between species, the ontogeny of the different hematopoietic cells from progenitors to maturity, as well as the genes and pathways driving differentiation are well conserved (Paik and Zon, 2010). There are two distinct waves of hematopoiesis in all vertebrates; a transient primitive wave supplying necessary macrophages and erythrocytes for early embryonic development, followed by the definitive wave that gives rise to the full complement of blood cells throughout an animal's lifetime. In mammalian development, the primitive hematopoietic stem cells (HSCs) appear within the blood islands in the embryonic yolk sac (Palis and Yoder, 2001). In zebrafish, these limited HSCs instead arise from the intermediate cell mass (ICM) within the ventral mesoderm, and, similarly, produce erythrocytes and other myeloid cells (Detrich et al., 1995). Expression of the transcription factors scl, gata2, lmo2, tif1γ, and fli1 promotes the primitive HSC lineage (Liao et al., 1998; Thompson et al., 1998; Ransom et al., 2004), while gata1 and spi1 (also known as pu.1) drive their differentiation into the erythroid and myeloid lineages, respectively (Detrich et al., 1995; Lieschke et al., 2002).

The mammalian definitive wave of hematopoiesis begins with true multipotent HSCs emerging from the ventral wall of the dorsal aorta in the aorta-gonad-mesenephros (AGM) region that then migrate to the fetal liver to proliferate and differentiate, and ultimately migrate to seed the bone marrow (Cumano and Godin, 2007). This process is mirrored in the zebrafish with the definitive HSCs also arising from the ventral wall of the dorsal aorta, and migrating to the caudal hematopoietic tissue (CHT) before seeding the kidney marrow, which is the zebrafish adult hematopoietic tissue (Burns et al., 2002; Jin et al., 2007). Definitive HSCs are true multipotent hematopoietic progenitors and are marked by their expression of the transcription factors runx1, c-myb, lmo-2, and scl (Thompson et al., 1998; Burns et al., 2002). Similar to primitive hematopoiesis, gata1 and tif1γ drive erythropoiesis (Detrich et al., 1995; Ransom et al., 2004) and spi1 and c/ebp1 drive myelopoiesis (Lyons et al., 2001; Lieschke et al., 2002). Unlike the primitive lineages, definitive HSCs also produce lymphoid cells through expression of rag1, rag2, ikaros, lck, and gata3 (Willett et al., 1997, 2001; Langenau et al., 2004). There are several functional and structural differences in the hematopoietic system of zebrafish compared to mammals, namely the location of the marrow, the lack of lymph nodes, and the rapid development and early reliance on the innate immune system (Novoa and Figueras, 2012; Renshaw and Trede, 2012). However, ultimately the blood cells of the zebrafish and human are molecularly very similar and thus have common genetic drivers of leukemia.

The conservation of the genes and pathways regulating hematopoiesis between humans and zebrafish, combined with the significant technical advantages provided by this model animal, make the zebrafish an ideal system for investigating hematological malignancies. In this review, we will examine the many leukemia models that have been developed within the zebrafish, and discuss the major findings made possible by each model that have advanced our understanding of human leukemia.

#### ZEBRAFISH LEUKEMIA MODELS: LYMPHOID ORIGIN

The first leukemia model in zebrafish was developed over 15 years ago using the lymphocyte-specific rag2 promoter driving the murine c-Myc oncogene to produce T-cell acute lymphoblastic leukemia (T-ALL) (Langenau et al., 2003). The success of that first step has spawned a variety of other models tied to different types of leukemia. Over time, these models have been altered and improved to fit the particular investigations of each project, and there are now multiple similar models available, each with their own strengths and weaknesses (**Table 1**). This section discusses models of lymphoid origin, and the major discoveries made possible by them.

### T-cell Acute Lymphoblastic Leukemia (T-ALL)

The majority of lymphoid leukemia models in zebrafish replicate T-ALL, partially due to the success of the rag2 promoter in driving that particular malignancy. Although rag2 is expressed in both T- and B-cell precursors in zebrafish (Langenau et al., 2004), only T-cell leukemias were initially identified from models utilizing this promoter. Interestingly, the Langenau lab has recently published a brief communication describing a subset of B-cell derived and bi-phenotypic leukemias produced from a rag2 promoter (Garcia et al., 2018), suggesting some of the research done on these T-ALL models may have unknown contributions from B-ALL as well.

The oncogene c-Myc is associated with many cancers and is one of the most frequently affected gene pathways in lymphoid leukemia (La Starza et al., 2014). The first T-ALL model was developed by Langenau et al. by expressing the murine c-Myc oncogene under the zebrafish rag2 promoter, with an EGFP tag for easy monitoring by fluorescent microscopy (Langenau et al., 2003). Tumors were generated in microinjected mosaic F<sup>0</sup> fish at similar rates to EGFP expression in control animals injected with rag2: EGFP, suggesting complete penetrance of tumor induction upon successful integration of the c-Myc transgene. These tumors grew rapidly, with a mean latency of 52 days post-fertilization (dpf), extensively infiltrating the entirety of the TABLE 1 | Zebrafish leukemia models of lymphoid origin.

fcell-06-00115 September 19, 2018 Time: 10:25 # 4


fish. Analysis of the expression profiles and T-cell receptor (TCR) rearrangements confirmed that the tumor cells derived from clonal expansion of transformed T lymphocyte precursors and originated in the thymus. Tumor cells could be transplanted into irradiated recipients and quickly grew new tumors that homed to the thymus before spreading throughout the animal. Overall, the tumors progressed, similarly, to human T-ALL, at an accelerated pace. However, most F1 progeny developed advanced disease well before reaching sexual maturity (mean latency 32 dpf), necessitating sperm collection and in vitro fertilization (IVF) to continue the stable transgenic line. Subsequent characterization of this model showed that the tumors express tal1/scl and lmo2, genes associated with a molecular subgroup of Myc-induced T-ALL in humans (Langenau et al., 2005).

To circumvent the necessity of IVF, Langenau et al. sought to create an inducible version of their model. They achieved this by inserting a loxP-DsRed2-loxP sequence cassette between the rag2 promoter and EGFP-mMyc oncogene, creating the rag2:loxP-dsRed2-loxP-EGFP-mMyc line (rag2: LDL-Emyc) (Langenau et al., 2005). This allowed for default red fluorescent expression with a switch to EGFP-mMyc expression in the presence of Cre recombinase. The disease in these animals was morphologically similar to that in the rag2:EGFP-mMyc model, but with significantly decreased incidence (6.5%) and delayed latency (mean 151 dpf). This was presumed to be due to incomplete recombination of the transgene, as evidenced by the persistence of red fluorescent expression within the tumor. To combat this, they developed a heat shock-inducible Cre line, hsp70: Cre (Feng et al., 2007). When combined with their Cre-inducible rag2: LDL-Emyc line and subjected embryos to heat shock at 3 dpf, the penetrance (81%) and latency (120 dpf) were closer to those of the original rag2: EGFP-mMyc model. This improved model allowed them to explore the molecular events governing the progression of the disease from the localized T-lymphoblastic lymphoma (T-LBL) to disseminated T-ALL. All of the Myc-induced models of T-ALL in zebrafish begin as T-LBL with thymic hyperplasia and localized outgrowth before advancing to T-ALL and expanding

into the circulation and other tissues. The investigation into this transition led them to combine the rag2:LDL-Emyc; hsp70:Cre model with a line overexpressing the survival gene bcl2 (Feng et al., 2010). This combination accelerated T-LBL induction by suppressing Myc-induced apoptosis. However, it also promoted homotypic cell adhesion through s1p1 and icam1 that prevented intravasation into the vascular space and restricted the tumor to the thymus. The tumor cells then proliferated until they exhausted their nutrient supply and underwent autophagy. Because AKT-signaling is known to promote T-cell migration and to suppress autophagy (Sotsios and Ward, 2000; Lum et al., 2005), they hypothesized that addition of constitutively active AKT could force progression to T-ALL. Indeed, when their Myc;Cre;bcl-2 embryos were injected with a myristolated-akt2 transgene, the resulting tumors rapidly advanced to T-ALL (Feng et al., 2010).

The importance of AKT signaling in zebrafish T-ALL progression is not surprising, given the PTEN-PI3K-AKT pathway is frequently disrupted in human T-ALL (Palomero et al., 2008; Gutierrez et al., 2009). Gutierrez et al. further investigated this connection with the aid of another Myc-induced model, the tamoxifen inducible rag2: Myc-ER line (Gutierrez et al., 2011). This model allows for conditional expression of the c-Myc oncogene only in the presence of 4-hydroxytamoxifen (4-OHT). When continually treated with 4-OHT these fish develop T-ALL, but upon cessation of treatment and loss of c-Myc expression, the tumor cells undergo apoptosis and the tumor rapidly regresses. However, when AKT signaling was increased through either loss-of-function mutations in pten or constitutively active akt2, the tumors lost their dependence on Myc expression and were able to continue progression after removal of 4-OHT treatment. Further investigation into the relationship between Myc and the AKT pathway revealed that Myc drove the expression of the proapoptotic protein bim, while the constitutively active myr-akt2 blocked that induction (Reynolds et al., 2014). Additionally, loss-of-function bim mutations allowed for increased persistence of T-ALL after cessation of 4-OHT treatment and Myc expression. Overall, these results suggest AKT-signaling enhances Myc-induced T-ALL progression via promotion of T-cell migration, suppression of autophagy, and inhibition of apoptosis.

Due to difficulty maintaining stable transgenic lines expressing c-Myc, an alternative approach was developed involving coinjection of the rag2-EGFP and rag2-mMyc transgenes into single-cell embryos (Langenau et al., 2008; Smith et al., 2010). In this model, the two transgenes randomly integrated into the genome to be co-expressed such that GFP expression was observed only in tumors and tumor induction only with GFP+ thymocytes. The resulting tumors followed the same pathology as the stable Myc-induced models. Smith et al. (2010) used this method to create tumors in clonal CG1 fish, demonstrating that they could be transplanted into syngeneic recipient CG1 fish without irradiation. This also allowed them to determine the frequency of leukemia stem cells (LSCs) present in these tumors through limit dilution analysis of the transplanted tumors. Each successful engraftment requires at least one LSC, and by transplanting different doses of cells, they were able to determine that 0.1–1.4% of the primary T-ALL tumor cells were LSCs. Transplantation of T-ALLs generated using this co-injection model was further investigated by Blackburn et al. (2014) who demonstrated that serial transplantation of T-ALL tumors led to spontaneous clonal evolution of monoclonal tumor subclones. As tumors were passaged from primary to secondary to tertiary recipients, some subclones evolved increased LSC frequency, growth, and/or resistance to therapy. Subclones with increased LSC frequency also displayed increased AKT phosphorylation, and treatment with an AKT inhibitor dramatically reduced their engraftment after transplant. Co-expression of myr-akt2 with Mycsignificantly increased proliferation of tumor cells, decreased latency after transplantation, and increased LSC frequency sixfold, and these effects are at least partially due to AKT's induction of mtorc1 expression. Additionally, the subclones that had evolved glucocorticoid resistance were resensitized to dexamethasone treatment by AKT inhibitors. Altogether, these results provided further evidence of the connection between Myc and AKT in T-ALL.

Another major player in the transformation of T-cell precursors to T-ALL is Notch1, which has activating mutations in over 65% of T-ALL patients (Weng et al., 2004). To further study the role of Notch1 in T-ALL, Chen et al. (2007) created a transgenic line expressing rag2:ICN1-EGFP, a GFP-tagged Notch1 intracellular domain which acts as a constitutively active transcription factor to drive Notch target gene expression. This line develops T-ALL, but at a lower incidence (40%) and higher latency (>11 months) than the Myc-driven tumors. However, in the presence of bcl2 overexpression, the incidence (60–80%) and latency (40 dpf to induction; 3 months to dissemination) were significantly enhanced and apoptosis was decreased. Blackburn et al. further demonstrated this by combining the rag2:ICN1-EGFP and rag2:cMyc models which accelerated leukemia onset and incidence (Blackburn et al., 2012). They concluded that Notch signaling expanded pre-leukemic clones that required Myc (or acquired secondary mutations) to transform, and that Notch signaling did not increase the overall frequency of LSCs. They also used this model to make cross-species microarray comparisons with mouse and human T-ALL to identify a common T-ALL gene signature and novel Notch gene expression profile present in humans that is regulated independently of Myc. These two studies suggest that Notch1 activation alone is not sufficient for induction of T-ALL and requires additional oncogene activation and/or tumor suppressor mutations.

Taking advantage of the ability to perform large-scale forwardgenetic screens in zebrafish to identify genetic modifiers of disease, Frazer et al. (2009) developed one such screen for causative mutations in T-ALL using ENU-mediated mutagenesis of an lck-EGFP line. This screen identified three mutant lines that developed outgrowth of the GFP-tagged thymus and subsequently T-ALL. Two of these lines, shrek (srk) and hulk (hlk), contained dominant mutations and one dubbed Oscar-the-grouch (otg) contained a recessive mutation. Homozygous fish from all 3 lines had incidences around 50% and time to tumor induction between 6 and 8 months. The mutated genes in these lines have not yet been reported, but the screen demonstrates the potential

for identification of genes driving different leukemias. The lab also developed a chemical screen to identify small molecules capable of eradicating immature T-cells, using the same lck-EGFP line (Ridges et al., 2012). They identified Lenaldekar (LDK; 1Hindole-3-carbaldehyde 8-quinolinylhydrazone) as a compound capable of killing both normal and T-ALL blasts in zebrafish, and showed it was effective in mouse xenograft and human primary leukemia cells as well.

#### B-Cell Acute Lymphoblastic Leukemia (B-ALL)

The TEL-AML1 (also known as ETV6-RUNX1) fusion protein results from t(12;21), the most common translocation in childhood cancer, present in ∼25% of B-cell acute lymphoblastic leukemia (B-ALL) (Romana et al., 1995). However, attempts to produce a model of B-ALL from this fusion gene were unsuccessful in mice (Andreasson et al., 2001). Sabaawy et al. (2006) created multiple lines expressing human TEL-AML1 from different promoters in zebrafish and were able to produce the only zebrafish model of B-ALL. Three different promoters were tested: the Xenopus ef1a (Xef1a) and zebrafish beta-actin (zba) for global expression, and zebrafish rag2 for lymphocyte specific expression. Both of the global promoters produced B-ALL tumors in ∼3% of fish with 8–12 months latency and similar molecular and morphological features to pediatric CD10+ B-ALL. The low incidence likely indicates the need for a secondary mutation for oncogenic transformation. They surmised that the rag2: TEL-AML1 fish did not develop tumors because the transformation occurs prior to the expression of Rag2 in the common lymphoid progenitor, and instead occurs in an earlier multipotent progenitor or hematopoietic stem cell in the global promoter lines. With the apparent T-cell bias of the rag2 promoter in zebrafish, it also seems possible that a different promoter of common lymphoid or B-cell progenitors may have more success. However, the recent discovery of B-ALL in the rag2: cMyc fish provides an opportunity for studying B-ALL in a more accessible model, with much shorter latency and higher incidence (Garcia et al., 2018).

### ZEBRAFISH LEUKEMIA MODELS: MYELOID ORIGIN

Following the initial success of the zebrafish ALL models, serious efforts began to recapitulate myeloid leukemias including myeloproliferative neoplasms (MPN) and acute myeloid leukemia (AML) in zebrafish. This was done largely through creating transgenic lines that expressed oncogenic fusion genes and mutations commonly found in patients with MPN and AML. This section discusses the features and major findings of the myeloid leukemia models developed to date in zebrafish (**Table 2**).

### Acute Myeloid Leukemia (AML) and Myeloproliferative Neoplasms (MPN)

Many hematological malignancies are driven by oncogenic fusion genes created after chromosomal translocations and these fusions can often be expressed in animal models or cell lines to drive transformation and oncogenesis. Zhuravleva et al. were the first to do so with a myeloid malignancy in zebrafish by creating transgenic fish expressing the MYST3/NCOA2 (MOZ/TIF2) fusion product under the spi-1 (pu.1) early myeloid promoter (Hsu et al., 2004) along with EGFP (Zhuravleva et al., 2008). This fusion protein is the result of the inv(8)(p11q13) chromosome abnormality found in human AML, and fuses two histone acetyltransferases (HATs). A small number of F<sup>0</sup> fish (1.1%) expressing the transgene developed AML after 14–26 months, characterized by expansion of myeloid blast cells and invasion of the kidney. This low incidence and long latency suggest that secondary mutations may be necessary to induce transformation.

Another model using the NUP98-HOXA9 (NHA9) fusion gene [t(7;11)(p15;p15)] was developed by Forrester et al. (2011) with an spi-1 promoter driving conditional expression of either EGFP or the transgene after heat shock by the hsp70-Cre line. This oncogenic fusion product is associated with poor prognosis in AML and chronic myeloid leukemia (CML) (Gough et al., 2011). Following heat shock at 24 hpf, NUP98-HOXA9;Cre embryos had perturbed hematopoiesis promoting myeloid fates, and also showed reduced apoptosis and cell cycle arrest in response to irradiation, correlating with increased levels of bcl2. 23% of NUP98-HOXA9;Cre fish developed myeloid tumors with a latency of 19–23 months. These tumors closely resembled the pathology of the polyclonal MPN found in NUP98-HOXA9-transgenic mice (Kroon et al., 2001). Further investigation into the model uncovered an increase in HSCs, as well as a dependency on meis1, the prostaglandin/cyclooxygenase pathway, and genome hypermethylation via dnmt1 for the fusion gene's oncogenic potential (Deveau et al., 2015). This dependency could be exploited through treatment with DNMT or COX inhibitors, or sub-therapeutic doses of either in combination with HDAC inhibitors. This study both revealed mechanistic details of the NHA9 oncogene and demonstrated the potential of zebrafish leukemia models in identification of new treatment combinations.

Because most leukemia oncogenes produce early detectable effects on hematopoiesis, along with the inherent advantages of the zebrafish model, it is possible to develop drug screens in preleukemic embryonic models. One such model was developed by Yeh et al. (2008) using the AML1(RUNX1)- ETO fusion oncogene under the heat shock responsive hsp-70 promoter. After heat shock, embryos accumulated noncirculating immature blast cells, with disruption of definitive hematopoiesis via loss of runx1 and cmyb expression, loss of gata1-expressing erythroid cells, and were ultimately driven to a myeloid-granulocytic fate. These effects were all downstream of AML1-ETO's suppression of scl, and could be reversed with scl overexpression. The transcriptional signature of AML1- ETO-expressing embryos closely paralleled that of human AML. Using the perturbation of embryonic hematopoiesis and the AML transcriptional signature as a readout of AML1-ETO oncogenic activity, they were able to develop a chemical screen for inhibitors that can rescue AML1-ETO's TABLE 2 | Zebrafish leukemia models of myeloid origin.

fcell-06-00115 September 19, 2018 Time: 10:25 # 7


effects (Yeh et al., 2009). This screen identified the COX and β-catenin pathways as vital to the function of AML1- ETO.

TEL(ETV6)-JAK2 fusion genes have been identified in both ALL and atypical chronic myelogenous leukemia (aCML), with slightly different translocations driving each, t(9;12)(p24;p13) and t(9;15;12)(p24;q15;p13), respectively (Peeters et al., 1997). Onnebo et al. (2012) created transgenic zebrafish lines expressing these different fusions under CMV or spi-1 promoters to better understand how they drive oncogenesis distinctly. These lines differ from other fusion gene transgenic lines in that the fusion proteins were generated from the zebrafish tel and jak2a genes combined to mimic two different human translocations found in T-ALL and aCML. Overall, the different genes behaved true to form, with the T-ALL fusion gene disrupting embryonic lymphopoiesis and the aCML fusion gene disrupting myelopoiesis similar to an MPN, driven by either CMV or spi-1. They were also able to demonstrate subtle differences in activity, with the T-ALL fusion gene showing greater enzymatic activity, but reduced downstream STAT activation and decreased sensitivity to JAK2 inhibition.

Although Zhuravleva et al. were the first to claim production of AML in a zebrafish model, the first myeloid malignancy was created in the Zon lab. Le et al. (2007) generated βactin: LGL-KRASG12D; hsp70-Cre zebrafish, with conditional global expression of an oncogenic KRAS inducible by heat shock. This model produced a variety of tumors following heat shock, including rhabdomyosarcoma, myeloproliferative neoplasm, intestinal hyperplasia, and malignant peripheral nerve sheath tumor. Although the heat-shocked fish had juvenile lethality, they discovered that the non-heat shocked adult fish developed MPN, likely due to the well-known "leakiness" of heat-shock promoters. The MPN-affected fish displayed classic disease characteristics such as expansion of myeloid progenitors, invasion of the marrow (kidney), and depletion of erythroid cells. Interestingly, these MPN cells could engraft after primary transplantation into irradiated recipients, but were unable to engraft after secondary transplantation, suggesting that they lack self-renewal capabilities. Another model utilizing an oncogenic RAS mutation was developed by Alghisi et al. (2013) inducing expression in the hemogenic endothelium prior to hematopoietic emergence. This fli1:GAL4-FF; UAS-GFP-HRASG12V line developed an MPN characterized by prominent expansion of the CHT, increased number of immature hematopoietic cells, and a block of myeloid differentiation in the kidney marrow. The Notch pathway was significantly downregulated and overexpression of the active NICD rescued the MPN phenotypes. They used this model to identify candidate genes both downregulated by Notch and upregulated by RAS that could be involved myeloid oncogenesis.

Similar to the connection of c-Myc to T-ALL, n-Myc is frequently upregulated in AML and is a poor prognostic marker. Shen et al. (2013) created a heat shock responsive zebrafish line expressing murine n-Myc, MYCN:HSE:EGFP, that simultaneously drives expression of EGFP. Following heat shock, n-Myc overexpression promoted immature myeloid blast cell expansion and enhanced the repopulating activity of myeloid cells. N-Myc enhanced primitive hematopoiesis by upregulating scl and lmo2 expression and promoted myelopoiesis by inhibiting gata1 expression and inducing spi1 and mpo expression. Many major cancer pathways were upregulated, such as cell cycle, glycolysis/gluconeogenesis, MAPK/Ras, and p53-mediated apoptosis. In contrast, mismatch repair and transforming growth factor β (TGFβ) signaling were downregulated. Overall, the model faithfully recapitulates AML phenotypes with high incidence (∼75%) and rapid onset (∼60 dpf).

Internal tandem duplications of the receptor tyrosine kinase FLT3 (FLT3-ITD) is a common mutation in AML and associated with poor prognosis and increased risk of relapse (Takahashi, 2011; Hou et al., 2013). It frequently coincides with mutations to the nucleophosmin NPM1 that restrict it to the cytoplasm (NPMc+). Lu et al. (2016) sought to investigate the interaction of these two mutations in AML by making transgenic lines expressing each under the myeloid spi1 promoter, spi1:FLT3-ITD-2A-EGFP and spi1:NPM1-Mut-PA. The FLT3-ITD mutant fish alone developed moderate myeloid hyperplasia at 6 months and some of these progressed to leukemia at 9 months. NPMc+ mutants had grossly normal hematopoietic composition. However, double mutants for both FLT3-ITD and NPMc+ progressed to leukemia by 6 months, demonstrating their synergistic effect in driving AML. In a different model using NPMc+ mRNA embryonic microinjections, Bolli et al. (2010) saw an increase in spi1<sup>+</sup> early myeloid progenitors, with a more pronounced effect in a p53 mutant line. NPMc+ expression resulted in increased erythromyeloid progenitors in the posterior blood island and c-myb/cd41<sup>+</sup> cells in the ventral wall of the aorta. They suggest these results may be relevant to human NPMc+ AML, where a multilineage expression pattern implies transformation of a multipotent HSPC.

Using a large-scale ENU mutagenesis screen, Peng et al. (2015) identified a line with a significant increase in HSPCs in hematopoietic organs, designated LDD731:CBLH382T. They determined the causal mutation was in the c-cbl gene, which is found frequently mutated in human MPN and acute leukemias and acts as a tumor suppressor by depressing growth factor and cytokine signals. The mutation was homozygous lethal at ∼15 dpf and led to an expansion of the myeloid/erythroid lineages in definitive hematopoiesis. Flt3 was necessary for this expansion, consistent with that observed in both mice and humans, suggesting flt3 signaling promotes HSPC proliferation and is regulated by c-cbl.

cAMP response element binding protein (CREB) is another frequently overexpressed gene in AML, however, it is unclear whether overactivation alone is sufficient to induce leukemia. Tregnago et al. (2016) generated a zebrafish model overexpressing creb with the spi1 myeloid promoter, which resulted in a disruption of myelopoiesis in 79% of adult fish with 66% progressing to a monocytic leukemia (latency 9–14 months) mirroring the human counterpart. These fish showed a transcriptional signature with 20 differentially expressed genes in common with pediatric AML, including the CCAAT-enhancer-binding protein-δ (c/ebpδ). Increased

c/ebpδ expression impaired myeloid differentiation which could be reversed through silencing of the creb-c/ebpδ axis. Identification of this creb-c/ebpδ axis in zebrafish AML led Tregnago et al to classify C/EBPδ expression as a new pediatric AML subgroup after validation in publicly available patient databases.

To study the role of interferon regulatory factor 8 (IRF8) in the pathogenesis of myeloid neoplasia, Zhao et al. (2018) created a missense mutation, irf81<sup>57</sup> that acted as a functional knockout. IRF8 is a critical transcription regulator for myeloid lineage commitment and closely tied to myeloid leukemia. irf8 mutants quickly developed MPN with expansion of myeloid precursors, which recurred after transplantation, and invasion of kidney marrow. Myeloid expansion was caused by both increased proliferation and decreased apoptosis. mertk expression was increased in irf8 mutants leading to hyperactivation of the ERK pathway. Transgenic mertk overexpression recapitulated the myeloid neoplasia and knockdown of mertk rescued irf8 mutant myeloid expansion. These results support mertk signaling as critical in the irf8-mediated regulation of myeloid proliferation and survival.

#### Myelodysplastic Syndrome

Myelodysplastic syndromes (MDS) are a group of diseases characterized by aberrant hematopoietic differentiation leading to cytopenias and increased blasts, and often splenomegaly and cytogenetic abnormalities (Gangat et al., 2016). Approximately 30% of MDS patients will eventually progress to AML or other leukemias, which are frequently more resistant to conventional therapies.

Somatic loss-of-function mutations of the 10–11 translocation 2 gene TET2 are frequently observed in patients with MDS. TET2 encodes a DNA methylcytosine oxidase that converts 5-methylcytosine (5 mC) to 5-hydroxymethylcytosine (5 hmC) to initiate the demethylation (and activation) of DNA. Gjini et al. (2015) created an enzymatically inert tet2 mutant zebrafish line through genome-editing technology. These fish had normal embryonic hematopoiesis, but developed progressive clonal myelodysplasia as they aged, eventually progressing to MDS by 24 months, with myeloid progenitor cell dysplasia and anemia. Decreased levels of 5 hmC were present in hematopoietic cells of the kidney marrow but not in other cell types, likely a result of compensation in non-hematopoietic tissues by other Tet family members.

The c-myb transcription factor is vital to hematopoietic proliferation and differentiation, and is closely associated with an array of hematological disorders. Liu et al. (2017) sought to better define its pathogenic role through characterization of a zebrafish model expressing a GFP-tagged c-myb mutant with increased activity, c-mybhyper. This hyperactive c-myb resulted in the dysregulation of cell cycle genes and subsequent proliferation of hematopoietic progenitor cells. Abnormal granulocyte expansion began embryonically and was maintained through adulthood, ultimately resulting in MDS. A small number of c-mybhyper fish developed AML or ALL and treatment with c-myb target drug flavopiridol relieved the MDS-like symptoms in both embryos and adult fish.

In addition to its previously discussed role in early myeloid progenitors, spi1 (pu.1) is an Ets-family transcription factor important in leukemogenesis. It is frequently impaired in AML either through decreased expression or loss-of-function mutations (Mueller et al., 2002; Dakic et al., 2007). Sun et al. (2013) used the Targeting Induced Local Lesions IN Genomes (TILLING) approach to create a hypomorphic spi1 mutant allele, dubbed pu.1G242D. These fish have expanded myelopoiesis by 3 dpf, with increased immature granulocytes within the CHT. By 18 months, immature myeloid cells were increased at the expense of the lymphoid population in both the kidney marrow and peripheral blood, consistent with an MDS or AML-like disorder. The antiproliferative drug cytarabine was able to relieve the myeloid expansion, while apoptosisinducing daunorubicin could not. This may indicate that spi1 associated neoplasms are more susceptible to drugs limiting their proliferation.

#### FUTURE DIRECTIONS AND EMERGING METHODS

The discoveries described in this review open numerous avenues for further research. Many of these models require further characterization and could uncover important pathways in leukemia initiation and progression. Drug screens utilizing these models can teach us much about the resistance and response to different therapies depending on the specific genetic drivers of the leukemia. The advent of effective CRISPR-Cas9 protocols has allowed for a rapid advancement in the creation of knockout and transgenic zebrafish investigating various genetic pathways and oncogene fusion products tied to human leukemia. This development will only continue to expand in scope, as ongoing research within the zebrafish field continues to uncover more genes and pathways associated with leukemia, as well as new discoveries made in the clinic that are converted into zebrafish models for further characterization.

One major emerging avenue of research is the identification and characterization of LSCs within hematopoietic malignancies. Cancer stem cells (CSCs), defined by their ability to regrow a tumor from a single cell, are implicated as the cause of cancer evolution, resistance to therapy, and relapse after therapy (Adorno-Cruz et al., 2015). Increased tumor heterogeneity and CSCs have been associated with resistance and relapse for many tumor types, including AML and ALL (Mullighan et al., 2008; Anderson et al., 2011; Notta et al., 2011; Ding et al., 2012). Identification and isolation of these cells is difficult because of a lack of defined surface markers, but there are some promising methods being developed. The side population assay has long been used to isolate normal tissue stem cells by exploiting their ability to export Hoechst dye (Goodell et al., 1996), and has more recently been shown to enrich for CSCs in many cancers (Hu et al., 2010; Britton et al., 2011; Richard et al., 2013). Side populations have also been defined in zebrafish hematopoietic cells and leukemia (Kobayashi et al., 2008; Pruitt et al., 2017), making it possible for LSCs to be further studied in zebrafish

leukemia models. Another similar protocol of enriching for stem cell activity is through the Aldefluor assay, which utilizes the increased aldehyde dehydrogenase (ALDH) activity common in stem cells to produce increased fluorescence from the aldefluor reagent (Storms et al., 1999; Ma et al., 2010). This assay can also be combined with the side population assay to isolate an even greater enrichment of stem cells (Pearce and Bonnet, 2007; Pierre-Louis et al., 2009). Genetic and functional characterization of the combined ALDHbright and side population in zebrafish leukemia models could uncover significant contributors to leukemia resistance and relapse.

Another potential method for defining LSCs within a tumor is through single cell sequencing. Single-cell RNA sequencing techniques are capable of discerning expression profiles of each cell within a population, allowing small subpopulations like LSCs to be characterized within a tumor (Zhang et al., 2016). Multiple microfluidic systems have been developed to produce single-cell expression data, with the Fluidigm system already used by Moore et al. (2016) in a zebrafish T-ALL model to identify a small population with reactivated expression of putative stem cell genes. The DropSeq system is an alternative with much higher throughput that could potentially identify very small subpopulations in tumors with lower LSC frequencies (Macosko et al., 2015). Both systems allow for characterization of the expression profiles of LSCs in the various leukemia models, which opens up exciting possibilities in discovering what drives

#### REFERENCES


the LSC subpopulation and their unique functions within the leukemia.

#### AUTHOR CONTRIBUTIONS

JB wrote the entirety of the review, with significant input from JdJ in the initial outline and also comments and revisions of the final version.

#### FUNDING

Investigator Award University of Chicago Women's Board Discovering the genetic drivers of leukemia initiating cells in T-cell acute lymphoblastic leukemia. Program Pilot Project Grant University of Chicago Comprehensive Cancer Center Genetic drivers of leukemia initiating cells in the side population of zebrafish T-cell leukemia.

#### ACKNOWLEDGMENTS

We would like to acknowledge the University of Chicago Women's Board and the University of Chicago Comprehensive Cancer Center for helping to fund our work in this field.

T cell acute lymphoblastic leukemia through Akt/mTORC1 pathway activation. Cancer Cell 25, 366–378. doi: 10.1016/j.ccr.2014.01.032





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

Copyright © 2018 Baeten and de Jong. 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.

# Mga Modulates Bmpr1a Activity by Antagonizing Bs69 in Zebrafish

Xiaoyun Sun<sup>1</sup> , Ji Chen<sup>1</sup> , Yanyong Zhang<sup>1</sup> , Mumingjiang Munisha<sup>2</sup> , Scott Dougan<sup>2</sup> \* and Yuhua Sun<sup>1</sup> \*

1 Institute of Hydrobiology, Chinese Academy of Sciences, Wuhan, China, <sup>2</sup> Department of Cellular Biology, University of Georgia, Athens, GA, United States

MAX giant associated protein (MGA) is a dual transcriptional factor containing both T-box and bHLHzip DNA binding domains. In vitro studies have shown that MGA functions as a transcriptional repressor or activator to regulate transcription of promotors containing either E-box or T-box binding sites. BS69 (ZMYND11), a multidomain-containing (i.e., PHD, BROMO, PWWP, and MYND) protein, has been shown to selectively recognizes histone variant H3.3 lysine 36 trimethylation (H3.3K36me3), modulates RNA Polymerase II elongation, and functions as RNA splicing regulator. Mutations in MGA or BS69 have been linked to multiple cancers or neural developmental disorders. Here, by TALEN and CRISPR/Cas9-mediated loss of gene function assays, we show that zebrafish Mga and Bs69 are required to maintain proper Bmp signaling during early embryogenesis. We found that Mga protein localized in the cytoplasm modulates Bmpr1a activity by physical association with Zmynd11/Bs69. The Mynd domain of Bs69 specifically binds the kinase domain of Bmpr1a and interferes with its phosphorylation and activation of Smad1/5/8. Mga acts to antagonize Bs69 and facilitate the Bmp signaling pathway by disrupting the Bs69–Bmpr1a association. Functionally, Bmp signaling under control of Mga and Bs69 is required for properly specifying the ventral tailfin cell fate.

#### Keywords: Mga, Bmp signaling, Zmynd11, Bs69, Bmpr1a, ventral tailfin

### INTRODUCTION

Bone morphogenetic proteins (BMPs) comprise a subgroup of the TGF-beta family of secreted signaling molecules. They transduce their signal by extracellular binding to membrane protein complex consisting of a type I receptor (BMPRI) and a type II receptor (BMPRII). Type I BMP receptor (BMPRI) activation leads to the phosphorylation and activation of Smad1/5/8. The pSmad1/5/8 form complex with SMAD4 and translocate to the nucleus to regulate transcription of downstream target genes. BMP signaling is known to control multiple important biological events, ranging from dorsal-ventral patterning, stem cell maintenance and differentiation to tissue homeostasis (Katagiri and Watabe, 2016). Given its importance in development and homeostasis, Bmp signaling is tightly regulated at the extra- and intracellular levels, by numerous factors such as Noggin, Chordin, Smad7, and Fkbp1A (Wang et al., 2014).MAX's giant associated protein (MGA) was first identified as a MAX interacting protein by a yeast two hybrid assay in a mouse embryonic day e9.5 and e10.5 cDNA library (Hurlin et al., 1999). Like other Myc family of transcriptional factors, MGA has a basic helix-loop-helix zipper (bHLHZip) domain that mediates dimerization with MAX, which is required for their specific DNA binding to E-box sequences. In addition to the bHLHZip domain, MGA contains a second

#### Edited by:

Ryan M. Anderson, Indiana University–Purdue University Indianapolis, United States

#### Reviewed by:

Saurabh Chattopadhyay, University of Toledo, United States Vasudevan Seshadri, National Centre for Cell Science (NCCS), India

#### \*Correspondence:

Scott Dougan dougan@uga.edu Yuhua Sun sunyh@ihb.ac.cn

#### Specialty section:

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

Received: 29 April 2018 Accepted: 10 September 2018 Published: 28 September 2018

#### Citation:

Sun X, Chen J, Zhang Y, Munisha M, Dougan S and Sun Y (2018) Mga Modulates Bmpr1a Activity by Antagonizing Bs69 in Zebrafish. Front. Cell Dev. Biol. 6:126. doi: 10.3389/fcell.2018.00126

DNA-binding domain, the T-box or T-domain. In vitro studies suggested that MGA could regulate transcription of promotors containing either E-box or T-box binding sites (Hurlin et al., 1999). MGA is thus proposed to function as a dual-specificity transcription factor that could regulate the expression of both the MAX-network and the T-box gene family genes. MGA:MAX heterodimers were often found as part of a large transcriptionsilencing complex E2F6-com.1 or as part of polycomb repressive complex 1 (PRC1) that catalyzes the monoubiquitylation of histone H2A (Ogawa et al., 2002; Gao et al., 2012). MGA:MAX were shown to repress developmental genes in somatic or embryonic stem cells by recruiting PRC1.6 complex to gene promotors (Endoh et al., 2017; Suzuki et al., 2017). Consistently, MGA depletion leads to the death of proliferating pluripotent ICM cells in vivo and in vitro, and the loss of self-renewal and pluripotency of embryonic stem cells (Washkowitz et al., 2015). Moreover, Mga mutation in somatic cells is associated with a variety of tumor or cancers, including aggressive lymphoma called Richter's Syndrome, which occurs in a minority of patients with chronic lymphocytic leukemia (De Paoli et al., 2013; The Cancer Genome Atlas Research Network, 2014; Jo et al., 2016). Together, these studies suggest that MGA functions as a tumor suppressor in normal tissues, presumably by antagonizing Myc oncogene or by recruiting PRC1 to target genes.

Because Mga deficient mice are embryonic lethal, the role of MGA in vertebrate embryogenesis and disease remains unclear. To overcome this issue, we and others took advantage of the zebrafish model system, which is useful for developmental biology studies because of its transparent embryo during early embryogenesis and it is also highly amenable for genetic studies. By Morpholino-mediated gene knockdown, Rikin and Evans (2010) showed that Mga plays essential role in organogenesis by regulating gata4 expression. We recently reported that Mga together with Smad4 and Max are required for the dorsal ventral patterning of zebrafish embryos by transcriptionally regulating Bmp2 expression in the extra embryonic tissue, yolk syncytial layer (YSL) (Sun et al., 2014). This was the first report showing that Mga is involved in the regulation of Bmp signaling in a vertebrate. Unlike other Myc family of transcriptional factors, Mga is predominantly localized in the cytoplasm throughout early zebrafish embryogenesis. This observation implies that Mga has important roles in the cytoplasm, and it may also regulate Bmp signaling independent of its transcriptional activities.

BS69 (ZMYND11) is a multidomain-containing (i.e., PHD, BROMO, PWWP, and MYND) protein that was originally identified as an adenoviral early region 1A-interacting protein (Ansieau and Leutz, 2002; Harter et al., 2016). Through its PHD–BROMO–PWWP domains, BS69 selectively recognizes histone variant H3.3 lysine 36 trimethylation (H3.3K36me3), modulates RNA Polymerase II elongation, and functions as RNA splicing regulator for intron retention (Guo et al., 2014; Wen et al., 2014). The MYND domain of BS69 seems to act as an important protein–protein interaction surface, through which BS69 interacts with a variety of chromatin regulators, including MGA (Velasco et al., 2006). Recently, a growing body of research has shown that BS69 localized in the cytoplasm or cytoplasmic membrane is involved in mediating multiple signaling pathways. For instances, BS69 physically interacts with LMP1 and negatively regulates LMP1 mediated JNK and Nf-κB activation (Hateboer et al., 1995; Chung et al., 2002; Wan et al., 2006; Ikeda et al., 2009). BS69 associates with lymphotoxin beta receptor (LTβR) and inhibits LTβR-mediated signaling transduction (Liu et al., 2011). It has been shown that BRAM1, an alternatively spliced form of BS69, may inhibit Bmp signaling by interacting with the type I BMP receptor 1A (Kurozumi et al., 1998; Morita et al., 2001; Wu et al., 2006). However, BRAM1 was thought likely to be an artificial product that generated from the library construction (Velasco et al., 2006). Therefore, it remains unclear whether BS69 is involved in the regulation of Bmp signaling.

Using zebrafish as a model, we revealed a cytoplasmic role for Mga in the regulation of Bmp signaling. We showed that Mga protein localized in the cytoplasm acts to antagonize Bs69 to facilitate the Bmp signaling pathway. Mechanistically, Mga binds to Bs69 and disrupts the Bs69–Bmpr1a association, thereby maintaining proper Bmp signaling that is required to properly specify zebrafish ventral tail fin.

### MATERIALS AND METHODS

#### Zebrafish Maintenance

Zebrafish (Danio rerio) were maintained at 28.5◦C on a 12 h light/12 h dark cycle. All procedures were performed with the approval of the Institute of Hydrobiology, Chinese Academy of Sciences, Wuhan, China.

### Generation of mga and bs69 Mutants Using TALEN or CRISPR/Cas9

Mga mutants (GenBank accession numbers MH853640-853641) were generated by TALEN technology as described (Chen et al., 2016). We identified potential TALEN-target sites in the coding sequence of zebrafish mga gene (NM\_001170739.1) using ZiFit 3.0<sup>1</sup> . The mga TALEN recognition sequences are: left TALEN 5<sup>0</sup> -CCATTGCAGCCCAGCCTG-3<sup>0</sup> and right TALEN 5 0 -GAATGAGACGAACAGTT-3<sup>0</sup> . Between the two binding sites is an 16-bp spacer (GAGGATGTCGAAGGTC). Genotyping was conducted using PCR followed by restriction enzyme digestion. The primers used were 5<sup>0</sup> -TTCTGACAACAGTATTTCCA-3<sup>0</sup> and 5<sup>0</sup> -CTCGTTCTAAACTCGGTTGACT-3<sup>0</sup> . Bs69 mutants (GenBank accession numbers MH853642-853643) were generated by CRISPR/Cas9 technology. gRNA was designed to target a site 5<sup>0</sup> -GGCTGATGTGGAACAGCTGT-3<sup>0</sup> in exon 15 of bs69 gene. Genotyping was conducted using PCR followed by restriction enzyme digest. The primers were 5<sup>0</sup> - CCCTTACAGTCTCCTCCTGTAT-3<sup>0</sup> and 5<sup>0</sup> -TGTTCTCCGC CTTCATCATTT-3<sup>0</sup> . Mutagenized F0 males were crossed to wildtype females to obtain F1 fish. The F1 heterozygous females were then crossed with wild type males to derive the F2 heterozygous. The F2 heterozygous fish were randomly intercrossed, yielding F3-offspring.

<sup>1</sup>http://zifit.partners.org/ZiFiT/

### Plasmid Constructions and Microinjections

fcell-06-00126 September 26, 2018 Time: 15:23 # 3

To make PCS2- version of constructs used in this work, cDNAs encoding zebrafish Mga, Bs69 and their mutants were generated by RT-PCR from cDNA libraries made from 8 hpf zebrafish embryos, and then cloned into the pCS2+ vector using the In-fusion HD Cloning kit (Clontech). The primers used are shown in **Table 1**. Mga, bs69 and bmpr1aa and mutant mRNAs were made using mMESSAGE mMACHINE <sup>R</sup> Kit (Ambion, TX, United States). mRNAs were injected to the zebrafish embryos at one-cell-stage by a microinjector (WPI, United States).

#### Cell Culture and Transfection

HEK293 or HEK293T cells were cultured with DMEM (BI) containing 10% fetal bovine serum (FBS, Gibco). Cell transfection was performed with Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Briefly, cells (about 10<sup>6</sup> viable cells) were seeded in 6-well plates without antibiotics in DMEM medium containing 10% FBS. Then the constructs expressing the tagged protein of interest or empty pCS2+ vector were transiently transfected into HEK293 cells. 48 h later, the cells were harvested for co-immunoprecipitation and Western blot analysis.

### Luciferase Assay

The BRE-luc reporter was kindly provided by Prof. Zongbin Cui. pTK-Renilla was kindly provided by Dr. Xing Liu (Liu et al., 2015). HEK293 or C2C12 cells (about 10<sup>5</sup> viable cells) were seeded in 24-well plates in DMEM medium containing 10% FBS for 24 h. The cells were then transiently transfected with indicated luciferase reporters using Lipofectamine 2000 (Invitrogen). pTK-Renilla was used as an internal control. After transfection, the cells were treated with BMP4 (10 ng/ml) or control for 16 h. The luciferase activity was measured with the Dual-luciferase Reporter Assay system (Promega).

#### Co-immunoprecipitation and Western Blot Analysis

The protocols for the co-immunoprecipitation and WB analysis were described (Sun et al., 2011). For pSmad1/5/9 WB analysis, the phosphatase inhibitor (Thermo, A32957) was used. For in vivo co-immunoprecipitation assay, whole cell lysates were prepared with the TEN buffer (50 mM Tris-HCL, 150 mM NaCl, 5 mM EDTA, 1% Triton X-100, 0.5% Na-deoxycholate, and ROCHE protease inhibitor cocktail). The cytoplasmic and nuclear fractions were prepared with a ProteinExt <sup>R</sup> Mammalian Nuclear and Cytoplasmic Protein Extraction Kit (TransGen, DE201-01). 1.5 mg Dynabeads protein G was conjugated with 10 µg anti-Mga antibodies, 10 µg anti-Bs69 antibody, 10 µg anti-FLAG antibody, 10 µg anti-HA antibody, or 10 µg IgG. The cell lysates and the antibody-conjugated Dynabeads were incubated overnight at 4◦C. After extensive washing, the beads-protein complex were boiled and the supernatant was loaded on a 8% PAGE gel for electrophoresis. The proteins were transferred to a PVDF membrane, followed by blocking with 5% (w/v) non-fat milk for 2-h at room temperature. Then the membrane was incubated overnight at 4◦C with the primary antibodies. Anti-MgaN (STTPSENLPADAR); anti-MgaI (EHSADKNTLKSSDQN); anti-MgaC (SPDSKDEIDIPPK) were made by Genscript (China). Anti-FLAG(1B10) and anti-HA (4F6) were purchased from Abbkine (China). Anti-pSmad1/5/9 (D5B10) was purchased from the Cell Signaling Technology. HRP-conjugated goat anti-rabbit IgG (GtxRb-003-E3EUR) was purchased from ImmunoReagents. Anti-mouse IgG, HRPlinked antibody (#7076s) was purchased from Cell Signaling Technology.

### Protein–Protein Interaction Assay Using Rabbit Reticulocyte Lysate System

FLAG or HA tagged Mga, Bs69 or Bmpr1a proteins were synthesized using the TnT coupled reticulocyte lysate system according to the manual (L5020, Promega, United States). Briefly, 1 µg of circular PCS2- version of plasmid were added directly to the TnT Lysate and incubated in a 50 µl reaction for 1.5 h at 30◦C. To evaluate the quality or quantity of the synthesized protein, 1 µl of the reaction products were subjected to WB assay.

For protein–protein interaction assay, 5–10 µl of the synthesized HA or FLAG tagged proteins were mixed in a 1.5 ml tube with the TEN buffer, and the mixture was shaken for 30 min at room temperature. Next, Co-IP or pull-down assay was performed using Dynabeads protein G coupled with FLAG or HA antibodies as described above (Sun et al., 2014).

#### Immunofluorescence Assay

Six to eight hpf zebrafish embryos were collected and fixed with 4% paraformaldehyde (PFA) overnight at 4◦C. Embryos were washed with PBST (0.1% Triton X-100) and then permeabilized


with acetone for 7 min at −20◦C. The embryos were rinsed with PBST, followed by 1 h blocking with solution (0.1% Triton X-100, 1% BSA, and 1% DMSO in PBS). Then the embryos were incubated with primary antibodies overnight rocking at 4 ◦C. After washing with the blocking solution, embryos were incubated with the secondary antibodies for 2 h at room temperature, followed by extensively washing. Nuclei was stained with DAPI. Primary antibodies diluted with blocking solution were anti-pSmad1/5/9 (1:300), anti-FLAG (1:300), and anti-HA (1:300). Secondary antibodies diluted with blocking solution were goat anti-mouse IgG(H+L) Alexa Fluor 555 (1:500, Molecular Probes), goat anti-rabbit IgG(H+L) Alexa Fluor 488 (1:500, Molecular Probes). The embryos were transferred to a Glass Bottom (NEST) and submerged with 75% glycerol. The images were taken under a Leica SP8 confocal microscope (Germany).

#### Whole-Mount in situ Hybridization

Whole-mount in situ hybridization (WISH) was performed as described previously (Sun et al., 2011). The DIG-labeled probes were generated with DIG RNA Labeling Kit (SP6/T7) (Roche) and Fluorescein-labeled RNA probes were made with Fluorescein RNA Labeling Mix, 10x conc. (Roche). Mga, chordin, and eve1 RNA in situ probes were described before (Sun et al., 2014). Zebrafish embryos were collected at different development stages and fixed with 4% PFA overnight at 4◦C. Following the WISH, the embryos were transferred to 6-well plates and submerged by 100% glycerol for imaging.

## RESULTS

### Mga Positively Regulates Bmp Signaling

To establish a genetic model to explore the developmental function of Mga, and avoid potential off-target effects of Morpholino Oligos, we generated mga mutant zebrafish by using TALEN technology (**Figure 1A**). Out of 12 potential founders, we identified two fish in which mga gene was mutated at the TALEN cleavage site in the exon 2. The identity of each mutation was confirmed by genotyping (**Figure 1B**).

FIGURE 1 | Mga mutant embryos exhibited mild dorsalized phenotype. (A) Schematic representation of the zebrafish mga gene, depicting the location of the TALEN targeting site. (B) Sequences around the TALEN targeting site, showing the TALEN-induced 5-bp deletion in mga (in red). (C) Phenotypes of 1 dpf wild type, mga mutant, and mga mutant embryos injected with 50 pg mga mRNA at one-cell-stage. The reduction of ventral tail fin was restored by injecting 50 pg mga mRNA into one-cell-stage of mga mutant embryos. Lateral view. (D) Immunoblot analysis of Mga and pSmad1/5 levels of lysates from 7 hpf control, mutant and mga mRNA restored embryos. (E) Quantification of pSmad1/5 levels of panel D based on three independent experiments. (F) pSmad1/5 gradient of wild type, mga > YSL morphants, and mga mutant embryos at 7 hpf. Dorsal to the right. (G) Chordin expression in mga mutant and control embryos at shield stage. Lateral view, and dorsal to the right. (H) eve1 expression in mga mutant and control embryos at 22 hpf. Lateral view, and dorsal to the right. (I) qRT-PCR transcript analysis of the indicated Bmp target genes in control and mga mutant embryos at 8 hpf. (J) Immunoblot analysis of pSmad1/5 levels of lysates from 8 hpf control and Mga overexpressing (OE) embryos. All experiments were performed in technical triplicate and are representative of multiple experiments. <sup>∗</sup>p < 0.05.

Each mutation results in an open reading frame-shift that leads to a premature stop codon. Western blotting was used to detect Mga protein in lysates from homozygous mutant and control embryos at 8 hpf, using Mga specific antibodies (Sun et al., 2014). As seen in **Figure 1D**, the band around 250 kDa was detected in lysates from control embryos, but was barely detected in lysates from mutant embryos. These results indicate that our mutant alleles are functional nulls. One mutation line with a 5 bp deletion in exon 2 of the mga gene, was used for most of the subsequent studies (**Figure 1B**).

The mutant embryos at 1 dpf appear largely normal except the loss or reduction of the ventral tail fin defect (**Figure 1C**). Because mga mutant embryos at 1 dpf exhibited the loss of ventral tailfin defect that resembled our previously characterized mga > YSL morphants, we went on to confirm that Bmp signaling was compromised in mga mutant embryos (Sun et al., 2014). Nuclear phosphorylated Smad1/5/8 is a direct intracellular readout of Bmp signaling. As expected, pSmad1/5 levels were indeed reduced in mutant embryos. Importantly, pSmad1/5 levels could be rescued by injecting 100 pg mga mRNA, demonstrating the specificity of observed phenotype (**Figures 1D,E**). A low but detectable level of phospho-Smad1/5 was still present in lysates of mutant embryos, suggesting that Mga is only required for higher Bmp activity. This was in accordance with the notion that the ventral tail fin formation is most sensitive to the reduction of Bmp signaling (Schumacher et al., 2011). We also compared the pSmad1/5 gradient among 7 hpf mga > YSL morphants, mga mutant and control embryos by immunofluorescence assay using anti-pSmad1/5/9 antibody. As seen from **Figure 1F**, mga mutant embryos and mga > YSL morphants both exhibited reduced pSmad1/5 gradient when compared with the control embryos. However, the pSmad1/5 gradient in mga mutant embryos was reduced to a greater extent than mga > YSL morphants (**Figure 1F**), suggesting that Mga cell autonomously regulates Bmp signaling in embryos in addition to its role in YSL (Sun et al., 2014). To further confirm that Bmp activity was reduced in mga mutants, whole mount in situ hybridization was performed to examine the expression pattern of dorsal marker chordin and the ventral marker eve1. As expected, chordin expression was slightly expanded in mutant embryos at shield stage, whereas eve1 expression was slightly reduced in mutant embryos at 22 hpf compared with the controls (**Figures 1G,H**). Moreover, the expression of well-known Smad-dependent Bmp target genes Id1 and bmp4 was down-regulated in mga mutant embryos (**Figure 1I**).

On the other hand, we overexpressed Mga by injecting 200 pg mga mRNA into one-cell-stage wild-type embryos. Beta-gal overexpressing embryos were used as controls. As seen in **Figure 1J**, pSmad1/5 levels were slightly elevated in Mga overexpressing embryos compared with control embryos, which further supports that Mga positively regulates Bmp signaling.

Taken together, we concluded that Mga is cell autonomously required for proper Bmp signaling that is important for specifying the ventral tailfin cell fate during zebrafish embryogenesis.

### Mga Interacts With Bs69 in Zebrafish Embryos

Our yeast two hybrid experiments have identified multiple Mga interacting proteins, including Smad1, Smad4, and type I Bmp receptors, suggesting that Mga could regulate Bmp signaling by physical association with the core components of Bmp signaling pathway (Sun et al., 2014). Type I Bmp receptor 1a (Bmpr1a) was of particular interest to us because previous studies have shown that bmpr1a mutant embryos at 1 dpf exhibited the loss or reduction of ventral tailfin defect that closely resembled our mga mutant embryos (Smith et al., 2011), and also that wild type embryos injected with mRNA encoding dominant negative Bmpr1a (dnBmpr1a) at one-cellstage had defects in the ventral tailfin formation at 1 dpf (Pyati et al., 2005) (**Supplementary Figure S1A**). Wild type embryos treated with 0.05 µM dorsomorphin or LDN193189, which are known potent Bmpr1a inhibitors, exhibited the loss or reduction of ventral tail fin defect at 1 dpf (**Supplementary Figure S1A**). Furthermore and most importantly, injection lower dose (25 pg) constitutive active bmpr1a (caBmpr1a) mRNA into one-cell-stage mga mutant embryos rescued the loss or reduction of the ventral tailfin phenotype (Nikaido et al., 1999) (**Supplementary Figure S1B**). Altogether these previous studies, along with our observations, strongly suggested a functional link between Mga and Bmpr1a. Unfortunately, we failed to detect reproducible interaction between Mga and Bmpr1a (data not shown). Transcriptional regulation of bmpr1a gene by Mga was ruled out, as bmpr1a transcript levels in mga mutant embryos were comparable to control embryos (**Figure 1I**).

It has been previously shown that mammalian MGA directly interacts with BS69, and that BRAM1, a possible spliced form of BS69, may be involved in the regulation of BMPR1A activity (Kurozumi et al., 1998; Ansieau and Leutz, 2002; Velasco et al., 2006; Wu et al., 2006). Therefore, we speculated that Mga may modulate Bmpr1a activity through Bs69 in zebrafish. To test these hypothesis, we firstly examined the gene expression pattern of mga, bs69, and bmpr1a during zebrafish early embryogenesis. The three genes have similar expression patterns from blastula to late organogenesis as reported (Wu et al., 2006; Rikin and Evans, 2010; Smith et al., 2011). During gastrulation, both bmpr1aa and mga transcription levels are high, whereas bs69 transcript levels are relatively low. During early somitogenesis stage, mga and bmpr1aa are strongly expressed in the trunk and tail region, whereas bs69 expression domain seems to be more restricted to the ventral region of the trunk. During organogenesis, all three genes are strongly expressed in the head and gut regions (**Supplementary Figure S1C**). The contrasting expression patterns between mga and bs69 genes imply that Mga may act to antagonize Bs69 to modulate Bmpr1a activity.

To investigate whether Mga regulates Bmpr1a activity through Bs69, we determined whether Mga associates with Bs69 in physiological conditions in zebrafish embryos. The cellular colocalization of Bs69 and Mga was examined in 7 hpf embryos

(**Figure 2A** and **Supplementary Figure S2A**). One-cell-stage wild-type embryos were injected with 100 pg HA-tagged bs69 mRNA, and embryos were collected for co-immunofluorescence assay using Mga and HA antibodies. Our IF data clearly showed that these two proteins were co-localized in both cytoplasm and nucleus. To determine whether Mga interacts with Bs69 in physiological conditions, we performed co-immunoprecipitation (co-IP) assay using Mga and HA antibodies. Cytoplasmic and nuclear fractions of lysates from 7 hpf embryos were prepared using a commercial TransGen kit, and were subjected to co-IP experiments. It was obvious that Mga interacts with HA-Bs69 in both cytoplasm and nucleus (**Figures 2B,C**).

Next, we mapped the interacting domain between Mga and Bs69 in 293T cells. We first confirmed that FLAG-Mga and HA-Bs69 interact with each other (**Figure 2D**). It was previously reported that the PXLXP motif of mammalian MGA binds to the MYND domain of BS69 (Ansieau and Leutz, 2002). We identified a conserved PXLXP motif (PKLVP) within the C-terminal region (amino acid 2702-2706) of zebrafish Mga protein (**Figure 2E**). Deletion of this PXLXP motif (FLAG-Mga1) abrogated Mga binding to HA-Bs69 (**Figure 2G**). Truncated Bs69 lacking the Mynd domain (HA-Bs691) did not co-immunoprecipitate with the FLAG-tagged Mga (**Figure 2H**). However, FLAG-tagged C-terminal fragment of Mga (FLAG-Mga-Cter) containing the PXLXP motif was sufficient to bind the Mynd domain of Bs69 (HA-Bs69-Mynd) (**Figure 2I**). To determine whether this interaction was direct or not, we used a reticulate lysate system to synthesize FLAG-Mga-Cter and HA-Bs69-Mynd. When they were mixed together, the anti-FLAG antibody readily pulldowned HA-Bs69-Mynd (**Figure 2J**).

Taken together, we concluded that zebrafish Mga physically associates with Bs69 in the physiological condition and this interaction is mediated by the PXLXP motif of Mga and the Mynd domain of Bs69.

### Bs69 Negatively Regulates Bmp Signaling

To understand the function of Bs69, we generated bs69 mutant zebrafish by CRISPR/Cas9 technology (**Figure 3A**). We designed gRNAs targeting exon 2 or 15 of bs69. Out of eight potential founders, we identified two fish in which bs69 gene was mutated around the CRISPR targeting site of exon 2. Both mutations result in an open reading frame-shift that leads to a premature stop codon. Out of nine potential founders, we identified two fish in which bs69 gene was mutated around the CRISPR targeting site

phenotyped according to the dorsal-ventral patterning index. (J) The phenotypes of Bs69 overexpressing embryos at 2 dpf. C1-4 dorsalized phenotypes according to DV patterning index. (K) Immunoblot analysis of pSmad1/5 levels of lysates from 8 hpf control and Bs69 overexpressing embryos. (L) Chordin expression in Bs69 overexpressing embryos at shield stage. Animal view, and dorsal to the right. (M) eve1 expression in BS69 overexpressing embryos at 22 hpf. Lateral view, and dorsal to the right. (N) qRT-PCR transcript analysis of the indicated Bmp target genes in 8 hpf control and Bs69 overexpressing embryos. All experiments were performed in technical triplicate and are representative of multiple experiments. <sup>∗</sup>p < 0.05.

of exon 15. Both mutations result in an open reading frameshift that leads to a truncated Bs69 protein that lack the entire Mynd domain. One mutant causes 1-bp insertion within the CRISPR targeting site, resulting in a ∼475aa truncated protein. Another mutant has 11-bp deletion within the CRISPR targeting site, resulting in a ∼480aa truncated BS69 protein (**Figure 3B**). The F3 zygotic bs69 mutants are viable and can be raised up to adulthood. When F3 female and male adults were intercrossed, maternal zygotic bs69 mutant embryos were obtained for further analysis.

The majority of bs69 mutant embryos at 1 dpf exhibited mild V1 ventralized phenotype, characterized by slightly reduced head region. Some of the bs69 mutant embryos at 1 dpf displayed missing notochord, and enlarged ventral cell types, indicating a V2 ventralized phenotype (Kishimoto et al., 1997) (**Figure 3C**). Smad1/5 phosphorylation was increased in 8 hpf bs69 mutants compared with controls, and this could be rescued by injecting bs69 mRNA into one-cell-stage mutant embryos (**Figures 3D,E**). To confirm that bs69 mutants had the ventralized phenotype, whole mount in situ hybridization was performed to examine the expression pattern of dorsal marker chordin and the ventral marker eve1. As expected, the chordin expression domain was decreased in mutant embryos at shield stage, while eve1 expression domain was expanded in mutant embryos at 22 hpf compared with control embryos (**Figures 3F,G**). Moreover, the expression of the known Bmp target genes Id1 and bmp4 was

examined in bs69 mutant and control embryos. As seen in **Figure 3H**, the expression of these genes was up-regulated in bs69 mutant embryos. Injection of 50 pg bs69 mRNA into onecell-stage bs69 mutant embryos largely rescued the ventralized phenotypes (**Figure 3I**).

Next, we overexpressed Bs69 by injecting bs69 mRNA into one-cell-stage wild-type embryos. Beta-gal overexpressing embryos were used as controls. The Bs69 overexpressing embryos at 24 hpf exhibited dorsalized phenotypes ranging from C1 to C4 dorsalization depending on the injected mRNA dose (**Figure 3J**). Smad1/5 phosphorylation was decreased in Bs69 overexpressing embryos (**Figure 3K**). Importantly, the DV patterning phenotype of Bs69 overexpressing embryos was similar to those of bmpr1a mutants or dnBmpr1a overexpressing embryos (Smith et al., 2011). To further confirm that Bs69 overexpressing embryos had dorsalized phenotypes, whole mount in situ hybridization was performed to examine the expression pattern of dorsal marker chordin and the ventral marker eve1. As expected, chordin expression domain was expanded in Bs69 overexpressing embryos at shield stage, while eve1 expression domain was reduced in Bs69 overexpressing embryos at 22 hpf, compared to controls (**Figures 3L,M**). Moreover, the expression of Id1 and bmp4 genes was down-regulated in Bs69 overexpressing embryos, which was similar to that of mga mutant embryos (**Figure 3N**). Taken together, we concluded that Bs69 is required for the dorsal ventral patterning of zebrafish embryos, and functions as a negative regulator of Bmp signaling.

#### Bs69 Regulates Bmp Signaling by Association With Bmpr1a

Next, we investigated the mechanism by which Bs69 negatively regulates Bmp signaling. We hypothesized that Bs69 may regulate Bmp signaling through Bmpr1a in zebrafish. We therefore examined whether Bs69 interacts with Bmpr1a in vivo. We injected mRNAs encoding HA-Bs69 and Bmpr1a-FLAG into one-cell-stage bs69 mutant or wild type embryos, and performed co-immunofluorescence and co-immunoprecipitation assays for 7 hpf embryos. Co-immunofluorescence data clearly showed that HA-Bs69 was co-localized with Bmpr1a-FLAG, and coimmunoprecipitation data demonstrated that HA-Bs69 interacts with Bmpr1a-FLAG (**Figure 4A**). Importantly, HA-Bs69 interacts with Bmpr1a-FLAG in the cytoplasmic fraction of embryonic lysate (**Figure 4B**).

Next, we mapped the interacting domain between Bs69 and Bmpr1a. pCS2-HA-Bs69 or pCS2-HA-Bs69-Mynd and PCS2- Bmpr1a-kinase-FLAG were transiently co-transfected into 293T cells, and co-IP experiments were performed using HA and FLAG antibodies **Figure 4C**. The Mynd domain of Bs69 was sufficient to interact with the kinase domain of Bmpr1a (**Figure 4D**). In contrast, Bs691 lacking the Mynd domain did not immunoprecipitate with Bmpr1a, and Bmpr1a1 lacking the kinase domain did not immunoprecipitate with Bs69 (**Figures 4E,F**).

The association of Bs69 and Bmpr1a in physiological conditions strongly suggested that Bs69 modulates Bmpr1a activity. Like its mammalian counterpart, zebrafish Bs69 also has these three conserved chromatin reader domains (**Figure 2F**). It is possible that zebrafish Bs69 regulates Bmp signaling by modulating chromatin or functioning as a transcriptional co-factor. To investigate the significance of Bs69–Bmpr1a association for the regulation of Bmp signaling, we took advantage of the truncated form of Bs69 (BS69- Mynd) lacking all the chromatin reader domains but containing the intact Mynd domain that was still able to interact with Bmpr1a. By overexpressing Bs69-Mynd or HA-Bs691, we were able to determine whether Bs69 regulates Bmp signaling by physical association with Bmpr1a. To this end, we injected 100 pg mRNAs encoding HA-tagged Bs69-Mynd or HA-Bs691 into one-cell-stage wild-type embryos, and collected embryos at 7 hpf or 24 hpf for subsequent assays. HA-Bs69- Mynd was localized in both nuclei and cytoplasm which was similar to HA-Bs69 (**Supplementary Figures S2A–C**). HA-Bs69- Mynd overexpressing embryos displayed dorsalized phenotypes that were similar to HA-Bs69 overexpressing embryos at 1 dpf (**Supplementary Figures S3A,B**). In contrast, HA-Bs691 overexpression did not cause obvious dorsalization of the embryos (data not shown). Next, we investigated whether Bs69- Mynd could rescue the DV patterning phenotype of bs69 mutant embryos. Injection of 50 pg bs69-mynd or bs69 mRNA into one-cell-stage bs69 mutant embryos largely rescued the ventralized phenotypes at 1 dpf (**Supplementary Figure S3C**). In contrast, injection of 50 pg bs691 mRNA into onecell-stage bs69 mutant embryos had no effect on the DV patterning.

Phosphorylation of Smad1/5/8 at the C-terminal SXS motif by Bmp type I receptor is one of the most critical events in the transduction of Bmp signaling. We hypothesized that Bs69 may negatively regulate Bmp signaling through suppressing Bmpr1a activity by interfering its phosphorylation and activation of Smad1/5. If this is the case, loss of Bs69 function should cause increased Bmp signaling, indicated by elevated pSmad1/5 activity; whereas overexpressing Bs69 should cause decreased Bmp signaling, indicated by diminished pSmad1/5 activity. Indeed, Western blot analyses of Smad1/5 phosphorylation indeed supported this hypothesis (**Figures 3D,J**). Overexpressing HA-Bs691 had no obvious effect on the Smad1/5 phosphorylation (data not shown).

Taken together, we concluded that Bs69 negatively regulates Bmp signaling by physical association with Bmpr1a, which interferes with its phosphorylation and activation of Smad1/5.

### Mga Binding to Bs69 Disrupts the Bs69–Bmpr1a Interaction

Because both Mga and Bmpr1a interact with Bs69 through its Mynd domain, we hypothesized that Mga modulates Bmpr1a activity through Bs69. To explore this, we examined the relationship between Mga–Bs69 and Bmpr1a–Bs69 interactions. We performed competitive protein-binding assay. The amount of HA-Bs69 co-immunoprecipitated with Bmpr1a-FLAG became reduced by the increased addition of Mga (**Figure 5A**), and the amount of HA-Bs69 co-immunoprecipitated with Mga became reduced by the increased addition of Bmpr1a-FLAG (**Figure 5B**).

These data suggested that Mga and Bmpr1a compete for the binding to Bs69.

Since Mga and Bmpr1a compete for the binding to Bs69, we speculated that Mga functions to maintain or enhance Bmp signaling by antagonizing Bs69 in physiological conditions. We injected into one-cell-stage wild-type embryos with 50 pg mRNA encoding HA-Bs69 or mixture of mRNAs encoding both HA-Bs69 and mouse or zebrafish MGAs. Embryos injected with 50 pg beta-gal mRNA were used for controls. Phosphorylation of Smad1/5 was detected by Western immunoblotting of lysates from 8 hpf embryos. As seen in **Figure 5C**, the pSmad1/5 levels in HA-Bs69 overexpressing embryos were reduced compared with control embryos. The pSmad1/5 levels were restored and even enhanced by simultaneously expressing either zebrafish or mouse MGAs, but not by Mga1 that is unable to interact with Bs69 (**Figures 5C,D**). Accordingly, zebrafish or mouse MGAs rescued the loss of ventral tailfin phenotype in Bs69 overexpressing embryos at 1 dpf, supporting that Bmp signaling under control of Mga and bs69 is required for specifying the ventral tailfin cell fate (**Figure 5E**).

That Mga co-localizes and interacts with Bs69 in the cytoplasm strongly suggested that Mga localized in the cytoplasm functions to maintain Bmp signaling through Bs69–Bmpr1a axis. To test this hypothesis, we went on to generate completely nuclear or cytoplasmic version of Mga mutants. Using the NetNES or NLStradamus servers (la Cour et al., 2004; Nguyen et al., 2009), three putative nuclear localization sequences (NLSs) and two putative nuclear export sequences (NESs) were identified for Mga (**Figure 6A**). We injected mRNA encoding FLAG-Mga1NES or FLAG-Mga-Cter into one-cell stage mga mutant zebrafish embryos, and collected 7 hpf embryos for assays described below. Our immunofluorescence assay showed that FLAG-Mga1NES was strictly localized in the nuclei, whereas FLAG-Mga-Cter was found to be localized only in the cytoplasm (**Figures 6B,C**). These data indicated that we have successfully generated nuclear or cytosolic version of Mga protein. If it was Mga in the cytoplasm that regulates Bmp signaling through Bs69–Bmpr1a axis, then only cytosolic but not nuclear version of Mga protein could rescue the DV patterning defect of mga mutant. Indeed, we found that FLAG-Mga-Cter but not FLAG-Mga1NES rescued the reduced ventral tailfin phenotype of 2 dpf mga mutant embryos (**Figure 6D**). Consistently, Smad1/5 phosphorylation was increased in 7 hpf mga mutant embryos by overexpressing FLAG-Mga-Cter but not by FLAG-Mga1NES (**Figure 6E**). Finally, FLAG-Mga-Cter but not FLAG-Mga1NES rescued the loss of ventral tailfin phenotype in Bs69 overexpressing embryos (data not shown). Together, these data strongly indicated that the role of Mga in the regulation Bmp signaling is mainly acting in the cytoplasm separately from its role as a DNA binding protein.

To further determine how Mga–Bs69 interaction affects Bmp signaling, luciferase activity assays were performed with a Bmpresponsive luciferase reporter BRE-luc. C2C12 cells were cotransfected overnight with pCS2-Mga, PCS2-Bs69, and BRE-luc, followed by 12 h serum starvation, and treated with BMP4 or left untreated for 16 h. As seen from **Figure 6F**, BMP4 treatment remarkably stimulated the Bmp-responsive BRE-luc activity and Bs69 inhibited it. When Mga and Bs69 were co-expressed, Mga substantially antagonized the inhibitory effect of Bs69 on the luciferase activity.

If Mga regulates Bmp signaling through Bs69–Bmpr1a axis, embryos simultaneously depleted of both Mga and Bs69 should have similar DV patterning phenotype to bs69 mutant embryos. To test this hypotheses, we depleted Mga by injecting 4–5 ng mga morpholino (mgaMO) into one-cell stage of bs69 mutant embryos. 4–5 ng mgaMO was shown to cause a reduction of ventral tailfin phenotype in wild-type embryos (Sun et al., 2014).

We found that mgaMO had no obvious effect on DV patterning phenotype of 1 dpf bs69 mutants (**Supplementary Figure S4**).

Together, our data suggested that Mga antagonizes Bs69 to enhance the phosphorylation and activation of Smad1/5/8 both in vitro and in vivo.

Previous work showed that there are cardiac laterality defects in bmpr1a mutant embryos (Smith et al., 2011). If Mga regulates Bmp signaling through Bs69–Bmpr1a axis, mga mutant or Bs69 overexpressing embryos should exhibit similar laterality phenotypes as bmpr1a mutant embryos. We therefore

investigated whether mga mutant or Bs69 overexpressing embryos had the cardiac laterality defects. Whole mount in situ hybridization was performed to examine the expression of a set of laterality genes, including spaw, lefty2 and cmcl2. Embryos treated with 0.05 µM dorsomorphin or LDN193189 were used as positive controls. A small percentage of mga mutant or Bs69 overexpressing embryos indeed displayed the left-right patterning defects similar to bmpr1a mutants or dorsomorphin treated embryos (**Supplementary Figures S5A,B**) (Smith et al., 2011). Importantly, Mga or Mga-Cter but not Mga1

performed in technical triplicate and are representative of multiple experiments.

partially rescued the laterality defects of mga mutant or Bs69 overexpressing embryos (**Supplementary Figure S5C**).

## DISCUSSION

In this work, we demonstrated that Mga protein localized in the cytoplasm regulates Bmp signaling at least partly by physically interacting with and antagonizing Bs69. We provided genetic and biochemical evidence that Bs69 is a negative regulator of

Bmp signaling. The Mynd domain of Bs69 binds to the kinase domain of Bmpr1a which interferes with its phosphorylation and activation of Smad1/5. Mga binds to Bs69 and disrupts the Bs69– Bmpr1a interaction which allows Smad1/5 to be phosphorylated, and proper Bmp signaling to be maintained. Functionally, the Bmp signaling under control of Mga is important for specifying the ventral tailfin cell fate in zebrafish embryos (**Figure 6G**).

#### Bs69 Functions as a Negative Regulator of Bmp Signaling by Association With Bmpr1a

BRAM1, previously thought as an alternatively spliced product of BS69, was first identified by a yeast two hybrid screen using BMPR1A as a bait in human cells (Kurozumi et al., 1998). However, based on the analysis of genomic sequences and Bs69 gene product, Velasco et al. (2006) argued that the BRAM1 cDNA is in fact an artificial chimeric product between Anks1 and Bs69 sequences that happened from a recombination event during the cDNA library construction. Zebrafish Bram1 cDNA, isolated from a cDNA library by RACE technique, is merely a C-terminal part of the full length Bs69 cDNA, as the proposed zebrafish bram1 gene encodes a peptide lacks the featured MLLEPPSPVPW sequences like its mammalian counterparts. Moreover, we failed to amplify any Bram1-like cDNAs from our zebrafish cDNA libraries. Thus, we think that no Bram1 type exists in zebrafish. In this context, we are the first to dissect the developmental function of Bs69 using zebrafish as a model. In this work, by loss of function and overexpression assays, we demonstrated that zebrafish Bs69 is a negative regulator of Bmp signaling by physically interacting with Bmpr1a. The Mynd domain of Bs69 is indispensible for this function as it mediates Bs69 binding to the kinase domain of Bmpr1a.

The majority of studies so far proposed that BS69 function as co-factor for transcriptional or chromatin regulation in the nucleus (Hateboer et al., 1995; Guo et al., 2014; Wen et al., 2014). In recent years, however, a growing body of work have demonstrated that BS69 has important roles in the regulation of signaling pathways in the cytoplasmic membrane. For instances, BS69 was shown to interact with multiple transmembrane proteins, including LMP1, LTβR, and BMPR1A (Kurozumi et al., 1998; Liu et al., 2011; this study). BS69 is also constitutively co-localized in the membrane lipid rafts in mammalian cells (Wan et al., 2006). Lipid rafts were proposed to function in membrane protein sorting and in the formation of signaling complexes, as well as in endocytic trafficking (Hartung et al., 2006). It is possible that Bs69 is localized in membrane lipid rafts in zebrafish cells and may be involved in the endocytosis of Bmp receptors or the formation of ligandreceptor complex. In the future, it will be interesting to look into this possibility.

### Mga Interacts With Bs69 to Regulate Bmp Signaling

MAX giant associated protein is a transcriptional factor containing both T-box and bHLH domains, and was proposed to regulate the expression of both Max-network or T-box family genes (Hurlin et al., 1999). The Myc family of transcriptional factors are known for their role in the control of cell cycle progression, cellular growth and proliferation (Gallant, 2006). The T-box family of transcriptional factors play key role in mesendoderm formation in vertebrate embryogenesis (Papaioannou, 2014). Surprisingly, our mga mutant zebrafish are viable, and can grow up to adult without obvious morphological defects, except the loss or reduction of ventral tailfin. This fact indicates that Mga is not critically required for cell cycle progression and cell proliferation in zebrafish, and that Mga has limited function as the member of the T-box family of transcriptional factors. Alternatively, there are certain compensation mechanisms to allow normal zebrafish development in the absence of Mga. In this context, our work revealed a different requirement of MGA for zebrafish and mice embryonic development.

Our previous work suggested that Mga:Max could transcriptionally regulate bmp2b expression by binding to its promotor or enhancer in YSL in zebrafish. Now, we show that Mga localized in the cytoplasm directly interacts with and antagonizes Bs69 to modulate Bmpr1a-mediated Bmp signaling. This surprising role of Mga protein is executed in the cytoplasm, does not require dimerization with Max, and is independent of its transcriptional or chromatin remodeling activities. This is consistent with the observation that Mga is predominantly localized in the cytoplasm throughout zebrafish early embryogenesis. Indeed, mouse MGA is also predominantly localized in the cytoplasm (data not shown), suggesting that similar mechanism for the regulation of Bmp signaling occurs in mammals. We found that zebrafish Mga physically associates with Bs69 in the physiological conditions and this interaction is mediated by the PXLXP motif of Mga and the Mynd domain of Bs69. Mga-Cter binding to Bs69 disrupts the Bs69–Bmpr1a association which allows proper Bmp signaling to be maintained. By applying mga morpholino, we depleted Mga in bs69 mutant background. And we found that embryos depleted of both proteins had a similar DV patterning phenotype to bs69 mutant animals. This data further supported our hypotheses that Mga regulates Bmp signaling through Bs69–Bmpr1a axis.

Our previous work showed that Mga-Cter also binds to Smad4 and Smad1, two core components of Bmp signaling pathway (Sun et al., 2014). It is possible that Mga, Bs69, and Smad4 could form a triplex or Bs69 competes with Smad4 for binding to Mga. In any case, Mga binding to Bs69 could even enhance Bmp signaling by simultaneously antagonizing Bs69 and promoting the formation of Smad4: pSmad1/5/8 complex at the cytoplasmic membrane via releasing or bringing Mga bound Smad4 or Smad1 to Bmpr1a. This may explain why expressing both Bs69 and Mga had significantly stronger effect on Bmp activity than expressing Mga only (**Figures 5D**, **6F**). Altogether, these data indicate that Mga functions to control Bmp signaling pathways in different cellular compartments, at different levels and through different mechanisms. To our knowledge, this is the first report showing that certain member of the Myc or T-box family of transcriptional factors regulates signaling pathways in the cytoplasm. Nevertheless, mga mutant embryos at 1 dpf displayed mild DV patterning defects, suggesting that Mga acts to fine-tune Bmp signaling in zebrafish.

In addition to the regulation of Bmp signaling, both Bs69 and Mga are known chromatin readers and remodelers, raising an interesting question whether Mga and Bs69 could link Bmp signaling to chromatin structure regulation or transcriptional elongation (Velasco et al., 2006; Guo et al., 2014). Answering this question will help us to understand how signal transduction pathways directly communicate with chromatin to change the epigenetic landscape or gene expression. With mga and bs69 mutant zebrafish in hand, this question is under investigation in our lab.

#### ETHICS STATEMENT

fcell-06-00126 September 26, 2018 Time: 15:23 # 13

This study was carried out in accordance with the recommendations of ethical approval had been obtained from Animal ethical committee of Institute of hydrobiology for the approval of animal experiments. The protocol was approved by the "Animal Ethical Committee."

## AUTHOR CONTRIBUTIONS

YS and SD designed the experiments. XS, JC, YZ, and MM performed the experiments. YS wrote the manuscript.

#### FUNDING

This work was supported by National Key Research and Development Program (2016YFA0101100), National Natural Science Foundation of China (31671526), and Hundred-Talent Program (CAS and IHB) (Y623041501).

#### REFERENCES


### ACKNOWLEDGMENTS

We thank Dr. Amar Singh from Molecular Medicine Center of UGA for critical comments of the manuscript, Gang Ouyang and Huan Xiong for excellent technical support, and Prof. Xiao Wuhan and Prof. Cui Zongbin from Institute of Hydrobiology for providing BRE-luc and SBE-luc reporter plasmids.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2018.00126/ full#supplementary-material

FIGURE S1 | (A) The phenotypes of 1 dpf wild type embryos treated with 0.05 µM DMSO, or 0.05 µM dorsomorphin or 0.05 µM LDN193189 starting at one-cell stage, or injected with 50 pg dnBmpr1a mRNA at one-cell-stage. (B) 50 pg caBmpr1a mRNA rescued the loss of ventral tail fin phenotype of mga mutant embryos. Shown are representative embryos at 2 dpf. (C) WISH for 72 hpf embryos using mga, bs69, and bmpr1aa probes. mga and bs69 probes are DIG-labeled, whereas bmpr1aa probe is fluorescein-labeled.

FIGURE S2 | (A) The co-localization of Mga and HA-Bs69-Mynd in 7 hpf embryos. (B) The co-localization of zebrafish Mga and HA-Bs69 in 293T cells. (C) The co-localization of zebrafish Mga and HA-Bs69-Mynd in 293T cells.

FIGURE S3 | (A) Dorsalized phenotypes of HA-Bs69-Mynd overexpressing embryos at 2 dpf. C1-4 classification according to DV patterning index. (B) Western blot analysis of lysates from 8 hpf HA-Bs69-Mynd overexpressing or control embryos. (C) Tail region of 2 dpf bs69 mutant embryos injected at one-cell-stage with 50 pg mRNAs encoding HA-Bs69-Mynd or β-Gal.

FIGURE S4 | (A) DV patterning phenotypes of bs69−/<sup>−</sup> mutant embryos at 1 dpf injected with 4 ng mgamisMO or mgaMO. V1-2 classification according to DV patterning index. (B) Quantification of (A) based on three independent experiments.

FIGURE S5 | (A) The cardiac laterality defects of mga mutant or Bs69 overexpressing embryos at 1 dpf revealed by WISH using spaw, lefty2, and cmlc2 probes. (B) Percentage of embryos that exhibited cardiac laterality defects. L, left; M, middle; R, right. (C) 50 pg mouse mga mRNA partially rescued the cardiac laterality defects of mga mutants at 1 dpf; 50 pg caBmpr1a mRNA partially rescued the cardiac laterality defects of Bs69 overexpressing embryos at 1 dpf.

Gallant, P. (2006). Myc/Max/Mad in invertebrates: the evolution of the max network. Curr. Top. Microbiol. Immunol. 302, 235–253.


progenitor cells in zebrafish. PLoS One 6:e27403. doi: 10.1371/journal.pone. 0027403


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

Copyright © 2018 Sun, Chen, Zhang, Munisha, Dougan and Sun. 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.

fcell-06-00126 September 26, 2018 Time: 15:23 # 14

## Analyzing Neuronal Mitochondria in vivo Using Fluorescent Reporters in Zebrafish

#### Amrita Mandal, Katherine Pinter and Catherine M. Drerup\*

*Unit on Neuronal Cell Biology, NICHD, National Institutes of Health, Bethesda, MD, United States*

Despite their importance for cellular viability, the actual life history and properties of mitochondria in neurons are still unclear. These organelles are distributed throughout the entirety of the neuron and serve many functions, including: energy production (ATP), iron homeostasis and processing, calcium buffering, and metabolite production, as well as many other lesser known activities. Given their importance, understanding how these organelles are positioned and how their health and function is maintained is critical for many aspects of cell biology. This is best illustrated by the diverse disease literature which demonstrates that abnormal mitochondrial movement, localization, size, or function often correlates with neural pathology. In the following methods article, we will describe the techniques and tools we have optimized to directly visualize mitochondria and analyze mitochondrial lifetime, health, and function in neurons *in vivo* using fluorescent reporters in the zebrafish. The zebrafish system is ideal for *in vivo* studies of mitochondrial biology as: (1) neuronal circuits develop rapidly, within days; (2) it is genetically accessible; and (3) embryos and larvae are translucent allowing imaging in a completely intact vertebrate nervous system. Using these tools and techniques, the field is poised to answer questions of mitochondrial biology in the context of neuronal health and function in normal and disease states.

Keywords: mitochondria, neuron, dynein, kinesin, zebrafish, axonal transport

#### INTRODUCTION

Mitochondria and the eukaryotic cells in which they reside are considered symbiotes. This now critical organelle is thought to have originally entered into existence as a bacteria that took up residence in another cell, likely a eukaryote. From that time until now, co-evolution of mitochondria and the cells of plants and animals have given rise to a situation in which each depends upon the other for survival. During this co-evolution, mitochondria have taken on many responsibilities in the cell. In addition to their well-known role in Adenosine Tri-Phosphate (ATP) synthesis, mitochondria function to buffer and store calcium, produce essential metabolites, synthesize signaling molecules, and regulate iron and iron processing. Despite these essential functions, we have only a minimal understanding of mitochondrial life history in cells. In particular, our knowledge of mitochondrial properties in neurons is specifically lacking. We will describe the tools developed by our lab and others that utilize zebrafish to study mitochondrial health, function, movement, and turnover in vivo, in order to understand mitochondrial life history in neurons.

#### Edited by:

*Gokhan Dalgin, University of Chicago, United States*

#### Reviewed by:

*Ross F. Collery, Medical College of Wisconsin, United States Rachel Elizabeth Moore, King's College London, United Kingdom*

> \*Correspondence: *Catherine M. Drerup katie.drerup@nih.gov*

#### Specialty section:

*This article was submitted to Molecular Medicine, a section of the journal Frontiers in Cell and Developmental Biology*

> Received: *19 July 2018* Accepted: *08 October 2018* Published: *25 October 2018*

#### Citation:

*Mandal A, Pinter K and Drerup CM (2018) Analyzing Neuronal Mitochondria in vivo Using Fluorescent Reporters in Zebrafish. Front. Cell Dev. Biol. 6:144. doi: 10.3389/fcell.2018.00144*

#### Mitochondrial Structure and Function

Mitochondria are double membrane-bound organelles. The canonical structure, starting from the outside, includes an outer membrane, inner membrane space, inner membrane and matrix. The curvature of the inner membrane creates the observed cristae. The most well-characterized function of mitochondria is producing ATP, the primary source of energy for cellular activities. This process occurs within the mitochondrial cristae. In neurons, the demand for ATP is incredibly high due to their high metabolic rate even at resting states. It is estimated that average activity in rodent gray matter, for example, uses ∼30 µmol ATP/g/min (Attwell and Laughlin, 2001). Since an average rat brain is ∼2 g, this means that every hour a rat brain is burning 3.6 mmol of ATP.

The generation of ATP in mitochondria occurs through the conversion of the pyruvate generated from glycolysis to acetyl CoA (the first step of the Kreb cycle; localized to the mitochondrial matrix). Acetyl CoA is then used to generate the carbon dioxide necessary to produce NADH/FADH2, the substrates of oxidative phosphorylation, i.e., ATP production. For this process to proceed, an oxidative potential must be maintained across the inner membrane. To create an oxidative potential, the inner membrane space is maintained at high hydrogen ion concentrations compared to the matrix, this creates a pH and charge gradient within mitochondria. Maintenance of mitochondrial matrix potential is tightly linked to mitochondrial health and several methods have been developed to monitor this aspect of mitochondrial biology. The most common method is the cationic dye tetramethylrhodamine ethyl ester (TMRE). This vital dye accumulates in the mitochondrial matrix due to its negative charge. Consequently, the fluorescence intensity of the mitochondrial TMRE is a readout of matrix potential and health. Though commonly used in cultured cells, the ability to measure matrix potential in vivo has, to date, been lacking. The maintenance of mitochondrial matrix potential is accomplished by a complex array of proteins intricately arranged in specific compartments within the organelle. Of particular importance are the respiratory chain complexes I-IV. These complexes maintain the proton and pH gradients across the inner membrane that are critical to ultimately power ATP synthase, generating ATP. Interestingly, a subset of the main components of the electron transport chain in mitochondria are encoded by the mitochondrial DNA, while the rest of the Complex I-IV components and the majority of the proteins that form and maintain the mitochondria are synthesized externally from nuclear DNA.

In addition to generating ATP, mitochondria serve as a calcium buffer and reservoir in cells. Calcium enters mitochondria through the largely unselective Voltage Dependent-Anion Channel (VDAC) (Gincel et al., 2001). Once in the inner membrane space, calcium is transported to the mitochondrial matrix by the mitochondrial calcium uniporter (MCU). For calcium release back to the cytoplasm, mitochondria utilize both sodium-dependent and independent calcium channels. Mitochondrial calcium levels are highly regulated and thought to rely on the cytoplasmic calcium concentration, in addition to other signals (Kirichok et al., 2004). Consequently, high calcium levels in the local microenvironment can result in rapid uptake of calcium into the mitochondrial matrix. In the mitochondria, calcium levels regulate mitochondrial functions as well as signaling molecules associated with cell death and cell survival (reviewed in Pivovarova and Andrews, 2010). Thus, calcium levels must be tightly regulated in mitochondria to ensure organelle function and cell viability are maintained.

Calcium regulation by mitochondria is especially critical in neurons. High cytosolic calcium levels have been linked to axonal and neural degeneration. This is likely due to the fact that mitochondria harbor cell death genes whose release regulates apoptosis. When calcium levels remain elevated in this organelle, mitochondria release these proteins and induce apoptosis, leading to loss of neural tissue. While high calcium levels on a prolonged basis lead to cell death, regulated elevation of calcium levels in mitochondria stimulates ATP synthesis (McCormack and Denton, 1989). Transient increases in cytoplasmic and consequently mitochondrial calcium levels commonly occur in neurons, particularly at synapses. Action potentials triggered by circuit activity lead to the activation of presynaptic calcium channels and calcium influx. Following calcium store release, this ion must be rapidly removed from the cytosol to regulate synaptic release. Both the endoplasmic reticulum (ER) and the mitochondria have been proposed to serve as intracellular stores that can rapidly buffer calcium after neuronal activity, though their relative contribution is still a source of active research and debate. It is likely that these organelles actually function in concert to regulate calcium ion levels as they are tightly coupled at regions of ER-contact sights shown to influence mitochondrial activity and signaling (Boehning et al., 2004; Cárdenas et al., 2010; Raturi et al., 2016). However, the role of mitochondria in calcium homeostasis has been difficult to address in vivo. We have optimized approaches to use genetically encoded calcium indicators (GECIs) to assay cytoplasmic and mitochondrial calcium levels in vivo. GECIs are commonly used to measure transient increases in intracellular calcium in neurons as an indicator of neuronal activity. These indicators typically consist of a calcium binding domain fused to one or two fluorescent proteins. The binding of calcium changes the fluorescence intensity of the signal. Common GECIs include GCaMP variants and GECOs. Particularly useful are the GECO color variants, including the red indicator R-GECO1, which has been used previously to study hair cell responses to mechanical stimulation (Maeda et al., 2014). The combination of calcium indicators with different spectral properties allows monitoring of whole cell calcium levels and subcellular compartments simultaneously as described below. As calcium serves to regulate neuronal activity, neuronal maintenance, and has a critical role in regulating mitochondrial movement, understanding the dynamics of mitochondrial-cytoplasmic calcium flux in vivo is of significant importance.

Finally, mitochondria can also act as signaling centers in cells. During the process of oxidative phosphorylation, mitochondria produce metabolites including NADH, FADH, succinate, reactive oxygen species (ROS), and many others. These molecules, sometimes thought of as byproducts, have been shown to induce cellular responses (Chandel et al., 1998; Sena and Chandel, 2012; Weinberg and Chandel, 2015; Weinberg et al., 2015). For instance, evidence in cancer cell lines has demonstrated that mitochondrial metabolites can in fact signal to regulate the growth and movement of these cells, making this organelle a target for cancer therapy (Wang et al., 2011; Weinberg and Chandel, 2015). In addition, work in neurons has demonstrated that mitochondrial positioning can regulate the localization of axon branching (Courchet et al., 2013; Spillane et al., 2013). While this could be attributed to the higher levels ATP in the microenvironment surrounding this organelle, it is entirely possible that the signaling molecules produced locally could regulate subcellular dynamics necessary for axon branching to initiate as well. With the incredible potential of mitochondrial metabolites to influence the local microenvironment, the maintenance and regulation of mitochondrial health and positioning within the neuron is of obvious importance. To date, it has been difficult to measure mitochondrial metabolite production in vivo. We have developed a protocol to use transient transgenic animals that express an indicator of chronic ROS production in neurons. This protein, Timer, is oxidation sensitive, switching its fluorescence spectrum based on oxidation state (Hernandez et al., 2013; Laker et al., 2014). With this new tool, we have the ability to assay chronic ROS in various neuronal compartments in normal states and with manipulation.

#### Mitochondrial Dynamics

Mitochondria are not static, but are rather quite dynamic organelles. One activity of particular importance for mitochondrial maintenance is the active interchange of mitochondrial components, known as mitochondrial dynamics. The term mitochondrial dynamics describes the fusion events that bring two mitochondria together as well as the fission events which produce two daughter mitochondria from a single parental organelle. Work on mitochondrial dynamics has shown a clear role for fusion in the maintenance of mitochondrial health. Studies in which mitochondrial fusion has been disrupted have shown that this leads to loss of mitochondrial DNA and subsequent mitophagy (Chen et al., 2005, 2010; Chen and Chan, 2009). Several mitochondrial proteins have been identified as necessary for mitochondrial fusion, including Mitofusin and OPA1 (Optic Atrophy 1; Alexander et al., 2000; Delettre et al., 2000; Zuchner et al., 2004). Loss of either protein results in highly fragmented mitochondria, degradation of these organelles, and subsequent neuronal cell death. The necessity of fusion is thought to be due to in its ability to replenish proteins and mitochondrial DNA quality and quantity, maintaining the organelle's health and function. Of similar importance is the process of mitochondrial fission. During fission, receptors on the mitochondrial outer membrane, such as Fis1 and Mff, recruit the dynamin-related protein Drp1 to mitochondria. Drp1 oligomerizes, resulting in a constricted point that then cleaves to generate two independent organelles. While the process of fission is clearly important for the maintenance of a healthy mitochondrial pool (Parone et al., 2008; Twig et al., 2008; Ban-Ishihara et al., 2013), the underlying mechanistic reasons for its necessity, particularly in neurons, are somewhat unknown. A subset of literature suggests that fission may be a way for the "unhealthy" parts of the organelle to be targeted for degradation. Evidence to support this includes the lower membrane potential observed in one of the two daughter mitochondria following division; however, the fate of the daughter mitochondria after fission is not clear. Therefore, the actual function of mitochondrial fission in the maintenance of a healthy mitochondrial pool in neurons is still an active area of investigation.

### Mitochondrial Transport

A process intricately related to mitochondrial dynamics is mitochondrial transport. Active transport of this organelle occurs in all cell types, but neurons are perhaps most keenly sensitive to the precise regulation of this process due to their large size and high metabolic demand. Neurons, unlike many cell types, have large processes which can stretch up to a meter away from their cell bodies in humans, as is the case for a subset of axons that make up the sciatic nerve. This means that their axon length to cell body diameter ratio is roughly 100,000 to 1. In order to form and maintain this enormous structure, proteins and organelles must be actively transported through the cell body and processes using molecular motors and their microtubule tracks. These tracks are unipolar in axons, with their plus or fast-growing end situated toward axon terminals. The directionality of these tracks governs the direction of movement by microtubule-based motor proteins. Anterograde axonal transport (away from the cell body/toward microtubule plus ends) utilizes a superfamily of Kinesin motors proteins (Pilling et al., 2006). Conversely, retrograde axonal transport (cell body-directed/toward microtubule minus ends) is primarily accomplished by a single motor protein complex, Cytoplasmic Dynein (Schnapp and Reese, 1989). The mechanics and regulatory mechanisms that control how and when these motors attach to and move mitochondria through axons is tightly regulated and a subject of intense investigation.

Anterograde mitochondrial movement is largely accomplished by the Kinesin-1 molecular motor. To bind to mitochondria, Kinesin-1 elicits the help of adaptor proteins Miro (a Rho GTPase) and Milton (also known as Trak1/2). Originally discovered in Drosophila, these proteins are essential for movement of mitochondria from the neuronal cell body into axons via anterograde transport (Stowers et al., 2002; Glater et al., 2006; Russo et al., 2009). A key discovery in the regulation of mitochondrial movement came due to the study of the structure of Miro in particular. Miro is a small Rho GTPase that contains EF hand structural domains. These EF hand domains bind calcium which elicits a conformational change in the protein that changes the propensity for the Kinesin motor to interact with microtubules, which is necessary for processive mitochondrial movement (Wang and Schwarz, 2009). It is likely that Miro and Milton have a role in retrograde transport of mitochondria as well as loss of either effects transport in both directions. In support of this, elevated calcium levels stop all mitochondrial movement (Russo et al., 2009; van Spronsen et al., 2013). Therefore, an understanding of how direction of mitochondrial transport is actively controlled is still lacking.

A step toward understanding how mitochondria move in either the anterograde or retrograde direction came recently as a result of a forward genetic screen in zebrafish. This genetic approach identified the protein Actr10 as essential for mitochondrial attachment to the retrograde motor protein complex (Drerup et al., 2017). Actr10 is a member of the dynein accessory complex dynactin. It is situated in the so called pointedend complex of dynactin, in an ideal place for cargo binding. Loss of this protein in zebrafish caused mitochondrial accumulation at microtubule plus ends due to failed mitochondrial retrograde transport (Drerup et al., 2017). Anterograde mitochondrial transport is not affected by loss of Actr10, further suggesting the importance of this protein specifically for regulation of retrograde mitochondrial movement. Below we describe the tools we have developed that allowed us to characterize the retrograde transport defect in this line. We also present the subsequent technologies we have adapted for use in this system which are allowing us to define the ultimate function of retrograde mitochondrial transport in neurons.

## METHODOLOGY AND RESULTS

### Using Zebrafish to Discover the Mechanisms and Function of Retrograde Mitochondrial Transport

Zebrafish larvae are an ideal system to identify and characterize the mechanisms that regulate axonal transport in vivo, as we and others have shown (Drerup and Nechiporuk, 2013, 2016; O'Donnell et al., 2013; Campbell et al., 2014; Paquet et al., 2014; Drerup et al., 2017). Embryos develop rapidly with primary neural circuits developed by 4 days post-fertilization (dpf). Zebrafish are also translucent through these stages, allowing visualization of cellular and sub-cellular phenomena in vivo. In addition, zebrafish are a genetically tractable system: they are amenable to forward and reverse genetic screens, transient and stable transgenic animals are easily generated, and a wealth of transgenic and mutant lines are available through community databases. Finally, zebrafish have well-characterized neural circuits for analysis of the functional implications of transport disruption. One well-characterized circuit is the afferent axons of the posterior lateral line (pLL) mechanosensory system.

The pLL is a sensory system in aquatic vertebrates that allows the animal to detect movement in the water around them (Dijkgraaf, 1963). This system is composed of clusters of sensory hair cells situated in neuromasts that are distributed along the flanks and throughout the head. The hair cells themselves have apical protrusions that bend in response to water movement. The mechanical force due to this bending activates hair cells and ultimately leads to the release of glutamate at the synaptic contact they make with the pLL afferent axon terminals (Obholzer et al., 2008). For our purposes, the pLL axons are ideal for studies of both the molecular regulation of mitochondrial transport and the impact that disruption of this process has on the function of the circuit. The pLL forms in the first several days of development with initial axon extension completed by 2 dpf (days postfertilization) and functional circuits by 4 dpf (**Figure 1**; Metcalfe, 1985; Metcalfe et al., 1985). The axons of the pLL nerve extend from the cell bodies of the pLL ganglion, located behind the ear, and to the tip of the tail in the longest cases. These axons are just under the skin and are largely planar, making visualization easy and reliable. Finally, unlike other sensory neurons, the axons of pLL neurons have stereotyped projections and welldefined axon terminals which allows rapid detection of any abnormalities in structure and function (Faucherre et al., 2009; Sarrazin et al., 2010; Dow et al., 2018). Because of the ease of visualization, mitochondrial transport can be imaged in vivo without removing neurons from their natural environment. In this way, neurons, myelinating glia, pre- and post-synaptic components, extracellular structural proteins and molecules, and growth factor support, as well as other more amorphous factors found in the surrounding tissue can be studied in an intact system. As neural circuit activity, growth factors, and myelination have all been proposed to effect mitochondrial transport (Chada and Hollenbeck, 2004; Kiryu-Seo et al., 2010; Ohno et al., 2011), the zebrafish pLL axons are highly suited for understanding the cellular mechanisms that regulate the movement of this organelle in vivo.

A critical question in the field of mitochondrial transport is the relevance of retrograde movement to mitochondrial health and neural circuit function. While anterograde mitochondrial transport is necessary to populate the axon and facilitates axon outgrowth (Morris and Hollenbeck, 1993; Han et al., 2016; Zhou et al., 2016), the precise function of retrograde mitochondrial transport is less clear. Evidence from cultured mammalian neurons has demonstrated that mitochondria with lower oxidative potential, thought to be a sign of failing health, are more likely to be transported in the retrograde direction (Miller and Sheetz, 2004; Lin et al., 2017). This work has been used to suggest that the main role of mitochondrial retrograde transport is to remove damaged organelles from the axon; however, there are conflicting reports that indicate that oxidative potential may not influence direction of mitochondrial movement (Suzuki et al., 2018). In addition to understanding the exact relevance of mitochondrial retrograde transport, there are several unexplored questions that are fundamental to axonal mitochondrial transport, health, lifetime, and function. First, how often do mitochondria move in axons? Second, does movement change with development? Third, does mitochondrial health and function differ between different regions of the cell? Finally, do mitochondria serve to regulate calcium levels differentially in subcellular compartments? Below, we will describe the tools we have developed to address these questions.

### Analyzing Mitochondrial Transport in the pLL

Two primary systems exist to analyze mitochondrial movement in axons in vivo in the zebrafish (O'Donnell et al., 2013; Paquet et al., 2014; Drerup and Nechiporuk, 2016; Drerup et al., 2017). Both use sensory axons and mark these organelles with fluorescent proteins to then monitor their active movement. The method developed independently by the Sagasti and Misgeld groups utilizes a GAL4:UAS system to drive expression of mitochondrial markers in axons of Rohon-Beard cells, a type of cutaneous sensory neuron in the early embryo (O'Donnell

deposits approximately five nascent neuromasts which will develop into the sensory organs of the pLL. As it migrates, the pLL primordium guides the axons of the pLL nerve to the tail. The axons stretch from the pLL ganglion (green; contains pLL cell bodies) which remains behind the ear (asterisk) to the terminal cluster neuromasts in the tail in the longest of cases. Images modified from Kimmel et al. (1995). Permissions to use these images were obtained from the copyright holder, John Wiley & Sons, Inc. (B) By 5 days post-fertilization (dpf; bottom) the zebrafish larva is freely swimming and feeding. This image is of a live transgenic zebrafish larvae at 5 dpf which carries the stable *TgBAC(neurod:egfp)nl*<sup>1</sup> transgene (hereafter referred to as *neurod:egfp*) that labels all pLL neurons with cytoplasmic green fluorescent protein (GFP). This larva is also a *5kbneurod:mito-TagRFP* transient transgenic, which expresses mitochondrially-localized TagRFP (shown in magenta) in the mitochondrial inner membrane space in neurons. Transient transgenics are chimeras, with only a subset of neurons expressing this construct. (B′ ) Transient expression of *mito-TagRFP* in a single pLL ganglion neuron (outlined) and the axon extending from it (arrowheads). (B′′ ) Mosaic expression allows imaging of a mitochondrial movement in a single axon of the pLL nerve. Yellow arrowhead points to a mitochondria moving in the retrograde direction. Red arrow indicates an anterogradely moving mitochondria (see Movie 1). (B′′′ ) This single axon forms a basket-like axon terminal that innervates half of the hair cells in this neuromast. Arrowheads point to individual mitochondria in the terminal innervating sensory cells which are not shown. (C) Screenshot of an imaging analysis session in MetaMorph. The dotted lines outline the axonal region analyzed. (D) Kymograph of this imaging session. Mitochondrial movement is represented by the sloped black lines. Vertical lines show pauses in movement. (D′ ) Schematic of this kymograph with representative anterograde movement bouts (blue lines) and retrograde movement bouts (pink) traced.

et al., 2013; Paquet et al., 2014). Because these axons arborize across the skin, mitochondrial localization and movement can be easily imaged in single axons in vivo. However, the precise synaptic targets of these cells are less clear and systems are not in place to monitor synaptic activity, complicating analyses of circuit structure and function. As an alternative system, we have optimized the pLL sensory system for interrogation of mitochondrial transport regulation and function in zebrafish axons. As described above, this system has stereotyped synapses and analysis of circuit function is routine (Zhang et al., 2016). Finally, several reagents, as outlined below, have been generated that allow targeting of mitochondria in single axons in the pLL nerve, allowing analysis of mitochondrial movement in vivo.

To visualize mitochondrial transport, we have optimized the system that we used previously to analyze lysosome and dynein movement in pLL axons (Drerup and Nechiporuk, 2013, 2016; Drerup et al., 2017). In this system, zebrafish zygotes can be injected with a plasmid encoding a mitochondrial targeting sequence derived from Cytochrome C Oxidase fused to an open reading frame encoding a fluorescent reporter. Expression is driven in neurons using a five kilobase portion of the neurod promotor (5kbneurod; Mo and Nicolson, 2011). As zebrafish embryos and larvae develop they exhibit mosaic expression of the reporter in pLL neurons (**Figure 1B**). At the stage of interest, the zebrafish are screened, and those with expression of the mitochondrial marker in one to two neurons of the pLL ganglion are selected for analysis. Individuals are then mounted on a coverslip in 1.5% low melt agarose and imaged at high magnification on a confocal microscope (e.g., a 63X, NA1.2 objective with a 100 × 100µm field of view on an LSM800 microscope). To capture mitochondrial dynamics, imaging is done in a single z-plane, in a 30–100µm length of axon, three to five times a second, as per Nyquist sampling requirements. The region imaged for analysis is chosen based on the ability to follow a single organelle through the region of interest, ensuring the visualized axon (∼1µm diameter) is kept in a single z plane (±0.44µm; **Figure 1C**; **Movie 1**). Axonal transport of mitochondria can be imaged in various regions of the expressing pLL axon using this set up, with a consistent area chosen between related experiments. Typically, this imaging is done in the middle portion of the trunk, between neuromasts two and four (see **Figure 1B**), near the end of the yolk sac extension after 2 dpf. For imaging during axon extension, we image a region approximately two thirds of the distance between the neuronal cell body and growth cone of the extending axon. Mitochondrial transport distance, velocity, and direction can then be assayed directly from these image sequences (Drerup and Nechiporuk, 2016; Drerup et al., 2017).

Mitochondrial transport parameters are measured using kymograph analysis in MetaMorph (BioVision). The transport session is analyzed to ensure that individual organelles can be tracked through the length of the region of interest analyzed and the axon region to be analyzed is identified in the program (**Figure 1C**). Kymographs of the fluorescent signal are then generated using MetaMorph (**Figure 1D**). These line traces represent the organelles imaged with longer organelles generating thicker traces. Each individual movement bout, represented by a slanted line that can be followed from beginning to end, is traced (**Figure 1D**′ ). The distance and time of a single movement bout are identified as the change in X and Y, respectively, on the associated kymograph. The slope then translates to the velocity of movement. Finally, the number of mitochondria moving or stationary is manually counted. While kymographs can be used to estimate these values, we have found that manual assessment is more consistent.

Using this general imaging scheme, we have analyzed mitochondrial transport parameters across development in zebrafish embryos and larvae. After injection of the mitochondrial reporter construct, embryos were raised to 30 hpf, 2, 4, or 5 dpf prior to mounting and imaging. These time-points were chosen specifically to match up with critical developmental stages in the pLL circuit. At 30 hpf, pLL axons are mid-extension and have active growth cone dynamics (Metcalfe, 1985). By 48 hpf, extension of the longest "pioneer" axons of the pLL is complete, but pLL axon terminals synapses are not yet formed with target hair cells (Metcalfe et al., 1985). By 4 dpf the pLL neural circuit is complete. At pLL axon terminals, there are synapses formed with hair cells and active neurotransmission can be observed between pLL axons and their hair cell targets (Zhang et al., 2018). Finally, at 5 dpf myelinating glia are present on the pLL nerve and have wrapped the axons, though compression of the myelin sheath is not yet complete (Monk et al., 2009). When we analyzed mitochondrial transport parameters across these developmental stages we found that the distance and time of anterograde and retrograde movement was maintained at a steady state from 30 hpf to 4 dpf. At 5 dpf there were slight but significant increases in both the time and distance of anterograde and retrograde movement bouts compared to other time-points (**Figures 2A,B**; ANOVA with post-hoc contrasts). At all developmental stages, the velocity of mitochondrial movement remained constant at ∼0.83 µm/s and ∼1.14 µm/s for anterograde and retrograde movement, respectively (**Figure 2C**). These values fall within the large window of reported mitochondrial axonal transport velocities among various systems (MacAskill and Kittler, 2010) and align closely with velocities observed in vivo in mouse sciatic nerve studies (Misgeld et al., 2007). The underlying reason for the enhanced bi-directional transport at 5 dpf is not clear but could rely on the establishment of stabilized microtubules, the addition of microtubule modifying proteins, or perhaps the expression of adaptor proteins necessary for long-distance transport.

When we focused our developmental analysis on the direction of mitochondrial movement, among all stages, we found no significant change in the proportion of mitochondria that were moving in the anterograde or retrograde direction. Additionally, during development, there was no change in the proportion of mitochondria that were stationary in these axons (**Figure 2D**; ANOVA with post-hoc contrasts). This is somewhat contradictory to literature analyzing mitochondrial transport in cultured neurons where days in vitro correlated with decreased mitochondrial movement (Kang et al., 2008). The precise reason for these differences is not entirely clear but could be due to a number of reasons. One possibility is the nature of the larval zebrafish. While the primary pLL is fully developed by 5 dpf, the fish continues to grow. This could mean that the axons are not in an entirely mature state by this time-point and do not express

all of the docking proteins required to decrease mitochondrial movement (Kang et al., 2008). Another factor that could account for the consistent mitochondrial population dynamics in the pLL axons is the in vivo nature of this analysis. Previous work on mitochondrial transport frequency has largely been done on in vitro culture systems or in the sciatic nerve of a mouse after removal of surrounding tissues. It is possible that disruption of the neuron's in vivo environment alters intracellular transport dynamics, particularly over long periods of time. Future work on adult neurons in zebrafish could allow us to differentiate between these possibilities but technically this is not feasible in the short-term.

#### Measuring Mitochondrial Lifetime in Axon Terminals

Mitochondria can be long-lived organelles, undergoing dynamic rearrangement to sustain themselves and their viability. Work in vitro in cultured Drosophila and chick neurons as well as in vivo imaging of mitochondria in the exposed sciatic nerve of mice has shown that these organelles largely exist in two pools; a stationary pool comprised of longer mitochondria and a smaller, actively moving mitochondrial pool (Miller and Sheetz, 2004; Misgeld et al., 2007; Narayanareddy et al., 2014). This is recapitulated in the zebrafish pLL axons, where stationary mitochondria (∼3.5µm) are longer than motile mitochondria (<2µm; **Figure 2E**). This demonstrates that there is consistency between experimental systems. Overall, analyses of mitochondrial transport dynamics have given us a substantial amount of information about the acute nature of mitochondrial movement. Unfortunately the majority of experiments analyzing stationary verses mobile mitochondrial pool have largely been done on time periods that span minutes. Therefore, little is known regarding how the properties of individual mitochondria within axons change over longer periods of time. In addition is not clear whether there are subtypes of mitochondria restricted within the cell body, axon, or axon terminal. To begin to address these questions, we have developed a stable transgenic line to express a photoconvertible protein, mEos (McKinney et al., 2009), in neuronal mitochondria [Tg(5kbneurod:mitomEos)y586]. mEos was originally discovered in coral and is known for its ability to be stably photoconverted from green to red using 405 nm illumination. It is widely used in developmental biology to label independent cells to determine the origin of various tissues and is now used to follow the persistence of intracellular structures as well (Lam et al., 2015). We have utilized it similarly to track the localization and persistence of mitochondria in neuronal compartments. For this work, we engineered a transgenic fish in which mitochondrially targeted mEos (mitomEos) is expressed in neurons using the 5kbneurod promotor (Mo and Nicolson, 2011) and the mitochondrial targeting sequence from Cytochrome C Oxidase (Fang et al., 2012). We have previously used this signal sequence to transiently tag mitochondria with red fluorescent protein successfully (Drerup et al., 2017 and **Figure 1**). Analysis of our transgenic line indicates that mitochondria are efficiently tagged with mEos in this line and can be visualized in pLL axon terminals (**Figure 3A**).

In order to analyze mitochondrial lifetime in the pLL sensory axons, we examined larvae at 4 dpf, after pLL axons have formed stable synapses. For conversion and analysis, larvae were mounted in 0.8% low melt agarose for imaging with a 40X, NA1.0 dipping objective on a LSM800 confocal microscope (Zeiss). Preconversion, mitochondria were easily visualized with 488 nm excitation (green), indicating unconverted mEos. After short stimulation with 405 nm light, mitochondria were now visualized with 568 nm excitation (red), confirming mEos photoconversion (**Figures 3A,B**). To examine mitochondrial lifetime, imaging was done immediately after photoconversion and again 24 h later in axon terminals. Analysis of red to green fluorescence intensity revealed that immediately after photoconversion the mEos signal was dominated by red fluorescence (**Figure 3B**). In contrast, after 24 h there was an almost complete loss of the converted (red) mEos relative to green in axon terminals, indicating that there are high levels of mitochondrial turnover at this site (**Figures 3A,B**). Next, we analyzed the timeline of mitochondrial turnover using time-lapse imaging following photoconversion at 4 dpf. With both 10 and 3 min intervals, we found relative old (red) to new (green) mitochondria in axon terminals plateaued by ∼3 h post-conversion (**Figures 3C,D**). This data implies a rapid turnover of mitochondria, within hours, in axon terminals.

Turnover could be due to either converted mitochondria moving out of the terminals, mitophagy in the axon terminal, or, perhaps, new (unconverted) mitochondria moving into the terminal, increasing the "green" signal. To differentiate between these possibilities, we analyzed the change in mean green and red fluorescence intensity immediately after photoconversion and compared it to the intensities after 24 h. We found that 24 h after photoconversion, there was an increase in green mEos labeled mitochondria, indicating ample addition of new organelles at the axon terminal. Simultaneously, there was a strong decrease in mean red fluorescence intensity, bringing the values back down to pre-conversion levels, representing loss of all if not almost all photoconverted organelles by 24 h post-conversion (**Figure 3E**). We then wanted to more precisely determine the temporal dynamics of mitochondrial gain and loss from axon terminals. For this analysis, we used timelapse imaging again to monitor the green and red mEos intensity at 3 min intervals immediately after the photoconversion described above. Strikingly, there was a sharp increase in green mEos within 3 h after photoconversion with a concomitant decrease in red fluorescence intensity (**Figure 3F**). This indicates both addition of new mitochondria and loss of old occurs within this time period and both contribute to the ratiometric changes shown in **Figure 3**. The lower fold change in the red channel in **Figure 3F** is likely due to incomplete photoconversion (see **Figure 3A**′ ) Together, our data support rapid turnover of mitochondria in pLL axon terminals, with addition of new organelles (green) and loss of photoconverted (red/old) organelles on the span of hours. This high level of mitochondrial loss from axon terminals was surprising as previous studies in cultured neurons demonstrated large stationary pools of mitochondria, anticipated to remain in place for extended periods of time (Kang et al., 2008). Our data argues instead that these organelles are very dynamic in the axon in vivo, with rapid exchange over a few hours. One still open question is the ultimate fate of the axonal mitochondria lost from axon terminals. Subsequent work using this line will address the relative contribution of mitophagy verses mitochondrial transport in the turnover of mitochondrial axon terminal populations. Additionally, we plan to explore the relative lifetime of mitochondria in other regions of the neurons, including the cell body, to identify any changes in mitochondrial population dynamics related to cellular compartment. Together, this work will shed light on the lifetime and turnover of mitochondria in neuronal compartments in vivo.

### Measuring Mitochondrial Health and Productivity in vivo

Mitochondrial health is often assessed by analyzing the potential/pH across the inner membrane. Indeed, if the mitochondrial machinery that maintains this hydrogen gradient is not maintained, mitochondria fail to function normally and are subject to degradation. The most common way to analyze mitochondrial membrane potential is the vital dye TMRE

FIGURE 3 | Analysis of mitochondrial lifetime in axons. (A) Expression of mitochondrially-localized mEos in the posterior lateral line of a 4 dpf *5kbneurod:mito-mEos* transgenic larvae. For our analysis we used *5kbneurod:mito-mEos* homozygotes to maximize mitochondria-localized mEos expression. This approach provides strong labeling of mitochondria in axon terminals. (A′ ) Illumination with 405 nm light converts mEos from green to red (shown in magenta). (A′′ ) Twenty-four hours post-conversion, minimal converted (red) mEos remains in the axon terminal. (B) The relative converted to unconverted mito-mEos illustrates an almost complete loss of converted mitochondria from axon terminals by 24 h post-conversion (ANOVA with *post-hoc* contrasts; \*\*\**p* < 0.0001; *n* = 15). (C,D) Time-lapse analysis of red to green *mito-mEos* in axon terminals after photoconversion shows a rapid turnover of mitochondria from the axon terminal, with most converted (red) mitochondria lost by 3 h post-conversion (*n* = 1 for C; *n* = 2 for D). (E) Quantification of the mean unconverted (green) and converted (red) mito-mEos in axon terminals shows both addition of new mitochondria and loss of converted mitochondria over a 24 h period (ANOVA with *post-hoc* contrasts; \*\*\**p* < 0.0001; \**p* < 0.01; *n* = 15). (F) Time-lapse analysis of green and red mEos mean fluorescence intensity (normalized to post-conversion values) immediately after photoconversion in two larvae (L) demonstrates increased green and decreased red mEos over time. Timelapse data from two larvae normalized to post-conversion values. Scale bar = 10µm.

(tetramethylrhodamine ethyl ester). This positively charged dye is highly attracted to the negative charge in the mitochondrial matrix, making it a marker of "healthy" mitochondria. We adapted a TMRE protocol used previously in the zebrafish pLL hair cells (Esterberg et al., 2014) for use in pLL sensory axons. For this experiment, we incubated zebrafish larvae at 4 dpf in 25µM TMRE in embryo media with 0.1% DMSO (dimethyl sulfoxide) for an hour in the dark. Larvae were subsequently washed three times in embryo media prior to being mounted in 1.5% low melt agarose and imaged with a 63X, NA1.4 objective on a confocal microscope (Zeiss LSM800). For analysis, mitochondrial TMRE was measured in axon terminals and the pLL ganglion after subtraction of nonneural tissue using the ImageJ Image Calculator function. This revealed a consistent elevation in TMRE mean fluorescence in axon terminals compared to cell body mitochondria (**Figures 4A,B,E**). As TMRE recruitment is proportional to negative change in the matrix, this could mean that axon terminal mitochondria have a stronger oxidative

*neurod:egfp* transgene expression). (C,D) Transient transgenesis using a *5kbneurod:mito-Timer* construct allows imaging of the oxidation-sensitive protein Timer in individual pLL neurons and axon terminals in larval zebrafish at 4 dpf. Green Timer is native while red Timer (shown in magenta) is oxidized. (E) Quantification of TMRE fluorescence intensity shows elevated TMRE labeling in axon terminals compared to cell bodies, an indicator of higher matrix potential (ANOVA; *n* = 7 larvae). (F) The red:green Timer ratio is slightly, but not significantly reduced in axon terminals (ANOVA; *n* = 16 larvae). Scale bar = 10µm.

gradient compared to those in the cell body. Alternatively, dye permeability or light scatter during imaging due to differences in tissue thickness could underlie this difference between pLL ganglion cell bodies and axon terminals. Therefore, we attempted to address the question of mitochodondrial health in another way, using the genetically encoded sensor, Timer (Laker et al., 2014).

Timer is a protein that fluoresces in the green spectrum when in its native conformation but switches to red upon oxidation (Laker et al., 2014). Therefore, the red (568 nm excitation) to green (488 nm excitation) ratio of Timer can be used to assess ROS production, an indicator of mitochondrial health. Excess ROS production is an indicator of oxidative stress and a failing mitochondrion prone to degradation. To express Timer in mitochondria, we used the mitochondrial targeting sequence described above to target it to the inner membrane space and expression was again driven using the minimal 5kbneurod promotor sequence. Similar to what we described above, injection of the 5kbneurod:mito-Timer construct resulted in mosaic expression of Timer in pLL neurons. Injected larvae were raised to 4 dpf, and we selected larvae with expression of Timer in a subset of pLL ganglion neurons. These larvae were mounted for imaging as described for TMRE. This analysis revealed no significant difference in the red to green Timer ratio in axon terminals verses the cell body; however, there was a trend toward lower red to green ratios, a potential indicator of lower oxidative stress, in axons terminals (**Figures 4C,D,F**; ANOVA p = 0.16). One important point to clarify, however, is that Timer does not just become oxidized based on elevated ROS in unhealthy mitochondria. ROS is a natural biproduct of oxidative phosphorylation. As mitochondrially localized Timer ages, it will naturally become oxidized as well, making the relative contribution of age verses increased oxidative stress difficult to fully distinguish without secondary methods, such as TMRE (Hernandez et al., 2013; Laker et al., 2014). Therefore, it is important to take the results of our mitochondrial turnover assays, TMRE staining, and mito-Timer analyses together. In combination, our work suggests that mitochondria in axon

terminals are rapidly turned over, have a higher potential across their inner membrane, and perhaps have lower ROS production than those in the cell body. This, in turn, would allow axon terminal mitochondria to have increased ATP production relative to those in the cell body.

The next question we wanted to address was in regard to the actual productivity of mitochondria in the cell body and axon terminal of pLL sensory axons. For this, we used the ATP:ADP dual ratiometric sensor PercevalHR (Tantama et al., 2013). PercevalHR was developed as an improved version of the original Perceval: a chimeric protein, composed of the ATP-binding pocket of the bacterial protein GlnK1 and a mutated form of the yellow fluorescent protein mVenus (Berg et al., 2009). This protein can be competitively bound by both ADP and ATP with differential excitation peaks (420 vs. 500 nm, respectively) making it a ratiometric sensor. We made a DNA construct to transiently express PercevalHR in neurons, with a p2a cleavable peptide sequence linking it to mCherry-CAAX (membrane localized red fluorescent protein) to visualize the neuron. This 5kbneurod:PercevalHRp2amCherry-CAAX construct, when injected into zebrafish zygotes, results in mosaic neuronal expression of the sensor in larval zebrafish. Quantification of the ATP:ADP ratio demonstrated consistent ATP:ADP ratios in axon terminals and the cell body of these neurons (**Figure 5**). Due to the somewhat variable nature of this transient transgenic approach, statistical comparisons between the cell body and axon terminals were challenging; however, as this sensor is ratiometric, our data imply that mitochondria in

these distinct cellular compartments are adequately converting ADP to ATP. The generation of a stable transgenic line expressing this sensor in neurons, which is currently in progress, will allow comparative studies on mitochondrial productivity in various cellular compartments in the near future.

### Mitochondrial Calcium Buffering in Neurons

The mitochondrial outer and inner membranes have calcium channels to rapidly take up this ion from the microenvironment surrounding this organelle. It has been proposed that this ability to locally modulate or buffer calcium could be of great importance to neurons, particularly at synapses (reviewed in Devine and Kittler, 2018). Synaptic activity requires active release of calcium from intracellular pools to facilitate neurotransmitter release. After activity, the calcium must then be sequestered to modulate the activity of the synapse. One method of calcium sequestration would be through mitochondrial uptake. In addition to calcium buffering, calcium entry into the mitochondria is also thought to stimulate mitochondrial productivity with elevated calcium levels increasing ATP synthesis; however, the mitochondrial calcium load must be properly regulated as prolonged elevation of mitochondrial calcium stimulates the release of pro-apoptotic factors and initiates cell death (reviewed in Strasser et al., 2000). Therefore, proper regulation of calcium by and in mitochondria is absolutely essential for mitochondrial and neuronal health.

To begin to study the relationship between cytoplasmic and mitochondrial calcium in neurons, we engineered two new transgenic lines. These lines are stable transgenics which express a green calcium indicator [G-GECO1; Tg(5kbneurod:G-GECO)nl19] in neuronal cytoplasm and a red calcium indicator [R-GECO1; Tg(5kbneurod:mito-R-GECO)nl20] in the mitochondrial inner membrane space (Zhao et al., 2011). Using confocal imaging as described for TMRE analysis, we can image chronic and acute changes in calcium ion abundance in and around mitochondria in these lines. This analysis revealed a striking difference between the neuronal cell body and axon terminals of the pLL sensory neurons. The mitochondria (red) to cytoplasmic (green) GECO ratio is increased in axon terminals compared to the cell body, indicating a higher calcium load in mitochondria compared to the surrounding environment in this compartment of the neuron (**Figure 6**). Together with the increased TMRE/lower Timer signal and rapid turnover, our data indicate that the axon terminal has an extremely dynamic population of mitochondria, that may be necessary to support synaptic activity and signal transduction.

#### REFERENCES


In summary, our work has developed a collection of novel tools to interrogate the lifetime, health, productivity and function of mitochondria in neurons in vivo in the embryonic and larval zebrafish pLL sensory system. Our data to date illustrate the importance of considering all compartments when analyzing mitochondrial activity and transport, particularly the differences between organelles residing in the cell body verses those in the synaptically active regions of the axon terminal. Future work using these and other tools developed by the community will allow us to come to a better understanding of the symbiotic relationship between this former prokaryote and the eurkaryotic cell it calls home.

#### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of NICHD Animal Care and Use Committee. The protocol (ASP 15-039) was approved by the NICHD ACUC.

### AUTHOR CONTRIBUTIONS

KP and AM contributed experimental design, data generation, and data analysis to this work. CD designed, conducted, and analyzed experiments and wrote the manuscript.

#### FUNDING

Funding for this work from the National Institute of Child Health & Human Development.

#### ACKNOWLEDGMENTS

We would like to thank Wesley Schnapp and Dane Kawano for thoughtful discussions on this work and Dr. Katie Kindt for feedback on this manuscript. We would also like to acknowledge Dr. Alex Nechiporuk who provided thoughtful comments on the development of this project and support during its initial stages.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcell.2018. 00144/full#supplementary-material

Movie 1 | Mitochondrial transport in a pLL axon. Mitochondria (labeled with TagRFP) are visualized moving in the anterograde (left to right) and retrograde (right to left) direction in a single pLL axon (see Figure 1). Scale bar = 10µm.

Flow. Metab. 21, 1133–1145. doi: 10.1097/00004647-20011000 0-00001


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

Copyright © 2018 Mandal, Pinter and Drerup. 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.

# Expanding the CRISPR Toolbox in Zebrafish for Studying Development and Disease

Kaili Liu, Cassidy Petree, Teresa Requena, Pratishtha Varshney and Gaurav K. Varshney\*

Functional and Chemical Genomics Research Program, Oklahoma Medical Research Foundation, Oklahoma City, OK, United States

The study of model organisms has revolutionized our understanding of the mechanisms underlying normal development, adult homeostasis, and human disease. Much of what we know about gene function in model organisms (and its application to humans) has come from gene knockouts: the ability to show analogous phenotypes upon gene inactivation in animal models. The zebrafish (Danio rerio) has become a popular model organism for many reasons, including the fact that it is amenable to various forms of genetic manipulation. The RNA-guided CRISPR/Cas9-mediated targeted mutagenesis approaches have provided powerful tools to manipulate the genome toward developing new disease models and understanding the pathophysiology of human diseases. CRISPR-based approaches are being used for the generation of both knockout and knock-in alleles, and also for applications including transcriptional modulation, epigenome editing, live imaging of the genome, and lineage tracing. Currently, substantial effort is being made to improve the specificity of Cas9, and to expand the target coverage of the Cas9 enzymes. Novel types of naturally occurring CRISPR systems [Cas12a (Cpf1); engineered variants of Cas9, such as xCas9 and SpCas9-NG], are being studied and applied to genome editing. Since the majority of pathogenic mutations are single point mutations, development of base editors to convert C:G to T:A or A:T to G:C has further strengthened the CRISPR toolbox. In this review, we provide an overview of the increasing number of novel CRISPR-based tools and approaches, including lineage tracing and base editing.

#### Edited by:

Gokhan Dalgin, The University of Chicago, United States

#### Reviewed by:

Mingyu Li, Xiamen University, China Máté Varga, Eötvös Loránd University, Hungary

> \*Correspondence: Gaurav K. Varshney gaurav-varshney@omrf.org

#### Specialty section:

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

> Received: 22 November 2018 Accepted: 24 January 2019 Published: 04 March 2019

#### Citation:

Liu K, Petree C, Requena T, Varshney P and Varshney GK (2019) Expanding the CRISPR Toolbox in Zebrafish for Studying Development and Disease. Front. Cell Dev. Biol. 7:13. doi: 10.3389/fcell.2019.00013 Keywords: Cas12a (Cpf1), lineage tracing, base editors, zebrafish, disease models, CRISPR/Cas9

#### INTRODUCTION

Information gained from the study of model organisms is essential to our understanding of human development and disease. Replication of a mutant phenotype in a gene knockout (inactivation of a gene in an animal model) is considered to be the gold standard approach to support candidate gene predictions in humans. Zebrafish (Danio rerio) is uniquely suited to this approach and has become one of the fastest growing model organisms, useful for both basic and translational research (Bradford et al., 2017).

Zebrafish are an attractive alternative to mouse models because they give rise to a large number of progeny and are amenable to high-throughput mutagenesis and drug screening approaches (Kettleborough et al., 2013; Varshney et al., 2013; Varshney and Burgess, 2014; Gallardo et al., 2015). In addition, zebrafish fertilization is external, and their transparent larvae can be monitored throughout embryogenesis, providing unique accessibility to embryonic lethal mutations.

The process of gene targeting in zebrafish is not as laborious as it is in mice, and the maintenance costs are 10X cheaper per animal (Varshney and Burgess, 2014). Importantly, zebrafish overcome an emerging technical issue in modeling disease pathology: many of the diseases studied today are multigenic, so disruption of a single gene may not produce a disease phenotype in any model system. However, combining genetic variants is straightforward in zebrafish, making it an ideal organism in which to model the functional consequences of multiple mutations. In addition, complementation studies in fish are relatively simple and allow for the direct testing of specific variants (not just knockouts) in a vertebrate system. The utility of zebrafish was further increased upon completion of the zebrafish genome-sequencing project (Howe et al., 2013); zebrafish and mammalian genes are highly conserved, and 70% of human disease genes have an ortholog in zebrafish (Howe et al., 2013). In zebrafish, many large-scale forward and reverse genetic screens have been performed using random mutagenesis methods - ENU or insertional mutagenesis (retroviral, Tol2, DS) - and the number of different genetic and molecular tools rapidly increased once the genome was sequenced (Amsterdam et al., 2011; Marquart et al., 2015; Quach et al., 2015; Seiler et al., 2015; Vrljicak et al., 2016). For decades, targeted gene knockouts were not possible in zebrafish, and its utility for validation studies of candidate genes was limited. This challenge was recently eliminated with the development of novel gene targeting approaches including ZFNs, TALENs, and CRISPR/Cas9 (Bedell et al., 2012; Jinek et al., 2012; Mali et al., 2013; Hsu et al., 2014; Varshney et al., 2015b); other techniques such as Structure Guided Nucleases (SGNs) have been shown to work for gene targeting in zebrafish but have not been adopted widely (Varshney and Burgess, 2016; Xu et al., 2016). With the transformative CRISPR/Cas9 approach, it is now possible to target any number of genes in an efficient and high-throughput manner (Gagnon et al., 2014; Shah et al., 2015; Varshney et al., 2015a). It is also possible to target multiple genes simultaneously; given that part of the zebrafish genome is duplicated, it is very useful in targeting 2 paralogs simultaneously (Jao et al., 2013).

CRISPR/Cas9 and other enzymes are not only being used to generate knockouts, introduce specific changes in the genome and repair mutant alleles, but are also being repurposed in other applications including transcriptional regulation, in vivo chromatin imaging, epigenome modulation, genome-wide knockout screens, etc. There are many reviews discussing the use of CRISPR-based approaches and the various technological developments in zebrafish (Varshney et al., 2015b; Li et al., 2016; Demarest and Brooks-Kayal, 2018). In this review we will focus on the latest development in expanding targeting coverage of gene targeting, base editing, transcriptional regulation, epigenome modulation, and lineage tracing.

### CRISPR-MEDIATED TARGETED MUTAGENESIS IN ZEBRAFISH

In 2012, a joint team from Jennifer Doudna and Emmanuel Charpentier's lab, and an independent team from Virginijus Siksnys's lab demonstrated that Cas9 from Streptococcus pyogenes or Streptococcus thermophilus together with CRISPR RNA (crRNA) can be guided to a target site to cleave DNA in vitro. Early the following year, George Church and Feng Zhang's labs utilized Cas9 from Streptococcus pyogenes and/or Streptococcus thermophilus to edit the genome in mammalian cells: they showed that single guide RNA (sgRNAs) can direct Cas9 to the target site to induce a double stranded break, which can then be repaired by either the non-homologous end joining (NHEJ) or homologydirected repair (HDR) pathways. An alternative repair pathway, microhomology-mediated end joining (aka Alt-EJ) (MMEJ; an error-prone repair mechanism that uses microhomologous sequences 5–25 bp in length) has also been shown to be activated by the double-stranded break induced by Cas9 (McVey and Lee, 2008; Ata et al., 2018). In the last 5–6 years, CRISPR-based genome editing tools have been used for many applications in a variety of cells, organisms and plants (Hsu et al., 2014). The use of simple and programmable CRISPR/Cas9 technology has completely transformed reverse genetics in zebrafish.

Zebrafish was the first vertebrate model used to demonstrate that CRISPR/Cas9 can efficiently edit the genome in vivo (Hwang et al., 2013) with up to 50% targeting efficiency. Another report demonstrated that CRISPR/Cas9 can be used to generate biallelic mutations in gata5 and etsrp, and the observed phenotypes in injected embryos can phenocopy genetic mutants (Chang et al., 2013). Using a codon optimized version of Cas9 with nuclear localization signals, Jao et al. (2013) showed that Cas9 can efficiently induce biallelic mutations when Cas9 mRNA and sgRNA are injected into one-cell stage embryos. The authors further showed that up to five genes can be targeted simultaneously, and all showed phenotypes associated with each gene (Jao et al., 2013). It is evident from these initial reports that CRISPR/Cas9 is so efficient at inducing biallelic mutations that it allows for the generation of phenotypes in injected embryos similar to antisense morpholinos. Several strategies have been used to screen for phenotypes in the F<sup>0</sup> generation in injected embryos; one such strategy used multiplexing to target multiple genes simultaneously and screen for phenotypes in F0. This approach was used to screen 48 genes and identify two novel genes involved in electrical-synapse formation (Shah et al., 2015). A similar strategy used a pool of four sgRNAs together with Cas9 protein to identify transcriptional regulators in cardiomyocytes; 50 candidate genes were screened and the role of zbtb16a in cardiac development was identified (Wu et al., 2018). Burger et al. (2016) demonstrated that the use of in vitro assembled Cas9 mCherry or EGFP fusion protein, and sgRNA together as a ribonucleoprotein complex can provide a visual readout for efficient microinjections for the analysis of mutant phenotypes in F<sup>0</sup> generation. These mutants were termed CRISPR-mediated mutants or crispants (akin to morphants; Burger et al., 2016). While these approaches to screen candidate genes by analyzing the expected phenotypes in injected embryos are efficient, in most cases a stable mutant is required for phenotypic analysis of gene function. Data from Shawn Burgess's lab targeting 89 genes show that genetic mutants can be generated with ∼28% germline transmission rates at a 99% success rate. This high germline transmission rate is four–fivefold higher than

that of other targeting approaches such as ZFNs, and TALENs (Varshney et al., 2015a).

Many groups have developed a streamlined workflow for generating mutants using CRISPR/Cas9 in a high-throughput manner (Gagnon et al., 2014; Varshney et al., 2015a, 2016a). The Burgess Lab addressed a few challenges in developing this workflow: First they developed a strategy to synthesize sgRNA by annealing two oligonucleotides that served as a template for in vitro transcription; this allowed for the synthesis of sgRNA in few hours with relatively low cost and is similar to the strategy was used by Gagnon et al. (2014). Secondly, the zebrafish genome is highly polymorphic, and it was predicted that this might cause multiple mismatches in the target sequence and prevent the sgRNA from binding efficiently. To address this, they sequenced the genome of the NHGRI-1 lab strain and identified more than 14 million variants. This data is available through UCSC genome browser track; while designing sgRNAs or PCR primers, variant regions of the genome can be avoided to maximize the success rate (LaFave et al., 2014). The third challenge they encountered was the identification of mutant alleles in a high-throughput manner. Several methods are currently used for the identification of mutants in zebrafish including DNA mismatch nuclease assays (Chang et al., 2013; Jao et al., 2013), restriction fragment length polymorphism (Hruscha et al., 2013) and sequencing (Gagnon et al., 2014; Varshney et al., 2015a; Burger et al., 2016), but none are amenable to high-throughput application. A method to determine the size of amplicons by fluorescent PCR was optimized to identify indels (Sood et al., 2013). This method uses three primers (gene-specific forward and reverse primers and a FAM-labeled primer) to amplify the regions around the target sites and resulting fluorescently labeled amplicons are mixed with a size standard (e.g., Rox400) to determine the amplicon size on ABI sequencing platform. This method can be applied in a highthroughput manner, and has resolution up to 1 bp (**Figure 1**) (Carrington et al., 2015; Varshney et al., 2015a).

Adopting CRISPR/Cas9 technology in a high-throughput manner for targeted mutagenesis has enabled geneticists to screen for large-numbers of genes with relatively modest resources, and generate disease models for ever-increasing candidate disease genes. The approaches have been widely applied: Pei et al. (2018) screened more than 200 candidate genes to identify genes involved in hair cell regeneration, and screens related to retinal regeneration or degeneration (Unal Eroglu et al., 2018) and several disease models including Niemman-Pick disease type C1;(Tseng et al., 2018), hearing disorders (Varshney et al., 2015a), congenital sideroblastic anemia with immunodeficiency, fevers and developmental delay (SIFD; (Giannelou et al., 2018), Mucolipidosis type IV (Li et al., 2017) have been performed. In zebrafish, phenotypes are generally studied in the F<sup>2</sup> generation where homozygous embryos are generated by breeding two heterozygous (F1) lines over ∼6–7 months. It has been shown that phenotypes can be screened in the F<sup>1</sup> generation in a non-Mendelian manner by inbreeding two founders, thus eliminating a generation and saving time (Varshney et al., 2015a). This could be an important strategy that may speed up the phenotypic screening of a large number of candidate disease genes. To generate knockouts, Cas9 is transiently expressed ubiquitously in one-cell stage embryos thus generating global knockouts, two independent studies have shown that Cas9 can also be expressed in a tissue-specific manner thus it is possible to inactivate genes in a specific tissues (Ablain et al., 2015; Yin et al., 2015). Yin et al. (2015) demonstrated that by using a heat-shock inducible and tissue specific promoters, the expression of Cas9 can be controlled both temporally and spatially [reviewed in (Li et al., 2016)]. They further characterized five U6 promoters to drive the expression of multiple guides thus adopting this approach for multiplex genome editing (Yin et al., 2015).

While CRISPR/Cas9 is an effective and simple tool for genetic manipulations, there are several concerns over its specificity as it has been shown to bind and edit unintended targets (e.g., Off targets) including inducing large deletions (Adikusuma et al., 2018). In zebrafish genetics, off-targets can easily be outcrossed away and a genotype-phenotype linkage must be established thus losing on-target activity by using Cas9 variants to achieve specificity should be considered. There are fewer studies in zebrafish that have tested the off-target effect in zebrafish. One such study detected off-target mutagenesis in only 1/25 off target sites in germline, another study showed off-target mutation rates from 1.1 to 2.5% (Hruscha et al., 2013; Varshney et al., 2015a). Many Cas9 variants such as Cas9-HF1, eSpCas9, evoCas9, HypaCas9, and others have been developed to increase the specificity of the Cas9 enzyme and thus reduce the off targets, however, these variant might also affect the on-target activities (Jamal et al., 2018).

We have summarized important CRISPR-based genome editing tools in **Table 1**. While SpCas9 can target multiple sites in the coding regions of the genome that is GC-rich, efforts are being made to expand target coverage by employing either orthologous Cas9 or evolving SpCas9 to identify different Protoacceptor Adjacent Motif (PAM) sequences.

#### ENGINEERED AND NOVEL NUCLEASES TO EXPAND THE TARGETING COVERAGE

Cas9 from Streptococcus pyogenes (spCas9) is the most popular and effective genome editing tool, and the sequences recognized by SpCas9 are limited by the specific and simple PAM (50 -NGG-3<sup>0</sup> ) requirement (Jinek et al., 2012). However, this specific PAM sequence may not be available near the target of interest. To expand the targeting coverage, researchers have identified additional, naturally occurring CRISPR nucleases that may have different PAM requirements. Additionally, spCas9 was engineered to recognize other PAM sequences, expanding the targeting coverage and allowing them to be used in orthogonal applications. These newly identified CRISPR nucleases may also address the challenge of delivering the large size of spCas9 (1,368 aa) as they may be smaller; they may also provide a homology template for in vivo therapeutic applications (Mout et al., 2017).

#### Orthologous Cas9

Many of the smaller-sized Cas9 nucleases discovered in different species can recognize different PAM sequences and varied lengths

of target sequences for in vivo genome editing (**Table 2**): the Staphylococcus aureus Cas9 (SaCas9, 1053 aa) is not only small in size, but also uses a different complex PAM (NNGRRT; (Muller et al., 2016). Many other Cas9 nucleases from different bacterial species are being used for in vivo genome editing: Cas9 from Neisseria meningitidis (NmCas9) requires NNAGAAW PAM; Cas9 from Streptococcus thermophilus (St1Cas9, 1121 aa and St3Cas9, 1388 aa) require NNAGAAW and NGGNG PAMs, respectively; Cas9 nuclease from Campylobacter jejuni (CjCas9, 984 aa) recognizes a 22-nt target sequence with NNNVRYAC and NNNNRYAC PAM (Kim E. et al., 2017).

### Engineered Cas9 Variants

Most of the orthologous Cas9 nucleases have long and complex PAM requirements that will limit the targeting range because they will occur less frequently in genomes. An alternative strategy to expand PAM specificity would be to engineer the SpCas9 to recognize other PAMs. Kleinstiver et al. (2015b) engineered SpCas9 based on the crystal structure of the enzyme, and the mutated Cas9 was tested for its ability to recognize different PAM sites. Engineered SpCas9 variants VRER (D1135V/G1218R/R1335E/T1337R) recognizing NGCG PAM, VQR (D1135V/R1335Q/T1337R) recognizing NGAN or NGNG PAM, and EQR (D1135E/R1335Q/T1337R) variants recognizing NGAG PAM were generated. All of these SpCas9 variants were able to target sequences that were not targetable by wild-type SpCas9 in human cells, but only the VQR variant was able to target sites with NGAG PAMs (20–43% efficiency in zebrafish; (Kleinstiver et al., 2015b). The efficiency of the VQR was further validated by showing its ability to target tyr and EGFP loci with 50 and 70% efficiency, respectively. Zebrafish codon-optimized versions of VQR and EQR SpCas9 generated by Shawn Burgess' lab are also available from Addgene (Varshney et al., 2016a).


TABLE 1 | Commonly used tools for CRISPR-mediated genome editing.

As described above, the majority of the Cas9 orthologs or variants have complex PAM requirements, and the frequency of these targets in the genome is limited. To circumvent this challenge, David Liu's lab used phage-assisted continuous evolution (PACE) to isolate 14 evolved SpCas9 variants (xCas9 3.0–3.13); one such variant (xCas9 3.7) was able to recognize a broad range of PAM, including NG, NNG, CAA, GAT, and GAA (Hu et al., 2018). The xCas9 3.7 variant was able to cleave multiple PAMs at much higher frequency than wildtype SpCas9: GAA and GAT PAM showed ∼5-fold, NGT ∼4.5-fold and NGC 2.1-fold efficiencies. Another variant, xCas9 3.6, showed the second-best editing efficiencies at fewer PAMs (Hu et al., 2018). SpCas9 was further engineered to generate a variant called SpCas9-NG that has a relaxed preference for the third nucleobases in the NGG PAM (Nishimasu et al., 2018). This variant had seven residues mutated (R1335V/L1111R/D1135V/G1218R/E1219F/A1322R/T1337R) in SpCas9; was capable of cleaving NGA, NGT, and NGG PAMs with more than 20% editing efficiency; and showed lower activity at NGC PAM. A comparison of editing efficiencies showed that spCas9-NG had higher editing efficiencies at NGA, NGT, and NGG sites, and xCas9 failed to edit NGC PAM targets (Nishimasu et al., 2018).

Similarly, Staphylococcus aureus Cas9 (SaCas9) was also modified using a molecular evolution strategy to recognize NNNRRT PAMs. This variant of SaCas9 is known as KKH SaCas9 (variant E782K/N968K, R1015H) and can further increase the SaCas9 targeting range by two–fourfold (Kleinstiver et al., 2015a). The KKH SaCas9 variant was able to recognize five independent targets in different genes with 10–90% efficiency, thus expanding targeting coverage further in zebrafish (Feng et al., 2016).

The Cas9 nuclease from Francisella novocida (FnCas9) is one of the largest nucleases identified thus far (1629 aa) and recognizes NGG PAM similar to SpCas9, but has failed to generate indels in mammalian cells. It is possible that microinjecting mouse zygotes with FnCas9 protein and a guide RNA ribonucleoprotein complex may induce target-specific indels; a variant of FnCas9 (E1369R/E1449H/R1556A) called RHA FnCas9 could recognize YG PAM (Hirano et al., 2016).

Recently, a homolog of SpCas9 in Streptococcus macacae (SmacCas9) has been described to recognize the 5<sup>0</sup> -NAAN-3<sup>0</sup> PAM. A variant of SmacCas9 (iSpy-macCas9) was engineered to maintain its specificity for adenine dinucleotide PAM while showing higher genome editing efficiency in vivo (Jakimo et al., 2018). An orthologous Cas9 protein from Streptococcus canis (ScCas9) with 89.2% sequence similarity to wild-type SpCas9 has also been characterized and shown to prefer a more minimal NNG (Chatterjee et al., 2018). An engineered version of ScCas9 (1Loop 1KQ) not only cleaves NGG PAM but also recognizes and edits NNGA PAM at a comparable rate, but it edits other NNGN PAMs with reduced efficiency. All of these engineered and orthologous Cas9 proteins have significantly expanded the targeted coverage.

### CRISPR/Cas12a (Cpf1)

The majority of class 2 and type II nucleases and their engineered versions described earlier have preference for GC-rich PAMs that limits the targeting of AT-rich sequences, for example most of the non-coding genome in zebrafish is AT-rich (Howe et al., 2013). Another class 2 and type V family of nucleases, originally described as Cpf1 and later renamed Cas12a (Shmakov et al., 2017), was discovered as an alternative effective genome-editing tool (Zetsche et al., 2015). Cas12a is different from SpCas9 in many ways (**Figure 2**): (i) Cas12a recognizes T-rich PAM located at the 5<sup>0</sup> end of the target DNA sequence, (ii) Cas12a is guided by a single CRISPR RNA (crRNA) that is shorter than that of SpCas9 and does not require trans-acting crRNA (tracrRNA), (iii) Cas12a TABLE 2 | Summary of Cas orthologs and variants.

fcell-07-00013 March 1, 2019 Time: 11:26 # 6


uses an ∼23 nt target sequence, (iv) Cas12a induces a doublestranded break in the target sequence via a staggered cut and ∼18 nt distal to PAM, generating a 4–5 nt 5<sup>0</sup> overhang, and (v) Cas12a has both DNAse and RNase activity; therefore it is capable of processing its own CRISPR array (Fonfara et al., 2016). As of now, 32 Cas12a orthologs have been described, and their genomeediting potential was screened. Cas12a from Francisella novicida (FnCas12a), Acidaminococcus sp. BV3L6 (AsCas12a), and Lachnospiraceae bacterial ND2006 (LbCas12a) exhibited robust editing in human cells, plants and many other model organisms, including zebrafish (Zetsche et al., 2015). Additionally, four other Cas12a orthologs [Thiomicrospira sp. Xs5 (TsCas12a), Moraxella bovoculi AAX08\_00205 (Mb2Cas12a), Moraxella bovoculi AAX11\_00205 (Mb3Cas12a), and Butyrivibrio sp. NC3005 (BsCas12a)] have been shown to induce indels in human cells, although only Mb3Cas12a was able to induce indels at a

rate comparable to AsCas12a and LbCas12a (Zetsche et al., 2017). AsCas12a and LbCas12a use TTTV PAM, while FnCas12a and Mb3Cas12a recognize the less-restrictive TTN and NTTN PAMs, respectively. FnCas12a also has been show to target sequences with KYTV PAM preference in mammalian cells (Tu et al., 2017). The Cas12a nucleases were further engineered by introducing mutations S542R/K607R and S542R/K548V/N552R to generate AsCas12aRR and AsCas12RVR variants, and G532R/K592R and G5323R/K538V/Y542R to generate LbCas12a RR or RVR variants, which can recognize non-canonical PAMs such as TYCV, TWTV, and TATV PAMs. Use of Cas12a in editing the zebrafish genome is not as straightforward as editing using SpCas9. Cas12a mRNA and crRNA targeting the tyr locus do not induce any indels in zebrafish at optimal temperature (28◦C) (Watkins-Chow et al., 2017). Further optimization revealed that Cas12a crRNAs are degraded rapidly after injection in onecell stage embryos; however, LbCas12a-crRNA ribonucleoprotein (RNP) complex can protect crRNA from degradation and can efficiently induce indels at a rate comparable to that of SpCas9 in zebrafish (Moreno-Mateos et al., 2017). LbCas12a is more effective in inducing indels than AsCas12a, and AsCas12a activity is temperature dependent in zebrafish. Heat shocking embryos after injection for 4 h at 34◦C significantly increased the mutagenic activities for AsCas12a and LbCas12a nucleases. (Moreno-Mateos et al., 2017) LbCas12a has been shown to achieve higher homology-directed repair compared to SpCas9. LbCas12a-mediated HDR is most efficient when an ssDNA donor template that is complementary to the target strand is provided (SpCas9 favors the non-target strand). Cas12a nucleases have expanded the targeted coverage many fold, which will help target non-coding regions that are AT rich in zebrafish (Moreno-Mateos et al., 2017).

#### Base Editing Tools

The majority of genetic diseases are caused by point mutations (single or multiple) that result in amino acid substitutions which generate non- or partially functional proteins. Studying these mutations in a model organism using gene knockout technology may not completely mimic the mutations found in human patients. Creating these mutations in zebrafish has been challenging: a targeted knock-in mutant is achieved via homologous recombination by delivering sgRNA and Cas9 together with either a single-stranded oligonucleotide or a donor plasmid containing the left and right homologous arms. Several strategies have been developed for introducing specific changes using knock-in technologies (Prykhozhij et al., 2018; Tessadori et al., 2018; Zhang et al., 2018), and reviewed in many publications (Albadri et al., 2017).

However, the success rate of homology-directed repair (HDR) is extremely low, so introducing specific changes in the genome has been difficult because repair machinery tends to favor nonhomologous end joining repair. Moreover, HDR requires the delivery of donor template to the target cells and precise repair of the genomic sequence. Recently, Jeffery Essner's lab described an optimized targeted knock-in strategy, called GeneWeld, in which they developed a series of donor plasmids for gene tagging [pGTag-plasmids for Gene Tagging (58)]. This strategy is based on the targeting of multiple genomic loci using donor plasmids with short homology arms (24–48 bp), and can be used to integrate cargos up to 2 kb in zebrafish with high efficiency [up to 50% germline transmission (58)]. This technology should allow for maximal integration of fluorescent tags.

For introducing point mutations, recent progress in CRISPRmediated base editing allows for the introduction of point mutations (conversion of G-C base pairs to A-T base pairs or vice-versa) without inducing a double-stranded break (**Figure 3**; Komor et al., 2016; Gaudelli et al., 2017).

The first-generation CRISPR base editor (BE1) uses catalytically inactive Cas9 (dCas9) fused with cytidine deaminase enzyme encoded by the rat APOBEC1 gene (Komor et al., 2016). The cytidine deaminase enzyme converts cytosine bases into uridine, which are then read as thymine during replication. The result is a conversion of cytosine to thymine that occurs within the five-nucleotide window. The second-generation base editor (BE2) is fused with a uracil glycosylase inhibitor (UGI) that prevents excision of uracil during repair; BE2 has marginally higher activity compared to BE1 but does induce indel formation because it contains dCas9not. To further improve editing efficiency, the catalytic His residue at position 840 (which nicks the non-edited strand to mimic newly synthesized DNA) was restored to create BE3, the most widely used base editor. BE3 is the most efficient of the three base editors, and may also induce indels due to its nicking capabilities (Komor et al., 2016).

A new version of BE3 – HF-BE3 – was developed by incorporating mutations in Cas9 known to increase specificity and decrease off-target editing; in practice HF-BE3 appears to have lower on-target editing efficiency (Rees et al., 2017). Delivering the BE3- ribonucleoprotein complex (RNP) results in more robust editing than using plasmid-mediated delivery. This efficiency of BE3 RNP was further tested in zebrafish to generate specific point mutations targeting the tyrosinase locus: two of the three BE3:sgRNA RNP complexes were able to induce substantial point mutations in vivo, with 4–5% editing efficiency (Rees et al., 2017).

David Liu's group continued their effort toward refining and improving the base editors. They engineered the next generation base editors (BE4) to increase the base conversion efficiency by 50%. BE4 editors have extended (32 aa) rAPOBEC1-Cas9n and Cas9n-UGI linker (9 aa), and fusion of an additional UGI to the C terminus with another 9-amino acid linker. The BE4 base editor was further refined by adding the bacteriophage Mu protein Gam, which binds to double-stranded breaks and reduces indel formation to less than 1.5%; this modified base editor is called BE4-Gam (Komor et al., 2017). To increase APOBEC1 expression, ancestral sequence reconstruction using 468 homologs of APOBEC1 was performed, and two ancestors (Anc689 and Anc687) were selected (Koblan et al., 2018). Codonoptimized bipartite NLS were added at the N- and C-termini, similar to BE4max, to create the AncBE4max variant that showed improved editing at multiple loci (Koblan et al., 2018).

In zebrafish, cytidine deaminase fused with Cas9 nickase was able to induce sequence-specific single base mutations from ∼9 to 28% efficiency at multiple loci with a low number of indels (Zhang et al., 2017). Authors targeted tyr gene causal gene for human ocular albinism (OA) and oculocutaneous albinism (OCA). A mutation p.P301L in the tyr gene has been identified in OCA patients. While they were not able to convert proline to leucine, proline was converted to three other amino acids: serine, alanine or threonine. Edited embryos showed the loss

of pigmentation in the eyes of injected embryos. Five other targets tested also converted cytosine to thymine with varying efficiency (Zhang et al., 2017). While the BE system works in zebrafish, the efficiency is low compared to knockouts, and further optimization is required to improve editing efficiency.

Similar to the BE system, the "Target-AID system" was developed by Japanese researchers. The Target-AID system is composed of nuclease-dead Cas9 or Cas9 nickase fused with activation-induced cytidine deaminase (AID) encoded by the PmcDA1 gene from sea lamprey. Target-AID can also induce cytosine to thymine conversion within a five-nucleotide window. The target-AID system was used in zebrafish to introduce premature stop codons (TAG or TAA) by converting cytosine to thymine. Two genes, chordin (chd) and one-eyed pinhead (oep) were targeted using this strategy, and the introduction of premature stop codons phenocopies the known genetic mutants (Tanaka et al., 2018).

Recently, another base editor (eA3A-BE3) fused with an engineered human APOBEC3A (eA3A) domain was shown to deaminate cytidines in a more controlled manner, and function according to a TCR > TCY > VCN (V = G, A, C, Y = C, T) hierarchy (Gehrke et al., 2018). The new base editor variant has shown comparable activities on cytidines in TC motifs, with reduced or no significant editing on cytidines in other motifs. Furthermore, eA3A-BE3 has shown low undesirable bystander mutations compared to other versions (Gehrke et al., 2018).

Existing cytosine deaminase base editors can target bases located between the 4th and 8th position in the target sequence. To expand the targeting window, a new base editor for programming larger C to U (T) scope (BE-PLUS) was developed. This new editor utilizes the SunTag system (Jiang et al., 2018); SunTag contains multiple copies of GCN4 peptide (each consisting of 19 residues) which is recognized by a single chain variable (scFV) antibody. BE-PLUS contains three components: nickase Cas9 fused at the C-terminus to 10 copies of GCN4 peptide (SunTag), scFv-APOBEC-UGI-GB1, and sgRNA. BE-PLUS induced fewer C-T conversions at positions 4–8, but converted C-T at 9–16 positions more efficiently. However, at positions 4–8, BE-PLUS was as efficient as the previously described BE3 (Jiang et al., 2018).

Third- and fourth-generation base editors (BE3 and BE4Gam) were further optimized by codon optimizing Cas9, as well as by adding a FLAG tag and NLS at the N-terminus. These modified base editors were shown to improve C-T conversion up to 50-fold compared with the original BE3 or BE4 base editors (Zafra et al., 2018). A novel method - CRISPR-SKIP - has been shown to program exon skipping by mutating splice acceptor sites using cytidine deaminase (Gapinske et al., 2018). The CRISPR-SKIP webtool can identify exons that can be skipped using this method, and it currently supports BE3, VQR-BE3, VRER-BE3, and SaKKH-BE3 variants (Gapinske et al., 2018). Michael Bassik's lab developed a novel base editor CRISPR-X, which uses an RNA aptamer (MS2) fused to sgRNA to recruit the cytidine deaminase to the target site and induce somatic hypermutation within a 100 bp window. This is a powerful approach for protein engineering because it can generate a diverse population of alleles that could be useful for directed evolution (Hess et al., 2016).

### Adenine Base Editors

Cytidine deaminase-based base editors convert C-T or G-A; there are no natural enzymes that can convert A-G or T-C. To address this problem, David Lius' lab developed an adenine base editor (ABE) to modify adenine bases. The existing adenosine deaminase TadA/ADAR enzymes do not act on double-stranded DNA. Using phage-assisted continuous evolution (PACE), multiple rounds of directed evolution led to the identification of Escherichia coli TadA that can use DNA as a substrate. The ABE consists of a nickase Cas9 fused with a heterodimer of wild-type TadA and engineered TadA enzymes, guided by sgRNA to the target site. Engineered TadA converts adenine (A) to inosine (I) on the DNA target; inosine is recognized as guanine during DNA repair or replication, thus converting adenine (A) to guanine (G) or thymine (T) to cytosine (C). Of the Liu lab's multiple versions from ABE 0.1 to ABE 7, ABE7.10 has been shown to convert AT to GC with approximately 50% efficiency in mammalian cells. The ABE7.10 variant converts bases at position 4 to 7, and ABE7.8 or ABE7.9 variants convert bases at positions 4–9. The ABE7.10 variant was further optimized to generate a new variant, ABEmax, by replacing SV40NLS to codon-optimized bipartite NLS at both the N- and C-termini. Modified ABEmax increases the base substitution rate from ∼1.5- to 2-fold without changing the editing window; however, the rate of indels slightly increased.

Both cytidine deaminase and adenine deaminase enzymes are further fused with different variants such as VQR, VRER, SaCas9KKH, and newly evolved Cas9 such as xCas9, iSpymacCas9, and SpCas9-NG (Kim Y. B. et al., 2017; Hu et al., 2018). Thus, both types of base editors will provide coverage to change all four bases in a targeted manner. As the new variants and new orthologs of nucleases evolve the targeting coverage will further expand covering all of the pathogenic variants. **Tables 3**, **4** summarizes the different base editing resources, targeting range of each base editors mentioned above, respectively.

### TRANSCRIPTIONAL MODULATION AND EPIGENOME EDITING

CRISPR/Cas9 has also been repurposed to modulate transcription and manipulate the epigenome. In order to apply the CRISPR system beyond inducing a double stranded break, the DNA cleavage activity of Cas9 nuclease must be inactivated. Cas9 from Streptococcus pyogenes (SpCas9) contains two nuclease domains – a RuvC-like domain and a HNH domain – both of which are required to induce a double stranded break (Jinek et al., 2012). Introducing mutations in the catalytic residues of both nuclease domains (D10A, H840A) will create a catalytically inactive version of the Cas9 (dCas9; Qi et al., 2013). Using sgRNA, dCas9 can be recruited to a specific target without inducing a DNA break. To modulate transcription, dCas9 was first fused with transcriptional activators (VP64, a synthetic tetramer of the Herpes Simplex Viral Protein or p65 a transcription factor activation domain) or transcriptional repressors [KRAB, a Kruppel-associated box and the transcriptional repressor of Kox1, or 4X mSin Interaction Domain (SID; Konermann et al., 2015)]. These fusion proteins

#### TABLE 3 | Resources for base editing.

fcell-07-00013 March 1, 2019 Time: 11:26 # 10


result in transcriptional activation (CRISPRa) or repression (CRISPRi) when targeted to the regulatory or coding regions of the gene.

Both CRISPRa and CRISPRi have been shown to work in modulating transcription of the target genes in zebrafish (Long et al., 2015). Two genes required for otic placode induction (fgf8a and foxi1) were targeted to demonstrate the utility of CRISPRa and CRISPRi in zebrafish. When sgRNAs targeting fgf8a coding


regions were co-injected with dCas9-KRAB (CRISPRi) fusion protein, the expression of fgf8a was reduced at 11 hpf, and smaller otic vesicles were observed at 32 hpf (Long et al., 2015). Similarly, when dCas9-VP160 (CRISPRa) together with either sgRNAs targeting fgf8a or foxi1 were injected in one-cell stage embryos, the expression of fgf8a and foxi1 was increased and the resulting animals showed enlarged otic vesicles.

The dCas9 was also fused with putative Eve repressor domain of zebrafish Evx1, and the resulting dCas9-Eve fusion together with three sgRNAs targeting sequences upstream of all zinc finger transcription factors (zfnl1s) were used to inhibit the transcription of the znfl1 in zebrafish (Dong et al., 2017). Decreased expression of znfl1 disrupts the formation of the posterior neuroectoderm in zebrafish gastrula, and the phenotype perfectly phenocopies that generated by the anti-sense morpholino (Dong et al., 2017).

In zebrafish it has been shown that mutants generated by targeting mutagenesis techniques, genetic compensation or transcriptional adaptation could all trigger the upregulation of related genes and compensate for the loss of the targeted gene. Such upregulation and compensation were not observed when antisense morpholinos were used, suggesting that downregulation of target genes using CRISPRi could be an alternate tool to study gene function.

Additionally, catalytically inactive Cas9 has been fused to various epigenetic effectors such as the catalytic core of the human acetyltransferase p300 which catalyzes acetylation of histone H3 lysine 27 (Hilton et al., 2015), histone demethylase (Kearns et al., 2015), histone deacetylase (HDAC) (Kwon et al., 2017) and many others (reviewed in Lau and Suh, 2018).

#### LINEAGE TRACING USING CRISPR/Cas9

A fundamental goal in developmental biology is to determine the origin of different cell types and tissues, and to establish their relationship in complex organisms. Lineage tracing is one method employed by developmental biologists to study the origin of cell types: techniques include dye based markers, nucleotide pulse-chase analysis, transplantation, sequencing somatic mutations, Cre-Lox and FLP-FRT based methods (reviewed in Kretzschmar and Watt, 2012). These methods can efficiently label cells at a single time-point to study large numbers of clonal populations in a complex animal, however, a detailed

lineage tree over time cannot be reconstructed; understanding how cells change over the time will help determine the mechanisms of disease progression. Recently, the CRISPR/Cas9 mediated genome editing technique was used to generate genetic scars (indels) in the genome which serve as genetic barcodes for use in the reconstruction of cell lineages in developing animals or adults. Using this principle, many innovative approaches have been developed including genome editing of synthetic target arrays for lineage tracing (GESTALT; McKenna et al., 2016), lineage tracing by nuclease-activated editing of ubiquitous sequences (LINNAEUS; Spanjaard et al., 2018), ScarTrace (Alemany et al., 2018), and memory by engineered mutagenesis with optical in situ readout (MEMOIR; Frieda et al., 2017). CRISPR based lineage tracing is being adopted in multiple model organisms including zebrafish (Schmidt et al., 2017; Kalhor et al., 2018; Raj et al., 2018; Spanjaard et al., 2018). GESTALT, first applied to the understanding of the origin of organ development in zebrafish, was developed in the labs of Jay Shendure and Alex Schier. The Schier Lab engineered 10 different target sequences (unique barcodes in the 3<sup>0</sup> UTR of DsRed) that are not found in the zebrafish genome to avoid any interreference with normal development. A transgenic line that drives the expression of DsRed under the ubiquitin promoter was generated. Ten sgRNAs that target the 10 unique sequences present in the transgenic lines were injected together with Cas9 protein in one-cell stage zebrafish embryos. Embryos were collected at different time points, and target regions were amplified using primers containing unique molecular identifier (UMI) to add UMI barcodes in the amplicons. (The process of UMI tagging helps in assigning individual sequencing reads back to the cell of origin). Sequencing confirmed the in vivo allelic diversity, and the recovered alleles were used to reconstruct the lineage tree. To investigate whether these barcodes can also be recovered in adult animals, several organs (brain, eyes, intestine, gills, heart, and blood) were collected and subjected to DNA sequencing to recover barcode information. It was concluded that most cells in different adult organs were derived from fewer embryonic progenitors; more than 98% of circulating blood in an adult zebrafish contains five common alleles, suggesting a highly clonal origin of the blood system in zebrafish. The GESTALT method was further modified by combining single-cell RNA sequencing to develop scGESTALT [single cell Genome Editing of Synthetic Target Arrays for Lineage Tracing; (Raj et al., 2018)]. The workflow for the cell lineage tracing and scRNA-seq experiment involves the introduction of sgRNAs to target exogenous sequences and the isolation of single cells at appropriate time points. This is followed by mRNA isolation, reverse transcription and cDNA amplification, library preparation, and sequencing both DNA and RNA; this method has been used to identify more than 100 different cell types during brain development. The scGESTALT method also allows barcodes to be recorded at post gastrulation stages by employing temporal control of Cas9 using a heat shock promoter and constitutively expressing sgRNA from the U6 promoter (Raj et al., 2018).

Unlike GESTALT, LINNAEUS, and ScarTrace approaches take advantage of existing transgenic lines carrying multiple integrations of a transgene – green fluorescent protein (GFP) or mCherry. Both LINNEAUS and ScarTrace combine lineage tracing with identification of cell types by singlecell transcriptomics. ScarTrace uses a zebrafish line carrying eight in-tandem copies of a histone–GFP transgene. sgRNA targeting the GFP with Cas9 protein is injected into one-cell stage embryos. Cas9 induces double-stranded breaks that when repaired by non-homologous end joining leave insertions or deletions (scars). During embryonic development, cells accumulate these scars and pass them on to future generations. When the scars are then sequenced, any adult cells containing identical scars must originate from a common progenitor cell. This method also defines cell types based on their transcriptome, thus cataloging both cell type and progenitor for each organ type. ScarTrace revealed that hematopoietic cells in the kidney marrow originated from fewer embryonic progenitors, and multiple progenitors give rise to specific cell types in the brain and eye. It was further revealed that a common progenitor produces both epidermal and mesenchymal cells of the caudal fin. Interestingly, this technique also showed how a progenitor cell commits to produce a left or right eye in zebrafish.

The LINNAEUS approach is similar to ScarTrace, however, LINNAEUS uses a zebrafish line carrying an RFP transgene with 16–32 independent integrations in the genome. The presence of independent integrations in different loci protect the scars from being removed or overwritten by Cas9. sgRNA targeting RFP and Cas9 were together injected into one-cell stage embryos; since RFP-targeting sgRNA generates indels (scars) and leave RFP non-functional, loss of RFP was used as a quality control to monitor the efficiency of editing and scar formation. At later time points, embryos were dissociated into single cells and RFP transcripts were sequenced to quantify the scar formation and the transcriptome from the same cell was sequenced by scRANAseq. Spanjaard and colleagues applied this approach to identify many different cell types from dissected adult organs including heart, liver, primary pancreatic islets and telencephalon. They found that immune cells from different organs can be grouped together in the lineage tree: analysis of cardiac and pancreatic cell types showed the early separation of myocardial and endocardial lineages.

MEMOIR method uses two different tools: Sequential single molecule Fluorescence In Situ Hybridization (smFISH) that reveals which specific genes are active in a particular cell, and CRISPR/Cas9 that generates indels (Frieda et al., 2017). MEMOIR uses bipartite genetic recording elements called barcoded scratchpads. Each scratchpad contains 10 repeat units, and sgRNAs and Cas9 targeting scratchpads induces indels. There is also a barcode adjacent to each scratchpad which can be identified by smFISH and allows for the recording of each lineage. This method has so far only been used in vitro to record the cellular history of mouse embryonic stem cells through multiple generations (Frieda et al., 2017).

The methods described above generate complex lineage trees using scarring accumulated over several hours, and each method has a limited number of scars. To overcome this limitation, a selftargeting homing guide RNA (hgRNA) system was developed that can induce scarring over a longer time period and further

increase complexity (Perli et al., 2016; Kalhor et al., 2018). Lineage tracing in mammals has been challenging compared to that in zebrafish. Kalhor et al. (2018) used hgRNAs to generate a mouse model for the study of cell lineages during early mouse development; hgRNA containing a targeting sequence with PAM was attached to a scaffold and allowed Cas9 to target the expression cassette. For in vivo cell lineage tracing, a transgenic mouse harboring 41 different homing guide RNA expression cassettes was created. This transgenic mouse was bred with a Cas9 expressing mouse strain to induce indels (barcodes). These barcodes can be used to track cells temporally and spatially (Kalhor et al., 2018).

These proof-of-principle studies have developed elegant lineage tracing strategies to establish lineage relationships and understand the fundamental mechanisms of cell differentiation under normal and pathological conditions (e.g., cancer metastasis) in complex model organisms including zebrafish. The development of CRISPR based lineage tracing methods is well-timed and complementary to the efforts toward building the human cell atlas (Regev et al., 2017). These approaches will have significant impact on our understanding of the origin of each cell type and how the adult body is developed from a single cell.

#### CONCLUSION AND FUTURE PERSPECTIVES

As described above the focus of recent research has been on developing strategies to improve Cas9 function, targeting coverage, and on-target efficiencies by reducing off target editing. CRISPR-based genome editing technologies have revolutionized biological research; CRISPR-related nucleases have been repurposed in many applications, and recent developments in

#### REFERENCES


base editing and lineage tracing have further increased their utility in studying development and human diseases. New and inexpensive sequencing technologies are accelerating the discovery of candidate disease genes and pathogenic variants. CRISPR has provided a variety of tools to precisely modify the genome in a targeted manner for a variety of applications including functional gene knockouts, targeted induction or correction of single point mutations, and epigenome editing. Recent work has been focused on refining the specificity and expanding the target coverage of Cas9; directed evolution has led to the discovery of multiple Cas9 variants that will significantly expand the targeting coverage. Furthermore, development of base-editing techniques is an important milestone in the study of pathogenic variants in animal models; they will not only accelerate the functional validation of candidate disease genes in a model organism, but also accelerate the development of therapeutic tools for the treatment of a wide range of human diseases. Development of CRISPR-based lineage tracing methods are revealing information that could have been challenging to uncover using traditional approaches including the discovery of novel cell types and the origin of cells in different organs and tissues in complex model organisms.

#### AUTHOR CONTRIBUTIONS

KL, CP, TR, PV, and GV researched the data. GV wrote the article. All authors read the article and approved it for publication.

#### FUNDING

This research was supported by a grant from NIH/COBRE GM103636 (Project 3 to GV).

end joining for precision gene editing applications. PLoS Genet. 14:e1007652. doi: 10.1371/journal.pgen.1007652


acetyltransferase activates genes from promoters and enhancers. Nat. Biotechnol. 33, 510–517. doi: 10.1038/nbt.3199


fcell-07-00013 March 1, 2019 Time: 11:26 # 13

Editing from Sanger Sequencing. CRISPR J. 1, 239–250. doi: 10.1089/crispr. 2018.0014


CRISPR-Cas9 targeting in vivo. Nat. Methods 12, 982–988. doi: 10.1038/nmeth. 3543


fcell-07-00013 March 1, 2019 Time: 11:26 # 14

Frontiers in Cell and Developmental Biology | www.frontiersin.org


**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, Petree, Requena, Varshney and Varshney. 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.

**155**

fcell-07-00013 March 1, 2019 Time: 11:26 # 15