# SIGNALLING PATHWAYS IN EMBRYONIC DEVELOPMENT

EDITED BY: Juan J. Sanz-Ezquerro, Andrea E. Münsterberg and Sigmar Stricker PUBLISHED IN: Frontiers in Cell and Developmental Biology

#### *Frontiers Copyright Statement*

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

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

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

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

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

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

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

ISSN 1664-8714 ISBN 978-2-88945-346-7 DOI 10.3389/978-2-88945-346-7

## 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

# **SIGNALLING PATHWAYS IN EMBRYONIC DEVELOPMENT**

Topic Editors:

**Juan J. Sanz-Ezquerro,** National Center for Biotechnology (CSIC), Spain **Andrea E. Münsterberg**, University of East Anglia, United Kingdom **Sigmar Stricker,** Freie Universität Berlin, Germany

Whole-mount in-situ hybridisation of a 4 day old chicken embryo for MyoD shows the developing muscles in the somites (striped pattern), the limbs and branchial arches.

Image by Sigmar Stricker.

The formation of a complex multicellular organism from a single cell is one of the most amazing processes of biology. Embryonic development is characterised by the careful regulation of cellular behaviours such that cells proliferate, migrate, differentiate and form tissues at the correct place and time. These processes are genetically controlled and depend both on the history of cells, their lineage, and on the activities of signalling pathways, which coordinate the cell interactions leading to organogenesis.

The aim of the Frontiers research topic "Signalling pathways in embryonic development" has been to provide a forum for experts in cell and developmental biology to share recent advances in the field of signalling during embryonic development. Sixteen articles in a variety of formats are united in this Topic, offering a valuable collection for researchers looking for an update in the knowledge of signalling pathways operating during embryogenesis. The works, focused mainly on vertebrates, explore different aspects of this theme from cell communication to organ formation and have implications for areas as distant as evolution or pathology.

Understanding developmental signalling pathways is important for several reasons. It gives us information about basic mechanisms of cell function and interactions needed for morphogenesis and organogenesis. It uncovers the basis of congenital malformations, since errors at any step of cell signalling during development are a major cause of defects. This fundamental insight gives us clues to understand the mechanisms operating in evolution that explain diversity in form and function. And finally, it allows the identification of possible causes of disease in the adult organism (such as cancer or degenerative diseases) pinpointing possible targets for therapeutic approaches.

**Citation:** Sanz-Ezquerro, J. J., Münsterberg, A. E., Stricker, S., eds. (2017). Signalling Pathways in Embryonic Development. Lausanne: Frontiers Media. doi: 10.3389/978-2-88945-346-7

# Table of Contents


*\*\*see also 157. A Novel Role for the BMP Antagonist Noggin in Sensitizing Cells to Non-canonical Wnt-5a/Ror2/Disheveled Pathway Activation*

## **1.4. Shh signalling**

*\*\*see 119. Sonic Hedgehog Signaling in Limb Development*

## **2. Organogenesis**

## **2.1. Ear development**

## *71 Autophagy in the Vertebrate Inner Ear*

Marta Magariños, Sara Pulido, María R. Aburto, Rocío de Iriarte Rodríguez and Isabel Varela-Nieto

*\*\*see also 36. Signaling and Transcription Factors during Inner Ear Development: The Generation of Hair Cells and Otic Neurons*

## **2.2. Epicardium development**

*80 More than Just a Simple Cardiac Envelope; Cellular Contributions of the Epicardium*

Angel Dueñas, Amelia E. Aranega and Diego Franco

## **2.3. Muscle development**

*91 Non-myogenic Contribution to Muscle Development and Homeostasis: The Role of Connective Tissues*

Sonya Nassari, Delphine Duprez and Claire Fournier-Thibault

*108 Pitx2 in Embryonic and Adult Myogenesis* Francisco Hernandez-Torres, Lara Rodríguez-Outeiriño, Diego Franco and Amelia E. Aranega

## **2.4. Limb development**


Juan A. Montero, Carlos I. Lorda-Diez, Javier Francisco-Morcillo, Jesus Chimal-Monroy, Juan A. Garcia-Porrero and Juan M. Hurle

*144 CaMKII Signaling Stimulates Mef2c Activity* **In Vitro** *but Only Minimally Affects Murine Long Bone Development* **in vivo**

Chandra S. Amara, Christine Fabritius, Astrid Houben, Lena I. Wolff and Christine Hartmann

*157 A Novel Role for the BMP Antagonist Noggin in Sensitizing Cells to Non-canonical Wnt-5a/Ror2/Disheveled Pathway Activation*

Ondrej Bernatik, Tomasz Radaszkiewicz, Martin Behal, Zankruti Dave, Florian Witte, Annika Mahl, Nicole H. Cernohorsky, Pavel Krejci, Sigmar Stricker and Vitezslav Bryja

## **2.5. Eye development**

## *\*\*see 49. WNT/b-Catenin Signaling in Vertebrate Eye Development*

## **3. Cells and their context**

*166 Mechanical Control of Myotendinous Junction Formation and Tendon Differentiation during Development*

Mauricio Valdivia, Franco Vega-Macaya and Patricio Olguín

## *174 Putting Cells into Context*

Sigmar Stricker, Petra Knaus and Hans-Georg Simon

# Editorial: Signaling Pathways in Embryonic Development

Juan J. Sanz-Ezquerro<sup>1</sup> \*, Andrea E. Münsterberg<sup>2</sup> and Sigmar Stricker <sup>3</sup>

<sup>1</sup> Department of Molecular and Cellular Biology, National Center for Biotechnology (CSIC), Madrid, Spain, <sup>2</sup> School of Biological Sciences, University of East Anglia, Norwich, United Kingdom, <sup>3</sup> Institute for Chemistry and Biochemistry, Freie Universität Berlin, Berlin, Germany

Keywords: signaling pathways, embryo development, mechanotransduction, notch signaling pathway, Wnt Signaling Pathway, Shh signaling

**Editorial on the Research Topic**

### **Signaling Pathways in Embryonic Development**

The formation of a complex multicellular organism from a single cell is one of the most amazing processes of biology. Embryonic development is characterized by the careful regulation of cellular behaviors such that cells proliferate, migrate, differentiate, and form tissues at the correct place and time. These processes are genetically controlled and depend both on the history of cells, their lineage, and on the activities of signaling pathways, which coordinate the cell interactions leading to organogenesis.

A limited number of key signaling pathways—Fgf, Hedgehog, Wnt, TGFß, Notch among the most important—operate during development, acting repeatedly at different times and in different regions in the embryo and eliciting diverse cellular responses. This raises the question of how cells integrate all the information they receive and can respond in cell type-specific ways to the same signals. Classical concepts in embryology such as organizers (groups of cells producing instructive signals) and competence (ability of cells to respond) can now be analyzed in molecular terms. In recent years many advances have been made in identifying the signals acting during embryo development and understanding their properties and functions, which is equally of relevance for human pathology and evolution. An important discovery is the conservation of signals and mechanisms, not only in evolutionary terms (similar genes and signals acting in distant organisms), but also in the repeated use of the same signaling pathways at different times and places in the embryos. Moreover, many of those mechanisms are involved in adult tissue homeostasis and regeneration.

Understanding developmental signaling pathways is important for several reasons. It gives us information about basic mechanisms of cell function and interactions needed for morphogenesis and organogenesis. It uncovers the basis of congenital malformations, since errors at any step of cell signaling during development are a major cause of defects. Fundamental insight also gives us clues to understand the mechanisms operating in evolution that explain diversity in form and function. And finally, it allows the identification of possible causes of disease in the adult organism (such as cancer or degenerative diseases) pinpointing possible targets for therapeutic approaches.

In this context, the aim of the Frontiers research topic "Signaling pathways in embryonic development" has been to provide a forum for experts in cell and developmental biology to share recent advances in the field of signaling during embryonic development. Sixteen articles in a variety of formats are united in this Topic, offering a valuable collection for researchers looking for an update in the knowledge of signaling pathways operating during embryogenesis. The works, focused mainly on vertebrates, explore different aspects of this theme from cell communication to organ formation and have implications for areas as distant as evolution or pathology.

Among the signaling pathways with important and widespread roles in development is the Wnt pathway, comprising a family of ligands with homology to wingless in Drosophila. Wnts can bind

Edited and reviewed by: Pedro M. Fernández-Salguero, University of Extremadura, Spain

> \*Correspondence: Juan J. Sanz-Ezquerro jjsanz@cnb.csic.es

#### Specialty section:

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

> Received: 02 August 2017 Accepted: 11 August 2017 Published: 31 August 2017

#### Citation:

Sanz-Ezquerro JJ, Münsterberg AE and Stricker S (2017) Editorial: Signaling Pathways in Embryonic Development. Front. Cell Dev. Biol. 5:76. doi: 10.3389/fcell.2017.00076 to multiple receptor complexes and trigger several downstream signaling cascades [including the so-called canonical WNT/βcatenin dependent signaling pathway, the non-canonical WNT/planar cell polarity (PCP), and the WNT/Ca2<sup>+</sup> pathways], illustrating how the same signal can elicit diverse cellular responses depending on the cell type, context, and developmental timing. Fujimura reviews the role of canonical Wnt signaling in eye development, highlighting the important roles it plays in patterning of ocular tissue, differentiation of retinal pigment epithelium, and morphogenesis of the optic cup. Importantly, mis-regulation of the signaling cascade can lead to eye malformations and disease. Gentzel and Schambony review a group of core intracellular effectors of the Wnt pathway, disheveled (DVL) proteins, which comprise three members in vertebrates. Although all DVLs share a common basic function in Wnt signaling, the expression patterns, and functions of the different isoforms are not totally redundant and have also diverged between different species, suggesting they play specific roles depending on the tissue distribution and specific interactions. Again, mutations in DVL genes can cause human congenital disease, highlighting their important role in development. Additionally, Berger et al. review the role of PTK7 (protein tyrosine kinase 7, a transmembrane receptor) in the fine-tuning of the Wnt signaling network. Its functions in establishing cell polarity, regulation of cell movements, and migration are also essential for development and disease, particularly in cancer and metastasis.

Another important signaling pathway is Notch, a transmembrane protein that mediates juxtacrine cell-cell communication. Notch has many functions in organ formation and adult homeostasis, including cell determination and stem cell maintenance. Carrieri and Dale review the particularly well-studied function of Notch in somitogenesis and also present recent data on the role of FBXW7 protein in regulating the turnover of Notch intracellular domain (NICD, the effector of the pathway), in development and cancer. This relates to an often-overlooked essential point in signaling, which is the termination of activation and resetting of the components, allowing the cells to become competent again. Multiple mechanisms of regulation exist (positive and negative feedback loops) that allow a fine control of signaling pathways at different steps of the intracellular cascades.

Crosstalk between the limited numbers of signaling pathways is a mechanism that allows cells to respond differently to the same signal, producing the diverse cellular behaviors that are needed to build tissues and organs. A new example of this is provided by Bernatik et al. reporting on the role of the BMP antagonist Noggin in sensitizing cells and potentiating the activation of non-canonical Wnt signaling in skeletal development. They also provide evidence for a genetic interaction between these two pathways, which are involved in human congenital malformations.

The role of specific signaling pathways in the formation of particular organs is discussed in other articles. Díez del Corral and Morales review the multiple roles of Fgf signaling in the developing spinal cord. This important structure of the nervous system arises from neural derivatives of an early neuromesodermal population located at the caudal part of the embryo. Extension of this region is coupled to spinal cord formation and several essential processes such as neurogenesis, ventral patterning or neural crest specification are controlled by Fgf signaling. These embryonic functions of Fgfs could be related to its ability to promote regeneration in the injured spinal cord of adults.

Signalling pathways often converge on controlling the expression of transcription factors, which regulate cell fate specification. The integration of Notch signaling and bHLH transcription factors during inner ear development is analyzed by Gálvez et al. which also highlight that these same mechanisms are involved in hair cell regeneration, opening avenues for possible therapeutic approaches in hearing impairment. Ear development is also the topic reviewed by Magariños et al. They present evidence for a crucial role of autophagy, the regulated process of degradation, and recycling of cellular components, in vertebrate inner ear formation.

The limb is a classic model in embryology and some of the most important discoveries related to the roles of signaling pathways in pattern formation, growth, and differentiation have been made studying limb development. Tickle and Towers review the role of Shh in this process, a paradigm of how signals control and integrate tissue pattern and growth. They also discuss the implications of this important pathway for congenital malformations in humans and for the generation of limb morphological diversity during evolution. Montero et al. also treat this evolutionary aspect in their perspective article. They present a detailed analysis of Sox9 expression in developing digits of several species. This transcription factor, regulated by signaling pathways such as BMPs, Tgfßs, or Fgfs is involved in formation of the chondrogenic template of the skeleton. Differences in Sox9 expression patterns among species that have specific morphologies may reflect differences in signaling pathways controlling its expression. Also related to skeletal development, Amara et al. show that the effects of Calcium/Calmodulin dependent kinase II (CAMKII), an effector for Ca+2 -dependent signal transduction, in promoting chondrogenic differentiation seems to be specific for chicken embryos. This function is not observed in the mouse, thus highlighting the existence of differences in signaling functions and regulation among different species.

Integration of extrinsic and intrinsic regulatory cues is essential for organ formation. Dueñas et al. review the role of signals, transcription factors and cellular processes in the formation of the epicardium. This is the external-most layer of the heart that serves not only as the outer cover for this organ, but also seems to play a role in regeneration. Thus, understanding the basis of its development may have important therapeutic implications. Two articles deal with muscle development. Hernandez-Torres et al. review the role of Pitx2 in embryonic and adult myogenesis. A hierarchy of transcription factors controls skeletal muscle differentiation and Pitx2 plays an important role in the regulation of this process. Importantly, it also seems to be involved in the establishment and function of satellite cells, the stem cells resident in adult muscle, thus opening new avenues for development of regenerative therapies. Additionally, Nassari et al. review the role of connective tissues in muscle development. Apart from the intrinsic molecular signals mentioned above, the interaction of muscle cells with surrounding tissues (bone, cartilage, tendon, and ligament) is critical for the correct assembly of the musculoskeletal system during development and for maintaining adult homeostasis.

An emerging theme in developmental biology is the control of tissue morphogenesis by physical forces (mechanotransduction). Valdivia et al. review the mechanical control of myotendinous junction formation and tendon differentiation, highlighting again the importance of the interplay between chemical and mechanical signaling during embryogenesis. In the same line, Stricker et al. provide a timely discussion reminding us that cells in embryos and adult organisms are not present in isolation, but embedded in extracellular matrices into complex tissues. Cells attach to the ECM and sense its mechanical properties. Typically, experimental in vitro conditions do not fully reproduce this environment, which is however critical for the physiological cellular responses to signaling cascades. The challenge for the future is to try and integrate as many interactions as possible when designing experiments.

We hope that the articles in this topic will be of interest to researchers working in development and cell biology, fuelling discussion on this area and opening new avenues for thinking and investigation.

## AUTHOR CONTRIBUTIONS

JS was the Guest editor of this Research Topic, inviting co-editors AM and SS and working with them to define the subjects to be treated. They identified and invited leaders in specific research fields to contribute their work to the Research Topic. They acted as handling editors of manuscripts in the topic. JS wrote the Editorial with input from the other co-editors.

## FUNDING

Research in AM laboratory is supported by BBSRC grants BB/N007034, BB/N002970 and BHF grant PG/15/77/31761. Research in JS laboratory is supported by grants SAF2013-45331- R and SAF2016-79792-R from spanish MINECO.

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

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

# The Multiple Roles of FGF Signaling in the Developing Spinal Cord

#### Ruth Diez del Corral 1, 2 \* and Aixa V. Morales <sup>1</sup> \*

<sup>1</sup> Department of Cellular, Molecular and Developmental Neurobiology, Cajal Institute, Consejo Superior de Investigaciones Científicas, Madrid, Spain, <sup>2</sup> Champalimaud Research, Champalimaud Centre for the Unknown, Lisbon, Portugal

During vertebrate embryonic development, the spinal cord is formed by the neural derivatives of a neuromesodermal population that is specified at early stages of development and which develops in concert with the caudal regression of the primitive streak. Several processes related to spinal cord specification and maturation are coupled to this caudal extension including neurogenesis, ventral patterning and neural crest specification and all of them seem to be crucially regulated by Fibroblast Growth Factor (FGF) signaling, which is prominently active in the neuromesodermal region and transiently in its derivatives. Here we review the role of FGF signaling in those processes, trying to separate its different functions and highlighting the interactions with other signaling pathways. Finally, these early functions of FGF signaling in spinal cord development may underlay partly its ability to promote regeneration in the lesioned spinal cord as well as its action promoting specific fates in neural stem cell cultures that may be used for therapeutical purposes.

Keywords: spinal cord, spinal cord injury, neuromesodermal progenitors, neural stem cells, patterning, neurogenesis, caudal extension, FGF

## INTRODUCTION

The spinal cord is the most caudal part of the nervous system which is responsible for body motion, including locomotion, somatosensation and the control of basic functions of the autonomous nervous system. During development, in addition to the neurons that reside within the spinal cord, it provides neural crest cells for the formation of sensory ganglia, ganglia of the autonomous system and for the enteric nervous system. A fundamental aspect of spinal cord development is its relation to the organs and muscles it innervates. Thus, spinal cord development appears highly coordinated in space and time with the caudal extension that accompanies the development of the whole body.

The spinal cord cells of vertebrates derive from a region initially specified as neuromesodermal progenitors (NMP) with mixed neural and mesodermal characteristics (Wilson et al., 2009; Henrique et al., 2015; Row et al., 2016), with the exception of those forming the floor plate which in amniotes derive from the node. In chick and mouse, this corresponds to a region of the epiblast adjacent to the early node and the rostral primitive streak. From this population, some cells remain in the ectoderm layer and form most of the spinal cord, while others gastrulate through the primitive streak to become part of the paraxial mesoderm (Wilson et al., 2009; Henrique et al., 2015). Later, with the closure of the caudal neuropore, the NMP region remains in the tailbud from which the caudal spinal cord and mesodermal populations segregate. Overall, this constitutes an ongoing process that takes several days to generate the complete rostrocaudal axis. Different aspects of spinal cord development such as initiation of neurogenesis, ventral patterning and neural crest specification and migration are conditioned by this caudal axis elongation (**Figure 1**). This

#### Edited by:

Juan Jose Sanz-Ezquerro, Centro Nacional de Biotecnología, Spain

#### Reviewed by:

Cristina Pujades, Pompeu Fabra University, Spain Gregg Duester, Sanford Burnham Prebys Medical Discovery Institute, United States Benjamin Louis Martin, Stony Brook University, United States

#### \*Correspondence:

Ruth Diez del Corral ruth.diezdelcorral@ neuro.fchampalimaud.org Aixa V. Morales aixamorales@cajal.csic.es

#### Specialty section:

This article was submitted to Signaling, a section of the journal Frontiers in Cell and Developmental Biology Received: 17 February 2017 Accepted: 11 May 2017 Published: 02 June 2017

#### Citation:

Diez del Corral R and Morales AV (2017) The Multiple Roles of FGF Signaling in the Developing Spinal Cord. Front. Cell Dev. Biol. 5:58. doi: 10.3389/fcell.2017.00058

is a complex process involving several signaling pathways and gene networks in which the FGF signaling pathway stands as a crucial regulator, maintaining cells in an immature state until they are displaced to a region where they are no longer influenced by it.

FGF signaling acts in numerous stages and tissues during embryonic development and the use of experimental approaches designed to manipulate the FGF signaling pathways at specific stages and tissues has been fundamental to overcome its early roles in implantation, gastrulation, and neural induction. These include treatment of tissue explants with FGF factors or pharmacological antagonists, expression of pathway inhibitors or truncated FGFR proteins that function interfering with the normal function in the neural tube cells in ovo and the use of conditional mouse mutants specifically removing FGFs or FGFR in the NMP and its derivatives.

Here, we review the contribution of FGF signaling in the initial process of spinal cord specification and elongation and then we cover the initiation of neurogenesis, ventral patterning, and neural crest specification and migration. We make an effort to identify and separate the different steps in these highly interconnected networks governing spinal cord extension and associated events, focusing on the influence of FGF signaling on the neural tissue. In addition, we have selected some of the evidence supporting the use of FGF to promote regeneration of the lesioned adult spinal cord both acting on spinal cord cells in vivo as well as to promote expansion of neural stem cells in vitro and their differentiation toward specific neuronal fates for their use for regenerative purposes.

Note: Gene symbols are italicized in all species, but there are specie-specific differences. Thus, gene symbols for human and chick appear all in upper-case; for mouse and rat with only the first letter in upper-case and for fish, gene symbols appear with all letters in lower-case. In the case of protein symbols, they are not italicized and all letters are in upper-case, except in fishes where only the first letter is upper-case (http://www.biosciencewriters.com/Guidelines-for-Formatting-Gene-and-Protein-Names.aspx). When referring to genes from several species they have been separated by a slash.

## FGF SIGNALING PATHWAY: EXPRESSION OF COMPONENTS IN THE DEVELOPING SPINAL CORD

Let's first start with a brief introduction of the components of the FGF signaling pathway in the context of spinal cord development. As most signaling pathways, the FGF pathway includes ligands, receptors, modulators, intracellular transducers, and final effectors (Ornitz and Itoh, 2015). The only components exclusive for the pathway are the ligands (up to 23 FGFs have been described in vertebrates) and their receptors of the tyrosin kinase (RTK) type (FGFR1–4 in vertebrates). Other more general players, which are also used by other signaling pathways, such as the pathway inhibitors SPROUTY2, SEF, DUSP6, and the transcription factor effectors of the ETV family, are particularly associated to this pathway as the corresponding mRNAs are highly expressed in regions with high FGF activity and in particular in the caudal NMP region (Chotteau-Lelievre et al., 2001; Karabagli et al., 2002; Corson et al., 2003; Harduf et al., 2005; Lunn et al., 2007). Interestingly, they are themselves downstream targets of the pathway and are thus considered its readouts and have been the basis for the development of pathway activity reporters (Molina et al., 2007; Ekerot et al., 2008). However, as these downstream targets of the pathway are not exclusively activated by the FGF pathway, they do not constitute definitive readouts of the activity of the FGF pathway. The identification of cells where the pathway is truly active is still one of the main difficulties in the analysis of FGF function, as none of the intracellular cascades is specific for FGF signaling and the difference with other RTK pathways may be in the fine tuning of the signaling properties.

The three main intracellular cascades that can mediate the FGF signal are: the RAS-MAPK, the PI3K-AKT and the PLPCγ pathways (**Figure 2**). High levels of MAPK phosphorylation are detected in the NMPs and surrounding area and these depend on the activation of FGF pathway (Lunn et al., 2007). Moreover, most of the effects resulting from FGFR inhibition in this region can also be observed following the inhibition of MEK (MAPK Kinase), suggesting this is the main FGFR downstream pathway in this region (Diez del Corral et al., 2002; Delfino-Machin et al., 2005; Lunn et al., 2007; Martinez-Morales et al., 2011; Olivera-Martinez et al., 2012; Morales et al., 2016).

A gradient of AKT phosphorylation has also been described in the region surrounding the node with higher levels caudally (Dubrulle et al., 2001) but the exposure to PI3K inhibitors does

not result in the same effects as blockade of FGFR signaling (Martinez-Morales et al., 2011) and thus the relevance of the AKT pathway in this context has not been addressed further.

The most comprehensive analysis of the expression patterns of FGF signaling related genes in spinal cord development has been performed in the chick. At the stages of chick spinal cord specification, several FGFs, including FGF3, FGF4, FGF8, FGF13, FGF18, are expressed in the caudal NMP region or surrounding tissues (Karabagli et al., 2002; Delfino-Machin et al., 2005). During later stages (during spinal cord elongation and including tailbud formation) FGF3, FGF4, FGF13 and FGF18 become restricted to the primitive streak while FGF8 is more broadly expressed in the streak, the adjacent NMP region and the ingressing mesoderm (Karabagli et al., 2002; Delfino-Machin et al., 2005). Expression of FGF8 is highly dynamic as those cells that progress from the NMP state to the spinal cord fate or from the presomitic mesoderm to the somitic mesoderm slowly downregulate their expression (**Figures 3A–C**). Expression of FGF4 and FGF8 in the NMP region (including caudal lateral epiblast and later, the tailbud) continues for several days but declines toward the final stages of somitogenesis and the cessation of axis elongation (Cunningham et al., 2011; Olivera-Martinez et al., 2012).

FGFR1-3 are initially present in the NMP zone (Karabagli et al., 2002; Lunn et al., 2007; Nishita et al., 2011), but later, only FGFR1 remains throughout the neural tissue including the NMP region while FGFR2 is absent there and becomes restricted to the neural tube, rostral to Hensen's node, and FGFR3 restricts to the neural tube adjacent to somites (Karabagli et al., 2002; Lunn et al., 2007; Nishita et al., 2011).

Similar expression patterns have been described in mouse for those genes analyzed. In mouse, expression of Fgf3, Fgf4, Fgf8, Fgf17, and Fgfr1 has been reported in and around the NMP region (Gofflot et al., 1997; Wahl et al., 2007; Anderson et al., 2016a). In zebrafish, in addition to fgf4 and fgf8, fgf17, fgf17b, and fgf24 (a zebrafish exclusive gene) have also been shown in or near the tailbud (Reifers et al., 2000; Draper et al., 2003; Cao et al., 2004; Akiyama et al., 2014).

Very little is known about the regulation of expression of FGFRs and FGFs. Most work has been done with FGF8, but the control of FGF8 transcription in the spinal cord NMP region and its progressive downregulation coordinated with embryonic axial extension still constitutes an unsolved enigma (**Figure 3**). Three regions have been identified with respect to FGF8 expression: the most caudal region where FGF8 is actively transcribed (NMP), a more rostral region where transcription is stopped but transcripts remain (known as the transition zone or the preneural tube) and a third most rostral region, adjacent to the mesoderm ready to segment where transcripts are no longer detected (**Figures 3A–C**).

Several signaling pathways have been shown to influence Fgf8 expression either promoting or decreasing Fgf8 levels (**Figure 3D**). The WNT/β-Catenin pathway is active in the caudal region (Aulehla et al., 2003; Olivera-Martinez and Storey, 2007; Cunningham et al., 2011) and manipulation of the pathway has been shown to affect Fgf8 expression. Reduced levels of Fgf8 have been shown in the Wnt3a mouse mutant vestigial tail (Aulehla et al., 2003) and a further reduction is observed in double Wnt3a/Wnt8a mutants (Cunningham et al., 2015b). Furthermore, altering the levels of β-Catenin in the PSM promotes changes in Fgf8 expression (Aulehla et al., 2008; Dunty et al., 2008). In fact, studies in mouse craniofacial development support a direct role for WNT in Fgf8 regulation throughout a conserved Tcf/Lef site 2.8 kb upstream of Fgf8 (Wang et al., 2011). However, no upregulation of FGF8 by WNT has been observed in chick spinal cord suggesting a more indirect regulation of FGF8 by the WNT in the context of caudal neural tube (Olivera-Martinez and Storey, 2007). Evidence has also been presented for the requirement of signals from the notochord. In particular, a reduced level of FGF8 is observed in the absence of the notochord that can be rescued by SHH supplementation (Resende et al., 2010).

An autoregulatory mechanism of FGF activating FGF8 has been suggested based on the ability of FGF8 to activate the transcriptional repressor NKX1.2 (previously known as SAX1; Bertrand et al., 2000) which when overexpressed can in turn result in increased FGF8 levels (Sasai et al., 2014). However, exposure of the neural explants or embryos to FGF does not result in activation of FGF8 expression and inhibition of FGF signaling does not result in decreased FGF8 or Fgf8 levels in chick and mouse, respectively (Harrison et al., 2011; Patel et al., 2013), raising the possibility that NKX1.2 may be involved in the stabilization of FGF8/Fgf8 transcripts.

In addition, although a localized source of a regulator of FGF8 caudal to Hensen's node has been ruled out (Dubrulle and Pourquie, 2004; Harrison et al., 2011), other caudally active pathways such the HMG-CoA reductase/mevalonate pathway (mediating steroid biogenesis) could be also playing a role in FGF8 expression activation and maintenance (Olivera-Martinez et al., 2014).

Recent efforts for the characterization of the Fgf8 gene regulatory region have led to the identification of several regions driving expression around the NMP region (Beermann et al., 2006; Marinic et al., 2013) and furthermore to the identification of CDX2 and TBRA as direct transcriptional activators (Amin et al., 2016). This could explain the decreased Fgf8 levels that have been observed in Cdx2 mutants (Savory et al., 2009) but additional activators may also be acting to regulate Fgf8, such as WNT/β–Catenin.

In addition to signals maintaining FGF8/Fgf8 expression in the caudal precursor region, progressive downregulation of FGF8/Fgf8 involves cessation of transcription in cells that exit the NMP region. Retinoic acid (RA; which is produced by somites and rostral presomitic mesoderm) has been shown to downregulate FGF8/Fgf8 and reduction in RA signaling (in a vitamin A deprived quail model and in Raldh2−/<sup>−</sup> mutants) results in a rostral expansion of the FGF8/Fgf8 expression domain (Diez del Corral et al., 2003; Molotkova et al., 2005; Vermot and Pourquie, 2005; Sirbu and Duester, 2006; Olivera-Martinez and Storey, 2007; Patel et al., 2013; Kumar and Duester, 2014; Cunningham et al., 2015a). This effect of RA has recently been attributed to direct binding of RA receptor (RAR) to an RA response-element (RARE) in the regulatory region in the Fgf8 promoter (Kumar and Duester, 2014; Cunningham and Duester, 2015). This constitutes one of the few examples described where RAR bound to RA would repress gene transcription. Furthermore, additional studies show how NCOR repressors are required for RA repression of Fgf8 (Kumar et al., 2016).

However, as the forced reduction in RA signaling only promotes a limited expansion of the FGF8/Fgf8 domain, additional mechanisms of transcriptional repression must be involved. One possible theoretical mechanism proposed would involve a caudal diffusing signal transcribed in NMP cells that would repress both its own transcription as well as FGF8/Fgf8 (Harrison et al., 2011). However, to date no such signal has been identified. Moreover, failure of FGF8/Fgf8 to be expressed more anteriorly in spinal cord and somites could be due to a lack of transcriptional activators such as TBRA, CDX, and WNT expression.

Interestingly, the change in the transcriptional state of Fgf8 locus from active to inactive is associated to a change in its nuclear position from a more central location in the NMP to a more peripheral position in neural tube cells (Patel et al., 2013). This location seems to be regulated by FGF signaling per se as inhibiting FGFR signaling results in a more peripheral location of Fgf8 transcription in the caudal region. However, in spite of the change of location, transcription of Fgf8 still occurs (Harrison et al., 2011; Patel et al., 2013) suggesting that location of the Fgf8 locus to the periphery is required but is notsufficient for cessation of expression.

As mentioned above, analysis of active transcription by in situ hybridization using intronic FGF8/Fgf8 probes has shown that the actively transcribing region is rather limited and that cessation of FGF8/Fgf8 transcription seems to be abrupt (Dubrulle and Pourquie, 2004). However, the high stability of the transcript (which can perdure more than 5 h) is such that a gradient of FGF8/Fgf8 transcripts can be generated. The mechanism accounting for this high FGF8/Fgf8 stability however has not been further explored.

Final steps in the regulation of FGF8/Fgf8 expression are the decrease in FGF8/Fgf8 levels associated to the trunk to tail transition and the termination of transcription in the tailbud associated to the termination of axis elongation. Mouse embryos with mutations associated to a prolonged trunk extension (Gdf11 loss of function mutants or overexpression of OCT4; Aires et al., 2016) show an abnormal increase in Fgf8 expression in the tailbud region. In the chick, RA derived from the tailbud is also important for the correct termination of FGF8 expression (Olivera-Martinez et al., 2012) while in the mouse other mechanisms seem to be responsible (Cunningham et al., 2011), as body axis extension continues for a much longer time to form the tail in mouse.

As FGFs are secreted factors, their distribution also depends on their diffusion and transport in the extracellular medium. This has been examined in detail in zebrafish embryos mostly in the context of gastrulation but this may be extended to other situations (reviewed in Bokel and Brand, 2013). There, the binding to heparan sulfates, important constituents of the extracellular matrix, is not only relevant for the activation of the receptor by the ligands but also has an influence on the spread of Fgfs (Yu et al., 2009). The shape of the gradient is also greatly influenced by degradation of Fgf8 that may be largely due to its endocytic removal (Scholpp and Brand, 2004).

Overall, a complex gene regulatory network is in place in the NMP region involving several interconnected signaling pathways that ensures that FGF signaling components are expressed at the appropriate levels for the control of a number of processes that take place as the neural tube extends to form the spinal cord. Given the temporally controlled exposure of cells to FGFs in and around the NMP region, the measurement of the activity of the pathway at the level of the receptor as well as at the different downstream components in fixed tissue and in vivo, as the axis extends caudally, would greatly improve our understanding of the coordination of morphogenetic movements and the control of tissue differentiation.

## FGFs AND THE ESTABLISHMENT OF THE CAUDAL NEUROMESODERMAL PROGENITORS AND OF THE SPINAL CORD IDENTITY

It has recently become clear that the spinal cord derives progressively from the caudal NMP region which is specified through FGF and WNT actions on sensitized epiblast cells around the primitive streak (Henrique et al., 2015). Specification of the NMP region is initiated at gastrulation stages (Muhr et al., 1999; Delfino-Machin et al., 2005; Nordstrom et al., 2006) within a region around the node and primitive streak characterized by expression of genes such as NKX1.2, CDX1, CDX2, CDX4, and HOXB8 where high MAPK signaling levels are present. This coincidence is maintained during axis extension (Delfino-Machin et al., 2005; Lunn et al., 2007) and reflects the activity of FGF in the control of the expression of those genes (Storey et al., 1998; Muhr et al., 1999; Bel-Vialar et al., 2002; Delfino-Machin et al., 2005; Nordstrom et al., 2006; Sasai et al., 2014). Thus, in the chick embryo, blockade of FGF signaling with a dominant negative form of FGFR (DN-FGFR) and with pharmacological inhibitors results in downregulation of NKX1.2 and HOXB8 in vivo and in explant cultures (Delfino-Machin et al., 2005). In double Fgf4; Fgf8 conditional mutant mice, there is a decrease in Wnt3a, Wnt5a, Cyp26a1, T-Bra in the NMP region (Naiche et al., 2011; Boulet and Capecchi, 2012). Similarly, Fgfr1 conditional mutants also show decreased levels of a number of NMP region genes, such as Gbx2 and Cyp26a1 (Wahl et al., 2007). All these results, from the current perspective that stresses the relevance of the NMP cells, suggest that FGF contributes in an important way to the specification of the NMP character in chick and mouse embryos, including genes expressed in NMP and its mesodermal derivatives (T-BRA) as well as those expressed in NMP and its neural derivatives (i.e., NKX1.2). In the same direction, in Xenopus and zebrafish embryos, expression of a dominant negative form of FGFR/Fgfr (DN-FGFR/DN-Fgfr) results in the loss of markers of NMP and its derivatives (Isaacs et al., 1994; Griffin et al., 1995; Holowacz and Sokol, 1999; Ota et al., 2009).

The NMP give rise to both spinal cord and mesodermal cells during an extended period of time and FGF levels also contribute to preserve the balance between the three cell types. For instance, double Fgf4; Fgf8 conditional mutant mice where defective signaling is restricted to the NMP and its derivatives display dramatic reduction of the presomitic mesoderm markers Tbx6 (Naiche et al., 2011; Boulet and Capecchi, 2012) and display ectopic neural tubes (Boulet and Capecchi, 2012) similar to the ones observed in Fgfr mutant chimeras (Ciruna et al., 1997). In chick, pharmacological inhibition of FGFR results in precocious and caudal expression of the neural tube specific gene SOX1 (Stavridis et al., 2010). On the other hand, situations with excessive caudal FGF8 signaling such as the Raldh2 mutant present an imbalanced NMP differentiation favoring mesodermal fate (Cunningham et al., 2015a).

This suggests a requirement of FGF signaling for the promotion of mesodermal or neuromesodermal vs. neural fates (Henrique et al., 2015). Most interestingly, FGF signaling has been shown recently to promote the expression of enzymes that drive the glycolytic metabolic state of the NMP region (Oginuma et al., 2017) that is in turn important for WNT signaling and for restraining the transition from a NMP state to a neural state (Oginuma et al., 2017). Later on, that glycolytic metabolic state in a gradient fashion also operates in presomitic mesoderm development (Bulusu et al., 2017).

In spite of FGF promotion of neuromesodermal and mesodermal fates, FGF signaling in combination with WNT signaling also appears to contribute to the activation and maintenance of the expression of the neural genes SOX2 and SOX3 through specific gene regulatory regions (Takemoto et al., 2006; Nishimura et al., 2012) and this might help to prevent the excess production of mesoderm precursors from the NMP (Yoshida et al., 2014).

The precise sequence of exposure of cells to FGF in combination with the other caudal signal WNT as well as the temporal dynamics within the cells may here determine whether cells are maintained in a NMP state, differentiate toward a mesodermal fate or toward a neural fate. This idea has been recently explored with experiments developing in vitro methods to generate a population of cells that co-express the NMP genes from mouse and human pluripotent stem cells by timed exposure to FGF2 in combination with WNTs (Gouti et al., 2014; Turner et al., 2014; reviewed in Henrique et al., 2015). This constitutes a good example of how the temporal exposure and competence to interpret FGF signals play an important role in specification of cell fates.

Once spinal cord cells leave the NMP region, FGF is not required for the maintenance of the spinal cord identity. Thus, the spinal cord specific homeobox transcription factor HOXB8, that initially requires FGF for its expression in the NMP (Delfino-Machin et al., 2005), remains actively expressed in spinal cord progenitors after the levels of FGF signaling have dropped during axis elongation.

The different mechanisms responsible for the role of FGF in specification of the NMP and then in the balance of mesodermal and neural derivatives may be related to the coactivity with other signals and/or to the temporal sequence of exposure and response of cells to FGF and other signals, as suggested by the cell culture experiments. All these crucial aspects certainly deserve now a thorough analysis within the developing embryo.

## FGFs AND THE CONTROL OF SPINAL CORD CAUDAL EXTENSION

The most striking feature of embryos where FGF signaling has been diminished (once the early lethality is overcome) is the truncation of the caudal embryonic axis, observed in mouse, Xenopus and zebrafish. Mouse Fgfr1−/<sup>−</sup> embryonic chimeras cannot gastrulate properly and mutant cells tend to accumulate in the tail displaying a short axis (Ciruna et al., 1997; Ciruna and Rossant, 2001). Similarly, Fgfr1 conditional mutant mice where defective signaling is restricted to the caudal NMP and derivatives (using a TBra- driven Cre-line), result in truncated axis at the level of sacral regions (Wahl et al., 2007). An even shorter axis is observed in the double Fgf4; Fgf8 conditional knock-out (Naiche et al., 2011; Boulet and Capecchi, 2012; either using a TBra- or a Hoxb1-driven Cre-lines). A shortened tail is also apparent in the Fgf3 null mutant embryos (Anderson et al., 2016a,b). Similarly, in Xenopus and zebrafish the overexpression of DN-FGFR/DN-Fgfr versions also result in truncated embryos (Griffin et al., 1995; Holowacz and Sokol, 1999) and in chick decreased elongation rates have been observed following blockade of FGFR (Benazeraf et al., 2010).

However, in all these situations the lack of FGF signaling affects specification of both mesodermal and neural derivatives and it is therefore not possible to assess whether the defect on elongation is a consequence of an alteration in gastrulation, in the specification of NMP, spinal cord or mesoderm, the result of abnormal motility in the mesoderm (Benazeraf et al., 2010) or whether there is a more specific requirement within the spinal cord population. Support for a more localized role of FGF signaling in spinal cord caudal extension came from analysis of cell distribution after electroporating a DN-FGFR1 construct in chick NMP region (Mathis et al., 2001; therein referred to as node region). In control experiments, cells could either remain in the NMP region and continue the backward displacement or get incorporated into the neural tube. However, cells with decreased FGF signaling had an increased probability to get incorporated in the neural tube and thus would not be part of the caudally displaced NMP region suggesting some changes in cell adhesion properties of those cells, at least indirectly. In addition, a role of FGF in the maintenance of proliferating cells could also contribute to the extension of the axis (Mathis et al., 2001).

In presomitic mesoderm, axis extension has been shown to involve differential motility of cells along the rostrocaudal axis in a space constrained by lateral boundaries (possibly the lateral plate), with cells moving more in caudal presomitic mesoderm than in the rostral part. Interestingly, in that context, FGF signaling has been shown to promote cell motility (Benazeraf et al., 2010; Lawton et al., 2013). As mentioned before, FGF is required for the transcription of rate limiting enzymes responsible for the glycolytic metabolic state of the NMP that has been shown to be important for cell motility and axis elongation (Oginuma et al., 2017). The mechanism of control of cell motility is still not known but it has been proposed to be related to the ability of localized glycolytic activity to ensure rapid production of ATP for actin polymerization in the forming protrusions of motile cells (Oginuma et al., 2017). In other contexts, FGF has been shown to have chemotaxis properties (Yang et al., 2002) and this has been suggested as an additional mechanism that could in theory contribute to axis extension (Harrison et al., 2011). In any case, given that the spinal cord is composed by epithelial cells and not by mesenchymal cells (as it is the case for presomitic mesoderm) it is unlikely that the same morphogenetic mechanisms are responsible for its extension which may be a more passive process driven by mesoderm.

Recent work on the generation of a population with NMP properties by differentiation of mouse embryonic stem cells (mESCs) in adherent cell culture has shown that these cell aggregates also have the ability to elongate in vitro and that this elongation requires FGF signaling, providing an in vitro system where this function can be further examined (Turner et al., 2014). In conclusion, there are still many unknowns in relation to the cellular process of spinal cord extension. Most likely, the combination of in vitro culture systems together with imaging techniques (both in vitro and in vivo), the use of biosensors to investigate metabolism in developing embryos (such as the PYRATES mouse line, Bulusu et al., 2017) and in silico simulations will greatly contribute to the understanding of the important role of FGF signaling in spinal cord extension.

## FGFs AND THE CONTROL OF CELL PROLIFERATION, CELL CYCLE EXIT AND NEURONAL DIFFERENTIATION

FGFs play important roles in cell survival and proliferation in many developmental contexts and in particular for neural stem cells and progenitors (Vaccarino et al., 1999; Storm et al., 2006; Maric et al., 2007). In the developing spinal cord, analysis of cell cycle exit (Sechrist and Bronner-Fraser, 1991) and of the early postmitotic marker NeuroM (Roztocil et al., 1997) revealed two regions with respect to cell proliferation. NeuroM<sup>+</sup> cells start to appear in the region flanked by somites while no NeuroM<sup>+</sup> cells are found in the more caudal region (the preneural tube) nor in the NMP region, coinciding with the region of influence of FGF signaling (Diez del Corral et al., 2002).

Exposure of the neural tube to FGF at a stage when some cells are already exiting the cell cycle can impair the generation of new NeuroM expressing cells (Diez del Corral et al., 2002) and by that way, the onset of neurogenesis. By following the fate of neural progenitors using time lapse imaging, it has been possible to analyze the changes in the dynamics of progenitors associated to FGF exposure (Wilcock et al., 2007). Neural progenitors and stem cells can normally experience three modes of division to give rise to neurons (N) and progenitors and stem cells (P): self-expanding, PP (i.e., giving rise to 2 progenitors or stem cells); self-replacing, PN (i.e., giving rise to a progenitor and a neuron); and self-consuming, NN (i.e., giving rise to 2 neurons). Previous studies in the developing cortex and spinal cord suggest that different modes are associated with different cell cycle duration times, with neuron generating divisions (PN or NN) characterized by a longer cell cycle than PP divisions (Takahashi et al., 1995; Calegari and Huttner, 2003; Calegari et al., 2005; Wilcock et al., 2007).

Upon exposure to FGFs, progenitors only go through PP divisions while no PN nor NN divisions could be observed (Wilcock et al., 2007). These FGF induced PP divisions exhibited the typical short PP cell cycle length while no changes in the range of cleavage plane orientation were observed. Interestingly, a subpopulation of cells was found dividing without contacting the apical membrane and with very short cell cycle times (Wilcock et al., 2007). These data support a role for FGF in the maintenance of cells characterized by a rapid cell cycle that can only generate further progenitors. Interestingly, within the embryo, shorter cell cycle lengths are observed in the region exposed to FGF with respect to the rostral neural tube (Olivera-Martinez et al., 2014) and several cell cycle genes are differentially expressed in the caudal vs. more rostral region and could be regulated by FGF (Lobjois et al., 2004; Olivera-Martinez et al., 2014). One example is CYCLIN D2, a cell cycle regulator specifically expressed in the chicken caudal neural plate that can be activated by and requires FGF signaling (Lobjois et al., 2004; Molina and Pituello, in press).

Although exposure to FGF can impede neurogenesis, blockade of FGF signal in explants is not sufficient to drive premature expression of the postmitotic and neurogenesis marker NeuroM (Diez del Corral et al., 2002). However, as discussed above, cells subject to interference with FGF signaling in the embryo tend to prematurely leave the NMP region (Mathis et al., 2001) where only proliferating cells are found and it remains to be assessed whether they have alterations in their type of division or cell cycle exit parameters.

A high level of aerobic glycolysis is known to facilitate cancer cell proliferation. Although no significant change in proliferation was observed by Oginuma et al. (2017) in embryos grown in the absence of glucose, more detailed analysis are required in order to determine a possible implication of the FGF dependent changes in metabolism in the control of proliferation during axis extension and more specifically within the spinal cord.

The contribution of FGF to the control of proliferation in the spinal cord discussed above is restricted to cells before or at the onset of neurogenesis and could be equivalent to the ability of FGF2 and FGF8 in the telencephalon to maintain the proliferative symmetrical PP divisions of neuroepithelial cells before the onset of neurogenesis (Raballo et al., 2000; Storm et al., 2006; Maric et al., 2007; Rash et al., 2013). Interestingly, the analysis of the telencephalon of mutant mouse embryos has revealed additional requirements for FGF signaling in proliferation of neurogenic lineages at different steps. At the start of telencephalon neurogenesis, neuroepithelial cells transform into radial glial cells, which divide asymmetrically to generate another radial glia and a postmitotic neuron or a basal progenitor (Gotz and Huttner, 2005) and this transition is promoted by FGF10 (Sahara and O'Leary, 2009). Finally, after neurogenesis has started, it has been demonstrated (using mutants for three FGF receptors) that FGF signaling is required to slow down the progression from radial glia to basal progenitors (Kang et al., 2009; Rash et al., 2011). Similar roles for FGF at later stages of spinal cord development remain to be explored (see below for functions during spinal cord adult neurogenesis).

In addition to a more direct action of FGF on the cell cycle, several FGF dependent pathways could mediate its influence on cell cycle exit and neuronal differentiation before the onset of neurogenesis in the spinal cord. FGF signaling is required for the expression of DELTA-1, an important component of the NOTCH signaling pathway involved in mutual inhibition in the NMP region and required to limit precocious cell cycle exit (Akai et al., 2005). Additionally, FGF signaling promotes WNT8a expression, which in turn prevents neuronal differentiation (Olivera-Martinez and Storey, 2007).

Manipulation of FGF signaling in chick embryo explants and the use of mouse mutants has shown that FGFs can reduce the levels of RA signaling, a neuronal differentiation promoter (reviewed in Diez del Corral and Morales, 2014) and this would also favor the maintenance of the progenitor state. Double Fgf4; Fgf8 conditional mutant mouse embryos exhibit increased caudal RARE-lacZ reporter expression (Naiche et al., 2011). But, at what level could FGF act on the control of RA signaling? FGF4 and FGF8 can repress the gene encoding the RA-synthesizing enzyme RALDH2 in the paraxial mesoderm (Diez del Corral et al., 2003). However, the contribution of this repression to the RA levels is probably partial since double Fgf4; Fgf8 conditional mutant mouse embryos do not exhibit increased Raldh2 expression (Boulet and Capecchi, 2012). FGF signaling is required for the caudal expression of the RA-degrading enzyme Cyp26a1 (Wahl et al., 2007) and this could also contribute to the control of RA levels similarly to what has been described in the context of the hindbrain (Gonzalez-Quevedo et al., 2010). FGF4 and FGF8 can also downregulate RARβ receptor levels in the spinal cord (Olivera-Martinez and Storey, 2007) and this would affect the sensitivity to RA levels. This receptor gene depends on RA for its activation (Olivera-Martinez and Storey, 2007) and thus its downregulation by FGF could be due to upregulation of Cyp26a1 but this has not been examined yet.

FGF signaling is also required to prevent precocious activation of PAX6 and IRX3 in chick and Pax6 in mouse (Bertrand et al., 2000; Diez del Corral et al., 2003; Patel et al., 2013), two transcription factors which promote neuronal differentiation (de la Calle-Mustienes et al., 2002; Bel-Vialar et al., 2007). Thus, FGF seems to contribute to a rather complex network that controls proliferation before the onset of neurogenesis maintaining an undifferentiated state. However, open questions still remain: does FGF signaling act differentially on the process of proliferation within NMP and then for promotion of self-renewal of neural progenitors? Does it act differently in the spinal cord than in telencephalon progenitors where it has also been involved in the appearance of intermediate progenitors? What are the cell cycle components modulated by FGF signaling in all these processes?

## FGFs AND PATTERNING OF SPINAL CORD ALONG THE ROSTRO-CAUDAL AXIS

Once the region of the neural plate giving rise to the spinal cord has been specified (in an FGF dependent way), FGF signaling has an additional role in the further regionalization of the spinal cord along the rostral-caudal axis. The spinal cord presents heterogeneity along the rostro-caudal axis responsible for differences in motor neuron subpopulations, interneuron distribution (Francius et al., 2013; Lai et al., 2016) or neural crest derivatives (Le Douarin et al., 2004). This regionalization, which has been mainly examined in motor neurons, is a consequence of the restricted rostro-caudal expression of Hox genes in progenitor cells and subsequently in the resulting postmitotic motor neurons (reviewed in Philippidou and Dasen, 2013).

Experiments in chick embryos have shown that exposure to FGF or electroporation of FGFs expressing constructs shifts rostrally the domain of expression of caudal HOX mRNAs (HOXB6, HOXC6, HOXB7, HOXB8, and HOXA9-B9-C9) in neural progenitors resulting in an increase in the protein levels of a subset of HOXB proteins (Bel-Vialar et al., 2002; Dasen et al., 2003). FGF signaling appears to act here by activating the transcription factor genes of the Cdx family, known to activate HOX/Hox gene expression, in particular cdx2 and cdx4 in zebrafish (Shimizu et al., 2006), CDX1 and CDX2 in the chick (Bel-Vialar et al., 2002), and Cdx1, Cdx2, and Cdx4 in mouse (van den Akker et al., 2002; Amin et al., 2016). Exposure to FGF not only has consequences in the expression of genes in progenitors but also in the resulting motor neurons (Liu et al., 2001; Dasen et al., 2003). Explants of neural tissue fated to give rise to cervical spinal cord do not express HOXC6, HOXC8, HOXC9, or HOXC10 after culture but their exposure to increasing FGF levels results in progressive activation of the production of these proteins suggesting that FGF works in a concentration dependent way. Considering that in the embryo, caudal cells are exposed to FGF for a longer period of time than rostral cells, but not necessarily to higher levels of FGF signaling, this concentration dependent effect has also been interpreted as an effect of the duration of exposure to the FGF morphogen. The mechanism to explain such concentration/time of exposure dependence is still not known but may involve the regulation of genes encoding transcription factors of the CDX family mentioned above.

The role of FGF signaling in this further caudalization, however, has not yet been ascertained by loss of function approaches and therefore, the extent of its contribution to patterning remains an open question. A possible contribution of FGF to rostro-caudal patterning of interneurons has also been suggested (Francius et al., 2013) but has not been explored yet.

## FGFs AND VENTRAL PATTERN (INTERMEDIATE, VENTRAL AND FLOOR PLATE)

Another regionalization process where FGF signaling plays an essential role is the patterning of the spinal cord along the dorso-ventral (DV) axis which is fundamental for the assignment of neuronal subtype identities such as motor neurons and the different interneuron subtypes (reviewed in Gouti et al., 2015). Specific combinations of transcription factors of the homeodomain and bHLH families are expressed in restricted domains along the DV axis (reviewed in Le Dreau and Marti, 2012). In the ventral/intermediate neural tube this is regulated by the SHH morphogen. The graded distribution of SHH, produced in the ventral midline, results in a graded activation of the pathway and the expression of target genes (reviewed in Briscoe and Small, 2015). In addition, cross-repressive interactions between target genes occur to further delimit and ensure gene expression in the appropriate domains (Briscoe et al., 2000; Kutejova et al., 2016).

During spinal cord caudal extension, SHH is expressed in the node and along the derived notochord while in the neural tissue it is expressed in floor plate (FP) cells at the level of the somitic mesoderm. Thus, cells in the preneural tube (the transient spinal cord population derived from NMP and adjacent to presomitic mesoderm) are initially exposed to notochord derived SHH and express some SHH target genes such as GLI1, PTCH1 and PTCH2 suggesting that at least low SHH signaling is achieved (Diez del Corral et al., 2003; Morales et al., 2016). However, neural progenitors in the preneural tube do not display expression of the complete repertoire of ventral identity genes, suggesting that the pathway is being modulated in this region. A role for FGF signaling in the control of ventral patterning was first inferred from its ability to repress PAX6, a gene expressed in an intermediate domain in the neural tube (Bertrand et al., 2000). Since that observation, a more complex picture has emerged showing that FGF signaling is crucial for controlling the onset of SHH signaling and ventral patterning in the spinal cord (Diez del Corral et al., 2003; Morales et al., 2016) and for the early specification of the most ventral fate, the FP (Sasai et al., 2014; **Figure 4**).

Forced maintenance of FGF signaling in preneural tube tissue, impairs not only PAX6 but also other ventral and intermediate patterning genes such as NKX6.1, NKX6.2, IRX3, and FOXA2 (Bertrand et al., 2000; Diez del Corral et al., 2003; Novitch et al., 2003). Conversely, interference with FGF signaling in chick embryos results in precocious caudal activation of PAX6 and IRX3 (Bertrand et al., 2000; Diez del Corral et al., 2003) and in the dorsal expansion of ventral markers such as OLIG2 and NKX6.1 (Morales et al., 2016). Reduced FGF signaling in mouse embryos results in precocious caudal Pax6 (Patel et al., 2013) and NKX6.1 expression as well as in alterations in the ventral patterning with an increase in the number of NKX6.1 expressing neural progenitor cells (Morales et al., 2016).

FGF would thus be repressing more or less indirectly the two types of SHH responding genes, ventral genes activated by SHH (FOXA2 and NKX6.1) and intermediate genes repressed by SHH (PAX6 and IRX3) (Briscoe et al., 2000). Repression of PAX6 and IRX3 and their mouse homologs seems to involve several mechanisms (**Figures 4B,C**). FGF signaling promotes chromatin compaction and peripheral nuclear position around the mouse Pax6 and Irx3 loci, a chromatin organization associated to transcriptionally inactive loci (Patel et al., 2013). In addition, repression of these genes appears to be mediated by transcriptional repressor NKX1.2, transcriptionally activated by FGF signaling in NMP region (Storey et al., 1998; Bertrand et al., 2000; Sasai et al., 2014; **Figure 4C**).

A molecular mechanism that accounts for the effect of FGF on genes relying on SHH for their expression has been identified recently (**Figure 4C**; Morales et al., 2016). FGF can activate the expression of PTCH2, one of the SHH receptors that also acts as an inhibitor of the SHH pathway, and can thus restrain expression of SHH targets. Experiments in chick explants have shown that PTCH2 is expressed in the preneural tube in a SHH and FGF dependent way indicating the existence of an enhanced feedback loop where SHH activates PTCH2 more efficiently in regions of high FGF signaling. This regulation also appears to be conserved in mouse as Fgfr1 conditional mutant embryos show extremely reduced Ptch2 levels (Morales et al., 2016). Surprisingly, however, the Ptch2 gene seems to be largely dispensable as no obvious phenotype has been identified yet in the mouse mutant (Holtz et al., 2013). It is possible that its function is only apparent when the development of the embryo is challenged, for example when the elongation process is altered. If elongation is arrested, the high levels of PTCH2 in the spinal cord precursor may maintain the levels of SHH signaling low and ventral patterning on standby mode until the elongation is restored (**Figures 4D,E**). In fact, Ptch2 is required to keep SHH signaling in check in situations of partial deficiency of the other member of the family, Ptch1, both in development and in tumorigenesis (Lee et al., 2006; Nieuwenhuis et al., 2006; Holtz et al., 2013; Zhulyn et al., 2015).

In addition, and probably as a result of its role on repressing ventral and intermediate genes, an important role for FGF on the specification of FP cells has been recently identified in chick

FIGURE 4 | Role of FGF signaling in the regionalization of the intermediate and ventral neural tube. (A) Diagram representing the caudal part of an embryo showing the region with active FGF signaling and the rostro-caudal level of the transverse section represented in (B). (B) Transverse section at the level of the preneural tube showing the neural tissue and the underlying presomitic mesoderm and notochord. FGF8 (produced by neural and mesoderm tissues) and Shh (produced by notochord) are represented in purple and green respectively and their influence on intermediate, ventral and floorplate specification is shown. The dashed arrow indicates that FGF provides competence for floor plate specification. (C) Gene regulatory network relating FGF and Shh signaling in the pre-neural tube where the two signals coincide. Data from chick, mouse or both are included in this figure. (D,E) Graphs to illustrate the hypothetical role of the regulation of Ptch2 by FGF during the initial establishment of the Shh signaling levels. The graphs represent the changing levels of SHH in time at a particular position (within the dorsoventral and rostrocaudal axis). (D) In embryos continuously extending their caudal axis, the levels of FGF (purple) at a particular position would decrease constantly while the levels of SHH (green, SHHNormal) would increase until they reach their maximum. The levels of Shh signaling thus also increase progressively. (E) In embryos where elongation is arrested for some time (shaded area), FGF levels would remain constant during the arrested period, while the levels of SHH (SHHPerturbed) would accumulate more rapidly due to the decreased amount of tissue generated through which SHH could diffuse (or be transported). If Shh signaling was dependent exclusively on SHH levels, the level of the signaling would also increase to levels higher than normal and this may result in the irreversible activation of its targets. However, the ability of FGF to activate Ptch2 and thus downregulate the Shh pathway could serve to limit Shh signaling levels to normal values. FGF signaling levels (decaying in time) are shown in purple.

(Sasai et al., 2014). FP territory, characterized by expression of the ARX1 protein, is induced by the highest levels of SHH that are only achieved in the cells closest to its source and also requires transient FGF exposure (Sasai et al., 2014). Here again, NKX1.2 plays an important role, providing competence to respond to high SHH levels and drive ARX1 expression. Given the repressive interactions between FP specific genes and the ventral and intermediate patterning genes (Cho et al., 2014; Kutejova et al., 2016), one important function for FGF signaling and NKX1.2 here would be to ensure that a region free of expression of nonfloor plate factors such as PAX6, IRX3 and NKX2.2 is established in the future FP region (Sasai et al., 2014). Nevertheless, the details of the gene regulatory network are still not elucidated as expression of ARX1 (and other definitive floor plate markers) is only apparent well after the FGF signaling levels have decayed. Here again, the system may be highly redundant as no obvious alterations in the FP have been reported in Nkx1.2 mutant mice (Simon and Lufkin, 2003), raising the possibility that the related Nkx1.1 gene could be also playing a role.

Antagonism of the FGF signaling pathway with the RA pathway is also important in the context of ventral patterning as RA is required for expression of several intermediate and ventral genes (chick NKX6.1, IRX3, PAX6, OLIG2; Diez del Corral et al., 2003; Novitch et al., 2003; Diez del Corral and Morales, 2014 and mouse Nkx6.1, Pax6, and Olig2; Molotkova et al., 2005). However, the temporal and quantitative contributions of both FGF and RA pathways in the modulation of ventral specification require a deeper analysis.

## FGFs AND NEURAL CREST SPECIFICATION

At the most dorsal part of the spinal cord, the development of a specific cell population also requires the participation of FGF signaling: the neural crest cells (NCCs). The neural crest is formed by a transient population of multipotent cells that arise from the dorsal neural tube. Once specified, NCCs undergo a process of epithelium to mesenchyme transition (EMT) that confers NCCs the ability to delaminate and migrate away from the dorsal neural tube, giving rise to NCC derivatives that include craniofacial skeleton, the peripheral nervous system (sensory neurons and glia, sympathetic neurons) and melanocytes, amongst others (Le Douarin and Kalcheim, 1999).

The process of neural crest formation implies the orchestration of a complex gene regulatory network. This involves signaling pathways and transcription factors that are responsible for the sequence of early induction of the NCC during gastrulation; the specification of the neural plate border; the expression of bona fide NCC transcription factors and the regulation of numerous downstream effectors involved in EMT, cell adhesion, and cell cycle control, amongst others (Morales et al., 2005; Sauka-Spengler and Bronner-Fraser, 2008). First, parallel to the induction and patterning of the neural plate that generates the central nervous system, at the border between the neural ectoderm and the non-neural ectoderm, the NCCs are specified through a series of steps controlled by FGF, WNT, and BMP signaling pathways (reviewed in Saint-Jeannet and Moody, 2014).

Transient exposure to FGF has been shown to allow neural tube cells to activate NCC markers in response to BMP (Sasai et al., 2014). This seems related to the ability of FGF to repress PAX6 and IRX3, two intermediate neural tube genes which can repress the NCC marker SNAIL (Sasai et al., 2014). It has been proposed that the repression of IRX3 and PAX6 by FGF, acting through activation of NKX1.2, is required for the early establishment of a territory competent to NCC specification (see the ventral patterning section for a further discussion on possible mechanisms for FGF regulation of PAX6 and IRX3). However, in FGF deficient conditions impaired NCC specification in vivo has not been reported yet. On the contrary, forced reduction of FGF signaling allowed neuroepithelial cells to prematurely initiate the expression of the early NCC specifier SNAIL2 at caudal levels (Martinez-Morales et al., 2011). This indicates that dorsal neuroepithelial progenitors in the caudal neural tube are maintained in an uncommitted non-NCC state in presence of strong FGF/MAPK signaling pathway (Martinez-Morales et al., 2011). Thus, in the elongating neural tube, as the dorsal neuroepithelial progenitors are progressively exposed to decreasing FGF signaling levels, they initiate the expression of neural crest specifier genes SNAIL2 and FOXD3.

Interestingly, upon reduction of FGF signaling, when those prematurely SNAIL2 expressing NCCs initiate the expression of other NCCs specifiers such as FOXD3, SOX5, and SOX10 they prematurely start EMT from the neural tube at mid-rostral PSM levels. Essentially, the regulated decrease in FGF signaling is primary responsible for the control of the initiation of NCC specification in the trunk, and as a consequence of that, it controls the timing of EMT and emigration. Subsequent development of trunk NCCs is highly dependent on the development of adjacent somites, which impose a segmented migration and organization to the trunk NCCs and to the derived peripheral nervous system (Sela-Donenfeld and Kalcheim, 1999). Considering that FGF signaling is important both for segmentation of the mesoderm and for the neural crest specification it would constitute an important mechanism of coordination of both tissues.

As it has been described above, FGF and RA signaling can act as opposite gradients, each one negatively regulating the activity of the other. In the context of NCC development, RA signaling produced by the somites does not appear to promote their specification but does trigger the EMT of already specified NCCs (Martinez-Morales et al., 2011). FGF and RA signaling control the timing of EMT and emigration in part through modulation of elements of the BMP and WNT signaling pathways, important signaling cascades operating in the dorsal neural tube (Sela-Donenfeld and Kalcheim, 1999; Burstyn-Cohen et al., 2004). Whereas, RA signaling triggers the initiation of WNT1 expression in the dorsal neural tube at levels where the NCCs are already specified, FGF signaling prevents the premature expression of WNT1 (Martinez-Morales et al., 2011).

Moreover, recently it has been established that another FGF ligand, FGF3, coming from the caudal presomitic mesoderm provides another level of regulation of BMP signaling in the spinal cord at tailbud stages. Fgf3 mutant embryos exhibit axis truncation, increase in neuroepithelial proliferation, delay in neural tube closure and premature neural crest formation (Anderson et al., 2016a). The removal of one copy of NOGGIN, a BMP antagonist, in Fgf3 mutants, exacerbated all the Fgf3 phenotypes including premature neural crest specification. Conversely, genetically decreasing BMP signaling in Fgf3 mutants, via loss of BMP receptor activity, ameliorates morphological defects (Anderson et al., 2016a).

In summary, the data discussed in this section show that there is a limited time window during which the onset of the NCC emigration can be modulated, once those cells have acquired the expression of the essential gene network of the NCC specification program. That window coincides with the region where FGF and RA gradients collide. This FGF function constitutes another example of a general FGF role in controlling the onset of differentiation of cell types as they are generated at the tail end, during trunk axial elongation. Again, the molecular mechanism that allows the cells to interpret and execute that temporal window imposed by FGF signaling is far from understood and remains an important open question within the developmental biology field.

## FGFs AND NEURAL STEM MAINTENANCE IN THE ADULT SPINAL CORD

As we have discussed, during the development of the nervous system, the generation of hundreds of subtypes of neurons and glial cells relies upon the relatively fast production, amplification, specification, and differentiation of a pool of neural progenitors and neural stem cells (NSCs). Surprisingly, this strategy is retained to some extent in niches in the adult nervous system throughout lifetime under physiological conditions to generate specific subtypes of neural cells in limited numbers.

Adult NSCs are maintained into adulthood in two main niches, the ventricular-subventricular zone (V-SVZ) adjacent to the lateral ventricles and the subgranular zone (SGZ) in the hippocampus (reviewed in Fuentealba et al., 2012; Christian et al., 2014). Nevertheless, cells with neural stem cell properties can be isolated from most regions of the adult central nervous system, including, for example, the spinal cord (Weiss et al., 1996; Shihabuddin et al., 1997).

In the adult spinal cord, the cells with neural stem cell properties are the ependymal cells (Johansson et al., 1999; Meletis et al., 2008; Barnabe-Heider et al., 2010; Pfenninger et al., 2011). They rarely proliferate under physiological conditions and they mostly give rise to ependymal progeny in vivo. It is unclear which signals are responsible for maintaining this population of ependymal cells. However, in other neurogenic niches such as the SGZ of the hippocampus dentate gyrus (DG), the specific deletion of all the FGF receptors that are expressed in DG (Fgfr1, Fgfr2, and Fgfr3) in adult precursor cells has shown that, FGF signaling is required for neural stem-cell maintenance while an activated FGF receptor expressed in all precursors can increase the number of neurons produced (Kang and Hebert, 2015). The requirement for FGF receptors in maintaining stem but not progenitor cells in the adult hippocampus is reminiscent of their role in maintaining cortical radial glial stem cells during development (Kang et al., 2009).

In spite of the limited expansion of spinal cord ependymal stem cells under normal physiological conditions, their proliferation is dramatically increased after spinal cord injury, giving rise to scar-forming astrocytes as well as to a small population of remyelinating oligodendrocytes (Johansson et al., 1999; Meletis et al., 2008; Barnabe-Heider et al., 2010). More importantly, the ependymal derived astrocytes are essential for repairing the lesions because if their formation is inhibited, the lesions grow deeper over time and a higher number of axonal tracts are lost (Sabelstrom et al., 2013).

The application of FGF2 has been shown to promote functional recovery after spinal cord injury (SCI) in rodents (Lee et al., 1999; Rabchevsky et al., 1999; Yan et al., 2000; Kim et al., 2006). In SCI the recovery is thought to be due to FGF promoting the proliferation of spinal cord neural stem and progenitor cells expressing PAX6, NESTIN, and SOX2 (Shihabuddin et al., 1997; Goldshmit et al., 2014), promoting neuronal survival (Teng et al., 1998, 1999), angiogenesis (Kang et al., 2013), and causing a reduction in injury volume (Lee et al., 1999; Rabchevsky et al., 1999). In addition, FGF2 may reduce glial scar formation and astrogliosis after SCI in the mouse model (Goldshmit et al., 2014). In this situation, FGF2 influences glial cell activation, generating a proregenerative radial/progenitor-like state rather than reactive astrocytes that form scar tissue that are inhibitory to axonal regeneration. It is unclear if these proliferating astrocytes could be derived from the neural stem ependymal cells.

FGF2 also reduces the inflammatory response, as it causes the reduction in macrophage infiltration and cytokine levels (Goldshmit et al., 2014). The reduction in macrophage infiltration may be due to the ability of FGF-2 to reduce the leakiness of the blood-spinal cord barrier after SCI (Kang et al., 2010). Moreover, in combination with transplanting specific cells (Meijs et al., 2004; Kuo et al., 2011; Guzen et al., 2012; Lu et al., 2012) or with special scaffold forming hydrogels FGF1 and FGF2 can provide a proregenerative effect and may have clinical applications in the treatment of SCI (Chen et al., 2015). In fact, FGF1 is currently in clinical trials in human patients with cervical SCI (Wu et al., 2011) and more recently also in combination with special devices and rehabilitation in patients with thoracic SCI (clinical trial, NIH reference NCT02490501).

Since SCI has multiple factors that determine the progress of the injury, a combinatorial therapeutic approach including FGF will most likely be required for the most effective treatment of SCI (reviewed in Siddiqui et al., 2015; Ahuja et al., 2016).

## FGFs PROMOTING NEUROGENESIS IN A DISH

As a complementary approach and as a way to overcome the limited capacity for self-repair of the mammalian nervous system, efforts are being made to boost the repair process by transplanting exogenous cells into sites of injury (Rosser et al., 2007). FGFs can be used to generate, expand, and differentiate neurons in vitro and therefore have a major role to play in such cell replacement therapies.

First, FGF2 together with EGF has been extensively used to promote proliferation and self-renewal of NSCs in vitro (Kilpatrick and Bartlett, 1993; Gage et al., 1995; Gritti et al., 1996; Qian et al., 1997; Nelson and Svendsen, 2006). FGF2 converts embryonic stem cells into neural stem cells characterized by rapid self-renewing and the potential to generate neurons, astrocytes, and oligodendrocytes. This acquired tripotent neural stem cell state, which does not exist in vivo, provide high proliferative capacity and glial differentiation potential to the treated cells (Palmer et al., 1999; Laywell et al., 2000; Zhang et al., 2001; Gabay et al., 2003; Hack et al., 2004; Pollard et al., 2008). Several studies then showed that FGF2 ventralizes cultured rodent NSCs/NPCs of dorsal origin and induces oligodendrocytes from NSCs derived from regions where oligodendrocytes are not present (Gabay et al., 2003).

FGF2 has also been proved to be involved in neuronal subtype specification, as it has been shown that in-vitro-expanded human fetal forebrain-derived NSCs can generate cholinergic neurons with spinal motor neuron properties when treated with FGF2 within a specific time window (Jordan et al., 2009). Moreover, ESC-derived motor neurons, grown using a differentiation program that relies on endogenous embryoid body-derived WNTS, FGFs, and HH signaling, and then grafted isochronically into chick spinal cord, settle in appropriate columnar domains and select axonal trajectories with a fidelity that matches that of their in vivo generated counterparts (Peljto et al., 2010). Under those differentiation conditions, it is not clear if increasing FGF levels would increase motor neuron yields without sacrificing the columnar and motor pool subtype diversity achieved.

In the last few years induced pluripotent stem cells (iPSCs) have provided a platform for studying basic human development and disease mechanisms and hold great potential for future cell therapies (Murry and Keller, 2008). Nevertheless, biomedical application of iPSCs depends on the availability of robust cell expansion and differentiation protocols. A recent example is the use of FGFR inhibitor (SU5402) that promoted iPSCs to commit to a NCC cell fate that express specific genes, including PAX3, SLUG, TFAP-2α, and TWIST1 (Jaroonwitchawan et al., 2016).

FGF is also required for the specification of cell types outside the embryonic spinal cord such as the midbrain dopaminergic neurons (Ye et al., 1998). Human pluripotent stem cells have also been successfully converted into dopaminergic neurons using a novel floor plate-based strategy that involves the use of SHH and WNT agonists together with FGF8 and these are efficiently engrafted in vivo using rat, mouse and monkey models (Kriks et al., 2011). This could be promising for the development of cell-based therapies in Parkinson's disease.

Finally, it also important to consider the oncogenic risk associated to the mitogenic potential of cells treated with FGFs in transplantation experiments. As a recent example, human cord blood-derived iPSCs have been differentiated into dopaminergic neurons using either FGF2 or BMP/TGF-β inhibitor for neural induction. After transplantation in hemiparkinsonian rats in vivo, proliferation still occurred in FGF2-derived grafts (but not in BMP inhibitor treated grafts), resulting in tumor-like growth (Effenberg et al., 2015). Similarly, those effects have also been described for neurospheres derived from hIPSCs and transplanted into spinal cord injured mice (Nori et al., 2015).

## FUTURE DIRECTIONS AND CHALLENGES

This review highlights the multiple steps in spinal cord development that are regulated by FGF signaling, which may be viewed as a sensor of caudal elongation serving to coordinate different aspects of spinal cord maturation to each other, to adjacent mesoderm and to axial elongation. Further analysis of FGF signaling deficiency in mouse would help ascertain the extent of its contribution to floor plate formation, early neurogenesis, rostro-caudal patterning and neural crest development.

The molecular mechanisms that link FGF signaling specifically to the different functions are still not fully identified but for most of its functions, specific transcriptional targets downstream of the pathway have been proposed. It has been shown that FGF influences transcription by changing the phosphorylation state of transcription factors such as those of the ETV family. The analysis of the regulatory regions of the proposed targets will confirm which of them are more directly regulated. FGF also has an influence on chromatin compaction and nuclear positioning of specific gene loci (Patel et al., 2013) and this may be due, at least in part to the ability of FGF to regulate chromatin modifiers such as histone deacetylase 1 (HDAC1) (Olivera-Martinez et al., 2014).

The detailed regulation of the pathway including the intracellular dynamics of the MAPK pathway with its positive and negative feedbacks (Lake et al., 2016) as well as the involvement of the other FGFR dependent cascades (AKT, PKC) in some of the processes described here also remains largely unexplored. The understanding of the mechanisms responsible for the maintenance of FGF8 and FGF4, the principal ligands in this context, in the NMP and adjacent regions and their progressive downregulation would provide a better insight into axis elongation.

So far, the majority of the literature relies on static views of the expression of ligands and pathway components at different developmental stages. However, it is clear that those are highly dynamic and thus the development of reliable biosensors to measure FGF activity in vivo would help to address fundamental questions such as the mechanisms underlying the temporal changes in the response of NMP and its derivatives to FGF.

Throughout the review, we have focused on the similarities that exist in the different vertebrate species but it would also be interesting to understand how FGF functions may have diverged to accommodate the different modes of spinal cord formation (Steventon and Martinez Arias, in press). Equally interesting would be to study the emergence, during chordate evolution, of a function of caudal FGF on development of the caudal neural tube. FGF signaling has been described in the tailbud of the amphioxus cephalochordate embryo and, although only a limited role in somitogenesis has been described (Bertrand et al., 2015), it would be interesting to assess its requirement in spinal cord development.

Some of the functions of FGF described in the development of the spinal cord may also contribute to the maintenance of the ependymal neurogenic niche present in the adult spinal cord and to the functional recovery after SCI shown in rodents and currently under study in humans. Furthermore, the role of FGFs in the maintenance and expansion of neural progenitors as well as their promotion of specific fates in vitro supports their therapeutical potential in regenerative biomedicine. The advances in understanding the detailed mechanism underlying FGF function during the development of the central nervous system, and in particular of the spinal cord, should serve to selectively potentiate some of its functions.

## AUTHOR CONTRIBUTIONS

RD and AM jointly conceived, organized and wrote the manuscript.

## ACKNOWLEDGMENTS

We thank Gil Costa (Champalimaud Research, Champalimaud Centre for the Unknown, Lisbon, Portugal) for illustrations in **Figures 1**, **3**, **4**. We acknowledge support of our work by a grant from the Spanish MINECO (BFU2014-57494-R) and from Champalimaud Foudation (to RD), and support of the publication fee by the CSIC Open Access Publication Support Initiative through its Unit of Information Resources for Research (URICI).

## REFERENCES


linked to mouse embryo mesoderm development. Dev Cell 40, 331–341. e334. doi: 10.1016/j.devcel.2017.01.015


motor neurons and improves respiratory function after experimental spinal cord injury. J. Neurosci. 19, 7037–7047.


a clinical trial. J. Neurosurg. Spine 15, 216–227. doi: 10.3171/2011.4.SPINE 10404


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

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

# Turn It Down a Notch

Francesca A. Carrieri and Jacqueline Kim Dale\*

Division of Cell and Developmental Biology, School of Life Sciences, University of Dundee, Dundee, UK

In the developing vertebrate embryo, segmentation initiates through the formation of repeated segments, or somites, on either side of the posterior neural tube along the anterior to posterior axis. The periodicity of somitogenesis is regulated by a molecular oscillator, the segmentation clock, driving cyclic gene expression in the unsegmented paraxial mesoderm, from which somites derive. Three signaling pathways underlie the molecular mechanism of the oscillator: Wnt, FGF, and Notch. In particular, Notch has been demonstrated to be an essential piece in the intricate somitogenesis regulation puzzle. Notch is required to synchronize oscillations between neighboring cells, and is moreover necessary for somite formation and clock gene oscillations. Following ligand activation, the Notch receptor is cleaved to liberate the active intracellular domain (NICD) and during somitogenesis NICD itself is produced and degraded in a cyclical manner, requiring tightly regulated, and coordinated turnover. It was recently shown that the pace of the segmentation clock is exquisitely sensitive to levels/stability of NICD. In this review, we focus on what is known about the mechanisms regulating NICD turnover, crucial to the activity of the pathway in all developmental contexts. To date, the regulation of NICD stability has been attributed to phosphorylation of the PEST domain which serves to recruit the SCF/Sel10/FBXW7 E3 ubiquitin ligase complex involved in NICD turnover. We will describe the pathophysiological relevance of NICD-FBXW7 interaction, whose defects have been linked to leukemia and a variety of solid cancers.

### Edited by:

Juan Jose Sanz-Ezquerro, Spanish National Research Council, Spain

#### Reviewed by:

Raj Ladher, TIFR National Centre for Biological Sciences, India Marc Vooijs, Maastricht University, Netherlands Moises Mallo, Instituto Gulbenkian de Ciência, Portugal

#### \*Correspondence:

Jacqueline Kim Dale j.k.dale@dundee.ac.uk

#### Specialty section:

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

Received: 22 September 2016 Accepted: 16 December 2016 Published: 18 January 2017

#### Citation:

Carrieri FA and Dale JK (2017) Turn It Down a Notch. Front. Cell Dev. Biol. 4:151. doi: 10.3389/fcell.2016.00151

Keywords: somitogenesis, embryonic development, signalling pathway, notch, FBXW7

## INTRODUCTION

The formation of a segmented body plan is a conserved feature of embryogenesis for all vertebrate species. This process leads to the formation of transient embryonic segments, called somites. Somites are precursors of vertebrae and ribs, associated skeletal muscles, and some dermis (Christ et al., 2007). Their formation is regulated by a molecular oscillator called the segmentation clock (Gibb et al., 2010; Oates et al., 2012; Benazeraf and Pourquie, 2013). Aberrations in this mechanism lead to human developmental disorders, such as spondylocostal dysostosis (Pourquie, 2011; Eckalbar et al., 2012). Some of these malformations originate from defects in Notch signaling, suggesting that this pathway is essential in controlling and regulating vertebrate segmentation.

This review aims to give a general overview of the importance of the Notch signaling pathway in the segmentation clock in addition to a description of our current understanding of the Notch pathway, particularly focusing on the turnover and regulation of the Notch intracellular domain.

## SOMITOGENESIS

Somitogenesis has been the topic of several outstanding reviews (Pourquie, 2001; Maroto et al., 2012; Oates et al., 2012; Benazeraf and Pourquie, 2013; Hubaud and Pourquie, 2014; Bailey and Dale, 2015), thus we will provide a general overview.

Early in development, segmentation initiates through the formation of repeated segments, or somites (Christ et al., 2007; Gibb et al., 2010). Somitogenesis is a cyclical and gradual process such that somites are sequentially pinched off in pairs from the anterior end of two rods of paraxial mesoderm, the presomitic mesoderm (PSM), lying on either side of the caudal neural tube (Gossler and De Angelis, 1998; Cambray and Wilson, 2007; Dequeant and Pourquie, 2008; Gibb et al., 2010; Maroto et al., 2012). The PSM is continuously replenished with progenitor cells located initially in both the epiblast adjacent to the primitive streak and the rostral primitive streak and later in the tail bud (Iimura et al., 2007; Gomez and Pourquie, 2009; Henrique et al., 2015), and thus the presomitic mesoderm preserves its length (Dequeant and Pourquie, 2008; **Figure 1A**).

The periodicity of this segmentation process is different from species to species: 30 min in zebrafish (Schroter et al., 2008), 90 min in chicken (Palmeirim et al., 1997), 2 h in mice (Tam, 1981), 6–8 h in human (William et al., 2007). Similarly, the total number of somites is a characteristic feature of each species: 31 pairs in zebrafish, 50 somite pairs in chicken, 65 in mice, and about 500 in some snakes.

The regulation of the periodicity of somitogenesis is governed by the segmentation clock, a molecular oscillator (Palmeirim et al., 1997) whose existence was first proposed in theoretical models such as the "Clock and Wavefront model" (Cooke and Zeeman, 1976). According to the model, a wavefront of maturation sweeps along the body axis concomitant with extension of the trunk and tail, governing maturation of the PSM to become somites. This positional information gradient within the PSM interacts with a smooth cellular oscillator (the clock), driving cells to oscillate between a permissive and a non-permissive state. Segmentation of the PSM only occurs when the maturation wavefront reaches a group of cells in a specific "permissive" clock phase (Cooke and Zeeman, 1976).

Over the last 20 years the theoretical "Clock and Wavefront model" has received significant experimental support. The wavefront of maturation is thought to rely on the intersecting gradients and cross-regulatory activities of three signal pathways, namely a caudo-rostral gradient of FGF and Wnt and rostrocaudal gradient of retinoic acid (RA). The determination front marks the point of intersection of these gradients, where the next prospective somite boundary will form (**Figure 1B**). These crossregulatory activities thereby regulate somite size. The activity of Wnt and FGF also controls cell maturation in the PSM. These roles have been reviewed elsewhere, thus will not be covered here (Aulehla et al., 2003; Dubrulle and Pourquie, 2004; Wahl et al., 2007; Aulehla and Pourquie, 2010; Hubaud and Pourquie, 2014).

It is well established that the rhythmicity of somitogenesis is regulated by the segmentation clock driving cyclic and dynamic expression of "clock genes" in the PSM, with a periodicity that matches somite formation. This feature is conserved among a variety of vertebrate species (Jiang et al., 2000; Cinquin, 2007; Dequeant and Pourquie, 2008; Gomez et al., 2008; Ozbudak and Lewis, 2008; Krol et al., 2011). The clock genes are components of the Notch, Wnt, and FGF pathways (Aulehla et al., 2003; Dequeant and Pourquie, 2008; Yabe and Takada, 2016), playing a reciprocal regulatory role in oscillatory gene expression (reviewed in Gibb et al., 2010; Maroto et al., 2012).

FIGURE 1 | Schematic representation of somitogenesis and the segmentation clock. (A) Pairs of somites bud off from the rostral end of the presomitic mesoderm (PSM) progressively during early development. The tail bud, a site of gastrulation that lies at the posterior end of the embryo, continuously "replenishes" the posterior end of the PSM with progenitor cells. The periodicity of segmentation is regulated by a molecular oscillator that drives cyclic gene expression from the posterior to the anterior tip of the PSM. The different colors represent domains of clock gene expression in different cycles. As time progresses in each cycle, the domain of clock gene expression shifts anteriorly while narrowing until it reaches the anterior limit of the PSM. The periodicity of this cyclic gene expression matches that of somite formation. An orange asterisk lies adjacent to each of the new pairs of somites formed in the time series—the first pair is formed after the blue wave of clock gene expression traverses the PSM and the second pair is formed after the pink wave of clock genes expression traverses the PSM from the tail bud to the anterior limit of the tissue. (B) Two mutually opposing gradients of retinoic acid (RA) and FGF/Wnt regulate the maturation wavefront within the paraxial mesoderm. Due to somite formation anteriorly and gastrulation at the caudal end of the PSM, cells within the PSM become progressively more anteriorly displaced, and, as a result, they are exposed to progressively lower levels of FGF/Wnt. There is a position within the PSM, termed the determination front, where cells are released from the effect of FGF and can respond to the segmentation clock and RA, embarking on their segmentation programme.

While the specific genes which oscillate may vary among species, the most highly represented pathway among the clock genes is the Notch (Krol et al., 2011).

Stemming from the observation that the proteins encoded by clock genes are predominantly unstable negative regulators of the pathway that activates them, it is believed that oscillatory gene expression relies on negative feedback loops of these unstable regulators, such as the two Notch target clock genes, Hes7, and Lunatic Fringe (Lfng) (Bessho et al., 2001a,b, 2003; Cole et al., 2002; Hirata et al., 2002; Dale et al., 2003; Serth et al., 2003; Kageyama et al., 2012; Okubo et al., 2012). It is particularly interesting that blocking Lfng oscillations disturbs somitogenesis in the thoracic and lumbar areas but not in more posterior areas of the embryo (Shifley et al., 2008), implying the role of Notch signaling in segmentation is not uniform along the axis.

In addition to negative feedback, oscillatory gene expression in the PSM also invokes positive feedback; Notch signaling regulates dynamic expression of Notch1 itself, whereas Wnt regulates dynamic expression of Dll1 (Bone et al., 2014).

As the most highly conserved pathway involved in the segmentation clock, a wealth of studies have focused on elucidating the fundamental role of Notch in somitogenesis and in the segmentation clock mechanism (Barrantes et al., 1999; Jiang et al., 2000; Bessho et al., 2001b, 2003; Dale et al., 2003; Julich et al., 2005; McGrew et al., 2008; Hubaud and Pourquie, 2014; Wahi et al., 2014; Liao and Oates, 2016). Notch is clearly required to synchronize oscillations between neighboring cells (Jiang et al., 2000; Shimojo et al., 2016). A question that arises is whether oscillations are actually necessary for the segmentation process to occur or whether just non-oscillatory activity of the Notch pathway is sufficient. Mutant mice or fish lacking Notch components all display severe segmentation defects (Conlon et al., 1995; Barrantes et al., 1999; Jiang et al., 2000; Liao and Oates, 2016). For example, the lack of the obligate transcription factor RBP-Jκ, in mouse, leads to lethality before day E10.5 and only the first few cervical somites are formed (Oka et al., 1995). A pivotal study conducted by Ferjentsik et al. pointed out that Notch activity, per se, is indeed essential for somite formation. Mutating crucial Notch pathway components, or using a complementary pharmacological approach, they demonstrated that in mouse Notch activity is crucial for the oscillatory activity of all clock genes, and thus essential for the formation of a segmented body axis (Ferjentsik et al., 2009) (see also Huppert et al., 2005).

## NOTCH SIGNALING PATHWAY

The Notch pathway is highly conserved among metazoans and mediates short range juxtacrine communication. The Notch locus was first cloned in Drosophila and it was found to encode a large single pass type I transmembrane protein (Wharton et al., 1985), whose epidermal growth factor (EGF) repeats mediate interaction with their canonical activators—two ligands, Delta, and Serrate, in the Delta-Serrate-Lag2 (DSL) family. Drosophila studies have contributed hugely to our current understanding of Notch (Artavanis-Tsakonas et al., 1999). The role of Notch in developmental processes of multicellular species has been extensively elucidated (Dumortier et al., 2005; Radtke et al., 2005; Aster, 2014). Notch signaling outcome mostly relies on the cellular context, and thus Notch affects stem cell maintenance, cell fate choice, cell differentiation, lineage progression, and apoptosis in a context-dependent fashion (Bray, 2006; Hori et al., 2013).

Despite its multiple roles and versatility, the Notch pathway is relatively simple and conserved across species (Artavanis-Tsakonas et al., 1999; Bray, 2006; Kopan and Ilagan, 2009). In mammals, there are four Notch receptors (NOTCH1-4) and five DSL ligands (JAG1-2 and Delta-like 1-3-4). Both receptors and ligands are single transmembrane proteins and thus to trigger the signaling cascade, cell-cell contact is required (D'souza et al., 2010; Andersson et al., 2011; Greenwald and Kovall, 2013).

The Notch receptor is typically comprised of: (i) 29– 36 EFG-like repeats in its extracellular domain, involved in ligand interaction; (ii) three juxtamembrane repeats (Lin-12- Notch, LIN), required for extra-intracellular domain interaction (located within the Negative Regulatory Region (NRR); (iii) the intracellular region, including seven ankyrin (ANK) repeats flanked by a PEST [rich in proline (P), glutamic acid (E), serine (S) and threonine (T) residues] and a transactivation (TAD) domain (**Figure 2C**).

During its maturation, Notch undergoes ligand-independent cleavage by a furin-like convertase in the trans-Golgi (Artavanis-Tsakonas et al., 1999; Fiuza and Arias, 2007; Hori et al., 2013). This first cleavage (the S1) results in the production of a heterodimeric receptor comprised of a transmembrane/intracellular fragment non-covalently bound to the Notch extracellular domain (NECD). Notch is thus presented to the cell surface as a heterodimer. The non-activated Notch receptor is constitutively internalized, ubiquitinated by Itch/AIP4 (a member of the Nedd4 family of HECT domain E3 ubiquitin ligases), and thus targeted for lysosomal degradation (Chastagner et al., 2008; Moretti and Brou, 2013).

To ensure correct folding and activity, during synthesis and secretion in the Golgi, NECD undergoes O-linked glycosylation and fucosylation (Rana and Haltiwanger, 2011). These two modifications on the EGF repeats modulate Notch activity by modulating interaction with the Delta or Serrate ligands. The reaction is catalyzed by three Fringe homologs (Lunatic, Manic, and Radical Fringe), recognizing specific amino acids in individual EGF repeats (Rampal et al., 2005). In vitro, in the signal-receiving cell, all Fringes enhance Dll1-Notch1 interactions with comparable effects in both trans- and cis- (Lebon et al., 2014). Rfng also enhances trans- and cisinteractions between JAG1 and Notch1, but these interactions are weakened by Lfng and Mfng. By contrast, JAG1 activation of Notch2 is potentiated by Lfng, thereby expanding the ligand-receptor combinations that are differentially modified by the Fringe enzymes (Hicks et al., 2000). In the context of somitogenesis, Lfng is the only family member expressed in the PSM. In most systems, Lfng acts in the receiving-cell to potentiate receptor activation by Delta-like ligands while reducing activation by Jagged ligands (Hicks et al., 2000; Yang et al., 2005; Kato et al., 2010). However, it has been suggested that LFNG protein may synchronize clock oscillations between

FIGURE 2 | (A) The Notch signaling pathway. In the Golgi, after being glycosylated by members of the Fringe family, pre-Notch is cleaved by a Furin-like convertase into the extracellular and intracellular domains (termed the S1 cleavage), resulting in a heterodimeric receptor with non-covalently associated domains that is transported to the plasma membrane. The Fringe modifications introduced in the Golgi affect sensitivity of the receptor on the signal-receiving cell to the DSL (Delta-Serrate-Lag2) ligands, in the signal-sending cell. Following ligand-receptor interaction, trans-endocytosis of the Notch extracellular domain, by the signal-sending cell, exposes the second so called S2 cleavage site, facilitating intramembranous cleavage by an ADAM (a disintegrin and metalloproteinase domain) protease, in the extracellular domain. S2 cleavage, in turns, exposes the S3 proteolytic cleavage site within the transmembrane domain, which is cleaved by the γ-secretase complex and liberates the intracellular domain of Notch (NICD), allowing it to translocate to the nucleus and thus activate transcription of target genes. In order to prevent inappropriate signaling from the pool of Notch that has not been activated by ligand, Notch receptor is continuously internalized into early endosomes and thus degraded. (B) Zoom-in into the nucleus of the signal-receiving cell (A). Once released into the nucleus, NICD binds the DNA-binding protein CSL as well as the SKIP protein. The trimeric complex thus recruits Mastermind-like protein (MAM), which recruits additional co-activators (not shown), required for the transcriptional regulation of Notch target gene expression. Kinases, such as CDK8 and GSK3β, phosphorylate (p) NICD on its PEST domain, rendering it susceptible to recognition by Fbxw7 E3 ligase, leading to ubiquitination (Ub) and subsequent degradation by the proteasome. In the absence of NICD, CSL associates with transcriptional co-repressors blocking Notch target gene activation. Target genes are repressed until more NICD is produced to re-initiate a new cycle of target gene expression. (C) Notch ligands and receptors. In the signal-receiving cell, the four mammalian Notch receptors (Notch1-4) are represented. They are expressed on the cell surface as heterodimers and characterized by epidermal growth factor (EGF)-like and LIN repeats in their extracellular region. The intracellular domain includes an RBP-Jκ-associated molecule (RAM) domain, seven ankyrin (ANK) repeats, two nuclear localization signals (NLS), a transactivation (TAD) domain (lacking in Notch3 and Notch4), and a Proline-, Glutamate-, Serine- , and Threonine-rich (PEST) domain. The five Notch ligands (Delta-like 1, 3, and 4 and Jagged 1 and 2) are represented on the surface of the signal-sending cell. Each ligand contains an EGF-like repeat region and a conserved domain DSL (Delta/Serrate/Lag). A conserved cysteine-rich (CR) domain is also present on Jagged1 and Jagged2. The DOS (Delta and OSM-11) domain, containing two atypical EGF repeats, is part of Dll1, JAG1 and JAG2 ligands. (D) SCFFbxw7 E3 ubiquitin ligase complex representation. The upper part of the figure shows Fbxw7 domains: a conserved dimerization motif, which mediates dimerization of the SCF complex and thus facilitates ubiquitin conjugation; the F-box, which binds the SCF complex through Skp1; the WD40, containing three specific amino acid residues, which binds the phosphorylated substrate. In the bottom part of the figure the SCFFbxw7 complex is schematically represented. In general, in the ubiquitin system, three enzymes are involved in the signaling cascade: the ubiquitin-activating E1, the ubiquitin-conjugating E2 and an E3 ubiquitin ligase. The first step is ATP-dependent and involves the binding of ubiquitin to E1. Ubiquitin is then activated and transferred to E2. The ubiquitin-E2 complex then interacts with a specific E3 (SCFFbxw7, refer to main text for a description), which recognizes the substrate (phosphorylated NICD, in this case) and facilitates transfer of the ubiquitin molecules to the substrate, leading to substrate degradation by the 26S proteasome.

neighboring cells by acting in the signal-sending cell to inhibit Notch1 activation by Dll1 (Okubo et al., 2012). Ligand binding in an adjacent cell triggers a second cleavage, mediated by the metalloprotease ADAM10 (A disintegrin and metalloprotease) at S2 site in the juxtamembrane extracellular domain, proximal to the Notch transmembrane domain (Mumm et al., 2000; Dyczynska et al., 2007; Bozkulak and Weinmaster, 2009; Gordon et al., 2009; Van Tetering et al., 2009; Weber et al., 2011; Groot et al., 2014). The cleaved NECD product, bound to the ligand, undergoes trans-endocytosis into the ligand-expressing Carrieri and Dale Turn It Down a Notch

cell (Kramer, 2000; Parks et al., 2000; Meloty-Kapella et al., 2012). The second cleavage exposes the third cleavage site, S3, within the membrane-tethered Notch fragment, and is thus a rate-limiting step for the third and final cleavage (Brou et al., 2000; Mumm et al., 2000). Upon cleavage at the S3 site by a γ-secretase complex, the Notch intracellular domain (NICD) is then released (Schroeter et al., 1998) and translocates into the nucleus to activate transcription of target genes (**Figure 2A**). Notch can be activated in the endosomal pathway, independently of its ligands, through the activity of Deltex, a Ring-domain ubiquitin ligase that binds to NICD. However, it is unclear how the Deltex-activation mechanism relates to that of ligandinduced signaling.

Notch signaling does not require the use of second messengers. The activity is exclusively driven by nuclear concentration of NICD (Struhl and Adachi, 1998; Ehebauer et al., 2006). In the nucleus, NICD binds a bi-functional transcription factor CSL [CBF1, Su(H), Lag-1], a DNA binding complex Mastermind (MAM), and a variety of other coactivators involved in the transcriptional activation of Notch target gene expression (Fryer et al., 2004; Kopan and Ilagan, 2009; Hori et al., 2013). The transcriptional co-regulator SKIP (Skiinteraction protein) and the histone acetylase p300 are recruited concomitantly to the promoter region of target genes promoting the assembly of the initiation and elongation complexes (Zhou et al., 2000; Wallberg et al., 2002; Fryer et al., 2004; Bray, 2006; **Figure 2B**). MAM also engages kinases that phosphorylate NICD (Wu et al., 2000; Kitagawa et al., 2001; Nam et al., 2003; Fryer et al., 2004), a crucial step in the regulation of NICD stability and activity (Ingles-Esteve et al., 2001; Espinosa et al., 2003; Fryer et al., 2004; Jin et al., 2009). The domain targeted is the C-terminal PEST domain that is phosphorylated by the cyclin C cyclindependent kinase-8 complex (Cyc:CDK8) and glycogen synthase kinase 3β (GSK-3β) (Espinosa et al., 2003; Fryer et al., 2004; Jin et al., 2009).

## FBXW7 AND ITS ROLE IN NICD TURNOVER

NICD phosphorylation leads to its ubiquitination, turnover, and degradation by the proteasome, defining the half-life of Notch signaling, allowing the cell once again to become ligandcompetent and resetting the signaling for a new cycle of activation (Le Bras et al., 2011). In the prevailing model, the ubiquitin ligase involved is the SCFFbxw7 [S phase kinaseassociated protein 1 (SKP1)-Cullin 1 (CUL1)-F-box] protein complex (Wu et al., 2001; Tsunematsu et al., 2004; Crusio et al., 2010). SCFFbxw7 is part of the RING-finger domain E3 family (Petroski and Deshaies, 2005). Briefly, Cullin 1 acts as a scaffold on which SKP1 and RBX1 subunits assemble. SKP1 is involved in the recruitment of F box proteins (FBXW7, in the case of NICD), and RBX1 recruits a cognate E2 (Hao et al., 2007; Skaar et al., 2013). Fbxw7 consists of three isoforms (α, β, and γ) generated by alternative splicing and the isoform α, shown to ubiquitinate NICD, is localized to the nucleus (Matsumoto et al., 2006; O'neil et al., 2007; Welcker and Clurman, 2008; Crusio et al., 2010). Two domains are functionally important in the FBXW7 protein: the F-box domain, binding SKP1 (Bai et al., 1996), and the seven WD40 repeats mediating recognition/binding to the target protein in a specific consensus phospho-motif, the Cdc4 phospho-degron (Thr-Pro-Pro-Xaa-Ser, in which Thr and Ser residues are phosphorylated; Koepp et al., 2001; Welcker et al., 2003; Hao et al., 2007; Skaar et al., 2013; **Figure 2D**). A number of these phospho-degrons have been identified in the NICD PEST domain. Intriguingly, an additional hNICD1 specific degron has recently been identified within the N-terminal region, distinct from the PEST domain that is not recognized by FBXW7 (Broadus et al., 2016). Moreover, the E3 ligase, Itch, promoting PEST domain-independent NICD1 degradation (Qiu et al., 2000), does not mediate NICD1 degradation through the N1-Box (Broadus et al., 2016).

## NICD-FBXW7 INTERACTION

Given the importance of Notch signaling in cell fate determination and cell cycle progression, it is not surprising that aberrations in the pathway lead to cancers and other diseases (Roy et al., 2007; Simpson et al., 2011; Wang et al., 2011; Kamath et al., 2012; Bolos et al., 2013; Huang et al., 2013; Lobry et al., 2013). Moreover, the pleiotropic nature of the pathway means the various Notch receptors can act as tumor suppressors for example in epithelial tumors or as oncogenes in leukemia and a variety of solid cancers (Radtke and Raj, 2003; Miele et al., 2006; Lobry et al., 2014; Alketbi and Attoub, 2015; Habets et al., 2015; Bonyadi Rad et al., 2016). From this vast literature we will focus here on activating mutations in Notch1 which are predominantly located in the extracellular heterodimerization (HD) domain resulting in ligand-independent exposure of the S2 cleavage site (Malecki et al., 2006; Van Tetering et al., 2009), or in the PEST domain, leading to constitutive activation of the pathway through increased NICD stability or in FBXW7, in line with its fundamental role in restricting the signaling strength/duration of the Notch pathway (Oberg et al., 2001; Tetzlaff et al., 2004; O'neil et al., 2007; Thompson et al., 2007; Wang et al., 2012; Bolos et al., 2013). For instance, Notch1 mutations occur in over 50% of both pediatric and adult Tcell acute lymphoblastic leukemia (T-ALL) cases (Malyukova et al., 2007; Erbilgin et al., 2010), while Fbxw7 mutations are found in up to 20% of T-ALL cases (Baldus et al., 2009; Mullighan, 2009). Furthermore, Notch1 mutations were found in diffuse large B-cell lymphoma (DLBCL), splenic marginal zone lymphoma (SMZL), Hadju-Cheney syndrome (Isidor et al., 2011; Simpson et al., 2011; Kiel et al., 2012), breast cancer (Wang et al., 2015), and in 12% of non-small-cell lung carcinomas (NSCLCs), of which half were in the PEST domain (Westhoff et al., 2009). In these conditions, Notch target genes are highly upregulated.

Considering the variety of pathological conditions associated with alterations of NICD and FBXW7, there is a limited understanding of the regulation of this interaction. Our current understanding stems from a study on Sel-10, the nematode homolog of Fbxw7, showing the two proteins bind directly to each other and FBXW7 negatively regulates Notch signaling (Hubbard et al., 1997). Using cell models, human and murine homologs of Sel-10 were shown to play a key role in regulating Notch signaling by driving NICD to ubiquitin-proteasome mediated degradation (Gupta-Rossi et al., 2001; Oberg et al., 2001; Wu et al., 2001). NICD ubiquitination relies on the PEST domain. Studies on three NOTCH4 variants suggested that Sel-10 preferentially binds to phosphorylated forms of the C-terminal domain of NOTCH4 (Oberg et al., 2001; Wu et al., 2001). However, the NICD-Sel10 interaction has only been observed under overexpression conditions in vitro. It remains to be shown if this interaction occurs in vivo, if NICD interacts with any other E3 ligases, how this interaction is regulated and whether it is contextdependent. The FBXW7 null mutant mice exhibit elevated levels of Notch4 intracellular domain and/or Notch1 intracellular domain alongside defects that are in alignment with a variety of roles identified for Notch in different developmental process such as cardiogenesis and vascular development. However, intriguingly, with respect to the segmentation clock, the absence of Fbxw7 seems to play a less major role in this process, at least according to the mutant phenotypes—although a detailed analysis has yet to be conducted (Tetzlaff et al., 2004; Tsunematsu et al., 2004). The results of these reports suggest that the mechanisms of NICD1 degradation during the somitogenesis process might actually rely on alternative (or redundant) mechanisms, highlighting again the need to further study alternative means of regulation of stability NICD1/degradation.

## REFERENCES


## CONCLUSIONS

In this review we provided a general overview of the critical role of Notch signaling in regulating the segmentation clock involved in somitogenesis. Notch activity is based on stability and turnover of its intracellular domain, NICD. This stability is regulated by phosphorylation of the PEST domain, targeting NICD to proteasome degradation upon recognition by the E3 ligase FBXW7. Mutations in the PEST domain, leading to aberrations in NICD stability, are the underlying cause of a number of solid and non-solid cancers and different genetic disorders. Therefore, uncovering the finer details of Notch pathway regulation merits attention, particularly because a wider comprehension of this process would provide further insights into the mechanisms involved in the onset of Notch-related diseases.

## AUTHOR CONTRIBUTIONS

FC and JD conceived the structure and content. FC wrote the initial draft document. FC designed and produced the figures. JD corrected and edited the document.

## ACKNOWLEDGMENTS

We would like to thank Laura D'Ignazio, Lucas Morales Moya and Ioanna Mastromina for helpful and constructive feedback. This work was supported by a Wellcome Trust PhD studentship to FC and a Wellcome Trust Strategic award [097945/Z/11/Z].

fringe during somite boundary formation in the mouse. Curr. Biol. 9, 470–480. doi: 10.1016/S0960-9822(99)80212-7


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

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

# Signaling and Transcription Factors during Inner Ear Development: The Generation of Hair Cells and Otic Neurons

### Héctor Gálvez, Gina Abelló and Fernando Giraldez \*

Developmental Biology, CEXS, Parc de Recerca Biomèdica de Barcelona, Universitat Pompeu Fabra, Barcelona, Spain

Integration between cell signals and bHLH transcription factors plays a prominent role during the development of hair cells of the inner ear. Hair cells are the sensory receptors of the inner ear, responsible for the mechano-transduction of sound waves into electrical signals. They derive from multipotent progenitors that reside in the otic placode. Progenitor commitment is the result of cell signaling from the surrounding tissues that result in the restricted expression of SoxB1 transcription factors, Sox2 and Sox3. In turn, they induce the expression of Neurog1 and Atoh1, two bHLH factors that specify neuronal and hair cell fates, respectively. Neuronal and hair cell development, however, do not occur simultaneously. Hair cell development is prevented during neurogenesis and prosensory stages, resulting in the delay of hair cell development with respect to neuron production. Negative interactions between Neurog1 and Atoh1, and of Atoh1 with other bHLH factors driven by Notch signaling, like Hey1 and Hes5, account for this delay. In summary, the regulation of Atoh1 and hair cell development relies on interactions between cell signaling and bHLH transcription factors that dictate cell fate and timing decisions during development. Interestingly, these mechanisms operate as well during hair cell regeneration after damage and during stem cell directed differentiation, making developmental studies instrumental for improving therapies for hearing impairment.

Keywords: atoh1, Neurog1, Hes and Hey factors, Notch signaling pathway, cell fate specification, hair cell regeneration

## THE INDUCTION OF NEURAL COMPETENCE IN THE OTIC PLACODE

The ear is one major sensory organ of the vertebrate head that is responsible for the senses of hearing, balance and acceleration. The vertebrate inner ear derives from the otic placode, a thickening of the head ectoderm. The formation of the inner ear requires a series of cell fate decisions and morphogenetic events with a precise temporal and spatial pattern (Fritzsch et al., 2006; Groves and Fekete, 2012). Mature sensory organs of the vestibular and auditory regions of the inner ear are formed by three cells types: hair cells (HC), supporting cells (SC), and neurons, which in amniotes derive from a common neurosensory pool of cells.

One crucial step during inner ear development is the specification of neurosensory progenitors and the diversification of the different cell types. This is probably the first developmental decision in the otic epithelium and it reflects the segregation of two functionally independent domains,

#### Edited by:

Juan Jose Sanz-Ezquerro, Consejo Superior de Investigaciones Científicas (CSIC), Spain

#### Reviewed by:

Andy Groves, Baylor College of Medicine, USA Donna M. Fekete, Purdue University, USA Bruce B. Riley, Texas A&M University, USA

> \*Correspondence: Fernando Giraldez fernando.giraldez@upf.edu

#### Specialty section:

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

> Received: 17 December 2016 Accepted: 02 March 2017 Published: 24 March 2017

#### Citation:

Gálvez H, Abelló G and Giraldez F (2017) Signaling and Transcription Factors during Inner Ear Development: The Generation of Hair Cells and Otic Neurons. Front. Cell Dev. Biol. 5:21. doi: 10.3389/fcell.2017.00021 one with neurosensory competence and another devoid of it (Abello and Alsina, 2007). The neurosensory domain gives rise to otic sensory neurons, sensory hair cells and supporting cells in chick and mouse (Satoh and Fekete, 2005; Raft et al., 2007). The expression of Sox3 and Sox2, Fgf10, and also that of members of the Notch pathway like Delta1, Hes5, and Lunatic Fringe is restricted to the neurosensory domain (Abelló et al., 2007). SoxB1 genes have a proneural function (See **Box 1**) and drive the expression of Neurog1 and Atoh1 (Jeon et al., 2011; Neves et al., 2012). The complementary non-neural domain shows two major patterning genes, Lmx1b and Iroquois1, and two members of the Notch pathway, Serrate1 and Hes1 (Abelló et al., 2007, 2010).

FGF and BMP signaling differentially regulate the expression of Sox3 and Lmx1, and their respective restriction to the anterior and posterior domains (Abelló et al., 2007; Schneider-Maunoury and Pujades, 2007). The regionalization of the otic placode into neurosensory and non-sensory territories requires also the functional integrity of the Notch pathway for its stabilization (Abelló et al., 2007). The non-sensory region of the otic placode receives signals that confer posterior identity (Bok et al., 2011). Retinoic acid (RA), which is known to posteriorize the embryonic body axis, is also required to specify the posterior character of the otic placode. Expression of RA synthesizing and degrading enzymes coincides with the AP boundary of the otic placode, and experiments in chicken and zebrafish have disclosed a developmental window during which the otocyst receives and is sensitive to RA posteriorizing signals (Bok et al., 2011; Radosevic et al., 2011).

Two main cell fate decisions are made sequentially during ear development. First, neurosensory progenitors produce either neuronal (neuroblast) or sensory precursors. Secondly, once neurons have delaminated, the progenitors that remain in the epithelia develop into either hair cells or supporting cells. The differentiation of neurons and hair cells is driven by the expression of, respectively, Neurog1 and Atoh1, two basic Helix-Loop-Helix (bHLH) proteins. Notch signaling plays a critical role in these two sequential decisions because it is instrumental in forcing precursors to adopt alternative fates by the mechanism of lateral inhibition (**Figure 1**).

Evidence in different species suggests that neurosensory progenitors are multipotent. Lineage analyses by viral tracing in chicken embryos demonstrated that bipotential neurosensory progenitor cells are present in the otic placode (Satoh and Fekete, 2005) and dye-labeling of otic placode progenitors showed that neurons and hair cells derive from the neurosensory domain of the otic vesicle (Bell et al., 2008). Furthermore, genetic fate mapping in mouse and chick indicates that vestibular sensory hair cells derive from Sox2 expressing progenitors residing in the neurosensory domain of the otic placode (Raft et al., 2007; Neves et al., 2012). In zebrafish, there are three progenitor pools, one specific to neurons, another specific to hair cells and a third one that can give either neurons or hair cells until later stages (Sapède et al., 2012), but all come from a population that expresses Atoh1b, suggesting that they also may share a common progenitor (Millimaki et al., 2007).

## SOX2 AND NEURAL COMPETENCE

Sox genes are transcription factors that belong to the High Moblity Group (HMG) box domain proteins (Kamachi and Kondoh, 2013). One subfamily of Sox genes is the SoxB group, which is split in turn into two sub-groups: SoxB1 that includes Sox1, Sox2, and Sox3, being only Sox2 and Sox3 expressed in the vertebrate inner ear (Neves et al., 2007; Abelló et al., 2010), and SoxB2 that comprises Sox14 and Sox21, which are transcriptional repressors (Kamachi and Kondoh, 2013) and from which only Sox21 is expressed during ear development (Freeman and Daudet, 2012).

Sox2 is critical for the specification of neurons and hair cells in the neurosensory domain of the otic placode (Kiernan et al., 2005; Neves et al., 2012). Sox2 is able to activate both Neurog1 and Atoh1, but it is downregulated in differentiated neurons and hair cells. Sox2 expression remains high in supporting cells, suggesting that this cell type retains progenitor properties (Neves et al., 2007; Evsen et al., 2013; Kamachi and Kondoh, 2013). The expression of Sox2 in the inner ear is driven by signals from the surrounding tissues. Like in other regions of the nervous system, FGF signaling is determinant for setting the onset of SoxB1 factors (Alsina et al., 2004; Sweet et al., 2011; Ono et al., 2014). In the inner ear, first Sox3 and then Sox2 expression depends on FGF signaling emanating from the underlying mesoderm, the hindbrain and probably from the otic placode (Schneider-Maunoury and Pujades, 2007; Groves and Fekete, 2012). Sox3,

#### BOX 1 | bHLH in vertebrates: has Atoh1 lost its proneural function?

What is a proneural gene? A proneural gene must fulfill three main characteristics (Hassan and Bellen, 2000): First, its expression precedes and coincides with the selection of neuronal precursor cells. Secondly, its function is both necessary and sufficient for the specification of a given neuronal lineage in a cell autonomous fashion. Finally, its loss of function results in the deletion (and its missexpression ectopic development) of a given lineage. Proneural genes were first identified in Drosophila peripheral nervous system development. The Achaete–Scute complex (AS-C) genes were identified as proneural genes encoding bHLH factors. Later on, atonal (Atoh1 in mammals) was identified by PCR (Jarman et al., 1993). Atonal in Drosophila is the master gene for the formation of chordotonal organs, which are mechano-receptors of insect muscles. Atonal gene selects the progenitors that give rise to the mature organs. Atonal loss of function abolishes chordotonal organs and its missexpression favors their ectopic formation (Jarman et al., 1993). Are Atoh1 and Neurog1, the vertebrate homologs of atonal, also proneural genes? Atoh1 and Neurog1 overexpression drives, respectively, ectopic hair cell and neuron formation (Izumikawa et al., 2005; Evsen et al., 2013), and their loss of function results in the lack of hair cells or neurons (Ma et al., 1998; Bermingham, 1999). However, their function is far more restricted and, like in other regions of the Nervous System, they do not provide a broad neural competence, but a far more restricted lineage selection (for example, HCs and SCs in the case of Atoh1 and the inner ear). The broad neural competence is rather dependent on SoxB1 (Azuara et al., 2006; Puligilla and Kelley, 2017). This shows a proneural identity crisis in vertebrate development and the taking over by SoxB1 proteins (Hassan and Bellen, 2000).

which is expressed in the neurosensory domain in chick is not detected in the mouse, where Sox9 is co-expressed along with Sox2 in the prosensory region (Mak et al., 2009).

## THE REGULATION OF NEUROG1 AND ATOH1

## The Regulation of Neurog1

Neurog1 (Neurogenin1) is a basic helix-loop-helix (bHLH) transcription factor that behaves as master regulator for neuronal differentiation in different vertebrates (Henrique et al., 1997; Ma et al., 1998; Alsina et al., 2004; Evsen et al., 2013). Neurog1 is an Atonal-related protein (ARP; Hassan and Bellen, 2000). On average, it shares with Atoh1 53% amino acid identity in the bHLH domain, and differs from Atoh1 in four basic domain residues (Sommer et al., 1996). Three neurogenins have been described in mammals. Neurog1 and Neurog2 function as neuroblast selector genes in mouse (Ma et al., 1998), but in the chicken and mouse inner ear, only Neurog1 is expressed during ear development (Ma et al., 1996; Evsen et al., 2013).

Sox2 is necessary for Neurog1 up-regulation in the otic epithelium (Jeon et al., 2011; Neves et al., 2012). In mice, Neurog1 is also activated by Six1 and Eya1 thatsynergize with Sox2 (Zheng et al., 2003; Ahmed et al., 2012b). The neurosensory domain has high Notch activity, and Jeon and colleagues showed that the enhancer of Neurog1 is activated by high levels of NICD (Notch Intracellular Domain), while Atoh1 enhancer is not (Jeon et al., 2011). This may favor that Neurog1 expression precedes Atoh1 in the otic vesicle (Neves et al., 2011). However, later in development Notch signaling represses Neurog1 expression in the cells that remain in the epithelium.

Neurog1 expression is controlled by different cis-elements located 5′ and 3′ to the Neurog1 coding sequence. These enhancers drive the expression of Neurog1 in midbrain, hindbrain, trigeminal ganglia, and ventral neural tube. For Neurog1 expression in the dorsal neural tube only a 5′ enhancer has been identified (Nakada et al., 2004). Another enhancer region drives Neurog1 activity to the VIII cochleavestibular ganglion (Murray et al., 2000). The configuration of these enhancers is similar to the cis-elements identified for Neurog2 (Simmons et al., 2001), suggesting that there is a tight regulation of the two Neurogenins. Nakada et al. (2004) speculated that possibly the conservation between Neurog1 and Neurog2 arises from gene duplication. The modular organization of Neurogenins cis-regulatory regions contrasts with the single enhancer regulation described for Atoh1 (Helms et al., 2000 and see below).

## The Regulation of Atoh1

Atoh1 expression is regulated by a downstream enhancer, which depends on its interaction with Atoh1. In other words, Atoh1 expression relies on its auto-regulation. This implies that crucial events in the developmental regulation of Atoh1 are the chromatin arrangements that allow the interaction of Atoh1 with its own enhancer, and also the activity of potential repressors that break this loop (See **BOX 2**). Work by Jane Johnson's lab discovered a region in the Atoh1 genome landscape that recapitulates Atoh1 expression during mouse and chicken development (Helms et al., 2000). Transgene expression in mouse identified a region that directed the expression to the neural tube, external granular layer (EGL) of the cerebellum from rhombic lip, and the developing hair cells of the cochlea and semicircular canals. The region contains a 1.7 Kb fragment located 3.4 Kb 3′ of the Atoh1 coding region that recapitulates the expression of Atoh1. This region is called the 3′Atoh1-enhancer (see **Figure 2A**).

Two regions within the 3′Atoh1-enhancer (3′Atoh1-enh) show a high homology between humans and mouse, and they were named Enhancer A and Enhancer B (EnhA and EnhB). The length of A and B is highly conserved in species like chicken, mouse, and human (Ebert, 2003), although the distance in between them varies among species. Interestingly, Helms et al. (2000) observed that transgenic embryos for Atoh1-enh/lacZ transgenic mice had no detectable β-gal activity in the Math1 null background, and this was shown to be also the case for the 3′Atoh1-enhancer-GFP reporter in the inner ear (Raft et al., 2007). This suggested that the activity of the 3′Atoh1 enh is dependent on Atoh1 expression and that autoregulation is one major mechanism for setting Atoh1 transcriptional activity (**Figure 2B**).

The 3′Atoh1-enh contains several E-boxes, which are sixnucelotide DNA sequences that bind bHLH proteins, like Atoh1, Neurog1, and Hes/Hey factors (CANNTG; (Massari and Murre, 2000)). The Enhancer A contains a degenerated E-box and the Enhancer B three E-boxes. E-boxes in Enhancer B are a class A, a

#### BOX 2 | The priming of Atoh1.

Priming of Atoh1 in the developing ear may be also promoted by other mechanisms rather than Atoh1 autoactivation. Changes in histone modifications can change the transcriptional hierarchy that controls cell differentiation (Azuara et al., 2006). The dynamic changes in the histone modifications H3K4me3/H3K27me3, H3K9ac, and H3K9me3 indicate that there is a progression from poised to active, and finally to repressive marks in the Atoh1 locus. This sequence correlates with the onset of Atoh1 expression and its subsequent silencing during the perinatal period (Stojanova et al., 2015). The inhibition of acetylation blocks Atoh1 mRNA expression in nascent hair cells, as well as the ongoing hair cell differentiation during embryonic organ of Corti development (Stojanova et al., 2015). Contrarily, histone deacetylase inhibition favors the expression of hair cell markers in mouse utricle progenitor cells (Hu and Wang, 2014). Cochlear explants treated with histone deacetylase inhibitor increase the levels of Atoh1 mRNA in early post-natal mice (Stojanova et al., 2015). This suggests that Atoh1 is poised during developmental stages and thereby ready to be activated. However, after birth the Atoh1 locus becomes methylated and cannot be transcribed when hair cells are damaged. The epigenetic status of Atoh1 locus during organ of Corti development shows a bivalent mark of the Atoh1 locus by H3K27me3 and H3K4me3, prior to the upregulation of Atoh1 (Stojanova et al., 2015). This is consistent with the idea that Sox2 poises/primes the Atoh1 locus until Atoh1 itself is able to bind to the E-box A and trigger Atoh1 expression.

class C and a reversed N-box. All three boxes are very close, and class A and N-box overlap (**Figure 2A**, the CAN region for C-, Aand N-boxes). As mentioned above, Helms et al. (2000) already identified the class A E-box located in Enhancer B as crucial for Atoh1 autoactivation. Besides the CAN region, the 3′Atoh1 enh region contains putative binding sites for a menagerie of transcriptional activators and repressors. Some of them like Sox2, Six1/Eya1, and β-catenin bind directly to the enhancer as shown by biochemical assay (Akazawa et al., 1995; Ebert, 2003; Mutoh et al., 2006; Briggs et al., 2008; D'Angelo et al., 2010; Shi et al., 2010).

Sox2 is sufficient to activate Atoh1 and to induce ectopic hair cell formation in the chick otocyst (Neves et al., 2012) and it is rapidly downregulated as hair cells differentiate. This downregulation is required for further maturation because sustained expression of Sox2 in Atoh1 expressing cells blocks the induction of hair cell markers such as Myosin-VIIa (Dabdoub et al., 2008). Six1 and its transcriptional co-activator Eya1 are expressed in the prosensory domain of the cochlea, and they bind directly to the 3′Atoh1-enh (Ahmed et al., 2012a). These two factors are sufficient to induce Atoh1 expression in competent regions and their activation is potentiated by Sox2 (Ahmed et al., 2012a). As described before, Neurog1 is also upregulated by Six/Eya with Sox2 (Ahmed et al., 2012b).

In summary, the crucial elements in setting neuronal and sensory competence are: (1) FGF signaling that induces the expression of Sox2 in the neurosensory domain of the otic placode, (2) Sox2 that drives the activation of neuronal and sensory master genes Neurog1 and Atoh1 and neurosensory competence, and (3) the onset of Neurog1 and Atoh1 and the determination of neurons and hair cells, respectively.

## NEURONS VS. HAIR CELLS: HOW DOES NEUROG1 COUNTERACT ATOH1?

Although Sox2 is able to activate both Neurog1 and Atoh1, Atoh1 expression remains undetected until HC differentiation (**Figure 2C**). Neves et al. (2012) hypothesized that this delay in Atoh1 expression may be explained by an incoherent feedforward loop (I-FFL) triggered by Sox2, where Sox2 activates both Atoh1 and Atoh1 repressors (Neves et al., 2012). Recent experiments using a conditional gain of function system in mouse support the model by showing that Sox2 is required for prosensory specification, but it must be downregulated to allow Atoh1 expression (Puligilla and Kelley, 2017). Major candidates to mediate the repression of Atoh1 include a variety of bHLH factors that are expressed throughout ear development such Neurog1 or Notch targets.

There is a mutual antagonism between Neurog1 and Atoh1 functions. Neurog1 null mice show a loss of sensory neurons, smaller sensory patches, and premature development of hair cells (Matei et al., 2005). Moreover reduced Neurog1 causes ectopic Atoh1 expression and that excess of Atoh1 suppresses Neurog1 (Raft et al., 2007). During development, Neurog1 overrides Atoh1 expression. Therefore, the functional antagonism between Neurog1 and Atoh1 is resolved in favor of Neurog1, the result being that neurons develop prior to hair cells. The molecular mechanism of this dominance of Neurog1 over Atoh1 is still unknown, but it seems crucial for understanding the timing of cell fate during ear development. Neurog1 is a transcriptional activator, suggesting that the counteractive interaction with Atoh1 is likely complex. In principle, Neurog1 may repress Atoh1 by the following mechanisms: (1) by preventing Atoh1 transcription, (2) by preventing Atoh1 mRNA translation, or (3) by post-translational mechanisms that result in modified Atoh1 protein levels or activity.

## Transcriptional Repression of Atoh1

Neurog1 and Atoh1 are two bHLH type II proteins (Massari and Murre, 2000). They are known to dimerize with type I bHLH like E47 and bind to E-box sequences resulting in activation of transcription (Jarman et al., 1993; Koyano-Nakagawa et al., 1998; Bertrand et al., 2002). One simple possibility for Neurog1 acting as a repressor of Atoh1 is that Neurog1 acts as a partial agonist for Atoh1. Neurog1 would compete for the class A E-box located in the 3′Atoh1-enh, resulting in a weak activation but impeding the stronger autoactivation by Atoh1. Atoh1 and Neurog1 may also compete for the same E-protein partners, like E47, the result being that Atoh1 is unable to bind DNA.

Neurog1 may repress Atoh1 transcription in an indirect manner, by activating transcriptional repressors of Atoh1. Among the targets of Neurog1, NeuroD is one major effector of Neurog1 in the ear, being essential for neuroblasts delamination (Ma et al., 1996; Huang et al., 2000; Kim et al., 2001) and for shutting down Sox2 expression in the neurons (Evsen et al., 2013). Conditional NeuroD deficient mice show that NeuroD suppresses Atoh1 expression in auditory-vestibular neurons as indicated by the ectopic expression of Atoh1 after NeuroD deletion (Jahan et al., 2010). However, during early stages of neurosensory development, Neurog1 is expressed homogeneously in the neurosensory epithelium, including hair cell precursors (Raft et al., 2007), and only those cells that express high levels of Neurog1 trigger lateral inhibition and delaminate from the epithelium. Therefore, it is likely that alternative mechanisms may prevent Atoh1 without necessarily driving neuronal differentiation (Sun et al., 2001; Fritzsch et al., 2006).

## Post-Transcriptional Regulation of Atoh1: mRNA Processing and Stability

The half-life of many mRNAs can fluctuate during development and mRNA stability depends on RNA-binding proteins that bind mRNAs (Day and Tuite, 1998; Knuckles et al., 2012). Also micro-RNAs (miRNAs) are known regulators of mRNA stability or translation efficiency and modify protein expression levels. Some miRNAs like the miR-183 family (miR-96, miR-182, and miR-183) are expressed at high levels in young hair cells and ganglion neurons (Weston et al., 2006; Li et al., 2010) and the manipulation of miR-183 levels modify the number of hair cells (Li et al., 2010, see Groves et al., 2013 for a review).

Regulation of translation and protein synthesis depends on initiation factors (eIFs), some of which are phosphoproteins susceptible of regulation (Day and Tuite, 1998), but little is known about their behavior during embryonic development. Those are potential candidates to regulate the reduction of Atoh1 induced by Neurog1; however, we have no information on whether they are modified by Neurog1.

## Post-Translational Regulation: Atoh1 Activity and Degradation

Degradation of bHLH proteins has been extensively documented in different model systems and it is accounted by phosphorylations in their C-terminus domain (Forget et al., 2014; Hardwick and Philpott, 2015; Quan et al., 2016). Atoh1 post-transcriptional downregulation has been reported during cerebellar granule neuron differentiation, where BMP2 and BMP4 inhibit proliferation and induce differentiation through proteosome mediated degradation of Atoh1 (Zhao et al., 2008). BMPs induce the expression of Id1 and Id2 that upon dimerization with Atoh1 target the complex for degradation. In cerebellar granule neuron progenitors, Shh prevents Atoh1 degradation by preventing the recruitment of Atoh1 by Huwe3, an E3 ligase (Forget et al., 2014). Atoh1 is degraded by the proteosome pathway when dimerizing with Huwe1 in HEK cells, and the conditional deletion of Huwe1 generates supernumerary HCs in the mouse cochlea (Cheng et al., 2016).

Atoh1 contains in the C-terminus a potential PEST sequence (Jarman et al., 1993). This is a peptide sequence rich in proline (P), glutamic acid (E), serine (S), and threonine (T). This sequence is associated with proteins that have a short intracellular half-life and it is hypothesized that the PEST sequence acts as signal for protein degradation. Atoh1 protein stability is very short and it is extinguished in 2 h after protein synthesis blockade (Forget et al., 2014; Cheng et al., 2016). Aminoacid residues located at the C-terminus region of Atoh1 protein that are susceptible to phosphorylation are conserved among different species (Mulvaney and Dabdoub, 2012). Like Atoh1, other bHLH proteins as Neurog2 and NeuroD4 are also less stable upon phosphorylation (Hindley et al., 2012; Hardwick and Philpott, 2015).

Cyclin-dependent kinases (Cdks) drive cell cycle progression and are known to target Serine Proline (SP) and Threonine Proline (TP) sites (Errico et al., 2010). bHLH proteins like Atoh1 and Neurog1 contain several putative ST and TP in their C and N-terminal regions. Phosphorylation in these SP/TP residues may be crucial for regulating activity and is linked to the cell cycle. For example, in Xenopus embryos and P19 cells, progenitor cells that divide rapidly show Neurog2 phosphorylation and degradation, whereas when cell cycle is lengthened, Neurog2 accumulates and activates down-stream targets (Ali et al., 2011). Cell cycle exit in the cochlea is dependent on the expression of the cyclin inhibitor p27kip (White et al., 2006), and it is possible that Atoh1 is degraded in dividing prosensory progenitors until p27kip expression and cell cycle withdrawal. One plausible mechanism is that Atoh1 protein is degraded in the presence of Neurog1. This type of regulation has been recently described for Atoh1 protein when targeted by the E-3 ubiquitin ligase Huwe1 (Cheng et al., 2016). Finally, Neurog1 may also interfere with Atoh1 translation (see **Figure 3**).

## NOTCH SIGNALING AND THE SINGLING OUT OF HAIR CELLS

Notch signaling is an evolutionarily conserved juxtacrine signaling pathway used by metazoans. It controls a broad spectrum of developmental processes in organisms ranging from sea urchins to humans (Artavanis-Tsakonas et al., 1999; Neves et al., 2013). Lateral inhibition is one major operation mode of the pathway by which a ligand-producing cell signals its neighbors to reduce ligand expression (see (Neves et al., 2013) for review on the different modes of operation of Notch during ear development). Lateral inhibition mediates binary cell fate decisions by ensuring that the cells adopt one of two alternative fates. In the inner ear Notch mediates the determination of two major cell types, neurons and hair cells. Driven by Sox2, progenitors residing in the neurosensory domain express Neurog1, some of them with enough strength so to unfold the neuronal program and become neuroblasts. Nascent neuroblasts express the ligand Delta-like1 (Dll1), which activates Notch1 in the neighboring cells and suppress Neurog1 expression. Neuroblasts delaminate, and the cells that remain in the neurosensory epithelium are fated to become sensory precursors (the prosensory patches). Later in development, some cells from the prosensory patches start expressing Atoh1, which

FIGURE 3 | How does Neurog1 force neurogenesis before sensorigenesis? Possible models for Neurog1 repression of Atoh1. Neurog1 prevents Atoh1 activation by binding to the CAN region of the 3′Atoh1 enhancer or by sequestering type I bHLH factors necessary for Atoh1 binding (1). Neurog1 reduces Atoh1 protein levels by mRNA degradation (2), inhibition of protein synthesis (3), or reducing protein stability by promoting Atoh1 protein to proteosome degradation (4), of other bHLH in the prosensory epithelia.

initiates a second round of lateral inhibition by which some precursors activate the ligands Delta1 (in mammals also Jag2) that drive lateral inhibition. The result is that those cells that express Atoh1 become hair cells and prevent the neighbors to do so, generating the typical mosaic of alternate cell types (Adam et al., 1998; Eddison et al., 2000; Neves et al., 2013).

Both during neurogenesis and hair cell generation, the action of Notch ligands results in the expression of the typical Notch targets like Hes and Hey factors (Petrovic et al., 2014, 2015). The most studied Notch canonical effectors are Hairy and Enhancer of Split (Hes) and Hairy and enhancer of split related (Hey). Hes and Hey genes belong to the type VI bHLH group. Seven Hes members have been identified in vertebrates (Hes1–7), while the Hey subfamily of genes encodes three members in mammals (Hey1, Hey2, and HeyL; Iso et al., 2001, 2003).

The core structure of Hes and Hey proteins contains a basic and Helix-loop-Helix domain and an Orange domain at the C-terminus region. The Orange domain serves as a region for protein-protein interactions and for partner selection (Iso et al., 2001). Hey proteins differ from the Hes subgroup by two striking features: first a glycine present in the basic domain of Hey proteins instead of a conserved proline in Hes proteins, which confers DNA-binding specificity (Leimeister et al., 1999). Secondly, the C-terminal WRPW motif that is characteristic of Hes proteins and allow Groucho co-repressor recruitment, is replaced with YRPW or YXXW (HeyL; Fisher et al., 1996). The Cterminal WRPW of Hes motif acts as polyubiquitination signal, making Hes proteins short-living (Hirata et al., 2002; Iso et al., 2003).

Hes factors bind with high affinity to E-box class C or N-box. Hey1, due to the presence of a glycine residue in the basic domain has preference to class C or class B E-boxes (Iso et al., 2003). The repressive function can be either active or passive. Active repression involves DNA binding, whilst in passive repression Hey/Hes proteins sequester bHLH type I family and impair their heterodimerization with class II bHLH (Iso et al., 2003).

During development, several Hes and Hey genes are expressed in the inner ear. Hes5 is the major Notch target expressed during lateral inhibition. It is detected in the precursors that are not selected as neurons or hair cells. Its expression correlates well with that of Dll1 in nascent neurons and hair cells (Petrovic et al., 2014). Hey1 is also expressed in the prosensory epithelium, concomitantly with Jagged1, and co-expressed with Hes5 during hair cell formation (Petrovic et al., 2014). Although Hey1 and Hes5 are direct Notch downstream targets, they differ in the level of Notch required for their activation.

Knockout mice of different Hey and Hes factors exhibit supernumerary hair cells in the cochlea, suggesting a repressor

activity measured in the conditions indicated (n = 3–4). Levels of electroporation were normalized by luciferase activity. Atoh1 activated its own enhancer and Hey1 prevented Atoh1 autoactivation. All three bHLH factors repressed 3′Atoh1 enhancer basal activity. (C) Hey1 (left) and Hes5 (right) also prevented Atoh1 autoactivation in P19 cells. Values of luciferase correspond to enhancer activity relative to the basal activity of the 3′Atoh1 enhancer in the conditions indicated in abscissa (n = 3). Values are normalized by renilla activity. Data displayed as Mean + S.E.M.

function during hair cell development. The combined loss of function of Hes5, Hey1, and Hes1 results in supernumerary hair cells (Tateya et al., 2011) and Atoh1 is upregulated after interference of Hey1/Hes5 expression with siRNAs (Du et al., 2013) or treatment with the Notch inhibitor DAPT (Ren et al., 2016). Notch inhibition of damaged sensory epithelia favors HC regeneration (Lin et al., 2011; Mizutari et al., 2013), suggesting that these factors may also regulate the ability to regenerate HCs (see below).

## THE REPRESSION OF ATOH1 BY HES AND HEY: IS IT ALL IN THE 3′ATOH1 ENHANCER?

Hey1 and Hes5 repress Atoh1 and silence the 3′Atoh1-enh (**Figures 4A,B**). In addition, both factors are able to block Atoh1 autoactivation (**Figure 4C**), suggesting that the repression of Atoh1 by Hes5 and Hey1 prevails upon its own activation. Accordingly, and parallel to 3′Atoh1-enh repression, Hey1 overexpression is sufficient to prevent HC generation in chick sensory epithelia (**Figure 5**). Taken together, these observations suggest that during development, Notch targets Hey1 and Hes5 act on the 3′Atoh1-enh and repress Atoh1 expression in prosensory precursors and supporting cells.

Hey1 is also able to prevent both, the basal activity and the autoactivation of EnhB of the 3′Atoh1-enh in chicken otic vesicles and P19 cells (**Figures 6A,B**). Hes5 can also prevent Atoh1 autoactivation (**Figure 6B**). Moreover, the importance of the CAN region of EnhB is illustrated by multimer reporter analysis showing that Hey1 requires the E-boxes flanking the class A E-box to act as a repressor (**Figure 6C**). The CAN multimer mimics the 3′Atoh1-enh repression promoted by Hey1, indicating that the minimal region of the enhancer to explain the repression is the CAN region.

Hey1 needs to bind DNA in order to repress Atoh1, since the mutation of Hey1 DNA binding domain abolishes repression (**Figure 6D**). However, the identification of the region bound by Hey1 and Hes5 has been difficult and still remains elusive. On the one hand, mutations of either the class C E-box or the Nbox of the CAN region are unable to prevent Atoh1 repression by Hey1 and Hes5 (**Figure 7**). This is in agreement with the results of ChIP-seq analysis performed on HEK 293, which shows that Hey1 does not bind to the 3′Atoh1-enh (Heisig et al., 2012). But it is also somehow surprising and suggests that there are alternative binding sites and/or mechanisms of repression for Hes and Hey factors. For example, it is possible that Hey1 blocks the transcription of Atoh1 by interfering with the class A E-box. This possibility is difficult to explore since the mutation of E-box A silences the 3′Atoh1-enh.

HCs or SC.

A recent study has shown that Hes5 and Hey2 are able to prevent Atoh1 expression by binding to the promoter region of Atoh1, and that the repression of Atoh1 in supporting cells depends on this interaction, rather than on the 3′Atoh1 enhancer, which would operate mainly for Atoh1 activation (Abdolazimi et al., 2016). Gene repression and binding by Hey1 is also dependent on the chromatin signature of the promoter regions. Heisig et al. (2012) found that Hey1 bound sequences overlapped with the presence of polymerase II and the active chromatin mark H3K4m3, characteristic of active and poised promoters. H3K4m3 chromatin marks are found in Atoh1 promoter and enhancer prior to Atoh1 upregulation (Stojanova et al., 2015). Therefore, the regulation of Atoh1 is likely to be dependent on multiple sites, and not only mediated by the 3′Atoh1 enhancer. The relationship between the Enhancer and promoter regions may be crucial to fully solve the complex regulation by Hey1.

EnhB activity with Atoh1 and Hey1-DN (Hey1 dominant negative) in P19 cells. Data displayed as Mean + S.E.M.

## FROM DEVELOPMENT TO REGENERATION

Hearing loss is a major problem affecting more than 360 million people in the industrialized world (WHO). It affects speech and language and leads to severe deficits in communication, and a strong negative impact in the quality of life. Hearing impairment is mainly caused by the failure of hair cells and/or otic neurons (sensorineural hearing loss), hair cell damage being the most frequent triggering factor. Hair cell damage arises from genetic defects, aging, noise, traumatic lesions, infections, or therapeutic substances. The main problem of hair cell damage is that, unlike other animal species, mammals are not able to regenerate hair cells of the auditory epithelia and there is no treatment for hearing deficiencies in humans.

In contrast to mammals, non-mammalian vertebrates like chicken, zebrafish, or lizards, are able to repair and heal

damaged sensory epithelia. In the chick, damaged hair cells trigger supporting cells to replace lost hair cells by two different mechanisms: (1) mitotic regeneration, where SC divides asymmetrically and one daughter cell remains as SC and another as HC, and (2) transdifferentiation of SC into HC. In transdifferentiation, HCs are generated at the expense of SCs, which become exhausted and hence, the epithelium is disorganized. The consequence is that, although HCs are recovered, hearing function is not (Stone and Cotanche, 2007). In birds, hair cell regeneration starts with direct transdifferentiation of SCs into HCs, followed by mitotic regeneration and the correct replacement of the sensory epithelium and auditory function (Roberson et al., 2004).

Although mammals have some capacity to regenerate hair cells in the vestibular organs and the early post-natal cochlea, the adult auditory organ is completely devoid of this capacity. The question arises as to what are the differences between birds and mammals that explain their different regenerative capacity. Are there signals that regulate SC quiescence and activation after HC loss in chicken? Are they similar to mammalian early post-natal regeneration? Why mammals lose the capacity of regeneration after birth?

Studies in the chick have shown that hair cell regeneration reuses mechanisms that operate during embryonic development. Several molecular pathways known to regulate embryonic hair cell progenitors are reactivated in mature chicken epithelia after HC loss. Upon HC damage, Atoh1 becomes reactivated in transdifferentiating and mitotically active SCs (Cafaro et al., 2007). Atoh1 reactivation is essential to form new hair cells, like it is to form hair cells during development (Bermingham, 1999). Notch signaling is down-regulated upon damage in the basilar papilla suggesting that in the mature organ it maintains a repressive state that prevents Atoh1 expression. In agreement, different laboratories have shown that treatment with Notch inhibitors favors Atoh1 reactivation and HC regeneration in the chick basilar papilla and also in the post-natal mammalian cochlea under certain conditions (Cafaro et al., 2007; Mizutari et al., 2013). The ability of SC to respond to Notch blockade dramatically declines after birth, and is lost by post-natal day 6 (Maass et al., 2015).

Human stem cells constitute a reasonable alternative to replace damaged hair cells. Major problems of this approach are the difficulty to deliver treated cells to the damaged areas and their limited ability to integrate in the epithelium. Several groups have developed protocols to differentiate hair cells by mimicking the hair cell development in the embryo. Although this has proved successful, the efficiency of the procedures in hair cell regeneration is still very low (Chen et al., 2012; Ronaghi et al., 2014). In contrast to the low efficiency in replacing hair cells, stem cell therapy has proven surprisingly effective at restoring auditory neurons. The first reports of otic guidance with monolayer cultured human ESCs (hESCs) revealed a propensity to differentiate along an otic neurogenic lineage rather than HC lineage (Chen et al., 2012). Different attempts to generate HCs in culture commonly face the problem that most cells go to the neuronal cell fate, making it very difficult to enrich them in HCs (Chen et al., 2012; Ronaghi et al., 2014). This problem is directly related to the question addressed in the present work. During early stages of development, Neurog1 prevails over Atoh1, thereby forcing neurogenesis and delaying sensorigenesis. This suggests that the default fate is to become a neuron and that sensory competence is silenced. If this is so, the consequence is that production relies mainly on relieving the repression of hair cell competence rather than on the expression of activators. The cellular context of conditionally derived stem cells may be similar to that in the embryo and interference with Neurog1 may open a way to improve the efficiency of HC production.

In summary, understanding the developmental mechanisms involving interactions among cell-to-cell signals and transcription factors is crucial for designing strategies for hearing repair. Developmental studies have shown that the connection between FGF signaling and neurosensory commitment relies on the induction of SoxB1 factors, which set the expression of Neurog1 and Atoh1. The further interaction between Neurog1 and Atoh1 is crucial for neuronal and hair cell specification and for setting the timing for cell diversification. Notch operates at several stages during ear development, but one of them is linked to proneural gene expression and the irreversible commitment to neuronal and hair cell fate of only a fraction of the competent progenitors. Its function is crucial for maintaining dormant the potential of prosensory cells to develop as hair cells, and to keep the regenerative potential of supporting cells. Further work is required to better understand the details of the molecular mechanisms of Atoh1 and Neurog1 regulation, and the development and regeneration of neurons and hair cells.

## ETHICS STATEMENT

The experiments reported in this paper were carried out on E2-E3 chick embryos according to national regulations following a protocol approved by the Ethics Committee of the PRBB.

## AUTHOR CONTRIBUTIONS

HG, GA, and FG: Planning, executing, and writing.

## REFERENCES


## FUNDING

La Marató and BFU2015-67400-P (MINECO/FEDER, UE).

## ACKNOWLEDGMENTS

We are grateful to Joana Neves, Jelena Petrovic, Thomas Schimmang, Marta Ibañes, Juan Camilo Luna, Diogo Castro, and Domingos Henrique for their contribution. We like to thank Marta Linares and Sara Calatayud for technical assistance.


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

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

# WNT/β-Catenin Signaling in Vertebrate Eye Development

### Naoko Fujimura\*

Laboratory of Eye Biology, BIOCEV Division, Institute of Molecular Genetics, Prague, Czechia

The vertebrate eye is a highly specialized sensory organ, which is derived from the anterior neural plate, head surface ectoderm, and neural crest-derived mesenchyme. The single central eye field, generated from the anterior neural plate, divides to give rise to the optic vesicle, which evaginates toward the head surface ectoderm. Subsequently, the surface ectoderm, in conjunction with the optic vesicle invaginates to form the lens vesicle and double-layered optic cup, respectively. This complex process is controlled by transcription factors and several intracellular and extracellular signaling pathways including WNT/β-catenin signaling. This signaling pathway plays an essential role in multiple developmental processes and has a profound effect on cell proliferation and cell fate determination. During eye development, the activity of WNT/β-catenin signaling is tightly controlled. Faulty regulation of WNT/β-catenin signaling results in multiple ocular malformations due to defects in the process of cell fate determination and differentiation. This mini-review summarizes recent findings on the role of WNT/β-catenin signaling in eye development. Whilst this mini-review focuses on loss-of-function and gain-of-function mutants of WNT/β-catenin signaling components, it also highlights some important aspects of β-catenin-independent WNT signaling in the eye development at later stages.

#### Edited by:

Andrea Erika Münsterberg, University of East Anglia, UK

#### Reviewed by:

Paola Bovolenta, Spanish National Research Council, Spain Gunnar Schulte, Karolinska Institutet, Sweden Andrea Streit, King's College London, UK

> \*Correspondence: Naoko Fujimura fujimura@img.cas.cz

#### Specialty section:

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

Received: 24 August 2016 Accepted: 09 November 2016 Published: 30 November 2016

#### Citation:

Fujimura N (2016) WNT/β-Catenin Signaling in Vertebrate Eye Development Front. Cell Dev. Biol. 4:138. doi: 10.3389/fcell.2016.00138 Keywords: retina, WNT, β-catenin, development, differentiation

## OVERVIEW OF EYE DEVELOPMENT IN MICE

During gastrulation, the eye field, a group of the retinal precursor cells, is specified within the anterior neural plate. At this stage, these cells are anteriorly and laterally surrounded by the telencephalic progenitor cells. Subsequently, the eye field is divided into two lateral parts, which extend toward the surface ectoderm and give rise to the optic vesicle (**Figure 1A**; Inoue et al., 2000; Cavodeassi and Houart, 2012; Heavner and Pevny, 2012). The head surface ectoderm thickens to give rise to the lens placode while the optic vesicle subdivides into three parts, namely the presumptive retinal pigment epithelium (RPE), the presumptive neural retina, and the presumptive optic stalk (**Figure 1B**). The optic vesicle subsequently invaginates together with the lens placode to form the double-layered optic cup (**Figure 1C**). The inner part of the optic cup gives rise to the neural retina, meanwhile the outer layer forms the RPE. The ciliary margin (peripheral part of the optic cup) develops to generate the iris and the ciliary body. The lens placodes progresses to form a hollow lens vesicle. Cells in the posterior region differentiate as primary lens fiber cells and elongate to fill the cavity, while the cells in the anterior region become proliferative lens epithelial cells (**Figure 1D**; Fuhrmann, 2008; Cvekl and Ashery-Padan, 2014; Fuhrmann et al., 2014). The retinal vessels arise from the optic nerve head shortly after birth and extend radically to the retinal periphery in the superficial retina. The vasculature then sprouts ventrally to form the deep vascular layer (Gariano and Gardner, 2005).

equator and differentiate as secondary lens fiber cells. Pink color represents the region where the activity of WNT/β-catenin signaling is active, green shows the source of WNTs, blue indicates the region where WNT/PCP signaling is active. (E, F) Schematic representation of WNT/β-catenin signaling in the early lens development and in the RPE development, respectively. E. The periocular mesenchyme secretes TGFβ, which signals to the non-lens surface ectoderm. WNT2b is induced by TGFβ and activates WNT/β-catenin signaling in order to suppress the lens fate by repressing expression of Pax6. In the lens placode, WNT/β-catenin is inhibited by Pax6 which initiates lens development. (F) The surface ectoderm secretes WNTs which activate WNT/β-catenin signaling in the RPE. This signaling induces expression of Otx2 and Mitf which in cooperation with Pax6 control the RPE developments.

## WNT SIGNALING

WNTs can couple to various receptors and trigger different downstream signaling cascades including the non-canonical WNT/planar cell polarity (PCP), WNT/Ca2+, and the canonical WNT/β-catenin signaling pathway, the focus of this review. WNT/β-catenin signaling is initiated by binding of the WNTs to the Frizzled/LRP5/6 receptor complex, which leads to the accumulation of β-catenin and nuclear translocation. In the nucleus, β-catenin interacts with the TCF/LEF family of transcription factors and regulates their target genes. In the absence of WNTs, β-catenin is phosphorylated by a "destruction complex" composed of multiple proteins, including AXIN2 and GSK3β, and targeted for degradation (Loh et al., 2016). In addition to its critical role as a transcriptional co-activator, βcatenin acts as a central component of the adherens junction by forming a link between cadherins and the actin cytoskeleton (Heuberger and Birchmeier, 2010). WNT/PCP signaling does not use β-catenin, but activates the Rho family GTPases and JNK pathway, which results in changes in cytoskeleton and cell polarity (Loh et al., 2016). WNT signaling is modulated by a number of WNT-sequestering proteins, such as DKKs and SFRPs, which prevent ligand-receptor interactions (Cruciat and Niehrs, 2013).

**Abbreviations:** pOV, presumptive optic vesicle; OV, optic vesicle; SE, head surface ectoderm; ME, extraocular mesenchyme; pRPE, presumptive retinal pigment epithelium; pNR, presumptive neural retina; pOS, presumptive optic stalk; LP, lens placode; RPE, retinal pigment epithelium; LPT, lens pit; OS, optic stalk; CM, ciliary margin; LE, lens epithelium; ON, optic nerve.

## THE LENS

WNT signaling plays essential roles in eye organogenesis (Fuhrmann, 2008). During lens development, WNT/β-catenin signaling is active in the periocular surface ectoderm and lens epithelium (Stump et al., 2003; Smith et al., 2005; Kreslova et al., 2007; Machon et al., 2010; Carpenter et al., 2015). Conditional deletion of β-catenin in the presumptive lens placode and surrounding head surface ectoderm results in abnormal lens morphogenesis due to cell-cell adhesion defects. Conversely, the lens induction in the β-catenin loss-of-function mutant is not affected as expression of lens-specific markers is maintained (Smith et al., 2005). Consistently, a null mutation in Lrp6, which is expressed throughout the eye at the optic vesicle stage, does not have a profound effect on the lens induction (Stump et al., 2003; Smith et al., 2005). Interestingly, ectopic lentoid bodies are formed in the periocular surface ectoderm, where WNT/βcatenin signaling is inactivated in β-catenin-deficient mutants. Although the adherens junction is disrupted, ectopic lentoid bodies are not observed in the E-cadherin/N-cadherin or Scribs conditional knockout mice generated using the same Cre line (Pontoriero et al., 2009; Yamben et al., 2013). Thus, formation of ectopic lentoid bodies is mediated by the inactivation of WNT/βcatenin signaling rather than by cell-cell adhesion defects. In addition, ectopic activation of WNT/β-catenin signaling by expression of constitutively active β-catenin leads to inhibition of the lens formation (Smith et al., 2005; Machon et al., 2010). Taken together, WNT/β-catenin signaling is not required for the lens fate determination, however it inhibits the lens formation and appears to suppress the lens fate in the periocular ectoderm. The precise regulation of WNT/β-catenin signaling is required to ensure the correct patterning of the ocular tissue.

WNT/β-catenin signaling is regulated by TGFβ signaling and Pax6 in the surface ectoderm at the optic vesicle stage (**Figure 1E**). The migrating neural crest cells inhibit the lens specification, while their ablation results in ectopic lens formation (Bailey et al., 2006). In chick embryos, the neural crest cells secrete multiple TGFβs which activate WNT/β-catenin signaling by inducing WNT2b in the adjacent non-lens ectoderm. The lens fate in presumptive lens ectoderm explants can be suppressed by the neural crest, constitutively active β-catenin, as well as TGFβ. Interestingly, the expression of lens markers is restored when these explants are cultured with TGFβ and WNTsequestering protein FZD8-CRD, a truncated and soluble form of the WNT receptor. This indicates that lens suppression by the neural crest-derived TGFβ is dependent on WNT/β-catenin signaling (Grocott et al., 2011). WNT2b null mice display no ocular defects and multiple WNTs are expressed in the surface ectoderm, therefore additional WNTs are required for the process in mice (Tsukiyama and Yamaguchi, 2012; Carpenter et al., 2015).

Pax6 is expressed in the presumptive lens placode and Pax6 null mutation results in failure of the lens formation (Hill et al., 1991; Grindley et al., 1995). It has been shown that Pax6 regulates the expression of Sfrp2, and Dkk1. In Pax6 deficient presumptive lens placode, Sfrp2 is down-regulated and WNT/β-catenin signaling is ectopically activated (Machon et al., 2010). However, it is unlikely that Sfrp2 acts as a downstream effector as lens induction is not affected in the Sfrp1−/−; Sfrp2−/<sup>−</sup> mice (Sugiyama et al., 2013). On the other hand, the role of Dkk1 in the lens induction remains elusive as Dkk1 null embryos lack the anterior head structure including the eyes (Mukhopadhyay et al., 2001). Interestingly, PAX6 ChIP sequencing using human neuroectodermal cells has shown that PAX6 binds to a variety of genes, which regulate WNT signaling (Bhinge et al., 2014). Further studies are necessary to understand how Pax6 counteracts WNT/β-catenin signaling.

At later stages of development, WNT/β-catenin signaling is required for the formation and maintenance of the lens epithelium (Stump et al., 2003; Cain et al., 2008; Martinez et al., 2009). Interestingly, WNT/β-catenin signaling is reduced in the lens epithelium of the Sfrp1−/−; Sfrp2−/<sup>−</sup> embryos (Sugiyama et al., 2013). SFRP1/2 are primarily characterized as WNT-sequestering proteins, however they can activate WNT/βcatenin signaling by facilitating the diffusion of WNTs or suppressing WNT/PCP pathway which can antagonize WNT/βcatenin signaling (Satoh et al., 2008; Mii and Taira, 2009). Additionally, Sfrp1/2 can also inhibit BMP and Notch signaling, which are required for lens development, thus mis-regulation of these signaling pathways might also be responsible for the defects in the Sfrp1/2-deficient lens (Misra and Matise, 2010; Esteve et al., 2011a).

Although WNT/β-catenin signaling is not required for the lens fiber development, there are indications that the alignment and orientation of lens fiber cells are dependent on the WNT/PCP signaling pathways (Chen et al., 2008; Sugiyama et al., 2010, 2011). In the lens overexpressing Sfrp2, the fiber orientation is severely disrupted and expression of components of the WNT/PCP pathway is down-regulated (Chen et al., 2008; Sugiyama et al., 2010). WNT5, which activates the PCP pathway is secreted from the lens epithelium and WNT5 promotes the directed behavior of lens fiber cells in the lens explants (Dawes et al., 2014).

## THE RPE

Signals from neighboring tissues are crucial for the accurate specification of the neural retina and the RPE within the optic vesicle. The dorsal optic vesicle receives signals from the extraocular mesenchyme and the head surface ectoderm to differentiate into the RPE (Fuhrmann et al., 2000; Martínez-Morales et al., 2004; Bharti et al., 2006; Steinfeld et al., 2013; Carpenter et al., 2015). During retinal development, WNT/βcatenin signaling is active in the dorsal optic vesicle which gives rise to presumptive RPE at the optic vesicle stage and is subsequently restricted to the peripheral RPE (Liu et al., 2006; Fujimura et al., 2009; Westenskow et al., 2009; Hägglund et al., 2013). The RPE transdifferentiates into the neural retina in the β-catenin-deficient RPE at the optic cup stage, as evidenced by loss of the RPE markers Mitf and Otx2 and by the ectopic expression of neural retinal markers, such as Chx10 and Rax (Fujimura et al., 2009; Westenskow et al., 2009; Hägglund et al., 2013). The β-catenin-deficient RPE preserves intact adherens junctions at the optic cup stage, although cell-cell adhesion is disrupted at later stages (Fujimura et al., 2009; Westenskow et al., 2009). Interestingly, γ-catenin, a paralog of β-catenin, can substitute β-catenin in cell adhesion complexes in various developmental contexts (Huelsken et al., 2000; Posthaus et al., 2002; Zhou et al., 2007). The lack of β-catenin in the adherens junctions might be compensated by γ-catenin as evidenced by the presence of γ-catenin in the β-catenin-deficient RPE at the optic cup stage. Thus, the transdifferentiation is probably caused by loss of WNT/β-catenin signaling (Fujimura et al., 2009). A similar phenomenon is observed in the optic cup derived from the mouse embryonic stem cell aggregates in vitro (Eiraku et al., 2011). Treatment with a WNT secretion inhibitor reduces the number of the RPE cells, while WNT3a promotes the RPE differentiation and suppresses the neural retina generation (Eiraku et al., 2011). Interestingly, ectopic activation of WNT/β-catenin signaling in the entire RPE also results in disruption of the RPE patterning. The peripheral RPE remains normal, while the central part, in which WNT/β-catenin signaling is ectopically active, loses expression of the RPE markers. In contrast to β-catenin-deficient mutants, the RPE is not transdifferentiated to the neural retina (Fujimura et al., 2009). Thus, the activity of WNT/β-catenin signaling is spatially and temporally regulated during the RPE development.

WNT/β-catenin signaling regulates RPE development in cooperation with Mitf, Otx2, and Pax6 (**Figure 1F**). Expression of Mitf and Otx2 is directly regulated by WNT/β-catenin signaling (Fujimura et al., 2009; Westenskow et al., 2009). Furthermore, ectopic expression of both Otx2 and β-catenin in the presumptive chick neural retina promotes the RPE fate while the ectopic expression of Otx2 or β-catenin alone is not sufficient. Therefore, β-catenin, together with Otx2, induces a change in cell fate from retinal progenitor cells to the presumptive RPE (Westenskow et al., 2010). Furthermore, β-catenin directly interacts with MITF and promotes Mitf-mediated transcription (Schepsky et al., 2006). A recent study has shown that PAX6 acts in synergy with βcatenin and MITF to activate the promoters of melanogenic genes Tyr and Trp-1 (Fujimura et al., 2015).

Although the identity of the specific WNTs involved in RPE development remains elusive, a recent study has shown that WNTs from the surface ectoderm are necessary for this process (Carpenter et al., 2015). During early eye development, the WNT transporter Wntless is expressed in the presumptive lens placode, the periocular surface ectoderm, the periocular mesenchyme at the optic vesicle stage, and it is also detected in the peripheral retina and the RPE at later stages (Carpenter et al., 2015). Conditional deletion of Wntless in the presumptive lens leads to inactivation of WNT/β-catenin signaling in the peripheral retina and periocular mesenchyme (Carpenter et al., 2015). Moreover, the number of RPE cells is reduced in Wntlessdeficient mice (Carpenter et al., 2015). Despite the presence of multiple WNTs and Wntless in the periocular mesenchyme, conditional inactivation of Wntless in the periocular mesenchyme and RPE does not affect the eye development or the activity of WNT/β-catenin signaling (Carpenter et al., 2015). It remains elusive how WNTs disperse from the periocular mesenchyme to the WNT-responsive tissue in the optic cup. There are, however, indications that heparan sulfate proteoglycans (HSPG) are involved in the distribution of WNTs within the eye. HSPGs are located on the cell surface and in the extracellular matrix and have been implicated in a number of signaling pathways including WNT (Sarrazin et al., 2011). In the context of WNT signaling transduction, HSPGs play an essential role in organizing the extracellular distribution of WNTs and they maintain the activity of WNTs by preventing their aggregation in the extracellular environment (Fuerer et al., 2010; Matsuo and Kimura-Yoshida, 2014). Interestingly, conditional deletion of Ext1, a key HSPG synthetic enzyme, in the periocular mesenchyme leads to severe ocular malformations including the defects in the peripheral RPE development (Iwao et al., 2010). It has not been shown whether WNT/β-catenin signaling is affected in the peripheral optic cup of the Ext1 deficient mice, however Ext1 is required for the activation of the WNT11/β-catenin pathway in Xenopus embryos (Tao et al., 2005). Thus, HSPG in the periocular mesenchyme might mediate the distribution of WNTs from the surface ectoderm.

## THE CILIARY MARGIN

WNT/β-catenin signaling is active in the developing ciliary margin or peripheral retina, but it is inactive in the central retina (Liu et al., 2003, 2007; Cho and Cepko, 2006). Several WNT signaling members, such as WNT2b, Frizzled-4 (FZD4), and Lef1 are expressed in the ciliary margin (Trimarchi et al., 2009). Overexpression of a constitutively active form of β-catenin leads to the expansion of the ciliary margin at the expense of the central retina (Cho and Cepko, 2006; Liu et al., 2007; Trimarchi et al., 2009). In addition, Axin2 null embryos display multiple ocular phenotypes including expansion of the ciliary margin (Alldredge and Fuhrmann, 2016).

Several studies indicate that WNT/β-catenin signaling activity in the peripheral retina is controlled by Sfrp1/2, Foxg1, and Sox2 (Matsushima et al., 2011; Esteve et al., 2011b; Fotaki et al., 2013). As mentioned above, it has been suggested that WNTsequestering proteins SFRP1/2 can activate WNT/β-catenin signaling (Bovolenta et al., 2008). In the Sfrp1−/−; Sfrp2−/<sup>−</sup> embryos, this signaling is inactive in the peripheral retina, which displays neural retinal characteristics (Esteve et al., 2011b). Conversely, restriction of WNT/β-catenin signaling to the ciliary margin has been shown to be mediated by Foxg1 and Sox2 (Matsushima et al., 2011; Fotaki et al., 2013). In Foxg1−or Sox2 deficient retina, WNT/β-catenin signaling are up-regulated in the peripheral retina and the ciliary margin expands at the expense of the neural retina (Matsushima et al., 2011; Fotaki et al., 2013). foxg1 suppresses WNT/β-catenin signaling by directly repressing the transcription of WNTs in the forebrain of zebrafish (Matsushima et al., 2011). SOX2 interferes with WNT/β-catenin signaling by binding β-catenin in the osteoblast lineage (Seo et al., 2011). Taken together, it is likely that multiple mechanisms control the activity of WNT/β-catenin signaling in the ciliary margin.

## THE DORSO-VENTRAL PATTERNING IN THE OPTIC CUP

In addition to the correct patterning of the lens and the RPE development, WNT/β-catenin signaling is required for the maintenance of the dorsal retinal identity (Veien et al., 2008; Zhou et al., 2008; Hägglund et al., 2013). Conditional inactivation of β-catenin in the early optic cup results in the down-regulation of dorsal retinal markers, such as Bmp4 and expansion of the ventral retinal markers, such as Vax2 (Hägglund et al., 2013). Similarly, loss of Lrp6 causes dorso-ventral patterning defects in the neural retina (Zhou et al., 2008). Consistently, the expression of dorsal retinal markers are attenuated in a transgenic fish which overexpresses dkk1 or dominant-repressor form of tcf3. This phenotype is rescued by LiCl, which promotes the accumulation of cytoplasmic β-catenin by inhibiting GSK3β (Veien et al., 2008). Thus, the role of WNT/β-catenin signaling in the dorso-ventral patterning within the retina seems to be evolutionarily conserved.

## THE RETINAL VASCULAR SYSTEM

WNT/β-catenin signaling plays an essential role in the retinal vascular development. In genetic disorders, such as Norrie disease and Familial Exudative Vitreoretinopathy, retinal hypovascularization is caused by loss-of-function mutations in the Norrin disease protein (Norrin), FZD4, or LRP5 genes. Norrin contains separate binding sites for FZD<sup>4</sup> and for LRP5 (Ke et al., 2013). Activation of FZD4/β-catenin signaling by Norrin requires the presence of either LRP5 or LRP6 (Ye et al., 2009). Although Lrp5 can compensate for the loss of Lrp6 (and vice versa) in the postnatal brain vasculature, Lrp5 plays a major role and Lrp6 plays a minor role in the retinal vascularization (Zhou et al., 2014; Huang et al., 2016). Norrin secreted from Müller glial cells binds to FZD<sup>4</sup> in the endothelial cells and regulates retinal vascular development (Xu et al., 2004; Junge et al., 2009; Ye et al., 2009; Wang et al., 2012). The retinal vascular defects caused by ablation of Norrin are rescued by stabilizing β-catenin, while ectopic expression of dominant negative Tcf4 in the endothelial cells mimics the phenotype. This indicates that Norrin/FZD<sup>4</sup> signaling acts via β-catenin signaling (Zhou et al., 2014). In addition, WNT/β-catenin signaling in the retinal vascular system is regulated by the EST transcription factor Erg, which plays a critical role in vascular development and angiogenesis (Birdsey et al., 2015). Erg controls WNT/β-catenin signaling by

## REFERENCES


promoting β-catenin stability and regulating transcription of FZD<sup>4</sup> (Birdsey et al., 2015).

β-catenin-independent WNT signaling pathway is also required for the retinal vascular system development (Stefater et al., 2011; Korn et al., 2014; Franco et al., 2016). The endothelial cells express preferentially non-canonical WNTs, such as WNT5a and WNT11. Conditional deletion of Wntless or WNT5a in the endothelial cells leads to significant decrease in vascular density due to excessive vessel regression (Korn et al., 2014; Franco et al., 2016).

## CONCLUSION

The activity of WNT/β-catenin signaling is tightly regulated during eye development and mis-regulation of the signaling results in multiple ocular malformations due to defects in the process of cell fate determination and differentiation. Studies of conditional knockout mice of various members of the WNT/βcatenin signaling pathway indicate that WNT/β-catenin signaling is essential for eye development by controlling the correct patterning of the ocular tissue, promoting the differentiation of the retinal pigment epithelium, controlling the morphogenesis of the optic cup, and maintaining the dorsal retinal identity. Further research is necessary to clarify the mechanisms through which WNT/β-catenin signaling integrates into the genetic regulatory networks controlling the eye development in the vertebrate.

## AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and approved it for publication.

## FUNDING

This work was supported by the Ministry of Education, Youth and Sports of CR within the LQ1604 National Sustainability Program II (Project BIOCEV-FAR).

## ACKNOWLEDGMENTS

I apologize to the investigators whose original work was not cited due to space constraints. I thank Drs. Z. Kozmik for advice, continuous support, and mentoring and N. Chambers and C. Pantzartzi for critical reading of the manuscript.

of the vertebrate eye. Pigment Cell Res. 19, 380–394. doi: 10.1111/j.1600-0749. 2006.00318.x


embryogenesis. Philos. Trans. R. Soc. Lond. B Biol. Sci. 369:20130545. doi: 10.


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

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

## PTK7 Faces the Wnt in Development and Disease

#### Hanna Berger <sup>1</sup> , Andreas Wodarz 2, 3 and Annette Borchers 1, 4 \*

<sup>1</sup> Department of Biology, Molecular Embryology, Philipps-Universität Marburg, Marburg, Germany, <sup>2</sup> Department of Anatomy I, Molecular Cell Biology, University of Cologne, Cologne, Germany, <sup>3</sup> Cluster of Excellence - Cellular Stress Responses in Aging-Associated Diseases, University of Cologne, Cologne, Germany, <sup>4</sup> DFG Research Training Group, Membrane Plasticity in Tissue Development and Remodeling, GRK 2213, Philipps-Universität Marburg, Marburg, Germany

PTK7 (protein tyrosine kinase 7) is an evolutionarily conserved transmembrane receptor regulating various processes in embryonic development and tissue homeostasis. On a cellular level PTK7 affects the establishment of cell polarity, the regulation of cell movement and migration as well as cell invasion. The PTK7 receptor has been shown to interact with ligands, co-receptors, and intracellular transducers of Wnt signaling pathways, pointing to a function in the fine-tuning of the Wnt signaling network. Here we will review recent findings implicating PTK7 at the crossroads of Wnt signaling pathways in development and disease.

Keywords: PTK7, Wnt signaling, planar cell polarity, cancer, neural tube defect, scoliosis

#### Edited by:

Sigmar Stricker, Free University of Berlin, Germany

#### Reviewed by:

Madelon Maurice, Utrecht University, Netherlands Frank Thévenod, Witten/Herdecke University, Germany

> \*Correspondence: Annette Borchers borchers@uni-marburg.de

#### Specialty section:

This article was submitted to Signaling, a section of the journal Frontiers in Cell and Developmental Biology Received: 23 December 2016 Accepted: 21 March 2017 Published: 05 April 2017

#### Citation:

Berger H, Wodarz A and Borchers A (2017) PTK7 Faces the Wnt in Development and Disease. Front. Cell Dev. Biol. 5:31. doi: 10.3389/fcell.2017.00031

## INTRODUCTION

PTK7 (protein tyrosine kinase 7) is an evolutionarily conserved transmembrane receptor with a broad range of functions in tissue development and homeostasis. Originally identified as a gene upregulated in colon carcinoma cells and accordingly named colon carcinoma kinase 4 (CCK-4) (Mossie et al., 1995) it was later shown to affect various aspects of cell-cell communication and movement. PTK7 controls tissue morphogenesis and patterning by affecting cell polarity, migration as well as tissue regeneration and wound healing (Lu et al., 2004; Shnitsar and Borchers, 2008; Caddy et al., 2010; Lee et al., 2011; Lander and Petersen, 2016). Additionally its function in adult tissue homeostasis is demonstrated by the fact that misregulation of PTK7 expression correlates with development of cancer and its progression to metastasis in various cellular contexts (reviewed in Dunn and Tolwinski, 2016). Furthermore, mutations in PTK7 have been implicated in scoliosis and human neural tube closure defects, demonstrating its clinical relevance (Hayes et al., 2014; Wang et al., 2015; Grimes et al., 2016). Since the first publication on PTK7/CCK-4 (Mossie et al., 1995) more than 20 years ago over 120 publications have followed. Although its signaling function is still far from being understood, recent findings provide compelling evidence that PTK7 is a regulator of Wnt signaling pathways. In this review we will summarize recent findings and take a look at PTK7's function in distinct Wnt signaling pathways.

Secreted glycoproteins of the Wnt family are key regulators of development and disease. Wnt ligands regulate a wide range of processes including primary embryonic axis specification, organogenesis and stem cell proliferation. Further, deregulated Wnt signaling has been implicated in various diseases like colon and breast cancer, melanoma, and neurodegenerative disorders (MacDonald et al., 2009; Clevers and Nusse, 2012; Anastas and Moon, 2013; Inestrosa and Varela-Nallar, 2014). Wnt ligands activate distinct downstream signaling pathways, and historically the first described, ß-catenin-dependent, signaling cascade is referred to as the "canonical" Wnt signaling pathway, while later discovered, ß-catenin-independent pathways were termed "non-canonical." Canonical Wnt signaling (Logan and Nusse, 2004; MacDonald et al., 2009) is activated by binding of the Wnt ligand to a receptor complex consisting of the seven-pass transmembrane Frizzled (Fz) receptor and the low-density lipoprotein receptor-related protein 6 (LRP6) (MacDonald and He, 2012). Wnt binding to the Fz/LRP6 receptor complex leads to inactivation of glycogen synthase kinase 3ß (GSK3ß) regulating various intracellular substrates. One of these is the transcriptional co-activator ßcatenin, which is phosphorylated and thereby targeted for proteasomal degradation. Thus, in the presence of Wnt ligands, ß-catenin is stabilized, enters the nucleus and regulates in combination with transcription factors of the Lef (lymphoid enhancer-binding factor) and Tcf (T cell factor) family the transcription of target genes. In contrast to canonical Wnt signaling, non-canonical Wnt signaling pathways encompass a complex network of signal transducers that do not activate ß-catenin, but use alternative modes of downstream signaling (reviewed in Niehrs, 2012). Here, we will focus on the planar cell polarity (PCP) pathway, as PTK7 has been implicated in its regulation.

The PCP pathway (Goodrich and Strutt, 2011; Yang and Mlodzik, 2015) determines the orientation of cells in the plane of an epithelium and is one of the best-characterized noncanonical Wnt signaling pathways. PCP was first described in Drosophila, where genetic screens discovered its function in the polarization of adult cuticular structures. According to mutant phenotypes showing wing hair polarity defects the genes Frizzled (Fz) and Disheveled (Dsh) were identified. Other core PCP proteins include the four-pass transmembrane protein Van Gogh (Vang, Strabismus), the atypical cadherin Flamingo (Fmi, Celsr) and intracellular components like Prickle (Pk) and Diego (Dgo). These proteins confer intra- and intercellular signaling, thereby aligning PCP in neighboring cells. Complementary studies in vertebrates revealed that these core PCP proteins are also required for the polarization of vertebrate tissues, like the orientation of hair follicles in the epidermis or the sensory hair cells in the inner ear (Montcouquiol et al., 2006; Simons and Mlodzik, 2008; Wallingford, 2012). Furthermore, these proteins are also involved in the polarized localization of cilia, microtubule-based protrusions that are found on the surface of most vertebrate cells and required for fluid movement during development and homeostasis (Wallingford, 2010; Wallingford and Mitchell, 2011). In addition to the polarization of tissues, loss of function studies using the mouse, zebrafish and Xenopus model systems demonstrated that PCP signaling also affects morphogenetic cell movements shaping the embryonic body. One of these is convergent extension, a cell movement whereby cells intercalate in a way that a tissue converges in one direction and extends in the perpendicular direction (Wallingford et al., 2002; Wallingford, 2012). Convergent extension is required to drive gastrulation and neural tube closure. Consequently, misregulation of PCP signaling leads to severe gastrulation and neurulation defects in mouse, zebrafish and Xenopus embryos. Since the discovery of vertebrate PCP phenotypes, these have also contributed to the identification of novel vertebrate regulators of PCP without previous knowledge of a Drosophila phenotype. One of these genes, which was identified by its mouse neural tube closure and inner ear hair polarity defect, is PTK7.

## PTK7 AFFECTS Wnt SIGNALING PATHWAYS

Vertebrate PTK7 is according to the current criteria a bona fide PCP regulator. Using a mouse gene trap-screen for transmembrane proteins with a function in neural development, PTK7 mutants were identified showing a combination of severe neural tube closure and inner ear polarity defects (Lu et al., 2004). Based on this mutant phenotype, which is typical for known regulators of PCP (Hamblet et al., 2002; Curtin et al., 2003; Montcouquiol et al., 2003), as well as its genetic interaction with Vangl2, PTK7 was added to the list of vertebrate PCP regulators. Further functional studies using mouse, zebrafish and Xenopus confirmed a function for PTK7 in processes that are regulated by PCP signaling, including convergent extension movements during gastrulation, neurulation and Wolffian duct elongation, as well as neural crest migration and wound healing (**Table 1**). Surprisingly, although PTK7 appears to be a core regulator of vertebrate PCP, classical PCP phenotypes have so far not been reported for the Drosophila orthologs of PTK7, off-track (otk), and off-track2 (otk2). These two genes, which are the result of a tandem gene duplication, function redundantly in the tubular morphogenesis of the male ejaculatory duct, leading to male sterility in the otk, otk2 double mutant (Linnemannstons et al., 2014). Intriguingly, mesoderm-specific knock-out of PTK7 in the mouse resulted in tubular morphogenesis defects in the Wolffian duct, again leading to male sterility (Xu et al., 2016). In both cases, tubular morphogenesis defects upon loss of Otk/Otk2 or PTK7 may be caused by the failure to properly execute convergent extension movements. Thus, although the Drosophila mutants do not display the classical PCP defects, PTK7/Otk may play an evolutionarily conserved role in the regulation of cell movements.

The molecular mechanism by which PTK7 affects PCP signaling is currently unclear. However, as PTK7 interacts with Wnt ligands and known Wnt receptors (**Table 2**) it likely affects PCP by functioning as a Wnt receptor. This is also supported by the structure of PTK7, which is highly reminiscent of receptor tyrosine kinases. PTK7 consists of seven extracellular immunoglobulin domains, a transmembrane domain, and an evolutionarily conserved tyrosine kinase homology domain. The kinase homology domain of PTK7 lacks catalytic activity (Miller and Steele, 2000; Kroiher et al., 2001), but serves as an interaction site for intracellular signaling molecules like ß-catenin, Dsh, and Src (Shnitsar and Borchers, 2008; Puppo et al., 2011; Andreeva et al., 2014). PTK7 interacts with distinct Wnt receptors including Fz7, LRP6, and Ror2 (Peradziryi et al., 2011; Bin-Nun et al., 2014; Linnemannstons et al., 2014; Martinez et al., 2015; Podleschny et al., 2015), indicating that PTK7 affects canonical and noncanonical Wnt signaling pathways. This is also reflected by its evolutionarily conserved interaction with different Wnt ligands that are supposed to signal via both canonical and non-canonical pathways (Peradziryi et al., 2011; Linnemannstons et al., 2014; Martinez et al., 2015). While its requirement for PCP signaling is firmly established, the function of PTK7 in canonical Wnt signaling remains controversial. PTK7 has been reported to activate canonical Wnt signaling in the context of Spemann's organizer formation (Puppo et al., 2011) and the specification

#### TABLE 1 | PCP phenotypes upon PTK7 loss of function in vertebrates.


#### TABLE 2 | PTK7 interaction partners with a known function in Wnt signaling.


of posterior neural tissue (Bin-Nun et al., 2014) in Xenopus embryos. However, PTK7 inhibits canonical Wnt signaling in Xenopus double axis and luciferase reporter assays (Peradziryi et al., 2011). This was confirmed by ptk7 mutant zebrafish, which showed an upregulation of ß-catenin target gene expression, suggesting that PTK7 functions in attenuating canonical Wnt signaling (Hayes et al., 2013). Conflicting results were also obtained analyzing the interaction of PTK7 with Wnt ligands using immunoprecipitation of overexpressed/tagged constructs. While we found interaction of PTK7 with canonical Wnt3a and Wnt8 but not non-canonical Wnt5a (Peradziryi et al., 2011) in Xenopus lysates, Martinez et al. observed an interaction with non-canonical Wnt5a, but not canonical Wnt1 (Martinez et al., 2015) in HEK293T cells. Some of these contradictions may be explained by receptor context. Using secreted proteins we showed that the extracellular domain of PTK7 requires the extracellular Fz7 domain for interaction with recombinant Wnt3a (Peradziryi et al., 2011). Conversely, Wnt5a binding may require the Ror2 co-receptor. Although Martinez et al. confirmed interaction of PTK7 and Wnt5a in cells that were depleted of Ror2 using a specific siRNA, there may still be sufficient endogenous Ror2 present to mediate binding. Thus, studies analyzing direct interaction of PTK7 and Wnt ligands are currently lacking. Furthermore, experiments testing Wnt binding of combinatorial PTK7 co-receptor complexes are required. As PTK7 is a versatile receptor interacting not only with Wnt coreceptors but also with plexin and VEGF receptors (reviewed in Peradziryi et al., 2012), the latter interactions may also contribute to tissue-specific differences. Thus, future research will have to elucidate how receptor context affects PTK7 signaling and its functions in distinct Wnt signaling pathways.

## PTK7 AND DISEASE

As PTK7 has a crucial function in the regulation of Wnt signaling pathways known to be essential for embryonic development and homeostasis, mutations in the human PTK7 gene are likely of clinical relevance. PTK7 was identified as a gene upregulated in colon carcinoma cells and appears to be misregulated in a variety of cancers (Dunn and Tolwinski, 2016). Furthermore, PTK7 mutations have recently been implicated in the etiology of neural tube defects and scoliosis (Hayes et al., 2014; Wang et al., 2015). Here, we will briefly describe these respective disorders and look at the human PTK7 gene variants identified in this context as well as their functional implications.

The connection between PTK7 and cancer has so far mostly been deduced on the basis of up- or downregulation of PTK7 in a variety of cancer types. PTK7 levels were reported to be increased in esophageal (Shin et al., 2013), gastric (Lin et al., 2012), colorectal (Lhoumeau et al., 2015), breast (Gartner et al., 2014), intrahepatic bile duct (Jin et al., 2014), prostate (Zhang et al., 2014), and lung carcinoma (Chen et al., 2014), as well as liposarcoma (Gobble et al., 2011). In other cancer types PTK7 was shown to be downregulated, including lung squamous cell carcinoma (Kim et al., 2014), ovarian carcinoma (Wang et al., 2014) and metastatic melanoma (Easty et al., 1997). While the mechanistic contribution of PTK7 to the respective tumor phenotypes is unclear at present, the upregulation of PTK7 in many tumor types makes it an attractive tumor marker and therapeutic target. Indeed, the first PTK7 specific reagents with potential clinical applications have now been published, including a PTK7-specific fluorescently labeled aptamer for in vivo detection of tumor tissue (Calzada et al., 2017). Very interestingly, PTK7 has recently been established as a marker for normal colon stem cells (Jung et al., 2015) and as a marker for tumor initiating cells in triple-negative breast cancer, ovarian cancer and non-small cell lung cancer (Damelin et al., 2017). The authors of the latter study also developed a PTK7-targeted antibody-drug conjugate and showed that its application reduces tumor initiating cells and induces sustained tumor regressions, paving the way for a PTK7-directed anti-tumor therapy (Damelin et al., 2017).

Neural tube defects are among the most common human birth defects affecting 1 per 1000 live births and are caused by environmental as well as genetic factors (Wilde et al., 2014). PCP genes are likely among the genetic factors contributing to the etiology of human neural tube closure defects as loss of function mutants of PTK7, Vangl, Celsr, Fz, Dvl, and Scribble result in the most severe neural tube closure defects called craniorachischisis (Gerrelli and Copp, 1997; Kibar et al., 2001; Hamblet et al., 2002; Curtin et al., 2003; Murdoch et al., 2003; Lu et al., 2004; Wang et al., 2006), whereby the neural tube fails to close from the midbrain-hindbrain boundary to the base of the spine. Indeed, rare mutations with a predicted damaging role were identified for a number of PCP genes including Vangl1/2, Celsr1, Fzd6, Dvl2, Prickle, and Scribble (Kibar et al., 2007; De Marco et al., 2014). Furthermore, the analysis of a cohort of 473 patients with various forms of neural tube defects identified six rare PTK7 sequence variants (Wang et al., 2015). Interestingly, five of these mutations are located in the extracellular domain of PTK7, which serves as interaction site for Wnt ligands as well as Fz7 and Ror2 receptors (**Table 2**; Peradziryi et al., 2011; Martinez et al., 2015; Podleschny et al., 2015). Whether these interactions are affected in the potentially pathogenic sequence variants is currently unclear and functional validation assays testing their efficiency to rescue for example Xenopus or zebrafish loss of function phenotypes are still missing. Nevertheless, the extracellular domain was shown to be important for promoting PCP and inhibiting canonical Wnt signaling. In fact, deletion of the extracellular domain abolished PTK7's ability to inhibit canonical Wnt signaling in Xenopus reporter assays (Peradziryi et al., 2011). Conversely, a membranetethered PTK7 extracellular fragment was sufficient to rescue excess Wnt/ß-catenin signaling and PCP morphogenesis defects in maternal-zygotic ptk7 mutant zebrafish (Hayes et al., 2013). Thus, these data point to PTK7 as a risk factor for neural tube closure defects and stress the functional importance of its extracellular domain.

In addition to neural tube defects, PTK7 has also been implicated in the pathogenesis of scoliosis, a complex genetic disorder characterized by a three-dimensional spinal curvature. Congenital scoliosis (CS) is apparent at birth and involves abnormal vertebrae development, while idiopathic scoliosis is diagnosed during adolescence and does not show vertebral malformations. Ptk7 mutant zebrafish were recently discovered as a model for congenital and idiopathic scoliosis. Maternalzygotic ptk7 (MZptk7) mutant zebrafish exhibit vertebral abnormalities at larval stages, phenotypically resembling congenital scoliosis. Further, zygotic ptk7 (Zptk7) mutants show late onset spinal curvatures consistent with the idiopathic form of scoliosis (Hayes et al., 2014). Analysis of maternal-zygotic mutants showed that PTK7 positively regulates PCP-dependent morphogenesis, while it attenuates ß-catenin-dependent canonical Wnt signaling (Hayes et al., 2013). Thus, segmentation and somite patterning are disturbed, likely causing the observed vertebral abnormalities. In contrast, zygotic ptk7 mutants did not show defects in segmentation and somite patterning, but developed late spinal curvatures resembling idiopathic scoliosis (Hayes et al., 2014; Grimes et al., 2016). They showed defects in ependymal cell cilia development leading to irregularities in the cerebrospinal fluid (CSF) flow. Moreover, the brain ventricles revealed a severe hydrocephalus, a condition associated with loss of cilia function. Consistently, the number of motile cilia was reduced and if cilia were present they lacked the correct polarization. Transgenic reintroduction of wild-type PTK7 in motile ciliated cell lineages rescued all phenotypes, proving a specific function of PTK7 in motile ciliated cells. The authors hypothesized that impaired cerebrospinal fluid flow due to abnormal cilia function is most likely the cause of scoliosis in ptk7-deficient zebrafish (Grimes et al., 2016). The connection of PTK7 to scoliosis was further evidenced by the isolation of a novel PTK7 mutation from a single patient suffering from idiopathic scoliosis. This mutation, hPTK7P545A, exhibits a proline to alanine substitution in the sixth extracellular immunoglobulin domain thereby affecting PCP and canonical Wnt signaling function (Hayes et al., 2014). In fact, in contrast to wild-type human PTK7, the hPTK7P545A failed to rescue PCP-dependent axial extension defects as well as nervous system patterning defects caused by Wnt8 overexpression (Hayes et al., 2014). Further, the mutant protein accumulated at the plasma membrane, indicating altered protein stability and/or trafficking of this mutant compared to the wild-type protein. As PTK7 forms co-receptor complexes with Fz7 and LRP6 (Peradziryi et al., 2011; Bin-Nun et al., 2014; Linnemannstons et al., 2014), which were shown to be subject to Wnt-dependent receptor complex trafficking (Yamamoto et al., 2006; Kim et al., 2008; Ohkawara et al., 2011), this is likely also the case for PTK7 containing receptor complexes. Thus, it is tempting to speculate that the proline residue in position 545—which is conserved in mammals—is required for interaction with Wnt ligands or coreceptors, respectively. Interestingly, this conserved P545 residue is also mutated in one of the six sequence variants identified in patients with neural tube closure defects. In a patient affected with myelomeningocele and interestingly also hydrocephalus,

## REFERENCES


which is indicative of a cilia-defect, the non-polar proline residue was changed to a positively charged arginine (Wang et al., 2015). These data indicate that this conserved residue is important for protein function and mutations are likely pathogenic. Future studies are required to elucidate the molecular pathomechanism.

## CONCLUSIONS

During the last two decades our understanding of the function of PTK7 has significantly advanced. Diverse biological processes that are regulated by PTK7 have been identified and its role in the establishment of polarity and coordinated cell movements has been acknowledged. Recent publications shed light on a literary "complex" function of PTK7 in Wnt signaling. While its role in non-canonical PCP signaling has been confirmed in different animal model systems and biological contexts, its function with respect to canonical Wnt signaling remains controversial. Possibly, these contradictory findings can be explained by the cell type-specific formation and subcellular localization of distinct co-receptor complexes. Further characterization of the formation and dynamics of these ligand-receptor complexes may help us to understand how PTK7 affects development as well as diseaserelated processes.

## AUTHOR CONTRIBUTIONS

AB conceptualized and wrote the main part of the paper. AW wrote the Drosophila and the cancer section and edited the manuscript; HB wrote the section on scoliosis and designed the two tables.


Biol. 20, 781–810. doi: 10.1146/annurev.cellbio.20.010403. 113126


embryonic development. Dev. Cell 2, 695–706. doi: 10.1016/S1534-5807(02) 00197-1


for tubular morphogenesis of the Wolffian duct. Dev. Biol. 412, 219–233. doi: 10.1016/j.ydbio.2016.02.029


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

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

# Dishevelled Paralogs in Vertebrate Development: Redundant or Distinct?

Marc Gentzel <sup>1</sup> \* and Alexandra Schambony <sup>2</sup>

<sup>1</sup> Molecular Analysis—Mass Spectrometry, Center for Molecular and Cellular Bioengineering (CMCB), TU Dresden, Dresden, Germany, <sup>2</sup> Developmental Biology, Biology Department, Friedrich-Alexander University Erlangen-Nuremberg, Erlangen, Germany

Dishevelled (DVL) proteins are highly conserved in the animal kingdom and are important key players in β-Catenin-dependent and -independent Wnt signaling pathways. Vertebrate genomes typically comprise three DVL genes, DVL1, DVL2, and DVL3. Expression patterns and developmental functions of the three vertebrate DVL proteins however, are only partially redundant in any given species. Moreover, expression and function of DVL isoforms have diverged between different vertebrate species. All DVL proteins share basic functionality in Wnt signal transduction. Additional, paralog-specific interactions and functions combined with context-dependent availability of DVL isoforms may play a central role in defining Wnt signaling specificity and add selectivity toward distinct downstream pathways. In this review, we recapitulate briefly cellular functions of DVL paralogs, their role in vertebrate embryonic development and congenital disease.

#### Edited by:

Andrea Erika Münsterberg, University of East Anglia, United Kingdom

#### Reviewed by:

Christine Hartmann, University of Münster, Germany Victoria Sherwood, University of Dundee, United Kingdom

> \*Correspondence: Marc Gentzel marc.gentzel@tu-dresden.de

#### Specialty section:

This article was submitted to Signaling, a section of the journal Frontiers in Cell and Developmental Biology Received: 31 January 2017 Accepted: 12 May 2017 Published: 26 May 2017

#### Citation:

Gentzel M and Schambony A (2017) Dishevelled Paralogs in Vertebrate Development: Redundant or Distinct? Front. Cell Dev. Biol. 5:59. doi: 10.3389/fcell.2017.00059 Keywords: Dishevelled, Wnt signaling, vertebrate embryonic development, embryonic expression, autosomal dominant robinow syndrome

## INTRODUCTION

The Dishevelled (dsh<sup>1</sup> ) phenotype has been described the first time in Drosophila close to 60 years ago (Fahmy and Fahmy, 1959) and the diverse molecular functions of Dishevelled (DVL) proteins still stimulate intensive research. To date DVL is considered the central intracellular effector of Wnt signaling pathways, which play key roles in establishing and patterning of the body axes and in the control of proliferation, differentiation, planar cell polarity (PCP), and cell movements throughout the animal kingdom. Although evidence is accumulating that Wnt pathways should be considered as a signaling network, individual pathways have been subdivided into the Wnt/β-Catenin pathway and the β-Catenin–independent pathways including Wnt/PCP, Wnt/Ca2+, and Wnt/STOP signaling, all of which involve DVL (reviewed in Kühl et al., 2000; Kohn and Moon, 2005; Macdonald et al., 2009; Niehrs and Acebron, 2012; van Amerongen, 2012; Collu and Mlodzik, 2015, **Figure 1A**).

DVL and its core functions in β-Catenin-dependent and -independent Wnt pathways are highly conserved. Notably, the genomes of Drosophila and most other invertebrates harbor only a single DVL gene. By contrast vertebrate genomes comprise genes for three DVL paralogs (DVL1-3), which have most likely arisen from two rounds of genome duplication (reviewed in Kasahara, 2007; Dillman et al., 2013). The still high degree of conservation among the three vertebrate DVL paralogs raises the question to which extent they have undergone functional diversification since obviously a single protein is sufficient to mediate DVL functions in invertebrates.

## CELLULAR FUNCTIONS OF DISHEVELLED PARALOGS

All DVL proteins share an N-terminal DIX domain, a PDZ and a DEP domain (Axelrod et al., 1998; Boutros et al., 1998; Li et al., 1999). Additional sequence motifs have been identified that provide interfaces for protein-protein interactions like the basic and proline-rich regions, or functional motifs as the nuclear export sequence (NES), a putative nuclear localization sequence (NLS), the—YHEL—motif that is required for internalization of the activated receptor complex and motifs that mediate the association of DVL with the cytoskeleton or with phospholipids (Capelluto et al., 2002; Penton et al., 2002; Itoh et al., 2005; Yu et al., 2007, **Figure 1B**). The lack of striking differences in the primary structure with no obvious additional or missing functional motifs and domains poses a challenge to predict functional differences and to develop assays that would reveal potential dissimilarities.

DVL interacts with the cytoplasmic interface of Frizzled receptors (Tauriello et al., 2012), regulates receptor internalization and maintenance of Frizzled membrane equilibrium through protein-protein interaction (Yu et al., 2007; Jiang et al., 2015) and serves as scaffold for numerous proteins including multiple protein kinases (reviewed in Gao and Chen, 2010; Mlodzik, 2016). Depending on the composition of the receptor complex and its interaction partners, DVL contributes to β-Catenin stabilization, inhibition of GSK3β or activation of β-Catenin-independent signaling cascades (Gao and Chen, 2010). DVL also enters the nucleus and interacts with TCF/c-Jun/β-Catenin or FOXK1/2 transcription factor complexes (Itoh et al., 2005; Gan et al., 2008; Wang et al., 2015). In addition, DVL proteins play a role in microtubule stabilization (Ciani et al., 2004), centrosome positioning and separation (Sepich et al., 2011; Cervenka et al., 2016) and mediate cross-talk with other signaling pathways such as Notch or NF-κB (Deng et al., 2010; Collu et al., 2012).

Notably, for most of the abovementioned DVL functions no preference for a DVL paralog has been detected, although some studies suggest dose-dependent effects (e.g., Cervenka et al., 2016). A different picture was observed for the role of DVL paralogs in PCP in ciliated cells, which is required to position the basal bodies. DVL1 was required for intact PCP signaling and localized asymmetrically in multiciliated cells in the epidermis of Xenopus tadpoles. DVL2 was concentrated in dots in vicinity to the basal bodies that led to a local concentration of active RhoA and was required for basal body positioning. Localization of DVL2 itself was mediated by Inturned and according to current knowledge neither of both proteins plays a role in ciliogenesis in the fly (Park et al., 2008). In the mouse node, DVL2 and DVL3 were apically localized and polarized to the posterior side. Positioning of the basal body and directional flow was disturbed or absent in DVL1/DVL2 or DVL1/DVL3 deficient embryos (Hashimoto et al., 2010). Moreover, polarized localization of DVL, planar polarity of basal bodies and their docking could be separated experimentally although the detailed mechanism remains elusive (Park et al., 2008; Vladar et al., 2009; Hashimoto et al., 2010).

## DVL PARALOGS IN VERTEBRATE EMBRYONIC DEVELOPMENT

## Embryonic Expression Patterns

In early Xenopus embryos, DVL2 and DVL3 are present maternally, whereas DVL1 expression is upregulated after the mid-blastula transition (Tadjuidje et al., 2011). At early gastrula stages, all three DVL paralogs are expressed in the prospective mesoderm including Spemann's organizer and, although weaker, in the ectoderm. Post-gastrula expression of DVL1 and DVL2 largely overlaps and is strongest in the neural tube, premigratory and migrating neural crest, as well as in the otic placode, the presomitic and somitic mesoderm. Notably, DVL3 was not expressed in the neuroectoderm but restricted to the paraxial mesoderm, the heart, cranial placodes, and at tadpole stages to the branchial arches (Gray et al., 2009).

In chicken embryos, only two DVL genes, DVL1 and DVL3, were identified. DVL3 was already expressed in day 2 embryos and showed broad expression in most embryonic tissues whereas DVL1 was upregulated only after day 2, i.e., after completion of neurogenesis, and showed a spatially restricted expression in the brain, strongest in the optic stalk, the olfactory bulb, and the ventral forebrain and spinal chord (Gray et al., 2009).

All murine DVL genes are maternally expressed in mouse oocytes and pre-implantation embryos, but interestingly individual protein levels differ considerably and dynamically from oocyte to blastocyst (Na et al., 2007). Post-implantation, DVL1 expression was detected in the mesoderm, but not in the node, at stage E7.5. Post-gastrula, DVL1 was expressed strongest in the neuroectoderm and later in the CNS, cranial and dorsal root ganglia, somites, the liver, kidney, intestine, and lung (Sussman et al., 1994). For DVL2, ubiquitous expression has been reported during embryogenesis in the mouse (Klingensmith et al., 1996). At E 7.5, DVL3 has also been detected ubiquitously, but shortly after showed elevated expression in the CNS and the somitic mesoderm, the notochord, heart, dorsal root ganglia, and branchial arches and in the limb buds (Tsang et al., 1996; Diez-Roux et al., 2011).

Phylogenetic analyses suggest that DVL1 separated first from the common ancestor of DVL2/3, which split into DVL2 and DVL3 in a second round of duplication (Gray et al., 2009). Consistent with corresponding functional divergence, expression of at least one DVL2/3 paralog in the oocyte and pre-blastula embryo and of DLV1 in the central nervous system appear to be conserved among vertebrates. Except for these conserved patterns, developmental expression of the three DVL genes is highly divergent within a species and among different species with DVL2 expression showing the highest variability.

## Developmental Function and Human Congenital Disease

Both, animal models and human congenital disease provide insights into the developmental function of vertebrate DVL paralogs. To date, DVL has predominantly been studied in the mouse and, to a much lesser extent, in Xenopus. Transgenic, single, and compound knock-out mouse models have been discussed comprehensively and in detail in a review by Wynshaw-Boris(2012, see also Supplementary Table 1), therefore we will focus here on human congenital disease and related phenotypes in animal models. In humans, mutations in DVL genes have been associated with neural tube closure defects (NTD) and autosomal-dominant Robinow Syndrome (ADRS) (De Marco et al., 2013; Bunn et al., 2015; White et al., 2015, 2016).

## Neural Tube Defects

During embryonic development, the neural tube is formed by elevation, convergence and fusion of the lateral neural folds to form a hollow tube. Morphogenesis and closure of the neural tube is affected by nutritional, environmental and genetic factors including Wnt/PCP signaling, which is illustrated by genetic association between NTD in humans and mutations in the PCP genes VANGL1, VANGL2, CELSR1, FZD6, and DVL2 (Cai and Shi, 2014; reviewed in De Marco et al., 2014). Notably, also mutations in DVL1 or DVL3 have been identified in humans with NTD (**Figure 2**), although the correlation was not significant (De Marco et al., 2013; Merello et al., 2013; Chen et al., 2016).

Consistently, neural tube closure also requires the same PCP factors in mouse, frog, and zebrafish (Darken et al., 2002; Hamblet et al., 2002; Jessen et al., 2002; Curtin et al., 2003; Formstone and Mason, 2005; Wang et al., 2006). DVL2−/<sup>−</sup> mice show NTD while single and compound DVL1 and DVL3 deficient mice do not, suggesting that among the three DVL genes, DVL2 is necessary and sufficient to mediate neural tube closure. However, DVL2−/<sup>−</sup> DVL3+/<sup>−</sup> and DVL1−/<sup>−</sup> DVL2−/<sup>−</sup> mice display much more severe NTD than DVL2−/<sup>−</sup> mice (Hamblet et al., 2002; Wang et al., 2006; Etheridge et al., 2008), which strongly suggests that DVL1 and DVL3 contribute directly or indirectly to neural tube closure. Along this line, maternal single knock-down of either DVL2 or DVL3 in Xenopus caused NTD (Tadjuidje et al., 2011), supporting a contribution of DVL2 and DVL3.

### Autosomal-Dominant Robinow Syndrome

Robinow syndrome is a rare genetic disorder characterized by mesomelic limb shortening, short stature, cranio-facial malformations, microgenitalia, and occasional cardiac outflow tract defects with either autosomal dominant or recessive inheritance (reviewed in Robinow et al., 1969; Patton and Afzal, 2002). Notably and despite the multifaceted clinical presentation of affected individuals, NTD have neither been described in ADRS nor the more severe recessive Robinow syndrome (RRS). Two recent studies have identified mutations in exon 14 of DVL1 and DVL3 as causative for ADRS (Bunn et al., 2015; White et al., 2015, 2016, **Figure 2**). In addition, ADRS is also associated with WNT5A whereas RRS is caused by loss-of-function mutations in ROR2 (Afzal et al., 2000; van Bokhoven et al., 2000; Person et al., 2010). Thus, the features of Robinow syndrome are generally considered as consequences of defective WNT5A/ROR2-mediated PCP signaling in multiple tissues (Wang et al., 2011) and are partially recapitulated in DVL deficient animal models.

Short stature and defects of the axial skeleton are likely related to impaired convergent extension movements of the paraxial mesoderm and defects in somitogenesis, which have also been reported for DVL2−/−, DVL1−/−;DVL2−/−, and DVL2+/−;DVL3−/<sup>−</sup> mice as well as for Xenopus embryos deficient of any DVL paralog (Hamblet et al., 2002; Etheridge et al., 2008; Gray et al., 2009; Gentzel et al., 2015), indicating that all DVL paralogs contribute to the development of the axial skeleton although DVL2 seems of particular importance.

The characteristic cranio-facial deformations seen in ADRS or RRS indicate defective development of the neural crest (NC), which gives rise to the majority of cranial cartilage and bone. In addition, a subpopulation of the NC contributes to the cardiac outflow tract (OFT). Cranio-facial malformations are also visible in DVL1 and DVL2 morphant Xenopus embryos and the abovementioned mice. The latter and additionally DVL3−/<sup>−</sup>

animals also show cardiac OFT defects. In Xenopus, DVL1 or DVL2 morphant embryos showed normal NC induction but defects in NC migration. The NC is present in DVL3−/<sup>−</sup> mice whereas the cardiac NC markers PITX2 and PLEXINA2 were decreased in mice lacking DVL2 (Hamblet et al., 2002; Etheridge et al., 2008; Gray et al., 2009), indicating differential roles of DVL2 and DVL3 in NC and cardiac development.

Interestingly, DVL1 mutations in humans affect predominantly cranio-facial development with little or no aberrations in body height, whereas in DVL3 and WNT5A associated ADRS craniofacial malformations are accompanied by short stature (Person et al., 2010; Bunn et al., 2015; White et al., 2015, 2016). In mouse, DVL1 is predominantly expressed in the neuroectoderm (Sussman et al., 1994) and as discussed above, knock-out models suggest dominant roles of DVL2 and DVL3 in the development of the axial skeleton and the heart. Although the spatial expression of the three DVL isoforms in human embryos is unknown, the differential prevalence of defects in the axial skeleton in DVL1 and DVL3 associated ADRS supports a prevailing role of DVL2/3 in the paraxial mesoderm in mammals.

### DVL Signaling in Embryonic Development

The recently characterized mutations in DVL1 and DVL3 are frameshift mutations, which alter and shorten the C-termini in the corresponding proteins (White et al., 2015, 2016). This C-terminal domain has been shown to interact with ROR2, a major receptor in β-Catenin independent Wnt signaling and affected in RRS, and is required for ROR2-mediated inhibition of Wnt/β-Catenin signaling (Witte et al., 2010). An initial study suggests a gain of Wnt/β-Catenin activity in ADRS (Bunn et al., 2015), thus it is conceivable that ADRS mutations in DVL1 and DVL3 might result in defective β-Catenin independent Wnt signaling and concomitantly upregulate Wnt/β-Catenin activity.

Malformations of the axial skeleton seen in DVL2−/<sup>−</sup> and DVL1−/−;DVL2−/<sup>−</sup> mice are reminiscent of the phenotypes in ROR2 or WNT5A deficient embryos that can be attributed to aberrant PCP signaling in the paraxial mesoderm (reviewed in Stricker et al., 2017). Genetic interactions between DVL2 and DVL3 with VANGL2 in the mouse further suggest that DVL2 acts in Wnt/PCP signaling in neural tube closure and in cochlea development (Wang et al., 2006; Etheridge et al., 2008). In addition, NTDs and OFT defects in DVL1−/−;DVL2−/<sup>−</sup> mice were similar to defects in VANGL2 mutants and rescued by a DVL21DIX transgene, which is defective in β-Catenin signaling but retains activity in PCP signaling (Wang et al., 2006; Sinha et al., 2012). Notably, also DVL3 KO mice develop OFT defects, but no genetic interaction with VANGL2 in OFT morphogenesis could be demonstrated, indicating a non-redundant function of DVL3 (Etheridge et al., 2008).

Wnt/β-Catenin signaling also contributes to the development of the paraxial mesoderm, heart and neural crest, and patterns the neural tube. However, defective Wnt/β-Catenin signaling results in patterning defects of the dorsal mesoderm and affects proliferation, expansion, or specification of dorsal neural tube progenitors and neural crest (NC) cells (Greco et al., 1996; Ikeya et al., 1997; Pinson et al., 2000; Lou et al., 2008; Seldin et al., 2008; Valenta et al., 2011). By contrast, in either single or compound DVL knock-out mice, Wnt/β-Catenin signaling, and early dorsal mesoderm markers were unaffected. Only in triple knock-out mice a marked downregulation of Wnt/β-catenin signaling has been observed (Etheridge et al., 2008; Hashimoto et al., 2010). Still, defective β-Catenin signaling in smaller cell populations cannot be excluded. One such example might be the cardiac neural crest in DVL2−/<sup>−</sup> embryos, in which the β-Catenin target PITX2 is downregulated (Hamblet et al., 2002; Kioussi et al., 2002). PITX2 is required for proliferation of cardiac NC, but also for the interaction between cardiac NC and second heart field cells (Kioussi et al., 2002; Ma et al., 2013) and, indirectly, for OFT morphogenesis via its target gene WNT11 (Zhou et al., 2007). OFT defects in DLV1−/−, DVL2−/<sup>−</sup> embryos were rescued by a DVL21DIX transgene (Sinha et al., 2012), however this does not exclude a role of DVL2 in β-Catenin signaling upstream of PITX2 since the transgene could also restore OFT morphogenesis downstream of WNT11.

Overall it appears that DVL function in β-Cateninindependent Wnt signaling is more sensitive to the loss or dysfunction of one or two paralogs and accounts for most of the developmental phenotypes in knock-out animal models and also for the features of ADRS.

### Induced Heart Defects

A number of studies indicate a specific role of DVL1 in cardiac remodeling and regeneration. DVL1 and CamKII are upregulated after induced myocardial infarction and heart failure indicating a role of Wnt/Ca2<sup>+</sup> signaling in regeneration (Chen et al., 2004; Ai et al., 2005; Bossuyt et al., 2008). Persistent pressure overload induced cardiac hypertrophy was attenuated in DVL1 knock-out mice, which has been attributed to lower Wnt/β-Catenin activity as well as decreased AKT activation (van de Schans et al., 2007). Consistently, DVL1 gain-offunction induced progressive cardiomyopathy (Malekar et al., 2010). Interestingly, evidence is accumulating that DVL1 is functionally associated with Wnt/Ca2<sup>+</sup> and CamKII signaling in cardiomyopathy (Malekar et al., 2010; Zhang et al., 2015), in excitory synapses in the rat spinal chord (Ciani et al., 2011) and in convergent extension movements in Xenopus gastrulation (Gentzel et al., 2015), indicating a functional specification of DVL1.

## PERSPECTIVES

Striking differences between different DVLs and species have been observed in temporal and spatial expression patterns. Lossof-function phenotypes of each single paralog in mouse as well as ADRS features associated with DVL1 or DVL3 mutations also differ, indicating some degree of divergence but also overlapping functions. In addition, expression of transgenes in a single knock-out background further supported partial redundancy and indicated a dose dependency. If the DVL paralogs would be fully redundant in function it might be speculated that the summed abundance of all paralogs is important for cell survival. But even in cell culture models, any single knock-down is effective and the cells apparently do not sense overall "DVL concentration" and do not compensate the down-regulation of one protein by upregulation of the others (Cervenka et al., 2016).

Functional redundancies however, do not connote biochemical identity. The observed differences could reflect differential expression levels, epistatic relations, or differential biochemical properties such as protein-protein interaction affinity and consequently also molecular function, which would also be consistent with dose-dependencies. The developmental phenotypes further indicate that β-Catenin independent Wnt pathways are more sensitive to the dose of individual DVL paralogs than β-Catenin signaling. This hypothesis is further supported by comparison of triple knock-out and triple-RNAi knock-down embryos. Whereas, in triple KO embryos early β-Catenin signaling is strongly reduced resulting in axis and mesodermal mispatterning, in the knock-down, in which ∼25–30% of each paralog were retained, early β-Catenin signaling was not affected, but the embryos showed strong morphogenetic defects in the dorsal mesoderm and neuroectoderm. Consistently, specific and different molecular functionality of the three DVL paralogs has been observed in ciliogenesis and the Wnt/Ca2<sup>+</sup> pathway, which were revealed in intact tissue or tissue models (Park et al., 2008; Gentzel et al., 2015).

Overall, the currently available data indicate that DVL expression and function have diverged to some degree apparently and consistent with phylogenetic models mostly between DVL1 and DVL2/3. Thus, depending on the cellular context, DVL paralogs exhibit both redundant and distinct functionality.

## AUTHOR CONTRIBUTIONS

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

## ACKNOWLEDGMENTS

The authors' original work was funded by German Research foundation (DFG) grants to AS (grant numbers SCHA965/6-1 and 6-2), MG was funded by the Max Planck Society (MPG).

## REFERENCES


## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fcell. 2017.00059/full#supplementary-material


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

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

## Autophagy in the Vertebrate Inner Ear

Marta Magariños 1, 2, 3, Sara Pulido1, 2, María R. Aburto1 †, Rocío de Iriarte Rodríguez <sup>1</sup> and Isabel Varela-Nieto1, 2, 4 \*

<sup>1</sup> Department of Endocrine and Nervous Systems Pathophysiology, Instituto de Investigaciones Biomédicas "Alberto Sols," CSIC-UAM, Madrid, Spain, <sup>2</sup> CIBERER, Unit 761, Instituto de Salud Carlos III, Madrid, Spain, <sup>3</sup> Departamento de Biología, Universidad Autónoma de Madrid, Madrid, Spain, <sup>4</sup> Instituto de Investigación Hospital Universitario La Paz (IdiPAZ), Madrid, Spain

Autophagy is a conserved catabolic process that results in the lysosomal degradation of cell components. During development, autophagy is associated with tissue and organ remodeling, and under physiological conditions it is tightly regulated as it plays a housekeeping role in removing misfolded proteins and damaged organelles. The vertebrate inner ear is a complex sensory organ responsible for the perception of sound and for balance. Cell survival, death and proliferation, as well as cell fate specification and differentiation, are processes that are strictly coordinated during the development of the inner ear in order to generate the more than a dozen specialized cell types that constitute this structure. Here, we review the existing evidence that implicates autophagy in the generation of the vertebrate inner ear. At early stages of chicken otic development, inhibiting autophagy impairs neurogenesis and causes aberrant otocyst morphogenesis. Autophagy provides energy for the clearing of dying cells and it favors neuronal differentiation. Moreover, autophagy is required for proper vestibular development in the mouse inner ear. The autophagy-related genes Becn1, Atg4g, Atg5, and Atg9, are expressed in the inner ear from late developmental stages to adulthood, and Atg4b mutants show impaired vestibular behavior associated to defects in otoconial biogenesis that are also common to Atg5 mutants. Autophagic flux appears to be age-regulated, augmenting from perinatal stages to young adulthood in mice. This up-regulation is concomitant with the functional maturation of the hearing receptor. Hence, autophagy can be considered an intracellular pathway fundamental for in vertebrate inner ear development and maturation.

Keywords: Atg4, Atg5, Beclin-1, cochlea, LC3, otic development, vestibular system

## AN INTRODUCTION TO AUTOPHAGY

Autophagy is a catabolic process that degrades the cytoplasmic content of a cell in lysosomes, returning energy, and molecular building bricks to the cell. Indeed, autophagy has a housekeeping role in cells as it is a way to eliminate damaged macromolecules, organelles, and pathogens. Since the initial description of autophagy by Christian de Duve in 1963, it has become more and more relevant as it has become implicated in a variety of physiological and pathological situations (Jiang and Mizushima, 2014). Indeed, three different types of autophagy are now recognized: (1) Macroautophagy (herein autophagy), where a double-membrane autophagosome forms and engulfs cytoplasmic content, subsequently fusing with the lysosome to form an autolysosome and releasing the autophagosome cargo into the lysosome lumen to be degraded by hydrolases; (2) Microautophagy, in which the cargo reaches the lumen by invagination of the lysosomal membrane;

### Edited by:

Andrea Erika Münsterberg, University of East Anglia, United Kingdom

#### Reviewed by:

Markus A. N. Hartl, University of Innsbruck, Austria Mai Har Sham, University of Hong Kong, Hong Kong

> \*Correspondence: Isabel Varela-Nieto ivarela@iib.uam.es

#### Present Address:

María R. Aburto, Institute of Cell Biology and Neuroscience, and BMLS, Goethe University Frankfurt, Frankfurt am Main, Germany.

†

#### Specialty section:

This article was submitted to Signaling, a section of the journal Frontiers in Cell and Developmental Biology Received: 23 January 2017 Accepted: 09 May 2017 Published: 26 May 2017

#### Citation:

Magariños M, Pulido S, Aburto MR, de Iriarte Rodríguez R and Varela-Nieto I (2017) Autophagy in the Vertebrate Inner Ear. Front. Cell Dev. Biol. 5:56. doi: 10.3389/fcell.2017.00056

and (3) Chaperone-mediated autophagy, exclusive to mammals, where proteins associated to chaperones bind to the LAMP2A lysosomal receptor and are delivered directly to the lumen (Tasset and Cuervo, 2016).

The formation of the autophagosome requires the activity of a set of proteins, most of them encoded by the autophagy related genes (ATG; **Figure 1A**). The formation of the autophagosome involves induction, nucleation, elongation, and completion. A specific subset of ATG proteins has been associated to each of these stages (Ariosa and Klionsky, 2016). As such, the ULK1/2 complex (ATG13, ATG101, FIP200) participates in induction and ULK1 activates the phosphatidylinositol 3-kinase complex (PI3KC: Beclin-1, Vsp34, Vps15, ATG14) to promote nucleation. Two ubiquitin-like conjugation systems contribute to elongate the phagophore: ATG12 (ATG12, ATG7, ATG10, ATG5, and ATG16L) and ATG8 (LC3, the mammalian homolog of ATG8). Both these complexes regulate the formation of LC3-II, the relative levels of which serve as a readout of the autophagic flux, along with SQSTM1/p62 that facilitates the entry of the cargo into the autophagosome. Accordingly, the SQSTM1/p62 levels are inversely correlated with those of LC3-II (Katsuragi et al., 2015; Klionsky et al., 2016). Finally, the ATG9 cycling system incorporates membranes from cell donor locations (Pavel and Rubinsztein, 2016). Following the completion of the autophagosome, its fusion with lysosomes requires the activity of proteins involved in other vesicular fusion events, such as the SNARE (soluble NSF attachment protein receptor) and HOPS (homotypic fusion and vacuole sorting proteins) complexes (Zhen and Li, 2015).

Autophagy can be induced by starvation, growth factor deprivation, hypoxia, or infections. These stimuli elicit an immediate response and long-term gene expression responses mediated by specific transcription factors like TFEB (transcription factor EB). TFEB acts as a master regulator of autophagy by up-regulating the expression of autophagy genes. Under nutrient-rich conditions, TFEB is phosphorylated by mTORC1 (mammalian target of rapamycin complex (1) and kept inactivate in the cytosol, mTORC1 also inhibiting autophagy by phosphorylation of ATG13, Füllgrabe et al., 2016; Napolitano and Ballabio, 2016).

Autophagy is a housekeeping mechanism that removes damaged molecules and organelles from the cell's cytoplasm, yet it also participates in the immune response, and it provides energy and molecules as building blocks when needed. Autophagy is essential during development, as it contributes to organ and tissue sculpting in Drosophila by facilitating cell death (Denton et al., 2012). Indeed, autophagy may promote largescale cytosolic self-digestion and the removal of certain pro-survival proteins (Yu et al., 2006). Thus, the final output of autophagy could be either positive or negative for the cell, and this depends on the intensity and duration of its induction.

## DEVELOPMENTAL AUTOPHAGY

Autophagy contributes to developmental tissue remodeling, responding to specific extrinsic, and intrinsic stimuli. For example, following fertilization of the mouse egg, autophagy removes maternal mRNA and proteins, allowing the egg to initiate its zygotic program (Tsukamoto et al., 2008; Yamamoto et al., 2014). Later on in development, autophagy drives the development of the nervous system, adipose tissue, osseous tissue, hematopoietic system, and the heart (Aburto et al., 2012a). The study of genetically modified mice has shed light on the roles played by the genes involved in autophagy. Ambra1 is an essential gene for the development of the mouse central nervous system, the deficiency of which impairs autophagy and induces aberrant neuronal proliferation (Fimia et al., 2007; Antonioli et al., 2015). Different mutations in genes that participate in the autophagy machinery have shown that autophagy is needed for terminal neuronal differentiation, and specifically for axonal outgrowth and guidance. For example, axon formation is disturbed in the cerebellar granule neurons of Ulk1−/<sup>−</sup> mice (Zhou et al., 2007) and more recently, ALFY, an adaptor protein between the cargo and the ATG proteins, was seen to be required for axon outgrowth in the brain and to establish neuronal connectivity (Dragich et al., 2016).

Atg mutants have provided evidence that autophagy is needed for the correct development of adipose, osseous and cardiac tissues, as well as for the differentiation of hematopoietic cells. Atg5 and Atg7 deficiency is associated with a reduction in thymocytes and B-lymphocytes (Pua et al., 2007), as well as reduced levels of adipocyte differentiation factors and decreased lipid adipose mass (Singh et al., 2009). In addition, Atg5 and Atg7 deficiency in the embryonic P19CL6 cells inhibited cardiac cell differentiation (Jia et al., 2014). However, in the analysis of the phenotypes associated to these mutations it should be considered that ATG proteins also fulfill functions that are not related to autophagy (Mauthe and Reggiori, 2016).

During development, autophagy facilitates rapid changes in intracellular composition, promoting the turnover of specific proteins, receptors, cytoskeletal components, or transcription factors necessary to define the different cell fates. It is also essential for the temporal dynamics of cell organelles, controlling their number, and quality (e.g., mitochondria). Finally, after birth and before the initiation of suckling behavior, the up-regulation of autophagy protects newborns from death by starvation (Kuma et al., 2004). Autophagy may not only supply energy at this stage but it may also help control oxidative stress (Schiaffino et al., 2008).

## AN INTRODUCTION TO INNER EAR ANATOMY

The mammalian inner ear is a complex sensory structure within the temporal bone that is composed of the cochlea and the vestibule, structures that are responsible for the senses of hearing and balance, respectively (**Figure 1B**). The auditory and vestibular organs contain the mechanosensory receptors that transduce mechanical stimuli into electrochemical signals that are transmitted to the brain by the fibers of the VIIIth cranial nerve. The auditory receptor is the organ of Corti in the scala media of the cochlea (Magariños et al., 2012, 2014), which is

FIGURE 1 | (A) Schematic view of the molecular steps of macroautophagy. Growth factors and nutrient-rich conditions activate mTORC1, a negative regulator of the ULK1/2 complex and TEFB. In turn, growth factor deprivation, inflammation, or nutrient starvation, activate the ULK1/2 complex, which phosphorylates and activates the PI3K complex III (PI3KC). The ATG9 cycling system provides membranes to form the autophagosome from different donor sources. Autophagosome formation also requires the action of two ubiquitin-like (Ubl) systems, ATG8-Ubl and ATG12-Ubl, required for the elongation and completion of the autophagosome. LC3 is converted into the cytosolic form, LC3-I, by cleavage of ATG4B, and into the membrane associated form, LC3-II, by conjugation with phosphoethanolamine via ATG5 (and the remaining components of the ATG12-Ubl system). SQSTM1/p62 (p62) binds to ubiquitinated proteins and carries them to the autophagosome (adapted from de Iriarte Rodríguez et al., 2015). (B) Anatomy of the adult mouse inner ear. (a) Lateral view showing a mammalian inner ear. (b,c) Detail of the vestibular macula (b) and cristae ampullaris (c), where sensory hair cells are labeled for myosin VIIa (green) and neurofilament (red). (d) Detail of the organ of Corti showing myosin VIIa positive hair cells (green) and SOX2 positive supporting cells (red). (e) The stria vascularis is visualized by labeling for Kir4.1 (green). Development of the mouse inner ear. The inner ear develops from the otic placode (f, E7.5). The otic placode invaginates to form the otic cup (g, E8-9), which later pinches off to form the otic vesicle or otocyst (h,i). Neural precursors delaminate from the ventral otocyst epithelium to form the acoustic-vestibular ganglion (AVG: g–i). The cochlear duct evaginates from the ventromedial region of the otic vesicle, and it will be innervated by the acoustic portion of the AVG, also known as the spiral ganglion (SG: yellow, j–m). The cochlear duct elongates and grows to form a coiled tube, the membranous labyrinth, which includes the primordium of the scalas media, vestibularis, and tympanic (j–m). At the cochlear duct the prosensory patch will become the primitive organ of Corti. Scale bars: (a) 0.5 mm; (b–e) 50µm. Co, cochlea; V, vestibule; Asc, Lsc and Psc, anterior, lateral and posterior semicircular canals; Do, dorsal; Cd, caudal; IHC, inner hair cells; OHC, outer hair cells; StV, stria vasculari; SpL, spiral ligament; SV, scala vestibule; SM, scala media; ST, scala tympani; LW, lateral wall; OC, Organ of Corti (adapted from Magariños et al., 2014).

formed by sensory hair cells and by non-sensory support cells (Deiters', Hensen's and Claudius') that maintain the ionic and metabolic cochlear homeostasis (Forge and Wright, 2002). There are two functional types of hair cells arranged in a stereotypic manner: one row of inner hair cells (IHC) and three rows of outer hair cells (OHC; Forge and Wright, 2002; Magariños et al., 2012). The IHC cells connect to bipolar auditory type I neurons of the spiral ganglion, whilst the OHC are innervated by type II neurons (Nayagam et al., 2011; Fritzsch et al., 2015). The axons of these neurons leave the spiral ganglion to form the cochlear division of the acoustic-vestibular nerve, which is responsible for transmitting the auditory information through a multisynaptic, ascendant pathway from the cochlea to the auditory cortex (Demanez and Demanez, 2003). HC stereocilia are bathed by endolymph, which maintains the unique ionic concentration required for mechanotransduction. The stria vascularis is located in the lateral wall of the scala media. This three-layered structure regulates cochlear ion transport and maintains the endocochlear potential (Patuzzi, 2011).

The vestibular system is formed by five sensory structures, three cristae located at the base of the semicircular canals and two maculae. Each of these structures has a similar organization, with sensory HC and non-sensory support cells innervated by the vestibular ganglion axons. The vestibule is responsible for sensing equilibrium, and for the perception of linear and angular acceleration, and of gravity (Highstein and Fay, 2004; Ekdale, 2016).

## THE REGULATION OF INNER EAR DEVELOPMENT BY EXTRACELLULAR FACTORS AND INTRACELLULAR SIGNALING NETWORKS

The development of the inner ear is initiated by the induction of the otic placode from the ectoderm lying between the rhombomeres 5 and 6 (Magariños et al., 2014; Whitfield, 2015). Otic placode induction is orchestrated from mesoderm signals that coordinate with intrinsic factors in the ectoderm. FGFs, Notch and WNT signaling play a key role during these initial events (Ohyama et al., 2006, 2007; Jayasena et al., 2008). The otic placode then invaginates to form the otic cup that will later detach and close to form the otocyst or otic vesicle. The otocyst is transient embryonic round structure whose multipotent cells will differentiate to produce most adult inner ear cell types (Bissonnette and Fekete, 1996; Sanchez-Calderon et al., 2007). The ventral region of the otocyst is specified by the Sonic hedgehog (Shh) secreted from the floor plate and notochord (Riccomagno et al., 2002, 2005), as well as through repression by the WNT signaling pathway (Groves and Fekete, 2012). Significantly, it is this region that will form the auditory portion of the inner ear. The vestibule develops from the dorsal otocyst, instructed by signals from the bone morphogenetic protein BMP4 (Chang et al., 2008) that antagonize Shh. Sensory HC, nonsensory support cells, plus the acoustic and vestibular neurons that contribute to the acoustic-vestibular ganglion (AVG) also arise from the otocyst. Finally, Notch signaling helps specify the prosensory domain (Daudet and Lewis, 2005; Hartman et al., 2010) and in combination with Atoh1 expression, it is involved in determining the HC and supporting cells (Mizutari et al., 2013).

Otic vesicle development requires the coordinated response to apoptosis, survival and proliferation signals. IGF-1 signaling, mainly through the RAF-MEK-ERK and PI3K/AKT pathways, fulfills a critical role in regulating these processes. In the chicken embryo, PI3K/AKT signaling regulates the number of otic neurons and it determines the timing of their generation (Aburto et al., 2012b). Moreover, both the RAF-MEK-ERK and PI3K/AKT pathways modulate AVG neuritogenesis (Magariños et al., 2010; Aburto et al., 2012b). Phosphatase and tensin homolog deleted on chromosome 10 (PTEN) is required to define the size of the neuroblast population (Kim et al., 2013) and it negatively regulates the AKT signaling pathway, as well as interacting with the WNT, Notch, and BMP pathways.

Cell cycle regulation is also essential for correct inner ear organogenesis. IGF-1, Notch, and WNT are among the signaling pathways involved in regulating the proliferation of otic progenitors (Magariños et al., 2014). Through the RAF-MEK-ERK pathway, IGF-1 promotes the cell cycle progression of otic progenitors (Sanz et al., 1999b; Magariños et al., 2010). Finally, the otocyst must undergo the morphogenetic changes that transform the simple pseudostratified otic vesicle epithelia into an extremely complex three-dimensional membranous labyrinth (Kelly and Chen, 2009). The neighboring mesenchymal cells will be responsible for generating the bony labyrinth (Chang et al., 2002).

## DEVELOPMENTAL AUTOPHAGY IN THE EMBRYONIC CHICKEN INNER EAR

Beclin-1 and Atg5 transcripts are expressed throughout the developmental stages in the chick when otic vesicles can be explanted and studied in organotypic cultures. Indeed, the Beclin-1 and LC3B proteins are present in the otic epithelium and the AVG (Aburto et al., 2012c; summarized in **Figure 2A**). Chemical and genetic inhibitors of autophagy demonstrate the importance of the autophagic flux for the development and cellular dynamics of the otocyst (Aburto et al., 2012c; **Figure 2A**). Inhibiting autophagy shows that it is required for the clearance of apoptotic cells and for cell cycle progression. Developmental apoptosis is an essential process during inner ear development (Fekete et al., 1997; Sanz et al., 1999a; Frago et al., 2003; León et al., 2004; Magariños et al., 2012), and both this cell death and the elimination of apoptotic cells require energy (Qu et al., 2007; Mellén et al., 2008). During development, autophagy provides ATP by degrading intracellular components and it thereby facilitates apoptosis. Impaired autophagy causes an accumulation of apoptotic cells that cannot be eliminated from the otic vesicle, a failing that can be reverted by adding ATP. The region where otic neural progenitors originate is the neurogenic zone, where the extracellular matrix is degraded to detach cells and the migrating detached cells accumulate

acoustic-vestibular ganglion (AVG) develops from the cultured otic vesicle after 20 h in serum-free culture medium (0S). (b) Autophagic flux is typically measured in Western blots to determine the LC3 ratio in the presence or absence of chemical inhibitors of autophagy (3-MA and CQ). (c) Otic vesicles incubated with an inhibitor of autophagy accumulate apoptotic cells, as evident by reduced staining for An-V in red and by increased TUNEL green spots (d–g). Aberrant AVG development is also seen (h,i), with fewer neuroblasts (TuJ-1, red), and (j,k) altered neurite outgrowth and pathfinding (G4, green). (f,g) Higher magnification of the boxed regions in (d) and (e), respectively. annexin-V, An-V; 3-methyladenin, 3-MA; chloroquine, CQ. Scale bars: (d–i), 150µm; (f), (g), 50µm; (j), (k) 300µm (adapted from Aburto et al., 2012c). (B) Autophagy in the postnatal and adult mouse inner ear. (a,c) Histograms showing the changes in Beclin-1, Atg4b, and Atg5 expression with age in the mouse vestibule (a) and cochlea (c), as determined by RT-qPCR. (b,d) Autophagic flux increases with age in the mouse inner ear. The LC3-II/LC3-I and SQSTM1/p62 (p62/β-actin) ratios were determined in Western blots of the vestibule (b) and cochlea (d) at E18.5 and P270. Significance: \*P < 0.05, \*\*P < 0.01, and \*\*\*P < 0.001 vs. E18.5; #P < 0.05, ##P < 0.01, and ###P < 0.001 vs. P0; and & P < 0.05 and && P < 0.01 vs. P30–60. E, embryonic day and P, postnatal day (adapted from de Iriarte Rodríguez et al., 2015).

autophagic vacuoles. Conversely, the inhibition of autophagy results in aberrant AVG phenotypes (Aburto et al., 2012c; **Figure 2A**). Therefore, autophagy is required for the migration of the epithelial neuroblasts from the neurogenic zone to form the AVG. In summary, the early development of the inner ear is one example of many where developmental autophagy plays a supporting role to apoptosis and migration (Di Bartolomeo et al., 2010; Wada et al., 2014; Lorda-Diez et al., 2015; Boya et al., 2016).

## AUTOPHAGY IN THE MOUSE INNER EAR

Autophagy is required for the development of the vestibular system in the mouse. Atg4 deficient mice have impaired balance, with different phenotypic penetrance from severe to mild vestibular alterations. The biogenesis of the otoconia is defective in both Atg4b−/<sup>−</sup> and Atg5−/<sup>−</sup> mice (Mariño et al., 2010), and otoconial impairment occurs in mice exposed to streptomycin ototoxicity (Takumida et al., 1997), which can inhibit autophagy (Levano and Bodmer, 2015) and increase cell damage in the inner ear due to oxidative stress (Guthrie, 2008). In fact, autophagy reduces the reactive oxygen species (ROS) in mice subjected to noise-induced hearing loss (Yuan et al., 2015). This crucial role of autophagy in eliminating ROS could explain the similarities between autophagy gene mutants and streptomycin-treated animals. However, increased ROS are not the only consequence of inhibiting autophagy during vestibular development, as otoconial biogenesis requires the secretion and assembly of specific proteins that are also affected by inhibiting autophagy (Mariño et al., 2010).

Autophagy plays a key role in newborn mice, and Atg5, Atg7, Atg9, and Atg16 null mice die soon after birth (Mizushima and Levine, 2010). The transcriptome of the E18.5 mouse cochlea shows that a wide variety of Atg genes are expressed, underlining the relevance of autophagy at perinatal stages (de Iriarte Rodríguez et al., 2015). Furthermore, several key genes of the autophagic molecular machinery (Becn1, Atg4b, Atg5, and Atg9) are expressed in the mouse vestibule and cochlea throughout development and adulthood (de Iriarte Rodríguez et al., 2015; summarized in **Figure 2B**). The expression of these genes is significantly enhanced from the perinatal stages (E18.5 and P0) to adulthood (P30) as the inner ear acquires its complete functionality (Rueda et al., 1996). A temporal analysis of autophagic proteolysis in the cochlea and vestibule confirms the induction of autophagy in adults rather than E18.5 embryos. Moreover, there is significantly less SQSTM1/p62 at P270 than at E.18.5, whilst the relative LC3-II levels increase in the cochlea and vestibule (de Iriarte Rodríguez et al., 2015; **Figure 2B**). Indeed, autophagosomes are clearly visible in adult neurons of the spiral ganglion but not at earlier stages. LC3B forms granular structures in the neuronal soma at P30 and onwards, yet not at E18.5 (de Iriarte Rodríguez et al., 2015). Autophagy is essential in neurons because they do not dilute their damaged molecules or organelles by proliferation. Thus, autophagy is required for detoxification and to manage damage (Son et al., 2012; Damme et al., 2015; He et al., 2016). Accordingly, the postnatal onset of hearing and the concomitant increase in neuronal activity is correlated with the induction of autophagy in the cochlea.

## THE INFLUENCE OF AUTOPHAGY ON INNER EAR HOMEOSTASIS AND AGING

Autophagy plays an additional role in inner ear homeostasis once development is concluded. Otic injury caused by a combination of aminoglycoside and loop diuretics augments aspects of autophagy (Taylor et al., 2008). Moreover, autophagy is activated by rapamycin alleviated ototoxicity in cisplatintreated rats (Fang and Xiao, 2014) and in mice exposed to an auditory insult (Yuan et al., 2015). Thus, autophagy helps maintain adult hearing in response to stress. Proteostasis is impaired during aging (López-Otín et al., 2013) and the stabilization of proteic events that is mostly provided by molecular chaperones also declines with age (Rodriguez et al., 2016). In addition, protein degradation systems control the levels of misfolded or aggregated proteins, the accumulation of which drives age-related neurodegenerative diseases like Parkinson's or Alzheimer's disease (Balchin et al., 2016). Thus, it is not surprising that the senescence-accelerated mouse prone 8 (SAMP8) mutant mice exhibit age-related hearing loss and autophagy stress (Menardo et al., 2012).

Our studies of 9 month-old Igf1−/<sup>−</sup> mice show they suffer defects in the proteostasis associated with aging. These Igf1−/<sup>−</sup> mice suffer a loss of hearing and a reduced lifespan, among other traits (Varela-Nieto et al., 2013). Hearing loss in Igf1−/<sup>−</sup> deficient mice is accompanied by a general failure of the hearing receptor (Riquelme et al., 2010), although the weaker autophagy gene expression in one-year-old cochlea may also contribute to this auditory phenotype (de Iriarte Rodríguez et al., 2015). However, the vestibular defects in the Igf1−/<sup>−</sup> mouse are milder than those found in the cochlea (Rodríguez-de la Rosa et al., 2015). Becn1, Atg4b, and Atg5 are more strongly expressed in 9-month-old Igf1−/<sup>−</sup> vestibules compared to those of wildtype mice. Thus, the induction of autophagy might provide Igf1−/<sup>−</sup> vestibules with some protection, as it does in Igf1−/<sup>−</sup> retinas (Arroba et al., 2016). After differentiation, hair cells do not regenerate in the mammalian cochlea, whilst vestibular hair cells do to a limited extent (Burns and Stone, 2016). The up-regulation of autophagy might be partially responsible for the different potentiality of vestibular and cochlear hair cells.

## CONCLUSIONS

During the development of the vertebrate inner ear, autophagy participates in cell remodeling and dynamics, and it contributes to the biogenesis of the vestibular otoconia. In the postnatal cochlea, the autophagy machinery is upregulated concomitant with the increase in neuronal activity at the onset of hearing. Autophagy becomes a housekeeping process in the adult inner ear, and it is a means to protect hearing during aging and in response to injury. Further work is needed to fully understand the role of autophagy in the inner ear and to explore the potential of modulating autophagy as a novel strategy to combat inner ear diseases.

## AUTHOR CONTRIBUTIONS

MM and IV designed and drafted the work and wrote the manuscript; MM, IV, SP, MA, and RdIR analyzed and interpreted the data, revised the manuscript critically and approved the version to be published; SP designed and performed the figures. All authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity

## REFERENCES


of any part of the work are appropriately investigated and resolved.

## ACKNOWLEDGMENTS

We warmly thank our present and former colleagues of the Neurobiology of Hearing group for sharing information and for useful discussions. SP is supported by a FPI predoctoral fellowship (BES-2015-071311) funded by MINECO and the European Social Fund. This work was supported by the European Commission FP7-PEOPLE-2013-IAPP TARGEAR and Spanish MINECO/FEDER SAF2014-53979-R to IV.


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

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

# More than Just a Simple Cardiac Envelope; Cellular Contributions of the Epicardium

### Angel Dueñas, Amelia E. Aranega and Diego Franco\*

Cardiac and Skeletal Muscle Research Group, Department of Experimental Biology, University of Jaén, Jaén, Spain

The adult pumping heart is formed by distinct tissue layers. From inside to outside, the heart is composed by an internal endothelial layer, dubbed the endocardium, a thick myocardial component which supports the pumping capacity of the heart and exteriorly covered by a thin mesothelial layer named the epicardium. Cardiac insults such as coronary artery obstruction lead to ischemia and thus to an irreversible damage of the myocardial layer, provoking in many cases heart failure and death. Thus, searching for new pathways to regenerate the myocardium is an urgent biomedical need. Interestingly, the capacity of heart regeneration is present in other species, ranging from fishes to neonatal mammals. In this context, several lines of evidences demonstrated a key regulatory role for the epicardial layer. In this manuscript, we provide a state-of-the-art review on the developmental process leading to the formation of the epicardium, the distinct pathways controlling epicardial precursor cell specification and determination and current evidences on the regenerative potential of the epicardium to heal the injured heart.

#### Edited by:

Juan Jose Sanz-Ezquerro, Consejo Superior de Investigaciones Científicas, Spain

#### Reviewed by:

Nicola Smart, University of Oxford, UK Manvendra K. Singh, DUKE-NUS Medical School, Singapore Nadia Mercader, University of Bern, Switzerland

> \*Correspondence: Diego Franco

dfranco@ujaen.es

#### Specialty section:

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

> Received: 31 January 2017 Accepted: 10 April 2017 Published: 01 May 2017

#### Citation:

Dueñas A, Aranega AE and Franco D (2017) More than Just a Simple Cardiac Envelope; Cellular Contributions of the Epicardium. Front. Cell Dev. Biol. 5:44. doi: 10.3389/fcell.2017.00044 Keywords: epicardium, proepicardium, non-coding RNAs, heart development, regeneration

The development of the heart is a complex process. The primitive heart tube is formed from cardiogenic mesoderm of the cardiac crescents, i.e., first heart field (FHF), while anterior and venous poles are derived from a subsequent subset of cardiogenic cells located medial to the cardiac crescents, dubbed second heart field (SHF; Kelly et al., 2001; Kelly and Buckingham, 2002). In addition, external cellular contributions to the developing heart will take place from this stage onwards. On the one hand, cardiac neural crest will colonize the most anterior parts of the heart playing a pivotal role on aortico-pulmonary septation (Kirby and Waldo, 1990, 1995). On the other hand, cell originating from the proepicardium (PE) will cover and infiltrate into the developing heart leading to distinct cellular subpopulations, such as endothelial and smooth muscle cells forming the coronary vasculature, endocardial cushion mesenchyme, cardiac fibroblasts, and of

**Abbreviations:** Bmp, Bone morphogenetic protein; CXCR4, Chemokine (C-X-C motif) receptor 4; Cre, Cre recombinase; CXCL12, Chemokine (C-X-C motif) ligand 12; Dkk1, Dickkopf-related protein 1; Dkk2, Dickkopf-related protein 2; EMT, Epitelial-to-mesenchymal transition; EPDCs, Epicardial derived cells; Fabp4, Fatty acid binding protein 4; Fgf, Fibroblast growth factor; Fgf10, Fibroblast growth factor 10; Fgfr2b, Fibroblast growth factor receptor 2b; FHF, First heart field; lcnRNA, Long non-coding RNA; MAPK, Mitogen-Activated Protein Kinase; Nfatc1, Nuclear factor of activated T-cells, cytoplasmic 1; NF-κB, Nuclear Factor κB; Nrg1, Neuregulin 1; Pcd4, Programmed Cell Death 4; PCP, Planar cell polarity; PE, Proepicardium; Pod1/Tcf21, Podocyte-expressed 1/Transcription factor 21; Raldh2, Aldehyde dehydrogenase family 1, subfamily A2; SHF, Second heart field; Tbx18, T-box homeobox 18; Tcf21/Pod1, Transcription factor 21/Podocyte-expressed 1; VEGF, Vascular endothelial growth factor; Wt1, Wilms tumor protein; Yap/Taz, Yes associated protein 1/Transcriptional coactivator with PDZ-binding motif.

course the adult epicardial lining (Winter and Gittenbergerde Groot, 2007; Gittenberger-de Groot et al., 2012). In this manuscript we will provide a state-of-the-art review on the developmental process leading to the formation of the PE/epicardium, the signaling pathways providing cell specification and fate determination to those epicardial precursor cells including the upcoming role of non-coding RNAs, and current evidences on the regenerative role of the epicardium as to heal the injured heart.

## INITIAL PHASES OF THE PROEPICARDIAL (PE) AND EPICARDIAL FORMATION; A JOURNEY TO THE DEVELOPING EMBRYONIC HEART

The proepicardium (PE) is a small protuberance that progressively develops within limiting boundaries between the hepatic and cardiac primordia. It is composed of an external epithelial lining configured as a cauliflower structure and an internal mesenchymal component (Virágh et al., 1993; Kálmán et al., 1995; Ratajska et al., 2008). A single PE anlage is observed at early developmental stages in zebrafish (Serluca, 2008) while in the sturgeon and in mice bilateral PE buds are formed subsequently merging into a single midline structure (Schulte et al., 2007; Icardo et al., 2009). Curiously, in chicken two PE primordia are formed, but interestingly the right PE anlage develops before the left one is visible (Schulte et al., 2007). These data suggest divergent evolutionary trends on the formation of the PE primordia and furthermore advocate that embryonic left-right signaling might play a role controlling PE formation (Schlueter and Brand, 2012).

Transcriptional heterogeneity is widely documented for the PE anlage, and in addition, cell specific markers for several of the PE/epicardial cell derivatives, such as endothelial (Poelmann et al., 1993; Mikawa and Gourdie, 1996; Cossette and Misra, 2011; Niderla-Bielinska ´ et al., 2015) and smooth muscle (Valder and Olson, 1994) cells have also been documented, suggesting an early heterogeneous compartmentalization. Subsequently after the formation of the PE a process of delamination and migration of the proepicardial cells occurs. This process will lead to external covering of the atrioventricular canal and the entire atrial and ventricular myocardial chambers as demonstrated by seminal studies using quail-chicken embryos (Pérez-Pomares et al., 1998, 2002; Vrancken Peeters et al., 1999; **Figure 1**). In zebrafish, this process is dependent on the pericardial fluid currents (Peralta et al., 2013, 2014; Plavicki et al., 2013, 2014). In mice, proepicardial cells are detached from the PE forming cysts that migrate to the developing cardiac chambers through the pericardial cavity (Männer et al., 2001; Hirose et al., 2006). These cysts randomly attach to the ventricular and atrial chambers and progressively expand until the final full coverage of the cardiac chambers is completed.

Once the PE cells migrate and cover the surface of the developing embryonic myocardium an epicardial-myocardial signaling crosstalk is initiated. This process is crucial for the correct development of both cardiac tissue layers. The epicardium is instructed to initiate an epithelial-to-mesenchymal transformation (EMT), detaching from the epithelial epicardial layer and migrating first into the subepicardial space. These cells subsequently invade the myocardial walls, giving rise to the epicardial derived cells (EPDCs) (Dettman et al., 1998). An additional source of subepicardial cells of hematopoietic origin is provided during embryonic development which further contributes to the heterogeneity of the embryonic and postnatal epicardium (Balmer et al., 2014). In the following chapters we provide a state-of-the-art review on the differential contribution of the embryonic epicardium in cardiovascular development and disease.

## CELL FATE AND CONTRIBUTION OF THE EMBRYONIC EPICARDIUM TO THE MATURE HEART

Epicardial derived cells once they go through the subepicardial space continue their journey into the developing heart. Seminal approaches using quail-chick chimeras demonstrated that quail EPDCs contribute to distinct cardiac cell lineages, such as endothelial and smooth muscle cells in the coronary vasculature, endocardial mesenchymal cells in the atrioventricular cushions and also cardiac fibroblasts (Poelmann et al., 1993; Dettman et al., 1998; **Figure 1**). Since the experimental model used was a heterologous chimera, multiple criticisms were arising as which was indeed the real contribution of these cells. Supporting evidences were generated using retroviral-defective cell lineage tracing experiments in chicken hearts providing similar results (Mikawa and Gourdie, 1996); i.e., vascular endothelial, smooth muscle, and cardiac fibroblasts. Contribution to endocardial cushions is scarce, although it has been proposed that these cells are important for the correct development of the atrioventricular junction and the annulus fibrosus (Lie-Venema et al., 2008; Zhou et al., 2010; Lockhart et al., 2014 ). More recently, a contribution to cardiac resident stem cells (mesenchymal-like) has also been reported (Chong et al., 2011). In all cases, contribution to the developing myocardium was never observed (Poelmann et al., 1993; Mikawa and Gourdie, 1996; Pérez-Pomares et al., 2002). Surprisingly, in vitro PE culture experiments demonstrated that cardiomyocytes could be derived from these precursor cell pools (Kruithof et al., 2006).

With the advent of the molecular era, genetic lineage tracing in mice assaulted the quest to understand the contribution of the PE/embryonic epicardium to the mature murine heart. Several lineage tracing approaches were documented, in most cases, using Cre/loxP conditional activation of the reporter genes. In this setting, Tbx18-lineage tracing demonstrated a contribution to all the previously reported EPDC-derived lineages but surprisingly, also to the cardiomyocyte lineage. Whereas, these studies claimed that epicardial Tbx18+ cells contributed in vivo to ventricular cardiomyocytes (Cai et al., 2008), it was previously reported that fetal cardiomyocytes also expressed Tbx18 (Franco et al., 2006; Christoffels et al., 2009; Zeng et al., 2011) and thus those Tbx18+ epicardial lineage tracing experiments were dubious.

Epicardial Wt1+ derived cells have also been reported to contribute to endothelial cells and to the myocardium (Zhou et al., 2008; Zhou and Pu, 2011). Evidence for Wt1+ cells in the embryonic heart has also been reported but excluding cardiomyocytes (Zeng et al., 2011) yet more recent evidence demonstrated that Wt1-derived cardiomyoctes can be traced in the developing heart before PE/epicardial formation (Rudat and Kispert, 2012; Cano et al., 2016), thus questioning the epicardial contribution to the developing cardiac muscle. On the other hand, prove that epicardial cells do not contribute to myocardium in zebrafish comes from Tcf21-tracing (Kikuchi et al., 2011) and transplant experiments (González-Rosa et al., 2012), in which a contribution to the perivascular beds is reported. While these data might support the notion that epicardial cells can contribute to the formation of cardiomyocytes in vivo, yet these evidences remain controversial, mainly because the limitation on the use of Cre-based techniques as a bone fide fate mapping approach (Christoffels et al., 2009).

Additional controversies have also arisen regarding the contribution of EPDCs to other vascular components. To date, it seems clear that EPDCs mostly contribute to cardiac fibroblasts and vascular smooth muscle cells, but their contribution to vascular endothelial cells have also been challenged by additional Cre-based fate mapping experiments. In fact, epicardial-derived Cre based lineage tracing in mice failed to provide substantial contribution to the developing vascular endothelium in mice (Merki et al., 2005; Cai et al., 2008; Zhou et al., 2008). Red-Horse et al. (2010) described that coronary endothelial lining was mostly entirely derived from the sinus venosus endothelium as revealed by an Apelin-Cre mice (Red-Horse et al., 2010; Tian et al., 2013), a process that is VEGF-dependent (Chen et al., 2014). However, additional evidences reported that ventricular endocardial cells also can contribute to the coronary vasculature (Wu et al., 2012) as revealed by Nfatc1-Cre lineage tracing. Furthermore, by the usage of novel proepicardial lineage tracing markers such as Scleraxis-Cre, Semaphorin-3D-Cre, and Fabp4- CreER drivers (Katz et al., 2012; He et al., 2014) a contribution to the coronary vasculature was also reported. In fact, reconciling evidences reported by Chen et al. (2014) determined that sinusvenous (SV) derived coronary vasculature mostly contributed to the dorsal and lateral coronary vasculature (∼70%) whereas the ventral aspects were mostly endocardial derived (∼70–80%), with just a small (∼20%) but uniform contribution from the epicardium. These data are in line with a recent report that similarly estimated a 20% contribution from the proepicardium (Cano et al., 2016). Interestingly, a significant proportion of SV-derived and endocardial-derived cells displayed overlapping patterns with PE-derived cells, suggesting a common lineage origin. These data support the notion that multiple precursor cell populations contribute to the formation of the cardiac vasculature in mice, in contrast to avian hearts, in which the epicardial-derived contribution is large and undisputed. Lineage relationships between these three distinct coronary vasculature components remain nonetheless to be fully elucidated in mice.

Over the last decade our understanding of the molecular regulation of epicardial derived cells has largely increased with TABLE 1 | List of transcription factors, growth factors are other distinct molecules involved in distinct phases of proepicardium/epicardium development.


the usage of conditional spatio-temporal deletion of discrete signaling pathways. Epicardial cells display distinct divergent and overlapping expression patterns of Wt1, Nfatc1, Tbx18, and Pod1 in the chicken and murine hearts (Braitsch et al., 2012), providing a heterogeneous panel of potentially distinct cardiac stem cells. Whereas, to date it remains elusive when and how epicardial cells becomes specific to their prospective lineage, it is increasing clear that multiple factors play pivotal roles in this process as summarized in **Table 1**. In particular, PDGFRβ is important for epicardial migration and for the development of coronary vascular smooth muscle cells (Mellgren et al., 2008; Bax et al., 2009; Smith et al., 2011), retinoic acid and VEGF primes endothelial vs. smooth muscle differentiation (Guadix et al., 2006; Tomanek et al., 2006; Azambuja et al., 2010) and Fgf signaling (Guadix et al., 2006), mainly through Fgf10 and Fgfr2b are essential for cardiac fibroblast formation (Vega-Hernández et al., 2011). In addition, Pod1/Tcf21 is regulated by retinoic acid and inhibits differentiation of EPDCs into smooth muscle cells in chicken and mice (Braitsch et al., 2012) while Wnt signaling is also important for epicardial specification, as Dkk1 and Dkk2 mouse mutants display impaired epicardial development (Phillips et al., 2011). Similarly PCP disruption is also critical in this context (Phillips et al., 2008) as well as MAPK kinase genetic inactivation (Liberatore and Yutzey, 2004; Craig et al., 2010a,b). Other signaling pathways, such as CXCL12/CXCR4 are also crucial for cardiac vascular development (Cavallero et al., 2015). Furthermore, Hippo signaling, mediated by Yap/Taz modulates Tbx18 and Wt1 expression in the epicardium controlling their contribution to the coronary vasculature (Singh et al., 2016). Several other molecules have also been reported to be critical for coronary artery formation, such as tenascin C (Ando et al., 2011) and nephrin (Wagner et al., 2011) particularly for smooth muscle recruitment to those cardiac vessels. Overall these findings highlight the complexity of distinct signaling pathways and molecules governing the coronary vasculature development.

## THE ROLE OF THE POSTNATAL EPICARDIUM IN THE INJURED HEART

Within the adult heart, the epicardium represents the outermost layer, which is a simple epithelial layer. For many years, the functional role of this layer has been neglected as it was considered as an external cover devoid of any functional meaning. The discovery that the epicardial precursors can differentiate to beating cardiomyocytes has branded the epicardium as a source of cardiac stem cells with great therapeutic potential (Wessels and Pérez-Pomares, 2004; Pérez-Pomares et al., 2006; Winter et al., 2009). In addition, it has been reported that the adult epicardium plays a pivotal role in cardiac regeneration (Bollini et al., 2011, 2015; Schlueter and Brand, 2012; Masters and Riley, 2014; Kennedy-Lydon and Rosenthal, 2015; **Figure 1**) as detailed below.

Seminal work by Kruithof et al. (2006) described that the embryonic chicken PE if placed in appropriate cell culture conditions, was capable of giving rise to beating cardiomyocytes. Such in vitro conditions could be further promoted by Bmp administration and blocked by Fgf signaling. Thus, these data opened out the possibility that the epicardium could serve as an in vivo source of potential cardiomyocytes if the appropriate signals would be instructed in vivo. Importantly, Smart et al. (2011) demonstrated that adult epicardial derived cells, if previously primed with thymosinβ4, eventually generated functionally beating cardiomyocytes in an ischemic heart, yet the proportion of de novo integrated cells was rather spurious and its instructive mechanism remains rather obscure (Gajzer et al., 2013). Nonetheless, as a proof of principle approach it was highly valuable. This work introduced a novel concept of an activated epicardium, a condition by which embryonic epicardial markers such as Wt1 and Tbx18 are re-expressed in the adult epicardium (Huang et al., 2012; van Wijk et al., 2012; Braitsch et al., 2013; Bollini et al., 2014; Aguiar and Brunt, 2015) in response to distinct biological stimuli such as thymosinβ4 (Smart et al., 2012; Smart and Riley, 2012), stem cell factor (SCF; Xiang et al., 2014), and prokineticins (Urayama et al., 2008) among others. In addition, this activated epicardium secretes paracrine factors that modulate myocardial injury response (Zhou et al., 2011; Foglio et al., 2015).

While it is documented that the human heart has a limited capacity to regenerate (Bergmann et al., 2009), it is also highly acknowledged that the newt heart can also widely regenerate by other means (Becker et al., 1974; Oberpriller and Oberpriller, 1974). Furthermore, the adult zebrafish heart can also regenerate (Poss et al., 2002) and the epicardium provides a pivotal role during this regeneration process (Gemberling et al., 2015; Wang et al., 2015). Molecular analyses have demonstrated that the epicardium becomes activated as soon as the heart is injured and such activation provides instructive signals that promote cardiomyocyte proliferation, revascularization, and tissue repair (Lien et al., 2006, 2012; Marín-Juez et al., 2016). During this process a transitory scar stage occurs and is subsequently replaced by fully functional and integrated cardiomyocytes (González-Rosa et al., 2011; Mercer et al., 2013; Itou et al., 2014; Marro et al., 2016).

Further analyses in this front identified that Wnt1/βcatenin is crucial promoting formation of cardiac fibroblasts and hence cardiac repair (Duan et al., 2012). Several studies have identified key molecules modulating this regeneration capacity. For example, Nrg1 acts as a mitogenic agent in cardiomyocytes following injury during cardiac zebrafish regeneration (Gemberling et al., 2015). Notch (Zhao et al., 2014), Raldh2 (Itou et al., 2014), and myocardial NF-κB (Karra et al., 2015) are also essential for heart regeneration in zebrafish. Hydrogen peroxide (Han et al., 2014) has been reported to prime heart regeneration and telomerase has been identified as instrumental for zebrafish regeneration (Bednarek et al., 2015), but still it remains to be established if these factors are modulated by the epicardium. More recently, it has been demonstrated that epicardial regeneration is guided by the cardiac outflow tract and hedgehog signaling (Wang et al., 2015) and single cell transcriptome of the epicardium has identified caveolin1 as an essential factor in regenerating zebrafish heart (Cao and Poss, 2016). Moreover, re-expression of epicardial developmental genes and enhanced EMT in response to injury has been widely demonstrated (Lepilina et al., 2006; Kim et al., 2010; González-Rosa et al., 2011; Schnabel et al., 2011). These data suggest that complex regulatory networks control zebrafish regeneration (Rodius et al., 2016) positioning the epicardium as a key tissue layer for regeneration. Thus, these data will be highly instrumental to search for novel ways to heal the injured heart.

In adult mice, the regenerative capacity is lost and the injured heart responds by generating a fibrous scar which is derived from pre-existing epicardial cells (Zhou et al., 2008; Duan et al., 2012) as well as de novo recruited bone marrow-borne circulating cells (Ruiz-Villalba et al., 2015). Interestingly, full regeneration is achieved at early developmental stages, i.e., on the first week of life, in which the epicardium (Porrello et al., 2011) is also a highly instructive player and thymosinβ4 priming increases the time window for mammalian heart regeneration (Rui et al., 2014). In addition a role for Wnt signaling has also been identified in the regenerating heart in mice (Mizutani et al., 2016). Recent evidence demonstrated that exosomal signaling from the epicardium is essential for myocardial maturation highlighting a pivotal role for clustering in this process (Foglio et al., 2015). All these efforts have provided the bases of heart regeneration. A giant step was recently reported by Wei et al. (2015) whom used reconstitution of epicardial follistatin-like1 expression in biomaterial patches to heal the adult injured heart, opening a novel way to regenerate the adult mammalian heart.

## AN UNEXPECTED EPICARDIAL DERIVATIVE WITH PARACRINE SIGNALING LEADING TO CAD AND AF

While it is highly acknowledged that the epicardial precursor cells, within the PE, and subsequently the EPDCs will give rise to distinct cardiovascular embryonic cell lineages, it has remained unexplored if the adult epicardium can generate additional cellular subpopulations. Recent evidences have demonstrated that intramyocardial adipose tissue is derived from the endocardium (Zhang et al., 2016), whereas adipose tissue around the heart, mainly at the venous, arterial connections, and atrial appendages is an adult epicardium derivative (Yamaguchi et al., 2015). Furthermore, cardiac adipose tissue deposition has recently been associated to distinct cardiovascular pathologies (**Figure 1**), such as coronary arteries diseases (Iwayama et al., 2014), atherosclerosis plaque disruption (Talman et al., 2014; Yamashita et al., 2014), and atrial fibrillation (Batal et al., 2010; Nakanishi et al., 2012; Gaborit et al., 2013). Although, these are early days to fully understand the molecular mechanisms linking epicardium, adipose tissue deposition, and cardiovascular pathologies, supporting evidences suggest that these cells can act as paracrine signaling center that, if impaired, can be the source of cardiovascular diseases (Langlois et al., 2010; Greulich et al., 2012).

## NON-CODING RNAs IN THE PE/EPICARDIUM

Over the last decade we have witnessed a revolution in the concept of the control of gene expression with the discovery of non-coding RNAs. Non-coding RNAs can be broadly classified according to the transcript size into long non-coding RNAs (lncRNAs) and small non-coding RNAs. Our current understanding of lncRNAs is still in its infancy with just a limited number of reports in the developing heart (Grote et al., 2013; Klattenhoff et al., 2013; Sauvageau et al., 2013; Zhu et al., 2014; Kurian et al., 2015). On the other hand, our knowledge on the functional role of small non-coding RNAs, in particular microRNAs, has been largely increased (Callis and Wang, 2008; Chen and Wang, 2012; Bonet et al., 2013; Philippen et al., 2015; Yan and Jiao, 2016). microRNAs are small non-coding RNA of 18–24 nt in length that by homologous base-priming are capable of blocking translation or degrading mRNA transcripts. microRNAs are transcribed by RNA polymerase II, 5′ capped and 3′ polyadenylated leading to mature microRNA by RNA endonucleases such as Drosha and Dicer (Aranega and Franco, 2015; Towler et al., 2015). Mature microRNAs are loaded into the RISC complex which can thereafter search for mRNA transcript base complementarity (Hammond, 2015; Shen and Hung, 2015). To date more than a 1,000 distinct microRNAs have been identified in humans, which are quite conserved among evolution. A seminal study by Singh et al. (2011) demonstrated that conditional ablation of Dicer, an RNAse processing enzyme, in the epicardium provoked impaired epicardial formation, thin-walled myocardium, and aberrant coronary vasculature formation. Thus, this study demonstrated a pivotal role for microRNAs in PE/epicardium development. A large array of studies have been reported in key developmental processes by which the PE/epicardium is formed, such as epithelialto-mesenchymal transition in cancer (see for recent reviews; Behbahani et al., 2016; Peng et al., 2016; Sulaiman et al., 2016; Zou et al., 2016) and also within the heart (Stankunas et al., 2010; Bonet et al., 2015) and cardiac regeneration (Porrello et al., 2013) but surprisingly only a short list of studies have been reported in PE/epicardium formation. miR-21 has been reported in numerous studies promoting fibrogenesis both during cardiac development and disease (Thum et al., 2008; Adam et al., 2012; Derda et al., 2015; Gupta et al., 2016). Brønnum et al. (2013a) has recently reported that miR-21 promotes fibrogenic EMT in epicardial cells by modulating Pcd4 and Sprouty-1 and these authors (Brønnum et al., 2013b) have also reported that islet-1 can influence miR-21 expression and therefore modulate cardiac fibrogenic EMT. Seeger et al. (2016) demonstrated that let-7 inhibition enhances the recruitment of epicardial cells after myocardial infarction promoting an improved cardiac function. Overall these studies demonstrate a nascent role for microRNAs in PE/epicardium formation, which might provide novel approaches to activate and prime epicardial cells for cardiac regeneration.

## CONCLUSIONS AND PERSPECTIVES

Over the last decade our understanding of the cellular contribution of the PE/epicardium has largely increased. Seminal works using quail-chick chimeras demonstrated a large plasticity for the EPCDs, contributing to the cardiac fibrous skeleton, the coronary vasculature and the developing atrioventricular valves (Poelmann et al., 1993; Wessels and Pérez-Pomares, 2004; **Figure 1**). However, with the advent of molecular tracing tools, multiple evidences demonstrated a rather more complex contribution and architecture to the coronary vasculature in mice. Cre-driven fate mapping can be pervasive and promiscuous tools, deriving in complex and in many cases controversial findings. We hope that either retrospective clonal analysis as previously reported for myocardial components (Meilhac et al., 2003, 2004a,b) or genuine prospective lineage tracing would serve to reconcile these findings in the PE/epicardial context. With no doubt one of the seminal work that prompted the interest of the epicardial lining in the context of cardiac stem cell and cardiac regeneration was reported by Kruithof et al. (2006) demonstrating that PE/epicardial cells could be generating cardiomyocyte in vitro opening the possibilities to unlock the myocardial lineage commitment in vivo. Thymosin beta4 was the first of these unlocking tools, providing an entry site to regenerate the heart using the epicardium as a cell source (Smart et al., 2011; Smart and Riley, 2012). In addition, bridging epicardial activation by follistatin-like1 into biomaterials provided additional convincing evidences on the feasibility of these approaches (Wei et al., 2015). New tools will be discovered in the near future.

In recent years a novel link between the epicardium and epicardial derived structures is emerging (**Figure 1**). Intriguingly, adipose fat deposition within the pericardiac regions has been linked to cardiac pathophysiologies such as coronary artery atherosclerosis and atrial fibrillation. To date the casual relationship remains enigmatic, yet a plausible embryonic link might be present since epicardial cells can differentiate into adipose tissue (Zhang et al., 2016) and epicardial cells contribute to both endothelial and smooth muscle components of the coronary vessels (Pérez-Pomares et al., 2002; Cano et al., 2016). However, our current understanding is still in its infancy and for sure we will witness additional cellular and molecular evidences deciphering the interplay between these rather apparent distinct cardiovascular entities.

While our cellular and molecular understanding of PE/epicardium/EPDC has greatly advanced in recently years, the discovery of novel levels of gene regulations, in particular those exerted by the non-coding RNAs, is called to change our molecular and signaling pathways schemes. The discovery that microRNAs are crucial to epicardial development is simply demonstrating the equally pivotal roles of these tiny molecules in other cardiovascular developmental contexts (Cordes and Srivastava, 2009; Chinchilla et al., 2011; Bonet et al., 2015). In addition to microRNAs, long-non-coding RNAs are also called to play pivotal role in cardiogenesis (Grote et al., 2013; Klattenhoff et al., 2013) and thus similarly in epicardial development. In coming years, additional routes would be discovered demonstrating the essential role of these

## REFERENCES


new players in epicardial biology both during development and disease.

## AUTHOR CONTRIBUTIONS

AD searched for literature records and wrote part of the manuscript. DF wrote part of the manuscript, editing, and approved the final manuscript version. AA read and suggested editing comments to the manuscript.

## FUNDING

DF and AA are partially funding by a grant of the Ministerio de Economia y Competitividad (BFU2015-67131-P) of the Spanish Government and the Consejeria de IEconomia y Conocimiento (CTS-446) of the Junta de Andalucia Regional Council.

epithelial to mesenchymal transition in cancer. Pathol. Oncol. Res. 23, 1–12. doi: 10.1007/s12253-016-0101-6


of the interventricular septum in the mouse heart. Dev. Biol. 294, 366–375. doi: 10.1016/j.ydbio.2006.02.045


organ give rise to coronary vascular endothelial cells. Dev. Cell 22, 639–650. doi: 10.1016/j.devcel.2012.01.012


immunohistochemical and quail-chick chimera study. Dev. Biol. 200, 57–68. doi: 10.1006/dbio.1998.8949


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

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

# Non-myogenic Contribution to Muscle Development and Homeostasis: The Role of Connective Tissues

#### Sonya Nassari, Delphine Duprez and Claire Fournier-Thibault\*

Developmental Biology Laboratory, IBPS, Centre National de la Recherche Scientifique UMR7622, Institut National de la Santé Et de la Recherche Médicale U1156, Université Pierre et Marie Curie, Sorbonne Universités, Paris, France

#### Edited by:

Juan Jose Sanz-Ezquerro, Consejo Superior de Investigaciones Científicas (CSIC), Spain

#### Reviewed by:

Peleg Hasson, Technion – Israel Institute of Technology, Israel Denitsa Docheva, Universitätsklinikum Regensburg, Germany Malcolm Logan, King's College London, UK

> \*Correspondence: Claire Fournier-Thibault claire.thibault@upmc.fr

#### Specialty section:

This article was submitted to Signaling, a section of the journal Frontiers in Cell and Developmental Biology Received: 20 December 2016

Accepted: 07 March 2017 Published: 23 March 2017

#### Citation:

Nassari S, Duprez D and Fournier-Thibault C (2017) Non-myogenic Contribution to Muscle Development and Homeostasis: The Role of Connective Tissues. Front. Cell Dev. Biol. 5:22. doi: 10.3389/fcell.2017.00022 Skeletal muscles belong to the musculoskeletal system, which is composed of bone, tendon, ligament and irregular connective tissue, and closely associated with motor nerves and blood vessels. The intrinsic molecular signals regulating myogenesis have been extensively investigated. However, muscle development, homeostasis and regeneration require interactions with surrounding tissues and the cellular and molecular aspects of this dialogue have not been completely elucidated. During development and adult life, myogenic cells are closely associated with the different types of connective tissue. Connective tissues are defined as specialized (bone and cartilage), dense regular (tendon and ligament) and dense irregular connective tissue. The role of connective tissue in muscle morphogenesis has been investigated, thanks to the identification of transcription factors that characterize the different types of connective tissues. Here, we review the development of the various connective tissues in the context of the musculoskeletal system and highlight their important role in delivering information necessary for correct muscle morphogenesis, from the early step of myoblast differentiation to the late stage of muscle maturation. Interactions between muscle and connective tissue are also critical in the adult during muscle regeneration, as impairment of the regenerative potential after injury or in neuromuscular diseases results in the progressive replacement of the muscle mass by fibrotic tissue. We conclude that bi-directional communication between muscle and connective tissue is critical for a correct assembly of the musculoskeletal system during development as well as to maintain its homeostasis in the adult.

Keywords: connective tissue, muscles, bones, tendons, development, regeneration

## INTRODUCTION

Skeletal muscle forms a highly complex and heterogeneous structure, which is part of the musculoskeletal system of the body. The process of generating muscles is defined as "myogenesis." This mechanism occurring during development is an important step in the establishment of the musculoskeletal system allowing its essential functions, for instance body motion or the ability to breath. Myogenesis occurs through successive and overlapping phases that ultimately give rise to correctly patterned muscles. In the first phase of myogenesis, which is called embryonic myogenesis, embryonic progenitors cells form primary muscle fibers, which constitute the scaffold of the muscles. During the second phase of myogenesis named fetal myogenesis, fetal progenitors fuse between themselves and with primary fibers to form secondary fibers and allow muscle growth. Both waves of myogenesis occur during embryonic development, and involve specific types of muscle progenitors cells. After birth, a third wave of myogenesis can be activated during muscle regeneration, which occurs after muscle damage. This step involves muscle stem cells, so-called muscle satellite cells, which contribute to muscle reconstruction by fusing with the existing muscle fibers or generating new muscle fibers (Stockdale, 1992; Tajbakhsh, 2009; Tedesco et al., 2010). Studies suggest that embryonic myogenesis is largely exhausted at the end of embryonic development, while fetal and perinatal phases of myogenesis persist to contribute to the majority of adult muscle stem cells (reviewed in Tajbakhsh, 2009).

The intrinsic molecular signals regulating the different waves of myogenesis have been well described in the literature. However, the interactions between muscles and adjacent tissues during development are not completely elucidated. During development and adult life, as part of the musculoskeletal system, muscles are closely associated with the other components of this system: bone, cartilage, tendon, ligament and irregular connective tissue, all of them emerging from the family of connective tissues. Although the interactions between the different components of the musculoskeletal system during development has been highlighted from the 1980's, more recent work has begun to decipher the molecular mechanisms underlying the importance of connective tissue in the regulation of developmental and regenerative myogenesis.

The scope of this review is to synthesize the data supporting the process of connective tissue-mediated muscle development and regeneration and to point out the active role of this so-called "supporting tissue" in muscle formation and repair. Indeed, defect in connective tissue-muscle interactions can lead to human pathology, as congenital diaphragmatic hernias, a birth defect of the diaphragm muscle (Merrell et al., 2015), or the Holt-Oram syndrome characterized by skeletal defects of the upper limbs and heart anomalies (Hasson et al., 2010). In addition, in skeletal muscle regenerative disorders (muscular dystrophies) as well as in aging (sarcopenia), the impairment of the muscle regenerative potential correlates with a progressive replacement of contractile mass by fibrotic and adipose tissues (reviewed in Farup et al., 2015). It is therefore necessary to better understand the interactions occurring between the different components of the musculoskeletal system. This would allow us to decipher the molecular mechanisms underlying muscle disorders not related to the impairment of intrinsic regulation of myogenesis.

## CONNECTIVE TISSUE DEVELOPMENT

## Different Types of Connective Tissues

In the body, the main role of **connective tissues (CTs)** is to support and connect organs together. CTs are primarily composed of fibroblasts and extracellular matrix consisting of amorphous gel-like and matrix fibers. The amorphous gel-like, named ground substance, mostly contains glycoproteins and proteoglycans, while the fibrous network is made of collagen and elastic fibers (Omelyanenko and Slutsky, 2013). Among the **supportive** CTs, two main types can be distinguished: the **specialized CT** and the **dense CT**. The specialized CT refers to bones and cartilage elements. The dense CT is further divided into two subtypes: the **dense regular CT** and the **dense irregular CT**, which refer respectively to tendon/ligament structures and to CT embedding organs (**Table 1**). The nature and function of these different CTs are predominately determined by the composition and organization of the extracellular matrix. In dense regular CT, fibroblasts produce a significant amount of collagen fibers that display a spatial organization, while in the dense irregular CT, fibroblasts produce collagen fibers that do not present any specific organization (Omelyanenko and Slutsky, 2013).

## Connective Tissue Formation

During embryonic development, undifferentiated mesenchymal cells, derived from mesodermal and mesectodermal (neural crest cells) origins, give rise to the different forms of CT: bones, cartilage, tendons, ligaments, and irregular CT (Wachtler et al., 1981). Head CTs originate from neural crest cells, while CTs of the body originate from paraxial or lateral plate mesoderm (**Figure 1**). The specification and differentiation processes of the different types of CTs is controlled by specific key transcription factors or signaling molecules. Irrespective to their embryological origins, transcription factors have been identified for the specification of each type of CT from undifferentiated mesenchymal cells (**Figure 2**).

### Specialized Connective Tissue (Bone and Cartilage)

The embryonic origins of cartilage and bone are multiple. Indeed, elements of the trunk, head and limb skeleton arise from three distinct embryonic structures, somites, neural crest cells and lateral plate mesoderm (**Figure 1**, Wachtler et al., 1981; Christ and Wilting, 1992; Noden and Trainor, 2005). The process of skeleton formation, which corresponds to the development of cartilage and bone elements, is initiated by the condensation of undifferentiated mesenchymal cells at the future sites of bones. Following condensation, mesenchymal precursors undergo either chondrocyte or osteoblast differentiation, giving rise respectively to cartilage or bone. Osteogenesis characterizes the process of ossification, which occurs through two different mechanisms. The process of intramembranous ossification corresponds to a direct transition from condensed undifferentiated-mesenchymal cells into osteoblasts (as described above). The second mechanism of bone formation is called endochondral ossification. It defines the replacement of cartilage by bone. In this case, chondrogenesis is the first step in a process that ultimately gives rise to bones. Intramembranous ossification occurs in bones of the skull, while other bones form by endochondral ossification (reviewed in Karsenty and Wagner, 2002).

Molecular mechanisms involved in cartilage and bone specifications are well understood (**Figure 2**). Members of the

#### TABLE 1 | Classification of the different types of connective tissues.


SOX (SRY-related HMG-box) transcription factor family are key players in the regulation of cartilage specification (Lefebvre et al., 1998). During mouse embryonic development, Sox9 presents a similar expression pattern to Col2a1, the main collagen in the cartilage extracellular matrix (Zhao et al., 1997). In mouse mutant embryos for Sox9, cartilage development fails. The complete absence of cartilage elements in Sox9 mutant mice highlights a role for Sox9 in the regulation of mesenchymal cell condensation and differentiation toward a cartilage fate (Bi et al., 1999; Akiyama et al., 2002). Moreover, it has been shown that Sox9 is required for the expression of two additional Sox genes, Sox5, and Sox6 that are co-expressed with Sox9 in committed cartilage cells (chondrocytes), (Lefebvre et al., 1998, 2001). Both Sox5 and Sox6 mutant mice show skeletal abnormalities, with no modification of Sox9 expression, demonstrating that Sox9 acts upstream of Sox5 and Sox6 (Smits et al., 2001).

Runx2 (Runt-related transcription factor 2) is a master gene for osteogenesis (Komori et al., 1997; Ducy et al., 1999). This transcription factor is specific to bone progenitor cell lineage (Ducy et al., 1999). Knockout mice for Runx2 show no osteogenesis. While cartilage elements are still present in Runx2 <sup>−</sup>/<sup>−</sup> mouse, all bones are missing, demonstrating the importance of Runx2 in bone specification (Komori et al., 1997). In contrast to Sox9, which is required for cartilage differentiation in addition to specification (Akiyama et al., 2002), Runx2 is not required for bone differentiation (Takarada et al., 2013). After osteogenic cell commitment, Runx2 activity has to be shut down to allow immature committed bone cells to become fully mature and to differentiate (Yoshida et al., 2004; Takarada et al., 2013; Adhami et al., 2014). Osterix (Osx) is also a key transcription factor in bone formation (Nakashima et al., 2002). Osx is specifically expressed in all bones (Nakashima et al., 2002) and is required for differentiation of bone progenitor cells. In mutant mouse

for Osx, no bone is observed, however Runx2 expression is maintained (Nakashima et al., 2002). Conversely, Osx expression is absent in Runx2 mutant mice (Nakashima et al., 2002). This indicates that Runx2 and Osx are involved in bone specification and differentiation, respectively.

Beside the specific transcription factors, major signaling pathways have also been demonstrated to be involved in skeletal development. Wnt pathway regulates skeleton differentiation through a cell-autonomous mechanism, which enhances osteoblast differentiation at the expense of chondrocytes (Day et al., 2005; Hill et al., 2005). Conditional inactivation of ßcatenin in mesenchyme blocks osteoblast differentiation and induces ectopic chondrocytes (Day et al., 2005). In addition, ß-catenin has been shown to control the expression of Sox9 and Runx2 in vitro (Day et al., 2005). The role of FGF signaling in skeletal development comes from the observations that FgfR3 and FgfR1 inactivation in mouse leads to achondroplasia and hypochondroplasia (Deng et al., 1996; Jacob et al., 2006). In both mutant mice, an expansion of the hypertrophic chondrocyte zone is observed, suggesting that FGF signaling is a negative regulator of chondrocyte proliferation (Deng et al., 1996; Jacob et al., 2006). Inactivation of one member of the Hedgehog family, Ihh (Indian Hedgehog), leads to a decrease in chondrocyte proliferation and a defect in osteoblast formation (Vortkamp et al., 1996). This effect is mediated by the interaction between Ihh and Parathyroid hormone, which maintains the rate between cell proliferation and differentiation (Kronenberg, 2006). Interestingly, RUNX2 induces Ihh expression which inhibits Runx2 expression by a feedback loop mechanism (Yoshida et al., 2004). Finally, BMPS (Bone Morphogenetic Proteins) are important regulators of chondrocyte differentiation (Kobayashi et al., 2005, reviewed in Li and Cao, 2006) and have been shown to regulate IHH expression in chick embryos (Zhang et al., 2003).

### Dense Regular Connective Tissue (Tendon)

Similarly to specialized CTs (bone and cartilage), tendons arise from distinct embryological origins depending on their position in the body (**Figure 1**). Tendons of the trunk originate from somites, more precisely from a subregion of the sclerotome named the syndetome (Brent et al., 2003), tendons of the craniofacial region derive from neural crest cells (Crane and Trainor, 2006; Grenier et al., 2009) and limb tendons derive from the lateral plate mesoderm (Kieny and Chevallier, 1979). Tendons attach muscles to bones by connecting muscle at the myotendinous junction and connecting bone at the enthesis, while ligaments connect bone to bone. The role of tendons is to transmit forces generated by muscle contractions to bones, in order to allow joint movements and maintain articular stability. The tendon extracellular matrix is rich in type I collagen fibers, which display a specific spatial organization parallel to the tendon axis. This specific organization lends mechanical properties to tendons (Benjamin and Ralphs, 2000). Ligaments are essential components of the skeletal joints. Their elasticity defines the range of motion of the joints, supports joint stability and protects joints and bones by their stretching capacities. Tendons and ligaments display similar structural collagen organization and molecular markers (Benjamin and Ralphs, 2000). However, genome-wide analysis identifies different levels of gene expression between adult tendons and ligaments (Pearse et al., 2009). However, tendon development has been more studied than ligament development (Tozer and Duprez, 2005).

In contrast to cartilage and bone, the master gene(s) involved in tendon specification during development is (are) still unknown. To date, Scx (Scleraxis) is the unique early tendon marker that has been described in vertebrates. Scx is specifically expressed in tendon progenitors and differentiated cells (Schweitzer et al., 2001). Scx mutant mice display severe tendon defects, leading to a severe impairment of limb and tail force-transmitting tendons, while anchoring tendons are less affected (Murchison et al., 2007). However, tendon progenitors are still present in Scx−/−, indicating that Scx is not the master gene driving tenogenesis during development. Two additional transcription factors have been identified to be involved in tendon formation, the homeobox transcription factor Mkx (Mohawk), (Ito et al., 2010; Liu et al., 2010) and the zinc finger transcription factor Egr1 (Early Growth Response 1), (Lejard et al., 2011, **Figure 2**). Both Mkx and Egr1 mutant mice display tendon defects associated with a decrease in Col1a1 expression levels and in type I collagen fiber number in tendons (Ito et al., 2010; Liu et al., 2010; Lejard et al., 2011; Guerquin et al., 2013). However, both Mkx and Egr1 are expressed after Scx during development and are not specific to tendons, since they are expressed in many other lineages (Rackley et al., 1995; Anderson et al., 2006).

TGFß (Transforming growth factor) and FGF (Fibroblast growth factor) signaling pathways have been shown to regulate tendon specification and differentiation at different places of the body (recently reviewed in Gaut and Duprez, 2016). As mentioned above, **axial** tendon progenitors arise from a somitic subcompartment named the syndetome. The syndetome compartment, localized at the interface between the sclerotome and myotome, is formed by Scx-expressing cells. In chick embryos, axial tendons do not develop in the absence of axial muscles, as demonstrated by the absence of tendons after dermomyotome removal (Brent et al., 2003). Chick axial SCX expression is induced in response to FGF signaling arising from the myotome, which concomitantly downregulates PAX1 expression in the sclerotome (Brent et al., 2003). In contrast to axial tendons, the initiation of **head and limb** tendons is independent of muscle. In the absence of limb or craniofacial muscles, Scxa/SCX/Scx expression is normally induced in limb and head of zebrafish, chick and mouse embryos (Schweitzer et al., 2001; Edom-Vovard et al., 2002; Grenier et al., 2009; Chen and Galloway, 2014). In chick limbs, SCX induction is known to be mediated via ectodermal signals, as shown by the absence of SCX expression after ectoderm removal (Schweitzer et al., 2001). BMP signaling from the limb mesenchyme represses SCX expression and overexpression of the BMP antagonist NOGGIN leads to ectopic SCX expression, indicating that tendon specification in chick limbs results from a balance between an unidentified factor coming from the ectoderm and BMP signaling from the mesenchyme (Schweitzer et al., 2001). TGFß is a key signaling molecule for tendon development. TGFß signaling isrequired and sufficient for Scx/SCX expression during development in chick and mouse embryos (Pryce et al., 2009; Havis et al., 2014, 2016), while FGF signaling is required and sufficient for SCX expression in undifferentiated chick limb cells but not in mouse limb cells (Pryce et al., 2009; Havis et al., 2014, 2016).

Although muscles are not necessary for head and limb tendon initiation, they are required for the maintenance of Scxa/SCX/Scx expression in tendons and for full tendon differentiation. In the absence of muscles, tendons degenerate in chick, mouse and zebrafish embryos (Kardon, 1998; Edom-Vovard et al., 2002; Grenier et al., 2009; Chen and Galloway, 2014). Moreover, overexpression of FGF4, which is normally expressed at the tips of muscles fibers, leads to ectopic expression of tendon-associated genes in chick limbs (Edom-Vovard et al., 2002; Eloy-trinquet et al., 2009). In addition, chick embryo immobilization decreases SCX expression in limb tendons and application of FGF4 or TGFβ ligands prevents SCX down-regulation consecutive to immobilization, demonstrating that FGF and TGFβ act downstream of mechanical forces to regulate tendon differentiation (Havis et al., 2016).

### Dense Irregular Connective Tissue

The irregular CT constitutes a protective envelop for the different organs of the body, by embedding and scaffolding organs, with scattered cells embedded in high extracellular matrix content. Irregular CT is present all around organs, but also inside organs. First studies on the differentiation of irregular CT have focused on the extracellular matrix composition. During development, Type I and type III collagen are both expressed in dense regular and irregular CTs, however type I collagen tends to replace type III collagen in adult tendons, while mature irregular CT is characterized by the expression of both type III and type VI collagen (Kieny and Mauger, 1984; Zhang et al., 2005; Gara et al., 2011; Stricker et al., 2012). Due to the lack of specific early molecular markers, the mechanisms driving irregular CT specification have been poorly investigated. However, the recent identification of transcription factors expressed in irregular CT has provided new insights into irregular CT formation and function (**Figure 2**).

The first marker identified in irregular CT fibroblasts is the transcription factor TCF4, belonging to the TCF/LEF family. In limbs of both mouse and chick embryos, Tcf4-expressing cells discriminate the lateral plate-derived mesodermal population from myogenic cells (Kardon et al., 2003; Mathew et al., 2011). When chick limb muscles differentiate, TCF4 expression is restricted to muscle CT (Kardon et al., 2003) and colocalizes with type I collagen. Expression of TCF4 in muscle CT persists at adult stages (Mathew et al., 2011). TCF4 misexpression in chick limbs leads to muscle patterning defects, highlighting a non-cell autonomous effect of muscle CT on muscles, in which TCF4-expressing fibroblasts define a pre-pattern that ultimately drive muscle patterning (Kardon et al., 2003). However, low levels of Tcf4 have been also observed genetically in myogenic cells (Murphy et al., 2011). BMP signaling has been shown to negatively regulate TCF4 expression (Bonafede et al., 2006), while Wnt signaling positively regulates TCF4 expression (Kardon et al., 2003) in developing chick limbs. TCF4 is also expressed dynamically in avian jaw muscle CT and has been shown to be regulated by neural crest mesenchyme (Tokita and Schneider, 2009).

The T-box transcription factor Tbx5 is another gene that has been characterized as expressed in fibroblasts constituting irregular CT. At early stage of mouse limb bud development (E11.5), Tbx5 is broadly expressed in lateral plate mesodermal cells in domains overlapping with bone, tendon and muscle progenitors (Hasson et al., 2007). Disruption of Tbx5 function in mice leads to disorganization of muscle CT during embryonic development (Hasson et al., 2010), which could be related to subtle alterations of muscle CT markers, as Tcf4 and the Osr genes (see below), (Hasson et al., 2010). Tbx5 positively regulates the expression of N-cadherin and ß-Catenin in muscle CT and as the expression levels of Wnt signaling targets are not affected in Tbx5 mutant, it seems noteworthy that Tbx5 mostly affects cell adhesion mechanisms independently of Tcf4.

Two orthologs of the Odd-Skipped genes, Osr1 and Osr2, has been described as expressed in the irregular CT in chick and mouse embryos (Stricker et al., 2006, 2012). Both genes are expressed in a variety of organs such as kidney, eye, branchial arches, and dermis (So and Danielian, 1999; Lan et al., 2001; Stricker et al., 2006). In the developing limb of mouse and chick embryos, Osr1 is expressed in all irregular CTs, displaying a partial overlap with Tcf4 (Stricker et al., 2006). Osr2, although widely expressed in irregular CT, shows prevalence for muscle CT (Stricker et al., 2006, 2012). Both genes are also expressed in the mesenchyme of branchial arches in chick (Stricker et al., 2006) and mouse (Liu et al., 2013) embryos. Forced expression of OSR1 or OSR2 in chick mesenchymal progenitor limb cells induces the expression of irregular CT markers such as COL3A1 and COL6A1 and down-regulates the expression of markers of cartilage (specialized CT) and tendon (dense regular CT), (Stricker et al., 2012). Conversely, OSR1 or OSR2 inactivation down-regulates COL3A1 and COL6A1 expression, while increasing cartilage formation in chick limb cells (Stricker et al., 2012). Similarly, specific inactivation of Osr1 in cranial neural crest cells result in the formation of an ectopic cartilage in the developing mouse tongue (Liu et al., 2013). OSR1 has been shown to bind Sox9 promoter and repress Sox9 expression, indicating that OSR1 prevents chondrogenesis in the mammalian tongue through repression of Sox9 expression (Liu et al., 2013).

## MUSCLE DEVELOPMENT

## Embryonic Origins of Skeletal Muscles

In vertebrates, all skeletal muscles derive from paraxial mesodermal cells (**Figure 3**; reviewed in Stockdale et al., 2000; Noden and Francis-west, 2006), with the exception of a small population of neck muscles that have been shown to derive from the lateral plate mesoderm (Theis et al., 2010). Most of the knowledge about the paraxial mesodermal origin of skeletal muscles was established thanks to Di-I labeling (Selleck and Stern, 1991) and chick-quail graft experiments (Couly et al., 1992; Ordahl and Le Douarin, 1992). These lineage studies showed that although skeletal muscles share a common mesodermal origin, muscle organization significantly differs depending on their rostro-caudal position in the embryo.

Head muscles originate from cranial paraxial mesoderm. Cranial paraxial mesoderm lacks any initial signs of segmentation and mesodermal cells will only be segregated once they reach the branchial arches concomitantly with cranial neural crest cells (**Figure 3**; reviewed in Noden and Francis-west, 2006). Three distinct groups of cranial muscles can be distinguished: the extraocular muscles, originating from the prechordal mesoderm, the branchiomeric muscles including the muscles of the jaw, anterior neck and face, arising from the paraxial mesoderm and the tongue and posterior neck muscles, deriving from anterior somites (Noden, 1983; Couly et al., 1992; Trainor et al., 1994).

Truncal paraxial mesoderm caudal to the head emerges from already segmented embryonic structures, the somites, that will give rise to two main compartments all along the truncal axis of the embryo, the sclerotome and the dermomyotome (**Figure 3**, reviewed by Christ and Ordahl, 1995). Limb and axial skeletal muscles originate from the dermomyotome. The dorsomedial part of the dermomyotome gives rise to the epaxial musculature corresponding to the back and intercostal muscles, while the ventrolateral part of the dermomyotome gives rise to the hypaxial musculature corresponding to the diaphragm, abdominal and limb muscles (Ordahl and Le Douarin, 1992). Few muscles from the most posterior part of the head, including tongue muscles and muscles of the posterior pharyngeal arches also develop from the somites (Noden and Francis-west, 2006).

## Molecular Cascades That Regulate Muscle Development

Lineage progression to establish skeletal muscle from a founder mesodermal cell in the embryo is common to all skeletal muscles. An undifferentiated mesodermal cell (fate is not acquired) will switch to a muscle progenitor state (fate being acquired) to finally end up as a differentiated muscle cell (functional entity). Such switches from an undifferentiated state to a fully differentiated state are regulated by the activation of different groups of transcription factors (**Figure 4**). Head, trunk and limb muscle progenitors are specified by different genetic programs, but once specified, myogenic cells use a common differentiation program.

In the body, myogenic specification requires Pax3 and Pax7 genes, belonging to the paired-box Pax family. PAX3 controls the delamination of epaxial myogenic progenitor cells (reviewed in Tajbakhsh and Buckingham, 2000). Moreover, the central domain of the dermomyotome gives rise to a PAX3/PAX7 progenitor population forming subsequent axial muscles. In Pax3/Pax7 double-mutant mice, somitic cells do not enter the myogenic program, resulting in defective skeletal muscles (Kassar-Duchossoy et al., 2005; Relaix et al., 2005). The acquisition of a myogenic fate depends on a second group of transcription factors, named the basic Helix-Loop-Helix (bHLH) Myogenic Regulatory Factors (MRFs). MRFs have the ability to trigger skeletal muscle differentiation in non-muscle cells in vitro (Weintraub et al., 1991) and in vivo (Delfini and Duprez, 2004). Myod1 (MyoD), Myf5, and Myf6 (Mrf4) are considered as the muscle determination factors (Kassar-Duchossoy et al., 2004), while MyoG (Myogenin) is associated with muscle differentiation (Hasty et al., 1993). However, both Myod1 and Myf6 (Mrf4) are also required for terminal muscle

transcription factors. PAX3 and PAX7 label the progenitor state. The MYF5, MYF6, MYOD myogenic factors label the entry into the myogenic program, MYOG is characteristic of differentiated multinucleated muscle cells.

differentiation (reviewed in Buckingham, 2006). Myf5 and Myf6 (Mrf4) regulate the entry of progenitor cells into the myogenic program when they delaminate from the lips of the dermomyotome to form the myotome, but subsequent hypaxial activation of Myf5 is Pax3-dependent (Bajard et al., 2006). Early expression of Myod1 depends on Myf5, Myf6, and Pax3 as in the Myf5/Myf6/Pax3 triple mutants, Myod1 expression is altered and skeletal muscles do not form in the trunk and limbs (Tajbakhsh et al., 1997).

In vertebrates, the myogenic program of the head differs from the body musculature. While the expression of the myogenic regulatory factors Myf5, Myod1, and Myog in the craniofacial muscles is similar to what is observed in trunk/limb muscles (Hacker and Guthrie, 1998), the genetic hierarchies operating upstream of the myogenic genes are different for head muscles (branchiomeric and extraocular muscles). Pax3 is not expressed in head muscles, and Pax7 does not appear critical as head muscles form in the Pax7 mutant mice (Relaix et al., 2004). While Myf6 is not necessary for cranial myogenesis, other transcription factors among which Tcf21 (Capsulin), MyoR, Tbx1, and Pitx2 regulate the myogenic factors to form the different craniofacial muscles (Tzahor, 2009). Tbx1 and Pitx2 have been shown to activate Myod1 and Myf5 in the head and inactivation of Tbx1 and Pitx2 in mice causes severe reduction of specific groups of head muscles (Kelly et al., 2004; Zacharias et al., 2011). In mutant mice for Tcf21 and MyoR, myogenic genes are not activated in branchiomeric muscles, and cells undergo cell death (Lu et al., 2002).

Myogenic factors are crucial intrinsic actors for correct development of muscle, however numerous studies have shown that their initiation and regulation also depends on secreted factors coming from the adjacent tissues. The influence of neural tube, neural crest cells, notochord and ectoderm on the formation of muscles has been previously extensively studied and showed that Shh, BMP, Wnt, FGF, and Notch signaling pathways participate to both axial and limb myogenesis (reviewed in Deries and Thorsteinsdóttir, 2016).

## Connective Tissue-Mediated Muscle Morphogenesis

CTs and muscles are closely related during embryonic development and adult stages, suggesting that interactions between these tissues might be essential for their development. Classical experiments in avian embryos have demonstrated that signals involved in muscle differentiation and patterning partly derived from surrounding tissues (Lance-Jones, 1988; Ordahl and Le Douarin, 1992; Kardon, 1998). Over the last years, thanks to the identification of specific molecular markers for the different types of CT, progress has been made in the dissection of mechanisms underlying the interactions between CT and muscle development. These data have shown that depending on their embryological origin and their position throughout the body, mechanisms and signaling pathways coming from the diverse types of CT influence spatially and temporally muscle morphogenesis.

## Specialized Connective Tissue-mediated Myogenesis (Bone and Cartilage)

**Limb muscles** and specialized CTs (bone and cartilage) do not exhibit direct physical interactions, as they are linked together via tendons. During limb development, processes regulating skeleton and muscle formation can be dissociated (Hasson et al., 2010; Li et al., 2010). Indeed, disruption of skeletogenesis, through the mutation in the LIM-homeodomain transcription factor Lmx1b in skeletal progenitors using the Sox9-Cre, has no effect on muscle development (Li et al., 2010). Similarly, inactivation of the BMP antagonist, Noggin, which is expressed in condensing cartilage and immature chondrocytes, leads to profound skeletal defects without affecting the early stages of myogenic differentiation (Tylzanowski et al., 2006). However, despite the fact that skeleton and muscle formation can be dissociated, it has been evidenced that skeleton-derived signals are required for proper myogenesis. Indeed, although no defect at the onset of myogenesis is observed in Noggin null-mutant mice, terminal muscle differentiation is impaired (Tylzanowski et al., 2006; Costamagna et al., 2016). The Indian hedgehog (Ihh) secreted factor which belongs to the Hedgehog family is secreted by developing chondrocytes (Vortkamp et al., 1996). In the absence of Ihh, muscles are affected (Bren-Mattison et al., 2011). As for Noggin null-mutant, muscle impairment is restricted to secondary myogenesis, resulting in a decrease in the muscle masses. Finally, in vitro experiments show that C2C12 myoblasts can be converted toward osteogenic lineage when exposed to BMPs (Lee et al., 2000).

**Axial muscles** develop from the myotomal compartment of the somite, which is formed by the delamination of cells deriving first from the dorsomedial lip of the dermomyotome and then from its caudal and rostral lips. This process is partly controlled by another somatic compartment, the sclerotome. During chick embryonic development, pioneer myoblasts, constituting the medial part of epithelial somites, express the receptor ROBO2, while its ligand SLIT1 is expressed in the caudal domain of the nascent sclerotome (Halperin-Barlev and Kalcheim, 2011). Loss-of-function assays targeting either ROBO2 or SLIT1 lead to similar results: disruption of the caudo/rostral migration of pioneer myoblasts and of myofibre formation, demonstrating that skeletal precursor-derived signals (sclerotome) regulate the myotome morphogenesis (Halperin-Barlev and Kalcheim, 2011). However, since the sclerotome give rise to both skeleton and tendon progenitors (syndetome), these experiments cannot discriminate between the effects of bone or tendon progenitors on muscle morphogenesis.

Skeletal elements in the **head** derive from the cranial neural crest cells (Couly et al., 1992). Using a HoxA1/HoxB1 doubleknockout mouse, it was shown that cranial neural crest cells fail to form and migrate into the second branchial arch. Despite the absence of neural crest cells (at the origin of skeletal progenitors), cranial myogenesis was initiated (Rinon et al., 2007). However, muscle patterning defects were observed, as evidenced by the expansion in Tcf21 (Capsulin) and Tbx1 expression (Rinon et al., 2007). Similarly, ablation of cranial neural crest cells in the chick embryo shows that early steps of head myogenesis are not impaired by the removal of skeletal progenitors but that expression of myogenic genes is expanded to fill the entire arch mesenchyme, suggesting that the nature of the interactions between cranial skeleton and muscles are conserved in chick and mouse embryos (Tzahor et al., 2003; Rinon et al., 2007). Analysis of the molecular mechanisms demonstrate that BMP and Wnt signaling are important actors involved in these interactions (Tzahor et al., 2003; Rinon et al., 2007). However, cranial neural crest cells give rise to skeleton, tendon and CT progenitors. It is then difficult to determine in these experiments whether cranial myogenesis is controlled by interactions coming from prospective bone, tendon or muscle CT.

## Dense Regular Connective Tissue (Tendon) As an Important Source of Signals during Muscle Development

Muscle and dense regular CT (tendon) displays interactions during their development. It is well established that tendon requires muscle to fully develop in chick, mouse and zebrafish embryos (reviewed in Gaut and Duprez, 2016). However, the influence of tendon on muscle development is less clear in vertebrates. During limb muscle development, muscle masses differentiate between tendon primordia. In experimentally tendon-depleted region in chick embryo, ectopic muscles form at the place where tendons normally develop, indicating the role of tendon in delimitating regions of muscle growth and differentiation (Kardon, 1998). The role of tendon cells on muscle development has been studied more in Drosophila. Drosophila tendon precursor cells are defined as a group of ectodermal cells, named the apodeme and characterized by the expression of the Early growth response (EGR)-like transcription factor Stripe (Frommer et al., 1996). Altering apodeme formation during the early steps of leg development affects the localization of myoblasts (Soler et al., 2016). Establishment of the myotendinous junctions also requires correct migration of myogenic cells toward tendon cells. This migration step is mediated through guidance cues delivered by tendon cells. In tendons, Stripe positively regulates the expression of the Slit gene (Volohonsky et al., 2007), coding for a secreted protein implicated in guidance cues during axonal migration (Wong et al., 2002). Slit is expressed by tendon cells, while its receptor Robo (Roundabout) is expressed in muscle (Kramer et al., 2001). Interestingly, Slit mutants present defects in muscle patterning (Ordan et al., 2015), revealing tendon-signaling requirement for proper muscle development. Tendon and muscle interactions via Slit/Robo is necessary for the migration arrest of muscle progenitors in Drosophila (Wayburn and Volk, 2009). The formation of the myotendinous junction in Drosophila also requires the transmembrane protein KON-TIKI, enriched at the tips of myotubes, and necessary to direct their migration and the subsequent recognition between muscle and tendon cells (Schnorrer et al., 2007). These data indicate that tendon cells are required for muscle morphogenesis through specific signals emanating from tendon cells and acting on myogenic cells. However, these signals remain to be elucidated during development of the vertebrate musculoskeletal system. In zebrafish, Tsp4b (thrombospondin-4) appears critical to orchestrate tendon extracellular matrix assembly necessary for muscle attachment at the myotendinous junction (Subramanian and Schilling, 2014). Although it has been shown that the vertebrate orthologs of Stripe, Egr1/2 are involved in vertebrate tendon differentiation (Lejard et al., 2011; Guerquin et al., 2013), there is no obvious defect in muscle formation in the absence of Egr1. Inactivation of Tsp4 in mice shows that thrombospondin-4 controls the deposition of extracellular matrix in both tendon and muscle and is necessary for the correct organization of collagen fibrils in tendon (Frolova et al., 2014). However, the absence of Tps4 also directly affects skeletal muscle structure, by controlling the expression of heparan-sulfate proteoglycans in muscle (Frolova et al., 2014). Finally, tendons have been shown to be required in late events of vertebrate muscle morphogenesis. Indeed, the translocation of myofibers to form the final position of the flexor digitorum superficialis muscle in the mouse forelimb is largely impaired in Scx mutant, showing that tendon is implicated in the final patterning and position of muscles (Huang et al., 2013).

## Dense Irregular Connective Tissue Establish a Pre-pattern for Muscle Differentiation

Most of our knowledge concerning myogenesis regulation by signals produced by the irregular CT has been established in the limb. Each step of limb muscle development is tightly regulated by signals among which some are derived from the irregular CT. The different steps are the following. Somitic-PAX3-positive cells migrate toward the limb bud, invading the limb mesenchyme. Once they reached their target sites, PAX-3 positive cells proliferate and organize into dorsal and ventral muscle masses. Muscle differentiation is then initiated, followed by muscle mass growth and splitting (reviewed in Duprez, 2002; Deries and Thorsteinsdóttir, 2016).

### Delamination and Migration of Muscle Progenitors

Delamination and migration of muscle progenitor cells from the ventrolateral lip of the dermomyotome are mediated via the tyrosine kinase receptor c-Met and its ligand, the Scatter Factor/Hepatocyte Growth Factor (SF/HGF), (Brand-Saberi et al., 1996; Heymann et al., 1996; Dietrich et al., 1999). Cells from the ventrolateral lip of the somite express c-Met, while SF/HGF is released by irregular CT progenitors in the limb mesenchyme. In Hgf or c-Met mutant mice, limb muscles are missing (Bladt et al., 1995). Dermomyotome development proceeds normally and migratory somatic precursors are correctly specified but they remain aggregated and fail to migrate toward limb buds (Dietrich et al., 1999). SF/HGF also regulates the migration of myogenic progenitors from occipital and cervical somites, giving rise to the tongue, diaphragm and shoulder muscles (Dietrich et al., 1999). These studies highlight the link between irregular CT and hypaxial muscle progenitors during the migration step of muscle morphogenesis. Other signaling pathways expressed in irregular CT are involved in the guidance of muscle progenitors to reach their target sites into the limb bud. The CXCL12 chemokine is expressed in restricted regions of limb bud irregular CT and has been shown to attract muscle progenitors, which expressed the chemokine receptor CXCR4 (Vasyutina et al., 2005). Ectopic expression of CXCL12 in limb mesenchyme of chick embryos, or inactivation of Cxcr4 in mouse embryos both give rise to aberrant localization of muscle progenitors in the limb (Vasyutina et al., 2005), demonstrating a chemoattractive role of CXCL12 positive-CT cells for Cxcr4 expressing muscle precursors. During their migration toward the limb, muscle progenitors also express the receptor EPHA4, while its ligand EPHRINA5 is expressed in specific areas of limb irregular CT (Swartz et al., 2001). Conversely to the chemoattractive role of CXCL12/CXCR4 signaling, EPHRINA5 acts as a repulsive signal for muscle cells expressing EPHA4 (Swartz et al., 2001), demonstrating that both chemoattractive and repulsive signals from irregular CT act simultaneously on muscle progenitors to restrict and define their pathway of migration. Finally, it is important for muscle progenitor cells to stay in an undifferentiated state during migration. It is likely that this step is regulated through secreted signals produced by the limb mesenchyme, however it is not clear yet which signaling exactly is involved in this process. Previous studies suggest that BMPs and FGFs secreted by limb irregular CT might be important to prevent differentiation in migrating cells by respectively inhibiting and promoting the expression of SF/HGF (Heymann et al., 1996; Pourquié et al., 1996; Scaal et al., 1999). In the chick embryo, FGF18 and retinoic acid, secreted by limb mesenchyme, control the timing of Myod1 and Myf5 expression in myogenic cells (Mok et al., 2014).

### Muscle Differentiation and Patterning

During the whole processes of limb muscle morphogenesis, irregular CT and muscles (progenitors or differentiated cells) are in close association. Kardon et al. (2003) identified TCF4 as a putative actor in the process of irregular CT-mediated muscle morphogenesis. TCF4 is expressed in the lateral plate-derived mesoderm in close association with limb muscles during their differentiation and patterning. In the absence of limb muscles, TCF4 expression pattern is unchanged, suggesting that TCF4 expression may serve as a pre-pattern for limb musculature. To verify this hypothesis, TCF4 gain- and loss-of-functions were performed in lateral plate-derived limb mesodermal cells. In all cases, muscle mispatterning was observed, demonstrating that TCF4 in irregular CT is important to establish the correct location of limb muscles (Kardon et al., 2003). Tcf4 deletion in mice also lead to aponeurosis defects (Mathew et al., 2011). However, TCF4 is also expressed at low level in myogenic cells and is involved in the intrinsic regulation of muscle fiber type differentiation in mice (Mathew et al., 2011).

Recently, the role of irregular CT has also been involved in the context of a common and often lethal muscle diaphragm defect, called congenital diaphragmatic hernia (CDH). Merrell et al. (2015) have shown that the pleuroperitoneal folds, which are transient embryonic structures, give rise to the diaphragm irregular CT. Muscle progenitor cells arising from the ventrolateral dermomyotome of the cervical somites migrate into the Tcf4-positive pleuroperitoneal cells which guide muscle cells to organize the diaphragm morphogenesis. Tcf4-positive CT cells also express Gata4, known to be mutated in CDH, and Gata4 inactivation in diaphragm CT leads to hernias similar to those observed in CDH, demonstrating that this congenital muscular disease is related to a defect in muscle irregular CT (Merrell et al., 2015).

As previously mentioned, the human Holt-Oram syndrome is characterized by limb and heart musculoskeletal defects and irregular CT disorganization. This syndrome is due to a mutation in the TBX5 gene, which is expressed in irregular CT during limb development (Hasson et al., 2010). Tbx5 deletion leads to a defect in irregular CT organization during embryonic development (Hasson et al., 2010). In these conditions, while the early steps of limb myogenesis are not affected, ectopic splitting of nascent muscle bundles is observed. Tbx5 inactivation leads to a disruption of muscle irregular CT, to an alteration of Tcf4 expression, but also a marked decrease of ß-catenin and N-cadherin at the membranes of muscular irregular CT cells (Hasson et al., 2010). In addition, deletion of ß-catenin in the limb mesenchyme leads to ectopic muscle splitting consistent with a model in which the N-cadherin/ß-catenin complex in the muscle CT is critical for muscle patterning (Hasson et al., 2010). Finally, Tbx5 deletion also alters the expression of mesenchymal secreted factors important in limb myogenesis, as CXCL12 and SF/HGF (Hasson et al., 2010). It is noteworthy that in synovial fibroblasts, Cxcl12 is a target of Tbx5 in human synovial fibroblasts (Karouzakis et al., 2014). Recently, it has been shown that the conditional deletion of another T-box gene, Tbx3, in the lateral plate mesoderm (using a Prx1-Cre transgene) leads to defects in myofiber formation in a subset of limb muscles in mice (Colasanto et al., 2016). These localized muscle defects are correlated with Tbx3 expression in a subset of limb bones, tendons and muscle CT. Similar muscle defects are observed in patients with TBX3 mutations that are responsible of the Ulnar-mammary syndrome (Colasanto et al., 2016). In addition to being expressed in limb skeletal elements, Hoxa11 gene is also expressed in mouse muscle CT and Hoxa11 inactivation disrupts limb muscle and tendon patterning in addition to the already known skeleton defect (Swinehart et al., 2013). Tendon and muscle phenotypes in heterozygous Hoxa11 mutants are independent of skeletal patterning as abnormal tendon and muscle patterning are observed in Hoxa11 mutants with normal skeleton (Swinehart et al., 2013). However, it cannot be excluded that, in this case, muscle mispatterning could be related to tendon abnormalities rather than to the muscle CT defect. Recently, Gu et al. (2016) have shown that in neonatal muscles, muscle interstitial cells activate NF-kB, which regulates EPHRINA5 to stimulate myoblast migration toward the end of growing fibers, where they subsequently fuse to contribute to muscle growth. These data show that muscle CT also contributes to the process of muscle maturation during neonatal development. However, these interstitial cells are characterized by the expression of NG2, a neural/glial antigen 2 expressed in pericytes and it cannot be excluded that these cells are of vascular origin (Gu et al., 2016).

Finally, differentiated muscle fibers also act on muscle CT formation. In mice deleted for Lox (Lysyl-oxidase), an enzyme regulating collagen organization and secreted from the myofibers, TGFβ signaling is decreased and promotes muscle CT differentiation at the expense of muscle tissue (Kutchuk et al., 2015).

## Connective Tissue Cell Involvement in Adult Muscle Homeostasis

In adult, skeletal muscle loss is observed in neuromuscular diseases, but also during aging, inactivity and chronic systemic disorders (i.e., diabetes, cancer, rheumatoid arthritis). The regenerative potential of skeletal muscle provides a compensatory response against such pathological muscle loss. The regenerative capacity of skeletal muscle relies on muscle stem cells (named satellite cells), which proliferate in response to exercise to facilitate muscle growth and remodeling, or following myotrauma to repair the injured muscle. Satellite cells are PAX3/7-positive progenitor cells located under the basal lamina that forms around muscle fibers of postnatal skeletal muscle. Satellite cells remain quiescent until the muscle is injured, when the lamina breaks down and activated satellite cells begin to proliferate before forming new muscle fibers (Relaix et al., 2005). Myf5 is detected in the majority of quiescent satellite cells (Cornelison and Wold, 1997; Beauchamp et al., 2000) and activation of satellite cells is accompanied by expression of Myod1 as well as higher levels of Myf5, leading to the downregulation of Pax7, activation of Myogenin, and new muscle fiber formation (Relaix et al., 2006, reviewed in Motohashi and Asakura, 2014). In the absence of Pax7-positive cells, the process of muscle regeneration failed and instead, fibrotic and fatty infiltration are observed, demonstrating the major contribution of muscle satellite cells in the formation of new muscle fibers (von Maltzahn et al., 2013). However, in response to muscle damage, non-myogenic cells can also participate to skeletal muscle regeneration, either by giving rise to myogenic stem cells or by stimulating the activation of resident muscle satellite cells.

A non-satellite cell population with myogenic capacity was first identified when it has been shown that bone-marrow-derived cells can participate directly to muscle regeneration (Ferrari et al., 1998). These cells, which normally reconstitute the hematopoietic lineage, can give rise to new satellite cells and myofibers during the muscle regeneration process (Asakura, 2012) and their transplantation into mdx mice (a model for Duchenne muscular dystrophy) improves muscle function (Sampaolesi et al., 2006). Similarly, a vascular progenitor population, which can be isolated from postnatal muscle, participate in muscle repair following arterial delivery in mice (Sampaolesi et al., 2003). Interestingly, pre-treatment of both mesenchymal bonemarrow stromal cells (Galvez et al., 2006) or vascular progenitors (Brzoska et al., 2012) with the CXCL12 chemokine improved the regeneration of injured muscle. CXCL12 is expressed in the adult muscle by the endomysium, i.e., the CT surrounding each muscle fiber (Hunger et al., 2012). Following muscle injury, CXCL12 secreted by muscle CT rapidly increases (Griffin et al., 2010) and chemoattracts both satellite cells and bone-marrowderived cells to actively participate to the regeneration process (Ratajczak et al., 2003). In this context, CXCL12 would not only chemoattract stem cells toward the injury site, but would also increase their fusion with native muscle fibers (Griffin et al., 2010). These results demonstrate that signals provided by muscle irregular CT are not only crucial for muscle morphogenesis during development but also mediate the processes of muscle regeneration in the adult.

More recently, a population of interstitial muscle cells with myogenic potential has been identified (Mitchell et al., 2010). These cells, characterized by the expression of the PW1/Peg3 gene and named PICs (PW1-positive interstitial cells) contribute to the satellite cell pool during muscle regeneration (Mitchell et al., 2010), although they do not express Pax3 or Pax7 (Pannérec et al., 2013). PICs can be subdivided into two distinct populations: PW1+ PDGFrα− and PW1+ PDGFrα+ cells. It has been establish that only PW1+ PDGFrα− PICs are associated with a myogenic potential while PW1+ PDGFrα+ cells give rise to adipocytes (Pannérec et al., 2013). Interestingly, PW1+ PDGFrα+ PICs express the pericyte marker NG2, indicating a possible overlap between these cells, and pericytes (Pannérec et al., 2013). Pericytes represent perivascular cells that are present in the muscle interstitium and associated with capillaries. They can be separated into two different populations: type-1 pericytes (NG2+ NESTIN− PDGFrα−) and type-2 pericytes (NG2+ NESTIN+ PDGFrα+), (Birbrair et al., 2013). Similarly to what has been described for PICs, the two different populations of pericytes have distinct cell fate potential: type-1 contribute to adipose tissue and type-2 to myogenesis (reviewed in Birbrair et al., 2014). Type-2 pericytes do not express Pax7, Myf5 and Myod1, but upregulate these markers before forming myotubes in regenerative conditions (Cappellari and Cossu, 2013).

Different studies also reported the participation of mesenchymal progenitors without myogenic capacity during muscle regeneration. These progenitors all arise from resident cells in the adult muscle interstitium (Joe et al., 2010; Uezumi et al., 2010). Based on the expression of PDGFRα, a cell population resident in the muscle interstitium has been isolated, which, under specific culture conditions, differentiate into fibroblasts, adipocytes or osteoblasts, but never give rise to muscle cells and has been named mesenchymal progenitors (Uezumi et al., 2010). Simultaneously, Rossi's group also identified a cell population with fibroblastic and adipogenic potential, but no myogenic potential (Joe et al., 2010). These progenitors were isolated on the basis of SCA1 and CD34 expression, and termed Fibro/Adipogenic progenitors (FAPs), (Joe et al., 2010). Interestingly, mesenchymal PDGFRα+ progenitors express SCA1 (Joe et al., 2010, Uezumi et al., 2010) and FAPs express PDGFRα, highlighting the possibility that mesenchymal progenitors and FAPs actually represent a unique progenitor population. FAPs/mesenchymal progenitors are activated upon muscle injury and promote myoblast differentiation in co-cultures (Joe et al., 2010), but also exhibit a strong adipogenic and fibrogenic potential in vitro, indicating a potential contribution of FAPs to fibrotic and adipose accumulation in diseased muscles (Uezumi et al., 2010). It is then proposed that a balance between satellite cell-dependent myogenesis and FAPs-dependent adipogenesis/fibrogenesis regulates muscle homeostasis and regeneration. After muscle injury, FAPs/mesenchymal progenitors start to proliferate before satellite cells and invade the space between regenerating muscle fibers, where they

generate factors promoting myogenesis. When regeneration proceeds efficiently, FAPs/mesenchymal progenitors are discarded from the tissue through apoptotic signals emanating from satellite cells. If regeneration fails, FAPs/mesenchymal progenitors persist and differentiate into adipocytes and fibroblasts, leading to fatty and fibrotic degeneration (reviewed in Natarajan et al., 2010; Judson et al., 2013). Depending on the surrounding environment, FAPs/mesenchymal progenitors will preferentially give rise to fibroblasts or adipocytes. Addition of TGFß to FAPs/mesenchymal progenitors in vitro induces the expression of fibrosis markers leading to fibroblastic differentiation at the expense of adipocyte differentiation (Uezumi et al., 2011). Interestingly, PDGFRα+ expressing FAPs/mesenchymal progenitors accumulate preferentially into fibrotic regions, suggesting a specific role for PDGFRα in muscle fibrosis (Uezumi et al., 2014). This hypothesis is supported by the observation that, in adult and embryonic mouse, an elevated level of PDGFRα leads to an abnormal increase in CT differentiation (Olson and Soriano, 2009).

The participation of irregular CT to muscle regeneration has been also highlighted by a recent set of experiments. CT fibroblasts identified by Tcf4 expression have been shown to proliferate close to muscle satellite cells following injury and conditional ablation of Tcf4-positive cells prior to muscle lesion leads to premature satellite cell differentiation, depletion of the early pool of satellite cells, and small regenerated fibers, indicating that Tcf4-positive fibroblasts participate in muscle regeneration (Murphy et al., 2011). It remains unclear whether a direct relationship exists between FAPs/mesenchymal progenitors and TCF4-positive cells. However, Tcf4-positive cells express PDGFRα (Murphy et al., 2011) and accumulating evidence suggests that FAPs/mesenchymal and irregular CT progenitors share common features (Sudo et al., 2007; Haniffa et al., 2009). Extracellular matrix components also contribute directly to the regenerative potential of muscle. Indeed, it has been shown that a fibronectin-rich fibrosis is essential during the initial step of regeneration to activate the proliferation of muscle satellite cells (Bentzinger et al., 2013). Irregular CT progenitors, FAPs and PICs could be potential sources of fibronectin and might contribute to the transient fibronectinrich promyogenic fibrosis during muscle regeneration. However, activated satellite myogenic cells themselves release fibronectin into their microenvironment and inactivation of fibronectin using a Myf5-Cre reporter impairs the regenerative potential of muscle, suggesting that this effect could be also related to a cellautonomous role of satellite cell derived-fibronectin (Bentzinger et al., 2013).

The importance of muscle CT has been also evidenced in muscle disorders. Indeed, mutations in COL6A1, COL6A2, and COL6A3 genes, which give rise to the main collagens expressed in muscle CT, have been observed in congenital Ullrich muscular dystrophy and in Behlem myopathy. Mutant mice for Col6a1 display alterations of muscle sarcoplasmic reticulum and mitochondria (Pan et al., 2014) and Col6a3 mutant mice display myopathic and connective tissue phenotypes similar to the Col6a1 null mice (Pan et al., 2013), demonstrating that collagen VI mutations result in disorders with combined muscle and connective tissue involvement. In addition, Col6a1 mutant mice showed delayed muscle regeneration and reduced satellite cell self-renewal. Transplantation of wild-type fibroblasts in muscles of Col6a1 mutant mice rescues muscle satellite cells, indicating that COL6A1 in the muscle environment can modulate satellite cell behavior (Urciuolo et al., 2013).

Finally, during muscle hypertrophic activity, satellite cells can regulate fibrogenic cell collagen expression via exosome secretion, showing that muscle cell progenitors can also act with their surrounding environment to facilitate tissue plasticity (Fry et al., 2016). Similarly, Abou-Khalil et al. (2015) have shown that Pax7-positive muscle satellite cells are involved in bone repair, providing a direct evidence of a muscle contribution to specialized CT (bone and cartilage) formation.

Taken together, these data evidence interactions between different cell populations promoting muscle progenitor activation during regeneration, with a central role of muscle irregular CT in this process. Changes in CT local environment may contribute to muscle pathologies and age-related loss of muscle stem cell competence by implicating pivotal signaling pathways and genes similar to those described to mediate the CT-dependent muscle morphogenesis during development.

## CONCLUSIONS

The development of skeletal muscle has been extensively studied for decades and most of the studies have first concentrated on the elucidation of the intrinsic mechanisms underlying the conversion of muscle progenitors toward a functional skeletal muscle organ. The identification of specific myogenic transcription factors has allowed us to decipher the importance of these intrinsic gene networks in the specification and differentiation of muscles during embryonic development. In parallel, the role of neighboring tissues on muscle morphogenesis has been investigated and highlighted the influence of the neural tube, notochord and ectoderm on the early steps of axial muscle morphogenesis, mostly via the effect of the secreted factors Wnts, BMPs, and SHH. More recently, the role of CTs in muscle morphogenesis has been investigated, thanks to the identification of transcription factors specifically expressed in the different types of CT surrounding (tendon) or composing (muscle CT) the developing muscle (previously reviewed in Hasson, 2011). These studies demonstrate that mesenchymal cells at the origin of the different CT types deliver information necessary for a correct muscle morphogenesis, from the early steps of myoblast migration and fusion to the late stages of muscle maturation. Secreted factors as BMPs, FGFs, and chemokines as CXCL12 are important in this dialogue between CTs and muscles, which also implicate a reverse interaction between both tissues, as muscle cells are necessary for the tendons to develop correctly and for the organization of irregular CT and bone in adult. A right balance between myogenic and CT cells is particular necessary during the muscle regeneration process. Indeed, impairment of the regenerative potential after injury or in neuromuscular diseases results in the progressive replacement of the muscle mass by fibrotic tissue (Farup et al., 2015). Thus, bi-directional communication between muscle and CT is critical for a correct assembly of the musculoskeletal system during development as well as to maintain its homeostasis in the adult.

## AUTHOR CONTRIBUTIONS

SN, DD, and CF have equally contributed to making the original plan and writing the manuscript.

## FUNDING

This work was supported by the Fondation pour la Recherche Médicale (FRM) DEQ20140329500, Agence Nationale de la

## REFERENCES


Recherche (ANR) ANR-12-BSV1-0038, Association Française contre les Myopathies (AFM) N◦ 16826. SN was part of the MyoGrad International Research Training Group for Myology and received financial support from the AFM (AFM 201505 32272).

## ACKNOWLEDGMENTS

We thank S. Gournet for illustrations.


Zhang, G., Young, B. B., Ezura, Y., Favata, M., Soslowsky, L. J., Chakravarti, S., et al. (2005). Development of tendon structure and function: regulation of collagen fibrillogenesis. J. Musculoskelet. Neuronal Interact. 5, 5–21.

Zhao, Q., Eberspaecher, H., Lefebvre, V., and De Crombrugghe, B. (1997). Parallel expression of Sox9 and Col2a1 in cells undergoing chondrogenesis. Dev. Dyn. 209, 377–386. doi: 10.1002/(SICI)1097-0177(199708)209:4<377::AID-AJA5> 3.0.CO;2-F

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

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

# Pitx2 in Embryonic and Adult Myogenesis

Francisco Hernandez-Torres 1, 2, Lara Rodríguez-Outeiriño1, 2, Diego Franco1, 2 and Amelia E. Aranega1, 2 \*

<sup>1</sup> Cardiac and Skeletal Myogenesis Group, Departmento de Biología Experimental, Universidad de Jaén, Jaén, Spain, <sup>2</sup> Cardiac and Skeletal Myogenesis Group, Fundación MEDINA Centro de Excelencia en Investigación de Medicamentos Innovadores en Andalucía, Granada, Spain

Skeletal muscle is a heterogeneous tissue that represents between 30 and 38% of the human body mass and has important functions in the organism, such as maintaining posture, locomotor impulse, or pulmonary ventilation. The genesis of skeletal muscle during embryonic development is a process controlled by an elaborate regulatory network combining the interplay of extrinsic and intrinsic regulatory mechanisms that transform myogenic precursor cells into functional muscle fibers through a finely tuned differentiation program. However, the capacity of generating muscle still remains once these fibers have matured. Adult myogenesis resembles many of the embryonic morphogenetic episodes and depends on the activation of satellite cells that have the potential to differentiate into new muscle fibers. Pitx2 is a member of the bicoid family of homeodomain transcription factors that play an important role in morphogenesis. In the last decade, Pitx2 has emerged as a key element involved in the fine-tuning mechanism that regulates skeletal-muscle development as well as the differentiation and cell fate of satellite cells in adult muscle. Here we present an integrative view of all aspects of embryonic and adult myogenesis in which Pitx2 is involved, from embryonic development to satellite-cell proliferation, fate specification, and differentiation. Those new Pitx2 functions on satellite-cell biology might open new perspectives to develop therapeutic strategies for muscular disorders.

Keywords: Pitx2, myogenic precursor cells, embryonic myogenesis, adult myogenesis, satellite cell and regeneration

## INTRODUCTION

Skeletal muscle is a heterogeneous tissue that represents between 30 and 38% of the human body mass (Janssen et al., 2000). It is composed of individual muscle fibers, diversified in size, shape, and contractile protein content, to fulfill the different functional needs of the vertebrate body such as maintaining body posture, locomotor impulse, or pulmonary ventilation. The genesis of skeletal muscle during embryonic development and postnatal life is a process controlled by an extremely elaborate regulatory network that combines the interplay of extrinsic (e.g., morphogens, neurohormonal input, muscle damage, etc.) and intrinsic elements (gene regulatory elements). The intrinsic elements form hierarchical interactions between transcriptional regulators, regulatory RNAs, and chromatin-remodeling factors. In this sense, during embryogenesis, muscle progenitors are specified by the sequential expression of a network of transcription factors composed of PAX3 and PAX7, and the basic helix-loop-helix (bHLH) myogenic regulatory factors (MRFs) MYOD,

#### Edited by:

Juan Jose Sanz-Ezquerro, Consejo Superior de Investigaciones Científicas, Spain

#### Reviewed by:

Solveig Thorsteinsdottir, Universidade de Lisboa, Portugal Eusebio Perdiguero, Pompeu Fabra University, Spain Thimios Mitsiadis, University of Zurich, Switzerland

> \*Correspondence: Amelia E. Aranega aaranega@ujaen.es

#### Specialty section:

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

> Received: 31 January 2017 Accepted: 12 April 2017 Published: 01 May 2017

#### Citation:

Hernandez-Torres F, Rodríguez-Outeiriño L, Franco D and Aranega AE (2017) Pitx2 in Embryonic and Adult Myogenesis. Front. Cell Dev. Biol. 5:46. doi: 10.3389/fcell.2017.00046 MYF5, MYF6 (also called MFR4), and MYOG (Bentzinger et al., 2012; Moncaut et al., 2013). In addition, during adult life the skeletal muscle has the ability to resume developmental mechanisms that compensate for the physiological turnover and damage in order to maintain tissue homeostasis (Schmalbruch and Lewis, 2000; Pellettieri and Alvarado, 2007). This adult myogenesis depends on the activation of satellite cells (SCs), that have the potential to proliferate, differentiate, and generate new fibers, or repair existing ones (Chargé and Rudnicki, 2004). It has been well-established that SCs are closely related to progenitors of embryonic origin (Gros et al., 2005; Relaix et al., 2005; Schienda et al., 2006; Hutcheson et al., 2009; Lepper and Fan, 2010). Thus, many similarities have been discovered between prenatal myogenesis and regeneration in the mature skeletal musculature, such as common transcription factors and signaling molecules (Tajbakhsh, 2009).

During the last two decades the homeobox transcription factor Pitx2 has emerged as a key element in the finetuning mechanism that regulates skeletal-muscle development. Concurrently, several recent experimental pieces of evidence point to the role of Pitx2 in SC biology. Here, we present an integrative view of the role of Pitx2 in prenatal and adult myogenesis (from embryonic development to SC proliferation), fate specification, and differentiation. Finally we discuss the potential therapeutic use of Pitx2 in the future.

## PRENATAL AND ADULT MYOGENESIS

In vertebrates, skeletal-muscle development is a biphasic process. A primary (embryonic) myogenesis takes place to generate primary muscle fibers, between embryonic day (E) 9.5 and E14.5 in the mouse. This is followed during fetal stages by a secondary myogenesis which gives rise to the bulk of skeletalmuscle fibers present at birth (Kelly and Zacks, 1969; Biressi et al., 2007; Tajbakhsh, 2009; Deries and Thorsteinsdóttir, 2016). All skeletal-muscle cells have the same underlying functions, although their progenitors within the paraxial mesoderm are spread throughout the embryo at the onset of myogenesis. This bears emphasizing since the genetic networks that control myogenesis present differences depending on the location of those myogenic precursors in the embryo.

## Embryonic Myogenesis: The Trunk and Limb Muscles

The muscles of the trunk and limbs derive from somites (**Figure 1A**), which are transient paraxial mesodermal structures that form pairwise on either side of the neural tube, following an anterior-posterior developmental gradient. The somite is initially a spherical unit of polarized epitheloid cells that soon after subdivides into two compartments, the ventral mesenchymal sclerotome and the dorsal epithelial dermomyotome. Shortly afterwards, myogenic precursor cells from the epaxial and hypaxial lips of the dermomyotome undergo an epithelialmesenchymal transition (EMT) and accumulate underneath, where they differentiate and elongate to form the myocytes of the myotome, the first myogenic structure to develop in the body (Buckingham and Relaix, 2015; Deries and Thorsteinsdóttir, 2016). The epaxial region of the myotome gives rise to the deep back muscles, whereas the hypaxial myotome is the source of body wall muscles and most other trunk muscles (Buckingham and Relaix, 2015; Deries and Thorsteinsdóttir, 2016). In segments adjacent to the limb-region cells of the hypaxial dermomyotome undergo an EMT, leave the epithelial structure, and migrate toward the fore and hind limbs to form dorsal and ventral muscle masses in the limb-bud mesenchyme, where they begin to differentiate and express muscle-specific genes (Biressi et al., 2007; Deries and Thorsteinsdóttir, 2016).

Cell commitment in the somite is highly dependent on a number of transcription factors which act in a hierarchical molecular cascade to orchestrate the specification, determination, and differentiation of myogenic precursors. In the genetic hierarchy that regulates the onset of trunk myogenesis, Pax3 and Myf5 play a dominant role (Buckingham and Relaix, 2015). Pax3 is already transcribed in pre-somitic mesoderm adjacent to the first somite (Schubert et al., 2001) and then throughout the newly formed somites. As somites mature Pax3 expression becomes confined to the dermomyotome (Goulding et al., 1991) and persists in myogenic progenitor cells that delaminate and migrate from the somite to more distant sites of myogenesis such as the limb (Buckingham and Relaix, 2015). Myogenic cells that have activated the myogenic determination genes Myf5/Myf6 and MyoD downregulate Pax3 and delaminate from the edges of the dermomyotome (Buckingham and Relaix, 2015). The epaxial myotome then start to form. This depends on the early epaxial activation of Myf5, which is driven by Wnt and Shh signaling, without any Pax3 and/or Pax7 requirement (Borello et al., 2006; Buckingham and Relaix, 2015). These cells do not activate MyoD but rather Myog and differentiate (Kablar et al., 2003). On the other hand, the activation of Myf5 in the hypaxial somite as well as in the limb depends on PAX3 (Bajard et al., 2006; Buckingham and Relaix, 2015). At this stage MYF6 also acts as a myogenic determination factor (Kassar-Duchossoy et al., 2004). The Myod gene is activated after the onset of Myf5 expression in the rest of the dermomyotome and limbs (Hu et al., 2008). Finally, the transcription factor MYOG is required for the onset of the expression of terminal differentiation genes needed for the fusion of myocytes and the formation of myotubes (Bentzinger et al., 2012).

## Embryonic Myogenesis: The Head Muscles

Although, all skeletal muscle throughout the body originates within paraxial mesoderm, in the head, identifiable compartments such as the somites in the trunk are not evident histologically or by most molecular criteria. This unsegmented head mesoderm is remodeled at the early stages of embryonic development (**Figures 1B,C**). The unsegmented head mesoderm gives rise to all craniofacial skeletal muscles, which can be cataloged as four distinct populations: extra-ocular (EOMs), branchial, laryngoglossal, and axial neck muscles (Noden and Francis-West, 2006; Tzahor, 2015). EOMs are formed by cells from the cranial paraxial mesoderm that migrate through the first branchial arch (FBA) as well as from the prechordal mesoderm (Jacob et al., 1984; Evans and Noden, 2006; Tzahor,

FIGURE 1 | Embryonic myogenesis (A) Schematic representation of somite maturation. Somites mature following an anterior to posterior developmental gradient (Modified from Gray's Anatomy. The Anatomical Basis of Clinical Practice, 40th Edition Standring, 2008): myogenic precursor cells arise from the epaxial and hypaxial lips of the dermomyotome after archive epithelial-mesenchymal transition (EMT) and migrate toward the limbs to form dorsal and ventral muscle masses where they begin to differentiate. (B,C) Head frontal and transverse planes of a mouse embryo between stages of development E7.5–8.75 and E8–9.25 in mouse. At an open neural plate stage, head mesoderm in a frontal plane includes the prechordal mesoderm and the paraxial mesoderm. When the neural tube closes dorsally and the endoderm ventrally, the prechordal mesoderm is integrated within the remaining paraxial mesoderm, which is located anterior to the somites. Dashed line illustrates the cutting plane. (D) Origins of skeletal muscles: Myogenic precursors arise from different paraxial mesoderm compartments. (E) Pitx2 expression domains at the E10.5 stage of development in mouse. NT, neural tube; NC, notochord; SM, somites; DMT, dermomyotome; ST, sclerotome; MT, myotome; LMP, limb muscle precursors; FL, forelimb; PAM, head paraxial mesoderm; PCM, prechordal mesoderm; PROS, prosencephalon; MES, mesencephalon; MET, metencephalon; SPM, splanchnic mesoderm; OFT, outflow tract of heart; HT, heart tube; EOM, extra-ocular muscles; BRM, branchial muscles; LGM, laryngoglossal muscles; HGC, hypoglossal cord; ANM, axial neck muscles; BM, back muscles; BWM, body wall muscles; FLM, forelimbs muscles; HLM, hind limbs muscles.

2015; **Figure 1D**). Branchial arch muscles are formed mainly by migrating cells from the cranial paraxial mesoderm and the lateral splanchnic mesoderm (Harel et al., 2009; Sambasivan et al., 2009; Tzahor, 2015). Laryngoglossal muscles develop from migratory myoblasts arising from occipital somites that form a condensed mesenchymal band, the hypoglossal cord, which elongates and similarly brings myoblasts ventral to pharynx (Hammond, 1965; Hazelton, 1970; Tzahor, 2015). Finally, in the transition zone between the head and the trunk are the axial neck muscles. They arise from medio-dorsal and latero-ventral domains of occipital and cervical somites (Noden, 1983; Couly et al., 1992; Matsuoka et al., 2005).

The genetic hierarchy governing primary myogenesis in the trunk does not appear to operate for head-muscle formation. Activation of the myogenic program in the head therefore depends on different upstream factors, responds differently to signaling pathways and also displays site-dependent regulation. Branchial-arch-derived muscles depend on Myf5/Myf6/Myod, whereas extra-ocular muscle formation is initiated by Myf5/Myf6 and in their absence cannot be restored by Myod (Tajbakhsh and Buckingham, 1999).

## Fetal Myogenesis

During fetal myogenesis, secondary fibers in trunk, limbs, and head are generated by the fusion of fetal myoblasts. Secondary fibers form initially at the site of innervation of the primary fiber and are surrounded by the same basal lamina as the primary fiber on which they lie (Duxson et al., 1989). The secondary myotubes remain attached for a short period to primary fibers and subsequently elongate and become independent fibers, which can be distinguished from primary fibers by their relative small size (Kelly and Zacks, 1969). Although, the genetic networks that rule this second stage of prenatal myogenesis is less understood, it is known that the MRFs MYF5, MYOD, and MYOG are also crucial, since in Myog−/<sup>−</sup> as well as Myf5−/−:MyoD−/<sup>−</sup> doublemutant secondary myogenesis is completely inhibited (Venuti et al., 1995; Kassar-Duchossoy et al., 2004).

## Adult Myogenesis

The regulatory inputs that orchestrate myogenesis during prenatal myogenesis are partially reactivated in adult muscle repair. In adulthood, the maintenance as well as the repair of muscle tissue are both directed mainly by SCs. These cells, originally identified via electron microscopy in 1961 by Alexander Mauro, are located underneath the basal lamina and adjacent to the plasma membrane of the skeletal-muscle myofiber (Mauro, 1961; **Figure 2**). In their quiescent state, SC express the transcription factor Pax7 and represent a genuine stem-cell population indispensable for skeletal-muscle repair (Lepper et al., 2011; Murphy et al., 2011; Sambasivan et al., 2011; Miersch et al.,

2017; Stuelsatz et al., 2017). It has been established that SCs in adult muscle represent a lineage continuum of the embryonic myogenic progenitor cells. Thus, while SCs of the body and limbs arise from somites, in common with the muscle that they are associated with (Armand et al., 1983; Gros et al., 2005; Relaix et al., 2005; Schienda et al., 2006), the SCs located in head muscles also originate from the cranial mesoderm (Harel et al., 2009). Within a context of physiologicalstimuli (physical exercise or pathological conditions) SCs become activated, proliferate, differentiate and fuse to form multinucleated myofibers in order to undergo proper myogenesis (Lepper et al., 2011; Murphy et al., 2011; Sambasivan et al., 2011; Miersch et al., 2017; Stuelsatz et al., 2017; **Figure 2**). In this regard, numerous studies have revealed a striking similarity between adult and embryonic myogenesis, where the core regulatory network composed of the MRFs MYF5, MYOD, MYOG, and MYF6 is mainly required (Bentzinger et al., 2012; Segalés et al., 2016; **Figure 2**).

## THE PITX2 GENE

The Pitx gene family includes three vertebrate paralogues, Pitx1, Pitx2, and Pitx3, which have been cloned in multiple organisms (Gage et al., 1999b; Knopp et al., 2013). These three genes encode transcription factors that belong to the bicoid-related subclass of homeodomain proteins (Gage et al., 1999b) The members of this family share an almost identical protein sequence within their homeodomains, varying mainly in the N-terminal region (Gage et al., 1999b; Knopp et al., 2013). Mutations or misregulation of Pitx1, Pitx2, and Pitx3 result in developmental disorders in humans, such as Facioscapulohumeral Muscular Dystrophy (FSHD; Dixit et al., 2007), Axenfeld-Rieger syndrome (Semina et al., 1996), and Anterior Segment Mesenchymal Dysgenesis (ASMD; Semina et al., 1998), respectively. Muscle expression of these genes during development has been systematically studied. Thus, Pitx1 is highly expressed in developing hind-limb-bud mesenchyme and is shown to determine hind-limb identity in mice (Lanctôt et al., 1999; Szeto et al., 1999), chicks (Logan and Tabin, 1999), and fish (Shapiro et al., 2004). On the other hand, Pitx3 is widely expressed in all skeletal muscles of the head, trunk and limbs (Semina et al., 1998; L'honoré et al., 2007). Curiously, despite its apparent importance in muscle development, the investigation of Pitx3−/<sup>−</sup> mice indicates that Pitx3 on its own is not required for myogenesis (L'honoré et al., 2007). In this scenario Pitx2, the third Pitx family member is strongly upregulated and appears to fully compensate for the loss of Pitx3 during muscle formation (L'honoré et al., 2007). Pitx2 is also able to control the growth ability of hind-limb mesenchyme together with Pitx1 (Marcil et al., 2003), indicating the importance of Pitx2 in the control of skeletal myogenesis during development.

In mice, the Pitx2 (Pituitary homeobox 2 or Paired-like homeodomain transcription factor 2) gene is mapped on chromosome 3 (3G3; 3 57.84 cM) (Gage and Camper, 1997) and is transcribed into three distinct isoforms: Pitx2a, Pitx2b, and Pitx2c. Pitx2a and Pitx2b share the same promoter while Pitx2c uses an alternative one upstream of exon 4 (Schweickert et al., 2000). In human, PITX2 is mapped on chromosome 4 (4q25) and maintains a similar genetic structure, but presents a fourth isoform (Arakawa et al., 1998; Cox et al., 2002). This fourth isoform is generated by the PITX2C alternative promoter and differential splicing, being able to suppress the transcriptional activity of the other PITX2 isoforms (Cox et al., 2002). All Pitx2 isoforms share a K50 DNA-binding homeodomain which binds to the consensus sequence TAATCC (Amendt et al., 1998; Chaney et al., 2005), thus being able to induce a transcriptional activation of Prl (Amendt et al., 1998) or Anf (Ganga et al., 2003) promoters. The Pitx2 gene was isolated independently by several research groups and designated as Otlx2 (Muccielli et al., 1996), Rieg (Semina et al., 1996), Ptx2 (Gage and Camper, 1997), Brx1 (Kitamura et al., 1997), and Arp1 (Arakawa et al., 1998). Although, most of these works focused on the role of this gene in the development of brain structures, the authors reported the expression of Pitx2 in the mesenchyme of the eye, the first and second branchial arches, the fore and hind limbs as well as the dermomyotome at somite stages E8.5 and 10.5 in mouse, and its equivalent stages in chicken (**Figure 1E**). Soon afterwards, a role for Pitx2 was also described in left-right asymmetry, being proposed as the molecular transducer of embryonic left-right signaling during early developmental stages at the level of organs such as heart, gut, and/or stomach (Logan et al., 1998; Ryan et al., 1998; Yoshioka et al., 1998; Campione et al., 1999).

## PITX2 WITHIN THE GENETIC HIERARCHIES THAT CONTROL MUSCLE DEVELOPMENT

## Pitx2 Function during Embryonic Myogenesis

### Pitx2 in Trunk and Limb Muscle Development

The first evidence involving Pitx2 in the molecular process controlling myogenesis was provided by Kitamura et al. (1999). These authors reported Pitx2 expression co-localizing in dermomyotomes with Pax3, a muscle specification marker playing a key role in delamination and migration of the somitic muscle progenitor cells to the limb buds (Goulding et al., 1994; Tajbakhsh et al., 1997). Later, Marcil et al. demonstrated the presence of PITX2 protein in the myoblasts of the limb bud, displaying an expression pattern similar to that of PAX3 and MYOD (Marcil et al., 2003). All these data suggested that PITX2 was involved in muscle patterning. A more detailed temporal and spatial analysis during initial muscle-cell-cluster formation, by using lacZ expression from a Pitx2 gene insertion, revealed the presence of a Pitx2-expressing cell cluster lateral to the dermomyotome (Shih et al., 2007b). This cluster first appeared at the forelimb level at E10.25. After E10.5, Pitx2(+/LacZ) expressing cells were then also detected on sections of the limbs. Curiously, Pax3 and the muscle-regulatory factors (MRFs) stained only subsets of Pitx2<sup>+</sup> cells within these areas, and virtually all Pitx2<sup>+</sup> cells in these areas express at least one of these known myogenic markers (Shih et al., 2007b). These observations led the authors to conclude that Pitx2 marks the muscle lineage more completely than any of the known markers does. In agreement with the interpretation that muscle progenitors express Pitx2, L'Honoré et al. (2007) found extensive co-labeling of myotome- and dermomyotome-proliferating cells with PITX2, PAX3, and with PAX7. Notably, they also observed PAX3-positive cells that have completed migration at the proximal limb bud also express PITX2 while not all PITX2-positive cells expressed PAX3. All these data suggest that Pitx2 might be a player within the molecular pathways controlling muscle-progenitor fate.

Sometime afterwards, additional information regarding the hierarchical position occupied by Pitx2 within the genetic cascade that control somite-derived myogenesis was inferred by using Pitx2−/−, Myf5nlacZ/nlacZ and Pitx2−/−;Myf5nlacZ/nlacZ double-mutant mice (L'honoré et al., 2010). In this work, the authors showed that PITX2 protein directly regulates Myod expression through binding to its core enhancer in wildtype limbs. In agreement, the authors described a delayed myogenic differentiation and a Myod down regulation in Pitx2−/<sup>−</sup> limb buds and proposed that this phenotype appears to be due to the failure to activate the Myod core enhancer. However, although the inactivation of Myf5 and Myf6 in Myf5nlacZ/nlacZ mutant embryos (Myf6 is inactivated in cis in this mutant; Tajbakhsh et al., 1997) did not affect Myod expression in limb buds, this inactivation in a Pitx2−/<sup>−</sup> background (Pitx2−/−;Myf5nlacZ/nlacZ) induced a synergic effect that resulted not in a downregulation but in almost a complete loss of Myod expression compared with Pitx2−/−;Myf5nlacZ/<sup>+</sup> embryos, where the presence of one active Myf5 allele prevented Myod loss in about 60% of myogenic precursors cells. These results imply that Myf5 and/or Myf6 cooperate with Pitx2 to control Myod expression during early limb-bud myogenesis (**Figure 3A2**). In contrast to limb-muscle cells, myotome expression of Myod was not delayed in Pitx2−/<sup>−</sup> embryos. Nevertheless, in Myf5nlacZ/nlacZ mutant embryos, Myod expression was delayed by ∼2 days. Therefore, the onset of Myod expression in the myotome does not appear to depend on PITX2 but mostly on MYF5/MYF6. Nonetheless the inactivation of Myf5 and/or Myf6 in a Pitx2−/<sup>−</sup> background (Pitx2−/−;Myf5nlacZ/nlacZ) led to an almost complete loss of Myod expression in myotome, as happened in limbs (L'honoré et al., 2010). These results indicate that MYF5 and/or MYF6 also cooperate with PITX2 to control Myod expression during myotome development (**Figure 3A1**).

In addition, it should be stressed that the analysis of Pax3 mutant Splotch mice revealed a deficit of Pitx2 expression restricted to the myotome (L'honoré et al., 2010). This deficit is not observed in neighboring mesenchyme, indicating that Pitx2 is downstream of Pax3 during myotome myogenesis. This is also supported by transcriptome analyses of Pax3GFP/<sup>+</sup> and Pax3GFP/PAX3−FKHR transgenic mice carried out by Lagha et al. (2010) since, in gain-of-function screens for PAX3 targets, they found an up-regulation of Pitx2 in somites but not in limb buds. Although, all these seminal works suggest that Pitx2 could be acting downstream of Pax3 and in parallel with Myf5, at least in the myotome, as noted above, not all PITX2-expressing cells were positive for PAX3, and limb expression of Pitx2 precedes Myf5 (L'honoré et al., 2010). Therefore, additional studies using conditional Pitx2 inactivation in specific myogenic cell populations would help to elucidate the function of Pitx2 in embryonic myogenesis.

Scientific evidence also relates Pitx2 to cell proliferation in myogenic cells and somite derivates. Notably, Pitx2 has been reported to be a target gene in the Wnt/Dvl2/beta-catenin pathway and operates in specific cell types to control proliferation by regulating expression of the growth-control genes Ccnd1, Ccnd2, and c-Myc (Kioussi et al., 2002; Baek et al., 2003). These authors established that the PITX2 N-terminal domain is required for its effects on proliferation in a myoblast cell line. We have previously demonstrated that Pitx2c is the main Pitx2 isoform expressed in Sol8 myoblasts and that overexpression of Pitx2c in Sol8 cells led to an increase in proliferative capacity and completely blocked terminal differentiation, mainly because high levels of Pax3 expression were maintained (Martínez-Fernández et al., 2006). Additional data in vivo have supported the role of Pitx2 in cell proliferation during myogenesis. In this sense, Abu-Elmagd et al. (2010) showed that Pitx2 loss of function in chicken embryos decreased the number of differentiated myocytes/myofibers in the somites, whereas Pitx2 overexpression increased myocyte/myofiber numbers, particularly in the epaxial region of the myotome. In agreement with Abu-Elmagd et al. and by using Pitx2c−/<sup>−</sup> mutant embryos, we have reported that Pitx2c plays a pivotal role in the control of the subtle equilibrium between proliferation and differentiation during trunk and limb myogenesis. This control is exercised by balancing Pax3+/Pax7+ myogenic population in vivo as well as regulating key myogenic transcription factors such as Pax3 through the repression of miR-27 (Lozano-Velasco et al., 2011; **Figures 3A1,A2**). This new function of Pitx2c mediated by miRNAs introduces a new level of complexity in the intricate regulatory network that governs myogenesis in the embryo.

### Pitx2 during Head-Muscle Development

As mentioned above, Pax3 controls the myogenic specification of muscle embryonic progenitors in trunk and limbs (Tajbakhsh et al., 1997). However, it has been proposed that, instead of Pax3, Pitx2 plays a major role as an upstream regulator of craniofacial myogenesis (Zacharias et al., 2011; Buckingham and Rigby, 2014). This is supported by the fact that EOM development is impaired in Pitx2 null mice (Gage et al., 1999a; Kitamura et al., 1999). However, in the early studies it was not evident whether this muscle dysgenesis in Pitx2 mutant mice resulted from an intrinsic defect in the developing myoblasts or was secondary to the loss of Pitx2 expression in the periocular mesenchyme. Other authors have subsequently suggested that this phenotype could be due to the Pitx2 effect on proliferation rate of myogenic precursors (Noden and Francis-West, 2006), in agreement with previously reported data (Kioussi et al., 2002; Martínez-Fernández et al., 2006). The hypothesis that Pitx2 plays a part in controlling cell proliferation in myogenic cells in this context is also supported by the fact that conditional inactivation of Pitx2 in neural-crest-derived cells does not affect the early differentiation of eye muscles (Evans and Gage, 2005), while conditional Pitx2 deletion in the mesoderm induces a downregulation of Myf5, Myf6, Myod1, and Myog expression and, therefore, blocks the onset of myogenesis of EOM (Zacharias

et al., 2011). In this regard, in 2009, Sambasivan et al. by analyzing double defective mutant mouse embryos Myf5(Myf6) (Myf5nlacZ/+, Myf5nlacZ/nlacZ) and Myf4−/<sup>−</sup> mutants, showed that Pitx2 cannot ensure survival and activation of Myod expression in EOM in the absence of both Myf5 and Myf6 (Sambasivan et al., 2009). Shortly afterwards, Zacharias et al. were able to inactivate the expression of Pitx2 in mesodermal EOM precursors by using a tamoxifen inducible UBC- CreERT<sup>2</sup> promoter (Zacharias et al., 2011). This inactivation clearly showed that Pitx2 is required for EOM precursor specification and survival, acting as an anti-apoptotic factor in the premyogenic mesoderm and subsequently activating the myogenic program in these cells through direct binding to Myf5 and Myod promoters (Zacharias et al., 2011). Taken together, all these data clearly suggest that Pitx2 is an upstream regulator of Myf5, Myf6, and Myod in EOM embryonic myogenesis (**Figure 3A3**).

Pitx2 is also expressed in the myogenic precursors of the FBA. Tbx1 expression on FBA premyoblast is required for specification leading to Myf5 and Myod1 activation in those cells (Kelly et al., 2004). Notably, systemic Pitx2 mutants, whether Pitx2−/<sup>−</sup> (Dong et al., 2006) or Pitx2LacZ/LacZ (Shih et al., 2007a), display a down regulation of Tbx1 expression in this structure, although Pitx2 expression is unaffected in Tbx1 null mutants (Dong et al., 2006). These data, together with the fact that Pitx2 directly interacts with Tbx1 regulatory elements (Shih et al., 2007a) suggest that Pitx2 is an upstream activator of Tbx1 in FBA. A fuller analysis of both systemic Pitx2 mutants reveals that the inactivation of Pitx2 in FBA results in increased cell death in the mesodermal core and loss of early premyoblast specification markers such as Six2, Tcf21, and MyoR (Dong et al., 2006; Shih et al., 2007a). Although, the role that Six2 could play in the myogenesis of the FBA remains elusive, Tcf21 and MyoR are known to be upstream effectors of Myf5, Myod, and Myog in these initial muscle-cell clusters (Lu et al., 2002). Jointly, these results indicate that Pitx2 controls the expression of Myod1 and the onset of myogenesis in FBA through Tbx1, Tcf21, and MyoR (**Figure 3A4**).

## Pitx2 during Fetal Myogenesis

Most of what is known about Pitx2 concerns early (embryonic) myogenesis. However, a new role for Pitx2 has recently been unraveled during fetal myogenesis. L'Honoré et al. by using Pitx2:Pitx3 double conditional mutants, have shown that Pitx2/3 control the expression of the antioxidant system through the regulation of Nrf1 and antioxidant enzymes during muscle differentiation (L'honoré et al., 2014a). Thus, Pitx2/3 depletion at the onset of differentiation induces an abnormal increase of reactive oxygen species (ROS) levels in differentiating myoblasts and leads to impaired myogenesis due to apoptosis of these cells. These results emphasize the role of Pitx2 controlling redox conditions during fetal myogenesis.

## Pitx2 Is Emerging as a Key Transcription Factor That Modulates Adult Myogenesis

During adult life the maintenance and repair of skeletal-muscle tissue is directed by SCs. The regulation of SC function in adults requires the redeployment of many of the regulatory networks fundamental for developmental myogenesis. Although, several efforts have been made during the last few years to disentangle the role of Pitx2 in embryonic and fetal stages of myogenesis, studies linking Pitx2 to adult myogenesis have only recently emerged and are still controversial.

The first evidence regarding Pitx2 expression in SCs was reported by Ono et al. (2010). These authors showed that all Pitx2 isoforms are expressed in proliferating SC-derived myoblasts. They analyzed SCs with a different ontology, comparing those of the extensor digitorum longus (EDL) of the limb with SCs from the masseter of the head (MAS). They found that Pitx2b and Pitx2c levels were higher in cells from the EDL than from the MAS, with Pitx2c being the main Pitx2 isoform expressed in proliferating limb SCs (Ono et al., 2010). Based on these distinct gene-expression profiles, the authors suggest that, even after activation and entry into the cell cycle, SCs retain an identity consistent with their ontogeny underlying their distinct properties. Subsequent studies have pointed out that Pitx2a, Pitx2b, and Pitx2c were expressed at very low levels in proliferating SCs, but increased during the early stages of myogenic differentiation. Meanwhile the constitutive expression of any Pitx2 isoform suppressed SC proliferation, with the cells undergoing greater myogenic differentiation (Knopp et al., 2013). However, additional evidence underlying the functional relevance of Pitx2 on SC proliferation has been reported. For example, Herbet et al. demonstrated that Pitx2 is crucial in maintaining the phenotype of myogenic precursor cells in the extraocular muscles (EOM; Hebert et al., 2013). In this analysis, the authors found that the higher levels of Pitx2 expression in EOM in comparison with limb muscles were concomitant with longer proliferative state in EOM-derived SCs as compared with limb cells. In addition, the knockdown of Pitx2 in SCs isolated from EOM slowed their proliferation rate, and a similar trend was seen for SCs isolated from tibialis anterioris muscle. These data led to the conclusion that Pitx2 helps maintain a proliferating pool of myogenic precursor cells. Finally, the authors highlight that this greater proliferative capacity may facilitate the repair of damaged EOM tissue, thereby contributing to the sparing of EOM in muscular dystrophies (Hebert et al., 2013).

More recently, a study conducted in our laboratory has provided additional information about the molecular mechanisms by which the Pitx2 transcription factor regulates cell proliferation in SCs (Lozano-Velasco et al., 2015). We have reported that Pitx2c expression is higher in early-activated SCs than in long-term activated ones, and our in vitro Pitx2c gain-offunction experiments have revealed that Pitx2c stimulates Ccnd1 and Ccnd2 expression, accelerating cell proliferation during early satellite-cell activation. Moreover, we have demonstrated that such Pitx2c effect on SCs proliferation is due to Pitx2c-mediated downregulation of the miRNAs miR-15b, miR-106b, miR-23b, and miR-503 (**Figure 3B**). The existence of the Pitx2-miRNA pathway controlling the expression of key regulatory cell-cycle genes in early-activated SCs revealed a role of Pitx2 in satellitecell activation. Although, muscle SCs are promising targets for cell therapies, the paucity of SCs that can be isolated or expanded from adult muscle tissue is limiting; thus these findings provide new molecular tools to overcome such a bottleneck. It bears noting that our analyses also showed that Pitx2c can increase Myf5 expression by down-regulating miR-106b (**Figure 3B**), thus expanding the Myf5<sup>+</sup> satellite-cell population and revealing a role for Pitx2c in promoting satellite-cell populations more primed for myogenic commitment (Lozano-Velasco et al., 2015). In this context it should be highlighted that in several muscular disorders such as muscular dystrophies, the progressive muscle wasting and weakness is often associated with exhaustion of muscle-regeneration potential. Therefore, the progressive loss of muscle mass has been attributed, at least partly, to the inability of muscle stem cells to efficiently regenerate tissue loss as the result of the disease (Berardi et al., 2014). Thus, critical for the development of effective strategies to treat muscle disorders is the optimization of approaches targeting muscle stem cells and capable of regenerating tissue loss as the result of the disease or as the result of normal muscle turnover (Bertoni, 2014). Notably, very recent reports have been pointed out that muscle stem cells should be considered as a therapeutic target for restoring muscle function in individuals with DMD (Chal et al., 2015; Dumont et al., 2015). Therefore, identification of new Pitx2 functions in the context of SC biology may significantly contribute to the clarification of the molecular and cellular mechanisms of skeletal-muscle regeneration and may help to develop therapeutic strategies for muscular disorders.

Notably, the analysis of adult single and double Pitx2:Pitx3 conditional mutant mouse lines targeted to the muscle stem-cell compartment revealed that double mutant SCs undergo senescence with impaired regeneration after injury, suggesting that Pitx2-mediated changes in ROS levels are required for differentiation of SCs (L'honoré et al., 2014b).

All these data provide new insight into the function of Pitx2 in the molecular mechanisms that control SC behavior and might thus have future application to enhance the regenerative capacity of these myogenic precursor cells. Further analysis using in vivo models could aid in understanding how the Pitx2 mediated effects on SCs can influence the kinetics of muscle regeneration.

## CONCLUSIONS AND FUTURE CHALLENGES

The data reviewed above show that Pitx2 is a comprehensive marker for cells undergoing myogenic progression, more so than any of the MRFs. This supports models that include a Pitx2 dependent pathway in virtually all skeletal muscles. Many pieces of experimental evidence have pointed out that Pitx2 is the first molecular signal specifying all myogenic precursors in the head muscles. However, although several works have characterized Pitx2 as a key transcription factor in the molecular cascade regulating trunk- and limb-muscle progenitors, additional work is needed to elucidate the function of Pitx2 in specification vs. determination during trunk and limb myogenesis. In addition, since seminal works have revealed that Pitx2 functions on myogenic cells may be due to Pitx2-mediated regulation of miRNAs, the role of Pitx2 in the post-transcriptional control of myogenesis should be further explored.

In parallel, the role of Pitx2 during adult myogenesis is beginning to be explored. Skeletal muscle has the ability to

## REFERENCES


repair and regenerate due to the presence of resident SCs. SC function in adults requires redeployment of many of the regulatory networks fundamental to developmental myogenesis. Currently, SCs are considered potential therapeutic targets for restoring muscle function in muscle degenerative disorders such as muscular dystrophies. Recent works indicate that Pitx2 is expressed in proliferating SCs and can promote differentiation of satellite-cell-derived myoblasts. Moreover, the identification of Pitx2-miRNA pathways that regulate satellite-cell behavior as well as the impact of Pitx2 on redox condition during satellitecell differentiation may open insights toward future applications to modulate satellite-cell fate during muscle regeneration. Therefore, these findings propose Pitx2 as a new player on skeletal-muscle satellite-cell biology and may help to develop therapeutic strategies for muscular disorder.

## AUTHOR CONTRIBUTIONS

FH, LR, and AA conceived of the structure and content. FH wrote the first draft document. FH and LR designed and produced the figures. LR and DF critically revised the manuscript for intellectual content. AA corrected, edited, and approved the final version of the document to be published.

## ACKNOWLEDGMENTS

This work was partially supported by grants BFU2012-38111, BFU2015-67131 (Ministerio de Economía y Competitividad, Gobierno de España), CTS-1614, P08-CTS-03878, BIO-302 (Junta de Andalucia), and AFM2012-16074 (AFM).


their commitment to a myogenic cell fate. Mol. Cell. Biol. 35, 2892–2909. doi: 10.1128/MCB.00536-15


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

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

# Sonic Hedgehog Signaling in Limb Development

#### Cheryll Tickle<sup>1</sup> \* and Matthew Towers <sup>2</sup> \*

<sup>1</sup> Department of Biology and Biochemistry, University of Bath, Bath, UK, <sup>2</sup> Department of Biomedical Science, The Bateson Centre, University of Sheffield, Western Bank, Sheffield, UK

The gene encoding the secreted protein Sonic hedgehog (Shh) is expressed in the polarizing region (or zone of polarizing activity), a small group of mesenchyme cells at the posterior margin of the vertebrate limb bud. Detailed analyses have revealed that Shh has the properties of the long sought after polarizing region morphogen that specifies positional values across the antero-posterior axis (e.g., thumb to little finger axis) of the limb. Shh has also been shown to control the width of the limb bud by stimulating mesenchyme cell proliferation and by regulating the antero-posterior length of the apical ectodermal ridge, the signaling region required for limb bud outgrowth and the laying down of structures along the proximo-distal axis (e.g., shoulder to digits axis) of the limb. It has been shown that Shh signaling can specify antero-posterior positional values in limb buds in both a concentration- (paracrine) and time-dependent (autocrine) fashion. Currently there are several models for how Shh specifies positional values over time in the limb buds of chick and mouse embryos and how this is integrated with growth. Extensive work has elucidated downstream transcriptional targets of Shh signaling. Nevertheless, it remains unclear how antero-posterior positional values are encoded and then interpreted to give the particular structure appropriate to that position, for example, the type of digit. A distant cis-regulatory enhancer controls limb-bud-specific expression of Shh and the discovery of increasing numbers of interacting transcription factors indicate complex spatiotemporal regulation. Altered Shh signaling is implicated in clinical conditions with congenital limb defects and in the evolution of the morphological diversity of vertebrate limbs.

Keywords: Sonic hedgehog, limb, digits, mouse, chick, positional information

## INTRODUCTION

Over 20 years ago the first evidence was presented that Sonic hedgehog (Shh), an orthologue of the Drosophila Hedgehog (Hh) gene, encodes the long sought after morphogen that specifies anteroposterior pattern in developing vertebrate limbs (Riddle et al., 1993). Grafting experiments in chick wing buds in the 1960s revealed that a group of morphologically indistinguishable mesenchyme cells at the posterior margin of the wing bud (the margin nearest the tail), later known as the polarizing region (or zone of polarizing activity), is an important cell-cell signaling center that controls development across the antero-posterior axis (Saunders and Gasseling, 1968). Tissue transplanted from the posterior margin of one chick wing bud to the anterior margin of another was shown to have the striking ability to duplicate the pattern of three digits, so that another set develop in mirror-image symmetry to the normal set. Based on these observations it was proposed that the

#### Edited by:

Andrea Erika Münsterberg, University of East Anglia, UK

#### Reviewed by:

Megan Davey, University of Edinburgh, UK Robert Hill, University of Edinburgh, UK Susan Mackem, National Cancer Institute, USA

\*Correspondence:

Cheryll Tickle cat24@bath.ac.uk Matthew Towers m.towers@sheffield.ac.uk

#### Specialty section:

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

> Received: 07 December 2016 Accepted: 08 February 2017 Published: 28 February 2017

#### Citation:

Tickle C and Towers M (2017) Sonic Hedgehog Signaling in Limb Development. Front. Cell Dev. Biol. 5:14. doi: 10.3389/fcell.2017.00014 polarizing region produces a diffusible morphogen that specifies antero-posterior positional values (Wolpert, 1969). These positional values are interpreted so that a structure, such as a digit with an appropriate identity, develops in the correct position.

The key pieces of evidence that Shh is the polarizing morphogen are that Shh transcripts were found to be localized to the polarizing region of the chick wing bud (**Figures 1a–f**) and that Shh-expressing cells grafted to the anterior margin of chick wing buds can produce the same effects as grafts of the polarizing region (Riddle et al., 1993). Earlier experiments revealed that tissue from the posterior margin of mammalian limb buds grafted to the anterior margin of chick wing buds could duplicate the pattern of chick wing digits (Tickle et al., 1976; Fallon and Crosby, 1977). This is explained by the finding that Shh is expressed at the posterior margin of mammalian limb buds (Echelard et al., 1993; Odent et al., 1999). Shh has now been shown to be expressed at the posterior margin of the limb buds of all vertebrates studied to date, including the fin buds of the most primitive chondrichthyan fishes such as the shark (Dahn et al., 2007).

Experiments in which the polarizing region was grafted to the anterior margin of another chick wing bud showed that polarizing region signaling also plays a role in controlling the width of the limb bud and that widening of the bud is required to specify a complete set of new antero-posterior positional values (Tickle et al., 1975; Smith and Wolpert, 1981). The earliest detected effect of a polarizing region graft was an increase in cell proliferation in adjacent mesenchyme in the host wing bud (Cooke and Summerbell, 1980). In addition, it was proposed that the polarizing region controls the production of a factor by the mesenchyme that maintains the apical ectodermal ridge over the region of the wing bud that will give rise to distal structures including the digits (Zwilling and Hansborough, 1956). The apical ectodermal ridge is a signaling region that rims the bud and is required for proximal-distal patterning and outgrowth and the laying down of structures along this axis; the extent of the apical ectodermal ridge across the antero-posterior axis controls the width of the wing bud and determines the number of digits that can form. The effects of the polarizing region on the apical ectodermal ridge also link antero-posterior and proximo-distal pattern formation. This explains the observation that polarizing region grafts made at later stages of development affect the antero-posterior pattern of more-distal structures (Summerbell, 1974).

Early experiments highlighted the complex relationship between the polarizing region and apical ectodermal ridge. In order for a polarizing region to signal, it has to contact the apical ectodermal ridge (Tickle et al., 1975) and this interaction is required in order for the polarizing region to maintain production of the apical ridge maintenance factor by the mesenchyme that will form distal structures. In addition, in the chick wing bud, the polarizing region itself demarcates the posterior limit of the apical ectodermal ridge and grafts of the polarizing region placed under the apical ectodermal ridge flatten it (Saunders and Gasseling, 1968). Interestingly, it has also been shown that the dorsal ectoderm of the wing bud, which produces a signal controlling the development of the dorsal pattern of structures (e.g., extensor muscles), is also required for the polarizing region to signal (Yang and Niswander, 1995). Thus, signaling along all three axes of the developing limb bud is integrated.

It has now been shown that Shh affects cell proliferation in the chick wing bud by controlling expression of genes encoding cell cycle regulators including D cyclins independently of the apical ectodermal ridge (Towers et al., 2008). Work on mouse limb development has shown that Shh controls expression of the Gremlin1 gene, which encodes the BMP antagonist that acts as the apical ridge maintenance factor (Zuniga et al., 1999). In addition, it has also been demonstrated that short-range Shh signaling can flatten the apical ridge above the polarizing region (Bouldin et al., 2010).

Experiments on chick wing buds have identified FGFs as the apical ectodermal ridge signals that promote outgrowth and also maintain Shh expression in the polarizing region (Laufer et al., 1994; Niswander et al., 1994). Genetic experiments in mouse have identified Wnt7a as the dorsalizing signal that also contributes to regulating Shh expression (Parr and McMahon, 1995). Loss of Wnt7a function in the mouse limb results in the transformation of dorsal to ventral fates and loss of posterior digits (Parr and McMahon, 1995). This second phenotype is consistent with a function for Wnt7a in controlling Shh expression since no digits form in the fore-limbs of Shh−/<sup>−</sup> mouse embryos and only a single digit—considered to be an anterior digit 1—is present in hind-limbs (Chiang et al., 1996).

In this review, we will emphasize the parallel contributions that experimental chick embryology and mouse genetics have played in providing the current picture of Shh function in the limb. We will provide an in-depth picture of how Shh specifies antero-posterior positional values in the limb buds of these two main vertebrate models and how this is integrated with its role in growth. We will consider how Shh expression in the limb is initiated, maintained and eventually extinguished and how cells respond to the Shh signal. We will finally review clinical conditions affecting the limb and examples of evolutionary diversification of limb morphology that are associated with changes in Shh signaling.

## SPECIFICATION OF ANTERO-POSTERIOR PATTERN

## Chick Wing

Detailed embryological experiments on the chick wing bud have been crucial in establishing the signaling parameters of the polarizing region morphogen. The polarizing region was first discovered in the chick wing bud, where it overlaps with a region of programmed cell death, known as the posterior necrotic zone (Saunders and Gasseling, 1962). Indeed, the original grafting experiments were designed to investigate how this region of cell death is controlled (Saunders and Gasseling, 1968). Tissue from the posterior margin of a chick wing bud was grafted to the anterior margin of a second wing bud and this resulted in a mirror-image pattern of digits across the antero-posterior axis. The normal chick wing has three digits (designated at this time as 2, 3, and 4) but following a polarizing region graft to the

anterior margin, six digits can develop in the pattern 4-3-2-2- 3-4. Note that recent evidence supports numbering of the digits as 1, 2, and 3 (Towers et al., 2011), and this numbering system is now generally accepted and will be used in this review. This grafting experiment provided an assay for polarizing activity and antero-posterior pattern that could readily be scored by the distinct skeletal morphology of each of the three digits of the chick wing. It should be noted that grafts of the polarizing region also affect the antero-posterior pattern of the wing forearm skeleton and soft tissues (Shellswell and Wolpert, 1977; Robson et al., 1994). Thus, following a polarizing region graft, two ulnae develop and the pattern of muscles is also duplicated. The myogenic cells of the muscle originate in the somites and migrate into the limb bud but the pattern of the wing muscles is dictated by the connective tissue, which is derived from the lateral plate mesoderm (Chevalier and Mauger, 1977). Therefore, the duplicated pattern of muscles following a polarizing region graft will be based on the response of the cells that give rise to the muscle connective tissue.

The experimental parameters determined for polarizing region signaling in the chick wing (reviewed in Towers and Tickle, 2009) are consistent with the suggestion that the polarizing region produces a long-range morphogen that sets up a concentration gradient across the antero-posterior axis of the wing bud and specifies positional values (Wolpert, 1969). According to this model, the positional values at particular threshold concentrations govern digit identity, with the highest threshold concentration in tissue closest to the polarizing region specifying the most-posterior digit, digit 3, and the lowest threshold concentration in tissue further away specifying the most-anterior digit, digit 1. Thus, any candidate molecule for the polarizing region morphogen must act in a concentrationdependent manner (Tickle, 1981) and provide a long-range signal (Honig, 1981).

The first defined molecule found to mimic the duplicating activity of polarizing region grafts was the vitamin A derivative, retinoic acid (Tickle et al., 1982, 1985) but it was subsequently shown that retinoic acid acts indirectly (Noji et al., 1991; Wanek et al., 1991) by inducing Shh expression (Riddle et al., 1993). There is now good evidence that Shh acts in a concentrationdependent fashion to induce digit duplications. When Shh– expressing cells, or beads soaked in bacterially produced ShhN protein (the active N-terminal fragment produced by autocatalytic cleavage of the large precursor Shh protein), are placed at the anterior margin of a chick wing bud, the extent of digit duplication depends on the number of Shh– expressing cells grafted or the concentration of ShhN protein in which the beads are soaked (Yang et al., 1997). Fewer Shhexpressing cells or lower concentrations of Shh elicit duplication of only the anterior digit 1 (Yang et al., 1997). Grafts of Shh-expressing cells that induce full digit duplications were also shown to result in two ulnae developing in the forearm together with a duplicated pattern of muscles (Duprez et al., 1999).

The original model for how antero-posterior values are specified in the chick wing bud did not consider the dynamic nature of the process, although experiments showed that the extent of duplication following a polarizing graft depended on the length of time that the graft was left in place (Smith, 1980). A similar time dependency was subsequently seen with Shh–soaked beads (Yang et al., 1997). Furthermore, fate mapping experiments showed that cells near a Shh-soaked-bead give rise to an anterior digit 1 when the bead is removed after a short time, but give rise to a more posterior digit (2) if the bead is left in place for longer (Yang et al., 1997). This process by which positional values of cells change over time in response to an increasing concentration of morphogen is known as promotion (see also (Gurdon et al., 1995). An alternative process in which wing bud cells acquire a stable positional value depending on the duration of Shh signaling and then are displaced by growth can be ruled out because an anterior digit 1 has been shown to arise in tissue which was not originally adjacent to a polarizing region graft (see Tickle, 1995).

The parameters of polarizing region discussed above were determined in experiments in which additional digits were induced following polarizing region grafts to the anterior margin. But what is the evidence that Shh acts long range and how does Shh signaling specify antero-posterior positional values during normal development of the chick wing? Measurements of Shh activity in slices taken from different positions across the bud using an in vitro cell-differentiation assay are consistent with there being a concentration gradient of Shh across the bud, with Shh activity of a posterior slice being 5–6 times higher than that of a middle slice (Zeng et al., 2001). Another indication that Shh spreads across the wing bud and provides a long range signal is that high levels of the transcripts of known direct gene targets of Shh signaling, including Ptch1 (encoding the main receptor for Shh), and Gli1 (encoding a transcriptional effector of Shh signaling) encompass the posterior two-thirds of the wing bud, including adjacent tissue in addition to the polarizing region (Marigo et al., 1996). It should also be noted that following a polarizing region graft or implantation of an Shh bead to the anterior margin of the chick wing, there is a burst of high level Ptch1 expression in the anterior part of the wing bud, which then subsides and is later followed by the establishment of a stable domain of high level Ptch1 expression (Drossopoulou et al., 2000). This suggests that cells could respond to and interpret two waves of Shh signaling; the first defining the size of the domain that can give rise to digits, and the second, promoting the growth of this domain and specifying positional values.

The temporal specification of positional values specified by Shh in normal wing development has been directly addressed by applying cyclopamine, a small molecule inhibitor of Hh signaling at the level of Smoothened to chick embryos, at a series of short time intervals after the onset of Shh expression in wing buds (Towers et al., 2011). Smoothened, a member of the Gprotein coupled receptor superfamily, is normally activated upon Shh binding to Ptch1, and this triggers of activation of the Gli family of transcription factors (see section on Mechanisms of Shh signaling). Application of cyclopamine about 4 h after the onset of Shh expression results in the development of just the anterior digit 1, the anterior and middle digits (1 and 2) develop when cyclopamine is added at 8 h while a complete set of digits (1, 2, and 3) develop when cyclopamine is added at 12 h (Towers et al., 2011). Furthermore, fate mapping experiments show that promotion is occurring with cells next to the polarizing region first being specified to form the anterior digit 1, then being promoted to form the middle digit 2 and finally the posterior digit 3 (**Figure 2A**).

The effects of Shh signaling on antero-posterior growth must be included in any comprehensive model for specification of antero-posterior pattern in the chick wing. Application of cyclopamine in the experiments described above demonstrated that Shh signaling has effects on both specification of anteroposterior positional values and growth because this treatment not only prevented promotion but also expansion of the region of the wing bud that will give rise to distal structures leading to the development of fewer digits (Towers et al., 2008, 2011). When growth alone is targeted by adding trichostatin A or colchicine, and following over-expression of the cyclin-dependent kinase inhibitor—p21Cip1—at a similar series of time points, fewer digits also develop, but because specification of positional values and promotion by Shh signaling are unaffected, the digits that develop are posterior digits (Towers et al., 2008). These experiments show that specification of antero-posterior positional values in the early chick wing bud is coupled with growth that determines the width of the wing bud.

The cyclopamine experiments also show that antero-posterior values are specified over a relatively short time period during early wing bud development. However, these values will not be interpreted in terms of digit identity until much later in development when the digit condensations develop (**Figure 2A**). When the Shh-expressing region is completely removed from the early wing bud at the time when the positional values that specify two digits are specified, truncated wings develop with posterior structures being preferentially lost (Pagan et al., 1996), showing the crucial importance of Shh signaling in stimulating anteroposterior expansion and maintaining the apical ectodermal ridge. Resulting skeletons bear resemblance to those of the wings of the chicken mutant Oligozeugodactyly (Ozd) that develop devoid of Shh (Ros et al., 2003). It is unclear why Shh continues to be expressed at the posterior margin of the chick wing bud long after the antero-posterior values have been specified (**Figure 2A** see section Termination of Shh expression).

## Chick Leg

The chick leg has four morphologically distinct digits (numbered 1, 2, 3, and 4 in antero-posterior sequence). Early grafting experiments demonstrated that chick leg buds also have a polarizing region but it was noted that when the leg polarizing region was grafted to a chick wing bud, a toe frequently developed in the duplicated wings (Summerbell and Tickle, 1977). It has since been demonstrated using grafts from the Green Fluorescent Protein-expressing transgenic chicken to make fate maps of the polarizing region that the chick leg polarizing region gives rise to the most posterior digit 4, whereas in the chick wing all the digits

FIGURE 2 | Comparison of models of Shh function in chick and mouse limbs. (A) Chick wing promotion model. Positional values of digits 1, 2, and 3 specified adjacent to polarizing region (blue shading) and promoted over 12 h through a series of increasingly posterior positional values by a concentration gradient of paracrine Shh signaling (graded blue shading—note coloring of polarizing region also shows strength of Shh expression. Shh terminated at around 60 h as digit condensations form by self-organization (black numbers). Colors of developing digits indicate a different positional value that cells were specified with. (B) Chick leg promotion model. Positional values of digits 1, 2, and 3 specified as (A) but polarizing region cells promoted through progressively anterior positional values over 16 h in response to time of autocrine Shh signaling (red numbers) and form digit 4. Shh terminated at around 60 h. (C) Mouse limb temporal expansion. Positional values of digits 1, 2, and 3 specified adjacent to the polarizing region by a gradient of paracrine Shh signaling over approximately 24 h– it is unclear whether promotion is involved (see A). Positional values of digits 4 and 5 specified in polarizing region sometime before Shh terminates at 60 h according to duration of autocrine Shh signaling. Shh terminates at around 60 h. (D) Mouse limb biphasic model. Positional values of digits 1, 2, 3, 4, and 5 specified by Shh, possibly by a gradient of paracrine signaling from the polarizing region in approximately 6 h. It is unclear whether promotion is involved and is possible in this time (see A), or if Shh levels can reach concentrations predicted required to specify posterior positional values. Shh signaling over the next 16 h required for specified digit progenitor cells to proliferate and form condensations in the order digit 1, 4, 2, 5, and 3 (purple numbers). (E) Mouse limb promotion model. Positional values of digits 1, 2, 3, and 4 specified as (B) and polarizing region enlarges sufficiently to give rise to digits 4 and 5 by self-organization. Note promotion model does not easily explain digit 5 patterning that requires a shorter exposure to form than digit 3 (see D). (F) Mouse limb truncated promotion model. Anterior positional values specified (1 and 2) specified by autocrine and paracrine signaling and then cells become refractory to further posterior promotion. Digits form by self-organization: 1, 2, and 3 from cells adjacent to polarizing region, digits 4 and 5 from the polarizing region.

come from tissue anterior to the polarizing region (Towers et al., 2011; see **Figure 1c**).

Shh is expressed at the posterior margin of chick leg buds for a similar duration to its expression in chick wing buds. Furthermore, it has been demonstrated by treating leg buds with cyclopamine that the positional values that specify the three anterior digits of the chick leg are promoted in response to paracrine Shh signaling in an identical fashion to those that specify the three digits of the chick wing (Towers et al., 2011). However, the positional value for the most posterior digit 4 is promoted in response to autocrine Shh signaling (**Figure 2B**). Thus, when Shh signaling was attenuated in the chick leg bud by cyclopamine 4 h after onset of Shh expression, two toes with digit 1 identities arose—one from the polarizing region, the other from adjacent anterior tissue, while when Shh signaling was attenuated after 8 h, three digits develop, toes with digit 2 identities from the polarizing region and adjacent cells and a toe with a digit 1 identity from cells further away, and so on, until by 16 h, all the antero-posterior positional values in the leg bud have been specified (Towers et al., 2011). These observations show that although it takes slightly longer to specify antero-posterior positional values in the leg compared to the wing, this process is nevertheless accomplished in the early leg bud, and, as in the wing bud, some considerable time elapses before these positional values are interpreted (**Figure 2B**). It should be noted that, in the Ozd chicken mutant, a single digit 1 forms in the leg (Ros et al., 2003).

## Mouse Limb

The mouse limb has five digits (1, 2, 3, 4 and 5 in antero-posterior sequence) and digits 2–5 all have three phalanges making them morphologically very similar. Fate maps of the mouse limb polarizing region made by tracing genetically labeled cells that have expressed Shh show that the two posterior digits of the mouse limb are entirely derived from the polarizing region, and while there is some contribution to digit 3, the two anterior digits come from cells outside of the polarizing region (Harfe et al., 2004).

Shh is expressed at the posterior margin of limb buds of mouse embryos between E9.5–E12.0 (60 h; Zhu et al., 2008, **Figure 2C**, note expression is between E10-E12.5 in hind-limbs). At E10.5, a graded distribution of Shh across the posterior third of the mouse hind limb bud has been detected by immunohistochemical analysis (Gritli-Linde et al., 2001) in keeping with paracrine Shh signaling specifying antero-posterior positional values as in the chick wing. Shh is expressed not only at the posterior margin but also at the anterior of the limbs of several polydactylous mouse mutants (Masuya et al., 1995) consistent with Shh functioning as a polarizing signal in mouse limbs. In contrast, in mouse embryos lacking Shh function, the limbs taper toward the tip, and only one digit-like structure (interpreted as digit 1) develops in the hind-limb, while no digits develop in the fore-limb (Chiang et al., 1996). This indicates that Shh is required for the outgrowth of the limb and for the development of structures distal to the elbow/knee in the mouse limb. It should also be noted that in mouse embryos lacking Shh function the development of muscles in this distal region of the limb is severely compromised (Kruger et al., 2001) Experiments in which Smoothened activity is deleted specifically in the prospective myogenic cells show that Shh signaling has direct effects on these cells; timing myogenic differentiation, promoting slow muscle differentiation and controlling their migration into the distal part of the limb (Anderson et al., 2012; Hu et al., 2012).

In chick limbs, antero-posterior positional values clearly relate to the identity of a digit that develops in an appropriate position. However, this is not readily observable in the mouse limb due to the difficulties in determining which digits are present in mouse limbs conditionally lacking Shh function. Therefore, there is currently no general consensus about the model which best reflects how positional values are specified in the mouse limb bud. The various models are now discussed below (also see **Figures 2C–F**).

The first formal model to be proposed for the mouse limb was the temporal expansion model (Harfe et al., 2004). In this model, anterior positional values for digit 2 (and in part for digit 3) are specified in a concentration-dependent fashion by paracrine Shh signaling and then, posterior positional values (for digits 4 and 5) by the duration of autocrine Shh signaling, which is governed by the proliferative expansion and then displacement of cells from the polarizing region (**Figure 2C**; specification of digit 1 is considered to be Shh-independent in the hind-limb). Consistent with the model, the restriction of paracrine signaling in a Dispatched mutant (see later section on Mechanisms of Shh signaling) resulted in loss of one digit, suggested to be digit 2. This model also gained support from the finding that when Shh expression was curtailed in the developing mouse limb, this resulted in only three digits developing. The authors identified these digits as being 1, 2, and 3 consistent with the prediction that digits should be lost in a posterior to anterior sequence (Scherz et al., 2007). A particular feature of this model is that it takes considerable time for all the antero-posterior positional values to be specified (**Figure 2C**), rather than over a short time in the early limb bud. Moreover, it does not take into account promotion through a transitory series of anterior to posterior positional values, which has been demonstrated to occur in the limb buds of the chick.

A later model was proposed by Zhu et al. (2008) based on the results of a more extensive set of experiments, in which Shh function was deleted at a series of different stages in mouse limb development. Again, digits were lost with progressively fewer digits developing when Shh function was deleted at earlier and earlier stages. However in this case, the authors suggested that the sequence of digit loss reflects the order in which digits form, with digits that form last being lost first. Thus, for example, they identified the digits in limbs with three digits as being 1, 2, and 4. If their identification of the digits is correct, a posterior digit has formed adjacent to an anterior digit, an outcome not predicted by any previous model. Based on their findings, they proposed a biphasic model for digit patterning—in which Shh has two functions (**Figure 2D**). In the first phase, Shh specifies positional values across the antero-posterior axis of the very early limb bud, possibly via a concentration gradient, while in the second phase Shh is required to support proliferation and survival of cells that will form the digits (Zhu et al., 2008). It is not clear whether this latter function is a separate direct function of Shh signaling or reflects an essential role of Shh signaling in maintaining sufficient apical ectodermal ridge signaling. According to this model, the resultant digit patterns when Shh function is deleted are due to loss of Shh compromising survival and proliferation of specified digit progenitor cells rather than failure to specify anteroposterior positional values (Zhu et al., 2008). Furthermore, positional values would have to be specified in the early mouse limb bud over a period of approximately 6 h (based on Ptch1 expression), which suggests that this process is not integrated with growth as in the chick wing.

The ability to observe promotion in chick limbs gives insights into the time required to specify positional values, but in the mouse limb, in which promotion is not readily observed, it is difficult to distinguish between the effects of Shh signaling on specification of positional values and survival and proliferation of the cells that will form the digit condensations. Indeed the time required for digit specification proposed by Zhu et al. does not appear consistent with a model in which antero-posterior positional values are promoted in response to the concentration and/or duration of Shh signaling. However, if one were to take promotion into account, a unifying model can be proposed (Towers et al., 2011). According to this proposal, positional values would be specified early in the mouse limb as suggested in the biphasic model. However, these would only be anterior positional values, which would then be promoted to posterior values by both paracrine and autocrine Shh signaling operating in parallel. Thus, the pattern of digits specified would depend on how far positional values have been promoted at the time at which Shh function is deleted in keeping with more conventional models for digit patterning. The digits that develop in the threedigit mouse limb when Shh signaling is curtailed would therefore be predicted to be 1, 2, and 2—a pattern that is readily observed in cyclopamine-treated chick legs (Towers et al., 2011), and occasionally in wings (Pickering and Towers, 2016). However, there are difficulties in applying a promotion model to the specification of digit 5 of the mouse limb as this would imply that it is the last digit of the pattern to be specified (**Figure 2F**), when in fact it forms before more-anterior digits (**Figure 2D**, see also discussion in Towers et al., 2011).

## INTERACTION BETWEEN POSITIONAL INFORMATION AND A TURING-TYPE MECHANISM

Although, it has been shown that Shh is the critical signal in controlling development across the antero-posterior axis of the limb, there is evidence that the periodic condensation of cells that will form the digits depends on an underlying Turing type selforganization mechanism independent of graded Shh signaling. In the basic Turing model, diffusible signals—one operating as an inhibitor, the other as an activator—interact to produce the pattern of digits and interdigits. Positional information and self-organization have been presented as competing models of digit development, when in fact the power of both processes operating together has been long recognized (see (Wolpert, 1989) and for original paper on reaction-diffusion (Turing, 1952).

The first indications that such a self-organization mechanism might be involved in limb development came from experiments in which it was shown that recombinant limb buds formed from disaggregated single cells, re-aggregated and placed back in an ectodermal jacket could still form digits (Zwilling, 1964; Pautou, 1973). Indeed, based on this latter study, one of the first computer simulations of limb development was developed (Wilby and Ede, 1975). Further experiments showed that when recombinant limbs were made from chick mesenchyme cells from the anterior halves of early chick leg buds, which would not include a polarizing region, and which would not normally give rise to digits, two or three morphologically similar digit-like structures developed (Hardy et al., 1995; Elisa Piedra et al., 2000). When a polarizing region was grafted into such recombinant limbs, however, the digits that developed had recognizable identities (MacCabe et al., 1973). These experiments elegantly revealed that positional information and self-organization are integrated in limb development. There is evidence that a self-organization mechanism also operates in mouse limb buds, as the limbs of mutant mouse embryos in which the Shh signaling pathway is non-functional have many morphologically similar digits (Litingtung et al., 2002; te Welscher et al., 2002; see Section– Measurement of Shh concentration and duration of signaling). Indeed, recent studies in the mouse limb have suggested that this mechanism is based on WNT signals acting as inhibitors and BMP signals as activators, that together, converge on the transcription factor Sox9 to generate a repeated series of digit condensations (Raspopovic et al., 2014).

Since digits 2–5 have similar morphologies in the mouse limb, particularly in regard to phalangeal count, one proposal is that self-organization plays a dominant process (Delgado and Torres, 2016). This scenario could for account for difficulties in applying a positional information model to the five digits of the mouse limb. Moreover, a recent study on developing chick wings has revealed how positional information and self-organization can interact and this could be relevant to understanding how the mouse digit pattern is specified. If chick wing buds are treated with cyclopamine under conditions in which the promotion of antero-posterior values is truncated, a series of morphologically similar digit 2s in a pattern 1-2- 2-2 can develop by self-organization (Pickering and Towers, 2016). It should be noted that the digit 2s were not of identical morphologies and sizes suggesting other factors control these finer aspects of development. In wings with multiple digit 2s, the most-posterior of these digits arises from cells of the polarizing region. An interpretation of these findings is that antero-posterior expansion mediated by a posteriorly extended apical ectodermal ridge has enabled a small pool of cells specified with the same positional value to produce a series of digit 2s by self-organization (Pickering and Towers, 2016). In extrapolating these data to the mouse limb, it has been suggested that a similar mechanism could account for the patterning of digits 1 through to 4 (Pickering and Towers, 2016; **Figure 2F**). In addition, the apical ectodermal ridge of the mouse limb completely overlies the polarizing region (Pickering and Towers, 2016), and an intriguing suggestion is that this could enable the cells of the polarizing region to expand sufficiently to give rise to two digits (4 and 5) by self-organization (**Figure 2F**; Pickering and Towers, 2016). The specification of the same positional value during mouse limb development could occur if cells become refractory to the levels/duration of Shh signaling at a certain point (**Figure 2F**). In support of such a mechanism operating in the mouse, there is not a simple linear relationship between position and level of positive Shh signaling in the limb bud as expected in a classical positional information model (Ahn and Joyner, 2004). However, even though mouse digits 2–5 are morphologically similar, it is clear that they still have different identities, with the cells of digit 4 being characterized by having many more receptors for both testosterone and estrogen than digit 2 thus determining the sexual dimorphism in digit length (Zheng and Cohn, 2011). Indeed, digit 5 in particular, has quite a different morphology to the other digits. Taken together, even if the cells that give rise to mouse digits 2–5 are specified with the same positional value that is interpreted so that they have the same phalanx number, other factors operate to give the digits their individual morphologies and hence identities. Additional support for a model in which loss of Shh signaling can increase digit number and also result in posterior digits developing with anterior traits has been provided by work on the fore-limbs and hind-limbs of the amphibian Xenopus tropicalis. Inhibition of Shh signaling at a series of developmental stages resulted in forelimbs occasionally developing with five digits rather than four (Stopper et al., 2016). In addition, hind-limbs often developed terminal claws on all five digits whereas in normal development claws are only present on digits 1, 2, and 3. Additional work is required to determine if other characteristics of these posterior digits are anteriorised such as phalange number.

The work of Pickering and Towers further highlights the complex relationship between the polarizing region and the apical ridge already mentioned (Niswander et al., 1994), and the importance of short-range reciprocal signaling between these structures in the formation of posterior digits in particular as observed in the mouse limb (Zuniga et al., 1999; Bouldin et al., 2010). Thus, in the chick wing, Shh signaling inhibits the overlying apical ridge and the polarizing region fails to produce digits, yet in the mouse limb, the overlying apical ridge is less sensitive to Shh signaling than in the chick wing (see also (Bouldin et al., 2010), and in persisting posteriorly, allows two digits to form—the chick leg appears to have an intermediate relationship allowing one digit to form. Such dynamic interplay between the polarizing region and apical ridge could have contributed to patterns of posterior digit loss during limb evolution (see Section on Evolutionary aspects of Shh signaling in the limb).

## MECHANISMS OF Shh SIGNALING

As indicated in the models outlined above, positional values in developing limbs are specified by paracrine Shh signaling, in which Shh acts as a long-range graded signal and in a concentration/time dependent fashion, or by the duration of autocrine Shh signaling. Therefore, the crucial questions are how a graded distribution of Shh arises, how the range of Shh signaling is controlled and how cells measure the concentration of Shh and the duration of Shh signaling.

## Long-Range Shh Signaling and Gradient Formation

Studies in developing mouse limbs have revealed general mechanisms that modulate the distribution of Shh protein in tissues. One factor is the addition of lipids. Following its autocatalytic conversion, Shh is secreted by cells as a modified form of ShhN with cholesterol added at the C-terminus and a palmitoyl group (as part of a thiol ester) at the N-terminus (known as ShhNp; p indicating that ShhN is processed; reviewed (Lee et al., 2016). In limb buds of mouse embryos in which the C-terminal processing domain of Shh is conditionally deleted so that the polarizing region produces ShhN instead of ShhNp, ShhN spreads further across the limb bud and additional digits develop anteriorly (Li et al., 2006). It should be noted that previous analyses also suggested that cholesterol modification extends the range of paracrine Shh signaling. Thus, mice limbs expressing ShhN that lacks cholesterol failed to form digits 2 and 3 (Lewis et al., 2001) consistent with a role for paracrine Shh signaling in specifying these digits (Harfe et al., 2004). Other data however are consistent with cholesterol modification restricting the spread of Shh. Thus, mice deficient in SREBP-2 that encodes a sterol regulatory element binding protein that regulates cholesterol production failed to up-regulate Ptch1, consistent with impaired Shh transport (Vergnes et al., 2016). Similar studies on mutant mice that are unable to palmitoylate Shh show that this modification is essential for long range signaling (Chen et al., 2004). Intriguingly, cholesterol has also recently been shown to be the endogenous activator of Smoothened (Huang P. X. et al., 2016). Because cholesterol plays such important roles in Shh signaling, changes in the availability of cholesterol can impact on the development of the limb and might explain the subtle alterations in the spacing of the digits that have been observed in the limbs of mice with a mutation in a gene encoding a protein required for cholesterol metabolism (Schmidt et al., 2009) and in the limbs of rat embryos treated with triparanol, an inhibitor of cholesterol biosynthesis (Gofflot et al., 2003). The membrane protein Dispatched1 is required for paracrine signaling by cholesterol–modified Shh (Tian et al., 2005). The restriction of the spread of the ligand in a Dispatched1 mouse mutant resulted in the loss of a digit, which was interpreted as being digit 2, and as already mentioned, provided crucial evidence for the temporal expansion model (Harfe et al., 2004).

Another mechanism that influences the range of Shh signaling is the binding of Shh to cell surface and extracellular proteins. A generic response to Shh in all tissues is transcriptional upregulation of genes encoding cell surface proteins such as Ptch1 and Hhip that bind Shh. The resultant increase in their expression in response to Shh creates negative feedback loops, that not only limit the spread of Shh by sequestering it at the cell surface, but also, in the case of Ptch1, because it inhibits Smoothened activity, dampens activation of the Shh pathway. In mice in which Ptch1 is conditionally inactivated in the limbs (Butterfield et al., 2009), and therefore the signaling pathway is activated independently of Shh, the hind-limbs have extra digits, but the fore-limbs have fewer digits. This difference between hind-limbs and fore-limbs appears to be due to the timing of activation of the signaling pathway, which is earlier in the mutant fore- limbs (Zhulyn et al., 2014).

In contrast to Ptch1 and Hhip1, the genes Cdo (CAMrelated/downregulated by oncogene), Boc (brother of Cdo) and Gas1 (growth arrest specific 1) encoding membrane associated proteins that bind Shh, are expressed in the anterior region of early limb buds and their expression is negatively regulated by Shh. Analysis of limb development in single or double mouse mutants suggest that Gas1 and Boc sustain paracrine Shh signaling at a distance from the polarizing region (Allen et al., 2007). ShhNp can also bind to heparan sulfate proteoglycans and the distribution of these and other extracellular proteins in the developing limb will affect the distribution of Shh. In Drosophila, the hydrolase notum that cleaves glypicans, a subfamily of heparan sulfate proteoglycans, promotes high-level Hh signaling in the wing. Interestingly, in the chick wing bud, Notum was identified in microarray experiments as being downstream of Shh signaling (Bangs et al., 2010), suggesting possible functional conservation.

One way in which Shh could spread across the limb bud is by diffusion (see Muller et al., 2013, for discussion on mechanisms of morphogen transport), although it has been questioned whether simple diffusion would be a sufficiently robust mechanism to generate a stable concentration gradient (Kerszberg and Wolpert, 2007). Mathematical modeling however showed that specification of positional values for the three digits of the chick wing can be simulated by simple diffusion of Shh from the polarizing region (Woolley et al., 2014). In the model, based on the results of (Drossopoulou et al., 2000), Shh specifies the initial size of the domain that will give rise to the digits and then provides positional information. The model incorporates promotion of positional values in a dose-dependent fashion over the observed time frame in a growing domain of the correct dimensions as determined experimentally (Towers et al., 2008). The model can be extended successfully to the specification of the positional values in the chick leg, even though digit 4 arises from the polarizing region. However, it is unclear whether Shh levels in the polarizing region could reach the predicted concentration required to specify digit 4 (assumed to be double that required to specify digit 3) and whether indeed there is a simple graded response to Shh signaling in the leg. It is therefore more plausible that digit 4 is specified by length of time that cells express Shh. The model cannot however be extended further to simulate easily specification of the fifth digit of the mouse limb.

Live imaging of chick wing buds showed that Shh can be transported along the external surface of specialized filopodia (similar structures in insects are called cytonemes). These filopodia extend up to 150 microns away from the polarizing region and a similar distance away from the receiving cell (Sanders et al., 2013) equating to about 300 microns, the initial size of the chick wing digit-forming field (Vargesson et al., 1997; Towers et al., 2008). Thus, direct cell-cell contacts can span the required range of Shh signaling. Furthermore, Boc and Cdo have been visualized in discrete microdomains on a subset of filopodia extending from Shh-responding cells. However, it is not clear whether this transport mechanism could produce robust graded signaling and indeed whether filopodia are required. The involvement of filopodia could however explain the apparently anomalous finding that grafts of cells expressing a membranetethered form of Shh (generated by fusing the integral membrane protein CD4 to the C-terminus of ShhN) can duplicate digits in the chick wing (Yang et al., 1997).

## Measurement of Shh Concentration and Duration of Signaling

It has been proposed that limb bud cells respond to paracrine Shh signaling in a concentration dependent fashion although length of exposure to the Shh signal also plays a role. So how do cells measure the concentration of Shh? The mechanism depends on the Shh-dependent processing of full-length Gli proteins, which act as transcriptional activators; in the absence of Shh signaling, Gli proteins are processed to short forms, which act as transcriptional repressors (reviewed in Lee et al., 2016). In normal chick and mouse limb buds, anterior cells not exposed to Shh contain high levels of Gli repressor, while in the posterior region of the limb, there is a gradient in the ratio of Gli activator/Gli repressor, higher posteriorly than anteriorly, reflecting the response to the Shh gradient across this part of the limb (Wang et al., 2000). There are three Gli genes, Gli1, Gli2, and Gli3 with the protein encoded by Gli1 acting exclusively as an transcriptional activator as it does not undergo processing into a repressor form. While functional inactivation of Gli1 and Gli2 in mice has little effect on limb development (Mo et al., 1997; Park et al., 2000), when Gli3, the major contributor to transcriptional repression, is functionally inactivated, Shh is expressed anteriorly and several additional morphologically similar digits form anteriorly while posterior digits are less affected (Wang et al., 2000). Unexpectedly, the limbs of Gli3 and Shh double knockout embryos are identical to the Gli3−/<sup>−</sup> limb buds showing that the function of Shh in the limb is to relieve repression by Gli3 and allow a patterned set of digits to develop from the posterior part of the limb (Litingtung et al., 2002; te Welscher et al., 2002). In the mouse limb, the gradient of Gli3 activity could only specify at most digits 1, 2, and 3 because Gli3 is not expressed in the polarizing region itself (Buscher and Ruther, 1998). Instead the initial response to autocrine Shh signaling would have to be mediated by Gli2, and consistent with this hypothesis, removing the function of Gli2 in a Gli3 mutant background, thus effectively inactivating all Gli function, results in the digits appearing morphologically similar (Bowers et al., 2012). This suggests that Gli3 mediates the response of cells in the limb bud to paracrine Shh signaling and Gli2 to autocrine Shh signaling. It should also be noted that the digits that form in single Gli3−/<sup>−</sup> mouse limbs (and also in compound Shh−/−/Gli3−/<sup>−</sup> mouse limbs) are thinner and more closely spaced together than in normal limbs, suggesting that Gli3 plays a role in regulating the digit period (Sheth et al., 2012, see section Interaction Between Positional Information and a Turing-type Mechanism). 5′Hoxa/d function also seems to be involved since the progressive titration of 5 ′Hox genes in the Gli3−/<sup>−</sup> background increases digit number and decreases the digit period still further (Sheth et al., 2012).

Surprisingly, chemical mutagenic screens to identify mutations causing polydactyly in mouse identified genes required for formation and functioning of primary cilia (Huangfu et al., 2003; Weatherbee et al., 2009; Ashe et al., 2012). In such mutants, many morphologically similar digits develop and this is because Gli processing takes place on primary cilia in vertebrate cells. Thus, absence of cilia is equivalent to functional inactivation of all three Gli genes. The classical chicken mutant, talpid3, with a range of defects including polydactylous limbs (Ede and Kelly, 1964) was found to have a mutation in a gene encoding a centrosomal protein required for formation of a primary cilium (Davey et al., 2006), and functionally inactivating the talpid3 gene in a mouse limb, leads to the development of many morphologically similar digits (Bangs et al., 2011). Another chicken mutant, talpid2, with the same range of defects including polydactylous limbs, was found to have a mutation in a gene encoding another ciliary protein—C2CD3 (Chang et al., 2014).

For autocrine Shh signaling, the duration of signaling is the most important parameter. Timing appears to be a general way of specifying positional values, but how cells in embryos measure time is little understood. Interestingly, a timing mechanism involving a cell cycle clock has been proposed to specify proximodistal positional values in the chick wing bud (Saiz-Lopez et al., 2015), although the most proximal positional values may be specified by retinoic acid signaling (Cooper et al., 2011; Rosello-Diez et al., 2011). The molecular nature of intrinsic timers is currently unknown and presents a widespread problem in developmental biology.

## INITIATION OF Shh EXPRESSION

A key discovery in understanding how Shh expression is localized to the posterior margin of the limb bud was identification of a cisregulatory element that controls limb-specific expression (Lettice et al., 2002). Analysis of Sasquatch, an insertional mouse mutant with limb polydactyly, in which Shh was expressed anteriorly as well as posteriorly in the limb, showed that the exogenous DNA construct had serendipitously disrupted an enhancer (Sharpe et al., 1999). This 1.7 Kb enhancer, which has become known as the ZRS (zone of polarizing activity regulatory sequence), is unexpectedly located in intron 5 of the LMBR1 (limb region 1) gene, which is almost 1 MB upstream of the promoter of the Shh gene. It is still not clear why insertion of the transgene into this particular region of the ZRS in Sasquatch leads to anterior Shh expression in the limb bud. In contrast, deletion of the entire ZRS region in mouse embryos results in loss of Shh expression in the limb buds resulting in limb truncations similar to those found in mouse embryos lacking Shh function (Sagai et al., 2005). It should be noted however, that the many other defects seen in mouse embryos lacking Shh function, which reflect the widespread functions of Shh signaling in organogenesis, are not present in the mouse embryos in which the ZRS is deleted.

The ZRS is of general interest as an example of a long-range enhancer—a cluster of three similar long-range enhancers also regulates Shh expression in the epithelial linings of the pharynx, the lung and the gut respectively (Sagai et al., 2009). 3D FISH and chromatin configuration assays showed close associations between the ZRS and the Shh locus in mouse limb bud cells compared to cells from other tissues (Amano et al., 2009). Curiously, transcriptional activity was not seen in all polarizing region cells suggesting that the cells may express Shh in pulses. One possibility is that Shh is expressed periodically during the cell cycle. In support of this, Shh expression is lost in chick wing buds treated with aphidicolin—an inhibitor of progression through Sphase (Ohsugi et al., 1997). More recently FISH and chromatin configuration assays together with super-resolution microscopy have revealed that the Shh locus loops out of its chromosome territory to make contacts with the ZRS in polarizing region cells in the mouse limb bud at the time Shh expression is activated (Williamson et al., 2016).

The ZRS provides an excellent reference point for deciphering the gene network that controls Shh expression in the limb and contains binding sites for the transcription factors, Hand2 (heart and neural crest derivatives 2; (Galli et al., 2010) and 5′ Hoxd proteins. The genes encoding these transcription factors are expressed in the posterior region of the early limb bud and when they are deleted in the mouse limb, Shh is not expressed. Conversely, when Hoxd13 is expressed throughout the mouse limb bud, there is an ectopic Shh domain and polydactylous limbs result (Zakany et al., 2004).

Expression of Hand2 and Hoxd genes is restricted to the posterior part of mouse limb buds by Gli3. In the mouse fore-limb-forming region, Hand2 expression is also repressed anteriorly by the Hox5 paralogous group genes (Xua et al., 2013), while Hand2 expression in the posterior region of the fore-limb-forming region is dependent on the Hox9 paralogous group genes, thus providing antero-posterior polarity prior to the transcriptional activation of the Shh gene (Xu and Wellik, 2011). Recently, it has emerged that GATA family transcription factors also contribute to supressing anterior expression of Shh (Kozhemyakina et al., 2014) as conditional removal of Gata4/6 in limbs of mouse embryos results in pre-axial polydactyly. Two distinct mechanisms have been proposed. One is that GATA transcription factors in complex with FOG co-factors bind directly to the ZRS enhancer while the other is that GATA6 may interact directly with GLI3 to promote repression of the vertebrate Hedgehog pathway and this may explain the formation of an additional anterior digit in the hindlimb (Hayashi et al., 2016).

Shh expression in the polarizing region is also controlled by FGF signaling from the apical ridge and FGF signaling has been shown to regulate the expression of the genes encoding the ETS translocation variant transcription factors ETV4 and ETV5. The genes encoding these transcription factors are expressed beneath the entire extent of the apical ectodermal ridge and suppress Shh expression outside of the polarizing region. These ETV transcription factors bind directly to sites in the ZRS. In the polarizing region, posteriorly expressed ETS1/GABPα binds to other sites in the ZRS and over-rides this inhibition and allows expression of Shh (Lettice et al., 2012). Wnt7a signaling from the dorsal ectoderm also contributes to controlling Shh expression but the mechanism is not yet known (Yang and Niswander, 1995).

The activity of the ZRS not only determines the location of cells expressing Shh in the developing limb bud but also the size of the Shh expression domain. In addition, an autoregulatory mechanism has been discovered in which Shh controls the number of polarizing region cells by regulating the size of the posterior necrotic zone (Sanz-Ezquerro and Tickle, 2000) via BMP2 signaling (Bastida et al., 2009) Taken together these mechanisms have the crucial function of controlling the levels of Shh signaling.

Lastly, retinoic acid derived from the flank also appears to be required for initiating Shh expression in limb buds. Shh expression is greatly reduced in the limb buds of vitamin A deficient quails (Stratford et al., 1999) and in chick wing buds following treatment with inhibitors of retinoic acid synthesis (Stratford et al., 1996). Mouse embryos in which a gene encoding an enzyme that generates retinoic acid was functionally inactivated died early and lacked fore-limbs. When these embryos were provided with retinoic acid so that development can proceed further, Shh was not restricted posteriorly in the rescued fore-limb buds suggesting that retinoic acid plays a role in determining antero-posterior polarity prior to activation of Shh expression (Niederreither et al., 2002; Zhao et al., 2009).

## TERMINATION OF Shh EXPRESSION

The failure of the positive feedback loop between the polarizing region and the apical ectodermal ridge has been proposed to terminate the duration of Shh expression in the chick wing. In this model, Shh up-regulates Grem1 by paracrine signaling, but cells displaced from the polarizing region by proliferative expansion are then unable to up-regulate Grem1 (the apical ridge maintenance factor; Scherz et al., 2004). This is proposed to create a tissue barrier that results in Shh being no longer able to up-regulate Grem1 at a distance, leading to de-repressed BMP signaling suppressing Fgf4 expression in the apical ectodermal ridge, that in turn, leads to loss of Shh expression in the polarizing region (Scherz et al., 2004). Tbx2 is proposed to be the factor that suppresses the posterior up-regulation of Grem1 in and around the polarizing region (Farin et al., 2013). In the absence of Tbx2, Grem1 expression expands posteriorly resulting in prolonged Shh expression and extra tissue growth indicated by the bifurcation of digit 4. It is unclear why this only occurs in the hind-limbs of these Tbx2 knockout mice. An alternative model for the mouse limb is that increased FGF signaling inhibits Grem1 expression leading to termination of the feedback loop (Verheyden and Sun, 2008).

A clock linked with the cell cycle has also been shown to be involved in timing the duration of Shh expression in the polarizing region of the chick wing bud with the clock being set once retinoic acid concentrations fall below a certain level. Thus, tissue transplantation experiments have shown that the chick wing polarizing region intrinsically times the duration of Shh expression irrespective of the extrinsic signaling environment (Chinnaiya et al., 2014). Indeed, Shh expression has been shown to terminate on time if the separation of Grem1 and Shh expressing cells is prevented (Towers et al., 2008). Furthermore, the inhibition of Shh signaling with cyclopamine in the chick wing leads to the premature loss of Shh expression in the presence of an Fgf4-expressing apical ectodermal ridge and Grem1 expression extending into the posterior part of the wing bud, thus suggesting that Shh autoregulates its own transcription in the polarizing region (Pickering and Towers, 2016). The mechanism by which this is achieved has not yet been elucidated.

## RESPONSE TO Shh SIGNALING IN THE LIMB

Many studies have provided information about the expression of individual genes that are affected by Shh signaling in the limb. For example, changes in gene expression have been observed in chick limb buds treated with Shh or cyclopamine, and in mouse limb buds in which Shh or Gli3 is functionally inactivated, or in which Gli3 processing does not occur, e.g., mutants with defective cilia. Microarray analyses have been carried out in both chick and mouse limbs (Vokes et al., 2008; Bangs et al., 2010). It has been estimated from one microarray study that 10% of the genes expressed in the early limb bud (about 1,000 genes) are downstream of Shh signaling (Bangs et al., 2010). Putative direct targets of Gli3 repression have been identified by ChIP seq analysis of limb bud nuclear extracts using transgenic mice expressing a tagged form of the Gli3 protein (Vokes et al., 2008). Further analysis has involved RNAseq (Lewandowski et al., 2015).

Analysis of this information has begun to uncover the gene regulatory network underlying the response to the Shh signaling pathway in the limb in addition to the generic suite of genes that encode proteins that enable or modulate Shh signaling. The genes in the network include those that are expressed posteriorly either due to positive regulation by Shh or because Shh relieves Gli3 repression; also those that are expressed anteriorly either due to negative regulation by Shh or because they are downstream of Gli3 repression (Bangs et al., 2010). A study involving analysis of gene expression patterns in the limb buds of Shh−/−, Gli3−/<sup>−</sup> double mouse mutants indicated that the expression of nearly all the putative Gli target genes identified by ChIP seq in the posterior mesenchyme of E10.5 mouse limb buds depends on Gli repressor activity rather than Gli activator activity (Lewandowski et al., 2015).

One generic class of potential target genes already mentioned comprises genes encoding cell cycle regulators such as N-myc and Cyclin D1 that are predominantly expressed posteriorly and Cyclin D2 that is expressed in the polarizing region, and that are likely mediate the effects of Shh on proliferation (Towers et al., 2008; Welten et al., 2011). Shh has also been shown to promote vascularisation of the chick wing bud via regulating expression genes encoding pro-angiogenic factors such as VEGF (Davey et al., 2007). There is evidence in the mouse limb, that transcription factor genes including 5 ′ genes in the Hoxa Tickle and Towers Shh in Developing Limbs

and Hoxd clusters, Sall1, and Tbx2/Tbx3 are putative direct targets of Shh and would be predicted to encode the positional information conferred by the autocrine/paracrine Shh signaling (Vokes et al., 2008). Experiments with cultured mouse limb buds suggest that Shh signaling is required for robust and continued expression of 5′members of the Hoxd cluster (Panman et al., 2006; Lewandowski et al., 2015) while mis-expression of Tbx2 and Tbx3 genes in the chick leg bud in the embryo has been reported to change digit identity (Suzuki et al., 2004).

Other putative direct Gli3 targets are genes involved in BMP signaling; Gremlin encoding the apical ridge maintenance factor and Bmp 2 expressed together with Bmp7, in the posterior region of the early limb bud (Vokes et al., 2008). There is a close relationship between Shh and Bmp2 expression elsewhere in vertebrate embryos, which is also conserved in Drosophila. For instance in the Drosophila wing imaginal disc, Hh secreted from the posterior compartment induces expression of the Bmp2 orthologue, Dpp, that encodes a long range signaling molecule regulating position-dependent expression of transcription factors such as Spalt and Omb, orthologues of Sall1 and Tbx2/3 respectively. Experiments in chick wing buds show that Bmpsoaked beads placed at the anterior margin of a chick limb do not induce digit duplications (Drossopoulou et al., 2000). However, when a bead soaked in a BMP antagonist was implanted at the anterior margin of the wing bud following implantation of an Shh-soaked bead, a series of morphologically similar digits developed anteriorly suggesting that BMP signaling is involved in digit promotion (Drossopoulou et al., 2000). In chick leg buds, BMP signaling is graded across the tip of the bud at the stage at which the digit condensations form in the so-called phalanxforming region (PFR—Suzuki et al., 2008). Grafting interdigital tissue to different positions between digit condensations and manipulating BMP signaling alters the morphology of the digits in terms of phalange number suggesting that it is BMPs produced by interdigital regions that are directly responsible for realizing digit-specific morphology (Dahn and Fallon, 2000). Recently, evidence has been presented that interdigital signaling may also be involved in regulating the morphogenesis of the digit condensations in mouse limbs (Huang B. L. et al., 2016).

## CLINICAL ASPECTS OF Shh SIGNALING IN THE LIMB

The increasing understanding of the molecular basis of anteroposterior pattern formation has led to insights into congenital malformations that affect the limb. Unsurprisingly, defects in Shh function have been found to underlie several inherited disorders. In particular, these include polydactyly: pre-axial polydactyly in which additional digits arise from the thumb-side of the hand, and post-axial polydactyly in which the additional digits arise from the little finger-side (Biesecker, 2011). Often these conditions are associated with syndactyly (fusion of the soft tissues between the digits).

Alterations in the coding sequence of the SHH locus are not known to form the basis of any congenital malformation of the limb—presumably because such lesions are not compatible with the development of other tissues. However, point mutations in the ZRS enhancer that would be predicted to lead to ectopic SHH expression specifically in the limb bud are found in human patients with pre-axial polydactyly type 1 (PPD1—OMIM 174400) and triphalangeal thumb polysyndactyly syndrome (TPTPS OMIM 174500) (see review Hill and Lettice, 2013). In TPTPS, additional digits can arise post-axially as well as pre-axially, suggesting that the normal regulation of SHH expression at the posterior margin of the limb is also perturbed. It remains to be determined how these point mutations affect the regulation of endogenous SHH expression. One possibility is that the levels and/or duration of SHH expression are increased and these lead not only to an additional digit pre-axially but also to overgrowth of the polarizing region and its subsequent development into additional post-axial digits—perhaps by self-organization (see section on Interaction between positional information and a Turing-type mechanism). A point mutation at a particular position in the ZRS is associated with Werner mesomelic syndrome in which there are distal arm and leg bone defects in addition to extra digits (VanderMeer et al., 2014). Unexpectedly, duplications of the ZRS have also been reported in individuals with TPTPS as well as the related condition Haastype polysyndactyly (OMIM 186200). Microduplications of the ZRS have also been detected in patients with Laurin-Sandrow syndrome OMIM 13750); the limb phenotype of these patients overlaps with the Haas-type polysyndactyly phenotype but can be distinguished by mirror-image polysyndactyly of the feet and duplication of the fibula (Lohan et al., 2014). In contrast, patients with a deletion involving exon 4 and portions of introns 3 and 4 of the LMBR1 gene, a region distinct from the ZRS, have a condition known as acheiropodia (OMIM 200500) in which elements distal to the elbow/knee fail to form in all four limbs. This condition not only resembles the phenotype of the limb buds of mouse embryos lacking Shh function but also that of the limbs of Ozd mutant chickens in which it has now been shown that a large part of the ZRS sequence is deleted (Maas and Fallon, 2004). Inborn errors in cholesterol metabolism can lead to limb anomalies, as might be expected given the importance of cholesterol in Shh signaling as already discussed. For example, post-axial polydactyly is found in patients with Smith-Lemli-Opitz syndrome (OMIM 270400) in which a mutation deactivates the function of 7-dehydrocholesterol reductase, which is the final enzyme in the metabolic pathway that generates cholesterol. Post-axial polydactyly is also seen at low frequencies in patients with other syndromes in which cholesterol biosynthesis is altered (Gofflot et al., 2003). Why post-axial polydactlyly occurs however is not clear.

Defects in the response to Shh signaling are found in syndromes that include polydactyly. For instance, the Pallister-Hall (OMIM 146510, Hill et al., 2007) and Grieg Cephalopolysyndactyly (OMIM 175700- Kalff-Suske et al., 1999) syndromes present with pre-axial and post-axial polydactyly and are caused by mutations in the GLI3 gene. The effects of these mutations are likely due to the de-repression of the Shh signaling pathway in the anterior part of the limb. Since the processing of full-length Gli3, occurs in primary cilia, syndromes known as ciliopathies, in which cilia function/structure is compromised, include polydactyly as part of their spectrum of defects—examples being, Bardet-Biedl syndrome (BBS– OMIM 209900, Forsythe and Beales, 2013) and Meckel-Gruber syndrome (OMIM 249000, Shaheen et al., 2013). Recently mutations in the TALPID3 gene, required for formation of cilia have been discovered in patients with Joubert syndrome (OMIM 21330) although these patients rarely show limb defects (Roosing et al., 2015; Stephen et al., 2015). Homozygous mutations in the TALPID3 gene have however been found in families affected by lethal ciliopathies associated with polydactyly (Alby et al., 2015), phenotypes more akin to those of the homozygous chicken mutants already mentioned in which the talpid3 gene was first identified.

Several clinical conditions are associated with mutations in putative gene targets of Shh signaling in the limb (see previous section on Response to Shh signaling, also reviewed Pickering and Towers, 2014). Sall1 encoding a transcription factor is expressed in the posterior region of the early chick and mouse limb buds but more widely at the base of the digital plate at later stages (Buck et al., 2001; Fisher et al., 2011). Mutations in SALL1 that produce a truncated protein with dominant negative activity have been detected in patients with Townes-Brockes syndrome characterized in the limb by pre-axial polydactyly and triphalangeal thumb (Kohlhase et al., 1998). A transgenic mouse model in which a truncated SALL1 protein is produced mimics the human limb phenotype (Kiefer et al., 2003). Inactivating mutations in the gene encoding the transcription factor Tbx3, which is expressed at high levels in stripes at both anterior and posterior margins of early chick and mouse limb buds (Tumpel et al., 2002; Emechebe et al., 2016) are seen in patients with Ulnar-mammary syndrome (OMIM 181450); the defects affect the development of posterior structures in the upper limb and include missing ulna, missing posterior digits and post-axial polydactyly. The same limb phenotype is seen in mouse Tbx3 mutant embryos (Davenport et al., 2003; Emechebe et al., 2016). Finally mutations in HOXD13 are associated with many clinical conditions in which there are digital abnormalities including polydactyly, syndactyly (fused digits) and brachydactly (short digits). Hoxd13 is another putative gene target of Shh signaling identified in the mouse limb and is expressed in the posterior region of early chick and mouse limb buds and then throughout the digital plate at later stages (Nelson et al., 1996). A complex spectrum of mutations in HOXD13-polyanaline tract expansions, truncating mutations and point mutations leading to amino acid substitutions have been identified (reviewed Goodman, 2002). Hoxd13 is likely to have several roles in digit development and the challenge is to understand how a particular genetic change leads to a particular phenotype.

## Shh SIGNALING AND LIMB REGENERATION

Adult urodele amphibians (newts and salamanders) can regenerate their limbs after amputation. Shh signaling occurs in adult urodele limbs during regeneration and understanding how Shh expression is activated in these adult tissues may be relevant in the context of stimulating growth and repair of tissues in damaged limbs. Following amputation of a newt limb, a mound of undifferentiated cells called the blastema forms at the stump surface and proliferation of blastemal cells replenishes the missing limb structures. Shh is expressed in posterior part of the newt limb blastema recapitulating embryonic expression in the limb bud (Imokawa and Yoshizato, 1997), and when regenerating salamander limbs were treated with cyclopamine, only one digit-like structure formed—similar to hind-limbs of Shh mutant mice (Chiang et al., 1996). Recently, it has been demonstrated that Shh, which is expressed in the posterior part of the salamander blastema is part of a reciprocal feedback loop via Grem1 and Fgf8 that are expressed in the anterior part of the blastema (Nacu et al., 2016). This feedback loop is required for outgrowth of the blastema and closely recapitulates the epithelial-mesenchyme signaling network that drives embryonic limb development. The demonstration that two signals, which can act at a distance—Shh and Fgf8—drive limb regeneration is at odds with a long standing model in which direct cell-cell interactions stimulate intercalary growth to even out disparate positional identities between anterior and posterior parts of the blastema (French et al., 1976). The size of the limb blastema is about 10 times that of embryonic limb buds, therefore it is not clear whether these signals could indeed act over the large distances involved.

Fate maps of the blastema showing which cells give rise to the digits and experiments addressing timing of specification of antero-posterior positional values could give important insights into whether digit regeneration is comparable to embryonic development. One possibility is that cells within a blastema maintain memory of their position along the antero-posterior axis and restore missing structures by a timing mechanism linked to proliferation. Evidence for such a cellular memory based on epigenetic modifications has been obtained in regenerating limb buds of Xenopus embryos (Hayashi et al., 2015). A timing model would dispense with difficulties in scaling long range gradients over considerable distances to restore missing positional values during regeneration and the role of Shh and Fgfs would be to maintain the outgrowth and the width of the blastema. It would also be useful to know the fate of polarizing region cells from embryonic urodele limb buds in adult limbs and regenerating limbs.

Unlike urodeles, anuran amphibians can only regenerate their limbs during embryonic stages. Interestingly, increased methylation of the ZRS enhancer during Xenopus development correlates with reduced capacity to regenerate the limb in the adult suggesting that epigenetic mechanisms limit this process by preventing re-expression of Shh (Yakushiji et al., 2007).

## EVOLUTIONARY ASPECTS OF Shh SIGNALING IN THE LIMB

The ZRS element located in the fifth intron of Lmbr1 gene that drives limb-specific Shh expression is well conserved at the sequence level in many vertebrates. The ZRS is an excellent candidate for evolutionary modifications that have resulted in changes in limb morphology because the rich diversity of limb morphologies could have evolved without affecting other features of the body plan. In support of this, mutations in the ZRS at a conserved ETS1 binding site in pythons have been described that appear to be responsible for the early loss of Shh expression and subsequent failure of limb bud outgrowth (Kvon, 2016; Leal, 2016). CRISPR/CAS9 gene editing approaches, in which the mouse ZRS was replaced by the python ZRS sequence, resulted in limb truncations similar to those obtained upon the complete removal of Shh function in the mouse limb (Kvon, 2016). As in pythons, Shh fails to be up-regulated in the hind-limbs of the spotted dolphin and is associated with reduced outgrowth, although the molecular basis of this has not been examined (Thewissen et al., 2006). Many described ZRS mutations to date, however, result in ectopic expression of Shh in the anterior part of the limb, and therefore the development of additional digits as in domesticated animals; for instance, Dorking's (Bouldin and Harfe, 2009) and Silkie chickens (Dunn et al., 2011) have an additional anterior digit in the leg and dogs and cats (notably Hemingway cats) have extra anterior digits in their fore-paws (Lettice et al., 2008).

Limbs with more than five digits have not been selected for during evolution suggesting there is little benefit in increasing digit number. Interestingly, the limbs of the earliest Devonian tetrapods such as Acanthostega and Icthyostega had up to eight digits (Clack, 2002). The mechanism by which such digit patterns would have been specified is of considerable interest. In having several digits, the limbs of such tetrapods superficially resemble the limbs of mouse Gli mutants, which have many digits that form by self-organization. However, the digits in these Devonian tetrapods display differences in phalangeal number suggestive of antero-posterior positional values specified by Shh in the early limb bud. Once pentadactyly was established in tetrapods, this has remained the basic plan, although occasionally limbs with so-called "sixth digits" have evolved. These sixth digits are in fact, adaptations of other limb bones, such as the overgrown wrist bone in the case of the mole's "paddle-like" limb (Mitgutsch et al., 2012). The chick leg has retained the basic pentadactyl phalangeal pattern in digits 1–4 and therefore is of special interest to the evolution of digit patterns. As we discussed earlier, a model in which Shh signaling specifies different positional values is sufficient to explain chick leg patterning. Thus, any deviations away from this model in the mouse limb would therefore suggest a derived mode of patterning digits 1–4 in the mammalian lineage.

Digit loss has commonly occurred over the course of evolution and alterations in Shh expression and response to Shh have been implicated. A striking example is seen in the wings of birds and the fore-limbs of their basal theropod dinosaur ancestors in which two digits have been lost during evolution (Sereno, 1999). Understanding this mode of digit loss has puzzled investigators for over 150 years because theropods appeared to have had digit identities 1, 2, and 3, but in the embryo at least, bird digits appear to arise from positions 2, 3, and 4 (Burke and Feduccia, 1997). Therefore, it was suggested in the so-called "frameshift" model that digits with the identities 1, 2, and 3 arise from positions 2, 3, and 4 of the bird wing (Wagner and Gauthier, 1999; Tamura et al., 2011), perhaps due to reduced Shh signaling levels/duration in limbs of the theropod ancestors of birds (Vargas and Wagner, 2009). However, the Green Fluorescent Protein fate-mapping experiments in chick wings (see **Figure 1c**) showed that in fact digits with the identities 1, 2, and 3 arise from embryonic positions 1, 2, and 3 that are found in tissue adjacent to the polarizing region (Towers et al., 2011). Therefore, it is not necessary to invoke a frameshift and suggests that the digits 4 and 5 of the dinosaur hand were simply lost and that bird wing digits should be numbered 1, 2, and 3 in line with the fossil record, as is now generally accepted. As already mentioned, in the chick wing bud, the posterior necrotic zone overlaps with the polarizing region. In the chick wing bud, the posterior necrotic zone is much larger than the corresponding zone in chick leg and mouse limb buds (Fernandez-Teran et al., 2006). Therefore, the loss of the two posterior digits in birds might be based on evolutionary changes in Shh signaling, in particular the autoregulatory mechanisms by which Shh signaling regulates apoptosis in the posterior necrotic zone of the wing bud (Sanz-Ezquerro and Tickle, 2000) and also proliferation (Chinnaiya et al., 2014). Interestingly, a recent study showed that an extension of the posterior part of the apical ectodermal ridge in the absence of Shh signaling was sufficient to enable the polarizing region to give rise to a digit in the chick wing. In such buds, the posterior necrotic zone was lost and this was accompanied by a dramatic increase in proliferation of polarizing region cells (Pickering and Towers, 2016).

Shh has also been implicated in digit loss in cow limbs in which only two digits form (3 and 4). It was revealed that Ptch1 is expressed in the very posterior of the bud and at low levels in response to Shh signaling, because of the degeneration of a cisregulatory enhancer. As a consequence, it is suggested that Shh fails to be sequestered and restricted to the posterior part of the cow limb bud resulting in more-or-less uniform Shh signaling which results in symmetrical and distally restricted anteroposterior gene expression patterns (Lopez-Rios et al., 2014). As a result, the two digits of the cow limb are also symmetrical and lateral digits are lost because the apical ectodermal ridge fails to extend sufficiently to support their outgrowth. Similarly, Ptch1 is also restricted to the posterior of the limb buds of pigs that develop four digits, two of which are prominent (digits 3 and 4; Cooper et al., 2014). However, camels do not display a posterior restriction and down-regulation of Ptch1 in their developing limb buds although they also produce two digits (3 and 4), suggesting another mechanism of digit loss (Cooper et al., 2014). An additional case of digit loss involves the limbs of different species of the Australian skink, Hemiergis (Shapiro, 2002). The shortened duration of Shh expression in these lizards correlates well with the extent of digit reduction—species with five digits express higher levels of Shh for a longer time than those with only two digits (Shapiro et al., 2003). Interestingly, digit reduction correlates with a reduction in cell proliferation. One possibility is that factors other than reduced Shh signaling could be involved. As yet no mutations have been reported in ZRS sequences of various Hemiergis clades. However, as further studies are required to understand how positional values are specified by Shh signaling in mammals and lizards, this means that it is difficult to interpret some of the patterns of evolutionary digit loss discussed in this section.

## FUTURE DIRECTIONS

It is now established that Shh has a pivotal function in vertebrate limb development and many details have been uncovered. Surprisingly however, there is still no consensus about how Shh specifies antero-posterior positional values in the limb. It remains possible that different combinations of transcription factors govern antero-posterior positional values, but it has been difficult to identify them because all the digits are made up of the same differentiated cell types. Therefore, a generegulatory network such as one operating downstream of Shh in the neural tube to specify distinct neural fates is unlikely to operate during limb development (Balaskas et al., 2012). It is also likely that the temporal regulation of the same sets of genes could contribute to specifying positional values. For instance, there is a clear relationship between Hoxd expression and thumb (digit 1) development, with cells that give rise to thumb the only cells that express Hoxd13 and not Hoxd12 (Vargas and Fallon, 2005). Therefore, since the cells that give rise to all the other digits express Hoxd12 and Hoxd13, a simple Hox code is unlikely to specify the digits, and perhaps timing of expression is the important determinant. Another challenge is to understand how the positional information conferred by Shh signaling is remembered and then interpreted so that digits with different identities arise in the proper places in the limb. In chick limbs, it is clear that the concentration/duration of Shh is sufficient to specify digit identity, however, this is not readily apparent in mammalian limbs because the digits are morphologically similar—at least in terms of phalangeal number. It will be

## REFERENCES


important to fill this gap in knowledge in order to apply the principles to developing human limbs and gain deeper insights into the basis of congenital limb defects and to evolutionary alteration in digit pattern. The analysis of the function of Shh in new animal models of limb development could help resolve issues regarding the relationship between positional values and digit identity. Further development of the CRISPR/Cas9 system should facilitate this.

An issue of general relevance is the mode of Shh transport in the limb and how a graded distribution of Shh is established. This may require further refinement of in vivo imaging techniques to visualize directly the distribution of Shh in real time. It also seems clear that the timing of Shh expression is another critical parameter that still needs to be addressed. Disentangling the relationship between autoregulatory mechanisms of intrinsic timing of Shh expression and extrinsic mechanisms could shed light on processes that ensure robustness of limb development and pattern scaling between different species.

## AUTHOR CONTRIBUTIONS

MT and CT jointly wrote the manuscript, MT formatted the Figures.

## ACKNOWLEDGMENTS

MT is supported by the Wellcome Trust (Grant number 202756/Z/16/Z).


and digit number in the vertebrate limb. Development 136, 3515–3524. doi: 10.1242/dev.037507


and Laurin-Sandrow syndrome. Clin. Genet. 86, 318–325. doi: 10.1111/cge. 12352


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

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

# Sox9 Expression in Amniotes: Species-Specific Differences in the Formation of Digits

Juan A. Montero<sup>1</sup> , Carlos I. Lorda-Diez <sup>1</sup> , Javier Francisco-Morcillo<sup>2</sup> , Jesus Chimal-Monroy <sup>3</sup> , Juan A. Garcia-Porrero<sup>1</sup> and Juan M. Hurle<sup>1</sup> \*

<sup>1</sup> Departamento de Anatomía y Biología Celular and IDIVAL, Universidad de Cantabria, Santander, Spain, <sup>2</sup> Departamento de Anatomía, Biología Celular, y Zoología, Universidad de Extremadura, Badajoz, Spain, <sup>3</sup> Departamento de Medicina Genómica y Toxicología Ambiental, Instituto de Investigaciones Biomédicas, Universidad Nacional Autónoma de México, Mexico City, Mexico

#### Edited by:

Juan Jose Sanz-Ezquerro, Consejo Superior de Investigaciones Científicas, Spain

#### Reviewed by:

Juan Carlos Izpisua Belmonte, Salk Institute, USA Alexander Omar Vargas, Universidad de Chile, Chile Karen Sears, University of Illinois at Urbana–Champaign, USA

#### \*Correspondence:

Juan M. Hurle hurlej@unican.es

#### Specialty section:

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

> Received: 03 January 2017 Accepted: 07 March 2017 Published: 23 March 2017

#### Citation:

Montero JA, Lorda-Diez CI, Francisco-Morcillo J, Chimal-Monroy J, Garcia-Porrero JA and Hurle JM (2017) Sox9 Expression in Amniotes: Species-Specific Differences in the Formation of Digits. Front. Cell Dev. Biol. 5:23. doi: 10.3389/fcell.2017.00023 In tetrapods the digit pattern has evolved to adapt to distinct locomotive strategies. The number of digits varies between species or even between hindlimb and forelimb within the same species. These facts illustrate the plasticity of embryonic limb autopods. Sox9 is a precocious marker of skeletal differentiation of limb mesenchymal cells. Its pattern of expression in the developing limb has been widely studied and reflects the activity of signaling cascades responsible for skeletogenesis. In this assay we stress previously overlooked differences in the pattern of expression of Sox9 in limbs of avian, mouse and turtle embryos which may reflect signaling differences associated with distinct limb skeletal morphologies observed in these species. Furthermore, we show that Sox9 gene expression is higher and maintained in the interdigital region in species with webbed digits in comparison with free digit animals.

#### Keywords: limb development, interdigit regression, chondrogenesis, skeletal progenitors, SOX9 transcription factor

The limb is an excellent model system to study the molecular basis of morphogenesis (Hinchliffe, 2002; Fabrezi et al., 2007). The skeletal pattern of the limb is conserved in tetrapods, yet differences in bone morphology are remarkable among different species (Kavanagh et al., 2013). Interpretations of skeletal limb diversification has been largely based on comparative developmental studies using histochemical or radiolabeling markers of initial stages of cartilage differentiation. From these approaches it has been proposed that the limb skeleton in tetrapods is generated by sequential branching and segmentation of a basic pattern representative of the distal segment of the fish fins, termed the "metapterygial axis." The advent and progress of molecular biology has provided new insights about the diversification of the limb skeletal morphology. For example, it has been shown that activation of signals responsible for skeletogenesis may be differentially regulated by transcriptional enhancer DNA sequences that are species-specific (Kvon et al., 2016). These studies explain major skeletal differences in evolutionary distant species such as the absence of limbs in snakes. However, differences between the fore- and the hind-limb in the same species or skeletal differences observed among closely related tetrapods might be regulated in a different fashion, such as timing differences in the expression of signaling molecules (Richardson et al., 2009; Moore et al., 2015; Zuniga, 2015).

Sox9 is a well known marker of the skeleton that precedes the appearance of cartilage blastemas (Wright et al., 1995; Healy et al., 1999; Chimal-Monroy et al., 2003; Kawakami et al., 2005; Lorda-Diez et al., 2011; Sensiate et al., 2014). Hence, Sox9 is expressed even in domains that represent skeletal pieces lost in the course of evolution of specialized species (de Bakker et al., 2013). Silencing Sox9 in mouse embryos causes loss of appendicular skeleton and increases programmed cell death (Akiyama et al., 2002). Sox9 overexpression promotes polydactyly (Akiyama et al., 2007). Furthermore, Sox9 along with BMP and WNT signaling are considered key regulators of digits formation through a self-organizing Turing mechanism (Raspopovic et al., 2014). Overall, such findings make Sox9 an excellent marker to detect signaling differences, later transduced into specific patterns of chondrification (Richardson et al., 2009), responsible for variations in the morphology of the appendicular skeleton. Based on the observation of in situ hybridizations, we have revised the pattern of Sox9 gene expression during digit development in reptilian (Mauremys turtle), avian (chick and duck), and mammal (mouse) species with different autopodial morphology to uncover signaling differences of potential interest to explain digit morphogenesis.

In chick embryos the expression of Sox9 shows differences between the wing (**Figures 1A–G**) and the leg bud (**Figures 1H–M**). In wing buds at stage HH22 (3.5 id) the expression of Sox9 marks the primary axis of the appendicular skeleton. In next stages, the initial domain extends proximally and distally (**Figures 1A–C**). Proximally, the domain forms the humerus primordium, and distally it shows a branching that establishes the primordium of the radius (**Figures 1C,D**). By stage HH24 (4 id) the primary axis is continued distally by the digital arch oriented toward the anterior margin of the bud. Between stages HH26-HH28 the digital arch undergoes a branching process to form each digit (**Figures 1E–G**). First branching forms digit 3 and a common branch that bifurcates to form digit 4, and a reduced domain reminiscent of a digit 5. The latest, is progressively reduced in size and expression intensity. The most anterior digit, is formed distally and aligned with the radial domain (**Figures 1F,G**).

In the leg bud the initial expression of Sox9 at stage HH22 appears divided into a posterior (primary axis) and an anterior domain for the tibia (**Figures 1H,I**). The femur is identifiable at stage HH25 coupled between the proximal end of the fibular and tibial domains (**Figure 1K**). The appearance of these skeletal domains at stage HH23 is accompanied by the formation of a nascent digital arch that occupies a posterior and distal position (**Figures 1K,L**). Initially, the expression is uniform and limited to the posterior half of the autopod but, in the following stages (HH25 and HH26), the expression progresses anteriorly and digits became identifiable as patches of higher expression (**Figures 1J–M**). Digits 3 and 4 are the most prominent at these stages while digits 2 and 5 are poorly defined areas where the expression of Sox9 is not very intense. Interestingly, the most anterior part of the autopod lacks Sox9 transcripts until stage HH26-HH27.

Both in the wing and in the leg bud, concomitantly with the intensification of Sox9 expression at stage HH26 in the digit blastemas, a carpal/tarsal arch of lower Sox9 expression level is formed. Carpal and tarsal pre-cartilages are individualized when digit blastemas are defined.

Expression of Sox9 in the mouse is similar in fore- and hind-limbs (**Figures 1N–S**). Initial expression of Sox9 occupies the whole central region of the early bud (**Figures 1N,O**).

Regionalization of this domain in stylopod, zeugopod and digital arch is due to the loss of transcripts from the central region at 10.5 pc (**Figures 1P,Q**). Due to this process, expression of Sox9 appears as a loop where the distal curved region constitutes the digital arch. The proximal part of the loop lengthens marking the position of the stylopod. The zeugopodial elements are identified as the lateral regions of the loop. In next stages the digit primordia appear as elongated domains of intensified Sox9 gene expression (**Figures 1R,S**). Digits 3 and 4 are the first to appear.

The skeletal domains of Sox9 in the Mauremys turtle are similar in fore and hind-limbs (**Figure 2**). At the beginning, a central ill-defined domain is transformed into a triangular domain with a posterior elongated vertex, which marks the stylopod (**Figure 2C**). The sides of the triangle form the zeugopodial domains, and the base corresponds with the digital arch. The expression of Sox9 in the digital arch becomes progressively intensified at discrete regions to form digit primordia (**Figures 2D,E**). Digits 3 and 4 are the most precociously identifiable while digit 1 is the last to appear, preceded by digit 5 (**Figures 2E–I**). In the course of digit development, the expression of Sox9 is progressively restricted to the digit tip and to the developing joints (**Figures 2E–I**).

Remarkably, the Mauremys turtle interdigital regions retain considerable levels of Sox9 expression not observed in chick and mouse embryos (**Figures 2F–I**). To ascertain if interdigital expression of Sox9 associates with the absence of interdigit

remodeling in the Mauremys turtle (**Figure 2A**), we compared the level of expression of Sox9 in the third interdigit of the leg bud of chick and duck embryos, as characteristic models of species with free and webbed digits respectively. As shown in **Figures 2J–L**, expression of Sox9 in the non-regressing interdigit of the duck was much higher than that of the chick embryo.

Detailed analysis of phylogenetically related but phenotypically different species have provided important cues about the mechanisms accounting for limb morphogenesis (Moore et al., 2015). Gene expression computational modeling have also provided insights on the molecular bases responsible for differences in limb skeletogenesis among vertebrates (Uzkudun et al., 2015). The consideration of the subtle Sox9 expression differences highlighted in this "perspective" assay, are consistent with heterochrony detected in the stages of chondrification (Richardson et al., 2009), and may help to improve our understanding of how digits differ in morphology. In all species, the expression of Sox9 marks the successive appearance of the stylopod, zeugopod, and autopod along the proximo-distal axis of the limb. The autopod includes the mesopodium (carpi/tarsi) and the acropod (digits); the specification of zeugopod and the acropod has been proposed to determine the mesopodial intermediate domain in between (Woltering and Duboule, 2010). Consistent with this hypothesis, the appearance of intensified expression of Sox9 marking the nascent digits precedes that of the carpal/tarsal domains that lie in the concavity of the digit arch.

The formation and expansion of the digit arch in the chick embryo is clearly distinct from that observed in mouse and turtle embryos. Consistent with the evolutionary model proposed by Shubin and Alberch (1986), in the avian limb the progressive appearance of the digit expression domains follows a polarized sequence from posterior to anterior, which is more accentuated in the wing bud. In contrast, the digital arch domain in mouse and turtle limbs appears occupying a central position in the autopod, and in the course of development expands uniformly to the margins of the bud. These differences raise doubts about the validity of current thought, which considers independent identities for each of the digits in the hand/foot of vertebrates. The consideration of such identities have implications for evolutionary hypothesis that consider digit 1, as the most distal element of a conserved skeletal axis modified in the course of evolution through branching and segmentation processes (see Cohn et al., 2002, for discussion). Sox9 expression domains precede the appearance of prechondrogenic blastemas that were formerly employed in traditional comparative embryonic studies. Hence, the differences among species observed here, support mechanisms of skeletal diversification based on the combination of a distinct distribution of signals with differences in the intrinsic properties of the skeletal progenitors of the autopod, likely associated with differential epigenetic signatures (Sheth et al., 2016). Both in the wing and in the leg of avian embryos, digits are different along the antero-posterior axis justifying the consideration of different digit identities according to their position and number of phalanxes. In contrast, digits of mouse and turtle embryos, expands from the centrally located digital arch toward the margins of the limb bud. In this model of digit arch expansion, there are not morphological landmarks that allow to assign specific identities to the central digits (2,3,4). It must be taken into account that the carpal/tarsal domains of Sox9 appear when the digit arch shows independent digit domains. Therefore, at these embryonic stages mesopodial domains cannot be taken as a primary reference to establish the identity of the digits. The only morphological differences observed among the pre-cartilaginous blastemas are located in the marginal digits (digits 1 and 5) where Sox9 domains exhibit a reduced size and appear at more advanced stages than the central digits.

The growth of the limb bud is regulated by a complex signaling network (Uzkudun et al., 2015), where Shh and Gremlin1 genes play an important role in digit specification (Sanz-Ezquerro and Tickle, 2003a; Zhu et al., 2008). Evolutionary or genetic deregulations of the Shh/Gremlin loop causes polydactylous (Norrie et al., 2014) or oligodactylous (Lopez-Rios et al., 2014) autopods. Consistent with our interpretation, central digits in these mutants, regardless of its number, are identical and indistinguishable from each other (Norrie et al., 2014). These findings make plausible that digit formation result of the selforganization of the limb mesenchyme (Cooper, 2015), within an autopod of dimensions and shape finely tuned by regulatory genes responsible for growth (Zhu et al., 2008).

Sox9 is target of signals controlling proliferation and differentiation of the skeletal progenitors, including FGFs, BMPs, TGFbetas, and Retinoic acid (RA). These signals are themselves closely regulated by the AER and the ZPA, to establish the pattern of limb skeletogenesis as well as the number of digits in the autopod. BMPs up-regulate the expression of Sox9 and promote differentiation of progenitors (Lorda-Diez et al., 2014; Norrie et al., 2014) and in conjunction with TGFβs and Activins induce the formation of extra-digits in the avian limb (Chimal-Monroy et al., 2003; Montero et al., 2008). FGFs are major determinants of digit size (Sanz-Ezquerro and Tickle, 2003b; Seki et al., 2015). FGFs inhibit chondrogenesis but expand the amount of Sox 9 positive skeletal progenitors and its overexpression in the limb results in the formation of extra cartilages, including extra-digits (Montero et al., 2001; Norrie et al., 2014), or extra-phalanxes (Sanz-Ezquerro and Tickle, 2003b). RA is a potent inhibitor of Sox9 gene expression (Weston et al., 2002) and RA inhibition in the autopod causes the formation of extra digits (Rodriguez-Leon et al., 1999). Hence, the pattern of Sox9 gene expression may reflect differences in the spatial distribution of signals within the limb bud mesoderm. According with this interpretation, avian digits may represent an evolutionary specialization of digit development consequence of a posterior polarization of signals responsible for limb outgrowth. In contrast, the pentadactyl autopod of mouse and turtle embryos may result from the uniform expansion (like opening a fan) of the signals that coordinate proliferation and differentiation of the skeletal progenitors.

## ETHICS STATEMENT

The animal care and handling, and all the experimental procedures were in accordance with the guidelines of the European Communities Council and the Spanish legislation and they were approved by the Service of Animal Health and Welfare of the Regional Government of Cantabria (Reference No. PI-10-15).

## AUTHOR CONTRIBUTIONS

Conceived and designed the experiments: CL, JF, JG, JC, and JH. Performed the experiments: CL, JM, JF, JC, and JH. Analyzed

## REFERENCES


the data: CL, JM, JG, JC, and JH. Writing of the manuscript: JM and JH.

## ACKNOWLEDGMENTS

Thanks are due to Prof. Virginio García-Martínez for help and suggestions. Supported by Grants BFU2014-54026P from the Spanish Science and Innovation Ministry to JH and NextVal NVAL15-07 from the IDIVAL to CL. **Figure 1L** has been previously published by Chimal-Monroy et al. (2003).

in the embryonic limb. PLoS ONE 6:e24546. doi: 10.1371/journal.pone.0 024546


progenitor differentiation. J. Cell Biol. 158, 39–51. doi: 10.1083/jcb.200 112029


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

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

# CaMKII Signaling Stimulates Mef2c Activity In Vitro but Only Minimally Affects Murine Long Bone Development in vivo

Chandra S. Amara, Christine Fabritius, Astrid Houben, Lena I. Wolff and Christine Hartmann\*

Department Bone and Skeletal Research, Medical Faculty of the University of Münster (WWU), Institute of Experimental Musculoskeletal Medicine, Münster, Germany

The long bones of vertebrate limbs form by endochondral ossification, whereby mesenchymal cells differentiate into chondrogenic progenitors, which then differentiate into chondrocytes. Chondrocytes undergo further differentiation from proliferating to prehypertrophic, and finally to hypertrophic chondrocytes. Several signaling pathways and transcription factors regulate this process. Previously, we and others have shown in chicken that overexpression of an activated form of Calcium/calmodulin-dependent kinase II (CaMKII) results in ectopic chondrocyte maturation. Here, we show that this is not the case in the mouse. Although, in vitro Mef2c activity was upregulated by about 55-fold in response to expression of an activated form of CaMKII (DACaMKII), transgenic mice that expressed a dominant-active form of CaMKII under the control of the Col2a1 regulatory elements display only a very transient and mild phenotype. Here, only the onset of chondrocyte hypertrophy at E12.5 is accelerated. It is also this early step in chondrocyte differentiation that is temporarily delayed around E13.5 in transgenic mice expressing the peptide inhibitor CaM-KIIN from rat (rKIIN) under the control of the Col2a1 regulatory elements. Yet, ultimately DACaMKII, as well as rKIIN transgenic mice are born with completely normal skeletal elements with regard to their length and growth plate organization. Hence, our in vivo analysis suggests that CaMKII signaling plays a minor role in chondrocyte maturation in mice.

Keywords: CaMKII, peptide inhibitor, chondrocyte maturation, hypertrophy, Mef2c, mouse model

## INTRODUCTION

Endochondral ossification is the process underlying the formation of the long bones in the vertebrate limbs (Erlebacher et al., 1995; Kronenberg, 2003). It starts with the condensation of mesenchymal cells that undergo chondrogenic differentiation, forming a cartilage template consisting of immature chondrocytes. These produce an extracellular matrix composed of proteoglycans, glycosaminoglycans, and glycoproteins. This template, which prefigures the future skeletal element, enlarges through chondrocyte proliferation (Akiyama and Lefebvre, 2011). In the next phase, chondrocytes in the center of the cartilage anlage stop proliferating and differentiate into prehypertrophic, Indian hedgehog (Ihh) expressing chondrocytes. Prehypertrophic chondrocytes enlarge further, becoming hypertrophic chondrocytes, which

#### Edited by:

Sigmar Stricker, Free University of Berlin, Germany

#### Reviewed by:

Joy Richman, University of British Columbia, Canada Pavel Krejci, Masaryk University, Czechia

\*Correspondence: Christine Hartmann christine.hartmann@ukmuenster.de

#### Specialty section:

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

Received: 25 October 2016 Accepted: 02 March 2017 Published: 16 March 2017

#### Citation:

Amara CS, Fabritius C, Houben A, Wolff LI and Hartmann C (2017) CaMKII Signaling Stimulates Mef2c Activity In Vitro but Only Minimally Affects Murine Long Bone Development in vivo. Front. Cell Dev. Biol. 5:20. doi: 10.3389/fcell.2017.00020 initiate synthesis of extracellular matrix containing type X collagen, encoded by the Col10a1 and Col10a2 genes (Kronenberg, 2003). The transition of chondrocytes from proliferating to prehypertrophic and then to hypertrophic cells is a critical step in determining the growth rate and size of skeletal elements (Kronenberg, 2003).

Previous studies in chicken and mouse identified a complex network of signaling pathways and transcription factors that regulate the different steps during endochondral bone formation (reviewed in Karsenty, 2008; Hartmann, 2009; Lefebvre and Bhattaram, 2010). A central regulatory node in the chondrocyte differentiation program is the Ihh-PTHrP (parathyroid hormone-related peptide) feedback loop (Vortkamp et al., 1996). Ihh, which is part of this feedback loop, is considered a master regulator of chondrocyte maturation and has multiple functions (Kronenberg, 2003; Mak et al., 2008). Numerous transcription factors such as Runx2 and Runx3, Mef2c, and Mef2d, as well as transcriptional cofactors such as β-catenin, promote chondrocyte hypertrophy (Inada et al., 1999; Kim et al., 1999; Hartmann and Tabin, 2000; Yoshida et al., 2004; Arnold et al., 2007; Guo et al., 2009).

It is not yet well established how the activity of these transcription factors is regulated by signaling events. Ca2+/calmodulin-dependent kinase II (CaMKII), is a calmodulin (CaM) binding serine/threonine kinase and important for Ca2+-mediated signal transduction (Colbran et al., 1989). Most vertebrates possess four different CaMKII genes (α, β, γ, and δ) giving rise to at least 38 different splice variants (Tombes et al., 2003). Two hallmarks distinguish CaMKII from other kinases: firstly, it acts as a multimeric holoenzyme composed of 4–14 heteromeric or homomeric subunits of the different isoforms of the four genes and secondly, its ability to autophosphorylate on the threonine 286 residue upon Ca2+/CaM binding (Soderling, 1996; Hudmon and Schulman, 2002; Colbran, 2004; Lantsman and Tombes, 2005; Rosenberg et al., 2006). Autophosphorylation relieves the enzyme from its Ca2+/CaM dependence. Alternatively, CaMKII can be activated by methionine oxidation (Erickson et al., 2008).

Various studies suggest that CaMKII signaling may play a role in skeletogenesis. All four genes are expressed in chicken and mouse chondrocytes (Taschner et al., 2008; Li and Dudley, 2009). Studies on human articular chondrocytes have suggested that CaMKII is involved in N-methyl-D-Aspartic acid (NMDA) receptor signaling, which is important for maintaining matrix integrity of joints (Salter et al., 2004; Shimazaki et al., 2006). CaMKII signaling has also been implicated in osteoblast and osteoclast differentiation (Quinn et al., 2000; Zayzafoon et al., 2005). In the chicken, we demonstrated previously using a retroviral system that the expression of a dominant active form of CaMKII (DaCaMKII), which mimics the autophosphorylated form, caused premature ectopic chondrocyte maturation, while the inhibition of CaMKII activity by a peptide inhibitor (rKIIN) delayed the hypertrophic program (Taschner et al., 2008). Li and colleagues suggested that the increasing CaMKII activity in the chondrocytes during their transition from the proliferative to the prehypertrophic state regulates Runx2 and β-catenin activity and thereby promotes chondrocyte hypertrophy (Li et al., 2011).

Retroviral driven expression in the chick system has the disadvantage that all mitotically active cells get infected. So besides the chondrocytes also the soft-tissue is infected. This makes it difficult to distinguish between cell-autonomous and non-cell-autonomous effects. Using a transgene approach in the mouse allowed us to overcome this problem. Hence, in order to gain more specific insights into the potential role of CaMKII in endochondral bone formation, we expressed an activated form of CaMKII (DaCaMKII) or the peptide inhibitor rKIIN under the control of the Col2a1 promoter primarily in chondrocytes. Based on our observations in the transgenic mouse models we conclude that modulation of CaMKII activity in the mouse has only an effect early in development at the onset of chondrocyte hypertrophy.

## MATERIALS AND METHODS

## Generation of Col2a1-Transgenic Mice

For the dominant active CaMKII transgene a cassette containing CaMKII-T286D C-terminal fused to eGFP (Taschner et al., 2008) followed at the 3′ end by two SV40polyA sequences was inserted via blunt-end cloning into the BamHI site of a plasmid carrying the Col2a1 promoter–rabbit β globin intron–Col2a1 enhancer element (a gift from Yoshi Yamada; (Nakata et al., 1993)). For the rKIIN transgene, a peptide inhibitor for CaMKII from rat (Chang et al., 1998, 2001) fused with eGFP at its Nterminus (Taschner et al., 2008) was cloned accordingly. The final transgenic constructs (**Figures 1A,B**) were excised with the restriction enzymes AfllII and SwaI, purified on agarose gel and eluted with DNA microinjection buffer. The linear transgene cassettes were microinjected into the pronucleus of B6CBAF<sup>1</sup> zygotes in the transgene facility of the IMP, Vienna, Austria (Hogan et al., 1994). The zygotes were implanted into pseudo-pregnant mice to obtain transgenic founder lines. To maintain the transgenic lines in a pure genetic background, the founder lines were crossed with C57Bl/6J females and the subsequent generations were backcrossed with C57Bl/6J animals over at least eight generations. Genotyping of transgenic mice and embryos was performed by PCR using ear biopsies and material from the embryonic tail respectively, in combination with transgene-specific primer pairs (listed in Supplementary Table 1). Transgene-copy numbers were determined using genomic DNA from ear biopsies of the different transgenic lines (two independent samples per line) by qPCR and normalized to actin and control genomic DNA from mice carrying one copy of the transgene in the Rosa26 locus (Amara and Hartmann, unpublished; Ballester et al., 2004). Primers are listed in Supplementary Table 2. Mouse experiments were performed in accordance with local, institutional and national regulations under the following licenses (84-02.05.2012.260, 84- 02.04.2015.A278, 84-02.05.50.15.022).

**Abbreviations:** CaMKII, Calcium/calmodulin-dependent kinase II; Gfp, Green fluorescent protein; ISH, in situ hybridizations; tg, transgenic; E, embryonic day.

## Mouse Husbandry and Embryo Processing

For timed pregnancies, heterozygous transgenic mice were interbred overnight and the plug day was designated as embryonic day 0.5 (E0.5). Embryos were harvested at the required stages, dissected and fixed overnight in 4% paraformaldehyde (PFA). For stages E18.5 and older, the skin above the limbs was removed prior to fixation. Fixed limbs were removed, washed in PBS, and tissue was dehydrated using increasing ethanol concentrations (25, 50, 75, 100%).

## In situ Hybridization, Histology and Skeletal Preparations

For in situ hybridization (ISH) and histological staining on sections, limbs were processed into paraffin using the Excelsior ES Vacuum Infiltration Processor (Thermo Scientific), embedded in paraffin and sectioned at 5 µm. Non-radioactive ISH were performed using digoxigenin (DIG)-labeled anti-sense RNA probes as previously described (Murtaugh et al., 2001). Probes for type 2 collagen (Col2a1), indian hedgehog (Ihh), osteocalcin (Ocn), osteopontin (Opn), type 1 collagen (Col1a1), and type 10 collagen (Col10a1) have been published previously (Hill et al., 2005). The Gfp probe was generated using a plasmid containing the eGFP coding region. All probes are available upon request. Histological stainings such as alcian blue, alcian blue/von Kossa, and hematoxylin/eosin were performed as previously described (Houben et al., 2016). Skeletal preparations were performed on 6 day old pups which were sacrificed by decapitation, skinned, eviscerated, fixed in 95% EtOH, and double-stained for alcian blue and alizarin red (McLeod, 1980). The length of isolated humeri and femora was calculated by using Zeiss image analysis software.

## RNA Isolation and qPCR

The stylo- and zeugopod regions of forelimbs were isolated from embryos at the required stage. Skin and soft tissue was removed and skeletal elements were dispersed using the Polytron PT 1200 E manual disperser with the aggregate PT-DA 03/2 EC-E50 (Kinematica). The RNA was isolated using the QIAGEN RNeasy micro kit according to the manufacturer's instructions. A total of 500 ng RNA was reverse transcribed using PrimeScript RT reagent kit (Takara, #RR037A) with oligo dT primer. For qPCR cDNA was diluted 1:10 in water. 3µl of diluted cDNA were mixed with 1µl of primer mix and 10µl SYBR Premix Ex TaqII (Tli RNaseH plus) (Takara, #RR820Q) in a final volume of 20µl. Gene expression was monitored using the BioRad CFX96 cycler and the following protocol: 95◦C 10 min, 45 × (95◦C 30 s, 60◦C 30 s, 72◦C 45 s + plate read), 72◦C 5 min, melting curve: 55◦C– 99◦C in 0.5◦C increments for 5 s + plate read. qPCR analysis was performed in duplicates and results were produced from at least three independent samples. Expression values were calculated using the comparative 1C(t) method and normalized to Gapdh and Actb expression. For primer sequences and product sizes see Supplementary Table 3.

## Isolation and Cultivation of Primary Chondrocytes

E13.5 appendicular skeletal elements from the stylo- and zeugopod were dissected in PBS supplemented with 1% P/S followed by a digestion with 0.1% type II collagenase (Gibco, #17101-015) and 0.3% dispase (Gibco, #17105-041) in medium (DMEM/F-12; 1% P/S) for a period of 60 min shaking at 100 rpm in a CO<sup>2</sup> incubator. Cells were filtered through a cell strainer, centrifuged and washed with culture medium (DMEM/F-12, 1% P/S, and 10% FCS). Cell pellets were then resuspended in culture medium, plated at 2.5–3 × 10<sup>4</sup> cells/cm<sup>2</sup> in tissue culture flasks and grown for 5 days.

## Luciferase Assays

Primary chondrocytes were seeded at a density of 3 × 10<sup>4</sup> cells/well in 48 well plates 1 day prior to transfection. Cells were transfected with the pGL4.23-Mef2-luc (a gift from Eric Olson) or the pGL4.10-Ose2-luc reporter plasmid (a gift from G. Karsenty) in combination with pRL-TK (Promega) as a control and pCDNA3.1-DaCaMKII or pCDNA3.1-rKIIN expression vectors. Total amounts of transfected plasmids (equivalent to 52 fM) in each group were adjusted by adding empty vector (pCDNA3.1+). Transfection was performed using jetPRIMETM (Polyplus transfection) according to manufacturer's instructions. Luciferase activities were measured 48 h after transfection following the protocol by (Hampf and Gossen, 2006). Luciferase measurements were normalized to the corresponding renilla activities for transfection efficiency. Experiments were performed in triplicates and repeated at least three times.

## Immunofluorescence

For immunofluorescent staining on cells, primary chondrocytes isolated from P6 C57Bl/6J knee epiphyseal regions were seeded on 0.1% gelatin coated glass coverslips. Cells were transfected with DaCaMKII::eGFP and eGFP expression vectors (equivalent to 0.12 pmol) using jetPRIMETM (Polyplus transfection). Forty-Eight hours after transfection, cells were fixed with 4% PFA/PBS for 20 min, treated with 0.5% TritonX-100 for 5 min and blocked with 10% normal goat serum (NGS) in PBS for 1 h at room temperature (RT). Incubation with β-catenin antibody (1:200 in 2% NGS/PBS, BD Biosciences, #610154, RRID:AB\_397555) was performed overnight at 4◦C. Coverslips were subsequently incubated with Alexa Fluor 647 goat anti-mouse IgG (Molecular Probes) at 1:200 in 2% NGS/PBS. Nuclei were counterstained with DAPI.

## Immunoprecipitation by Magnetic Beads

HEK293 cells were co-transfected with Mef2c-Myc and DaCaMKII::eGFP at 4:3 ratio or FLAG-HDAC4 and DaCaMKII::eGFP at 1:3 ratio using CaPO<sup>4</sup> transfection method. Cells were lysed in ice-cold lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% NP-40 with proteinase and phosphatase inhibitors). After centrifugation, a total amount of 150 µg of protein from the supernatant was subjected to immunoprecipitation using 25µl magnetic beads coupled with GFP monoclonal antibody (MBL, #D153-11) under the following conditions: 15 min at 4◦C on a rotating wheel (10 rpm), washed 4 times with wash buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% NP-40) for 5 min at RT. For the pull-down of FLAG-HDAC4 using GFP-magnetic beads the conditions were as following: 500µg of protein from the supernatant was subjected to immunoprecipitation using 50 µl magnetic beads and bound for 45 min at 4◦C on a rotating wheel (10 rpm), followed by 3 washes (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.05% NP-40) for 30 s at RT. Magnetic beads were then boiled for 2 min in 25 µl Laemmli's sample buffer and supernatant of the beads was loaded on a 10% SDS-PAGE to separate the immunoprecipitated proteins.

## Immunoblots

Transfected primary chondrocytes were lysed with ice-cold lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% NP-40 with proteinase and phosphatase inhibitors). 50µg of protein lysates were run on a 10% SDS-PAGE and transferred to a 0.45 µm PVDF-membrane by semi-dry transfer (PerfectBlueTM, PeqLab). The membrane was blocked with 5% milk and incubated with the appropriate primary antibodies anti-GFP (1:15000, Abcam, #ab13970, RRID:AB\_300798), anti-Myc (1:1000, Cell Signaling Technology, mAb #2276, RRID:AB\_331783), pan-CaMKII (1:1000, Cell Signaling Technology, mAb #4436S, RRID:AB\_10545451), anti-Flag (1:1000, Sigma-Aldrich, #F1804, RRID:AB\_262044), and proteasome 20 (P20) (1:15000, Abcam, #ab3325, RRID:AB\_303706), followed by incubation with the respective, species-specific HRP-coupled secondary antibodies (1:1000 and 1:5000). ECL substrate was used for signal development (Amersham, #RPN2106) on X-ray film (Amersham).

## Image Acquisition

Histological images were acquired using Zeiss AxioImager M2 equipped with an AxioCam MRc 6.45µm color camera (Zeiss, Jena). Images of embryos were acquired using a Zeiss Stereo discovery V8 equipped with Zeiss plan Apo S, 0.63X lens. Immunofluorescent images were acquired using Zeiss AxioImager M2 equipped with an Apotome2 and an AxioCam MRm 6.45µm monochromatic camera (Zeiss, Jena) using the Zen software (Zeiss, Jena).

## Statistical Analysis

All statistical analyses were performed using GraphPad Prism software 6.0 (RRID: SCR\_002798). Data are displayed as mean values ± standard error of the mean (SEM). Statistical significance of differences (P-value) was determined by the twotailed, unpaired Student's t-test.

## RESULTS

## In Vitro: CaMKII Signaling Affects β-Catenin Localization and Alters Runx2 and Mef2c Activity

Based on previous studies in chicken, potential molecular mechanisms of how CaMKII signaling regulates chondrocyte hypertrophy have been suggested (Li et al., 2011). These included increased nuclear localization of β-catenin and Runx2 in chondrocytes with activated CaMKII signaling (Li et al., 2011). Thus, we investigated whether the molecular mechanism in mouse chondrocytes may be similar. Mouse chondrocytes, like chicken chondrocytes, express all four CaMKII isoforms, and also here the CaMKIIβ isoform was expressed at lower levels than the other three isoforms (**Figure 1A**, see also Taschner et al., 2008; Li et al., 2011). Similar to what was observed in chicken, β-catenin was localized in the nucleus in mouse chondrocytes transfected with a DaCaMKII::GFP expression plasmid (**Figure 1B**, white arrows). This was not observed in the GFP-transfected control cultures (**Figure 1B**, left panel). We also analyzed whether Runx2 transcriptional activity was altered by the presence of DaCaMKII or rKIIN in mouse chondrocytes. For this, the Runx2-dependent OSE2-luc reporter (Ducy and Karsenty, 1995) was co-transfected into primary chondrocytes with increasing amounts of either DaCaMKII or rKIIN expression plasmid. OSE2-luc reporter activity was upregulated in response to increasing amounts of DaCaMKII, while expression of rKIIN led to a downregulation of the reporter activity in a concentration independent manner (**Figure 1C**). To rule out the possiblity that modulation of CaMKII activity affects the expression levels of Runx2 itself, we examined the Runx2 expression levels in the OSE2-luc assay by qPCR. Here, Runx2 expression was not affected in cells transfected with the DaCaMKII or rKIIN expression plasmids (**Figure 1D**).

Another transcription factor promoting hypertrophy is Mef2c (Arnold et al., 2007). The MEF2-luc reporter (Naya et al., 1999) was activated over 50-fold by DaCaMKII at the highest concentration used (**Figure 1E**). An increase in CaMKII protein levels corresponding to the increased amount of transfected expression plasmid was demonstrated by western blot (**Figure 1E**, lower panel). The DaCaMKII expression levels were also determined by qPCR using a primer that detects the mouse as well as the transgenic rat Camk2a transcript. Here, the Camk2a expression levels were normalized to the endogenous Camk2a expression levels in control-transfected chondrocytes. This quantitative approach revealed an over 100-fold increase in Camk2a levels the cells transfected with 30 fM DaCaMKII (Supplementary Figure 1). Overexpression of rKIIN on the other hand, led, with the exception of the highest concentration, to a downregulation of the MEF2-luc reporter activity independent of the concentration (**Figure 1E**). To our surprise, at the highest concentration even a two-fold activation was observed (**Figure 1E**). Again, the endogenous Mef2c expression levels were not significantly affected by either DaCaMKII or rKIIN expression in the transfected primary chondrocytes (**Figure 1F**). Given the strong effect on Mef2c activity we tested whether CaMKII may directly interact with Mef2c. Co-immunoprecipitation assays in HEK293 cells using tagged proteins revealed that the activated form of CaMKII physically interacts with Mef2c (**Figure 2A**). Here we used GFP-magnetic beads and pulled on the DaCaMKII::GFP fusion protein and detected bound Mef2c::Myc (**Figure 2A**). Conversely, using Myc-magnetic beads and pulling on Mef2c::Myc the DaCaMKII::GFP fusion protein was detected by immunoblot in the bound fraction (data not shown).

Given previous findings that (a) the CaMKIIδ isoform can interact with and influence histone deacetylase 4 (Hdac4) activity in cardiomyocyte hypertrophy (Backs et al., 2006, 2009), (b) Hdac4 controls chondrocyte hypertrophy by interacting with and inhibiting the activity of Runx2 and Mef2c (Vega et al., 2004; Arnold et al., 2007), and (c) our observation that DaCaMKII modulates Mef2c as well as Runx2 activity, we addressed whether DaCaMKII also physically interacts with Hdac4 or whether it alters its repressive activity. For this, HEK293 cells were co-transfected with FLAG::Hdac4 and DaCaMKII::GFP. Co-IP studies with GFP-coupled magnetic beads revealed no physical interaction for Hdac4 in the pull-down lysate (**Figure 2B**). Luciferase assays on primary chondrocytes transfected with MEF2-luc, Mef2c and Hdac4 resulted in the inhibition of Mef2 activity. Yet, this inhibitory effect was not altered by the addition of increasing amounts of DaCaMKII (**Figure 2C**). This suggests that the DaCaMKII does not modulate Hdac4 activity. In essence, our results show that the activated form of CaMKII positively influences the activity of the transcription factors Mef2c and to a lesser extent Runx2 activity and that this effect is probably not mediated via an inhibitory effect on Hdac4. Furthermore, CaMKII physically interacts with Mef2c suggesting that it may possibly modulate its transcriptional activity by phosphorylation.

## In Vivo: Generation of Chondrocyte Specific Transgenic Mice

In order to determine the in vivo role of CaMKII modulation in mouse endochondral ossification, we employed a transgenic approach using the Col2a1 promoter/enhancer element to drive transgene expression in chondrocytes (**Figures 3A,B**). From previous studies in the chicken we knew that eGFP tagged transgenes are functional (Taschner et al., 2008). Hence, in order to visualize transgene expression and to facilitate the distinction between heterozygous and homozygous embryos, eGFP tagged transgenes were used. For the DaCaMKII::eGFP construct two independent transgenic founder lines (Tg1 and Tg2- DaCaMKII) were established and for the eGFP::rKIIN construct three independent founder lines were obtained (Tg1, Tg2, and Tg3-rKIIN; **Figures 3A,B**). All founder lines transmitted the transgene to the F1 generation. The transgenic lines

differed in the number of transgenes integrated in the genome (**Figure 3A**). For a first characterization of the transgenic lines, heterozygous transgenic mice of the independent founder lines were intercrossed to isolate embryos at embryonic day (E) 15.5. Contrary to our expectations, no obvious gross morphological differences were detected within the offspring of the various intercrosses (**Figures 3C,D**). Genotyping of the offspring was performed examining the GFP intensity using a fluorescent stereomicroscope and confirmed by conventional PCR-based genotyping (**Figures 3C,D**; data not shown). Heterozygous and homozygous transgenic embryos were classified based on the fluorescence intensity (**Figures 3C,D**). As expected, GFP activity was detected in the cartilaginous regions of the skeletal elements, particularly visible in the skull and limbs at sites corresponding to Col2a1 expression domains (**Figures 3C,D**; Cheah et al., 1991). This was confirmed by ISH with RNA antisense probes for Col2a1 and Gfp on adjacent sections of transgenic limbs (**Figures 3E,F**). As expected, no Gfp signal was detected in humeri of nontransgenic (wild-type) littermates (**Figures 3E,F**). These results demonstrated that the transgene is indeed expressed in a Col2a1 promoter/enhancer-dependent manner. Furthermore, the GFP signal intensity in embryos as well as the Gfp ISH signal on sections allowed us to distinguish heterozygous and homozygous transgenic littermates.

In the chicken a subset of the chondrocytes that expressed DaCaMKII prematurely differentiated into prehypertrophic and subsequently into hypertrophic chondrocytes outside their normal expression domains (Taschner et al., 2008; Li et al., 2011). Hence, we performed ISH on E15.5 sections through the limbs using Col2a1 as a marker for more immature chondrocytes, Ihh as a marker for prehypertrophic cells, and Col10a1 as a marker for hypertrophic cells. Yet, no expression of the maturation markers Ihh or Col10a1 was observed outside their normal expression domains and no obvious differences were observed with respect to the individual domain sizes, distance between the domains, or the expression levels of these three markers comparing non-transgenic with heterozygous and homozygous DaCaMKIItg littermates (**Figures 4A,B**). Similar, negative results were

taken at the same magnification under fluorescent stereomicroscope (Zeiss). (E,F) Non-radioactive Col2a1 and Gfp ISH on E15.5 humeri isolated from DaCaMKII-tg mice (E) and isolated from rKIIN-tg mice (F) and corresponding control (non-transgenic, +/+) littermates showing that the Gfp riboprobe signal overlaps almost completely with that of Col2a1. Signal intensity for the Gfp anti-sense riboprobe differs between heterozygous and homozygous transgenic embryos. SA, splice acceptor site; pA, poly A.

obtained performing the analogous analysis on sections of E15.5 humeri from non-transgenic, heterozygous, and homozygous rKIIN-tg littermates (**Figures 4C,D**). Here, we did not observe an obvious delay in chondrocyte maturation, as one could have expected based on the effect of rKIIN overexpression in chicken.

The final step in endochondral ossification is the remodeling of the cartilage template into bone. Thus, in order to see whether this remodeling process was altered, we analyzed the longitudinal extension of the ossified zone by performing ISH analysis for the osteoblastic markers collagen 1 (Col1a1), osteopontin (Opn), and osteocalcin (Ocn) on sections of E18.5 DaCaMKII- and rKIINtg humeri and corresponding non-transgenic littermates. No significant alteration to the longitudinal extension of the ossified zone for either of the three analyzed markers was observed when comparing the humeri of the transgenic embryos expressing either DaCaMKII (**Figures 5A,B**) or rKIIN with humeri of their respective, non-transgenic controls (**Figures 5C,D**). Similar to what we had observed at E15.5, no morphological alterations with regard to the shape, organization or appearance of the prehypertrophic and hypertrophic chondrocytes were detected in the growth plates of the E18.5 DaCaMKII- and rKIINtg humeri compared to their corresponding non-transgenic littermate controls (Supplementary Figures 2A,B). In addition, the total length, as well as the length of the mineralized region was measured in alician blue/alizarin red stained humeri of 6-day-old pups. Here again, no differences regarding the two parameters were observed in comparison to non-transgenic littermate controls (Supplementary Figure 3).

## Modulation of CaMKII Signaling Affects the Onset of Chondrocyte Maturation

As there were no obvious effects visible in the older mouse limbs, we addressed whether early steps of skeletogenesis would potentially be affected. In the forelimb the anlagen of humerus, radius and ulna are visible at E11.5. At this stage the cells in the center of the humerus begin to express Ihh (data not shown), while they start to express Col10a1 around E12.5. In humeri of E12.5 DaCaMKII transgenic limbs the marker for prehypertrophic chondrocytes, Ihh, was expressed in a slightly broader domain compared to non-transgenic, control littermates (**Figures 6A–C**). The effect was more prominent in humeri of homozygous (6/8) than heterozygous (8/11). When we analyzed the expression of Col10a1, as a marker for hypertrophic chondrocyte differentiation, weak Col10a1 expression was first detected in the humerus at E12.5-E13.0 in wild-type limbs. In the transgenic specimens, in which we had observed a broadend Ihh domain we noticed on adjacent sections a more intense staining for Col10a1 in a broader domain (**Figure 6D**). Again

the effect was more obvious in the homozygous specimens. Next we asked whether the increased Col10a1 expression was accompanied by the histological appearance of hypertrophic chondrocytes. For this, we examined E12.5 and E13.5 limbs by alcian blue staining. On alcian blue stained sections, the hypertrophic chondrocytes within the cartilage elements appear lighter in color due to their increase in size and vacuolization. Histological examination of E12.5 humeri (n = 2) revealed no apparent signs of hypertrophic chondrocyte differentiation in the transgenic limbs (Supplementary Figure S4A). In E13.5 DaCaMKII-tg humeri, a moderate size increase of the zone of hypertrophic chondrocytes was detected compared to the nontransgenic control (**Figure 6E**, n = 3). Consistent with the in vivo ISH results, qPCR analysis of material from E12.5 DaCaMKII transgenic and non-transgenic control limbs revealed an increase in Ihh and Col10a1 expression levels in the transgenic limbs in comparison to the control (**Figure 6F**). These results suggest that the onset of chondrocyte maturation is slightly accelerated in the transgenic limbs.

We then analyzed rKIIN transgenic limbs at equivalent stages. At E12.5, no obvious differences were detected with respect to the size or intensity of the Ihh expression domains between transgenic and non-transgenic embryos (Supplementary Figure S4B). However, at E13.5 the two expression domains of Ihh were not yet separated in the humeri of homozygous transgenic animals and still closer together in the heterozygous transgenic animals compared to non-transgenic littermate controls (**Figure 6I**). Furthermore, the Col10a1 expression domain was reduced in size in the humeri of the transgenic animals compared to non-transgenic littermate controls (**Figure 6J**). Together, his suggests that the onset of chondrocyte maturation is slightly delayed when CaMKII signaling is antagonized. Again, the effect was more pronounced in homozygous (9/9) than in heterozygous transgenic specimens (11/13) compared to nontransgenic specimens (**Figures 6I,J**). Histological examination of alcian blue stained E13.5 humeri confirmed that the decrease in the size of the Col10a1 expression domain in homo- and heterozygous humeri was associated with reduced hypertrophic chondrocyte differentiation (**Figure 6K**). Quantification of Ihh and Col10a1 expression levels by qPCR using material from E13.5 forelimbs corroborated the dose-dependent effects of CaMKII-signaling inhibition, which was statistically significant

with respect to Col10a1 expression (**Figure 6L**). Together, our results suggest that modulation of CaMKII signaling in the mouse affects the onset of chondrocyte maturation during the early stages of endochondral ossification.

ISH with a Gfp riboprobe confirmed transgene expression at all stages examined and at least according to the ISH transgene expression was not silenced after E13.5 (Supplementary Figure 5). In order to determine why the expression of the DaCaMKII and rKIIN-transgenes may not lead to a major phenotype at later stages we examined the transgene expression level in the two DaCaMKII-transgenic lines, DaCaMKII-Tg1 (1 copy) and DaCaMKII -Tg2 (19 copies), using qPCR and RNA isolated from the chondrogenic elements of E12.5 limbs. Transgene expression levels were estimated by comparing the DaCaMKII::GFP transgene expression level to the endogenous expression levels of the four CaMKII isoforms and the levels were normalized to the Camk2g expression level. Here, we noticed that the transgene expression levels in the Tg1 line were similar to the Camk2a endogenous levels in the wild-type control. Yet, they were lower than the endogenous expression levels of the γ and δ isoforms. Interestingly, the Camk2a expression levels were increased significantly in the transgenic skeletal elements of the DaCaMKII-Tg1 line. In contrast, transgene expression level in the Tg2 line with 19 copies did not even reach the Camk2a endogenous levels. As such, the transgene copy number did not correlate with the transgene expression levels in the two independent lines (Supplementary Figures 6A,B).

## DISCUSSION

Based on previous work in chicken it has been proposed that activation of the endogenous CaMKII activity controls the onset of the prehypertrophic and hypertrophic chondrogenic program. In proliferating chondrocytes, the phosphorylated form of CaMKII is not detectable and possibly constantly dephosphorylated under the influence of PTHrP signaling (Li et al., 2011). As a consequence, the phosphorylated form of CaMKII is limited to the prehypertrophic and hypertrophic chondrocytes in chicken and mouse (Li et al., 2011). In the chicken, ectopic expression of an activated form of CaMKII, was able to override the endogenous CaMKII activity or lack thereof in proliferating chondrocytes and led to premature and ectopic activation of the prehypertrophic/hypertrophic program in cells outside of the normal maturation zones (Taschner et al.,

FIGURE 6 | Modulation of CaMKII signaling affects the onset of chondrocyte maturation in early skeletal development. (A–D) Representative, alternating sections through humeri of E12.5 control (non-transgenic, +/+), hetero- (Tg/+), and homozygous (Tg/Tg) DaCaMKII-tg littermates. (A) Gfp ISH revealing transgene expression in heterozygous and homozygous animals but not in non-transgenic littermate controls. (B) Col2a1 ISH confirming that the transgene expression is restricted to the Col2a1 positive domain. (C) Ihh ISH defining the prehypertrophic zone, which is enlarged in the homozygous and heterozygous transgenic embryos. (D) Col10a1 ISH marking the onset of hypertrophic chondrocyte differentiation. Col10a1 expression was more intense in the transgenic embryos compared to control (non-transgenic) littermates. (E) Alcian blue/eosin stained representative images of sections through E13.5 humeri of control (non-transgenic; +/+), hetero- (Tg/+), and homozygous (Tg/Tg) DaCaMKII-tg embryos. Images below show magnifications of the boxed areas in the panel above. Due to their increase in volume and vacuolization of the cells, the hypertrophic chondrocytes appear whiter. The region of hypertrophy is enlarged in the transgenic humeri compared to wild-type littermate control. (F) qPCR analysis for Ihh and Col10a1 normalized to Gapdh and Actb using material from E12.5 control, hetero- and homozygous DaCaMKII-tg forelimbs reveals a dose-dependent increase in the expression of these two genes. Gene expression levels are plotted relative to control. (G–J) Representative, alternating sections through humeri of E12.5 (G,H) and E13.5 (I,J) control (non-transgenic; +/+) and hetero- (Tg/+), and homozygous (Tg/Tg) rKIIN-tg littermates. (G) Gfp ISH revealing transgene expression in heterozygous and homozygous animals but not in non-transgenic littermate controls. (H) Col2a1 ISH confirming that the transgene expression is restricted to the Col2a1 positive domain. (I) Ihh ISH defining the prehypertrophic zone, which is reduced in a transgene dose-dependent manner in the hetero- and homozygous transgenic embryos. (J) Col10a1 ISH defining the hypertrophic zone, showing that hypertrophic differentiation is even more delayed. (K) Histological alcian blue/eosin stained representative images of sections from control, hetero- and homozygous rKIIN embryos: here, fewer and smaller hypertrophic, whiter cells are visible in hetero- and homozygous transgenic humeri compared to wild-type littermate controls. (L) qPCR analysis for Ihh and Col10a1 normalized to Gapdh and Actb using material from E13.5 control, hetero- and homozygous rKIIN-tg forelimbs revealing a dose-dependent reduction in the expression of Ihh and Col10a1. Gene expression levels are plotted relative to control. (F,L) n refers to the number of independent biological samples. \*\*p < 0.01, \*\*\*p < 0.001, n.s., not significant. Error bars indicate ± SEM.

2008; Li and Dudley, 2009; Li et al., 2011). In mouse long bone development, the phenotypic effects of CaMKII activation in proliferating, type II collagen-expressing chondrocytes were very mild and in vivo, effects could only be detected developmentally around the onset of chondrocyte hypertrophy at E12.5–E13.5. Here, our data suggest that in agreement with previous findings in chicken, DaCaMKII in mouse activated the prehypertrophic/hypertrophic program prematurely at early stages of endochondral ossification, while down-regulation of endogenous CaMKII activity interfered with the onset of the prehypertrophic and hypertrophic program. Interestingly, the phenotypic changes caused by CaMKII activity modulation at the onset of chondrocyte maturation did not accumulate over time despite the fact that the transgene was continuously expressed in type II collagen-producing chondrocytes (Supplementary Figure 5). Yet, in contrast to the overexpression experiments in chicken, no premature maturation of chondrocytes outside their normal maturation zones or shortening of the limbs was observed. Hence, the phenotypic consequences of ectopic activation of CaMKII are quite distinct between chicken and mouse (Taschner et al., 2008; Li and Dudley, 2009; Li et al., 2011). Possible explanations for the phenotypic discrepancies between chicken and mouse could be that on the one hand in the chicken not only chondrocytes also the soft tissue is infected and that this could contribute to the phenotype. On the other hand, it is likely that the retroviral driven expression levels of the transgenes were much higher in chicken. For retroviral driven transgenes over 100-fold expression levels have been reported (Geetha-Loganathan et al., 2014; Nimmagadda et al., 2015). In the mouse, we did not detect phenotypic differences between the independent transgenic lines despite the fact that they varied in copy numbers (**Figure 3A**), which can be explained by the fact that transgene expression levels were similarly low in both cases (about one-fold of the Camk2a endogenous levels; Supplementary Figure 6). Hence, the lack of a major phenotype in the transgenic mice may be associated with the relatively low transgene expression levels compared to the expression levels that have been reached by retroviral expression in the chicken. Although, of course, the actual levels of retroviral-driven DaCaMKII transgene expression were not determined in the chicken experiments performed previously (Taschner et al., 2008; Li and Dudley, 2009; Li et al., 2011). A possible explanation for the observed transient phenotype at the early stages of skeletal development maybe that the onset of chondrocyte maturation may present a window of opportunity for the transgenes to excert a mild effect accelerating, respectively delaying chondrocyte maturation. And that this effect is later on compensated as the chondrocyte maturation program comes under the transcriptional control of many regulatory factors and feed back mechanisms as development progresses (Hartmann, 2009; Kozhemyakina et al., 2015).

At the molecular level, we observed in vitro that CaMKII robustly affects the transcriptional activity of the transcription factor Mef2c and to a lesser extent also Runx2 transcriptional activity. Mef2c and Runx2 start to be expressed as the chondrocytes undergo hypertrophy (Arnold et al., 2007). Mef2c and Runx2 activity are both negatively regulated by the class II histone deacetylase HDAC4 (Vega et al., 2004; Kozhemyakina et al., 2009; Correa et al., 2010). In vascular smooth muscle cells, CaMKIIdelta2 regulates Mef2 transcriptional activity through HDAC4/5 (Ginnan et al., 2012). Yet, our results indicate that activated CaMKII increases Mef2c activity by an HDAC4 independent mechanism. As the activated form of CaMKII physically interacts with Mef2c, the underlying mechanism may involve phosphorylation of Mef2c protein by CaMKII. In different cell types other kinases, such as p38 MAPK and ERK5, have also been shown to phosphorylate Mef2c enhancing its transcriptional activity (Han et al., 1997; Kato et al., 1997). Yet, there is the obvious discrepancy between the in vitro results where a robust stimulatory effect on Mef2c transcriptional activity was observed and the subtle in vivo effects. For the in vitro luciferase experiments an expression vector was used that drives DaCaMKII expression under the control of the CMV promoter, which drives high levels of expression in mammalian cells. The high transgene expression levels in vitro (Supplementary Figure 1) may be an explaination for the strong in vitro effect on Mef2c activity and in contrast to the only mild effect in vivo where transgene expression levels were at least a 100-fold lower. Furthermore, Li and colleagues proposed that the endogenous CaMKII activity is opposed by an inhibitory gradient, which is under the control of PTHrP signaling (Li et al., 2011). PTHrP signaling via cAMP and protein kinase A negatively regulates chondrocyte hypertrophy through HDAC4 mediated inhibition of Mef2c activity (Kozhemyakina et al., 2009). Yet, even in vitro, DaCaMKII signaling was not able to overcome or to partially revert the effect of HDAC4 on Mef2c activity (**Figure 2C**). The PTHrP expression itself is under the regulatory control of Ihhsignaling (Vortkamp et al., 1996; St-Jacques et al., 1999). As such the slightly increased Ihh expression observed in the E12.5 DaCaMKII transgenic limbs may upregulate PTHrP signaling and counteract a possibly direct, positive effect of DaCaMKII on the Mef2c activity via HDAC4. Hence, one possible scenario may be that increased levels of PTHrP signaling override the activation of Mef2c by DaCaMKII at later stages of development.

## AUTHOR CONTRIBUTIONS

CA: Data collection, assembly, analyses, interpretation, and manuscript writing. CF: Performed the co-immunoprecipitation studies. AH: Performed qPCR studies determining transgene copy numbers and expression levels, and proof-read the manuscript. LW: Performed histological analysis on E18.5 material. CH: Conception and design, data interpretation, and manuscript writing.

## FUNDING

Part of the work was conducted at the Institute of Molecular Pathology in Vienna which is funded by Boehringer Ingelheim. AH was supported by a grant from the Deutsche Forschungsgemeinsaft [HA 4767/2-1] and LIW is supported by a grant from the BMBF [Overload-PrevOP, SP07]. The funding sources had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

## ACKNOWLEDGMENTS

We thank Christian Theussl for pronuclear injections and Joanna Sherwood for checking the English in the manuscript. We also thank the reviewers for their valuable input.

## REFERENCES


## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fcell. 2017.00020/full#supplementary-material


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

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

,

## A Novel Role for the BMP Antagonist Noggin in Sensitizing Cells to Non-canonical Wnt-5a/Ror2/ Disheveled Pathway Activation

Ondrej Bernatik <sup>1</sup> , Tomasz Radaszkiewicz <sup>1</sup> , Martin Behal <sup>1</sup> , Zankruti Dave<sup>1</sup> , Florian Witte<sup>2</sup> Annika Mahl <sup>2</sup> , Nicole H. Cernohorsky <sup>3</sup> , Pavel Krejci 1, 3, Sigmar Stricker <sup>2</sup> \* and Vitezslav Bryja1, 4 \*

### Edited by:

Gregory Kelly, University of Western Ontario, Canada

#### Reviewed by:

Terry Van Raay, University of Guelph, Canada Ole-Morten Seternes, University of Tromsø, Norway

#### \*Correspondence:

Sigmar Stricker sigmar.stricker@fu-berlin.de Vitezslav Bryja bryja@sci.muni.cz

Specialty section:

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

> Received: 30 January 2017 Accepted: 13 April 2017 Published: 04 May 2017

#### Citation:

Bernatik O, Radaszkiewicz T, Behal M, Dave Z, Witte F, Mahl A, Cernohorsky NH, Krejci P, Stricker S and Bryja V (2017) A Novel Role for the BMP Antagonist Noggin in Sensitizing Cells to Non-canonical Wnt-5a/Ror2/ Disheveled Pathway Activation. Front. Cell Dev. Biol. 5:47. doi: 10.3389/fcell.2017.00047 <sup>1</sup> Faculty of Sciences, Institute of Experimental Biology, Masaryk University, Brno, Czechia, <sup>2</sup> Institute for Chemistry and Biochemistry, Freie Universität Berlin, Berlin, Germany, <sup>3</sup> Department of Biology, Faculty of Medicine, Masaryk University, Brno, Czechia, <sup>4</sup> Department of Cytokinetics, Institute of Biophysics AS CR, v.v.i., Brno, Czechia

Mammalian limb development is driven by the integrative input from several signaling pathways; a failure to receive or a misinterpretation of these signals results in skeletal defects. The brachydactylies, a group of overlapping inherited human hand malformation syndromes, are mainly caused by mutations in BMP signaling pathway components. Two closely related forms, Brachydactyly type B2 (BDB2) and BDB1 are caused by mutations in the BMP antagonist Noggin (NOG) and the atypical receptor tyrosine kinase ROR2 that acts as a receptor in the non-canonical Wnt pathway. Genetic analysis of Nog and Ror2 functional interaction via crossing Noggin and Ror2 mutant mice revealed a widening of skeletal elements in compound but not in any of the single mutants, thus indicating genetic interaction. Since ROR2 is a non-canonical Wnt coreceptor specific for Wnt-5a we speculated that this phenotype might be a result of deregulated Wnt-5a signaling activation, which is known to be essential for limb skeletal elements growth and patterning. We show that Noggin potentiates activation of the Wnt-5a-Ror2-Disheveled (Dvl) pathway in mouse embryonic fibroblast (MEF) cells in a Ror2-dependent fashion. Rat chondrosarcoma chondrocytes (RCS), however, are not able to respond to Noggin in this fashion unless growth arrest is induced by FGF2. In summary, our data demonstrate genetic interaction between Noggin and Ror2 and show that Noggin can sensitize cells to Wnt-5a/Ror2-mediated non-canonical Wnt signaling, a feature that in cartilage may depend on the presence of active FGF signaling. These findings indicate an unappreciated function of Noggin that will help to understand BMP and Wnt/PCP signaling pathway interactions.

Keywords: noggin, Wnt5a, non-canonical Wnt pathways, BMP signaling, brachydactyly, Ror2

## INTRODUCTION

Limb bud development and the concomitant formation of limb skeletal structures are regulated by the intricate interplay and integration of various signaling pathways, with major roles played by the Shh, BMP, FGF, and Wnt/β-catenin pathways (reviewed for example in, Robert, 2007; Zuniga, 2015). The BMP signaling pathway is of pivotal importance especially for skeletal development. The analysis of inheritable human hand malformation syndromes has been instrumental in understanding the contribution of BMP signaling and other pathways for skeletal development. One example are the brachydactylies, a group of inheritable syndromes that are characterized by shortening or absence of phalanges. Most brachydactyly subtypes are caused by mutations in BMP signaling components or factors that, at different levels, intersect with BMP signaling. Therefore brachydactylies have been interpreted in terms of a molecular disease family (Stricker and Mundlos, 2011). This hypothesis predicts that overlapping phenotypes are likely caused by mutations affecting components that show a close functional interaction within a common signaling network.

Intriguingly, two closely related brachydactyly subtypes, BDB1 and BDB2, are caused by mutations in ROR2 or NOGGIN, respectively (Oldridge et al., 2000; Lehmann et al., 2007). While NOG is well known as a secreted BMP antagonist, ROR2 is an atypical receptor tyrosine kinase that is involved in the inhibition of Wnt/β-catenin signaling (Mikels and Nusse, 2006). In developing digits, Ror2-mediated Wnt/β-catenin inhibition allows BMP-mediated digit outgrowth (Witte et al., 2010). In addition, Ror2 is a Wnt (co)receptor, mainly for Wnt-5a, acting in non-canonical Wnt signaling (Oishi et al., 2003; Schambony and Wedlich, 2007). Recently, activation of the non-canonical Wnt/planar cell polarity (PCP) pathway by Wnt-5a and ROR2 was shown to be critically involved in the regulation of limb skeleton development (Gao et al., 2011; Wang et al., 2011; Ho et al., 2012; Kuss et al., 2014). Moreover, a separate set of mutations in ROR2 causes autosomal recessive Robinow syndrome (RS), which is characterized by diverse malformations including the axial and limb skeleton (Afzal et al., 2000; van Bokhoven et al., 2000). A dominant form of RS is caused by mutations in Wnt/PCP components DVL1, DVL3, and WNT-5A, it is therefore believed that the developmental defects seen in Robinow syndrome are caused by a deregulation of Wnt-5a/Ror2/PCP signaling (Stricker et al., 2017).

The skeletal elements of the limbs are formed by endochondral ossification. In this process a cartilage template is formed that mediates growth of the skeletal element and becomes later replaced by bone. This process is dependent on the formation of stacked columns of proliferating chondrocytes oriented perpendicular to the longitudinal axis of the growing skeletal element (Romereim and Dudley, 2011). Deregulation of PCP signaling in proliferating chondrocytes leads to perturbation of column formation, and to arbitrary chondrocyte orientation that ultimately leads to skeletal malformations typically resulting in a shortening and widening of the skeletal elements (Ahrens

Based on the close phenotypic overlap of human brachydactyly-causing mutations in ROR2 and NOG, we hypothesized that NOG may directly interact with the Wnt-5a/Ror2 pathway. We show here a subtle genetic interaction of Noggin with Ror2 during mouse limb development. Mechanistically, we provide evidence that Noggin can sensitize cells to Wnt/PCP pathway activation mediated by ROR2, providing first evidence for a yet uncharacterized level of cross-talk between BMP and Wnt/PCP signaling.

## MATERIALS AND METHODS

## Mouse Lines and Phenotypical Analysis

Ror2+/<sup>−</sup> (Takeuchi et al., 2000) and Nog+/<sup>−</sup> (McMahon et al., 1998) were maintained as heterozygous lines and intercrossed to yield compound mutants. Timed matings were set up and embryos were collected at E18.5. Skeletal preparations were performed as described previously (Mundlos, 2000). All animal procedures were carried out in accordance with European Union and German law. Animals were maintained in the SPF mouse facility of the Max Planck Institute for Molecular Genetics, Berlin under license from the Landesamt für Gesundheit und Soziales (LAGeSo) under license numbers ZH120 and G0346/13.

## Cell Culture and Treatments

Ror1−/<sup>−</sup> Ror2−/<sup>−</sup> mouse embryonic fibroblasts (MEF) were derived from Ror1 flox/flox Ror2 flox/flox MEF cells as described previously (Ho et al., 2012). MEF and RCS cells were propagated in DMEM, 10% FCS, 2 mM L-glutamine, 50 units/ml penicillin, and 50 units/ml streptomycin. RCS cells were seeded in 24 well plates, grown for 24 h and treated as indicated. Following reagents: Wnt-5a (R&D systems, 645-WN-010), Noggin (R&D Systems, 1967-NG-025), FGF2 (5 ng/ml, R&D Systems) and Wnt-C59 5 µM (Tocris Bioscience, 5148) were used for treatment. Wnt-5a conditioned media was produced from L Wnt-5a cells (ATCC CRL-2814) according to ATCC instructions. RCS cells intended for WB analysis were treated by FGF2 for 48 h, then were treated by the porcupine inhibitor Wnt-C59 (to reduce background autocrine Wnt activity), Noggin and Wnt-5a in indicated doses for additional 24 h. Total time of FGF2 treatment was 72 h.

## Western Blotting

Lysates for western blotting were prepared as follows: Growth medium was removed and cells were directly lysed in 100 mM Tris/HCl (pH 6.8), 20% glycerol, 1% SDS, 0.01% bromophenol blue and 1% 2-mercaptoethanol. Western blotting was performed according to manufacturer's instructions with minor adjustments [SDS-PAGE run on 150 V, transfer onto PVDF membrane 1 h on 100 V, both steps on ice (BIO-RAD)]. Antibodies were from Santa Cruz Biotechnologies: anti-Dvl2 (dephospho-Dvl2)–sc8026, anti-beta-Actin–sc1615-R, anti-Dvl3 sc8027 and from Cell Signaling Technologies: anti-Dvl2–CS3224. Anti-Ror2 antibody was a gift from Henry Ho (UC Davis) (Ho et al., 2012). Phosphorylation status of Dvl2 and Dvl3 was quantified by densitometric analysis of Western Blot in three independent replicates using Fiji distribution of ImageJ software as described (Bernatik et al., 2014). For pDvl/Dvl rations the peak area for the upper band representing P-Dvl was divided by the peak area of the lower band (Dvl). Data was analyzed by paired t-test (GraphPad Prism).

## Dual Luciferase Assay

RCS cells were transfected using pRLtkLuc and Super8X TopFlash plasmid. 9µg Super8X TopFlash and 3 µg pRLtkLuc plasmid were mixed with 38.4 µl of Fugene6 (E2691, Promega) in 1200 µl of DMEM. Cells were treated by 0.3% collagenase type II (GIBCO, cat.no.17101015) before transfection, 50 µl of transfection mixture and 500 µl of collagenase treated RCS cells in DMEM were used per 1 well of 24 well plate. Transfection was carried out overnight, cells were treated according to the experimental scheme for 20 h, and samples were processed by Dual-Luciferase <sup>R</sup> Reporter Assay System according to the manufacturer instructions (Promega, E1960).

## RESULTS

## Noggin Genetically Interacts with Ror2

To get a first indication whether Ror2 and Noggin might functionally interact we generated compound mutants for Ror2 and Noggin. Ror2+/<sup>−</sup> mice (Takeuchi et al., 2000) were crossed to Noggin+/<sup>−</sup> mice (Brunet et al., 1998; McMahon et al., 1998). Heterozygous inactivation of either Ror2 or Noggin does not result in any skeletal alteration (**Figure 1A**). In Ror2+/−;Nog+/−compound heterozygotes the overall appearance of the limb skeleton was normal; however the skeletal elements of the stylopod (the humerus) and the zeugopod (radius and ulna) showed a consistent small lateral expansion (**Figure 1A**, width of skeletal elements in wild type and single mutants indicated in yellow, width in compound mutant indicated in orange). All skeletal elements showed a tendency toward widening at both metaphyseal sides, however statistical significance was only reached for the distal humerus and radius, respectively. This feature was not seen in single heterozygotes, indicating genetic interaction between Nog and Ror2.

Ror2−/<sup>−</sup> mice are a model for RRS, recapitulating several of its features including mesomelic limb shortening as well as mild brachydactyly (Schwabe et al., 2004). Ror2−/<sup>−</sup> mice have shortened digits, however all phalanges (two in the thumb/digit 1, three in digits 2–5) as well as the interphalangeal joints separating the phalanges are present (Takeuchi et al., 2000; Schwabe et al., 2004; Schwarzer et al., 2009) (**Figure 1B**). Noggin heterozygous mice have phenotypically normal digits. When one allele of Noggin was removed on the Ror2−/<sup>−</sup> background, shortening of phalanges was further increased. In digit 3 the appearance of 3 individual phalanges, which were smaller than those in the Ror2−/−, was preserved. In digits 2 and 5 loss of one Noggin allele on the Ror2−/<sup>−</sup> background led to loss of an individual phalanx 2, concomitant with a longer phalanx 3, indicating failure of distal joint formation. Distal joint fusion is also a feature seen sometimes in BDB1 (ROR2 mutation) and frequently in BDB2 (NOG mutation). In addition, joint fusions are the hallmark of proximal symphalangism 1A (SYM1A) and multiple synostosis syndrome (SYSN1), two conditions caused by a different set of NOG mutations (Stricker and Mundlos, 2011). Altogether the compound mutants support the notion of a genetic and functional interaction of Ror2 and Nog in skeletal development.

## Noggin Potentiates Wnt/PCP Signaling in a Ror2-Dependent Manner

In digit formation, Ror2 acts in part via inhibition of β-catenin signaling leading to derepression of BMP/SMAD signaling in a structure called phalanx-forming region (Witte et al., 2010). Evidence however has accumulated that in addition or in parallel to this function Ror2 and its paralog Ror1 are both required for Wnt-5a/PCP signaling activation during digit development (Gao et al., 2011; Ho et al., 2012). Our genetic interaction experiments cannot distinguish the origin of the interaction seen, i.e., whether it originated from Nog function in the BMP pathway, or a yet uncharacterized role in the Wnt-5a/PCP pathway. Noggin thus might not only influence activity of BMP, but also of Wnt-5a-Ror2 pathway. To test if Noggin is able to activate Ror2 we treated mouse embryonal fibroblasts (MEF) with increasing doses of Noggin. The activation of endogenous Ror2 can be monitored as a phosphorylation-dependent mobility shift on Western blotting (Oishi et al., 2003). As we show in **Figure 2A**, even in the highest concentrations used (1,500 ng/ml) Noggin did not induce phosphorylation of Ror2 and was unable to promote phosphorylation of Ror2 induced by its cognate ligand Wnt-5a. This suggests that at the receptor level Noggin is unable to act either directly as a ligand for Ror2, or indirectly.

In the next step we tested if Noggin can promote any of the Ror2-downstream events. A robust readout of noncanonical Wnt pathways activation is the Wnt-5a-induced phosphorylation of Disheveled (Dvl) 2, an event dependent on the Ror1 and Ror2 receptors (Ho et al., 2012). We took advantage of an anti-Dvl2 antibody that recognizes only the inactive, dephosphorylated form of Dvl2 in MEF cells (Gonzalez-Sancho et al., 2013). Disappearance of non-phosphorylated Dvl2 currently represents one of the most sensitive tools for visualization of Dvl2 phosphorylation and hence Wnt/PCP pathway activation. When we treated MEF cells with increasing doses of Wnt-5a, the non-phospho Dvl2 signal disappeared (**Figure 2Bi**), indicative of activated Wnt-5a-Ror-Dvl signaling. No such phenotype was observed when cells were treated by Noggin, confirming our previous observation that Noggin itself is not able to activate signaling via Ror2 (**Figure 2Bii**). However, when cells were treated with 100 ng/ml of Noggin, we could clearly observe stronger effects of Wnt-5a on Dvl2 activation (compare **Figure 2Bi** vs. **Figure 2Biii**). This indicates that Noggin can sensitize MEF cells to Wnt-5a/Ror2 signaling. To confirm this observation, we treated cells with 25 ng/ml of Wnt-5a, which is a suboptimal dose unable to trigger Dvl2 activation (**Figure 2Bi**). When cells pre-treated by 25 ng/ml of Wnt-5a were supplemented with increasing doses of Noggin, activation of Dvl2 was observed in a dose dependent manner (**Figure 2Biv**),

indicating that presence of Noggin can reveal biological activity of previously sub-threshold Wnt-5a concentrations.

All these data suggest that Noggin, despite its inability to activate Ror2 on its own, can efficiently potentiate the Wnt-5a-Ror2 signaling axis and sensitize cells to low amounts of Wnt-5a. Ror2 can have redundant function with closely related Ror1 (Ho et al., 2012) that can also bind Wnt-5a. To confirm that the effects of Noggin are indeed dependent on Ror1/Ror2, Ror1−/<sup>−</sup> Ror2−/<sup>−</sup> double knockout MEF cells were isolated from conditional Ror1/Ror2 knockout mice (as described in Ho et al., 2012). Individual clones were tested by Western blotting (**Figure 2C**) and one of the Ror1/Ror2 double negative clones (#13) was further used for functional analysis. When Ror1/Ror2 deficient MEF cells were treated with 30 ng/ml of Wnt-5a and 500 ng/ml of Noggin simultaneously, no shift of Dvl2 mobility (upper panels) or effects on non-phospho Dvl2 (middle panel) was observed, in contrast to wt MEF where Dvl2 was activated by the combination of Wnt-5a (30 ng/ml) and Noggin (500 ng/ml) (**Figure 2D**). This data show that Noggin is able to potentiate the activation of the Wnt-5a-Ror2 signaling circuit and demonstrate that the observed Noggin/Wnt-5a synergism toward Dvl2 is dependent on Ror1/Ror2.

## FGF2-Induced Chondrocyte Growth Arrest Enables Noggin-Mediated Wnt/PCP Potentiation in RCS Cells

The genetic interaction between Ror2 and Noggin observed in mice as well as the skeletal involvement in human syndromes characterized by NOG and ROR2 mutations pointed toward the importance of a functional Noggin-Ror2 interaction for skeletal development. To test the Noggin-Ror2 synergy in a model system that is more relevant to skeletal development we decided to use the rat chondrosarcoma (RCS) cell line.

RCS chondrocytes maintain a fully differentiated proliferating chondrocyte phenotype in culture, manifested by abundant production of cartilaginous extracellular matrix rich in sulfated proteoglycans and collagen type 2, but not collagen type 10 characteristic for hypertrophic chondrocytes (Mukhopadhyay et al., 1995). Moreover, RCS chondrocytes faithfully recapitulate FGF-receptor 3 (FGFR3) signaling in the growth plate cartilage. Many essential features of FGFR3 signaling in the growth plate cartilage, such as the FGF-mediated chondrocyte growth-arrest have been unraveled using the RCS chondrocyte model system (Aikawa et al., 2001; Dailey et al., 2003; Krejci et al., 2005).

To define this experimental system, we first investigated whether the FGF-induced growth arrest in RCS cells is influenced by addition of Noggin and Wnt-5a. Noggin, Wnt-5a and/or their combination did not induce a growth arrest by themselves, and also did not modulate the FGF-induced growth arrest of RCS cells (**Figure 3A**). We also wanted to exclude that any possible observations in RCS cells are caused by modulation of canonical Wnt pathway that was shown to oppose Wnt/PCP pathway in chondrogenesis. Since it was shown that RCS cells are responsive to canonical Wnt ligands, e.g., Wnt3a (Krejci et al., 2012), we tested whether Noggin and Wnt-5a treatment alters the canonical Wnt pathway in RCS cells using TopFlash reporter assay. These results (**Figure 3B**) showed that Noggin and Wnt-5a could not activate or inhibit the canonical Wnt pathway even though RCS cells responded well to canonical Wnt ligands such as Wnt-3a (**Figure 3B**). We conclude that combined treatment of RCS cells with Noggin/Wnt-5a does not influence FGF2 induced growth arrest or the canonical Wnt signaling pathway in RCS cells.

Finally, we analyzed whether RCS cells respond to combined Noggin/Wnt-5a treatment similarly to MEF cells. As we could not detect any signal by using the dephospho-Dvl2 antibody used in MEF cells (not shown), we have used an alternative readout—electrophoretic mobility shift of Dvl induced by Wnt-5a. Such mobility shift indeed represents a phosphorylation and can be effectively abrogated by alkaline phosphatase (AP) treatment (**Figure 3C**). Using this readout we next tested whether Noggin could potentiate the response to Wnt-5a in RCS cells similarly as was observed in MEF cells. When RCS cells were treated by combination of Wnt-5a and Noggin, no potentiation of Wnt-5a-Ror2 signaling was observed (**Figure 3D**, quantified in **Figure 3E**), and only the highest dose of Wnt-5a triggered

(72 h), graph shows average and SD from two independent experiments, \*\*\*p < 0.001 [One-way ANOVA (ANalysis Of VAriance) with post-hoc Tukey test]. (B) Treatment of RCS cells by FGF2, Noggin and Wnt-5a does not activate canonical Wnt pathway analyzed by TopFlash reporter system. Treatment with Wnt3a was used as a positive control. RLU—relative light units, graph shows average and SD from two independent experiments, \*\*\*p < 0.001 [One-way ANOVA (ANalysis Of VAriance) with post-hoc Tukey test]. (C) Alkaline phosphatase (AP) treatment can remove the electrophoretic mobility shift of Dvl3 induced by FGF2/Noggin/Wnt-5a treatment, which suggests that the mobility changes (used in D–G) are caused by phosphorylation. (D) Wnt-5a can activate downstream signaling—visible as phosphorylation-dependent shift (p-Dvl) of Dvl2 and Dvl3—at 50 ng/ml and this effect is not positively modulated by the addition of Noggin (125 ng/ml). (E) Quantification of p-Dvl/Dvl ratios for Dvl2 and Dvl3 from three independent experiments. (F) Similar experiment as in (D) but RCS cells were pre-treated also by FGF2 (5 ng/ml) for total 72 h to induce FGFR3-mediated growth arrest. Under these conditions 24 h treatment by C59, Noggin (125 ng/ml) and Wnt-5a can dramatically induce the Wnt-5a-induced activation of Dvl2 and Dvl3. (G) Quantification of three independent experiments. \*p < 0.01 (paired t-test).

phosphorylation of Dvl2 and Dvl3. However, when RCS cells were pre-treated with FGF2 for 2 days in order to induce growth arrest (Krejci et al., 2010), Noggin dramatically improved the response of RCS cells to low doses of Wnt-5a (**Figure 3F**, quantified in **Figure 3G**). Importantly, acute treatment of RCS cells with FGF2, Noggin and Wnt-5a was unable to induce such "sensitization" (data not shown). These data thus argue that the synergism between Noggin and Wnt-5a-Ror2 is not a proximal effect of FGF2-induced signaling or an inhibition of the canonical branch of Wnt signaling but is rather induced by cell changes caused by prolonged FGF2 treatment and cell cycle arrest.

## DISCUSSION

Signaling pathways do not operate as standalone units but functionally cooperate and interact. Inspired by the phenotypic resemblance of BDB1 and BDB2, inheritable syndromes caused by mutations in ROR2 or NOGGIN, respectively, we decided to study how Noggin, an inhibitor of BMP pathway, and noncanonical Wnt signaling, driven by Ror2 receptor, can interact. We could show that Noggin increased biological activity of Wnt-5a and rendered cells sensitive to Wnt-5a concentrations otherwise not causing cellular responses. This function was dependent on the presence of Ror2, but Noggin did not elicit a signal on its own via Ror2.

Our study does not elucidate the molecular mechanism behind this interaction. One mechanism may involve BMP receptor type 1 b (Bmpr1b), which is mutated in BDA2 (Lehmann et al., 2003). In vitro, Ror2 and Bmpr1b were shown to interact and Ror2 is phosphorylated by Bmpr1b (Sammar et al., 2004, 2009). The functional consequence of this phosphorylation remains unclear but one can speculate that the effects of Bmpr1b on Ror2 are controlled by BMP ligands, whose active concentration is controlled by Noggin. Another possibility, which we were, however, not able to prove (data not shown) can be formation of Noggin-Wnt-5a-Ror2 ternary complex with the increased signaling capacity in comparison to Wnt-5a-Ror2 only. As another alternative, Noggin can, via regulation of BMP pathway, control signaling competence or cell surface amount of Ror2—here a possible point of crosstalk can be represented by Smurf family E3-ligases, which were reported to control both BMP pathway (negatively) as well as Wnt/PCP pathway (positively) (Narimatsu et al., 2009).

The importance of the BMP pathway and its tight regulation by antagonists for digit development is underscored by the fact that the majority of human brachydactylies are caused by mutations in different members of this signaling network (reviewed in Stricker and Mundlos, 2011). A necessity for integration of BMP and Wnt/β-catenin pathways has been reported for numerous developmental processes (Itasaki and Hoppler, 2010). For example, in digit outgrowth, BMP/SMAD signaling is fine-tuned by inhibition from the Wnt/β-catenin pathway, which itself is kept in check by Ror2 (Witte et al., 2010). Non-canonical (or alternative) Wnt pathways regulate entirely different aspects of tissue development compared to the Wnt/β-catenin pathway, but are connected with the BMP pathway as well, albeit the connection has not been studied to the same depth (Narimatsu et al., 2009; Schille et al., 2016). In developing limbs, Wnt/PCP signaling was involved in both digit shaping and outgrowth (Gao et al., 2011; Wang et al., 2011; Ho et al., 2012). Altogether this substantiates that both BMP and non-canonical Wnt pathways are required and act in concert during the establishment of the limb skeleton. Ror2 appears to be a pivotal intersection point between these two pathways.

Our work on RCS chondrocytes, a cell model for chondrocyte growth and differentiation that to some extent recapitulate the behavior of developing limb growth plate cartilage (Krejci et al., 2012) showed that Noggin could potentiate Wnt-5a-Ror2 pathway activity much more effectively when growth arrest was induced by FGF2 stimulation. It was previously shown in RCS chondrocytes that the FGF pathway can stimulate phosphorylation of LRP6, a co-receptor of the Wnt/β-catenin pathway (Krejci et al., 2012; Buchtova et al., 2015). We speculated that FGF signaling might be involved in activation of Wnt-5a-Ror2 in RCS cells, as it is known that Wnt/β-catenin and noncanonical Wnt pathways receptors can be activated by common mechanisms (Bryja et al., 2009; Grumolato et al., 2010). However, Wnt/β-catenin is likely not involved in the Noggin/Wnt-5a/Ror2 crosstalk in RCS cells because no differences in the activity analyzed by the TopFlash reporter system were observed.

Where can such FGF-dependent Noggin-induced activation of Wnt-5a-Ror2 signaling pathway in chondrocytes take place in vivo? In limb cartilage development, Wnt/PCP signaling appears to be involved at two steps: during condensation of cartilage elements, especially the digits (Gao et al., 2011; Wang et al., 2011; Ho et al., 2012), and for establishing cartilage growth plate morphology (Ahrens et al., 2009; Li and Dudley, 2009; Kuss et al., 2014; Romereim et al., 2014). In the first scenario, Wnt-5a is required for digit formation, and mice deficient for Wnt-5a form rudimentary digits (Yamaguchi et al., 1999). The Wnt-5a null phenotype is recapitulated by either Ror1/Ror2 double null mutants (Ho et al., 2012) or Ror2/Vangl2 double null mutants (Gao et al., 2011), clearly establishing that a Wnt-5a/Ror2/PCP pathway is necessary for digit formation. Noggin is expressed in forming cartilage condensations (Brunet et al., 1998) and could hence facilitate this process. During digit outgrowth, FGFs are expressed in the apical ectodermal ridge (AER). FGF signaling from the AER is thought to keep distal mesenchymal cells proliferating and undifferentiated (ten Berge et al., 2008). In vitro, FGFs inhibit chondrogenesis (Buchtova et al., 2015), but on the other hand application of FGF beads can induce ectopic digit formation in vivo (Montero et al., 2001). One possibility is that FGF signaling that acts at a distance from the AER on prechondrogenic cells provides competence for Noggin activity toward the Wnt-5a/Ror2/PCP pathway, and is thus enforcing PCP signaling in cells undergoing chondrogenic differentiation. In the growth plate, both Wnt-5a and Ror2 are essential for cellular polarity (Yang et al., 2003; Schwabe et al., 2004), and Wnt-5a acts via a PCP pathway (Gao et al., 2011; Kuss et al., 2014). Noggin is expressed throughout the growth plate (Brunet et al., 1998), and FGF signaling, which is a major regulator of growth plate chondrocyte proliferation, is active here as well (Horton et al., 2007).

In summary our data pinpoint a novel, yet unappreciated role for Noggin in sensitizing cells to Wnt-5a. The cellular mechanism by which Noggin accomplishes this effect on the Wnt-5a-Ror2 pathway remains to be elucidated.

## AUTHOR CONTRIBUTIONS

PK, SS, and VB designed research; MB, TR, OB, ZD, FW, AM, NC, and PK performed research; all authors analyzed data; and OB, SS, and VB wrote the paper.

## ACKNOWLEDGMENTS

This work was supported by the Czech Science Foundation (15-21789S, 17-16680S, 17-09525S). TR and ZD are supported by the Marie Curie ITN WntsApp. PK was supported by

## REFERENCES


Ministry of Education, Youth and Sports of the Czech Republic (KONTAKT II LH15231) and Ministry of Health of the Czech Republic (15-33232A, 15-34405A). This work was supported by the Deutsche Forschungsgemeinschaft (DFG SFB 577).

regulation of chondrocyte proliferation and extracellular matrix homeostasis. J. Cell Sci. 118, 5089–5100. doi: 10.1242/jcs.02618


development. Development 135, 3247–3257. doi: 10.1242/dev.023176


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

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

# Mechanical Control of Myotendinous Junction Formation and Tendon Differentiation during Development

Mauricio Valdivia, Franco Vega-Macaya and Patricio Olguín\*

Program in Human Genetics, Faculty of Medicine, Institute of Biomedical Sciences, Biomedical Neurosciences Institute, University of Chile, Santiago, Chile

The development of the musculoskeletal system is a great model to study the interplay between chemical and mechanical inter-tissue signaling in cell adhesion, tissue morphogenesis and differentiation. In both vertebrates and invertebrates (e.g., Drosophila melanogaster) the formation of muscle-tendon interaction generates mechanical forces which are required for myotendinous junction maturation and tissue differentiation. In addition, these forces must be withstood by muscles and tendons in order to prevent detachment from each other, deformation or even losing their integrity. Extracellular matrix remodeling at the myotendinous junction is key to resist mechanical load generated by muscle contraction. Recent evidences in vertebrates indicate that mechanical forces generated during junction formation regulate chemical signaling leading to extracellular matrix remodeling, however, the mechanotransduction mechanisms associated to this response remains elusive. In addition to extracellular matrix remodeling, the ability of Drosophila tendon-cells to bear mechanical load depends on rearrangement of tendon cell cytoskeleton, thus studying the molecular mechanisms involved in this process is critical to understand the contribution of mechanical forces to the development of the musculoskeletal system. Here, we review recent findings regarding the role of chemical and mechanical signaling in myotendinous junction formation and tendon differentiation, and discuss molecular mechanisms of mechanotransduction that may allow tendon cells to withstand mechanical load during development of the musculoskeletal system.

## Keywords: tendon cells, myotendinous junction, mechanical forces, morphogenesis, mechanoresponse

## INTRODUCTION

Living cells and tissues are in a constant state of isometric tension allowing them to respond to mechanical cues (Ingber, 1997; Wang et al., 2001; Mammoto and Ingber, 2010). During embryogenesis, mechanical stress is generated within the tissue and by its interaction with external factors and/or other tissues. Shear stress generated by blood flow modulates blood vessels morphogenesis, regulates the fate acquisition of arteries and veins, and is required for the development of the hematopoietic system (le Noble et al., 2004; Adamo et al., 2009; North et al., 2009). In addition, hemodynamic forces are required for heart morphogenesis. Disturbing blood flow at either the inflow or outflow tracts of the zebrafish heart results in several defects including abnormal formation of third chamber and heart looping (Hove et al., 2003). Furthermore, mechanotransduction mechanisms and its role in development are evolutionary conserved across

#### *Edited by:*

Juan Jose Sanz-Ezquerro, Consejo Superior de Investigaciones Científicas (CSIC), Spain

#### *Reviewed by:*

Peter Bräunig, RWTH Aachen University, Germany Krzysztof Jagla, Institut National de la Santé et de la Recherche Médicale (INSERM), France Juan Jose Perez-Moreno, University of Cambridge, UK

> *\*Correspondence:* Patricio Olguín patricioolguin@med.uchile.cl

#### *Specialty section:*

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

> *Received:* 30 December 2016 *Accepted:* 07 March 2017 *Published:* 23 March 2017

#### *Citation:*

Valdivia M, Vega-Macaya F and Olguín P (2017) Mechanical Control of Myotendinous Junction Formation and Tendon Differentiation during Development. Front. Cell Dev. Biol. 5:26. doi: 10.3389/fcell.2017.00026 species. In zebrafish and Drosophila, mechanical cues generated during gastrulation (epiboly in zebrafish, and mesoderm invagination in flies) induce β-Catenin release from E-Cadherin based junctions, and translocation to the nucleus of mesodermal cells, where it promotes gene expression changes and cell specification (Farge, 2003; Desprat et al., 2008; Brunet et al., 2013).

The development of muscle-tendon attachment is a great model to study the role of chemical and mechanical signaling between tissues in morphogenesis and differentiation (Schweitzer et al., 2010; Subramanian and Schilling, 2015). During embryogenesis, tendon cells attach to the developing muscle through the Extracellular Matrix (ECM) forming a specialized junction called Myotendinous Junction (MTJ) (Schweitzer et al., 2010; Subramanian and Schilling, 2015). MTJ development relays mainly on the interaction of Integrins and ECM molecules secreted by tendons and muscles, although, other proteins, like Dystroglycan and Kon-tiki (Kon) also contribute to the formation of the MTJ. While Dystroglycan participates on muscle binding to the ECM, Kon controls muscle guidance and attachment to muscle attachment sites (Pérez-Moreno et al., 2014; Weitkunat et al., 2014; Maartens and Brown, 2015; Subramanian and Schilling, 2015). Strain generated by the contraction of the developing muscles contributes to MTJ maturation and muscle and tendon differentiation (Weitkunat et al., 2014; Havis et al., 2016). Here we will review recent evidences regarding the role of mechanical signaling in tendon differentiation and MTJ formation in vertebrates and Drosophila. Additionally, we will discuss the mechanisms of mechanoresponse that may allow tendon cells to sense and respond to mechanical load during development of the muscle-tendon interaction.

## THE ROLE OF MECHANICAL AND CHEMICAL SIGNALING IN VERTEBRATE TENDON DIFFERENTIATION

Mechanical control of tendon differentiation and remodeling has been widely studied in vertebrates (reviewed in Shwartz et al., 2013). Tendons are formed by ECM, composed principally by strong collagens fibril arrays, and a type of fibroblast termed tenocyte (Subramanian and Schilling, 2015). In response to mechanical forces, tenocytes secrete collagens and proteoglycans, modifying ECM composition and elastic properties (Chen X. et al., 2012; Li et al., 2015). These changes confer tendons with the ability to resist mechanical load generated during muscle contraction and to form functional attachments to bones (Evans and Barbenel, 1975; Kjaer and Kjær, 2004; Maeda et al., 2011; Schwartz et al., 2013; Havis et al., 2016). How force is sensed by tenocytes and transduced into a cellular response? Recent studies on the development of the MTJ shed lights into this problem. In chicks and mice, the morphogenesis of the limb MTJ is divided in two phases (Subramanian and Schilling, 2015). The first phase is independent of muscle derived signals (Pryce et al., 2009). Here, the initial expression of Scleraxis (Scx), a tendon-specific bHLH transcription factor that promotes tendon differentiation and tenocyte specification (Alberton et al., 2012; Chen L. et al., 2012; Li et al., 2015), is stimulated by Fibroblast Growth Factor (FGF) and Transforming Growth Factor-beta (TGFβ) through MAPK/ERK and SMAD2/3 signaling pathways, respectively (Schweitzer et al., 2001; Havis et al., 2014; **Figure 1A**). Scx mutant mice display disrupted tenocyte differentiation leading to disorganized ECM, however, tenocyte precursor cells are still specified, indicating that other genes are required for early specification (Murchison et al., 2007). During the second phase of tendon differentiation, the interaction with the developing myofiber is mandatory to maintain the expression levels of Scx and other tendon markers (Havis et al., 2016). Pharmacological inhibition of muscle contraction disturbs tendon differentiation, even in presence of FGF and TGFβ, diminishing the levels of Scx. Moreover, force exerted by muscles on tendons is required for the activation of FGF and TGFβ at the muscle-tendon interface, maintaining the expression levels of Scx, leading to tendon terminal differentiation (Maeda et al., 2011; Havis et al., 2016).

TGFβ-ligands are secreted bound to TGFβ-binding proteins which form a complex with the large latency complex (LLC) in the ECM, capturing TGFβ and precluding its binding to TGFβ-receptors (Wipff et al., 2007; Maeda et al., 2011; **Figure 1A**). Shearing forces generated during muscle contraction may stimulate TGFβ release from the LLC through its degradation by proteases, allowing its binding to the receptor (**Figure 1A**). Moreover, it may promote the activation of Integrin signaling through the binding of the RGD motifs present on the latency TGFβ binding proteins associated to LLC (Munger and Sheppard, 2011; Subramanian and Schilling, 2015; **Figure 1A**). TGFβ signaling maintains Scx expression under normal muscular-load regime in mice (Maeda et al., 2010, 2011) and in response to mechanical stress promotes expression of Integrins (Popov et al., 2015). Thus, different mechanotransduction mechanisms appear to function at the ECM levels, activating either TGFβ or Integrin signaling. In vertebrates, recent evidence have shown that mechanical forces appears to be required for muscle development. Mechanical force driven by muscle contraction is necessary to maintain the pool of muscle progenitors during chick fetal myogenesis (de Lima et al., 2016), and in vitro studies suggest that strain drives mesenchymal stem cells differentiation into myoblasts (Lisio et al., 2014; Lemke and Schnorrer, in press).

## THE ROLE OF MECHANICAL SIGNALING IN *DROSOPHILA* MYOTENDINOUS JUNCTION FORMATION AND TENDON DIFFERENTIATION

In contrast to vertebrates, Drosophila displays an exoskeleton instead of an internal skeleton and its connection with muscles relays on epithelial cells of ectodermal origin called tendon cells, which are analogs to vertebrate tendons (Fernandes et al., 1996; **Figure 1B**). Similar to vertebrates, signals emanated from tendon cells are required for MTJ formation, both during embryogenesis and metamorphosis (Costello and Wyman, 1986; Fernandes et al., 1991; Wayburn and Volk, 2009; Ordan

as Thrombospondin (Tsp) and Laminin (Lam), are secreted to the MTJ. In tendon cells, SrA is expressed and SrB expression diminishes. (IV) Myotube compacts generating mechanical stress on the system triggering myofibrillogenesis. (V) Sarcomeres are formed, and muscle elongate back toward tendon cells.

et al., 2015). In order to resist mechanical load, tendon cells modify their elastic properties deploying an array of polarized microtubules and actin filaments that stretch along their apical-basal axis, from the exoskeleton attachment site to the MTJ (Subramanian et al., 2003; Alves-Silva et al., 2008).

The development of the interaction between the Indirect Flight Muscles (IFMs) and the tendon cells of the dorsal thorax (notum) is an interesting model to study the role of mechanical signaling in tissue morphogenesis and cell differentiation (Olguín et al., 2011; Weitkunat et al., 2014). The notum develops from a monolayer epithelium, from which a subset of epithelial cells differentiates as analogs to vertebrate tendons, serving as bridges between the flight muscles and the exoskeleton (Fernandes et al., 1991; Weitkunat et al., 2014). At early stages of tendon differentiation, tendon precursors are specified by the activity of the isoform B of the Stripe transcription factor (SrB), which is required and sufficient to specify tendon cells (Volk and VijayRaghavan, 1994; Frommer et al., 1996; Becker et al., 1997; **Figure 1C**). The stripe homologous in vertebrates, Egr1 and Egr2, are required for tendon terminal differentiation, specifically to promote the expression of ECM proteins (Frommer et al., 1996; Lejard et al., 2011; Guerquin et al., 2013), however, as Scx, they are not strictly required for tendon specification (Lejard et al., 2011; Guerquin et al., 2013). Once specified, embryonic tendon cells provide initial attracting cues to the myotube and secrete Slit, a ligand that binds Robo receptor, which is expressed at the tips of myotubes (**Figure 1C**; Kramer et al., 2001; Ordan et al., 2015). Whether Slit acts as a chemoattractant in this context, remains to be elucidated. During this first stage of myotendinous system development, myotubes extend bipolar extensions that migrate toward their tendon targets, conversely, tendon cells extend processes that interact with the myotube extension tips (**Figure 1C**; Vega-Macaya et al., 2016). Muscle migration requires the accumulation of Kon, a single pass transmembrane protein, on the muscle leading ends (**Figure 1C**; Estrada et al., 2007; Schnorrer et al., 2007). Loss of function of Kon in the ventral longitudinal muscles causes abnormal projection of filopodia, altering the myotube migration pattern (Schnorrer et al., 2007). Following, in a second stage, myotubes secrete Vein, a short range signaling molecule that binds to the epidermal growth factor receptor (EGFR) expressed in tendon cells, promoting SrB expression (Yarnitzky et al., 1997; **Figure 1C**). High levels of SrB induce Slit secretion and Leucine Rich repeat Transmembrane protein (LRT) expression, which bind to Robo and are both required for muscle migration arrest (**Figure 1C**; Wayburn and Volk, 2009; Ordan and Volk, 2015, 2016). Slit acts as a short range repellent signal that arrests muscle migration. This mechanism depends on Slit cleavage by Amontillado, a Pheromone Convertase 2 homolog, sequestering Slit on the tendon cell membrane, stopping muscle migration (Ordan et al., 2015; Ordan and Volk, 2016). In a third stage, the MTJ starts forming mainly through the association of Integrin with ECM proteins secreted by tendon and myotube (Chanana et al., 2007; Subramanian et al., 2007; Gilsohn and Volk, 2010; **Figures 1B,C**). The musclespecific αPS2βPS Integrin binds to Thrombospondin (Tsp) and its regulator Slow, conversely, Laminin (Lam) associates with the tendon-specific αPS1βPS Integrin (Gotwals et al., 1994; Martin et al., 1999). The induction of SrA isoform and the decrease of SrB expression levels is essential to promote the expression of tendon specific differentiation genes such as Delilah (Dei), a transcription factor that promotes βPS expression, and shortstop/kakapo (Shot), a plakin that connects the actin cytoskeleton to microtubules, regulating the elastic properties of tendon cells (Subramanian et al., 2003; Schweitzer et al., 2010). Thus, during this stage EGFR and Integrin signaling promotes junction formation and terminal differentiation of tendon cells.

During metamorphosis, developing tendons and muscles express the same combinations of Integrin subunits and secrete extracellular matrix components such as Tsp, forming stable hemiadherent junctions (Subramanian et al., 2007; Gilsohn and Volk, 2010; Weitkunat et al., 2014). Following, IFMs compaction, driven by Myosin Heavy Chain (MHC) motor activity, generates mechanical strain at the MTJ (Weitkunat et al., 2014; **Figure 1C**). In addition, the overlying notum epithelium migrates toward anterior through a still unknown mechanism, which may contribute to the mechanical strain generated between these tissues (Bosveld et al., 2012). Recently, it has been shown that mechanical strain at the MTJ is required for myofibrillogenesis, indicating that mechanical signaling is also required for muscle morphogenesis (Weitkunat et al., 2014). In response to muscle compaction, tendon extensions attached to the myotube elongate (Weitkunat et al., 2014; **Figure 1C**). During this process, MTJ must be able to withstand mechanical load, and tendon cells might regulate its elastic properties in order to maintain its integrity and shape.

## MEMBRANE MECHANORECEPTORS AND MECHANICAL SIGNALING AT THE MYOTENDINOUS SYSTEM

At focal adhesions, the Integrin signaling pathway might be triggered in response to deformation or changes in the rigidity of the ECM (outside-in activation) (Takagi et al., 2003; Campbell and Humphries, 2011). In absence of external forces, Integrins remain in a restings state, associated with Filamin (**Figure 2A**). Mechanical stimuli may cause the opening of the extracellular domains of the Integrin heterodimer, which is transmitted to its cytoplasmic portion where it could recruit the actin binding protein Talin, although it is not the most characterized mechanism of Integrin signaling (Nieves et al., 2010; **Figure 2B**). The activation of Integrins also results in the recruitment of several other proteins, like Src kinases, promoting cell proliferation and migration (Arias-Salgado et al., 2003). Importantly, Src activates Rho signaling pathway, which through Rho-kinase (ROCK) induces the phosphorylation of the myosin regulatory light chain (MRLC) and the contraction of the acto-myosin network, building up tension at the focal adhesions (Arthur et al., 2000; Arias-Salgado et al., 2003). The Integrin signaling cascade may be activated also by an insideout mechanism (Otoole et al., 1994; Vinogradova et al., 2002). There is evidence that certain proteins, like Talin, are able to respond to mechanical deformation (Lee et al., 2007; del Rio et al., 2009). In vitro studies have shown that Talin has cryptic vinculin interacting domains that are exposed by deformation (Lee et al., 2007; del Rio et al., 2009). Stretching of the actin cytoskeleton may be directly transmitted to Talin, releasing its Vinculin binding site, triggering the recruitment of Talin and Vinculin toward Integrins, promoting adhesion.

Integrin-ECM interaction plays an important role in the formation of the vertebrate and Drosophila MTJ (Brown, 2000; **Figure 1A**). Drosophila mutant embryos for either βPS Integrin or tsp show detachment of developing muscle fibers from tendons, due to the loss of the αPS2βPS-Tsp interaction (Chanana et al., 2007; Subramanian et al., 2007; **Figure 1B**). Moreover, Talin mutants display similar defects suggesting that Talin-related signaling is required for functional MTJ formation (Brown et al., 2002).

Similar to Focal adhesion, activation of Rho signaling downstream of Integrins appears to be indispensable for MTJ formation (reviewed in Geiger and Bershadsky, 2002). We have recently shown that Drosophila Rho-kinase (DRok) loss of function in tendon cells results in diminished phosphorylation of MRLC and abnormal βPS localization and Tsp accumulation at the MTJ, suggesting that DRok could be part of the insideout mechanism of Integrin activation (Vega-Macaya et al., 2016). Interestingly, ROCK activity appears to be required for stretchinduced tenocyte differentiation from human Mesenchymal Stem Cells (hMSCs) (Xu et al., 2012). Stretching of hMSCs elicited enhanced expression of Scx, Collagen I and II, among other tendon specific genes. The addition of a ROCK inhibitor results in an attenuated expression of these genes (Xu et al., 2012). Whether DRok activity contributes to tendon

cell differentiation in response to mechanical forces, through regulation of gene expression in Drosophila remains to be explored. In addition to its role in MTJ maturation, DRok regulates the orientation of tendon extensions toward IFMs during recognition stage, enabling the correct attachment to the muscle fibers (Vega-Macaya et al., 2016; **Figure 1C**). DRok mutant tendon cells display miss-oriented tendon extensions, resulting in irregular attachments to the muscle fiber. Tendon extensions appears to be unable to resist the pulling forces generated by IFMs compaction, resulting in muscle detachment and death (Vega-Macaya et al., 2016). How DRok regulates tendon recognition of the myotube ends remains to be elucidated.

In conclusion, the membrane mechanoreceptor model explains how forces are sensed and transduced at the MTJ, but how tension exerted by muscle compaction is withstood by the whole tendon cell is still unclear.

## ACTIN CROSSLINKERS AS INTRACELLULAR MECHANOSENSORS AND REGULATORS OF THE ACTIN NETWORK

In vitro and in vivo experiments show that mechanical perturbation of cell shape causes a redistribution of actin crosslinkers and a rearrangement of the actin network (Gardel et al., 2004; Chaudhuri et al., 2007; Luo et al., 2013). Studies in Drosophila epithelial cells, Dyctiostelium discoideum and mammalian cells have demonstrated that mechanical deformation of the plasma membrane results in accumulation of crosslinking and motor proteins such as Filamin and myosin, respectively, to the perturbation site in distinctive ways (Fernandez-Gonzalez et al., 2009; Luo et al., 2013; Schiffhauer et al., 2016; **Figure 2C**). Myosin is recruited to regions under dilation stress, counteracting cell deformation by contraction of the acto-myosin filaments (**Figure 2C**). On the other hand, Filamin is recruited to sites subjected to shear stress (Luo et al., 2013; Schiffhauer et al., 2016; **Figure 2C**).

In contrast to Myosin, Filamin does not act as a contractile unit; instead, it enhances elasticity of the actin network to allow cell shape adaptation and remodeling (Luo et al., 2013; Schiffhauer et al., 2016). Filamin is a large actin-binding protein that works as a dimer (Noegel et al., 2004). Each Filamin monomer binds to one actin filament forming orthogonal and elastic actin networks by dimerization via their C-terminal immunoglobulin-like domains (Tseng et al., 2004; Pudas et al., 2005; **Figures 2A–C**). Both, Jitterbug, one of the two Filamins present in Drosophila, and non-muscle Myosin II (MyoII) are required to maintain the shape and polarity of tendon cells and partially co-distribute with actin filaments and Shot (Olguín et al., 2011). Interestingly, Shot loss of function display similar epithelial deformation phenotypes to Jbug (Olguín et al., 2011), suggesting that both microtubule and actin arrays that stretched along the apical-basal axis of tendon cells are required to withstand mechanical load.

At the signal-transduction level, Filamin acts as a scaffold for other actin regulatory proteins (Popowicz et al., 2006). In monocytes, Filamin recruits the small GTPases of the Rho family, their effectors and regulators (Leung et al., 2010). In migrating mammalian cells, Filamin recruits ROCK (Ueda et al., 2003), which may promote acto-myosin network contraction and stabilization by activation of the myosin regulatory light chain, α-Adducin and LIMK (Maekawa et al., 1999; Zhang et al., 2003). During cell migration, Filamin also interacts with the Integrin beta subunit, keeping it in a resting state, preventing focal adhesion formation (Liu et al., 2015; **Figure 2A**). A proposed

## REFERENCES


mechanism is that after Integrin interaction with a stiffer ECM, Filamin dissociates from Integrin cytoplasmic domain leading to Talin and Vinculin recruitment in its place, reinforcing adhesion (Nieves et al., 2010; **Figure 2B**).

Based on these evidences, Filamin could play a dual role in tendon cell mechanoresponse during MTJ formation: as a molecular scaffold for actin regulators at the MTJ, and as regulator of tendon cell elastic properties at specific cellular regions. Moreover, Filamin redistribution may regulate its role as a scaffold at the MTJ.

## CONCLUDING REMARKS

The ability of cells and tissues to respond to mechanical stress during development is crucial to shape organs and the whole individual. The combination of molecular tools that allows to measure in developing animals, mechanical stress across developmental fields, dynamic signaling pathway activity and cytoskeleton organization will be key to unveil the interplay between mechanical and chemical signaling during embryogenesis, including the formation of the musculoskeletal system.

## AUTHOR CONTRIBUTION

MV and FV contributed equally to this work. MV wrote sections of the manuscript, then contributed to its editing and final formatting. FV made **Figures 1**, **2**. PO contributed to writing, editing materials written by MV and FV, and final integration of the various sections.

## FUNDING

This work was supported by DRiDANS, PIA ACT-1401 and Biomedical Neuroscience Institute, Iniciativa Científica Milenio, ICM P09015F.

## ACKNOWLEDGMENTS

We thank Gonzalo Olivares for advice and discussion during the preparation of the manuscript.

beta cytoplasmic domain. Proc. Natl. Acad. Sci. U.S.A. 100, 13298–13302. doi: 10.1073/pnas.2336149100


and migration during embryonic and imaginal development. J. Cell Biol. 145, 191–201. doi: 10.1083/jcb.145.1.191


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

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

# Putting Cells into Context

Sigmar Stricker <sup>1</sup> \*, Petra Knaus <sup>2</sup> \* and Hans-Georg Simon<sup>3</sup> \*

<sup>1</sup> Musculoskeletal Development and Regeneration Group, Institute for Chemistry and Biochemistry, Freie Universität Berlin, Berlin, Germany, <sup>2</sup> Cell Signaling and Regeneration Group, Institute for Chemistry and Biochemistry, Freie Universität Berlin, Berlin, Germany, <sup>3</sup> Department of Pediatrics, Feinberg School of Medicine, Northwestern University and Stanley Manne Children's Research Institute, Chicago, IL, USA

Keywords: cell signaling, mechanobiology, extracellular matrix, artificial matrices, material sciences

## CELLS LIVE IN A COMPLEX WORLD

It may sound blatantly obvious, but we have to remind ourselves occasionally that in vivo cells experience an environment with a level of complexity far beyond experimental reach. The developing organism is a highly complex system, where each cell receives a multitude of cues of diverse nature at any given time point. Only the comprehensive integration of all these multivalent interactions determines the actual signaling state and hence the behavior of a cell.

#### Edited by:

José Lozano, University of Málaga, Spain

#### Reviewed by:

Stefano Biressi, University of Trento, Italy Vania Braga, Imperial College London, UK

#### \*Correspondence:

Sigmar Stricker sigmar.stricker@fu-berlin.de Petra Knaus knaus@chemie.fu-berlin.de Hans-Georg Simon hgsimon@northwestern.edu

#### Specialty section:

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

> Received: 16 January 2017 Accepted: 21 March 2017 Published: 05 April 2017

### Citation:

Stricker S, Knaus P and Simon H-G (2017) Putting Cells into Context. Front. Cell Dev. Biol. 5:32. doi: 10.3389/fcell.2017.00032

The analysis of biological questions is mainly inspired by a reductionist approach adopted from the "exact sciences," where it has been proven immensely successful. That is, we are used to break down our experimental setup to a manageable number of variables. This of course is inherently contradictory to the complexity of biological systems. While simplification may be the only viable option for the experimenter to dissect biological function down to detail, it has also influenced our perspective toward the experimental systems applied. For example, studies of intracellular signaling pathways are typically performed with cultured cells. Culturing cells in an in vitro setting became a standard model system in biomedical research and with it in cell and developmental biology. These simplified systems allow for the dissection of molecular interactions and pathways and are aimed to deepen and mechanistically understand cellular behavior. While cell cultures have generated a wealth of information into cellular function, the data obtained in vitro frequently are in conflict with in vivo observations. One reason for this discrepancy is that these analyses focus on the cell as a closed functional system, thus conceptually unhinging it from its environment.

In a living organism, cells are embedded in extracellular matrix (ECM) with diverse but also organ-specific properties. Cells attach to the ECM mainly via focal adhesions, which on the inside are linked to the cytoskeleton (**Figure 1A**). The ECM has long been seen as a mere scaffold providing support and shape; however, in the past decades it has become clear that the ECM has also an instructive character (see Adams and Watt, 1993; Tsang et al., 2010). Like a color palette ECMs come in many shades with different molecular composition, resulting in manifold chemical and physical characteristics (Rozario and DeSimone, 2010). We focus here on a rather simple but often overlooked property that provides tissues with their rigidity or elasticity. It is these mechanical properties that emerged as a decisive factor mediating information flow (see Mammoto et al., 2013). There is a multitude of interactions between cells and their ECM and it is now well accepted that cells perceive the substrate's mechanical cues and integrate them into their intracellular signal transduction pathways, gene expression and cell fate decisions. In this sense cells can be both writers and readers of ECM and its cues, implying crosstalk between cells via the ECM. Besides determining mechanical tissue properties, cells build a matrix with spatial decoration of specific growth factors to thereby modulate their local availability. Emerging data even suggest that intracellular signaling pathways integrate external biomechanical cues directly by altering the phosphorylation state of cytosolic signaling proteins (Kopf et al., 2012, 2014; Ashe, 2016).

## BIOMECHANICAL PROPERTIES OF THE ECM AND ITS INFLUENCE ON CELLS

For a long time it has been known that cells can, based on intrinsic sensing mechanisms, differentially respond to, for example, growth factor signaling (e.g., Nakagawa et al., 1989). The importance of the mechanical aspect, however, has only gained wider attention in recent years. In 2006, a landmark study from the Discher group showed that in vivo tissues exhibit an elastic modulus in the range of below 1 kPa (brain) up to over 100 kPa (ossified bone). Moreover, they provided the first evidence that this property has an instructive character on the behavior of progenitor cells, specifically mesenchymal stem cells (MSCs). The cells sensed their microenvironment by attaching and applying force to the substrate. Plating MSCs on polyacrylamide substrates of varying stiffness revealed a differentiation potential correlating with the stiffness of the in vivo tissue; in other words, MSCs seeded on extremely soft substrates differentiated along the neural fate, while cells seeded on hard substrates differentiated along the osteogenic fate, and intermediate stiffness substrates supported differentiation along a myogenic fate (Engler et al., 2006).

Experiments using dynamic modulation of substrate stiffness further revealed that cells initially cultured on soft (0.5 kPa) or stiff (40 kPa) polyacrylamide hydrogels and then transferred to gels of the opposite stiffness had the capacity to revert their gene expression profile from neurogenic to osteogenic, and vice versa (Lee et al., 2014). However, while the cells displayed a remarkable potential switching lineage specification, MSCs transferred from stiff to soft substrates maintained elevated osteogenesis markers; thus, they kept a memory of their previous culture conditions indicating a certain degree of irreversible, likely epigenetically fixed, lineage commitment.

Just recently it has been discovered that partial matrix stress relaxation is another fundamental signal in cell-ECM interactions (**Figure 1B**). Stress relaxation means that the force cells exert on the ECM dissipates, and over time ECM resistance decreases. Chaudhuri et al. engineered alginate polysaccharide hydrogels that are, independent from their initial elastic modulus, also tunable in their viscoelasticity. Thus, they mimic the remodeling of the matrix microenvironment over time. MSCs embedded in 17 kPa-stiff hydrogels with a rapid rate of stress relaxation demonstrated enhanced spreading, proliferation and osteogenic differentiation (Chaudhuri et al., 2016). In addition to force dissipation it is likely that remodeling of the ECM by Matrix Metalloproteinase (MMP) activity contributes to this behavior in vivo.

How do cells perceive these stimuli and translate them into transcriptional activity? It was shown that mechanotransduction of ECM stiffness toward MSC differentiation critically depends on YAP (yes-associated protein) and TAZ (transcriptional coactivator with PDZ-binding motif) signaling. In this context, YAP/TAZ were not activated by the "canonical" Hippo/LATS cascade, but by cytoskeletal tension and Rho-GTPase activity (Dupont et al., 2011). Interestingly, deregulation of YAP/TAZ signaling has been linked to disease conditions characterized by ECM stiffness changes such as fibrosis (Liu et al., 2015), and in this context apparently a circuit with TGFβ and WNT signaling pathways exists (Piersma et al., 2015). It is noteworthy that the interplay of ECM stiffness and mechanosensing itself impinges on the expression of profibrotic genes, driving a feed forward vicious cycle (Parker et al., 2014).

Another example are muscle resident stem cells, the so-called satellite cells (SCs). SCs reside in a specific niche underneath the myofiber's rigid basal lamina where they are kept in a quiescent state that is dependent on different factors including the collagen, glycoprotein and proteoglycan-rich ECM (e.g., Brohl et al., 2012; Bentzinger et al., 2013). In an injury situation SCs are activated and form new muscle, but importantly they also self-renew. After isolation and culture, the expansion of SCs in vitro and their expression of myogenic transcription factors was shown to be influenced by the elasticity of the culturing substrate (Gilbert et al., 2010), a feature recently utilized to create artificial niches maintaining satellite cell quiescence ex vivo (Quarta et al., 2016).

The importance of the biochemical and biophysical properties of extracellular matrices on myogenesis has been coherently demonstrated in a vertebrate in vivo/in vitro regeneration model. During amphibian limb and cardiac regeneration, the collagen/laminin-rich matrix typical for differentiated tissues is temporarily replaced by a transitional matrix of reduced stiffness composed of hyaluronic acid, tenascin-C and fibronectin that is surprisingly similar to the type found in developing structures (Calve et al., 2010; Mercer et al., 2013). Employing these regeneration-permissive ECMs in in vitro cultures, Calve et al. demonstrated an instructive role of distinct ECM components promoting cell fragmentation, proliferation, migration and differentiation of ex vivo skeletal muscle cells (Calve et al., 2010). In addition, using a polydimethylsiloxane (PDMS)-ECM culture system that allowed for the modulation of both, stiffness and matrix composition they could further demonstrate that ECM type and substrate stiffness over a range of 2–100 kPa combine to control migration as well as differentiation state of skeletal muscle cells (Calve and Simon, 2012).

Mechanical and biophysical properties of the ECM are provided by coordinated synthesis and secretion of matrix components with protein or sugar backbones and biologically active epitopes that result in a network of different biomolecules. Dynamic post-translational modifications including MMP cleavage further shape a characteristic local signaling environment. These native 3D structures also serve as versatile surfaces for the binding of growth factors—either in their active form or inactive preform—which, often as a consequence of mechanical stress, will be released in a spatially controlled manner. Mimicking these complex in vivo conditions using a 3D bioreactor with a collagen scaffold as a simplified in vitro culture system, the Knaus and Petersen groups provided evidence that biomechanical signaling is directly integrated into the BMP/Smad pathway (Kopf et al., 2012). Coapplication of mechanical stress and BMP stimulation resulted in increased and prolonged phosphorylation of Smads, the direct target of the transmembrane BMP receptor kinases. As a consequence, distinct target genes, including known mechanotransducers, were upregulated in a synergistic manner.

## CONSEQUENCES FOR IN VITRO STUDIES

Clearly, studying the complex in vivo interactions of cellgrowth factor, cell-cell, and cell-matrix interactions and their downstream intracellular signal transduction and gene expression pathways, we will also in the future have to rely on simplified in vitro culture systems. As by default cells apparently integrate mechanical and biochemical inputs, the cellular behavior experimentally determined is in consequence dependent on the in vitro culture conditions and not necessarily reflect cellular behavior seen in vivo.

The standard tissue culture method is still the plastic dish, with an elastic modulus of approximately 10<sup>6</sup> kPa way out of the physiological range. When a more natural environment is desired, plastic dishes are at best coated with a thin layer of mostly collagenous matrices such as gelatine or MatrigelTM . This, however, rather serves as a functionalization of the surface toward better cell adherence rather than altering the mechanical properties of the substrate. In light of the growing body of evidence from the emerging field of mechanobiology we have to change course.

Time has come to move on to more comprehensive in vitro culture systems that better simulate the complex in vivo conditions. Recent approaches employing engineered biopolymers as mimetics of the natural environment provide new opportunities to develop more physiological cell culture procedures. The material sciences have made available a range of different tested hydrogels; of particular interest are those made from biologically inert polymers including polyacrylamide, PDMS, alginate, and polyethylene glycol (PEG). All of these synthetic polymers allow, to various degrees, for the tuning of stiffness over a range of 2–40 kPa (similar to that observed in natural tissues), presentation of native matrix-derived peptide epitopes, and/or binding and release of growth factors. Ideally, these cell culture models would be in a 3D architecture resembling the in vivo context as closely as possible. However, building a perfect mimetic of the in vivo environment is virtually impossible in a standard cell culture experiment when analyzing, for example, intracellular signaling using current routine reporter assays. It therefore appears as a minimal requirement for a more comprehensive experimental design to at least consider the biomechanical properties of the tissue of origin, i.e., the mechanical modulus. A realistic rational approach could be 2D culturing techniques with softer synthetic matrix substrates (as compared to the hard plastic dish) that mimic the in vivo viscoelastic conditions, which we think would greatly improve the reliability of ex vivo/in vitro experimentation and improve comparability to in vivo data. It is important to note that the fabrication of such biomimetic matrices in the laboratory is still challenging and coupled to an operating expense that clearly hinders their standard application. However, custom products are beginning to enter the market and it is foreseeable that a panel of matrix solutions will become available in the near future tailored to many if not most individual experimental needs.

Importantly, embracing more physiological cell culture conditions might generate a fundamentally new understanding of how extracellular cues, both insoluble and soluble, are integrated and stored to guide cellular behavior. These immediate biological goals would further help to achieve current and future therapeutic challenges in humans (see Sommerfeld and Elisseeff, 2016).

## AUTHOR CONTRIBUTIONS

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

## ACKNOWLEDGMENTS

Research in the Stricker lab on this topic is funded by the Berlin-Brandenburg School for Regenerative Therapies (BSRT). In the Knaus lab this research topic is funded by the DFG (FOR2165)

## REFERENCES


and the Berlin-Brandenburg School for Regenerative Therapies (BSRT). This manuscript was inspired during a sabbatical stay of HGS at the Knaus lab, who was supported through a Kennedy Sabbatical Award of Northwestern University's Department of Pediatrics and a gift of the Family of Michael P. Polsky.

activation and fibrosis. Am. J. Physiol. Lung. Cell Mol. Physiol. 308, L344–L357. doi: 10.1152/ajplung.00300.2014


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

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