The Role of Chordin and Twisted Gastrulation in Regulating BMP Family Growth Factors

A thesis submitted to the University of Manchester for the degree of Doctor of Philosophy in the Faculty of Life Sciences

2013

Helen Troilo

Contents

1. Introduction 21

1.1 Bone Morphogenetic Proteins 22

1.1.1 Bone Morphogenetic Protein Structure 22

1.1.2 BMP Receptor Binding 23

1.1.3 Smad Signalling 24

1.1.4 Non-Canonical BMP Signalling 26

1.1.5 Effects of BMP Signalling 26

1.2 Chordin 26

1.2.1 Chordin Binding and Kinetics 27

1.2.2 Chordin Structure 29

1.2.3 Structure of the Antagonistic Complex 31

1.3 Twisted Gastrulation 32

1.4 Tolloid Metalloproteinases 34

1.5 Other Modulators of the BMP-Chordin-Tsg Ternary Complex 36

1.5.1 BAMBI 36

1.5.2 Ont1 37

1.5.3 Sizzled 37

1.5.4 Integrins 37

1.5.5 Proteoglycans 37

1.5.6 Fibronectin 38

1.6 Development 38

1.6.1 Chordin, Tsg and BMPs in Embryo Patterning 38

1.6.2 Chordin Shuttling 41

1.6.3 Differences in Insect Models 42

1.6.4 Differences in Mammals 43

1.6.5 Chordin in the Development of Other Species 46

1.7 Mathematical Modelling of the Complex in Development 47

1.7.1 Modelling Strategies 47

1.7.2 Computer Models of Ternary Complex Function 47

1.8 Medical Implications 48

1.8.1 Congenital Disorders 49

1.8.2 Cancer 50

1.8.3 Vascular Disease 51

1.8.4 Arthritis 52

1.8.5 Bone Disorders 53

1.8.6 Delivery Methods 53

1.8.7 Safety 54

1.8.8 Industrial Applications 55

1.9 Aims 55

2. Materials and Methods 58

2.1 Materials 59

2.1.1 Buffers and Media 59

2.1.2 Bacterial Strains 59

2.1.3 Expression Cells 59

2.1.4 Plasmid Vectors 59

2.2 Molecular Cloning 59

2.2.1 Recombinant DNA Constructs 59

2.2.2 Mutagenesis Primers 60

2.2.3 Polymerase Chain Reaction 61

2.2.4 Agarose Gel Electrophoresis and Gel Extraction DNA Purification 62

2.2.5 Construct Ligation into Cloning Vector 63

2.2.6 Construct Ligation into Expression Vector 63

2.2.7 Bacteria Culture and Transformation 63

2.2.8 DNA Purification from Bacteria 64

2.2.9 DNA Sequencing 64

2.2.10 DNA Storage 64

2.3 Protein Expression and Purification 64

2.3.1 HEK293-EBNA Cell Culture 64

2.3.2 Lipofectamine Transfection 65

2.3.3 HEK-293 EBNA Storage 65

2.3.4 Protein Expression in Serum Free Media and Storage 65

2.3.5 Nickel Affinity Chromatography 66

2.3.6 Protein Concentration by Centrifugation 66

2.3.7 Size Exclusion Chromatography 66

2.3.8 Protein Concentration Estimation 67

2.3.9 Stable Complex Formation 67

2.4 Protein Characterization 67

2.4.1 Polyacrylamide Gel Electrophoresis 67

2.4.2 Western Blotting 67

2.4.3 Circular Dichroism 68

2.4.4 PNGase F Digestion 68

2.4.5 MASS Spectrometry Identification 68

2.5 Biomolecular Analysis 69

2.5.1 Multi-Angle Light Scattering 69

2.5.2 Analytical Centrifugation 69

2.5.2.1 Velocity Sedimentation 69

2.5.2.2 Sedimentation Equilibrium 70

2.5.3 Transmission Electron Microscopy 70

2.5.4 Small Angle X-Ray Scattering 70

2.5.5 Structural Data Analysis 71

2.5.6 Surface Plasmon Resonance 71

3. Results Chapter 1: Biomolecular Analysis of Chordin 72

3.1 Summary 73

3.2 Cloning and Expression of Chordin Constructs 73

3.3 Protein Purification and Fragment Separation 75

3.4 Circular Dichroism 77

3.4.1 Prediction of Secondary Structure of Full Length Chordin using SRCD 77

3.4.2 CD Secondary Structure Prediction of Chordin ∆C and ∆N 79

3.5 Surface Plasmon Resonance 80

3.5.1 Kinetic SPR Analysis 80

3.5.2 Equilibrium SPR Analysis 82

3.6 Multi-Angle Light Scattering 83

3.7 Chordin forms a Reversibly Associating Dimer 85

3.8Sedimentation Velocity Analytical Ultracentrifugation of Chordin ∆N 87

3.9 Discussion 87

4. Results Chapter 2: Nanostructure of Chordin and

Chordin Fragments 91

4.1 Summary 92

4.2 Small Angle X-ray Scattering from Full Length Chordin 92

4.2.1 Gnom Fit to Experimental Data from SAXS 93

4.2.2 Guinier Analysis and Kratky Plots 94

4.2.3 Ab initio Modelling of Full Length Chordin 94

4.2.4 SAXS Analysis using Low Concentration Chordin 96

4.3 Small Angle X-ray Scattering of Chordin Fragments 98

4.3.1 Data Analysis 98

4.3.2 Ab Initio Modelling of the Chordin Fragments 100

4.4 Transmission Electron Microscopy Models of Chordin Fragments 101

4.4.1 TEM Class Averages from Single Particle Analysis 101

4.4.2 3D Reconstruction of the Chordin Constructs From TEM Data 103

4.4.3 Fourier Shell Correlation Curve Interpretation 104

4.5 Ab initio Model Docking and Structure Cross-Validation 105

4.5.1 Bead Modelling of Chordin SAXS Data using Hydropro 105

4.5.2 Prediction of Hydrodynamic Properties from TEM Models 106

4.5.3 Fitting of SAXS and TEM models of Full Length Chordin 108

4.6 Discussion 108

5. Results Chapter 3: Structure and Biomolecular Analysis of

Twisted Gastrulation and its Interaction with Chordin 110

5.1 Summary 111

5.2 Expression and Purification of Tsg 111

5.2.1 Cloning of Tsg 111

5.2.2 Tsg Expression and Purification 112

5.3 The Tsg Constructs are Glycosylated, Folded and are Able to Bind

to Chordin 113

5.3.1 Circular Dichroism Analysis of Tsg Secondary Structure 113

5.3.2 Surface Plasmon Resonance Confirms Binding 115

5.3.3 Treatment of Tsg with PNGase-F 116

5.4 Small Angle X-Ray Scattering and Modelling of Tsg 116

5.4.1 Gnom Fit to the Experimental Data of from SAXS 116

5.4.2 Guinier Analysis Indicates a Monodisperse Sample 118

5.4.3 Kratky Plot Analysis 119

5.4.4 Ab initio Modelling of Tsg 120

5.5 Biomolecular Characterization of Tsg Using MALS and AUC 122

5.5.1 Multi Angle Light Scattering Shows that Tsg is a Stable Monomer 122

5.5.2 Velocity Analytical Ultracentrifugation of Native Tsg 123

5.5.3 Bead Modelling of Tsg Using Hydropro 124

5.6 Fluorescence Resonance Energy Transfer 125

5.6.1 Cysteine and Histidine Tag Linked Dyes 125

5.6.2 Amide-Linked Dyes 126

5.7 Production of a Stable Tsg-Chordin Complex 128

5.8 Chordin ∆C forms Fibres in the Presence of Tsg 129

5.9 Discussion 130

6. Final Discussion 132

6.1 Data Summary 133

6.2 Comparison to Previous Literature 135

6.2.1 Twisted Gastrulation Structure 135

6.2.2 Chordin Structure 135

6.2.3 The Effect of Tolloid Cleavage on Chordin 137

6.2.4 Interactions of the Ternary Complex 139

6.3 Future Directions 141

6.3.1 Double-Tagged Chordin to Further Improve Purity 141

6.3.2 CHRD 1-4 Domains 141

6.3.3 vWC-vWC Interactions 142

6.3.4 Interactions of the Ternary Complex 142

6.3.5 Tolloid-Chordin Complex 143

6.3.6 Differential Regulation of BMPs by Tolloid Cleavage 144

6.4 Summary 144

7. References 146

Appendix I: Vector Maps 168

Appendix II: Buffer List 170

Appendix III: DNA and Peptide Sequences 172

Appendix IV: Mass Spectrometry Peptide Hits 175

Word Count: 46,990

List of Figures

1. Introduction

Figure 1.1 BMP structure and processing 23

Figure 1.2 Schematic diagram showing BMP signalling pathways 25

Figure 1.3 Chordin domains 27

Figure 1.4 vWC domain structure 30

Figure 1.5 Twisted gastrulation schematic diagrams 34

Figure 1.6 Chordin cleavage sites 35

Figure 1.7 The BMP morphogen gradient 38

Figure 1.8 Effects of BMP gradient disruption in zebrafish patterning 40

Figure 1.9 Interaction between crossveinless-2 and the complex 42

Figure 1.10 Chordin knockout in mice 44

Figure 1.11 Twisted gastrulation knockout in mice 46

Figure 1.12 BMP involvement in atherosclerotic calcification 52

Figure 1.13 Protein constructs to be developed for study 57

3. Results Chapter 1: Biomolecular Analysis of Chordin

Figure 3.1 Cloning of chordin constructs 74

Figure 3.2 Chordin purification 76

Figure 3.3 CD trace of full length chordin 78

Figure 3.4 CD traces for chordin ∆C and ∆N 80

Figure 3.5 Kinetic analysis of chordin-BMP-2 binding 81

Figure 3.6 Equilibrium analysis of chordin-BMP-2 binding 83

Figure 3.7 MALS analysis of chordin 84

Figure 3.8 Equilibrium AUC of chordin 85

Figure 3.9 Kd analysis using velocity AUC 86

Figure 3.10 Ultracentrifugation profile of chordin ∆N 87

4. Results Chapter 2: Nanostructure of Chordin and Chordin Fragments

Figure 4.1 Full length chordin Gnom fit and P(r) plot 93

Figure 4.2 Full length chordin Guinier and Kratky plots 94

Figure 4.3 Ab initio models of full length chordin 95

Figure 4.4 SAXS analysis of low concentration chordin 97

Figure 4.5 Gnom fits, P(r) plots and Guinier plots for the chordin constructs 99

Figure 4.6 Chordin fragment Ab initio models 100

Figure 4.7 Superimposed ab initio models of chordin 101

Figure 4.8 Class averages from single particle analysis of TEM Data 102

Figure 4.9 EM models from single particle reconstruction TEM 103

Figure 4.10 Fourier Shell Correlation Curves 104

Figure 4.11 Hydromic output Data from TEM ab initio models 107

Figure 4.12 Fitting of SAXS and TEM Models 108

5. Results Chapter 3: Structure and Biomolecular Analysis of Twisted

Gastrulation and its Interaction with Chordin

Figure 5.1 Tsg purification 113

Figure 5.2 Tsg constructs are folded, glycosylated and able to bind chordin 114

Figure 5.3 Gnom fit and P(r) plots for Tsg 117

Figure 5.4 Guinier plots of the SAXS scattering Data 119

Figure 5.5 Kratky plots of the SAXS scattering Data for Tsg 120

Figure 5.6 Ab initio Tsg models 121

Figure 5.7 MALS analysis of Tsg 123

Figure 5.8 Analytical ultracentrifugation profile of native Tsg 124

Figure 5.9 Schematic diagram summarizing the FRET experiments 125

Figure 5.10 FRET analysis of Tsg binding 127

Figure 5.11 Production of chordin-Tsg complexes 129

Figure 5.12 Chordin ∆C-Tsg fibres 130

Discussion

Figure 6.1 Schematic diagram of selective cleavage hypothesis 136

Figure 6.2 Schematic diagram of fragment inactivation hypothesis 138

Figure 6.3 Chordin Domain Layout and Complex 140

Appendix I: Vector Maps

Figure 1 pCR2.1-TOPO cloning vector 168

Figure 2 pCEP Pu-AC7 expression vector 169

Appendix III: DNA and Peptide Sequences

Figure 1 Annotated chordin sequence 172

Figure 2 Full length chordin construct peptide sequence 173

Figure 3 Tsg construct DNA sequence 174

Figure 4 Tsg construct peptide sequence 174

List of Tables

2. Materials and Methods

Table 2.1 Primers used to produce DNA constructs 60

Table 2.2 Mutagenesis primers 61

Table 2.3 PCR cycling parameters 62

3. Results Chapter 1: Biomolecular Analysis of Chordin

Table 3.1 Mass spectrometry identification 77

Table 3.2 Summary of SRCD Data 79

Table 3.3 Summary of CD Data 80

Table 3.4 MALS analysis of chordin 84

Table 3.5 CD comparison table 88

4. Results Chapter 2: Nanostructure of Chordin and Chordin Fragments

Table 4.1 Hydropro calculations 105

Table 4.2 Summary of Hydromic predictions 107

Table 4.3 Summary of hydrodynamic properties 109

5. Results Chapter 3: Structure and Biomolecular Analysis of Twisted

Gastrulation and its Interaction with Chordin

Table 5.1 Table summarizing analysis of CD and SRCD Data for Tsg 115

Table 5.2 Automated SAXS Data analysis 118

Table 5.3 Summary of the structural properties of Tsg 131

Abstract

The Role of Chordin and Twisted Gastrulation in Regulating BMP Family Growth Factors

A thesis submitted to the University of Manchester for the degree of Doctor of Philosophy in the Faculty of Life Sciences

Helen Troilo 24th May 2013

Bone morphogenetic proteins (BMPs) are growth factors of the TGF-β superfamily. They are essential to early embryonic patterning, and are of interest in a range of pathologies. The secreted glycoprotein chordin binds to BMPs and antagonizes their signalling by preventing them from interacting with their cell surface receptors. This antagonism is relieved by cleavage of chordin at two specific points by tolloid family metalloproteinases. Twisted gastrulation has a dual role in regulating this pathway. It binds to both chordin and BMPs strengthening their interaction but also promotes cleavage of chordin by tolloids through an unknown mechanism.

In this study the structures of chordin and twisted gastrulation are investigated using a range of in solution techniques including analytical ultracentrifugation, small angle X-ray scattering and multiangle light scattering. These show that chordin is bowl shaped with the C-terminal region looping back toward the N-terminus while twisted gastrulation is more linear. It is shown for the first time that the vWC domains of chordin have self-affinity and that the C-terminal vWC containing region is flexible. We propose a mechanism by which this flexibility may allow more than one binding arrangement between chordin and BMPs depending on BMP specificity. This in turn allows for selective regulation by tolloids. In addition potentially novel functions for the chordin fragments including conformational change and fibre formation in the presence of Tsg were investigated.

Declaration

No portion of the work referred to in this thesis has been submitted in support of an application for another degree or qualification of the University of Manchester or any other university or institute of learning.

Copyright Statement i. The author of this thesis (including any appendices and/or schedules to this thesis) owns any copyright in it (the “Copyright”) and she has given the University of Manchester the right to use such Copyright for any administrative, promotional, educational and/or teaching purposes. ii. Copies of this thesis, either in full or in extracts, may be made only in accordance with the Copyright, Designs and Patents Act 1988 (as amended) and regulations issued under it or, where appropriate, in accordance with licensing agreements which the University has from time to time. This page must form part of any such copies made. iii. The ownership of any Copyright patents, designs, trademarks and other intellectual property (the “Intellectual Property”) and any reproductions of copyright work in the thesis, for example graphs and tables (“Reproductions”), which may be described in this thesis, may not be owned by the author and may be owned by third parties. Such Intellectual Property and Reproductions cannot and must not be made available for use without the prior written permission of the owner(s) of the relevant Intellectual Property and/or Reproductions. iv. Further information on the conditions under which disclosure, publication and commercialisation of this thesis, the Copyright and any Intellectual Property and/or Reproductions described in it may take place is available from the Director of the Wellcome Trust Centre for Cell-Matrix Research and the Dean of the Faculty of Life Sciences, in the university IP policy (see http://documents.manchester.ac.uk/DocuInfo.aspx?DocID=487), in any relevant Thesis restriction declarations deposited in the University Library, The University Library’s regulations (see http://www.manchester.ac.uk/library/aboutus/regulations) and in the University’s policy on Presentation of Thesis.

Acknowledgements

Dr Clair Baldock (Supervisor)

Dr Hilary Ashe (Advisor)

All members of the Baldock and Kielty Labs

Dr Thomas Jowitt and Mrs Marjorie Howard (University of Manchester Biomolecular Analysis Facility)

Dr Richard Collins (University of Manchester Electron Microscopy Facility)

The staff of the Diamond Light Source (Oxford) and Petra III

I would like to take this opportunity to thank my parents Carlo and Rosalind Corby, and my husband Febo for their love and support during my PhD studies.

The work described here was funded by the BBSRC and the Wellcome Trust.

List of Abbreviations

ADMP Anti-dorsalizing Morphogenetic Protein

AUC Analytical ultracentrifugation

BAMBI BMP and Activin Membrane Bound Inhibitor

BMP Bone morphogenetic protein

CCN Connective tissue growth factor, Cysteine rich protein,

and Nephroblastoma overexpressed gene

CD Circular Dichroism

Chordin ∆C Chordin fragment generated by cleavage of vWC4

Chordin ∆N Chordin fragment generated by cleavage of vWC1

CHRD Chordin domains

CTF Contrast Transfer Function

CUB Complement 1r/s, Uegf and BMP1

Dmax Maximum Diameter

DMEM Dulbecco’s Modified Eagle Medium

DMSO Dimethyl sulfoxide dNTP diNucleotide Triphosphate dpp decapentaplegic

DRAGON Name derived from Dorsal Root Ganglion 11

DSD Diaphanospondylodystosis

EBNA Epstein-Barr virus nuclear antigen

ECL Enhanced Chemiluminescence

EGF Epidermal growth factor

FBS Foetal Bovine Serum

18 f/f0 frictional ratios

FOP Fibrodysplasia Ossificans Progressiva

FSC Fourier Shell Correlation

FRET Fluorescence Resonance Energy Transfer gbb glass-bottom boat

GDF Growth and Differentiation Factor

HEK Human Embryonic Kidney Cells

HPLC ` High Performance Liquid Chromatography kDa kilo Daltons

MALS Multiangle light scattering

MAPK Mitogen Activated Protein Kinase

MES 2-(N-morpholino)ethanesulfonic acid

MOPS 3-(N-morpholino)propanesulfonic acid

MTLL Mammalian tolloid like

Nel Neural epidermal growth factor (EGF)-like molecule

NRMSD Normalized root mean square deviation

NSD Normalized standard deviation

Ont1 Olfactomedin–Noelin–Tiarin protein 1

PBST Phosphate Buffered Saline with Tween20

PCR Polymerase chain reaction

PKC Protein Kinase C

PMSF Phenylmethanesulfonyl Fluoride

P(r) Intramolecular distance distribution function

PVDF Polyvinylidene difluoride

Rg Radius of gyration

Rh Hydrodynamic radius

19 rhBMP recombinant human Bone Morphogenetic Protein

RMSD Root Mean Square Deviation

S Svedberg (10-13 seconds)

o S20W Sedimentation coefficient corrected to pure water at 20 C

Sapp Apparent sedimentation

SAXS Small angle X-ray scattering

Scw Screw

SD1/2 Subdomain 1 and 2

SDS Sodium Dodecyl Sulfate

Smad Some mothers against decaptapentaplegic

SOC Super Optimal broth with Catabolic repressor

Sog Short gastrulation

SOG Sog domains (Drosophila equivalent of CHRD domains)

SPR Surface plasmon resonance

SRCD Synchrotron Radiation Circular Dichroism

TBST Tris Buffered Saline with Tween20

TEM Transmission electron microscopy

TGF-β Transforming Growth Factor-β

Tris tris(hydroxymethyl)-aminomethane

Tsg Twisted gastrulation

UV Ultra-violet

VSMC Vascular Smooth Muscle Cells vWC von Willebrand factor C homology domain

1. Introduction

1.1 Bone Morphogenetic Proteins

Bone morphogenetic proteins (BMPs) are a family of growth factors involved in both adult processes and early development, all of which bind to the same family of cell surface receptors. These receptors activate Smad transcription factor signalling intracellularly which leads to regulation of target genes. Most notably they regulate dorsoventral patterning of the early embryo, tissue repair and bone deposition1. They are widely expressed and are also able to diffuse significant distances in vivo however their activity is modulated by other secreted proteins including chordin, twisted gastrulation and noggin. BMP signalling and BMP regulation is very highly conserved which has led to it being studied in a broad range of species2.

1.1.1 Bone Morphogenetic Protein Structure

BMPs are small (~14kDa) single domain proteins which form hetero- and homodimers shown in Figure 1.1A. There are at least sixteen known variants with overlapping and distinct expression patterns and effects. The monomers lack a hydrophobic core but are held together by three intrachain disulfide bridges forming a cysteine knot. A hydrophobic interface between monomers then stabilises the dimer3. The main structural variation between types of BMPs are secondary structure elements in the loop regions and in the surface charges in two helix-finger clefts and two central cavities. The dimer surface features positive, negative and hydrophobic regions. It is these hydrophobic regions which are responsible for binding receptors and are the primary targets for antagonists4, 5. An additional effect of these localized charge and hydrophobic patches is unusual solubility properties and a tendency toward aggregation3. BMPs are expressed with a prodomain which is cleaved prior to secretion, shown in Figure 1B. Some BMPs, such as BMP-7 and BMP-9 have been shown to be secreted as stable complexes with their non-covalently associated propeptides, which may have a role in targeting and blocking interaction with antagonists, but do not prevent interaction with receptors6. In addition, differential processing of some pro-BMPs (such as pro-BMP-4) appear to have a role in altering BMP specificity7.

Figure 1.1: BMP structure and processing. (A) Front and side view ribbon representation of the BMP-2 dimer showing the wrist and knuckle binding epitopes. α-helices are shown in pink, β-sheets in gold and the disulfide bonds of the cysteine knot in cyan3, 8. Figure generated using Jmol9. (B) Schematic diagram of expression and intracellular processing of pro-BMP BMPs are expressed with a large N-terminal pro-domain which is cleaved prior to secretion. Some BMPs remain non-covalently associated with their prodomain6.

1.1.2 BMP Receptor Binding

BMP dimers bind to two type I and two type II receptor serine/threonine kinases called bone morphogenetic protein receptors (BMPRs) bringing them into close proximity. There are three known type I receptors (BMPRIA, BMPRIB and activin receptor IA) and three type II receptors (BMPRII, activin receptor IA and activin receptor IB). Different combinations of receptors have some overlapping and some distinct effects even with the same ligand4. BMPs have two types of hydrophobic BMPR binding domain, shown in Figure 1.1A, the knuckle epitope which binds to type II receptors and the wrist epitope which binds to type I receptors8. Both BMPs and their receptors undergo conformational change in their binding interfaces prior to association. For example; upon interaction with BMP-2, BMPRIa adapts to fit a prehelix binding “hotspot” loop in the wrist epitope of BMP-25. The same ligand may

23 adapt slightly differently to different receptors which may explain the ligand-receptor promiscuity which is characteristic of the BMP family5, 10. In addition to their signalling receptors BMPs also interact with pseudoreceptors (e.g. BAMBI) and coreceptors (e.g. DRAGON and endoglin)11.

1.1.3 Smad Signalling

Activation of the Smad transcription factors (called MAD in invertebrates) is known as the canonical BMP signalling pathway shown in Figure 1.2. BMP-bound BMPRII transphosphorylates BMPRI which goes on to phosphorylate receptor Smads -1, -5 and -8, activating them. These heteroligomerize with co-Smad (Smad-4) and translocate to the nucleus where they modulate the activity of target genes. Smads work both by interacting with other transcription factors and by displacing them12. Inhibitory Smads (Smads -6 and -7) may associate with receptor Smads instead of co-Smad to block signalling. The pathway can also be interrupted by other mechanisms such as phosphorylation by mitogen activated protein kinases (MAPK)13.

Figure 1.2: Schematic diagram showing BMP signalling pathways. (A) Canonical BMP signalling through Smad transcription factors. Activated BMPRs phosphorylate Smads 1, 5 and 8 which bind to co-Smad (Smad4) and translocate to the nucleus to regulate transcription11 ,12. (B) Non-canonical BMP signalling pathways influence transcription through promoting protein kinase C (PKC) and blocking MAPK signalling13, 14, 15. Apoptosis can also be induced through initiating caspase cascades16.

1.1.4 Non-Canonical BMP Signalling

The non-canonical mechanisms of BMP signalling include via the p38 MAPK pathway which interacts with the Smad pathway13, 15. In addition the BMPRII carboxy terminal domain promotes apoptosis and cell death following activation by BMPs, through activation of caspases -3, -8 and -9 and down regulation of the pro-survival factor Bcl-216. It is thought that this is, at least in part, mediated by BMP activation of protein kinase C14. The non- canonical pathways have not been fully mapped and it is expected that BMPs will be found to signal through other uncharacterized mechanisms17. BMPs also have indirect effects on other signalling pathways as a result of gene activation and suppression, for example they suppress nodal and antagonize transforming growth factor β (TGF-β) signalling by promoting catabolism of TGF-β receptors18.

1.1.5 Effects of BMP Signalling

BMP signalling controls many processes both inside and outside the cell. Through direct and secondary expression responses, BMP signalling ultimately exerts some degree of influence on almost all aspects of cell behaviour. This in turn means it is involved in a wide range of physiological processes including adipocyte generation19, stress response erythropoiesis20, stem cell self-renewal, angiogenesis11 and osteoblast generation21. The effects are strongly dependant on the specific BMP and BMPR combinations involved4. For example BMP-2 activates a different set of receptors in blood vessel development to BMP-10, with the result that one combination promotes angiogenesis and the other blood vessel maturation11.

1.2 Chordin

Chordin, is a 120kDa secreted glycoprotein which binds to and antagonizes BMPs -2, -4, -7 and anti-dorsalizing morphogenetic protein (ADMP) by preventing them from interacting with their receptors. This limits BMP signalling and indirectly promotes the activity of sonic hedgehog and nodal by removing BMP antagonism of these pathways1. Chordin, like other antagonists such as noggin and gremlin, covers the receptor-binding interfaces of BMP. Structural features shared between these antagonists and the BMPs themselves (such as cysteine knots) suggest that they derived from a common ancestral protein in a simpler organism22. The antagonism of BMPs by chordin is relieved by tolloid family metalloproteinases to yield fragments which retain reduced biological activity1. The chordin family share many structural and functional similarities and include sog (the Drosophila equivalent of chordin), the chordin-like proteins, crossveinless-2 and kielin-like proteins. All contain repeating cysteine-rich von Willebrand Factor C (vWC) homology domains which are the binding sites for BMPs1.

1.2.1 Chordin Binding and Kinetics

Chordin has four vWC and four chordin specific domains (CHRD) domains shown in Figure 1.3A. While the function of the CHRD domains has yet to be determined, the vWC domains are the binding sites responsible for all known chordin interactions. For chordin, crossveinless-2 and chordin-like 2 it has been shown that they bind to the same sites on BMPs as the BMPRs do; the knuckle, wrist or both epitopes8. Mutations of key residues in the knuckle and wrist epitope of BMP-2 have demonstrated that chordin binds mainly to the knuckle epitope, with mutations in the wrist having little or no effect on affinity8. Chordin has been shown to bind BMP-4 in the nM range and chemical crosslinking experiments indicate that this is in a one chordin to one BMP dimer ratio23. Overall chordin binds to BMP-2 most strongly with BMPs -7 and -4 having lower affinity and with weak affinity for ADMP8. It is not yet clear whether chordin can bind to BMP heterodimers, where it may bind to one BMP partner but not the other (such as BMP-2/-6 heterodimers)24. It is therefore possible that chordin has a wider capacity for BMP antagonism than has been demonstrated so far.

Figure 1.3: Chordin domains. (A) Schematic diagram showing the domain layout of chordin and known interaction sites with its major binding partners bone morphogenetic proteins (BMP), crossveinless-2, twisted gastrulation (Tsg) and Olfactomedin–Noelin–Tiarin protein 1 (Ont1). (B) Possible binding modes for chordin and BMPs; (i) similar to the predicted mode for crossveinless-2 with 2:2 stoichiometry, (ii) expected for chordin and BMP-2, (iii) expected for chordin and BMP-4 a 1:2 stoichiometry predicted from crosslinking data.

With the exception of vWC2, all the cysteine rich domains of chordin show some binding affinity for BMPs8 but individual vWC domains have different specific BMP preferences. The specificity of vWC1 is promiscuous but vWC3 binds strongly to BMP-2 but only weakly to BMP-7. Conversely vWC4 binds strongly to BMP-7 but only weakly to BMP-2. Interestingly in sog, despite the similarity in function and domain layout, vWC3 and 4 seem to be the dominant BMP binding domains25. In chordin most individual vWC domains tested can bind to BMPs with dissociation constants in the nanomolar range with similar or slightly reduced affinity to full length chordin26. The domains can produce similar phenotypes to chordin when overexpressed in vivo though with a reduction in biological activity which appears disproportionate to the similarity in binding affinity27.

The binding of other chordin family members varies in affinity and specificity. Chordin-like 1 differs from chordin in that it appears to be dependent on twisted gastrulation for its BMP antagonistic activity while chordin-like 2 binds preferentially to the wrist epitope of BMPs instead of the knuckle8. The individual vWC1 domains of crossveinless-2 interact with BMP dimers in a 2:1 ratio though it is has not been shown whether the full length protein behaves the same way. While increased chordin always reduces BMP signalling levels in vitro, crossveinless-2 produces a biphasic response. BMP signalling increases with increasing crossveinless-2 to a maximum but decreases at higher concentrations of crossveinless-2. This is thought to be a mechanism where crossveinless-2 binds to both BMPs and their receptors delivering one to the other, but segregates them at higher concentrations. The exact concentration at which the second phase predominates varies between BMPs so it is possible for crossveinless-2 to promote signalling of one BMP type whilst antagonizing another28

As well as binding to BMPs, vWC homology domains often have affinity to each other and are becoming recognised as a common interaction domain between many vWC containing proteins. This is highly significant because many proteins even outside the chordin family feature similar domains including Neural epidermal growth factor (EGF)-like molecule (Nel) family proteins, cysteine-rich protein connective tissue growth factor, and Nephroblastoma overexpressed gene (CCN) family proteins and amnionless29. Members of the chordin family are known to interact via this mechanism, for example chordin-like 2 was thought to be unable to bind chordin when first characterized30 but it has since been shown that through its vWC domains it does8. The most significant of these vWC-vWC interactions are between chordin and crossveinless-2. Crossveinless-2 binds to and internalizes both BMP (through a clathrin dependant mechanism involving noggin) and chordin but not the chordin BMP- ternary complex31. Crossveinless-2 holds the ternary complex at the cell surface where BMPs can be freed by proteinase cleavage of chordin. This interaction forms the basis of a pathway in which chordin targets BMPs to specific tissues which is important in development. The vWC domains also have binding partners in addition to BMPs and each, other including twisted gastrulation and Ont1 as shown in Figure 1.3A.

1.2.2 Chordin Structure

Prior to this study, the structure of chordin had not been investigated, however the structure of vWC1 of the chordin related protein crossveinless-2 has been determined in complex with BMP-232, 33 . The vWC domains of crossveinless-2 are very similar in sequence to those of chordin34. The domains are characterized by ten cysteine residues spaced in a conserved pattern with a tryptophan residue (the loss of which reduces anti-BMP function) between the first two cysteines25. These domains form a cysteine knot topology similar to noggin and BMPs22. It is thought that variation within the vWC domains determines strength of BMP binding, while the binding specificity is determined by the surrounding residues25.

The first vWC domain of crossveinless-2 has a non-globular rod-like structure of ~50x24Å with no hydrophobic core33. Figure 1.4 shows crossveinless-2 vWC1 divided into an N- terminal subdomain (SD1), C-terminal subdomain (SD2) and a short N-terminal clip domain. SD1, which consists of a three stranded anti-parallel β-sheet and the clip are responsible for BMP binding33. Five hydrogen bonds form between the clip and BMP-2. Little conservation is needed in the clip because the bonds are formed with peptide backbone. It is unlikely that chordin uses a clip in its vWC domains as expression of chordin vWC SD1 and SD2 subdomains with and without their flanking regions had equivalent BMP affinity33. The activity of SD1 has been shown to be consistent in chordin and chordin-like 235. SD2 of crossveinless vWC1 binds to chordin36, blocking chordin activity to promote BMP signalling37. SD1 is related to the type I domain of fibronectin but SD2 has no structural homologs35.

Figure 1.4: vWC domain structure. (A) Schematic diagram showing the secondary structure arrangement of vWC1 of crossveinless-233 (B) Ribbon representation of crossveinless-2 vWC1 showing subdomain structure positioning. β-sheets are shown in gold and the disulfide bonds of the disulfide bonds in cyan33. Figure generated using Jmol9. (C) Ribbon representation of crossveinless-2 vWC1 in complex with BMP-2 showing co-operative binding. The vWC1 domains are shown in orange and green, the BMP dimer subunits are shown in purple and blue. SD1 binds to the knuckle epitope while the clip hooks over the wrist33.

Although there are many members of the chordin family with similar BMP antagonist functions38 only chordin contains four CHRD domains between vWC1 and vWC3. The CHRD domains, are uncharacterized structurally and functionally. Within their sequence are four potential N-linked glycosylation sites with none elsewhere in chordin39. The CHRD domains are unique to chordin and some bacteria so the domain boundaries had to be estimated based on the exon-intron structure and similar domains in bacterial proteins with single CHRD domains40. These domains are defined by a conserved C-terminal GE[I/L]RGQP[V/I/L] motif. PSI-PRED predicts an all β-sheet structure and this is supported by limited homology of the CHRD domains to the β-barrel immunoglobin-like folds of the Cu- Zn superoxide dismutase family. In some CHRD domain containing bacterial proteins the copper co-ordinating histidines of the Cu-Zn superoxide dismutase family are present but in chordin they are not, so functional conservation is unlikely. In the SOG domains of Drosophila sog, the N-terminal and C-terminal sequences within each domain are circularly permutated. The result is that, since each domain is adjacent to the next, SOG motifs can be found in chordin and CHRD domains can be found in sog40. Truncated forms of sog appear to have higher biological activity the more of the SOG domains remain intact and vWC1 of sog can antagonize both heterodimers and homodimers of BMP (unlike full length sog which is only thought to antagonize BMP heterodimers). Taken together these findings indicate that the SOG domains may have roles in both self-inhibitory activity and target specificity41.

Chordin has six splice variants which are tissue specifically expressed39. Variants 1 and 2, found only in the liver, are highly truncated with only part of vWC1 present and a short N- terminal tail. Variant 3 found in the ovaries has vWC1, CHRD1 and part of CHRD2. Variant 4 is truncated after SD1 of vWC3 while variants 5 and 6 have mid sequence deletions of 40 and 88 amino acid residues respectively (the exact boundaries of the variants can be found in Appendix III). These variants are found in a broad range of adult tissues along with full length chordin, particularly in the brain, liver and ovaries39. Constructs similar to these variants were found to be significantly less effective in vivo39. Although the role of the splice variants have not been determined specifically for chordin, chordin-like 2 has ten known alternative splicing patterns which are predicted to have a role in cell differentiation30.

1.2.3 Structure of the Antagonist Complex

Figure 1.4C shows BMP-2 bound to vWC1 of crossveinless-2. Seventy percent of this interaction is hydrophobic and there are no polar bonds. This may explain why only the cysteine and typtophan residues are highly conserved between the vWC domains of the chordin family, and the rest of the sequence is variable33. The N-terminal subdomain, SD1, binds to the knuckle epitope of BMP-2 (the same as chordin) and the clip folds over the wrist epitope. This is thought to strengthen the binding of these two small interfaces as well as blocking BMP binding to both types of receptor. Most of the non-polar binding occurs

31 between BMP-2 and SD1 with the clip reliant on hydrogen bonding33. As this clip is not shared by chordin and chordin-like vWC domains, it is possible that co-operative binding of multiple vWC domains in the full length protein compensates for the absence of the clip domain.

1.3 Twisted Gastrulation

Twisted gastrulation (Tsg) is a secreted protein which has a dual role in BMP regulation. It can bind to both BMPs and chordin strengthening the antagonistic complex26, 42. The Tsg- sog complex binds more strongly to BMPs than either component alone43. However Tsg also increases the rate of chordin cleavage by tolloids freeing BMP, and it promotes degradation of the remaining chordin fragments26. Whether the pro- or anti- BMP activity is dominant depends on physiological circumstances, particularly tolloid concentration. For example it is thought Tsg may act as a BMP agonist in skeletogenesis and a BMP antagonist in T-cell development44. In addition Tsg can bind to and modulate chordin-like 1 and chordin-like 2 and although it is disputed whether or not it binds directly to crossveinless-2 (crosslinked ternary complexes have been produced36 but no binding was detected by SPR8 so this may be an artefact) but there is some evidence that the BMP-crossveinless-2 interaction may be enhanced by Tsg8, 36. It has also been speculated that Tsg may function independently of chordin family proteins to keep BMP heterodimers in solution45.

Though prior to this study the structure of twisted gastrulation had not been investigated, it has previously been shown to be glycosylated with one heavier and one lighter glycosylated variant. This glycosylation has a significant impact on function, with deglycosylation of Tsg greatly reducing BMP binding and variation in glycosylation between species also affecting affinity46. Tsg is expressed ubiquitously in human adult tissues47. Humans have only one Tsg gene, a trait shared with other mammals and some insect species48. In flies and frogs, Tsg and Tsg3 are expressed more in early development with Tsg2 expressed later (for example during wing development in flies49 and the tadpole stages in Xenopus50). In all species the protein function is highly conserved; human and mouse Tsg can substitute in Drosophila for fly Tsg. Likewise Drosophila Tsg has the same effect in fish as zebrafish Tsg indicating functional equivalence42.

The structural mechanism of action of Tsg is not clear. It may act as a scaffold by binding to both chordin and BMP and bringing them closer together, induce conformational change in chordin to facilitate BMP binding or both. A conformational change in chordin induced by Tsg may explain why an extra chordin cleavage site is found in mice in the presence of Tsg51. It is also unknown why the presence of Tsg appears to increase the rate of chordin fragment removal but it may increase fragment interaction with crossveinless-2 resulting in

32 endocytotic degradation31 (which does occur with unbound chordin, but not the BMP-chordin complex).

Tsg can bind to the BMP-binding vWC domains of chordin and to its vWC containing cleavage fragments. According to one study Tsg binds to chordin and individual vWC domains with similar strength8, though there are conflicting data which suggests that Tsg can only bind to chordin if the C-terminal cleavage site is intact26. One study suggests that Tsg has ~60x higher affinity for chordin-like-2 than chordin8. Tsg also binds to BMPs -2, -7 and GDF5 with particularly high affinity for BMP-78 (early studies indicate that fly Tsg2 has different specificities52) and competes with individual vWC domains for BMP binding instead of forming a ternary complex. However Tsg cannot displace full length chordin from binding to BMPs42, 53. It is thought that Tsg interacts with the ternary complex as a dimer53. The larger natural cleavage fragments of chordin have been previously shown to form a ternary complex with BMP and Tsg26.

Tsg only binds to the wrist epitope of BMP, unlike chordin which binds to the knuckle epitope8. In theory this alone could antagonize BMP receptor binding, but studies have shown that Tsg can only compete with the BMP-BMPR interaction at a 20 fold molar excess so it is highly unlikely to antagonize BMP-BMPR interaction in vivo8. Figure 1.5A shows a domain schematic diagram for Tsg (the full sequence is shown in Appendix III) which consists of two domains which some papers have suggested may be connected by a variable hinge region49. The BMP binding activity of Tsg is found in the cysteine-rich N- terminal domain which is similar to the BMP-binding vWC domains of chordin. The isolated N-terminal domain can interact with BMPs but not with chordin. The C-terminal alone cannot interact with chordin either, indicating that both are required53, 54. In chordin the SD1 subdomains of vWC domains are the binding site for Tsg35. Figure 1.5B shows possible arrangements for the BMP-chordin-Tsg ternary complex8.

Figure 1.5: Twisted gastrulation schematic diagrams. (A) Schematic diagram of the domain structure of twisted gastrulation. (B) Possible modes of Tsg interaction with the BMP-2- chordin complex; (i) 1:1 Tsg monomer:chordin, (ii) 2:1 Tsg monomer :chordin. (ii) Tsg binding to chordin as a dimer and (iv) Tsg binding to different vWC domains to BMP-2 and/or only binding to one component of the BMP-2-chordin complex.

1.4 Tolloid Metalloproteinases

The tolloids are a family of zinc-dependant metalloproteases which process a range of substrates. Like chordin, tolloids are highly functionally and structurally conserved being found from humans to animals as simple as hydra55. Tolloids have an N-terminal astacin-like protease domain followed by calcium binding epidermal growth factor (EGF) domains and C1r/C1s-Uegf-BMP-1 (CUB) domains. Tolloids cleave at specific recognition sites characterized by an aspartic acid residue in the P1’ position56. Tolloids are secreted in an inactive form with a prodomain which is removed by furin to activate tolloid57.There are three tolloid genes in humans; mammalian tolloid-like 1 (mTLL1), mammalian tolloid-like 2 (mTLL2)58 and mammalian tolloid which also produces shorter splice variants BMP1 (a misnomer resulting from initial copurification of the enzyme with BMPs) and BMP1His (which has a sequence rich in histidine residues). Their structure is highly conserved and important to function. The non-catalytic domains appear to have an essential role in enzyme activity59 and dimerization excluding substrates appears to control specificity in mammalian tolloid60.

Chordin is cleaved by tolloids at two specific sites and a summary of the resultant fragments is shown in Figure Fragment 1.6A. In mice in the presence of twisted gastrulation, tolloids

34 also cleave chordin at a third site to yield additional fragments but this is not thought to be conserved in humans51. The cleavage fragments of chordin retain biological activity, which is much weaker, possibly due to the loss of co-operative binding51. In chordin-like 1 and zebrafish chordin-like, tolloid cleavage occurs between vWC2 and vWC3. Mutating the conserved aspartic acid residue in the 1P’ position to asparagine and tyrosine to phenylalanine generates cleavage resistant, biologically active chordin in fish61.

Figure 1.6: Chordin cleavage sites. (A) Schematic diagram of chordin showing the position of tolloid cleavage sites in the presence and absence of twisted gastrulation (red arrows). (B) Schematic diagram of the fragments produced through tolloid cleavage of chordin. (C) Recognition sites of tolloid in chordin across species, sog and two non-chordin tolloid targets for comparison. The ubiquitous aspartic acid residue is shown in red and other highly conserved residues in the cleavage site are shown in green 43, 58, 61.

Sog has similarly positioned cleavage sites but unlike chordin cannot be cleaved unless already bound to BMP. This is at least partly a property of the cleavage site of sog, as if the site is replaced with a chordin cleavage site, free sog becomes susceptible to tolloid cleavage, albeit weakly43. The sog cleavage site features the conserved aspartic acid residue in the P1’ position but has little other similarity to that of chordin. As with zebrafish, mutation of this residue greatly reduced cleavage by tolloids. Significantly, the cleavage resistant sog did not show discernible differences in binding to BMPs and twisted gastrulation43.

Interestingly, carboxy-truncated sog constructs retaining vWC1 and varying amounts of intact SOG domain region, have the capacity to block dpp/gbb (decapentaplegic and glass bottomed boat- Drosophila equivalent of BMPs) activity more effectively than full length sog, and have been dubbed supersog accordingly. When the SOG domains of the constructs were intact, their effectiveness as BMP antagonists was even higher, raising the possibility that the ∆C fragment of chordin may also be a more efficient BMP antagonist than full length chordin41. The supersog constructs do not match known cleavage fragments so it is not certain whether sog is processed to supersog-like fragments in vivo, or if splice variants produce supersogs, though Western blots of embryo extracts show sog fragments of many lengths so it is possible41.

Other known tolloid functions involve cleaving proteins to their mature forms and releasing growth factors from latent complexes. Tolloids proteases are able to cleave collagen propeptides to their mature form cleaving the C-terminal propeptides of major fibrillar collagens I, II and III and the N-terminal propeptides of the minor fibrillar collagens V and XI as well as non-fibrillar collagen VII55. Tolloid also cleaves the prodomains to activate apolipoprotein A1 (the major protein of high density lipoprotein)62, the crosslinking enzyme lysyl oxidase, the anchoring protein laminin-555, dentin matrix protein-163 and the small leucine rich proteoglycans osteoglycan55, biglycan and decorin64. Tolloids release growth and differentiation factors (GDFs -8 and -11) and TGF-β from their latent complexes65. They also cleave prolactin growth hormone and prolactin placental lactogen to release a 17kDa fragment with angiogenesis suppression effects66. In general BMP-1 cleaves proteins in positions between domains, rather than mid-domain cuts65.

1.5 Other Modulators of the BMP-Chordin-Tsg Ternary Complex

1.5.1 BAMBI

BMP and activin membrane bound inhibitor (BAMBI) is a BMPR-I-like decoy receptor that lacks a functional kinase domain67. It antagonizes BMP signalling by binding to BMPs and BMPR-II to form an inactive complex. BAMBI has very rapid turnover by auto lysosomal degradation and may trap-sink BMPs68. However although it behaves as a pseuodoreceptor

36 for BMPs, it is thought to activate Smad signalling in response to other members of the TGF- β superfamily69, 70. Knockout of BAMBI in mice has no apparent effect on embryonic development, postnatal survival71 or general behaviour72. However the mice show an exaggerated response to pain indicating that BAMBI may be involved in pain transmission or perception through modulating BMP activity72.

1.5.2 Ont1

First identified in chick embryos73, Ont1 behaves as an adaptor protein bringing chordin and tolloid together to increase the rate of chordin cleavage but does not appear to interact with Tsg or BMPs directly. Ont1 has a biphasic dose response, increasing BMP activity by promoting chordin degradation but at higher concentrations protecting chordin from cleavage by isolating chordin and tolloids on separate scaffolds. Whether it is ever sufficiently concentrated in vivo to inhibit tolloid cleavage of chordin is uncertain74.

1.5.3 Sizzled

Secreted frizzled (sizzled) and its homologue sizzled-2 are relatives of the Wnt receptor frizzled75. They stabilise chordin (which they do not bind to directly76) by acting as competitive inhibitors of tolloids in zebrafish and Xenopus76, 77. Misexpression of sizzled does not appear to affect BMP signalling in chordin knockout animals indicating that all of its BMP antagonistic activity is chordin dependant75. A cysteine rich domain, sometimes called a frizzled domain75 is required for BMP-1 inhibition77. In mammals this effect is not conserved, with sizzled having no effect on chordin and enhancing tolloid pro-collagenase activity78. Crescent is a close relative of sizzled and can partially compensate for its loss in Xenopus79 and also acts as a competitive inhibitor of tolloids77, 79.

1.5.4 Integrins

Chordin and integrin α 3 are coexpressed during development. Chordin binds to integrin α-3 on the cell surface and is internalized as a result. It is not yet known if this is in order to degrade chordin and promote BMP signalling or to facilitate transcytosis80.

1.5.5 Proteoglycans

Chordin binds to heparin and cell surface heparan sulphate proteoglycans. This binding limits chordin diffusion and may mediate its cellular uptake or modulate BMP antagonism by

37 chordin81. The tolloid cleavage fragments of chordin can also bind in this way. It is also thought that heparin sulfate may also trap BMPs either directly or through chordin. Biglycan is a small leucine rich proteoglycan which can, like Tsg, bind to BMPs -2 and -4 and chordin, strengthening the antagonistic complex82.

1.5.6 Fibronectin

Fibronectin binds to BMP-1 with physiologically relevant strength and enhances BMP-1 activity against several of its substrates including chordin. There are several pieces of evidence that fibronectin, which is highly abundant in the extracellular matrix may act as a general scaffold to facilitate BMP-1-chordin interaction. Fibronectin and BMP-1 colocalize in vitro, chordin processing is reduced in fibronectin negative cells despite levels of the protease remaining constant and fibronectin binds to both chordin and BMP-1 in vitro83.

1.6 Development

1.6.1 Chordin, Tsg and BMPs in Embryo Patterning

In early development bilaterian animals develop along the anterior-posterior, left-right and dorsal-ventral axes which determine basic body layout and the location of structures, shown in Figure 1.7A. Establishing asymmetry is controlled by gradients of morphogens such as nodal, BMPs and Wnt family members. One of the first events in establishing asymmetry is the nuclear localization of β-catenin on the dorsal side of the embryo. This induces the expression of BMP antagonists, particularly chordin and noggin from the dorsal centre leading to a gradient of high ventral BMP and low dorsal BMP, as shown in the diagram in Figure 1.7B84. This initial establishment of the gradient is assisted in some species by maternal deposition of chordin at the dorsal pole85. Experiments showing the effect of targeted knockdown of chordin in the cells which form the Kupffer’s vesicle in fish and chordin knockout in chicks show that chordin is also required for correct left-right axis patterning, probably independent of its effect on the dorsal-ventral axis86, 87. The relative importance of chordin and noggin in gradient initiation varies between species88 and they are partially redundant, yet despite their functional similarity they are structurally very different. Noggin is much smaller, a 64kDa homodimeric glycoprotein with a two-fold axis of symmetry resembling BMPs. It has an acidic N-terminal and a cysteine-rich C-terminal with a very loose similarity to the vWC domains of chordin89.

Figure 1.7: The BMP morphogen gradient. (A) Diagram showing the direction of body axes patterning. (B) Diagram of BMP gradient produced by BMP antagonism in the early embryo and structures produced as a result. (C) Schematic diagram showing balance of the gradient by the dorsal and ventral synexpression groups 23, 79, 84, 90.

Figure 1.7C shows how the gradient is balanced. The dorsal centre secretes chordin, noggin, crescent and Ont1, while the ventral centre (which is sometimes called the tail organizer)91 secretes crossveinless-2, sizzled and BAMBI. Both ends secrete BMPs, tolloids and Tsg. This means that both ends secrete proteins with similar biochemical activities in synexpression groups84, 92 which enables them to compensate for each other93. For example, if BMP levels fall ADMP transcription is increased from the dorsal centre so Smad signalling is restored. When BMP levels from the dorsal centre rise, sizzled transcription increases from the ventral centre which inhibits tolloid-mediated chordin cleavage thereby

39 indirectly antagonizing BMP. This feature of the BMP gradient has been described as a “molecular seesaw” where the independent dorsal and ventral organizers are oppositely regulated allowing the BMP gradient to self-regulate94. This makes the gradient extremely robust, to the extent that in Xenopus from a dorsal half-embryo BMP can re-establish the dorsal-ventral axis and form near normal patterning95.

Figure 1.8 shows the dorsalizing effect of heterozygotic BMP-2 mutation in zebrafish. As long as BMP-2, -4, -7 or ADMP continue to signal, some degree of dorsal-ventral patterning is retained. However when all four are knocked out, simultaneously the axis is completely lost and the whole embryo consists of central nervous system with a little spinal cord tissue94. When chordin is knocked down, mutants are severely ventralized as shown in Figure 1.896. Raised Tsg levels in development usually promote BMP signalling through increasing chordin cleavage and are therefore ventralizing (as shown in zebrafish in Figure 1.8)97. However, owing to its ability to enhance chordin anti-BMP effects as well as increase tolloid chordinase activity, Tsg has the opposite effect when tolloid levels are lower leading to dorsalization in species such as Xenopus98.

Figure 1.8: Effects of BMP gradient disruption in body patterning in zebrafish development at 24h. (A) Wild type and BMP-2b+/- embryos also showing that BMP haploinsufficiency leads to dorsalization97. (B) Wild type and ventralized chordin morpholino embryos96. (C) Wild type and ventralized Tsg embryos97. Images in this figure have been altered to remove original background and figure labels.

Individual chordin vWC domains can induce dorsalization in Xenopus when introduced alone showing the same biological activity as chordin only significantly weaker27. In one experiment a 32-fold higher molar concentration of vWC1 mRNA was needed to induce a comparable phenotype53. The vWC1 phenotype can be partially rescued by coinjection with Tsg (but not by coinjection with xolloid RNA indicating that it is the fragment and not endogenous full length chordin being antagonized), showing that Tsg may have a role in fragment cleanup beyond increasing the chordinase activity of tolloids53.

Chordin-like 199 and chordin-like 2100 have different expression patterns to chordin, being found primarily in the neural-plate and developing cartilage respectively. However when the single chordin-like gene is knocked down in zebrafish alongside chordin, the extent of ventralization exceeds that of either knock down alone indicating partial functional redundancy101. Other modulators of the pathway have been shown to alter the BMP gradient in a chordin dependent way. For example, without Ont1 to promote cleavage, chordin localization is far less precise in developing embryos73, 90 while introducing biglycan promotes antagonistic complex formation leading to dorsalization82.

1.6.2 Chordin Shuttling

Beyond simply sequestering BMP in the tissue in which it is expressed, chordin facilitates “shuttling” of BMPs to other tissues. A chordin-dependant flow of tagged BMP-4 from the dorsal to the ventral centre has been shown in Xenopus95. The result is a localized build-up of inactive BMP. Tolloid mediated cleavage may then allow spatially and temporally controlled mass liberation93. This idea partially stems from an in situ hybridization study in which an accumulation on BMP expression occurs where chordin expression is highest102. This transportation complex is localized partly by crossveinless-2. Figure 1.9 shows a proposed mechanism, in which crossveinless-2 binds to chordin and chordin cleavage fragments when bound to BMPs103, 104.

When BMPs bind to crossveinless-2 in complex with antagonists like noggin and gremlin this induces endocytosis. However, this does not happen when BMPs are bound to chordin. Instead the antagonistic complex is sequestered at the cell surface, where tolloid cleavage of chordin can later release BMPs31. The mechanism is potentially highly efficient because crossveinless-2 binds preferentially to free chordin fragments which can then be internalized36. This would mean that following release of BMPs by tolloid cleavage, the residual chordin fragment antagonist activity would be removed rapidly. Recent studies have shown that in Drosophila the Dpp/Scw-Sog shuttling complex (the Drosophila equivalent of BMP-chordin complex) assembles on collagen IV and Tsg binding disrupts the interaction with collagen IV releasing the complex2.

Figure 1.9: Schematic diagram showing interaction between the BMP-chordin-Tsg complex and crossveinless-2. The complex attaches to membrane-anchored crossveinless-2 but is not internalized. Tolloid cleavage of chordin relieves BMP antagonism allowing it to bind to BMPRs31, 36.

1.6.3 Differences in Insect Models

The role of the chordin family in development has also been mapped in insect models, primarily in Drosophila. It is thought that the dorsal-ventral axis was inverted over the course of evolution so that in insects the direction of the BMP gradient is opposite to that of mammals53. In both flies and beetles, sog mutant embryos are severely dorsalized while in vertebrates chordin loss is ventralizing48. Other components (such as Tsg in Drosophila) have been shown to have corresponding inverted expression between the animal poles53. Another key difference is that sog is only susceptible to tolloid cleavage when bound to BMPs and this has a role in stabilising the gradient. Sog modified to be cleaved in the absence of BMP like chordin, was found to change the steeper BMP gradient found in flies to a more shallow gradient as found in vertebrates. This led to differences in tissue size and cell fate and more variable patterning which may be required in more complex organisms but is disadvantageous in Drosophila43.

A hypothesis explaining differences in the effect of sog on different BMPs has been proposed suggesting that dpp and scw homodimers bind less strongly in the BMP-sog-Tsg ternary complex and are less antagonized, however they also produce a lower signal

42 response from their receptors. The heterodimers induce a stronger response but also bind more strongly to the antagonist complex, and the net flux of the antagonist complex toward the dorsal midline leads to a higher proportion of heterodimers shuttled to the dorsal region. Different BMP gene activation at high and low signalling levels then leads to specification105.

As with vertebrates, there is significant variation in the activity of specific pathway components despite the pattern of gradient formation being largely conserved. For example, Drosophila have several paralogues of the tolloid gene while simple single-tolloid systems are found in other insect species48. The role of Tsg appears to differ substantially between insects. Flies have both Tsg and Tsg2 whereas in bees, mosquitos52 and Tribolium only one Tsg-like protein has been found. In Tribolium, Tsg is essential for correct early BMP signalling and is apparently sog-independent, but is developmentally dispensable in Drosophila48. This might be due to the latter having multiple Tsg genes, as although both Tsg and Tsg3 were simultaneously knocked out, Tsg2 remained. The justification for this was that Tsg2 is not thought to be expressed until very late in embryogenesis48. This late expression of Tsg2 in the developing wing may mean that, in contrast to many other systems, it is not involved in dorsal-ventral patterning. However, when overexpressed with sog, BMPs are generally more strongly antagonized than with sog alone52 so it is possible that in the absence of Tsg, Tsg2 is expressed at higher levels to compensate. Unexpectedly, the role of Tsg appears to vary with BMP type. Neither dpp nor gbb overexpression can be rescued by sog alone, but dpp can be rescued by sog and Tsg together but not by Tsg alone, while gbb can be rescued by Tsg alone but this is reduced to a partial rescue if sog is coexpressed52.

1.6.4 Differences in Mammals

Chordin appears to be less essential in mammalian development, though it has only been studied in-depth in mouse models and the requirement for chordin is highly dependent on genetic background106. Early experiments showed inviability (predicted from similar experiments in other species) but with surprisingly normal early development107. These experiments showed a range of defects in organ development, vasculature108, oedema, limb deformity109, and pharyngeal defects107 but with normal axis formation and body plan. In addition the apparent ventralizing effect of chordin knockout on the dorsal mesoderm also did not reach full penetrance107, 108. One thing all these early experiments had in common was that they were performed in a specific inbred strain, and chordin knockout in outbred strains produced fully viable phenotypes88, 106. These mice showed no developmental abnormalities, shown in Figure 1.10, except for mild defects of the chondrocrianium and cervical vertebra88. This discrepancy in phenotype was later revealed to be the result of a chance mutation of Tbx1 in the inbred strain for which chordin turned out to be a modifier106.

Figure 1.10: Chordin knockout in mice. Mouse embryos with a gestational age of 10- 12.5dpc showing (A) Wild type, (B) Chordin knockout109, (C) noggin knockout110 and (D) Chordin;noggin double knockout109. The chordin knockout embryos present a near-normal phenotype probably due to functional redundancy with noggin. Co-knockout with noggin produces a more severe phenotype than noggin knockout alone with loss of eye, forebrain and facial structures109. Images in this figure have been altered to remove original figure labels.

It is likely that the reduced role of chordin in patterning in mice is due to functional redundancy. Though noggin null mice are always unviable, heterozygous mutants are relatively normal, but if chordin is also absent patterning of the noggin+/- mice is disrupted though defects are limited to the head region88, 111. Chordin and noggin double mutant mice show a phenotype even more severe than noggin knockout alone, shown in Figure 1.10. The pups are resorbed by the mothers relatively early in pregnancy showing holoprosencephaly at the mildest and aprosencephaly (a total absence of forebrain) at the

44 most severe109. This indicates that the function of chordin is conserved, but is no longer essential in normal mice.

Little is known about the role of chordin in later development. Chordin is highly expressed in almost all major mouse organs including the heart, lung, brain and gut, suggesting that chordin plays a role in organogenesis112. There is some evidence that its role in shuttling BMPs to crossveinless-2 may be conserved113, 114. In particular it is expressed at high levels in condensing and differentiating cartilage112. However, the only striking deficiencies in outbred chordin null mice are neurological. In the post-natal mouse brain chordin is highly expressed in the cerebellum and hippocampus112 (an observation also made in humans115 indicating a role in synaptic plasticity). There is both physical evidence for this, chordin mutant axon terminals having significantly more docked vesicles and chordin null mice demonstrating faster learning and enhanced synaptic plasticity (associated with short term memory). It is thought that the increased rate of learning is due to increased neurotransmitter release116.

It is important to note that although chordin appears to be dispensable in early mouse embryogenesis this would need to be shown in multiple species before we could say that it is dispensable in mammals and ultimately in humans. One striking piece of evidence that it might not be dispensable is that noggin (which appears to have taken over the role of chordin in mice) is sometimes mutated in humans who, despite being born with severe joint abnormalities, have correct dorsal-ventral axis patterning117. In addition adaptive redundancy has not been ruled out; it is possible that in the absence of chordin expression of other anti- BMP factors from the dorsal and ventral synexpression groups are increased in mice to compensate. It is also possible that because noggin binds to some BMPs which chordin cannot89, the respective roles of chordin and noggin may have changed in mice due to the evolving roles of BMPs themselves. Alternatively the reason may be functional redundancy between chordin and antagonists other than noggin such as the chordin-like proteins.

Interestingly, twisted gastrulation produces a much clearer phenotype in mice than chordin deficiency. Tsg is expressed from gastrulation onwards in mice23. Perinatal mortality is seen in more than half of Tsg knockout mice, and though a limited number survive to adulthood and breed, major developmental abnormalities are observed118. Figure 1.11A shows a Tsg - /- mouse with a control littermate displaying the most prevalent defects in survivors. These mice are in smaller size119, 120 and have vertebral abnormalities most noticeable in the tail44. The lymphatic system is underdeveloped with organ size reduced120. They are also prone to osteoporosis120, lymphopenia44 and holoprosencephaly23. These mutants are also prone to severe hydrocephaly shown in Figure 1.11B with enlarged brains, gait imbalance and closed eyes118.

Figure 1.11: Twisted gastrulation knockout in mice. (A) Wild type and twisted gastrulation knockout mice showing reduced size and kinked tail due to skeletal defects120. (B) Hydrocephaly in Tsg knockout mice showing larger head and brain size in the knockout mouse compared to the wild type118.

1.6.5 Chordin in the Development of Other Species

The effect of the chordin-Tsg-BMP complex on development has been widely studied in a range of other models. It has been shown to be involved in establishing the dorsal-ventral axis in hemichordates121, C. Intestinalis92, arachnids122, cephalochordates123, and basal chordates124. In emu and chickens chordin was found to function similarly to other vertebrates57, 125 but in leeches a different subset of BMPs to the ones antagonized by chordin are the main dorsal-ventral patterners126. The pathway has also been investigated in radiata, where BMP and nodal gradients are required for oral-aboral specification, BMP being pro-aboral127. Chordin is not required for this (or even for shuttling as BMP is highly diffusible without it128) but is needed for neural development and patterning of the oral-aboral boundary region127. Interestingly in sea anemones BMP and chordin are expressed on the same side of the embryo indicating that the function of chordin may have changed over the course of evolution129. These variations show that despite conservation in the pathway it has adapted to suit the requirements of different modes of embryogenesis.

1.7 Mathematical Modelling of the Complex in Development

1.7.1 Modelling Strategies

Mathematical models have been used to test mechanisms of BMP regulation from expression and secretion to transport through the extracellular space and interaction with BMPRs130. Uses of the models in this field have been wide ranging and have included predicting a mechanism for the biphasic dose responses of crossveinless-228, calculating that heterodimerization of BMPs is selectively favourable because it protects the robustness of the gradient105 and comparing BMP in different systems (e.g. the wing) through the models developed for them131. Of particular relevance to this study, modelling has predicted that without BMP antagonists the level of BMP-BMPR interaction would be constantly very high, suggesting that antagonists such as chordin may be needed not only for antagonism but also for keeping BMP-BMPR interaction within physiologically useful kinetic parameters128.

Modelling the gradient is difficult as is both the issue of crosstalk and the large number of components involved in the pathway, the action of many of which is context dependant. The morphology of the embryos themselves also pose a major difficulty in modelling as they affect the distances chordin would need to diffuse in order to shuttle BMPs. Reaction diffusion equations in general are highly dependent on scale132 with the core of morphogen patterning models being based on the L/λ ratio where L represents the size of the system (λ represents the diffusivity and degradation rate constant)133, 134. However, robustness and scale invariance are important in models, and a frequent experimental observation is that embryo development is remarkably resistant to scale changes105.

1.7.2 Computer Models of Ternary Complex Function

The first model able to account embryo scaling in the chordin-BMP complex was developed by Ben-Zvi for the Xenopus embryo. Xenopus dorsal half-embryos had been shown to retain the capacity to generate a well proportioned embryo which previous models were unable to explain. The composition of this nine parameter model included ADMP, a single generic BMP, chordin and the Xenopus tolloid, Xlr, and allowed to vary over two orders of magnitude. Even with so few pathway components involved the model generated 26,000 solutions of which 1100 had correct dorsal-ventral polarity but only 21 were able to scale a dorsal half-embryo95. This model strongly supported a shuttling role for chordin because networks which failed to scale were the ones where the gradient of chordin remained uniform but in the ones that succeeded the ligand accumulated at the ventral pole95. Interestingly it predicts that chordin is degraded faster when coupled to BMP than it does alone although this has only been shown experimentally for Drosophila sog95.

Zebrafish has been the other major vertebrate system for modelling the role of chordin in dorsal-ventral patterning. Zhang et al developed a model based on a 3D dome intended to reproduce the shape of an embryo at 30% epiboly which is the stage approximately between the end of blastula formation and the beginning of gastrulation. This more realistic model even used reflective boundary conditions to differentiate between the outer surface and the animal-yolk interface, factoring in limited diffusion of pathway components into the yolk. This study focussed on the role of initial asymmetries (maternal deposition) in creating synergistic feedback loops of gene expression, which the authors argue are key in establishing the gradient. The basic model indicates a mechanism consistent with experimental observation that without maternal chordin mRNA deposition the gradient is almost able to recover because feedback loops become the main driving force in establishing the gradient once the influence of maternal deposition wanes85.

One model, which supports the shuttling mechanism by suggesting that Dpp must be widely diffusible in the presence of Sog but tightly localised in its absence, also raised some intriguing possibilities about the function of chordin family members. The model, developed by Eldar et al, was a nine parameter model similar to the Ben-Zvi Xenopus model, but allowed for more possibilities as each parameter was allowed to vary over four orders of magnitude instead of two. The gradient still stabilised if unbound sog was not cleaved and, more interestingly, a requirement for robustness is that sog can bind to and capture BMPs even when they are already associated with receptors. One particular requirement for robustness is being able to store excess signal molecules in a restricted domain135.

A later Drosophila model by Mizutani et al included receptor mediated endocytosis as a parameter and concluded that BMP is able to diffuse significant distances without chordin85. The most recent study looking at the patterning differences between sog and chordin adapted an existing model105, 131 to test what the outcome might be if sog, like chordin, were cleavable without bound BMPs43. The result was an overall reduction in BMP signalling due to increased sog degradation and a reduction in localised BMP accumulation due to loss of BMP shuttling by sog. The effect was so pronounced that even increasing the levels of sog could not compensate. The group proposes that this difference between sog and chordin allowed for a more robust BMP morphogen gradient in insects and a more variable gradient in vertebrates.

1.8 Medical Implications

Owing to the powerful influence of BMP family members on such a broad range of signalling pathways and developmental processes, it is unsurprising that their regulation is of significant research interest in connection to a diverse range of pathologies. Recombinant human (rh) BMP-2 and -7 can be produced in E.coli in a biologically active form which is

48 approved for limited clinical use136. There is a substantial body of evidence that the sphere of influence of this pathway incorporates congenital disorders, cancer, vascular disease, arthritis and bone disorders. The BMP pathway has also been implicated in other conditions, for example crossveinless-2 in kidney disease137, BAMBI in chronic obstructive pulmonary disorder138, BMP-2 in renal interstitial fibrosis18 and Tsg-null mice have been suggested as a model for studying some mechanical aspects of human dwarfism44. BMP-2 can induce transcription of follicle stimulating hormone β-subunit which could have implications for hair loss139, while BMP-4 has been shown to be an adipogenic factor and elevated in obesity (though uncertain whether this is a cause or effect of excessive weight gain)19. The pathway was initially linked to Cornelia de Lange syndrome but other causes were later discovered.

It is likely that an improved understanding of BMP-chordin interaction may prove clinically significant, for example in developing selective antagonists, predicting mutation consequences or developing antagonist-resistant recombinant BMPs which are more effective in clinical use (noggin resistant BMPs are available already)89. It has also been suggested that chordin could be used therapeutically as a way of generating neuronal cells for transplantation because of its neuronal specifying properties140, and understanding its mechanism of action may help with this. Despite the fact that chordin appears to have a reduced developmental role in mice, chordin and twisted gastrulation are ubiquitously expressed in mouse adult tissues47 and in humans chordin expression is particularly high in the liver, cerebellum and female reproductive tract39. This section will discuss the best documented pathological links to the pathway; congenital disorders, cancer, vascular disease, arthritis and bone disorders in addition to BMP delivery methods, safety and industrial applications.

1.8.1 Congenital Disorders

The BMP signalling pathway has a role in disrupting a number of developmental processes leading to congenital disorders. Loss of BMP-7 coupled with half or complete loss of Tsg causes sirenomelia in mouse models141. Fusion of the legs occurs in 1 in 100,000 live human births and is caused by incorrect formation of the ventral mesoderm141. Chordin-like 1 mutation causes inherited X-linked megalocornea which is identifiable in the eye at birth and later causes degeneration of the cornea and presenile cataracts and can lead to loss of vision142. This is thought to be due to localized dysregulation of BMP signalling. Interestingly, memory tests performed by some of these patients indicate that the enhanced short-term plasticity and spaciotemporal processing previously observed in chordin knockout mice116 may be an effect conserved in humans lacking chordin-like 1.

Chordin null mice with the tbx1 mutation are susceptible to neural crest defects causing disorders of the heart and neck similar to DiGeorge Syndrome108, and Velo-Cardio-Facial

Syndrome patients. In particular a single vessel heart outflow instead of the usual two is characteristic in these disorders107. This is of special interest because in human patients with Tbx1 mutation, some have particular symptoms which are lacking in others, for which a second mutation in chordin may be responsible106. However, in mice chordin is close to the hotspot region of DNA for DiGeorge syndrome related mutations, called the critical region. In contrast, in humans it is on a different chromosome meaning that it may not be connected in human patients106.

The most serious of the congenital disorders linked to this pathway is holoprosencephaly in which the brain fails to develop into two hemispheres. Tsg-/- mice sporadically display this phenotype, which is far more common when one copy of BMP-4 is also lacking120. This is possibly due to Tsg activity in the choroid plexus part of the brain where BMPs are the dominant branch of the TGF-β superfamily118. Although the mechanism is not known, if may be linked to a BMP effect on p53 activity as the craniofacial defects in mice can be partially rescued by reduced p53 expression143. Tsg1 is present in the developing foetal brain118 and one possible mechanism of action is that it effects nodal signalling144. In human prosencephaly, nodal mutation is common and promoted by BMP antagonism by noggin and chordin. Tsg has been linked directly to human holoprosencephaly at locus 4145 however large scale screening of patients for alterations to the Tsg gene found only minimal evidence of a link, suggesting that the Tsg mutation as a direct cause or modifying factor for the condition in humans is rare146.

Chordin itself does not match any loci for holoprosencephaly but may signal upstream of other loci or represent a novel locus145. Chordin is not found in the choroid plexus but chordin-like 1 and brorin (brain-specific chordin-like) are, indicating a similar mechanism of action118. However, despite chordin not being in this region of the brain, mouse chordin-/- /noggin-/- mice produce similar phenotypes to human holoprosencephaly (though neither knockout produces this phenotype alone)88, 145.

1.8.2 Cancer

There is a wealth of evidence connecting BMP signalling to cancer. However the precise nature of the relationship is unclear and may be context dependent. BMPs have been shown to be involved in many different types of cancer including lung147, 148, colon149 and cerebellum150 cancer. They also have been linked to metastasis (both migration151 and reattachment152) and angiogensis153 in cancer progression making the pathway a promising target for anti-cancer therapy. Chordin and noggin treatment of hepatocellular carcinoma cells also appears to suppress their migration and invasiveness154. In addition twisted gastrulation has been shown to be overexpressed along with BMP-4 in some cancer cell lines155.

From the early stages of many types of cancer, BMPs appear to be connected to cell proliferation and tumour growth. In medulloblastoma, a common malignant brain tumour in children, BMP-2 signalling is thought to antagonise tumour proliferation by opposing Sonic hedgehog-dependent proliferation and by causing neurons to terminally differentiate so that they can no longer divide. Most of the major components of the BMP signalling pathway are reduced in tumour cells compared to their healthy counterparts150. Statistical evidence shows BMP also appears to be a negative regulator in cancers of the gut149. Mutations in BMP-2 and BMP-4 and their receptors are associated with increased risk of colon and rectal cancer, while loss of BMP is common in sporadic colon cancers and half of juvenile polyposis patients carry mutations in the pathway149. BMPs have been shown to affect cell migration in the vascular development of some species via the transcription factor GATAa92 and both Tsg and BMP-4 deficient mice have defects of the major blood vessels120.

1.8.3 Vascular Disease

The BMP signalling pathway is heavily involved in vascular health. A hormone used for anaemia treatment called erythropoietin functions via a BMP dependant mechanism. Some patients become resistant to erythropoietin treatment and it has been suggested that the BMP pathway could be manipulated as a substitute20. In humans, excessive BMP signalling can lead to anaemia while deficient signalling caused by mutations in pathway components, particularly receptors, is a known cause of hereditary hemorrhagic telangiectasia and pulmonary arterial hypertension11, 16. The mechanism for this is not always clear, however BMPs can induce apoptosis in pulmonary artery smooth muscle cells by activating caspases, releasing cytochrome C and repressing the anti-apoptotic protein Bcl-2. Twisted gastrulation loss has also been shown to reduce apoptosis in some cell types which appears to support the role for BMP in this pathway as Tsg effects in vivo are predominantly pro- BMP15. This mechanism requires the carboxy tail of BMPR II; if this is truncated or mutated the cells become resistant to apoptosis, and such alterations have been identified in familial idiopathix pulmonary arterial hypertension patients16.

BMPs are particularly involved in the calcification aspect of cardiovascular disease, a process which may be regulated through similar mechanisms to calcification in new bone formation. Vascular calcification is a leading cause of death in patients with chronic renal failure. BMP-2 and-7 can reduce proliferation of pulmonary artery smooth muscle cells, which may reduce calcification as these can transdifferentiate into osteoblast-like cells16, 156. BMP-2 is overexpressed and chordin downregulated in these dedifferentiated cells, and BMP-2 overexpressing mouse models are prone to calcification. BMPs can promote osteoblast differentiation and it is thought that BMP-2 increases calcification by promoting differentiation to an osteoblast-like state157. A summary of the predicted pathways is shown in Figure 1.12.

Figure 1.12: Schematic diagram showing the presumed mechanisms of BMP involvement in atherosclerotic calcification. Dedifferentiated vascular smooth muscle cells (VSMCs) migrate into the atherosclerotic intima in blood vessel walls. There BMP signalling is promoted and BMP antagonists suppressed. BMPs then promote the differentiation of osteoprogenitor cells into osteoblast-like cells which produce ALP and initiate mineral crystal precipitation157.

1.8.4 Arthritis

BMPs are associated with arthritis. Many experiments in animals have shown a protective and even reverse effect on the progression of arthritis. Rabbits with anterior cruciate ligament transections were treated with BMP-7 to see if it would prevent osteoarthritis. Matched pair analysis (where the opposite knee on the same animal is used as a control) appeared to show a protective effect with no apparent side effects such as ectopic bone formation158. These data though promising were not wholly convincing, particularly in regard to the histological osteoarthritis score in which the standard deviation (SD) was extremely high at the mid-late time points and the small sample size (n=3). Rats subjected to forced strenuous running on treadmills for long periods were used to mimic mechanical stress induced osteoarthritis159. There are more effective preventatives being trialled, however there is some evidence that BMP-7 may induce matrix deposition as well making it a potential treatment for more severe osteoarthritis159.

1.8.5 Bone Disorders

The BMP signalling pathway has a major role in bone disorders, both in loss through osteoporosis and in excessive bone development in Fibrodysplasia Ossificans Progressiva (FOP). Loss of balance between BMPs and their antagonists can determine whether fractures heal or not160 and because BMPs, particularly BMP-7, have the capacity to induce new bone formation they have frequently been proposed as an attractive target for promoting bone regeneration in humans6. In addition, in spinal surgery BMPs are used as a substitute for bone grafting because they carry a reduced risk of pseudoarthritis and intense post-operative pain161. In dentistry, fitting dental implants in patients with low bone quality can be facilitated by improving bone quality with BMPs beforehand162. In the future the potential of BMPs to enhance angiogenesis may also play a role in using them for bone regrowth163.

Increased BMP signalling causes FOP, a rare but incurable and life-limiting condition in which ectopic bone is deposited throughout the patient’s life, gradually welding the joints together and preventing movement164. Antagonism of BMP may offer therapeutic benefit in this condition. In cases of classic FOP a mutant form of the BMPR I receptor, ALK2 is almost always seen165. This mutant receptor is constitutively active and though responsive to BMPs can signal without it. The near normal prenatal development of individuals with the mutation is explained by the weakness of the constitutive signal166.

Other components of the pathway that have been identified as therapeutic targets for bone disorders include Tsg, which when knocked down in mice leads to osteoporosis and deformed bones120 and particularly crossveinless2 which is associated with a condition called diaphanospondylodystosis (DSD)114. DSD is a rare recessive condition which causes abnormal formation of the skeleton and results in perinatal death due to respiratory insufficiency. Mutations in crossveinless-2 have been found in many affected individuals and the phenotype is similar to that of the chordin null mouse discussed in section 1.6. It is thought that the phenotype results from an early abnormal mesenchymal differentiation leading to distorted growth of the vertebrae114.

1.8.6 Delivery Methods

BMP-2 and BMP-7 are licensed for limited clinical administration, in particular for surgery as a substitute for bone grafts167. There is still a question of the best way to manipulate the pathway, either directly or via antagonists, and the way in which to administer it. Direct injection of either BMPs or an antagonist would be unwise because in most cases (FOP being the major exception) the effect should be limited to a specific area and these extracellular signalling molecules are diffusible. The ideal carriers for any protein should be target specific, easy to administer, sterile, non-toxic and unlikely to promote an allergic

53 reaction. Particularly for BMPs the carrier should be biodegradable but protect BMPs, which have a short life in vivo, for long enough to produce the required effect while releasing them slowly167. The three types of carriers are (1) biological carriers such as collagen and agarose which are good for biocompatibility but bad for contamination risks, (2) inorganic materials such as calcium salts and bioglass and (3) synthetic materials such as hydrogel163.

Implants are an interesting option as they not only deliver the BMPs and retain them at the desired site, but they also provide a scaffold for cells of regenerated tissue167. Delivery methods have been optimised for BMP specifically, for example Bioverit II couples macroscopic roughness with porous silica coatings to give very effective delivery, with reduced potential for high dose side effects168. This is a likely risk with many delivery methods which gives a temporary high release when first administered before settling to the desired release rate167. Significantly this method completely immobilises BMP-2 via silane linkers instead of slowly releasing it, meaning that it cannot diffuse to non-target tissues168.

Adenoviral delivery methods have also been explored as a method for localised activity for BMP pathway components, including Tsg, for combating bone loss in osteoporosis169 and for improving bone quality in dentistry prior to inserting dental implants162. In rabbits, this method was more successful than BMP releasing implants in preparing the jaw for dental implants162. In one experiment an adenoviral vector was used to overexpress Tsg in osteoclasts, which inhibited osteoclast differentiation blocking their activity in a highly specific and localized way169. There are significant risks with this system however including ectopic expression if the virus infects the wrong cell type or promoter and there is the possibility of provoking a strong immune response162. Combinations of different BMPs, depending on the ratio and dose strength, can either favour osteoclast or osteoblast production. Because bone repair requires both deposition and resorption of bone this is highly significant for BMP mediated therapy21. In addition BMP-2 and stromal cell-derived factor-1 can both regenerate bone in isolation but when released in implants simultaneously the effect was greatly enhanced163.

1.8.7 Safety

The safety of BMP use in therapy is of significant concern, particularly in relation to ectopic bone formation and its potential role in cancer progression. Although BMPs -2 and -7 are FDA approved, they have only been applied clinically relatively recently so it is not yet known what the long term side effects may be161. In some trials use of BMP in animal joints to treat injury has not produced any obvious side effects170 but in other experiments using BMP to regrow damaged bone, unwanted bone growth has been produced163. This may, however, simply be a case of improving delivery methods. Carriers such as fibrin glue have been suggested as a means of inhibiting ectopic bone formation from therapeutic BMPs161.

A sample of almost 16,000 patients in whom BMP-2 or BMP-7 were used for spinal surgery was followed up after a year and a half. The study found no increased risk of pancreatic cancer compared to a larger control group who had the same surgery without the protein171. This study was commissioned because in 2004 Wyeth, a BMP-2 manufacturer, found a higher rate of the cancer of those treated with BMP-2 than those who were not in a routine safety review. However this was a small trial and was specific for this type of surgery as other surgeries using BMP-2 produced no such correlation21. Though reassuring, the ratios of BMP-2 and BMP-7 administered are not recorded and given the differing biological and clinical effects of the two BMPs, this renders the study somewhat questionable171.

1.8.8 Industrial Applications

As well as medical, there are industrial applications for understanding chordin structure. For example BMP production is extremely expensive and in order to make it more biologically effective, antagonist resistant versions have been developed for noggin172. Understanding BMP interaction with chordin may help in the design of chordin resistant BMPs.

A better understanding of this mechanism may also have implications in agriculture. In goats, prolifically reproducing breeds have higher serum BMP-4 levels than less prolific breeders. It was also found that different naturally occurring BMP polymorphisms in different breeds in different geographical areas were associated with birth rate173. The mechanism is not clear but BMP-4 is required for follicle stimulating hormone release in females174 and spermatogenesis in males173. If different polymorphisms of BMP-4 affect fertility then this information could be used to improve the reproductive capacity of goats, and potentially other livestock173.

1.9 Aims

BMP regulation by chordin and Tsg is highly important in development and the pathway is of interest in a range of pathologies. It has been extensively studied in cells, organisms and in silico, however some of the observations are still difficult to explain. In particular, although cleavage of chordin by tolloids is the primary mechanism for relief of BMP antagonism in vivo, it is unclear structurally why this should be the case when cleavage leaves the binding domains intact. In addition it is not known what, if any, biological function the residual cleavage fragments have. It is clear that they are still capable of some level of BMP antagonism in vivo as overexpression replicates the effects of chordin27, 53. Furthermore it is now well documented how Tsg produces its dual effect on BMP signalling (strengthening the complex, increasing the rate of fragment cleanup and promoting chordin cleavage) but it is not known what effect it has on chordin to achieve this.

These knowledge gaps have particular relevance for computer modelling of the gradient and for therapeutic use of BMP. For example chordin could potentially be the basis for the design of highly selective BMP antagonists, but it would be essential to know why the full length is effective and the fragments are not. Conversely, to produce chordin resistant rhBMPs for therapeutic use (as has been done for noggin) it would not be suitable to alter the BMP- chordin binding site because it is the same as the BMPR interaction site and this could alter specificity with unpredictable consequences. However because of the large size of chordin, steric hindrance by a strategically placed side group might achieve the same effect.

This project aims to take a structural approach to looking at two key questions;

 How does cleavage of chordin by tolloids inactivate chordin when the BMP binding domains remain intact and functional? Cleavage could reduce chordin affinity for BMPs or Tsg directly. It could also change the way it interacts with Tsg, for example Tsg might no longer strengthen BMP-chordin interaction. Cleavage could also increase the rate of chordin internalization or prevent co-operative binding between multiple vWC domains.  What is the mechanism of action for twisted gastrulation? It may work as a molecular scaffold bringing chordin, BMP and tolloids into close proximity to strengthen the complex and promote chordinase activity. Alternatively it may induce conformational change in chordin making it more accessible to BMPs and tolloids. Observations suggesting potential conformational change include that chordin-like 1 is dependent on Tsg for its BMP antagonistic activity, indicating that Tsg does more than act as a scaffold8. In addition in mice, Tsg introduces an entirely new cleavage site in chordin suggesting that the shape may have changed to make it accessible.

The specific aims of this project were;

 The initial aims were to generate and purify twisted gastrulation and chordin constructs including the major cleavage fragments of chordin and a cleavage resistant full length chordin, (shown in Figure 1.13) in a mammalian expression system.  These constructs would then be used to investigate the structural properties of chordin and Tsg using multi-angle light scattering, electron microscopy, analytical ultracentrifugation and small angle X-ray scattering.  Another aim was to show specific binding affinities between BMP-2, Tsg and chordin and its fragments using surface plasmon resonance. A key reason for doing this is that a major difficulty in studying mammalian chordin is separating its larger cleavage fragments from the intact protein. As this issue is not directly addressed in

previous literature analysing BMP-chordin binding, it is possible that if there were a significant difference in affinities these fragments may affect the binding constants.  The final aim was to generate novel data for chordin complexes, in particular looking at how twisted gastrulation interacts with chordin using electron microscopy and fluorescence resonance energy transfer.

Figure 1.13: Protein constructs to be developed for study of the structural interaction between twisted gastrulation and the BMP-antagonist chordin. Thr = thrombin cleavage site. His = poly-histidine tag. All constructs are C-terminal poly-histidine tagged for nickel affinity purification. (A) N-terminal tolloid resistant full length chordin to study full length chordin with less contamination by fragments produced by endogenous tolloids from which it cannot be separated in solution. This will use two point mutations at the cleavage site; Y152  F152 and D154  N154. (B) Chordin ∆N and (C) chordin ∆C which begin and terminate at the same N- and C- terminal cleavage sites respectively. (D) Twisted gastrulation. (E) Twisted gastrulation without a thrombin cleavage site to minimize the size of the tag in applications where it is not removed.

2. Materials and Methods

2.1 Materials

2.1.1 Buffers and Media

All buffers were prepared using ddH2O purified using the Mili-Q Water System (Millipore) and degassed. A full list of buffer compositions is given in Appendix II.

2.1.2 Bacterial Strains

JM109 competent cells were used for all applications (Alliance Bio).

2.1.3 Expression Cells

Human Embryonic Kidney (HEK) containing the Epstein-Barr virus Nuclear Antigen (EBNA) cassette provided by Dr Richard Kammerer (Paul Scherrer Institut) were used for all applications. Full length human chordin containing a myc tag at the N-terminus and a His-tag at the C-terminus in the pCEP-Pu vector was cloned by Dr Richard Berry in the same cell line175.

2.1.4 Plasmid Vectors

The pCEP-Pu vector (Invitrogen) conferring resistance to ampicillin in bacteria and puromycin in mammalian cells was used to express chordin and cleavage resistant chordin with the native signal peptide. Chordin ∆C, chordin ∆N, and Tsg were inserted into the pCEP-Pu AC7 vector (Invitrogen) which differs from the pCEP-Pu vector by the presence of a BM40 signal peptide. Both signal peptide types are cleaved prior to secretion. Full maps of both vectors can be found in Appendix I.

2.2 Molecular Cloning

2.2.1 Recombinant DNA Constructs

Cloning primers were designed terminating in a C or G residue to include the features described in Project Aims, with a complimentary sequence. Owing to the high G/C content around the target sites of chordin, redundant sequencing was used to change some G/C bases to A/T without altering the peptide sequence to reduce primer secondary structure formation. Altered residues are shown in red in Table 2.1. For insertion into the AC7 vector an additional base is required following the 5’ restriction endonuclease site in the insert to keep the construct in-frame, shown in cyan in Table 2.1. All primers were synthesised using the Invitrogen custom oligomer service.

o o Construct Forward Primer Sequence Tm C Reverse Primer Tm C 5’-3’ Sequence 5’-3’

Chordin CTCGTCGCTAGCCGCT 80 CTCGTCGCGGCCGCTC 84 ∆C GGACCCGAGCCTCCAG AATGGTGGTGATGGTG TGCTGCCCATCCGTTC GTGGCTGCCTCTGGGC TGAG ACCAGAGCCTGCATGG GGTCCCCCA

Chordin CTCGTCGCTAGCCGAC 77 CTCGTCGCGGCCGCTC 82 ∆N AGAGGAGAGCCAGGC AATGGTGGTGATGGTG GCTGAGGAGC GTGGCTGCCTCTGGGC ACCAGAGAGCCTTCGG CTTCTTTCTCC

Tsg-His CTCGTCGCTAGCGTGT 69 CTCGTCGGATCCTTAAT 72 AACAAAGCACTCCTGT GGTGGTGATGGTGGTG GCTAG AAACATGCAGTTCATAC ATTTGAC

Tsg-Thr- CTCGTCGCTAGCGTGT 69 CTCGTCGGATCCTTAAT 76 His AACAAAGCACTCCTGT GGTGGTGATGGTGGTG GCTAG GCTCCTCTGGGGACCA GAAACAT

Table 2.1: Table showing the forward and reverse primers used to produce DNA constructs, melting temperature (Tm). All primers were purified by reverse phase chromatography. Bases altered to avoid secondary structure formation are shown in red, polyhistidine tags are shown in green, thrombin cleavage sites are shown in yellow, the extra base to keep the construct in frame in the vector is shown in cyan and restriction endonuclease recognition sites are shown in purple.

2.2.2 Mutagenesis Primers

To produce tolloid resistant chordin, point mutations were introduced changing two residues at the N-terminal cleavage site, 152Y to F and 154D to N respectively. The P1’ aspartic acid residue was chosen because in drosophila sog43 and zebrafish61 chordin mutation of this residue reduced cleavage of chordin by tolloids and this residue is conserved between

60 almost all known tolloid substrates56. To minimize changes that might affect structure just this residue was mutated first, however this did not succeed in reducing cleavage. A second residue, 152Y was then changed to 152F, because in zebrafish mutating both sites produced a cleavage resistant but importantly also biologically active chordin61.

The primers used to produce this construct (chordin-FN) are shown in Table 2.2. Primers were phosphorylated using 0.5µl T4 polynucleotide kinase (PNK from New England Biolabs) with 5mM primer in PNK buffer (New England Biolabs) at 37oC for 1 hour to enable compatibility with the KAPA HiFi Hotstart system (KAPA Biosystems). All primers were synthesised using the Invitrogen custom oligomer service.

o o Construct Forward Primer Sequence Tm C Reverse Primer Sequence Tm C 5’-3’ 5’-3’

Chordin 154D CATCGCAGTTATAGCAAC 82 CTGGCTCCCCGCGGTTG 82  N CGCGGGGAGCCAG CTATAACTGCGATG

Chordin 152Y GGAGCATCGCAGTTTTAG 72 CTCCCCGCGGTTGCTAA 72 F CAACCGCGGGGAG AACTGCGATGCTCC

Table 2.2: Table showing forward and reverse mutagenesis primers used for inserting each point mutation purified by high performance liquid chromatography. Altered bases to introduce the point mutations are highlighted.

2.2.3 Polymerase Chain Reaction

Chordin ∆C, chordin ∆N, Tsg-his, and Tsg-thr-his were amplified using PCR according to the manufacturer’s instructions with the Expand High Fidelity PCR System (Roche). All chordin constructs were cloned from full length chordin while the template for Tsg-his and Tsg-thr-his was image clone number Swissprot Q9GZX9 (Bioscience Gene Service). The PCR reaction mix was composed of 3mM forward primer, 3mM reverse primer, 30ng template DNA,

100µM of each dNTP and 5% DMSO in a 50µl reaction (volume made up with H2O). The PCR cycles for each construct are shown in Table 2.3. Chordin-FN was produced by introducing point mutations at positions 154 and 152 sequentially. The reactions were carried out using KAPA HiFi Hotstart Ready Mix (KAPA Biosystems) according to the manufacturer’s instructions with 5µM of each primer, 1ng of template (full length chordin in vector pCEP-Pu) and 5% DMSO in a 25µl reaction volume. The PCR cycle conditions are shown in Table 2.3. All PCR reactions were run in a 2720 Thermal Cycler (Applied

Biosystems). Following PCR the template plasmid was digested by 50 units of Dpn1 (New England Biolabs) in NEB buffer 4 for 30 minutes at 37oC to ensure only mutated PCR product remained.

Construct First Hold 25 Cycles Final Hold

Denaturing Denaturing Annealing Extension Extension

Chordin ∆C 95oC 2min 95oC 1min 55oC 30s 72oC 4min 72oC 7min

Chordin ∆N 95oC 2min 95oC 1min 55oC 30s 72oC 4min 72oC 7min

Tsg-His 94oC 2min 94oC 30s 55oC 30s 72oC 3min 72oC 7min

Tsg-Thr-His 94oC 2min 94oC 30s 55oC 30s 72oC 3min 72oC 7min

First Hold 16 Cycles

Chordin-N154 95oC 2min 98oC 20s 65oC 15s 72oC 72oC 5min 7.5min

Chordin Y152 95oC 5min 98oC 20s 65oC 15s 72oC 72oC 5min 7.5min

Table 2.3: Cycling parameters for standard PCR reactions for chordin ∆C, chordin ∆N, Tsg- thr-his and Tsg-his and for the mutagenesis PCR reactions for chordin-N154 and chordin- Y152.

2.2.4 Agarose Gel Electrophoresis and Gel Extraction DNA Purification

Samples were run at 150V for 45 minutes in Gel Loading Dye Blue (New England Biolabs) on a 1% agarose gel with 1 in 100 SYBERsafe stain (Invitrogen). Bands were visualized using UV and compared to DNA molecular mass markers (Bioline, New England Biolabs and Promega) to estimate molecular mass. Bands were visualised using UV light or blue light excitation. Bands were excised if DNA to be purified from the agarose gel and transferred to an Eppendorf tube. DNA was then extracted on the principle of nucleic acid adsorption to silica in the presence of high salt, using a QIAEX II Gel Extraction Kit (Qiagen)

62 according to the manufacturer’s instructions. DNA was resuspended in 20µl ddH2O, or in DNA elution buffer (Appendix II) for samples intended for freezing.

2.2.5 Construct Ligation into Cloning Vector

Purified constructs were inserted into the ampicillin and kanomycin resistance cloning vector pCR2.1-TOPO, using the TopoTA cloning kit (Invitrogen) according to the manufacturer’s instructions. The vector was supplied linearised with a single adenine base overhanging complimentary to the thymine base overhang left by Taq DNA polymerases, enabling ligation at 16oC overnight using a final concentration of 0.05U µl-1 T4 DNA ligase (Roche).

The 50µl reaction volume contained 65mM Tris HCl, 5mM MgCl2, 5mM DDT, 1mM ATP in ddH2O.

2.2.6 Construct Ligation into Expression Vector

Constructs and expression vectors were digested separately using 0.25-0.5U µl-1 Nhe1/Not1 (New England Biolabs) for the chordin constructs or Nhe1/BamH1 for Tsg in buffer 4 (New England Biolabs) for 3 hours at 37oC with 50µl DNA. 0.05 U µl-1 calf intestinal derived alkaline phosphatase (Roche) was added to the vector digestion mix and the digestion continued for 15 minutes, before enzyme inactivation at 65oC for 5 minutes. Constructs and vector were purified using agarose gel electrophoresis and gel extraction as described in Section 2.2.4 and ligated as described in Section 2.2.5.

2.2.7 Bacteria Culture and Transformation

20-100µl aliquots of XL1 blue supercompetent cells were thawed and incubated on ice with up to 250ng of plasmid DNA for 30 minutes. Cells were incubated at 42oC for 45 seconds and immediately returned to ice for a further 5 minutes. Room temperature SOC medium (see Appendix II) was then added to the aliquots to a final volume of 250µl and bacteria were allowed to recover for 1 hour at 37oC with shaking at 200rpm. 100-200µl of transformed cells were then spread over 100µg ml-1 ampicillin agar plates (see Appendix II). The cloning vector carries a β-galactosidase expressing gene which is disrupted when DNA is inserted into the multiple cloning site. This allows identification of self-ligated vector hosting colonies by blue-white screening. Plate surfaces were spread with 40µl IPTG (Bioline) and 40µl xGal (Bioline) which is cleaved by β-galactosidase to give a blue product, Plates were incubated overnight at 37oC. Single white colonies from these plates were grown overnight in 100µg ml-1 ampicillin or chloramphenicol LB broth volume (see Appendix II) with 200rpm shaking to amplify the vector for purification.

2.2.8 DNA Purification from Bacteria

Single colonies of plasmid hosting bacteria grown on selective plates were used to inoculate 4ml in LB broth containing 100µg ml-1 ampicillin (see Appendix II). The cultures were grown overnight at 37oC with 200rpm shaking. Bacteria were pelleted at 10,000rpm for 3 minutes at 4oC in a table top microcentrifuge. The supernatant was discarded and bacteria were resuspended in buffer P1 (Qiagen). DNA was extracted based on alkaline lysis of the bacteria and adsorption of DNA onto silica in the presence of high salt using a Plasmid MiniPrep or MidiPrep (Qiagen) according to the manufacturer’s instructions. Purified DNA was eluted or resuspended in ddH2O or 10mM Tris, pH8.5 if intended for freezing.

2.2.9 DNA Sequencing

Constructs were sequenced using BigDye Terminator Cycling sequencing (Applied Biosystems) on an ABI Prism 3100 Genetic Analyser by the University of Manchester sequencing facility. Samples were prepared for sequencing by adding 200µg DNA template with 4pM primer made up to a final volume of 10µl with ddH2O.

2.2.10 DNA Storage

Purified DNA was stored in 10mM Tris-Cl pH8.5 at -20oC. For long term storage plasmids were transformed into XL1 Blue Supercompetent cells (Agilent Technologies) as described in Section 2.2.7. 100µl aliquots of cells were then placed in 1.5µl Eppendorf tubes, with 100µl 50% glycerol and flash frozen in liquid nitrogen. Cells were then immediately transferred to -80oC where they are stored for up to 1 year.

2.3 Protein Expression and Purification

2.3.1 HEK293-EBNA Cell Culture

HEK 293-EBNA cells were cultured in T75cm2 or T225cm2 culture flasks (Costar) with a o canted neck and vented cap, incubated at 37 C in 5% CO2 in 13ml or 25ml DMEM4 culture media respectively (Invitrogen). Standard media was supplemented with 0.002% each of penicillin-streptomycin solution (Sigma), 300mg ml-1 G148 (Sera Laboratories International) and 9% foetal bovine serum (FBS) (Fisher Scientific). Confluent cells were detached from the flask surface using either 2ml or 5ml 0.025% trypsin solution (Sigma Aldrich) depending on the flask size, for up to 5 minutes. The cell suspension was then diluted with an equal volume of culture media and centrifuged at 1,600rpm for 4 minutes. The supernatant was discarded and the pelleted cells resuspended in 6ml culture media which was applied to a

64 single T225cm2 or 2ml to each T75cm2 flask. Culture media was then added to each flask to make up a final volume of 25ml for a T225cm2 or 13ml for a T75cm2 flask.

2.3.2 Lipofectamine Transfection

16µl lipofectamine solution (Invitrogen) and 40µg DNA were incubated in separate tubes each with 200µl serum reduced Optimem media (Invitrogen) for 30 minutes at 25oC. The contents of both tubes were combined and incubated at 25oC for a further 15 minutes. The solution was then added to 60-80% confluent HEK 293-EBNA cells in 4ml Optimem media o and incubated at 37 C in 5% CO2 overnight. The transfection media was then replaced with 13ml standard culture media without antibiotics for a 24 hour recovery period. Both cloning vectors used, pCEP-Pu and pCEP-PuAC7, confer puromycin resistance to stably transfected cells. Following the recovery period, 2mg ml-1 puromycin (Invitrogen) was included in the media for selection and maintenance of the stably transfected cell line.

2.3.3 HEK-293 EBNA Storage

Following treatment with trypsin (as described in Section 2.3.1), the supernatant was discarded and the pellet resuspended in 4ml cell freezing media containing DMSO, DMEM, and FBS (Invitrogen Recovery- exact composition not provided). 0.5ml aliquots were placed in 1.6ml cryogenic tubes (Sarstedt Aktengesellschaft & Co) and cells were transferred to - 80oC in a Mr Frosty freezing box (NALGENE) to freeze at a rate of 1oC per minute. Cells were stored at -80oC for use within 6 months or transferred to liquid nitrogen for long term storage.

2.3.4 Protein Expression in Serum Free Media and Storage

Media was removed from confluent cells and cells were washed twice with Dulbecco’s phosphate buffered saline (Sigma Aldrich). Serum free media DMEM4 (Sigma Aldrich) and HAMS F-12 with L-Glutamine (Sigma Aldrich) were mixed in a 1:1 ratio. 0.002% penicillin- streptomycin solution (Sigma) and a 2mM L-glutamine supplement was added (Sigma Aldrich). For chordin constructs media was also supplemented with 50mM of the BMP-1 inhibitor L-arginine (Sigma Aldrich), to reduce chordin cleavage and increase yields175. After 2-3 days media was collected and spun at 3000rpm for 10 minutes at 4oC to remove cell debris. A protease inhibitor cocktail of 0.3mM N-ethylmaleimide and 0.5mM phenylmethane sulporyl fluoride (PMSF) was added immediately to collected media. Media not intended for immediate use was frozen at -20oC and thawed at 40C for use.

2.3.5 Nickel Affinity Chromatography

1ml and 5ml HisTrap HP nickel affinity columns (GE Healthcare) were equilibrated with HisTrap binding buffer (see Appendix II) containing 10mM imidazole for 5 column volumes at 4oC. Media was prepared by dialysing overnight against 5L of HisTrap binding buffer (see Appendix II). Imidazole was added to the media to a concentration of 10mM and for chordin constructs 2M urea was also added to prevent smaller fragments (produced by endogenous tolloids secreted by the cells) from binding and copurifying with the immobilized construct. The media was passed through the column at a rate of 1 column volume/min. The column was then washed with at least 5 column volumes of binding buffer. Protein was eluted with HisTrap elution buffer (see Appendix II) containing 500mM imidazole and collected in 1ml fractions. When required, the polyhistidine tag was removed by thrombin digestion at room temperature for 10 hours in 2mM CaCl2 following nickel affinity purification. The sample was then passed over the nickel affinity column a second time to remove uncleaved protein and the cleaved tag.

2.3.6 Protein Concentration by Centrifugation

Vivaspin centrifugal concentrator columns (Sartorius) were blocked in 0.5% milk in Tris buffered saline (see Appendix II) with 0.01% tween (TBST) for 30 minutes. The milk was removed, and columns equilibrated by spinning in the same buffer as the protein at 8000g. The buffer was removed and protein added to the column and spun at 8000g to the required final volume.

2.3.7 Size Exclusion Chromatography

A Superdex 200 10/300GL gel filtration column (GE Healthcare) was equilibrated with filtered, degassed ddH2O for 1 hour on an ÄKTA purifier system (GE Healthcare) at a flow rate of 0.5ml min-1 and a pressure limit of 1.5mPa. The column was then equilibrated with filtered and degassed Tris-HCl buffered saline for a further 2 hours. 0.5-1ml of protein sample was applied to the column and collected in 0.5ml elution fractions. In most cases the peak was collected, concentrated and size exclusion purified a second time to achieve the required purity. Chordin constructs which were cleaved and copurified with vWC4 with a His- tag required an initial size exclusion step in 2M urea to separate the vWC4 domain from the larger construct. Depending on the intended application, HEPES or sodium phosphate could be substituted for Tris-HCl as the buffering agent.

2.3.8 Protein Concentration Estimation

The concentration of protein was estimated using a nanodrop 2000 (Thermo Scientific) at 280nm assuming 1 absorbance unit is equal to 1mg ml-1. This value was then adjusted based on the extinction coefficient (calculated from the peptide sequence using EXPASY) of the protein measured to account for different proportions of aromatic residues in the peptide sequence.

2.3.9 Stable Complex Formation

Purified chordin with the histidine tag removed and Tsg-his were incubated in a 1:3 chordin:Tsg molar ratio for 30 minutes at room temperature. Chordin and the chordin-Tsg complex are too similar in size to separate by size exclusion chromatography, so the chordin-Tsg mixture was passed over a nickel affinity column. Unbound chordin flowed through while Tsg and the chordin-Tsg complex were trapped and eluted. The complex was then purified by size exclusion chromatography to separate the complex from unbound Tsg. The complex was used immediately following purification for electron microscopy to avoid dissociation.

2.4 Protein Characterization

2.4.1 Polyacrylamide Gel Electrophoresis

Protein samples were run on 4-12% NuPAGE Bis-Tris gels (Invitrogen) in MOPS or MES buffer (see Appendix II) at 200V for 35 minutes in a Novex MiniGel Tank (Invitrogen) according to the manufacturer’s instructions. Proteins were mixed with SDS sample loading buffer (see Appendix II) with 5% v/v β-mercaptoethanol and heated to 70oC for 7 minutes. 5- 10µl protein standard markers (Invitrogen, New England Biolabs) were run alongside the gels for molecular mass estimate. Unless the gels were intended for Western blotting, bands were visualised using InstantBlue Stain (Expedion Protein Solutions) and dried by soaking in 20% methanol, 2% glycerol then placing between two sheets of DryEase Mini Cellophane (Invitrogen).

2.4.2 Western Blotting

Following electrophoresis, protein bands and standard markers were transferred to a PVDF membrane using an XCell Surelock box in a Novel MiniGel Tank (Invitrogen). The gel, EPAGE blotting pads (Invitrogen) and filter paper were soaked in transfer buffer (see Appendix II) for 10 minutes. PVDF membrane was cut to size and activated in 100%

67 methanol for 5 seconds before soaking in transfer buffer for 10 minutes. The box was assembled according to the manufacturer’s instructions, locked into the tank, and run in transfer buffer at 30V for 2 hours. Following transfer membranes were blocked in 5% milk, 0.01% TWEEN-20 in TBST (see Appendix II) for 1 hour at room temperature or 4oC overnight. Mouse α-Histidine primary antibody (R&D Systems) was diluted 1:1,000 in blocking solution and incubated with rocking on the membrane for 1 hour at room temperature.

Donkey α-mouse secondary antibody (R&D Systems) was diluted 1:10,000 in blocking solution and incubated with rocking on the membrane for 1 hour at room temperature. After each antibody step the membrane was washed 3 times with phosphate buffered saline (see Appendix II) for 5 minutes. After the final wash ECL Western Blotting Substrate (Abcam) was used to cover the blot and the reaction allowed to proceed for 1 minute. The blot was then exposed to BioMAX MR Film (Kodak) in a darkroom for 30 seconds to 10 minutes. The blot was developed using a JP33 X-ray Film Processor (JPI Healthcare Solutions).

2.4.3 Circular Dichroism

Purified proteins were loaded into a Jasco J715 spectropolarimeter and the difference in absorbance of left- and right-handed circularly polarized light (190-260nm) measured using the sample buffer as a reference. The shape of the curve was compared to known patterns to verify that the proteins were folded and comparing to the CD spectra of proteins with known structures using the algorithms Contin, Selcon3, CDSSTR and k2D provided by the Dichroweb server176.

2.4.4 PNGase F Digestion

Proteins were mixed with a final concentration of 0.04U ml-1 of the endoglycosidase PNGase-F at 37oC for 24 hours (48 hours for Tsg). Complete removal of N-linked glycosylation was then verified using SDS-PAGE and comparing to deglycosylation under denaturing conditions.

2.4.5 MASS Spectrometry Identification

Proteins were separated using SDS PAGE and stained as described in Section 2.4.1. Bands were excised and sent to the University of Manchester Biomolecular Analysis Core Facility where proteins were digested with trypsin, reduced and alkylated and analysed using a HCT Ultra MS (Bruker Daltronics). Detected fragments were compared to SwissProt (version

2011-05) and Uniprot (version 2011-05) databases to match them to known protein fragment patterns.

2.5 Biomolecular Analysis

2.5.1 Multi-Angle Light Scattering

Protein molecular mass was estimated using Multi-Angle Light Scattering. Proteins were first separated by size using a Superdex200 size exclusion column at a flow rate of 0.72ml per minute in degassed 10mM Tris, 150mM NaCl, pH7.4 buffer on a Biol HPLC pump (Dionex). Protein was passed through an EOS 18 laser photometer with the 13th detector replaced by a QELS detector (Wyatt Technologies) for simultaneous measurement of hydrodynamic radius. The sample then passed to a Optilab rEX refractive index detector (Wyatt

Technologies). The hydrodynamic radius, molecular mass moments MN, MW, and MZ and concentration of the peaks were analysed using Astra 5.3.2 software (Wyatt Technologies).

2.5.2 Analytical Centrifugation

2.5.2.1 Velocity Sedimentation

Wavelength for sedimentation velocity were optimized using a Camspec M501 single beam scanning colorimeter (Spectronic) taking a wavelength higher than 230nm where absorbance is between 0.5-1.5 absorbance units. Sedimentation velocity studies were carried out using a Beckman XL-A ultracentrifuge with an An60Ti 4 hole rotor with two-sector centrepieces and quartz glass windows, where one sector was loaded with the sample buffer as the reference and one sector with sample protein. Samples were analysed at 25,000rpm at 20oC with protein absorbance recorded at radial positions between 5.8 and 7.2cm in a vacuum. The time between scans was set at 90 seconds for 200-300 scans and a radial step size of 0.003cm. A three minute delay was fixed before the first scan to ensure that the rotor was fully accelerated to 25,000rpm. Data were interpreted using Sedfit software177 with a model based distribution of Lamm equation solutions C(s) to obtain a 178 sedimentation coefficient corrected to a standard value in water (S20W) . SEDNTERP software179 was used to calculate the partial specific volume from amino acid composition and the buffer density and viscosity. The frictional ratio (f/fo) and hydrodynamic radius (Rh) 179 were calculated using SEDNTERP from S20W .

2.5.2.2 Sedimentation Equilibrium

Sedimentation equilibrium was performed using the equipment described above with three different protein concentrations at a single speed. Concentrations were optimized using a Camspec M501 single beam scanning colorimeter (Spectronic) taking a higher concentration up to 0.5 absorbance units at 280nm and lower concentration where absorbance is 0.2 units at 230nM and a third concentration in between scanned at 230nM. The proteins were spun at 10,000prm at 4oC with protein absorbance recorded at radial positions between 7 and 7.1 in a vacuum. An approach to equilibrium method was employed where cells were scanned at 16 hours and then every two hours for 20 replicate scans up to 20h. Time between scans was set at 1 hour with a radial step size of 0.001cm. Sedphat software (Schuck, P.)177 was used to perform global analysis of these data by non-linear regression.

2.5.3 Transmission Electron Microscopy

Carbon coated Cu400 grids (Agar Scientific) were glow discharged using a K100X glow discharger (Emitech) at 25mA negative current for 90s. Within 30 minutes, 5µl of 15µg ml-1 sample was applied to the grid and allowed to adsorb for 15s. Grids were washed twice with sample buffer or ddH2O and negatively stained with 2.5% uranyl acetate for 15s. Data were recorded using an FEI Technai BioTwin transmission electron microscope operating at 100kV, 23,000X (2.5Å/pixel) with a defocus range around 0.5-1 microns. Particles were selected using the swarm function in Eman2180 using a box size of 96 pixel (Technai data) or 72 pixel (Polara data). The interactive tuning function was then used to determine accurate contrast transfer function (CTF) correction. Particle sets were built from these data, bad particles removed and class averages generated. Near-duplicate class averages were not included to avoid bias. The class sum images were then used to generate a 3D model which was underwent multiple rounds of refinement to generate a final model. The models were then assessed using Fourier Shell Correlation.

2.5.4 Small Angle X-Ray Scattering

Small angle X-ray scattering data were collected at 1-12mg/ml on the Diamond Light Source (I22) and PETRA III (P12) beamlines. Samples were maintained at 10oC during exposure using the standard sample holders at each beamline. Data were checked for radiation damage and undamaged data (sometimes at different concentrations) merged using Primus181. Ab initio models were calculated using either DAMMIN182 software or GASBOR183 software over a minimum of 20 repeats. For each program the models were then combined into a single model and averaged using DAMAVER and DAMFILT software to produce a

70 final ab initio model184. All software packages were part of the ATSAS suite produced by EMBL (Hamburg).

2.5.5 Structural Data Analysis

Hydrodynamic properties of the EM and SAXS models were estimated using Hydromic185 and Hydropro186 respectively. The input values were derived from sequence (molecular mass and partial specific volume) and the bead or ab initio model files (Hydromic converts the EM model to a bead model to use in the calculations). Predicted values can then be cross compared both between types of model and to experimental data, e.g. the hydrodynamic radius and sedimentation coefficient from AUC.

2.5.6 Surface Plasmon Resonance

Ligand protein for the solid phase was immobilised at a concentration of 100nM in SPR immobilization buffer pH 4.5 (see Appendix II) at a flow rate of 12µl min-1 using a CM5 sensor chip in a Biacore 3000 (GE Healthcare). For kinetic analysis, analyte protein was diluted to a range of concentrations in the 1-1500nM range in 10mM HEPES, 150mM NaCl, pH7.4 buffer (see Appendix II) and passed over the chip at a rate of 30µl min-1 for 300 seconds. For each set of dilutions a blank was run prior to the protein and one reference lane containing only buffer was run at the same time. Samples were double-referenced against a single lane of the chip activated and deactivated with no ligand immobilized. Lanes were regenerated using 1M NaCl and 50mM NaOH solution at a rate of 30µl min-1 for 180 seconds. Data were analysed by kinetic analysis using Biacore X-Control software

3. Biomolecular Analysis of Chordin

3.1 Summary

In order to gain insight into the nanostructure of chordin and its fragments; a mutant tolloid cleavage-resistant full length chordin construct and constructs of chordin cleaved at the C- terminal site (∆C) and N-terminal site (∆N) were produced. The mutation of the tolloid cleavage site reduced, but did not eliminate ∆N fragments in full length chordin. However, coupled with L-arginine inhibition of endogenous tolloids in the expression media, the proportion of fragment produced in the sample was reduced to very low levels. Circular dichroism was used to confirm that the constructs were folded and make estimates of secondary structure composition. It was also shown that human chordin is a glycoprotein with N-linked sugars. Surface plasmon resonance (SPR) was then used to compare the binding of full length chordin and the fragments to BMP-2, confirming binding activity in all three constructs. Multi-angle light scattering (MALS) was then used to determine the molecular mass and some of the hydrodynamic properties of full length chordin. From SPR and MALS data it was predicted that chordin may form a reversibly associating dimer. This is investigated in the final section of the chapter using analytical ultracentrifugation (AUC).

3.2 Cloning and Expression of Chordin Constructs

To study chordin structure, chordin with a mutated tolloid cleavage site was produced. The aim was to produce a chordin construct which could not be cleaved by endogenous tolloids in the expression media to allow production of a sample of full length chordin which did not contain any smaller fragments. Since the construct is purified by a C-terminal histidine tag cleavage fragments which retain the tag are co-purified. The larger tagged fragment, ∆N, is too similar in size and chemical properties to separate from the full length which has posed problems in previous studies175. In addition, constructs mimicking the natural cleavage products of chordin by tolloid metalloproteinases were cloned into the expression vector pCEP-Pu AC7 and expressed in HEK 293-EBNA cells. Figure 3.1A shows the PCR data for the chordin constructs. The two point mutations in the N-terminal cleavage site of chordin were inserted sequentially. Due to the ubiquitous presence of an aspartic acid residue in the P’ position of almost all known tolloid substrates it was tested whether substitution of this alone would be sufficient to prevent cleavage. No significant reduction in cleavage was observed however, so a second point mutation was introduced which has been shown in zebrafish to result in tolloid resistant, biologically active chordin61.

Figure 3.1: Cloning of chordin constructs. (A) Agarose gel electrophoresis analysis of PCR products: (i) wild type chordin (positive control, 2952bp), ∆C (2646bp) and ∆N (2469bp). (ii) Chordin in pCEP-Pu expression vector following Dpn1 digestion with the first point mutation (D154->N154). (iii) Chordin in pCEP-Pu expression vector following Dpn1 digestion with the second point mutation (Y152->F152). (B) Agarose gel of uncut and cut vector containing the constructs showing vectors and inserts consistent with the predicted size. (i) Wild type chordin positive control and chordin-FN. (ii) Chordin ∆C and chordin ∆N constructs. (C) Western blot using mouse anti-polyhistidine primary antibody. (i) Chordin-FN expression media one day post-transfection and 4 x concentrated chordin-FN expression media. Chordin without the mutated cleavage site shown for comparison. (ii) Chordin ∆C and chordin ∆N expression media. vWC4 copurified with ∆N by the histidine tag but can be separated using size exclusion chromatography.

Figure 3.1B shows insertion of fragments of the correct size into the expression vector. These were verified by sequencing to rule out PCR induced mutations. Following transfection, Western blots using antibodies for the poly-histidine tags showed fragments produced of the expected size, shown in Figure 3.1C. Although resistant to cleavage by tolloid metalloproteinases, when concentrated it became apparent that some limited cleavage of the mutant chordin was still occurring. However when the mutant chordin was

74 expressed in the presence of 50mM L-arginine to partially inhibit tolloids, it was possible to reduce this contamination to a trace. Chordin ∆C had a large proportion of terminal fragment which again could be reduced by addition of L-arginine (see Introduction Figure 1.6 for a diagram of the cleavage fragments). Pure terminal fragment could be produced by digestion of the larger constructs with BMP-1 and size exclusion to remove the cleaved vWC domains.

3.3 Protein Purification and Fragment Separation

The constructs were purified using nickel affinity chromatography to separate them from other components of the expression media by their histidine tags. Proteins were immobilized in HisTrap binding buffer (see Appendix II) containing 10mM Tris-HCl, 500mM NaCl and 10mM imidazole pH7.4 and eluted in HisTrap elution buffer containing 500mM imidazole. They were then further purified using size exclusion chromatography to separate them from aggregated material (shown in Figure 3.2A). The buffer used for this depended on the intended application. Unless otherwise stated 10mM Tris-HCl, 150mM NaCl at pH7.4 was used. The purified proteins are shown in Figure 3.2B and C, and were harvested in the presence of 50mM L-arginine. Figure 3.2B also shows PNGase-F digestion of full length chordin revealing N-linked glycosylation. The potential N-linked glycosylation sites determined from sequence are all located within the four CHRD domains. For applications such as surface plasmon resonance the thrombin cleavage site was used to remove the histidine tag by thrombin cleavage to avoid it engaging in non-specific binding. Western blotting using an anti-histidine antibody was used to ensure complete removal of the tag, shown in Figure 3.2D. Following digestion the antibody no longer recognises the chordin constructs demonstrating that the tag is no longer present.

Figure 3.2: Chordin purification. (A) Size exclusion purification of full length chordin monitored by UV absorption at 280nm showing a single purification peak. (B) SDS PAGE of mutant chordin, native and digested by PNGase-F demonstrating the presence of N-linked glycosylation. (C) Chordin ∆C, chordin ∆N and the terminal fragment purified from ∆C. In the case of chordin ∆C where the N-terminal cleavage site was not mutated there is a higher proportion of cleavage fragment present than in the case of full length chordin. (D) Western blots showing complete removal of the polyhistidine tag from full length chordin and chordin ∆N.

Bands were excised from the gel and identified using mass spectrometry to confirm that the purified protein was chordin. Analysis of matches to the databases did not reveal any chordin peptides before the N-terminal cleavage site in the case of chordin ∆N or after the C- terminal site for chordin ∆C (see Appendix IV). The smaller bands contained chordin peptides, confirming that they are copurifying cleavage fragments rather than contaminants.

Matches were also found for trypsin, keratin, heatshock protein and galectin 3 binding protein introduced during the course of the mass spec sample preparation. The results are summarized in the Table 3.1.

SwissProt Hits IPI Human Hits Mutant chordin chordin isoform 1 human (20) chordin isoform 1 human (26) (upper band) trypsin pig (3) keratin type I human (2) keratin type I human (2) keratin type II human (1) keratin type II human (1) Mutant chordin chordin isoform 1 human (3) chordin isoform 1 human (25) (lower band) trypsin pig (1) keratin type I human (5) keratin type I human (1) keratin type II human (1) keratin type II human (4) Chordin ∆C chordin isoform 1 human(17) chordin isoform 1 human (19) heatshock protein zebrafish (3) heatshock protein human (3) galectin 3 binding protein human galectin binding protein human (3) (2) Chordin ∆N chordin isoform 1 human (21) chordin isoform 1 human (27) trypsin pig (3)

Terminal chordin isoform 1 human (4) chordin isoform 1 human (4) fragment

Table 3.1: Table showing mass spec ID results for FN-chordin and the constructs chordin ∆C, ∆N and the terminal fragment from comparison to the SwissProt database (all species) and IPI database (human only). The proteins expressed and purified, and the fragments copurified with them are chordin. The peptide matches are shown in Appendix IV.

3.4 Circular Dichroism

3.4.1 Prediction of Secondary Structure of Full Length Chordin using SRCD

In order to confirm folding and predict secondary structure, chordin was analysed using circular dichroism at a sensitivity of 100mdeg and a rate of 20nm min-1. Expression levels of full length chordin are very low and at the concentrations available normalized root mean standard deviation (NRMSD) was unacceptably high when these data from a lab CD source were analysed using algorithms on the Dichroweb server176. Synchrotron radiation circular dichroism (SRCD) was used instead with sensitivity of 100 mdeg scanned at a rate of 60 nm min-1 and 0.1 mg ml-1 because SRCD minimizes the signal to noise ratio at lower

77 concentrations. Due to the sensitivity of SRCD to salt in the buffer at low wavelengths, these data were collected in pure water with no buffering agent at pH7.4. The trace (Figure 3.3) shows a dip at around 210nm and a steady absorbance at higher wavelengths indicating a folded protein. The shape of the graph is characteristic of a protein with mixed secondary structure types with a deep trough at around 208nm indicating high proportion of β-sheet.

Figure 3.3: CD trace of full length chordin from SRCD showing mdeg as a function of wavelength (nm). The shape of the trace indicates a folded protein with a high proportion of β-strand secondary structure, but some α-helix.

These data were analysed using Contin (average of all matching solutions), K2d, Selcon3 and CDSSTR on the Dichroweb server. A summary of the secondary structure predictions is provided in Table 3.2. The averaged results indicate that chordin is composed of 34% β- sheet and 9.7% α-helix with the remainder being turns and unordered regions. According to the crystal structure of the related vWC1 domain of crossveinless-2, the β-strands in the domain are relatively short33 and chordin has several long linker regions (see Appendix III) so the high proportion of unordered and connector structure is unsurprising. There is reasonable consistency between the predictions shown in Table 3.2 which indicate a largely β-strand protein but with a significant proportion of α-helix. The vWC domains are predicted

78 to be composed entirely of sheet and turns so it is probable that the helical regions are found in the CHRD domains which are so far structurally uncharacterised.

Helix Strand Turns Unordered NRMSD Contin 0.107 0.363 0.208 0.351 0.126 Selcon3 0.052 0.374 0.269 0.39 0.178 CDSSTR 0.11 0.33 0.23 0.31 0.063 K2d 0.12 0.34 N/A 0.55 0.104 Mean 0.097 0.34 0.236 0.35 N/A

Table 3.2: Analysis summary of SRCD data for full length chordin using the Contin, Selcon3, CDSSTR and K2d algorithms showing mean of the predicted values. Calculation of the mean of the unordered region excludes K2d because the algorithm does not distinguish between turns and unordered regions.

3.4.2 CD Secondary Structure Prediction of Chordin ∆C and ∆N

The chordin ∆C expresses at higher levels than full length chordin, so a lab CD source was used to make secondary structure predictions at 0.4mg ml-1. The SRCD sample for chordin ∆N was found to have aggregated so the lab CD source at 0.1mg ml-1 was used. This technique is less sensitive to salts so standard buffer (10mM Tris, 150mM NaCl, pH7.4) was used. The traces, shown below in Figure 3.4, are similar to that of full length chordin as would be expected. The results of sequence analysis using Contin (average of all matching solutions), K2d, Selcon3 and CDSSTR on the Dichroweb server are shown in Table 3.3. These data have a low signal to noise ratio and as a result NRMSD values are high for Contin and Selcon3. As the difference between these constructs and full length chordin is the absence of a vWC domain it was predicted that the proportion of β-strand would fall and the proportion of α-helix would rise. This is what these data show although the rise in helical proportion is more than expected, particularly in the case of ∆N.

Figure 3.4: CD traces for chordin ∆C and ∆N showing mdeg as a function of wavelength (nm). The shape of the trace indicates a folded protein with a high proportion of β-sheet secondary structure. In the chordin ∆N sample the signal to noise ratio is high which may account for high NRMSD values in the analysis.

Chordin ∆C Chordin ∆N Helix Strand Turns Unordered NRMSD Helix Strand Turns Unordered NRMSD Contin 0.05 0.061 0.335 0.547 0.397 0.105 0.258 0.292 0.346 0.301 Selcon3 0.176 0.273 0.233 0.343 0.436 0.207 0.278 0.233 0.315 0.578 CDSSTR 0.11 0.32 0.25 0.32 0.018 0.48 0.24 0.07 0.22 0.002 K2d 0.29 0.11 N/D 0.59 0.118 0.25 0.29 N/A 0.47 0.138 Mean 0.16 0.191 0.273 0.403 N/A 0.26 0.267 0.198 0.338 N/A

Table 3.3: Table summarizing analysis of CD spectra for the large chordin fragments using the Contin, Selcon3, CDSSTR and K2d algorithms. Mean calculation for the unordered regions excludes K2d, which does not distinguish between turns and unordered regions. Variation between predictions from the different algorithms is relatively high.

3.5 Surface Plasmon Resonance

3.5.1 Kinetic SPR Analysis

Previous in vivo chordin studies have revealed that the biological activity of chordin fragments is less than that of full length chordin27, 39. One possible reason for the reduced biological activity of the chordin fragments is reduced affinity for BMPs. To investigate this, rhBMP-2 produced in E.coli (kindly provided by Thomas Mueller, University of Wuerzburg) was immobilised on a CM5 sensor chip in SPR immobilization buffer pH4.5 (see appendix II)

80 and the binding affinity of full length chordin and the fragments compared over a concentration range of 15.6-1000nM in a running buffer of 10mM HEPES, 150mM NaCl, pH7.4. Curves were fitted using the Langmuir 1:1 on/off model, shown in Figure 3.5A. However model fitting of the curves and, as a result, Kd calculations were inaccurate with χ2 values between 50 and 150 for both models. A bivalent model which fits two binding sites independently and two state model which accounts for conformational change were also used but the accuracy of these fits was even lower. Estimates were 17nM for ∆N and 109nM for full length chordin. This indicates increased affinity of the fragment, however compared to previous studies which place the Kd of full length chordin for BMP in the low nM range the estimate for full length chordin was unrealistically weak8, 27. Similar results were found for chordin ∆C (data not shown) in the mid-nM range.

Figure 3.5: Kinetic analysis of chordin-BMP-2 binding. (A) Binding of 0-1000nM full length chordin and chordin ∆N to immobilized BMP-2 using surface plasmon resonance showing 1 to 1 fitting. (B) Normalized response shown as a percentage of response from the strongest analyte as a function of concentration. The curve suggests a more rapid on-rate for chordin ∆N than the full length chordin.

To determine if cleavage at the N-terminal tolloid recognition site of chordin causes reduced binding affinity to BMP-2, the results were compared directly by plotting the response difference against concentration. This was done by taking the maximum binding at a fixed

81 timepoint (180s) after injection for the strongest binder (1000nM ∆N) and plotting the other analytes as a percentage of this. This graph, shown in Figure 3.5B, shows that response increases with increasing concentration of ∆N more than full length chordin which indicates that cleavage does not result in significant loss of affinity. It is not possible without an accurate fit to draw any more significant conclusions from these data than this, however these data agree with previous studies which also do not show a major difference between binding strength of full length chordin and some of the constructs containing C-terminal binding domains8, 27.

3.5.2 Equilibrium SPR Analysis

While these data show that the chordin constructs are able to bind BMP strongly, thereby verifying that the proteins produced are functional, accurate Kds were not determined. To attempt to overcome this, equilibrium analysis was used. However, it was found that even using long binding times (up to 12 minutes) at high concentrations (1000nM) of analyte with low (50RU) quantities of bound ligand, binding did not come to equilibrium (representative curves are shown in Figure 3.6A). It was predicted from this that chordin may have some self-affinity, resulting in chordin interacting with BMP-bound chordin on the chip. To counter this a small amount of urea in the media was used to block non-specific binding, although this could also weaken the BMP-chordin interaction. Ideally the solution would be to use BMP-2 as the analyte, however the insolubility of the BMP-2 prevented this. Repeats were done varying binding times and buffer conditions for chordin ∆N to attempt to optimize the experiment but equilibrium was never reached. Analysis of the best fit, obtained using 0.25M urea, shown in Figure 3.6B gives a Kd of 27nM (χ2 = 10) which is similar to that obtained for full length chordin in previous studies (37nM)8.

Figure 3.6: Equilibrium analysis of chordin-BMP-2 binding with chordin as the analyte. (A) Representative equilibrium binding curves comparing full length chordin and chordin ∆N. Binding fails to reach equilibrium even at 1µM concentration, possibly due to self-affinity in chordin. (B) Plot of maximum response of chordin ∆N in 0.25M urea against concentration. Binding was still not fully saturated, however this gave the most accurate Kd result (χ2 = 10) of 27nM.

3.6 Multi-Angle Light Scattering

To determine the size and oligomeric state of chordin and its fragments, these were analysed using multi angle light scattering (MALS) in line with a Superdex 200 size exclusion column. Samples were run in 10mM Tris-HCl, 150mM NaCl, pH7.4. Figure 3.7A shows the result for full length chordin. The experimentally determined mass was 101kDa which is close to that predicted from the sequence (102kDa). Chordin would be expected to be slightly larger owing to the presence of N-linked glycosylation, however SDS-PAGE analysis indicates that sugars only account for a very small proportion of total mass. However the fragments have an experimentally determined mass which is larger than would be expected based on sequence. It is possible that this may be in part due to reversibly associating

83 dimers of chordin. The experimental masses of the fragments were found to be 101kDa, 93kDa and 86kDa for chordin ∆C, chordin ∆N and the terminal fragment respectively, summarized in table 3.4. For full length chordin a radius of gyration (Rg) of 5.4 and a hydrodynamic radius (Rh) of 4.4 were also measured using the combined MALS-QELS system.

Figure 3.7: MALS analysis of (A) full length chordin and (B) the fragment constructs chordin ∆C (red), chordin ∆N (purple) and the terminal fragment (blue). The elution volume is plotted against normalized differential refractive index and mass. In most cases the experimentally determined mass is larger than predicted from sequence which may be due to a combination of glycosylation and reversibly associating dimer.

Full Length Chordin ∆C Chordin ∆N Terminal Chordin Fragment Predicted Size (kDa) 102 92 87 76 Size (kDa) 101 101 93 86

Polydispersity Mw/Mn (%) 0.3 0.9 4 1

Polydispersity Mz/Mn (%) 0.6 1 7 1

Table 3.4: The predicted sizes for the constructs based on peptide sequence compared to the experimentally determined mass from MALS. Homogeneity estimates are shown based on the calculated polydispersity (%) Mw/Mn and Mz/Mn of the samples.

3.7 Analytical Ultracentrifugation shows that Chordin forms a Reversibly Associating Dimer

An analytical ultracentrifugation (AUC) approach was used to confirm that chordin forms a reversibly associating dimer. Chordin ∆N was used as this construct does not contain any cleavage fragments post-purification and is therefore easier to analyse when only one species and oligomers are present. Samples (in 10mM Tris, 150mM NaCl pH7.4) were centrifuged at three speeds (8,000, 14,000 and 20,000rpm) to achieve equilibrium at concentrations of 0.22, 0.61 and 2.44µM. Data were analysed using monomer-dimer shown in Figure 3.8 and monomer-nmer self association models in the Sedphat software177. The monomer-dimer model provided the best fit to these data shown in Figure 3.8 which at 0.22µM predicted Kd 76µM (RMSD = 0.036) and at 2.44µM 9.9µM (RMSD = 0.02µM). Although it was possible to assess these data and conclude that chordin has self-association in the low-mid µM range it was not possible to bring the system fully to equilibrium before the sample began to aggregate, leading to protein sedimenting into the bottom of the cell.

Figure 3.8: Equilibrium AUC processing of two representative chordin ∆N samples showing absorbance as a function of radial position, (A) 0.22µM and (B) 2.44µM. Fits to the raw data are shown in the top panel and the residuals are shown in the bottom panel at 8000rpm (blue), 14,000rpm (yellow) and 20,000rpm (red). Due to aggregation during the long equilibrium centrifugation times (16-20h at 20oC) it was not possible to bring the system to a stable equilibrium.

As an alternative approach which required the protein to spend less time at room temperature, a series of concentration AUC experiments were performed using velocity

AUC. The drawback of this method for measuring an equilibrium reaction is that the sedimentation boundaries separate for monomer and dimer which can lead to changes in species distribution over the course of the experiment and therefore reduced accuracy. As with equilibrium AUC buffer salt concentration was kept close to physiological (10mM Tris, 150mM NaCl, pH7.4) The c(S) plots shown in Figure 3.9A show distinct monomer and dimer peaks and an increase in the proportion of dimer with increasing sample concentration. There is also a significant amount of higher order species formation as well some loss of mass conservation, probably due to aggregation. This meant that is was not possible to fit these data to a standard model.

Figure 3.9: Kd analysis using velocity AUC. (A) Sedimentation velocity AUC of chordin ∆N at a range of concentrations showing continuous size distribution (C(s)) against Sapp. At higher concentrations the proportion of dimer increases compared to the proportion of monomer. The monomer sediments with S values of ~5S and the dimer at ~7S. (B) Curve fitted to a plot of differential sedimentation coefficient distribution (g*s) against concentration where g*s is the distribution function of apparent sedimentation values.

To obtain a Kd value, the differential sedimentation coefficient distribution for each dataset was compared to concentration. From this a non-linear regression curve was produced shown in Figure 3.9B. At half maximum g*s these data indicate a Kd of approximately 3.3µM. At this level of affinity it is likely that, while some dimer formation will occur under physiological conditions, chordin will be primarily monomer in tissue systems. Analysis of a construct of the CHRD1-4 domains using SAXS, MALS and AUC by Dr Richard Tunnicliffe later confirmed that the CHRD domains do not dimerize and therefore the self affinity of chordin and its fragments is a property of the vWC domains178, 187.

3.8 Sedimentation Velocity Analytical Ultracentrifugation of Chordin ∆N

Initially the sedimentation behaviour of chordin was analysed using chordin ∆N as a model. This was not practical for the other chordin constructs because the contaminating cleavage fragments were so close in size and shape, that their sedimentation boundaries did not separate sufficiently. High salt (10mM Tris-HCl, 500mM NaCl, pH7.4) was used in order to minimize the proportion of dimer. Figure 3.10 shows the sedimentation profile for chordin ∆N. The size, predicted using Sedfit, was 70kDa, which is smaller than the size calculated from the amino acid sequence (87kDa) and MALS data (93kDa). This may be because the monomer and dimer were separated fully using this technique. The hydrodynamic radius was calculated as 4.7nm (similar to 4.4nm calculated for full length chordin from MALS) and f/f0 was 1.56, indicating a protein which is reasonably compact but not globular. This is consistent with a linear protein which bends around on itself. The dimer peak had an estimated MW of 140kDa which is lower than expected but unlikely to be accurate from such a low absorbance. The sedimentation coefficient was calculated to 4.52S (Svedberg) which is within the range of reasonable values for a protein of this size.

Figure 3.10: (A) Analytical ultracentrifugation profile of chordin ∆N showing C(s) against Sapp. These data show that the protein is primarily monomeric but with a significant proportion of dimer. Monomer peak: S20W = 4.52. Mw = 70kDa, f/fo = 1.56 and Rh = 4.7nm.

3.9 Discussion

Mutant full length chordin and chordin constructs mimicking the tolloid produced cleavage products of chordin were cloned and expressed in HEK 293-EBNA cells. The identities of the constructs were confirmed using Mass Spectrometry which also confirmed that smaller

87 products were fragments co-purifying with chordin rather than impurities. Removal of the polyhistidine tag for experiments where this is desirable such as surface plasmon resonance was confirmed using Western blotting with an anti-histidine antibody. PNGase-F digestion of chordin demonstrates for the first time that chordin is a glycoprotein. The candidate N-linked glycosylation sites are found in the CHRD domains and the sugars may play a role in solubilising chordin, which is prone to aggregation even with the sugars present. Glycosylation is an emerging area of research in terms of functional roles and it is possible that the sugars may have an effect on chordin function.

The secondary structure of chordin was predicted using circular dichroism. SRCD data for full length chordin gave an estimate of 9.7% α-helix and 35% β-sheet. It was possible from what is known about chordin vWC domains that the secondary structure might have been all β, so a high proportion of β-sheet was expected. It is probable that the α-helical secondary structure resides in the structurally uncharacterized CHRD domains, as the crystal structure of vWC1 from crossveinless23 (and other related vWC homology domains188, 189) shows that they are β-sheet containing domains. For that reason it was predicted that when a vWC domain is lost (vWC1 and vWC4 depending on which site tolloid cleaves) that the proportion of β-sheet in the resulting structure would be lower and the proportion of α-sheet correspondingly higher. To put this into context Table 3.5 compares the percentage of β- sheet predicted from the full length CD data without the regions before and after each cleavage site, if it is assumed that approximately half the structure removed following cleavage contains β-sheet and the other half turns and unordered flexible loops (proportions estimated based on domain boundaries and sequence shown in Appendix III).

∆C Predicted ∆C Actual (%) ∆N Predicted ∆N Actual (%) (%) (%) β-sheet 32 19 32 27 α-helix 11 16 11 26

Table 3.5: Comparison of the predicted difference in proportion of secondary structure components following tolloid cleavage compared to the experimentally derived secondary structure predictions for the chordin fragments.

Even assuming that the entire sequence removed by cleavage is composed of β-sheet the changes in proportions are still too high to be the result of a direct proportional loss alone. The fragment CD data are noisy (particularly ∆N) and require repeating with SRCD. However a potential difference in secondary structure raises two possibilities; that cleavage of chordin results in conformational change at the secondary structure level which is likely due to known functional differences between full length chordin and the fragments, or that

88 the constructs are folding differently as an artefact of independent expression. It would be interesting to perform tolloid cleavage of full length chordin during SRCD data collection to analyse conformational arrangement.

To ascertain if ∆C and ∆N were functional in terms of BMP binding and to show whether cleavage reduces BMP binding ability SPR was used to compare binding kinetics. The curves produced for kinetic analysis would not fit to any standard model and the interaction could not be brought to equilibrium. This may be due to the formation of higher order oligomers of chordin once bound to BMP-2, a hypothesis supported by the unusually steep off-rate which could be expected if an aggregate of chordin molecules which had attached to each other one by one detached from BMP-2 as a single mass. While it is clear that all the constructs bind readily to chordin it is not possible to determine what proportion of this derives from self-affinity. However in 0.25M urea, which minimized this problem, a Kd of 27nM was determined for ∆N which is similar to the 37nM Kd obtained for full length chordin from previous studies.

These SPR results led to the conclusion that chordin may have some self affinity, and as a first step toward investigating this MALS was used to determine its oligomeric state. The mass of chordin was very close to the predicted size based on sequence analysis and in addition an Rg of 5.4nm and an Rh of 4.4nm were recorded which are useful values for comparison to structural analysis techniques such as small angle X-ray scattering (SAXS), AUC and TEM. However the cleavage fragments were all significantly larger than predicted, and larger than could be reasonably accounted for by N-linked glycosylation. This led to the possibility that chordin may form reversibly associating dimers.

To investigate this possibility equilibrium AUC was initially used, however chordin proved too unstable to last the duration of the experiment without aggregating. As an alternative the average sedimentation across a range of concentrations was calculated and fitted. This analysis gave a Kd of 3.3µM which suggests that chordin monomer is likely to be the dominant form in vivo however at peak chordin concentrations in the developing embryo a proportion of chordin may form dimer under physiological conditions190. It is not clear whether full length chordin forms dimers as readily as the fragments given that it did not show the same increased mass from the MALS data, however without a sample entirely free from ∆N cleavage fragment it is not possible to show that full length chordin is monomeric because a proportion of the sample solution will contain chordin ∆N.

Analytical ultracentrifugation was used to calculate hydrodynamic properties for chordin ∆N giving a predicted Rh of 4.7nm, compared to 4.4nm for full length chordin determined by MALS. These are reasonably similar values, though it may be that conformational change in chordin following cleavage causes ∆N to have a higher hydrodynamic radius. Taken together with the apparent change in secondary structure from CD, it was decided to

89 investigate this further using in solution structural analysis discussed in the following chapter.

4. Nanostructure of Chordin and Chordin Fragments

4.1 Summary

Following on from the production and initial characterization of chordin and its cleavage fragments, small angle x-ray scattering (SAXS), analytical ultracentrifugation (AUC) and transmission electron microscopy (TEM) were used to find the structure of chordin. Existing structural data for chordin are limited to the crystal structure of vWC1 of a related protein, crossveinless-2. However, in the literature it is a common assumption based on the binding of BMPs to vWC domains that chordin is a U-shaped protein which binds to a single BMP dimer through two interfaces, one at the N- and one at the C-terminus of chordin. In this chapter the low resolution structure of chordin is shown which provides clues as to its binding behaviour with BMPs.

The cleavage fragments produced by tolloid metalloproteinases were investigated. These consist of chordin ∆C which terminates before vWC4, ∆N which lacks the vWC1 domain and the terminal fragment which consists only of the four CHRD domains and vWC2-3. All these fragments were found to form dimers at the high concentrations required for SAXS which was explained on the basis that they all contain vWC homology domains which are thought to have self-affinity8, 31. Values were obtained for the size, shape and radius of gyration for all these constructs using SAXS. Hydrodynamic radius, sedimentation coefficient and the frictional ratio was also shown for monomeric ∆N using AUC. These methods were then compared using the programs Hydropro and Hydromic which can be used to calculate theoretical hydrodynamic properties from SAXS and TEM data respectively.

4.2 Small Angle X-ray Scattering from Full Length Chordin

Small angle X-ray scattering (SAXS) data were collected for full length chordin in a range of concentrations from 1mg ml-1 to 5 mg ml-1 and the buffer (10mM Tris, 150mM NaCl, pH7.4) subtracted data merged using PRIMUS. These data describe the relationship between the scattering intensity (LogI) and the scattering wave vector (q). An X-ray wave with vector ki encounters a protein in solution (the scattering centre) and the wave is scattered at a new vector of ks. The scattering wave vector (q) is given as ks = ki + q. The magnitude may be expressed as q=4πsinƟ/λ where 2Ɵ is the scattering angle, and λ is the wavelength. Scattering of the X-ray is determined by the properties of the protein such as size, shape and hydration. Analysis of the small angle scattering behaviour of the protein in solution can therefore be used to experimentally determine physical properties of the protein including the radius of gyration (Rg), molecular mass and the approximate low resolution shape.

4.2.1 Gnom Fit to the Experimental Data from SAXS

In order to enrich the sample for full length chordin with minimal ∆N contamination, mutant full length chordin was used. This coupled with L-Arginine tolloid inhibition during harvesting resulted in a sample which was almost entirely full length. The transform program Gnom was then used to read the one dimensional scattering curve and create an ideal fit to raw experimental data. These data are shown in the SAXS profile in Figure 4.1A giving LogI as a function of q. The signal-to-noise is high at higher resolution, and a cutoff point was chosen at 0.37 A-1. Gnom calculates a distribution function P(r) as a function of distance using the fitted data, shown in Figure 4.1B. This gives a maximum dimension (Dmax) of 18 nm and a double peak which could be indicative either of dimer formation or two distinct “lobes” of the protein.

Figure 4.1: Full length chordin Gnom fit and P(r) plot. (A) SAXS profile for full length mutant chordin showing merged data from 1 mg ml-1 and 5 mg ml-1 samples and the Gnom fit to experimental data. (B) Distribution of intramolecular distances; P(r) plot calculated using Gnom, indicating a 180Å (18nm) Dmax.

Automated analysis at the beamline predicts molecular mass using BSA as a reference and was found to be in the range of 90-124kDa, rising at higher concentrations indicating the formation of dimer. At the highest concentration used for data collection (6mg ml-1) the estimated molecular mass was 194kDa which is close to double the molecular mass of monomeric chordin. As it was clear that most of the signal from this sample was coming from dimer which could not easily be cross compared to our other data this concentration was not processed further. Estimated Rg and Dmax were also higher for the 5mg ml-1 sample (6.08nm and 21.3nm respectively) than for the 1mg ml-1 sample (5.5nm and 18.5nm). Increases in both these values would point to a monomer-dimer equilibrium between chordin molecules tending more toward the dimer in the concentrated sample.

4.2.2 Guinier Analysis and Kratky Plots

The Guinier plot for SAXS data shown in Figure 4.2A is linear suggesting a near monodisperse, unaggregated sample. The Guinier approximation of Rg was 5.8nm from these combined which as expected is a value between the two values calculated by the automated software at the beamline. It is also similar to the 5.4nm Rg value derived from MALS data. The Kratky plot (Figure 4.2B) shows a prominent peak at a very low angle (0.3nm) indicating a well folded protein. There is a shoulder to this peak and three subsequent peaks. This is generally due to either dimer formation or a multidomain protein with flexible linkers. In the case of chordin it is likely to be a combination of the two; dimers have been shown in the previous chapter to form in the µM range but chordin also has long peptide linkers either side of the four CHRD domains linking them to the vWC domains which potentially allow for movement between domains.

Figure 4.2: Full length chordin Guinier and Kratky plots. (A) Guinier plot for the SAXS data collected for full length mutant chordin showing linear dependence. (B) Kratky plot showing a prominent peak at low angles indicating a folded protein.

4.2.3 Ab initio Modelling of Full Length Chordin

Ab initio models for full length chordin were generated using DAMMIN182, a modelling program which takes a ball of spheres or “dummy atoms” and over multiple iterations fits a model which best matches these raw data. This was repeated using GASBOR183, which is a similar reconstruction program but uses a chain of dummy atoms instead. Each model was repeated twenty times assuming P1 symmetry. Average Χ2 for these models are 0.67 ± 0.001 and 0.64 ± 0.003 from DAMMIN and GASBOR respectively. These structures were combined into single averaged models using DAMSEL, DAMAVER and DAMFILT184 to produce final models. Average normalized spatial discrepancy (NSD) was 0.966 ± 0.06 for the DAMMIN and 3.581 ±0.24 for the GASBOR model. These values indicate that DAMMIN

94 produces accurate models and GASBOR less so which may be due to the dummy atoms that are confined in GASBOR to a peptide chain compatible conformation which limits where the model can move them to obtain the best fit.

The DAMMIN model shown in Figure 4.3A is a bowl shape with an additional region of extra density. The overall dimensions for the model are 21.5 x 11 x 6nm. This is slightly longer than predicted and it is not clear if the additional density is the product of results skewed by the presence of contaminating dimer. The GASBOR model shown in Figure 4.3B is not an exact match to the DAMMIN model which would be expected given the very high NSD, however the overall proportions are roughly the same which supports the reliability of the DAMMIN model. Ab initio models were then generated using DAMMIN assuming P2 symmetry, shown in Figure 4.3C. For this model average Χ2 is 0.67 ± 0.004 and NSD is 1.089 ± 0.028.

Figure 4.3: Ab initio models of full length chordin generated from SAXS data shown in three orthogonal views. (A) DAMMIN model assuming P1 symmetry. (B) GASBOR model assuming P1 symmetry showing broadly similar proportions to the DAMMIN model. (C) DAMMIN model assuming P2 symmetry which is possible if the protein is dimeric.

The P1 and P2 symmetry models generated are very different but both are plausible. In the P1 model it appears that chordin is a bowl shape with a region of extra density which could

95 be domains joined by a linker while the P2 model would seem to indicate two relatively compact “c” shaped halves in an end to end conformation. It is also possible from the P2 model that two chordin monomers are locked in an S-shaped conformation. Molecular mass estimated from the automated beamline software coupled with the known monomer-dimer equilibrium in chordin means that some dimer is certainly present but it is not clear to what extent this is influencing the model. A P2 model using GASBOR produced a similar two “c” shapes but in a cis- orientation which is not dissimilar to the DAMMIN P1 model (data not shown).

4.2.4 SAXS Analysis using Low Concentration Chordin

To try to establish to what extent dimerization influences the model, the analysis was performed again using only data from a very low (0.5 mg ml-1) chordin concentration. Analysing SAXS data from a sample with proportionally less dimer will not produce a model entirely free from dimer influence. However, it should give an indication of how much impact the presence of dimer has and which features of the model are affected when it is reduced. This modelling must be done using a limited q-range as these data at higher q had a high signal to noise ratio due to the lower concentration. In this case, Rg was estimated by the automated beamline software at 5.7nm and Dmax at 17.8nm. Figure 4.4A-D shows the associated plots from this individual run. These data shown in the Gnom fit are noisy at higher resolutions. However the P(r) plot while showing the same peak at 5nm does not have the second distinct peak seen in Figure 4.4B meaning that the second peak probably corresponds to chordin dimer. The Guinier plot shows a linear dependence between LogI and q2- which shows that the sample is of reliable quality and not aggregated. Unlike the P(r) plots, the Kratky plots are similar in shape between glycosylated and deglycosylated Tsg. Interestingly the proportion of shoulder is significantly higher relative to the main peak. This could indicate that it reflects a flexible region rather than a dimer.

Figure 4.4: SAXS analysis of low concentration chordin. (A) SAXS profile for 0.5mg ml-1 full length mutant showing the Gnom fit to experimental data. (B) P(r) distance distribution plot calculated using Gnom, indicating a 17.8nm Dmax. (C) Guinier plot for SAXS data showing linear dependence. (D) Kratky plot showing a prominent peak at low angles. (E) DAMMIN generated ab initio model from ten repeats shown in three orthogonal views.

The ab initio model from these data generated using P2 symmetry DAMMIN from 10 repeats is shown in Figure 4.4E. NSD was 0.496 ± 0.019 and average χ2 was 1.276. It resembles a half-way model between the P1 and P2 symmetry models generated from the higher concentration data, though lacking the higher resolution detail. Importantly there is no significant reduction in either lobe of the protein which would indicate that they were not an artefact of dimer formation. From this it is possible to form the tentative assumption that the models generated are primarily representative of monomeric chordin and that rather than the predicted U-shape, chordin has a curled conformation.

4.3 Small Angle X-ray Scattering of Chordin ∆C, ∆N and the Terminal Fragment

4.3.1 Data Analysis

SAXS data were collected for the cleavage fragments of chordin ∆N, ∆C and the terminal fragment for comparison to full length chordin. Since all three constructs contain vWC homology domains, all three are expected to form dimers like full length chordin. The terminal fragment was run at a 1mg ml-1 concentration while data for the larger fragments were merged from samples of 1-6mg ml-1. These data were then processed as described for full length chordin. The Gnom fit to the experimental data are shown in Figure 4.5A. Though still usable, these data for the terminal fragment is of lower quality than the other SAXS data in this investigation. Although size exclusion was used to remove aggregate prior to beam exposure either the need to digest at 37oC over a period of two days in order to remove all the chordin ∆C or inherent instability in the terminal fragment causes it to aggregate rapidly. These data for the larger fragments fit well and are of high quality.

Figure 4.5: Gnom fits, P(r) plots and Guinier plots for the chordin fragments. (A) SAXS data plots for the cleavage fragments of chordin showing Gnom fit to the experimental data. Owing to high expression of the larger fragments, very high concentration samples could be used. (B) P(r) plots for ∆C, ∆N and the terminal fragment predicting a Dmax of 18.5nm, 22.5nm and 19.5nm respectively. (C) Guinier plots showing data with linear dependence.

The P(r) plots shown in Figure 4.5B are interesting in that they do not predict Dmax in the order of size one would expect (∆C > ∆N > terminal fragment). Instead Dmax of ∆C is

18.5nm which is very similar to that obtained for full length chordin while the terminal fragment is longer (19.5nm) despite being of smaller mass and ∆N is substantially longer (22.5nm). This means that either the ∆N and terminal fragment data are more representative of the dimer than full length and ∆C or that the absence of vWC4 results in a more linear conformation (i.e. the proteins are smaller in mass but more extended). A high, prominent second peak in full length chordin at ~7nm was associated with a more dimeric mixture. While all three constructs have this double peak it is more pronounced in ∆N indicating that the difference is due to proportionally more dimer. Rg was calculated to be 5.3nm, 6.4nm and 5.6nm for ∆C, ∆N and the terminal fragment respectively, and again the higher Rg of the smaller fragments relative to ∆C would indicate influence of oligomers in the solution.

4.3.2 Ab Initio Modelling of the Chordin Fragments

The DAMMIN generated P1 ab initio models are shown n Figure 4.6. They are reasonably similar to the GASBOR generated models and in P2 symmetry the shape suggests two “c” shaped monomers in a cis-conformation (data not shown). The chordin ∆C data were produced by merging data from samples ranging in concentration from 1-6mg ml-1. From this we were able to generate a primarily monomeric ab initio model shown in Figure 4.6A. The shape generated for chordin ∆N and the terminal fragment are more skewed by the presence of dimer than the full length and ∆C models accounting for the elongation of the models. Their shape indicates that chordin dimers form in an end to end cis-conformation shown in Figure 4.6B and C. NSD values were 0.91 ± 0.038 for ∆C, 0.873 ± 0.137 for ∆N and 0.661 ± 0.022 for the terminal fragment.

Figure 4.6: Chordin fragment Ab initio models generated using DAMMIN shown in three orthogonal views for (A) chordin ∆C, (B) chordin ∆N and (C) the terminal fragment. (D) Superimposed chordin ∆C and full length mutant chordin showing an almost identical structure with a small region of extra density in the full length.

100

Figure 4.6D shows the full length and chordin ∆C models superimposed. They are almost identical, bar a small amount of extra density in the full length chordin as would be predicted. To determine whether the extra density could be accounted for by the domain absent in chordin ∆C, vWC4, the solved structure for the WC1 domain of crossveinless-2 was docked into the structure. This is shown in Figure 4.7A and the domain fits neatly into the additional density. It is therefore likely that the structures are monomeric and that the vWC2-4 region loops back toward the CHRD domains rather than the U-shaped chordin structure suggested in the literature. Figure 4.7B shows the same model aligned with an ab initio model of the CHRD domains alone (produced by Dr Richard Tunnicliffe) showing them fitting with room for vWC1 at the opposite terminal.

Figure 4.7: (A) Superimposed ab initio models of chordin ∆C (yellow) and full length mutant chordin (green) with vWC1 from crossveinless-2 showing that the region of extra density in full length chordin is the correct size and shape to fit a vWC domain. (B) The same model superimposed with an ab initio model of the CHRD domains alone (pink) and a second vWC domain to represent vWC1.

4.4 Transmission Electron Microscopy Models of Chordin Constructs

4.4.1 TEM Class Averages from Single Particle Analysis

To support the SAXS models and obtain higher resolution data, full length chordin and the larger fragments were looked at using transmission electron microscopy (TEM). Samples negatively stained with 5% uranyl acetate were absorbed to carbon coated grids at 10µg ml- 1. Images were recorded at 30,000X (2.1Å/pixel) on a Tecnai Biotwin at 120kV. The images

101 were taken using a Gatan Orius CCD camera with a 0.5-1.5µm defocus range. Owing to the extremely high affinity of ∆N to the grids, even short adhesion times at very low concentrations produced crowded grids, so images were collected in the least dense regions of the best grids to avoid particle overlap. Grids were produced using a range of buffer conditions of which 10mM Tris-HCl, 150mM NaCl, pH7.4 was found to be the most effective. Data were processed using the program EMAN2180. Using a combination of manual and semi-automated swarm picking, 3300 particles were picked for full length chordin, 3800 for chordin ∆C and 2500 for chordin ∆N, from 25 images.

CTF was corrected and 32 class averages were generated for full length chordin, ∆C, and ∆N. To avoid biasing the dataset from preferential positions of the protein adhering to the grid, 15 unique views were selected from these averages, shown below in Figure 4.8. Particle sizes appear broadly similar to the models generated from SAXS, although the conformation appears less elongated. The longest distances were ~16nm for full length chordin, ~17nm ∆C and ~14nm for ∆N. These are shorter distances than those determined using SAXS and indicate that either chordin is adhering to the grid in a way which does not present lengthways views, or that chordin adheres to the grid in a different conformation.

Figure 4.8: Selected class averages from single particle analysis of TEM data. (A) Full Length Chordin; from 32 class averages generated using Eman2 from 3300 particles selected from 25 images. (B) Chordin ∆C from 32 class averages of 3800 particles from 25 images and Chordin ∆N from 32 class averages of 2500 particles from 25 images. Box size = 126 pixels at 2.15Å/pixel.

102

4.4.2 3D Reconstruction of the Chordin Constructs From TEM Data

From the selected class averages ten initial 3D models were generated from eight iterations assuming C1 symmetry. These models were reasonably consistent with each other showing bowl shapes for all three constructs. A single representative model was chosen for each construct to be used as the starting model for 3D reconstruction. Particles which had been CTF corrected by phase flipping were used for a total of 8 refinement iterations. Post- processing, models were subject to lowpass Gaussian filtering. All the models produced a curved two-lobed shape, with full length chordin appearing slightly larger. Interestingly the models, although similar, do not directly superimpose, suggesting conformational change following cleavage.

Figure 4.9: EM models from single particle reconstruction TEM for (A) Full length chordin (B) chordin ∆C and (C) ∆N shown in three orthogonal views.

The models produced are similar to each other but in a different conformation to the SAXS models. Like the SAXS models, they are composed of two lobes with a linker region between them. However these models are closer to the U-shape predicted in the literature, whereas the SAXS models are more linear. This may be due to the way chordin adheres to the grid, with flexibility between the two domains resulting in a different conformation when

103 immobilized. Alternatively it may be that the influence of dimerization in the SAXS experiments leads to an exaggerated Dmax and the ab initio modelling programs stretch the model to try to match this.

4.4.3 Fourier Shell Correlation Curve Interpretation

Fourier shell correlation (FSC) curves are plotted to assess resolution and the noise levels present in final model reconstruction. It represents the FSC between two maps generated by separating the odd and even halves of these data and reconstructing them separately. The plots for full length chordin and the two fragments are shown in Figure 4.10 where the thin lines show the FSC between each iteration and the previous iteration. As the reconstructions converge the alignment parameters remain unchanged between iterations (although there is always some noise). In each model the lines converge suggesting that sufficient iterations have been performed as the model is no longer changing significantly with each iteration. The thick line is the FSC curve which using a cut-off value of 0.5 shows a resolution of 30Å for full length chordin, 32Å for chordin ∆C and 33Å for chordin ∆N.

Figure 4.10: Fourier Shell Correlation Curves for (A) Full length chordin, (B) Chordin ∆C and (C) ∆N showing convergence of the iterations and Even/Odd FSC (thick line). At an FSC cutoff value of 0.5 resolution is 30Å, 32Å and 33Å respectively.

104

4.5 Ab initio Model Docking and Structure Cross-Validation

4.5.1 Bead Modelling of Chordin SAXS data using Hydropro

The SAXS structures generated were further analysed using Hydropro186, a program which estimates the hydrodynamic properties of the protein structure from the ab initio PDB files generated and known properties of the protein including the molecular mass, partial specific volume and solution properties. For this modelling all solutions are assumed to be in water at 25oC. These calculations are useful for cross validating data as the values obtained, such as sedimentation coefficient and Rg, can be compared to experimentally derived values. Table 4.1 shows the values predicted from Hydropro alongside the experimental Rg from SAXS. The experimental and calculated values are consistent however the predicted sedimentation coefficient of chordin ∆N (3.5S) is significantly lower than the value obtained from analytical ultracentrifugation (4.52S) which is itself similar to the values for full length chordin and ∆C.

Predicting ∆N sedimentation from the ab initio SAXS model using Hydropro is inherently problematic because the influence of dimer results in comparing velocity AUC data on monomer to SAXS data on monomer/dimer. Essentially the calculation uses the long, elongated structure of the ab initio model but with a molecular mass of a monomer and this combination of low mass and increased friction leads to an underestimate of sedimentation. Using the molecular mass of the dimer for ∆N in the calculation gives an Rg of 6.3nm but an

S20W estimate of 7S.

Rg Experimental Rg Hydropro S20W S20W Hydropro (nm) (nm) Experimental (S) (S) Chordin 5.8 5.5 N/A 4.79 Chordin ∆C 5.3 5.4 N/A 4.26 Chordin ∆N 6.4 6.3 4.52 3.5 Terminal 5.6 5.8 N/A 3.08 Fragment

Table 4.1: Comparison of the hydrodynamic properties of chordin and its fragments from

Hydropro calculations and experimentally derived data showing low predicted S20W values from the ab initio models influenced by dimer in the SAXS experiment.

105

The AUC values obtained for monomeric ∆N would be expected to be broadly similar to the Hydropro calculated values from the monomeric ab initio models for full length chordin and ∆C. This is indeed the case, with full length chordin predicted to sediment marginally faster as would be expected because it is larger, and ∆C with a lower sedimentation coefficient than the experimental value for ∆N, possibly due to conformational differences between the two similarly sized fragments. The ab initio SAXS model of the terminal fragment, like ∆N, is dimeric and so has a lower than expected Hydropro prediction of sedimentation coefficient.

4.5.2 Prediction of Hydrodynamic Properties from TEM Models Using Hydromic

To further cross-compare structural data obtained from the techniques used in this chapter, Hydromic185 was used to predict hydrodynamic properties from the TEM models. Hydromic constructs a bead model from TEM data (shown in Figure 4.11Ai-iii). Then, like Hydropro, it uses bead-model calculations to generate predictions from this model, such as the molecular mass and partial specific volume and hydrodynamic properties. Figure 4.11 shows the P(r) plots and scattering function calculated from the models. The theoretical scattering curves (Figure 4.11 B) are similar in shape to the SAXS scattering curves. The P(r) plots (Figure 4.11 Ci-iii) are different to the SAXS P(r) plots in several ways. The maximum distance is lower because of the less extended conformation and instead of one large peak with a shoulder, two very distinct peaks are present each one representing one of the two lobes of the structure.

106

Figure 4.11: Hydromic output data from TEM ab initio models. Ab initio structural data are converted to a model composed of beads which can then be used for bead modelling calculations. (Ai) full length chordin (Aii) chordin ∆C (Aiii) chordin ∆N. (B) Theoretical scattering function, P(h), of the three models. (C) Distance distribution plots for (i) full length chordin (ii) chordin ∆C (iii) chordin ∆N.

The values calculated for the theoretical hydrodynamic properties of the models are summarized in Table 4.2. Dmax and Rg are slightly smaller than those obtained from SAXS owing to the conformational differences between the two models. The Rh, Rg and S20W are, however, similar to those obtained experimentally through MALS and AUC. This lends support to the idea that the elongation of the SAXS models are due to dimer in the sample. The Hydromic, Hydropro and experimental values are shown together in summary Table 4.3.

Full Length Chordin ∆C Chordin ∆N Rh (nm) 5 4.9 4.77 Rg (nm) 4.8 5.5 4.86 Dmax (nm) 14.6 16.8 16.3

S20W (S) 5 4.68 4.44

Table 4.2: Summary of Hydromic predictions for hydrodynamic properties of the chordin constructs; hydrodynamic radius (Rh), Radius of gyration (Rg), maximum length (Dmax) and sedimentation coefficient (S20W).

107

4.5.3 Fitting of SAXS and TEM Models of Full Length Chordin

Although similar in size, the SAXS and electron microscopy models for chordin are not directly superimposable. Very similar conformations using both methods have been obtained independently for the three different constructs. It is probable that the region between the fourth CHRD domain and the second vWC domain is a flexible linker. This flexible linker may explain variations in conformation and it is possible that the most abundant chordin conformation on the grid is different to that in solution. However, it may also be that the higher proportion of dimer in the more concentrated SAXS sample biases the model toward a more elongated conformation.

Figure 4.12: EM reconstruction of full length chordin superimposed with the SAXS ab initio model shown in three orthogonal views

4.6 Discussion

This chapter presents the first structural data for the extracellular BMP antagonist chordin, summarized in Table 4.3. Using SAXS models, we show that unbound chordin has an elongated structure. Previously it was predicted that chordin has a hairpin shape to enable co-operative binding between the vWC domains at each terminal and BMP. The TEM structures of chordin were arc-shaped which is compatible with the co-operative binding mechanism predicted in the introduction, but different to the SAXS model. This is probably due to flexibility between the N-terminal vWC1-CHRD4 lobe and the vWC2-4 lobe. It may be that this flexibility results in alternate conformations with a more linear alignment in solution than the conformation on the grid.

108

Chordin ∆C ∆N SAXS (experimental) Dmax (nm) 18 18.5 22.5 Rg (nm) 5.8 5.3 6.4 SAXS (Hydropro) Rg (nm) 5.5 5.4 6.3 S20W (S) 4.79 4.26 3.5 TEM (Hydromic) Dmax (nm) 14.6 16.8 16.2 Rg (nm) 4.8 5.5 4.86 Rh (nm) 5 4.9 4.77 S20W (S) 5 4.68 4.44 AUC (A) / MALS (M) S20W (S) - - 4.52 (A) Rh (nm) 4.4 (M) - 4.7 (A) Rg (nm) 5.4 (M) - - f/fo - - 1.56 (A)

Table 4.3: Summary of the predicted and experimentally determined hydrodynamic properties of chordin and its fragments derived from SAXS, TEM, MALS and AUC.

From this it can be predicted that binding of full length chordin to BMP-2 might lock the C- terminus of the protein into a more rigid position. Previously it has been unclear which of the C-terminal vWC domains bind to BMPs, because vWC3 and 4 have strong affinity for different types of BMP. It is possible that flexibility in the C-terminal region allows for alternative co-operative binding to BMPs. For example when binding to BMP-2 the conformation could be aligned to allow access to vWC3 at the same time as vWC1, but for BMP-4 conformational change could allow co-operative binding with vWC4.

109

5. Structural and Biomolecular Analysis of Twisted Gastrulation and its Interaction with Chordin

110

5.1 Summary

One of the primary aims of this study was to characterize the chordin regulator Tsg in order to gain insight into its interaction with chordin. To facilitate this, recombinant human Tsg with a polyhistidine tag was cloned and expressed in HEK cells. Although this expression system produces relatively low yields it has the advantage of featuring human glycosylation patterns which have been shown to be relevant in BMP binding activity46. In addition, Tsg has extensive disulfide bonding for a protein of its size which would be challenging to reproduce accurately in a bacterial expression system. The presence of secondary structure, glycosylation and bioactivity of the construct were verified using CD, PNGase-F digestion and SPR respectively. This confirms the construct produced as an accurate replica of human Tsg in terms of both structure and function. The biomolecular analysis techniques applied to chordin in the previous chapters were then used to investigate the structural properties of Tsg.

Biomolecular analysis of Tsg using SAXS, MALS and AUC were used to determine many of the physical characteristics of Tsg. These include size, shape, radius of gyration, hydrodynamic radius, sedimentation coefficient and frictional ratio. This is both interesting because it is the first study of the structure of Tsg and also it lays the groundwork for preparing the chordin-BMP-Tsg ternary complex. These three techniques produced consistent values, cross-validating each other. These data were then further analysed using bead modelling to combine the results from the different techniques into a single working model. Interaction of Tsg with chordin was attempted using fluorescence resonance energy transfer (FRET) and a chordin-Tsg complex was produced to model the complex using transmission electron microscopy (TEM) single particle analysis producing a number of interesting results.

5.2 Expression and Purification of Tsg

5.2.1 Cloning of Tsg

Tsg constructs, Tsg-his and Tsg-thr-his (with a thrombin cleavage site for tag removal) were produced from image clones using PCR as described in the materials and methods section. Analysis of the PCR reactions by agarose gel electrophoresis showed bands of the expected sizes (690 and 706 bp for Tsg-His and Tsg-Thr-His respectively) shown in Figure 5.1A. The bands were then purified from the gel and inserted into the mammalian expression vector pCEP-Pu AC7 and amplified in an E.coli host. To check the size of the inserts, the plasmids isolated following transfection were digested with Not1 and Xho1 which cut restriction sites on either side of the insert. This digestion yielded fragments of the expected size shown in Figure 5.1B, indicating that the constructs were successfully ligated into the vector. The presence of the inserts was further confirmed using DNA sequencing. Colonies containing

111 vectors free from any PCR induced copy errors in the Tsg sequence were selected and the plasmids were used for transfection into HEK 293-EBNA cells.

5.2.2 Tsg Expression and Purification

Following transformation of the expression vector into HEK 293-EBNA cells, media from the cell culture was tested by Western Blot for the presence of Tsg using a mouse anti- polyhistidine tag antibody. This showed single bands of secreted protein with a histidine tag at ~30kDa, which is slightly larger than the sequence predicted size of 25kDa, shown in Figure 5.1C. Tsg was then purified using the histidine tag by nickel affinity purification shown in Figure 5.1D. When required, deglycosylated Tsg-Thr-His was treated with thrombin to remove the His-tag and the digested protein was then passed through a nickel affinity column a second time to remove any undigested protein. Removal of the tag was verified using Western blotting. Tag removal under native conditions was only possible using deglycosylated Tsg as presumably in its native form it contains sugars which block access to the cleavage site. At higher concentrations of Tsg a second band of >35kDa can be seen. This is the alternative glycosylation pattern of Tsg documented in previous studies46. Densitometry analysis of gels from three separate purifications gave an average ratio of 7:1 major to minor form.

Tsg was further purified by size exclusion chromatography at 0.5ml min-1 using a superdex 200 column to remove large protein aggregates and trace contaminants of different sizes that co-purified with Tsg or bound to the nickel column. The protein was continually monitored by UV absorbance at 280nm as it eluted from the column. This separated it from aggregated protein (found in the void volume at 8ml). Deglycosylated Tsg elutes from the column later than native Tsg indicating a reduction in size. In addition the process of deglycosylation did not result in an increase in the proportion of void volume to eluted protein indicating that this form is stable. Figure 5.1D shows purified Tsg following size exclusion chromatography showing a very pure sample which is required for many of the techniques used in this chapter.

112

Figure 5.1: Tsg purification. Agarose gel separation of DNA bands showing; (A) PCR products from the amplification of Tsg-His and Tsg-Thr-His from an image clone and (B) digestion of Tsg constructs from the pCEP-Pu AC7 vector. (C) Western blot of Tsg construct expression media samples using mouse anti-polyhistidine tag primary antibody. (D) Coomassie stained SDS-PAGE gel showing nickel affinity purified Tsg constructs with band sizes of ~30kDa (asterisk) and ~35kDa (arrow) for the more heavily glycosylated form.

5.3 The Tsg Constructs are Glycosylated, Folded and Able to Bind to Chordin

5.3.1 Circular Dichroism Analysis of Tsg Secondary Structure

In order to establish the presence of secondary structure in purified Tsg thereby confirming that the secreted peptide is folded, 0.4 mg ml-1 Tsg was analysed by CD in 10mM Tris-HCl, 150mM NaCl, pH7.4. Sensitivity of 100mdeg scanned at a rate of 20nm min-1 was used. A good trace with relatively little noise was obtained from averaging 20 buffer subtracted runs. The double dip observed between 205 and 225 nm is characteristic of a protein with mixed secondary structure, shown in Figure 5.2Ai. However the deeper dip toward the lower wavelengths in that range is indicative of β-strand or random coil dominance in the structure. Synchrotron radiation circular dichroism (SRCD) was also used because of its higher signal to noise ratio using 1mg ml-1 Tsg and sensitivity of 100 mdeg scanned at 60 nm min-1. The trace produced shows that the signal to noise is improved however the increased susceptibility of SRCD to salt interference at lower wavelengths meant that even at low salt (25µM), data below 195nm were unusable. In addition, after four repeats the shape of the

113 traces changes significantly (data not shown) indicating radiation damage. The shape of this trace as an average of four buffer subtracted repeats was similar to the standard CD trace indicating a strand based secondary structure see Table 5.1.

Figure 5.2: Tsg constructs are folded, glycosylated and able to bind chordin. (A) CD profiles of Tsg showing mdeg vs wavelength scanned (nm) using (i) standard CD and (ii) SRCD. The profiles show a dip between 205 and 225nm consistent with a folded protein, with the characteristic double dip with a deeper trough at 208nm typical of proteins containing mixed α/β secondary structure. (B) Binding of 0-1500nM Tsg to immobilized chordin using surface plasmon resonance showing the fit to a 1:1 interaction model (Langmuir). (C) Coomassie stained SDS-PAGE gel showing the results of PNGase-F digestion. The first lane shows PNGase-F enzyme alone, the second lane shows undigested Tsg and the third lane shows PNGase-F digested Tsg. Following digestion the band appears to be the size expected from the amino acid sequence (25kDa) and is sharper indicating loss of sugars.

Protein secondary structure proportions were analysed using Contin, Selcon3, CDSSTR and k2D provided by the Dichroweb server176. The results from each algorithm are shown in Table 5.1 with normalized root mean square deviation (NRMSD) and the average of each algorithm excluding K2d because of high NRMSD. The results and predicted reliability are

114 very similar for the SRCD data despite the 190-195nm region being unreadable due to buffer interference. The values obtained from the Contin, Selcon3 and CDSSTR algorithms are consistent and predict a protein consisting primarily of β-strand with some α-helix. This is interesting as the N-terminal domain of Tsg has limited sequence homology to the vWC domains of chordin family proteins which have been shown in crossveinless-2 to be composed of β-strands.

CD SRCD Helix Strand Turns Unordered NRMSD Helix Strand Turns Unordered NRMSD Contin 0.087 0.322 0.219 0.327 0.126 0.14 0.323 0.22 0.318 0.149 Selcon3 0.076 0.302 0.233 0.387 0.239 0.164 0.309 0.198 0.330 0.245 CDSST 0.11 0.33 0.24 0.32 0.057 0.13 0.31 0.25 0.3 0.053 R K2d 0.18 0.29 N/A 0.54 0.407 0.15 0.31 N/A 0.55 0.339 Mean 0.091 0.318 0.231 0.345 N/A 0.146 0.313 0.223 0.316 N/A

Table 5.1: Table summarizing analysis of CD and SRCD data for Tsg using the Contin, Selcon3, CDSSTR and K2d algorithms. The mean values show broadly similar predictions although these data from the SRCD have some structure classed as α-helix which in lab source CD data were classed as unordered. Mean values exclude K2d owing to high NRMSD values.

5.3.2 Surface Plasmon Resonance confirms binding between chordin and Tsg

In order to confirm the biological activity of the Tsg construct produced, surface plasmon resonance (SPR) was used to record binding between Tsg and chordin. Full length chordin for the solid phase was immobilised at a concentration of 100nM in SPR immobilization buffer (Appendix II) at a flow rate of 12µl min-1 to a total of 1000 response units on a CM5 sensor chip (GE Healthcare) in a Biacore 3000 (GE Healthcare). Analyte Tsg was in a concentration range of 0-1500nM in 10mM HEPES, 150mM NaCl, pH7.4 buffer (Appendix II) and passed over the chip at a rate of 30µl min-1 for 180 seconds as shown in Figure 5.2B. For each set of dilutions a blank was run prior to the protein and a referenced lane containing only buffer was run at the same time. Samples were double-referenced against a single lane of the chip activated and deactivated with no ligand immobilized. Lanes were regenerated using 1M NaCl and 50mM NaOH solution at a rate of 30µl min-1 for 180 seconds. Using a 1:1 binding model, the Kd was calculated to be 26nM showing a strong association with a χ2 value of 0.176. Previous estimates put Kd with immobilized chordin at 1200nM, however the same group found that the association was much stronger when Tsg was immobilized (Kd = 50nM) which is the same order of magnitude as our estimate8. The

115 weaker value from this group is likely to be an artefact of chordin immobilization blocking some binding surface and the low nM range more representative of in solution affinity.

5.3.3 Treatment of Tsg with PNGase-F Enables Deglycosylation of the Protein Under Non- Denaturing Conditions

Tsg ran at a higher molecular mass on SDS-PAGE gels than was predicted from the amino acid sequence and the bands also had a smeared appearance. As the sequence of Tsg has several potential N-linked glycosylation sites, it was predicted that this may be due to attachment of glycans. This was confirmed by digestion with PNGase-F (Figure 5.2C) with the resultant decrease in size confirming that human Tsg does carry N-linked glycosylations. It was later shown by another group that in mice N-linked glycosylation was the only type of glycosylation present46. Following deglycosylation, the dual bands seen in Tsg samples become a single band showing that the difference between size in the two isoforms expressed is due entirely to the presence of N-linked glycans.

As the sugars of Tsg have been shown to be critical in determining its BMP binding properties46, it was decided to investigate the structural differences between the glycosylated and deglycosylated forms. It was found that the glycan links on Tsg are accessible and the protein can be fully deglycosylated under non-denaturing conditions using a small volume of PNGase-F. After subsequent size exclusion purification the PNGase-F itself was no longer detectable by SDS-PAGE. The thrombin cleavage site of the Tsg construct is shielded by the sugars but accessible when the sugars are removed. By repeating SPR with deglycosylated Tsg with the histidine tag removed and obtaining similar binding data (data not shown), it was possible to rule out that the his-tag or sugars altered binding to chordin.

5.4 Small Angle X-Ray Scattering and Modelling of Tsg

5.4.1 Gnom Fit to the Experimental Data from SAXS

SAXS data were collected for Tsg (3mg ml-1) and deglycosylated Tsg (1mg ml-1) in 10mM Tris-HCl, 150mM NaCl at pH7.4. Scattering profiles detailing scattering intensity (LogI) varying with inverse wavelength of the scattering wave vector are shown in Figure 5.3A. The fit produced by Gnom is good and indicates that the models produced from it will be representative of the experimental data. Gnom was further used to plot the computed distribution function P(r) as a function of distance shown in Figure 5.3B giving a Dmax of 10.9nm for native and 8.9nm for deglycosylated Tsg. This difference may in part be due to an increased hydration radius in native Tsg as well as an actual difference in overall molecule length. In the deglycosylated P(r) distance distribution plot it is just possible to discern a shoulder on the peak which may indicate some flexibility between the two domains

116 of Tsg though this is not apparent in native Tsg. The Gnom output files were then used for further analysis and as the basis for ab inito modelling of Tsg.

Figure 5.3: Gnom fit and P(r) plots for Tsg. (A) SAXS profiles for native Tsg (i) and deglycosylated Tsg (ii) showing Gnom fit and the experimental data. (B) P(r) distance distribution plots calculated with Gnom indicating Dmax of 10.9nm and 8.9nm for native and deglycosylated Tsg respectively. In the deglycosylated plot a slight shoulder can be seen off of the main peak at ~4nm though this is absent in the native Tsg data.

Table 5.2 shows the results of automated analysis on the beamline following buffer subtraction. Using BSA as a reference molecular mass estimates were produced, giving values of 42kDa and 29kDa for native and deglycoslated Tsg respectively. Compared to an amino acid sequence based estimate of 25kDa this indicates that Tsg is monomeric in solution even at high concentrations. Predictably, the radius of gyration (Rg) which describes the mass distribution of the protein around its centre of gravity calculated from the slope of the plot, is higher with the N-linked glycosylations intact. At higher concentrations (5mg ml-1) Tsg begins to aggregate, so it was not possible to determine whether concentration has any

117 effect on Rg. However as 1mg ml-1 is already much more concentrated than Tsg is likely to be in any in vivo situations it is unlikely that there is a physiological Tsg dimer.

Sample Concentration Estimated MW Rg (nm) Dmax (nm) mg ml-1 (kDa) Tsg 1 - 3.11 10.9 Tsg 3 42 3.10 10.9 Deglycosylated 1 29 2.54 8.9 Tsg

Table 5.2 Automated data analysis at the Petra III Light Source. Rg and Dmax are consistent between 1-3mg ml-1 native Tsg and both values are considerably lower with the N-linked glycosylations removed from Tsg. Unless the glycans were located at the furthest length of the protein, which would be unlikely given that the N-linked glycosylation candidate sites are found mid-peptide sequence, the difference in Dmax may be the result of increased hydration attracted by the sugars.

5.4.2 Guinier Analysis Indicates a Monodisperse Sample

Guinier plots of the SAXS data (Figure 5.4) show a linear dependence at low q2 values indicating that the samples are monodisperse with little aggregation. This supports the reliability of these data and the values and models calculated from them. The plot is given as the square of q and the inverse of I(q) (scattering intensity as a function of q). Native Tsg shows a more linear Guinier plot at the lowest concentrations than deglycosylated Tsg. This was a somewhat surprising result since the two distinct glycosylation patterns of Tsg mean that Tsg should be more monodisperse without the sugars. This discrepancy is probably due to additional noise in the more dilute deglycosylated sample. In general though, data quality of both samples is reliable and the difference is small. Rg was calculated using Primus from the slope of this curve (Rg2/3) to give values of 2.99nm and 2.48nm for native and deglycosylated Tsg respectively which are similar to the automated beamline analysis.

118

Figure 5.4: Guinier plots of SAXS scattering data for native Tsg (A) and deglycosylated Tsg (B) from which Rg can be calculated to 2.99nm and 2.48nm respectively. The intensity is higher from native Tsg data owing to higher concentrations used and it is this, possibly coupled with inherent decreased stability in the deglycosylated sample, which accounts for the more linear Guinier plot for the former.

5.4.3 Kratky Plot Analysis

Kratky plots are given as the square of the scattering vector multiplied by the scattering intensity (q2I) as a function of the scattering vector. A double peak is usually indicative of either a dimer or a multidomain protein with a flexible hinge region. Sequence analysis in the literature suggests that Tsg is likely to have such a flexible region between its two domains, however the shape of the Kratky plot for native Tsg in Figure 5.5A shows no signs of this. Interestingly the plot for deglycosylated Tsg (Figure 5.5B) shows a shoulder after the main peak. As Tsg is clearly monomeric, this cannot be accounted for by dimer formation and may indicate that flexibility between the two domains increases in the absence of the glycosylations. This suggests that in addition to their role in BMP binding, the glycans may have a stabilizing effect on the structure of Tsg.

119

Figure 5.5: Kratky plots of SAXS scattering data for Tsg (A) and deglycosylated Tsg (B). In the plot for deglycosylated Tsg a slight shoulder is visible which may be indicative of multiple domains with a flexible linker.

5.4.4 Ab initio Modelling of Tsg and Comparison of Native and Glycosylated Structures

Ab initio models of Tsg in its native and deglycosylated forms were generated using DAMMIN182 and are shown in Figure 5.6. GASBOR183 which is a similar ab initio reconstruction program that uses a chain-like ensemble of dummy residues was used to cross validate the deglycosylated model and produced an almost identical structure. Each model was generated twenty times using P1 symmetry and the results pooled into a single model using DAMSEL, DAMAVER and DAMFILT184 to produce a final model. Average normalized spatial discrepancy (NSD) from DAMMIN was 0.52 ± 0.01 for native Tsg and 0.6 ±0.01 for deglycosylated Tsg. A value closer to zero indicates high consistency between models whereas higher values (over one) are obtained for highly dissimilar models. These values indicate a good degree of reliability in how the program interprets these data. Average Χ2 for these models is 1.086 ± 0.012 and 0.93 ± 0.01 for native and deglycosylated Tsg respectively.

120

Figure 5.6: Ab initio Tsg models. GASBOR generated (NSD= 1.288 ±0.05, χ2=0.95 ± 0.03) (A) and DAMMIN generated (NSD= 0.6 ±0.01, χ2=0.93 ± 0.01) (B) ab initio models of deglycosylated twisted gastrulation. (C) DAMMIN generated ab initio model of native Tsg shown in three orthogonal views (NSD= 0.52 ± 0.01, χ2=1.086 ± 0.012). (D) Native Tsg superimposed with the DAMMIN ab initio model generated for deglycosylated Tsg showing the region of density where the N-linked glycosylations are located.

The models show a diamond-shaped structure which is relatively compact. Comparing the glycosylated and deglycosylated protein in Figure 5.6D there is a clear region of additional density in the native Tsg which is not present in the deglycosylated model. This additional density can be accounted for by the glycans and it is located in the centre of the molecule which is where glycosylation is predicted they would be from the peptide sequence. Contrary

121 to previous suggestions, based on amino acid sequence homology comparisons, that the two domains of Tsg are linked by a flexible hinge region, the protein was found to be relatively compact. The removal of the sugars may allow for increased flexibility by lifting steric hindrance from the mid-region of the protein.

5.5 Biomolecular Characterization of Tsg Using MALS and AUC

5.5.1 Multi Angle Light Scattering Shows that Tsg is a Stable Monomer

Tsg has been previously predicted to bind to chordin in a 2:1 ratio. As a result free Tsg is frequently assumed to be a dimer, however the scattering data from SAXS appear to show that Tsg is monomeric even at high concentrations. As has been determined, the construct is folded and functional, so this is unlikely to be the result of misexpression. To confirm its monomeric state and to determine the exact mass of glycosylation, native and deglycosylated Tsg were compared using multi angle light scattering (MALS) after passing through a superdex 200 size exclusion column. The calculated size for both forms show that Tsg exists as a stable monomer in solution and trace oligomerization was not detected. The results, shown in Figure 5.7, show the elution volume plotted against differential refractive index and predicted molecular mass. The peak elution shifts by 1.2ml following deglycosylation indicating a marked decrease in size. This was confirmed by predicted MW of 30kDa for glycosylated and 25kDa for deglycosylated Tsg which is consistent with the results from SDS PAGE. The predicted MW for native Tsg is higher at earlier elution volumes within the peak whereas for deglycosylated Tsg it remains the same throughout. This could be due to the alternative glycosylation patterns for Tsg discussed in section 5.1.2 which result in a mixture containing a small proportion of higher MW glycoform.

122

Figure 5.7: MALS analysis of Tsg comparing the glycosylated (blue) and deglycosylated (red) constructs. The elution volume is plotted against differential refractive index (RI) and predicted MW. The deglycosylated protein elutes later and with a lower calculated MW than the native protein.

5.5.2 Velocity Analytical Ultracentrifugation of Native Tsg

To obtain hydrodynamic radius (Rh), sedimentation coefficient (S20W) and frictional ratio (f/f0) values for Tsg to enable bead modelling, native Tsg was analysed using velocity ultracentrifugation. A large peak is shown in Figure 5.8 with a MW of 33.2kDa and an S20W of -13 2.72S (Svedberg or 10 s). This MW estimate is higher than that obtained from MALS but lower than predicted from automated SAXS data analysis. However MALS is probably more accurate because it predicts size based on light scattering whereas in AUC bases size prediction on the changing sedimentation boundary which is more subject to conformation based bias and the SAXS estimates are broad based on comparison to BSA.

123

Figure 5.8: Analytical ultracentrifugation profile of native Tsg showing C(s) and Sapp. These data show a single species of 33.2kDa with an S20W of 2.72S and a trace of possible dimer -3 of 61.4kDa with an S20W of 4.15. Root mean square deviation (RMSD) = 4.9x10 showing a good data fit.

f/f0 is 1.62 (1 being a perfect sphere) which supports the moderately elongated model obtained from SAXS data analysis. The hydrodynamic radius was found to be 3.23nm which is similar to the radius of gyration (3.1nm) obtained from SAXS. Rh is the radius of an equivalent hard sphere sedimenting at the same rate as the protein while Rg describes the mass distribution of the protein around its centre of gravity. Due to the way the values are calculated an Rh which is higher than Rg indicates a more globular protein while Rg being higher indicates a linear shape. That the two are relatively similar suggests that the conformation is between the two; compact but slightly elongated. There is a trace second peak with a predicted MW of 61.4kDa (although with such weak absorbance the size prediction in this case is unreliable) which is likely to be either the higher MW glycoform or trace dimer.

5.5.3 Bead Modelling of Tsg Using Hydropro

SAXS ab initio models are composed of an array of beads. It is possible to make predictions about the in solution behaviour of a protein based on this and sequence derived values for the molecular mass and partial specific volume of a protein, and the viscosity and density of the solvent. These values were calculated using SEDNTERP179 assuming that all cysteine residues in Tsg are covalently linked. Hydropro186 predicted a value for Rg of 3.32nm for native Tsg and a sedimentation coefficient of 1.54S. While Rg was similar to the experimentally derived value, S20W was significantly lower than expected. This is likely to be because the molecular mass and partial specific volume derived from peptide sequence do not take into account additional density from the glycosylations. Using instead the MW

124 experimentally calculated from velocity AUC this value increases to 2.03S which is closer to the experimentally derived value of 2.72S.

5.6 Fluorescence Resonance Energy Transfer

Fluorescence resonance energy transfer (FRET) was attempted to demonstrate the binding ratio between Tsg and chordin by attaching a donor dye to one population of Tsg and an acceptor dye to a second sample. Figure 5.9 shows a summary of how the experiment was designed. In theory when the two samples were mixed there would be no FRET when the donor dye was excited, but if the two populations were mixed with chordin and more than one Tsg protein bound to chordin then some complexes would have both a donor and an acceptor tagged Tsg attached, bringing them into close enough proximity to produce FRET. In all FRET experiments Tsg was dialysed into 10mM HEPES buffer with 150mM NaCl at pH7.4 and separated from unbound dye using a PD-10 desalting column (GE LifeSciences).

Figure 5.9: Schematic diagram summarizing the FRET experiments. Two populations of Tsg, one tagged with a donor dye, the other with an acceptor dye are mixed and excited but do not come into close enough proximity for energy released by the donor to be used by the acceptor. In the presence of chordin, when two Tsg bind it acts as a scaffold bringing the two populations together resulting in FRET when the donor is excited.

5.6.1 Cysteine and Histidine Tag Linked Dyes

Initially a dye was screened that would use disulfide bonding to a free cysteine residue, however this was unsuccessful indicating that all 24 cysteine residues in Tsg are already

125 involved in disulphide bonding. The absorbance spectra of size exclusion elutions Tsg in 10mM HEPES buffer with 150mM NaCl at pH7.4 using a PD-10 desalting column (GE LifeSciences) were measured to determine if the dye had bound. High absorbance at ~280nm indicates the presence of protein. In this instance scans indicating the presence of eluted protein did not also indicate the presence of dye showing that the dye was not attached to the protein. Dye absorbance in later elutions without protein corresponding to unbound dye was observed. Next a nickel-NTA tagged dye was tried which would bind via the histidine tag. This was not the preferred method owing to the relatively weak binding of his-tag linking dyes, however it is unlikely that a single dye molecule at the C-terminus would interfere with any Tsg-chordin interaction. Binding of the NTA-tagged dye was unsuccessful, as it did not co-elute with the protein.

5.6.2 Amide-Linked Dyes

Dyes were then attached using amide linkage and Tsg was separated from unbound dye in 10mM HEPES buffer with 150mM NaCl at pH7.4 using a PD-10 desalting column (GE LifeSciences). In the initial trial this proved successful however there are 10 Lys residues in Tsg and there was concern that this could potentially sterically interfere with binding to chordin. In order to reduce this, dyes were attached at pH7 to try to encourage preferential binding to the N-terminal free amide group rather than all lyseine. Figure 5.10(A) and (B) show the absorbance spectra of size exclusion elutions for the donor dye (Alexa 488) and acceptor dye (Alexa 555) respectively. This time the dyes attached successfully as protein and dye co-eluted and there was a reduction in absorbance between these fractions and the free dye containing fractions collected later.

126

Figure 5.10: FRET analysis of Tsg binding. (A and B) Attachment of fluorescent dyes to Tsg populations. Dyes were bound and the samples subject to size exclusion chromatography. Elutions were collected in 1ml fractions and scanned using a spectrophotometer to detect the presence of protein and dye by absorption at different wavelengths. Earlier fractions showed protein co-eluting with dye, little absorbance in the middle fractions and absorbance only at wavelengths corresponding with the dye at higher elution volumes indicating successful attachment of the dyes to the protein. (C and D) Emission spectra of Tsg-chordin FRET experiment. (C) Donor and unlabelled Tsg with chordin (curve D), acceptor and unlabelled Tsg with chordin (curve A) and donor and acceptor Tsg together with chordin. (D) Combined curve A + curve D and with the curve for donor and acceptor Tsg together with chordin.

The donor dye was excited at 490nm, and the emission spectra measured. Owing to the overlap between the absorbing wavelengths of the donor and acceptor dyes shown in Figure 5.10 the acceptor dye has an emission spectrum in the absence of the donor dye. Were the two dyes to be in close enough proximity to achieve FRET, one would expect the emission peak of the donor dye to fall as energy released is absorbed by the acceptor dye, and the emission peak of the acceptor dye to rise as it is excited by the donor dye. Figure5.10C shows the initial experiment comparing three samples, one with chordin, donor-tagged Tsg

127 and acceptor-tagged Tsg; chordin, donor-tagged Tsg and untagged Tsg; chordin, acceptor- tagged Tsg and untagged Tsg. The sum of the second two spectra should the same curve shape as the first spectrum (shown in Figure 5.10D) indicating FRET did not occur. For FRET to occur, chordin would have to bind the Tsg in a 2:1 ratio or higher and bring the donor and acceptor dyes into close enough proximity to generate FRET. Figure 5.10D shows that this did not happen, there is a general reduction in absorbance due to decreased energy availability as the dyes absorb light. As a control chordin was also tagged with acceptor dye and incubated with donor tagged Tsg (data not shown). Since this mixture also failed to produce FRET it is likely that the presence of dye interferes with binding of Tsg to chordin.

5.7 Production of a Stable Tsg-Chordin Complex

Pure complexes of chordin fragments and Tsg were generated using the removable histidine tags and size exclusion chromatography with a Superdex 200 column (GE Healthcare). Chordin ∆C and ∆N were detagged using thrombin and passed multiple times through a nickel affinity column to separate any undigested protein. Detagged chordin constructs were incubated with a 4x molar excess Tsg-his and the mixture run through a his-trap column so that only chordin which was bound to Tsg would attach to the column. The complexes, eluted in 500mM imidazole, 150mM NaCl, 10mM Tris, pH7.4 were then further purified by size exclusion chromatography in 10mM Tris buffer with 150mM NaCl at pH7.4 order to separate unbound Tsg. Densitometry indicates that the complex eluted in a molar ratio of 1:1 chordin fragment:Tsg. The results of the purification are shown in Figure 5.11 with Tsg present in the chordin fragment fractions showing that it remains bound through the purification process and is stable.

128

Figure 5.11: Production of chordin-Tsg complexes. (A) Size exclusion purification of chordin-Tsg complexes. Eluate monitored by UV absorption (280nm), plots showing mAU against elution volume (ml) for (i) chordin ∆C and (ii) chordin ∆N. (B) SDS-PAGE analysis of the elution fractions for (i) chordin ∆C and (ii) chordin ∆N.

5.8 Chordin ∆C forms Fibres in the Presence of Tsg

An interesting feature of the chordin ∆C-Tsg complex grids is the formation of fibres along the grid shown in Figure 5.12A. These fibres had a regular, repeating pattern shown in Figure 5.12B. The dimensions of these fibres and the distance between repeats were measured by intensity of staining measured and plotted in graphs against distance. The fibres were found to have a repeating distance of approximately 4nm (Figure 5.12C). This result was reproducible and not seen when combining chordin ∆N with Tsg suggesting that this is not a staining artefact, however whether it is physiological has yet to be established. The buffer used was 10mM Tris-HCl, 150mM NaCl, pH7.4.

129

Figure 5.12: Chordin∆C-Tsg fibres. (A) Transmission electron micrograph image of chordin ∆C-Tsg complex on a grid showing free complex and fibres. (B) A chordin ∆C-Tsg fibre showing regular repeating units. (C) Average of 41 fibres showing distance along the fibres (nm) vs the intensity of the stain with a repeating distance of approximately 4nm.

5.9 Discussion

Functional Tsg was expressed and purified from HEK 293-EBNA cells. The construct is predicted from CD to consist of primarily strand secondary structure (~30%) with a small proportion of helix (~10%). It binds to immobilized chordin at physiological salt and pH with a Kd of 26nM which indicates strong association with the same order of magnitude as a value obtained from a previous study8. From biomolecular analysis of the chordin modulator Tsg, the study has determined many of the structural properties of both native and deglycosylated Tsg. These are summarized in Table 5.3. These data coupled with the ab initio models from SAXS data show a heavily glycosylated, almost globular protein which behaves as a stable monomer in solution.

130

Native Tsg Deglycosylated Tsg

MW (kDa) 30 25 Dmax (nm) 10.9 8.9 Rg (nm) 3.14 2.5 Rh(nm) 3.23 ND

S20W 2.72 ND f/f0 1.62 ND

Table 5.3: Summary of experimentally predicted values for the structural properties of Tsg. Molecular mass estimate from MALS, Dmax and Rg values derived from SAXS data and Rh,

S20W and f/f0 from AUC. ND = not determined.

Consistent with these data showing Tsg as a free monomer, it was not possible to detect binding of Tsg to chordin in a ratio higher than 1:1. However it is possible that two Tsg monomers bind to chordin independently. Unfortunately, the flexibility of the C-terminal domains of chordin meant that it was not possible to see the extra density directly using TEM and the propensity of chordin to form dimers meant that if could not be done using SAXS. Full length chordin and Tsg complexes could not be produced in sufficient quantities for MALS or AUC analysis. Due to the absence of vWC1 ∆N would only be expected to bind one Tsg and light scattering data from ∆C and Tsg complexes were uninterpretable, possibly due to fibre formation.

Formation of a stable complex with chordin has not enabled direct modelling of this complex for the reasons stated above, however it does pave the way for modelling a ternary complex of chordin, BMP and Tsg. In addition, it has been shown that on carbon coated grids chordin ∆C forms fibres in the presence of Tsg, an effect not observed with ∆C alone or with ∆N and Tsg. It would be interesting to determine whether this effect occurs in vivo as it would show a previously undocumented function for both Tsg and the chordin fragments. It is possible that it could be related to fragment turnover or some mechanism of BMP sequestration within the extracellular matrix.

131

6. Final Discussion

132

6.1 Data Summary

Constructs of chordin and its modulator Tsg were produced from HEK 293-EBNA cells. Their folding was verified using circular dichroism and binding activity confirmed using SPR. CD estimates of secondary structure for both proteins were approximately 10% α-helix and 30% β-sheet according to database comparisons. The dominance of β-sheet in the structure was predicted from related proteins, in particular the all-β crystal structure of vWC1 of crossveinless-133. It is probable that the vWC domains of chordin are also entirely composed of β-strand and turns and therefore it is likely that the helical regions are located within the CHRD domains. SPR analysis shows binding between Tsg and chordin of Kd 26nM which is 8 stronger than previous estimates, although in a similar range . Binding of chordin to BMP-2 was also estimated to be in the low nm range and although it was not possible to fit these data to Langmuir 1:1, bivalent or two state models, it was determined that there is not a significant loss of affinity following chordin cleavage by tolloids.

Tsg was found to be a stable monomer from SAXS, MALS and AUC both with and without its N-linked sugars. In AUC there was a trace of possible dimer but this was only a very small proportion at higher than physiological concentrations and may have been a more heavily glycosylated isoform instead. It is possible that Tsg either forms a dimer in the presence of BMPs or binds as two independent monomers to chordin, BMPs or both. However, the evidence for Tsg binding as a dimer primarily stems from crosslinking experiments where Tsg crosslinks with itself (giving the appearance of dimer) even in the absence of BMPs and chordin53. As Tsg is a monomer in solution this is almost certainly an artefact of non-specific crosslinking. It is interesting that it is free in solution because it means that two separate binding events and diffusion of two separate monomers need to be accounted for in computational models.

In contrast, SPR indicated that chordin and its fragments have significant self-affinity. This was verified using SAXS where the Rg increasing at higher chordin concentrations and AUC with two distinct C(s) distribution peaks seen. It was not possible to measure the affinity of the dimer using equilibrium AUC due to the inherent instability of chordin. However by comparing velocity AUC at different chordin concentrations (a method established by Schuck, P. (2003))178 it was possible to derive a Kd of 3.3µM. At this level of affinity it is probable that, while some chordin dimerization may occur under physiological conditions, monomeric chordin is the dominant form in vivo.

Both chordin and Tsg were found to be glycoproteins from PNGase-F digestion, with sugars contributing approximately 16% of the molecular mass of Tsg. For secreted proteins glycosylation is frequently associated with improved solubility, though other functional roles are an area of expanding study46, 191. This study of the structural properties of chordin and Tsg has determined the low resolution structure of both proteins with a Dmax of 18nm and 10.9nm respectively. Although Tsg is a third of the molecular mass of chordin it is almost two

133 thirds of the length because the Tsg structure is linear while the chordin structure loops back on itself. In addition, chordin has a large globular lobe at the N-terminus formed by the CHRD domains, whereas in Tsg the mass is more evenly distributed along the length of the protein. This is interesting because depending on binding orientation, Tsg may have a large exposed surface area when bound to chordin which might facilitate interaction with other pathway components such as BMPs and tolloids.

SAXS ab initio structures of glycosylated and deglycosylated Tsg were compared because of the apparent effect of the sugars on BMP-2 binding46. As would be expected from sequence, the glycosylation on Tsg sits between the two lobes of the protein, which is a diamond shape in its native form. It was shown that native Tsg is relatively compact and lacking the previously predicted flexible linker between the two domains. We also found by SPR that deglycosylation of Tsg does not appear to affect Tsg-chordin binding, whereas it has been shown to affect BMP-2 binding46. These data suggest that the Tsg binding sites for chordin and BMP-2 are distinct. Distinct binding regions for BMP and chordin on Tsg do allow for the possibility of a scaffold mechanism of action for Tsg90. It is also possible that binding of Tsg helps to stabilize the flexible regions of chordin following BMP binding.

By aligning SAXS ab initio structures of full length and cleavage fragment chordin constructs it was possible to predict the location of some domains. This alignment shows that chordin has a bulky near-globular N-terminal region consisting of vWC1 and the CHRD domains and a more linear C-terminal consisting of vWC2-4 which loops back. Sequence analysis shows that there are long connecting sequences between the vWC domains and the CHRD domains but short connections between domains of the same type (see annotated chordin sequence, Appendix III) so it makes sense that they would form distinct structures within the protein. The Rg from SAXS and Rh from AUC of full length chordin were determined to be 5.8nm and 4.7nm respectively similar to the values determined by MALS (5.4nm and 4.4nm) which helps to cross validate the model. It was also shown that the nanostructure of chordin does not change significantly following cleavage by tolloid metalloproteinases.

EM and SAXS structures for chordin and its fragments indicate that there is some flexibility between the N- and C- terminal lobes. This is interesting because it is likely that this flexibility enables co-operative BMP binding by vWC1 and the C-terminal vWC domains. The N- and C-terminal domains of chordin would both be able to bind a separate monomer from the same BMP dimer. The flexibility also means that vWC3 and vWC4 may both be able to interact with a BMP dimer already bound to vWC1. Given the higher affinity of vWC3 for BMP-2 and of vWC4 for BMP-7, this flexibility would allow for differential regulation of chordin antagonism of different types of BMPs which would allow finer control of signalling in the BMP morphogen gradient.

134

6.2 Comparison to Previous Literature

6.2.1 Twisted Gastrulation Structure

Little has been previously documented about the structure of Tsg, except that it has two distinct domains separated by a long linker sequence and N-linked glycosylations which play a role in binding affinity for BMPs46. The domains are an N-terminal BMP binding domain, similar to the vWC domains of chordin, and a C-terminal domain. Both domains are required for chordin binding53, 54 indicating that the Tsg binding sites for BMP and chordin are distinct. Some papers have also suggested that the connecting peptide sequence may be a variable hinge region allowing flexibility between the two domains49. Our data support existing studies which indicate that the chordin and BMP-binding domains of Tsg are distinct because deglycosylation of Tsg does not affect chordin binding in the same way that it affects BMP binding46. However the compact, stable structure obtained by SAXS indicates that there is no flexible hinge between the N- and C-terminal domains. The location of the bulky sugars in this region between the two lobes of the protein would also not facilitate flexibility in this region. Our predictions for a primarily β-sheet secondary structure for Tsg, based on circular dichroism, supports predictions that the structure of the N-terminal domain is similar to that of the predominantly β-sheet containing structures of vWC domains49.

6.2.2 Chordin Structure

This study constitutes the first direct structural study of chordin although components of related structures have been solved. The structure of vWC1 of the chordin-related protein, crossveinless-2, has been shown using X-ray crystallography in complex with BMP-233 as have the structures of vWC domains in fibronectin188 and collagen IIA189. These domains are similar in sequence to chordin and are thought to be similar in structure as well. In particular the cysteine residues of vWC domains are conserved suggesting a consistent pattern of disulfide bonding. The main difference between vWC1 of crossveinless-2 and the vWC domains of chordin is a short “clip” domain which is absent in chordin. Various vWC domains have also been shown to bind to each other in chordin family proteins and are emerging in the literature as a common interaction domain between different vWC containing proteins8, 31. Some structural predictions have previously been made for the four CHRD domains as well. PSI-PHRED predicts that they have an all-β structure and this is supported by the limited homology between CHRD domains and the immunoglobin-like folds of the Cu-Zn superoxide dismutase family, which are β-barrels40.

Our CD data for chordin indicates a primarily β-sheet structure with a small proportion of α- helix in contrast to previous predictions. The α-helical secondary structure is probably located in the CHRD domains as vWC structures have so far been consistently β-sheet between different proteins33, 188, 189 while predictions for CHRD domain structure were based

135 purely on sequence. It is not possible to thread the vWC domains into the model without knowing the structure of the CHRD domains, however the known binding domains of chordin and predicted binding ratio of 1:1 have led previous papers to assume a horseshoe-shape for chordin in schematic diagrams. This model has arisen because it would be necessary for chordin to wrap around the BMP dimer in order for vWC1 and the vWC2-4 domains to bind co-operatively. The apparent flexibility of the vWC2-4 region allows for both looping around and for either vWC3 or 4 to bind depending on which BMP form. The current model is represented by a schematic diagram in Figure 6.1. This model allows for the possibility of selective deactivation of chordin by tolloids. For example, ∆C may be an equally effective BMP-2 antagonist but less effective in blocking BMP-7 activity allowing for selective regulation.

Figure 6.1: Schematic diagram suggesting how flexibility between the N- and C-terminal regions of chordin could allow binding of either vWC3 or vWC4 to BMPs. If the structure of chordin was rigid, only vWC3 or vWC4 could bind to the same BMP dimer as vWC1, not both. Following cleavage of the C-terminal site, vWC3 which has high affinity for BMP-2 remains but vWC4 which has high affinity for BMPs -4 and -7 is lost.

As with many other vWC domain-containing proteins, chordin demonstrates a degree of self- affinity. It has been shown previously that chordin interacts with other vWC domain- containing chordin relatives through the vWC domains8, 31. Some vWC domain-containing proteins (e.g. crossveinless-2) are thought to be involved in endocytic turnover of chordin and it is likely that vWC-vWC interaction is required for this31, 36. We first showed chordin

136 dimerization using SAXS and SPR and then confirmed it with AUC, with our data showing a dissociation constant in the low µM range. While this affinity is not particularly strong, it confirms that the vWC mediated protein interaction observed in other proteins is conserved to at least some degree in chordin. This is significant because vWC-vWC domain interactions may mediate chordin binding to other vWC domain-containing proteins, which are not part of the chordin family. Put in a wider context it also provides another example of vWC-vWC interaction for what is increasingly becoming accepted as a common binding motif. Arguably, in addition to their other functions, vWC domains are a type of adaptor domain.

6.2.3 The Effect of Tolloid Cleavage on Chordin

Studies have shown two conflicting aspects of the relief of BMP-antagonism through tolloid cleavage of chordin. On the one hand, underexpression of tolloid produces a phenotype consistent with chordin over-expression96 and the reduced biological activity of truncated chordin is well documented27. In addition to this, tolloid-resistant chordin expressed in vivo has a stronger dorsalizing effect than wild-type chordin showing that cleavage by tolloids is an essential mechanism for countering its activity61. On the other hand, individual chordin domains retain the ability to antagonize BMPs and can produce similar phenotypes to chordin27. In addition the affinity between BMPs and some chordin fragments compared to full length chordin is relatively similar8. For this reason it is not immediately obvious why cleavage of chordin by tolloids is an effective means of antagonizing it. Currently there are two leading theories which are not mutually exclusive; that the chordin fragments are removed from the extracellular space more rapidly through endocytotic uptake and that loss of co-operative binding following cleavage allows bound BMP to be more readily displaced from chordin.

We investigated whether conformational change in the nanostructure may provide some explanation for reduced biological activity, but the cleavage fragments of chordin appear similar in shape to full length chordin although slightly smaller. This suggests that a mechanism, which does not directly affect the chordin-BMP interaction may be at work. The mechanism previously suggested is that the fragments are more susceptible to endocytic degradation than full length chordin which explains the reduced biological activity of the fragments in whole-body systems53, 81 despite little difference in binding. Chordin binds to a number of candidate proteins for mediating cellular uptake including cell surface heparin sulphate proteoglycans81 and crossveinless-231. It is possible that the vWC-vWC domain interaction discussed previously is a means of chordin interaction with these cell surface proteins leading to its internalization. It may be that cleavage of chordin leaves the binding surfaces of these domains more exposed and able to dock with larger proteins mediating their uptake, shown in Figure 6.2. In addition vWC1 domain overexpression phenotypes can

137 be partially rescued by co-overexpression of Tsg. This raises the possibility that Tsg may have a role in chordin antagonism involving cleanup of the larger cleavage fragments as well as increasing the chordinase activity of tolloids and dislodging individual vWC domains from BMPs53.

Figure 6.2: Schematic diagram showing a possible alternative mechanism for the antagonism of chordin activity by tolloid cleavage, based on observations from the literature. Our structure for full length chordin indicates that it loops back on itself. It is accessible to small proteins aided by flexibility. Although full length chordin is able to bind to larger proteins such as crossveinless-2 there is evidence that, perhaps due to less steric interference, chordin fragments may bind more readily which in turn may contribute to their endocytotic uptake36.

Alternatively a loss of co-operative BMP binding by the chordin fragments may play a role in reducing BMP antagonism. Binding only one BMP monomer leaves the other half of the dimer free to interact with a cell surface receptor. Once the BMP has bound to a receptor and the complex is immobilized on the cell surface it may be easier for a second receptor to displace the remaining antagonistic chordin domain. This would render the antagonism less effective. By this mechanism, chordin fragments would bind to free BMPs as readily as full length chordin but would nevertheless be less efficient at antagonizing the BMP-BMPR interaction. This seems a likely mechanism for chordin ∆N and the individual vWC domains

138 produced by tolloid cleavage, however in the case of chordin ∆C it is unlikely that it would have a significant impact on BMP-2 activity as discussed above.

6.2.4 Interactions of the Ternary Complex

Previous studies of the interaction between chordin, BMP-2 and Tsg have centred around measuring binding affinity. There is no direct comparison available for interactions with the chordin fragments, though most individual vWC domains tested can bind to BMPs with dissociation constants in the nM range8. In the case of supersog (truncated forms of the Drosophila equivalent of chordin with enhanced BMP-antagonistic activity) it was found that the more the SOG domains of the constructs were removed, the better the BMP antagonist41. This raised the possibility that chordin ∆C might also be a better antagonist of BMP-2 than full length chordin. It was also suggested that ∆C was not bound by Tsg but our studies show that this is not the case. Our data challenge some of the assumptions about the ternary complex. We show that free Tsg is not, as was previously assumed, a dimer although this does not mean that it cannot bind in the predicted 2:1 ratio to chordin. Unfortunately Tsg does not appear to stabilize the flexibility in chordin indicating that it does not bind as a dimer to either end of chordin. As a result we were unable to get an electron microscopy structure of the complex.

Figure 6.3A shows the predicted domain layout of chordin. The fitting is straightforward as there is only one area of density in which the CHRD1-4 SAXS model can be fitted. The fourth vWC domain is either unresolved due to flexibility or there is sufficient density in the model for it to curve back around which may suggest that it binds to vWC3. This is interesting as it raises the possibility of partial auto-antagonism which may increase chordin binding specificity. Bioactivity assays could be used to determine whether chordin ∆C (missing vWC4) has increased BMP-2 antagonistic activity, which would support this hypothesis. Figure 6.3A also shows BMP2 docked into chordin, demonstrating that the distances between the N- and C-termini are suitable for co-operative binding. This is in contrast to the SAXS data which suggests that the two termini cannot bind to BMP simultaneously. The most probable explanation for this is that flexibility between the N- and C-terminal lobes allows them to come into close enough proximity to bind BMPs co- operatively. A schematic diagram of the predicted final assembly of the chordin-BMP-Tsg ternary complex is shown in Figure 6.3B, in which vWC1 binds one BMP monomer and vWC3 or 4 can bind to the other BMP monomer. Twisted gastrulation binds to chordin and may also bind to BMP-2 at the same time, acting as a molecular scaffold.

139

Figure 6.3: (A) EM model of full length chordin with CHRD1-4 SAXS model (cyan) and vWC1 from crossveinless-2 fitted (vWC1=yellow, vWC2=green, vWC3=blue, vWC4=purple). BMP-2 (red) is shown docked. Shown in three orthogonal views. (B) Schematic diagram showing predicted ternary complex layout of BMP, chordin and Tsg.

Interestingly in the presence of Tsg chordin ∆C forms fibres on the TEM grid while full length chordin and ∆N do not. In the case of chordin ∆N it would not be possible to form end to end fibres as the N-terminal vWC domain is absent. It is not clear, however, why full length chordin would not form the fibres. This property of chordin has not been documented before and while it is reproducibly seen in EM further evidence is needed to determine whether or not it is a physiological phenomenon. If it is, this may be the mechanism by which Tsg increases endocytotic uptake of cleaved chordin, causing an accumulation of fragment in these long fibres. Since ∆C also binds to individual vWC domains (purification in urea is required to separate it from vWC1) fragments which are unable to form fibres themselves may also adhere to these chains. Unpublished data produced by the group show that individual vWC4 is monomeric but that vWC2-3 dimerizes, so the fibres may only be seen in ∆C because that is the only construct with C-terminal vWC3 domain exposed.

140

6.3 Future Directions

6.3.1 Double-Tagged Chordin to Further Improve Purity

One difficulty with the project was that we were unable to purify a chordin construct completely free from cleavage fragment contamination. This made it impractical to investigate full length chordin using analytical ultracentrifugation which would be interesting not only for determining the hydrodynamic radius of chordin more accurately, but also to determine whether full length chordin is more or less prone to dimerization than the fragments. In addition it is not certain whether full length chordin forms end to end dimers or whether the dimers seen in SAXS (which tends to be dominated by the largest component in the mix) are trace amounts of chordin ∆N dimer. A possible solution to this is to use a chordin construct with a different affinity tag at each end. The protein could then be purified in a two step process removing chordin ∆N and chordin ∆C one at a time as they would only have one tag each. The Strep-II tag seemed an optimum choice for combining good yields with a small eight residue tag (WSHPQFEK) which would be unlikely to interfere with folding or function.

There are also other investigations that would be made possible with double tagged chordin. Firstly it would be easier to interpret SPR experiments with a more homogenous sample. One particularly interesting follow on experiment from this study would be to determine whether BMPRs can displace cleaved chordin from BMP more easily than full length chordin. This could be done by immobilizing chordin, binding BMP and then running BMPR as a second analyte. If the Koff of BMP increased at a faster rate for chordin fragments than full length in the presence of BMP that would strongly support that mechanism. Tags at different ends of chordin could also be used to monitor cleavage rates by tolloids independently. This could be used to show whether BMP-1 has a preference for one site over the other or a preferred order of cleavage, by using antibodies to the two different fragments to quantify production of each. It would also be interesting to see whether this varies in the presence of modulators like Tsg.

6.3.2 CHRD 1-4 Domains

The CHRD domains are a unique fold to chordin and some bacterial proteins. Their structure and function is undetermined. There are a number of reasons to study the CHRD1-4 domains, one of which is the possibility that unlike full length chordin it may be feasible to crystallize this construct as it would be a more homogenous sample. Rigid body modelling with this structure and the crystal structure of vWC1 could then be used to map the overall chordin structures. Cloning into a bacterial expression system may facilitate this by removing probable need for deglycosylation and increasing yield although it is possible that they may

141 be too large and complex to reconstitute easily. This would provide valuable information as to the positioning of the binding domains and the interaction surfaces.

If the vWC domains are the common interaction domains of chordin and other cysteine-rich domain containing proteins, it would be helpful to show that the CHRD domains do not oligomerize (preliminary results indicate that they do not). In addition it raises the possibility of determining a function for the CHRD domains. Through pull-down assays it may also be possible to identify novel binding partners of the CHRD domains which could be confirmed through SPR and functional studies in tissue systems. This would be useful both to show their function in chordin (and perhaps reveal new functions for chordin) and also to predict possible functions for the bacterial proteins which contain them.

6.3.3 vWC-vWC Interactions vWC-vWC homology domain interactions are emerging as a common binding mechanism between the proteins containing them. Now that we have established that chordin vWC domains have self affinity, chordin could be screened for binding using SPR against other known vWC containing proteins to look for potential interaction partners. As discussed previously, there are no obvious structural or binding reasons why tolloid cleavage of chordin reduces BMP binding affinity, the vWC domains would be the next avenue for investigating this mechanism. Cleavage by tolloids may expose these domains making them more available for interaction with other vWC domain containing proteins involved in their cellular uptake such as crossveinless-2.

It would be interesting to start by determining whether vWC domain self affinity is stronger in the fragments than in full length chordin. AUC comparing the dissociation constant of ∆N to individual vWC domains and pure full length chordin would be a starting point for this. It may be that because the C-terminal domains of chordin loop back toward the N-terminus that the vWC domains of full length chordin are less exposed for binding to large proteins. In addition SPR could be used to show whether the fragments interact more strongly with internalizing proteins like crossveinless-2 (there is already some evidence that they do31, 36). This would be strong evidence in favour of an internalization mechanism of chordin inactivation following tolloid cleavage.

6.3.4 Interactions of the Ternary Complex

The analysis of complexes including chordin-BMP-2 were delayed in this project due mainly to the insolubility of BMP-2 expressed in a bacterial expression system. Various conditions were used to counter this insolubility including varying pH, salt, titration to form the complex gradually and the use of chaotropic agents but all were unsuccessful. As an alternative

142 strategy, BMP-2 was cloned into the same expression system as Tsg and chordin in the hope that glycosylation would increase solubility. A complication of BMP expression is that the C-terminal tail of BMP-2 is internal, and the pro-domain (which is cleaved prior to secretion) is required for correct folding and processing making tagging at either terminal impractical. An untagged BMP-2 construct in HEK cells has proved more soluble and a future aim is to use tagged chordin to bind to untagged BMP-2 in expression media to enable isolation of the complex.

The fibres formed by chordin ∆C in the presence of Tsg require further investigation. Close inspection of the fibres reveals that the regular repeats appear similar in shape to the individual chordin particles but it is possible that the formation of the fibres is an artefact of grid preparation. Mathematical models provide some support for a long range chordin assembly because a particular requirement for robustness is being able to store signal molecules in a restricted space135. Initially it would need to be shown that the effect is not an artefact of preparation, for example using a different staining method. It would be very interesting to see what the function and structure of the fibres are but it would be necessary to establish whether they are physiologically relevant first. This would involve showing that they can exist, for example by over expressing chordin ∆C and Tsg in vitro and using immunofluorescence to look for fibres. The next step would be showing whether they exist in vivo. This would not be straightforward as turnover of chordin ∆C in vivo is expected to be very high so the fibres would be likely to be unstable however it might be possible through cryosectioning.

6.3.5 Tolloid-Chordin Complex

With the cleavage resistant chordin construct, chordin-FN, and a tolloid construct with reduced catalytic activity produced previously60, it may be possible with crosslinking to study the structural properties of a tolloid-chordin complex. A potential difficulty with this is that there are two cleavage sites on chordin so there may be two binding locations. In this event, a solution could be to use the same primers to mutate the cleavage site of chordin ∆C or alternatively mutate the second cleavage site of chordin ∆N. The effectiveness of the latter has not been tested, however there would not be the issue of contaminating fragments, leading to a more homogenous sample. With either construct there would then be only one tolloid cleavage site reducing the potential modes of chordin-tolloid interaction. Of particular interest would be to see the binding regions of tolloid and chordin, and to see if regions distinct from the cleavage/catalytic sites are involved in the binding. In relation to this project it would also be useful to show whether chordin-tolloid interaction changes in the presence of Tsg or BMP-2.

143

6.3.6 Differential Regulation of BMPs by Tolloid Cleavage

We have shown that there is flexibility between the CHRD region and the C-terminal vWC2- 4 region. This could allow vWC3 or vWC4 to bind in co-operation with vWC1 rather than one or the other. This is interesting in the context of tolloid regulation because if either domain can bind co-operatively, removal of vWC4 by tolloids could reduce antagonism of BMP-7 and not BMP-2 as explained in Section 6.2.2. This adds an extra potential layer of complexity to fine tuning the BMP gradient in development because BMPs antagonized by chordin could be selectively reactivated. It may even be that whether the fragments remain and continue to have antagonistic activity or whether they are degraded rapidly following cleavage is tissue and timing dependent during embryogenesis.

There are various ways that this could be tested. One method would be to study chordin in complex with BMP-2 compared to a chordin-BMP-7 or -4 complex but the drawback is that vWC domains are small and vWC3 and 4 are adjacent so it may not be immediately obvious whether there is a difference at low resolution. Alternatively the purified constructs could be compared using in vitro analysis to compare their effects on cell signalling. For example the BMP antagonistic activity full length chordin and ∆C in a cell culture model could be compared with both BMP-2 and BMP-7. If ∆C had a weaker antagonistic effect on BMP-7 than full length chordin but no difference with BMP-2 this would be strong evidence in favour of this model.

6.4 Summary

The first aim of this project was to generate and purify in a mammalian expression system a toolkit of constructs for the structural study of the BMP-2 antagonist chordin and to understand if Tsg and tolloids regulate the structure. To achieve this aim, chordin resistant to cleavage by endogenous tolloid metalloproteinases was produced in a mammalian expression system. In addition chordin constructs with sequence boundaries mimicking the larger cleavage fragments of chordin and twisted gastrulation were also generated. These constructs were successfully cloned, expressed and confirmed to be folded by CD and had binding ability confirmed by SPR. These techniques determined that both proteins have a predominantly β-sheet secondary structure and a Kd for binding between chordin and Tsg of 26nM which is stronger than previous estimates though within the same order of magnitude.

We show that chordin has a bowl-shaped structure with flexibility in the vWC2-4 region allowing for differential regulation of specific BMPs through chordin cleavage by tolloids. The structural and hydrodynamic properties have been investigated using a range of in solution techniques including SAXS, MALS and AUC to give corroborating evidence. The low resolution structure of twisted gastrulation has also been determined showing a bulky mass

144 of glycan in the mid-region of the protein, the presence of which does not appear to have an effect on chordin binding.

The second two aims were to investigate the mechanism of action of twisted gastrulation on chordin and how tolloid cleavage reduces the biological activity of chordin while leaving the BMP binding domains intact. It was found that cleavage of chordin does not result in a significant change in nanostructure or BMP binding affinity, leaving more rapid endocytotic degradation of chordin fragments than full length chordin as the leading mechanistic hypothesis. Our studies suggest a potential link between these two lines of investigation in that Tsg appears to cause chordin fragment fibres to form which may be a means of accumulating fragments to aid endocytotic uptake. This requires further investigation to show whether it is a physiological phenomenon.

There are three main branches of future research following on from this project. The first is to gain more structural insights into chordin. This could be achieved by solving the structure of one or more CHRD domains allowing for rigid body modelling into the nanostructure. Forming a chordin-BMP complex would ideally lock the flexible regions into a more rigid conformation through co-operative binding, allowing an EM structure for the whole chordin molecule. The second is to investigate the possibility of selective antagonism of different types of BMP by tolloid cleavage of the C-terminal vWC domain. Bioactivity assays could be used to determine whether chordin ∆C has a stronger antagonistic effect on BMP-2 than on BMPs -4 or -7. Finally investigation of the formation of ∆C fibres in the presence of Tsg, which may prove to be a novel function, perhaps involved in the cleanup of cleavage fragments.

145

7. References

146

1. Gazzerro, E. and Canalis, E. (2006) Bone morphogenetic proteins and their antagonists. Rev Endocr Metab Disord, 7: 51-65.

2. Sawala, A., Sutcliffe, C. and Ashe, H. (2012) Multistep molecular mechanism for bone morphogenetic protein extracellular transport in the Drosophila embryo. PNAS, 109(8): 11222-11227.

3. Scheufler, C., Sebald, W. and Hulsmeyer, M. (1999) Crystal structure of human bone morphogenetic protein-2 and 2.7 A Resolution. J Mol Biol, 287: 103-115.

4. Yin, H., Yeh, L., Hinck, A. and Lee, J. (2008) Characterization of ligand-binding properties of the human BMP type II receptor extracellular domain. J Mol Biol, 378: 191-203.

5. Klages, J., Kotzsch, A., Coles, M., Sebald, W., Nickel, J., Muller, T. and Kessler, H. (2008) The solution structure of BMPR-IA reveals a local disorder-to-order transition upon BMP-2 binding. Biochemistry, 47: 11930-11939.

6. Sengle, G., Ono, R., Lyons, K., Bachinger, H. and Sakai, L. (2008) A new model for growth factor activation: type II receptors compete with the prodomain for BMP-7. J Mol Biol, 381: 1025-1039.

7. Sopory, S., Kwon, S., Wehrli, M. and Christian, J. (2010) Regulation of Dpp activity by tissue-specific cleavage of an upstream site within the prodomain. Developmental Biology, 346: 102-112.

8. Zhang, J., Huang, Y., Qiu, L., Nickel, J. and Sebald, W. (2007) von Willebrand factor type C domain-containing proteins regulate bone morphogenetic protein signalling through different recognition mechanisms. J Biol Chem, 282: 20002-20014

9. Jmol: an open-source Java viewer for chemical structures in 3D. http://www.jmol.org.

10. Allendorph, G., Vale, W. and Choe, S. (2006) Structure of the ternary signalling complex of a TGFβ superfamily member. PNAS, 103: 7643-7648.

147

11. David, L., Feige, J. and Bailly, S. (2009) Emerging role of bone morphogenetic proteins in angiogenesis. Cytokine and Growth Factor Reviews, 20: 203-212.

12. Meng, X., Chung, A. and Lan, H. (2013) Role of the TGFβ/BMP-7/Smad pathways in renal diseases. Clinical Science, 124(4): 243-254.

13. Javelaud, D. and Mauviel, A. (2005) Crosstalk mechanisms between the mitogen- activated protein kinase pathways and Smad signaling downstream of TGF-beta: implications for carcinogenesis. Oncogene, 24(37): 5742-5750.

14. Hay, E., Lemonnier, J., Fromique, O. and Marie, P. (2001) Bone morphogenetic protein- 2 promotes osteoblast apoptosis through a Smad-independent, protein kinase C-dependent signaling pathway. J Biol Chem, 276(31): 29028-29036.

15. Qi, X., Li, T., Hao, J., Hu, J., Wang, J., Simmons, H., Miura, S., Mishina, Y. and Zhao, G. (2004) BMP4 supports self-renewal of embryonic stem cells by inhibiting mitogen activated protein kinase pathways. PNAS, 101: 6027-6032.

16. Lagna, G., Nguyen, P., Ni, W. and Hata, A. (2006) BMP-dependent activation of caspase-9 and caspase-8 mediates apoptosis in pulmonary artery smooth muscle cells. Am J Physiol Lung Cell Mol Physiol, 291: 1059-1067.

17. Su, D. et al. (2011) Role of p38 MAPK pathway in BMP4-mediated Smad-dependent premature senescence in lung cancer cells. Biochem J, 433: 333-343.

18. Yang, Y., Liu, Y., Chuang, L., Guh, J., Lee, T., Liao, T., Hung, M. and Chiang, T. (2009) Bone morphogenetic protein-2 antagonizes renal interstitial fibrosis by promoting catabolism of type I transforming growth factor-β receptors. Endocrinology, 150: 727-740.

19. Son, J. et al. (2011) Association of serum bone morphogenetic protein 4 levels with obesity and metabolic syndrome in non-diabetic individuals. Endocrine Journal, 58: 39-46.

148

20. Millot, S. et al (2010) Erythropoietin stimulates spleen BMP-4 dependent stress erythropoiesis and partially corrects anaemia in a mouse model of generalized inflammation. Blood, 116: 6072-6081.

21. Wutzl, A., Rauner, M., Seemann, R., Millesi, W., Krepler, P., Pietschmann, P. and Ewers, R. (2010) Bone morphogenetic proteins 2, 5, and 6, in combination stimulate osteoblasts but not osteoclasts in vitro. Journal of Orthopaedic Research, 28: 1431-1439.

22. Groppe, J. et al. (2002) Structural basis of BMP signalling inhibition by the cysteine knot protein noggin. Nature, 420: 636-642.

23. Chang, C. et al. (2001) Twisted gastrulation can function as a BMP antagonist. Nature, 410: 483-487.

24. Valera, E., Isaacs, M., Kawakami, Y., Izpisua Belmonte, J. and Choe, S. (2010) BMP-2/6 heterodimer is more effective than BMP-2 or BMP-6 homodimers as inductor of differentiation of human embryonic stem cells. PLoS ONE, 5(3): 11167.

25. Yu, K., Kang, K., Heine, P., Pyati, U., Srinivasan, S., Biehs, B., Kimelman, D. and Bier, E. (2004) Cysteine repeat domains and adjacent sequences determine distinct bone morphogenetic protein modulatory activities of the drosophila sog protein. Genetics, 166: 1323-1336.

26. Larrain, J, Oelgeschlager, M., Ketpura, N,m Reversade, B., Zakin, L. and Robertis, E. (2001) Proteolytic cleavage of chordin acts as a switch for the dual activities of twisted gastrulation in BMP signalling. Development, 128: 4439-4447.

27. Larrain, J., Bachiller, D., Lu, B., Agius, E., Piccolo, S. and Robertis, E. (2000) BMP- binding modules in chordin: a model for signalling recognition in the extracellular space. Development, 127: 821-830.

28. Serpe, M. et al. (2008) The BMP binding protein crossveinless-2 is a short-range, concentration-dependent, biphasic modulator of BMP signalling in Drosophila. Dev Cell, 14: 940-953.

149

29. Abreu, J., Coffinier, C., Larrain, J., Oelgaschlager, M. and Robertis, E. (2001) Chordin- like CR and the regulation of evolutionarily conserved extracellular signalling systems. Gene, 287: 39-47.

30. Oren, A,, Toporik, A., Biton, S., Almogy, N., Eschel, D., Bernstein, J., Savitsky, K. and Rotman, G. (2004) hCHL2, a novel chordin-related gene, displays differential expression and complex alternative splicing in human tissues and during myoblast and osteoblast maturation. Gene, 331: 17-31.

31. Kelley, R. et al (2009) A concentration-dependent endocytic trap and sink mechanism converts Bmper from an activator to an inhibitor of BMP signalling. Journal Cell Biology, 184: 597-609.

32. Qiu, L., Zhang, J., Kotzsch, A., Sebald, W. and Mueller, T. (2008) Crystallization and preliminary x-ray analysis of the complex of the first von Willebrand type C domain bound to bone morphogenetic protein 2. Crystallization Communications, 64: 307-312.

33. Zhang, J., Qiu, L., Kotzsch, A., Weidauer, S., Patterson, L. and Hammerschmidt, M. (2008) Crystal structure analysis reveals how the chordin family member crossveinless-2 blocks BMP-2 receptor binding. Developmental Cell, 14: 739-750.

34. Coffinier, C., Ketpura, N., Tran, U., Geissert, D. And Robertis, I. (2002) Mouse crossveinless-2 is the vertebrate homolog of a drosophila extracellular regulator of BMP signalling. Mechanisms of Development, 1198: 179-184.

35. Fujisawa, T., Huang, Y., Sebald, W. and Zhang, J. (2009) The binding of von Willebrand factor type C domains of chordin family proteins to BMP-2 and Tsg is mediated by their SD1 subdomain. Biochemical and Biophysical Research Communications, 385: 215-219.

36. Ambrosio, A., Taelman, V., Lee, H., Metzinger, C., Coffinier, C. and Robertis, E. (2008) Crossveinless-2 is a BMP feedback inhibitor that binds chordin/BMP to regulate Xenopus embryonic patterning. Dev Cell, 15(2): 248-260.

150

37. Zhang, J., Patterson, L., Qiu, L., Graziussi, D, Sebald, W. and Hammerschmidt, M. (2010) Binding between crossveinless-2 and chordin von Willebrand factor type c domains promotes BMP signalling by blocking chordin activity. PLoS ONE, 5(9): e12846.

38. Balemans, W. and Hul, W. Extracellular regulation of BMP signalling in vertebrates: a cocktail of modulators. Developmental Biology, 250: 231-250.

39. Millet, C., Lemaire, P., Orsetti, B., Guglielmi, P. and Francois, V. (2001) The human chordin gene encodes several differentially expressed spliced variants with distinct BMP opposing activities. Mechanisms of Development, 106: 85-96.

40. Hyvonen, M. (2003) CHRD, a novel domain in the BMP inhibitor chordin, is also found in microbial proteins. Trends Biochem Sci, 28: 470-473.

41. Yu, K., Srinivasan, S., Shimmi, O., Biehs, B., Rashka, K., Kimelman, D., O’Connor, M. and Bier, E. (2000) Processing of the drosophila sog protein creates a novel BMP inhibitory activity. Development, 127: 2143-2154.

42. Ross, J. (2001) Twisted gastrulation is a conserved extracellular BMP antagonist. Nature, 410: 479-483.

43. Peluso, C., Umulis, D., Kim, Y., O’Connor, M. and Serpe, M. (2011) Shaping the BMP morphogen gradient through enzyme-substrate interactions. Developmental Cell, 21: 375- 383.

44. Nosaka, T., et al (2003) Mammalian twisted gastrulation is essential for skeleto- lymphogenesis. Molecular and Cellular Biology, 23: 2969-2980.

45. Shimmi, O., Ralston, A., Blair, S., and O’Connor. (2005) The crossveinless gene encodes a new member of the twisted gastrulation family of BMP binding proteins which, with short gastrulation, promotes BMP signalling in the crossveins of the drosophila wing. Developmental Biology, 282: 70-83.

151

46. Billington, C. et al (2011) Glycosylation of twisted gastrulation is required for BMP binding and activity during craniofacial development. Front Physiol, 2: 59.

47. Fenwick, M., Mansour, Y., Franks, S. and Hardy, K. (2011) Identification and regulation of bone morphogenetic protein antagonists associated with preantral follicle development in the ovary. Endocrinology, 152: 3513-3526.

48. Fonseca, R., Zee, M. and Roth, S. (2010) Evolution of extracellular Dpp modulators in insects: the roles of tolloid and twisted-gastrulation in dorsoventral patterning of the tribolium embryo. Developmental Biology, 345: 80-93.

49. Vilmos, P., Sousa-Neves, R., Lukacsovich, T. and Marsh, L. (2005) Crossveinless defines a new family of twisted gastrulation-like modulators of bone morphogenetic protein signalling. EMBO reports, 6(3): 262-267.

50. Oelgeschlager, M., Tran, U., Grubisic, K. and Robertis, E. (2004) Identification of a second xenopus twisted gastrulation gene. Int J Dev Biol, 48(1): 57-61.

51. Scott, I., Blitz, I., Pappano, W., Maas, S., Cho, K. and Greenspan, D. (2001) Homologues of twisted gastrulation are extracellular cofactors in antagonism of BMP signalling. Nature, 410: 475-478.

52. Vilmos, P., Sousa-Neves, R., Lukacsovich, T. and Marsh, J. (2005) Crossveinless defines a new family of twisted gastrulation-like modulators of bone morphogenetic protein signalling. EMBO Reports, 6: 262-267.

53. Oelgeschlager, M., Larrain, J., Geissert, D. and Robertis, E. (2000) The evolutionarily conserved BMP-binding protein twisted gastrulation promotes BMP signalling. Nature, 405: 757-763.

54. Oelgeschlager, M., Reversade, B., Larrain, J., Little, S., Mullins, M. and Robertis, E. (2003) The pro-BMP activity of twisted gastrulation is independent of BMP binding. Development, 130: 4047-4056.

152

55. Gaoxiang, G and Greenspan, D. (2006) Developmental roles of the BMP1/TLD metalloproteinases. Birth Defects Research, 78: 47-68.

56. Bode, W., Gomis-Ruth, F., Zwilling, R. and Stocker, W. Structure of astacin and implications for activation of astacins and zinc-ligation collagenases. Letters to Nature, 358: 164-167.

57. Reynolds, S., Zhang, D., Puzas, E., O’Keefe, R., Rosier, R. and Reynolds, P. (2000) Cloning of the chick BMP1/tolloid cDNA and expression in skeletal tissues. Gene, 248: 233- 243.

58. Scott, I. (1999) Mammalian BMP-1/tolloid-related metalloproteinases, including novel family member mammalian tolloid-like 2, have differential enzymatic activities and distributions of expression relevant to patterning and skeletogenesis. Developmental Biology, 213: 283-300.

59. Canty, E., Garrigue-Antar, L. and Kadler, K. (2006) A complete domain structure of drosophila tolloid is required for cleavage of short gastrulation. The Journal of Biological Chemistry¸281: 13258-13267.

60. Berry, R., Jowitt, T., Ferrand, J., Roessle, M., Grossman, G., Canty-Laird, E., Kammerer, R. Kadler, K. and Baldock, C. (2009) Role of dimerization and substrate exclusion in the regulation of bone morphogenetic protein-1 and mammalian tolloid. PNAS, 106: 8561-8566.

61. Xie, J. and Fisher, S. (2004) Twisted gastrulation enhances BMP signalling through chordin dependent and independent mechanisms. Development, 132: 383-391.

62. Chau, P., Fielding, P. and Fielding, C. (2007) Bone morphogenetic protein-1 (BMP-1) cleaves human proapolipoprotein A1 and regulates its activation for lipid binding. Biochemistry, 46: 8445-8450.

153

63. Steiglitz, B., Ayala, M., Narayanan, K., George, A. and Greenspan, D. (2004) Bone morphogenetic protein-1/Tolloid-like proteinases process dentin matrix protein-1. JBC, 279: 980-986.

64. Marschall, Z. and Fisher, L. (2010) Decorin is processed by three isoforms of bone morphogenetic protein-1 (BMP1). Biochem Biophys Res Commun, 391: 1374-1378.

65. Ge, G. and Greenspan, D. (2006) BMP1 controls TGFβ1 activation via cleavage of latent TGFβ-binding protein. JCB, 175: 111-120

66. Ge, G., Fernandez, C., Moses, M. and Greenspan, D. (2007) Bone morphogenetic protein 1 processes prolactin to a 17-kDa antiangiogenic factor. PNAS, 104: 10010-10015.

67. Grotewold, L., Plum, M., Dildrop, R., Peters, T. and Ruther, U. Bambi is coexpressed with BMP-4 during mouse embryogenesis. Mechanisms of Development, 100: 327-330.

68. Xavier, S., Gilbert, V., Rastaldi, M., Krick, S., Kollins, D., Reddy, A., Bottinger, E., Cohen, C. and Schlondorff, D. (2010) BAMBI is expressed in endothelial cells and is regulated by lysosomal/autolysosomal degradation. PLoS ONE, 5: e12995.

69. Pils, D., et al. (2010) BAMBI is overexpressed in ovarian cancer and co-translocates with Smads into the nucleus upon TGF-β treatment. Gynecologic Oncology, 117: 189-197.

70. Lin, Z., Gao, C., Ning, Y., He, X., Wu, W. and Chen, W. (2008) The pseudoreceptor BMP and activin membrane-bound inhibitor positively modulates Wnt/β-catenin signalling. J Biol Chem, 283: 33053-33058.

71. Chen, J., Bush, J., Ovitt, C., Lan, Y. and Jiang, R. (2007) The TGF-β pseudoreceptor gene bambi is dispensable for mouse embryonic development and postnatal survival. Genesis, 45: 482-486.

72. Tramullas, M. et al. (2010) BAMBI (bone morphogenetic protein and activin membrane- bound inducer) reveals the involvement of the transforming growth factor-β family in pain modulation. The Journal of Neuroscience, 30: 1502-1511.

154

73. Sakuragi, M., Sasai, N., Ikeya, M., Kawafa, M., Onai, T., Katahira, T., Nakamura, H. and Sasai, Y. (2006) Functional analysis of chick ONT1 reveals distinguishable activities among olfactomedin-related signalling factors. Mechanisms of Development, 123: 114-123.

74. Inomata, H., Haraguchi, T. and Sasai, Y. (2008) Robust stability of the embryonic axial pattern requires a secreted scaffold for chordin degradation. Cell, 134: 854-865.

75. Yabe, T., Shimizu, T., Muraoka, O., Bae, Y., Hirata, T., Nojima, H., Kawakami, A., Hirano, T. and Hibi, M. (2003) Ogon/secreted frizzled functions as a negative feedback regulator of BMP signalling. Development, 130: 2705-2716.

76. Lee, H., Ambrosio, A., Reversade, B. and Robertis, E. (2006) Embryonic dorsal-ventral signalling: secreted frizzled-related proteins as inhibitors of tolloid proteinases. Cell, 124: 147-159.

77. Muraoka, O., Shimizu, T., Yabe, T., Nojima, H., Bae, Y., Hashimoto, H. and Hibi, M. (2006) Sizzled controls dorso-ventral polarity by repressing cleavage of the chordin protein. Nature Cell Biology, 8: 329-338.

78. Kobayashi, K. et al (2008) Secreted frizzled related-2 is a procollagen C proteinase enhancer with a role in fibrosis associated with myocardial infarction. Nature Cell Biology, 11: 46-55.

79. Ploper, D., Lee, H. and Robertis, E. (2011) Dorsal-ventral patterning: crescent is a dorsally secreted frizzled-related protein that competitively inhibits tolloid proteases. Developmental Biology, 15: 317-328.

80. Larrain, J., Brown, C. and Robertis, E. (2003) Integrin α3 mediates binding of chordin to the cell surface and promotes its endocytosis. EMBO reports, 4: 813-818.

81. Jasuja, R., Allen, B., Pappano, W., Rapraeger, A. and Greenspan, D. (2004) Cell- surface heparin sulphate proteoglycans potentiate chordin antagonism of bone morphogenetic protein signalling and are necessary for cellular uptake of chordin. Journal of Biological Chemistry, 279: 51289-51297.

155

82. Moreno, M., Munoz, R., Aroca, F., Labarca, M., Brandan, E. and Larrain, J. (2005) Biglycan is a new extracellular component of the chordin-BMP4 signalling pathway. EMBO, 24: 1397-1405.

83. Huang, G., Zhang, Y., Kim, B., Ge, G., Annis, D., Mosher, D. and Greenspan, D. (2009) Fibronectin binds and enhances activity of bone morphogenetic protein 1. Journal of Biological Chemistry, 284: 25879-25888.

84. Robertis, E. and Plouhinc, J. (2010) Systems biology of the self-regulating morphogenetic gradient of the xenopus gastrula. CSH Perspectives, 1(2).

85. Zhang, Y., Lander, A. and Nie, Q. (2007) Computational analysis of BMP gradients in dorsal-ventral patterning of the zebrafish embryo. Journal of Theoretical Biology, 248: 579- 589.

86. Aamar, E. and David, I. (2010) Sox17 and chordin are required for the formation of kupffer’s vesicle and left-right asymmetry determination in zebrafish. Developmental Dynamics, 239: 2980-2988.

87. Garriock, R., Czeisler, C., Ishii, Y., Navetta, A. and Mikawa, T. (2010) An anteriorposterior wave of vascular inhibitor downregulation signals aortae fusion along the embryonic midline axis. Development, 137: 3697-3705.

88. Anderson, R., Lawrence, A., Stottmann, R., Bachiller, D. and Klingensmith, J. (2002) Chordin and noggin promote organizing centers of forebrain development in the mouse. Development, 129: 4975-4987.

89. Krause, C., Guzman, A. and Knaus, P. (2011) Noggin. Int J Biochem Cell Biol, 43: 478- 481.

90. Harland, R. (2008) A protein scaffold plays matchmaker for chordin. Cell, 134: 718-719.

156

91. Agathon, A., Thisse, C. and Thisse, B. (2003) The molecular nature of the zebrafish tail organizer. Nature, 424: 448-452.

92. Christiaen, L., Stolfi, A. and Levine, M. (2010) BMP signalling coordinates gene expression and cell migration during precardiac mesoderm development. Developmental Biology, 340: 179-187.

93. Blitz, I., Shimmi, O., Wunnenburg-Stapleton, K., O’Connor, M. and Cho, K. (2000) Chordin a long-range- or short-range-acting factor? Roles for BMP-1 regulated metalloproteinases in chordin and BMP4 autofeedback loop regulation. Developmental Biology, 223: 120-138.

94. Reversade, B., and Robertis, E. (2005). Regulation of Admp and Bmp2/4/7 at Opposite Embryonic Poles Generates a Self-Regulating Morphogenetic Field. Cell, 123, 1147-1160.

95. Ben-Zvi, D., Shilo, B., Fainsod, A. and Barkai, N. (2008) Scaling of the BMP activation gradient in Xenopus embryos. Nature, 453: 1205-1212.

96. Jasuja, R., Voss, N., Ge, G., Hoffman, G., Lyman-Gingerich, J., Pelegri, F. and Greenspan, D. (2006) Bmp1 and mini fin are functionally redundant in regulating formation of the zebrafish dorsoventral axis. Mechanisms of Development, 123: 548-558.

97. Little, D. and Mullins, M. (2004) Twisted gastrulation promotes BMP signalling in zebrafish dorsal-ventral axial patterning. Development, 131: 5825-5835.

98. Wills, Al., Harland, R. and Khokha, M. (2006) Twisted gastrulation is required for forebrain specification and cooperates with chordin to inhibit BMP signalling during X. Tropicalis gastrulation. Developmental Biology. 289: 166-178.

99. Coffinier, C., Tran, U., Larrain, J. and Robertis, E. (2001) Neuralin-1 is a novel chordin- related molecule expressed in the mouse neural plate. Mechanisms of Development, 100: 119-122.

157

100. Nakayama, N. et al. (2001) A novel chordin-like protein inhibitor for bone morphogenetic protein expressed preferentially in mesenchymal cell lineages. Developmental Biology, 232: 372-387.

101. Branam, A., Hoffman, G., Pelegri, F. and Greenspan, D. (2010) Zebrafish chordin-like and chordin are functionally redundant in regulating patterning of the dorsoventral axis. Developmental Biology, 341: 444-458.

102. Sidi, S., Goutel, C., Peyriéras, N. and Rosa, F. (2003) Maternal induction of ventral fate by zebrafish radar. Developmental Biology, 100(6): 3315-3320.

103. Zakin, L., Metzinger, C., Chang, E., Coffinier, C. and Robertis, E. (2008) Development of the vertebral morphogenetic field in the mouse: interactions between crossveinless-2 and twisted gastrulation. Dev Biol, 323: 6-18.

104. Zakin, L. and Robertis, E. (2010) Extracellular regulation of BMP signalling. Curr Biol, 9;20(3): 89-92,

105. Shimmi, O., Umulis, D., Othmer, H. and O’Connor, M. (2005) Facilitated transport of a Dpp/Scw heterodimer by sog/Tsg leads to robust patterning of the drosophila blastoderm embryo. Cell, 120: 873-886.

106. Choi, M. and Klingensmith, J. (2009) Chordin is a modifier of Tbx1 for craniofacial malformations of 22q11 deletion syndrome phenotypes in mouse. PLoS Genetics, 5: e1000395.

107. Bachiller, D., Klingensmith, J., Shneyder, N., Tran, U., Anderson, R., Rossant, J. and Robertis, E. (2003) The role of chordin/bmp signals in mammalian pharyngeal development and DiGeorge syndrome. Development, 130: 3567-3578.

108. Delot, E., Shneyder, N., Zhang, H. and Bachiller, D. (2007) Abnormal venous and arterial patterning in chordin mutants. Developmental Dynamics, 236: 2586-2593.

158

109. Bachiller, D. et al. (2000) The organizer factors chordin and noggin are required for mouse forebrain development. Nature, 403: 658-661.

110. McMahon, J., Takada, S., Zimmerman, L., Fan, C., Harland, R. and McMahon, A. (1998) Noggin-mediated antagonism of BMP signalling is required for growth and patterning of the neural tube and somite. Genes and Development, 12: 1488-1452.

111. Anderson, R., Stottmann, R., Choi, M. and Klingensmith, J. (2006) Endogenous bone morphogenetic protein antagonists regulate mammalian neural crest generation and survival. Developmental Dynamics, 235: 2507-2520.

112. Scott, I., Steiglitz, B., Clark, T., Pappano, W. and Greenspan, D. (2000) Spatiotemporal expression patterns of mammalian chordin during postgastrulation embryogenesis and in postnatal brain. Developmental Dynamics, 217: 449-456.

113. Zakin, L., Chang, E., Plouhinec, J. and Robertis, E. (2010) Crossveinless-2 is required for the relocalization of chordin protein within the vertebral field in mouse embryos. Developmental Biology, 347: 204-215.

114. Funari, V. et al, (2010) BMPER mutation in diaphanospondylodysostosis indentified by ancestral autozygosity mapping and targeted high-throughput sequencing. The American Journal of Human Genetics, 87: 532-537.

115. Pappano, W., Scott, I., Clark, T., Eddy, R., Shows, T. and Greenspan, D. (1998) Coding sequence and expression patterns of mouse chordin and mapping of the cognate mouse chrd and human CHRD genes. Genomics, 52: 236-239.

116. Sun, M., Thomas, M., Herder, R., Bofenkamp, M., Selleck, S. and O’Connor, M. (2007) Presynaptic contributions of chordin to hippocampal plasticity and spatial learning. The Journal of Neuroscience, 27: 7740-7750.

117. Gong, Y. et al (1999) Heterozygous mutations in the gene encoding noggin affect human joint morphogenesis. Nature Genetics, 21: 302-304.

159

118. Sun, M., Forsman, C., Sergi, C., Gopalakrishnan, R., O’Connor, M. and Petryk, A. (2010) The expression of twisted gastrulation in postnatal mouse brain and functional implications. Neuroscience, 169: 920-931.

119. Ikeya, M., Nosaka, T., Fukushima, K., Kawada, M., Furuta, Y., Kitamura, T. and Sasai, Y. (2008) Twisted gastrulation mutation suppresses skeletal defect phenotypes in crossveinless 2 mutant mice. Mechanisms of Development, 125: 832-842.

120. Zakin, L. and Robertis, E. (2004) Inactivation of mouse twisted gastrulation reveals its role in promoting BMP4 activity during forebrain development. Development, 131: 413-424.

121. Lowe, C. et al (2006) Dorsoventral patterning in hemichordates: insights into early chordate evolution. PLoS Biol, 4(9): e291.

122. Akiyama-Oda, Y. and Oda, H. (2006) Axis specification in the spider embryo: dpp is required for radial-to-axial symmetry transformation and sog for ventral patterning. Development, 133: 2347-2357.

123. Kozmikova, I., Smolikova, J., Vlcek, C. and Kozmik, Z. (2011) Conservation and diversification of an ancestral chordate gene regulatory network for dorsoventral patterning. PLoS One, 6: e14650.

124. Onai, T., Yu, J., Blitz, I., Cho, K. and Holland, L. (2010) Opposing nodal/vg1 and bmp signals mediate axial patterning in embryos of the basal chordate amphioxus. 344: 377-389.

125. Nagai, H., Mak, S., Weng, W., Nakaya, Y., Ladher, R. and Sheng, G. (2011) Embryonic development of the emu, Dromaius novaehollandiae. Developmental Dynamics, 240: 162- 175.

126. Kuo, D. and Weisblat, D. (2011) A new molecular logic for BMP-mediated dorsoventral patterning in the leech Helobdella. Current Biology, 21: 1282-1288.

160

127. Bradham, C., Oikonomou, C., Kuhn, A., Core, A., Modell, J., McClay, D. and Poutstka, A. (2009) Chordin is required for neural but not axial development in sea urchin embryos. Developmental Biology, 328: 221-233.

128. Lapraz, F., Besnardeau, L. and Lepage, T. (2009) Patterning of the dorsal-ventral axis in echinoderms: insights into the evolution of the BMP-chordin signalling network. PLoS Biology, 7: e1000248.

129. Saina, M., Genikhovich, G., Renfer, E. and Technau, U. (2009) BMPs and chordin regulate patterning of the directive axis in a sea anemone. Developmental Biology, 106: 18592-18597.

130. Umulis, D., O’Connor, M. and Blair, S. (2009) The extracellular regulation of bone morphogenetic protein signalling. Development, 136: 3715-3728.

131. O’Connor, M., Umulis, D., Othmer, H. and Blair, S. (2006) Shaping BMP morphogen gradients in the Drosophila embryo pupal wing. Development, 133: 183-193.

132. Umulis, D. (2010) Organism-scale modelling of early Drosophila patterning via bone morphogenetic proteins. Dev Cell, 18: 260-274.

133. Reeves, G., Muratov, C., Schupbach, T. and Shvartsman, S. (2006) Quantitative models of developmental pattern formation. Developmental Cell, 11: 289-300.

134. Kicheva, A., Pantazis, P., Bollenbach, T., Kalaidzidis, Y., Bittig, T., Julicher, F. and Gonzalez-Gaitan, M. (2007) Kinetics of morphogen gradient formation. Science, 315: 521- 525.

135. Eldar, A., Dorfman, R., Weiss, D., Ashe, A., Shilo, B. and Barkai, N. (2002) Robustness of the BMP morphogen gradient in Drosophila embryonic patterning. Nature, 19: 304-308.

136. Einem, S., Schwarz, E. and Rudolph, R. (2010) A novel two-step renaturation procedure for efficient production of recombinant BMP-2. Protein Expression and Purification, 73: 65-69.

161

137. Ikeya, M., Fukushima, K., Kawada, M., Onishi, S., Furuta, Y., Yonemura, S., Kitamura, T., Nosaka, T. and Sasai, Y. (2010) Cv2, functioning as a pro-BMP factor via twisted gastrulation, is required for early development of nephron precursors. Developmental Biology, 337: 405-414.

138. Dromann, D. et al. (2010) The TGF-beta-pseudoreceptor BAMBI is strongly expressed in COPD lungs and regulated by nontypeable Haemophilus influenza. Respiratory Research, 11: 67.

139. Ho, C., Zhou, X., Ishina, Y. and Bernard, D. (2011) Mechanisms of bone morphogenetic protein 2 (BMP2) stimulated inhibitor of DNA binding 3 (Id3) transcription. Molecular and Cellular Endocrinology, 332: 242-252.

140. Gratsch, T. and O’Shea, S. (2002) Noggin and chordin have distinct activities in promoting lineage commitment of mouse embryonic stem (ES) Cells. Developmental Biology, 245: 83-94.

141. Zakin, L., Reversade, B., Kuroda, H., Lyons, K. and Robertis, E. (2005) Sirenomelia in BMP7 and Tsg compound mutant mice: requirement for BMP signalling in the development of ventral posterior mesoderm. Development, 132: 2489-2499.

142. Webb, T. et al (2012) X-linked megalocornea caused by mutations in CHRDL1 identifies an essential role for ventroptin in anterior segment development. The Americal Journal of Human Genetics, 90: 247-259.

143. Billington, C. et al. (2011) The molecular and cellular basis of variable craniofacial phenotypes and their genetic rescue in twisted gastrulation mutant mice. Developmental Biology, 355: 21-31.

144. Yang, Y., Anderson, R. and Klingensmith, J. (2010) BMP antagonism protects nodal signalling in the gastrula to promote the tissue interactions underlying mammalian forebrain and craniofacial patterning. Human Molecular Genetics, 19: 3030-3042.

162

145. Klingensmith, J., Matsui, M., Yang, Y. and Anderson, R. (2010) Roles of bone morphogenetic protein signalling and its antagonism in holoprosencephaly. American Journal of Medical Genetics, 154: 43-51.

146. Kauver, E. et al (2011) Minimal evidence for a direct involvement of twisted gastrulation homolog 1 (TWSG1) gene in human holoprosencephaly. Molecular Genetics and Metabolism, 102: 470-480.

147. Choi, Y., Kim, S., Park, K., Oh, S., Seo, J., Shin, S., Kim, J. and Kim, Y. (2011) The serum bone morphogenetic protein-2 level in non-small-cell lung cancer patients. Med Oncol, 29: 582-588.

148. Lee, K. et al (2011) Effects of the bone morphogenetic protein binding protein spp24 (secreted phosphoprotein 24kD) on the growth of human lung cancer cells. Journal of Orthopaedic Research, 29: 1712-1718.

149. Slattery, M., Lundgreen, A., Herrick, J., Kadlubar, S., Caan, B., Potter, J. and Wolff, R. (2011) Genetic variation in bone morphogenetic protein and colon and rectal cancer. International Journal of Cancer, 130: 653-654.

150. Hatten, M. And Roussel, M. (2011) Development and cancer of the cerebellum. Trend in Neuroscience, 34: 134-142.

151. Secondini, C., Wetterwald, A., Schwaninger, R., Thalmann, G. And Cecchini, M. (2011) The role of BMP Signalling in the development of prostate cancer osteolytic bone metastasis. PLoS, 6: e16078

152. Peart, T., Corres, R., Valdes, Y., DiMattia, G. and Shepherd, T. (2012) BMP signalling controls the malignant potential of ascites-derived human epithelial ovarian cancer spheroids via AKT kinase activation. Clin Exp Metastasis, 29: 293-313.

153. Cunha, S. and Pietras, K. (2011) ALK1 as an emerging target for anti-angiogenic therapy of cancer. Blood, 21: 451-472.

163

154. Maegdefrau, U. and Bosserhoff, A. (2012) BMP activated Smad signalling strongly promotes migration and invasion of hepatocellular carcinoma cells. Experimental and Molecular Pathology, 92: 74-81.

155. Johnston, J., Al-Bahrani, R., Abuetabh, Y., Chiu, B., Forsman, C., Nagamori, S., Leng, R., Petryk, A. and Sergi, C. (2012) Twisted gastrulation expression in cholangiocellular and hepatocellular carcinoma. J Clin Pathology, 65: 945-948.

156. Davies, M., Lund, R. and Hruska, K. (2003) BMP-7 is an efficacious treatment of vascular calcification in a murine model of atherosclerosis and chronic renal failure. J Am Soc Nephrol, 14: 1559-1567.

157. Nakagawa, Y. et al (2010) Paracrine osteogenic signals via bone morphogenetic protein-2 accelerate the atherosclerotic intimal calcification in vivo. Journal of the American Heart Association, 30: 1908-1915.

158. Hayashi, M., Muneta, T., Ju, Y., Mochizuki, T. and Sekiya, I. (2008) Weekly intra- articular injections of bone morphogenetic protein-7 inhibits osteoarthritis progression. Arthritis Research and Therapy, 10: 118.

159. Sekiya, I., Tang, T,. Hayashi, M., Morito, T., Ju, Y., Mochizuki, T. and Muneta ,T. (2009) Periodic knee injections of BMP-7 delay cartilage degeneration induced by excessive running in rats. Journal of Orthopaedic Research, 27: 1088-1092.

160. Kloen, P., Lauzier, D. and Hamdy, R. (2012) Co-expression of BMPs and BMP- inhibitors in human fractures and non-unions. Bone, 51: 59-68.

161. Wellington, K., Hsu, M., Jeffrey, C. and Wang, M. (2008) The use of bone morphogenetic protein in spine fusion. The Spine Journal, 8: 419-425.

162. Hsiao, W. et al. (2010) Intramarrow BMP4 delivery improves local bone quality in femurs of ovariectomized rabbits. Journal of Periodontology, 82: 854-862.

164

163. Ratanavaporn, J., Furuya, H., Kohara, H. and Tabata, Y. (2011) Synergistic effects of the dual release of stromal cell-derived factor-1 and bone morphogenetic protein-2 from hydrogels in bone regeneration. Biomaterials, 32: 2797-2811.

164. Groppe, J., Wu, J., Shore, E. And Kaplan, F. (2011) In vitro analysis of the dysregulated R206H ALK2 kinase-FKBP12 interaction associated with heterotypic ossification in FOP. Cells Tissues Organs, 194: 291-295.

165. Shore, E. et al (2006) A recurrent mutation in the BMP type I receptor ACVR1 causes inherited and sporadic fibrodysplasia ossificans progressive. Nature Genetics, 38: 525-527.

166. Shen, Q. et al (2009) The Fibrodysplasia Ossificans Progressiva R206H ACVR1 mutation activates BMP-independent chondrogenesis and zebrafish embryo ventralization. The Journal of Clinical Investigation, 119: 3462-3472.

167. Calori, G., Donati, D., Bella, C. and Tagliabue, L. (2009) Bone morphogenetic proteins and tissue engineering: future direction. Injury, 40: 67-76.

168. Ehlert, N., Hoffmann, A., Luessenhop, T., Gross, G., Mueller, P., Stieve, M., Lenarz, T. and Behrens, P. (2011) Amino-modified silica surfaces efficiently immobilize bone morphogenetic protein 2 (BMP2) for medical purposes. Acta Biomaterialia, 7: 1772-1779.

169. Pham, L., Beyer, K., Jensen, E., Rodriguez, J., Davydova, J., Yamamoto, M., Petryk, A., Gopalakrishnan, R. and Mansky, K. (2011) Bone morphogenetic protein 2 signaling in osteoclasts is negatively regulated by the BMP antagonist, twisted gastrulation. Journal of Cellular Biochemistry, 112: 793-803.

170. Hurtig, M., Chubinskaya, S., Dickey, J. and Rueger, D. (2008) BMP-7 protects against progression of cartilage degeneration after impact injury. Journal of Orthopaedic Research, 27: 602-611.

171. Mines, D., Gu, Y., Doug, K. and Cooper, G. (2011) Recombinant human bone morphogenetic protein-2 and pancreatic cancer: a retrospective cohort study. Pharmacoepidemiology and Drug Safety, 20: 111-118.

165

172. Song, K. (2010) Identification of a key residue mediating bone morphogenetic protein (BMP)-6 resistance to noggin inhibition allows for engineered BMPs with superior agonist activity. JBC, 285: 12169-12180.

173. Chu, M., Lu, L., Feng, T., Di, R., Cao, G., Wang, P., Fang, L., Ma, Y. and Li, K. (2010) Polymorphism of bone morphogenetic protein 4 gene and its relationship with litter size of Jining Grey goats. Mol Biol Rep, 38: 4315-4320.

174. Moll, F. et al (2006) Chordin is underexpressed in ovarian tumours and reduces tumour cell motility. The FASEB Journal, 20: 240-250.

175. Berry, R. (2009) Structure and functional regulation of mammalian BMP-1/TLD family proteases. The University of Manchester.

176. Whitmore, L. and Wallace, B.A. (2004) DICHROWEB: an online server for protein secondary structure analyses from circular dichroism spectroscopic data. http://dichroweb.cryst.bbk.ac.uk/html/home.shtml

177. Sedfit and Sedphat: analytical software for the interpretation of AUC data. http://www.analyticalultracentrifugation.com/sedphat/references.htm

178. Schuck, P. (2003) On the analysis of protein self-association by sedimentation velocity analytical ultracentrifugation. Analytical Biochemistry, 320:104-124

179. SEDNTERP: analytical software for the calculation of buffer and peptide properties from composition and sequence. http://bitcwiki.sr.unh.edu/index.php/Main_Page

180. Tang, G., Baldwin, P., Mann, D., Jiang, W., Rees, I. and Ludtke, S. (2007) EMAN2: an extensible image processing suite for electron microscopy. J Structural Biol, 157: 38-46.

181. Konarev, P., Volkov, V., Sokolova, A., Koch, M. and Svergun, D. (2003). PRIMUS - a Windows-PC based system for small-angle scattering data analysis. J Appl Cryst. 36: 1277- 1282.

182. Svergun, D. (1999) Restoring low resolution structure of biological macromolecules from solution scattering using simulated annealing. Biophys J. 76: 2879-2886.

166

183. Svergun, D., Petoukhov, M. and Koch, M. (2001) Determination of domain structure of proteins from X-ray solution scattering. Biophys. J., 80: 2946-2953.

184. Volkov, V. and Svergun, D. (2003). Uniqueness of ab-initio shape determination in small-angle scattering. J. Appl. Cryst. 36: 860-864.

185. Torre, J., Valpuesta, J. and Carrascosa, J. (2001) HYDROMIC: Prediction of hydrodynamic properties of rigid macromolecular structures obtained from electron- microscopy images. Eur. Biophys. J, 30: 457-462

186. Ortega, A., Amoros, D. and Torre, J (2011) Prediction of hydrodynamic and other solution properties of rigid proteins from atomic- and residue-level models. Biophys. J. 101: 892-898

187. Lebowitz, J., Lewis, M. And Schuck, P. (2002) Modern analytical ultracentrifugation in protein science: A tutorial review. Protein Sci, 11: 2067-2079.

188. Williams, M., Phan, I., Harvey, T., Rostagno, A., Gold, L. and Campbell, I. (1994) Solution structure of a pair of fibronectin type 1 modules with fibrin binding activity. J Mol Biol, 235: 1302–1311.

189. O'Leary, J., Hamilton, J., Deane, C., Valeyev, N., Sandell, L. and Downing, A. (2004) Solution structure and dynamics of a prototypical chordin-like cysteine-rich repeat (von Willebrand Factor type C module) from collagen IIA. JBC, 279: 53857–53866

190. Piccolo, S., Sasai, Y., Lu, B. And Robertis, E. (1996) Dorsoventral patterning in Xenopus: inhibition of ventral signals by direct binding of chordin to BMP-4. Cell, 86: 589- 598.

191. Haltiwanger, R. and Lowe, J. (2004) Role of glycosylation in development. Annu Rev Biochem, 73: 491-537.

167

Appendix I: VECTOR MAPS

pCR2.1-TOPO Cloning Vector

F1 Ori: replication origin in phage

Kanamycin: kanamycin resistance ORF

Ampicillin: ampicillin resistance ORF pUC ori: replication origin in bacteria

Plac: wild type pLac promoter

MCS: multiple cloning site lacZα:encodes β-galactosidase (Nb. LacIq repressor present in host strain)

168

pCEP-Pu/pCEP-Pu-AC7 Expression Vector

OriP/EBNA1: replication origin in eukaryotic cell lines stimulated by EBNA1 enabling maintenance in EBNA1 positive cells

AMP: ampicillin resistance ORF

ColE1: replication origin in bacteria

PSV40: promoter for pac in eukaryotic cell lines

Pac: pac gene conferring puromycin resistance

SV40pA: 3’-polyadenylation signal

CMV: enhancer

BM40 signal: signal peptide for secretion, PcepPu-AC7 vector only

169

Appendix II: BUFFER COMPOSITIONS

All buffers listed, except manufacturer supplied, were prepared using ddH2O purified using the Mili-Q Water System (Millipore) and degassed.

1. Agar Plates: 1% tryptone w/v, 1% yeast extract w/v, 2% agar w/v, 10mM NaCl. Solution autoclaved, after cooling 100µg ml-1 ampicillin or chloramphenicol added.

2. Ampicillin LB Broth: 1.6% tryptone w/v, 1.6% yeast extract w/v, 15mM NaCl. Solution autoclaved, after cooling 100µg ml-1 ampicillin or chloramphenicol added.

3. 1X Buffer 4 (New England Biolabs): 50mM potassium acetate, 20mM Tris acetate, 10mM magnesium acetate, 1mM DTT, pH7.9.

4. Buffer P1 (Qiagen): 50mM Tris-Cl, 10mM EDTA, 100µg ml-1 RNase A, pH8.

5. DNA Elution Buffer: 50mM Tris, pH8 (Qiagen).

6. Gel Drying Solution: 20% methanol v/v, 2% glycerol v/v.

7. 1X Gel Loading Dye Blue (New England Biolabs) 62.5mM Tris-HCl, 2% SDS, 10% glycerol and 0.01% bromophenol blue w/v, pH6.8.

8. HisTrap Binding Buffer: 10mM Tris-HCl, 500mM NaCl, 10mM imidazole, pH7.4.

9. HisTrap Dialysis Buffer: 10mM Tris-HCl, 500mM NaCl, pH7.4.

10. HisTrap Elution Buffer: 10mM Tris-HCl, 500mM NaCl, 500mM imidazole, pH7.4.

11. MES Buffer: 50mM 2-(N-morpholino)ethanesulfonic acid (MES), 5mM Tris-HCl, 0.1% SDS w/v, 1mM EDTA, pH7.3.

12. MOPS Buffer: 50mM 3-(N-morpholino) propanesulfonic acid (MOPS), 50mM Tris- HCl, 0.1% SDS w/v, 1mM EDTA, pH7.7.

13. Phosphate Buffered Saline: 5mM sodium phosphate (3.5:1 ratio

Na2HPO4:NaH2PO4.H2O), 150mM NaCl, pH7.4.

14. 1X PNK Buffer (New England Biolabs): 70mM Tris-HCl, 10mM MgCl2, 5mM dithiothreitol, pH7.6.

15. Protease inhibitor cocktail: 0.3mM N-ethylmaleimide, 0.5mM phenylmethyl sulphonyl fluoride (PMSF).

170

16. SDS Sample Loading Buffer: 62.5mM Tric-HCl, 2% SDS, 0.01% bromophenol blue w/v, 25% glycerol v/v, 5% β-mercaptoethanol v/v, pH6.8.

17. SOC Medium: 2% tryptone, 0.5% yeast extract w/v, 10mM NaCl, 2.5mM KCl, 10mM

MgCl2, 10mM MgSO4, 20mM glucose.

18. SPR Analyte Buffer: 10mM HEPES, 150mM NaCl, 0.005% TWEEN-20 pH7.4.

19. SPR Immobilization Buffer: 50mM sodium acetate, pH 5.5

20. SPR Regeneration Buffer: 1M NaCl, 50mM NaOH.

21. 1X T4 DNA Ligase Reaction Buffer: 50mM Tris-HCl, 10mM MgCl2, 10mM dithiothreitol, 1mM ATP, pH7.5.

22. Transfer Buffer: 25mM Tris-HCl, 192nM glycine, 20% methanol, pH8.3.

23. Tris Buffered Saline: 10mM Tris-HCl, 150mM NaCl, pH7.4.

24. Western Blot Blocking Solution: 5% milk w/v, 0.01% TWEEN-20 in PBS, pH7.4.

171

Appendix III: Construct Sequence and Properties

Annotated Human Chordin DNA and Peptide Sequence

Figure Appendix 1: Annotated sequence of human chordin as determined from cDNA and corresponding amino acid residues. The signal sequence is shown in yellow and the signal peptide cleavage site is marked by a pink arrow39. Blue represents vWC domains and green represents the CHRD domains40. The SD1 and SD2 subdomain boundaries of the BMP binding domains vWC1 and vWC3 are marked by blue stars39. Splice variants are shown in cyan39. Predicted N-glycosylation sites are underlined in orange. The tolloid cleavage sites predicted from cross-species conservation are shown in red with a red arrow marking the point of cleavage43, 61.

172

Full Length Chordin Construct Peptide Sequence

M P S L P A P P A P L L L L G L L L L G S R P A R G A (NATURAL SIGNAL PEPTIDE) G P E P P V L P I R S E K E Q K L I S E E D L (myc tag) E P L P V R G A A G C T F G G K V Y A L D E T W H P D L G E P F G V M R C V L C A C E A P Q W G R R T R G P G R V S C K N I K P E C P T P A C G Q P R Q L P G H C C Q T C P Q E R S S S E R Q P S G L S F E Y P R D P E H R S Y S (tolloid cleavage site) D R G E P G A E E R A R G D G H T D F V A L L T G P R S Q A V A R A R V S L L R S S L R F S I S Y R R L D R P T R I R F S D S N G S V L F E H P A A P T Q D G L V C G V W R A V P R L S L R L L R A E Q L H V A L V T L T H P S G E V W G P L I R H R A L A A E T F S A I L T L E G P P Q Q G V G G I T L L T L S D T E D S L H F L L L F R G L L E P R S G G L T Q V P L R L Q I L H Q G Q L L R E L Q A N V S A Q E P G F A E V L P N L T V Q E M D W L V L G E L Q M A L E W A G R P G L R I S G H I A A R K S C D V L Q S V L C G A D A L I P V Q T G A A G S A S L T L L G N G S L I Y Q V Q V V G T S S E V V A M T L E T K P Q R R D Q R T V L C H M A G L Q P G G H T A V G I C P G L G A R G A H M L L Q N E L F L N V G T K D F P D G E L R G H V A A L P YC G H S A R H D T L P V P L A G A L V L P P V K S Q A A G H A W L S L D T H C H L H Y E V L L A G L G G S E Q G T V T A H L L G P P G T P G P R R L L K G F Y G S E A Q G V V K D L E P E L L R H L A K G M A S L M I T T K G S P R G E L R G Q V H I A N QC E V G G L R L E A A G A E G V R A L G A P D T A S A A P P V V P G L P A L A P A K P G G P G R P R D P N T C F F E G Q Q R P H G A R W A P N Y D P L C S L C T C Q R R T V I C D P V V C P P P S C P H P V Q A P D Q C C P V C P E K Q D V R D L P G L P R S R D P G E G C Y F D G D R S W R A A G T R W H P V V P P F G L I K C A V C T C K G G T G E V H C E K V Q C P R L A C A Q P V R V N P T D C C K Q C P V G S G A H P Q L G D P M Q A (tolloid cleavage site) D G P R G C R F A G Q W F P E S Q S W H P S V P P F G E M S C I T C R C G A G V P H C E R D D C S L P L S C G S G K E S R C C S R C T A H R R P A P E T R T D P E L E K E A E G S L V P R G S (thrombin cleavage site) H H H H H H (poly-histidine tag) Stop

173

Twisted Gastrulation Construct DNA Sequence

atgaagttacactatgttgctgtgcttactctagccatcctgatgttcctgacatggcttccagaatcactgagctgtaacaaagcact ctgtgctagtgatgtgagcaaatgcctcattcaggagctctgccagtgccggccgggagaaggcaattgtcctgctgtaaggagt gcatgctgtgtcttggggccctttgggacgagtgctgtgactgtgttggtatgtgtaatcctcgaaattatagtgacacacctccaactt caaagagcacagtggaggagctgcatgaaccgatcccttctctcttccgggcactcacagaaggagatactcagttgaattgga acatcgtttctttccctgttgcagaagaactttcacatcatgagaatctggtttcatttttagaaactgtgaaccagccacaccaccag aatgtgtctgtccccagcaataatgttcacgcgccttattccagtgacaaagaacacatgtgtactgtggtttattttgatgactgcatg tccatacatcagtgtaaaatatcctgtgagtccatgggagcatccaaatatcgctggtttcataatgcctgctgcgagtgcattggtc cagaatgtattgactatggtagtaaaactgtcaaatgtatgaactgcatgttttaaagctggtccccagaggagccaccaccatca ccaccattaa

Twisted Gastrulation Construct Peptide Sequence

M K L H Y V A V L T L A I L M F L T W L P E S L S C N K A L C A S D V S K C L I Q E L C Q C R P G E G N C S C C K E C M L C L G A L W D E C C D C V G M C N P R N Y S D T P P T S K S T V E E L H E P I P S L F R A L T E G D T Q L N W N I V S F P V A E E L S H H E N L V S F L E T V N Q P H H Q N V S V P S N N V H A P Y S S D K E H M C T V V Y F D D C M S I H Q C K I S C E S M G A S K Y R W F H N A C C E C I G P E C I D Y G S K T V K C M N C M F L V P R G S (thrombin cleavage site) H H H H H H (poly-histidine tag) Stop

174

Appendix IV: Mass Spectrometry Peptide Hits

IPI Human Database peptide hits are highlighted in yellow. No hits are expected in the red sequence for chordin ∆N, while none are predicted in the purple sequence for chordin ∆C.

Chordin-FN Upper Band mpslpappapllllgllllgsrpargagpeppvlpirsekeplpvrgaagctfggkvyaldetwhpdlgepfgvmrcvlcaceapq wgrrtrgpgrvsckNIKPECPTPACGQPRqlpghccqtcpqerssserqpsglsfeyprdpehrsysdrgepgaeerar gdghtdfvalltgprsqavararvsllrsslrfsisyrrldrptrirfsdsngsvlfehpaaptqdglvcgvwravprlslrllraeqlhvalvt lthpsgevwgplirhralaaetfsailtlegppqqgvggitlltlsdtedslhflllfrglleprsggltqvplrlqilhqgqllrelqanvsaqe pgfaevlpnltvqemdwlvlgelqmalewagrpglrisghiaarkscdvlqsvlcgadalipvqtgaagsasltllgngsliyqvqvv gtssevvamtletkpqrrdqrtvlchmaglqpgghtavgicpglgargahmllqnelflnvgtkdfpdgelrghvaalpycghsar hdtlpvplagalvlppvksqaaghawlsldthchlhyevllaglggseqgtvtahllgppgtpgprrllkGFYGSEAQGVVKd lepellrhlakgmaslmittkgsprgelrgqvhianqcevgglrleaagaegvralgapdtasaappvvpglpalapakpggpgr prdpntcffegqqrphgarwapnydplcslctcqrrtvicdpvvcpppscphpvqapdqccpvcpekqdvrdlpglprsrdpge gcyfdgdrswraagtrwhpvvppfglikcavctckggtgevhcekvqcprLACAQPVRVNPTDCCKQCPVGSGA HPQLGDPMQADGPRgcrfagqwfpesqswhpsvppfgemscitcrcgagvphcerddcslplscgsgkesrccsrct ahrrpapetrtdpelekeaegs

Chordin-FN Lower Band mpslpappapllllgllllgsrpargagpeppvlpirsekeplpvrgaagctfggkvyaldetwhpdlgepfgvmrcvlcaceapq wgrrtrgpgrvscknikpecptpacgqprqlpghccqtcpqerssserqpsglsfeyprdpehrsysdrgepgaeerARGD GHTDFVALLTGPRsqavararvsllrsslrfsisyrrldrptrirfsdsngsvlfehpaaptqdglvcgvwravprlslrllraeqlh valvtlthpsgevwgplirhralaaetfsailtlegppqqgvggitlltlsdtedslhflllfrglleprSGGLTQVPLRLQILHQGQ LLRelqanvsaqepgfaevlpnltvqemdwlvlgelqmalewagrpglrisghiaarkscdvlqsvlcgadalipvqtgaagsa sltllgngsliyqvqvvgtssevvamtletkpqrrdqrtvlchmaglqpgghtavgicpglgargahmllqnelflnvgtkdfpdgelr GHVAALPYCGHSARhdtlpvplagalvlppvksqaaghawlsldthchlhyevllaglggseqgtvtahllgppgtpgprr LLKGFYGSEAQGVVKDLEPELLRhlakgmaslmittkgsprgelrGQVHIANQCEVGGLRLEAAGAE GVRalgapdtasaappvvpglpalapakpggpgrprDPNTCFFEGQQRPHGARWAPNYDPLCSLCTCQ RrtvicdpvvcpppscphpvqapdqccpvcpekqdvrdlpglprsrdpgegcyfdgdrswraagtrwhpvvppfglikCAV CTCKGGTGEVHCEKVQCPRLACAQPVRVNPTDCCKQCPVGSGAHPQLGDPMQADGPRG CRfagqwfpesqswhpsvppfgemscitcrcgagvphcerDDCSLPLSCGSGKesrccsrctahrrpapetrtdpele keaegs

Chordin ∆C mpslpappapllllgllllgsrpargagpeppvlpirsekeplpvrgaagctfggkVYALDETWHPDLGEPFGVMRCV LCACEAPQWGRrtrgpgrvsckNIKPECPTPACGQPRqlpghccqtcpqerssserQPSGLSFEYPRdp ehrSYSDRGEPGAEERARGDGHTDFVALLTGPRsqavararvsllrsslrfsisyrrldrptrirFSDSNGSVL FEHPAAPTQDGLVCGVWRavprlslrllraeqlhvalvtlthpsgevwgplirhralaaetfsailtlegppqqgvggitlltls

175

dtedslhflllfrglleprSGGLTQVPLRLQILHQGQLLRelqanvsaqepgfaevlpnltvqemdwlvlgelqmalew agrpglrisghiaarkscdvlqsvlcgadalipvqtgaagsasltllgngsliyqvqvvgtssevvamtletkpqrrdqrTVLCHM AGLQPGGHTAVGICPGLGARGAHMLLQNELFLNVGTKDFPDGELRGHVAALPYCGHSARH DTLPVPLAGALVLPPVKsqaaghawlsldthchlhyevllaglggseqgtvtahllgppgtpgprrLLKGFYGSEA QGVVKDLEPELLRhlakGMASLMITTKGSPRGELRGQVHIANQCEVGGLRLEAAGAEGVRal gapdtasaappvvpglpalapakpggpgrprdpntcffegqqrphgarWAPNYDPLCSLCTCQRrtvicdpvvcppp scphpvqapdqccpvcpekqdvrdlpglprSRDPGEGCYFDGDRswraagtrWHPVVPPFGLIKcavctckg gtgevhcekvqcprlacaqpvrvnptdcckqcpvgsgahpqlgdpmqadgprgcrfagqwfpesqswhpsvppfgemsci tcrcgagvphcerddcslplscgsgkesrccsrctahrrpapetrtdpelekeaegs

Chordin ∆N mpslpappapllllgllllgsrpargagpeppvlpirsekeplpvrgaagctfggkvyaldetwhpdlgepfgvmrcvlcaceapq wgrrtrgpgrvscknikpecptpacgqprqlpghccqtcpqerssserqpsglsfeyprdpehrsysdrgepgaeerARGD GHTDFVALLTGPRsqavararvsllrsslrFSISYRRldrptrirFSDSNGSVLFEHPAAPTQDGLVCGV WRavprlslrllraeqlhvalvtlthpsgevwgplirhralaaetfsailtlegppqqgvggitlltlsdtedslhflllfrglleprSGGLT QVPLRLQILHQGQLLRelqanvsaqepgfaevlpnltvqemdwlvlgelqmalewagrpglrisghiaarkscdvlqsv lcgadalipvqtgaagsasltllgngsliyqvqvvgtssevvamtletkpqrrdqrTVLCHMAGLQPGGHTAVGICPG LGARGAHMLLQNELFLNVGTKDFPDGELRGHVAALPYCGHSARHDTLPVPLAGALVLPPV KsqaaghawlsldthchlhyevllaglggseqgtvtahllgppgtpgprrLLKGFYGSEAQGVVKDLEPELLRhlak gmaslmittkgsprgelrGQVHIANQCEVGGLRleaagaegvralgapdtasaappvvpglpalapakpggpgrprd pntcffegqqrphgarWAPNYDPLCSLCTCQRrtvicdpvvcpppscphpvqapdqccpvcpekqdvrdlpglprS RDPGEGCYFDGDRSWRAAGTRWHPVVPPFGLIKCAVCTCKGGTGEVHCEKVQCPRLACA QPVRvnptdcckQCPVGSGAHPQLGDPMQADGPRgcrfagqwfpesqswhpsvppfgemscitcrcgagv phcerDDCSLPLSCGSGKesrccsrctahrrpapetrtdpelekeaegs

Twisted Gastrulation mklhyvavltlailmfltwlpeslscnkALCASDVSKCLIQELCQCRPGEGNCSCCKecmlclgalwdeccdcv gmcnprNYSDTPPTSKSTVEELHEPIPSLFRaltegdtqlnwnivsfpvaeelshhenlvsfletvnqphhqnvsv psnnvhapyssdkehmctvvyfddcmsihqckISCESMGASKYRWFHNACCECIGPECIDYGSKtvkcm ncmf

176