UNLOCKING SYNTHETIC BIOMATERIALS:

MANUFACTURE OF STRUCTURAL BIOGENIC MATERIALS VIA 3D-PRINTED

ARRAYS OF BIOENGINEERED CELLS

A DISSERTATION

SUBMITTED TO THE DEPARTMENT OF MECHANICAL ENGINEERING

AND THE COMMITTEE ON GRADUATE STUDIES

OF STANFORD UNIVERSITY

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

Diana Marron Gentry

February 2015

http://creativecommons.org/licenses/by-nc-sa/3.0/us/

This dissertation is online at: http://purl.stanford.edu/hp337sb3635

© 2015 by Diana Marron Gentry. All Rights Reserved.

Re-distributed by Stanford University under license with the author.

This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 United States License.

ii

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Thomas Kenny, Primary Adviser

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Lynn Rothschild, Co-Adviser

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Sheri Sheppard

Approved for the Stanford University Committee on Graduate Studies.

Patricia J. Gumport, Vice Provost for Graduate Education

This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.

iii

iv

Abstract

Many complex, biologically-derived materials have extremely useful properties (think wood or silk), but suf-

fer from production, manufacturing, and processing limitations. Cells naturally specialize in making complex

biomaterials on a micro scale. This work explores a technology concept combining this ability with the re-

cently emergent fields of synthetic biology and additive manufacturing in which the end product is a nonliving

biomaterial with human-designed shape, structure and composition. A 3D printer capable of printing living

cells with near single-cell resolution is used to create 3D-structured arrays of cells bioengineered to secrete

different materials. The cells produce the materials in rates and quantities determined by human-controlled

stimuli. A proof of concept is described consisting of a two-material array of non-structural proteins. Each

step in the end-to-end demonstration has been proven to reach the minimum level of critical functionality.

Adding a vast new set of biomaterials, both natural and newly designed, to the traditional metal, plastic and

ceramic material toolkit has applications limited by our future imagination; this work is an important first

step on that path.

v

vi

Preface

I have had the opportunity to work on a number of distinct projects as part of my Ph.D. thesis work. My

primary thesis work has been on a proof of concept of a novel manufacturing method for non-living, structural

biomaterials, which composes the bulk of the thesis (Section I). Because the other projects gave me experience

and insights which fed into the biomaterials development, I have also included a writeup of the some of the

work originally done for these other projects. I hope that these sections will help answer some of the most

common questions I receive, including “What does a mechanical engineer know about synthetic biology?”

My thesis sections are ordered by importance to the primary work, rather than chronological order. For

clarity, and also to make the attributions of the various laboratories and working groups I’ve participated in

explicit, a brief summary of my graduate trajectory is presented below.

I began my thesis work in the Telerobotics Lab under Gunter Niemeyer in late 2006. I worked jointly with

other students, particularly Probal Mitra, on several studies involving haptics, model-based mediation, wave

variable control, and user task completion; this joint work is presented in Section II. I was individually

responsible for the design and implementation of a user study on task completion in variously virtualized

environments; this work is presented following the joint work in the same section.

After circumstances ended that project track in 2008-2009, I became involved in an effort led by Lynn Roth-

schild, of NASA Ames Research Center, to explore the upper extent of Earth’s biosphere. This eventually

forked into two distinct projects, the design of a survey system (flight payload hardware, sampling guide-

lines, and analysis protocols) for high-altitude biological sampling and the implementation of an extended

settling time suspension chamber for lab-based aerobiological testing. Both of these projects are described in

a separate section not included as part of this document, but available on request.

vii

The high-altitude biological sampling project originated in early 2008 as an effort to design a payload for an

amateur rocketry non-profit, Mavericks Civilian Spaceflight. The project was quickly expanded to include

development of sampling guidelines and analysis protocols after the interdependencies between the three

parts became apparent. I was responsible for the design, construction and testing of the payload, including

building appropriate calibration hardware; for the development and testing of sampling guidelines; and for

working with a yet-to-be-appointed biologist to develop and refine the analysis protocols. I used a commercial

avionics board from G-Wiz, including its µC/OS-II-based operating system, for the first iteration of the

sampling payload. After Mavericks withdrew as our flight provider in 2010, I continued the project by

redesigning the payload for use with high-altitude balloons using open-source hardware and software. I re-

modified this second iteration to share payload space on a sounding rocket flight via UP Aerospace with the

PhoneSat team in 2011. I also performed ground-based air sampling and analysis in 2010-2011 in Brazil via

an arrangement with the Universidade de Sao Paulo.

The extended settling time suspension chamber project originated in mid-2009 in discussions of how lab-

based tests might be able to complement the sampling effort. A primary goal was to determine whether

existing organisms go through a complete life cycle while airborne (so-called ’aerophiles’) or whether the

role of the atmosphere in Earth’s biosphere is limited to transport and short-term residence. After consider-

able further discussion in 2010 and 2011, the experimental phase of this project was initiated in April 2012

with the goal of designing a lab-based device to demonstrate that microbes can metabolize and reproduce in

the absence of stable surface liquid, or, in other words, while aerosolized. I identified and evaluated several

different possible approaches and chose a final design based on electrostatic suspension and optical detec-

tion. I built and tested the device using commercial hardware from PASCO as a foundation. The microbes

intended for use with the detection system were tagged first with genetic parts identified and transferred by the

2011 Stanford-Brown iGem Team (a luciferin/luciferase indicator) and later with parts isolated by the 2012

Stanford-Brown iGem Team (an RFP/GFP dual indicator). I performed characterization and quantification

of these indicators.

The biology I had begun to learn as part of these projects led to my being assigned to another project in

2011, to 3D-print a multicellular organism. The goal was to take a relatively undifferentiated multicellular

organism, such as a placozoan or a poriferan, characterize its 3D structure, dissociate its cells, 3D print the

cells, and show that the cells could reaggregate into an organism whose structure was correlated with the

3D-printed shape. (That work is not included in this document, but the report is available upon request.) I

had an insight during this project, after realizing that the non-living, extracellular materials created by the

cells within the organism were a significant challenge to extraction and purification of the living cells. What

viii

if creating such materials were the goal? Non-living, structural biomaterials have a wide variety of promising

properties, but generally cannot be manufactured in human-useful configurations.

Based on this insight, I wrote a NASA Ames Center Innovation Fund Grant proposal in 2012, with Lynn sub-

mitting it as principal investigator. The grant was awarded, which allowed the addition to the team of Ashley

Micks, another Stanford engineering graduate student with whom I worked closely throughout the project’s

duration. The expanded team wrote, submitted, and received a competitive renewal of the CIF in early 2013

and a NASA Innovative Advanced Concepts grant later that same year. During the course of this work, I

received significant advice and input (particularly on the synthetic biology front) from Kosuke Fujishima and

Lucas Hartsough; areas which were joint or partial responsibilities are indicated in their respective sections.

This project, which became my thesis project, is presented in Section I.

ix

x

Contents

Abstract v

Preface vii

I Structural Biomaterials from 3D-Printed Cell Arrays 1

1 Background 3

1.1 The Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.3 Technical Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.4 State of the Art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.5 Project & Design Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2 Synthetic Biomaterials: A Proof of Concept 21

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.2 Peformance Metric Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.3 Material Template Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.4 Cell Engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.5 Print Medium Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

xi

2.6 3-D Printing System Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

2.7 Print Substrate Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

2.8 Printing Demonstration & Performance Characterization . . . . . . . . . . . . . . . . . . . 57

2.9 Results & Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3 Future Development Pathways 63

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.2 Short Term Implementations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.3 Long Term Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

4 Mission Context Feasibility/Benefit Analyses 77

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

4.2 An ISS Replacement Part . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

4.3 A Long-Term Mars Habitat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

4.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

5 Summary & Conclusions 105

Appendices 109

A Plasmid Maps & Sequences 111

References 113

II Haptics 129

6 Background 131

6.1 Teleoperation and Humans in the Loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

6.2 Telerobotic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

6.3 Virtual Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

6.4 Telepresence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

6.5 Model-Mediated Teleoperation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

xii

7 User Perception and Preference in Delayed Model-Mediated Telemanipulation 143

7.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

7.2 User Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

7.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

8 User Preference and Performance in Manipulation of Virtual Environments 161

8.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161

8.2 User Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

8.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178

9 Conclusions 181

9.1 Task Performance and System Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

9.2 User Preferences and Behavior . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183

9.3 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184

References 187

xiii

xiv

List of Tables

1.1 Examples of natural structural biomaterials . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1.2 TRL definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.1 Bioprinting performance metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.2 Material selection ratings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.3 Commercial hardware and software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.4 Print parameter (dispersion) test data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

2.5 Performance metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.1 Potential materials for future implementation . . . . . . . . . . . . . . . . . . . . . . . . . 65

3.2 Production stimulus and form factor. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3.3 Material delivery options. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

4.1 In situ resource grades. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

4.2 Material compatibility grades. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

4.3 ISS biomaterial manufacturing requirements . . . . . . . . . . . . . . . . . . . . . . . . . . 81

4.4 Fixed and per unit costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

4.5 Tensile strength and mass comparisons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

4.6 Labor costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

4.7 Habitat skin material options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

xv

4.8 Habitat column material options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

4.9 Material grades . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

4.10 ISRU options and effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

4.11 Labor costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

4.12 Tensegrity compressive material limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

4.13 Tensegrity buckling material limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

4.14 Tensegrity tensile material limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

5.1 TRL advancement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

7.1 Final haptic/visual selections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

8.1 Types of virtualizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164

8.2 Results summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

xvi

List of Figures

1.1 Inputs and outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.2 Multidisciplinary challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.3 Potential space mission context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.4 Biomaterial sizes and scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.1 Saccharomyces cerevisiae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.2 Material type options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

2.3 Promoter concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2.4 Stimulus method options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.5 The GAL1 promoter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

2.6 Delivery method options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.7 Material binding options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2.8 Plasmid assembly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.9 Fluorescence expression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.10 Protein binding test . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.11 Viscosity tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.12 Aggregation tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

2.13 Cell growth demonstration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

xvii

2.14 Expression tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.15 Hardware components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

2.16 Flowchart for bioprinting software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

2.17 Example print parameter test data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2.18 Post-printing survival . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

2.19 Substrate design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

2.20 Cell growth demonstration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

2.21 Cell toxicity demonstration control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

2.22 Cell toxicity demonstration array . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

2.23 Protein binding gel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

2.24 Print test results with beads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

2.25 RFP and GFP yeast print . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

2.26 Design dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.1 Substrate printing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.2 An alternate array construction method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

3.3 Suggested technology development roadmap . . . . . . . . . . . . . . . . . . . . . . . . . . 75

A.1 Plasmid map: yeGFP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

A.2 Plasmid map: yeRFP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

6.1 Elements of teleoperation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

6.2 Systems for teleoperation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

6.3 Interactivity and feedback in teleoperation . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

6.4 Virtual world systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

6.5 Model-mediated telemanipulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

xviii

7.1 The virtual workspace . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

7.2 Pilot test results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

7.3 User preference results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

7.4 Location identification success by method . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

7.5 Impact velocity by method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

7.6 Impact velocity by location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

7.7 Peak force by method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

7.8 Peak force by location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

7.9 Task completion time by location . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

7.10 User strategy, slow and stiff . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

7.11 User strategy, tapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

7.12 User strategy, move and wait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

7.13 User effects, consistent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

7.14 User effects, consistent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

8.1 Drawing task templates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

8.2 Example bounding boxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

8.3 Example virtual session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

8.4 Example drawing measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

8.5 Spacing data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

8.6 Offset data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

8.7 Skew data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173

8.8 Scaling data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174

8.9 Error rate data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

8.10 An ‘unforced’ error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176

8.11 Per-user trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

8.12 Consistency of skew and scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

xix

xx

Part I

Structural Biomaterials from

3D-Printed Cell Arrays

1

2

1 Background

1.1 The Concept

Biomaterials are the natural materials produced by and integrated into living systems, as well as artificial

materials developed to mimic them [145]. They include simple molecules like sugars, complex polymers like

collagen, and organic-inorganic composites ranging from macroscale mixtures like microbial mats [70] to the

molecularly bound biominerals of bone and nacre [21].

Biomaterials have many tantalizing features. In nature, they are manufactured entirely from accessible, com-

mon molecular components extracted without human oversight from air, sunlight, ground- or seawater, soil,

and other organic matter. These basic components are combined with molecular precision into a vast, still

undercharacterized set of materials with different mechanical, electrical, and chemical properties, often with

internally varying properties customized to match the functional requirements at the micron or smaller level.

Structural biomaterials, such as the nonliving parts of bone, tooth, wood, shell, and so forth, are often stronger,

more deformation-resilient, and more fracture-resistant for their mass than traditional human-produced struc-

tural materials such as metals, plastics and ceramics.

Although early human tool use made much use of such structural biomaterials, their modern industrial use is

limited by three major factors. Firstly, they cannot be reliably produced at large scale; a forest, for example,

may hold a large quantity of wood, but the yield of material suitable for structural use will be significantly

reduced by trunk and branch size and shape, knots, grain irregularities, scars from natural damage, and so

forth. Secondly, production and crafting is typically high-overhead compared to other modern alternatives;

producing wool from sheep requires unpolluted pasture, water, and air as well as a significant number of

person-hours for care and collection, a limitation that becomes much more apparent in, for example, a space

mission case. Thirdly, there is a profound lack of tools able do the kind of microscale functional customization

that natural systems perform, leaving the strongest advantage of biomaterials inaccessible to human use.

3

This work is an multidisciplinary investigation of a new concept, a manufacturing technique that combines

3-D printing technology with genetic engineering of living cells for material production. A 3-D printer

with near single cell resolution to deposit a 3-D array of bioengineered cells according to a human-designed

digital template, such as the shape of a desired product. The cells are bioengineered to secrete the molecular

components of the desired biomaterials, such as proteins and polysaccharides, in amounts and rates that are

controlled by external, human-controlled physicochemical cues. The cell array deposits its materials onto

a substrate, which both provides additional material to be integrated (e.g., metal ions) and helps bind the

materials together. Afterward, the cells and non-integrated substrate are washed away, leaving a finished,

nonliving product with microscale structure and precision determined by the original digital template. This

creation of “synthetic biomaterials” would allow humans to work with an vast new set of complex material

building blocks well outside our traditional metal, plastic and ceramic toolkit.

1.2 Motivation

1.2.1 Model Systems for Biomarker Formation

The study of biosignatures, which are markers left behind by the presence of life, is one of the foundations

of both biogeochemistry and astrobiology. Many long-lived biosignatures, such as ancient stromatolites and

trace fossils [138], are the inorganic remainders of organic-inorganic composites (OICs) once formed by

living systems.

Distinguishing between biosignatures and similar but non-biogenic geochemical formations is key to inter-

preting the history of life on Earth. Naturally occurring OICs were likely important in early life chemistry;

microbial production of OICs is responsible for many biogeochemical markers; and the use of OICs in multi-

cellular tissues and organs has been a necessary part of life’s increasing complexity and adaptation. Similarly,

understanding biological systems’ synthesis and incorporation of such composites, and the different situations

in which certain composites are likely to be found, is a keystone of astrobiological biosignature investiga-

tions; developing reliable methods for distinguishing between biogenic and abiogenic inorganic markers, is

one of the top goals of NASA’s astrobiology program [49].

Fortunately, life makes extensive use of many different OICs, ranging from macroscale mixtures like the

extracellular polymer/sediment matrix of microbial mats [70] to the molecularly integrated biominerals of

4

amino acids

ions and salts

simple sugars

piezo-electric

dispenser

head

3D-printed

cell arraysdispensing

controller

positioning

controller

3-axis

micropositioning

system

organic-inorganic

laminate

novel biomaterial

finished biomaterial

composite part

Figure 1.1: By combining the natural ability of cells to make highly complex

biomolecule assemblies from simple molecular base materials inputs with the ability

of modern 3-D printers to create arrays with near single cell precision, it becomes

possible to make non-living, structural biomaterials in human-specified forms.

diatoms, foraminifera, and nacre [21]. However, there is a significant lack of model systems for studying

the biological production of OICs under controlled conditions. Artificial microbial mats have been produced

by growing samples from existing natural mats [70], but the reliance on reproducing already characterized

colonies limits this approach’s use in modelling hypothesized prehistoric or extraterrestrial life. Although

larger-scale life incorporating OICs, such as animals with skeletons, can be grown in the laboratory, the

overwhelming complexity of multicellular organism development prevents their use in the study of OIC

production in isolation. There is therefore a pressing need for appropriate model systems for research on

in vivo OIC formation; by simulating the networks of cells hypothesized to have left biosignatures on the

early Earth or elsewhere, we can better understand the nature of such cells, their environment, and their

development.

Common laboratory cell strains can be bioengineered to produce the component materials of OICs at con-

trollable rates (e.g., [24, 176]). The question is whether standardized 3-D arrays of such cells are suitable for

use as model systems in the study of biological OIC formation under varying conditions. One potential min-

imal example of a material for investigation is citric acid and copper, which form a tridentate metal-organic

framework; a more complex one is chitin and calcium carbonate, which can form secondary and tertiary

structures [56].

5

An

aly

sis

Synthetic Biology

+

(a) genes are constructed for

material secretion

(b) artificial

plasmids are

transformed into

yeast

(c) transformed cells secrete

desired materialMedium Design

(d) suspend cells

in support

medium

3D Printing Substrate & Purification Design

(f) cell array secretes

and binds desired

material

(g) cells are removed,

leaving finished material

(h) novel biomaterial’s

properties are analyzed

promoter terminatorsecretion binding

material 1

material 2

material 3

(e) deposit cells in 3D array

Figure 1.2: Implementing this technology to the proof-of-concept level faces several

different challenges, many of which span disciplines as diverse as genetic engineer-

ing, embedded systems design, material science, and biophysical chemistry.

If it can be shown that standardized arrays of bioengineered cells can be useful models systems for the study of

biogenic OICs, it would be immediately applicable to laboratory simulation and reproduction of biosignature-

producing systems, such as microbial mats and metal-encrusting bacteria. The same model systems, with the

additional variable of a substrate on which the cells are deposited, could also be used to simulate early life in

geological microstructures or endolithic populations.

1.2.2 Novel Structural Designs

New materials are one of the primary drivers of new technological capabilities. Advanced materials can

provide the same strength and toughness as existing materials with lower mass; improved resilience, failure

tolerance, and damage recovery abilities; and even enable entirely new types of structural design. In a broad

sense, these desires are the same as those addressed in nature through the evolution of structural biomaterials:

they are strong, lightweight, tolerant of sudden and unpredictable loads, self-healing, and adaptable to a wide

variety of functions from a few basic components. These properties emerge from the materials’ defining

characteristic of multiscale, hierarchical levels of structure; an excellent example is nacre, part of mollusk

shell, which has at least six levels of organization (see Figure 1 in [107]).

The tight interdependence between the layout of each level of microstructure and the desired macroscale

properties of the overall materials requires complex control cycles between each cell creating the material

6

amino acids

simple sugars novel biomaterial

3D-positionable microdispensing system

sunlight

carbon dioxide

local mineralsmineral-extracting cells

photosynthetic

cells

metal ions &

trace minerals

piezo

dispenser dispensing

controller

positioning

controllers

3D-printed

cell arrays

organic-inorganic

laminate

finished biomaterial

composite part

material-

secreting

engineered

cells

digital

material

template

Figure 1.3: An example of how this technology concept might be deployed in a Mars

habitat context. By combining the biomaterials manufacturing technology with other

synthetic biology techniques that provide in situ resource creation and extraction,

advanced composites can be created from essentially raw molecular input.

and the cell’s microenvironment, including neighboring cells, external physicochemical signals, and the par-

tially formed structural deposit. We currently lack the ability to either manufacture existing biomaterials in

human-useful form factors or to reproduce these microstructural properties using more traditional material

components; we can neither make wood grow so that its grain is optimized to suit a chair of a certain shape,

nor create plastics that have a directional microstructure like the grain of wood. Traditional structural design

assumes essentially constant material properties and calculates maximum allowed loading conditions accord-

ingly. The ability to manufacture synthetic biomaterials would invert this paradigm; it would allow creation

of fine-grained material properties optimized to suit a given shape and load.

Although other ‘materials design for manufacturing’ projects have attempted to achieve property configura-

bility by using large-scale assemblies of repeating metal or plastic macroscale components [35, 77], gaining

the ability to manipulate microscale or molecular material components would be a new frontier in manufac-

turing. One major consequence would be the ability to do single-piece manufacture of what are now complex

assemblies: imagine an airplane chassis from a single, continuous material, with no bolts, no rivets, no welds,

no laminates. The material’s properties would be customized at a molecular level to allow it to be dense, rigid

and load-bearing where necessary yet light and flexible wherever possible, drastically reducing total mass.

Further, the material could transition seamlessly into a flexible gas barrier, creating functional unification be-

tween tensile, compressive, and pressurization support. Inherent ‘joints’ of customized directional flexibility

along particular axes could allow the entire structure to fold up like an umbrella for transport.

7

Such multilayered material functionality is common in natural structural biomaterials such as arthropod cuti-

cle [178]. Other light, tough, strong structural biomaterials include the feathers of birds [19], the bony plates

of seahorse tails [142], and porcupine quills [119]. The ability of these natural structures to resist fracture

and absorb impact or deformation energy has already been identified as highly desirable for human applica-

tions [32, 112, 114]. Manufacture of synthetic biomaterials with microscale material property optimization

would have profound effects on the entire field of structural design and mechanical engineering. Although

the work described here can only lay the basis for such future advances, identifying this need and justifying

future development of the technology concept driving it will spur future work as well.

1.2.3 Spatial Control of Gene Expression

The manufacture of synthetic biomaterials requires fine-scale (single-cell) spatial regulation of gene control.

This project is novel in seeking to use this capability for material production; however, it has applicabil-

ity to many other fields, including pharmaceutical screening [196], study of cell differentiation [18], tissue

engineering [139], and gene therapy [113]. This study of different material production control techniques,

particularly their different temporal and spatial resolutions, will inform efforts in these other fields.

The work described here also requires making 3-D spatial arrays of cells with near single cell resolution. As

with regulation of gene control, this capability bears on several other fields, including tissue engineering [45],

creation of living model systems for biomedical studies [5, 169], and high-speed bioassays [101]. By demon-

strating and charactizing the use of existing printing media, substrates, and mechanical parameters for a new

application, this work establishes a body of knowledge that will be highly relevant to other research groups.

1.2.4 Potential Aerospace Implications

Upmass requirements are the single most significant limitation of current space mission capability. The limits

of what can economically be launched and safely landed severely restrict current missions and disqualify

many others from consideration. For example, a lunar or Mars base would require heavy manufacturing

equipment and construction materials far beyond current transportation capacity. Any long-term human space

presence requires periodic replenishment, adding massive cost overhead. Even autonomous missions, such as

orbiters and flybys, often sacrifice science goals for heavy radiation and thermal protection. This problem is

compounded for planetary missions in which mission instrumentation or equipment must be landed safely, as

8

proteins

sugars

biogenic silica (diatoms)

nacre (mollusks)

microbial mats

biogenic aragonite (calcareous sponges)

wood

size

com

ple

xity

Figure 1.4: Biomaterials cover a wide range of size and complexity scales, rang-

ing from simple biomolecules to macroscale organic-inorganic composites. (Non-

original images are public domain.)

planetary landing structures must absorb a tremendous amount of energy which requires a great deal of mass

(particularly at joints) given current material strength limitations; this further reduces the amount of mission

payload mass that can be transported for a given launch cost.

The manufacture of synthetic biomaterials would both reduce upmass and allow the construction of a new

category of strong, lightweight, multifunctional structures, creating novel biocomposites and laminates with

customized physical properties, vastly increasing the use of extraterrestrial in situ resources in place of trans-

ported materials, and enabling just-in-time manufacturing of needed parts in off-Earth environments.

1.2.5 Other Benefits

Creation of novel, bio-inspired materials is likely to have many additional applications which cannot yet

be predicted. As mentioned earlier, many biomaterials are highly multifunctional; as later research efforts

build on the framework established here, many new and innovative applications are likely to be developed

as other groups begin to implement the technology concept in other fields. For instance, photonic crystals

have been inspired by the structure of butterfly wing scale [120, 192], and synthetic OICs hold great promise

for applications such as biosensing, gas catalysis, and carbon sequestration [167]; many more are listed in

section 1.4.

9

A much longer-term possibility is a beneficial environmental impact. Many compressive-strength biomateri-

als, such as those mineralized with calcium carbonate, are natural CO2 sinks. On Earth, the most common

traditional engineering material they would replace is concrete, the manufacture of which is a major CO2

source. Large-scale adaptation of this technology could fundamentally alter the climate change calculus.

1.3 Technical Background

1.3.1 Definition of Structural Biomaterials

There is considerable debate over a precise definition of the term biomaterial; it is deliberately used it in

its broadest possible useful sense here, to mean the natural materials produced by and integrated into living

systems as well as artificial materials developed to mimic them [145]. In size, biomaterials range from a

single nanometer, for simple molecules like sugars, to micron-sized cell walls, to millimeter-sized sponge

spicules, to meter-sized bones, to trees that can reach 100 m in height. In complexity, they range from largely

amorphous soft matrices to highly structured, multi-layer composites of organic and inorganic constituents.

There are many examples of naturally occurring materials showing the independent combinations of these

characteristics (Figure 1.4); many proteins form highly ordered folded assemblies, and fist-sized sponge ex-

oskeletons may be essentially bulk calcium carbonate deposits.

The focus of this work is non-living, structural biomaterials. In nature, these are the materials which are pro-

duced by, or integrated into, living systems to provide a physical, mechanical function – protection, leverage,

support, and so on. The distinction between living tissue and non-living biomaterial is typically blurry in nat-

ural systems; an animal bone, for example, is a living organ that consists of active cells which have generated

a complex, partially mineralized tissue, and it plays important hormonal and metabolic roles in addition to

its mechanical one [160]. For purposes of this work, the structural biomaterial associated with bone is that

extracellular part of the tissue which would, at least in theory, have essentially the same mechanical prop-

erties if all living cells were removed: a highly ordered composite of proteins (collagen, etc.) and biogenic

minerals (calcium phosphate). Most structural biomaterials are rarely, if ever, found in a purely non-living

state in nature; it is because the technology described here aims to create such forms of these materials that

they are termed ‘synthetic biomaterials.’

Table 1.1 lists several examples of natural structural biomaterials and their most important constituents. A

very common, though by no means dominant, type of arrangement consists of long-chain tough fibers such

10

as collagen or keratin embedded in a softer and more compliant matrix. In some cases, such as spider silk, a

single protein may fill both roles by alternating between different folding configurations; there are many more

examples of composite materials in which the fibers and matrix are different material types. Most structural

biomaterials are entirely organic, in the chemical sense of consisting of carbon-based compounds such as pro-

teins, polysaccharides, and so forth. However, the most rigid structural biomaterials are those which have an

additional inorganic component. These biominerals are formed through biologically-controlled crystalliza-

tion of inorganic materials, such as calcium salts; this process of biomineralization is generally poorly under-

stood [21], but there are known to be several different mechanisms involved in different species [15, 56, 157].

1.3.2 Properties of Structural Biomaterials

Naturally occurring structural biomaterials have several common traits relevant to the challenges and oppor-

tunities of this work (see [107, 118, 181] for extensive reviews):

• Hierarchical structure. This, perhaps more than anything else, defines the class of structural

biomaterials of most interest. Nacre, part of mollusk shell, is an excellent example: it has

at least six different spatial scales of organization, with ‘rods’ bound together to form ‘tiles’,

‘tiles’ into ‘shelves’, and so on [107]. The impressive toughness and resilience of many struc-

tural biomaterials emerge from their ability to distribute loading or impact energy between

each of these microstructured levels.

• Strong functional customization. The control over multiple levels of microstructure allows

similarly fine control over the emergent macroscale material properties. For example, the the

non-living materials that compose insect [32] and crustacean [143] exoskeletons are optimized

at a molecular level to allow the structure to be dense, rigid and load-bearing where necessary

yet light and flexible wherever possible; the outer surface transitions to a flexible vapor barrier,

preventing dehydration; and the inner surface can transition to muscle or other attachments.

• Robust interfacing of distinct component materials. Many biominerals have organic and inor-

ganic components interwoven in a highly ordered way on the molecular level (subsection 1.2.1).

At a larger scale, many biological structures have continuous, gradient transitions between ma-

terial phases [100, 180] – for example, the collagen fibers of rigid bone are essentially part of a

11

Ta

ble

1.1

:N

atu

rally

occu

rring

stru

ctu

ralb

iom

ate

rials

may

have

co

mp

lian

t,rig

id,

org

an

ic,

an

din

org

an

icco

mp

on

en

tsin

any

co

mb

ina

tion

.It

isw

orth

no

ting

how

ma

ny

diffe

ren

tstru

ctu

res,

with

wild

lyd

ifferin

gm

ech

an

ica

lp

rop

ertie

s,

ca

nb

e

co

nstru

cte

dfro

me

sse

ntia

llyth

esa

me

ba

sic

co

mp

on

en

ts.

Stru

cture

Exam

ple

Su

pp

ort

(Matrix

)S

tructu

re(O

rgan

ic)S

tructu

re(In

org

an

ic)N

otes

cyto

skeletal

fibers

all(?)cells

–fi

lamen

tous

pro

teins

–[1

05

]

mag

neto

som

esbacteria

lipid

s,pro

teins

–iro

noxid

esor

sulfi

des

[95

]

frustru

lediato

ms

silaffins

–silica

[42

]

micro

bial

mats

micro

bial

mats

poly

saccharid

es–

calcium

carbonate

[70

]

silksilk

worm

ssericin

fibro

in–

[92

]

silksp

iders

spid

roin

(amorp

hous)

spid

roin

(crystallin

e)–

[141

]

spicu

lesiliceo

us

sponges

silicateins

–silica

[158

]

spicu

lecalcareo

us

sponges

collag

en–

calcium

carbonate

[ 157

]

shell

crustacean

sresilin

chitin

calcium

carbonate

[143

,177

]

shell

mollu

sks

gly

copro

teins,

poly

saccharid

esch

itincalciu

mcarb

onate

[107

,110

]

feather

bird

sunknow

nkeratin

–[1

82

]

tooth

enam

elverteb

rates–

–calciu

mphosp

hate

[57

]

cartilage

anim

alspro

teogly

cans

collag

en,

elastin–

[81

]

horn

,hoof

verteb

rateskeratin

(amorp

hous)

keratin

(crystallin

e)–

[175

]

bone,

scaleverteb

ratesgly

copro

teins

collag

encalciu

mphosp

hate

[ 160

,182

]

wood

plan

tslig

nin

cellulo

se–

12

continuous piece with the attached collagen-based tendons – which eliminates the weak points

caused by seams, bolts, laminates, and so on in traditional human manufacture.

• Combinatorial complexity from basic molecular constituents. The basic building blocks of

life are carbon, hydrogen, oxygen, nitrogen, phosphorus and sulfur. Proteins and polysac-

charides have an astonishing array of properties despite having only a few dozen component

monomers. There are relatively few inorganic components found in biomineralization – silica,

calcium salts, and a few metals (primarily deposited by bacteria in sedimentary structures)

comprise the vast majority. It is differences in the microstructures formed by these materials

that are responsible for the incredible diversity of mechanical properties in structural bioma-

terials. This common basis, combined with biology’s inherent self-replicating capability and

necessary use of immediately accessible, energetically favorable resources, makes biomateri-

als very mass- and energy-efficient to produce.

• Low mass. Structural biomaterials are near-universally less dense than traditionally-used man-

ufacturing materials such as metals and ceramics. On many engineering measures – tensile

strength, fracture resistance, elasticity, and so forth – they are also lighter for the same per-

formance when tested under the directional loading conditions for which they are optimized

(Table 4.5, Table 4.8).

• Low energy manufacturing requirements. Biogenic materials are limited in manufacturing

requirements by the amount of energy that can be spared by living systems. This is typically

orders of magnitude lower than that required by, for example, casting metal or heat-molding

plastic. This is achieved by, e.g., use of catalysis, bottom-up self-assembly, and repeated use

of naturally stable monomers.

• Life-friendly environment manufacturing requirements. Essentially all biomaterials can be

made at Earth-surface-normal pressure and temperature, near-neutral pH, moderate osmotic

pressure, and so on (although cells may create internal microenvironments where the chemical

environment can be quite different). Their manufacture and processing thus poses much less

risk to human workers and the local and global environment than many current chemically or

physically harsh processes used in industry.

However, there are three current roadblocks to human use of structural biomaterials:

13

• High production overhead. Natural biomaterials are produced by living systems, which re-

quire a complete ecological support structure. Although the few biomaterials we continue

to use on a large scale are typically produced by living systems with other secondary uses –

raising a sheep to produce wool is, energetically speaking, absurdly expensive, but sheep can

also produce milk and meet – it is an insurmountable barrier to the wider adoption of most

otherwise promising materials.

• Non-predictable microstructure. The process by which living systems fine-tune the microstruc-

ture of the materials they produce are poorly understood. We also lack the tools to predict the

emergent mechanical properties of the overall material from its microstructure. Our inability

to reproduce or predict the formation of these microstructures has led to an assumption that

they make such materials unsuitable for human use.

• Non-controllable macrostructure. Life produces materials in the shape it needs. The enamel of

tooth is found only in the shape of a tooth, the rachis of feather in the shape of a feather. Given

the way the microstructures of the material are optimized to suit the function of the overall

structure, the natural shape cannot simply be cut down to fit human needs without losing its

mechanical benefits.

1.4 State of the Art

1.4.1 3-D Living Cell Printing

Additive manufacturing is a class of fabrication processes in which successive layers of material are laid down

and bound together according to a 3-D digital design. It comprises a variety of rapid prototyping and small-

run manufacturing techniques, such as fused deposition modelling, powder sintering, and stereolithography;

in recent years, some industrial firms have adapted the same approaches to larger scales. The term 3-D

printing, which originally referred to a specific subset of additive manufacturing technologies, is increasingly

commonly used to refer to all additive manufacturing, particularly in the popular press; however, since several

different additive manufacturing techniques will be discussed in this section, the more general term is used.

The focus here is on additive manufacturing techniques as they use, or are used for, biological and biomedical

purposes; for a more general technical review, the reader is directed to a source such as [59].

14

Historically, additive manufacturing has been limited to either polymeric materials (plastics and waxes),

which can be melted or photocured together, or thin sheets of materials like paper first cut and then bonded

into laminates. More recent approaches have added metal and ceramics to the toolkit, primarily through

techniques that deposit the materials in powdered form and then either fuse them to a secondary substrate or

melt them together. Precision of material placement for these techniques varies with cost and implementation

choices, but often approaches the microscale [155].

All of these approaches have two limitations relevant to this technology concept. Firstly, the final material

is made of the feedstock used – in other words, if the material laid down in layers is metal powder with a

binding substrate, than the final piece will be made of porous metal mixed with binding substrate. Secondly,

the material used must be relatively homogenous, as the process of fusing bulk smaller parts (whether powder

particles, droplets of liquid, or solid laminate layers) into a larger whole is destructive to any internal material

structure smaller than the scale of the parts. (Ceramics and metals do have small-grain crystalline structure,

but these must re-form after powder fusion to be continuous; for this reason, fine temperature and positioning

control during deposition, as well as post-processing with temperature treatments, are often necessary to

prevent development of unplanned variation in the material properties of the final piece.) Thus, they cannot

be used to make hierarchically structured materials such as the chitin and calcium carbonate complex of

shell [56].

There have been two major drivers of the adoption of additive manufacturing technology for biological or bio-

compatible materials over the past two decades. The first is the desire for microvolume cell or biomolecule

spatial positioning, useful for compact and high-throughput biological or pharmaceutical assays [47, 101,

189, 191]. This approach tends to focus on single-cell precision and deposition speed, and is often only

concerned with 2D spatial patterning. The second is the desire to construct replacement or testbed living

tissues, a major goal of the field of tissue engineering [67] (also reviewed in [9, 48]). In this area, work

typically focuses either on fabricating a non-living, biocompatible substrate onto which living cells can be

seeded, in the hopes that they will grow over and bind into the substrate to form a 3-D tissue, or on depositing

layers of activated cell clusters and stimulating them to form attachments to each other, a closer mimicry of

natural tissue growth [85, 170].

A third, more recent driver is the discovery that the three-dimensional structure of groups of cells, both uni-

cellular microbial cultures [36, 179] and multicellular tissues and tumors (reviewed in [93]), has a profound

effect on their behavior. These techniques have hewed closely to those developed for tissue engineering,

primarily creating biocompatible scaffolds – typically a combination of hydrogel [5], support proteins such

15

as collagen or its derivative gelatin [41, 169], and sometimes synthetic polymers [45] – seeded with cells

to create 3-D living cell clusters for purposes of testing environmental and pharmaceutical effects. Various

depositional technologies have been developed for these scaffolds and cells, including optical gel cross-

linking [41], chemical hydrogel cross-linking [163], liquid jet-based methods (reviewed in [146]), and laser

thermal/photomechanical ejection [12].

The primary limitation of these ‘bioprinting’ technologies, where the work presented here is concerned, is the

lack of integration between single-cell resolution deposition and and biocompatible substrate construction.

Although some hydrogel-based tissue engineering approaches can embed single cells, or small cell clusters,

in a substrate [188], living tissues are not spatially organized on a single-cell level. In tissue engineering,

therefore, the goal is to encourage cell-cell binding and production of extracellular matrix among functional

units of cell clusters, and the cells’ ability to proliferate over and migrate into the scaffold is often considered

an important feature. Although much of the work on substrate printing is directly applicable to this work, in

existing systems typically no provision is made for fine-level control over the structure of the final cell array,

which is necessary for this new technique.

1.4.2 Synthetic Biology for Material Production

Synthetic biology is the creation of new, designed systems using biological parts – biomolecules, cells, tis-

sues, and organisms [156]. Although the boundaries of the field have expanded in recent years to include

subjects previously considered independent, such as protein engineering and functional genomics, a common

synthetic biology application involves re-ordering and re-linking DNA sequences to form a new gene or set

of genes, inserting the modified sequences into cells, and characterizing and make use of the resulting change

in cell functionality.

The combination of synthetic biology and material production has largely focused on engineering microor-

ganisms to increase their production of bulk biomolecules, such as pharmaceuticals (e.g., [37, 40]), biofu-

els [38, 78], and biopolymers which can be processed into so-called ‘bioplastics’ [133]. The goal may be to

increase the rate at which the microorganisms make the desired biomolecule, to increase their ability to ex-

crete it into solution (reducing the amount of post-production work to isolate the target product), to increase

their survival at high levels of the target compound, or any other improvement to the final product yield. A

more directly relevant application has been the engineering of microbes to bind or sequester ions from their

surroundings; although microbes which naturally perform these functions for metal ions are widely used

16

in so-called ‘biomining’ [153], synthetic biology work to improve yield or the variety of ions that can be

economically harvested is still primarily at the proof-of-concept stage [11, 96].

Application of synthetic biology to tissue engineering (which, as discussed in subsection 1.4.1, has been the

primary area of development for 3-D printing of living cells) has been sparse. Some work has been done,

similar to the biomolecule production approaches above, on making microorganisms produce supporting pro-

tein ingredients for tissue scaffolds [62]; these proteins are purified and mixed into a gel substrate to produce

the final scaffold material. A slightly different use is engineering microorganisms to produce regulating or

signalling proteins; when these proteins are purified and distributed within a scaffold, they can be used to

stimulate or otherwise control biological growth after the scaffold is seeded with cells [66].

The closest work in the existing field of synthetic biology to the concept presented here is the use of spatially

structured arrays of cells to regulate gene expression. In this approach, cells are spatially patterned along with

physical cues or causes of changed cell behavior as a means of creating spatial differentiation in function.

This technique has been implemented with cells and gene-vector viruses [113, 139], patterns of two different

cell types [18], and cells and plasmids [196].

The use of synthetic biology to enable spatially controlled, non-living, macroscale material production has

been essentially unexplored prior to this work.

1.4.3 Biomimetic Materials

As described in section subsection 1.3.2, among the many factors that distinguish structural biomaterials from

traditional human-made materials, two stand out. One is the many levels of hierarchical organization. The

other is the micro-level integration of different, molecularly complex constituent materials. Much current

materials research focuses on ‘biomimetic materials’, a term which encompasses two related goals.

The first goal is reproduction of the mechanical properties of structural biomaterials; this is the usage common

in materials science and engineering, and work in this area tends to focus on producing hierarchical organiza-

tion using non-biogenic materials (see [20, 152] for reviews). Examples include designs for new mixed-phase

reinforced ceramics based on the structure of sponge spicules [112] and structuring the fiber grain directions

of glass-fiber resin composites to match the directionality of chitin fibers in insect cuticle [31].

17

One intriguing set of techniques in this area has been the direct use of biogenic materials as templates for

additional material assembly. Titanium dioxide structures formed by coating and then burning away butterfly

wing scales have been shown to be effective photoanodes [192]; a similar approach was used to generate

antireflective structures from fly eyes [80]. More unusually, carbon nanotubes have been shown to self-

assemble into honeycomb structures directly onto butterfly wings, resulting in composite biogenic-artificial

material structures with unusual thermal and electrical characteristics [122].

The second goal is production of biocompatible materials that can substitute for or supplement the function

of natural materials; this is the usage common in tissue and biomedical engineering, and work in this area

tends to focus on producing one or two levels of structural function using existing biogenic materials. An

example is the formation of collagen-based scaffolds for tissue engineering with control of porosity in the

10−4 m range [100].

More interdisciplinary work includes the creation of new composites of existing biogenic materials. For

instance, silk fibers have been integrated with hydroxyapatite, the key mineral in bone, to create a human-

designed composite biomaterials mimicking the bone/ligament interface [71, 190]. Such work establishes

that there is a desire for techniques which can create novel biomaterial composites; the technology concept

presented here aims to provide a general framework for this application, rather than the collection of ad hoc

techniques used in existing work.

More directly relevant to this work, self-assembling structures of viruses genetically engineered to be crys-

tallization nuclei have been used to control the orientation of microstructures formed by hydroxyapatite in

solution [187]. Here, genetic engineering is used to create what is, in essence, a scaffold providing a template

for material self-assembly. This is highly relevant to the substrate design section of this work, and offers a

potential path for future development.

Another modern form of biomimicry in structural materials is tensegrity, a promising new approach to struc-

tural design using inherent load-balancing between tensile and compressive elements (reviewed in [83])

with greatly reduced mass [23]. The internal mechanics of many structural biomaterials, with compressive-

strength mineralized phases embedded in tensile polymers at many spatial scales, bear striking similarity to

(and in some cases are the inspiration for) modern tensegrity designs [7, 112]. Although these biomaterials,

such as the non-living parts of bone, have been highly optimized by nature for biological tensegrity applica-

tions such as skeletal mechanics [33], our inability to create them in human-controlled forms has prevented

us from investigating their use in artificial structures. A significant amount of work has gone into developing

18

Table 1.2: NASA’s Technology Readiness Level definitions and their relationship to

the project objectives and this document.

TRL Definition Reference

‘early’ TRL 2(a) technology concept formulated

chapter 1(b) documented speculative practical application

TRL 2 exit (a) feasibility/benefit analysis chapter 4

‘early’ TRL 3 (a) proof-of-concept demonstration chapter 2

‘late’ TRL 3(a) analytical prediction of key parameters

future work(b) laboratory validation of predictions

one- or two-level mechanical models of existing biomaterials [31, 118], including understanding continuous

material gradient ‘joints’ [180]; however, this has not yet been integrated with the dynamic modelling tools

under development for understanding tensegrity structures at human-made scales. Tensegrity designs, while

some years off, are therefore a highly complementary long-term future application for synthetic biomaterials.

1.5 Project & Design Objectives

The overall goal of this project was to demonstrate the possibility of manufacturing human-designed, non-

living biomaterials by using 3-D-printed arrays of bioengineered living cells. As this work was supported

by the National Aeronautics and Space Administration (NASA), the goal was defined within the context

of NASA’s defined space mission Technology Readiness Levels (TRLs) [108] as advancing the technology

concept from TRL 2 to TRL 3. The definitions of these levels as applied to this specific technology concept,

and their relation to the chapters of this thesis, are shown in Table 1.2.

The overall goal thus consisted of three objectives:

1. A proof-of-concept demonstration integrating all critical components of the biomaterial pro-

duction technology. The end deliverable was chosen to be a grid of two differently-fluorescing

proteins, with a pattern demonstrated to correlate to the digital material template used to print

the original cell array. Showing end-to-end critical functionality of each component part of

our technology concept was necessary to validate the core technology concept. This was the

majority of the work performed, and is presented in chapter 2.

19

2. A proposed design of this technology for follow-on work. Recommendations for hardware,

biological, and material implementations for the next stages of technology development are

presented in chapter 3 based on the lessons learned from the proof of concept work.

3. A feasibility and benefit analysis of this technology in two specific space mission contexts.

To span the range of potential applications, the space mission contexts were chosen to be

(1) a ‘minimal working example’ making a finished biomaterial part aboard the ISS, and (2)

‘cradle-to-grave’ use at a Mars human habitat. This work is presented in chapter 4.

20

2 Synthetic Biomaterials: A Proof of Concept

2.1 Introduction

The major objective of this project was a proof-of-concept demonstration of the technology concept. To

achieve this, it is necessary to demonstrate critical functionality for each component of the technology

(Figure 1.2): the cell strains, the print medium, the additive manufacturing hardware and software (consisting

of the micropositioning and microdispensing subsystems), the print substrate, and any post-processing and

analysis steps. The objective was therefore broken down into milestones reflecting these components:

1. to identify appropriate performance metrics for each component of the technology (presented

in section 2.2);

2. to choose an appropriate demonstration material pattern that will allow measurement of the

chosen performance metrics (presented in section 2.3);

3. to identify a suitable implementation choice for each component (each component presented

separately in section 2.4 through section 2.7);

4. to demonstrate critical functionality of all components (each component presented separately

in section 2.4 through section 2.7);

5. to integrate the functional components into a complete system (presented in section 2.8);

6. to demonstrate functionality of the complete system (presented in section 2.8); and

7. analyze the performance of the complete system, including correlation of the final material

product to the original material pattern (presented in section 2.9).

21

2.2 Peformance Metric Selection

Two sets of performance metrics were necessary. The first are the minimum values necessary to indicate

critical functionality of each component. These are derived from two different basic requirements. The first

is the need to distinguish between all cell-made material and the desired extracellular, bound material; the

second is the need to verify that the material has a pattern correlated to the print template. The size of the host

cell strain (subsection 2.4.1) and the limitations of the material yield and cell survival quantification methods

available to the project were the primary determining factors for these minimum metrics.

The second set of metrics are those representing the current state of the art in bioprinting and biomaterial

fabrication, as described in section 1.4. While meeting each of desired metrics is not necessary to meet these

to achieve the overall proof of concept, and in many cases the reported values in the literature are not directly

comparable to the minimum metrics due to differences in cell size and type or deposition method, they

provide a useful set of guideposts regarding the context of the technology and the likely ease of improving

the performance of different components in future work.

Cell resolution metrics were originally described in terms of voxels (the smallest mapped element of a three-

dimensional volume, as a pixel is the smallest element of a mapped two-dimensional area); however, after

the decision was made to implement the 3-D printing of cells using piezoelectric deposition (section 2.6),

these were converted to metrics in terms of droplets, as the mapping from printed, deformable droplets to true

voxels is inexact. The “pattern completion” metric refers to the percentage of specified locations at which

material should be deposited at which material is present at a detectable level; as such it is affected by cell

survival, but also encompasses material yield and analytic sensitivity.

Both sets of originally derived metrics are listed in Table 2.1. For comparison, the table with the final mea-

sured metrics is presented as Table 2.5.

2.3 Material Template Selection

As there are many established technologies for three-dimensional printing of living cells (section 1.4), for

ease of analysis, the proof of concept material pattern was chosen to be a two-dimensional array to improve

ease of verification without loss of generality. The presence of so many composites on the original list of

22

Table 2.1: The performance metrics derived for the proof of concept.

Metric Minimum Notes Desired Notes

positioning precision 1 cell diameter ≈ 10 µm a 1 µm [129]b

dispensing volume ≤ 1000 cells“voxel”

c ≈ 1 cell“voxel”

[101]

“voxel” size enc. 1000 cells = 1 nL a 10 pL [188]

cell survival 50% d 90% [188, 191]

pattern completion 20% [4] 70% [188]e

a Based on measurements of Saccharomyces cere-

visiae (subsection 2.4.1).b Nanometer-precision positioning systems exist (e.g.,

[163]); metric approximates current scale of fabri-

catable biomaterial features.c Maximum density quantifiable post-printing with

available equipment.d Minimum required for detection with available

equipment.e Higher rates reported are for larger (e.g., mam-

malian) cells.

motivating materials (Table 1.1) indicated that the pattern should consist of at least two different materials.

Furthermore, as the verification and analysis techniques available to the project were limited to visible and

fluorescent microscopy and absorbance and fluorescence spectrophotometry, the pattern needed to allow for

verification of all performance metrics (Table 2.1) within these constraints.

The resulting choice was a square grid of twenty-five (five by five) droplet locations alternating between two

different materials tagged with fluorescent proteins. The template is shown in Figure 2.24.

2.4 Cell Engineering

2.4.1 Host Cell Strain Selection

There are two primary desired qualities for the host cell strain. The first is size-driven; the functional cellular

units (either single cells, as unicellular organisms, or activated clusters of cells with a unified function, as in

multicellular tissues) should be no smaller than the printing system’s resolution. This constraint is derived

from the need to achieve resolution of one functional cellular unit (FCU) per voxel resolution in constructing

the cellular arrays. Being able to control FCU spacing as an independent variable is necessary, in turn, to

determine whether the ultimate goal of using the spatial pattern of the cell array to control the formation of

material microstructure is possible; although the work presented here is limited to a non-structural proof of

concept, a key secondary goal was to ensure that the same system could be used for follow-on work.

23

The second initial desired quality for the host cell strain was that it have a history of established use. At a

minimum, this would be a cell type which can be easily cultured in the laboratory; ideally, this would be a

cell type with a strong history in molecular biology and genetic engineering. This criterion was particular

important given the short timeline of the work (the funding was for one year) and that it was to be performed

by researchers beginning the project without a background in molecular biology or genetic engineering; there

would not be time or resources to develop new plasmid backbones or other support biological tools.

A third criterion was identified after a review of the list of potential material types for implementation

(subsection 2.4.2). Nearly all of the most promising structural materials are made by multicellular organ-

isms, all of which are eukaryotes; the internal genetic machinery of eukaryotes is substantially different from

that of prokaryotes. Although genetic engineering of prokaryotes is generally much simpler, and there is a

longer history of established tools for it, the decision was made to use a eukaryotic cell strain for the initial

proof of concept, thus avoiding the necessity of re-implementing all of the genetic engineering work when

moving to implementation of structural materials in a later project.

Investigation of commercially available microdeposition systems showed that the typical minimum droplet

size of commercial systems is in the picoliter range, which corresponds very roughly to a droplet diameter of

10 µm. This is substantially larger than typical prokaryotes (around 1 µm for E. coli), on the small side for

single-celled eukaryotic species, and much smaller than cells from most multicellular organisms. Thus, the

combination of these three criteria narrowed down the potential cell strains to existing unicellular eukaryote

model organisms.

The following options for host cell strains were then identified and evaluated:

• Saccharomyces cerevisiae. A common, spherical yeast approximately 5 – 10 µm in diameter

Figure 2.1. It grows very quickly, is robust under standard laboratory conditions, and has an

extensive history in genetic engineering. This was the final selection.

• Schizosaccharomyces pombe. Another common yeast, very similar to S. cerevisiae. It was not

chosen due to its rod-like shape, which makes it exceed the size criterion along one dimension

(although not in total volume) and would have complicated evaluation of performance metrics

based on cell diameter (Table 2.1).

• Thalassiosira pseudonana. A frequently studied diatom. It was a promising option due to its

natural deposition of silica, which would have meant the host cells produced external material

24

Figure 2.1: Saccharomyces cerevisiae under 1000x DIC magnification. The cells

are roughly spherical in a typical state, although they will elongate prior to division.

(Public domain image.)

in their natural state without the need for further bioengineering. However, the ability to con-

trol the cells’ silica deposition, and suppress it in order to introduce genes for other materials,

is not established. Thus, it was not chosen for the initial study because (as with all the prokary-

otic options examined) it would have required re-implementing the genetic engineering in a

different host cell strain later to allow for work with other materials.

The most promising choice, given the evaluation criteria, was Saccharomyces cerevisiae. The selection was

cemented by the availability of a commercially available plasmid backbone, pYES2.1 (Life Technologies),

which is specialized for use in S. cerevisiae and included several other features necessary for our work (see

details in subsection 2.4.5). This saved the project a substantial amount of time in achieving the final proof

of concept.

The final choice was an S. cerevisiae ura3 strain, meaning that the strain lacked the ability to synthesize

its own uracil (one of the basic amino acids). Making the strain dependent on user-supplied uracil is a

contamination and selection control.

With the cell strain selected, the next step was to characterize those of its physical parameters that were

relevant to later implementation choices.

25

Table 2.2: The materials evaluated for use in the proof of concept.

Material Type Part Available? Optical Verification?

chromogenic proteins protein yes no

fluorescent proteins protein yes yes

silk protein partial potential

oak other organic no no

pine other organic no no

cellulose other organic partial no

lignin other organic partial no

bone mineral OIC no potential

rubber other organic no no

cork other organic no no

mollusk shell OIC no no

2.4.2 Material Selection

As with the choice of host cell strain, when materials were evaluated for implementation in the proof of

concept, a key criterion was that it have an established history of implementation in genetic engineering. For

material choice, this meant that it corresponded to a known genetic part, or coding DNA sequence. After

deciding that the proof of concept would consist of two materials (section 2.3), this criterion was expanded

to include that both materials should be expressible using the same plasmid backbone. Being able to do all

necessary genetic engineering through design, assembly, and insertion of a single plasmid greatly simplified

the cell engineering workflow.

The other primary criterion was ease of verification. Most structural biomaterials, in their natural state, can

only be distinguished and their internal structure identified with electron microscopy, which the project did

not have access to. The project did, however, have access to high-quality visible and fluorescence microscopy

tools (Zeiss’s Axio Imager Z1, with Lumen Dynamics’ X-Cite series 120 fluorescent light source), as well as

absorbance and fluorescence spectrophotometry (Molecular Devices’ SpectraMax Plus 384 and SpectraMax

Plus Gemini XS).

The last major decision, after establishing these criteria, was whether to use a protein directly as the final

material. Broadly speaking, genes within a cell code for proteins; some of these proteins are structural in and

of themselves, such as keratin and collagen, but others work indirectly to cause the formation of non-protein-

based structural materials, such as polysaccharides (chitin, cellulose, etc.), inorganic crystalline materials

(silica, calcium carbonate) or even sequestered metal ions (Figure 2.2).

26

(a)(b) (c)

Figure 2.2: Assuming that the material is made by a cell and is dependent on a

genetic part, there are multiple levels of complexity available. The simplest (a) is

a protein-based material encoded by a single gene. Some protein-based materials

have multiple subunits controlled by different genes (b). At the highest level of com-

plexity (c), a composite biomaterial may consist of multiple protein-based materials

as well as inorganic materials deposited by the cell.

A number of biogenic materials were identified and evaluated based on these criteria (Table 2.2). The number

of materials with established genetic parts was by far the most significant determining factor, reducing the

choices to essentially two families of non-structural proteins, one with differing visible colors (chromogenic

proteins), one with differing fluorescence properties (fluorescent proteins). The project was initially not able

to procure coding regions for the chromogenic proteins; thus, the final selection was the family of fluorescent

proteins. Later testing revealed that expression in S. cerevisiae of four chromogenic proteins was below the

minimum threshold (not included here), so it would not have met the second criterion even if it had been

available earlier.

The decision was thus made to use two differently-colored fluorescent proteins for the proof of concept. Sev-

eral DNA sequence registries, including GenBank and the Registry of Standard Biological Parts, were used to

search for coding sequences that were appropriate for the chosen host cell strain. Funding constraints limited

the initial selections to what could be extracted from cell strains previously implemented for other co-located

projects by generous colleagues; these were GFPmut3 [44] and mRFP1 [27]. After initial implementation,

expression quantification of GFP yield was well below the desired level. A second round of implementation

with sfGFP [137] did not substantially improve results.

An timely opportunity for no-cost gene synthesis through the commercial company Gen9 allowed the project

to have new coding regions synthesized from partially codon-optimized sequences generated from those

published in [43] (green) and [89] (red). As these were coding regions for proteins demonstrated to function

in yeast, with some additional codon optimization for S. cerevisiae using Gen9’s toolset, they were expected

27

to have substantially improved performance. These sequences, included in Appendix A, are referred to

hereinafter as yeGFP and yeRFP (for ‘yeast-enhanced’) and proved sufficient for use in all subsequent work

presented here.

An additional advantage to this material choice is that the final material product is itself fluorescent, rather

than requiring fusion of a secondary fluorescent protein for detection. Further, these yeGFP and yeRFP

regions can act as such fusion fluorescence tags for future non-fluorescent proteins to provide optical veri-

fication of their placement. Thus, future work with structural proteins will be able to take advantage of the

same yeRFP/yeGFP implementations as fluorescent tags to improve material analysis.

The literature search and downselection process strongly highlighted the very limited material options avail-

able. The future of this technology concept will therefore have a strong dependence on future advancement

in genetic sequencing and functional identification.

2.4.3 Material Stimulus Method Selection

The technology concept requires the ability to control the production of material by the cells. For the proof

of concept, at a minimum, this requires the ability to prevent material production prior to printing, in order to

ensure that the final material product is truly the result of the cells’ activity in the array. However, ultimately,

there will be a need to be able to regulate material production on a fine-grained level to be able to characterize

its effect on material structure formation.

Although gene expression regulation can be complex, particularly in eukaryotes such as the host cell strain

chosen here, its basis is the part of the gene known as the promoter (Figure 2.3). There are many established

promoter sequences for S. cerevisiae in the literature. The first choice is whether to use a constitutive pro-

moter, or one whose effect is determined by external stimuli; if the latter, the second choice that must be

made is what the external stimuli is to be.

A review was conducted of a significant number of promoters available in yeast (see overviews in [29, 135,

166], among others). The most clear conclusion was that there was no advantage to choosing a constitutive

promoter, as there are equally strong or stronger inducible promoters and the additional process overhead

added by needing to time printing to avoid accumulation of secreted material in the print medium was a

significant cost.

28

stimulus

exp

ressio

n

stimulus

exp

ressio

n

stimulus

exp

ressio

n

(a) (b) (c)

Figure 2.3: Broadly speaking, promoters are the part of a gene which control ex-

pression. The simplest type is (a) a constitutive promoter, which is essentially in-

sensitive to environmental conditions. Many promoters are (b) inducible, controlling

expression of their linked coding regions in response to a particular stimulus. Some

are also (c) repressible, allowing for an effect like a programmable dimmer switch.

Given the decision to use an inducible promoter, the next key question was what type of stimulus to use

(Figure 2.4). The reviewed promoters included those which are controlled by heat, different wavelengths of

light, and a wide variety of chemical stimuli (phosphates, alcohols, metal ions, and sugars, among others).

Thermal stimulus was discounted because of the undesirable effects of changes in temperature on the vis-

cosity of the printing solution and print substrate. Optical stimulus had several attractive features, including

the potential of finest-scale control over material production, but both optical gene regulation systems exam-

ined [159, 164] were extremely complex and not feasible to implement on this project’s short timeline. The

decision was therefore made to use a chemical-based stimulus.

A key realization from this examination of promoters was that the type of stimulus best suited to a particular

implementation of this technology concept would vary based on the form factor of the material being pro-

duced. A material that does not require small-scale variation in material properties – for instance, a structural

element in tension – can be produced at much lower cost and effort using thermal or chemical stimulus; a

more specialized piece, such as a gear or buckle, might require a scanning laser optical stimulus system to be

fully optimized. Table 3.2 summarizes these conclusions.

Of the remaining promoters, the GAL1 promoter, pGAL1, offered several notable features. It is well-studied,

and its peak expression effect is so strong that it is sometimes used as a baseline for comparing the effective-

ness of other promoters [29, 135]. In nature, pGAL1 regulates galactose metabolism. Galactose is a sugar

that requires more energy to break down (i.e., is less desirable as a food source) than common dextrose.

pGAL1 represses expression in the presence of dextrose and induces expression in the presence of galactose,

leading to preferential consumption of dextrose (Figure 2.5). Placing pGAL1 upstream of the material coding

region would therefore allow suppression of material production by placing the cells in a dextrose-containing

29

(a) (b) (c)

Figure 2.4: Material production can be triggered via, among other choices: (a)

chemical stimulus, which can be implemented by adding the trigger to the print

substrate; (b) thermal stimulus, in which an exogenous temperature change after

printing is used; and (c) optical, in which post-printing cells are exposed to different

intensities or wavelengths of light.

medium, then stimulation of material production by printing the cells onto a galactose-containing substrate.

This functional unification between the act of printing and the act of material production stimulus eliminated

several potentially costly development steps in the project workflow.

The final material stimulus method was therefore chosen to be inducible chemical stimulus, implemented

by pGAL1 and dextrose/galactose presence. The depth of the connection between printing process control

and material stimulus method, and the importance of achieving functional unification wherever possible, was

another key aspect of the technology concept identified here.

2.4.4 Material Delivery Method Selection

Material delivery method is the means by which the final material products (proteins, secondary materials,

etc.) are removed from the cell and delivered to the print substrate (if one is present). Some secondary

materials, such as metal ions sequestered by surface-expressed proteins, are deposited entirely outside of the

cell to begin with, making production and delivery unified into a single function. For materials which are

manufactured within the cell, there are two broad approaches to delivering it to the external environment

(Figure 2.6).

One class of approach to material delivery is to redirect existing cellular mechanisms. The general process of

transporting molecules outside of a cell is called secretion; in eukaryotes such as the host cell strain used here,

secretion may occur through several pathways, but the classical pathway for secretion of proteins involves

short ‘signal’ sequences which are part of the initially formed protein. These secretion tags signal the cell’s

internal transport mechanisms to pass the protein to and through the cell membrane; some secretion tags are

30

dextrose

galactosedextrose

expre

ssio

n

dex + gal

expre

ssio

n

galactose

expre

ssio

n

promoter terminatormaterialgal dex

dex

expre

ssio

n

gal

promoter terminator

Figure 2.5: In nature, the GAL1 promoter is attached to the gene that makes an

enzyme involved in metabolism of galactose, a sugar that provides less energy than

dextrose; the promoter has two methods of response, both suppressing expression

in the presence of dextrose and promoting expression in the presence of galactose.

cleaved as part of the transport process, leaving the un-tagged protein as the final extracellular product. There

also exist tags for membrane-binding, which cause proteins to be bound to the external membrane of the

cell without being released into the extracellular environment; such surface-expressed proteins can bind to or

otherwise interact with external compounds, an additional potential category of delivery mechanism.

The other class of approach is to forcibly remove the material, typically by killing and rupturing (lysing) the

cells; this mechanism is commonly used for industrial production of biocompounds such as pharmaceuticals.

Existing secretion pathways operate on proteins, or a few other specifically recognized compounds; cell lysis

is the only other established delivery method for non-protein or non-tagged materials.

Initially, the primary criterion was simply to use an established method of material delivery, as the project

timeline did not allow for extensive additional protocol development. Having already decided to use a protein

as the end material (subsection 2.4.2), sequestration was not a relevant approach, although it remains of

interest for future work involving secondary materials.

A review of existing work on related methods of biomolecule production (subsection 1.4.2) showed that, of

the remaining possibilities, cell rupture was typically used for large biomolecules and fibers, such as silk

and ‘plastic’ polymers, whereas genetic parts for secreting smaller proteins in yeast have been used in the

production of pharmaceuticals for decades [30, 197].

Given the relatively small size of the proteins chosen as target materials, secretion was a natural choice.

However, this stage of the proof of concept served to strongly emphasize that the choice of implementation

31

(a) (b) (c)

Figure 2.6: The material can be delivered by to the external environment via,

among other choices: (a) secretion, in which the material has a self-cleaving tag

that causes the cell to export it outside its membrane; (b) sequestration, in which

the cell expresses a protein that causes it to gather a material present in the exter-

nal environment either within or close to itself; and (c) lysing, in which the cell has

to be exogenously broken open to release the desired material.

for material delivery method is deeply intertwined with the selection of target material. The results of the

initial survey are shown in Table 3.3; this is an area that will require further characterization and development.

Given the choice of secretion as the material delivery method, the next step was to determine how to im-

plement this method in the chosen host strain and with the chosen material coding region. The secretory

mechanisms of eukaryotes are much more complex than those of prokaryotes, and secretion often requires

several different genetic mechanisms to work in concert.

A literature review identified a consensus artificial secretion signal protein section that had been proven to

work for a wide variety of proteins in S. cerevisiae [40]. This header sequence (pre-pro-peptide) attaches to

the material protein of interest and signals the cell that the protein should be secreted; the header sequence

is cleaved as part of the secretion process, leaving the desired protein intact in the extracellular environment.

The DNA sequence corresponding to this amino acid sequence was modified slightly to account for limita-

tions of the DNA synthesis capabilities available to the project and inserted upstream of the material coding

region. This sequence is included in Appendix A.

2.4.5 Material Binding Method Selection

Once the material has been delivered from the cell, via secretion or another method, it must be bound either

to other deposited material (as in material self-assembly) or to the print substrate. This is necessary both to

ensure a continuous material product and to enable the cells to be removed prior to material analysis.

32

The mechanisms used in nature to cause continuous macroscale material growth appear to involve fine control

of the cells’ immediate microenvironment (pH, chemical composition, and so forth), but are currently poorly

characterized. One of the key strengths of this technology concept is that it treats this functionality as a

black box; it is not necessary now to understand how it works as long as it is possible to stimulate the cell to

reproduce the desired behavior, and when the genetic tools are developed in the future to allow customization

of this the behavior, the technology’s potential will only be expanded. Therefore, although this part of the

technology concept is theoretically the most wide-open in terms of potential implementations, this proof of

concept was limited to the simplest material binding methods currently well-understood.

There were three criteria on which the decision of binding method was made. The first, as with the previous

implementation choices, was that it correspond to an established genetic part, a necessity given the timeframe

and budget. The second was that it not interfere with material production or delivery. The last was that it

require minimal pre- and post-processing, as reducing the overhead of biomaterial production is a key selling

point of the technology concept.

The natural place to start was by surveying existing protein purification methods, as the choices already made

regarding material selection, stimulus method, and delivery method simplified the material binding step of

the proof of concept to the task of separating a secreted protein from its parent cell. This survey quickly

identified the concept of protein affinity tags – short protein sequences added to a main protein that bind to a

specific complementary substance – as a suitable implementation.

The initial choice was a polyhistidine tag [74], which binds to nickel or cobalt ions. This method was

implemented as an additional DNA sequence downstream of the material coding region. Detectable binding

was achieved; however, cytotoxicity assays showed that the presence of nickel ions in the print substrate had

a significant negative effect on material production due to stress on the printed cells (section 2.8). This stage

of the proof of concept was therefore re-implemented with a second type of affinity tag, an antibody-binding

tag (V5 epitope [165]); this approach borrows from the technique of western blotting. Characterizing the

binding efficiency of this new implementation will be an important part of future work.

As with material stimulus and delivery achieved earlier, the implementation of ‘on contact’ affinity tags as

the material binding choice lets the act of printing also be the trigger for material binding. In the proof of

concept, therefore, printing is the trigger for all three downstream steps. Such a functional unification is

a significant advantage of the chosen component implementations. This observation highlights the need to

address all steps of the technology concept simultaneously when making implementation decisions for future

work.

33

(a) (b) (c)

Figure 2.7: Protein-based materials can be given an ’affinity tag’ which binds to a

substance present in the print substrate. Polyhistidine tags (a) bind to certain kinds

of metal ions, such as nickel or cobalt; antibody-epitope tags (b) bind to specific

antibodies. A more advanced option is (c) is self-assembly, in which a seed material

is embedded in the substrate to allow self-assembling protein or other biomolecule

constructions to form.

2.4.6 Assembly and Testing

With all of these choices of genetic sequences determined, the final step was to assemble each of these parts

and test the efficiency of the final engineered cell strain.

Several different established methods were used to assemble the five genetic parts (Figure 2.8). The first

was the TA cloning method [75] for which the chosen pYES2.1 plasmid (subsection 2.4.1) was originally

designed; this produced plasmids whose insert held only the GAL1 promoter, a coding region for yeGFP or

yeRFP, and a binding tag. These plasmids were transformed into E. coli, the E. coli screened for presence of

the plasmid via resistance to ampicillin, bulk plasmid extracted from the resulting culture of E. coli, and the

extracted plasmid transformed into the host strain of S. cerevisiae.

These yeFP-expressing yeast were verified to fluoresce red or green when exposed to galactose and deprived

of dextrose (Figure 2.9), although there was (as expected) a low level of leakage expression in the dextrose

cultures. The yeRFP cells were examined with fluorescence microscopy using the Zeiss Filter Set 43 (exci-

tation 545 nm, 25 nm width; emission 605 nm peak, 70 nm width); fluorometer tests showed that the settings

of 542 nm excitation, 612 emission, 590 nm cutoff allowed for the cleanest signal, so these were used for all

subsequent fluorometry. The equivalent values for the yeGFP cells were the Zeiss Filter Set 53 (excitation

488 nm, 10 nm width; emission 525 nm peak, 50 nm width) and fluorometer settings 470 nm excitation, 510

emission, 495 nm cutoff.

With verification completed, a new plasmid insert was required that included the secretion tag. The secre-

tion tag, yeFP coding regions, and plasmid backbone were assembled using the Golden Gate technique [53];

34

(a)

(b) (c) (d)

Figure 2.8: Our final plasmid design had five distinct parts (a). Through a variety

of assembly methods, several partial plasmids were generated (b,d) as well as the

complete plasmid (c).

phase contrast

phase contrast

red fluorescence red fluorescence

green fluorescence green fluorescence

dextrose dextrose galactose

RF

Pce

llsG

FP

ce

lls

Figure 2.9: The RFP and GFP cells both show strong expression in galactose

medium and minimal expression in dextrose medium. The non-fluorescence (phase

contrast) micrographs are included to give a sense of overall cell density.

35

this resulted in several partial plasmids, which later became useful as controls. After our first implemen-

tation, using the GFP and RFP sequences provided in [43, 89] proved to have relatively low fluorescence

(subsection 2.4.2), we adjusted the sequences to re-balance their codon frequency and had the new sequences

synthesized from scratch; these new sequences were again inserted into the plasmid using TA cloning.

After the first binding method proved to be of marginal efficiency (subsection 2.4.5), the decision was made

to express the gene construct in E. coli as well to provide a baseline for protein quantification; therefore,

a consensus ribosome binding site (used in [52], sequence given in [161]) was added first with the Flexi

digestion/ligation system (Promega), and then re-done with site-directed plasmid mutagenesis [102] after

the restriction site scar from the Flexi system proved problematic. This was the final modification to the

sequence, which is listed in full in Appendix A.

Expression of the S. cerevisiae cells bearing the final plasmids was qualitatively verified in the same manner as

with the first plasmid assemblages. The next test was to quantify the expression and to demonstrate secretion.

RFP and GFP cultures were grown in either dextrose or galactose for seven days. Each day, their density

(number of cells per mL) and fluorescence (relative to Thermo-Fisher’s Fluoro-Max 1 µm polymer micro-

spheres as a standard, part number R0100 for yeRFP and G0100 for yeGFP) were measured and recorded.

Dividing the fluorescence by the cell density gives a relative measure of mean fluorescent protein produced

per cell. Each day, an aliquot of cells from the culture was spun down and the supernatant (remaining liquid

free of cells after centrifugation at 500g for three minutes) was also measured for fluorescence; this provided

a relative measure of how much fluorescent protein was being secreted into the culture medium, rather than

simply contained within the cells. As these tests were done in conjunction with the final print medium and

substrate tests, the results and discussion are presented in Figure 2.14.

To demonstrate the effectiveness of material binding, small pieces of materials being evaluated for use as

print substrate were soaked in drawn-off supernatant produced as above. The substrates were then stained

with Coomassie Brilliant Blue (Sigma-Aldrich), a blue dye which binds to protein, and washed to remove

unbound dye according to the manufacturer’s instructions. The dyed substrates were measured for blue light

absorption to quantify the amount of bound protein (Figure 2.10). The results show that the presence of

nickel in the substrate has a clear protein-binding effect. However, the lowest level of nickel at which binding

was observed was also, in the post-printing expression tests (section 2.7), the highest level of nickel at which

detectable material expression occurred.

36

0

5

10

15

20

0M

Ni

5m

MNi

10m

MNi

50m

MNi

Pure

Ni

Nylon

Incre

ase

inD

ye

Re

ten

tio

n(R

LU

)

Relative Protein Retention

Figure 2.10: A simple test using a blue dye which binds to proteins was used to

test the differences in protein retention on several different substrates with the same

surface area. The amount of dye bound was quantified by blue light absorption.

In summary, the tests of the cell engineering component of the proof of concept show that the target choice of

material is easily detectable; that its expression is strong, and the stimulus method provides adequate control;

and that the delivery method is effective. The binding efficiency was detectable, which was the minimum bar,

but required further investigation in combination with our specific print substrate (subsection 2.7.3).

2.5 Print Medium Design

2.5.1 Medium Basis Selection

The printing medium must keep the cells alive and in a non-material-producing state prior to printing; provide

protection during printing; not interfere with cell expression post-printing; and work together with the print

substrate to form a physical support structure for the cells and to minimize cell motion (drift or migration)

after printing.

Given the choice of host cell strain (subsection 2.4.1), the printing medium needed to be an aqueous solution

containing all the elements of a yeast minimal growth medium, minus uracil. Given the choice of mate-

rial stimulus control (subsection 2.4.3), it needed to include dextrose as a carbon source, to prevent material

37

0

2

4

6

8

10

12

10%, 22 ◦

C

10%, 30 ◦

C

100%, 21 ◦

C

20%, 38 ◦

C

Vis

co

sity

(cP

or

mP

a·s

)

Measured Viscosities, PEG

0

2

4

6

8

10

12

25%, 22 ◦

C

50%, 21 ◦

C

100%, 21 ◦

C

20%, 38 ◦

C

50%, 39 ◦

C

Vis

co

sity

(cP

or

mP

a·s

)

Measured Viscosities, glycerol

0

2

4

6

8

10

12

0.50%, 22 ◦

C

0.75%, 22 ◦

C

1.50%, 22 ◦

C

Vis

co

sity

(cP

or

mP

a·s

)

Measured Viscosities, alginate

0

2

4

6

8

10

12

1.0%, 21 ◦

C

0.5%, 32 ◦

C

Vis

co

sity

(cP

or

mP

a·s

)

Measured Viscosities, agarose

Figure 2.11: Four different gel bases were tested for viscosity under a variety of

concentration and temperature conditions. The dashed line represents the limit of

the microdispensing system.

production prior to printing. (This is an excellent illustration of how the implementation choices of this tech-

nology concept are interdependent.) As a control against bacterial contamination, ampicillin, an antibiotic

which does not affect fungi such as yeast, was also a necessary component. The remaining choice was then

only whether the print medium would contribute to the physical support structure, meaning have a gelling

basis, or not.

The microdispensing system (section 2.6) has a viscosity limit of 10 cP (= 10 mPa·s), which means that the

print medium must be printed in a state not substantially more viscous than water. Thus, if the print medium

is to contribute to the physical support structure of the cell array, it must be triggered to gel after after printing.

As discussed in subsection 1.4.1, a number of biocompatible printing support scaffold materials have been

developed. Since the ingredient used as the gelling basis has by far the largest effect on total print medium

viscosity, the viscosity of all of the potential gelling bases were measured with a dip cup (Gardco) at different

concentrations under varying temperature conditions (Figure 2.11).

38

Two materials stood out as having viscosities well within the suitable region. Alginate is a polysaccharide

with an extensive history of use in biofabrication [149, 163, 188, 193]. Agarose, also a polysaccharide, is

better known from its standard use in gel electrophoresis, but has a track record in biofabrication applications

as well [63, 131, 185]. Agarose gels below a certain threshold temperature; alginate gels in the presence

of calcium ions. As adding temperature control to the printing process was not feasible given the parallel

development of the hardware and software with the cell engineering and medium design, alginate was chosen

as the print medium basis. Because the calcium exposure can be provided by printing the medium onto a

calcium-containing substrate, this choice also provides functional unification between the act of printing and

the substrate gelling; this is a substantial simplification to the workflow.

2.5.2 Anti-Aggregation Performance

The microdispensing system (section 2.6) has a particulate limit of 10 µm diameter. Experience has shown

that that it is very sensitive to the presence of particles larger than this and frequently clogs, sometimes

requiring expensive replacement of the nozzle, if the solution to be printed is not passed through a 10 µm

filter immediately prior to loading. Given that the target print solution is a cell suspension, and yeast cells

fairly rapidly form aggregates, the rate of yeast cell settling in the print medium was a parameter of high

importance.

The workspace of the micropositioning system is one cubic centimeter (1×10−6 m3). Combining this figure

with a desired minimum resolution of one cell diameter per droplet and the minimum droplet size and max-

imum droplet production rate of the microdispensing system (section 2.6) yields a maximum print run time

estimate of one hour.

A test was conducted of the settling rate of yeast cells in the print medium. A transparent container was

filled with a cell suspension of a density equivalent to an average of one cell per minimum droplet diameter

and placed in a spectrophotometer such that the absorbance near the top of the suspension volume would be

continually measured. Absorbance, in this type of setup, is proportional to cell density, so as the process of

settling depopulates the upper portion of the suspension downward, the measured density at the top of the

solution will decrease.

Two different concentrations of alginate, combined with the other basic ingredients of the print medium,

were tested in this manner (Figure 2.12). A suspenion in 1.5% alginate (w/v) resulted in no recorded settling

39

0

0.5

1

1.5

2

0 10 20 30 40 50 60

OD600

D

Time t (s)

S. cerevisiae Settling vs. Time, 1 Hour

0.75% alginate

1.5% alginate

0

0.5

1

1.5

2

0 100 200 300 400 500 600 700

OD600

D

Time t (min)

S. cerevisiae Settling vs. Time, 12 Hours

0.75% alginate

1.5% alginate

Figure 2.12: Cell settling was tested in an alginate-based medium on 1-hour (l)

and 12-hour (r) timeframes. The y-axis is in base-10 optical density, which is pro-

portional to absorbance on a log scale.

within a one-hour timeframe. Cells suspended in 0.75% alginate demonstrated a slight settling effect apparent

after approximately 35 minutes; however, the equivalent change in density at one hour was calculated to be

lower than the uncertainty in cell density due to hemocytometer counting. It was therefore determined that

both concentrates met all settling and aggregation criteria. As lower viscosities were preferred for improved

printing precision, 0.75% alginate was chosen as the basis for the print medium.

2.5.3 Cell Growth and Expression

To determine whether the print medium design had any negative affect on cell growth, the RFP-expressing

cells, the GFP-expressing cells, and reverted (non-wild-type) cell strains were grown as liquid cultures in

the print medium for a week. Aliquots were taken once a day over the course of the week and their optical

densities measured. Based on the general agreement of the three curves (Figure 2.13), the modifications

necessary to enable material production do not appear to affect cell growth when the cells are not actively

producing the target material.

With growth established, the next step was to determine whether the print medium formulation was sufficient

to fully suppress material production, and whether the equivalent galactose concentration in the print substrate

would stimulate it. Therefore, a follow-up experiment was conducted in which the cell strains were grown

in either the print medium, which contains dextrose, or a modified version containing galactose. (This was

chosen over growing the cells on dextrose and galactose versions of the print substrate as, at the time of work,

the final print substrate basis had not yet been determined.) Sample aliquots were taken once a day over the

40

0

0.2

0.4

0.6

0.8

1

1.2

1.4

0 20 40 60 80 100 120 140 160

OD600

D(∝

cells

mL

)

Time t (hours)

Density of S. cerevisiae in Dextrose vs. Time, 7 Days

unaltered

yeRFP

yeGFP

Figure 2.13: Cell density measured over a week for our unmodified host cell strain,

our RFP-expressing cells, and our GFP-expressing cells in our print medium. Points

represent means of three replicates; bars represent one standard deviation above

and below. Compare with Figure 2.20.

course of the week. Each sample had its optical density and fluoresence measured. The samples were then

spun down and the fluorescence of the supernatant measured. Fluorescence measurements were normalized

to fluorescence standards as described in subsection 2.4.6.

The effects of dextrose and galactose on secreted material yield are very clear: at all timepoints after 48 hours,

the supernatant of the galactose cells shows significantly greater expression (Figure 2.14). Two other points

were worth noting from this data. Firstly, the yeGFP reads as approximately twice as bright as the yeRFP,

which can be seen by comparing both the levels of the background ‘leakage’ expression in the presence of

dextrose and the respective increase in the presence of galactose. The yeRFP-expressing cells, however,

appeared brighter in fluorescence micrographs with equivalent exposures. This is most likely an artifact of a

better match in peak emission between the yeGFP and the green fluorescent standard than the yeRFP and the

red fluorescent standard. Secondly, the yeRFP appears to take slightly longer to accumulate to a steady level

than the GFP; this may indicate that it is more stable in the extracellular print medium. In both cases, future

work would be required for verification.

These results clearly show that the functionality of the print medium and its integration with the host cell

strains, an important step towards the overall proof-of-concept demonstration. They also establish that a

41

0

1

2

3

4

5

6

0 20 40 60 80 100 120 140 160

Flu

ore

sce

nce

Φ

Time t (hours)

Extracellular RFP Expression vs. Time, 7 Days

dextrose

galactose

0

1

2

3

4

5

6

0 20 40 60 80 100 120 140 160

Flu

ore

sce

nce

Φ

Time t (hours)

Extracellular GFP Expression vs. Time, 7 Days

dextrose

galactose

Figure 2.14: The amount of secreted fluorescent protein was quantified by mea-

suring the fluorescence of the supernatant. This measurement was normalized to a

fluorescent standard and then divided by the number of cells present in the original

solution to create Φ, a measure of the average amount of secreted fluorescence

per cell. Points represent means of three replicates; bars represent one standard

deviation above and below.

typical timeframe for material deposition is on the order of two days (48 hours). Determination of this

parameter allowed addition of a section on time costs and savings to the mission feasibility/benefit analysis

(chapter 4).

2.5.4 Cell Drift and Gelling Measurements

As the positioning of the cells is one of the primary control variables for the technology concept, quantifying

and reducing the uncertainty in it is key. In keeping with the ‘one cell per voxel’ ideal resolution criterion,

a criterion was established of keeping cell drift after printing to within one cell diameter, or 10 µm for S.

cerevisiae.

A short qualitative demonstration comparing cell motion in ordinary aqueous medium to that in different

gelling bases (glycerol, agarose, and alginate with and without calcium) on an agar substrate was conducted

in order to determine the amount of post-printing cell drift likely to occur. Videos taken at 200x magnification

showed that the baseline cell drift, for cells printed in aqueous medium, was on the order of several hundreds

of microns. Agarose reduced this to essentially zero, as it gelled completely upon cooling. Glycerol reduced

to drift by approximately an order of magnitude; alginate’s effect without calcium was barely noticeable.

Alginate with calcium, like agarose, gelled upon printing and reduced cell drift to a largely undetectable

level. (Video files are available upon request.)

42

This short demonstration indicated that the alginate-based print medium reduced cell drift to within accept-

able parameters.

2.6 3-D Printing System Design

2.6.1 Hardware & Software Design

Additive manufacturing systems typically consist of two major parts: a dispensing subsystem, which handles

feeding the print medium (material) at the appropriate rate and time, and a positioning subsystem, which

controls the motion of the build platform, material feed, and any other components necessary. In some

designs, the positioning subsystem is physically integrated with the dispenser and moves it relative to a

stationary build platform on which the finished piece is deposited; in others, the positioning subsystem moves

the build platform, the dispenser head is stationary, and only software integration is required between the two.

As reviewed in section 1.4, there are a number of commercially available microdispensing systems capable

of printing living bacterial cells (very roughly, 1 µm diameter) at ten picoliter resolution (very roughly, a

droplet diameter of 10 µm); for cells the size of yeast, more typical resolutions are on the order of hundreds of

picoliters (very roughly, a droplet diameter of 100 µm). Spatial resolution of positioning systems is typically

finer: on the order of 1–2 µm for DC-motor-based solutions and into the nanometers for piezo-based systems.

To determine the overall design requirements for the hardware and software needed for this proof of concept,

a concept cell printing run was developed that consisted of a double-walled cylinder with an outer diameter of

1 cm. This form factor was chosen as a simplified version of several load-bearing elements found in animals,

including sponge spicules and small vertebrate bones. Based on this concept print run, the set of performance

parameters for the micropositioning and microdispensing subsystems reproduced below was derived.

Printing system requirements: A combination of motion positioners, motion verification sensors (encoders),

controller/driver electronics, and software, some or all of which may be integrated with each other, sufficient

to allow commanded positioning to the following specifications:

• Range of motion in X direction (along print substrate surface) ≥ 10 mm

• Range of motion in Y direction (along print substrate surface, perpendicular to Y) ≥ 10 mm

43

• Range of motion in Z direction (perpendicular to print substrate surface) ≥ 10 mm

• Resolution (encoder or other sensor limit) ≤ 0.01 µm

• Minimum incremental motion ≤ 0.1 µm

• Positional inaccuracy (measured directly, or inferred from maximum of roll/pitch/yaw at 10

mm, as on-axis bi-idirectional repeatability, compensated backlash, or equivalent figure) ≤ 2

µm per 1 mm travel

• Controller/driver interface allows for direct, low-level commands (e.g., via command-line in-

terface – note that a proprietary pipe from 3-D modelling software does not meet this specifi-

cation)

• Controller/driver interface specifications are compatible with real-time operation (e.g., RS–

232 – note that USB may not meet this specification)

• Command set for controller/driver is well-documented and said documentation is made avail-

able

• Unix support for controller/driver – at a minimum, C libraries are provided

• Must be compatible with external hardware mounting of dispensing subsystem and printing

substrate

• Must accept commands via same in-house-developed software as dispensing subsystem

Dispensing system requirements: A combination of dispensing hardware, dispensing verification sensors (e.g.

camera), controller/driver electronics, and software, some or all of which may be integrated with each other,

sufficient to allow commanded dispensing to the following specifications:

• Must dispense droplets of volume ≤ 50 pL to 100 pL

• Must have fluid reservoir of ≥ 25 µL

• Must reliably load and dispense fluids containing particulates ≥ 10 µm in diameter

44

• Must be compatible with fluids with the following properties: viscosity ≥ 0.9 mPa·s, temper-

ature 5–40 ◦C

• Droplet throughput ≥ 5 Hz

• No bio-incompatible wetted parts

• Dispenser head empty/clean/reload cycle time ≤ 1 hr

• Must be controllable via direct interface (e.g., serial) or triggered via external signal (to allow

for synchronization with positioning system timing)

• Command interface is compatible with real-time operation (e.g. RS–232 – note that USB may

not meet this specification)

• Must be accept direct, low-commands to the following degree: trigger dispensing, query dis-

pensing status, or dispensing complete signal

• Must be compatible with external hardware mounting of positioning subsystem and printing

substrate

• Must accept commands via same in-house-developed software as positioning subsystem

• Must include droplet sensing/metrology system (e.g. focused optical feed to CCD) to allow

in-house-developed software to perform droplet verification analysis and adjustment

An extensive search of commercially available solutions for 3-D printing of living cells, including positioning

subsystems, dispensing subsystems, and complete systems, was conducted. A short list of potential devices

was prepared and each evaluated (Table 2.3).

There were very few options available on the microdispensing side; most systems reviewed with droplet size

resolution in the appropriate range were designed for pure liquids and their manufacturers could not or would

not provide guidance on the system’s anticipated performance with particulates (such as cells) in solution.

Of those who did, only one, MicroDrop, made a system which provided low-level access to basic commands

rather than requiring a proprietary software interface; since the subsystem needed to suitable for integration

into a new hardware prototype, this feature was a must-have.

45

Ta

ble

2.3

:T

he

sh

ort

list

of

co

mm

erc

ial

op

tion

sfo

r3

-Dp

rintin

go

fliv

ing

ce

llsre

vie

we

d,

with

co

nclu

sio

ns

reg

ard

ing

su

itab

ilityfo

rth

ep

roo

f-of-c

on

ce

ptd

em

on

stra

tion

.

Org

an

izatio

nS

ub

sytem

Con

clusio

n

Atten

sion

positio

nin

g,

disp

ensin

gpositio

nin

gsu

bsy

stemhas

insu

fficien

treso

lutio

n;

disp

ensin

gsu

bsy

stemnot

com

patib

lew

ith

oth

erpositio

nin

gsu

bsy

stems

Sonoplo

tdisp

ensin

gor

com

plete

positio

nin

gsy

stemhas

only

2ax

esof

travel;

disp

ensin

gsy

stemis

desig

ned

for

micro

well

plates

and

not

adap

table

HP

disp

ensin

gin

com

patib

lew

/so

lutio

ns

contain

ing

particu

lates

Lab

cyte

disp

ensin

gm

inim

um

dro

plet

sizeto

olarg

e

CH

Instru

men

tsco

mplete

min

imum

dro

plet

sizeto

olarg

e;in

suffi

cient

OE

Msu

pport;

no

pro

vid

eddata

on

perfo

rman

ce

with

solu

tions

contain

ing

particu

lates

nS

crypt

com

plete

com

pan

yw

ould

not

pro

vid

esu

fficien

tin

form

ation

toguaran

teereq

uired

perfo

rman

cew

ithout

anN

DA

Micro

Dro

pdisp

ensin

gor

com

plete

positio

nin

gsy

stemhas

insu

fficien

treso

lutio

n;

disp

ensin

gsy

stemm

eetsall

specifi

cations

New

port

positio

nin

gm

eetsprecisio

nsp

ecificatio

ns,

but

no

Clib

rariesor

Unix

support

Physik

Instru

men

tepositio

nin

gpositio

nin

gsy

stemm

eetsall

specifi

cations

Lum

enL

abs

positio

nin

gin

suffi

cient

resolu

tion

Mak

erBot

positio

nin

g,

disp

ensin

gin

suffi

cient

resolu

tion,

min

imum

increm

ental

motio

n;

min

imum

dro

plet

sizeto

olarg

e

Fab

@H

om

epositio

nin

gin

suffi

cient

resolu

tion

Org

anovo

com

plete

no

accessto

low

-level

com

man

ds;

insu

fficien

tO

EM

support;

inco

mpatib

lew

ithin

-house

soft-

ware

Thor

Lab

spositio

nin

gin

suffi

cient

range

of

travel

46

(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser(2a) piezoelectric dispenser

(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir(2c) cell suspension reservoir

(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform(1b) cell deposition platform

(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages(1a) micropositioning axis stages

(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers(1c) positioning controllers

(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller(2b) pressure controller

Figure 2.15: Our current hardware setup, with important components marked.

The final dispensing solution was the MicroDrop MD-E–6010 system. The system uses contactless piezo-

electric dispensing: the liquid to be dispensed is loaded into a narrow glass column with a piezoelectric

actuator near the nozzle end, and when the actuator is energized with a voltage pulse, the resulting pressure

wave in the glass forces a small droplet to be dispensed. The lack of moving parts, or wetted parts other than

the glass column, was a significant benefit to ease of sterilization and contamination control.

Having made the microdispensing subsystem selection considerably narrowed the choices suitable for the

micropositioning subsystem, as the two subsystems needed to be physically and logically integrated through

an in-house framework. The Physik Instrumente C–863.11/M–111.1DG controller/motor combination was

chosen due to its sharing the ability for control over RS–232 serial communications with the MicroDrop

system. Three independent motor stages were chosen and assembled in an orthogonal configuration to allow

for three-axis (XYZ) positioning control in a 1 cm3 workspace.

After receipt of the commercial-off-the-shelf hardware, the next task was hardware and software integration

of the two subsystems. A laser-cut acrylic frame with adjustable spacing to mount the micropositioning

stages and microdispenser head (Figure 2.15) was designed and built.

As shown in Figure 2.16, the software written for this project1 can take keystroke input from the user, pri-

marily for manually filling and emptying of the dispenser head, or read input files that describe the pattern to

be printed in a simple text format. The software translates the received commands into low-level commands

to the printing system, handled using separate serial port connections for the dispensing system and the po-

sitioning system. In the version of the software used with the proof-of-concept demonstration, the input files

1The software specifications and design were determined by D. Gentry, and the coding and verification were performed by A. Micks.

47

Table 2.4: Mean and standard deviation of number of cells per two droplets for a

range of dispenser head voltages and pulsewidths. Print medium had a concentra-

tion of 1.2×107 cells

mL. Each set of numbers represents five replicates.

50 µs 100 µs 150 µs

50 V 32.8 ± 15.1 28.6 ± 8.0 0 ± 0

100 V 72.0 ± 31.9 55.8 ± 28.8 37.0 ± 9.8

150 V 58.4 ± 24.6 66.2 ± 39.5 81.6 ± 7.0

include the XYZ locations where drops are to be deposited, the number of drops at each location, and the

voltage and pulsewidth to be used with the piezoelectric actuator.

2.6.2 Resolution & Cell Number

To optimize the print resolution, the following parameters were evaluated: voltage commanded, pulsewidth

commanded, starting concentration of particulates, and number of droplets per location. A battery of tests

was conducted using 1, 6, and 10 µm polymer spheres, as well as cell cultures passed through a 10 µm filter,

was conducted, an example of which is shown in Figure 2.17. The different parameter combinations were

evaluated on the bases of consistency of number of cells placed (low standard deviation) and consistency of

pattern completion (low number of zero-cell locations). Results for one of the consistency tests is shown in

Table 2.4.

The criterion of pattern completion ruled out most of the lower voltage and pulsewidth combinations. Of the

remainder, the combination of 150 µs pulsewidth at 150 V had the lowest standard deviation in number of

cells per droplet, so it was chosen for all future work (unless otherwise specified). Mean number of cells per

droplet could then be affected as needed by diluting the starting cell concentration.

2.6.3 Cell Survival and Expression

The microdispensing system is a contactless piezoelectric system that works by sending a pressure wave

through a thin glass tube containing the cell culture. This process is likely to cause some stress and damage

to the cells. Typical results in the literature for similar systems have achieved cell survival rates exceeding

90% [47, 101]; rates can be higher for systems which print multiple cells per drop [116], as the larger amount

of liquid present around each cell provides a protective effect.

48

Sta

rt

Pri

nt

instr

uctio

ns

toscre

en

Wh

atis

the

use

rin

pu

t?

Re

ad

infill

tim

eR

ea

din

em

pty

ing

tim

eR

ea

din

text

file

top

rin

tfr

om

co

ord

ina

tes

Re

ad

inxyz

Isth

ein

pu

tva

lid?

Isth

ein

pu

tva

lid?

Clo

se

pro

gra

m

Fill

for

tha

tnu

mb

er

ofse

co

nd

sE

mp

tyfo

rth

at

nu

mb

er

ofse

co

nd

sP

rin

ta

cco

rdin

gto

tha

tfile

Go

toth

atp

ositio

n

Pri

nt

err

or

toscre

en

Re

turn

tosta

rt

f

ep

m

q

no

no

no

no

ye

sye

sye

sye

s

Any

oth

er

inp

ut

Isth

ein

pu

tva

lid?

Isth

ein

pu

tva

lid?

Pri

nt

err

or

toscre

en

Pri

nt

err

or

toscre

en

Pri

nt

err

or

toscre

en

Re

turn

tosta

rt

Fig

ure

2.1

6:

Aflow

ch

art

for

the

so

ftw

are

tha

tco

ntr

ols

the

bio

pri

nte

r,ta

kin

ga

nin

pu

tfile

tha

td

escri

be

sth

ep

rin

tp

atte

rn

an

dse

nd

ing

co

ntr

olsig

na

lsto

the

po

sitio

nin

ga

nd

dis

pe

nsin

gsyste

ms.

49

Beads per Five Droplets vs. Voltage, Pulsewidth

0

1

2

3

4

0 2 4 6 8 10 12

0

1

2

3

4

0 2 4 6 8 10 12

0

1

2

3

4

0 2 4 6 8 10 12

0

1

2

3

4

0 2 4 6 8 10 12

0

1

2

3

4

0 2 4 6 8 10 12

0

1

2

3

4

0 2 4 6 8 10 12

0

1

2

3

4

0 2 4 6 8 10 12

0

1

2

3

4

0 2 4 6 8 10 12

0

1

2

3

4

0 2 4 6 8 10 12

Occu

rre

nce

s

20

0V

15

0V

10

0V

Number of Beads

100 µs 150 µs 200 µs

Figure 2.17: One set of plots showing the number of 10 µm beads per print lo-

cation for a range of dispenser head voltages and pulsewidths. Print medium had

a concentration of 1.9×104 beads

mL. Each plot represents six replicates. A battery of

these tests was conducted to determine the optimal printing parameters.

50

0

1

2

3

4

5

Flu

ore

sce

nce

(φ)

Yeast Survival via Dead Stain Post-Printing

RFP GFP

dead printed dead printed

Figure 2.18: RFP and GFP cells were treated with a fluorescent dye that only

stains dead cells. The fluorescence level resulting from staining printed cells were

compared to staining an equivalent number of heat-killed cells to establish a floor

for the number of cells killed by printing.

A lower limit for the viability of dispensed cells was established by using SYTOX Green (Molecular Probes),

a nucleic acid stain that can only enter cells with compromised membrane integrity (and which are therefore

generally non-viable). A ‘killed control’ was created by leaving a 1 mL aliquot of the cell strain suspended

in PBS (a neutral salt buffer) in a 70–80 ◦C water bath for 10 minutes [123]. A 25 µL aliquot of the same

suspension was then loaded into the dispensing system and printed into a sample container. Both the printed

cells and the killed control were stained with 5 µM SYTOX Green following the manufacturer’s protocols.

The entire 25 µL printed cell sample and 25 µL of the killed control were then each diluted into 200 µL

of PBS to reduce background fluorescence. The 200 µL final samples were measured in the fluorometer to

determine the amount of overall fluorescence (normalized as before) and then in the spectrometer to determine

the cell density, allowing calculation of the amount of fluorescence present per cell (φ ). The results are shown

in Figure 2.18.

Because the efficiency of SYTOX Green in staining non-viable cells cannot be assumed to be the same

between heat-killed cells and cells killed by the printing process, whose mechanisms of damage are not

perfectly understood, the results can only be used to set a floor for survival: a minimum survival rate of

40% for the yeRFP-expressing yeast and 60% for the yeGFP-expressing yeast. While the true rate is likely

substantially higher, this is a sufficient demonstration to prove that enough cells survive to achieve our proof

of concept metrics (Table 2.1).

51

2.7 Print Substrate Design

2.7.1 Substrate Composition Selection

There are five basic functions that the print substrate must provide, either alone or in conjunction with the

print medium:

• keep the cells alive after printing;

• allow (in combination with the print medium) for a sufficiently low level of cell ‘drift’ post-

printing;

• support the printed cells in an actively secreting state for long enough to allow a sufficient

material yield;

• bind the secreted material so that it can be isolated; and

• allow removal (washing away) of the cells post-secretion.

As with the print medium (section 2.5), the majority of the print substrate components were determined by

earlier implementation choices (Figure 2.19). The choice of S. cerevisiae ura3 as host cell strain required an

aqueous component and the presence of yeast nitrogen base plus uracil-deficient amino acid supplements as

a selective nutrient source. The choice of the GAL1 promoter for material production control required the

presence of galactose as a carbon source and secretion expression stimulant. The initial choice of nickel-

binding affinity tags as material binding sites required the presence of nickel ions. The choice of alginate as a

gelling basis meant that the substrate also had to contain calcium ions. The open-air print process meant that

ampicillin was necessary to prevent bacterial contamination during the active material deposition phase.

A number of different gel bases were tested for their ability to provide biocompatible physical support, in-

cluding nitrocellulose membrane, positively charged nylon membrane, alginate, glycerol, nickel-plated alu-

minum, agarose, agarose with a surface coat of nickel-NTA beads, agar. Agarose was chosen after tests

with the yeRFP and yeGFP cells determined that it had acceptably low interference with fluorescence detec-

tion. The remaining tasks in finalizing the print substrate were centered around determining the appropriate

concentrations of nickel and calcium.

52

Figure 2.19: Many of the substrate components were determined by a priori

choices, including the presence of nickel ions, galactose, calcium ions, and the

yeast nitrogen base plus selective dropout supplements.

2.7.2 Post-Printing Growth and Expression

To test that the cells would be able to produce sufficient amounts of material after printing, growth and fluo-

rescence assays of cells manually deposited on the print substrate were conducted. The first test simply com-

pared growth in the presence of galactose (Figure 2.20) to growth in the presence of dextrose (Figure 2.13),

following the same experimental protocol described in subsection 2.5.3. The growth tests showed slower

growth on the galactose-containing (non-gelled) print substrate compared to the dextrose-containing print

medium – as expected due to the lower efficiency of galactose as a metabolic energy source compared to

carbon – but no other ill effects; the maximum population was slightly higher in galactose, despite its slower

overall growth.

The second test compared per-volume fluorescence Φ, measured as described earlier in subsection 2.5.3,

in two different concentrations of alginate (as would be present when the cells in the print medium were

dispensed onto the substrate) at a variety of concentrations of calcium and nickel. The readings taken without

calcium or nickel are shown in Figure 2.21; as the alginate gels in the presence of calcium, the fluorescent

readings taken at all other concentrations show significant noise in comparison.

The fluorescence data from the full battery of tests is shown in Figure 2.22. The presence of calcium appeared

to have essentially no negative effect on material production. The presence of nickel, however, was clearly

harmful at all but the lowest concentration. Interestingly, the higher concentration of alginate appears to offer

some protective effect against the higher nickel concentrations; this may indicate that the alginate, which

53

0

0.2

0.4

0.6

0.8

1

1.2

1.4

0 20 40 60 80 100 120 140 160

OD600

D(∝

cells

mL

)

Time t (hours)

Density of S. cerevisiae in Galactose vs. Time, 7 Days

unaltered

yeRFP

yeGFP

Figure 2.20: Cell density measured over a week. Experimental procedure is the

same as in Figure 2.13 with the non-gelled print substrate. The modifications nec-

essary to enable material production do not appear to affect cell growth when the

cells are actively producing.

0

0.2

0.4

0.6

0.8

1

1.2

0 10 20 30 40 50 60 70

Flu

ore

sce

nce

Φ

Time t (hours)

yeGFP Expression in Alginate vs. Time

0.75%1.5%

Figure 2.21: The normalized per-volume fluorescence Φ of cells on the print sub-

strate was measured over time to determine whether the presence of calcium and

nickel in the substrate would reduce material production. This is the reading from

the control, with neither calcium nor nickel added.

54

yeGFP Expression in Alginate vs. Time0.75%

1.5%

Flu

ore

sce

nce

Φ

0.1

00

MC

a0

.05

0M

Ca

0.0

20

MC

a0

.01

0M

Ca

0.0

05

MC

a0

.00

1M

Ca

0.0

00

MC

a

Time t (hours)

0.000 M Ni 0.001 M Ni 0.002 M Ni 0.005 M Ni 0.010 M Ni

Figure 2.22: The normalized per-volume fluorescence Φ of cells on the print sub-

strate was measured over time at different concentrations of calcium and nickel. As

the presence of calcium causes gelling of the alginate, which can interact with the

fluorescence measurement in unpredictable ways, these readings are very noisy.

takes up some of the calcium ions as part of the gelling process, was also also taking up nickel ions (which

have the same charge) at a lower efficiency.

Based on these results, the final choices for the proof-of-concept demonstration were chosen to be 1 mM

nickel chloride and 20 mM calcium chloride.

2.7.3 Material Binding and Washing

To further characterize the polyhistidine tag binding efficiency, which was qualitatively demonstrated in

subsection 2.4.5, the Dynabeads His-Tag Isolation and Pull-Down (Life Technologies) was performed. In

55

L 1 2 3 4 5 6 7 8 9

Figure 2.23: The protein bands resulting from a polyhistidine tag binding test. The

lanes are, from left to right: (L) SeeBlue Plus2 protein ladder, (1) RFP lysis product,

(2) GFP lysis product, (3) RFP flow-through, (4) GFP flow-through, (5) RFP wash,

(6) GFP wash, (7) RFP elution product, (8) GFP elution product, (9) wild-type elution

product. Arrow marks expect position of yeGFP and yeRFP.

this assay, the product of cell lysis (containing all proteins produced by the cells) are flowed past beads

with a cobalt surface, which binds polyhistidine tags. The beads are then extracted and washed to remove

non-specific binding products; the washed beads, now holding only those proteins with a strongly binding

polyhistidine tag, are finally treated with an elution buffer to remove the purified proteins.

Extracts from all four stages (lysis, flow-through, wash, and elution) of the purification of yeRFP, yeGFP, and

wild-type yeast were run on an electrophoresis gel, the image of which is shown in Figure 2.23. The lysis

solutions show no visible bands due to their high dilution factor. The lanes for the flow-through solutions

show bands corresponding to other proteins, but not to the molecular weight corresponding to RFP or GFP,

showing that if RFP or GFP is present, it is well bound by the cobalt beads. The wash solutions also show

no bands, indicating that none of the other proteins produced by the host cell strains had detectable binding.

The elution solution lanes for the RFP and GFP cells each showed a band of the right molecular weight to be

the fluorescent protein which was not present in the wild-type yeast.

The combination of quantity of tagged material (protein) produced by the host cell and the binding efficiency

of the polyhistidine tags to the cobalt Dynabeads is clearly therefore sufficient for detection. However, cobalt

was shown to be even more toxic to the host cells than nickel in early biotoxicity assays (data not shown).

56

(a) (b)

Figure 2.24: (a) The material template, printed in two colors to represent the two

different cell types. The beads are 6 µm in diameter. (b) The results of the two-color

grid print test with simulated cells, with the original material template overlaid. All

locations received at least one ’cell’ except for one, the farthest right red location in

the second row from the bottom. Image magnification is 250x.

Thus, although this material demonstration is sufficient for a proof of concept, it also establishes that binding

using polyhistidine tags and metal ions (whether nickel or cobalt) is not a promising implementation for future

development. Future work will be required to investigate other possible alternatives, such as antibody-epitope

pairs and protein-protein (or protein-polysaccharide) binding domains.

2.8 Printing Demonstration & Performance Characterization

The genetic engineering was by far the largest part of the proof-of-concept work. In order to allow work on

other parts of the project to proceed in parallel, the functionality of all other components (Figure 1.2) was

tested and verified using 6 µm polystyrene beads (Polysciences) in two different colors, red and blue, to

represent the two cell strains.

The red and blue beads were suspended in the print substrate, loaded into the printing system, and dispensed

in the five-by-five grid pattern onto the print substrate. The result, shown next to the original template, is

shown in Figure 2.24. Several of these print test runs were performed and analyzed using microscopy. The

results were compiled into the form of performance metrics and compared against the originally generated

metrics in Table 2.5.

57

Once the genetic engineering was completed for yeRFP- and yeGFP-expressing yeast, the cells were printed

in the same alternating pattern as the beads, and micrographs of the resulting grid were taken on a fluorescence

microscope using a different filter to view each fluorescent protein. Due to the limited field of view in the

fluorescent microscope, the grid was scaled down to a three by three array. The micrographs are overlaid to

show both fluorescent fields in Figure 2.25, making the alternating pattern is apparent.

This, in combination with the material production and binding tests shown earlier, demonstrates critical end-

to-end functionality of the proof-of-concept system.

2.9 Results & Discussion

The final proof-of-concept demonstration was a two-material (non-structural) pattern: a “checkerboard” pat-

tern of red and green fluorescent proteins. Overall, the demonstration showed that each component of the

technology identified in Figure 1.2 had at least one feasible implementation.

The process of creating the proof-of-concept demonstration, as intended, brought to the forefront a number of

needed answers and new questions about the technology concept. Given the overall scope of the technology,

it was apparent from the early conceptual stage that the technical implementation of the concept would be

application-dependent, but this work has clearly demonstrated the significant process-design effort involved

in optimizing the approach for a given material product. This is, in part, a consequence of its interdisciplinary

nature; each design decision has meaningful biological, chemical, physical and mechanical consequences that

must be taken into account. The degree of interdependence between the different implementation decisions

involved is substantial; nearly every choice made substantially impacted the options available at every other

step. The most important linkages are represented in Figure 2.26.

The work has clearly demonstrated that there is at least one complete biological, material, and mechani-

cal ‘chassis’ that meets the performance requirements of this technique. Comparing the final results to the

originally derived metrics, the minimum requirements are met in every case (Table 2.5). Although the proof-

of-concept prototype does not meet the state of the art in most cases, the work performed here identified that

in many of these such a level of performance was not necessary; an example is the larger cell print resolution

requirement derived from the larger size of the host cell strain. Further, many of the implementation choices

can be fine-tuned in future work; suggestions for improved alternate implementations and future development

pathways are presented in chapter 3.

58

Figure 2.25: yeRFP-expressing yeast cells were printed in an alternating grid with

yeGFP-expressing yeast cells. The width of the array is 800 x 800 µm, sized to

fit the fluorescent microscope’s field of view. (top) Grid imaged separately with red

and green fluorescent filters. (bottom left) Grid imaged with phase contrast. (bottom

right) Composite image of red and green fluorescence.

59

Table 2.5: The performance metrics derived for the proof of concept compared to

the final measured metrics. The minimum was met or exceeded in every case.

Metric Minimum Desired Current

positioning precision 1 cell diameter ≈ 10 µm 1 µm 2 µm

dispensing volume ≤ 1000 cells“voxel”

≈ 1 cell“voxel”

≤ 10 cell“voxel”

“voxel” size enc. 1000 cells = 1 nL 10 pL 20 pL

cell survival 50% 90% ≈ 50%

pattern completion 75% 95% ≥ 95%

Another overall takeaway is the importance of functional unification in process control. In this proof of

concept, the physical act of printing was able to be the trigger for material production, material delivery,

substrate gelling and support, and material binding. Without this simplification, the demonstration would

have been far more time-consuming to create. This combination of physical cues is something that needs to

be taken into account very early in the implementation design process.

The individual component steps also yielded several key findings.

The cell engineering is by far the most labor-intensive and complex part of the technology concept’s im-

plementation. It requires choosing a host cell to be printed, a material to be deposited, and identifying and

assembling genetic parts for the material, material stimulus, material delivery, and material binding. It ac-

counted for approximately three quarters of the total project effort. Thus, it is recommended that in future

work, time and resources be allocated accordingly.

The selection process for the material stimulus method generated insights regarding the relationship between

stimulus type, spatial and temporal resolution, and material form factor (Table 3.2). This is a vital piece

of information for any future work in this area. Similarly, the material delivery method selection process

generated significant refinement of the understanding of the link between delivery and material type, although

this is a weaker determination.

Another realization of note was that the concept of a ‘voxel’, as the unit of three-dimensional printing, is

a significant oversimplification of a parameter in this technology. Cell size, cell functional unit size (if

different from cell size), material organization scale, dispensing droplet volume, ‘number resolution’ (the

precision with which the number of cells in a droplet can be controlled), positioning accuracy, and stimulus

spatial resolution are all part of the ‘resolution’ of a printing system creating arrays of living cells. Although

the final print performance (Table 2.5) remains expressed in voxel resolution, as the original metrics were

expressed in these terms, in future work more specific figures of merit will need to be derived and justified.

60

Figure 2.26: The linked decision-making pathways of each of the implementation

choices. Understanding these interdependencies is key for future work on the tech-

nology concept.

Although it may not be relevant in industrial or commercial application, for research purposes, detailed ma-

terial analysis is an important additional step. Many of the choices made here were affected by the frequent

need for immediate, low-cost material deposition verification (fluorescence microscopy or fluorometry). Fu-

ture work will be necessary to determine if there is a need to monitor production rates this closely for struc-

tural material deposition – for example, for real-time feedback to stimulus control – or if this step can be

eliminated once the technology has reached a more mature level.

Finally, implementation of the proof of concept allowed the observation that the time costs of the technology

are not necessarily intuitive. As shown in Figure 2.14, there is a certain amount of time (approximately

two days) necessary for the cells to reach maximum material deposition. However, this time cost does not

scale with the size of the material; once the cell array has been printed at the correct cell density, the delay

for material deposition will always be two days, whether the cell array is millimeters or tens of meters in

size. The time cost that scales with form factor and size is the printing time. The current prototype prints

approximately 1 mL of array in 1 hour with approximately 1 µm resolution; however, this ratio will be

substantially improved for material and stimulus types that require a lower cell spacing resolution. Moving

to a more parallelized print system (akin to having multiple jets in an inkjet print head), or one specialized

for volume throughput, will also affect this part of the overall time cost.

61

These insights point to two primary outstanding needs in enabling technology. The first is for identification

and characterization of genetic parts for new material types. This includes both existing structural biomateri-

als – especially those with multiple constituent materials – and, further in the future, de novo gene design. A

particular need is genetic tools to control delivery methods other than secretion; most of the most promising

biomaterials have inorganic phases at least partially deposited outside of the cell, a mechanism currently out-

side the scope of fine control with synthetic biology. Being limited to secretion-related genetic parts, or using

a black box approach to extracellular deposition without human-created controls, will prevent this technology

concept from reaching its full potential.

The second key need is for better modelling and predictive tools for complex materials. Current technology

is just beginning to scratch the surface of predictive functionality tools for protein folding; modelling prop-

erties and self-assembly of synthetic biomaterials will require understanding the interactions between several

additional layers of structure beyond that. Such tools will need to be developed to, one day, allow de novo

hierarchical material design.

62

3 Future Development Pathways

3.1 Introduction

A secondary objective was to survey options and provide recommendations for hardware, biological, and ma-

terial implementations for the next stages of technology development. One of the main benefits of achieving

the proof-of-concept demonstration (chapter 2) and space mission feasibility/benefit analysis (chapter 4) is

answering those questions about cost and benefit that can be answered with the current state of knowledge

and bringing to light those that cannot – in other words, to let us know what is and isn’t known. This technol-

ogy concept is innovative, multidisciplinary, and extremely broad in potential scope; therefore, a significant

part of both the proof of concept and the mission scenario analyses consisted of identifying the unknowns

encountered raised and evaluating future paths to addressing them.

This survey is presented in two sections. The first consists of short-term recommendations for follow-on

work, particularly new or alternative potential implementations of the different components of the technology

concept. The second section contains discussion of the longer-term needs identified over the course of the

proof-of-concept work, including advances in other fields, and describing potential pathways forward in those

areas.

3.2 Short Term Implementations

3.2.1 Target Materials

Two primary factors determine the suitability of existing biomaterials for near-term future implementation.

The first is how advantageous the material’s mechanical properties are in comparison to existing engineering

63

material choices. The second is the degree to which the genes and mechanisms responsible for material

production have been identified. Both of these factors were evaluated in a survey of a range of promising

structural biomaterials in Table 4.5, Table 4.8, and Table 4.9.

Based upon this work, the short list of materials most promising for future implementation as part of this

technology concept includes spider silk, silkworm silk, keratins, cellulose, mollusk shell, and crustacean

shell. The two major factors of concern are firstly, whether the genetic parts responsible for their production

have been identified, and secondly, whether they have been expressed in another organism, an important

second step towards standardization.

Of these materials, spider silk comes closes to having a pre-identified genetic part; one of its major con-

stituent proteins has been cloned and expressed in E. coli [186]. Silkworm silk is a close second, as a

significant percentage of its major proteins have had their corresponding genes sequenced [171, 194]. Ker-

atins are a broad family of structural proteins found in most land vertebrates; some human keratins have been

sequenced [68, 99] and expressed in E. coli [69], though they are more typically expressed in mammalian

cell lines for biomedical purposes. Spider silk has the advantage that a single one of its major proteins can

alternate between tensile and compressive forms through being folded into either crystalline or amorphous

phases [174]. Pairs of certain keratins will spontaneously assemble into fibers [154], although control of the

properties of the resulting fibers is uncertain.

Cellulose, which is a polysaccharide, not a protein, is produced in plants by a complex arrangement of

membrane-bound proteins formed by at least ten different genes [173]. Some bacteria produce a different

form of cellulose [82]; though some of the genes responsible for the total synthesis pathway have been

identified and cloned [84, 168], their full synthesis function has not been reproduced. As cellulose is not

an ordered material, its production will not allow testing of the hypothesis that microscale material structure

control is possible with this technology concept; however, the genetic bases for the formation of the highly

ordered materials mollusk shell and crustacean shell are at a much lower state of knowledge.

A longer term avenue of investigation could be the investigation of empirical control of biomineralization

by printed living mollusk or insect shell-lining cells; this would lose the precise control aspect of the overall

technology concept, but might be a promising first step towards the most promising compressive materials.

However, for the short term, the recommendation is that future work focus on either spider silk or a keratin

protein pair, with cellulose as a fallback. These recommendations are summarized in Table 3.1.

64

Table 3.1: The potential materials proposed for immediate follow-on investigation,

in order of implementation preference.

Material Identified Expressed Notes

spider silk ≥ 2 major proteinsa one protein [58, 186]

silkworm silk ≥ 2 major proteinsa only partial sequences [171, 172, 194]

keratin 2 major proteinsa certain pairs [68, 99, 154]

cellulose (bacterial) ≥ 3 genes in combinationb no full synthesis function [84, 168]a Common, but not universal, structure of the material family.b Function of proteins in production not yet fully understood.

3.2.2 Host Cell Strains

One of the reasons behind the choice of S. cerevisiae as the host cell strain for the original proof of concept

was its high potential for extensibility to future work (subsection 2.4.1). It is recommended that immediate

follow-on work use the complete biological chassis detailed here, as its performance is well-characterized

and well above the minimum performance requirements (section 2.8). However, there are two other microor-

ganisms worth noting that may be of interest for particularly specialized applications.

• Escherichia coli is the most common model organism for prokaryotic genetic engineering.

Although it cannot be positioned as living cells with single cell resolution in three dimensions

given current dispensing technology, technology exists capable of fabricating 2D and 3-D

arrays of larger aggregates of bacteria [41, 116], and it can be grown relatively precisely in

2D configurations through clever use of small-scale configuration of nutrients and growth

factors [91]. It may therefore be suitable for use for a two-dimensional application, such

as creation of fine channels or ‘wires’ of material, or for creation of materials without scale

hierarchies smaller than tens of microns.

• Thalassiosira pseudonana. A frequently studied diatom reviewed for use in the proof of con-

cept due to its natural deposition of silica (subsection 2.4.1). Diatoms produce external, finely

patterned silica in their natural state, although the genetic tools necessary to give useful con-

trol over the cells’ silica deposition have not yet been established. However, if the target

material is silica, it is worth experimenting with diatoms to see whether simply printing them

in given patterns, or printing them onto particularly patterned substrates, is sufficient for the

requirements of the given application.

65

3.2.3 Support Medium & Substrate

The selection of a new target material will require design of a new support medium and substrate. Given

the recommendation to continue to use the existing host cell strain for immediate future work, it is unlikely

that the basic requirements (water, sugar, trace metals, amino acids, contamination control) will substantially

change. This leaves as the remaining concerns the requirements that the medium and substrate, either alone

or in combination, provide the raw material necessary for material formation, and that they are compatible

with the material production stimulus, delivery method, binding method, and verification.

Spider silk, silkworm silk, and keratin are all proteins; thus, no additional components beyond those required

for cell growth would be required. Pure polysaccharides can similarly be made without supplementation.

Biomineralized materials would require a richer source of the mineral components, such as calcium carbon-

ate for mollusk shell. Further, assembly of many structural biomaterials, particularly polysaccharides and

biominerals, requires secondary ‘helper’ proteins and even substrates that are not part of the final material; to

reduce the dependence on finding genetic parts for each of these intermediates, which may not yet have been

identified or characterized, such secondary materials could be provided as part of the substrate as well.

The dextrose/galactose pairing of the print medium and substrate in the current work was a necessary con-

sequence of the use of the GAL1 promoter. Those working with applications which use a different material

stimulus method are advised to simply use dextrose as a carbon source in both, as it results in faster growth.

The proof of concept work found no interference between the use of secretion as a delivery method and either

the use of alginate in the print medium or the use of agarose in the print substrate. The most likely alternative

delivery method for short-term future work is cell lysis; if this is done by heat, the agarose may melt. There are

photocured biocompatible substrates (see references cited in section 1.4) which may eliminate this difficulty

if encountered.

Binding is another important function of the medium and substrate. This is the area of functionality most

likely to be in need of application-specific adjustment in short-term future work; however, most alternative

implementations of binding can be achieved simply by changing the binding additive in the substrate, and

thus are covered below in subsection 3.2.7.

Material characterization and verification is another consideration. In the proof-of-concept work, autofluores-

cence of both the medium and substrate was a significant obstacle to obtaining clean verification of material

66

Figure 3.1: Combining printing the substrate, with differing chemical compositions,

with printing the cells will allow us finer control over material production and depo-

sition.

expression, deposition, and binding; even the purity of the agarose was an important factor in improving

micrograph quality. Alterations to the medium and substrate should be checked to ensure they do not conflict

with material verification, particularly if fluorescence tags or fluoresence-tagged antibodies are being used as

a verification tool.

3.2.4 Production Stimulus

Over the course of this work, a number of potential material production control stimuli have been surveyed

(Table 3.2). The primary factors to consider when choosing a control stimulus for future work are, firstly, how

well the type of stimulus corresponds to the desired form factor, and secondly, whether there is an existing

genetic part (promoter or other regulatory sequence) that corresponds to it.

Form factor matters because different types of stimuli have different spatial and temporal resolutions. Tem-

perature is easy to control in a large gradient, either linear or exponential, but hard to change rapidly or over

short distance scales. Chemical exposure can either be changed rapidly using fluidics, or achieve high spatial

resolution if the substrate is printed with different compositions. Optical stimulus can achieve high spatial

and temporal resolution over a 2D space at a low cost, but 3-D spatial resolution within a partially opaque

cell array becomes much more costly.

67

Table 3.2: Different material production stimuli match different form factors.

Stimulus Form Factors

chemical internal cavities, large solid shapes

optical sheets, fine surface structures

thermal density or composition gradients

There are relatively simple heat-responsive eukaryotic promoters [162], and a wide variety of chemically

sensitive ones that respond to salts, metal ions, sugars, and so on (e.g., [29, 135]). The optically sensitive

regulatory systems that have been identified are substantially more complex, involving cascades of multiple

mechanisms [159, 164].

Since much short-term work will likely focus on thin material sample sections, all three options are likely to

be relatively inexpensive to produce in terms of process control. This leaves the labor cost of implementing

the genetic engineering for a given stimulus as the determining factor. Therefore, chemical control appears

to be the most attractive option for short-term work, with implementing optical stimulus as the next logical

step. Implementation of high spatial resolution stimuli might also lower the minimum cell printing resolution

for certain types of materials, although the strength of the effect would have to be determined experimentally.

With immediate future work likely to rely on chemical stimulus, it is worth mentioning the possibility of 3-D

printing the substrate as well as the cells (Figure 3.1). The printing hardware is compatible with the current

print substrate, so multiple-material printing capability is the only remaining barrier to achieving the ability

to create spatial patterns of different chemical stimuli. An upgrade of the printing hardware to this level is

underway.

3.2.5 Hardware & Software

The most immediate short-term improvement to the bioprinting hardware and software is the addition of

multi-material printing. Currently, this is achieved by multi-pass printing – loading and printing one medium,

then emptying the dispensing system and loading and printing a second medium onto the same substrate.

This leads to the challenges of properly aligning each pass, desynchronization of material production due

to different cell strains sitting out for different periods of time prior to incubation, and to poor scalability

stemming from a linear increase in print time with the number of materials used. The addition of a second

print head is planned to alleviate these concerns, although implementation of this upgrade will not take place

68

before the cessation of the current phase of work. Future investigators are encouraged to identify the need for

this capability early and build it into the system from the start if required.

The software used in the current bioprinting setup is very much a functional placeholder. The control program

developed for this work integrates serial communication with the controllers of the micropositioning and

microdispensing subsystems to send commands stored in a basic text file. The text file is generated from a

set of developed GNU Octave [132] scripts that use a pre-set array configuration to generate a droplet pattern

based on a given size and number of print locations. One major future improvement will be integrating the

software with standard 3-D CAD formats or programs to allow the input files to be generated from arbitrary

array geometries.

The microdispensing technology will likely be sufficient for all short-term future work unless the host cell

strain is changed. Slightly smaller cells can be printed with smaller-aperture glass needles using the same

piezoelectric actuator, although this will not be sufficient to allow single-cell printing of typical-size bacteria.

Larger cells would likely require moving to a more traditional liquid dispensing system, such as an automated

pipette or (for large cell aggregates) peristaltic systems. The same micropositioning system could likely be

used with any of these technologies, as long as an appropriate mounting system was designed and constructed.

The demonstrated resolution of the micropositioning system should make it sufficient for all future cellular

array work within a workspace of one cubic centimeter. For applications requiring larger arrays, the simplest

upgrade is to use the same mechanical approach (high-quality DC motors and encoders) with larger ranges

of motion; three-axis position systems with ranges of motion over several meters exist, albeit generally with

reduced resolution. In the longer term, more radically different deposition methods exist; for example, a

deposition system integrated into a mobile platform that moves along the structure as it is constructed [79].

Another potential approach is to break down large pieces into many repeating smaller pieces which can be

manufactured using a small workspace and assembled robotically to full scale [35].

It should be remembered that not all possible cell array construction approaches require full 3-D printing

capability. One possible implementation would replace deposition entirely with something analogous to

extrusion, in which a thin cross-section of cells is used to produce material that is continuously drawn away

as it is deposited (Figure 3.2). The cells could either be bound to a substrate (e.g., through binding of surface-

expressed proteins) and continuously moved away from the deposited material, or the material functionalized

(e.g., through a magnetic tag) so that it can be drawn away from the cells without affecting cell placement.

69

Figure 3.2: A hypothetical alternative approach to cell array construction: if either

the cells are or material are functionalized so that they can be selectively subjected

to an external force, then a thin cross-section of cells could be used to produce a

much larger piece of material that is continuously drawn away as it is deposited.

3.2.6 Material Delivery

A number of potential material delivery methods (Table 3.3) were surveyed. The primary factors to consider

when choosing a delivery method for future work are, firstly, how well the type of delivery corresponds

to the desired material type, and secondly, whether there is an existing genetic part (secretion tag or other

pre-processing sequence) that corresponds to it.

The relationship between material type and delivery method is discussed in subsection 2.4.4. Of the options

surveyed, by far the best studied and least controlled approach is cell lysing. The ‘classical’ secretion path-

ways are also relatively well understood, although it is known that cells also transport materials to the external

environment by other means. Metal or inorganic salt sequestration and deposition, in contrast, is generally

poorly understood, although it is often assumed to involve either secreted or surface-expressed proteins. The

entry ‘sequestration’ in Table 3.3 is intended to represent this combined class of mechanism.

Given the present state of knowledge, it is recommended that short-term future work use secretion as a de-

livery method, with cell lysing as a secondary option for materials incompatible with known secretion tags.

Implementation of inorganic materials, whether through extracellular deposition or true control of biominer-

alization, is dependent on future work in other fields identifying the necessary biosynthesis mechanisms.

70

Table 3.3: Different material delivery methods match different material types.

Method Material Examples

secretion proteins, silica, other biominerals

sequestrationa copper, iron, nickel, other metals

lysing silk, cotton, other fibers, bioplasticsa Term used generally to cover a variety of extracellular deposition

mechanisms.

3.2.7 Material Binding

The material binding method, both for material binding to the print substrate and for material binding to

neighboring cell-produced material(s), is the implementation most likely to require future investigative work

to improve performance. The available alternatives in the near-term future depend on the type of material

used.

Protein-based materials, such as spider silk or keratin, can be engineered to include ‘affinity tags.’ These

are short peptide sequences, typically placed at the beginning or end of the protein, that bind strongly to

a particular substrate. The polyhistidine tag used in the proof of concept is an affinity tag that binds to

nickel or cobalt ions. Other affinity tags are epitopes to which antibodies will bind. The antibodies or ions

themselves can be bound to a larger substrate material to provide more flexibility. Short complementary

peptide sequences can be used similarly to provide the same functionality.

Beyond small affinity tags, there are many known protein domains (large functional regions) which bind to

particular substrates. These include binding to other proteins [10, 136], to sequence- or shape-specific DNA

regions [106, 111], RNA regions [50], lipids [60], particular carbohydrates including chitin and cellulose [51,

98, 140], and more exotic materials such as metals [17, 90, 103, 195], often in combination. Although de novo

design and production of such protein binding domains for arbitrary substrates is in a preliminary state, there

are a number of known binding domains that could be used if compatible with the target material. Future

development of such design capability will substantially expand the range of possible implementations of

material binding method.

Lastly, many biogenic materials, such as some molecular crystals, lipid bilayers, and certain polymers [61],

are self-assembling if the necessary environmental conditions are met. If the target material is known to

exhibit such behavior, then seeding the print substrate with ‘seed’ assemblages of such structures to initiate

the assembly process is another possibility worth investigation.

71

3.3 Long Term Approaches

Given that the technology concept is innovative, multidisciplinary, and extremely broad in potential scope,

it is to be expected that a significant part of the proof-of-concept work has been identifying specific relevant

unknowns relevant and investigating options to characterize them. In section 3.2, the focus was on immediate

work that could be done to advance the proof-of-concept implementation. In this section, a list is presented

of those longer-term questions identified by the proof-of-concept work which must be answered to gain a

complete understanding of this technology’s potential costs and benefits. The items are presented in rough

order of estimation of time to future emergence, given current trends in research and development in the field.

• Single-cell printing resolution. The only limiting factor in the cell array construction was

the low repeatability of the microdispensing system. Although the droplet volume was very

precise, the variability in the number of cells contained in each droplet causes undesirably

high variability in the number of cells placed at each location. Possible solutions include

installing a monitoring system that can detect the number of cells printed in real time and re-

print over the same location until the desired number is reached. This is not outside the scope

of integration with existing commercial devices in the next few years, as the metrology system

would essentially be a simplified flow cytometer.

• Micropositioning resolution. The best achievable resolution of DC-motor-based systems is in

the 1 µm range, which exceeds the diameter of most eukaryotic cells. Finer resolutions can

be achieved with piezo-based systems for work which requires prokaryotic cells; the limiting

factor here, as discussed above, will be droplet resolution and repeatability.

• Stimulus control. Control of the stimuli used to regulate cell material production and structur-

ing must be at a fine enough spatial and temporal scale. Chemical stimuli are limited only by

droplet size, if the print substrate is pre-printed to include the stimulus. Optical stimulus can

be over an order of magnitude more precise in space, and resolved on microsecond timescales,

if technology comparable to that in a scanning laser microscope is used; however, opaque or

semi-opaque supports or scaffolds will sharply limit the depth at which it can be used. Other

forms of contactless stimulus are worth investigation, but are comparatively poorly character-

ized. For instance, many microbes naturally respond to near-visible (ultraviolet or infrared)

electromagnetic radiation, which have very different transmittance curves than visible light

72

in many biologically relevant materials. Others respond to magnetic fields. Less traditional

options include electromagnetic radiation at lower frequencies, or even gravitational fields,

if the technology is used in a zero-gravity environment (or one simulated by centrifugation).

Quantifying these limits will be important for determining the cost and potential benefit of

implementing each.

• Identification of new genetic parts corresponding to stimulus, delivery, and target materials.

Although rapid generation of libraries of characterized genetic parts is a common goal in

synthetic biology, few existing efforts focus on genes related to structural materials. Fewer

still focus on new control or delivery mechanisms related to material production. Although

one can use black box empirical techniques, such as growing the cells which natively express

a material in close to their original environment, deep material customization will remain

elusive until the commensurate genetic tools are available.

• In situ resource extraction. A strong selling point of this technology concept from a sustain-

ability or space mission perspective is its ability to build a wide variety of materials from basic,

locally extractable resources. This advantage will only be further magnified as in situ resource

utilization technologies, both on Earth and in space, improve.

• Compliant materials modelling tools. The combination of compliant and rigid elements in

natural biomaterials gives them their extreme toughness, resilience, and energy-dissipating

ability. Existing material stress-strain modelling tools do not handle these properties well [22].

This is a current area of materials modelling research, due to its applications to emerging

robotics and architectural concepts such as those based on tensegrity.

• Modelling tools for self-assembling materials. Although this need shares some common

ground with software tools such as predictive protein folding models and computational nano-

materials design, the many additional layers of structure in the most highly ordered biomate-

rials will require a specialized toolset. This is a novel requirement created by this technology

concept; estimated time to development is unknown.

• Identification of biomaterial mechanical properties in a relevant environment. Both terrestrial

and space-based applications can require structural materials that are robust to extremes of

temperature, humidity, chemistry, and radiation. How much of a challenge this will be to the

usefulness of synthetic biomaterials is a key unknown. In the short term, it can be addressed

73

by testing natural materials under simulated environmental conditions. However, in order to

predict the performance of new materials created with this technology concept as part of the

design and synthesis process, this function will ultimately need to be added to the proposed

modelling tools referred to above.

• Material properties for design tools. Most structural design tools, such as finite element anal-

ysis software, assume essentially constant material properties and calculate maximum allowed

loading conditions accordingly. To fully tap this technology concept’s potential, this process

has to be inverted; the key ability is calculation of fine-grained material properties to suit a

given shape and load.

• Design of new genetic parts. De novo gene design and synthesis is still an emerging area.

Ultimately, it will be necessary to develop these tools in order to create materials fully cus-

tomizable down to the molecular level. For encoding target materials, it is likely fifteen to

twenty years before this becomes a bottleneck for the technology concept; however, novel

design of binding regions could be highly advantageous in a much shorter timeframe. Regard-

less, this technology concept will not achieve its full potential without both capabilities.

A suggested technology development roadmap is shown in Figure 3.3.

74

Fig

ure

3.3

:T

his

tech

no

log

yd

eve

lop

me

nt

roa

dm

ap

follo

ws

cu

rre

nt

estim

ate

sa

nd

pro

jectio

ns

ab

ou

tth

esta

teo

fth

e

invo

lve

dfie

lds

un

de

ra

be

st-

ca

se

sce

na

rio

with

asu

bsta

ntia

lre

se

arc

he

ffo

rtto

wa

rds

bri

ng

ing

this

tech

no

log

yco

nce

ptto

fru

itio

n.

Row

s,

sta

rtin

gfr

om

top

:th

isp

roje

ct’s

an

ticip

ate

dfu

ture

pa

th;

ad

ditio

na

lsyn

the

tic

bio

log

ya

nd

bio

en

gin

ee

rin

g

wo

rk;

mo

de

ling

an

dd

esig

nto

ols

ad

van

ce

me

nts

;d

eve

lop

me

nts

ino

the

rfie

lds.

75

76

4 Mission Context Feasibility/Benefit Analyses

4.1 Introduction

One of the secondary objectives of this work (section 1.5) was a feasibility and benefit analysis of this tech-

nology in a space mission context. The potential applications of this technology are sufficiently broad that

two potential missions, chosen to span a variety of radiation, thermal, gravitational, and in situ resource avail-

ability challenges, were chosen for investigation. The first, a near-term possibility, was creating a structural

replacement part aboard the International Space Station (section 4.2); the second, a much longer-term mission

design, was providing a long-term Mars human habitat with construction materials (section 4.3).

For each mission context, the costs and requirements of using currently existing technology (the ‘reference

mission’ scenario) and using this proposed technology (the ‘proposed mission’ scenario) were compared.

As many of the requirements are derived from the materials chosen for implementation, the schemes shown

in Table 4.1 and Table 4.2 were developed and used to evaluate the materials of the reference mission for

use in the proposed mission. Table 4.1 is used to rate materials according to their availability, with prefer-

ence for those that can be produced using existing in situ resource utilization (ISRU) technology, but taking

into account anticipated future ISRU developments. Table 4.2 is used to rate materials according to their

compatibility with the technology concept; the emphasis is on state of readiness for the genetic engineering

component of the technology, as the proof-of-concept work showed that this is by far the most significant

criterion (section 2.9).

The feasibility and benefit analyses focused on the following requirements:

• Material types. Each mission’s needed materials according to the scheme in Tables 4.1 (whether

the raw ingredients are available) and 4.2 (whether the material can be produced by cells). Po-

tential replacements were identified for materials with a C or D rating.

77

Table 4.1: In situ re-

source grades.

Grade Definition

A directly available in situ

B present in situ, extractable

C present in situ, not extractable

D not present in situ

Table 4.2: Material compatibility

grades.

Grade Definition

A standardized genetic part

B observed natural genetic part

C theoretical de novo genetic part

D not producible by cells

• Material form factors. The gross form factors (beams, sheets, laminates, etc.) correspond-

ing to the material needs were also identified; this information is needed to determine the

workspace, speed, and resolution requirements of the hypothetical print hardware.

• Mass and energy. A combination of data from the proof of concept and projections based on

flown space hardware will provide models for the upmass and energy budget of each proposed

mission scenario.

• Labor, time, and effort. Observations from the proof of concept and projections based on

known automated solutions to similar technology challenges will inform the estimates of hu-

man effort required for each proposed mission scenario.

4.2 An ISS Replacement Part

4.2.1 Mission Description and Assumptions

The International Space Station (ISS) is a human-built structure in low Earth orbit continuously occupied by

humans and resupplied through launches from Earth. As all parts and contents of the ISS must be sent up from

Earth in an expensive and carefully controlled launch, the ISS’s components are exhaustively documented and

extremely well characterized. The reference mission scenario was chosen to be replacement of a discrete part

(subsection 4.2.2), manufactured on Earth and sent up during a routine resupply mission. The proposed

mission scenario consisted of replacing a redesigned part made from hypothetical synthetic biomaterials with

one manufactured onboard, including all labor and equipment necessary to perform manufacture of synthetic

biomaterials.

78

As this mission context was chosen to represent a minimal, short-term implementation of this technology

concept, several simplifying assumptions were made. Firstly, in both scenarios, it is assumed that the original

part (the one to be replaced) was designed and manufactured on Earth and included as part of the original ISS

module assembly. In the proposed scenario, it is therefore also assumed that the necessary material design

and template files for fabrication of the synthetic biomaterial part already exist.

Secondly, the proposed mission scenario assumes that sufficient advance planning was performed to ensure

that all infrastructure necessary for manufacture of all synthetic biomaterial parts was included on the ISS;

thus, although the mass and energy requirements for this infrastructure are included as costs, it is assumed

that all cell lines have already been engineered and stored as cryostocks (small frozen seed cultures) and

that the required design and command files are stored locally rather than having to be transferred via remote

communications. (Remote, automated gene assembly and cell culture is likely to be a proven technology

within a decade, but was not included here in the interests of presenting a minimum, immediate-term benefits

case.)

Thirdly, the hypothetical mass, energy, and labor specifications for the synthetic biomaterials manufacture in-

frastructure are assumed to be equal to the benchtop hardware, and experiential process of researchers without

a background in microbiology, molecular biology or genetic engineering educating themselves through lit-

erature searches and hands-on trial and error, used in the proof of concept. This is clearly too conservative,

as complete packages which perform cell culture have been fit inside satellites weighing less than 5 kg and

occupying less than 0.005 m3 [121], and targeted training to use existing 3-D printers, including loading

existing design files and prepared, contained ‘feedstock’ cartridges, can often be performed in an afternoon;

however, extrapolation from the proof of concept effort will at least provide a clear upper bound.

4.2.2 Target Materials Types and Form Factors

A number of potential parts were evaluated for use in the mission definition. These included fluid lines and

connectors, micrometeorite shielding patches, air filters, hand tools such as wrenches, and medical supplies.

In the interests of simplicity, the final part chosen was a tether designed for extravehicular activities (EVA),

composed of a Nomex band with aluminum hooks (SED33104391–301 in the EVA Tools and Equipment

Reference Book [124]).

79

The Nomex portion of the tether was chosen as the focus for this mission scenario. It is modelled as a

compliant strap with constant cross-section and properties along its length. It is assumed to be loaded in

uniaxial tension.

4.2.3 Feasibility, Costs and Benefits

The work in this section was performed jointly by D. Gentry and A. Micks. D. Gentry defined the analyses to

be done and identified the reference sources. A. Micks collected the reference data from the sources and ran

the original analyses. A. Micks and D. Gentry documented the original results jointly. D. Gentry performed

the expanded analyses and wrote the documentation necessary to create this presentation of the work.

Mass & Energy Requirements

The tether design in the reference mission is the same as that specified in [124]: a band of Nomex, 0.025

m wide and 2.146 m long when fully extended without strain, massing 0.42 kg. In its planned function, the

tether is is initially folded up and held by stitching that breaks under a 34 kg load; after these stitches break,

the tether has a load limit of 635 kg.

In the reference scenario, a replacement tether is flown up on a routine resupply mission. The reference re-

supply mission is modelled on the SpaceX CRS–2, a commercial ISS cargo mission provided by the company

SpaceX using the Falcon 9 (v1.0) launch vehicle and Dragon cargo spacecraft [127]. In this mission, a pay-

load of 677 kg was delivered to the ISS1 [126] out of a total launch mass of approximately 333,000 kg [65].

As the significant variabilities in launch timing, availability, payload space, and so on are beyond the scope

of this analysis, this simplified figure of combined payload mass ratio of 492:1 will be used to represent the

‘mass cost’ of resupply from Earth.

The energy costs of resupply are the energy required to reach orbit and the fractional cost of the capital and

upkeep of the infrastructure necessary. The propellant equivalent to the energy required to achieve orbit is

already counted as a mass cost. It is further assumed that parts resupply only displaces other cargo, i.e., that

if less payload mass were required for resupply, the extra capacity would be used for additional science or

mission payload mass. Thus, the energy cost of resupply in the reference mission scenario is treated as zero.

1A theoretical capacity of 3310 kg is often reported in promotional materials, but the demonstrated value is used here.

80

Table 4.3: Conservative mass and power estimates for the equipment required for

on-site synthetic biology and storage of cell cultures for manufacture of biomaterials,

according to the proposed mission scenario.

Item Mass (kg) Power (W)

fixed costs

freezer 300 3350

incubator 70 0

micropositioning system 1.2 0

microdispensing system 5.0 0

control computer 2.0 0

fluidics integration 1.5 0

per unit costs

incubator 0 236

micropositioning system 0 1.8

microdispensing system 0 50

control computer 0 50

fluidics integration 0 36

dry consumables 0.008 0

In the proposed mission scenario, additional infrastructure is required to be aboard the ISS for on-site bioma-

terial manufacture, some of which will be under constant use (e.g., cryostocks must be frozen until needed),

some of which will be used only during manufacture (e.g., the printing control computer). Table 4.3 shows

highly conservative estimates of these mass and power infrastructure requirements based on the benchtop

implementations used in the proof of concept. (The exception is an estimate for fluidics to connect the cell

regrowth system to the printing system; this transfer is currently done by hand, but would require automation

in a space mission environment.)

The calculated mass cost of 380 kg is dominated by the temperature-controlled conditions required for cryo-

stock storage and cell regrowth, as is the background additional power cost of 3350 W. General-purpose

cold stowage and incubation systems already aboard the ISS perform similar functions with greater mass,

but lower power [26]; a single-purpose, optimized microbial growth system, such as those used in nanosatel-

lite payloads like GeneSat–1 [94] and O/OREOS [128], would have the same capabilities at a fraction of

the mass, power, and volume. The calculated per-print energy cost is a far more reasonable 374 W, again

reducible to a fraction of that with a mission-optimized design. (For comparison, the ISS’s solar arrays can

generate 84–120 kW [1].)

In addition to the per-print energy costs, there are per-print mass costs. In particular, the cells require aqueous

growth medium and substrate. While the exhausted medium can be reprocessed after use to reclaim much

of the water, new dry nutrients must be added to produce the next batch. A print run requires, as a rough

81

Table 4.4: Conservative total mass and power estimates for the fixed and per-unit

costs of the reference and proposed mission scenarios.

Fixed Costs Mass (kg) Power (W)

reference 0 0

proposed 379.7 3350

Per Unit Costs Mass (kg) Power (W)

reference 0.42 0

proposed 0.188 373.8

estimate, an equal combined volume of medium and substrate to the volume of the final piece, and both are

approximately 3% dry ingredients with the density of water; thus, for this piece occupying approximately

0.27 L, approximately 8 g of consumables are needed beyond the mass incorporated into the final piece.

The final question of the proposed mission scenario is what the mass of a tether constructed from synthetic

biomaterials meeting the same load requirements would be. Table 4.5 compares the suitability of a range of

biogenic and non-biogenic materials to the reference materials of Nomex and aluminum 6061-T6 (“Al”). The

original tether design masses 0.42 kg; from the volume estimates, it is assumed that the aluminum accounts

for two-thirds of that mass and the Nomex tether the remainder. Given the published tensile yield strength and

density of each material, and modeling each as axially symmetric and loaded in pure tension, it is possible

to calculate both the area ratio (the cross-sectional area needed to meet the same tensile strength as either Al

or Nomex) and the mass ratio (the mass per unit length at that cross-sectional area compared to the reference

material).

Of the biogenic materials rigid enough to form a hook, oak (loaded along the grain) would meet the same

tensile requirements with the lowest mass ratio (0.5). Of the potential tether replacement materials, silk is the

equivalent at 0.31. (The values listed for fibrous biomaterials are for individual fibers; while the premise of

synthetic biomaterials is that such performance can be extended to larger, microstructure-optimized pieces,

that advantage would not extend to non-biogenic fibrous materials such as Kevlar.) Thus, structurally opti-

mized oak- and silk-based synthetic biomaterials would enable the manufacture of a tether with equivalent

performance that masses 0.18 kg, less than half the reference mass.

Given the high fixed costs of enabling the manufacture of synthetic biomaterials aboard the ISS, it is natural

to ask how many parts would need to be manufactured before the proposed mission scenario’s costs ‘break

even.’ The mass cost of the proposed mission scenario exceeds that of the reference mission scenario after

1618 printings, equivalent to 291 kg of printed product. At that point, the payload mass savings is, of course,

82

380 kg, equivalent to 186,900 kg of launch mass. This comes at an energy cost of 167 GJ (assuming that the

incubator must be run for five days, and the printing system for 180 minutes, per print run).

Labor & Effort Requirements

In the reference mission scenario, the labor and effort costs are only those associated with preparing, launch-

ing and receiving the replacement part. These costs are assumed, in the long run, to be roughly equivalent to

the costs of preparing, launching and receiving the infrastructure necessary for biomaterial fabrication. Given

that the mass of parts that must be replaced must be approximately equal to the mass of the infrastructure

for the proposed mission scenario to be advantageous (section 4.2.3), and that that mass easily fits within a

single resupply cargo payload [126], this is reasonable.

The labor and effort costs for the proposed mission scenario can be divided into three major parts: the time

required to design, build and space-qualify the necessary hardware, software, and operational protocols; the

time required to train astronauts and ground support crew in the use of the resulting system; and the time

required to operate the system.

The system itself consists of fluidics, piezoelectric actuators, cell culture systems, temperature-controlled

storage, small-scale automated positioning systems, and a control interface. Each of these are types of sys-

tems that have been flown as part of integrated payloads or modules before (e.g., [2, 76, 87]), so the devel-

opment effort is not expected to exceed that of a typical biological science payload. While this is typically

substantial (1–3 years) in absolute terms, it is quite low in terms of the range of overhead to be expected when

switching technologies in a space mission context.

The training time, based on experience in the proof of concept, is estimated to be approximately 10 hours if the

unit is completely self-contained (fluidics integration between cold storage, incubation, and printing system)

and 50 hours if, in a more realistic short-term scenario, crew are required to manually transfer self-contained

cartridges between subsystems. If print runs are performed less than once per month, a once-per-month

refresher of one hour is recommended. Given that the proof of concept work was performed by personnel

with substantial, broad technical and research education but without prior experience in microbiology or

molecular biology, a reasonable representation of the skill basis and cost of a typical ISS crew, these estimates

appear reasonable.

83

Ta

ble

4.5

:C

om

pa

riso

no

fth

ere

qu

ired

cro

ss-s

ectio

na

la

rea

Aa

nd

ma

ss

mo

fN

om

ex

an

da

lum

inu

m6

06

1to

alte

rna

tive

ma

teria

ls,

inclu

din

gb

iog

en

ica

nd

trad

ition

ale

ng

ine

erin

gm

ate

rials

,fo

rth

ep

urp

ose

of

alo

ad

-be

arin

gte

the

r.V

alu

es

for

fib

rou

sb

iom

ate

rials

are

for

ind

ivid

ua

lfib

ers

;o

the

rfib

rou

sm

ate

rials

are

liste

da

sco

mp

osite

s.

Va

lue

sfo

rw

oo

da

refo

r

dry

wo

od

alo

ng

the

gra

in.

Va

lue

sfo

rstru

ctu

red

bio

ma

teria

lsa

ssu

me

no

flaw

so

rd

efe

cts

.

Materia

lstren

gth

σt

(Pa)

den

sityρ

(kg/m

3)N

om

exA

ratio

Nom

exm

ratio

Al

Ara

tioA

lm

ratio

Notes

Kev

lar49

2.8

0×10

91450

0.2

30.2

40.0

90.0

5[1

15

]

silicon

carbid

e2.6

0×10

92560

0.2

50.4

60.0

90.0

9[7

3]

carbon

fiber

2.2

0×10

92000

0.3

00.4

30.1

10.0

8[8

]

silk2.0

0×10

91300

0.3

30.3

10.1

20.0

6[8

]

cellulo

se1.0

0×10

91500

0.6

50.7

10.2

40.1

3[8

]

titaniu

m(allo

y)

8.3

0×10

84730

0.7

82.6

80.2

90.5

0[1

6]

Nom

ex6.5

0×10

81380

1.0

01.0

00.3

70.1

9[1

15

]

steelA

ISI

302

2.6

0×10

87920

2.5

014.3

50.9

22.7

0[1

6]

alum

inum

6061-T

62.4

0×10

82710

2.7

15.3

21.0

01.0

0[1

6]

com

pact

bone

1.6

0×10

81850

4.0

65.4

51.5

01.0

2[2

8,46

]a

nacre

1.3

5×10

82950

4.8

110.2

91.7

81.9

4[1

3]

b

oak

(pin

)1.1

2×10

8630

5.7

82.6

42.1

40.5

0[ 9

7]

copper

7.0

0×10

78910

9.2

959.9

53.4

311.2

7[1

6]

pin

e(p

ondero

sa)5.7

9×10

7400

11.2

33.2

54.1

50.6

1[ 9

7]

nylo

n6,6

4.5

0×10

71140

14.4

411.9

35.3

32.2

4[1

6]

poly

carbonate

3.5

0×10

71200

18.5

716.1

56.8

63.0

4[1

6]

rubber

1.4

0×10

71000

46.4

333.6

417.1

46.3

3[ 8

]

concrete

4.6

9×10

62400

138.7

3241.2

751.2

245.3

6[1

44

]a

brick

2.0

0×10

61900

325.0

0447.4

6120.0

084.1

3[1

04

]ac

cork

1.1

5×10

6160

565.2

265.5

3208.7

012.3

2[ 1

47

,148

]a

aR

epresen

tative

valu

ech

osen

from

repo

rtedran

ge.

bD

ensity

of

pu

rearag

on

iteu

sedas

con

servativ

eestim

ate.c

Den

sityn

ot

giv

enin

reference;

estimated

from

com

mercial

sou

rces.

84

Table 4.6: Conservative labor estimates for each step of biomaterial manufacture

in an ISS mission context.

Step Time (h)

reconstitution and inoculation 2

system loading 1

print operation 2

array transfer to incubation 1

piece removal and post-processing 2

disposal and reclamation 2

The final labor and effort consideration is the time required to operate the system. Again, based on the

experience of the proof of concept, this is estimated to consist of three stages. In the first stage, the desired

cell lines are removed from cold storage, placed in growth media, and incubated for two days (48 hours); this

stage includes the time necessary to reconstitute the growth media, print media, and substrate. In the second

stage, the grown cells, print media, and print substrate are loaded into the printer, the desired design template

is loaded, and the print run is performed; afterwards, the printed array is placed in an incubator for three

days (72 hours). In the third stage, the finished piece is removed from the remains of the array and all waste

disposed of.

Time estimates for each step of the operation are provided in Table 4.6. The minimum time assigned to a

step is one hour to allow for the additional overhead of conducting even basic pick-and-place operations in

zero gravity. The time for the print run is based upon the speed of the prototype system used in the proof of

concept and the volume of the target piece. The estimates for post-processing (removal of extra substrate, etc.

from the finished piece) and reclamation (processing of remaining substrate and medium to remove water)

are loose, as these steps are not required until a higher state of development than what was achieved in the

proof of concept work.

Other Considerations

The quantitative analysis here covers the mass, energy, and labor costs incurred by use of this technology

concept in one space mission environment. There are additional factors for which quantitative characteri-

zation will not be possible without future investment in development that are nonetheless worth qualitative

consideration at this stage.

One is the compounding effect of a part manufacturing technology which can be used to produce multiple

types of material and form factors. This is not captured well here due to the focus on a single part, and

85

the assumption that the part breaks infrequently enough that each part requires an individual production run.

The values shown in Table 4.3 and Table 4.6 make it clear that the energy, time, and labor costs both drop

dramatically if multiple arrays, or single arrays designed to deposit multiple parts, are produced, a simple

operational change. The infrastructure mass cost need only be counted once against the mass savings from

replacing any number of parts with synthetic biomaterial equivalents, and the per unit mass, energy, and time

costs remain essentially constant no matter the final piece. Further, the payload volume of the required per

unit mass is nearly minimized, as the same basic materials (water, nutrients, trace minerals) can be used to

produce any number of end biomaterials and shipped in bulk rather than in predetermined form factors.

The other major factor is the potential of other unique properties of synthetic biomaterials, beyond basic

stress-strain and mass characteristics. As described earlier, many structural materials are far more fracture-

resistant and more resilient in the face of deformation than traditional materials. They often have markedly

different stress-strain responses and failure modes as well. The possibility of using these characteristics to

improve mission robustness, efficiency, and capacity, rather than simply duplicate existing functionality with

greater repair and replacement capability, strongly warrants further investigation.

4.2.4 Summary

As shown in the previous sections, upgrading existing materials to optimized synthetic biomaterials can

reduce the mass of load-bearing parts substantially – over half, in the case studied here. If the further benefit

of on-board manufacture and replacement is desired, the upper limit of the mass required is about 380 kg.

Extrapolating from the mass ratios in Table 4.5, the break-even point for mass reduction is likely 300–400 kg

of material printed; how rapidly this is encountered depends on how broadly the technology is implemented,

but would take over a thousand print runs if only used for small, infrequently replaced parts. The energy

required is significant, but well within the capacity of the ISS’s solar cells, and pales in comparison to the

potential reduction in launch mass.

86

4.3 A Long-Term Mars Habitat

4.3.1 Mission Description and Assumptions

The second space mission context chosen was a long-term Mars human settlement. Unlike the ISS, this

analysis relies on a purely hypothetical reference mission scenario. The basic assumptions about mission

architecture are drawn from the Human Exploration of Mars Design Reference Architecture 5.0 (“DRA”)

from NASA’s Mars Architecture Steering Group [125].

This analysis focuses on the design and construction of a human habitat, the centerpiece of a Mars human

settlement. The requirements for a habitat on the Martian surface are strenuous. The thin Martian atmosphere

requires that the habitat be airtight and mechanically strong when pressurized to Earth atmospheric pressure;

the low temperatures, that it provide thermal insulation to reduce the energy required for heating; the high UV

flux, that it provide radiation shielding to its inhabitants; and the dust and potential for micrometeorites, that it

be puncture-resistant. Given the extraordinary transport costs of mass to the Martian surface, it must accom-

plish all of these while being as light as possible. As few traditional engineering materials can be optimized

for more than one of these requirements, existing space structures are constructed from several functional

layers: in the case of an Extravehicular Mobility Unit, for instance, fiberglass, aluminum, and stainless steel

for structural support, neoprene bladder for airtightness, Dacron restraint for the bladder, aluminized Mylar

insulation, and an orthofabric outer layer (a mix of Gore-Tex, Kevlar, and Nomex) for thermal insulation and

puncture resistance [124].

The primary type of habitat considered here is the expandable habitat, a concept which has demonstrated

several benefits over rigid structures for aerospace applications (see [25] for more details):

• The flexibility of expandable structures means that they can be packaged more compactly and

in a wider variety of geometries, and can therefore be launched in smaller, cheaper vehicles.

• Expandable structures lend themselves to the thin sheet form factor, which is better suited to

composite weaves and other materials that have higher strength to mass ratios than traditional

rigid materials.

87

• Structural supports, inherently rigid or inflated to rigidity, can be positioned such that the

composition of the structure varies from one point to another, concentrating the strength of the

system where needed and saving mass elsewhere.

• Expandable structures are flexible, and can therefore be more stress tolerant, deforming instead

of breaking as a rigid system might if this is desirable behavior.

• Expandable alternatives have often proven to have lower development and manufacturing costs

than their rigid counterparts.

These advantages are reminiscent of those offered by the use of synthetic biomaterials themselves, making

the combination particularly appealing. Specifically, a design is considered in which high-tensile-strength

materials provide the skin over a frame made of high-compressive strength materials. For both the reference

and proposed mission scenario, the habitat is modelled on a design study for an inflatable lunar habitat [130]

consisting of modular units, each one of which is a ring of identical, rigid, circular-cross-section columns

supporting a pressurized “skin.”

The original reference habitat design uses Kevlar 49 both as the tension-stressed dome material and as the

compression-stressed column material through the use of inflatable, overpressurized columns. As the use of

inflated columns converts the material requirements for the column into tensile strength as well, and one of

the purposes of this analysis is to cover a broad range of potential applications for synthetic biomaterials, this

work uses a modified reference habitat design with compressive aluminum columns.

Radiation is handled through the straightforward, commonly suggested approach of constructing the habitat

far enough underground that the local regolith provides sufficient shielding; it is assumed that the typical

shielding effects of the Martian and lunar regolith are comparable [6, 117], and that wherever the density is

lower, the regolith will be layered more thickly, making the shielding regolith mass per unit area of the habitat

the same.

One of the hypotheses behind this technology concept is that synthetic biomaterials can incorporate the prop-

erties of multiple materials into a single composite layer tailored to the desire form factor. Biomaterials,

in particular, are generally highly puncture- and fracture-resistant. Therefore, the feasibility/benefit analysis

presented here covers an outer structure providing both puncture resistance and mechanical support. Ra-

diation shielding is provided by regolith cover as described above; the remaining requirements of thermal

88

insulation and airtightness are assumed to be handled by a non-modelled inner structure in both the reference

and proposed cases.

4.3.2 Target Materials Types and Form Factors

The habitat design requires two material types and form factors: columns loaded in compression due to the

weight of regolith, and sheets (the dome covering) loaded in tension due to pressurization. The dome is

modeled as a spherical section, and the columns (four per habitat unit) as having constant circular cross-

section, loaded in uniaxial compression, and each identical in length and loading conditions.

4.3.3 Feasibility, Costs and Benefits

Mass & Energy Requirements

The first focus of this analysis is the tension-loaded skin of the habitat dome. For a thin spherical section

of material with radius of curvature r pressurized to p (note that this is the gage pressure, i.e., the difference

between the interior and exterior pressures), the tensile stress σt is related to the material thickness tm by

Equation 4.1 [16].

σt =pr

2tm(4.1)

The reference habitat design specifies a radius of curvature of 6.1 m and an internal pressure of 69 kPa (10

psi). The surface pressure of Mars is a mean of 636 Pa, with a minimum of approximately 400 Pa depending

on location, season, and weather [183], so as a conservative assumption it is treated as zero and the gage

pressure equal to the internal pressure. The radius of curvature and length and width of the unit both being

6.1 m means the area of the spherical section of the roof2 is area of 141 m2.

If the pressure and radius are taken as fixed, then the minimum required material thickness – and hence the

mass for a given material density – is inversely proportional to the maximum allowable stress, which will be

2Estimated from the solid angle of a right regular pyramid with four sides, each of length 6.1 m, and whose slant height – corre-

sponding to the radius of curvature – is also 6.1 m. This angle is 3.79 steradians.

89

the material’s tensile yield strength multiplied by a safety factor. Kevlar 49 has a tensile strength of 2.8×109

Pa [115]. Thus, using a safety factor of 1.5, the reference design dome of Kevlar 49 requires a 0.11 mm thick

sheet massing 23 kg.

Table 4.5 shows the tensile properties of several potential synthetic biomaterial substitutes and traditional

engineering materials. The same materials are presented in Table 4.7 with the allowable stress, required

thickness, and required mass if used to replace the reference material for the habitat dome. Silk is the clear

biomaterial choice, requiring just slightly more than the mass of the reference dome at 0.16 mm thickness

and 28.9 kg.

The second focus of this analysis is the compression-loaded columns of the habitat. The compressive strength

σc required to support a load of force Fc is shown in Equation 4.2, where A is the cross-sectional area.

σc =Fc

A(4.2)

In the reference habitat design, the necessary regolith shielding was determined to be 3.3 m thick. Mars is

generally subject to a lower radiation flux than the moon, but the same shielding thickness will be used in

the proposed scenario as a conservative estimate. Martian regolith has an average density of 1500kg

m3 [6], so

a 3.3 m thick layer over a cross-sectional square 6.1 m to a side is a load of 184,190 kg. Assuming each of

the four columns supports an equal fraction of the load, and given that the gravitational acceleration on the

surface of Mars is 2.3 times that on the surface of the moon [183, 184], the compressive loading in both the

reference and proposed mission scenarios is 171 kN per column. With aluminum columns 6.1 m in height

and a safety factor of 1.5, this requires that each column mass 16.3 kg. (Rigidity and buckling would also be

significant issues to consider, but they are not included in this analysis.)

Table 4.8 shows the compressive strength and other properties of several potential synthetic biomaterial sub-

stitutes and traditional engineering materials with the allowable stress, required thickness, and required mass

if used to replace the reference material for the habitat dome. Nacre is the clear choice, requiring less than

half the mass of the reference columns at 7.7 kg per column.

The remaining question is that of in situ resource utilization (ISRU). Although silk is not quite as light as

Kevlar 49 for the same performance, silk is a biogenic material, and Mars – through sunlight, CO2, H2O, and

regolith – contains all the necessary basic chemical species for life. Although many of these species are in

90

Ta

ble

4.7

:T

he

req

uire

dth

ickn

ess

t ma

nd

ma

ss

mfo

rte

nsile

sh

ee

tsfo

ra

Ma

rsh

ab

ita

tm

issio

nsce

na

rio

usin

gd

iffe

ren

t

ma

teri

als

,so

me

tra

ditio

na

l,so

me

bio

log

ica

llyd

eri

ved

.V

alu

es

for

fib

rou

sb

iom

ate

ria

lsa

refo

rin

div

idu

al

fib

ers

;o

the

r

fib

rou

sm

ate

ria

lsa

relis

ted

as

co

mp

osite

s.

Va

lue

sfo

rw

oo

da

refo

rd

ryw

oo

da

lon

gth

egra

in.

Va

lue

sfo

rstr

uctu

red

bio

ma

teri

als

assu

me

no

flaw

so

rd

efe

cts

.

Mate

rial

stre

ngth

σt

(Pa)

den

sity

ρ(k

g/m

3)

stre

ssσ

ma

x(P

a)

t m(m

)m

ass

m(k

g)

Kev

lar

49

2.8

0×

10

91450

1.8

7×

10

91.1

3×

10−

423.0

5

silk

2.0

0×

10

91300

1.3

3×

10

91.5

8×

10−

428.9

3

carb

on

fiber

2.2

0×

10

92000

1.4

7×

10

91.4

3×

10−

440.4

6

sili

con

carb

ide

2.6

0×

10

92560

1.7

3×

10

91.2

1×

10−

443.8

3

cell

ulo

se1.0

0×

10

91500

6.6

7×

10

83.1

6×

10−

466.7

7

Nom

ex6.5

0×

10

81380

4.3

3×

10

84.8

6×

10−

494.5

0

oak

(pin

)1.1

2×

10

8630

7.4

9×

10

72.8

1×

10−

3249.4

8

tita

niu

m(a

lloy)

8.3

0×

10

84730

5.5

3×

10

83.8

0×

10−

4253.6

5

pin

e(p

onder

osa

)5.7

9×

10

7400

3.8

6×

10

75.4

5×

10−

3307.5

0

alum

inum

6061-T

62.4

0×

10

82710

1.6

0×

10

81.3

2×

10−

3502.5

9

com

pac

tbone

1.6

0×

10

81850

1.0

7×

10

81.9

7×

10−

3514.6

5

nac

re1.3

5×

10

82950

9.0

0×

10

72.3

4×

10−

3972.6

3

nylo

n6,6

4.5

0×

10

71140

3.0

0×

10

77.0

2×

10−

31127.5

9

stee

lA

ISI

302

2.6

0×

10

87920

1.7

3×

10

81.2

1×

10−

31355.8

5

poly

carb

onat

e3.5

0×

10

71200

2.3

3×

10

79.0

2×

10−

31526.0

6

rubber

1.4

0×

10

71000

9.3

3×

10

62.2

5×

10−

23179.3

0

copper

7.0

0×

10

78910

4.6

7×

10

74.5

1×

10−

35665.5

1

cork

1.1

5×

10

6160

7.6

7×

10

52.7

5×

10−

16192.7

2

concr

ete

4.6

9×

10

62400

3.1

2×

10

66.7

4×

10−

222799.9

3

bri

ck2.0

0×

10

61900

1.3

3×

10

61.5

8×

10−

142284.6

7

91

Ta

ble

4.8

:T

he

req

uire

da

rea

Aa

nd

ma

ss

mfo

rth

eco

mp

ressive

ele

me

nts

(co

lum

ns)

of

for

aM

ars

ha

bita

tm

issio

n

sce

na

riou

sin

gd

iffere

nt

ma

teria

ls,

so

me

trad

ition

al,

so

me

bio

log

ica

llyd

erive

d.

Va

lue

sfo

rw

oo

da

refo

rd

ryw

oo

da

lon

g

the

gra

in.

Va

lue

sfo

rstru

ctu

red

bio

ma

teria

lsa

ssu

me

no

flaw

so

rd

efe

cts

.

Materia

lstren

gth

σc

(Pa)

den

sityρ

(kg/m

3)stress

σm

ax

(Pa)

area

A(m

2)m

ass

m(k

g)

Notes

nacre

6.0

0×10

82950

4.0

0×10

84.2

8×10−

47.6

9[1

4]

a

titaniu

m(allo

y)

9.0

0×10

84730

6.0

0×10

82.8

5×10−

48.2

2[ 1

6]

alum

inum

6061-T

62.6

0×10

82710

1.7

3×10

89.8

7×10−

416.3

1[1

6]

pin

e(p

ondero

sa)3.6

7×10

7400

2.4

5×10

76.9

9×10−

317.0

5[9

7]

den

tin1.9

0×10

82140

1.2

7×10

81.3

5×10−

317.6

2[3

9,109

]

com

pact

bone

1.6

0×10

81850

1.0

7×10

81.6

0×10−

318.0

9[ 2

8,54

]b

nylo

n6,6

9.5

0×10

71140

6.3

3×10

72.7

0×10−

318.7

8[ 1

6]

steelA

ISI

302

6.5

5×10

87920

4.3

7×10

83.9

2×10−

418.9

2[1

6]

oak

(pin

)4.7

0×10

7630

3.1

3×10

75.4

6×10−

320.9

7[9

7]

poly

carbonate

8.5

0×10

71200

5.6

7×10

73.0

2×10−

322.0

9[ 1

6]

brick

8.0

0×10

71900

5.3

3×10

73.2

1×10−

337.1

6[ 1

04

]b

c

concrete

7.0

0×10

72400

4.6

7×10

73.6

6×10−

353.6

5[1

44

]b

copper

2.2

0×10

88910

1.4

7×10

81.1

7×10−

363.3

7[ 1

6]

enam

el6.2

0×10

72970

4.1

3×10

74.1

4×10−

374.9

5[3

9,109

]

cork

7.0

0×10

5160

4.6

7×10

53.6

6×10−

1357.6

3[1

47

,148

]b

aD

ensity

of

pu

rearag

on

iteu

sedas

con

servativ

eestim

ate.b

Rep

resentativ

evalu

ech

osen

from

repo

rtedran

ge.

cD

ensity

no

tg

iven

inreferen

ce;estim

atedfro

mco

mm

ercialso

urces.

92

Table 4.9: Material compatibility, with biogenic production and in situ resource ex-

traction, for the materials examined in both mission contexts. See Table 4.1 and

Table 4.2.

Material Bioengineering Compatibility In Situ Availability

silk A C

compact bone B C

pine (ponderosa) B C

oak (pin) B C

cork B C

nacre B C

enamel B C

dentin B C

cellulose B C

rubber B C

concrete C B

brick C B

carbon fiber C C

copper D C

silicon carbide D C

polycarbonate D D

titanium (alloy) D D

aluminum 6061-T6 D D

steel AISI 302 D D

nylon 6,6 D D

Nomex D D

Kevlar 49 D D

markedly lower quantities or less energetically available states than on Earth, it nonetheless is theoretically

possible that any biogenic material could be manufactured on-site using in situ extractable materials rather

than transported from Earth.

The reference mission architecture already investigates the costs and benefits for transport of significant

ISRU infrastructure [125] for human life support and other needs. Although the future development paths

and potential of ISRU are uncertain, if the assumption is made that ISRU sufficient for human life support

is possible, it is reasonable to extrapolate from the conditions required for synthetic biomaterial manufacture

and model a “best case” ISRU scenario in which no additional ISRU payload mass would be required per unit

manufactured.

Table 4.2 and Table 4.1 were shown earlier to establish the scales used to rate potential materials for com-

patibility with this approach to synthetic biomaterial manufacture and for potential ISRU harvesting. These

ratings, for the set of materials examined in both mission contexts, are shown in Table 4.9. Only silk receives

93

Table 4.10: The required mass per habitat unit for three habitat material palettes

under zero or theoretical full in situ resource utilization assumptions.

Design no ISRU full ISRU

reference 88.2 88.2

least mass 53.8 23.0

all biomaterial 59.7 0

a grade above “B” for bioengineering compatibility, as some of its constituent proteins are available as genetic

parts (subsection 3.2.1); however, even this falls short of true standardization. As a result, all the remaining

naturally produced biomaterials are rated “B.” Some non-biogenic materials receive a “C” due to one or more

of their ingredients potentially being biogenic, such as binders or filling for concrete or brick.

The grades for ISRU compatibility are lower, reflecting that field’s nascent state. Concrete and brick are rated

“B” due to the availability of regolith for aggregate, though the other ingredients are more challenging. The

remaining materials are rated “C” if they can be produced from the available chemical species with heavy

processing, including biological synthesis; only those materials which cannot be produced short of an Earth-

scale industrial mining or refining venture are rated “D.” The two most promising biomaterials, silk and nacre,

are rated within the acceptable range.

The mass cost per habitat unit under two ISRU scenarios – the “best case” and current or “zero case” – and

for three dome designs – the reference design of Kevlar 49 and aluminum, the lowest mass option of Kevlar

49 and nacre, and the all-biomaterial design of silk and nacre – are shown in Table 4.10. Without ISRU,

the all-biomaterial design represents a mass savings of 32%, and the lowest-mass hybrid design represents a

mass savings of 39%. With ISRU, the all-biomaterial design has a potential mass cost of zero.

If the per unit mass cost is reduced to zero through ISRU, then the payload mass and energy costs of im-

plementing the technology in this mission context are reduced to the fixed cost of the infrastructure. From

section 4.2.3, the upper limit to this in an ISS mission context is estimated to be 380 kg; the larger pieces

here would benefit from a printing unit with a larger workspace, so the infrastructure cost is approximated as

400 kg. With a mass savings of 28.5 kg per habitat unit using a design entirely of synthetic biomaterials, the

infrastructure cost would be offset if the habitat consists of at least 14 domes – fewer if the infrastructure can

be used to make other parts and materials.

With the energy cost of transport already represented by the payload mass savings, the only energy concern

of note is the energy cost of operating the system. Having already assumed a printing system with a larger

94

workspace to account for the larger dimensions of the piece to be manufactured, it follows that it is fair

to assume the energy costs per subsystem do not substantially change from the estimates in section 4.2.3.

The operating time requirements are discussed in section 4.3.3; using the estimates in Table 4.11, the energy

cost per dome unit is estimated to be 109 MJ (less if multiple domes are printed from a single round of

reconstituted stocks and incubated in parallel). This cost should be well within the capacity of habitat-scale

Mars-surface solar arrays.

Labor & Effort Requirements

The labor and effort costs consist of the time required to design, build and transport the necessary materials,

hardware, software and operational protocols; the time required to train astronauts and ground support crew

in the use of the resulting system, including habitat assembly; and the time required to operate the system

and actually construct the habitat on Mars. (The previous line of assumptions regarding ISRU are followed

in assuming that ISRU use for biomaterial manufacture does not require additional labor and cost overhead

beyond what is necessary for other forms of mission support.)

In the first area of concern, transport, both the reference and proposed mission scenarios assume transport of

materials from Earth: the reference mission scenario assumes that all materials needed for habitat construc-

tion are transported as payload cargo, and the proposed mission scenario assumes that an equivalent mass

of synthetic biomaterial manufacture infrastructure is transported. Again as with the ISS scenario, the back-

ground development costs of bringing this technology concept to space mission readiness are not included in

the mission costs, as similar costs would be incurred by any new technology. The labor and effort of proofing

the reference habitat dome design is assumed to be equal to that of the proposed habitat dome design, as they

differ only in material of construction. Therefore, there are no labor and effort advantages or disadvantages

between the reference and proposed scenarios in this area.

In the short-term mission context, the training effort for a self-contained system was estimated to be 10 hours.

In this context, as the system is assumed to be further along in development, it will likely have a higher degree

of automation, but also be substantially more complicated, as it will have to be designed to handle a broader

array of materials over a more adjustable workspace (as well as potentially interface with ISRU systems). As

an upper limit, it should not be more complicated to operate than the manually operated workflow, so the

training effort is estimated to be 50 hours, with a once per month refresher of 2 hours.

95

Table 4.11: Labor estimates for each step of biomaterial manufacture in a Mars

habitat mission context.

Step Time (h)

reconstitution and inoculation 2

system loading 5

print operation 10

array transfer to incubation 5

piece removal and post-processing 10

disposal and reclamation 2

The operational labor costs are assumed to remain roughly the same as those in Table 4.6, given the coun-

terbalancing assumptions about greater system complexity and increased automation. However, the habitat

dome consists of five pieces. As the habitat dome is likely to be one of the largest structural units produced,

it is reasonable to assume that the biomaterial fabrication system can produce its largest piece in a single

production run. Therefore, each habitat dome can be made in five print runs; if the reconstitution, growth,

and incubation steps performed in parallel, then the labor cost should scale to those shown in Table 4.11.

The assembly and construction labor costs are highly dependent on assumptions made about degree of au-

tomation and the provision of heavy equipment from Earth. However, analysis of the effects on the use of

this technology concept can be substantially simplified by assuming that the construction process requires the

same amount of time regardless of the structural material used. This assumption is reasonable given that the

materials will be fulfilling the same structural role in the same general form factor. Therefore, there are no

labor and effort advantages or disadvantages between the reference and proposed scenarios in assembly and

construction labor costs.

Other Considerations

The mass, energy, and labor characterizations here make many simplifying assumptions. Several are worth

further discussion, although quantitative analyses will require further testing and development work.

The mass and energy estimates, as with the ISS mission context analysis (section 4.2), are based on the

values of the prototype and supporting equipment used in the proof of concept. Miniaturization, automation,

and design optimization is expected to substantially reduce these values, in some cases to a few percent of

their current values. For instance, one of the major mass costs is cold-temperature storage. Given the low

surface temperatures on Mars, about −50 ◦C on average [86], it should be possible to substantially reduce the

96

cooling requirements by taking advantage of ambient low temperature, particularly if stabilized a few meters

underground. Similarly, estimates of time and space tradeoffs for dispensing and positioning systems are

similarly based on extrapolation from existing systems, rather than mission specialization. A design study

of a potential mission package that simply adapts existing technology to best meet the space mission criteria

would be of significant value in refining these mass and energy projections.

The compounding effect of in situ resource utilization technology on the space mission potential of this

technology concept is substantial. One of the primary advantages of structural biomaterials is that life can

make them using base ingredients which, on Earth, are plentiful and energetically available. Whether this

advantage extends to off-Earth environments depends on the efficiency at which ISRU techniques can make

those same base ingredients available. While ISRU techniques for extracting water in a Martian environment

have been well-studied, published concepts for other compounds in bioavailable forms are scarcer. A more

in-depth review of relevant ISRU capabilities would be necessary to properly evaluate the potential costs and

benefits.

This analysis relies strongly on material substitution as a means of calculating mass and energy savings. How-

ever, the substitutions carried out here are based largely on published values for existing biomaterials. One of

the most appealing aspects of this technology concept is its potential to allow fine microstructure control of

structural biomaterials; such an ability would allow construction of materials with properties optimized to a

given loading condition whose performance ought to far exceed those of naturally grown materials repurposed

to human ends. The property tables used here approximates this by using figures for natural biomaterials in

optimal loading conditions – along the grain for wood, single fibers for fibrous materials, an absence of de-

fects or flaws for regularly structured materials such as nacre, and so forth – but the final material properties

achievable with true structural optimization cannot be calculated without improvements in existing material

modelling tools. This will also allow the incorporation of factors such as rigidity and bending, which were

not considered here, into a design analysis.

There is a second limitation of reliance on material substitution. Any reference design will have been created

with the properties and limitations of existing materials in mind. If it is possible to optimize material structure

for loading, then it also becomes possible to optimize structural design to take advantage of this capability.

Existing structural design and modelling tools do not adequately capture this dimension of working with

synthetic biomaterials. While a true analysis will have to wait for improved design and modelling tools, to

give a preliminary idea of what the effect might be, a third material substitution calculation was run using

a proposed space mission tensegrity structure, a planetary lander and exploration robot [3]. Human-scale

97

tensegrity structures share many features with smaller-scale biomaterial internal structures, such as inherent

tensile-compressive load balancing and whole-structure passive energy dissipation [33, 83], so they can be

considered a biomaterial-inspired design.

This design consists of compressive members (struts) connected by tensile members (cables). In addition

to the tensile and compressive yield strengths of the materials used, this analysis also takes into account the

potential for the compressive members to buckle. Euler’s equation for critical buckling load in a column

under uniaxial compression is shown in Equation 4.3 [16], where Fcrit is the applied load, E is the material’s

elastic modulus, I is the compressive member’s cross-sectional area moment of inertia, L is the length of the

column, and K is a factor used to account for whether the ends of the compressive member are fixed, pinned,

or free.

Fcrit =π2EI

(KL)2(4.3)

The loads that the compressive and tensile elements were designed for are given in [23, 150] as 800 N.

The compressive elements are modelled as hollow tubes 1 m in length with circular cross-sections and inner

diameters of 3 mm; K is set to 2. The required areas and masses for both the compressive and buckling limits,

using a safety factor of 1.5, for the same list of materials used previously as potential compressive substitutes

are shown in Table 4.12 and Table 4.13.

For the compressive elements, buckling rather than compressive yield proved to be the limiting factor. Within

this limitation, both types of wood analyzed were by far the most mass-efficient for the required performance,

with nacre second but still lighter than aluminum, the highest-ranking non-biogenic material.

The same reference design was used to run the equivalent analysis on the tensile elements, the cables. The

results are shown in Table 4.14. (A few materials, such as polycarbonate, which were not sensical in this

context are not included.) As the rover design uses cables whose load-bearing lengths are adjustable, the

required masses are given as kilograms per unit length.

For the tensile elements, silk is the most mass-efficient of the biomaterial options, although it remains slightly

heaver than Kevlar 49 for the same performance. However, spider silk is notable for its unusual stress-

strain responses, including substantial elongation before failure and hysteresis [64]. The importance of these

properties, and how they can be used to best advantage, is another unknown that must be addressed in future

work through the use of improved design optimization and soft-body modelling tools (section 3.3).

98

Ta

ble

4.1

2:

Th

ere

qu

ire

dm

ass

ma

nd

are

aA

for

the

co

mp

ressiv

ee

lem

en

tso

fa

ten

se

gri

ty-b

ase

dro

ver,

ba

se

do

nco

m-

pre

ssiv

estr

en

gth

,u

sin

gd

iffe

ren

tm

ate

ria

ls,

so

me

tra

ditio

na

l,so

me

bio

log

ica

llyd

eri

ved

.V

alu

es

for

fib

rou

sb

iom

ate

ria

ls

are

for

ind

ivid

ua

lfib

ers

;o

the

rfib

rou

sm

ate

ria

lsa

relis

ted

as

co

mp

osite

s.

Va

lue

sfo

rw

oo

da

refo

rd

ryw

oo

da

lon

gth

e

gra

in.

Va

lue

sfo

rstr

uctu

red

bio

ma

teri

als

assu

me

no

flaw

so

rd

efe

cts

.

Mate

rial

stre

ngth

σc

(Pa)

den

sity

ρ(k

g/m

3)

stre

ssσ

ma

x(P

a)

are

aA

(m2)

mass

m(k

g)

nac

re6.0

0×

10

82950

4.0

0×

10

82.0

0×

10−

60.0

059

tita

niu

m(a

lloy)

9.0

0×

10

84730

6.0

0×

10

81.3

3×

10−

60.0

063

alum

inum

6061-T

62.6

0×

10

82710

1.7

3×

10

84.6

2×

10−

60.0

125

pin

e(p

onder

osa

)3.6

7×

10

7400

2.4

5×

10

73.2

7×

10−

50.0

131

den

tin

1.9

0×

10

82140

1.2

7×

10

86.3

2×

10−

60.0

135

com

pac

tbone

1.6

0×

10

81850

1.0

7×

10

87.5

0×

10−

60.0

139

nylo

n6,6

9.5

0×

10

71140

6.3

3×

10

71.2

6×

10−

50.0

144

stee

lA

ISI

302

6.5

5×

10

87920

4.3

7×

10

81.8

3×

10−

60.0

145

oak

(pin

)4.7

0×

10

7630

3.1

3×

10

72.5

5×

10−

50.0

161

poly

carb

onat

e8.5

0×

10

71200

5.6

7×

10

71.4

1×

10−

50.0

169

bri

ck8.0

0×

10

71900

5.3

3×

10

71.5

0×

10−

50.0

285

concr

ete

7.0

0×

10

72400

4.6

7×

10

71.7

1×

10−

50.0

411

copper

2.2

0×

10

88910

1.4

7×

10

85.4

5×

10−

60.0

486

enam

el6.2

0×

10

72970

4.1

3×

10

71.9

4×

10−

50.0

575

cork

7.0

0×

10

5160

4.6

7×

10

51.7

1×

10−

30.2

743

aR

epre

sen

tati

ve

val

ue

cho

sen

fro

mre

po

rted

ran

ge.

99

Ta

ble

4.1

3:

Th

ere

qu

ired

ma

ss

ma

nd

are

aA

for

the

co

mp

ressive

ele

me

nts

of

ate

nse

grity

-ba

se

dro

ver,

ba

se

do

n

bu

cklin

glim

it,u

sin

gd

iffere

nt

ma

teria

ls,

so

me

trad

ition

al,

so

me

bio

log

ica

llyd

erive

d.

Va

lue

sfo

rfib

rou

sb

iom

ate

rials

are

for

ind

ivid

ua

lfib

ers

;o

the

rfib

rou

sm

ate

rials

are

liste

da

sco

mp

osite

s.

Va

lue

sfo

rw

oo

da

refo

rd

ryw

oo

da

lon

gth

egra

in.

Va

lue

sfo

rstru

ctu

red

bio

ma

teria

lsa

ssu

me

no

flaw

so

rd

efe

cts

.

Materia

lela

sticm

od

ulu

sE

(Pa)

den

sityρ

(kg/m

3)seco

nd

area

mom

ent

I(m

4)area

A(m

2)m

ass

m(k

g)

Notes

pin

e(p

ondero

sa)8.9

0×10

9400

5.4

6×10−

80.0

008

0.3

2[9

7]

oak

(pin

)1.1

9×10

10

630

4.0

9×10−

80.0

007

0.4

3[9

7]

nacre

9.0

0×10

10

2950

5.4

0×10−

90.0

002

0.6

9[1

3]

alum

inum

6061-T

67.0

0×10

10

2710

6.9

5×10−

90.0

003

0.7

3[1

6]

titaniu

m(allo

y)

1.1

5×10

11

4730

4.2

3×10−

90.0

002

0.9

6[1

6]

com

pact

bone

2.0

0×10

10

1850

2.4

3×10−

80.0

005

0.9

7[4

6]

a

concrete

3.0

0×10

10

2400

1.6

2×10−

80.0

004

1.0

2[1

6]

steelA

ISI

302

1.9

0×10

11

7920

2.5

6×10−

90.0

002

1.2

1[1

6]

brick

1.0

0×10

10

1900

4.8

6×10−

80.0

008

1.4

3[ 7

2]

a

nylo

n6,6

2.8

0×10

91140

1.7

4×10−

70.0

014

1.6

5[ 1

6]

copper

1.2

0×10

11

8910

4.0

5×10−

90.0

002

1.7

7[1

6]

poly

carbonate

2.4

0×10

91200

2.0

3×10−

70.0

016

1.8

8[ 1

6]

cork

3.8

0×10

7160

1.2

8×10−

50.0

127

2.0

2[ 1

48

]

den

tin1.6

5×10

92140

2.9

4×10−

70.0

019

4.0

5[3

9]

enam

el1.3

4×10

92970

3.6

3×10−

70.0

021

6.2

6[ 3

9]

aR

epresen

tative

valu

ech

osen

from

repo

rtedran

ge.

100

Ta

ble

4.1

4:

Th

ere

qu

ire

dm

ass

ma

nd

are

aA

for

the

ten

sile

ele

me

nts

of

ate

nse

gri

ty-b

ase

dro

ver,

ba

se

do

nte

nsile

str

en

gth

,u

sin

gd

iffe

ren

tm

ate

ria

ls,

so

me

tra

ditio

na

l,so

me

bio

log

ica

llyd

eri

ved

.V

alu

es

for

fib

rou

sb

iom

ate

ria

lsa

refo

r

ind

ivid

ua

lfib

ers

;o

the

rfib

rou

sm

ate

ria

lsa

relis

ted

as

co

mp

osite

s.

Va

lue

sfo

rw

oo

da

refo

rd

ryw

oo

da

lon

gth

egra

in.

Va

lue

sfo

rstr

uctu

red

bio

ma

teri

als

assu

me

no

flaw

so

rd

efe

cts

.

Mate

rial

stre

ngth

σt

den

sity

ρ(k

g/m

3)

stre

ssσ

ma

x(P

a)

are

aA

(m2)

mass

m(k

g/m

)

Kev

lar

49

2.8

0×

10

91450

1.8

7×

10

95.3

6×

10−

87.7

7×

10−

5

silk

2.0

0×

10

91300

1.3

3×

10

97.5

0×

10−

89.7

5×

10−

5

carb

on

fiber

2.2

0×

10

92000

1.4

7×

10

96.8

2×

10−

81.3

6×

10−

4

sili

con

carb

ide

2.6

0×

10

92560

1.7

3×

10

95.7

7×

10−

81.4

8×

10−

4

cell

ulo

se1.0

0×

10

91500

6.6

7×

10

81.5

0×

10−

72.2

5×

10−

4

Nom

ex6.5

0×

10

81380

4.3

3×

10

82.3

1×

10−

73.1

8×

10−

4

oak

(pin

)1.1

2×

10

8630

7.4

9×

10

71.3

3×

10−

68.4

1×

10−

4

tita

niu

m(a

lloy)

8.3

0×

10

84730

5.5

3×

10

81.8

1×

10−

78.5

5×

10−

4

pin

e(p

onder

osa

)5.7

9×

10

7400

3.8

6×

10

72.5

9×

10−

61.0

4×

10−

3

alum

inum

6061-T

62.4

0×

10

82710

1.6

0×

10

86.2

5×

10−

71.6

9×

10−

3

nac

re1.3

5×

10

82950

9.0

0×

10

71.1

1×

10−

63.2

8×

10−

3

nylo

n6,6

4.5

0×

10

71140

3.0

0×

10

73.3

3×

10−

63.8

0×

10−

3

stee

lA

ISI

302

2.6

0×

10

87920

1.7

3×

10

85.7

7×

10−

74.5

7×

10−

3

rubber

1.4

0×

10

71000

9.3

3×

10

61.0

7×

10−

51.0

7×

10−

2

copper

7.0

0×

10

78910

4.6

7×

10

72.1

4×

10−

61.9

1×

10−

2

cork

1.1

5×

10

6160

7.6

7×

10

51.3

0×

10−

42.0

9×

10−

2

101

One final question is the dependence of the properties of synthetic biomaterials on temperature, humidity,

and other environmental factors. Mitigation for some factors is easy and inexpensive – Kevlar, for instance,

is highly susceptible to UV radiation, and is typically used encapsulated in another material for that reason

even on Earth – but while the habitat design here could be altered to keep the biomaterial components within

Earth-normal ranges if necessary, the technology concept is substantially more useful in a space mission

context if its products can be used on the Martian surface. Most biomaterials have not been tested at such

extremes of condition; more information is necessary to understand the risks and required responses.

4.3.4 Summary

Synthetic biomaterials have the potential to reduce the mass of the reference habitat columns by nearly 53%

over the reference material of aluminum and by approximately 6% over the lightest non-biomaterial option,

titanium alloy. The reference habitat skin remains slightly more effective than the next best option, silk,

which is 25% heavier, in a simple material replacement scenario. Thus, the lightest-mass option is nacre

columns combined with Kevlar 49 skin, for 53.8 kg per unit or a mass savings of 39%. In the presence of

ISRU, synthetic biomaterials become increasingly attractive; in a best case ISRU scenario, a habitat design

consisting of the lightest biomaterials (nacre and silk) would have a per-unit cost of zero mass and a fixed

payload cost of no more than 400 kg. In such a scenario, the technology would pay for itself if the habitat

consisted of at least 14 domes, or fewer if synthetic biomaterials were also used in other habitat components.

The importance of implementing multiple material types and developing complementary ISRU technology

in expanding this technology concept’s potential in a space mission environment is clear.

The labor costs of biomaterial manufacture in a Mars environment are highly dependent on how many pieces

can be batched and run in parallel, as the technology itself relatively easy to learn from an operational stand-

point. Although many unknowns remain regarding the degree of automation that is feasible on a multi-year

timeline, the labor costs are not expected to be substantially greater than those of any other materials manu-

facture or assembly technology.

A few short-term follow-on studies are recommended to answer some of the most pressing remaining ques-

tions about the technology concept’s potential. Firstly, a design study of a complete end-to-end system

consisting of existing commercial technology modified for a particular use would help the mass and energy

estimations. Secondly, a critical review of existing and planned ISRU approaches in light of the resources

scarcest on Mars and most important to the manufacture of the top few structural biomaterials would provide

102

significant guidance and clarity to the ‘all or nothing’ estimations used here. Thirdly, an investigation of the

properties of the most promising structural biomaterial candidates at low temperatures, under UV radiation,

in extremes of desiccation, and under other typical off-Earth environments would determine what mitigation

strategies, and at what cost, are necessary for implementation in different space mission contexts. Lastly,

a review of existing materials analysis, structural design, and soft-body modelling tools would help further

illuminate the most-needed directions of future development.

4.4 Conclusions

Synthetic biomaterials offer the potential of a substantial mass reduction per part fabricated in both the ISS

mission context and the Mars habitat mission context, as well as improved mission robustness due to the

possibility of manufacturing multiple parts from the same infrastructure in both cases.

In the ISS mission context, for small or infrequently replaced parts, the technology concept offers the best

mass tradeoff if the biomaterial parts are manufactured on Earth and then transported as ordinary payload

resupply. However, on-board manufacture becomes more promising as the mass of parts to be replaced

increases. Additional benefit in the on-board manufacturing case accrues from the ability to print replacement

or new parts more rapidly than a resupply mission can be organized, as long as the cell strains necessary for

the particular material are already on board. Analysis of existing part failure rates, and the projected part

failure rates of equivalent biomaterial-based designs, will require future work.

In the Mars mission context, material substitution using synthetic biomaterials offers a potential mass savings

of over one third without on-site manufacture. However, on-site manufacture becomes rapidly more attractive

with the possibility of using ISRU-harvested resources, and with the added robustness that comes from being

able to manufacture replacement parts in an environment where resupply windows are periodic and transport

costs prohibitively expensive. The ultimate potential mass cost may be no more than the initial infrastructure

investment, with an energy cost well within the range of current solar power technology.

In addition to these quantitative results, performing this space mission context analysis also produced some

conceptual insights. Four key challenges, in particular, have been brought to light. The first is the unknown

performance of potential synthetic biomaterials in a space environment. The second is the tremendous un-

certainty surrounding the potential of ISRU technologies. The third is the lack of good compliant materials

models. The fourth is the lack of suitable materials/structure design optimization tools.

103

Although the mechanical behavior of natural structural biomaterials at non-Earth-standard temperatures and

other conditions is largely unexplored (silk being a notable exception [141]), it has been hypothesized that it

will be significantly different (e.g., [64]), particularly since many biomaterials’ properties are highly sensitive

to their water content [118]. This is an unknown that can be addressed directly with environmental tests on

natural samples.

In situ resource utilization for space mission support has been proposed for many years [55]. It is most often

associated with human life support, where it is one of NASA’s major identified technological needs [88, 125],

but has also been suggested for support of non-crewed missions where it could provide fuel, propellant, and

so forth [151]. Some proposed mission concepts include the use of microbes to assist with in situ resource

extraction [134]. Determining which approaches are most complementary to this technology concept, both

in terms of resources provided and in terms of shared infrastructure requirements, will require an in-depth

investigation of current literature and on-going projects.

Many biomaterials are more fracture-resistant than traditional engineering materials because they are able to

absorb energy through non-destructive bending, deformation, or other types of physical compliance [7, 13,

34, 112, 142, 148]. This behavior is poorly represented in traditional materials modelling and analysis and

consequently is often ignored in standard design tools and approaches. Addressing this unknown will require

modification of existing models or development of entirely new software tools.

Lastly, the most powerful potential advantage our technology concept offers is the ability to do micron-

scale material property customization. Traditional materials design analysis tools tend to assume fixed, and

spatially constant, material properties, and solve for maximum loading conditions. This technology concept,

one day, may be able to turn this philosophy on its head by providing the necessary material properties

at each point in a desired structure to support a given range of loading conditions with minimized mass.

Needless to say, the design tools to do this ‘reversed’ calculation have not yet been created, and so it is

not yet possible to say what the maximum possible benefit of our technology concept might be under ideal

conditions. Addressing this unknown will require modification of existing specialized software tools or re-

implementation of traditional solving methods (e.g., finite element analysis) using existing general systems.

104

5 Summary & Conclusions

A two-material array of non-structural proteins provides proof of concept of biomaterials manufacturing using

3-D printed arrays of bioengineered cells (chapter 2). Each step in the end-to-end demonstration (Figure 1.2)

has been proven to reach the minimum level of critical functionality (Table 2.5). The current host cells are S.

cerevisiae, a yeast, genetically engineered to secrete the target materials, fluorescent-tagged proteins, when

exposed to galactose. The current print medium and substrate are a dextrose-containing alginate medium

deposited onto a calcium- and galactose-containing agar surface. The calcium gels the alginate, and the intro-

duction of galactose when the cells contact the substrate triggers the material secretion. This design unifies

the act of printing with the act of creating a physical support for the cells and providing the material pro-

duction stimulus, greatly simplifying the end-to-end-process. The biological chassis and printing hardware

created as part of this work can be re-used for future work by inserting a material coding region upstream

of the fluorescent tag. Overall, this work shows that the technology concept is sound at the proof of concept

level.

The survey of future development pathways proved extremely informative in light of the lessons learned

from the proof of concept work (chapter 3). Distinguishing between the levels of functionality provided

by production of structural proteins, other polymers such as polysaccharides, and true organic-inorganic

composites such as bone and mineralized shell was a necessary step in understanding the benefits potentially

available from future development investment. Surveying the state of knowledge of the precise mechanisms

involved in the formation of both non-protein-based structural materials, such as chitin and cellulose, and the

inorganic phase of biominerals, allowed quantified estimations of the previously qualitative descriptions of

this technology concept’s reliance on advances in other fields. These are significant advances in formulating

specific applications for the technology concept.

Two space mission cost and benefit analyses, as well as a complementary tensegrity materials analysis, are

also complete (chapter 4). As was appropriate given the information available prior to the completion of the

105

Table 5.1: NASA’s Technology Readiness Level definitions and their relationship to

the work presented .

Technology Component Prior Status Current Status

material substition effects 2 3

on site manufacturing effects 2 3

novel material design effects 1 1

material optimization effects 1 2

host cell engineering 2 3

plasmid backbone engineering 2 3

biomaterial production 2 3

structural material production 1 2

material production control 2 3

organic material delivery control 2 3

inorganic material delivery control 1 2

material binding control 2 2

pint medium 2 3

3D printing system 2 3

print substrate 2 2

2D material verification 2 3

3D material verification 2 2

proof of concept, these focus on the potential benefits from material substitution and payload mass reduction.

These simplified calculations show that the technology concept is best used in situations where a significant

amount of structural mass of different types and form factors needs to be produced or printed, or where short-

term resupply of structural parts is not feasible. The use of structural biomaterials has the potential to reduce

payload mass by at least a third in both cases. If on-site manufacturing and in situ resource utilization (in

a planetary environment) are included, then the transport mass can be reduced still further, although future

studies on ISRU capacity and complementarity will be necessary to quantify this potential.

In many technology development projects, the implementations of the proof of concept which demonstrate

critical functionality also provide pathways for future development. The highly multidisciplinary nature of

this project, particularly the biological aspect of it, presents a challenge to this straightforward picture.

The most clear example of this in the work presented here is the fact that the polyhistidine tag material binding

method worked sufficiently well for a proof of concept demonstration, but is too cytotoxic for use in future

work where work with structural materials will require a much greater amount of the substrate to bind the

target material. Another, less obvious, example, is the case of genetic parts for material production. One of

the clearest lessons of this work is that despite the large number of known genetic parts that correspond to

non-structural materials, such as the proteins used in this work, sequences for structural organic proteins, let

106

alone biomineralization pathways, are still far from standardization (Table 4.9). These realizations require

further subdivision of the overall concept into more detailed development areas.

Similarly, although the feasibility/benefit analysis provided here describes the benefits from a material substi-

tution standpoint, not all potential benefits to a technology concept as broad in scope as this one are apparent

this early in development. Both the future pathways survey and the proof of concept work highlighted that

the true potential of this technology concept cannot be quantified without modification of existing materials

modelling tools to take into account the possibility of positional materials properties customization.

Returning to the concept of Technology Readiness Levels (section 1.5), the work presented here has simul-

taneously advanced one potential set of applications of the technology concept from TRL 2 to TRL 3 and

also identified a previously unknown set of applications and advanced it from TRL 1 to TRL 2. The overall

TRL advancement breakdown in shown in Table 5.1. The work has also identified the key areas necessary

for both short-term and long-term advancement, and made recommendations for specific future work in the

most promising directions. Overall, the results support the promise and potential of synthetic biomaterial

manufacture.

107

108

Appendices

109

A Plasmid Maps & Sequences

Figure A.1: The plasmid backbone used for the yeGFP-producing yeast cells.

Figure A.2: The plasmid backbone used for the yeRFP-producing yeast cells.

111

Final inserted sequence for yeGFP, including secretion tag, His-tag, and V5 epitope (930 bp):

ATGAAAGTTTTGATTGTTTTGTTGGCTATTTTCGCTGCTTTGCCATTGGCTTTGGCTCAACCAGTTATTTCTACTA

CTGTTGGTTCTGCTGCTGAAGGTTCTTTGGATAAAAGAATGTCTAAGGGTGAAGAATTGTTCACTGGTGTTGTCCC

AATTTTGGTTGAATTAGATGGTGATGTTAATGGTCACAAGTTCTCTGTCTCCGGTGAAGGTGAAGGTGATGCTACT

TACGGTAAGTTGACCTTAAAGTTCATTTGTACTACTGGTAAGTTGCCAGTTCCATGGCCAACCTTAGTCACTACTT

TCGGTTACGGTGTTCAATGTTTCGCTAGATACCCAGATCATATGAAGCAACATGACTTCTTCAAGTCTGCCATGCC

AGAAGGTTACGTTCAAGAAAGAACTATTTTCTTCAAGGATGACGGTAACTACAAGACCAGAGCTGAAGTCAAGTTC

GAAGGTGATACCTTGGTTAATAGAATCGAATTGAAGGGTATTGATTTCAAGGAAGATGGTAACATTTTGGGTCACA

AGTTGGAATACAACTACAACTCTCACAATGTTTACATCATGGCTGACAAGCAAAAGAATGGTATCAAGGTTAACTT

CAAGATTAGACACAACATTGAAGATGGTTCTGTTCAATTGGCTGACCATTACCAACAAAATACTCCAATTGGTGAT

GGTCCAGTCTTGTTGCCAGACAACCATTACTTGTCCACTCAATCTGCCTTATCCAAAGATCCAAACGAAAAGAGGG

ACCACATGGTCTTGTTGGAATTCGTTACTGCTGCTGGTATTACCCATGGTATGGATGAATTGTACAAGAAGGGCGA

GCTTCGAGGTCACCCATTCGAAGGTAAGCCTATCCCTAACCCTCTCCTCGGTCTCGATTCTACGCGTACCGGTCAT

CATCACCATCACCATTGA

Final inserted sequence for yeRFP, including secretion tag, His-tag, and V5 epitope (924 bp):

ATGAAAGTTTTGATTGTTTTGTTGGCTATTTTCGCTGCTTTGCCATTGGCTTTGGCTCAACCAGTTATTTCTACTA

CTGTTGGTTCTGCTGCTGAAGGTTCTTTGGATAAAAGAATGGTTTCGAAGGGTGAGGAGGATAACATGGCTATCAT

CAAGGAGTTTATGAGATTTAAGGTTCATATGGAGGGTTCGGTTAACGGTCATGAGTTTGAGATCGAGGGTGAGGGT

GAGGGTAGACCATATGAGGGTACTCAAACTGCTAAGTTGAAGGTTACTAAGGGTGGTCCATTACCATTTGCTTGGG

ATATCTTGTCGCCACAATTTATGTATGGTTCGAAGGCTTATGTTAAGCATCCAGCTGATATCCCAGATTATTTAAA

GTTGTCGTTTCCAGAGGGTTTTAAGTGGGAGAGAGTTATGAACTTTGAGGATGGTGGTGTTGTTACTGTTACTCAA

GATTCGTCGTTACAAGATGGTGAGTTTATCTATAAGGTTAAGTTGAGAGGTACTAACTTTCCATCGGATGGTCCAG

TTATGCAAAAGAAGACTATGGGTTGGGAGGCTTCGTCGGAGAGAATGTATCCAGAGGATGGTGCTTTAAAGGGTGA

GATCAAGCAAAGATTGAAGTTAAAGGATGGTGGTCATTATGATGCTGAGGTTAAGACTACTTATAAGGCTAAGAAG

CCAGTTCAATTACCAGGTGCTTATAACGTTAACATCAAGTTGGATATCACTTCGCATAACGAGGATTATACTATCG

TTGAGCAATATGAGAGAGCTGAGGGTAGACATTCGACTGGTGGTATGGATGAGTTATATAAGAAGGGCGAGCTTCG

AGGTCACCCATTCGAAGGTAAGCCTATCCCTAACCCTCTCCTCGGTCTCGATTCTACGCGTACCGGTCATCATCAC

CATCACCATTGA

112

References

[1] International Space Station: Solar arrays. http://www.nasa.gov/mission_pages/station/

structure/elements/solar_arrays.html, 2013. Accessed: Nov 29, 2014.

[2] Bioculture system validation. http://www.nasa.gov/mission_pages/station/research/

experiments/1180.html, 2014. Accessed: Nov 29, 2014.

[3] Agogino, A., SunSpiral, V., and Atkinson, D. Super Ball Bot - Structures for Planetary Landing and

Exploration. National Aeronautics and Space Administration, 2013. NIAC Phase I Final Report.

[4] Agresti, J., Antipov, E., Abate, A., Ahn, K., Rowat, A., Baret, J., Marquez, M., Klibanov, A., Griffiths,

A., and Weitz, D. Ultrahigh-throughput screening in drop-based microfluidics for directed evolution.

Proceedings of the National Academy of Sciences of the United States of America, 107(9):4004–4009,

2010. doi:10.1073/pnas.0910781107.

[5] Albrecht, D., Underhill, G., Wassermann, T., Sah, R., and Bhatia, S. Probing the role of multicellular

organization in three-dimensional microenvironments. Nature Methods, 3(5):369–375, 2006.

[6] Allen, C., Morris, R., Jager, K., Golden, D., Lindstrom, D., Lindstrom, M., and Lockwood, J. Martian

regolith simulant JSC Mars-1. In Lunar and Planetary Science XXIX, p. 1690. Lunar and Planetary

Institute, 1998.

[7] Allison, P., Chandler, M., Rodriguez, R., Williams, B., Moser, R., Jr, C., Poda, A., Lafferty, B.,

Kennedy, A., Seiter, J., Hodo, W., and Cook, R. Mechanical properties and structure of the biological

multilayered material system, Atractosteus spatula scales. Acta Biomaterialia, 9(2):5289–5296, 2013.

doi:10.1016/j.actbio.2012.11.005.

[8] Ashby, M., Gibson, L., Wegst, U., and Olive, R. The mechanical properties of natural materials. I.

Material property charts. Proceedings: Mathematical and Physical Sciences, 450(1938):123–140,

1995.

113

[9] Atala, A., Kasper, F., and Mikos, A. Engineering complex tissues. Science Translational Medicine,

4(160):160rv12, 2012.

[10] Bach, I. The LIM domain: regulation by association. Mechanisms of Development, 91(1):5–17, 2000.

[11] Bae, W., Chen, W., Mulchandani, A., and Mehra, R. Enhanced bioaccumulation of heavy metals by

bacterial cells displaying synthetic phytochelatins. Biotechnology and Bioengineering, 70(5):518–524,

2000.

[12] Barron, J., Wu, P., Ladouceur, H., and Ringeisen, B. Biological laser printing: a novel technique for

creating heterogeneous 3-dimensional cell patterns. Biomedical Microdevices, 6(2):139–147, 2004.

[13] Barthelat, F. and Espinosa, H. An experimental investigation of deformation and fracture of nacre –

mother of pearl. Experimental Mechanics, 47(3):311–324, 2007.

[14] Barthelat, F., Li, C., Comi, C., and Espinosa, H. Mechanical properties of nacre constituents and their

impact on mechanical performance. Journal of Materials Research, 21(08):1977–1986, 2006.

[15] Bazylinski, D., Frankel, R., Heywood, B., Mann, S., King, J., Donaghay, P., and Hanson, A. Controlled

biomineralization of magnetite (Fe 3O4) and greigite (Fe 3S 4) in a magnetotactic bacterium. Applied

and Environmental Microbiology, 61(9):3232–3239, 1995.

[16] Beer, F., Johnston Jr., E., and DeWolf, J. Mechanics of Materials. McGraw-Hill, 2006.

[17] Berg, J. Zinc fingers and other metal-binding domains. Elements for interactions between macro-

molecules. Journal of Biological Chemistry, 265(12):6513–6516, 1990.

[18] Bhatia, S., Balis, U., Yarmush, M., and Toner, M. Microfabrication of hepatocyte/fibroblast co-

cultures: role of homotypic cell interactions. Biotechnology Progress, 14(3):378–387, 1998.

[19] Bodde, S., Meyers, M., and Mckittrick, J. Correlation of the mechanical and structural properties of

cortical rachis keratin of rectrices of the Toco toucan (Ramphastos toco). Journal of the Mechanical

Behavior of Biomedical Materials, 4(5):723–732, 2011. doi:10.1016/j.jmbbm.2011.01.010.

[20] Bond, G., Richman, R., and McNaughton, W. Mimicry of natural material designs and processes.

Journal of Materials Engineering and Performance, 4:334–345, 1995. doi:10.1007/BF02649071.

[21] Boskey, A. Biomineralization: conflicts, challenges, and opportunities. Journal of Cellular Biochem-

istry Supplements, 30-31:83–91, 1998.

[22] Brodland, G. Finite element methods for developmental biology. International Review of Cytology,

150:95–118, 1994.

[23] Bruce, J., Sabelhaus, A., Chen, Y., Lu, D., Morse, K., Milam, S., Caluwaerts, K., Agogino, A., and

SunSpiral, V. SUPERball: Exploring tensegrities for planetary probes. In 12th International Sym-

posium on Artificial Intelligence, Robotics and Automation in Space. Canadian Space Agency et al.,

114

2014.

[24] Butt, T. and Ecker, D. Yeast metallothionein and applications in biotechnology. Microbiological

Reviews, 51(3):351–364, 1987.

[25] Cadogan, D., Stein, J., and Grahne, M. Inflatable composite habitat structures for lunar and Mars

exploration. Acta Astronautica, 44:399–406, 1999.

[26] Campana, S. and Melendez, D. Cold stowage flight systems. In 41st International Conference on

Environmental Systems, pp. 2011–5261. American Institute of Aeronautics and Astronautics, 2011.

doi:10.2514/6.2011-5261.

[27] Campbell, R., Tour, O., Palmer, A., Steinbach, P., Baird, G., Zacharias, D., and Tsien, R. A monomeric

red fluorescent protein. Proceedings of the National Academy of Sciences of the United States of

America, 99(12):7877–7882, 2002. doi:10.1073/pnas.082243699.

[28] Carter, D. and Hayes, W. Bone compressive strength: the influence of density and strain rate. Science,

194(4270):1174–1176, 1976.

[29] Cartwright, C., Li, Y., Zhu, Y., Kang, Y., and Tipper, D. Use of beta-lactamase as a secreted reporter

of promoter function in yeast. Yeast, 10(4):497–508, 1994.

[30] Chang, C., Matteucci, M., Perry, L., Wulf, J., Chen, C., and Hitzeman, R. Saccharomyces cerevisiae

secretes and correctly processes human interferon hybrid proteins containing yeast invertase signal

peptides. Molecular and Cellular Biology, 6(5):1812–1819, 1986.

[31] Chen, B., Peng, X., Cai, C., Niu, H., and Wu, X. Helicoidal microstructure of Scarabaei cuticle and

biomimetic research. Materials Science and Engineering: A, 423(1):237–242, 2006.

[32] Chen, B., Peng, X., Wang, W., Zhang, J., and Zhang, R. Research on the microstructure of insect

cuticle and the strength of a biomimetic preformed hole composite. Micron, 33(6):571–574, 2002.

[33] Chen, C. and Ingber, D. Tensegrity and mechanoregulation: from skeleton to cytoskeleton. Os-

teoarthritis and Cartilage, 7(1):81–94, 1999. doi:10.1053/joca.1998.0164.

[34] Chen, P., McKittrick, J., and Meyers, M. Biological materials: functional adaptations and bioinspired

designs. Progress in Materials Science, 57(8):1492–1704, 2012.

[35] Cheung, K. and Gershenfeld, N. Reversibly assembled cellular composite materials. Science,

341(6151):1219–1221, 2013. doi:10.1126/science.1240889.

[36] Cho, H., H., J., Campbell, K., Melke, P., Williams, J., Jedynak, B., Stevens, A., Groisman, A., and

Levchenko, A. Self-organization in high-density bacterial colonies: efficient crowd control. PLoS

Biology, 5(11):e302, 2007.

[37] Choi, S., Paik, H., Lee, S., Nihira, T., and Hwang, Y. Enhanced productivity of human lysozyme by

115

pH-controlled batch fermentation of recombinant Saccharomyces cerevisiae. Journal of Bioscience

and Bioengeering, 98(2):132–135, 2004.

[38] Choi, Y. and Lee, S. Microbial production of short-chain alkanes. Nature, 502(7472):571–574, 2013.

[39] Chun, K., Choi, H., and Lee, J. Comparison of mechanical property and role between enamel and

dentin in the human teeth. Journal of Dental Biomechanics, 5:1758736014520809, 2014. doi:10.

1177/1758736014520809.

[40] Clements, J., Catlin, G., Price, M., and Edwards, R. Secretion of human epidermal growth factor from

Saccharomyces cerevisiae using synthetic leader sequences. Gene, 106(2):267–271, 1991.

[41] Connell, J., Ritschdorff, E., Whiteley, M., and Shear, J. 3D printing of microscopic bacterial commu-

nities. Proceedings of the National Academy of Sciences of the United States of America, 2013.

[42] Coradin, T. and Lopez, P. Biogenic silica patterning: simple chemistry or subtle biology? Chem-

biochem, 4(4):251–259, 2003. doi:10.1002/cbic.200390044.

[43] Cormack, B., Bertram, G., Egerton, M., Gow, N., Falkow, S., and Brown, A. Yeast-enhanced green flu-

orescent protein (yEGFP): a reporter of gene expression in Candida albicans. Microbiology, 143:303–

311, 1997.

[44] Cormack, B., Valdivia, R., and Falkow, S. FACS-optimized mutants of the green fluorescent protein

(GFP). Gene, 173:33–38, 1996.

[45] Cui, X., Breitenkamp, K., Finn, M., Lotz, M., and D’Lima, D. Direct human cartilage repair using

three-dimensional bioprinting technology. Tissue Engineering: Part A, 18(11-12):1304–1312, 2012.

[46] Currey, J. Physical characteristics affecting the tensile failure properties of compact bone. Journal of

Biomechanics, 23(8):837–844, 1990.

[47] Demirci, U. and Montesano, G. Single cell epitaxy by acoustic picolitre droplets. Lab on a Chip,

7(9):1139–1145, 2007.

[48] Derby, B. Printing and prototyping of tissues and scaffolds. Science, 338(6109):921–926, 2012.

[49] Des Marais, D., Nuth, J., Allamandola, L., Boss, A., Farmer, J., Hoehler, T., Jakosky, B., Meadows,

V., Pohorille, A., Runnegar, B., and Spormann, A. The NASA astrobiology roadmap. Astrobiology,

8(4):715–730, 2008. doi:10.1089/ast.2008.0819.

[50] Draper, D. Themes in RNA-protein recognition. Journal of Molecular Biology, 293(2):255–270, 1999.

doi:10.1006/jmbi.1999.2991.

[51] Drickamer, K. C-type lectin-like domains. Curr Opin Struct Biol, 9(5):585–590, 1999.

[52] Elowitz, M. and Leibler, S. A synthetic oscillatory network of transcriptional regulators. Nature,

403(6767):335–338, 2000.

116

[53] Engler, C., Gruetzner, R., Kandzia, R., and Marillonnet, S. Golden gate shuffling: a one-pot DNA

shuffling method based on type IIs restriction enzymes. PLoS ONE, 4(5):e5553, 2009.

[54] Evans, F. and Lissner, H. Tensile and compressive strength of human parietal bone. Journal of Applied

Physiology, 10(3):493–497, 1957.

[55] Fairchild, K. and Mendell, W. Report of the In Situ Resources Utilization Workshop. National Aero-

nautics and Space Administration, 1988.

[56] Falini, G. and Fermani, S. Chitin mineralization. Tissue Engineering, 10(1-2):1–6, 2004. doi:10.1089/

107632704322791646.

[57] Fincham, A., Luo, W., Moradian-Oldak, J., Paine, M., Snead, M., and Zeichner-David, M. Enamel

biomineralization: the assembly and disassembly of the protein extracellular organic matrix. In

Teaford, M., Smith, M., and Ferguson, M., editors, Development, Function and Evolution of Teeth,

pp. 37–61. Cambridge, 2000.

[58] Gaines IV, W. and Marcotte Jr., W. Identification and characterization of multiple Spidroin 1 genes

encoding major ampullate silk proteins in Nephila clavipes. Insect Molecular Biology, 17(5):465–474,

2008. doi:10.1111/j.1365-2583.2008.00828.x.

[59] Gibson, I., Rosen, D., and Stucker, B. Additive Manufacturing Technologies: Rapid Prototyping to

Direct Digital Manufacturing. Springer, 2010.

[60] Gillooly, D., Simonsen, A., and Stenmark, H. Cellular functions of phosphatidylinositol 3-phosphate

and FYVE domain proteins. The Biochemical Journal, 355:249–258, 2001.

[61] Gilroy, J., Gadt, T., Whittell, G., Chabanne, L., Mitchels, J., Richardson, R., Winnik, M., and Manners,

I. Monodisperse cylindrical micelles by crystallization-driven living self-assembly. Nature Chemistry,

2(7):566–570, 2010. doi:10.1038/nchem.664.

[62] Girotti, A., Reguera, J., Rodrıguez-Cabello, J., Arias, F., Alonso, M., and Matestera, A. Design and

bioproduction of a recombinant multi(bio)functional elastin-like protein polymer containing cell adhe-

sion sequences for tissue engineering purposes. Journal of Materials Science: Materials in Medicine,

15(4):479–484, 2004.

[63] Gordeev, A., Chetverina, H., and Chetverin, A. Planar arrangement of eukaryotic cells in merged

hydrogels combines the advantages of 3-D and 2-D cultures. BioTechniques, 52(5):325–331, 2012.

doi:10.2144/000113861.

[64] Gosline, J., Guerette, P., Ortlepp, C., and Savage, K. The mechanical design of spider silks: from

fibroin sequence to mechanical function. Journal of Experimental Biology, 202:3295–3303, 1999.

[65] Guest, A. and Pelton, J. Major launch systems available globally. In Pelton, J., Madry, S., and

117

Camacho-Lara, S., editors, Handbook of Satellite Applications, pp. 1207–1219. Springer, 2013.

[66] Haidar, Z., Hamdy, R., and Tabrizian, M. Delivery of recombinant bone morphogenetic proteins for

bone regeneration and repair. part B: Delivery systems for BMPs in orthopaedic and craniofacial tissue

engineering. Biotechnology Letters, 31(12):1825–1835, 2009.

[67] Hamid, Q., Snyder, J., Wang, C., Timmer, M., Hammer, J., Guceri, S., and Sun, W. Fabrication

of three-dimensional scaffolds using precision extrusion deposition with an assisted cooling device.

Biofabrication, 3:1–12, 2011.

[68] Hanukoglu, I. and Fuchs, E. The cDNA sequence of a human epidermal keratin: divergence of se-

quence but conservation of structure among intermediate filament proteins. Cell, 31(1):243–252, 1982.

[69] Hatzfeld, M. and Weber, K. The coiled coil of in vitro assembled keratin filaments is a heterodimer of

type I and II keratins: use of site-specific mutagenesis and recombinant protein expression. Journal of

Cell Biology, 110(4):1199–1210, 1990.

[70] Havemann, S. and Foster, J. Comparative characterization of the microbial diversities of an artificial

microbialite model and a natural stromatolite. Applied and Environmental Microbiology, 74(23):7410–

7421, 2008. doi:10.1128/AEM.01710-08.

[71] He, P., Sahoo, S., Ng, K., Chen, K., Toh, S., and Goh, J. Enhanced osteoinductivity and osteocon-

ductivity through hydroxyapatite coating of silk-based tissue-engineered ligament scaffold. Journal of

Biomedical Materials Research Part A, 101A(2):555–566, 2012.

[72] Heindl, R. and Pendergast, W. Young’s modulus of elasticity at several temperatures for some refrac-

tories of varying silica content. Journal of Research of the National Bureau of Standards, 13:851–862,

1934.

[73] Hinoki, T., Snead, L., Lara-Curzio, E., Park, J., and Kohyama, A. Effect of fiber properties on me-

chanical properties of crystalline silicon carbide composites. In Lin, H.-T. and Singh, M., editors,

26th Annual Conference on Composites, Advanced Ceramics, Materials, and Structures: Ceramic

Engineering and Science Proceedings, volume 23(3), pp. 511–518. John Wiley & Sons, Inc., 2002.

doi:10.1002/9780470294741.ch60.

[74] Hochuli, E., Bannwarth, W., Dobeli, H., Gentz, R., and Stuber, D. Genetic approach to facilitate

purification of recombinant proteins with a novel metal chelate adsorbent. Nature Biotechnology,

6(11):1321–1325, 1988.

[75] Holton, T. and Graham, M. A simple and efficient method for direct cloning of PCR products using

ddT-tailed vectors. Nucleic Acids Research, 19(5):1156, 1991.

[76] Honda, Y. Study of the effects of space flight on the aging of C. elegans (space aging).

118

http://www.nasa.gov/mission_pages/station/research/experiments/1081.html, 2014.

Accessed: November 29, 2014.

[77] Hovsepian, S. Digital material skins: for reversible reusable pressure vessels. Master’s thesis, Mas-

sachusetts Institute of Technology, 2012.

[78] Howard, T., Middelhaufe, S., Moore, K., Edner, C., Kolak, D., Taylor, G., Parker, D., Lee, R., Smirnoff,

N., Aves, S., and Love, J. Synthesis of customized petroleum-replica fuel molecules by targeted

modification of free fatty acid pools in Escherichia coli. Proceedings of the National Academy of

Sciences of the United States of America, 110(19):7636–7641, 2013.

[79] Hoyt, R., Cushing, J., Slostad, J., Jimmerson, G., Moser, T., Kirkos, G., Jaster, M., and Voronka,

N. SpiderFab: An architecture for self-fabricating space systems. In SPACE 2013 Conference &

Exposition, pp. 2013–5509. American Institute of Aeronautics and Astronautics, San Diego, USA,

2013. doi:10.2514/6.2013-5509.

[80] Huang, J., Wang, X., and Wang, Z. Bio-inspired fabrication of antireflection nanostructures by repli-

cating fly eyes. Nanotechnology, 19:5602, 2008.

[81] Huey, D., Hu, J., and Athanasiou, K. Unlike bone, cartilage regeneration remains elusive. Science,

338(6109):917–921, 2012. doi:10.1126/science.1222454.

[82] Iguchi, M., Yamanaka, S., and Budhiono, A. Bacterial cellulose – a masterpiece of nature’s arts.

Journal of Materials Science, 35:261, 2000. doi:10.1023/A:1004775229149.

[83] Ingber, D. Tensegrity I. Cell structure and hierarchical systems biology. Journal of Cell Science,

116:1157–1173, 2003.

[84] Iyer, P., Liu, Y., Deng, Y., Mcmanus, J., Kao, T., and Tien, M. Processing of cellulose synthase (Ac-

sAB) from Gluconacetobacter hansenii 23769. Archives of Biochemistry and Biophysics, 529(2):92–

98, 2013. doi:10.1016/j.abb.2012.12.002.

[85] Jakab, K., Norotte, C., Damon, B., Marga, F., Neagu, A., Besch-Williford, C., Kachurin, A., Church,

K., Park, H., Mironov, V., Markwald, R., Vunjak-Novakovic, G., and Forgacs, G. Tissue engineering

by self-assembly of cells printed into topologically defined structures. Tissue Engineering: Part A,

14(3):413–421, 2008.

[86] Jakosky, B., Nealson, K., Bakermans, C., Ley, R., and Mellon, M. Subfreezing activity of mi-

croorganisms and the potential habitability of Mars’ polar regions. Astrobiology, 3:343, 2003. doi:

10.1089/153110703769016433.

[87] Jessup, J. Cellular biotechnology operations support systems. http://www.nasa.gov/mission_

pages/station/research/experiments/388.html, 2014. Accessed: Nov 29, 2014.

119

[88] Kennedy, K., Alexander, L., Landis, R., Linne, D., Mclemore, C., and Santiago-Maldonado, E. Human

Exploration Destination Systems Roadmap. National Aeronautics and Space Administration, 2010.

[89] Keppler-Ross, S., Noffz, C., and Dean, N. A new purple fluorescent color marker for genetic studies

in Saccharomyces cerevisiae and Candida albicans. Genetics, 179(1):705–710, 2008. doi:10.1534/

genetics.108.087080.

[90] Kershaw, C., Brown, N., and Hobman, J. Zinc dependence of zinT (yodA) mutants and binding of zinc,

cadmium and mercury by ZinT. Biochemical and Biophysical Research Communications, 364(1):66–

71, 2007. doi:10.1016/j.bbrc.2007.09.094.

[91] Khan, M., Fon, D., Li, X., Tian, J., Forsythe, J., Garnier, G., and Shen, W. Biosurface engineering

through ink jet printing. Colloids and Surfaces B: Biointerfaces, 75(2):441–447, 2010. doi:10.1016/j.

colsurfb.2009.09.032.

[92] Ki, C., Um, I., and Park, Y. Acceleration effect of sericin on shear-induced beta-transition of silk

fibroin. Polymer, 50(19):4618–4625, 2009. doi:10.1016/j.polymer.2009.02.017.

[93] Kim, J. Three-dimensional tissue culture models in cancer biology. Seminars in Cancer Biology,

15(5):365–377, 2005.

[94] Kitts, C., Hines, J., Agasid, E., Ricco, A., Yost, B., Ronzano, K., and Puig-Suari, J. The GeneSat-1

microsatellite mission: A challenge in small satellite design. In 20th Annual AIAA/USU Conference

on Small Satellites, pp. SSC06–IV–8. Utah State University, American Institute for Aeronautics and

Astronautics, 2006.

[95] Komeili, A., Li, Z., Newman, D., and Jensen, G. Magnetosomes are cell membrane invaginations

organized by the actin-like protein MamK. Science, 311(5758):242–245, 2006. doi:10.1126/science.

1123231.

[96] Kotrba, P., Doleckova, L., de Lorenzo, V., and Ruml, T. Enhanced bioaccumulation of heavy metal ions

by bacterial cells due to surface display of short metal binding peptides. Applied and Environmental

Microbiology, 65(3):1092–1098, 1999.

[97] Kretschmann, D. Mechanical properties of wood. In Wood Handbook: Wood as an Engineering

Material, pp. 1–46. U.S. Department of Agriculture, U.S. Forest Service, Forest Products Laboratory,

2010.

[98] Kuranda, M. and Robbins, P. Chitinase is required for cell separation during growth of Saccharomyces

cerevisiae. Journal of Biological Chemistry, 266(29):19758–19767, 1991.

[99] Lersch, R. and Fuchs, E. Sequence and expression of a type II keratin, K5, in human epidermal cells.

Molecular and Cellular Biology, 8(1):486–493, 1988.

120

[100] Levingstone, T., Matsiko, A., Dickson, G., O’Brien, F., and Gleeson, J. A biomimetic multi-layered

collagen-based scaffold for osteochondral repair. Acta Biomaterialia, 10(5):1996–2004, 2014.

[101] Liberski, A., Delaney, J., and Schubert, U. “One cell – one well”: a new approach to inkjet printing

single cell microarrays. ACS Combinatorial Science, 13(2):190–195, 2011.

[102] Liu, H. and Naismith, J. An efficient one-step site-directed deletion, insertion, single and multiple-site

plasmid mutagenesis protocol. BMC Biotechnology, 8:91, 2008.

[103] Liu, T., Reyes-Caballero, H., Li, C., Scott, R., and Giedroc, D. Multiple metal binding domains

enhance the Zn(II) selectivity of the divalent metal ion transporter AztA. Biochemistry, 46(39):11057–

11068, 2007. doi:10.1021/bi7006367.

[104] Lourenco, P., Almeida, J., and Barros, J. Experimental investigation of bricks under uniaxial tensile

testing. Journal of the British Masonry Society, 18(1):1–10, 2005.

[105] Lowe, J. and Amos, L. Evolution of cytomotive filaments: the cytoskeleton from prokaryotes to

eukaryotes. The International Journal of Biochemistry & Cell Biology, 41(2):323–329, 2009. doi:

10.1016/j.biocel.2008.08.010.

[106] Luger, K., Mader, A., Richmond, R., Sargent, D., and Richmond, T. Crystal structure of the nucleo-

some core particle at 2.8 A resolution. Nature, 389(6648):251–260, 1997. doi:10.1038/38444.

[107] Luz, G. and Mano, J. Biomimetic design of materials and biomaterials inspired by the structure of

nacre. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering

Sciences, 367(1893):1587–1605, 2009. doi:10.1098/rsta.2009.0007.

[108] Mankins, J. Technology Readiness Levels: A White Paper. Advanced Concepts Office, Office of Space

Access and Technology, 1995.

[109] Manly, R., Hodge, H., and Ange, L. Density and refractive index studies of dental hard tissues: II.

Density distribution curves 1,2. Journal of Dental Research, 18(3):203–211, 1939. doi:10.1177/

00220345390180030301.

[110] Marin, F., Luquet, G., Marie, B., and Medakovic, D. Molluscan shell proteins: Primary structure,

origin, and evolution. In Current Topics in Developmental Biology, pp. 209–275. Elsevier, 2008.

[111] Marmorstein, R. and Fitzgerald, M. Modulation of DNA-binding domains for sequence-specific DNA

recognition. Gene, 304:1–12, 2003.

[112] Mayer, G. New toughening concepts for ceramic composites from rigid natural materials. Journal of

the Mechanical Behavior of Biomedical Materials, 4(5):670–681, 2011.

[113] McConnell, K., Slater, J., Han, A., West, J., and Suh, J. Microcontact printing for co-patterning cells

and viruses for spatially controlled substrate-mediated gene delivery. Soft Matter, 7:4993, 2011.

121

[114] McKittrick, J., Chen, P., Tombolato, L., Novitskaya, E., Trim, M., Hirata, G., Olevsky, E., Horste-

meyer, M., and Meyers, M. Energy absorbent natural materials and bioinspired design strategies: A

review. Materials Science and Engineering: C, 30(3):331–342, 2010.

[115] Mera, H. and Takata, T. High-performance fibers. In Ullmann’s Encyclopedia of Industrial Chemistry,

pp. 1–26. Wiley-VCH Verlag GmbH & Co., 2000.

[116] Merrin, J., Leibler, S., and Chuang, J. Printing multistrain bacterial patterns with a piezoelectric inkjet

printer. PLoS ONE, 2(7):e663, 2007.

[117] Meyer, C. Lunar regolith. In NASA Lunar Petrographic Educational Thin Section Set, pp. 46–48.

National Aeronautics and Space Administration, 2003.

[118] Meyers, M., Chen, P., Lopez, M., Seki, Y., and Lin, A. Biological materials: a materials science

approach. Journal of the Mechanical Behavior of Biomedical Materials, 4(5):626–657, 2010. doi:

10.1016/j.jmbbm.2010.08.005.

[119] Meyers, M., McKittrick, J., and Chen, P. Structural biological materials: critical mechanics-materials

connections. Science, 339(6121):773–779, 2013. doi:10.1126/science.1220854.

[120] Michielsen, K. and Stavenga, D. Gyroid cuticular structures in butterfly wing scales: biological pho-

tonic crystals. Journal of the Royal Society Interface, 5(18):85–94, 2008. doi:10.1098/rsif.2007.1065.

[121] Minelli, G., Kitts, C., Ronzano, K., Beasley, C., Rasay, R., Mas, I., Williams, P., Mahacek, P., Shepard,

J., Acain, J., Hines, J., Agasid, E., Friedericks, C., Piccini, M., Parra, M., Timucin, L., Henschke, M.,

Luzzi, E., Mai, N., McIntyre, M., Ricks, R., Squires, D., Storment, C., Tucker, J., Yost, B., Defouw,

G., and Ricco, A. Extended life flight results from the GeneSat-1 biological microsatellite mission.

In 22nd Annual AIAA/USU Conference on Small Satellites, pp. SSC08–II–4. American Institute of

Aeronautics and Astronautics, 2008.

[122] Miyako, E., Sugino, T., Okazaki, T., Bianco, A., Yudasaka, M., and Iijima, S. Self-assembled carbon

nanotube honeycomb networks using a butterfly wing template as a multifunctional nanobiohybrid.

ACS Nano, 7(10):8736–8742, 2013. doi:10.1021/nn403083v.

[123] Molecular Probes, Inc. FungaLight CFDA, AM/Propidium Iodide Yeast Vitality Kit, 2005.

[124] National Aeronautics and Space Administration. EVA Tools and Equipment Reference Book, 1993.

[125] National Aeronautics and Space Administration. Human Exploration of Mars, Design Reference Ar-

chitecture 5.0 - Addendum, 2010.

[126] National Aeronautics and Space Administration. SpaceX 2 Cargo Manifest, 2013.

[127] National Aeronautics and Space Administration. SpaceX CRS-2 Mission Press Kit, 2013.

[128] Nicholson, W., Ricco, A., Agasid, E., Beasley, C., Diaz-Aguado, M., Ehrenfreund, P., Friedericks,

122

C., Ghassemieh, S., Henschke, M., Hines, J., Kitts, C., Luzzi, E., Ly, D., Mai, N., Mancinelli, R.,

Mcintyre, M., Minelli, G., Neumann, M., Parra, M., Piccini, M., Rasay, R., Ricks, R., Santos, O.,

Schooley, A., Squires, D., Timucin, L., Yost, B., and Young, A. The O/OREOS mission: First science

data from the space environment survivability of living organisms (SESLO) payload. Astrobiology,

11:951, 2011. doi:10.1089/ast.2011.0714.

[129] Nielson, R., Kaehr, B., and Shear, J. Microreplication and design of biological architectures using

dynamic-mask multiphoton lithography. Small, 5(1):120–125, 2009. doi:10.1002/smll.200801084.

[130] Nowak, P., Sadeh, W., and Morroni, L. Geometric modeling of inflatable structures for lunar base.

Journal of Aerospace Engineering, 5:311, 1992.

[131] O’Connor, S., Stenger, D., Shaffer, K., and Ma, W. Survival and neurite outgrowth of rat cortical

neurons in three-dimensional agarose and collagen gel matrices. Neuroscience Letters, 304(3):189–

193, 2001.

[132] Octave Community. GNU Octave 3.2.3, 2010.

[133] Olivera, E., Carnicero, D., Jodra, R., Minambres, B., Garcıa, B., Abraham, G., Gallardo, A., Roman,

J., Garcıa, J., Naharro, G., and Luengo, J. Genetically engineered Pseudomonas: a factory of new

bioplastics with broad applications. Environmental Microbiology, 3(10):612–618, 2001.

[134] Olsson-Francis, K. and Cockell, C. Use of cyanobacteria for in-situ resource use in space applications.

Planetary and Space Science, 58(10):1279–1285, 2010.

[135] Partow, S., Siewers, V., Bjørn, S., Nielsen, J., and Maury, J. Characterization of different promoters

for designing a new expression vector in Saccharomyces cerevisiae. Yeast, 27(11):955–964, 2010.

doi:10.1002/yea.1806.

[136] Pawson, T. Protein modules and signalling networks. Nature, 373(6515):573–580, 1995. doi:10.1038/

373573a0.

[137] Pedelacq, J., Cabantous, S., Tran, T., Terwilliger, T., and Waldo, G. Engineering and characterization

of a superfolder green fluorescent protein. Nature Biotechnology, 24(1):79–88, 2006. doi:10.1038/

nbt1172.

[138] Perry, R., McLoughlin, N., Lynne, B., Sephton, M., Oliver, J., Perry, C., Campbell, K., Engel, M.,

Farmer, J., Brasier, M., and Staley, J. Defining biominerals and organominerals: Direct and indirect

indicators of life. Sedimentary Geology, 201:157, 2007. doi:10.1016/j.sedgeo.2007.05.014.

[139] Pirone, D., Qi, L., Colecraft, H., and Chen, C. Spatial patterning of gene expression using surface-

immobilized recombinant adenovirus. Biomedical Microdevices, 10(4):561–566, 2008.

[140] Poole, D., Morag, E., Lamed, R., Bayer, E., Hazlewood, G., and Gilbert, H. Identification of the

123

cellulose-binding domain of the cellulosome subunit S1 from Clostridium thermocellum YS. FEMS

Microbiology Letters, 78(2-3):181–186, 1992.

[141] Porter, D., Vollrath, F., and Shao, Z. Predicting the mechanical properties of spider silk as a model

nanostructured polymer. The European Physical Journal E: Soft Matter and Biological Physics,

16(2):199–206, 2005. doi:10.1140/epje/e2005-00021-2.

[142] Porter, M., Novitskaya, E., Castro-Cesena, A., Meyers, M., and McKittrick, J. Highly deformable

bones: Unusual deformation mechanisms of seahorse armor. Acta Biomaterialia, 9(6):6763–6770,

2013. doi:10.1016/j.actbio.2013.02.045.

[143] Raabe, D., Sachs, C., and Romano, P. The crustacean exoskeleton as an example of a structurally and

mechanically graded biological nanocomposite material. Acta Materialia, 53(15):4281–4292, 2005.

[144] Rashid, M., Mansur, M., and Paramasivam, P. Correlations between mechanical properties of high-

strength concrete. Journal of Materials in Civil Engineering, 14(3):230–238, 2002.

[145] Ratner, B., Hoffman, A., Schoen, F., and Lemons, J. Biomaterials science: a multidisciplinary en-

deavor. In Biomaterials science: An introduction to materials in medicine, pp. 1–19. Elsevier, Inc.,

2004.

[146] Ringeisen, B., Othon, C., Barron, J., Young, D., and Spargo, B. Jet-based methods to print living cells.

Biotechnology Journal, 1(9):930–948, 2006.

[147] Rosa, M. and Fortes, M. Rate effects on the compression and recovery of dimensions of cork. Journal

of Materials Science, 23(3):879–885, 1988.

[148] Rosa, M. and Fortes, M. Deformation and fracture of cork in tension. Journal of Materials Science,

26(2):341–348, 1991.

[149] Rouillard, A., Berglund, C., Lee, J., Polacheck, W., Tsui, Y., Bonassar, L., and Kirby, B. Methods

for photocrosslinking alginate hydrogel scaffolds with high cell viability. Tissue Engineering Part C:

Methods, 17(2):173–179, 2011. doi:10.1089/ten.TEC.2009.0582.

[150] Sabelhaus, A., Bruce, J., Caluwaerts, K., Chen, Y., Lu, D., Liu, Y., Agogino, A., SunSpiral, V., and

Agogino, A. Hardware design and testing of SUPERball, a modular tensegrity robot. In 6th World

Conference of the International Association for Structural Control and Monitoring, pp. 1–11. Euro-

pean Association for the Control of Structures, 2014.

[151] Sanchez, J. and McInnes, C. Assessment on the feasibility of future shepherding of asteroid resources.

Acta Astronautica, 73:49–66, 2012. doi:10.1016/j.actaastro.2011.12.010.

[152] Sarikaya, M. Biomimetics: materials fabrication through biology. Proceedings of the National

Academy of Sciences of the United States of America, 96(25):14183–14185, 1999.

124

[153] Schippers, A., Hedrich, S., Vasters, J., Drobe, M., Sand, W., and Willscher, S. Biomining: metal

recovery from ores with microorganisms. In Advances in Biochemical Engineering/Biotechnology.

Springer-Verlag, Berlin, 2013.

[154] Schnabel, J., Weber, K., and Hatzfeld, M. Protein-protein interactions between keratin polypeptides

expressed in the yeast two-hybrid system. Biochimica et Biophysica Acta, 1403(2):158–168, 1998.

[155] Sequin, C. Rapid prototyping: a 3D visualization tool takes on sculpture and mathematical forms.

Communications of the ACM, 48(6):66–73, 2005.

[156] Serrano, L. Synthetic biology: promises and challenges. Molecular Systems Biology, 3:158, 2007.

doi:10.1038/msb4100202.

[157] Sethmann, I. and Worheide, G. Structure and composition of calcareous sponge spicules: a review and

comparison to structurally related biominerals. Micron, 39(3):209–228, 2008. doi:10.1016/j.micron.

2007.01.006.

[158] Shimizu, K., Cha, J., Stucky, G., and Morse, D. Silicatein α: cathepsin L-like protein in sponge biosil-

ica. Proceedings of the National Academy of Sciences of the United States of America, 95(11):6234–

6238, 1998.

[159] Shimizu-Sato, S., Huq, E., Tepperman, J., and Quail, P. A light-switchable gene promoter system.

Nature Biotechnology, 20(10):1041–1044, 2002.

[160] Shore, S., Unnikrishnan, G., Hussein, A., and Morgan, E. Bone biomechanics. In Winkelstein, B.,

editor, Orthopaedic Biomechanics, pp. 1–46. CRC Press, 2012.

[161] Singh, V. Recent advancements in synthetic biology: current status and challenges. Gene, 535(1):1–11,

2014.

[162] Slater, M. and Craig, E. Transcriptional regulation of an hsp70 heat shock gene in the yeast Saccha-

romyces cerevisiae. Molecular and Cellular Biology, 7(5):1906–1916, 1987.

[163] Song, S., Choi, J., Park, Y., Hong, S., Lee, J., Ahn, C., Choi, H., and Sun, K. Sodium alginate hydrogel-

based bioprinting using a novel multinozzle bioprinting system. Artificial Organs, 35(11):1132–1136,

2011. doi:10.1111/j.1525-1594.2011.01377.x.

[164] Sorokina, O., Kapus, A., Terecskei, K., Dixon, L., Kozma-Bognar, L., Nagy, F., and Millar, A. A

switchable light-input, light-output system modelled and constructed in yeast. Journal of Biological

Engineering, 3:15, 2009.

[165] Southern, J., Young, D., Heaney, F., Baumgartner, W., and Randall, R. Identification of an epitope on

the P and V proteins of simian virus 5 that distinguishes between two isolates with different biological

characteristics. Journal of General Virology, 72(Pt 7):1551–1557, 1991.

125

[166] Sun, J., Shao, Z., Zhao, H., Nair, N., Wen, F., Xu, J., and Zhao, H. Cloning and characterization of a

panel of constitutive promoters for applications in pathway engineering in Saccharomyces cerevisiae.

Biotechnology and Bioengineering, 109(8):2082–2092, 2012. doi:10.1002/bit.24481.

[167] Sun, L., Li, J., Park, J., and Zhou, H. Cooperative template-directed assembly of mesoporous metal-

organic frameworks. Journal of the American Chemical Society, 134(1):126–129, 2012. doi:10.1021/

ja209698f.

[168] Sunagawa, N., Fujiwara, T., Yoda, T., Kawano, S., Satoh, Y., Yao, M., Tajima, K., and Dairi, T.

Cellulose complementing factor (Ccp) is a new member of the cellulose synthase complex (terminal

complex) in Acetobacter xylinum. Journal of Bioscience and Bioengineering, 115(6):607–612, 2013.

doi:10.1016/j.jbiosc.2012.12.021.

[169] Szot, C., Buchanan, C., Freeman, J., and Rylander, M. 3D in vitro bioengineered tumors based on

collagen I hydrogels. Biomaterials, 32(31):7905–7912, 2011.

[170] Taboas, J., Maddox, R., Krebsbach, P., and Hollister, S. Indirect solid free form fabrication of local and

global porous, biomimetic and composite 3D polymer-ceramic scaffolds. Biomaterials, 24(1):181–

194, 2003.

[171] Takasu, Y., Yamada, H., Tamura, T., Sezutsu, H., Mita, K., and Tsubouchi, K. Identification and

characterization of a novel sericin gene expressed in the anterior middle silk gland of the silkworm

Bombyx mori. Insect Biochemistry and Molecular Biology, 37(11):1234–1240, 2007. doi:10.1016/j.

ibmb.2007.07.009.

[172] Tatematsu, K., Kobayashi, I., Uchino, K., Sezutsu, H., Iizuka, T., Yonemura, N., and Tamura, T.

Construction of a binary transgenic gene expression system for recombinant protein production in the

middle silk gland of the silkworm Bombyx mori. Transgenic Research, 19(3):473–487, 2010. doi:

10.1007/s11248-009-9328-2.

[173] Taylor, N., Howells, R., Huttly, A., Vickers, K., and Turner, S. Interactions among three distinct CesA

proteins essential for cellulose synthesis. Proceedings of the National Academy of Sciences of the

United States of America, 100(3):1450–1455, 2003. doi:10.1073/pnas.0337628100.

[174] Termonia, Y. Molecular modeling of the stress/strain behavior of spider dragline. In Elices, M.,

editor, Structural Biological Materials – Design and Structure-Property Relationships, pp. 337–349.

Pergamon Press, 2000.

[175] Trim, M., Horstemeyer, M., Rhee, H., El Kadiri, H., Williams, L., Liao, J., Walters, K., McKittrick,

J., and Park, S. The effects of water and microstructure on the mechanical properties of bighorn sheep

(Ovis canadensis) horn keratin. Acta Biomaterialia, 7(3):1228–1240, 2011. doi:10.1016/j.actbio.2010.

126

11.024.

[176] Vaughn, P., Galanis, M., Richards, K., Tebb, T., Ramshaw, J., and Werkmeister, J. Production of

recombinant hydroxylated human type III collagen fragment in Saccharomyces cerevisiae. DNA and

Cell Biology, 17(6):511–518, 1998.

[177] Vincent, J. Arthropod cuticle: a natural composite shell system. Composites Part A: Applied Science

and Manufacturing, 33(10):1311–1315, 2002.

[178] Vincent, J. and Wegst, U. Design and mechanical properties of insect cuticle. Arthropod Structure &

Development, 33(3):187–199, 2004. doi:10.1016/j.asd.2004.05.006.

[179] Volfson, D., Cookson, S., Hasty, J., and Tsimring, L. Biomechanical ordering of dense cell

populations. Proceedings of the National Academy of Sciences of the United States of America,

105(40):15346–15351, 2008.

[180] Wegst, U. Bending efficiency through property gradients in bamboo, palm, and wood-based com-

posites. Journal of the Mechanical Behavior of Biomedical Materials, 4(5):744–755, 2011. doi:

10.1016/j.jmbbm.2011.02.013.

[181] Wegst, U. and Ashby, M. The mechanical efficiency of natural materials. Philosophical Magazine,

84(21):2167–2186, 2004.

[182] Weiss, I. and Kirchner, H. Plasticity of two structural proteins: alpha-collagen and beta-keratin. Jour-

nal of the Mechanical Behavior of Biomedical Materials, 4(5):733–743, 2011. doi:10.1016/j.jmbbm.

2011.02.008.

[183] Williams, D. Mars fact sheet. http://nssdc.gsfc.nasa.gov/planetary/factsheet/

marsfact.html, 2014. Accessed: 30 Nov, 2014.

[184] Williams, D. Moon fact sheet. http://nssdc.gsfc.nasa.gov/planetary/factsheet/

moonfact.html, 2014. Accessed: Nov 30, 2014.

[185] Wylie, R., Ahsan, S., Aizawa, Y., Maxwell, K., Morshead, C., and Shoichet, M. Spatially controlled

simultaneous patterning of multiple growth factors in three-dimensional hydrogels. Nature Materials,

10(10):799–806, 2011. doi:10.1038/nmat3101.

[186] Xia, X., Qian, Z., Ki, C., Park, Y., Kaplan, D., and Lee, S. Native-sized recombinant spider silk

protein produced in metabolically engineered escherichia coli results in a strong fiber. Proceedings

of the National Academy of Sciences of the United States of America, 107(32):14059–14063, 2010.

doi:10.1073/pnas.1003366107.

[187] Xu, H., Cao, B., George, A., and Mao, C. Self-assembly and mineralization of genetically modifiable

biological nanofibers driven by beta-structure formation. Biomacromolecules, 12(6):2193–2199, 2011.

127

[188] Xu, T., Kincaid, H., Atala, A., and Yoo, J. High-throughput production of single-cell micropar-

ticles using an inkjet printing technology. Journal of Manufacturing Science and Engineering,

130(2):020601.1–024503.5, 2008. doi:10.1115/1.2903064.

[189] Xu, T., Petridou, S., Lee, E., Roth, E., Vyavahare, N., Hickman, J., and Boland, T. Construction of

high-density bacterial colony arrays and patterns by the ink-jet method. Biotechnology and Bioengi-

neering, 85(1):29–33, 2004.

[190] Yang, M., Shuai, Y., Zhang, C., Chen, Y., Zhu, L., Mao, C., and Ouyang, H. Biomimetic nucleation of

hydroxyapatite crystals mediated by Antheraea pernyi silk sericin promotes osteogenic differentiation

of human bone marrow derived mesenchymal stem cells. Biomacromolecules, 15(4):1185–1193, 2014.

[191] Zhang, K., Chou, C., Xia, X., Hung, M., and Qin, L. Block-cell-printing for live single-cell printing.

Proceedings of the National Academy of Sciences of the United States of America, 111(8):2948–2953,

2014.

[192] Zhang, W., Zhang, D., Fan, T., Gu, J., Ding, J., Wang, H., Guo, Q., and Ogawa, H. Novel photoanode

structure templated from butterfly wing scales. Chemistry of Materials, 21(1):33–40, 2008.

[193] Zhao, Y., Yao, R., Ouyang, L., Ding, H., Zhang, T., Zhang, K., Cheng, S., and Sun, W. Three-

dimensional printing of Hela cells for cervical tumor model in vitro. Biofabrication, 6(3):035001,

2014. doi:10.1088/1758-5082/6/3/035001.

[194] Zhou, C., Confalonieri, F., Medina, N., Zivanovic, Y., Esnault, C., Yang, T., Jacquet, M., Janin, J.,

Duguet, M., Perasso, R., and Li, Z. Fine organization of Bombyx mori fibroin heavy chain gene.

Nucleic Acids Research, 28(12):2413–2419, 2000.

[195] Zhou, Y., Yang, W., Kirberger, M., Lee, H., Ayalasomayajula, G., and Yang, J. Prediction of EF-hand

calcium-binding proteins and analysis of bacterial EF-hand proteins. Proteins, 65(3):643–655, 2006.

doi:10.1002/prot.21139.

[196] Ziauddin, J. and Sabatini, D. Microarrays of cells expressing defined cDNAs. Nature, 411(6833):107–

110, 2001.

[197] Zsebo, K., Lu, H., Fieschko, J., Goldstein, L., Davis, J., Duker, K., Suggs, S., Lai, P., and Bitter,

G. Protein secretion from Saccharomyces cerevisiae directed by the prepro-alpha-factor leader region.

Journal of Biological Chemistry, 261(13):5858–5865, 1986.

128

Part II

Haptics

129

130

6 Background

6.1 Teleoperation and Humans in the Loop

Teleoperation, or operation of a device in a remote environment (Figure 6.1), has been used for decades in

fields ranging from space exploration [62] to minimally invasive surgery [38]. Although improvements in

mobile computing technology have made it possible in recent years for many such remote tasks to be fully

automated, there remain many applications where safety, legal, and practical concerns require that a human

operator be involved in teleoperation. Such human-in-the-loop teleoperation may come in many forms, from

direct manual input over a control link to high-level ‘checks’ on autonomously generated behavior.

Because both the human user’s experience and the ability of the remote device to complete its assigned

task are critical, and interdependent, aspects of human-in-the-loop teleoperation, it lies at the intersection

of the broader fields of virtual environments and telerobotics. Virtual environments here are defined very

broadly (following, e.g., [68]) as those which are experienced by a human through mediation by human-

made technology, including graphics displays, audio speakers, haptic devices, and the like; thus, a camera

feed of a remote object displayed on a screen and an artificially generated image of the same object displayed

on the same screen are both considered virtual visual environments, and everything to the right of the operator

humanoperator

interpretation communications interpretation remoteenvironment

localdevice

remotedevice

Figure 6.1: The elements of a teleoperation setup. The interpretation block may

be a simple pass-through or may represent signal filtering, command interpretation,

predictive modeling, etc. Time delay is typically introduced in the communication

block.

131

in Figure 6.1 can be considered a virtual environment as well. Telerobotics is defined here as the study of

the systems used to carry out teleoperation, typically; thus every element of Figure 6.1 can be included when

viewed from a task-performance perspective. The two parent fields tend to approach the study of human-in-

the-loop teleoperation from significantly different philosophical and linguistic standpoints, most obviously

in the meaning and importance of telepresence (e.g., [39]); however, studies in both fields typically share an

underlying common goal of enabling better human activity (e.g., [67]).

6.2 Telerobotic Systems

Human-in-the-loop systems for teleoperation can take a wide variety of forms, from remote-controlled un-

derwater vehicles [59] to augmented surgical tools [70].

Although each teleoperation system consists of a local environment and remote environment, linked by com-

munications, with a tool of some sort at either end (Figure 6.1), real-world implementations can take a wide

variety of forms. The remote environment may be a few feet away, or on another planet; the ‘tools’ may

be a human hand, joystick, keyboard, wearable motion tracking system, robotic arm, or vehicle; the type

of information relayed between the two may be physical data about the device(s) such as force and motion,

physical data about the environment(s) such as visual or haptic sensing, or more abstract information such as

system states, pre-programmed actions, or autonomous behaviors. For example, a pantograph may be consid-

ered as an extremely simple telerobotic system, where the local and remote environments are connected by

a mechanical linkage that conveys position information, the local tool is the user’s hand, and the remote tool

is a pencil. An interacting swarm of autonomous small aircraft, with each unit responding to locally sensed

information and the overall swarm receiving high-level commands relayed wirelessly from a human operator,

may also be considered as a single telerobotic system, albeit at the other end of the complexity spectrum.

Traditional telerobotics defines systems as either unilateral, in which the information relayed from the remote

environment is purely passive (visual, sound, IR, etc.), or bilateral, in which the return path includes force

or motion information from the remote device. (In order for the local-side human to be in the loop, active

information must relayed in one form or another from the local to the remote location.) The reason for this

division stems from the all-important criterion of stability: systems with an active feedback path may gain

energy, leading to undesired, uncontrollable, or unpredictable behavior. However, bilateral devices have been

shown to be preferred by users [33] and to improve remote task performance [52, 55]; thus, the problematic

effects of haptic feedback on system stability have been extensively studied [24].

132

Figure 6.2: Systems for teleoperation may be simple, such as the pantograph, a

tool for reproducing drawings at different scales, or complex, such as a swarm of

interacting, autonomous aircraft controlled from the ground.

The usability of bilateral telerobotic systems is affected by factors such as loss of information during trans-

mission, differences in bidirectional bandwidth requirements, time lag in force and motion feedback, and

dynamic mismatches between the robot’s internal forces and external contact forces. Human-in-the-loop task

performance is further challenged by command-response delay, mismatch between remote and co-located en-

vironments, incompleteness of sensed information, and perceived system latency. A wide variety of control

techniques have been developed to mitigate the negative effects of these factors; the degree of remoteness,

most often encapsulated in the time delay necessary to send information between the two environments, is a

powerful factor in determining which approach is most appropriate.

Time delay is universally agreed to decrease user teleoperation performance. Transmission delay decreases

user performance linearly for small delays (< 5 s) and typically leads to development of distinct user strategies

such as ‘move and wait’ at around 0.7 s [23]. ‘Move and wait’ is a very common strategy for dealing with

perceived delays, and its particular implementation is typically very consistent for a given user [64]. User

error rate has previously been shown to increase with time delay as well [37]. Human perception of delay

is not automatically equivalent to empirical measures of delay; for instance, it has been shown that the types

of movement users make in a time-delayed haptic system affect their ability to perceive delay more strongly

than the nature of the haptic feedback from the environment [54].

Control schema focusing on high-bandwidth, bilateral relay of motion and force information between devices,

as generalized in the ‘four-channel’ system architecture [32], can provide users with accurate reproduction

of the contact forces and impedances experienced by the remote device. However, both stability and perfor-

mance criteria limit these methods to situations with very short communication delays [8, 17, 62], often well

below human perception thresholds (< 10 ms). This general approach can be made more robust to delay

133

by changing the type of information sent between devices to preserve passivity, as in wave variable encod-

ing [50, 51]. When delays are less than approximately one second, such passivity-based methods provide

good system performance [31].

With delays of several seconds, where human perception is substantially impacted, the traditional force/

motion feedback approach is no longer usable for human-in-the-loop applications. Desynchronization of

actions, sensory feedback, and reactions leads to either human-induced instability or inefficient behaviors

such as ‘move and wait’ strategies (reviewed in [62]). One approach to this problem is to use an internal

predictor, or a term which takes into account the current known state of the remote environment and attempts

to predict its response to a given local input; this predictor can be used to generate immediate feedback

and correspondingly adjust the delayed feedback when it arrives [4]. Other approaches included augmented

systems, which combine more traditional telerobotic systems with some degree of external virtual prediction

of the remote environment (such as a computer-manipulated video feed) to guide the user [2, 16]. Such

predictive systems are dependent on having an accurate model of the remote environment in advance of

sensory feedback; this is rarely the case in situations calling for human-in-the-loop control, however, as

otherwise the tasks to be completed could be automated without the need for human judgment [58].

For more extreme cases of delay, such as those encountered in deep underwater environments or in space

exploration, direct human control is no longer practical. A greater degree of autonomy on the part of the

remote robot becomes necessary, and the human input becomes more supervisory in nature [61]; commands

are given in the form of higher-level tasks (e.g., “drive to location A using route planning strategy B”).

Supervisory control has been shown to work well in a wide variety of situations, even without extreme time

delay, as long the remote environment is generally stable with respect to the task goals understood by the local

autonomous unit (e.g., the identification of location A does not change faster than route planning strategy B

can adapt). At the furthest extreme, at the boundary between teleoperation and remote autonomous operation,

robots may rely primarily on their own sensors and programmed behaviors with intermittent human operator

intervention [15].

It should be clear from the above examples that increasing remoteness generally calls for increasing levels

of abstraction to the human user (Figure 6.3). A relatively recent approach, intermediate between predictive

control and supervisory control, proposed to improve user task performance in telerobotic systems with large

time delays but slowly or infrequently changing remote environments is model-mediated teleoperation [48],

which is discussed more extensively in section 6.5.

134

< 10 ms

> 1 s

> 1 min

tim

ed

ela

y

de

gre

eo

fa

bstr

actio

n

complete

none

supervisory control

model-mediated control

predictive display

wave-variable encoding

force/position feedback

autonomous behavior with updates

partial

Figure 6.3: Higher levels of abstraction at the user level, corresponding to in-

creased autonomy at the robotic level, are necessary as the ’remoteness’ of the

environment – and hence the communications time delay – increases.

6.3 Virtual Environments

As mentioned in section 6.1, for purposes of this work, a virtual environment is defined (following, e.g., [68])

as one which is experienced by a human through mediation. It has been pointed out that to some degree, all

human experience with the external world is mediated through the mechanisms of our biological senses [34];

however, the use of the word here is limited to mediation via human-made technology. Virtual environ-

ments can be purely visual, such as those experienced through a television screen or head-mounted display

(Figure 6.4); audio, via headphones or ‘surround sound;’ tactile, via a force-feedback-enabled mouse or game

controller; or any combination of the same. Some innovative virtual technologies include use of additional

sensory mechanics such as proprioception [10], kinesthesia [40], or contact temperature [6].

Realism is defined as the degree to which the sensory information produced by a virtual environment matches

that which the user would expect to receive in the equivalent unmediated physical environment; the use of

the user’s sensory expectations, rather than the physical properties of the ‘real’ environment, as the standard

is important, as virtual environments may not have a physical equivalent – there is no such thing as a real

counterpart to a virtual dragon, but a user’s expectations about how such a virtual object should look, move,

sound, smell, and behave will still determine how ‘realistic’ he or she perceives it to be. The two criteria

often overlap, as human biological senses and our perceptual models and mechanisms evolved to deal with

a particular set of physical environment types, but the differentiation can be used to our advantage; for

instance, deliberate mismatch between high-information-content visual input and low-information-content

haptic information has been used to increase the apparent size of a virtual workspace [7].

The types of sensory information made available in the virtual environment, the types of interactions in which

135

Figure 6.4: Virtual worlds may be experienced in a variety of ways. Interactive text-

based virtual worlds are heavily mediated as sensory experiences; those conveyed

through head-tracking displays are intended to be more immersive.

users are able to engage with it, and the characteristics of individual users have long been known to be a com-

plex web of factors affecting users’ experience in virtual environments [68]. For convenience, these three

broad groups of factors can be remembered as information, interaction, and idiosyncracy. Early research

in teleoperation typically focused on the effects of information type, breadth, and depth, as the generally ex-

tremely limited remote communications bandwidth available were held to be the primary challenge; however,

as we enter an era both of rapidly increasing global telecommunications capability and rapidly decreasing lim-

its on real and virtual graphic, audio, and haptic rendering power, an increasing amount of work has shifted

to understanding the remaining factors of interactivity and idiosyncrasy. All three factors are interdependent,

particularly in their effects on user task performance.

As an example of dependence between these factors, it is generally held that reduction in sensory information

bandwidth and/or quality has been shown to decrease user performance in teleoperation, although there is

some evidence that this can be partially mitigated with training [53]. However, not only do humans have

asymmetrical input and output sensory information bandwidth requirements, but we may also change our

previously determined requirements in response to a given task [3]. This idiosyncratic behavior is affected

by the interactiveness of the task, as repeated interaction with an environment, whether real or virtual, serves

as an implicit source of feedback on perception and motion accuracy; it has been shown that ‘walking’ back

and forth between two points in a virtual environment substantially improves users’ ability to estimate more

136

general virtual distances [73].

The category of sensory information can further be subdivided into breadth – that is, the number of sen-

sory modalities affected – and depth, the amount of information made available to each sense. The process

of combining multi-modal sensory information into a perception is termed sensory integration (reviewed

in [14]). Redundant information provided to multiple sensory modalities has been held to increase the user’s

unconscious confidence in the ‘realness’ of a situation. Creating sensory redundancy in teleoperation sys-

tems has been shown to improve user performance, although this effect is likely task-dependent [56]. It has

also been argued that increased realism of rendered virtual environments may in fact lead to decreased task

performance, either through false confidence on the part of the users [43] or due to an increase in distract-

ing, unnecessary information present [52]. One study has shown that users’ estimation of virtual distances

becomes worse in the case of both larger and smaller virtual representations of previously experienced real

environments, and decreases only when the virtual environment matches the real environment; this provides

some support for the idea that users’ expectation of the ‘realness’ of a virtual environment improves task

performance, although given that surveys of users did not show a difference in reported experienced telep-

resence, such an effect would have to be primarily unconscious [25]. Conversely, another study showed no

effect on distance estimation with improved graphical quality [71], although it is unclear how well ‘realistic’

graphics alone translate into ‘convincingly real’ environments from a user perspective.

Moving beyond the criterion of realism, many virtual environments designed for teleoperation deliberately

add to or alter the presentation of the remote environment in the hopes of improving user task performance.

Increasing sensory breadth slightly has been shown to compensate for extremely reduced sensory depth, as

in adding a small amount of haptic feedback to predominant, but very low-bandwidth, graphic feedback [42].

The general view is that information sensed through multiple sensory modalities provides redundancy and

thus both robustness to noisy or low-bandwidth environments and greater user confidence in perceptual accu-

racy. For instance, one study showed that visual cues may help correct haptic biases in stiffness estimation,

but haptic cues may correct visual bias in size estimation [78]; another showed that visual cues [66] com-

pletely overrode haptic information in estimating deformation. Another relevant factor is prior experience;

users may use both outside knowledge, such as the fact that certain tapping sounds are typically associated

with certain types of material, and prior knowledge, such as how a particular haptic device portrays stiffness,

when deciding whether audio or haptic information is more reliable [11].

Given the presence of sensory breadth, consistency, or lack thereof, of information gained from different

sensory modalities has a strong effect on user perception and behavior. Sensory conflicts can include not

137

only external visual, audio, or haptic information, but internal proprioceptive and kinesthetic conflicts with

any of the above as well. The interference of time delay with a user’s ability to correlate multiple-modality

sensory information has been identified as one of the reasons why it generally correlates with decreased task

performance [19].

Vision is generally held to dominate other senses in the presence of intersensory conflict, a phenomenon

referred to as ‘visual capture’ (see [75] and references therein); some have proposed that this is due to vision

typically having denser and more precise information content, and thus implicitly presumed greater reliability,

than touch or other sense information [77]. This is supported by evidence that when the variance of visual

information content is artificially increased above that present in haptic information, users begin to more

heavily weight the latter [13].

The relationship between sensory integration, environment perception, and adaptation to tasks within an

environment is complex, and affected by users’ expectations, beliefs and prior experiences as well as external

information. Perceived temporal synchrony, perceived spatial coincidence, and prior experiential information

about likely sources of stimuli all play a role in how a user integrates information from an environment (e.g.,

attempting to determine whether an unexpectedly felt contact came from a nearby solid fixture or more distant

moving robotic arm). Users’ expectations and beliefs about what type of information it is ‘reasonable’ for an

environment to contain and what information is relevant to the task they have been asked to perform, as well

as explicit instructions given by trainers or experimenters, have all been shown to have an effect on the types

of sensory information users prioritize in the presence of conflict (see [18, 20, 21, 45, 80] and references

therein).

Beyond unintentional sensory conflict, artificial sensory substitution – e.g., transforming force magnitude to

a series of auditory tones – has also investigated. In some cases, it has been shown to improve user task

performance, particularly when the substituted sense has a larger stable dynamic range [41]. However, this

is dependent upon both the type of task and the type of environment; auditory substitution has been shown

to be more effective in tasks requiring handling of deformable objects than for rigid objects [55]. A study

of sensory substitution and augmentation (e.g., providing a bar graph of current contact force) in a primarily

visual virtual environment showed that users’ sense of presence and overall haptic task performance increased

with a small amount of additional-channel sensory information, but decreased in the presence of multiple

forms of sensory augmentation; however, this effect was reversed in the absence of haptic feedback [52].

Similarly, the use of artificial visual depth cues (e.g., rendered shadows) has been shown to improve task

138

performance in the presence of haptic feedback but to reduce it in its absence [43]. Thus, multiple sensory

channels can compete as well as complement.

User idiosyncrasy applies not only to reaction to different sensory modalities [68], but also to different tasks.

Completing the three-way interaction between information, interaction, and idiosyncracy, virtual environ-

ments’ effect on task performance is also dependent on both the nature of the task and the type of sensory

information present. For instance, one study has shown that the use of virtual graphic-only environments

improved part assembly task performance, both in terms of completion time and of reduced maximum con-

tact forces [52]. Another showed that for haptic-only tasks, different classes of tasks (for instance, touching

two points and then reproducing the distance between them vs. accurately returning to a point previously

touched) had substantially different accuracy and error rates [29]. Haptic-only constraints in a mixed graphic/

haptic environment have shown to be more useful in pick-and-place tasks than graphic-only constraints, with

graphic-only constraints actually causing a significant decrease in task performance [74]. Some effects are

relatively consistent across tasks and users; for instance, users typically underestimate distances in virtual

environments by a significantly greater margin than those in real environments. Echoing the example used

earlier, this underestimation effect is amenable to correction with task-specific explicit feedback [57] or more

general implicit feedback from repeated success or failure through environment interaction [49].

Despite this wealth of user studies and data, many of the factors affecting user task performance, including the

most likely types of errors made, are not well characterized. One study showed that when attempting to match

the size of two objects, users tended to underestimate the size of virtual objects when making estimates using

haptic information, but to overestimate them when using visual (graphically rendered) information; across

comparable sensory constraints, this effect was greatest when both objects were virtual, significant when one

was virtual and the other real, and not present when both were real [79]. A second indicated that the types of

body, arm, and hand motions made by a user when initially investigating a haptically rendered environment,

rather than type or quality of information provided about the environment itself, had the strongest effect on

users’ ability to accurately estimate and reproduce distance [29]. Yet a third, studying the tendency of users to

consistently under-estimate distances by a greater margin in virtual environments than in real environments,

found that estimation could be improved by adding haptic control of perceived distance traversal, but that the

type of control was not significant [76]. Such seemingly inconsistent results are widespread in the literature,

and have lead to the recognition that ‘task performance’ is not a monolithic metric; calls have been for

standard sets of ‘benchmark tests’ for user performance in task environments, but it is questionable whether

such measurements will override variability in user characteristics [67].

139

6.4 Telepresence

As mentioned earlier, the concept of telepresence is frequently encountered and notably controversial in both

the fields of telerobotics and virtual environments. It is a concept that seems to have arisen from the empirical

observation that users of teleoperation systems often have a harder time completing remote tasks – as simple

as parsing information or as complex as multi-part mechanical assembly – than the sum of changes in tools,

interfaces, sensory inputs, etc., appears to account for; there is a general sense that some element, a ‘being-

there-ness’, is missing. Approaches to defining telepresence beyond simply claiming its absence run the

gamut from philosophical [39] to social [34] to cognitive [12] to mechanical [32].

Many of the competing approaches to telepresence fall into one of two camps. The first stems from the as-

sumption that the teleoperation system’s goal is to be interfere as little as possible with the ideal case of local,

unmediated task performance. Telepresence is considered a consequence of realism, or ‘convincing-ness’,

and the telerobotic system is thought of as an obstacle to, or at best an imperfect medium for, transmitting

information. Transparency, the faithful reproduction of position, velocity, force, and other physical informa-

tion between the local and remote devices [24, 32, 50] is thus considered to be a critical goal, and definitions

of telepresence beyond quantifiable transparency are sometimes treated as questionable [12, 36].

The second treats the teleoperation system as a tool that must be suited to the current specific goal, taking into

account task characteristics, user characteristics, human sensory constraints, human motor constraints, and

the telerobotic and virtual environment characteristics. Telepresence is considered an experiential quality, the

user’s sense of mental and physical presence in the remote work, and to be best assayed through user reports

and questionnaires [60, 67, 76]. In this paradigm, telepresence will often, but not always, be a primary system

design goal, depending on the task at hand. Although telepresence’s causative relation to task performance

remains arguable, studies support the idea that both increase with the realism of the virtual environment [65].

The sensation of telepresence is likely an emergent phenomenon resting on many underlying perceptual and

cognitive factors; for instance, some have argued that ‘inverse’ telepresence, the false feeling that one is

in a mediated reality such as a film or a TV show, is an increasingly common experience due to the better

and wider distribution of movies, video games, and so forth having created a greater cultural familiarity

with telepresence [72]. Of the factors related to self-reported telepresence that have been rigorously studied,

a sense of spatial presence (i.e., feeling as if one is physically present in the remote environment) has been

identified as by far the strongest contributor [60]. Many authors have identified a general need for an objective

measure of a system’s ability to create a sense of telepresence; standardized subjective questionnaires [76],

140

tests of reflexive human response [19], and the ability of users to distinguish between ‘noisy’ real and virtual

environments [63] have all been proposed, but a generally accepted metric remains elusive – not surprising

given the lack of a generally accepted definition of the concept itself.

As telepresence is best recognized by its absence, virtualness can be defined as its positive converse, a mea-

sure of the consequences of the technological mediation between the user and the target environment. The

second approach to telepresence described above can be used to further extend this definition as an expe-

riential quality, the user’s awareness of the mental and physical links between themselves and the remote

work. Whereas telepresence is often spoken of as a binary state – users either achieve an experience of unity

between the local and remote environments, or don’t [22, 51] – there is an intuitively apparent spectrum of

virtualness, and likely multiple interacting spectra when multiple sensory modalities are engaged.

6.5 Model-Mediated Teleoperation

A relatively recent approach, intermediate between predictive control and supervisory control, has been pro-

posed to improve user task performance in haptic telerobotic systems with large time delays but slowly or

infrequently changing remote environments. This approach is model-mediated teleoperation [48]. In this im-

plementation of teleoperation (Figure 6.5), the user’s input as motion and forces is transmitted from the local

device to the remote device as usual. At the remote side, the device obeys the incoming commands while

simultaneously making use of sensor data (e.g., force sensors, rangefinders, cameras) to generate information

about the remote environment. As new data about the environment is learned, the updated information is

returned to the local side. This data is used to form a model of the remote environment, which may be as sim-

ple as a contact location or as complex as detailed estimates of shapes, stiffnesses, and dynamic properties.

This virtual ‘model’ environment is rendered at the local side haptically, and potentially graphically, audi-

bly, and in other ways, for immediate user interaction and feedback. When the expected rate or occurrence

frequency of remote environment change is significantly less than the system delay and update rate, users

are provided with an intuitive interface which, because of the local virtual model, allows them to generate

force and motion commands without feeling any delay-related damping or lag between local user action and

remote environmental reaction.

Model-mediated teleoperation is a user-centered approach. By virtue of the fact that it uses a virtual model

to address the problems of time delay, it is uniquely situated at the intersection of the study of both virtual

141

humanoperator

proxy location delay

remoteenvironment

modelenvironment

localdevice

remotedevice

delay

commandinterpretation

environmentinterpretation

Figure 6.5: In model-mediated telemanipulation, data from remote device-

environment interactions is used to create a local model of the remote environment.

The user interacts with this model, and the interactions are used to generate com-

mands for the remote device.

environments and telerobotic systems. The complete control the system designer has over the nature of the

virtual model and its presentation to the user offers great potential for improving time-delayed teleoperation,

following previous innovations such as creating perceptually larger workspaces in physically constrained

environments [7] and providing depth information in monoscopic displays [28, 43]. However, it also requires

a detailed understanding of user response to, and behavior in, virtual environments. Here two studies of

model-mediated teleoperation are presented, focused on methods of creating the virtual model based on

sensed input, handling model-environment mismatch, and processing the user’s input before applying it to

determine how these variables affect users’ ability to complete teleoperation tasks.

142

7 User Perception and Preference in Delayed

Model-Mediated Telemanipulation

7.1 Motivation

Model-mediated teleoperation is a user-centered approach to teleoperation first presented by P. Mitra in [48].

As shown in Figure 6.5, in a model-mediated system, user motion in the remote environment causes the

remote device to acquire new data regarding the objects it encounters. When this information arrives locally,

the model (virtual world) in which the user is operating may have to be updated; however, altering the

location and nature of virtual constraints currently being applied may generate abrupt changes both in the

graphical representation of the virtual world and in the force feedback supplied to the user. This may cause

loss of telepresence, undesirable reactions (startlement) in the user, and reduced task performance due to

discrepancies between the virtual display and the users expectations. Under certain circumstances – e.g.,

creating a new virtual surface at a position which the local device already occupies, resulting in a sudden,

strong applied force – system passivity will be violated, resulting in potential instability.

One proposed approach to mitigating the undesirable effects of model update under such circumstances is

the use of transition periods. During a transition, some level of error between the local and remote models

is tolerated, allowing users to become safely and intuitively accustomed to the updated environment. There

are many different possibilities for transition types: for instance, they may be subtle, fading in new forces

and images as the old are faded out; or explicit, giving the user a short force impulse or displaying animated

graphics to draw attention to a change. The choice of transition method will depend upon users preferences

and their ability to adapt to changing environment models. Of particular interest is the interaction of transition

method with stability and passivity. Modern haptic interfaces often ensure stability in the face of time delay by

providing passive feedback [1, 5, 35, 50]. Sudden changes to the virtual floor location may violate passivity

143

physical master &

virtual proxy

model floor physical floor

slave

one-second delay

Figure 7.1: The one degree of freedom (vertical) system virtual workspace used

in the user study, represented schematically and graphically. The master is repre-

sented in the virtual world by a proxy.

and hence bring stability into question, but from a user perspective, active responses may be acceptable and

even beneficial [30, 69]. Matching active responses to a user’s intuition appears to be important; previous

work indicates that users dislike unexpected sensory input [33].

P. Mitra and D. Gentry (the author of this thesis) designed and conducted a study of user performance and

preference for different transition methods in model-mediated teleoperation. The purpose of this user study

was to compare the effectiveness of three basic transition method philosophies: making transitions subtly

(e.g., as hidden as possible from the user), explicitly alerting the user to changes, or directly forcing changes

upon the user. Combinations of haptic and graphic transitions from each of these three categories was pre-

sented to a group of volunteers; physical data from user behavior was collected from the local device as well

as user preference rankings. Both data sets were analyzed and the conclusions are presented here.

7.2 User Study

7.2.1 Experimental Setup

To reduce variables in the system other than transition method, a one-degree-of-freedom teleoperation system

was constructed consisting purely of vertical motions and forces. The remote environment consists of a rigid

144

floor; since only vertical motion is considered, the remote device is either in free space or in contact with the

floor. The remote device senses the previously unknown floor location when it makes contact, or determines

that the initially sensed floor is gone when it passes the previous location without contact. This updated

location information is sent to the (physical) local device, where it is introduced to the user by means of a

model transition.

The user interface consisted of a SensAble Technologies ‘PHANToM’ haptic device [40] and a custom graph-

ical interface. In the graphical representation accompanying the haptic model (Figure 7.1), a sphere is used

to represent the current commanded location of the remote device; non-interfering ‘shelves’ mark the starting

height of the remote device between each study round; and the haptic floor’s location is represented by a solid

box below the sphere. When data from the remote device indicates that the estimated height of the floor has

changed, users are presented with the new virtual floor location using one of the transition methods described

below. The local device’s commanded position is tracked in the virtual environment by a mathematical proxy

to allow greater control over its collision response [47].

As in [52], to eliminate the possibility of variability in the physical behavior of a floor, the motion of the

remote device and location of the remote floor were simulated instead of using a physical robot and rigid

object. The position of the remote ‘device’ is calculated from the local device’s motion using information

regarding the simulated remote device’s dynamics and the programmed (but unknown to the model) location

and rigidity of the floor. All communication between the local and simulated remote devices was subjected

to a one second delay in each direction.

The goal of the user study was to evaluate and compare the three transition philosophies (subtle, explicit,

or forced) in both user preference and performance. An initial pilot study (subsection 7.2.2) with two male

college-age volunteers was conducted on 30 possible graphic-haptic transition methods was conducted, The

pilot test used the same model-mediated teleoperation system as the full user study. Each volunteer was asked

to complete 30 rounds, one for each graphic-haptic transition pair. In every round, the subject iterated three

times through all three possible floor locations (for a total of seven transitions) and gave a qualitative rating,

on a scale of 1 to 4, of preference to that iteration’s graphic-haptic transition method.

The pilot study yielded a set of six transitions to be tested in the full study. A sample set of twelve college-

age adults was tested. Each user performed six sets of tasks, one for each of the transition methods. Each

set comprised nine task rounds. A round began with the user near the top of the workspace, between the

two shelves shown in Figure 7.1. The task described to the user was to bring the remote device to rest

145

on the remote floor. At the beginning of each round, the remote floor location was set to one of three

randomly ordered values: high (near the maximum height of the workspace), low (near its minimum height),

or absent (outside the workspace); the model floor, as expected between continuous tasks, remained at the last

known location from the previous round. When the user decided that the task for the round was successfully

completed, they pressed a button to record their time and their estimate of the remote floor location (either

high, low, or absent). To reduce possible learning effects, each user was allowed a set of unrecorded training

rounds iterating through each transition type and floor location before beginning the recorded study. The order

in which the transition types were tested were structured for each user as in a randomized block design [44].

Four quantitative metrics were extracted from the haptic data to evaluate task completion performance: impact

velocity at the moment of remote contact with the floor, peak forces exerted by the remote device, time to task

completion, and remote device/floor penetration depth when the user indicated task completion (i.e., that the

local device was resting on the model floor). A fifth qualitative metric was recorded to gauge user preferences

for different transition methods: at the end of each completed nine-round set, the user was asked to give a

rating between 1 and 4 for the current transition method.

7.2.2 Transition Methods

The presence of the model in model-mediated teleoperation allows the system designer full control over the

haptic and graphic feedback, including transition methods, presented to the user. As described previously,

for purposes of this user study, three distinct approaches to presenting updates to the user were considered:

making transitions subtle to (intended to be unnoticed by) the user, explicitly alerting the user to a model

change, or directly forcing the user into the new model. Five possible graphic and and six possible haptic

transitions effects were initially identified:

Visual Transitions:

1. Jumping. The model graphic floor is removed from the old location or displayed at the new

location as soon as updated information indicates that its position has changed.

2. Tracking. When the new detected floor location is above the local device’s current location,

the model graphic floor is displayed just below the local device’s location until the device rises

above the remote floor’s new detected location, at which point the model floor becomes fixed

146

there. When the new detected floor location is below the local device’s current location, it is

immediately displayed. This transition avoids the illusion of the local device passing through

the floor.

3. Fading. The model graphic floor is faded in at the correct new location, and the old one faded

out, over a short period of time.

4. Flashing. The visual display is flashed white for a brief moment whenever model updates

occur, and then the model graphic floor is displayed at the new location.

5. Moving. The model graphic floor slides up or down from the old location to the new one at a

fixed velocity. This transition provides a continuous model floor presence.

Haptic Transitions:

(a) Tracking. Analogous to the tracking graphic transition. When the new detected floor location

is above the local device’s current location, the model haptic floor is continually placed just

below the device until the device rises past the new correct height; for a floor location detected

below the local device’s current location, the model haptic floor jumps immediately to the new

height.

(b) Moving. Analogous to the moving graphic transition. The virtual floor slides up or down from

the old location to the new one at a fixed velocity.

(c) Cautious. When the remote floor location is detected above the current local device’s location,

this method is like the tracking method. When the remote floor location is detected beneath

the current local device, the model haptic floor is only lowered when the user is not in contact.

This transition avoids sudden disappearances of the haptic floor.

(d) Pulsing. Analogous to the flash graphic transition. A short (10 ms) force pulse is generated

in the local device whenever an update regarding remote floor location is received; then the

tracking haptic transition is applied.

(e) Friction. When the remote floor location is detected above the current local device’s location,

this method is like the tracking method. When the remote floor location is detected beneath

147

the current local device, the model haptic floor is only lowered when the user pushes into it

with a certain minimum force, simulating Coulomb friction.

(f ) Stiffening. The model floor is immediately implemented at the new correct location; to pre-

serve stability, the stiffness of the new model floor is increased from zero over a short period

of time.

This yields a total of 30 possible graphic-haptic transition methods. Software, using freeglut and a haptics

interface library previously developed in-house, was written in C to implement all transition methods on a

PC running CentOS with the RTAI extension. This resulted in the virtual environment shown in Figure 7.1.

All transitions were designed to take 100 frames; running at 60 Hz, this is equivalent to 1.6 s.

A pilot test was run as described above to produce a refined list of transition methods for use in the full

user study. The pilot data (Figure 7.2) was used to select six transition methods from the original 30 while

ensuring that each of the three basic approaches (subtle changes, explicit alerts, or forced changes) was

represented. Because subtlety, to some degree, depends upon the subjective perception of the user, multiple

subtle transitions were included. The graphic methods flashing and jumping and the haptic method stiffening

were dropped entirely based on negative pilot user response. The haptic method cautious was highly ranked,

but dropped after users were unable to tell the difference between it and the simpler tracking transition. Using

the lettering and numbering scheme from the lists above, the final combinations used are shown in Table 7.1.

7.2.3 Results and Discussion

The pilot study (Figure 7.2) provided some initial insights that proved to hold consistent in the full user

study. Both pilot volunteers expressed a strong dislike for explicit transition methods, and commented that

methods which created a graphic/haptic mismatch in perception (for instance, feeling a floor where none was

displayed) were confusing. This is reflected in the rankings, as tested with the Newman-Keuls comparison

method. With a significance level of α = 0.05, the flashing graphic transition method was ranked lower than

all other methods, and the jumping method was ranking higher than flashing but lower than all remaining

methods; the other three methods were not statistically different. The haptic transition methods were less

clearly ordered; the stiffening transition was significantly lower than tracking, moving, and cautious, but the

others were not distinguishable.

148

1

1.5

2

2.5

3

3.5

4

Jumping Tracking Fading Flashing Moving

Me

an

Use

rR

an

kin

g

Graphic Transition Type

User Rankings by Graphic Transition, Pilot Study

1

1.5

2

2.5

3

3.5

4

Tracking Moving Cautious Pulsing Friction Stiffening

Me

an

Use

rR

an

kin

g

Haptic Transition Type

User Rankings by Haptic Transition, Pilot Study

Figure 7.2: Mean user rankings from the pilot study grouped by graphic and haptic

transition methods. Bars represent variances.

149

Table 7.1: The haptic and visual transition methods chosen for the user study.

Method Haptic Visual Transition Type

(a2) tracking tracking subtle

(a3) tracking fading subtle

(b2) moving tracking subtle

(b5) moving moving subtle

(d2) pulsing tracking explicit

(e2) friction tracking forced

The volunteers in the pilot study both commented that rounds in which the floor appeared above its last

known location were more difficult, confusing, or otherwise negative, so the pilot study data was also tested

for location (floor high, low, or absent) or location change (floor higher or lower). A one-way analysis of

variance (ANOVA) [44] showed that floor location was significant (p = 0.002). Cases in which the floor was

absent received overall lower ratings (mean 1.97, compared to 2.38 for low and 2.16 for high); a follow-up

Newman-Keuls means comparison showed that the absent case was significant compared to the low case (p =

0.001), but the high and low cases were not distinguishable. A two-tailed t-test indicated that location change

was significant (p = 0.002) as well. This last result contradicts the pilot users’ verbal comments, as cases in

which the floor was raised received a higher mean rating (2.27) than cases in which it was lowered (1.97).

As expected, a two-way ANOVA determined that both haptic and graphic transition methods had strong

effects on pilot user ranking (p < 0.001). Graphic and haptic methods interacted strongly as well (p = 0.006),

indicating that the full user study is correct to treat each graphic/haptic pair as a distinct method.

The full user study’s preference rankings cover a smaller range (Figure 7.3) than the pilot study. This is likely

a consequence of the removal of the lowest-ranked methods between the two studies. Nonetheless, a one-

way ANOVA still determined that transition method was a significant factor (p = 0.007). The Newman-Keuls

means comparison indicates that transition method b2 was preferred to methods e2, d2, and a2; no other

difference in preference was significant at a level of α = 0.05. Interestingly, all four distinguishable methods

used the tracking visual method, indicating that within that visual context, the moving haptic method was

preferred over tracking, pulsing, and friction.

The user study measured physical performance data as well as user preference. As described in subsection 7.2.1,

five task performance metrics were analyzed: success rate of user identification of floor location (absent, low

or high), remote device/floor impact velocity, peak remote force, task completion time, and remote device/

floor penetration depth.

150

1

1.5

2

2.5

3

3.5

4

a2 a3 b2 b5 d2 e2

Me

an

Use

rR

an

kin

g

Transition Method

User Rankings by Transition Method, User Study

Figure 7.3: Mean user rankings from the full user study grouped by methods. Bars

represent variances.

0

0.2

0.4

0.6

0.8

1

a2 a3 b2 b5 d2 e2

Fra

ctio

nC

orr

ectId

en

tifica

tio

n

Transition Method

Floor Location Identification Success by Transition Method

Figure 7.4: Mean success fraction at identifying the floor location (absent, low, or

high) from the full user study grouped by methods. Bars represent variances.

151

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

a2 a3 b2 b5 d2 e2

Imp

actV

elo

city

(ms

)

Transition Method

Floor Impact Velocity by Transition Method

Figure 7.5: Mean impact velocity from the full user study grouped by methods. Bars

represent variances.

The success rate of floor location identification was first tested with one-way ANOVAs using user identity,

transition method, floor location (high, low, or absent), and floor location change (higher or lower than the

previous round) as factors. Floor location change was not significant. Transition method was at borderline

significance (p = 0.05), with a2 and e2 slightly lower than the others (Figure 7.4). Floor location was sig-

nificant (p = 0.03); using the same follow-up means comparison, users performed better at identifying floor

location when the floor was absent (mean 97% correct) than when it was present (mean 92% correct) (p =

0.01), but the difference between the two present cases (high or low floor) was not significant. User identity

was also significant (p = 0.04), indicating that some users were, overall, less accurate than others.

Impact velocity is zero by definition for rounds in which the floor was absent, so these rounds were removed

and the same suite of tests run on the remaining data. Floor impact velocity was not affected by transition

method at α = 0.05 (Figure 7.5). Floor location was significant (p = 0.02); high floors had a greater mean

impact velocity (0.094 ms

) than low floors (0.077 ms

). Floor location change was of borderline significance

(p = 0.06), with rounds in which floors had been lowered having lesser impact velocity (0.076 ms

) than those

in which it had been raised (0.090 ms

). User identity was an extremely strong factor (p < 0.0001), indicating

that some users were consistently moving faster than others regardless of other factors.

152

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

absent low high

Imp

actV

elo

city

(ms

)

Floor Location

Floor Impact Velocity by Floor Location

Figure 7.6: Mean impact velocity from the full user study grouped by floor location.

Bars represent variances.

0

0.5

1

1.5

2

2.5

3

3.5

a2 a3 b2 b5 d2 e2

Pe

ak

Fo

rce

(N)

Transition Method

Peak Force by Transition Method

Figure 7.7: Mean peak force from the full user study grouped by methods. Bars

represent standard deviations.

153

0

0.5

1

1.5

2

2.5

3

3.5

absent low high

Pe

ak

Fo

rce

(N)

Floor Location

Peak Force by Floor Location

Figure 7.8: Mean peak force from the full user study grouped by floor location. Bars

represent standard deviations.

Peak force, like floor identification location, was analyzed first by one-way ANOVAs for user identity, tran-

sition method, floor location, and floor location change. Transition method was not significant at α = 0.05

(Figure 7.7). Floor location was highly significant (p = 0.0005); a follow-up Newman-Keuls means com-

parison indicates that rounds in which the floor was absent (1.17 N) had substantially lower peak force than

either the low (1.5 N) or high (1.4222 N) floor case, with the low and high cases not significantly different

(Figure 7.8). This likely reflects the frequent tendency of users to press into the perceived resistance of the

floor upon encountering it, although certain users demonstrated distinctly different reactions (see below).

Floor location change was also significant (p = 0.02), with rounds in which the floor was lowered having

greater peak force (1.44 N) than those in which it had been raised (1.27 N). Again, user identity was an ex-

tremely strong factor (p < 0.0001), indicating that some users were consistently hitting the floor harder than

others.

Remote device/floor penetration depth, like impact velocity, is only defined for rounds in which a floor is

present. Limited to this data, transition method, floor location and floor location change were not significant

at α = 0.05. User identity, however, was once more an extremely strong factor (p < 0.0001), indicating that

some users consistently drove deeper into the haptic floor than others before halting the search task.

154

0

5

10

15

20

absent low high

Ta

sk

Co

mp

letio

nT

ime

(s)

Floor Location

Task Completion Time by Floor Location

Figure 7.9: Task completion time from the full user study grouped by floor location.

Bars represent standard deviation.

Task completion time, by contrast, was strongly affected by both floor location and floor location change

(p < 0.0001). Absent floors resulted in the longest mean completion time (7.8 s), followed by low floors

(7.1 s) and then high floors (6.0 s); all three were significantly different at α = 0.05. Given the total span

of the workspace (0.25 m) and the typical speed of the user’s motion (estimated around 0.075 ms

from the

impact velocity data), each of these is well over twice the time required to simply move the distance to the

floor, indicating that the simple distance difference is not the only cause of the change. Additionally, floors

which had been lowered resulted in significantly faster mean completion times (6.3 s vs. 7.6 s), contrary to

what would be expected if the users were simply taking longer to explore a larger open space. Once again,

transition method was not a significant factor at α = 0.05, but user identity was a very strong effect (p <

0.0001).

The dependence on user identity of all tested metrics reflects an emergence of distinct task completion strate-

gies. At least three distinct strategies were observed. The first was the classic ‘move and wait’ approach [62],

in which a user makes a motion with the local device, then pauses and waits at least one round-trip delay to

receive updated information. Data from a user following this strategy can be seen in Figure 7.12, showing

consistent impact velocities arising from the user’s nearly constant velocity during move stages. The second

155

0

0.5

1

1.5

2

2.5

3

3.5

a2 a3 b2 b5 d2 e2

Ma

xim

um

Fo

rce

(N)

Transition Method

Maximum Force by Method

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

a2 a3 b2 b5 d2 e2Im

pa

ctV

elo

city

(ms

)

Transition Method

Impact Velocity by Method

0

5

10

15

20

a2 a3 b2 b5 d2 e2

Ta

sk

Co

mp

letio

nT

ime

(s)

Transition Method

Task Completion Time by Method

Figure 7.10: This user followed a very slow, stiff search strategy, resulting in con-

sistently high peak forces and low impact velocities. (3 N is the maximum reliable

recorded force.) The user’s average task completion time was over 1.5 times the

mean.

-0.5

0

0.5

1

1.5

2

2.5

3

3.5

a2 a3 b2 b5 d2 e2

Ma

xim

um

Fo

rce

(N)

Transition Method

Maximum Force by Method

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

a2 a3 b2 b5 d2 e2

Imp

actV

elo

city

(ms

)

Transition Method

Impact Velocity by Method

0

5

10

15

20

a2 a3 b2 b5 d2 e2

Ta

sk

Co

mp

letio

nT

ime

(s)

Transition Method

Task Completion Time by Method

Figure 7.11: This user followed a strategy of quick, light taps over the workspace,

resulting in wide variety in impact velocity, low peak forces, and low task completion

times.

-0.5

0

0.5

1

1.5

2

2.5

3

3.5

a2 a3 b2 b5 d2 e2

Ma

xim

um

Fo

rce

(N)

Transition Method

Maximum Force by Method

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

a2 a3 b2 b5 d2 e2

Imp

actV

elo

city

(ms

)

Transition Method

Impact Velocity by Method

0

5

10

15

20

a2 a3 b2 b5 d2 e2

Ta

sk

Co

mp

letio

nT

ime

(s)

Transition Method

Task Completion Time by Method

Figure 7.12: This user followed a ’move and wait’ strategy, resulting in relatively

consistent impact velocities and forces and relatively long task completion times.

156

-0.5

0

0.5

1

1.5

2

2.5

3

3.5

a2 a3 b2 b5 d2 e2

Ma

xim

um

Fo

rce

(N)

Transition Method

Maximum Force by Method

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

a2 a3 b2 b5 d2 e2

Imp

actV

elo

city

(ms

)

Transition Method

Impact Velocity by Method

0

5

10

15

20

a2 a3 b2 b5 d2 e2

Ta

sk

Co

mp

letio

nT

ime

(s)

Transition Method

Task Completion Time by Method

Figure 7.13: This user’s behavior was largely consistent across transition methods.

-0.5

0

0.5

1

1.5

2

2.5

3

3.5

a2 a3 b2 b5 d2 e2

Ma

xim

um

Fo

rce

(N)

Transition Method

Maximum Force by Method

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

a2 a3 b2 b5 d2 e2

Imp

actV

elo

city

(ms

)

Transition Method

Impact Velocity by Method

0

5

10

15

20

a2 a3 b2 b5 d2 e2

Ta

sk

Co

mp

letio

nT

ime

(s)

Transition Method

Task Completion Time by Method

Figure 7.14: This user’s behavior varied noticeably across transition methods.

strategy, shown in Figure 7.11, was to move from the top of the workspace to the bottom with quick repet-

itive taps. This strategy results in significant variation in impact velocity and peak forces, as both become

determined by at which part of the tap stroke the user encounters the floor, and low task completion times.

The third, and most distinct, strategy was a ‘slow and stiff’ search (Figure 7.10), in which the user descended

continuously from the top of the workspace at a very low velocity, resisting any forces exerted by the local

device, until the transition was complete. This strategy results in very low impact velocities, consistently high

force values (equivalent to the highest the haptic device is capable of producing), and long task completion

times.

For some users, transition method itself produced little variation in either strategy or performance. Con-

versely, other users showed marked differences in performance depending on transition method. Figure 7.13

shows one user who was relatively insensitive to transition method; Figure 7.14 shows one whose behavior

was significantly affected. Sensitivity to transition method itself appears to be a user-dependent effect.

7.3 Conclusions

User rankings were largely consistent between the pilot study and the full study. Both showed that users

disliked explicit or forced transition methods; e2 and d2 were both ranked very low. Dislike of methods with

low information content also appears to be a possible trend – in methods a2 and e2, which complete the set

157

of low-ranked methods, neither the graphics nor the haptics indicate the new final location of the floor (at

most, only a corrective motion in the new direction is perceived) until the transition is completed. Although

the effect is borderline, this may be related to the decrease in floor location identification success in these to

methods. The fact that all four methods with distinct rankings used the same visual transition is consistent

with previous work showing that visual input often overrides other senses in the presence of conflict.

Conversely, although previous studies indicated that graphic/haptic information mismatch is received poorly

by users, this was not observed here. The high-ranked b2 used the tracking visual method and the moving

haptic method, which will not agree at any point except the beginning and end state; the low-ranked a2

used tracking for both visual and haptic information, thus providing consistent information throughout the

transition. It appears that either other factors were of greater importance to users, or that the inconsistent

physical information was not perceived as inconsistent.

The imperfect match between user’s stated reactions and rankings is worth reinforcing in light of the latter

possibility. Users frequently made comments along the lines of “I don’t like this because it doesn’t do what

I expect it to;” however, a modern pool of users will have far more previous experience with the ‘rules’ of

common virtual worlds than with the real effects of haptic systems with time delay. This affects what they

perceive as realistic or reasonable in ways that system designers may not have an intuition for. More than one

user asked if the virtual floor could be made to disappear ‘more like in a video game,’ such as by changing

color and ‘blinking out,’ even though they correctly described it as a representation of a physically present

object.

This effect may be related to the fact that the two strategies observed to emerge here other than the classic

‘move and wait’ consist of ‘gaming the system.’ The ‘tapping’ strategy relies on moving in bursts shorter than

the round-trip time delay. This allows the user to trigger a visual transition which they can then passively

observe while out of range of the concordant haptic effect; the user never feels the effect of the remote

device first unexpectedly encountering the remote floor, but simply observes the corresponding visual update,

then moves the master device to the now-updated virtual floor. The ‘slow and stiff’ strategy exploits the

limitations of the local haptic device and the user’s assumed knowledge (based on their experimentation in

the first, unrecorded rounds) that nothing ‘bad’ happens if they force the virtual master through the virtual

floor, despite this being contrary to the task instructions.

There was little connection between user preference and any of the physical task metrics (impact velocity,

peak force, floor penetration and completion time). This is consistent with the observed tendency of most

158

users to use the same task completion strategies in all circumstances; faced with an unfamiliar task and setup,

users typically repeated their first successful behavior rather than continue to experiment to improve task

performance. Based on user feedback, at least part of the general dislike of ‘forced’ transition methods stems

from the way they force users away from their preferred strategies.

The strong dependence of task performance on floor location and floor location change was unexpected.

Floor location was highly relevant for all metrics except for floor penetration depth; floor location change was

relevant for impact velocity, peak force, and task completion time, as well as for user preference in the pilot

study. The former is a reminder that small variations in tasks can lead to large differences in performance.

The latter is a strong indication that, within the context of model-mediated teleoperation, different transition

methods may be appropriate for ‘positive’ model updates (adding a constraint) than for ‘negative’ model

updates (removing a constraint). A follow-up study evaluating this approach is highly recommended.

159

160

8 User Preference and Performance in

Manipulation of Virtual Environments

8.1 Motivation

Haptics and telerobotics aim to connect a user to a virtual or remote environment. In addition to whatever

force or motion information is typically relayed between a local device controlled by the human and a remote

device present in the remote environment, in most practical applications additional information from the

remote environment is also made available to the user, either directly (such as line-of-sight to the remote

location) or through a separate data gathering and presentation system (such as a TV feed). These extra

information channels may be direct, where the data is presented to the user as gathered (such as a video loop),

or there may be additional intermediate processing ranging from a graphical presentation of a virtual wall to

an internally generated intermediate model of the remote world [48]. This collection of information about

the remote environment, which may be real or virtual itself, is the equivalent of a local virtual environment

perceived by the user and with which she interacts.

Consider the ease with which humans perform everyday manipulation tasks when interacting directly with a

real environment. Such performance still far exceeds teleoperation capabilities, both real [36] and virtual [9].

For example, a human can typically perform the simple action of placing an fragile object gently on a table

so that it does not fall, tip or break without conscious effort. However, attempting a similar task with a

telerobotic system rapidly reveals shortcomings.

Proposed explanations for the persistent awkwardness of teleoperation tend to fall into two camps. One

focuses on the lesser amount of information available to the user [9, 16, 26, 61], such as loss of high fre-

quencies, point instead of multiple regions of contact, lack of depth perception, and other such changes. The

161

other focuses on changes in the user’s behavior [12, 22, 27, 36, 79], such as different use of various sensory

modalities (typically haptic and visual) or the increased cognitive load required to maintain an internal model

of the virtual world. These two effects interact; for example, a user finding the visual representation of an

object unconvincing may then perceive the object being handled as not ‘real’ and become more willing to

risk it or handle it roughly. This can lead to users adapting strategies that ‘game the system’ in teleoperation,

improving the apparent user experience while decreasing or circumventing the desired task metric [46].

The indirect, behavioral effects are difficult to isolate, in part because they may alter or shape the user’s

perceptual experiences [12, 79]. Telerobotic systems may also have non-obvious direct physical effects; for

instance, all linkages have preferred directions of motion, regardless of whether they are powered or providing

active feedback, and the user may be guided by this without realizing it.

Furthermore, human performance also greatly depends on task awareness and cognitive effort, and it has been

argued [12] that the user’s awareness of the virtual-ness of the environment may increase cognitive load even

when performing physically unaffected tasks. For unfamiliar actions, performance generally increases with

cognitive effort; however, for a highly familiar or normally unconscious action, such as balancing to reach for

a distant object, it is not uncommon for a person to encounter difficulty when asked to consciously perform

the same set of individual motions. While interacting with a telerobotic system is generally an unfamiliar

task and users are thus consciously focused on their physical movements, the effects of virtualization – the

introduction of awareness of the technology between the user and the remote environment – on unthinking

or reflexive user actions or action components, such as recovering from an off-balance state, may also be

present, and these effects may not be obvious.

The user study presented here explores the effect of different levels and types of virtualization on a user’s per-

formance, as gauged by tasks requiring both conscious and unconscious attention to precision and accuracy.

A simple task of tracing lines with one’s finger (“fingerpainting”) is introduced, and the tasks both of drawing

simple shapes and of making accurate marks are gradually virtualized by introducing a haptic device, virtual

paint, virtual surface, and indirect, dis-located visual feedback.

162

8.2 User Study

8.2.1 Experimental Setup

The criteria for task selection were that it be intuitive, manually simple, and easily virtualizable, while pro-

ducing measurable performance parameters, such as spatial accuracy and precision. The paradigm of finger-

painting (tracing shapes with one’s finger) was chosen, in addition to its suitability on these counts, for its

general familiarity. The experimental setup consisted of a haptic interface device suspended above a hori-

zonal display surface, a flat monitor screen overlaid with a thin piece of protective acrylic. An additional

monitor in the standard upright configuration was placed next to the flat surface. Users were asked to produce

a ‘painting’ consisting of simple shapes.

Four potential methods of ‘virtualization’ for this task were possible:

• Haptic. The task may be performed with an unaided hand, with the presence of a passive

haptic interface device, or with the same haptic interface device providing active feedback.

The haptic device is a SensAble Technologies “PHANToM” with thimble/gimbal interface,

allowing three-degrees-of-freedom active force feedback. To keep the tactile sensations com-

parable, an unattached thimble of the same size and weight is worn when painting unaided.

Both thimbles are covered with a piece of latex glove to provide protection from paint.

• Paint. The task may be performed using real, physical paint, or ‘virtual’ paint, created by

displaying a trace on a flatscreen monitor.

• Surface. The task may be performed using physical paper, on a physical surface that is not

paper (an acrylic overlay over the flat-screen monitor), or on a virtual surface, a virtual floor

generated by the haptic device.

• Visual. The task may be performed using direct visual feedback (watching the paint be applied

to the paper), indirect but co-located visual feedback (the ‘paint’ is displayed on the flat-

screen monitor flush beneath the painting surface), or dis-located visual feedback (the ‘paint’

is displayed on a second monitor separate from the painting surface).

163

Table 8.1: The haptic and visual methods of virtualization chosen for the user study.

Method Haptic Device Paint Type Surface Type Visual Feed

Method 1 none real real (paper) direct

Method 2 off real real (paper) direct

Method 3 off virtual real (acrylic) co-located

Method 4 on virtual virtual co-located

Method 5 off virtual real (acrylic) dis-located

Method 6 on virtual virtual dis-located

Preliminary tests were conducted to determine the details of implemention. For the physical drawings, sev-

eral types of paint and paper, including high-quality tracing paper (to allow a virtual painting to be shown

underneath at the same time) and wax paper (to more closely match the slipperiness of the acrylic and the

virtual surface) were tested; it was determined that densely mixed water-color paint and standard, acid-free

drawing paper (Strathmore 80 lb. 400 series) produced results most consistent between users. Different com-

binations of simple shapes were tested as well; the three-shape configuration shown in Figure 8.1, consisting

of square-circle-square, was chosen to allow a fair comparison with an explicit space-division task and so

that the effects of drawing a shape twice could be examined. All drawing surfaces and templates were 9 by

12 inches.

A small pilot study (4 users) was conducted to test the feasibility of combining the different methods of

virtualization. This resulted in the set of six virtualization methods shown in Table 8.1. They are numbered

in order from “least virtual” to “most virtual.”

The first method, Method 1, without any virtualization, asks users to paint on a horizontal surface with real

paint on real paper. As mentioned above, a thimble and glove are worn to assure tactile stimulation will

remain uniform across methods. Otherwise, the users perform the tasks entirely naturally.

In Method 2, the users use the haptic device, without active feedback, with physical paint and and a paper

surface. This method preserves the physical interactions of the system and adds the mechanical dynamics

and work space constraints of the haptic device.

Method 3 removes the real paper to reveal a horizontal flat panel display underneath. Using position in-

formation from the haptic device, virtual paint colors the screen on contact with the monitor, matching the

appearance of a real painting. No virtual haptic feedback is implemented and all physical interactions remain

passive.

164

In Method 4, a virtual surface is placed just above the monitor surface, such that contact forces are now

generated artificially. This necessarily alters the surface properties in both stiffness and friction such that

physical interaction experiences are changed.

Finally, Methods 5 and 6 add an alternate display to Methods 3 and 4. The virtual paint is visually rendered

on a vertical (standard) monitor placed behind the drawing surface, so that the (horizontal) drawing surface

and the (vertical) surface upon which the paint appears are separated but remain within the user’s field of

view. As such, Method 6 is the most virtualized environment, incorporating only artificial and dis-located

feedback.

Each volunteer was shown a template consisting of (left to right) a square, a circle, and a square, centered and

evenly spaced, and asked to draw the three shapes “as shown” twice for each method, once in their “natural

direction” and once in the “reversed” or mirrored direction. Typically, the natural motion was the three

shapes drawn left-to-right with each shape drawn clockwise, and the non-natural direction right-to-left with

each shape drawn counterclockwise; a minority of users preferred L-to-R counterclockwise and hence did R-

to-L clockwise. Regardless, each natural motion was L-to-R and each reversed R-to-L. This was designed to

remove any directional bias in the centering or sizing of the shapes when the directional results were averaged

during analysis. In both cases, users were given a reasonable time limit of one minute, but not informed that

equal spacing and sizing of the objects would be evaluated. This task is referred to hereinafter as the shape

task. A second follow-on task attempted to make the evaluation of geometric spacing and regularity explicit.

Users were asked to draw two vertical lines dividing the surface into thirds. This task is referred to hereinafter

as the division task. After completing all twelve 3-shape drawings, users then performed the division task six

times, once for each virtualization method.

Sixteen male subjects, ages 19–29, volunteered for the study and were tested. To attempt to reduce the effect

of individual user variation on the method results, the six methods were assigned to users as for a randomized

block design (RBD) [44]. Each volunteer was given the six methods as described above in random order,

and for each method, L-to-R or R-to-L was randomly chosen to be first. Once the initial shape tasks were

completed, the 6 division tasks were completed using the methods in the same order.

Before beginning the first method, users were given time to familiarize themselves with the haptic device

and the general setup. They were asked to push around on the different surfaces, including the virtual floor,

explore the limits of the workspace of the device (which did not quite cover the entire drawing surface

in the y-axis direction) and given the opportunity to ask questions. This was done to reduce the effects

165

Figure 8.1: The templates provided for the users for the 3-shape task (left) and

division task (right).

of inexperience and learning on the methods assigned earlier in the session. Most reported that they felt

immediately comfortable with and intuitively understood the setup and what they were being asked to do.

The template shapes shown to the user for both tasks are shown in Figure 8.1. For the shape task, users were

asked to draw the shapes “as shown”, but were not told that they would be measured or asked to take care with

accuracy. For the division task, users were explicitly asked to divide the paper into thirds, in addition to being

shown the template, but again were not given explicit explanation of what parameters would be measured. To

give a rough guideline of expectations, they were told that they should not spend more than about a minute

on each drawing. Almost all users came in well under that time for all tasks.

Data from the shape drawings was taken for analysis by determining the bounding boxes of each shape and

the bounding box of the drawing area as a whole. Figure 8.2 shows an example from Method 2 with the

measured bounding boxes displayed. The methods using paper produced physical drawings which were

physically measured using a transparent overlay; the methods using virtual surfaces produced pixel maps

which were processed in software. For Method 2, which produced both physical drawings and pixel maps,

the software results were used to generate the data after a comparison of both measurement methods for a

few sample results indicated that they were satisfactorily similar.

From this data, the following performance metrics were measured (see Figure 8.4 for variable definitions):

• Horizontal spacing inaccuracy. The magnitude of the deviation of the horizontal spacing from

an even spacing. For the shape task, the horizontal distances between all three shapes were

166

Figure 8.2: An example of a physical drawing produced by a user, with the bound-

ing boxes of each shape added to demonstrate how the drawings were measured.

Scanned image has been edited to obscure user-specific information.

100 200 300 400 500 600 700 800 900

100

200

300

400

500

600

Figure 8.3: An example of a virtual painting session. Axes marked are in pixels,

and were not visible to the user.

167

measured and the ratio between this measurement and an even division of the three shapes was

calculated (| 1

2 htot−hi|12 htot

). For the division task, this is simply the magnitude of the deviation of

the two marks from an even one-third division (not shown).

• Vertical offset inaccuracy. The magnitude of the deviation of the vertical spacing between

the three shapes from a flat line. The magnitude of the vertical distance between the centers

of the left shape and middle shape and the centers of the middle shape and right shape were

measured; this value was then divided by the average height of all three shapes to reduce

per-user effects ( viw

).

• Shape skew. The deviation from “square-ness” of each shape. The difference between the

height and width of each shape was measured, and this measurement divided by the height of

the shape to reduce per-user effects ( li−wili

).

• Shape scaling. The deviation from the provided template size of each shape. The size (height

times width) of each shape was measured, and this measurement was divided by the actual

size of the template shapes ( liwiLW

).

An additional metric was calculated by blind scoring (that is, without knowledge of which paintings corre-

sponded to which virtualization parameters):

• Errors. The number of “obvious mistakes” made in a drawing: overlapping shapes, the wrong

shape in the wrong place, shapes that run off the edge of the page, and shapes that are sig-

nificantly discontinuous. For instance, the drawing in Figure 8.3 had an error score of 2 (one

significant discontinuity, one runoff).

8.2.2 Results and Discussion

Three possible questions can be addressed by the collected data. The first is whether particular types of

virtualization (haptic, visual, etc.) have stronger effects on overall task performance than others. The second

is whether virtualization has stronger effects on certain performance parameters than others. These first two

questions were determined by one-way analysis of variance (ANOVA) [44]. The third question is whether

168

100 200 300 400 500 600 700 800 900

100

200

300

400

500

600

l3

w3

htot

h1

h2

v1

v2

Figure 8.4: An example of a virtual drawing with the measured parameters illus-

trated.

certain combinations of types of virtualization had a multiplicative effect on task performance (in other words,

whether the combination of any two virtualization types resulted in a greater change than could be accounted

for by the addition of the previously determined effects of each type); this was determined by examination of

the interaction term in two-way ANOVA between each combination of virtualization type. Unless otherwise

noted, the threshold for statistical significance was taken to be α = 0.05. Additionally, all metrics discussed

below were tested for per-user effects (whether, for that metric, certain users consistently had stronger or

weaker responses than others); these effects are noted where they were found.

Horizontal spacing inaccuracy, when analyzed by method, increased more rapidly with increasing virtualiza-

tion for the division (explicit) task than for the shape (implicit) task (Figure 8.5, top); this effect is significant.

However, the overall difference in mean inaccuracy between the two tasks is highly significant (p < 0.001),

likely reflecting a true difference in task difficulty – for instance, that users had less difficulty reproducing

an existing, visually marked distance than mentally estimating two equal spaces before marking either. This

indicates that the choice of followup task was not sufficiently comparable to the main task to allow further

inference between the two. The shape and division tasks were then analyzed separately by virtualization type

(Figure 8.5, bottom). For the shape task, only the effect of the dislocated visual feedback was significant. For

the division task, the presence of the active haptic device, virtual writing surface, and dislocated visual feed

were all significant factors; it is interesting that the ‘intermediate’ virtual stages (passive haptic device, clear

surface overlay, co-located but indirect visual feed) showed no effect in either case. Per-user effects were

significant for the division task, and the per-user effects interacted significantly with the location of the visual

feed as well (p < 0.01).

169

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

M 1 M 2 M 3 M 4 M 5 M 6Me

an

Ho

rizo

nta

lS

pa

cin

gD

evia

tio

n

Method

Horizontal Deviation by Method

ShapeDivision

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

Haptic Paint Surface Visual DirectionMe

an

Ho

rizo

nta

lS

pa

cin

gD

evia

tio

n

Virtualization Type

Horizontal Deviation by Virtualization Type, Shape Task

nat.

unnat.direct

col.

dis.

paperacrylic

virtual

real

virtualnooff

on

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

0.16

0.18

Haptic Paint Surface VisualMe

an

Ho

rizo

nta

lS

pa

cin

gD

evia

tio

n

Virtualization Type

Horizontal Deviation by Virtualization Type, Division Task

directcol.

dis.

paperacrylic

virtual

real

virtual

nooff

on

Figure 8.5: Data for horizontal spacing inaccuracy, by method (t) and by virtualiza-

tion factor (b) for the shape (l) and division (r) tasks. Bars indicate variances.

170

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

M 1 M 2 M 3 M 4 M 5 M 6

Ve

rtic

alO

ffse

tF

acto

r

Method

Vertical Offset Magnitude by Method

MeanLeftRight

0

0.02

0.04

0.06

0.08

0.1

0.12

0.14

Haptic Paint Surface Visual Direction

Me

an

Ve

rtic

alO

ffse

tF

acto

r

Virtualization Type

Mean Vertical Offset Magnitude by Virtualization Type

nat.

unnat.

directcol.

dis.

paper

acrylic

virtual

real

virtual

no

offon

Figure 8.6: Data for vertical offset, by method (t) and by virtualization factor (b).

Bars indicate variances.

171

Vertical offset inaccuracy (Figure 8.6) shows a less clear trend, although Methods 5 and 6 are significantly

less accurate than Methods 1–4. The offset was calculated for each pair of shapes separately to determine

whether this was a contributing factor; as it was not, the overall mean was used to examine the effects of

each virtualization type. As with horizontal inaccuracy, location of visual feed was the only virtualization

type with a significant effect. Per-user effects were significant; interestingly, per-user interaction effects were

significant for the presence and activity of the haptic device, but not for the location of the visual feed.

Shape skew (Figure 8.7) was a much more highly variable metric. In nearly all cases, shapes were ‘skinny’

(positive skew, meaning width was less than height). The metric was examined by each shape to determine

if placing (center/edge) or shape (square/circle) played a role, but neither was a significant factor; the mean

was then used to examine the effects of method and virtualization type. The passive presence of the haptic

device, and the presence of the acrylic overlay, were both associated with much skinnier figures; interestingly,

neither the active haptic device nor the dislocated visual feed caused a similar effect. Per-user effects were

highly significant (p < 0.001); this is the only metric for which per-user interactions were observed with

every virtualization type (though not with direction), and all were at a level of p < 0.001, indicating that

shape skew and the way it is affected by changes in task environment are both highly idiosyncratic.

Shape scaling (Figure 8.8) showed nearly the opposite behavior, as it was highly consistent. Shapes got

smaller as the level of virtualization increased; this effect is highly statistically significant for Methods 3–6 (p

< 0.002) and consistent across all three shapes. All virtualization types had a statistically significant negative

effect (direction did not). This is also the only metric in which virtualization type interactions were observed,

between the haptic and visual virtualization types (p < 0.03). Significant per-user effects were observed (p

< 0.001), and interacted only with direction and visual location.

The last metric examined was the occurrence of errors (Figure 8.9). This metric shows a weakly increasing

trend with virtualization, although since the occurrence of errors, overall, was very low (less than one per task)

and this user study is relatively small, the variance is high. Nonetheless, all four virtualization types showed

a strong and statistically significant effect (p < 0.02); in all cases the ‘least virtual’ type implementation was

associated with the lowest occurrence of errors, indicating that the presence of the haptic device (whether

active or passive), the presence of a floor above the work surface (whether physical or virtual), and the pres-

ence of indirect visual feedback (whether co-located or dis-located) all led to poorer task performance. The

higher error rate for ‘passive’ virtualization (presence of haptic device, clear paper overlay, collocated visual

display) strongly suggests that users found virtualization in some way distracting even when not introducing

major physical changes to the task; the user simply became more likely to make “stupid mistakes” (in the

172

-0.1

-0.05

0

0.05

0.1

M 1 M 2 M 3 M 4 M 5 M 6

Sh

ap

eS

ca

ling

Fa

cto

r

Method

Shape Skew Factor by Method

MeanLeft

MiddleRight

-0.1

-0.05

0

0.05

0.1

Haptic Paint Surface Visual Direction

Me

an

Skew

Fa

cto

r

Virtualization Type

Mean Skew Factor by Virtualization Type

nat.unnat.

directcol.

dis.paper

acrylic

virtualreal

virtualno

off

on

Figure 8.7: Data for skew, by method (t) and by virtualization factor (b). Bars

indicate variances.

173

-0.4

-0.2

0

0.2

0.4

M 1 M 2 M 3 M 4 M 5 M 6

Sh

ap

eS

ca

ling

Fa

cto

r

Method

Shape Scaling Factor by Method

MeanLeft

MiddleRight

-0.4

-0.2

0

0.2

0.4

Haptic Paint Surface Visual Direction

Me

an

Sh

ap

eS

ca

ling

Fa

cto

r

Virtualization Type

Mean Shape Scaling Factor by Virtualization Type

nat.unnat.

direct

col.

dis.

paper

acrylicvirtual

real

virtualnooff

on

Figure 8.8: Data for scaling, by method (t) and by virtualization factor (b). Bars

indicate variances.

174

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

M 1 M 2 M 3 M 4 M 5 M 6

Nu

mb

er

of

Err

ors

Method

Number of Errors by Method

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

Haptic Paint Surface Visual Direction

Nu

mb

er

of

Err

ors

Virtualization Type

Number of Errors by Virtualization Type

nat.

unnat.

direct

col.dis.

paper

acrylic

virtual

real

virtual

no

off

on

Figure 8.9: Data for errors, by method (t) and by virtualization factor (b). Bars

indicate variances.

175

Figure 8.10: One of the physical drawings produced by a user, showing an ‘un-

forced error’: the locations of the squares and circles are switched. Scanned image

has been edited to obscure user-specific information.

words of one volunteer, responding to the error shown in Figure 8.10) unless they made an effort beyond what

seemed necessary to focus. Per-user effects and per-user interactions with visual location were also observed.

In addition to the above statistical results, a few qualitative differences of note also became apparent during

the study. Consistent with the strong dependence of shape skew and scaling on user identity, many users had

a human-recognizable drawing ‘style’, to the point where it was possible to identify most (label-redacted)

drawings done by that user from the general drawing pool. For example, one user whose drawing is shown

in Figure 8.11 was consistently too small, too crowded and too far to one side or the other; another consis-

tently overestimated the first margin (i.e., began too far to the right to fit three shapes of the size of the first

square when going L-to-R, and too far to the left when going R-to-L). This ‘first margin’ error, dependent

on direction, was noted among a number of users. Another clear trend was that drawn squares that were

close to accurate (low skew, no errors) were generally too tall (positive skew), but attempted squares that

were not (high skew, one or more errors) were generally too wide (negative skew); the effect can be seen in

Figure 8.12. This may be where the preferred direction of motion of the haptic device came into play.

As a final note, user reaction to the virtual floor (implemented with simple PD control) varied significantly.

Some were unable to tell that it was present without peering underneath to verify that they were not actually

in contact with the acrylic; others said that it didn’t feel real, with the most common complaint being that it

was ‘slipperier’ than either the paper or the acrylic.

176

100 200 300 400 500 600 700 800 900

100

200

300

400

500

600

100 200 300 400 500 600 700 800 900

100

200

300

400

500

600

Figure 8.11: Certain behaviors, such as scaling, underestimating spacing, and

underestimating margins, were consistent across users and relatively unaffected by

methods. (l) a virtual painting session (Method 5, L-to-R) demonstrating significant

recentering and scaling; (r) a virtual painting session (Method 3, L-to-R) showing

the tendency of users to set their first margin too wide.

100 200 300 400 500 600 700 800 900

100

200

300

400

500

600

100 200 300 400 500 600 700 800 900

100

200

300

400

500

600

Figure 8.12: A comparison between two virtual painting sessions using Method 3,

showing correlation between skew and direction of scaling. (l) Method 3, R-to-L,

shapes are neat, but too tall; Method 5, L-to-R, shapes are messy and too wide.

177

8.3 Conclusions

Returning to the three originally posed questions, the first, whether certain types of virtualization have

stronger task performance effects that others, can be answered immediately. Visual display location has

an overwhelmingly larger effect than any other form of virtualization. This holds true for every parameter

examined except for shape skew, where the collocated visual display decreased task performance but not the

dislocated display. Interestingly, for error rate, both the collocated and dislocated displays were associated

with decreased performance, but for the remaining three tasks, only the dislocated display was statistically

significant. The only other virtualization type with a significant effect was the haptic device, whether active or

not; it is a much weaker second factor affecting many, but not all, performance measures, and the significance

of its effect was not consistent between users.

The second question was whether virtualization has stronger effects on certain performance parameters than

others, or, put another way, how robust the task performance studied here metrics are. Error rate and shape

scale were both sensitive to all virtualization factors, while skew was sensitive towards none. Horizontal and

vertical inaccuracy were both affected only by visual display location. Only horizontal spacing inaccuracy

showed robustness across different users. Vertical offset inaccuracy and error rate showed user dependence,

but for both metrics it was the weakest of the statistically significant variables; for shape skew and scale,

which user was performing the task had a greater effect on the metric than any virtualization type. Interest-

ingly, there was an overall lack of effects due to the order or type of shape and direction.

The third question regarded the presence of multiplicative, or ‘interaction’ in a statistical sense, effects in the

presence of multiple virtualization types. Surprisingly, these were essentially non-existent, appearing only

in the shape scaling data for the combination of haptic and visual virtualization. This indicates that the task

performance effects of different virtualization types are essentially independent from each other. However,

interaction effects with user identity were significant for at least one virtualization type for all metrics other

than horizontal inaccuracy, indicating that user response to virtualization types is highly personal. User/

factor interaction effects were observed slightly more often with visual and haptic virtualization types than

with surface or paint types.

Several further conclusions can be drawn from this work. The sensitivity of the various metrics to different

types of virtualization provides guidance on matching tasks to telerobotic system design. Changes in visual

display clearly dominate the other forms of virtualization studied here; every effort should be made to ensure

that visual display location is at least co-oriented with the haptic workspace. The particular sensitivity of the

178

Table 8.2: A summary of the effects of the different virtualization factors on each

measured metric. User dependency indicates the degree to which the metric was

affected by the particular user rather than the test methods. User/factor interactions

lists those virtualization types for which the effect was significantly greater for some

users than for others.

Metric Significant Factors User Dependency User/Factor Interactions

horizontal inaccuracy visual none none

vertical inaccuracy visual moderate haptic, surface

skew none strong haptic, paint, surface, visual

scaling all strong haptic, visual

errors all moderate visual

horizontal spacing and vertical offset metrics to visual display location indicates that system designs where

visual dislocation is unavoidable are less suitable for tasks which require precise, repeated horizontal and

vertical object placement. Similarly, the high sensitivity of scale means that tasks which require consistent

scaling activities must be carefully adapted to the type of virtualization introduced and tested to ensure the

final motion scale is as desired. Conversely, the general insensitivity of shape skew to virtualization indicates

that tasks which are sensitive to skew, such as drawing, writing, or spatial alignment tasks, are promising

candidates for teleoperation.

The strongest takeaway from the data is the high level of user effects and particularly the presence of user/

factor interaction effects. This clearly reinforces the significance of user idiosyncrasy in telerobotic task

performance. As the study here was limited to less than an hour of use, the persistence of these per-user

effects and their relation to the ‘new-ness’ of the telerobotic setup is unclear; this, and whether the per-user

variance can be mitigated with training, warrants additional investigation. Another implication of the high

degree of user-dependent effects is the importance of establishing the sensitivity of one’s metrics before using

them to choose or alter telerobotic system design. If only scale, or spacing accuracy, had been measured here,

or the same battery of metrics tested with a smaller user base, incorrect assumptions about the suitability of

various types of virtualization for this task would have been drawn.

The final lesson from this work is that there is a clear increase in general error rate when any form of virtual-

ization is introduced to this type of task. Virtualization begins to affect the user’s actions with the introduction

of a haptic or visual device, even when no information has yet been removed from the environment, and there

is a general increase in error rates when multiple types of virtualization are combined. These trends sup-

port the idea that a user’s unconscious coordination and focus are diminished in the presence of a virtual

environment. Whether the effect holds true for different task types is a promising subject for follow-on work.

179

180

9 Conclusions

9.1 Task Performance and System Design

The two user studies presented here clearly show that idiosyncratic user preference and strategy development

is at least as large a factor in successful virtual task performance, including model-mediated teleoperation, as

traditional informational system metrics such as fidelity of graphic or haptic transmission. Per-user effects, as

a category of factor, are frequently not tested or reported in the teleoperation literature. However, across both

user studies presented here, they were significant for every tested metric but one, with horizontal spacing

inaccuracy the only exception. This has important implications for telerobotic human-in-the-loop system

design, system testing, and system implementation.

In the area of telerobotic design, there are three key takeaways from this work. Firstly, a system must be

matched to the task the users will be expected to perform. Consistent with previous work, both studies indi-

cated the overriding importance of visual information, the first study in terms of user preference, the second in

terms of user behavior; thus, wherever possible, visual display location should be, at a minimum, intuitively

aligned with the haptic workspace. Although some degree of virtualization will always be unavoidable, sys-

tem designers may have the ability to prioritize realism in certain areas and should be aware of which areas are

task-critical. For instance, previous work has shown that users tend to significantly underestimate distances

in virtual workspaces; this is consistent with the finding here that scaling becomes strongly negative with

increasing virtualization of all studied types. In a task where accurate scaling is important, then, more em-

phasis on distance-related training, cues, and realistic behavior is warranted than in a task primarily sensitive

to skew, which the work here shows to be generally robust to virtualization.

Secondly, there is a wealth of potential in system personalization. Given that users have starkly different

preferences and reactions to physical system parameters, the option to allow ‘user preferences’ similar to

181

those seen in a typical software-only application may well lead to improved user experiences, if not quantita-

tive performance metrics. User interface variables such as visual display location, virtual surface firmness or

smoothness, vibration magnitude or frequency, and even (to use model-mediated teleoperation as an example)

model transition methods are well within a system designer’s ability to leave as user-configurable variables

without altering the teleoperation system’s underlying hardware or communications.

Thirdly, realism and virtualism are both questions of consistency between perception and expectation, but

both perception and expectation will vary between users, tasks, and systems. It is clear from the first user

study that users’ expectations of the behavior of a given system may or may not correspond to measured

physical behavior. Several users perceived the virtual floor and device in the first study as less real because

they behaved less like a video game, and video games formed their internal frame of reference for how a

‘real’ virtual environment should behave. This lead to an assumed lack of consequences for the equivalent

of ‘breaking the game’ by (in the users’ view) exploiting the effects of time delay, device strength, etc.

Similarly, it is clear from the second study that users’ perception of the system’s properties, such as its size,

can be significantly altered by seemingly minor changes such as the presence of an inactive haptic device.

This leads to the second set of implications, telerobotic system testing. The obvious approach of choosing

metrics that directly reflect the desired task, such as placement accuracy, or measure information fidelity,

such as transparency, may not be sufficient; such metrics may not be consistent across tasks, users, and im-

plementations. Both studies strongly drive home this point; only one metric (horizontal spacing inaccuracy)

across both studies did not show statistically significant user dependence, and for most metrics, user identity

was the strongest statistical factor. Although only the second user study tested for interactions between user

identity and other factors, they were present for all metrics but horizontal spacing, indicating that a small user

study may produce counter-representative results. The sensitivity of one’s metrics should be established with

thorough user testing. A good metric for evaluating task performance on a system intended for multiple users

should be relatively insensitive to user effects.

The final set of implications are those that concern system implementation, including task instruction and user

training. The first study showed that task instruction alone (‘the floor and object to be placed are fragile’) will

not counteract a contradictory apparent system behavior (‘the display shows nothing bad happening when I

punch through the floor, and the haptic device doesn’t exert forces large enough to stop me’). Further, a user’s

first experience with a new system and task appears to be highly likely to ‘set’ a pattern of task strategies that

the user later resists altering; both systems and user training procedures should be designed to naturally elicit

the desired behavior wherever possible.

182

9.2 User Preferences and Behavior

Compounding the issue, both user studies presented here indicate that user preference is, at best, only partly

correlated to task performance. The information on user preference captured in both studies is broadly consis-

tent with previous work in the telerobotics literature – the first showing that users dislike abrupt or unexpected

sensory input, the second showing emergence of non-continual strategies by one second of delay and later

adherence to these strategies – placed in the context of virtual environments, where general user tendencies

for weighting of visual information, distance underestimation, and so on are well established.

In light of users’ comments, ratings, and task performance metrics, a few general trends can be described.

Firstly, it appears that users prefer that at least some explicit information about changes to the environment,

including model updates in model-mediated teleoperation, be presented. Secondly, the users appear to prefer

this explicit information to be presented in way that allows them to look for it (including as part of simply

watching the virtual environment) and choose how to respond, rather than having visual or haptic constraints

forced on them.

This preference for control, or at least the illusion of it, is also apparent in the first user study due to the

alignment between the duration of each transition in the first user study and the round-trip time delay. Many

users adopted a behavior of reacting to the beginning of a transition by moving the master to a neutral position

and waiting for the transition to complete, in effect creating a ‘triggered’ move-and-wait strategy. Given that

model-mediated teleoperation is designed for systems where the remote environment changes infrequently

relative to the communication time delay, such a strategy effectively functions to suspend and restore the

user’s illusion of working in a non-delayed state.

The parameters that determine the emergence of such task completion strategies are not well understood.

Here, it was shown that at least three distinct strategies emerged; the well-studied ‘move and wait’, and the

newly described ‘tapping’ and ‘slow and stiff’ behaviors. The two new behaviors reported here are clearly

responsive to the particulars of the virtual environment used: the ‘tapping’ strategy to the time delay and

model mediation, ‘slow and stiff’ to the limitations of the haptic device. This raises the possibility of using

the virtual environment to ‘nudge’ users towards particular strategies considered desirable given the task at

hand. Whether such an approach can overcome users’ generally strong adherence to previously adopted

strategies is an important question: one hypothesis for the dislike of insistent or constraining information

described above is that it tends to force users away from their preferred behaviors. Additionally, as the

183

studies here were limited to an hour or less of use of a new system, the persistence of such training effects

will need to be characterized.

As usual, the question of the importance of telepresence, or virtualness, remains controversial. The second

study here showed that many metrics decreased even in the presence of ‘passive’ virtualization, such as

the presence of an inactive haptic device or a visual display co-located with the workspace; these trends

support the idea that a user’s unconscious coordination and focus are diminished in the presence of a virtual

environment. It remains to be seen, however, whether this difference can be explained by the effect of split

or differently focused attention or has some other cause. Additionally, although the second study showed

that the effect of the different sensory modalities of virtualization appear additive, with a general absence

of significant interactions, user preference for types of sensory information itself may shift with context;

the first study demonstrated that users may complain about perceived mismatches between haptic and visual

information when they are consistent, yet perceive no mismatch when they are inconsistent. This reinforces

the difference between matching a user’s expectation about the behavior of the virtual world and matching

the veridical physical behavior of the remote world.

The control over the user’s perceived work environment – the model – allowed by model-mediated teleopera-

tion is critical to both matching transition type to type of update and for adapting task presentation to suit user

particularities, the two major conclusions presented here. These results underline the promise and versatility

of model-mediated teleoperation.

9.3 Future Work

There are a number of avenues for future investigation suggested by this work.

The strongest conclusion apparent from both studies presented here is the sensitivity to user identity of a

great many metrics, such as object impact velocity, vertical placement accuracy, and horizontal motion skew,

that might otherwise be obvious choices for measuring task performance. A standard criterion for metric

robustness across users, perhaps defined through a set of tests and analyses, would be extremely useful; one

of the reasons for the endurance of transparency as a criterion for telerobotic system design, as opposed to the

more controversial telepresence, is its clear definition and ability to be analytically verified. A clear corollary

to the same conclusion, again demonstrated in both studies, is the sensitivity of many task performance

184

metrics to apparently small changes in setup. For example, rounds in which the floor was absent or present,

or rounds in which it had been raised or lowered compared to the previous round, showed a strong effect on

the measured task criteria in the first study, both unforeseen effects. A clear definition for robustness to this

kind of effect, including guidance on designing tests to quantify it, is also a worthwhile goal.

The correlation, if any, between task performance and user preference also deserves to be examined in light of

these two results. The first user study asked volunteers to rank the transitions after a completed set of rounds,

rather than after each round; based on the results of the pilot data analysis, in which finer-grained preferences

were recorded, a follow-up study examining the effect of small task differences (e.g., floor height) on user

preference is clearly indicated. Given the robustness of horizontal spacing accuracy to user identity, a task

relying primarily on this metric is a strong potential candidate.

As mentioned earlier, both user studies support the idea that adding the capacity for user personalization –

the ability to adjust things such as display location, spatial scaling, and surface ‘feel’ – to a telerobotic system

is likely to improve the user experience and potentially reduce the system’s perceived virtualness. A study

evaluating this possibility is strongly recommended.

There is a general need of better characterization of the relationship between different types of virtualization,

time delay, task requirements, and emergent user strategy. Better data is particularly needed on the possibility

of using the virtual environment to guide users into discovering and adopting desired strategies. The studies

here were not able to address the emergence of behaviors in a controlled way, nor, by using repeated sessions

or other methods, address the duration of learning or learned behaviors’ persistence. A follow-on research

effort to answer these questions is clearly justified.

The final area for short-term future work indicated here is the root cause of the changes in user behavior seen

in virtual environments. This work provides some support for the idea that the user’s overall ability to focus

on a given task diminishes in the presence of virtualization. Although the explanation that the awareness

of the technological mediation causes split cognitive effort is appealing, a more rigorous study addressing

the question is necessary. As many other, more complex tasks require a great deal more unconscious effort

than the one studied here, such as balancing upright, keeping oneself centered in a changing viewfield, or

maintaining a constant grip on an unsteady object, a similar approach to this but focused around these or

similar abilities is one clear possibility.

One area for longer-term future work indicated here is the relationship between user preference and the

cognitive effort demands described above. The first study presented here showed an complex relationship

185

between user preference and performance in need of further clarification. The second demonstrated the high

sensitivity of user task performance to differences in user perception. A follow-up study using a system

like the second to address a question like the first is highly recommended. By unifying the conclusions of

the first and second studies presented here, such a work would present a clear and compelling study of user

perception, preference, and performance in teleoperation.

186

References

[1] Anderson, R. and Spong, M. Bilateral control of teleoperators with time delay. IEEE Transactions on

Automatic Control, 34(5):494–501, 1989.

[2] Bejczy, A., Kim, W., and Venema, S. The phantom robot: predictive displays for teleoperation with time

delay. International Conference on Robotics and Automation, 1:546–551, 1990. doi:10.1109/ROBOT.

1990.126037.

[3] Brooks, T. Telerobotic response requirements. Systems, 1990.

[4] Ching, H. and Book, W. Internet-based bilateral teleoperation based on wave variable with adaptive

predictor and direct drift control. Journal of Dynamic Systems, Measurement, and Control, 128(1):86–

93, 2006.

[5] Chopra, N. and Spong, M. Adaptive coordination control of bilateral teleoperators with time delay.

IEEE Conference on Decision and Control, pp. 4540–4547, 2004.

[6] Citerin, J., Pocheville, A., and Kheddar, A. A touch rendering device in a virtual environment with

kinesthetic and thermal feedback. Proceedings of the 2006 IEEE International Conference on Robotics

and Automation, pp. 3923–3928, 2006.

[7] Conti, F. and Khatib, O. Spanning large workspaces using small haptic devices. First Joint Eurohaptics

Conference and Symposium on Haptic Interfaces for Virtual Environment and Teleoperator Systems,

2005.

[8] Daniel, R. and McAree, P. Fundamental limits of performance for force reflecting teleoperation. The

International Journal of Robotics Research, 17(8):811–830, 1998. doi:10.1177/027836499801700801.

[9] De Boeck, J., Cuppens, E., Weyer, T., Raymaekers, C., and Coninx, K. Multisensory interaction

metaphors with haptics and proprioception in virtual environments. Third Nordic Conference on

Human-Computer Interaction, pp. 189–197, 2004. doi:10.1145/1028014.1028043.

187

[10] De Boeck, J., Raymaekers, C., and Coninx, K. Exploiting proprioception to improve haptic interaction

in a virtual environment. Presence, 15(6):627–636, 2006.

[11] DiFranco, D., Beauregard, G., and Srinivasan, M. The effect of auditory cues on the haptic perception

of stiffness in virtual environments. Proceedings of the ASME Dynamic Systems and Control Division,

61:17–22, 1997.

[12] Draper, J. and Blair, L. Workload, flow, and telepresence during teleoperation. International Conference

on Robotics and Automation, pp. 1030–1035, 1996.

[13] Ernst, M. and Banks, M. Humans integrate visual and haptic information in a statistically optimal

fashion. Nature, 415(6870):429–433, 2002.

[14] Ernst, M. and Bulthoff, H. Merging the senses into a robust percept. Trends in Cognitive Sciences,

8(4):162–169, 2004. doi:10.1016/j.tics.2004.02.002.

[15] Gat, E., Desai, R., Ivlev, R., Loch, J., and Miller, D. Behavior control for robotic exploration of planetary

surfaces. IEEE Transactions on Robotics and Automation, 10(4):490–503, 1994.

[16] Gu, J., Augirre, E., and Cohen, P. An augmented-reality interface for telerobotic applications. Sixth

IEEE Workshop on Applications of Computer Vision, 2002.

[17] Hashtrudi-Zaad, K. and Salcudean, S. Analysis and evaluation of stability and performance robustness

for teleoperation control architectures. International Conference on Robotics and Automation, 2000.

[18] Helbig, H. and Ernst, M. Knowledge about a common source can promote visual-haptic integration.

Perception, 36(10):1523–1534, 2007. doi:10.1068/p5851.

[19] Held, R. and Durlach, N. Telepresence, time delay and adaptation. In Pictorial Communication in

Virtual and Real Environments, pp. 232–246. Taylor & Francis, 1991.

[20] Heller, M., Calcaterra, J., Green, S., and Brown, L. Intersensory conflict between vision and touch:

The response modality dominates when precise, attention-riveting judgments are required. Perception

& Psychophysics, 61(7):1384–1398, 1999.

[21] Hershberger, W. and Misceo, G. Touch dominates haptic estimates of discordant visual-haptic size.

Perception & Psychophysics, 58(7):1124–1132, 1996.

[22] Hirche, S., Bauer, A., and Buss, M. Transparency of haptic telepresence systems with constant time

delay. Conference on Control Applications, pp. 328–333, 2005.

[23] Hoffmann, E. Fitts’ law with transmission delay. Ergonomics, 35(1):37–48, 1992.

[24] Hokayem, P. and Spong, M. Bilateral teleoperation: An historical survey. Automatica, 42(12):2035–

2057, 2006. doi:10.1016/j.automatica.2006.06.027.

[25] Interrante, V., Ries, B., Lindquist, J., and Anderson, L. Elucidating factors that can facilitate veridical

188

spatial perception in immersive virtual environments. IEEE Proceedings of Virtual Reality, 2007. doi:

10.1.1.110.8929.

[26] Kahol, K., McDaniel, T., and Panchanathan, S. Methodology for efficient perception in exclusively hap-

tic environments. International Workshop on Haptic Audio Visual Environments and their Applications,

pp. 140–145, 2006.

[27] Kammermeier, P. and Buss, M. A human perception model for multi-modal feedback in telepresence

systems. International Conference on Systems, Man, and Cybernetics, 4:1020–1025, 1999.

[28] Kim, W., Tendick, F., and Stark, L. Visual enhancements in pick-and-place tasks: Human operators

controlling a simulated cylindrical manipulator. Journal of Robotics and Automation, 3(5):418–425,

1987.

[29] Klatzky, R. and Lederman, S. How well can we encode spatial layout from sparse kinesthetic contact?

Haptic Interfaces for Virtual Environment and Teleoperator Systems, 2003.

[30] Kuchenbecker, K., Fiene, J., and Niemeyer, G. Improving contact realism through event-based haptic

feedback. IEEE Transactions on Visualization and Computer Graphics, 12(2):219–230, 2006. doi:

10.1109/TVCG.2006.32.

[31] Lawn, C. and Hannaford, B. Performance testing of passive communication and control in teleoperation

with time delay. Proceedings of the IEEE International Conference on Robotics and Automation, pp.

776–783, 1993.

[32] Lawrence, D. Stability and transparency in bilateral teleoperation. IEEE Transactions on Robotics and

Automation, 9(5):624–637, 1993. doi:10.1109/70.258054.

[33] Lecuyer, A., Megard, C., Burkhardt, J., Lim, T., Coquillart, S., Coiffet, P., and Graux, L. The effect of

haptic, visual and auditory feedback on an insertion task on a 2-screen workbench. Proceedings of the

Immersive Projection Technology Symposium, 2002.

[34] Lombard, M. and Ditton, T. At the heart of it all: The concept of presence. Journal of Computer-

Mediated Communication, 3, 2006. doi:10.1111/j.1083-6101.1997.tb00072.x.

[35] Lozano, R., Chopra, N., and Spong, M. Passivation of force reflecting bilateral teleoperators with time

varying delay. Proceedings of the Mechatronics Forum, pp. 954–962, 2002.

[36] Lumelsky, V. On human performance in telerobotics. Transactions on Systems, Man and Cybernetics,

21(5):971–982, 1991.

[37] MacKenzie, I. and Ware, C. Lag as a determinant of human performance in interactive systems. Pro-

ceedings of the INTERACT ’93 and CHI ’93 Conference on Human Factors in Computing Systems, pp.

488–493, 1993. doi:10.1145/169059.169431.

189

[38] Madhani, A., Niemeyer, G., and Salisbury Jr., J. The black falcon: A teleoperated surgical instrument for

minimally invasive surgery. International Conference on Intelligent Robots and Systems, pp. 936–944,

1998.

[39] Mantovani, G. and Riva, G. “Real” presence: how different ontologies generate different criteria for

presence, telepresence, and virtual presence. Presence, 8(5):540–550, 1999.

[40] Massie, T. and Salisbury, J. The PHANTOM haptic interface: A device for probing virtual objects. Pro-

ceedings of the ASME Winter Annual Meeting, Symposium on Haptic Interfaces for Virtual Environment

and Teleoperator Systems, 55(1):295–300, 1994.

[41] Massimino, M. Improved force perception through sensory substitution. Control Engineering Practice,

3(2):215–222, 1995.

[42] Massimino, M. and Sheridan, T. Variable force and visual feedback effects on teleoperator man/machine

performance. Proceedings of the NASA Conference on Space Telerobotics, 1:89–98, 1989.

[43] McKnight, S., Melder, N., Barrow, A., Harwin, W., and Wann, J. Perceptual cues for orientation in a

two finger haptic grasp task. Proceedings of the First Joint Eurohaptics Conference and Symposium on

Haptic Interfaces for Virtual Environment and Teleoperator Systems, pp. 549–550, 2005.

[44] Mendenhall, W. and Sincich, T. Statistics for Engineering and the Sciences. Prentice Hall, 1995.

[45] Misceo, G. and Taylor, N. A “unity assumption” does not promote intersensory integration. Experi-

mental Psychology, 2011.

[46] Mitra, P., Gentry, D., and Niemeyer, G. User perception and preference in model mediated telemanipu-

lation. Second Joint EuroHaptics Conference and Symposium on Haptic Interfaces for Virtual Environ-

ment and Teleoperator Systems (World Haptics 2007), pp. 268–273, 2007. doi:10.1109/WHC.2007.122.

[47] Mitra, P. and Niemeyer, G. Dynamic proxy objects in haptic simulations. Conference on Robotics,

Automation and Mechatronics, 2:1054–1059, 2004.

[48] Mitra, P. and Niemeyer, G. Model-mediated telemanipulation. The International Journal of Robotics

Research, 27(2):253–262, 2008. doi:10.1177/0278364907084590.

[49] Mohler, B., Creem-Regehr, S., and Thompson, W. The influence of feedback on egocentric distance

judgments in real and virtual environments. Proceedings of the 3rd Symposium on Applied Perception

in Graphics and Visualization, pp. 9–14, 2006. doi:10.1145/1140491.1140493.

[50] Niemeyer, G. and Slotine, J. Stable adaptive teleoperation. Journal of Oceanic Engineering, 16:152–

162, 1991.

[51] Niemeyer, G. and Slotine, J. Telemanipulation with time delays. The International Journal of Robotics

Research, 23:873–890, 2004. doi:10.1177/0278364904045563.

190

[52] Petzold, B., Zaeh, M., Faerber, B., Deml, B., Egermeier, H., Schilp, J., and Clarke, S. A study on

visual, auditory, and haptic feedback for assembly tasks. Presence, 13(1):16–21, 2004. doi:10.1162/

105474604774048207.

[53] Ranadive, V. Video Resolution, Frame Rate and Grayscale Tradeoffs under Limited Bandwidth for

Undersea Teleoperation. Ph.D. thesis, Massachusetts Institute of Technology, 1980.

[54] Rank, M., Shi, Z., and Hirche, S. Perception of delay in haptic telepresence systems. Presence,

19(5):389–399, 2010.

[55] Richard, P., Burdea, G., Gomez, D., and Coiffet, P. A comparison of haptic, visual and auditive force

feedback for deformable virtual objects. Proceedings of the International Conference on Automation

Technology, pp. 49–62, 1994.

[56] Richard, P. and Coiffet, P. Human perceptual issues in virtual environments: sensory substitution and

information redundancy. IEEE International Workshop on Robot and Human Communiction, pp. 301–

306, 1995. doi:10.1109/ROMAN.1995.531976.

[57] Richardson, A. and Waller, D. The effect of feedback training on distance estimation in virtual environ-

ments. Applied Cognitive Psychology, 19(8):1089–1108, 2005. doi:10.1002/acp.1140.

[58] Rosenborough, J. Aiding Human Operators with State Estimates. Ph.D. thesis, Massachusetts Institute

of Technology, 1988.

[59] Rust, I. and Asada, H. The eyeball ROV: Design and control of a spherical underwater vehicle steered

by an internal eccentric mass. 2011 IEEE International Conference on Robotics and Automation, pp.

5855–5862, 2011.

[60] Schubert, T., Friedmann, F., and Regenbrecht, H. The experience of presence: Factor analytic insights.

Presence, 10(3):266–281, 2001.

[61] Sheridan, T. Telerobotics, automation, and human supervisory control. The MIT Press, 1992.

[62] Sheridan, T. Space teleoperation through time delay: review and prognosis. IEEE Transactions on

Robotics and Automation, 9(5):592–606, 1993.

[63] Sheridan, T. Further musings on the psychophysics of presence. Presence, 2:1073–1077, 1994.

[64] Sheridan, T. and Ferrell, W. Remote manipulative control with transmission delay. IEEE Transactions

on Human Factors in Electronics, 4(1):25–29, 1963.

[65] Slater, M. and Wilbur, S. A framework for immersive virtual environments (FIVE): Speculations on the

role of presence in virtual environments. Presence, 6(6):603–616, 1997.

[66] Srinivasan, M., Beauregard, G., and Brock, D. The impact of visual information on the haptic perception

of stiffness in virtual environments. Proceedings of the ASME Dynamic Systems and Control Division,

191

58:555–559, 1996.

[67] Stanney, K., Mourant, R., and Kennedy, R. Human factors issues in virtual environments: A review of

the literature. Presence, 7(4):327–351, 1998.

[68] Steuer, J. Defining virtual reality: Dimensions determining telepresence. Journal of Communication,

42(4):73–93, 1992.

[69] Tanner, N. and Niemeyer, G. High-frequency acceleration feedback in wave variable telerobotics. IEEE

Transactions on Mechatronics, 11(2):119–127, 2006.

[70] Taylor, R., Jensen, P., Whitcomb, L., Barnes, A., Kumar, R., Stoianovici, D., Gupta, P., Wang, Z.,

Dejuan, E., and Kavoussi, L. A steady-hand robotic system for microsurgical augmentation. The

International Journal of Robotics Research, 18(12):1201–1210, 1999.

[71] Thompson, W., Willemsen, P., Gooch, A., Creem-Regehr, S., Loomis, J., and Beall, A. Does the quality

of the computer graphics matter when judging distances in visually immersive environments? Presence,

13(5):560–571, 2004.

[72] Timmins, L. and Lombard, M. When “real” seems mediated: Inverse presence. Presence, 14(4):492–

500, 2005.

[73] Waller, D. and Richardson, A. Correcting distance estimates by interacting with immersive virtual

environments: effects of task and available sensory information. Journal of Experimental Psychology,

14(1):61–72, 2008. doi:10.1037/1076-898X.14.1.61.

[74] Wang, Y. and MacKenzie, C. The role of contextual haptic and visual constraints on object manipulation

in virtual environments. CHI Letters, 2(1):532–539, 2000.

[75] Welch, R., Widawski, M., Harrington, J., and Warren, D. An examination of the relationship between

visual capture and prism adaptation. Perception & Psychophysics, 25(2):126–132, 1979.

[76] Witmer, B. and Singer, M. Measuring presence in virtual environments: A presence questionnaire.

Presence: Teleoperators and virtual environments, 7(3):225–240, 1998.

[77] Witten, I. and Knudsen, E. Why seeing is believing: merging auditory and visual worlds. Neuron,

48(3):489–496, 2005. doi:10.1016/j.neuron.2005.10.020.

[78] Wu, W., Basdogan, C., and Srinivasan, M. Visual, haptic, and bimodal perception of size and stiffness

in virtual environments. Proceedings of the ASME Dynamic Systems and Control Divison, 67:19–26,

1999.

[79] Wuillemin, D., Doorn, G., Richardson, B., and Symmons, M. Haptic and visual size judgements in

virtual and real environments. First Joint Eurohaptics Conference and Symposium on Haptic Interfaces

for Virtual Environment and Teleoperator Systems, 2005.

192

[80] Zung, B., Butter, E., and Cashdan, S. Visual-haptic form recognition with task delay and sequenced

bimodal input. Neuropsychologia, 12(1):73–81, 1974.

193