A New Electrode Technology for Neuro-Robotic Interfaces Alik Sunil Widge CMU-RI-TR-07-03

Self-Assembled Monolayers of Polythiophene “Molecular Wires”: A New Electrode Technology for Neuro-Robotic Interfaces Alik Sunil Widge CMU-RI-TR-07-03

Submitted in partial fulﬁllment of the requirements for the degree of Doctor of Philosophy in Robotics

The Robotics Institute and Center for the Neural Basis of Cognition Carnegie Mellon University Pittsburgh, Pennsylvania 15213

January 3rd, 2007

Dr. Yoky Matsuoka, Thesis Committee Chair Dr. Kaigham Gabriel, Co-advisor Dr. Xinyan ”Tracy” Cui, University of Pittsburgh (external) Dr. Carl Lagenaur, University of Pittsburgh (external) Dr. Lee Weiss

Abstract

This thesis presents the proof of concept of a new type of electrode for interfaces between living nervous systems and electronic devices (“neuro-robotic interfaces”). Such interfaces have long been pursued due to their high clinical and scientific value. However, progress has been hindered by inade- quate performance of the implanted electrodes that bridge biological, ion- based electricity and analog/digital electronics. These electrodes provoke inflammatory reactions in surrounding tissue and often cause detrimental effects when stimulating neurons to deliver information. The most promising approach to improving electrode biocompatibility involves electrically polymerizing conductive polymers with biomolecules on metal electrodes. These coatings improve mechanical biocompatibility, attract neurons to the electrode, and lower electrode impedance. They are nevertheless limited by delamination of polymer from the electrode and inability to pattern or control the composition of the polymer/biomolecule blend. The new electrode technology described herein addresses those limitations through two innovations: the use of self-assembled monolayer (SAM) technology to bind polythiophene conductive polymers to metal electrodes, and the design of lipophilic polythiophenes (“molecular wires”) that can insert into a cell membrane and provide stable intracellular electrical access. Thiol-based SAMs should increase coating robustness and will permit better controllability and patterning, while still offering the same biocompatibility as electropolymerized coatings. Intracellular access that does not kill the target neuron will permit gentler and more specific stimulation and higher fidelity recording. Polythiophene SAMs are thinner than electrodeposited polymer films, and thus do not decrease impedance to the same degree, but should be able to compensate for this by allowing intracellular stimulation. Data from atomic force microscopy, cell culture, and impedance spectroscopy are combined to show that functionalized polythiophenes can form SAMs and that these SAMs have the appropriate biological and electrical activity for a neuro-robotic interface electrode. The feasibility of polymer-based intracellular electrophysiology is demonstrated through artificial lipid bilayer experiments and atomistic molecular dynamics simulations of the membrane-polymer interface. Taken together, these studies constitute proof of concept for the “molecular wire” SAM electrode and represent the first steps towards development and deployment of this new interface technology.

2 Acknowledgements

All science is ultimately a collaborative effort, but this thesis in particular has depended on a small army of collaborators, assistants, friends, mentors, and fellow travelers. Naming them all would take a section that dwarfs the remainder of the document, so I will instead begin by apologizing to all those not named, either due to space limitations or the imperfections of human memory. The list of thanks is headed by my advisors, Drs. Yoky Matsuoka and Ken Gabriel, for being supportive of the direction I chose and for assisting me in making the connections necessary to access the appropriate materials and equipment. Credit is also due to the Robotics Institute as a whole, for encouraging interdisciplinary research that ranges into fields far removed from traditional robotics, and for taking better care of its students than any other graduate department I have ever encountered. Beyond that general gratitude, I must acknowledge that the majority of my work was conducted in other investigators’ laboratories, and without their generous donations of equipment, materials, and advice, none of this could have been done. Dr. Victor Weedn, formerly of Carnegie Mellon, provided the seed ideas that led to this project and arranged for laboratory space within the Molec- ular Biotechnology & Imaging Center. Dr. Malika Jeffries-El, also formerly of Carnegie Mellon and now at the University of Iowa, synthesized or supervised the synthesis of all the polymers used in these experiments, devoting a substantial percentage of her effort over four years with minimal rec- ompense. The atomic force microscopy of Chapter 4 and the thermal evaporations of Chapters 4-6 all used equipment in the laboratory of Dr. James Schneider of Carnegie Mellon. Dr. Schneider also provided critical advice on experimental approaches and interpretation of the results. Every- thing I know about neuronal cell culture, and certainly everything presented in Chapter 5, is due to instruction and guidance from Dr. Carl Lagenaur of the University of Pittsburgh, who also made his laboratory available for that work. Similarly, my knowledge of impedance spectroscopy and understanding of electrochemistry is largely due to Dr. Xinyan “Tracy” Cui, also of the University of Pittsburgh, who provided the equipment and software used in the experiments of Chapter 6. The artificial lipid bilayer work of Chapter 7 arose from discussions with Dr. Schneider and was performed at the National Institute of Standards and Technology, using the equipment and voluminous technical assistance of Drs. John Kasianowicz and Martin Misakian. In Dr. Kasianowicz’s case, this generosity extended to opening his own home to me as a place to stay while the experiments were underway. The results of the bilayer experiments would be of greatly reduced significance if not for the accompanying molecular dynamics simulations described in Chapters 8 and 9. These relied on expert guidance from Dr. Maria Kurnikova of Carnegie Mellon, as well as tens of thousands of hours of computer time using her personal resources and her grant allocations from the Pittsburgh Super- computing Center. Finally, credit is due to several undergraduate research assistants who mastered toxic chemicals, designed new equipment, researched new techniques, and traveled across the length and breadth of Pittsburgh to move this work forward. Chief among these are Kelly Collins, Ololade Olakanmi, Megan Tzeng, and Mengyao “Mona” Zhe. Work such as this involves many experimental and logistical barriers, and I faced many periods of deep frustration. In these times, I was lucky enough to have colleagues and organizations who helped me to vent that frustration, channel it into positive advocacy for my fellow students and future patients, and to grow as a person and a leader along the way. Again, with my deepest apologies to all those who are not listed here, I would like to single out a few from that multitude:

3 • The representatives and leaders of the Carnegie Mellon Graduate Student Assembly, especially Matt Cronin, Brian Fifarek, Kim Murday, Rob Reeder, and Miriam Rosenberg-Lee. Without their early support and mentorship, much of the rest of this list would not have happened. • The National Association of Graduate-Professional Students, and particularly Doris Dirks, Jim Masterson, and Serge Egelman. Although my work with NAGPS introduced me to only a small part of the Federal legislative process, I came away with a new understanding of the interaction between science, education, and politics. • The American Medical Association and its Medical Student Section, where, through the assistance of many patient and tolerant friends and mentors, I managed to polish away some of my rougher edges and learn to be a more effective leader. My year as AMA-MSS Chair unquestionably slowed my scientific progress, but has left me with skills and friendships that will affect my career for decades. Special thanks are due to Adam Gordon, Parag Parekh, Sadeq Quraishi, and David Winchester for their guidance, and to Anne Christensen, Chris DeRienzo, Ben Galper, Michael Katz, Brad Lancaster, Steve Sherick, Heather Smith, and Hannah Zimmerman for working and fighting by my side throughout these five years. • Mary Moore and Suzanne Lyons Muth, two Carnegie Mellon staff whose doors were always open and whose desks were always stocked with candy when I needed administrative assistance, advice, or simply someone to talk to. Without professionals like Suzanne and Mary, the Robotics Institute would grind to a halt in short order. • My colleagues in the Medical Scientist Training Program, especially Pedram Afshar, Lou Ghanem, Audrey Lau, Rod Tan, and Ron Trible. Their advice, fellowship, and trailblazing has helped me clear many administrative barriers and reminded me that even M.D./Ph.D. training does eventually end.

The financial support for this work is as heterogeneous as the technical and moral support. My tuition and stipend were funded in the first year by the Center for the Neural Basis of Cognition (NIH institutional training grant T32 N507433-03), in years two through four by a National Defense Science & Engineering Fellowship funded by the Army and administered by the American Society for Engineering Education, and subsequently by an Ruth L. Kirchstein Individual National Research Service Award fellowship (NIH training grant F30 NS051866). The costs of the research were borne in large part by my advisors and collaborators, but we also received seed funding from the American Medical Association Foundation and from the Paralyzed Veterans of America Spinal Cord Research Foundation (grant SCRF 2275-01). The computational studies of Chapters 8-9 were made possible by grants to Drs. Matsuoka and Kurnikova of supercomputer time on the Cray XT3 and Terascale Computing System of the Pittsburgh Supercomputing Center. At an even more fundamental level of support, I thank my parents, Drs. Sunil and Toeruna Widge, for the years of education and support that led me to this point, and my father in particular for sharing with me advice and “war stories” from his own Ph.D. studies. In the final year and a half of this work, I have also been blessed with the love and companionship of a beautiful and intelligent woman, Ms. Jennifer Rauktis. She has endured the unpredictable schedule of laboratory work, a year of regular travel, and hours of conversations in arcane jargon. Her unwavering support and belief in me, along with her baking skills, provided the fuel and motivation to bring these studies to a close. Finally, and most importantly, I thank God for bringing me into contact with all of the people named and not named above, and for guiding me to both my many obstacles and the means to cross over them. There is something special and sacred in even the smallest act of discovery, and I have been privileged to experience that many times over the course of this work.

4 Contents

1 Introduction 15 1.1 Brief Rationale ...... 15 1.1.1 What are Neuro-Robotic Interfaces? ...... 15 1.1.2 Clinical Value of Neuro-Robotic Interfaces ...... 15 1.1.3 Scientiﬁc Value of Neuro-Robotic Interfaces ...... 16 1.2 Overview of This Thesis ...... 17

2 Background 19 2.1 Review of Current Neuro-Robotic Interface Approaches ...... 19 2.1.1 Nerve Cuﬀs ...... 20 2.1.2 Sieve Electrodes ...... 21 2.1.3 Microelectrode Arrays ...... 23 2.2 Limitations of Current Interface Technologies ...... 24 2.2.1 General Trends ...... 24 2.2.2 Proposed Mechanisms Underlying Limitations ...... 25 2.3 Known Techniques for Countering Electrode Limitations ...... 26 2.3.1 Increasing Material Biocompatibility ...... 27 2.3.2 Increasing Speciﬁcity and Decreasing Current/Charge Injection ...... 30 2.3.3 Countering Probe Micromotion ...... 33 2.4 Summary ...... 35

3 Details and Rationale for Proposed Technology 37 3.1 Self-Assembly of Conductive Polymers ...... 38 3.2 Lipophilic Polythiophenes for Intracellular Access ...... 38 3.3 Why Polythiophene? ...... 39 3.4 Roadmap of Experiments ...... 40

4 Nanostructure of EHPT Self-Assembled Monolayers 43 4.1 Methods and Materials ...... 43 4.1.1 Atomically Flat Gold Substrates for Atomic Force Microscopy ...... 44 4.1.2 Formation of Self-Assembled Monolayers on Gold Surfaces ...... 45 4.1.3 Protein Coupling into Self-Assembled Monolayers ...... 46

5 4.1.4 Tapping Mode Atomic Force Microscopy ...... 46 4.1.5 Exploratory SPM Techniques ...... 47 4.2 Results ...... 47 4.2.1 Tapping Mode AFM ...... 47 4.2.2 Self-Assembly of Non-Thiolated EHPT ...... 50 4.2.3 Exploration of Non-Tapping AFM Techniques ...... 50 4.3 Discussion ...... 53

5 Biocompatibility of Protein-Containing EHPT SAMs 57 5.1 Introduction ...... 57 5.2 Methods ...... 57 5.2.1 Translucent Gold-Coated Culture Substrates ...... 58 5.2.2 Formation of Self-Assembled Monolayers ...... 59 5.2.3 Protein Coupling into Self-Assembled Monolayers ...... 59 5.2.4 Primary Neuron Culture on SAMs ...... 59 5.2.5 Immunostaining and Cell Counting ...... 60 5.3 Results ...... 61 5.4 Discussion ...... 61

6 EHPT-Containing SAMs Lower the Impedance of Coated Electrodes 65 6.1 Introduction ...... 65 6.2 Methods ...... 65 6.2.1 EcoMEA Preparation for SAM Formation ...... 65 6.2.2 SAM Formation on Gold-Coated EcoMEAs ...... 66 6.2.3 Impedance Measurements ...... 67 6.3 Results ...... 67 6.3.1 Eﬀects of Pure and Mixed SAMs on MEA Impedance ...... 67 6.3.2 Removal and Re-Formation of EHPT SAMs ...... 69 6.3.3 Lack of SAM Formation With Unthiolated EHPT ...... 69 6.4 Discussion ...... 71

7 EHPT Raises the Electrical Conductance of Artificial Lipid Bilayers 73 7.1 Introduction ...... 73 7.2 Methods ...... 73 7.2.1 Creation and Measurement of Artificial Lipid Bilayers ...... 73 7.2.2 Introduction of PTs to Artificial Bilayers ...... 75 7.3 Results ...... 75 7.3.1 EHPT Increases BLM Conductance ...... 75 7.3.2 HPT Does Not Alter BLM Conductance ...... 77 7.3.3 Conductance Changes Caused By EHPT Diminish Over Time ...... 77 7.4 Discussion ...... 77

6 8 Development of Parameters for Polythiophene Molecular Modeling 79 8.1 Introduction ...... 79 8.2 Methods ...... 80 8.2.1 Ab initio Parameter Development ...... 80 8.2.2 Replica Exchange MD in Implicit Solvents ...... 83 8.3 Results ...... 84 8.3.1 Geometry and Partial Charges ...... 84 8.3.2 Rigid Rotor Potential in AMBER and GAUSSIAN ...... 86 8.3.3 Replica Exchange MD In Implicit Solvents ...... 88 8.4 Discussion ...... 96

9 Molecular Dynamics of Polythiophenes In Lipid Bilayers 99 9.1 Introduction ...... 99 9.2 Methods ...... 99 9.2.1 Simulation Parameters and Coordinates ...... 99 9.2.2 Steered Molecular Dynamics ...... 101 9.2.3 Umbrella Sampling ...... 102 9.3 Results ...... 103 9.3.1 Steered Molecular Dynamics: Reaction Path and Initial Energy Estimates . . 103 9.3.2 Umbrella Sampling: Accurate Potential of Mean Force ...... 109 9.4 Discussion ...... 116

10 Conclusions and Future Work 119 10.1 Summary of Results ...... 119 10.2 Future Directions ...... 120 10.2.1 Critical Path for Technology Development ...... 120 10.2.2 Mechanisms, Tools, and Expanded Techniques ...... 122 10.3 Final Thoughts ...... 126

Glossary 127

References Cited 131

7 8 List of Figures

1.1 First patient to receive a Cyberkinetics BrainGate implant ...... 16 1.2 Rehabilitation Institute of Chicago prosthetic neuro-robotic arms ...... 16 1.3 Schematic of proposed neuro-robotic interface ...... 17

2.1 Standard cylindrical nerve cuff ...... 21 2.2 Flat Interface Nerve Electrode (reshaping cuff) ...... 21 2.3 Schematic of sieve (regeneration) electrode ...... 22 2.4 Photograph of sieve (regeneration) electrode ...... 22 2.5 “Michigan” style neural probes ...... 22 2.6 Utah Intracortical Electrode Array ...... 22 2.7 Slanted (peripheral) Utah microelectrode array ...... 24 2.8 Schematic of microfluidic neural probe ...... 24 2.9 SEM of electrode before polypyrrole ...... 29 2.10 SEM of electrode after polypyrrole ...... 29 2.11 Cell growth on polypyrrole-coated electrodes ...... 29 2.12 Repeat units of polythiophene, polypyrrole, and poly(3,4-ethylenedioxythiophene) . 31 2.13 Schematic of “patch clamp” technique ...... 34 2.14 Photograph of patch clamping ...... 34 2.15 Schematic of patch clamp microarray ...... 34 2.16 Ratchet mechanism for Sandia mobile neural probe ...... 35 2.17 Sandia mobile neural probe ...... 35

3.1 Schematic of proposed neuro-robotic interface, reprised from Figure 1.3 ...... 37 3.2 Poly(alkylthiophene) repeat unit ...... 40 3.3 Illustration of regioregular polymerization ...... 40

4.1 Assembly schematic for SAM-based neuron/electrode interface ...... 44 4.2 AFM images: Bare gold, height ...... 48 4.3 AFM images: Bare gold, phase ...... 48 4.4 AFM images: EHPT SAM, height ...... 48 4.5 AFM images: EHPT SAM, phase ...... 48 4.6 AFM images: MHA SAM, height ...... 48

9 4.7 AFM images: MHA SAM, phase ...... 48 4.8 AFM images: 1:1 EHPT:MHA, height ...... 49 4.9 AFM images: 1:1 EHPT:MHA, phase ...... 49 4.10 AFM images: MHA with coupled protein, height ...... 49 4.11 AFM images: MHA with coupled protein, phase ...... 49 4.12 AFM images: 1:1 EHPT:MHA with coupled protein, height ...... 49 4.13 AFM images: 1:1 EHPT:MHA with coupled protein, phase ...... 49 4.14 AFM images: EHPT after UV-O3 treatment, height ...... 50 4.15 AFM images: EHPT after UV-O3 treatment, phase ...... 50 4.16 AFM images: Unthiolated EHPT, height ...... 51 4.17 AFM images: Unthiolated EHPT, phase ...... 51 4.18 AFM images: Unthiolated EHPT after sonication, height ...... 51 4.19 AFM images: Unthiolated EHPT after sonication, phase ...... 51 4.20 AFM images: Bare gold, deflection ...... 52 4.21 AFM images: Bare gold, friction ...... 52 4.22 AFM images: EHPT SAM, deflection ...... 52 4.23 AFM images: EHPT SAM, friction ...... 52 4.24 AFM images: MHA SAM, deflection ...... 52 4.25 AFM images: MHA SAM, friction ...... 52 4.26 STM of bare gold ...... 54 4.27 STM of EHPT SAM ...... 54 4.28 STM of MHA SAM ...... 54

5.1 Assembly schematic for SAM-based neuron/electrode interface ...... 58 5.2 Neurite outgrowth on pure and mixed EHPT/MHA SAMs at 3 and 7 DIV ...... 62 5.3 Attachment of neurons on pure and mixed EHPT/MHA SAMs at 3 DIV ...... 62 5.4 Neuron culture photographs: MHA SAM ...... 63 5.5 Neuron culture photographs: 1:2 EHPT:MHA SAM ...... 63 5.6 Neuron culture photographs: 1:1 EHPT:MHA SAM ...... 63 5.7 Neuron culture photographs: 2:1 EHPT:MHA SAM ...... 63 5.8 Neuron culture photographs: EHPT SAM ...... 63

6.1 MultiChannel Systems EcoMEA ...... 66 6.2 Impedance of electrodes before and after EHPT SAM ...... 68 6.3 Impedance of electrodes before and after MHA SAM ...... 68 6.4 Impedance of electrodes before and after 1:1 EHPT:MHA SAM ...... 69 6.5 Impedance changes for removal and re-application of EHPT SAM ...... 70 6.6 Lack of impedance changes with unthiolated EHPT “SAM” ...... 71

7.1 EHPT repeat unit ...... 74 7.2 HPT repeat unit ...... 74

10 7.3 Schematic of artiﬁcial bilayer experimental chamber ...... 74 7.4 Bilayer conductance changes induced by EHPT ...... 76 7.5 Lack of bilayer conductance changes induced by HPT ...... 76 7.6 Decay of EHPT-induced bilayer conductance changes ...... 76

8.1 EHPT dimer atomic nomenclature ...... 82 8.2 HPT dimer atomic nomenclature ...... 82 8.3 All-atom EHPT decamer ...... 83 8.4 All-atom HPT decamer ...... 83 8.5 Rigid rotor potentials computed for bithiophene ...... 89 8.6 Replica exchange: EHPT implicit water raster ...... 91 8.7 Replica exchange: EHPT implicit lipid raster ...... 91 8.8 Replica exchange: HPT implicit water raster ...... 91 8.9 Replica exchange: HPT implicit lipid raster ...... 91 8.10 Replica exchange: EHPT implicit water temperature history ...... 92 8.11 Replica exchange: EHPT implicit lipid temperature history ...... 92 8.12 Replica exchange: HPT implicit water temperature history ...... 92 8.13 Replica exchange: HPT implicit lipid temperature history ...... 92 8.14 End-to-end distance of EHPT in implicit water and lipid ...... 93 8.15 End-to-end distance of HPT in implicit water and lipid ...... 93 8.16 EHPT backbone angle distribution in implicit solvents ...... 95 8.17 HPT backbone angle distribution in implicit solvents ...... 95

9.1 Work curves for steered MD pulls of EHPT ...... 104 9.2 Work curves for steered MD pulls of HPT ...... 104 9.3 Force curves for steered MD pull of EHPT ...... 106 9.4 Force curves for steered MD pull of HPT ...... 106 9.5 SMD insertion trajectories for EHPT and HPT ...... 107 9.6 3D plot of insertion pathways for EHPT and HPT during SMD ...... 108 9.7 SMD removal trajectories for EHPT and HPT ...... 110 9.8 Potential of mean force for membrane insertion of EHPT and HPT ...... 111 9.9 Umbrella sampling conﬁgurations for EHPT and HPT ...... 113 9.10 3D plot of EHPT and HPT distribution during umbrella sampling ...... 114 9.11 EHPT backbone angle distribution in implicit and explicit solvents ...... 115 9.12 HPT backbone angle distribution in implicit and explicit solvents ...... 115

10.1 Schematic of in vitro electrophysiology experiment to follow up thesis ﬁndings . . . . 121

11 12 List of Tables

8.1 Optimized geometries of the thiophene ring ...... 85 8.2 Oligothiophene inter-ring dihedral angles ...... 86 8.3 Partial charges for ethylhexyl- and hexylthiophene monomers ...... 87 8.4 Energies of rotation about the bithiophene inter-ring bond ...... 90 8.5 Replica exchange rates for implicit solvent PT simulations ...... 90 8.6 Comparison of EHPT and HPT conformations in implicit solvents ...... 96

9.1 Comparison of EHPT and HPT conformations in implicit and explicit solvents . . . 116

13 14 Chapter 1

Introduction

1.1 Brief Rationale

1.1.1 What are Neuro-Robotic Interfaces?

The term “neuro-robotic interface”, as used in this document, means a technology that couples the electrical activity of living neurons to the input and/or output of an electronic device. These interfaces come in many forms, but almost all involve the implantation of one or more electrodes in close contact with neurons in a target area of the nervous system. This presents a host of technical challenges, from basic materials biocompatibility to the difficulty of establishing meaningful communication through relatively few electrode channels. Nevertheless, the tremendous clinical and scientific value of neuro-robotic interfaces has made them an active area of research for decades. The Neural Prosthesis Program at the National Institutes of Health has existed to guide research on neuro-robotic interfaces since 1972, and individual researchers were laying the groundwork of the field well before the establishment of a formal program. As shown briefly below and in detail in Chapter 2, the state of the field has advanced substantially and is accelerating, but there remain at least as many open problems as solved problems.

1.1.2 Clinical Value of Neuro-Robotic Interfaces

As of this writing, the clinical value of neuro-robotic interfaces still consists largely of unrealized potential. Cochlear implants, which transmit electrical signals to the inner ear to restore hearing, are the most widely used and successful example of a direct interface between electronics and the nervous system [1]. A distant second is functional electrical stimulation (FES), in which implanted electrodes stimulate nerves and muscles to restore function to paralyzed limbs and viscera. Although advances are being made, most FES systems currently implanted in patients are open-loop controllers that rapidly produce fatigue in the stimulated muscles [2]. No FES system currently available is able to restore anything resembling dexterous manipulation or normal standing/walking. The prior paragraph notwithstanding, the clinical value of neuro-robotic interfaces has advanced substantially in recent years, driven largely by the development of techniques to microfabricate high- density microelectrode arrays (described further in Chapter 2). Clinical trials are now underway with high-channel-count electrode arrays implanted in human motor cortex [3]. The first patient results show humans able to perform a variety of cursor control tasks, with performance improvement over time. Based on what has already been achieved in primate research (described below), it should be possible in the next 5-10 years for humans to control not only computer cursors, but also artificial effectors such as robotic arms. Arms that are controlled by residual muscular signals have recently been demonstrated, as shown in Figure 1.2. Tests are also underway of retinal electrode arrays to

15 Figure 1.1: Matthew Nagle Figure 1.2: Rehabilitation Institute of Chicago arms

Examples of human applications of neuro-robotic interfaces. In Figure 1.1, Matthew Nagle, the ﬁrst human implanted with the Cyberkinetics microfabricated electrode array. In Figure 1.2, two patients with Rehabilitation Institute of Chicago prosthetic arms that are controlled by electrical signals from chest muscles. Images are courtesy of Cyberkinetics and RIC, respectively.

restore visual function [4], and stimulatory prostheses for the visual cortex are in primate trials [5]. These present and near-future clinical applications of neuro-robotic interfacing exist alongside another active body of research that seeks to induce the nervous system to regenerate and self- heal following acute and chronic injuries. While such research is highly valuable, it is unlikely to ever eliminate the need for neuro-robotic interfaces. Even when it occurs, regeneration often “miswires” the body, such as wiring sensory neurons to motor targets. Electrical interfaces will likely be necessary to help remap the nervous system to adjust to the new wiring pattern, and recent results suggest that stimulation during regeneration can help reduce the degree of inappropriate reconnection [6].

1.1.3 Scientiﬁc Value of Neuro-Robotic Interfaces

While the clinical applications of neuro-robotic interfacing are still emerging, these same technologies have been generating valuable scientific data for years. The most well-known example is the work of multiple teams who are using signals from primate motor cortex to control robot arms [7–9]. In addition to providing guidance for future clinical studies, these platforms are providing data about how the human brain approaches the task of motor control, which in turn can inform roboticists’ efforts to construct artificial controllers. The same work is also helping to map the information content of the brain, such as the recent discovery that signals predicting an intention to move can be found in monkey parietal cortex [10]. As interface technology advances and is able to be implanted into more areas, it will become possible to instrument the nervous system and gain insight into a wide variety of motor, sensory, and cognitive functions. The same techniques can also be used to train and control animals to perform tasks in human-inaccessible environments, as has been demonstrated with remote-controlled insects and rats [11, 12]. The electrodes and associated electronics used for implantable neuro-robotic interfaces can also control devices based on the signals from arrays of harvested neurons cultured in a dish. Such in vitro interfaces have already been used to drive a simple mobile robot, detect features in an image, and serve as chemical sensors [13–15]. As techniques for culture maintenance and patterning improve, these biological controllers may find uses within robotics in domains where effective algorithms have not yet been developed. Alternatively, they may inform the development of “neuromorphic” systems that implement the same network-style computations in analog electronics. Within neuroscience,

16 A B

Figure 1.3: Schematic of the neuro-robotic interface proposed in this thesis. (A), overview of the concept, wherein a living neuron rests on a metal electrode surface that is coated with a conductive polymer and protein blend. (B), detail of indicated area in (A), showing the hypothesized intracellular recording mechanism based on polymer “molecular wires” penetrating the lipid bilayer. Both concepts are expounded on in more detail in Chapter 3. interfaces to cultured neurons will allow study of the properties of small networks of deﬁned geometry, thus elucidating the basic mechanisms underlying the complex computations performed by the nervous system.

1.2 Overview of This Thesis

The present and potential benefits of neuro-robotic interfaces are substantial, but realization of those benefits is still hampered by the limitations of the implantable electrodes that form the actual interface between the biological and electronic systems. This thesis describes the development of a new type of electrode, based on self-assembled monolayers (SAMs) of an electrically conductive polymer chemically bound to the electrodes of a microfabricated array. The concept is illustrated in Figure 1.3, and will be explained in detail in Chapter 3. Briefly, two of the principal difficulties plaguing existing interfaces are insufficient biocompatibility and an insufficiently specific connection between neuron and electrode; both of these concepts contain many subtleties that are explored in depth in Chapter 2. The electrode concept developed in this thesis offers increased biocompatibility with no detrimental effect on electrode properties, and preliminary data indicate that it can implement a new type of intracellular recording that will be far more specific and information-rich than anything achieved to date in vivo. In the following chapters, the rationale for this particular approach is laid out in detail, followed by the proof of concept of the essential components. Chapter 2 reviews in depth the current approaches to neuro-robotic interfacing with a focus on the different electrode configurations and materials used, followed by a review of the limitations of those approaches and the biological mechanisms hypothesized to underlie those limitations. This is followed in Chapter 3 with an explanation of the approach taken in this thesis and a justification of why this approach should improve on what has previously been accomplished. Chapter 4 presents data demonstrating the success of the self- assembly fabrication technique as applied to this system. Chapter 4 also shows the nanostructure of the polymer electrode surface, which will help to explain the results of later chapters. In Chapters 5 and 6, it is shown that this polymer-based electrode is biocompatible with cultured mammalian neurons, and that owing to the specific composition of the film, this biocompatibility does not negatively impact the electrical impedance of the electrodes (and may even improve it). Chapters 7 through 9 demonstrate the feasibility of a new approach to intracellular recording, combining experimental data from work with artificial membranes (Chapter 7) with high-resolution computer simulations (Chapters 8 and 9). The thesis concludes in Chapter 10 with a summary of the demonstrated results and a roadmap of future work that would be appropriate to further develop this proof of concept into a robust and usable technology.

17 18 Chapter 2

Background

2.1 Review of Current Neuro-Robotic Interface Approaches

This section reviews the various types of implantable and in vitro systems that have been developed to interface neurons and electronics. It is focused primarily on the electrodes that implement the interface between biological, ion-based electrical signals and engineered, electron-based signals. This review does not cover non-invasive systems based on electroencephalography (EEG), as these systems are extremely unlikely to become a dominant interface technology due to their bandwidth limitations. Previous reports indicate that EEG-based systems can transfer at most 25 bits per minute (0.42 bits per second, bps) [16] of information out of the nervous system. The first generation of invasive electrode systems were able to achieve 3 bps [17], and more modern decoding strategies have reached 6.5 bps [18]. While EEG techniques can also benefit from improved decoding strategies, the application of such techniques has still not pushed EEG information transfer beyond 1 bps [19]. Deep-brain stimulation (DBS) systems are also not reviewed. Although clinically interesting and valuable, these systems do not attempt to decode neural signals or inject information into the nervous system, and therefore face entirely different operating requirements from true neuro-robotic interfaces. Existing interface electrodes fall into three broad categories: nerve cuffs, regeneration/sieve electrodes, and microfabricated electrode arrays. The latter category, arrays, can be further subdivided into arrays designed for implantation and arrays designed for in vitro use with cultured neural networks. When evaluating any of these technologies, a good electrode will have three properties:

1. Universal. The electrode should be usable anywhere in the nervous system, peripherally or centrally, with minimal modifications to its design. The complex architecture of higher nervous systems will demand some degree of application-specific customization, but this should be achievable without substantial changes to the fabrication process. 2. Selective. If recording, the electrode should be able to detect and separate out the signals from individual neurons/axons (“single units”). If stimulating, the electrode should ideally activate only one neuron, although it may be acceptable if small groups are activated instead. It would be desirable for the population of neurons stimulated/recorded to remain constant over time and not drift due to electrode motion. 3. Biocompatible. The electrode must not damage the surrounding tissue, cause the death of nearby neurons, or provoke a foreign body response that walls it off from the target neurons. Biocompatibility is a complex concept with many subdomains; this review is concerned largely with materials biocompatibility (ensuring that the materials of electrode construction do not provoke negative reactions) and stimulus biocompatibility (ensuring that stimulation delivered through an electrode does not have detrimental effects on the

19 stimulated tissue). There is a secondary concern for what may be called “mechanical” biocompatibility (preventing negative eﬀects due to micromotion of the electrode), but this is primarily a materials problem and will be discussed within that context.

These three criteria will be used in the following sections to review the major types of neuro- robotic interface electrode and to highlight the advantages and shortcomings of each.

2.1.1 Nerve Cuﬀs

The nerve cuff is the simplest possible implant scheme – an electrode rolled into a cylindrical cuff that is attached around a peripheral nerve. An example of a nerve cuff is shown in Figure 2.1. Cuffs are simple to implant, and find widespread use within functional electrical stimulation (FES) [2,20,21]. Cuffs can also be considered biocompatible; since they do not cause penetrating injury of nervous tissue, and since the electrode sites are relatively far away from the target axons, any foreign body reaction that occurs does not have a detectable effect on electrode performance and may actually improve results by holding the cuff in place. However, the nerve cuff’s simplicity is also its weakness, as the design is not able to achieve high selectivity. Each peripheral nerve contains thousands of axons running to different muscle fiber populations, and cuffs in general are not able to make fine discriminations between these axons. This makes cuffs only marginally useful for recording, and their primary application has been for stimulation. Unfortunately, due to the lack of selectivity, stimulation through a cuff activates the entire nerve or a large fraction, all at once. Even if the stimulus amplitude is tuned to below the threshold of most of the axons in the nerve, the first to be recruited will be the largest (the well- known “size principle”). This is not helpful, as the largest axons innervate the “fast twitch” muscle fibers, which tire rapidly. The high-endurance “slow twitch” fibers are innervated only by small axons that are the hardest to activate with a cuff. The net result is that stimulation with a nerve cuff tends to cause rapid muscle fatigue [22]. There have been attempts to improve the clinical and experimental value of nerve cuffs by increasing their selectivity. Newer cuffs contain multiple electrode sites, allowing for “current steering” – the shaping of the applied electric field to selectively activate individual fascicles (small bundles of related axons) within the nerve. However, current steering is not able to achieve sufficient spatial precision to counteract the size principle recruitment and subsequent fatigue detailed above [23]. Another recent development in cuff electrodes is the “flat” or “reshaping” cuff, an example of which is depicted in Figure 2.2. This cuff applies a gentle pressure to the nerve over a period of weeks to months, slowly reshaping the nerve geometry to a flat “ribbon cable” configuration. This brings buried fascicles to the surface of the nerve and closer to the active electrode, which in turn ensures that the cuff is able to selectively activate the fibers of specific fascicles (and thus activate individual muscles and/or specific groups of motor units within those muscles). Flat cuffs have even been demonstrated in one study to be able to discriminate axons within a fascicle, as long as those axons are separated by over 200 µm [24]. It is not clear yet whether this reshaping process can truly be considered biocompatible. Initial studies of reshaping cuffs showed that they did not alter activation threshold (as measured by stimulating with a non-cuff electrode) or signal quality of axons in the rat sciatic nerve [25], but histological studies in this same nerve have shown increased intraneural deposition of collagen and other connective tissue, even when thresholds do not change [26]. Nerve cuffs’ relative biocompatibility, ease of use, and potential for modest selectivity will likely continue to make them the electrode of choice for FES applications for several years to come. However, they are ultimately not a universal neuro-robotic interface. They must be wrapped around a cylindrical bundle of axons, which precludes their use in essentially all of the central nervous system. Moreover, it is unlikely that cuffs will ever reach single-unit selectivity, which further limits their usefulness as new neuro-robotic applications demand access to more information-rich signals.

20 Figure 2.1: Standard nerve cuﬀ Figure 2.2: Flat Interface Nerve Electrode

Two examples of nerve cuff electrodes. On the left, in Figure 2.1, a standard cylindrical cuff in its rolled configuration. (Image courtesy of Medical University of Vienna.) On the right, in Figure 2.2, the Flat Interface Nerve Electrode (FINE) cuff, which slowly reshapes the geometry of a peripheral nerve in order to achieve more selective stimulation. (Image from [25].) Both types are relatively biocompatible, and the FINE is able to achieve modest selectivity, but neither is a universal interface, since they cannot be used in the central nervous system.

2.1.2 Sieve Electrodes

“Sieve” or “regeneration” electrodes are also targeted to the peripheral nervous system. A schematic of one such electrode is provided in Figure 2.3, with a photograph of an actual device in Figure 2.4. The sieve electrode depends on the presence of a peripheral nerve that has been severed, either by a traumatic injury or by deliberate surgical intervention. Axons in cut nerves naturally regenerate back towards their original targets. If a meshwork of holes is placed in the path of the regeneration, single axons and small groups will migrate through the holes. If the holes are lined by electrode sites, the device should be able to stimulate and record small groups of axons in a manner far more selective than a nerve cuff [20, 27, 28]. Although more selective than cuffs, sieves still possess difficulties. First and foremost is their inability to be used outside the peripheral nervous system. Although sieves have been placed in regenerating optic nerves (technically a central structure), they cannot be used in the brain and spinal cord proper [20,29]. These parts of the central nervous system do not regenerate significantly after an injury, and certainly do not regenerate axons to the lengths observed peripherally. Therefore, sieve electrodes, like cuff electrodes, do not meet the universality criterion. Even in the peripheral nervous system, the performance to date of sieves is suboptimal. In rats, muscles innervated by sieve-implanted nerves only recover about 30% of their pre-implant strength and show increased sensory latency [30]. The problem appears to be that current sieve designs interfere with nerve regeneration instead of aiding it; a blank tube produces better regeneration results than a sieve [31]. These barriers appear to be standing in the way of clinical applications for sieve electrodes; as of this writing, there are no known publications or reports of the use of such electrodes in humans. Sieve electrodes also perform poorly on the biocompatibility criterion. The poor regeneration is a clear sign that the devices are not seen as benign by peripheral nerves. Even with redesigns to increase the amount of open area through which axons can pass, only 7 of 10 implanted rats show any regeneration, and only 3/10 are able to produce recordings [32]. Stimulation delivered through sieves also requires levels of current and charge that may be unsafe, due to the need to deliver stimulation through the insulating myelin sheath around peripheral axons. Studies have reported currents from 30 to 100 µA, with corresponding injected charge of 6 to 20 nC [27, 33].

21 Figure 2.3: Schematic of a sieve electrode, from [27]

Figure 2.4: Photos of a sieve, from [28]

Sieve electrodes. Figure 2.3, schematic showing the regenerating nerve ﬁbers. Figure 2.4, a photograph showing an actual device and its central electrode area. Both sections of the photograph are measured against the same scale bar, which represents 1 mm.

Figure 2.5: “Michigan” style neural probes Figure 2.6: Utah Intracortical Electrode Array

Examples of the two principal microfabricated electrode arrays used for neuro-robotic interfaces. In Figure 2.5, examples of neural probes available from the University of Michigan’s Center for Neural Communications Technology. Probes are of silicon construction with 1 to 4 exposed gold active sites at the tip of each shank (gold circles). In Figure 2.6, micrograph of the Utah Intracortical Microelectrode Array, as described in [34]. The tips of the spikes are the active electrode sites.

22 2.1.3 Microelectrode Arrays

As depicted in Figures 2.5 and 2.6, there exist a wide variety of spike-like probes and arrays for implantation into the nervous system. While the precise geometry differs, all designs involve a small active metallized site that is integrated into a larger system of shanks and circuit traces to carry signals from the active site out of the body. These spike arrays represent the current state of the art in neuro-robotic interfaces, and of all the technologies discussed in this chapter, they come closest to meeting the three criteria. Although they are not entirely one-electrode-to-one-neuron selective, it is usually possible to separate out individual neurons using software processing [35–37]. Universality is at least theoretically achievable, leaving biocompatibility as the principal hurdle. The performance of these arrays degrades over time due to insufficient compatibility, as discussed further in Section 2.2. When used for stimulation, some arrays only require relatively low current amplitude and charge delivery, but other applications have required over 100 µA and 40 nC [38]. Recent efforts in microelectrode array technology have focused on expanding the variety of available arrays to increase their universality. Changes to the probe geometry have enabled use in the peripheral nervous system. This includes a “slanted” Utah-style array [39,40] (shown in Figure 2.7) and a multiple-sites-per-shank configuration of the Michigan-style probe [41]. While single-unit performance is still difficult to achieve, particularly with the Michigan designs, these probes have shown promise. They may be more effective than nerve cuffs for FES applications, because the numerous electrode sites can be used to interleave stimulation among multiple motor units, thus reducing the overall fatigue and achieving a more physiological recruitment profile [42]. Another design targeting peripheral applications is the longitudinal intrafascicular array, which is a Kevlar or other flexible fiber with electrodes patterned onto the cylinder [43]. The initial designs did not fare well, but the technology has been substantially revised in the past decade [44], even producing multi-channel versions [45]. These electrodes have successfully been used by human amputees to control computer cursors and artificial arms in a laboratory environment [46, 47]. Other advances in microelectrode arrays have focused on altering the probes to improve im- plantability and durability. Shape modifications such as sharpening the shank lead to easier tissue penetration [49]. Materials improvements have focused on either building stronger probes (mostly through ceramics) [50, 51] or building extremely flexible probes from polymer substrates [52, 53]. There is particular interest in fabricating extremely long probes that can reach subcortical and deep spinal structures. Methods for achieving this have included writing electrodes on optical fibers [54, 55], electroplating of structural nickel layers [56], and machining probes from blocks of metal using electrodischarge machining [57]. Finally, due to the detrimental effects of tethering probes to the skull via wires and cables (see below regarding micromotion), there has been interest in developing fully wireless systems [58]. Methods have been developed to integrate microfabricated probe arrays with CMOS-based electronics and/or small circuit boards [59, 60]. Other investigators have demonstrated optical (as opposed to radiofrequency) power and data transfer in vivo, which could offer higher bandwidth without cross-channel interference [61]. The same technology that creates implantable microelectrode arrays has also been used to create arrays for in vitro use [62–67]. These arrays are not meant to establish an interface to a living animal, but instead work with harvested and cultured neurons. They can be as simple as a square grid of electrodes, or be optimized to match the architecture of a brain slice [67]. Alternatively, if the target geometry is not well-understood, there exist designs for ultrasmall scanning probes similar to the implantable microelectrode spike array [68]. In vitro arrays can be used for many purposes, from controlling mobile robots to biosensing [13, 15]. With proper apparatus design, cultures can live for months, possibly even years [63, 69]. The most common application for in vitro microelectrode arrays at present is supporting and probing cell networks for basic neuroscience. Random networks grown from ordinary dissociated neurons can be used to test the effect of neurotransmitters or growth factors [64], and can be coupled to silicon “model neurons” to study phenomena such as oscillatory loops [70]. By applying one of the

23 Figure 2.7: Slanted Utah array, from [39]

Figure 2.8: Fluidic Michigan probe, from [48]

Variations on the neural probes of Figures 2.5-2.6. Figure 2.7, the Utah Slanted Electrode Array. The varying lengths of the shanks allow the array to access all the fascicles of a peripheral nerve. Figure 2.8, a schematic for a recent integrated microfluidic/electronic probe from the University of Michigan. The integrated fluidic channels allow the probe to deliver neurotransmitters or soluble growth factors, the advantages of which are discussed further in Subsection 2.3.1. many known techniques for patterning cultured neurons, it is also possible to create networks with defined geometry, either for studying their function or implementing a computation that cannot be well-expressed in digital hardware or software [71–75]. However, for the purposes of this thesis, the most valuable aspect of in vitro arrays is their close relation to implantable microelectrodes. Because these cell-culture arrays are fabricated from the same materials as the implants, they can be used as inexpensive models for those same implants. This property forms the basis of the impedance experiments in Chapter 6.

2.2 Limitations of Current Interface Technologies

2.2.1 General Trends

The majority of performance studies for implantable neuro-robotic interfaces have focused on microelectrode arrays. Moreover, as noted above, only spike-like arrays are able to meet the universality criterion and approach the specificity criterion. Therefore, this section will focus on the effects of microelectrode arrays’ failure to meet the biocompatibility criterion and the underlying causes of that failure, while the next section will discuss ways to improve microelectrode biocompatibility. The ultimate outcome of insufficient biocompatibility is probe failure, i.e., the inability of any electrode on the entire array to record usable signals from any neurons. However, the time and path to failure can be highly variable depending on the array, the anatomical site, the implantation technique, and the underlying host phenotype. In recent reports, probes in monkey motor cortex had a roughly 60% chance of being able to record activity when tested [76]. The figures for rats are more variable; one study with Michigan-type probes reported that only 4 of 6 implants had any recording sites at six-months post-implant [58], whereas another using similar probes reported

24 over 90% still operative within the same timeframe [37]. In one study of peripheral “slanted Utah” arrays in cat sciatic nerve, total failure occurred by six months, and was accompanied by visible histologic damage on necropsy [40]. Even when probes do not completely fail, they can behave in a suboptimal manner that substantially decreases the interface’s usefulness. One of the most common undesirable behaviors is ensemble shift – a day-to-day variability in the population of recordable neurons and the tuning characteristics of those neurons. This has been documented (along with substantial trial-to-trial variance in electrode impedance and signal-to-noise ratio) in monkeys using the Cyberkinetics array system [77], and has also been seen in the recent human trials [3]. This shifting requires a technician to re-tune the interface for the patient at every recording session, which is an acceptable burden in initial trials, but could not be tolerated in a widely-used clinical device. It is worth noting that microelectrode array performance is not entirely a function of biocompatibility; surgical technique can improve or worsen outcomes. For instance, in 1998, implanted Utah arrays had 40% total array failure by 6 months, and even successful arrays had fewer than 50% active electrodes at that timepoint [35]. In 2005, Utah-conﬁguration implanted arrays had 60-80% active sites at nearly two years post-implant [77]. However, with years of accumulated experience in implanting these arrays, it is likely that we will soon reach the limits of improved operative technique, and the community will then look to materials design to further improve performance and reliability.

2.2.2 Proposed Mechanisms Underlying Limitations

The mechanisms underlying probe failure and ensemble shifting are still not fully understood, but are believed to involve changes in the tissue around the implant. Within weeks of implanting a microelectrode array, a fibrous sheath or scar of reactive astrocytes forms around the implant [78]. This sheathing is accompanied by a decrease in the number of neurons in the implant vicinity and a slow increase in the impedance of the electrodes [36]. There remains some debate as to whether the impedance increase is due to the non-conductive nature of the scar tissue or whether it involves protein deposition on the electrode sites, although recent demonstrations of electrode “cleaning” in vivo support the latter theory [79]. The nature of the neuronal depletion is also debated. Neurons are known to be motile, so migration of neurons away from the zone of inflammation is certainly possible. On the other hand, one study found no increase in neuron density outside the depletion zone (as might occur if neurons had migrated short distances), and on this basis argued that neurons around the implant are being killed by inflammation [80]. In some sense, the question is moot: either way, inflammation is increasing the distance between the electrode and the nearest neurons, which substantially decreases the chance of being able to record signals or stimulate without causing damage. Therefore, to solve the problem, one must either prevent the inflammation or reverse its secondary effects (e.g., enrich the neuron density in the implant vicinity). Attempts to improve the nervous system’s tolerance of implanted microelectrodes have been hampered by lack of knowledge of the true cause of the observed inflammatory reaction. The simple act of implantation damages and distorts blood vessels, and vascular damage can trigger neuron damage through both cell signaling and local anoxia [81]. Even microfabricated electrodes are large relative to biological structures and are made of materials not normally found in higher organisms, so they may provoke a foreign body reaction through simple materials incompatibility. Some materials may also be at fault through their mechanical properties. Silicon, glass, and metal are all common materials of microelectrode array construction, and they are all extremely stiff compared to the surrounding tissue. It has been theorized that this “modulus mismatch” may be able to mechanically activate astrocytes and microglia. This effect may be exacerbated by microscale motion of the implant. The nervous system is subject to cyclic forces from the cardiovascular and respiratory systems, and a rigid probe will not be able to compliantly respond to these forces in the same way that the neighboring tissue can. Finite-element modeling of microelectrode probes in tissue supports this hypothesis and shows that a freely-moving probe will exert substantial strain for up to 100 µm around itself [82, 83]. This micromotion may be particularly troublesome as devices move from rat models to primates to humans, as the larger bodies cause larger motions [37].

25 In addition to the above mechanisms, which are due to the physical characteristics of the implant, neuro-robotic interfaces that stimulate neural tissue have additional modes of bio-incompatibility. There continues to be debate and investigation to determine which stimulus parameters drive these effects, but the two principal factors identified to date are total charge per stimulus and stimulus current amplitude. Injecting more charge than a given electrode can handle causes potentially deleterious redox reactions (Faradaic current) at the electrode surface; the underlying mechanisms are reviewed well in [84]. The precise chemical species that causes tissue damage is unknown (and presumably varies depending on the electrode material), but hypothesized mechanisms include free radicals that damage cells and microscopic bubbles from water electrolysis that cause mechanical damage [84, 85]. Some electrodes deposit metal particles in the stimulated tissue when over-pulsed, potentially triggering foreign body reactions [86]. Overly large stimulus currents can drive Faradaic reactions if applied with a sufficient pulse width, but can also cause negative effects that are not correlated with histological damage. Neurons stimulated with large currents can become unresponsive or greatly diminished in responsiveness over a period of days to months [85], a phenomenon which has been dubbed “Stimulus Induced Depression of Neuronal Excitability”, or SIDNE [87, 88]. SIDNE can begin within the first hour of stimulation [87], and can last for over two weeks after stimulation is discontinued [89]. It does not increase over time, but once the target tissue is less sensitive than baseline, the subject will inevitably lose some ability to discriminate between similar stimuli [88]. The hypothesized mechanism underlying SIDNE is over-activation of neurons that fall in the overlap zone between the stimulatable fields of two or more closely-spaced electrodes. The larger the applied current, the further that current can spread, and thus the larger the zones of potential overlap. Support for this hypothesis comes from studies in which stimulation was delivered through electrodes deliberately spaced far apart; in these studies, less SIDNE was observed compared to a closer spacing [89]. However, even this protocol does not prevent SIDNE; some neurons are activated transsynaptically by the stimulus coming from each electrode, and these doubly-activated neurons become refractory.

2.3 Known Techniques for Countering Electrode Limitations

Techniques have been suggested to counter all of the hypothesized root causes of probe failure and decreased tissue responsiveness. Incompatible materials and mechanical mismatch can both be addressed by coating the interface probe in a more compatible interface material and/or fabricating it entirely out of such materials. Micromotion can also be countered by careful materials selection, since the nearby tissue can be caused to tightly adhere to the probe and fix it in place. It may be possible to counter still more effects of micromotion by making the probe itself mobile. Solving the problems caused by excessive stimulus charge and/or current is more difficult, but the best approach is to further increase the closeness and specificity of contact between neuron and electrode. (This effectively means that the specificity criterion of Section 2.1 must be tightened such that present microelectrode arrays do not meet it.) If the electrodes can be brought into more intimate contact with the targets (i.e., reducing or eliminating the zone of neural depletion that is presently observed), those targets can be stimulated using lower currents. The use of lower currents will also decrease the total injected charge; when combined with a careful choice of electrode material, this can eliminate undesirable Faradaic reactions. This section reviews the approaches that have been tried to date to implement all of the above ideas, and also briefly reviews the areas where each approach still requires improvement. In addition to the mechanisms described below, the inherent plasticity of the nervous system can also work around some of the difficulties of neuro-robotic interfacing, particularly in the central nervous system. The real-space receptive fields of neurons in different brain regions are not as purely topographically mapped as is suggested in introductory textbooks – mapping of visual cortex with a Utah-type array showed substantial and non-linear deviation from the expected map [90]. Moreover, the nervous system responds to the presence of a functional implant by altering those maps. Monkeys with motor cortical implants are able to rapidly de-associate and re-associate the

26 activity of motor neurons with either forearm movements or the movements of a brain-controlled robotic arm [91]. Stimulatory arrays are particularly able to cause cortical remapping, as has been demonstrated in motor cortex [92], visual cortex [93] and with cochlear implants [94]. Nevertheless, for the cortex to remap, there must be living, electrically responsive neurons in the vicinity of the implant, and in order to provide those neurons, the limitations identiﬁed in Section 2.2 must be addressed.

2.3.1 Increasing Material Biocompatibility

One approach to building a more biocompatible neuro-robotic interface is to construct the entire probe (or the vast majority of it) from a soft and biocompatible material. This is the motivation behind, e.g., the polyimide probes described in [52]. The principal difficulty with this strategy is that, being soft and flexible, these probes buckle under the force of insertion unless individual incisions are made in the pia mater for each shank. Such an implantation technique is not acceptable or feasible for wide-scale use. Probes can be given a partial silicon backing, or dissolvable materials can be used to temporarily stiffen probes for insertion (the best known material being common table sugar) [95, 96]. Even if implantation is feasible, though, there remains a secondary difficulty: the mobility of the part of the probe that is not embedded in tissue. Some electronics, bond pads, etc. will have to protrude above the surface of the brain, spinal cord, or nerve in which the interface is implanted. Computer models suggest that flexible probes will not induce substantial strain around the shank, but the strain at the tissue surface will be much greater than that observed with a stiffer probe [83]. The better strategy, therefore, is to move away from probes where any single material performs both structural and compatibility functions, and instead to investigate coatings that can be applied over a stiff probe to improve its biocompatibility.

Hydrogels and Layer-by-Layer

“Hydrogel” refers here to any solid substance that swells when immersed in water and whose volume when fully swollen is composed largely of water. The solid component of these coatings is usually a biologically inspired polysaccharide, which tends to be well-tolerated by cells. The swollen gel is also very soft and compliant, which resolves the problems caused by mechanical strain. The biocompatibility of hydrogel-coated neural probes has not yet been verified in vivo, but is expected to be high, and the coatings are able to remain on probes when inserted into guinea pig brain [97]. There are some concerns that a thick layer of gel over the electrode sites will raise impedance and decrease recording capability, but this does not appear to occur with the gels tested to date [97]. The principal limitation of the hydrogel technique (assuming that the gels will not delaminate from the probe after longer implantation durations) is that a thick coating increases the cell-electrode separation. As noted above, increasing this distance will cause a stimulating electrode to require greater currents, which in turn may cause neurons to become refractory and/or cause Faradaic current reactions that would destroy or delaminate the gel. Some evidence for the latter effect has been seen in studies of conductive polymer electrodeposition in hydrogel environments [97]. A related technique, which avoids the concerns of delamination and cell-electrode separation, is placing the hydrogel into small wells contained on the probe shank [98]. While these wells will not resolve the problems of tissue/electrode mechanical mismatch, they can release drugs or other soluble factors that reduce inflammation in the local area [99]. Since peripherally administered dexamethasone reduces astrocytic scarring, the drug should work even better when released and concentrated in the implant area [100]. Local drug release is also more likely to be effective because the probe is surrounded by a chronic inflammation process, against which a single drug bolus is of only partial value [101]. In studies to date, the presence of dexamethasone-releasing wells caused the impedance of probe electrodes to be lower than in control implants [98, 99]. However, this lowered impedance did not correlate with any improvement in the quality of recordings, suggesting that the controlled-release strategy alone is not able to solve the inflammation problem.

27 Yet a third technique in this general family is layer-by-layer deposition. In this method, the coating film is built up from monolayers of two oppositely charged materials; each dip in one material prepares a base to which the other will adhere. It is easy to trap biomolecules in such films and thereby increase electrode compatibility, and indeed, example layer-by-layer films have proven to be biocompatible in neuron culture [102, 103]. It is as of yet unclear whether they can survive implantation, whether they will degrade in vivo, and whether the presence of multiple alternating charge layers will alter an electrode’s ability to pass capacitive current (and thus negatively impact stimulating and recording). Until these questions are resolved, layer-by-layer deposition remains a promising technology, but not one proven for neuro-robotic interfaces.

Integrated Microﬂuidics

In a close variant of the hydrogel well concept, anti-inflammatory or neurite-attracting factors can be delivered in liquid form through a probe that has integrated microfluidic channels. Initial versions of these “garden sprinkler” configurations have been designed [104], and newer versions include shutters to reduce unwanted leakage and flowmeters to measure delivery rates [48]. An example of a more advanced probe was shown in Figure 2.8. To date, these probes have been validated as able to deliver fluid during acute testing sessions, both in tissue phantoms [105] and in guinea pig brain [106]. Validation in chronic applications has not yet been published, and will require development of further components such as fluid reservoirs, micropumps, and heat dissipation schemes. However, once these issues are solved, the performance of fluidic-based probes will presumably be somewhere between that of the hydrogel wells and soluble growth factor approaches such as the cone electrode (see below).

Conductive Polymer Coatings

Conductive polymer films represent a very promising technology for increasing neuro-robotic interface biocompatibility. These films are deposited by passing a current through an electrode that is immersed in a solution of the monomer. If the solution also contains biomolecules, those molecules will become trapped within the film. The net result is a film that is soft (providing mechanical biocompatibility) and laced with cell adhesion molecules (enhancing material biocompatibility) [107]. Moreover, since the polymer is electrically conductive, the effective electrode surface area is increased, leading to a hundredfold decrease in electrode impedance [108]. This increases the electrode’s ability to deliver charge in a capacitive (non-Faradaic) regime, thus making it possible to stimulate more safely. Recording performance also improves when using conductive polymer films, but this does not correlate with the electrode impedance [109]. (This is not entirely surprising, since the bare metal electrodes are already of sufficiently low impedance to record good neural signals.) In- stead, the mechanism is believed to be the tight neuron/electrode contact induced by the presence of adhesion-promoting molecules. Images of the growth of these films and the positive response of cultured neurons are shown in Figures 2.9-2.11. Similarly good biocompatibility has been observed in vivo [109]. The three principal limitations of electrodeposited conductive polymer films are lack of control over film composition, film delamination, and slow loss of conductivity. The film structure can be determined to some degree by the use of a dissolvable template [110], but the precise proportion of biomolecules to polymer cannot readily be controlled, nor can their geographic distribution. While this may seem minor, there is evidence that neurons prefer very specific spatial distributions of adhesion sites [111]. As differences are discovered between different peptide dopants, the optimal strategy may involve controlling the distribution of those dopants on a micrometer to nanometer scale [112]. It would therefore be desirable to develop a conductive polymer method that allows for more precise control of film composition and structure. The precise mechanism of film delamination is not well-known, but is presumed to be due to the saline environment (culture media or extracellular fluid) attacking the interface between polymer and

28 Figure 2.9: Before polypyrrole deposition Figure 2.10: After polypyrrole deposition

Figure 2.11: Cell growth on PPY-coated electrodes

Electropolymerization of polypyrrole (PPY) films and their effects on cell growth. Figure 2.9, scanning electron micrograph of the bare gold electrode on a Michigan-type probe. Figure 2.10, similar electrode after deposition of a PPY film. Electrode surface area has been massively increased due to the film roughness. Figure 2.11, neuronal response to PPY films when doped with biomolecules. This photograph is a actin stain of human neuroblastoma cells cultured on a coated probe. No cell growth is seen on the right half, where polymer was formed using acetic acid as a counterion. On the left, extensive attachment and growth is found when the adhesive peptide CDPGYIGSR is used as the counterion. All images in these figures are from [107].

29 metal electrode. The polymer is only attached to the electrode surface by intermolecular electrostatic and van der Waals forces, which can in time be overcome. Roughening of the metal surface through techniques such as electroplating can improve adhesion, but the non-metallic compounds present in common electroplating solutions may interfere with biocompatibility [113]. A more tightly bound conductive polymer formulation would therefore also be desirable. The final limitation is slow loss of conductivity, in which the polymer slowly becomes non- conductive and the electrode impedance rises. This is seen after soaking in water for extended periods, or after multiple high/low voltage cycles, as are found during cyclic voltammetry and neural stimulus pulse trains. While the problem has been observed with polypyrrole (PPY), the most commonly-used polymer, it is not observed with poly(3,4-ethylenedioxythiophene) (PEDOT) [109, 114]. The structures of these different polymers are shown in Figure 2.12. PEDOT is equally effective as PPY at lowering electrode impedance and producing improved recording performance [115,116]. However, it is more difficult to synthesize, and requires harsher solvents that could denature delicate biomolecules. Polythiophene, which is related to PEDOT and is also depicted in Figure 2.12, should offer similar resistance to slow conductivity loss, and will be discussed more in Chapter 3.

Nanotubes

Coatings of carbon nanotubes would be expected to roughen the electrode surface and improve performance much as conductive polymers do, and these coatings have been studied as improvements to neuro-robotic interface probes. There is evidence for decreased impedance with high capacitance [117], and for biocompatibility with cultured neurons when the nanotubes are properly functionalized [117–119]. However, unlike conductive polymers, nanotubes are a stiﬀ material [117]. This may lead to the same deleterious eﬀects of mechanical mismatch that are seen with silicon and metal. The concept of neuro-robotic interfaces based on nanotubes is still fairly new, and much remains to be discovered, so it is still too early to judge the utility of the nanotube approach. At the present time, though, nanotube coatings have not been proven to produce a good interface electrode.

Self-Assembled Monolayers

Self-assembled monolayers (SAMs) are formed by soaking a metal electrode in a solution of a long- chain organic molecule that has a metal-binding end-group. The most commonly-known chemistry is that of alkanethiols on gold, which can readily be used to link peptides and proteins to metal surfaces [120, 121]. Alkanethiol SAMs have been used to construct impedance biosensors based on neuron-like PC12 immortalized cells [122], but have not been used in in vivo neuro-robotic interfaces to date. The advantage of SAMs is that most can be patterned at very high resolution, using methods ranging from photolithography to microcontact printing [74, 123–125]. Because of this, they have been used for in vitro applications, where they can produce high-ﬁdelity patterns in cultured neural networks [126]. These patterns increase the number of electrodes that are able to record spikes, and could also be used to study the behavior of deﬁned network geometries [127,128]. Since most SAM technology depends on the presence of long-chain alkanes, they likely have not been pursued for in vivo applications due to their tendency to insulate the coated electrodes. As will be described in the following chapters, that limitation can be overcome by changing the composition of the SAM.

2.3.2 Increasing Speciﬁcity and Decreasing Current/Charge Injection

As noted above, increasing the closeness and speciﬁcity of contact between neuron and electrode is the best strategy for improving stimulus biocompatibility. Most methods that improve contact will also increase material biocompatibility, since they entail the creation of an electrode that neurons can tolerate. This interaction between multiple aspects of biocompatibility and between the biocompatibility and speciﬁcity criteria is useful, as technologies developed to solve one problem may also address several others.

30 S * * S

Polythiophene n

H N * * N H Polypyrrole n

S * * S

O O PEDOT

Figure 2.12: Repeat units of polythiophene (PT), polypyrrole (PPY), and poly(3,4- ethylenedioxythiophene) (PEDOT).

31 Inducing Neurite Growth Onto Probes

Since the need to use high currents and inject large amounts of charge when stimulating is caused by too large a separation between neuron and electrode, the obvious approach is to reduce that separation. This approach is validated by in vitro results, which show that attracting neurons and/or neurites to an electrode improves the ability to record signals [127, 128]. Similar support is provided by in vivo evidence from probes coated with PPY/peptide or PEDOT/peptide blends, for which recording quality does not correlate with electrode impedance but is theorized to correlate with close approach of neurons [109, 116]. Several of the above-reviewed approaches to increase material biocompatibility also increase neurite growth onto and near the coated electrode. These approaches include conductive polymers, self-assembled monolayers, and (in theory) fluidics. Hydrogels loaded with a growth factor may also be able to stimulate neurite sprouting and induce cell migration to the electrode sites, but this has not been observed with hydrogel well probes loaded with nerve growth factor (NGF) [98]. It should be noted once again that if neurite growth occurs too quickly, or if it is required to occur through a thick coating, growth can be hindered and/or cell death caused due to the lack of vasculature supplying oxygen and nutrients. One growth-promoting approach not mentioned in prior Sections is the “cone electrode”. This is a glass cone containing two silver wire electrodes and a relatively large reservoir of NGF. Once implanted, the NGF slowly diffuses from the electrode, inducing a large number of neurites to grow into the cone. The two microwires then record action potentials from these neurites. This electrode has been implanted in humans, and can achieve single-unit recordings without failing over a period of years [129]. Anecdotal reports at recent conferences indicate that cone electrodes implanted in human speech-motor areas may be able to drive a computerized phoneme producer and thus create a “speech prosthesis”. However, the availability of only two data channels substantially limits the cone electrode’s use as a neuro-robotic interface. Achieving good control of, e.g., a cursor on a computer screen can require the use of local field potentials and/or the addition of electromyographic (EMG) signals from neck and shoulder muscles [130,131]. While this is adequate for the current primitive applications of neuro-robotic interfacing, future generations of applications will likely demand interfaces able to supply single units at a high channel count.

Current Steering

Steering involves applying an electrical stimulus through multiple electrodes simultaneously, as opposed to the more common approach of applying a simple pulse train through a single electrode. The essential concept is that neither electrode applies a stimulus that is strong enough to fire neurons in its vicinity, but the sum of the two stimuli in a zone of overlap is sufficient to specifically activate neurons within that zone. To date, current steering has been demonstrated in nerve cuffs [23] and in cochlear implants [132]. It is able to target objects about the size of a fascicle, but is not yet able to pick out individual neurons or axons [23]. Single-unit performance may be achievable with high-precision multielectrode arrays, but this has not yet been demonstrated. It would be a difficult endeavour, as the location of surrounding neurons is never precisely known, and tissue in- homogeneity will alter the applied electric field away from what is intended. In short, while current steering techniques are useful, particularly for FES applications, they would require major advances in understanding and modeling of the neuron/tissue/electrode interface in order to be usable for high-resolution information transfer.

Intracellular Stimulation and Recording

It is possible to establish an extremely specific and intimate connection by placing an electrode inside the target cell. The original implementations of this technique used a very fine wire or sharpened glass electrode. In recent decades, direct penetration techniques have given way to the “patch clamp” technique, illustrated in Figures 2.13-2.14. Briefly, a glass micropipet is carefully

32 pulled under controlled heat and force to produce a tip with a diameter of a few micrometers. The pipet is then filled with a saline solution and a small wire is inserted into the pipet bore. When the tip of the pipet is placed up against a cell and a small amount of vacuum is applied, a strong seal (with a resistance at least 1 GΩ) forms between pipet and cell membrane. A little more suction will rupture the membrane in the sealed zone, allowing direct electrical connectivity (through the pipet solution and the saline intracellular fluid) between the wire and the cell. In this configuration, an experimenter can monitor not only the normal action potentials of the neuron, but also small subthreshold changes in membrane voltage that may contribute to the cell’s computational function. Stimulation can be effected with nanoampere currents (compared to microamperes for all the electrodes described in Section 2.1). The technique is principally used in vitro, but a skilled experimenter can perform patch recording in the nervous system of a living animal. Intracellular stimulation in rat motor cortex has been shown to activate complex motor programs that would be valuable for a motor prosthesis [133]. The fatal flaw of all intracellular recording is that it cannot be done for any significant length of time without killing the target cell. As long as a pipet or other electrode is maintaining a hole in the membrane, vital molecules will diffuse out of the cell, and the cell will slowly be filled with extracellular medium or pipet solution. This will deprive the cell of the necessities of life, and it will die within a few hours. Patch clamping has been implemented in an array format for in vitro cell-based biosensors that rely on the use of immortalized cell lines [134–136]. An example of this “patch clamp on a chip” scheme is in Figure 2.15. However, as long as intracellular recording cannot be achieved without killing the target neurons, it cannot be used for true long-term neuro-robotic interfacing. Ideally, one should therefore seek a technique that provides intracellular access without the fatal membrane disruption. This thesis provides the basis of such a technique, the details of which will also be explained in Chapter 3.

Neurotransistors

Although most neuro-robotic interfaces use standard metal electrodes that convert electronic current to ionic current through the well-known Helmholtz double layer effect (reviewed in [84] and [137]), one group of investigators has proposed a different scheme. In their “neurotransistor” model, the neuron membrane and a buried electrode form two parts of a parallel-plate capacitor. Stimulation is achieved by charging the capacitor, and recording is achieved by coupling the artificial plate to the gate of a transistor [138,139]. In theory, no current should flow to or from the electrode, and thus the problems of charge injection and large currents should disappear. Arrays of recording transistors have been demonstrated [140], and the system has proven able to record from and stimulate invertebrate neurons in vitro [141, 142]. Unfortunately, the results with mammalian neurons in vitro have not been as promising. The recording transistors are able to sense local field potentials, but not the action potentials of single units [143]. In hippocampal slices, transistor/capacitor arrays have been able to successfully stimulate and record an increase in activity. However, achieving this success requires coating the entire device with a layer of metal to increase its capacitance, and once that coating is in place, stimulation requires voltages that may cause Faradaic current to flow [144]. These difficulties derive from the differences in neuron biology between invertebrates and vertebrates, with mammalian neurons tend- ing to be smaller and to have more complex ion channel distributions. The neurotransistor approach may someday be able to achieve effective stimulation and recording without risking deleterious redox reactions, but much development is still needed.

2.3.3 Countering Probe Micromotion

No deﬁnitive solutions are known for the problem of probe micromotion. Improvements in surgical technique are expected to improve matters. The development of fully wireless neural probes should also reduce micromotion, since some of the motion is likely due to the implant being “tethered” to

33 Figure 2.13: Patch clamping schematic Figure 2.14: Patch clamping in vitro

Figure 2.15: Patch clamp on a chip, from [136]

Illustration of the “patch clamping” technique for recording the intracellular potential of neurons. Figure 2.13, schematic drawing of the technique, illustrating the rupture of the cell membrane by a thin-tipped glass pipette and the formation of a high-resistance seal. Figure 2.14, photograph of a neuron in culture being recorded via patch pipet. Both images courtesy of the Institut f¨ur Biologische Informationsverarbeitung, Germany. Figure 2.15 illustrates how the same technique can be applied to a cell captured on a small hole in a microﬂuidic “patch clamp on a chip” system.

34 Figure 2.16: Sandia mobile probe ratchet Figure 2.17: Sandia mobile probe

The Sandia mobile neural probe. Figure 2.16, the MEMS ratchet mechanism that allows the probe shank to move along its long axis. Figure 2.17, a photograph of a three-shank probe, showing the shanks in different stages of extension. Both images taken from [146]. the skull by its cables. Constructing a probe entirely from soft materials could achieve compliant motion, as suggested at the start of Subsection 2.3.1, but as also noted in that same Subsection, this could merely increase the amount of strain applied to the surface of the tissue. The most likely techniques at the present time are those detailed in Subsection 2.3.2 for increasing neuron growth onto and around the implant. Until micromotion can be prevented, the next best strategy is to ensure that electrically active tissue is firmly attached to the probe so that it can survive the inflammatory reaction. One recent development not mentioned in the preceding sections is the creation of actively mobile neural probes. Using advanced fabrication processes, probes have been built that are able to move their shanks up to 2 mm, with a precision of about 1 µm [145]. Although the speed of motion is not sufficient for the probe to move in tune with cardiac/respiratory pulsations, anecdotal reports suggest that these mobile probes are able to automatically reposition their electrode sites in order to re-acquire neurons after experiencing failure. This is more of a workaround than a solution to the problem of motion-induced inflammation, but it is a valuable tool until such time as implant-associated inflammation can be prevented. The principal difficulty of mobile probes is the complexity of their actuation mechanisms. The initial design exposed the drive gears to the external environment, where blood clots and gliosis jammed them [145]. Newer mechanisms are more robust, but this robustness comes at the cost of a nearly tenfold decrease in positional accuracy [146]. Further re-engineering and refinement of the actuation mechanism will no doubt improve upon the state of the art, but it appears unlikely that active motion will be able to prevent local inflammation from occurring or fully mitigate the effects of that inflammation.

2.4 Summary

This chapter has introduced biocompatibility, selectivity, and universality as the three key criteria for evaluating neuro-robotic interface electrodes. Based on those criteria, the spike- or shank-based microfabricated electrode arrays described in Section 2.1.3 are the best interface currently available, primarily because they are the only known technology to meet the universality criterion. These arrays are still not suﬃciently biocompatible for advanced interface applications. In order to reach the desired level of biocompatibility, they must also improve their selectivity, because the deleterious eﬀects that appear when delivering stimulation through these electrodes are believed to trace back

35 to a lack of selectivity/specificity. In reviewing the approaches that have been taken by prior investigators to improve multielectrode array biocompatibility, conductive polymer coatings stand out as particularly promising. These coatings offer enhanced mechanical and materials compatibility. By increasing electrode area and drawing neurons closer to the electrode, they also decrease the chance that stimulation will cause unwanted Faradaic reactions or lowered neuron responsiveness. However, conductive polymer films still require solutions to the problems of delamination and composition control. Intracellular recording also offers advantages on the selectivity front, but all known intracellular techniques cause cell death in a matter of hours. In the following chapter, a new type of conductive polymer electrode will be proposed to address the limitations of existing coatings while also offering a prospect for long-term intracellular recording.

36 Chapter 3

Details and Rationale for Proposed Technology

As was revealed during the review of Chapter 2, no presently-available electrode technology fully satisfies the criteria for a good neuro-robotic interface. The principal difficulty is insufficient biocompatibility, with inadequately intimate/specific contact between electrode and neuron being part of that phenomenon. Conductive polymer films doped with adhesive biomolecules are able to improve biocompatibility, and intracellular recording gives extremely intimate contact with neurons, but both technologies have demonstrated weaknesses that prevent them from being full solutions to the neuro-robotic interface problem. This thesis presents a new concept in electrode materials, diagrammed in Figure 3.1, that addresses those weaknesses and provides a new platform for future development of neuro-robotic interface electrodes. The central idea is the use of self-assembled monolayers (SAMs) containing a tailored lipophilic poly(alkylthiophene) (PT). This in- novation offers a potential solution to the problem of polymer film delamination and forms the basis of a new type of intracellular recording. The remainder of this chapter explains how polythiophene SAMs address the known limitations of conductive polymer coatings, how lipophilic polythiophenes can be used for intracellular electrical access, and why polythiophenes are an appropriate polymer for this research.

A B

Figure 3.1: Schematic of the proposed neuro-robotic interface, reprised from Figure 1.3. (A), overview of the concept, wherein a living neuron rests on a metal electrode surface that is coated with a self-assembled monolayer blending conductive polymer (polythiophene) and adhesion molecules. (B), detail of indicated area in (A), showing the hypothesized intracellular recording mechanism based on polymer “molecular wires” penetrating the lipid bilayer.

37 3.1 Self-Assembly of Conductive Polymers

In Subsection 2.3.1, it was noted that conductive polymer films suffer from two major limitations: a tendency to delaminate from the coated electrode, and a lack of patternability. Self-assembled monolayers based on the sulfhydryl (thiol) group bypass both these limitations. They are chemically bound to noble metal substrates and are patternable by a wide variety of techniques, the most common being photolithography [123, 147] and microcontact printing [124, 148]. Thiol-based SAMs are robust under biological conditions [149–151]. The energy of the sulfur-gold bond is roughly 40 kcal/mol, stronger than the estimated gold-gold binding energy of 23 kcal/mol [152, 153]. More- over, organothiol binding to a metal substrate can often be refreshed electrochemically [154–156]. This particular SAM chemistry is most often used with long-chain alkanethiols, which can bear free carboxyl or nitrogen groups that are then usable to link adhesion molecules to the metal surface [121, 148, 157–162]. However, as was also noted in Chapter 2, alkanethiol SAMs are insulating, and thus would decrease electrode performance if used in vivo. Fortunately, thiol groups can be linked to many other organics besides long-chain alkanes, and this includes conductive polymers. The method was known for oligo(phenylene-ethylene) [163,164], and was developed for PTs in the preliminary work that preceded this thesis [165, 166]. It has also been shown that the individual “molecular wires” in these films do not couple to each other, but act as independent parallel channels to pass current from the underlying metal substrate to the medium at the tips of the molecules [167, 168]. SAMs formed purely from a conductive polymer would not be expected to be biocompatible. Neuro-robotic interfacing through a SAM therefore requires that the SAM have the biocompatibility of an alkanethiol/protein SAM and the conductivity of a polymer SAM or electrodeposited polymer film. This is easily achievable through the well-known “mixed monolayer” phenomenon. SAMs formed from a solution containing multiple self-assembling molecules will contain all of the chemical species, with the relative proportions in the SAM being related to the relative proportions in the solution [121, 124, 169]. Therefore, as diagrammed in panel A of Figure 3.1, this thesis proposes and demonstrates the development of electrodes based on a mixed SAM of a PT and a protein-bound alkanethiol. In addition to providing a solution for the issues of delamination and patternability, the multi-step assembly procedure (detailed further in Chapters 4 and 5) avoids exposing the biomolecules to harsh organic solvents. This, in turn, allows the use of more complex adhesion molecules (whole proteins instead of peptide fragments), which should produce a more effective response from the neurons. The presence of these adhesion molecules should also reduce the effects of probe micromotion. If nearby neurons grow axons that then attach to the electrode sites, those axons will be able to move compliantly and maintain their connection to the probe. There do exist techniques to covalently couple protein onto electrodeposited coatings, usually by including a component that bears free carboxyl or amino groups [170, 171]. However, these are not truly bound films – the protein is bound to the polymer, but the polymer is still only connected to the metal electrode by intermolecular forces, and is subject to delamination at any time. This thesis goes beyond that limitation by ensuring that a true bond is formed between polymer and metal. Because polythiophenes have the same backbone as PEDOT and can be synthesized in a regioregular fashion (see Section 3.3), they should also show the same resistance as PEDOT to slow conductivity loss, overcoming the third identified limitation of conductive polymer electrodes [172].

3.2 Lipophilic Polythiophenes for Intracellular Access

The use of PT SAMs oﬀers an interesting opportunity, in that the membrane of the attached neuron is drawn very close (50-100 nm) to the electrode [173, 174]. With this level of proximate contact, intracellular recording could easily be achieved, if substantial membrane disruption could be avoided. Neurons are clearly able to tolerate small holes in their membranes; they contain both gated and constantly-open ion channels, as well as other protein pores, which maintain regular exchange of

38 substances between the intracellular and extracellular spaces. Therefore, if a small (and presumably lipophilic) conductive molecule could be inserted into a cell membrane while coupled to a larger electrode, it could provide access to intracellular signals without killing the target cell. Combining that line of thought with the fact that PTs can be synthesized with many hydrophobic side chains, this thesis proposes the concept of inserting individual PT “wires” through the neuron membrane to access the intracellular potential. The concept is diagrammed in panel B of Figure 3.1. It is supported by previous demonstrations of a polymer that can insert itself into lipid membranes, although that polymer was designed to seal the membranes of damaged cells [175]. This “molecular wire” recording technique would simultaneously address many of the present neuro-robotic interface problems identified in Chapter 2. By maintaining intimate contact between neuron and electrode at all times, the effects of micromotion and ensemble shift would be reduced or eliminated, thus reducing or eliminating electrode failure. A stimulatory neuro-robotic interface would be ensured of one-to-one electrode-to-neuron connection, thus eliminating the current overlap/overexcitation mechanisms believed to underlie loss of neuronal excitability. Moreover, as suggested by results in rats, this highly specific stimulation may be more able to activate complex motor programs [133]. While the concept of PT-based intracellular access is simple, the hydrophobic outer residues of membrane channels are highly evolved and fairly well-conserved, suggesting that not just any PT will do; the side chains must be chosen to optimize compatibility. Preliminary solubility data, using hexanes as a proxy for membrane lipids, were able to identify some candidate polymers for study. In this thesis, two model polymers are used: poly(3-(2-ethylhexyl)-thiophene) (EHPT), which is soluble in hexanes, and poly(3-hexylthiophene) (HPT), which is only poorly soluble in hexanes. These two models can be used to test the concept that PTs can insert into lipid bilayer membranes and conduct current across those membranes. Further work will then be needed to optimize the polymer selection and appropriately balance any trade-offs between, e.g., lipid solubility, conductivity, and ease of synthesis.

3.3 Why Polythiophene?

The preceding sections have explained the value of conductive polymer SAMs as a potential neuro- robotic interface electrode and suggested how such SAMs could be used to establish intracellular recording. However, the question still remains: why polythiophene? There is a wide universe of conductive polymers, but polythiophenes are particularly well-suited for this work. They combine two valuable traits: high conductivity and high versatility. Polythiophenes, when synthesized in a regioregular head-to-tail (RR-HT) fashion (as shown in Figure 3.2), are among the most highly conductive polymers available [176]. Figure 3.3 shows how the RR-HT coupling is the essential ingredient for this high conductivity. Connecting monomers with long side chains in a head-to-tail fashion encourages the rings to remain roughly coplanar (a small dihedral angle between the ring planes). Rings with small dihedral angles relative to each other have a high degree of π-orbital overlap, which delocalizes electrons along the backbone, providing the necessary mechanism for electrical conduction. This same eﬀect is not readily achieved with other conductive polymers such as PPY. The pyrrole monomer can be functionalized with alkyl chains in a similar fashion, but attempts to polymerize poly(alkylpyrrole) tend to fail – pyrrole monomers are more reactive than thiophene monomers, and will react with trace impurities to produce tarry, low molecular weight product [172]. As noted in Section 3.1, PTs are also more stable than PPYs once synthesized, in part due to the lower reactivity of the monomer [172]. Polythiophenes are also the most versatile of the conductive polymers. Once synthesized, they can be functionalized with many end groups, as mentioned in Section 3.1. The side chains can also be tailored to have a wide range of properties. This thesis concerns itself with two relatively simple alkyl side chains, but a much wider variety is available, including side chains with chemically active groups [177]. Through block or random copolymerization, it is possible to combine multiple properties (such as hydrophobicity and hydrophilicity) within the same PT molecule while still

39 R

S * * S

* * n * * R S S S S Figure 3.2: Poly(alkylthiophene) Head-to-Tail Head-to-Head repeat unit

Figure 3.3: Illustration of regioregular polymerization

Illustrations of the chemical structure of poly(alkylthiophene)s and the importance of regioregular head-to-tail (RR-HT) coupling. Figure 3.2 shows the RR-HT repeat unit in the energetically favored anti conformation. Figure 3.3 illustrates why the RR-HT conﬁguration optimizes PT conductivity. The head-to-tail conﬁguration allows the side chains to avoid interacting, even if the backbone angle rotates into the syn conformation diagrammed here. By contrast, regioirregular polymers or regioregular non-HT polymers will include head-to-head (HH) couplings. As shown by the arrow in Figure 3.3, HH coupling between thiophene rings ensures that the side chains will either collide or be highly strained as the inter-ring bond rotates. This energetically unfavorable phenomenon will force the rings away from coplanarity, decreasing the atomic orbital overlap and lowering the polymer’s conductivity. achieving a high degree of polymerization [178]. Properly designed side chains can move beyond solubility and make the polymer’s conductivity sensitive to the presence of ions, small organic molecules, and even biomacromolecules [179]. While not critical to this thesis, this wide range of options will be important for future studies. This work presents a proof of concept, whereas future studies of PTs for neuro-robotic interfacing (or other biosensing) will need to develop PTs optimized for particular applications. Having a larger chemical space to search makes it more likely that these future investigators will be able to locate a PT that meets their needs.

3.4 Roadmap of Experiments

In summary, polythiophenes offer greater stability and versatility than previously-demonstrated neuro-robotic interface polymers, particularly in the availability of a large number of side chains and end groups. That versatility will be exploited in this work to combine the advantages of self- assembled monolayers and conductive polymer films while negating the disadvantages of each. It will also be used to pave the way for a new type of intracellular recording. These new approaches should produce a new polymer-based electrode that is as biocompatible as existing polymer-based electrodes, but that is more robust and offers a more specific connection between electrode and neuron. The two components of this new electrode technology, self-assembled PT monolayers and PT- based transmembrane conduction, will be addressed in series. To address the former, three things must be shown: that PTs can form SAMs, that these SAMs can be biocompatible, and that the biocompatible SAMs lower electrode impedance just as electrodeposited films do. End-thiolated PTs will be shown to be capable of forming pure and mixed SAMs that can support coupled proteins in Chapter 4, and the nanostructure of those SAMs will be analyzed. Biocompatibility with primary rodent neurons (one of the most stringent in vitro models) is demonstrated in Chapter 5, and the impedance of those same films is measured and described in Chapter 6. The model PTs used in

40 this thesis are not able to provide quite the same level of impedance decrease as is seen with thick films of PPY or PEDOT, but their other advantages may compensate for that limitation, and better choice of polymer should improve the performance. For intracellular electrical access, the critical points to demonstrate are insertion of a PT into the lipid bilayer and proof of its electrical conductivity when inserted. These will be tackled in reverse order, using a simplified model system so as to minimize confounding factors. Chapter 7 shows that current flows across an artificial lipid bilayer when it is exposed to EHPT, but not when it is exposed to HPT. To prove that this current is due to EHPT inserting into the membrane, Chapters 8 and 9 show the results of high-resolution computational modeling of the PT/membrane system. Taken together, these results not only prove the concept of PT-based intracellular recording/stimulation, but also provide in silico and in vitro methods for easier evaluation of future candidate polymers before they advance to cell-based or in vivo testing.

41 42 Chapter 4

Nanostructure of EHPT Self-Assembled Monolayers

The first step in constructing the proposed PT SAM electrode is verification that thiolated PTs are able to form pure and mixed SAMs, and that proteins can be coupled into those SAMs to provide biocompatibility. Many methods exist to detect the presence of a thin film on a metal surface, including Fourier transform infrared spectroscopy (FTIR), X-ray photoelectron spectroscopy (XPS), electrical impedance measurements (EIS, discussed further in Chapter 6), and quartz crystal microbalance gravimetry (QCM). However, in this case, the question is not merely whether a film is present – the nanostructure of the film is also important. The conceptual sketch given in Figure 3.1 shows the PT molecular wires standing roughly straight up from the electrode surface. While this is certainly the most common and expected configuration for organic SAMs, it is not the only possibility – PTs may be flexible enough to “lie down” on the surface instead. If that occurs, coverage density will be much lower and the proposed electrode may not function as desired. Testing for this possibility requires a technique that can physically map the SAM’s surface. The gold standard technique for assessing the topology of a film at the nanoscale is scanning probe microscopy (SPM), in which a nanometer-resolution tip is scanned across a sample of interest. Many SPM methods exist, and their various merits and methods of interrogating the sample will not be reviewed here. This work primarily uses atomic force microscopy (AFM), in which the probe makes physical contact (either continuously or in brief “tapping” contacts) with the surface. Brief investigations were also made using scanning tunneling microscopy (STM), in which the probe detects surface height by the magnitude of a tunneling electrical current, but the available equipment is not capable of resolving the SAMs under study. This chapter presents the results of AFM investigations of pure and mixed EHPT SAMs and discusses the implications of those results for the biocompatibility and impedance studies of future chapters. Methodological issues are also discussed in detail, with particular attention to techniques for preparing the atomically flat gold substrates needed for AFM studies of SAMs. Some exploratory data not directly relevant to the main investigative thrust were also collected, relating to the self- assembly of non-thiolated EHPT and preliminary attempts to dislodge or delaminate the observed thiolated EHPT SAMs. The unsuccessful STM investigations are also reviewed in order to establish the limits of what can be achieved with the equipment and techniques available.

4.1 Methods and Materials

This section describes the preparation of appropriate substrates for AFM of PT SAMs, the formation of those SAMs, and the AFM parameters used to study the resulting samples. Formation of PT SAMs and their subsequent functionalization with protein represent two steps in a larger chain for

43 Self-assemble organic monolayers from solution (a) Gold Electrode (b) Gold Electrode

Activate SAMs with reactive esters

Couple protein

(d) Gold Electrode (c) Gold Electrode

Plate primary rodent neurons and incubate

(e) Gold Electrode

Figure 4.1: Assembly schematic for SAM-based neuron/electrode interface. (a) and (b), formation of mixed EHPT/MHA SAM by incubation in organic thiol solution. (c), formation of reactive NHS esters on MHA carboxyl groups. (d), capture of adhesion-promoting biomolecules by NHS ester reaction with free amino groups. In this chapter, the protein is a model antibody with no adhesion activity. (e), quenching of unreacted esters and plating of primary neurons that bind to the coupled adhesion molecules. Step (e) is not performed in this chapter, and will be considered in Chapter 5.

assembly of the proposed PT SAM electrode. The complete assembly process is diagrammed in Figure 4.1.

4.1.1 Atomically Flat Gold Substrates for Atomic Force Microscopy

In order to visualize the topology of EHPT SAMs (or any SAM), the underlying substrate must be “atomically flat”, with roughness on the order of a few atom diameters. For the gold substrates used in thiol SAM chemistry, this implies a crystalline Au(111) surface. The atomically flat surfaces for this work were prepared through a template-stripping procedure adapted from [180]. Freshly- cleaved ruby mica sheets were heated to 465◦ C under a vacuum of ≤ 10−6 torr on the stage of a Denton Systems thermal evaporator. 1,300 A˚ of gold (99.99% purity, Kurt J. Lesker Corporation) were deposited by evaporation from a tungsten boat at a rate of 1 A/s˚ for the first 200 A˚ and 3 A/s˚ thereafter. The deposited gold was further annealed overnight at the deposition temperature. After annealing, the stage was turned off and the gold-coated mica sheet was allowed to cool for at least an hour before venting of the chamber and recovery of the sample. The mica sheet was then cut into approximately 1 cm squares. To prepare template-stripping “sandwiches”, 8 mm diameter round glass coverslips (ProSciTech) were washed in 8 N nitric acid, glass-distilled water, acetone (stored over molecular sieves), and absolute ethanol to remove oil and other contaminants. The mica squares were placed onto a hotplate set at 90◦ C, with the gold-coated side facing up. A small drop of solvent-resistant photocurable

44 epoxy (SU-8 2007, Microchem) was then applied to the center of each mica piece and allowed to rest briefly to evaporate solvent. One coverslip was then placed atop each epoxy drop; with the aid of the heating, the SU-8 rapidly reflowed to form a thin layer between glass and gold. The SU-8 was cured by exposure to light from a 3.3 mW/cm2 UV arc lamp for at least 1 hour (total dose approximately 12 J/cm2). Once the completed sandwiches were formed, they were stored for periods of days to months before use. At the time of use, gentle mechanical pressure applied to the corners of the mica square caused the glass, SU-8, and gold to delaminate from the mica, exposing a clean, high-quality Au(111) surface. These surfaces were immediately placed into SAM-forming solutions to minimize contamination. The epoxy used to form the templated sandwiches is a critical component of the process. The initial process given in [180] used an epoxy that swells in strong organic solvents, and took advantage of that property – the original method of stripping the gold from the mica was a brief solvent soak. While acceptable for applications in which the SAM will then be deposited from gentler solvents such as ethanol, this is entirely unacceptable for the application here, where the gold/epoxy/glass sample must spend at least 8 hours immersed in a strong organic solvent. Such extensive swelling will also delaminate the epoxy from the glass and may even dissociate the gold/epoxy interface. The result is a gold surface that is still very flat, but that is rolled up into a tube due to internal film stresses, and therefore unusable for AFM or any other application. SU-8 was used for this work due to its well-known impermeability to most solvents. However, even SU-8 will occasionally delaminate, especially if the glass coverslip is not completely clean and dry. This led to the discarding of several otherwise-good samples along the course of these experiments. Further testing determined that EpoTek 353ND (Epoxy Technology) is an even better choice – it is as resistant to solvent, but does not require the heating or UV exposure steps, which likely reduces internal film stresses and helps avoid delamination. For the sake of consistency, all samples reported in this chapter were prepared with SU-8, but EpoTek 353ND would be recommended to any investigator seeking to replicate or continue this work.

4.1.2 Formation of Self-Assembled Monolayers on Gold Surfaces

Self-assembled monolayers were formed on template-stripped gold from solutions containing mixtures of 16-mercaptohexadecanoic acid (MHA, Sigma, received as 90% and recrystallized from HPLC- grade chloroform) and end-thiolated poly(3-(2-ethylhexyl)thiophene) (EHPT, MW approximately 3,000 g/mol, synthesized following the method of [166]). All solutions were prepared in HPLC-grade toluene, which was chosen as the strongest solvent that had no observable swelling effect on SU-8. Solutions for single-component SAMs were 3 mg/ml of either component (approximately 1 mM in EHPT and 10.4 mM in MHA). The mixed 1:1 SAM used in these experiments was 3 mg/ml in both components. The molecular proportions in the final SAM were expected to be different from those in solution. Higher molecular weight compounds chemisorb faster from mixed solutions, implying that EHPT should be enriched above its solution fraction [169, 181]. Solutions were prepared in glass scintillation vials that were cleaned with a “base piranha” solution (1 part 30% hydrogen peroxide, 1 part ammonium hydroxide, 5 parts distilled water, heated to 80◦ C) to remove all traces of organics, then rinsed thoroughly with distilled water and dried overnight in a 120◦ C oven. This solution is recommended over the more traditional piranha formulation of peroxide and sulfuric acid due to its better safety profile – acid piranha is a spontaneous reaction that will proceed until all reactants are gone, whereas base piranha can be rapidly quenched and/or restarted by the removal or addition of heat. Given acid piranha’s legendary capacity to react violently and/or explode, the controllable base piranha is less troublesome, particularly in the hands of inexperienced undergraduates. Freshly stripped templated gold samples, as described above, were immersed in SAM-forming solutions in sealed vials for at least 36 hours and protected from light during SAM formation. After SAM formation, substrates were rinsed thoroughly in toluene and chloroform to remove unbound material. They were then soaked for at least 30 minutes in fresh toluene to further loosen

45 excess molecules, rinsed again in toluene and absolute ethanol, then blown dry with pre-purified nitrogen. All rinses and drying were conducted with extreme care, as the gold can be sheared off the substrate with excessive force, rendering the sample unusable. Samples were handled with fine-point forceps, and every effort was made to handle only the edges, so as to leave intact gold and SAM in the center of each circle. Once dry, each sample was placed into an individual carrier to shield it from dust, and the carrier was wrapped in aluminum foil to protect the SAM from damage by UV light. SAMs were stored in this fashion for up to 3 months while images were acquired.

4.1.3 Protein Coupling into Self-Assembled Monolayers

SAMs containing MHA were protein-functionalized in order to mimic the protein-presenting SAMs tested for biocompatibility in the work of Chapter 5. Following a protocol adapted from [121] and [157], the terminal COOH groups of MHA were activated to highly reactive NHS esters, which then captured primary amine groups on protein. The SAM-bearing gold samples were immersed for one hour in an absolute ethanol solution of 45 mM N-hydroxysuccinimide (NHS, Sigma) and 68 mM N-(3-Dimethylaminopropyl)-N’-ethylcarbodiimide hydrochloride (EDAC, Sigma). They were rinsed with ethanol and glass-distilled water, then gently blown dry with nitrogen. 30-50 µl of 100 µg/ml goat anti-mouse IgG (Sigma) in phosphate-buﬀered saline (PBS) was spotted onto the center of each activated SAM and allowed to couple for one hour. Excess protein and salt were removed by a brief rinse in glass-distilled water, followed by another gentle nitrogen drying.

4.1.4 Tapping Mode Atomic Force Microscopy

SAMs on template-stripped gold were mounted onto stainless steel AFM pucks using adhesive disks, and were retained on these pucks until the completion of all experiments. Images were acquired using a Digital Instruments MultiMode AFM with NanoScope IIIa controller, using an “EV” vertical- engage scanner column. The majority of images described in this chapter were acquired in tapping mode, using silicon cantilevers with nominal force constants of 42 N/m (Veeco/Digital Instruments). Before acquiring any images, the cantilever was cleaned for 10 minutes in a UV-ozone (UV-O3) cleaner. The cantilever and sample were then mounted in the AFM head, and the instrument was sealed with a measurement hood. The hood was purged with dry nitrogen, and both sample and cantilever were allowed to equilibrate in this environment for approximately 10 minutes before imaging began. All imaging sessions began with the acquisition of a bare template-stripped gold image in order to verify that the AFM tip was in good condition and free of attached particulates. Initial engagement with the sample occurred using the tuning parameters suggested by the Digital Instruments SPM software (Version 5.12r3). Once true engagement was verified, the drive voltage and deflection setpoints were adjusted to minimize the force applied to the sample. This was done based on the theory that these soft SAMs should be quite deformable, and application of any more force than necessary might distort the observed topography. To minimize the sample force, the deflection setpoint was manually increased (to at least 1 V of measured deflection signal) and the drive voltage manually decreased while watching the oscilloscope traces of the AFM. When loss of surface engagement was observed (signified by a flat and featureless oscilloscope trace), the drive voltage was increased by two or three steps (allowing the software to automatically control the step size) to restore engagement. Generally, the drive voltage would need to be further increased as the session continued, presumably due to effects of cantilever heating. Images were acquired using a 1 Hz scan rate. Faster scan rates obscured some details, while slower rates caused the controller to partially lose engagement over the course of one image, distorting and blurring the observations.

46 4.1.5 Exploratory SPM Techniques

While tapping mode AFM was chosen as the primary imaging mode, other modalities were also considered. Contact mode AFM was tested to see if the observed information was any different from that seen under tapping mode. Triangular silicon nitride cantilevers were used for contact imaging, with deflection (height) and friction images acquired simultaneously. The cantilever was scanned across the sample at 90◦ to the cantilever’s long axis, in order to maximize the friction. Efforts were made to minimize the applied force as for tapping mode, but acquiring a good friction image required a higher contact force. STM is able to resolve single atoms under optimal conditions, and thus was tested on these samples in an attempt to visualize individual molecules within the SAM. The samples were mounted on steel pucks as before, but silver-filled epoxy was applied around the edge of the puck and in contact with the upper gold surface, in order to establish electrical contact to the gold and the SAM. The quality of contact varied, but never showed a resistance above 15 Ω. STM tips were mechanically formed from Pt/Ir wire by partially cutting the wire and then pulling free a new tip. A 2 V bias was applied between tip and sample, and the tunneling current setpoint was 1-2 nA (the lowest achievable without a specialized low-current STM head).

4.2 Results

4.2.1 Tapping Mode AFM

Tapping mode AFM images of pure and mixed SAMs of EHPT and MHA, along with the bare template-stripped gold (TSG) substrate are shown in Figures 4.2-4.7. As seen in Figures 4.4-4.5, EHPT SAMs produce a granular surface appearance, with grains roughly 10-20 nm in diameter and with an apparent height of 10-15 nm. The precise height of these grains cannot be determined from these images, as the underlying substrate is effaced. This new roughened surface has an RMS roughness of 1.179 nm (compared to an RMS roughness of 0.481 nm on bare TSG) and a calculated surface area of 260,944 nm2 in a 500 nm scan square (compared to 253,808 nm2 for TSG alone). By contrast, SAMs formed purely from MHA, shown in Figures 4.6-4.7, are barely distinguishable from the bare TSG. This is to be expected at the present resolution, since individual MHA molecules are much smaller than the cantilever tip and should form a tightly packed monolayer that mirrors the underlying topography. MHA’s presence is nonetheless verifiable by its ability to participate in protein coupling reactions, the results of which are seen in Figures 4.10-4.11. After coupling, the height variation across the surface increases and the underlying gold structure is obscured. Figures 4.8-4.9 illustrate the effect of SAM formation from a mixed solution. These images show characteristics of both the MHA and EHPT pure SAMs, but favoring EHPT. The same 10-15 nm tall granules are seen as were found with EHPT, but the fissures between gold terraces are now also visible. Once the protein coupling reaction is performed (Figures 4.2.1-4.13), this surface appears almost identical to a pure MHA SAM with coupled protein. To test the robustness of these SAMs, a pure EHPT SAM was subjected to 10 minutes of UV-O3 cleaning. The result of this treatment is shown in Figures 4.14 and 4.15. The same granular appearance is seen as in the untreated EHPT SAM of Figures 4.4-4.5, indicating that this level of exposure was not able to remove the SAM. At the same time, there is an impression of finer (smaller diameter) grains, and the grains are shorter (maximum height on the order of 10 nm, compared to 15 nm for untreated). This suggests that the SAM did experience some damage during the cleaning process, which is to be expected given the long duration of exposure. The images presented in this section were acquired from films that were stored for up to 3 months, as noted above. During this time, several of the films were re-imaged in order to improve the quality of the final report. No changes were observed during this period, providing verification that the storage method was adequate to preserve the SAMs for study.

47 Figure 4.2: Bare gold, height Figure 4.3: Bare gold, phase

Figure 4.4: EHPT SAM, height Figure 4.5: EHPT SAM, phase

Figure 4.6: MHA SAM, height Figure 4.7: MHA SAM, phase

Atomic force microscopy of pure EHPT and MHA SAMs on template-stripped Au(111). All images were acquired in tapping mode with a scan speed of 1 Hz. 4.2 and 4.3, bare gold. Wide and ﬂat terraces with smaller subgrains are seen; the tall specks are presumed to be dust particles adsorbed to the surface. 4.4 and 4.5, EHPT SAM. The underlying terraces are concealed by this thick, granular ﬁlm. 4.6 and 4.7, MHA SAM. This tightly-packed thin organic monolayer essentially mirrors the underlying terrace structure.

48 Figure 4.8: 1:1 EHPT:MHA, height Figure 4.9: 1:1 EHPT:MHA, phase

Figure 4.10: MHA with coupled protein, Figure 4.11: MHA with coupled protein, height phase

Figure 4.12: 1:1 EHPT:MHA with cou- Figure 4.13: 1:1 EHPT:MHA with coupled protein, height pled protein, phase

Atomic force microscopy of mixed EHPT:MHA SAMs, continuing from previous page. 4.8 and 4.9, mixed 1:1 EHPT:MHA SAM. This strongly resembles an EHPT SAM, with polymer granules seen across the entire surface. Unlike pure EHPT, the underlying gold is not entirely eﬀaced. 4.10 and 4.11, MHA SAM after IgG coupling. Large globular protein molecules (likely including small aggregates) cover the surface. 4.2.1 and 4.13, mixed 1:1 EHPT:MHA SAM after protein coupling. The surface now matches the MHA/protein image, showing the same irregularly globular structures. 49 Figure 4.14: EHPT after UV-O3 treat- Figure 4.15: EHPT after UV-O3 treatment, height ment, phase

Tapping mode AFM of EHPT SAMs after 10 minutes of UV-O3 treatment. EHPT is still present, but the grain diameter is slightly reduced, and the grains are about 5 nm shorter than an untreated SAM. The underlying terraced structure of the gold substrate is not made visible despite this damage to the SAM.

4.2.2 Self-Assembly of Non-Thiolated EHPT

In an attempt to verify that the EHPT SAMs observed in this work were formed by binding of the terminal thiol to gold, the SAM-forming protocol was also carried out using a 3 mg/ml solution of unthiolated EHPT. AFM images of this “unthiolated SAM” are shown in Figures 4.16 and 4.17. These films had a tendency to acquire very large dust grains (which may also be large EHPT aggregates) that obscure the image, but in between these large features one can see similar grains to those seen in the thiolated EHPT SAMs of Figures 4.4-4.5. This indicates that binding may not be entirely through the terminal thiol, and that other mechanisms (nonspecific adsorption of aggregates, or binding of the sulfur on the thiophene backbone) are playing a role. To test the “nonspecific adsorption of aggregates” theory, a separate “unthiolated SAM” sample was sonicated for 5 minutes in fresh toluene before imaging. The images from that sample are shown in Figures 4.18 and 4.19. The large grains of Figures 4.16-4.17 are gone, revealing a field of grains very similar to that seen with thiolated EHPT in Figures 4.4-4.5. Some dark lines are also visible in the height image, which may be fissures between gold terraces, but this is not certain. It appears that, at least on these flat and clean gold surfaces, mechanisms other than thiol self-assembly can drive formation of an EHPT monolayer or near-monolayer.

4.2.3 Exploration of Non-Tapping AFM Techniques

Contact AFM images of pure EHPT and MHA SAMs are shown in Figures 4.20-4.25, and are qualitatively similar to their counterparts in Figures 4.2-4.7. (Deflection images are equivalent to height images, and friction images are roughly equivalent to phase images.) The bare TSG shows the same wide flat terraces, with occasional dust grains. EHPT shows the same granular “bump field”, with equivalent grain diameter and height between contact and tapping modes. The principal difference is that the grains on contact imaging are less well-resolved, and the image overall contains an impression of streaks (even after the erasure of occasional scan lines). This is most likely due to the greater force necessary to obtain a friction image, which can cause the tip to drag and catch in the polymer. The MHA SAM contact images in Figures 4.24-4.25 show similar frictional effects. Like their tapping mode counterparts, they mirror the structure of the underlying gold terraces, but the image appears blurry compared to Figures 4.6 and 4.7. This generally poorer image quality was the reason for using tapping mode to acquire the majority of images in this chapter.

50 Figure 4.16: Unthiolated EHPT, height Figure 4.17: Unthiolated EHPT, phase

Figure 4.18: Unthiolated EHPT after Figure 4.19: Unthiolated EHPT after 5 min sonication, height 5 min sonication, phase

Tapping mode AFM of ﬁlms formed from EHPT without a terminal thiol group. 4.16 and 4.17 show the images as obtained after forming and gentle rinsing. Large particles (which may be EHPT aggregates) obscure the picture, but some grains are seen, similar to what was observed with SAMs from thiolated EHPT. 4.18 and 4.19 are another sample (formed from the same solution) after 5 minutes of ultrasonic cleaning in toluene. With the dust particles removed, the grains are much more clearly visible. The height image (4.18) shows some faint suggestions of crevasses demarcating Au(111) terraces, but these cannot be seen on the corresponding phase image (4.19). The overall impression is that EHPT can form granular monolayers on template-stripped gold even without a terminal thiol group.

51 Figure 4.20: Bare gold, deﬂection Figure 4.21: Bare gold, friction

Figure 4.22: EHPT SAM, deﬂection Figure 4.23: EHPT SAM, friction

Figure 4.24: MHA SAM, deﬂection Figure 4.25: MHA SAM, friction

Contact mode AFM studies of pure EHPT and MHA SAMs. When compared with their counterparts of Figures 4.2-4.7, these images reveal the same structure (grains and pillars for EHPT, mirroring of the gold surface for MHA). However, the image quality is lower, with the images appearing streaked and blurred. This lower quality is attributed to greater friction between polymer and AFM tip.

52 In contrast to the contact mode studies, the STM images of Figures 4.26-4.28 show no resemblance to the tapping mode images of Figures 4.2-4.7. All three images show the terraced structure of the template-stripped gold, with no EHPT grains visible. In fact, the MHA image has more height variance (presumed due to dust grains) than the EHPT image. None of the images shows features that can be attributed to individual atoms. This suggests that the STM tip is not tunneling current through the SAM molecules, but is instead pushing into the SAM in order to tunnel current to the underlying gold. While this is an unfortunate result, it is not unexpected – visualization of SAMs, particularly insulating SAMs like MHA, requires a low-current STM head capable of reliably measuring currents on the order of 1-5 pA. The STM head available for this study had a lower limit of 1-2 nA, three orders of magnitude higher. Without the higher-resolution equipment, there was little chance of obtaining valuable data from the STM studies.

4.3 Discussion

The AFM results show a roughening of the gold surface by EHPT SAM formation, as expected and desired. This increased surface area should lower the impedance of a coated electrode, providing one of the principal advantages of conductive polymers for neuro-robotic interfacing. Based on the EHPT SAM’s ability to resist UV-O3 degradation, EHPT-containing SAMs are expected to be robust and able to survive cell culture conditions. The observed film morphology is similar to non-alkylated polythiophene films that have been electrochemically grown from surface-immobilized precursors, although these SAMs are thinner and show a smaller grain size [182]. The images show a grain height of 10 to 15 nm, but this may be an underestimate of the true value, since the force of the cantilever may be perturbing the film even at the low drive voltages used. The grains have a diameter between 10 and 20 nm. Given that the diameter of an EHPT molecule should be on the order of 2 nm, the observed grains are suspected to be aggregates, not single molecules. This is unsurprising, as EHPT is known to form intermolecular aggregates even in dilute solution [183]. Alternatively, they may be single-molecule intramolecular aggregates. Some investigators have theorized that regioregular PTs exposed to poor solvents will undergo a one-dimensional collapse, forming a shortened helical structure [184]. AFMs of these putative helical aggregates show grains not dissimilar to those seen in this work [184, 185]. Resolution of the would require a method that can resolve individual PT chains. STM is the logical choice, but the STM available is not capable of atomic resolution on PT or MHA SAMs. The nature of the observed grains thus remains a question for future investigations. The mixed 1:1 EHPT:MHA SAM demonstrates the expected effect of EHPT enrichment in the SAM (as compared to its relative concentration in solution). These images do not measure composition, and it is possible that there is substantial MHA present between the EHPT aggregates, but the overall impression from Figures 4.8-4.9 is of a SAM that is principally EHPT. At the same time, the presence of MHA in this SAM is clear from its ability to bind protein, as illustrated in Figures 4.2.1-4.13. The level of protein binding makes the mixed SAM appear essentially identical to the pure MHA SAM, despite its high EHPT content. This is an encouraging finding, since it suggests that these mixed SAMs will be able to combine high conductivity with high biocompatibility, a prediction that will be explored and at least partly confirmed in the following chapters. The observation that non-thiolated EHPT can produce similar SAMs is somewhat concerning, particularly since the thiolated EHPT samples used in this work did contain a substantial fraction of unfunctionalized polymer. However, upon careful comparison of the images from thiolated and unthiolated EHPT, the unthiolated sample produces a sparser field of grains, indicating that it is not coupling as readily. This would not be surprising, since the only available coupling element on an unthiolated PT is the sulfur of the backbone rings, and those sulfur atoms are already participating in multiple bonds. Even more important is to remember that the gold substrates used here are extremely clean, flat, and pure. Real electrodes never present anything approaching this surface quality – they are a much rougher gold deposited by sputtering or evaporation, and their gold often has adsorbed contaminants even after cleaning. A relatively unfavorable (and highly sterically hindered) binding between thiophene-ring sulfur and gold may happen on a template-stripped surface

53 Figure 4.26: STM of bare gold

Figure 4.27: STM of EHPT SAM Figure 4.28: STM of MHA SAM

Scanning tunneling microscopy (STM) studies of pure SAMs on template stripped gold. Although the pattern of terraces is different in each figure, the EHPT grains are not visible in Figure 4.27, and the alkyl chains of MHA cannot be seen in Figure 4.28. This indicates that the STM equipment available for this work was not able to effectively image these monolayers, and is instead imaging the underlying gold surface.

54 but be impossible on an electrode. In fact, this is precisely what will be seen when the same compounds are tested in Chapter 6. The observation of a field of grains on exposure of TSG to unthiolated EHPT can therefore be considered of lesser consequence, and does not detract from the ultimate conclusions of this work. In summary, this chapter has demonstrated that end-functionalized EHPT can bind to a gold surface, form mixed SAMs with MHA, and bind protein to those mixed SAMs using standard surface chemistry. Preliminary evidence has also been presented for the robustness of the films. In order to reach applications in neuro-robotic interfacing, it still remains to be shown that these films can achieve the biocompatibility and impedance improvements seen with other conductive polymer electrode coatings. Demonstrating these properties is the subject of the following two chapters.

55 56 Chapter 5

Biocompatibility of Protein-Containing EHPT SAMs

5.1 Introduction

With verification that EHPT can form SAMs on gold surfaces and that those SAMs can be functionalized to contain covalently bound proteins, the next step is proving that the protein-containing SAMs are biocompatible. In theory, one could simply implant coated electrodes into rat or primate brain, then sacrifice animals at successive timepoints and perform histology. However, that is a very expensive and time-consuming proposition, and it does not leave much leeway for fine-tuning inputs such as the SAM formation protocols or the choice of protein(s). A more logical first step for proof of concept is to assess compatibility using simpler in vitro methods, i.e., cell culture. Testing biocompatibility in vitro also removes potentially confounding variables from the system, including interactions of SAM-forming solvent with the materials of an MEA, effects of the MEA shape and micromotion, effects of surgical technique, and so on. Once the decision is made to test in vitro, the choice of cell culture protocol determines the generalizability of the results. There exist many immortalized cell lines that are used as models for neurons in cell culture, the most common being the PC12 pheochromocytoma line and the neuroblastoma lines. These cells are relatively easy to culture, and since they are continually dividing, one always has a ready supply. However, this ease comes at a price – in the process of mutating to become immortal and robust, neuron-like cell lines may have lost key characteristics of real neurons. Moreover, because they are so robust, they may be able to grow on a substrate that true neurons would find toxic. This work therefore takes a more rigorous approach by using primary neurons – non-mitotic cells harvested from late embryonic mouse brains. Primary neurons are harder to prepare, available in limited supply, and much more likely to die in culture and/or fail to respond to a substrate. However, if a positive response is seen, it is much more likely to correlate with a positive response in vivo. This chapter presents the results of culturing primary mouse neurons atop EHPT SAMs with varying levels of neuron adhesion protein, demonstrating their biocompatibility and shedding light on the significance of the AFM results from Chapter 4.

5.2 Methods

The process for forming SAMs for cell culture closely follows the one used for AFM samples in Chapter 4, and is provided for reference in Figure 5.1 (identical to Figure 4.1). The principal diﬀerences are the addition of the ﬁnal cell culture step, the use of translucent gold-coated coverslips instead of opaque template-stripped substrates, and the use of an adhesion protein instead of a

57 Self-assemble organic monolayers from solution (a) Gold Electrode (b) Gold Electrode

Activate SAMs with reactive esters

Couple protein

(d) Gold Electrode (c) Gold Electrode

Plate primary rodent neurons and incubate

(e) Gold Electrode

Figure 5.1: Assembly schematic for SAM-based neuron/electrode interface, reprised from Figure 4.1. (a) and (b), formation of mixed EHPT/MHA SAM by incubation in organic thiol solution. (c), formation of reactive NHS esters on MHA carboxyl groups. (d), capture of adhesion-promoting biomolecules by NHS ester reaction with free amino groups. In this chapter, the anti-NCAM antibody N-CAM13 is used. (e), quenching of unreacted esters and plating of primary neurons that bind to the coupled adhesion molecules. model antibody. Earlier parts of the protocol will be presented only to the degree that they diﬀer from what was done in Chapter 4, and the principal attention in this section will be given to the culture technique.

5.2.1 Translucent Gold-Coated Culture Substrates

To monitor cultured cells without ﬁxing and staining them, transparent or at least translucent substrates are needed so that transmitted-light phase contrast microscopy can be used. At the same time, the culture substrate must contain a layer of gold in order to support SAM formation. The solution is to deposit a thin layer of metal on one face of a glass coverslip. This will dim the transmitted image, but will still allow standard microscope techniques if the layer is thin enough. 8 mm glass coverslips (ProSciTech) were cleaned by serial washing in 8 N nitric acid, glass-distilled water, acetone (stored over molecular sieves), and absolute ethanol. Dry slips were mounted with polyimide tape (Fisher Scientiﬁc) onto silicon carrier wafers (Silicon Quest) and placed into the vacuum chamber of the same evaporator used for template-stripped gold deposition. The tape was positioned to cover a very small part (about 1/8 the total area) of the coverslip. Polyimide tape was chosen based on its lack of outgassing at high vacuum and its ability to withstand the high temperatures of evaporation. Earlier versions of this protocol attempted to attach the slips with unexposed photoresist, which was found to bubble and char during metal deposition and often caused coverslip breakage during removal.

58 30 A˚ of chromium were deposited as an adhesion layer (Cr-coated tungsten rods, Kurt J. Lesker Corporation) followed by 100 A˚ of gold, without breaking vacuum between films. Both films were deposited at a rate of 2.5 A/s.˚ Deposition rates and film thicknesses were monitored by the evaporator’s integrated QCM sensor. These film parameters produce a silver-gold colored translucent film with a slight reddish tinge on the chromium side. (The color difference can be used to reliably identify which surface is metallized.) After deposition, the films were cooled for approximately 30 minutes before breaking vacuum. Coated coverslips were stored in covered boxes, still attached to the strip of polyimide tape, until ready for use. Just before use for SAM formation, slips were subjected to 5 minutes of oxygen plasma cleaning, then demounted from the tape by careful handling with two sets of fine-tip forceps. In order to help identify slips with different SAMs, an identifying character was scratched onto the back of each slip using a diamond scribe. These gold-coated coverslips can also be cleaned using a base piranha solution (see Chapter 4 for recipe), but the vigorous foaming and gas production of that solution often delaminates the thin and fragile gold films. Oxygen plasma cleaning was therefore preferred as a gentler but equally effective method.

5.2.2 Formation of Self-Assembled Monolayers

Formation of pure and mixed EHPT and MHA SAMs on the coverslips followed the same protocol as described in Chapter 4. Once SAM coating and toluene/chloroform/ethanol rinsing was complete, the prepared SAMs were not stored, but were placed immediately into the EDAC/NHS/ethanol esterification solution. This continual immersion in non-aqueous solvents helped ensure sample sterility. For the work in this chapter, new mixed SAM compositions were tested in addition to the 1:1 EHPT:MHA mixture described in Chapter 4. These new mixed SAMs contained multiples of the base concentrations of 3 mg/ml in each component. For instance, the 1:2 EHPT:MHA SAM solution was prepared from 3 mg of EHPT and 6 mg of MHA per ml of toluene. This ensured that neither component dropped below 3 mg/ml concentration in the final solution, as lower concentrations might have diminished SAM-forming effectiveness.

5.2.3 Protein Coupling into Self-Assembled Monolayers

Protein coupling for cell culture again followed almost the same protocol described in Chapter 4. After NHS ester formation, the SAMs were not washed with water; they were instead rinsed twice with absolute ethanol and then allowed to dry in the bottom of a sterile polystyrene dish. The spotted protein was 30 µl of 100 µg/ml antibody against the neural cell adhesion molecule NCAM (antibody N-CAM13, produced in-house from hybridomas, available from BD PharMingen). N-CAM13 was diluted in phosphate-buffered saline (PBS), pH 7.3, dialyzed against PBS to remove any residual primary amines, and sterilized by filtration through an 0.22 µm syringe filter. Coupling was allowed to proceed for two hours, compared to one hour for the AFM experiments. The dish was sealed with Parafilm during protein coupling in order to prevent evaporation of these very small droplets. After the two hour coupling time, the dish was twice flooded with cold sterile PBS under vigorous agitation to rinse away unbound N-CAM13. Unreacted NHS esters were quenched by further incubation for at least 30 minutes in a fresh flood of PBS. SAMs of pure EHPT were subjected to this same protocol before culture; while this was not expected to lead to protein coupling, nonspecific adsorption is likely.

5.2.4 Primary Neuron Culture on SAMs

Timed-pregnant female Swiss Webster mice at 18 days gestation (Charles River Laboratories) were euthanized by CO2 inhalation and cervical dislocation, followed by harvesting of embryos by Caesar- ian section and dissociation of cortical neurons according to the protocol of [186]. Brieﬂy, embryos were removed from amniotic sacs and decapitated with scissors in cold sterile saline. The cranium was opened with ﬁne-tip scissors and/or forceps, and the cortical hemispheres exposed. Cortices

59 were lifted off the midbrain and hindbrain using forceps and placed into a culture tube of cold Hibernate-E (BrainBits LLC) or unsupplemented Neurobasal (Invitrogen) medium. When roughly four hemispheres were collected in one tube, the medium was removed and replaced with a pre- measured trypsin aliquot. Cortices were digested with trypsin for 10 minutes, then mechanically dissociated in DNAse. Dissociation was achieved by sequential trituration using two fire-polished Pasteur pipets, one at 2/3 of original diameter and one at 1/3 original diameter. After settling out non-dissociable particulates and digesting for four minutes, the DNAse supernatant was pipetted off and centrifuged for 5 minutes at 500-600 rpm to pellet out neurons. This pellet was resuspended in Neurobasal with B-27 supplement and 0.5 mM Glutamax, a stable glutamine analog (both from Invitrogen). A sample of this final suspension was counted in a hemocytometer under phase contrast, and the suspension was diluted with complete medium to a plating concentration of 0.5 × 106 phase-bright cells per ml. In some cases, the cortices were not immediately dissociated, but were stored for up to two days at 4◦ C in Hibernate-E with B-27 and 0.5 mM Glutamax. This medium has previously been demonstrated to permit storage of cortical tissue for up to one week without loss of neuronal viability [187]. Dissociation was then performed immediately before plating cells onto SAM-coated substrates. For the final plating of cells, coverslips were transferred to a 24-well culture plate, and 0.5 µl of the dilute neuron suspension was plated on each coverslip, corresponding to a density of roughly 250 cells/mm2. One hour after plating, each well was filled with approximately 0.5 ml of Neurobasal ◦ with B-27 and Glutamax. Cultures were maintained at 37 C in 5% CO2, and 50% of the medium was replaced after 3 days. All cultures were performed under an animal experimentation protocol reviewed and approved by the Institutional Animal Care and Usage Committees of both Carnegie Mellon University and the University of Pittsburgh.

5.2.5 Immunostaining and Cell Counting

At 3 and 7 days in vitro (DIV), coverslips were fixed and stained to assess neurite outgrowth. Cells were fixed for 10 minutes in 4% paraformaldehyde (Sigma), then immunolabeled for 1 hour with a primary rat monoclonal antibody against the mouse M6 neuronal antigen. Nonspecific binding was blocked by prior incubation with 10% bovine serum and 1% bovine serum albumin in Tris-buffered saline. Following removal of the primary antibody by washing in PBS, stained cells were incubated with a Cy3-labeled goat anti-rat secondary antibody for 1 hour, then washed again. All washes were performed by dipping the coverslip sequentially into two small beakers of PBS. After staining was complete, the back side (glass side) of each coverslip was gently wiped with a water-soaked Kimwipe to remove residual salt. Coverslips were then mounted with the gold face down on a glass microscope slide using a small drop of Mowiol mounting medium. Mounted slips were allowed to dry overnight before visualization. Neurons were visualized and photographed at 400x magnification under oil immersion, with confirmation of neuronal identity by M6 fluorescent staining. It should be noted that this high magnification and the use of oil immersion are necessary to visualize the neurons. Because the cells are on a gold substrate (which quenches the fluorescence signal) and are being visualized through the translucent (but attenuating) gold layer, the signal is weak and a high numerical aperture is needed. 30 fields were measured for each condition at each timepoint, with the fields being drawn from at least three coverslips. Neurite length was measured from captured fluorescence images using the program IPLab (version 3.9.4 r2 for Macintosh), calibrated using an image of a micrometer captured with the same optics. The longest process completely captured within each field was considered to be the primary neurite, and its length was measured from the end of the hillock to the most distal extent of the growth cone. In the event that a single neuron could not be captured within a single image, it was still considered as one field. After a field was captured in fluorescence microscopy for neurite length measurement, the optics were switched to phase contrast mode for counting of cell bodies (which stain only weakly with M6). Cell attachment counts were thereby taken from the same fields used for neurite length measurement.

60 Statistical analyses were performed in Microsoft Excel. Overall statistical significance of the results at 3 DIV and 7 DIV was assessed with a one-factor ANOVA; individual groups were then compared by the Neuman-Keuls procedure. Growth within a condition between 3 DIV and 7 DIV was tested with a two-tailed Student’s t-test for unequal sample variances. No significance level correction was employed, since this set of tests was pre-planned. The initial hypothesis was that the mixed SAMs tested should show performance somewhere in between that of pure EHPT or MHA SAMs, and that individual conditions should differ from themselves between 3 DIV and 7 DIV.

5.3 Results

Figure 5.2 displays the biocompatibility of various SAM formulations, as assessed by the willingness of primary neurons to extend neurites on those surfaces. All conditions show continued growth and viability up through 7 DIV, as demonstrated by significant increases in neurite length between these two timepoints (p < 0.0314). At 3 DIV, differences between conditions have not fully emerged; the overall ANOVA is significant (p = 8.213 × 10−8), but the only condition that differs from the others in a Neuman-Keuls pairwise analysis is the pure EHPT SAM (p < 4.624 × 10−5). At 7 DIV, the outgrowth ANOVA remains significant (p = 2.155×10−7), but three levels are now apparent. MHA now shows greater neurite growth compared to any other condition (p < 0.0087), and pure EHPT shows significantly less compared to any other (p < 0.0233). It should be kept in mind that the “pure” EHPT SAM has still been allowed to adsorb N-CAM13. If this adsorption step is eliminated, there is essentially zero survival on these surfaces. There is no significant difference between any of the mixed SAM conditions at 3 DIV or 7 DIV (p > 0.3287), although each group (and the pure SAMs) does show significant outgrowth between 3 and 7 DIV (p < 0.032). While it does appear from Figure 5.2 that there is a trend of increasing outgrowth with increasing MHA content in the mixed SAMs, this is not significant by linear trend analysis (p > 0.1722). Cell attachment to the various SAM surfaces was also tested as a measure of biocompatibility. Results at 3 DIV are reported in Figure 5.3; no measurements were taken from the 7 DIV cultures, as the staining protocol used often caused detachment of the cell bodies from the surface (leaving the neurites firmly attached and well-stained). The overall one-way ANOVA of the 3 DIV attachment data shows a significant between-groups difference with p < 0.01. However, this is entirely due to the decreased attachment seen on the 1:1 EHPT:MHA and 2:1 EHPT:MHA SAMs; these are the only groups that differ from their neighbors in the pairwise Neuman-Keuls analysis, with the largest p-value being p = 0.0176 for 2:1 EHPT:MHA compared with 1:2 EHPT:MHA. It is particularly noteworthy that there is no significant difference between the attachment to EHPT SAMs and the attachment to MHA SAMs, with the means being almost identical (p > 0.99). Given the results seen in the neurite outgrowth measurements and the substantially different morphologies observed on EHPT and MHA (see Figures 5.4-5.8), the lack of difference in neuron attachment suggests that attachment is not a sensitive measure of biocompatibility in this situation, and instead is measuring other factors (e.g., quality of dissociation or of plating). The neurite length measurements are reflected in the morphology of cells on the different SAM formulations, as illustrated in Figures 5.4-5.8. Decreasing MHA (and thus decreasing protein) content can be seen to produce shorter axons and, in the extreme, neurites that are no more than a very broad growth cone. Also visible in each image are scattered cell bodies that have not produced neurites. These cells contribute to the attachment counts reported in Figure 5.3, and their presence can explain the insensitivity of the attachment measure to the underlying SAM composition.

5.4 Discussion

The results presented in Figures 5.2 and 5.4-5.8 demonstrate that mixed SAMs of EHPT and MHA are suﬃciently neurocompatible to sustain continued neurite outgrowth up to 7 DIV. As predicted,

61 Neurite Outgrowth on Mixed SAMs of EHPT and MHA 250 3 DIV 7 DIV * 200 **

150

100 LengthNeuriteof Primary (um)

0 MHA 1:2 EHPT:MHA 1:1 EHPT:MHA 2:1 EHPT:MHA EHPT

Figure 5.2: Neurite outgrowth on pure and mixed EHPT/MHA SAMs at 3 and 7 DIV. Average length of each cell’s primary (longest) neurite is reported. At 3 DIV, only the EHPT group differs significantly from the others (p < 4.624 × 10−5). By 7 DIV, pure MHA shows significantly greater outgrowth than all mixed SAMs (p < 0.0087, denoted by *) and pure EHPT (with protein adsorbed) shows significantly less growth (p < 0.0233, denoted by **). There is no significant difference at 3 DIV or 7 DIV among the three mixed SAMs tested (p > 0.3287).

Cell Attachment on Mixed SAMs of EHPT and MHA 20

18 3 DIV

8 MeanCells perField

0 MHA 1:2 EHPT:MHA 1:1 EHPT:MHA 2:1 EHPT:MHA EHPT

Figure 5.3: Attachment of neurons on pure and mixed EHPT/MHA SAMs at 3 DIV. The reported count is the mean number of cells per high-power field, as assessed by phase contrast microscopy. There is an overall significant between-groups effect by ANOVA (p < 0.01), largely due to the 1:1 and 2:1 EHPT:MHA SAMs showing a marked decrease in attachment. Of note, pure EHPT and MHA show no difference in attached cell count, having nearly identical means (p > 0.99).

62 Figure 5.4: MHA SAM Figure 5.5: 1:2 EHPT:MHA SAM

Figure 5.6: 1:1 EHPT:MHA SAM Figure 5.7: 2:1 EHPT:MHA SAM

Figure 5.8: EHPT SAM (protein adsorbed)

Fluorescence microscope images of mouse cortical neurons at 3 DIV on the same pure and mixed SAM formulations reported in Figure 5.2. All images are 400x magniﬁcation with Cy-3 staining against the M6 neural antigen; the scale bar represents 20 µm. As the EHPT:MHA ratio increases, neurite outgrowth decreases. Of particular note is (5.8), which illustrates the short neurites with broadened growth cones that are common on the pure EHPT SAMs.

63 the quantitative performance of mixed SAMs falls in between that of pure SAMs of either component, and is significantly different from both. While longer-term survival of neurons on SAMs was not assessed in this work, prior experience with primary neuron cultures suggests that a culture lasting until 7 DIV can continue indefinitely as long as the medium is regularly changed. No difference was seen in cell attachment between SAMs, but, as noted above, this is most likely because attachment is not a sensitive measure of biocompatibility. Once the plating droplet is applied, neurons will settle on the surface through gravity, and can be held there by stiction forces (given the gentle handling applied to the plated slips in these experiments). The number of neurons that extend neurites may also be a poor measure – this depends on the cells not being permanently traumatized by the dissociation process, a factor which varies from run to run even with an experienced investigator. This leaves the measure of how well cells perform when they are able to extend neurites, and as seen here, that measure is sensitive. It is interesting and somewhat surprising that there is no significant difference in outgrowth between the three mixed SAMs tested. However, this may be explained by referring back to Fig- ures 4.8-4.9, which show the composition of the 1:1 EHPT:MHA SAM before protein coupling. These SAMs had a very similar appearance to that of a pure EHPT SAM, and this was interpreted as indicating that EHPT incorporates into the SAM faster than MHA. Following that line of rea- soning, EHPT will be enriched (relative to its solution fraction) in all three mixed SAMs studied here. Therefore, the composition of these SAMs is likely not as different as one would infer from the forming solutions. The AFM studies of Chapter 4 also revealed that coupling of protein to mixed SAMs can produce a surface that is not distinguishable from a pure MHA/protein surface by AFM. This effect could also blur the difference between different mixed SAM compositions. The fact remains that SAM composition is controllable, as evidenced by the different morphologies seen in Figures 5.4-5.8. These differences could most likely be made significant by using more extreme concentration differences or by increasing the number of samples. The principal barrier to performing such experiments is that they would require large amounts of thiolated EHPT, a compound which is highly nontrivial to synthesize. The experiments reported here and in the preceding and following chapters collectively consumed almost the entire supply of thiolated EHPT that could be obtained over the span of two years. The net result remains that mixed EHPT:MHA SAMs are biocompatible with primary neurons, which bodes well for future in vivo testing. This biocompatibility suggests that if EHPT does insert into neuron membranes (as will be demonstrated in Chapters 7 through 9), it does not cause substantial harm to the cells by doing so. Glia were present in the cultures but did not proliferate excessively, which might predict reduced astrocytic scarring around an implanted EHPT-coated electrode. (Low proliferation is to some degree attributable to the use of a serum-free medium, but more recent studies have shown that astrocytes do proliferate in Neurobasal/B27 [128].) The biocompatibility of the mixed SAMs compares favorably to the pure EHPT SAM, which does not support substantial neurite outgrowth. The mixed SAMs have therefore succeeded in meeting two of the three criteria laid out in Chapter 3 for an improved polymeric neuro-robotic interface electrode: they form robustly bound films and they achieve biocompatibility. It only remains to show that they can also lower electrode impedance, and that will be the topic of the next chapter.

64 Chapter 6

EHPT-Containing SAMs Lower the Impedance of Coated Electrodes

6.1 Introduction Electrodeposited PPY and PEDOT films most likely improve electrode recording performance by drawing neurons and their neurites closer to the electrode, as was reviewed in Chapter 2. The biocompatibility results of Chapter 5 indicate that EHPT-based SAMs (and, presumably, PT SAMs in general) should achieve the same effect. However, improved biocompatibility is only half of the benefit from conductive polymer coatings. The other half is lowered electrode impedance. While lower impedance may yet be shown to contribute to improved recording, its greater value is in safer stimulation. Existing conductive polymer films lower impedance by increasing the electrode area, which lowers both the resistive and capacitive components of the impedance (lower resistance and higher capacitance). This should permit passage of more charge through capacitive (non-Faradaic) current, which is believed to not be harmful to the surrounding tissue. PT SAMs should show similar effects in order to be an effective electrode material. That said, it is unlikely that PT SAMs will be able to reach the hundredfold impedance drops observed with the electrodeposited films. SAMs are much thinner films, and do not have the internal porosity of electrodeposited polymers, which makes it improbable that they can produce the same surface area. This chapter explores what can, in fact, be achieved with pure and mixed EHPT SAMs, using an in vitro electrode model. The model chosen is the EcoMEA 60-electrode array from MultiChannel Systems, shown in Figure 6.1. This is an array designed to perform electrophysiology on neural cell cultures. It therefore bears a central well (formed by a glass ring glued to the electrode base) that can contain about 2 ml of fluid and whose floor has an array of electrodes, providing a convenient base for SAM formation and impedance measurement. EcoMEAs have two particular advantages: they are formed largely of glass-epoxy materials that are not damaged by the solvents used in these SAM experiments, and they use gold for the electrode metallization. While the SAM protocol could be adapted to work with other in vitro MEAs, it is usable in EcoMEAs without alteration from prior chapters.

6.2 Methods

6.2.1 EcoMEA Preparation for SAM Formation

Preliminary work with EcoMEAs revealed that although the electrodes are gold, they possess some property (likely impurities) that can interfere with eﬀective SAM formation. To correct this diﬃculty,

65 Figure 6.1: The EcoMEA used in the experiments described in this chapter. On the left, a photograph showing the general conﬁguration, with 60 contact pads surrounding the central electrode well. (This is not a true EcoMEA, but a related design from Ayanda Biosystems.) On the right, a close-up photo of the well of an EcoMEA, showing the electrode traces converging to form the array. Electrodes are 100 µm in diameter, with 700 µm between them. This photo does not show the large-area reference electrode that is integrated into the array as electrode #38. Images courtesy of Ayanda Biosystems and MultiChannel Systems, respectively.

a thin coating of high-purity gold was applied to the array electrodes before use, following the method of [126]. Arrays (received without the central glass ring) were cleaned with 5 minutes of oxygen plasma treatment. They were then masked with aluminum foil, with the exception of a roughly 0.64 cm2 square in the center that contained the electrodes. Masked arrays were then mounted on the stage of the same thermal evaporator used to deposit gold on mica in Chapter 4 and coverslips in Chapter 5. The exposed portion of the array was then coated with 30 A˚ Cr and 100 A˚ Au using the same deposition parameters given in Chapter 5 for coverslips. Owing to the recessed structure of the electrodes, the thinness of the coating, and the well-known shadowing eﬀect observed in thermal evaporation of metal, no shorting occurs between electrodes as a result of this technique. After remetallization, a glass ring was mounted onto each MEA using Epo-Tek 353ND epoxy (Epoxy Technologies) and cured for at least 2 hours at 90◦ C.

6.2.2 SAM Formation on Gold-Coated EcoMEAs

Before SAM formation, MEAs were cleaned by a further 5 minutes of oxygen plasma. SAMs were formed on MEAs by ﬁlling the central well with approximately 1.5 ml of a pure or mixed SAM- forming thiol solution. These solutions were prepared in the same manner described in Chapters 4-5. A 3 mg/ml solution of unthiolated EHPT was also tested, to follow up the observation from Chapter 4 that this compound may be able to form SAMs. The ﬁlled MEA was then incubated for at least 36 hours in a sealed jar alongside a vial of toluene that maintained atmospheric saturation and prevented solution evaporation. Without this supplemental solvent, the liquid in the well will evaporate to near-dryness, preventing good SAM formation. It is worth noting that even with the solvent-resistant epoxy 353ND, SAM formation was not a completely reliable process. About one time in six, a failure of the epoxy to properly adhere to the substrate caused the ring to detach and the SAM-forming solution to leak out. Pre-roughening of the surface (by scratching with a diamond scribe) and oxygen plasma cleaning (of both ring and MEA) before joining may help, but the evidence is not clear. After SAM formation, the MEA well was vigorously rinsed at least 5 times each with toluene and chloroform to remove all unbound material, followed by at least 30 minutes further incubation

66 with fresh toluene. MEAs were allowed to dry for at least 12 hours to minimize solvent trapping in the monolayer. For preparation of protein-coupled SAMs, the well was ﬁlled with an NHS/EDAC solution (prepared as per Chapter 4) for one hour. The solution was drained, the well rinsed thoroughly with absolute ethanol, and a 100 µg/ml solution of FITC-conjugated goat anti-mouse IgG (Sigma) in PBS was spotted onto the gold-coated area. This is the same model protein used in Chapter 4. Protein coupling was allowed to proceed for at least one hour before extensive rinsing of the well with glass-distilled water.

6.2.3 Impedance Measurements

Electrical impedance spectroscopy (EIS) was performed using a Gamry Instruments FAS2/Femtostat potentiostat controlled by the Gamry Framework software running on a dedicated PC. MEAs were mounted using a custom connector that allowed selection of any of the 60 electrodes by turning one of three 32-position rotary switches. The central well was ﬁlled with PBS and used to construct a three- electrode cell with an individual MEA electrode as the working electrode, a large-diameter platinum wire as the counter electrode, and a saturated calomel electrode as the reference. Impedance was then measured with a 10 mV RMS amplitude sinusoid at a DC bias of 0 V versus the reference electrode, taking four measurements per decade from 100,000 to 10 Hz. Before measurement, MEAs were allowed to equilibrate overnight in PBS to standardize the SAM hydration state and ensure thorough wetting of all electrode wells. The equilibration PBS was replaced with fresh electrolyte immediately before measurement.

6.3 Results

6.3.1 Eﬀects of Pure and Mixed SAMs on MEA Impedance

The impedance spectrum of an EcoMEA before and after pure EHPT SAM formation is shown in Figure 6.2. Electrode impedance is significantly lower after coating, by about 40% (p < 10−12). A slight positive phase shift is seen, suggesting that capacitor-like components of the impedance fall further than do resistor-like components, but the overall shape of the phase curve does not change. A Nyquist plot of the same spectrum, also in Figure 6.2, shows an overall decrease in impedance without major changes in the curve shape, consistent with the expected surface area increase. Both “before” and “after” curves show a somewhat linear shape with a phase angle close to 45◦, suggesting that diffusion of charge carriers to and from the electrode is playing an important role in both cases. The pure MHA SAM, shown in Figure 6.3, has almost precisely the opposite effect to the pure EHPT SAM. The impedance at the biologically relevant frequency of 1 kHz is now significantly increased by approximately 2-fold (p < 9.8×10−10). There is a positive phase shift and flattening of the phase curve, indicating a decrease in electrode capacitance. Addition of coupled protein does not significantly alter the impedance, but does enhance that previously-observed phase shift. Turning to the Nyquist plot, there is a clear shift of the curve towards the real axis once the MHA SAM is applied, indicating that the resistive component of the impedance is also increasing. Others have argued that this is due to formation of a SAM with pinhole defects, effectively lowering the electrode surface area [155]. EIS of a mixed SAM formed from a 1:1 w:w EHPT:MHA mixture in solution is shown in Fig- ure 6.4. The impedance decrease is almost identical to that caused by pure EHPT, approximately 45% (p < 9.7 × 10−11). The positive phase shift, by contrast, is almost identical to that seen with MHA (+12◦, p < 10−12), although the curve shape is not visibly altered. Coupling protein to this SAM again does not alter the impedance magnitude, but causes a further positive phase shift just as was seen with MHA. The same characteristics are visible on the Nyquist curve in Figure 6.4, where the SAM curve is a hybrid of the observations for pure EHPT and pure MHA.

67 Impedance Spectrum of Gold Electrodes Before and After EHPT SAM Nyquist Plot of Gold Electrodes Before and After EHPT SAM 6 10 300

5 10 Ohm 4 250 10

1 2 3 4 5 10 10 10 10 10 200 Bare Gold (n=57) EHPT SAM (n=56) Hz

Bare Gold (n=57) EHPT SAM (n=56) 150 -ZImaginary (kOhm) -20 100 -30

-40

Deg 50 -50

-60

1 2 3 4 5 10 10 10 10 10 50 100 150 200 250 300 Hz ZReal (kOhm)

Figure 6.2: Impedance of electrodes across an MEA before and after EHPT SAM formation. Bode plot on the left, Nyquist plot on the right. SAM formation lowers the mean impedance magnitude at 1 kHz by 45% (from 20 KΩ to 11 KΩ, p < 10−12) and causes a positive phase shift of roughly 4◦ (p < 0.0018). The Nyquist plot again shows the overall impedance decrease, with a slight shift of the curve towards the real axis at low frequencies.

Impedance Spectrum of Gold Electrodes Before and After MHA SAM Nyquist Plot of Gold Electrodes Before and After MHA SAM 6 10

5 10 700

Ohm 4 10 600

Bare Gold (n=57) MHA SAM (n=57) 1 2 3 4 5 10 10 10 10 10 500 MHA SAM with Coupled IgG (n=52) Hz

Bare Gold (n=57) 400 MHA SAM (n=57) MHA SAM with Coupled IgG (n=52)

-ZImaginary (kOhm) 300

-30 -40 200

Deg -50 -60 100 -70

1 2 3 4 5 10 10 10 10 10 100 200 300 400 500 600 700 Hz ZReal (kOhm)

Figure 6.3: Impedance of electrode array after coating with a pure MHA SAM. In the Bode plot (left), the raw SAM more than doubles the impedance magnitude at 1 kHz (from 20 to 42 KΩ, p < 9.8 × 10−10) and causes a signiﬁcant decrease in capacitance (phase shift of -61◦ to -49◦, p = 3.5 × 10−9). Coupling of protein to this SAM causes no signiﬁcant change in impedance magnitude (p = 0.0538) and further shifts the phase (shift of +4◦, p = 0.0005). In the Nyquist plot (right), the MHA SAM shifts the plot towards the real axis, indicating that the resistive component of impedance has increased even more than the capacitance.

68 Impedance Spectrum of Gold Electrodes Before and After Mixed EHPT:MHA SAM Nyquist Plot of Gold Electrodes Before and After Mixed EHPT:MHA SAM 6 10 550 5 10 500

Ohm 4 10 450

400 Bare Gold (n=57) 1 2 3 4 5 10 10 10 10 10 1:1 EHPT:MHA SAM (n=56) Hz 350 1:1 EHPT:MHA SAM with Coupled IgG (n=57)

Bare Gold (n=57) 300 1:1 EHPT:MHA SAM (n=56) 1:1 EHPT:MHA SAM with Coupled IgG (n=57) 250 -ZImaginary (kOhm) 200 -20 150 -40

Deg 100

-60 50

1 2 3 4 5 10 10 10 10 10 50 100 150 200 250 300 350 400 450 500 550 Hz ZReal (kOhm)

Figure 6.4: Impedance after coating with a mixed 1:1 SAM. The mixed SAM lowers impedance approximately 45% (20 KΩ to 11 KΩ, p < 9.7×10−11) and shifts the phase positive by 12◦ (-61◦ to -49◦, p < 10−12). Protein coupling to the MHA component does not alter the magnitude (p = 0.3192), but further shifts the phase (+5◦, p = 0.0003). Like the Bode plot, the Nyquist plot is a blend of the EHPT and MHA spectra – it shows the decreased impedance caused by EHPT, but also shows the shift towards the real axis (altered capacitance/resistance balance) caused by MHA.

6.3.2 Removal and Re-Formation of EHPT SAMs

To demonstrate that the changes observed are due to the polymer and not some extraneous effect, one set of electrodes was subjected to SAM removal (by oxygen plasma ashing) and reapplication. The results of this set of experiments are shown in Figure 6.5. It should be noted that these data are from an early series of experiments with variant methodology; the MEA was not given an oxygen plasma clean before application of the first SAM, and neither the first nor the second SAM was pre-soaked in PBS overnight before data acquisition. On the first application of a pure EHPT SAM, the impedance decreases by about 50%, as was seen with the MEA of Figure 6.2. The surprise comes when the SAM is stripped by a brief plasma clean – there is no change in impedance magnitude, but there is a shift in the phase curve, and the new curve is similar in shape to that of the original “bare gold” curve. When a second SAM is then applied to the cleaned electrodes, impedance again drops by 50%, and the phase curve shifts positive once again. The likely explanation is that the intermediate oxygen plasma clean not only removed the first SAM, but also removed residual organic contaminants, lowering the electrode impedance. It was this observation that led to the policy of always performing a plasma clean before SAM formation.

6.3.3 Lack of SAM Formation With Unthiolated EHPT

In Chapter 4, it was seen than non-thiolated EHPT, when applied to a clean template-stripped gold surface, could form what appeared to be a SAM, possibly with lower coverage than the normal SAMs formed from thiolated EHPT. Figure 6.6 shows the results of applying the same compound to an EcoMEA, and the results do not match what was seen under AFM. In this case, the impedance magnitude does not change (p > 0.94), and the phase shows a discernible shift only at lower frequencies. The Nyquist plot also does not show changes consistent with EHPT; the curve tilts away from the real axis, implying a possible decrease in capacitance or surface area. This fits with the hypothesis first raised in the Discussion of Chapter 4: unthiolated EHPT can form a film on a very clean, very flat gold surface, but cannot effectively alter a rough and less-pure surface. At the most, unthiolated EHPT acts as yet another organic contaminant adsorbed to the electrode surface.

69 Impedance Spectra of EHPT SAM Formation, Removal, and Reapplication 6 10

5 10 Ohm 4 10

1 2 3 4 5 10 10 10 10 10 Hz

Bare Gold (n=51) EHPT SAM (n=50) 2 Min O2 Plasma (n=51) Re-formed EHPT SAM (n=52)

-20

-40 Deg

-60

1 2 3 4 5 10 10 10 10 10 Hz

Figure 6.5: Removal and re-application of an EHPT SAM. Application of the SAM (blue circles to green squares) reduces impedance by approximately 13 KΩ. Exposure to an oxygen plasma removes the SAM, but also removes organic contaminants, leading to essentially no total change in impedance (green squares to red diamonds). Re-exposure to the SAM-forming solution (red diamonds to cyan triangles) further lowers impedance by 7 KΩ, to 6.94 KΩ, the expected factor-of-two improvement. Similarly, the positive phase shift of roughly 20◦ disappears after plasma treatment and reappears following formation of a new SAM.

70 Impedance Spectrum of Unthiolated EHPT (EHPT-X) SAM Nyquist Plot of Gold Electrodes Before and After EHPT-X 6 10 300

5 10 Ohm 4 250 10 Bare Gold (n=59) EHPT-X (n=59)

1 2 3 4 5 10 10 10 10 10 200 Hz

Bare Gold (n=59) EHPT-X SAM (n=59) 150 -ZImaginary (kOhm)

-20 100 -30

-40

Deg 50 -50

-60

1 2 3 4 5 10 10 10 10 10 50 100 150 200 250 300 Hz ZReal (kOhm)

Figure 6.6: Application of an unthiolated EHPT (EHPT-X) solution to an electrode array. No change in impedance magnitude is seen versus the bare gold (p > 0.94). There is a slight phase decrease of 2◦, which is statistically signiﬁcant (p < 0.013), but which is much smaller than the 15-20◦ drop caused by a thiolated EHPT SAM. The Nyquist plot reﬂects this phase change (particularly at lower frequencies) with a shift towards the imaginary axis.

6.4 Discussion

The results of this chapter demonstrate that EHPT is able to lower the impedance of metal electrodes, as predicted by the AFM data of Chapter 4. While the AFM data showed a surface area change of only 4%, this does not reflect the actual surface area seen by the electrolyte. Mobile ions are able to probe much smaller surface features than are visible to the relatively larger AFM tip, and further study would be necessary to properly correlate AFM measurements with EIS. Similarly, the increase in electrode impedance and decrease in effective capacitance after MHA SAM formation is consistent with previous electrochemical studies of alkanethiol SAMs [154, 188–190]. The effect can be attributed to a decrease in the double-layer capacitance due to an increased electrode/electrolyte separation, based on the tightly-packed and insulating nature of these SAMs. Coupling protein to this SAM produces an impedance decrease that is not significant, but is borderline (p = 0.0538). The reason for this decrease is unclear; other authors have attributed it to decreased polarization and diffusion impedances [191]. Disruption of SAM packing by the coupled protein is a theoretically possible mechanism, but is less likely. That should shift the phase curve back towards the bare-metal baseline, and the opposite trend is observed in Figure 6.3. To be consistent with the results of Chapters 4 and 5, the 1:1 EHPT:MHA SAM should blend the electrical properties of its two components. This is precisely what is observed. The impedance decrease is equal to that caused by EHPT alone, and the phase shift matches that caused by MHA alone. (The lack of change in the shape of the phase curve shows that performance is not exactly equal to MHA.) The behavior after protein coupling also mirrors the MHA SAM. It is encouraging that this mixed SAM can raise biocompatibility without producing a higher impedance. However, it is also surprising, since the MHA should have some insulative effect. One possible explanation is that the negative charge of the COOH groups on MHA acts as a dopant for the polythiophene, increasing its conductivity to compensate for the lower amount actually present. Another (albeit less supported by the AFM data) is that MHA molecules are acting to disrupt the observed EHPT aggregates, which would decrease grain size and increase surface area, again compensating for the lower overall amount of EHPT in the SAM. Further elucidating the mechanisms behind this observation would require EIS studies of more film compositions, most likely combined with equivalent circuit modeling to quantify the changes in various circuit parameters. Models are known for polythiophene thick films [192,193] and for alkanethiol SAMs [194], but combining these is not trivial.

71 The data also indicate that the changes observed are specifically due to SAM formation, and that this SAM formation requires the presence of a terminal thiol group. It is not at all surprising that the EHPT SAM can be ashed off and then re-applied, but the curves of Figure 6.5 do help show that it is the EHPT which is responsible for the improved impedance. The failure of unthiolated EHPT to produce the same impedance decrease is also quite encouraging; it supports the contention that unthiolated PTs should only be able to bind through their ring sulfurs, and that such a binding would be highly disfavored on any “real world” gold surface. The one partial disappointment is that EHPT SAMs do not achieve nearly the same level of impedance drop that is available from electrodeposited coatings. As noted in the Introduction to this chapter, that result is not unexpected. It is also not necessarily a fatal flaw. The impedance performance of these films can undoubtedly be improved by choosing PT variants with higher molecular weight (which will further increase surface area), finding ways to dope the film, and otherwise tuning the system. The robustness and patternability of thiol-based films may also be a sufficient advantage to be worth trading off against impedance. It may even be possible to eventually match the electrodeposited films’ impedance by combining self-assembly and electrodeposition, as is described further in Chapter 10. It is important to reiterate that impedance lowering is not necessarily the key to good recording, as long as the initial electrode impedance is not so large as to produce excessive noise. Improvements in recording beyond that point come from good biocompatibility, which has already been shown in Chapter 5. Good stimulation is related to lower impedance, but these EHPT SAMs are not intended for extracellular stimulation. As laid out in Chapter 3, the second key concept in this thesis is the development of PT-mediated intracellular electrical access. If intracellular stimulation is achieved, it will not matter that EHPT passes less capacitive current than PPY or PEDOT, because orders of magnitude less current will be needed. It follows that the impedance improvement seen here is good, but that the real challenge is to prove that membrane insertion and intracellular stimulation will be possible with PT-based electrodes. The next three chapters will do precisely that, finishing the proof of concept for the utility of PT-based neuro-robotic electrodes.

72 Chapter 7

EHPT Raises the Electrical Conductance of Artiﬁcial Lipid Bilayers

7.1 Introduction

The preceding chapters have focused on the properties of polythiophenes acting en masse in a SAM, and have demonstrated that EHPT SAMs can at least be competitive with electrodeposited films. In this and the following two chapters, the spotlight shifts away from whole-film effects to the behavior of individual molecules, and in particular, the ability of individual PT chains to enter a lipid bilayer. As noted at the end of Chapter 6, this capability may be very important if PT SAMs are to be a viable technology for electrical stimulation. Previous chapters also tested a single polythiophene, EHPT, on the grounds that PTs with different alkyl side chains are not likely to show substantial differences in their SAM properties. That assumption does not hold when dealing with single molecules. As noted in Chapter 3, one of the specific hypotheses underlying this thesis is that PTs will only be able to enter a lipid membrane if they possess appropriately tuned side chains. Therefore, this chapter and the two that follow will use a paradigm of comparison between EHPT (whose branched side chain creates high solubility) and HPT (whose straight side chain leads to lower solubility). The repeat units of both polymers are diagrammed in Figures 7.1 and 7.2 for reference. In keeping with the approach taken in the prior chapters, the question of PT interactions with membranes will be treated in a simplified in vitro model to remove extraneous variables. The model used here is the “bilayer lipid membrane” or “black lipid membrane” (BLM) method, in which a cell- membrane-like bilayer of phospholipids is formed between two aqueous compartments. By exposing these artificial bilayers to EHPT and HPT and measuring the DC conductivity of the bilayer, it is possible to determine whether the polymers can establish an electrical bridge by self-assembling into the membrane.

7.2 Methods

7.2.1 Creation and Measurement of Artiﬁcial Lipid Bilayers

BLMs were constructed in a polytetraﬂuoroethylene (PTFE) experimental chamber as described in [195] and shown in Figure 7.3. Partitions made of PTFE ﬁlm (0.025 mm thick with hole diame-

73 S S * * * * S S

n n

Figure 7.1: EHPT repeat unit Figure 7.2: HPT repeat unit

Structure of the two polymers discussed in this chapter, EHPT (Figure 7.1) and HPT (Figure 7.2). EHPT is hypothesized to be lipophilic and able to enter a lipid membrane, whereas HPT is not. This diﬀerence is hypothesized to derive from EHPT’s branched side chain, which gives this polymer greater disorder in all phases.

Figure 7.3: Schematic diagram of the experimental chamber. (A), PTFE chamber partitioned by a thin septum. (B), close-up of septum showing microscale hole containing solvent-free artiﬁcial lipid bilayer. Holes used in these experiments ranged from 80 to 100 µm. Figure adapted from [195].

74 ters that varied from 80 µm to 100 µm) were used to divide the chamber into two halves. During an experiment, one half chamber (chosen arbitrarily) faces the experimenter and is termed the “front”. The lipid for forming the membrane, lyophilized 1,2-diphytanoyl-sn-Glycero-3-Phosphocholine (Di- PhyPC, Avanti Polar Lipids) was dissolved in pentane at 10 mg lipid per 1 ml pentane. Lipid bilayers were then formed across the hole using a technique modified from [196]. Initially, saline solution (1 M NaCl in these experiments) partially filled each half of the chamber (volume approximately 6 ml). Approximately 20 µl of lipid in pentane was applied to the surface of the aqueous phase, with the liquid just below the hole in the partition. To facilitate membrane formation, the rim of the hole was conditioned with hexadecane using a solution of hexadecane in pentane (1:100 v/v). After allowing the pentane to evaporate, the fluid levels were slowly raised in sequence to a point just above the hole to form the lipid bilayer. The additional saline used to raise the level was introduced into the chamber via PTFE tubes connected between the chamber and syringes filled with saline, and the process was monitored visually by observing the fluid levels through a microscope. During the raising of the fluid levels, an AC voltage was applied across the partition. Membrane formation was detected by a sudden step increase in the capacitively coupled current. A pair of Ag/AgCl electrodes connected to an Axon Instruments headstage and amplifier were used to apply computer-controlled voltages to the membrane and to measure currents through the membrane. Applied voltages ranged from 200 mV to -200 mV in steps of 10 mV. Currents were recorded after the application of each voltage and averaged over a 200 ms window. Recorded signals were lowpass filtered with a 1 kHz cutoff before being digitized for recording. The PTFE chamber and electrodes are located inside a mu-metal box on a vibration isolation table to eliminate pickup of environmental signals.

7.2.2 Introduction of PTs to Artiﬁcial Bilayers

Regioregular head-to-tail EHPT and HPT (both of average molecular weight 3,000 g/mol and synthesized as described in [166]) were dissolved in N,N-dimethylformamide (DMF) to saturation. Bi- layers were formed as described above, and the current-voltage (I-V) characteristic of the bilayer was recorded. One of the PT solutions and/or pure DMF was then injected in aliquots of 100 or 200 µl into the aqueous subphase of one half chamber under continual stirring. The system was allowed to settle and equilibrate for under ﬁve minutes, after which time another I-V characteristic was recorded. The capacitively coupled current was re-checked after each injection to ensure that the bilayer itself was not compromised. In the event that a bilayer was weakened or disrupted, it could often be re-established immediately by lowering and raising the saline level in one or both chambers. Reformation did not appear to aﬀect the conductance behavior of the BLMs.

7.3 Results

7.3.1 EHPT Increases BLM Conductance

Figure 7.4 shows the results of adding DMF with and without EHPT to an artificial lipid bilayer. Introduction of pure DMF into the aqueous phase that contacts one side of the BLM has no effect on the average conductance, as shown by curve B in Figure 7.4 (which essentially overlays A). Injection of an equivalent amount of the polymer solution in the same half-chamber, however, produces an elevenfold increase in conductance (ignoring the values around 0 V, which are larger). This increase is seen even though the majority of the injected polymer did not enter or associate with the bilayer. The small volumes of DMF used here are readily miscible with the 6 mL of saline that fills each well of the chamber, and thus the DMF and polymer appear to mostly intermix with and remain in the aqueous phase. It should be noted that although no increase in average current response to a voltage pulse is seen in the presence of DMF alone, transient spike-like increases in current do occur when constant voltages are applied over tens of seconds after application of DMF. These may represent transient membrane disruptions.

75 8 5 A: Base Membrane A: Base Membrane B: Membrane + 200 uL DMF B: Membrane + 300 uL HPT back C: Membrane + 200 uL EHPT C 4 C: Membrane + 200 uL EHPT back, 100 uL front (membrane reformed) 6 C

4 2

2 1 B A A,B 0 Current (pA) Current (pA) 0

-1 -2

-2

-4 -3

-6 -4 -250 -200 -150 -100 -50 0 50 100 150 200 250 -250 -200 -150 -100 -50 0 50 100 150 200 250 Voltage (mV) Voltage (mV) Figure 7.4: Conductance changes in a DiPhyPC Figure 7.5: Lack of conductance change when BLM after injection of pure solvent and poly- lipophilic EHPT is replaced by non-lipophilic mer solution. (A), I-V characteristic of the base HPT. (A), I-V characteristic of the base BLM. BLM. (B), BLM after injection of 200 µL of the (B), BLM after injection of 300 µL of a satu- solvent DMF. Essentially zero current is passed rated solution of HPT in DMF. No signiﬁcant except at the highest available voltages in both conductance change is seen. (C), BLM after fur- A and B.(C), BLM after further injection of ther injection of 300 µL of saturated EHPT in 200 µL of a saturated solution of EHPT in DMF. DMF (200 µL back, 100 µL front) and reforma- Conductance has increased by at least one order tion of both membrane leaﬂets. Conductance is of magnitude, with supralinear increases at the now comparable to that seen in Figure 7.4. extreme voltages.

15 A: Membrane + 200 uL EHPT front B: Membrane + EHPT, 12 Minutes After Injection D C: Membrane + EHPT, 22 Minutes After Injection D: Membrane + Additional 105 uL back 10 E: Membrane + EHPT, 8 Minutes After Injection

E A 5 B

C 0 Current (pA)

-5

-10

-15 -250 -200 -150 -100 -50 0 50 100 150 200 250 Voltage (mV) Figure 7.6: Decay of increased BLM conductance over several minutes. These curves represent the same DiPhyPC BLM described in Figure 7.4. (A), I-V characteristic of membrane injection of 200 µL of EHPT in DMF to the front half-chamber. This repeats Curve C of Figure 7.4. (B), BLM after 12 minutes. No significant changes in conduction are seen. (C), BLM after 22 minutes since injection. The current has now returned to the value for an unmodified membrane. (D), BLM after adding 105 µL of EHPT to the back half-chamber. This further addition of polymer produces a larger current than that seen with the original 200 µl. (E), BLM after eight minutes of further equilibration. Conductance has now decayed significantly from its original value.

76 The increase in current flow appears to depend on the concentration of polymer present. Injection of only 100 µl did not produce a significant effect but 200 µl did. It was not possible to directly assess the effects of injecting a full 300 µl of EHPT into a single half chamber, as this volume of EHPT in DMF solution showed a marked tendency to destabilize and permanently disrupt the membrane. (Interestingly, the same tendency is not seen with 300 µl of HPT solution.) However, it was possible to inject 200 µl into one half chamber and 105 µl into the other, which increased the current to nearly 13 pA (26 times greater than the approximately 0.5 pA across the unmodified membrane).

7.3.2 HPT Does Not Alter BLM Conductance

As seen in Figure 7.5, injection of 300 µl of HPT did not produce significant changes in conductance, whereas 300 µl of EHPT (split between front and back chamber to mitigate membrane disruption) did. The HPT in DMF was visually verified to mix with the aqueous phase just as the EHPT solution did. However, HPT was slightly less soluble in DMF than EHPT, as judged by a lighter color of the saturated solution. This is consistent with HPT’s overall lesser solubility, and the results of Figure 7.5 fit the hypothesis that EHPT can insert while HPT cannot.

7.3.3 Conductance Changes Caused By EHPT Diminish Over Time

One membrane was followed after EHPT exposure to track changes in the eﬀect over time, and Figure 7.6 shows the results of that investigation. As shown in curves A through C, one BLM showed a decay back to its base characteristics within 22 minutes from the initial exposure. This decay did not occur at a constant rate, as an I-V recording at 12 minutes showed no signiﬁcant change. Moreover, further application of another 105 µl to the opposite half chamber raised the conductance above its prior peak, as seen in D. When the membrane was again allowed to rest, the current again declined to roughly half its maximum, this time within only eight minutes. Stirring of the solution did not further increase the conductance at any point during these recordings.

7.4 Discussion

The data from these BLM studies establish that EHPT is able to significantly increase the current across a lipid bilayer, whereas HPT is not. The concordance with predictions based on solubility is very encouraging, as it argues for an effect driven by membrane insertion. These data do not conclusively prove the insertion mechanism, but there is nothing in the observations that contradicts it. No steady-state leak current is seen from the DMF alone, so solvent effects are not responsible. At most, the DMF is aiding polymer entry by causing tiny “current spike” membrane disruptions. Another alternate explanation is that EHPT molecules are assembling into pore-like structures or otherwise disrupting the membrane to allow passage of ions. However, EHPT has no hydrophilic component that could form an ion-conducting channel, and this mechanism does not explain the lack of effect with HPT. The same problem occurs with arguments that these observations represent EHPT aiding conduction through the aqueous phase. Moreover, undoped EHPT is a semiconductor at best, and is less conductive than the saline. Overall, the most likely mechanism is transfer of electrons along EHPT chains dissolved in the membrane, presumably through formation of electrical double layers to convert between ionic and electronic current. The observed decay in conductance over time is an interesting, but not unexpected effect. As can be seen in Figure 7.3, the BLMs used here consist of a bilayer surrounded by a thicker annulus of bulk phospholipid. If EHPT is entering the lipid phase of the bilayer as hypothesized, it is presumably able to diffuse within that phase much as it would in a hydrocarbon solvent. The greater volume of the annulus will favor diffusion of EHPT into that region over time, depleting the bilayer region of polymer and raising the conductance back towards its original value. This would not be a concern in the final device as laid out in Chapter 3; EHPT in the device will be tethered as SAMs and unable

77 to migrate. Moreover, real cell membranes have no annulus. It should be noted that the largest currents passed through the bilayer in this study were about 13 pA at 200 mV applied potential. In order to depolarize and fire a neuron, one would need currents at least ten times larger. However, the methods used here are probably not a fair measure of what could be achieved with a tethered SAM on a metal electrode. There are many extra impedances, from the solution resistance to the presumed electrical double layer on either side of the lipid bilayer, that would not exist in the proposed device. There is also the question of EHPT’s doping state – as noted in Chapter 5, being coupled into an anion-containing SAM could potentially dope the polymer and increase its conductivity. The currents observed here are therefore not a cause for concern. In summary, this chapter has shown that EHPT is capable of increasing current conduction across a lipid bilayer. While a real membrane is far more complex than the single-lipid model used here, it is also more robust. The fact that the applied EHPT did not lyse the membrane, even with substantial amounts of organic solvent present, is a very positive sign that neurons will be able to tolerate having EHPT chronically inserted into their membranes. However, these data lack one major component: proof that the observed effect is genuinely due to EHPT bridging the membrane. As discussed above, there are no truly compelling alternative explanations, but the absence of a defined alternative does not prove the insertion mechanism. Solid evidence in favor of insertion is still needed as the last component for proof of the device concept given in Chapter 3. That evidence will be given in Chapter 9, and the following chapter will explain in more detail the necessary methods.

78 Chapter 8

Development of Parameters for Polythiophene Molecular Modeling

8.1 Introduction

The results of Chapter 7 support the hypothesized intracellular recording mechanism, but they do not conclusively show that this mechanism underlies the observed increase in bilayer conductance. It is possible that other polythiophene/bilayer interactions are responsible, such as one involving polymers coating the membrane surface but not actually inserting. The ideal proof of the proposed mechanism would be an imaging method that allowed visualization of polymer chains and bundles inside a lipid bilayer, but such a method is not readily available and well-validated. This chapter and the following one present an alternate approach: atomic-resolution simulation of the PT/bilayer system. By simulating the system for sufficient periods of time, it is possible to derive estimates of the free energy change involved in inserting EHPT and/or HPT into a lipid bilayer. The qualitative and quantitative results of that estimation, combined with study of the configurations of the simulated atoms, can elucidate the molecular mechanisms underlying the BLM results. The method used in these two chapters is molecular dynamics (MD), a computational chemistry technique that has been used since the late 1970s to study the structure and energetics of biomacromolecules [197]. Briefly, the MD approach treats individual atoms as spherical particles interacting in a three-dimensional simulation “box” through repulsive and attractive two-body potentials that capture the electrostatic (Coulombic) and van der Waals (VDW) interatomic forces. Covalent bonds within a molecule are represented as linear springs that govern bond stretching, angle bending, and rotation about a bond axis. Given these forces, the known atomic masses, and a set of starting positions, the evolution of the system can be followed over time by integrating the Newtonian equations of motion. The simulation’s usefulness is then limited primarily by the computer power available. In order to accurately model atomic behavior, an MD simulation must employ a timestep smaller than the fastest motion in the system, which is usually the stretching vibration of atomic bonds. Moreover, some force terms (such as the non-bonded forces) are non-linear functions that can rapidly undergo large changes in magnitude and/or reversals in sign. This effectively produces a set of stiff equations, which also require a small timestep for numerical stability. The net result of these constraints is that MD of biomolecular systems uses simulation timesteps on the order of femtoseconds (10−15 seconds). Simulations of large systems (such as fully hydrated lipid bilayer patches) are therefore generally limited to examining nanosecond timescales, even with the use of supercomputers and/or high-performance clusters. However, this timescale is sufficient to capture processes such as the folding of small proteins or receptor-ligand docking. Moreover, as will be discussed in Chapter 9, there also exist enhanced sampling techniques that allow a careful investigator to obtain thermodynamic information about processes that occur on microsecond or

79 even millisecond timescales. MD simulations have been used by other investigators to model a variety of PT behaviors. Early simulations focused on predicting PT conformation in different solvation regimes, a problem that has some relevance to the present work [198–200]. MD methods as implemented in programs specifically designed to work with regular periodic lattices have also enjoyed good success at modeling the structure and pressure/temperature responses of doped and undoped solid-state PT films [201–205]. More recent applications have included modeling ion migration during doping [206], predicting the orientation and dipole of an amphiphilic PT monolayer at an air/water interface [207, 208], and interpreting/predicting scanning tunneling microscopy results [209]. MD has also been extensively used to model lipid bilayer membranes, and recent advances in computing power have enabled MD simulations of small molecules inserting into bilayers. The past few years have seen MD studies of the insertion of the mellitin peptide (a pore-forming toxin found in bee venom) [210,211], antimicrobial peptides [212], helical voltage sensors [213], the WALP model hydrophobic peptide [214], and large channel proteins (using a “coarse-grained” model) [215, 216]. The same techniques can also study molecules anchored to the membrane but not inserted [217], the extraction of membrane lipids by phospholipase enzymes [218], and the structure and function of transmembrane channels [219, 220]. In short, the techniques and parameters needed to simulate the modification and/or disruption of membranes by small-to-medium-sized molecules are well- disseminated and sufficiently validated to be applied to the question of polythiophene insertion into bilayers. Before MD can be applied to that problem, it is necessary to consider the forcefield – the set of bond spring constants, atomic charges, VDW radii, and other such parameters needed to completely specify all possible interactions between simulated atoms. Since a large portion of the activity in the MD field is focused on biomolecules, there exist well-tested parameters for proteins, DNA, lipids, and water. The same is not true for polythiophene – some generic forcefields do include atom types that can be used to model PTs, but they are unlikely to fully capture the complex behaviors of this unusual class of molecules. Most of the investigations mentioned above have used these generic parameters or have slightly modified known parameters for a related molecule, e.g. modeling thiophene using parameters designed for furan. That strategy has been able to achieve reasonable agreement with experimental data, but a more rigorous approach to parameterization would likely improve the reliability of the simulations and their results. Therefore, this chapter concerns itself with the derivation of polythiophene-specific forcefield parameters from higher levels of theory and the validation of those parameters in MD simulations of isolated polythiophenes. The following chapter will then demonstrate the use of the derived and validated parameters to model PT interactions with lipid bilayers.

8.2 Methods

8.2.1 Ab initio Parameter Development

The simulations described here and in Chapter 9 used the Generalized AMBER Force Field (GAFF), as supplied with AMBER 7 [221], to describe the majority of the polythiophene molecule. GAFF provides reasonable values for the force constants for the bonds within each monomer and the VDW interactions of the individual atoms. Starting from this base, three things are still required to construct MD simulations of polythiophenes:

1. The equilibrium values for the bond lengths, bending angles, and torsion angles within the monomer and polymers, which are used to produce both the final forcefield and the initial atomic structure files for simulations. 2. The effective charge on each atom (“partial charge”), assuming that the overall charge of the polymer remains zero.

80 3. A quantitative description of the energy of rotation about the inter-ring bond, which is known to be the most ﬂexible degree of freedom in the polymer [222, 223].

All three can be obtained by judicious use of ab initio computational quantum chemistry methods, in which the behavior of the molecule’s electrons are explicitly considered (instead of being lumped into Coulombic and VDW interactions as they are in MD methods). Ab initio calculations are far too computationally intensive to be used for simulation of polymer dynamics, but they can provide information that can then be used to parameterize the simpler and more computationally efficient representations used in MD. To produce initial structure files, pentamers of plain polythiophene, poly(3-hexylthiophene) (HPT), and poly(3-(2-ethylhexyl)thiophene) (EHPT) were constructed and their geometries optimized using GAUSSIAN03 [224] at the Hartree-Fock level of theory with the 6-31G* basis set (HF/6-31G*). Partial charges were then derived from the HF/6-31G* electrostatic potential grid using the Restrained ElectroStatic Potential (RESP) method, as implemented and automated by the ANTECHAMBER and RESP components of the AMBER suite [225]. The optimized and charge- assigned pentamer was split into “N-terminal”, “C-terminal”, and “central” (derived from the center ring) monomer residue files, discarding the information from the second and fourth rings. Only the nearly-planar all-anti conformation was optimized for each pentamer, based on others’ findings that bond lengths and other parameters do not significantly change in other near-minimum conformations [223]. A pentamer is expected to be more than sufficient to capture the geometry and electron structure of the general polymer, as previous ab initio studies have found that geometries do not change much once the length exceeds three rings [222, 226]. This particular basis set and calculation method were chosen for numerous reasons. Perhaps the most important is that many of the existing AMBER forcefields were derived using this same level of theory and basis set, and it is explicitly suggested in the AMBER documentation that new molecules be parameterized in this way. However, there also exist prior results showing that HF/6-31G* is a reasonable choice. Less computationally intensive methods such as semi-empirical techniques or density functional theory are sometimes able to reproduce the molecular geometry correctly, but are not able to accurately describe the energy of torsion about the inter-ring bond [227–231]. By contrast, HF/6-31G* optimizations and rotations have been shown by other investigators to match well with experimental results [232], and the use of higher levels of theory (e.g., MP2) does not substantially change the relative energies of different conformations [233]. With the geometries and partial charges determined, the ANTECHAMBER and PARMCHK programs were used to automatically assign GAFF atom types and corresponding force constants to the atoms of each monomeric residue. A new atom type “cp” was defined for the C2 and C5 carbons, while the C3 and C4 ring carbons were allowed to remain as GAFF type “ca” (see Figures 8.1 and 8.2 for nomenclature). The majority of forcefield parameters for the new “cp” atom type were copied from the GAFF parameters for “ca”, but the equilibrium bond lengths and angles were altered to better match the geometry found from the HF/6-31G* optimization. As previously noted, the inter-ring dihedral angle (S1-C2-C5’-S1’ or C3-C2-C5’-C4’ in Figures 8.1 and 8.2) is the primary degree of freedom through which polythiophenes alter their conformations. In some cases, it has even been possible to predict oligomer structure solely from the potential of rotation about this bond [223]. In order to properly parameterize this torsion for the AMBER forcefield, the HF/6-31G* rotational potential of bithiophene (with no alkyl side chains) was calculated in GAUSSIAN03. The dihedral angle between the two rings was rotated over a full 360◦ in increments of 5◦, with no geometry relaxation after the initial structure optimization. This “rigid rotor” method was chosen based on prior investigators’ observations that use of a “relaxed rotor” in bithiophene does not significantly change bond or angle parameters at each step, nor does it significantly alter the observed energy profile [234,235]. It is also more amenable to the type of cosine fitting described below, since a relaxed rotation would be difficult to mimic in AMBER. The full 360◦ rotation (as opposed to the more commonly-seen 180◦ rotation from anti to syn conformation) was undertaken based on previous reports that the rotor potential is not completely symmetric [229]. This appears to be the first instance where the bithiophene rotor potential has been calculated and reported in

81 CA6 CA6 CA5 CA5

CA4 CA4 CA3 CA3

CA2 CB2 CA2 C C A1 CB1 A1

C4 C3 C4 C3

S1' S1' H ' H ' C5 C2 2 C5 C2 2 C ' C ' H5 C5' 2 H5 C5' 2 S1 S1

C4' C3' C4' C3'

CB1'

CA1' CB2' CA1' CA2' CA2'

CA3' CA3' CA4' CA4'

CA5' CA5' CA6' CA6'

Figure 8.1: EHPT dimer nomenclature Figure 8.2: HPT dimer nomenclature

Diagram of ethylhexyl (EHPT, Figure 8.1) and hexyl (HPT, Figure 8.2) PT dimers to illustrate the ring and side chain atom nomenclature used in this chapter. To enhance the clarity of the drawing, hydrogen atoms of the side chains are not shown. These side chain hydrogens are numbered sequentially beginning on the carbon closest to the ring. For instance, side chain carbon CA1 will bear the hydrogens numbered HA1 and HA2.

such detail; all previous reports used an increment of 10◦ or greater, and only show the potential for a 180◦ rotation. Once the ab initio rotor potential was obtained, Fourier series decomposition was used to fit it to a series of cosine terms appropriate for use in an AMBER-compatible forcefield. Such decompositions have previously been shown to have good success in fitting thiophene rotor potentials with only a few terms [231,233]. Fitting proceeded by setting the torsional spring constant to zero and performing a 360◦ bithiophene rigid rotation in the derived AMBER forcefield. The resulting potential captured the effect of VDW and electrostatic interactions, as well as any effect of bond stretching and angle bending – all the forcefield terms except the resistance of the inter-ring bond itself. When the zero-spring-constant AMBER potential was subtracted from the HF/6-31G* potential, the residual values constituted an accurate description of the energy change that is attributable to electronic structure changes caused by torsion about the central bond. Fourier decomposition was applied to these residuals to obtain a set of terms of the form

Kn ∗ (1 + cos(nθ − φ)) (8.1)

for n=1 to 4, using both cosines (φ = 180◦) and sines (φ = 270◦). The obtained coefficients were then split between the S1-C2-C5’-S1’ and C3-C2-C5’-C4’ dihedral angles in the forcefield file, using four terms to describe each. That fitted potential was used in both the EHPT and HPT forcefields without further modification. Because the membrane model of Chapter 9 uses a united-atom representation for aliphatic carbons, a united-atom model was also derived for polythiophenes. The hydrogens of the alkyl side chains (but not the ring backbone) were unified with their carbons, summing the partial charges together. VDW radii for the new unified atoms were taken from the lipid forcefield used in Chap- ter 9, as were the bond, angle, and torsion parameters for these united-atom alkanes. Because these united-atom parameters require a different scaling of nonbonded forces (SCNB=8) than do the standard AMBER parameters, the Fourier decomposition of the previous paragraph was repeated in order to properly parameterize the united-atom model.

82 Figure 8.3: All-atom EHPT decamer Figure 8.4: All-atom HPT decamer

Examples of the EHPT and HPT decamers used for replica exchange simulations.

8.2.2 Replica Exchange MD in Implicit Solvents

Once parameters for the two PTs of interest were developed, the next step was the validation of those parameters in a membrane-free environment. Since EHPT and HPT are known to behave differently in good and poor solvents, the chosen test case was comparison of the behavior of the two polymers in water (bad solvent) and lipid (presumed good solvent). This same test could also provide some information about what to expect in simulations involving a lipid bilayer. For computational efficiency, these initial tests did not explicitly represent a box of solvent around the polymer, but instead used an implicit continuum dielectric model. Specifically, simulations were carried out using the revised generalized Born model described in [236], as implemented in AMBER 9 (the IGB=5 option). EHPT and HPT decamers were each simulated in implicit water (external dielectric constant 78.5) and a generic implicit saturated hydrocarbon (external dielectric constant 2.0). No surface area correction terms were used (GBSA=0). The polymer systems simulated are shown in Figures 8.3 and 8.4. In order to accelerate configurational sampling, all four implicit solvent MD simulations were performed using the replica exchange technique. Replica exchange, first formulated for molecular dynamics by Sugita and Okamoto [237], is a computational technique in which multiple parallel copies (“replicas”) of a system of interest are simulated at different temperatures. At pre-specified intervals, the simulation engine attempts to “exchange” two replicas by swapping the temperatures at which they are simulated. The probability p of a successful exchange is given by:

exp(−∆) if ∆ < 0 p = (8.2) 1 if ∆ ≥ 0

with

1 1 ∆ = ( − )(E1 − E2) (8.3) kBT1 kBT2

where E1 and E2 are the potential energies of the current conﬁgurations of the replicas under consideration, T1 and T2 are their current temperatures, and kB is the Boltzmann constant. There exist numerous other formulations of the replica exchange technique, including versions that oper- ate on Monte Carlo simulations [238], use restraining “umbrella” potentials [239], employ simulated annealing between exchanges [240], or exchange based on some parameter other than temperature [241–243]. However, the fundamental shared characteristic is that they all allow more eﬃcient sampling of the free energy landscape of a system than could be achieved with non-interacting parallel computations. For further review of these methods and their application, particularly in the

83 domain of simulated protein folding (which also often employs implicit solvent models), see [244] and [245]. Twelve replicas were used for each of the four simulations, with their temperatures selected according to the optimal spacing algorithm of Rathore et al. [246]. The temperatures were spaced to give a roughly 20% acceptance rate, which has been shown to provide the greatest sampling efficiency in replica exchange [247]. The temperature span for EHPT ran from 300 K to 709.27 K, while HPT ran from 300 K to 744.0 K. MD simulations were carried out in the SANDER module of AMBER 9 [248], with configuration information written and exchanges attempted every 1 ps of simulated time. The SHAKE algorithm was used to constrain bonds involving hydrogen, and an integration timestep of 1 fs was used due to the elevated temperatures applied to some replicas. The RESPA algorithm was used to only evaluate slowly-varying forces at every other timestep. A non-bonded interaction cutoff of 20 A˚ was used, which allowed the side chain of each monomer unit to “see” as far as the side chains of its neighbors. Temperature was maintained using a Langevin thermostat with a coupling constant of 0.5 ps−1. Each replica-exchange simulation was run for 5 ns (5000 attempted exchanges), giving a total simulation time of 240 ns. Simulation of all four systems required a total of approximately 1,400 processor-hours on the Terascale Computing System of the Pittsburgh Supercomputing Center.

8.3 Results

8.3.1 Geometry and Partial Charges

Table 8.1 compares the ring geometry obtained from other groups’ experimental and ab initio theoretical studies of oligothiophene to the geometry obtained from both the HF/6-31G* pentathiophene optimization and the AMBER energy minimizations using the forcefield derived from the HF/6-31G* results. As can be seen, the agreement is generally good, with the root-mean-squared (RMS) error between empirical atom-atom distances and these calculated ab initio distances being 0.024 A.˚ RMS error between empirical and ab initio bending angles is 4.05◦ when angles about the inter-ring C2-C5’ bond are included in the calculation, or 0.71◦ if they are not. This is not a surprising result; the empirical geometries for thiophene and terthiophene were measured in the solid phase, where intermolecular packing interactions would be expected to affect the bending and torsion angles about the inter-ring bond (the most flexible degrees of freedom). By contrast, the calculations reported here implicitly assumed a gas phase molecule in vacuum, where these same parameters would be expected to be more relaxed. Because the torsion (and associated bending) about the inter-ring bond is relatively free in polythiophene and is also sensitive to the molecule’s environment, it is important to compare the values obtained in these calculations with empirical results that were also obtained in the gas phase. Two electron diffraction studies of gas-phase bithiophene have been carried out; the first reported an inter-ring dihedral angle of 146◦ [250], while the second reported 148±3◦ [232]. The values reported here show good agreement with those experiments; the minimum of the HF/6-31G* rigid- rotor potential lies at 145.93◦, and the same dimer minimized with the derived AMBER forcefield has a dihedral angle of 145.07◦ (Table 8.1). The center rings of pentamer structures from ab initio and AMBER minimizations show slightly more planar angles, at 150.15◦ and 151.56◦, respectively. These are still within the experimental margin of error. Moreover, one could expect a pentamer to be more planarized than a dimer, since the increasing length should allow greater electron delocalization along the backbone, which will lower the energy of planar conformations [251]. It is also valuable to examine the effect of the side chains on the inter-ring dihedral angles. There are no known empirical data on how alkyl side chains affect the gas-phase geometry of oligothiophenes; all known studies have been conducted on crystals, where non-bonded interactions with other chains cause almost perfect planarization (leading to the well-known high conductivity of poly(alkylthiophene)s in the solid state) [252]. Other investigators have carried out ab initio studies of methylated and ethylated thiophene dimers with the 6-31G** and 3-21G* basis sets [229,234,253].

84 Ring Geometry from Prior Studies Empirical HF/6-31G* Empirical Empirical MP2/6-31G* Monomer Average [223, Dimer [232] Trimer [230] Average [233] [249] 229, 232, 233] S1-C2 1.71 1.73 1.74 1.74 1.73 C2-C3 1.37 1.37 1.38 1.35 1.38 C3-C4 1.42 1.45 1.41 1.43 1.42 C4-H4 1.08 1.12 1.07 1.09 C5-H5 1.08 1.12 1.07 1.08 C2-C5’ N/A 1.46 1.45 1.46 1.45 S1-C2-C3 111.47 111.80 110.70 110.97 111.10 S1-C5-C4 111.47 112.30 110.70 111.60 111.62 S1-C5-H5 119.85 119.83 121.43 119.99 S1-C2-C5’ N/A 121.90 119.85 122.79 120.88 C2-S1-C5 92.20 91.70 92.40 91.59 92.05 C2-C3-C4 112.45 111.90 113.30 113.08 112.72 C2-C3-H3 123.28 123.12 123.38 122.82 C3-C2-C5’ N/A 126.30 110.70 128.33 128.50 S1-C2-C5’-S1’ N/A 148.00 173.80 147.80 142.24

Ring Geometry from This Work GAUSSIAN AMBER AMBER HF/6-31G* All-Atom All-Atom Pentamer Dimer Pentamer S1-C2 1.74 1.74 1.74 C2-C3 1.35 1.35 1.35 C3-C4 1.43 1.39 1.39 C4-H4 1.07 1.08 1.08 C5-H5 1.07 1.08 1.08 C2-C5’ 1.46 1.47 1.47 S1-C2-C3 110.73 110.41 110.15 S1-C5-C4 110.70 110.72 110.15 S1-C5-H5 120.26 120.45 120.40 S1-C2-C5’ 120.86 121.00 121.14 C2-S1-C5 91.89 91.25 91.62 C2-C3-C4 113.32 114.02 114.03 C2-C3-H3 123.37 123.20 123.34 C3-C2-C5’ 128.41 128.59 128.70 S1-C2-C5’-S1’ 150.15 145.07 151.56

Table 8.1: Optimized geometry of the thiophene ring and side chain from empirical studies and selected ab initio studies (upper table), compared with the geometries obtained in this study (lower table). Distances are in angstroms and angles are in degrees. Atom nomenclature is given in Figures 8.1 and 8.2. For oligomers of degree three and greater, the reported values are for the central ring, with the exception of the terminal C5-H5 distance. The values from both the ab initio calculations and the molecular dynamics parameterization agree well with the empirical results and other groups’ ab initio calculations on similar molecules; RMS error is 0.024 A˚ for distances and 0.71◦ for angles (not including angles about the inter-ring C2-C5’ bond). Bending and torsion about that inter-ring bond are the principal locus of disagreement, and including them in RMS error calculations raises angle error to 4.05◦.

85 Inter-Ring Torsions from Prior Studies Empirical Methyl Methyl Dimer Methyl Dimer Ethyl Dimer Trimer (Solid) [254] HF/3-21G* [234] HF/6-31G** [234] HF/3-21G* [253] 172.85 106.35 113.75 108.00 Inter-Ring Torsions from This Work EH Pentamer EH Pentamer EH Pentamer HF/6-31G* AMBER All-Atom AMBER United-Atom 121.09 158.51 154.84 Hexyl Pentamer Hexyl Pentamer Hexyl Pentamer HF/6-31G* AMBER All-Atom AMBER United-Atom 122.55 159.225 141.88

Table 8.2: S1-C2-C5’-S1’ dihedral angles (in degrees) for alkylated oligothiophenes as reported by other groups and as observed in this work. The ab initio results indicate that the longer side chains used here are more planarizing than the shorter chains studied in prior work, although the rings remain substantially rotated relative to each other. Results from energy minimizations in the AMBER forceﬁeld show more planar conﬁgurations than the ab initio. Neither these calculations nor other groups’ results matches the empirical solid-state value, but gas phase calculations would not be expected to produce a good match.

All of these calculations have indicated that adding alkyl side chains causes gas-phase oligothiophene to deplanarize, and comparison of methyl and ethyl chains with the same basis set and software indicates that longer chains cause more deplanarization [229, 253]. The results of adding all-atom or united-atom ethylhexyl or hexyl chains to pentathiophenes are shown in Table 8.2, and follow the same trend seen in others’ calculations. The values found in HF/6-31G* optimizations of alkylated pentamers are slightly higher (more planar) than those observed in previous studies of dimers, possibly due once again to the increased electron delocalization of longer oligomers. More importantly, the AMBER minimizations (using the default steepest descent/conjugate gradient method) of those same molecules show substantially less deplanarization than the HF optimizations. There is no definitive explanation for this observation at this time. However, it is worth considering that relaxed-rotor potentials of alkylated thiophene dimers show a broad and flat potential well encompassing both syn and anti conformations (in contrast to the well-defined minima of the plain bithiophene rotor potential; see below) [229,234,253]. It is possible that the AMBER minimizations became trapped in local minima along the bottom of this potential well and were not able to reach the global minimum. Partial charges derived with the RESP procedure are shown in Table 8.3, and are generally unremarkable. As expected, the molecules are overall apolar, with the carbons and sulfurs of the thiophene rings showing higher partial charges (greater electron delocalization and polarizability) than the aliphatic side chains.

8.3.2 Rigid Rotor Potential in AMBER and GAUSSIAN

Figure 8.5 shows the HF/6-31G* rigid rotor potential for bithiophene, and compares it to the AMBER rotor potential obtained using the all-atom and united-atom derived forcefields. The cosine fit gives an excellent match, with the RMS error between AMBER and ab initio energies being 0.047 kcal/mol for the all-atom fit and 0.063 kcal/mol for the united-atom fit. The shape is also consistent with that observed by other groups using a self-consistent field approach and similar basis sets, and using both rigid and relaxed rotations [232–234]. The potential shows two sets of minima, corresponding to syn-like and anti-like configurations. As would be expected, the syn-like conformation has higher energy due to the close approach of the two sulfur atoms. The minima differ

86 Partial Charges for Ethylhexyl Monomers Partial Charges for Hexyl Monomers N-Term Central C-Term N-Term Central C-Term S1 -0.037 S1 -0.108 S1 -0.049 S1 -0.040 S1 -0.060 S1 -0.006 C2 -0.104 C2 -0.007 C2 -0.234 C2 -0.130 C2 -0.075 C2 -0.310 H2 N/A H2 N/A H2 0.222 H2 N/A H2 N/A H2 0.230 C3 0.220 C3 -0.033 C3 -0.001 C3 0.230 C3 0.026 C3 0.004 C4 -0.329 C4 -0.163 C4 -0.117 C4 -0.370 C4 -0.255 C4 -0.150 H4 0.184 H4 0.126 H4 0.122 H4 0.200 H4 0.173 H4 0.150 C5 -0.188 C5 0.09 C5 0.048 C5 -0.150 C5 0.110 C5 0.040 H5 0.222 H5 N/A H5 N/A H5 0.210 H5 N/A H5 N/A CA1 -0.240 CA1 -0.14 CA1 -0.271 CA1 -0.100 CA1 -0.050 CA1 -0.100 HA1 0.050 HA1 0.04 HA1 0.075 HA1 0.030 HA1 0.031 HA1 0.040 HA2 0.050 HA2 0.04 HA2 0.075 HA2 0.030 HA2 0.031 HA2 0.040 CA2 0.421 CA2 0.393 CA2 0.391 CA2 0.120 CA2 0.107 CA2 0.150 HA3 -0.040 HA3 -0.023 HA3 -0.033 HA3 -0.010 HA3 0.003 HA3 -0.020 CA3 -0.435 CA3 -0.475 CA3 -0.363 HA4 -0.010 HA4 0.003 HA4 -0.020 HA4 0.090 HA4 0.097 HA4 0.066 CA3 -0.080 CA3 -0.115 CA3 -0.075 HA5 0.090 HA5 0.097 HA5 0.066 HA5 0.016 HA5 0.020 HA5 0.004 CA4 0.074 CA4 0.103 CA4 0.084 HA6 0.016 HA6 0.020 HA6 0.004 HA6 0.003 HA6 -0.004 HA6 -0.007 CA4 -0.010 CA4 0.003 CA4 -0.007 HA7 0.003 HA7 -0.004 HA7 -0.007 HA7 -0.001 HA7 -0.002 HA7 -0.002 CA5 0.133 CA5 0.142 CA5 0.135 HA8 -0.001 HA8 -0.002 HA8 -0.002 HA8 -0.010 HA8 -0.014 HA8 -0.018 CA5 0.180 CA5 0.183 CA5 0.180 HA9 -0.010 HA9 -0.014 HA9 -0.018 HA9 -0.030 HA9 -0.031 HA9 -0.035 CA6 -0.250 CA6 -0.277 CA6 -0.267 HAA -0.030 HAA -0.031 HAA -0.035 HAA 0.057 HAA 0.061 HAA 0.057 CA6 -0.220 CA6 -0.239 CA6 -0.230 HAB 0.057 HAB 0.061 HAB 0.057 HAB 0.050 HAB 0.050 HAB 0.050 HAC 0.057 HAC 0.061 HAC 0.057 HAC 0.050 HAC 0.050 HAC 0.050 CB1 -0.133 CB1 -0.1 CB1 -0.025 HAD 0.050 HAD 0.050 HAD 0.050 HB1 0.042 HB1 0.037 HB1 -0.004 HB2 0.042 HB2 0.037 HB2 -0.004 CB2 -0.019 CB2 -0.047 CB2 -0.058 HB3 0.000 HB3 0.008 HB3 0.007 HB4 0.000 HB4 0.008 HB4 0.007 HB5 0.000 HB5 0.008 HB5 0.007

Table 8.3: Partial charges for ethylhexylthiophene and hexylthiophene monomers, in electron units. The charges reported here are for the all-atom model; the charges for the united-atom model simply sum the aliphatic hydrogen charges into the corresponding carbons. See Figures 8.1 and 8.2 for atomic nomenclature.

87 in energy by only 0.64 kcal/mol in the best case, and the barriers between them are relatively small, which implies that bithiophene and higher polythiophenes should show a mixture of syn and anti backbone angles. This is consistent with solution NMR and gas-phase electron diffraction studies showing that bithiophene exists in a mixture between syn and anti conformations, with the anti conformation slightly favored (by 8 ± 4% in the gas phase) [232, 255]. The energy barriers in this rigid rotor potential are compared in Table 8.4 to those observed by other groups in experimental and computational studies. The height of the highest barrier has been quantified by NMR in a liquid crystal solvent as 5±2 kcal/mol [256]; the value of approximately 2.2 kcal/mol observed in this work is slightly outside that range, but one would expect the rotational barrier to be lower in gas phase than in a highly ordered solvent. For the energy difference between the anti and syn conformers, there is little agreement in the literature – measurements with different techniques have ranged over almost an order of magnitude, from 0.18 to 1.16 kcal/mol [232,256,257]. Four values for this gap can be extracted from each curve in Figure 8.5, depending on which minima one chooses to compare. All of the calculated values fall within the experimental range. The agreement between the bithiophene rotor potential found in this work and that observed in other investigators’ calculations is qualitatively good, but the actual energy values differ slightly, as also seen in Table 8.4. The observed maximum barrier height of 2.2 kcal/mol agrees better with the HF/6-31G** calculations of Hernandez et al. than with the slightly lower values of prior HF/6- 31G* calculations [231–234]. The observed gap between syn and anti minima can be either higher or lower than previous ab initio calculations, depending on which of the four values one chooses. The majority of these differences are likely attributable to the use of different calculation techniques. Prior groups have performed relaxed-rotor potential calculations, rotating only over 180◦, with a minimum spacing of 10◦ between sample points. By contrast, the potentials of Figure 8.5 are rigid rotations over a full 360◦ with 5◦ spacing. This higher resolution reveals the aforementioned asymmetry of the rotor potential and may be able to sample deeper minima, while the lack of atomic position relaxation would be expected to produce higher energies at the peaks, where atoms are undesirably close together.

8.3.3 Replica Exchange MD In Implicit Solvents

Raster plots for all replicas and temperature histories for selected replicas of all four replica exchange simulations are shown in Figures 8.6-8.13. The corresponding exchange rates are given in Table 8.5. Taken together, these figures and table show that all replicas achieved exchange rates above 20%, and that all appear to have performed an adequate random walk in temperature space over the 5 ns of simulation time. A few replicas show exchange rates above 30%, which is slightly higher than optimal, but not so high as to have compromised the quality of the simulation. Figures 8.14 and 8.15 show the potential of mean force for EHPT and HPT in implicit water and lipid, plotted against the end-to-end distance of the simulated decamer. Numeric values for the minima are given in Table 8.6. PTs are well-known to form aggregates, particularly in poor solvents [176, 183], and it has been suggested that the first step in this process is the collapse of an individual chain upon itself to form “intramolecular aggregates” [184, 185]. Such collapse brings the two ends of the molecule closer together, and thus the end-to-end distance can be viewed as a measure of the ability of the medium to solvate the polymer under study. In the simulations reported here, both polymers show a lower energy in water than in lipid for conformers with lower end-to-end distance. This implies that an aqueous environment should favor aggregation, as predicted. The other notable feature of these PMFs is the difference in response when the solvent is changed from implicit water to implicit lipid. Both molecules show a shift towards more extended conformations, implying that lipid is a more favorable solvent than water, but EHPT shifts by approximately 1.5 A,˚ whereas HPT shifts by only 0.5 A.˚ While not a large effect, this does suggest that EHPT is more “at home” in non-polar lipid environments than HPT is. Another useful property to examine is the distribution of the backbone dihedral angles. The bithiophene rotor potential presented above describes the “natural” tendencies for the mutual ori-

88 Rigid Rotor Potential of Bithiophene, ab initio vs. AMBER 2.5

1.5

1 Energy (kcal/mol)

0.5

HF/6-31g* 0 AMBER SCNB=2 AMBER SCNB=8

-0.5 -150 -100 -50 0 50 100 150 C3-C2-C5-C4 Dihedral Angle (degrees)

Figure 8.5: Rigid rotor potential of plain bithiophene, as computed by ab initio HF/6-31G*, compared to the same rigid rotor potential computed in AMBER using the all-atom (SCNB=2) and united-atom (SCNB=8) parameterizations. RMS error for the all-atom ﬁt is 0.047 kcal/mol, and for the united-atom ﬁt is 0.063 kcal/mol.

89 Empirical Energies Solution Gas Phase e− Gas Phase NMR [256] Diﬀraction [232] Fluorescence [257] Trough-to-Peak Barrier 5±2 Not given Not given Anti-to-Syn Barrier 0.200 0.180 1.16±0.13 Other ab initio Energies HF/6- HF/6-31G** MP2/6-31G* 31G* [234] [231, 233] [231–233] Trough-to-Peak Barrier 1.931 2.200 1.836 Anti-to-Syn Barrier 0.708 0.700 0.505 Energies from This Work GAUSSIAN AMBER AMBER HF/6- All-Atom United-Atom 31G* Trough-to-Peak Barrier 2.228 2.173 2.153 Anti-to-Syn Barrier 0.640 to 1.135 0.626 to 1.176 0.608 to 1.140

Table 8.4: Relative energies from the rigid rotor potentials plotted in Figure 8.5, compared with results from other groups. The “trough to peak” value is the height of the highest barrier in the entire potential, while the “anti to syn” value reports the distance between the two local minima corresponding to the syn and anti conformations of bithiophene. All energies are given in kcal/mol. The AMBER parameterizations match well with the ab initio calculation, but both sets of calculations show a slightly higher trough-to-peak energy and a slightly lower anti-to-syn energy than previous ab initio results using the same basis set. However, the energies calculated in this work do fall within the broad range of empirical values.

EHPT Water EHPT Lipid HPT Water HPT Lipid Temp (K) Rate Temp (K) Rate Temp (K) Rate Temp (K) Rate 300.00 21.5% 300.00 20.8% 300.00 26.7% 300.00 26.4% 317.28 33.2% 317.28 33.2% 315.00 38.6% 315.00 37.4% 345.37 22.7% 345.37 23.8% 345.75 24.2% 345.75 23.5% 375.78 22.6% 375.78 23.3% 378.50 23.9% 378.50 25.4% 408.71 23.8% 408.71 23.2% 415.00 25.0% 415.00 26.4% 444.37 23.9% 444.37 23.3% 453.50 27.5% 453.50 28.6% 482.97 24.2% 482.97 23.8% 494.50 27.1% 494.50 27.4% 524.77 25.0% 524.77 24.4% 541.00 26.0% 541.00 26.4% 570.02 25.1% 570.02 25.5% 591.00 26.7% 591.00 27.9% 619.02 25.2% 619.02 25.2% 645.00 26.9% 645.00 27.3% 672.06 35.1% 672.06 35.8% 705.00 38.0% 705.00 38.3% 709.27 22.7% 709.27 23.5% 744.00 24.7% 744.00 25.0%

Table 8.5: Exchange rates for the four implicit solvent replica exchange simulations. All four show exchange rates that are mostly between 20% and 30%, in the range that produces an optimal exploration of temperature space.

90 EHPT Implicit Water REMD EHPT Implicit Lipid REMD

700 700

600 600

500 500 Temperature (K) Temperature Temperature (K) Temperature 400 400

300 300 2000 2100 2200 2300 2400 2500 4500 4600 4700 4800 4900 5000 Exchange Attempt Exchange Attempt

Figure 8.6: EHPT implicit water raster Figure 8.7: EHPT implicit lipid raster

HPT Implicit Water REMD HPT Implicit Lipid REMD

700 700

600 600

500 500 Temperature (K) Temperature

400 (K) Temperature 400

300 300 1000 1100 1200 1300 1400 1500 3300 3400 3500 3600 3700 Exchange Attempt Exchange Attempt

Figure 8.8: HPT implicit water raster Figure 8.9: HPT implicit lipid raster

Raster plots, with one dot per successful exchange, sampled from 500 ps (from a run length of 5 ns) from each of the four implicit solvent replica exchange simulations (EHPT and HPT in water and lipid). Each plot demonstrates regular exchanging of all the replicas within that simulation.

91 EHPT Implicit Water REMD EHPT Implicit Lipid REMD 750 750

700 700

650 650

600 600

550 550

500 500 Temperature (K) Temperature (K) 450 450

400 400

350 350

300 300

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Exchange Attempt Exchange Attempt

Figure 8.10: EHPT implicit water replica tem- Figure 8.11: EHPT implicit lipid replica temperature history perature history

HPT Implicit Water REMD HPT Implicit Lipid REMD 750 750

700 700

650 650

600 600

550 550

500 500 Temperature (K) Temperature (K) 450 450

400 400

350 350

300 300

0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 Exchange Attempt Exchange Attempt

Figure 8.12: HPT implicit water replica tem- Figure 8.13: HPT implicit lipid replica temperature history perature history

Temperature histories over the full 5 ns run for two randomly-selected replicas (solid line and dotted line in each panel) from each of the four implicit solvent replica exchange simulations. Each replica visits all temperatures multiple times over the course of the simulation, indicating that the exchange rate between replicas was suﬃcient to produce a random walk that fully explored temperature space.

92 PMF of EHPT Configuration in Implicit Solvent Replica Exchange 5

4.5 0.6 EHPT Implicit Water 0.5 EHPT Implicit Lipid 4 0.4 0.3 3.5 0.2 0.1 3 0 -0.1 2.5 34 34.5 35 35.5 36 36.5 37 37.5

2 Energy (kcal/mol)

1.5

0.5

0 20 22 24 26 28 30 32 34 36 38 Polymer End-to-End Distance (Angstroms)

Figure 8.14: PMF for EHPT in implicit water and lipid

PMF of HPT Configuration in Implicit Solvent Replica Exchange 5

HPT Implicit Water 4.5 HPT Implicit Lipid 0.6 0.5 4 0.4

3.5 0.3 0.2 3 0.1 0 2.5 -0.1 34 34.5 35 35.5 36 36.5 37 37.5 2 Energy (kcal/mol)

1.5

0.5

0 20 22 24 26 28 30 32 34 36 38 Polymer End-to-End Distance (Angstroms)

Figure 8.15: PMF for HPT in implicit water and lipid

Potentials of mean force for the end-to-end distance of EHPT and HPT in implicit water and lipid solvents. The insets display detail of the potential well around 34 to 37 A.˚ PMFs were computed and errors estimated using the WHAM code of John Chodera, as described in [258]. Both polymers show lower energy for shorter conﬁgurations in water than in lipid, suggesting that aqueous solvation favors aggregation. Extended conformations are more favored for EHPT in lipid than HPT in lipid (as seen in the insets), consistent with the empirically known higher solubility of EHPT in alkanes.

93 entation of thiophene rings in a polymer backbone. If different solvents change the molecule’s propensity to adopt certain conformations, this will be observed in the dihedral angle distribution as a deviation from the “ideal” distribution predicted by the rotor potential. In this case, the potential of Figure 8.5 shows effectively two minima, with the lower-energy minimum corresponding to an anti-like configuration and the higher-energy minimum corresponding to a syn-like configuration; one would therefore expect a dihedral angle distribution from a PT simulation to show two peaks, with the peak at anti-like angles higher than that at syn-like angles. It should be kept in mind, however, that the simple presence of side chains can shift and flatten the torsional potential (as described in Subsection 8.3.2), and that this phenomenon will interact with solvent effects to produce the final distribution. The backbone dihedral angle distributions observed in these implicit solvent replica exchange simulations are shown in Figure 8.16 and 8.17, and the angles corresponding to the peaks are in Table 8.6. All four simulations show the expected two peaks corresponding to syn and anti orientations, but there are also visible shifts in the relative peak magnitudes across polymer and solvent conditions. EHPT is consistently biased towards an anti configuration, whereas HPT is biased towards the syn configuration (although the two peaks are roughly equal height for HPT in lipid). Both polymers show an increase in the probability of anti-like angles when moving from water to lipid. Since the anti configuration is favored in the potential for free rotation in vacuum, one can consider simulations that show more of an anti bias to represent polymers that are in a more “relaxed” or “free” environment. By this interpretation, the simulation results confirm that lipid is a better solvent than water for both EHPT and HPT, and that EHPT is overall more soluble than HPT, all consistent with the polymers’ known properties. Both peaks in all four backbone dihedral distributions are shifted by 50-60◦ from the fully planar syn and anti configurations. This is consistent with the effect of alkyl side chains described in Subsection 8.3.1, and non-planar configurations have also been observed for hydrophilic PTs in water, suggesting that deplanarization is not a solvent-driven effect per se [207]. Ab initio studies of PT polarizability suggest that even with this substantial deplanarization, there should still be modest π orbital overlap and electron conductivity [251]. That prediction can be further evaluated by examining the persistence length of the polymers. The persistence length is defined as half the Kuhn length, which in turn can be considered as the average length over which the polymer does not bend. Since bends interrupt electronic conjugation and decrease conductivity, longer persistence lengths imply greater conductivity, and comparisons of the persistence length between two simulations can provide a rough estimate of the relative conductivity of the polymers in those simulations. For the short molecules simulated here, which contain only a few Kuhn lengths, the persistence length λ is best estimated by a modification of the method given in [259]:

hR2i λ = o (8.4) 2n × Lo

2 where hRoi is the mean squared end-to-end distance, n is the degree of polymerization, and Lo is the length of a monomeric unit (taken to be 3.9 A˚ in this case). For light-scattering measurements, the formula is usually couched in terms of molecular weight and monomer mass, since the precise degree of polymerization is not directly known. In the case of simulation, n is known, and the weight terms can be canceled out. The persistence lengths for the four implicit solvent simulations were calculated using 5 ns of trajectory snapshots, taking at each timestep the configuration of the replica that was visiting the 300 K commanded temperature. These results are shown in Table 8.6. As would be expected from the end-to-end distance PMFs, both polymers show longer persistence lengths in lipid than they do in water, and EHPT has a higher persistence length in lipid than does HPT, due to EHPT’s more extended conformation. All of the reported persistence lengths are significantly different from each other (p < 0.0001 by Student’s t-test). However, it is worth nothing that these low p-values are due to the ability of MD simulations to generate an arbitrarily large number of datapoints, and that there is a great deal of overlap between the persistence length distributions of the four simulations.

94 Distribution of EHPT Backbone Dihedral Angles 0.035

EHPT Implicit Water EHPT Implicit Lipid 0.03

0.025

0.02

Probability 0.015

0.01

0.005

0 0 20 40 60 80 100 120 140 160 180 Angle (degrees)

Figure 8.16: Distribution of backbone dihedral angles for EHPT

Distribution of HPT Backbone Dihedral Angles 0.035 HPT Implicit Water HPT Implicit Lipid 0.03

0.025

0.02

Probability 0.015

0.01

0.005

0 0 20 40 60 80 100 120 140 160 180 Angle (degrees)

Figure 8.17: Distribution of backbone dihedral angles for HPT

Distribution of dihedral angles between the individual rings of the EHPT and HPT backbones during 5 ns of implicit solvent replica exchange molecular dynamics. All samples were taken from the 300 K “virtual trajectory”.

95 EHPT Water EHPT Lipid HPT Water HPT Lipid Primary Peak (degrees) 115-122 116-123 55-64 56-61* Secondary Peak (degrees) 59-62 57-65 115-123 116-125* End-to-End PMF 35.0 35.0-36.5 35.0-35.5 35.0-36.0 Minimum (A)˚ Persistence Length 3.36±1.59 3.80±1.34 3.34±1.64 3.68±1.30 (300K Snapshots)

Table 8.6: Dihedral angle distribution peaks, minima of the end-to-end distance potential of mean force, and persistence length (as calculated from the 300K trajectory) for the implicit solvent replica exchange simulations of EHPT and HPT. Persistence length is reported in terms of number of repeat units, and the range spanned by two standard deviations is also reported. For dihedral distributions, (*) denotes a run where the primary and secondary peaks are of approximately equal height. HPT is found to favor a more syn orientation, whereas EHPT favors more anti backbone angles. EHPT shows more tendency towards extended conformation than HPT when in implicit lipid. This is associated with a prediction of greater conductivity (greater persistence length) in lipid, although the diﬀerence is much smaller than the width of the distribution.

The more important point is that all four observed lengths hover around roughly 3.5 repeat units, giving a Kuhn length of approximately 7 thiophene rings. This implies a polymer that is behaving largely as a rigid “twisted rod” and that, even in the worst case of HPT in water (mean persistence length of 3.34 repeat units) is not collapsed into a “ring” conformation with the two ends nearly touching. It is also noteworthy that the persistence length varies substantially over the course of the simulation, with the ends of the distribution for all four polymers encompassing an extremely coiled form (persistence length roughly 1.5 units) and a fully extended form (persistence length roughly 5 units). At the same time, empirical measurements of the persistence length of regioregular PTs in good solvent (chloroform) using UV-visible absorption spectroscopy tend to observe lengths of 10 repeat units or greater, substantially above what is seen here [260, 261]. One property not clearly reflected in these data is the well-known “solvatochromic” effect, where some regioregular polythiophenes change colors (due to changes in backbone angle and resulting changes in electron delocalization) when solvent quality is altered from good to poor. This property is known to hold for HPT, and has been theorized to occur through the formation of the previously- mentioned helical intramolecular aggregates [184, 185]. Solvatochromism has not been reported in the literature for EHPT, and has not been observed in chloroform/methanol tests of the EHPT samples used in previous chapters. If the helical aggregation mechanism does indeed drive HPT solvatochromism, one could argue that it is seen in the results of Figures 8.16 and 8.17; a PT helix should show predominantly syn angles [184], and HPT does show more syn angles than EHPT, particularly when in water (which should be a non-solvent). However, solvatochromism should also manifest as changes in the electronic conjugation length (as reflected in the persistence length). Shorter conjugation lengths will cause the polymer to absorb light at shorter wavelengths, implying a redder color. Since solvatochromism is associated with a shift to purple colors in poor solvents, one would expect greater conjugation (longer wavelength absorption) in those solvents. This effect is not seen for EHPT or HPT in these replica exchange simulations – as noted above, both polymers show longer persistence lengths in lipid than in water, opposite to the expected trend. The evidence for solvatochromism in these simulations is therefore ambivalent at best.

8.4 Discussion

This chapter has presented the development and validation of a new set of parameters that should enable accurate molecular dynamics simulations of polythiophenes. Although the results presented

96 here are for EHPT and HPT, the same parameters could be used to study any arbitrary PT as long as the partial charges are properly assigned. In the course of proving the new device concept, this work has therefore produced a useful new tool for studying the molecular structure of PTs in a variety of environments. The AMBER parameters derived from ab initio computations show good geometrical agreement with both empirical and previous ab initio results, which further indicates that they can be relied on to predicting the configuration of other polymers. The HF/6-31G* rotor potential and its AMBER cosine fit are also well-matched to empirical data. Although these potentials do not match perfectly with the results obtained by prior investigators, the differences in results can largely be explained by subtle differences in methodology. It should also be noted that the discrepancies in energies are extremely small, about 0.2 kcal/mol in the worst case. This small barrier is unlikely to have a detectable effect on the behavior of a simulated system, as it is of the same order of magnitude as the thermally induced potential energy fluctuations of a system at room temperature. That relatively free rotation will be important, since the minimum-potential- energy configurations of EHPT and HPT as predicted by AMBER (in Table 8.2) are substantially different from those predicted in the HF/6-31G* optimization. In addition to locating the correct dihedral angle minima, the replica exchange simulations show other features that match well with expectations. Both EHPT and HPT are seen to adopt more extended conformations in lipid and more collapsed conformations in water. Consistent with its greater overall solubility, EHPT is able to achieve a more extended state than HPT. Both polymers appear to favor a “twisted rod” conformation, with EHPT showing a predominance of anti-like angles and HPT a predominance of syn-like angles. This may imply that HPT is attempting to form a helical aggregate, particularly in water. PT helices are predicted to have at least 12 units per turn, so the full helix is not observable in these decamer simulations [184]. The relatively short length of the simulated molecules (compared to the high degrees of polymerization obtainable in real syntheses) may also explain why these data show only a slight collapse on moving from lipid to water solvation. These short rod-like molecules do not have much capacity for collapse, whereas longer molecules would have more flexibility to fold back upon themselves and stabilize bent conformations. Not all of the results from the replica exchange simulations agree with expectations. The persistence lengths observed in all four simulations are shorter than those empirically measured for PTs in good solvents (which should be the shortest lengths possible). The persistence lengths also show an inverse solvatochromic effect, with both polymers showing more conjugation in lipid than in water. These deviations from expectation most likely do not represent flaws in the forcefield, but can instead be attributed to limitations of the generalized Born model (and implicit solvent models in general). While variations in the solvent dielectric constant do have an effect on PT conformations (as evidenced by these observations of different behaviors in different implicit solvents), real solvents also alter molecular configuration through electrostatic and VDW interactions between solvent and solute. Lipid solvents in particular will also exhibit a high viscosity, since it is energetically unfavorable for the polymer to displace many long alkyl chains. Finally, there is the fact that the empirical data being used as a reference were collected using long-chain PTs interacting with each other in solution, whereas the data presented here represent short-chain PTs that are unable to experience intermolecular interactions with neighboring polymers. Without any possibility of aggregation or other intermolecular attraction, it is unsurprising that these results are not able to fully reproduce the range of known PT properties. In summary, these results demonstrate a straightforward and reliable method for producing forcefields appropriate for molecular dynamics studies of polythiophenes. Preliminary implicit solvent simulations show good agreement with expectations and/or empirical data on simple measures such as backbone angles, but deviate from expectations on more complex measures related to polymer conductivity. Much of this is likely due to limitations of the implicit solvent regime, which by ne- cessity must sacrifice accuracy in order to achieve its impressive computational efficiency. In the next chapter, explicit solvents (water and bilayer) will be introduced, and the results from those simulations will be shown to be more consistent with empirical data. Through this explicit modeling of the different components of the bilayer, the results of Chapter 7 can also be explained.

97 98 Chapter 9

Molecular Dynamics of Polythiophenes In Lipid Bilayers

9.1 Introduction

In this chapter, the polythiophene parameters developed in Chapter 8 are applied to simulate PT interactions with an explicit lipid bilayer, with a particular focus on explaining the diﬀerent behaviors of EHPT and HPT seen in the artiﬁcial bilayer experiments of Chapter 7. This chapter also introduces the use of steered molecular dynamics (SMD) and umbrella sampling, two more techniques for enhancing the sampling of key reaction coordinates during an MD simulation. These techniques are applied to answer two key questions:

1. What is the energy proﬁle of EHPT or HPT insertion into a lipid bilayer? 2. What conformation does EHPT or HPT adopt at its minimum-energy point?

The original hypotheses laid out in Chapter 3 and the results of Chapter 7 suggest that EHPT inserts into the bilayer while HPT does not. If this is the case, energy analysis of the MD simulations in this chapter should indicate that EHPT insertion will be favorable (negative free energy of insertion) while HPT will not (positive free energy of insertion). The answers to the second question will provide a qualitative explanation for the quantitative answers to the ﬁrst question. An initial guess at the results (following the hypothesis just presented) might have EHPT dissolved in and bridging the lipid phase of the membrane, while HPT remains in aqueous solution or associated with the bilayer surface.

9.2 Methods

9.2.1 Simulation Parameters and Coordinates

All simulations in this chapter used the same system: a square patch of phospholipid bilayer in a box of explicit water, with an EHPT or HPT chain interacting with that bilayer in various conﬁgurations. The necessary parameters for polythiophenes were presented in Chapter 8. Parameters for water (the TIP3P “triangular water” model) and a well-equilibrated “solvation box” are available as part of the standard AMBER forceﬁeld. This left the phospholipid parameters and initial structure to be determined. The lipid parameters and bilayer starting structure used in this work were those of Dr. D. Peter Tieleman, as available at http://moose.bio.ucalgary.ca/ and described in [262]. These

99 parameters have been successfully used to model a number of interactions between membranes and hydrophobic macromolecules, including the action of membrane-disrupting antimicrobials [212], the insertion of transmembrane voltage-sensitive helices [213], the behavior of transmembrane peptides when the bilayer is strained [263], and the function of transmembrane ion channels [219]. The specific lipid used in these studies was dimyristoylphosphatidylcholine (DMPC). DMPC has slightly shorter tails than typical neuronal membrane lipids, but has the useful property that its gel-to- liquid transition temperature of 296 K is well below the target simulation temperature of 310 K. Since real membranes are believed to always be in the liquid crystalline phase (due to the presence of cholesterol and other fluidizing factors), DMPC was the best choice for this experiment. Other longer-chain alternatives such as dipalmitoylphosphatidylcholine (DPPC), tend to have transition temperatures well in excess of 310 K. Because the original DMPC parameters were designed for the GROMACS [264] forcefield, modifications were necessary to make them compatible with the AMBER-format polythiophene parameters. The parameters were translated into AMBER-appropriate units and the partial charges replaced with the RESP partial charges given in [265] for greater consistency. As is appropriate for RESP charges and AMBER forcefields, 1-4 electrostatic interactions were scaled by a factor of 1.2 (SCEE=1.2). 1-4 VDW interactions were scaled by a factor of 8 (SCNB=8.0), which is more consistent with the united-atom representation of aliphatic carbons that is used in the Tieleman parameters. As noted in Chapter 8, a corresponding united-atom version of the polythiophene parameters was created by re-tuning the cosine fit of the ab initio rigid-rotor potential and using the Ryckaert-Bellemans united alkane torsion potential [266] for the side chains. The alkyl side chains of EHPT and HPT therefore used the same parameters as the alkyl tails of DMPC. These united-atom PT parameters were used in all the simulations reported in this chapter. Before use in production simulations, a bilayer structure derived from that used in [262] (128 DMPC lipids with 3655 waters, equilibrated for 1 ns) was re-heated with the revised parameters using a Langevin thermostat at 310 K and re-equilibrated for 300 ps. The reheating was done with constant box dimensions of 62 × 62 × 67 A˚3 simulation box, as given in the original file. This gives an area of 60.06 A˚2/lipid, consistent with the empirically measured value [267]. Fixing the area at the correct value appeared to normalize other bilayer structural factors. For instance, the thickness of the hydrocarbon layer (empirically measured as about 12.5 A˚ [267]) had a mean value of 12.58 A˚ in the final 100 ps of equilibration. For polythiophene/membrane simulations, a minimum-energy conformation (as judged from end- to-end distance) was extracted from the replica exchange simulations of EHPT and HPT and con- verted to the united-atom representation. This PT configuration was then placed parallel to the bilayer normal, with its N-terminal and C-terminal rings at least 12 A˚ away from either leaflet (including the periodic image). Extra water was added to the simulation box in order to fully solvate the polymer, yielding a total of 25,800 atoms for EHPT and 25,840 atoms for HPT. These structure file manipulations and visualization of all simulation results were performed using the VMD program [268]. The completed polymer/bilayer configurations were minimized (with the polymer backbone and one glycerol atom of each lipid restrained) to remove close contacts. 100 ps of further heating and equilibration was performed after minimization, with the same restraints in place. All simulations of PTs near the bilayer were conducted in a constant volume ensemble, with a simulation box size of 62 × 62 × 89 A˚3. Electrostatic interactions were handled using the particle mesh Ewald (PME) method, with the direct sum and VDW interactions truncated at 12 A.˚ All bonds involving hydrogen were constrained with the SHAKE algorithm, allowing an integration timestep of 2 fs. In order to enhance the computation speed, the program NAMD (version 2.6b) was used for all simulations in this chapter [269]. NAMD supports the AMBER forcefield format and calculation options, but provides better performance than AMBER’s PMEMD executable on this particular system.

100 9.2.2 Steered Molecular Dynamics

Before describing the SMD method and its application to the problem at hand, it is worth explaining why the replica exchange method used in Chapter 8 was not applied to the DMPC/PT/water system. The fundamental problem is the relatively poor scalability of replica exchange. When the standard model of exchanging temperatures is used, the number of replicas needed to span a given temperature range scales as approximately the square root of the number of degrees of freedom (i.e., the number of atoms) [243]. The EHPT decamer of Chapter 8 contained 312 atoms. That same decamer in the DMPC/EHPT/water box just described required 25,800 atoms, implying roughly 9 times more replicas, for a total replica count between 100 and 110. A ninefold increase in computational effort is still tractable, but this does not take into account the vastly greater processor time needed per simulation timestep for the explicit solvent simulations. The implicit solvent EHPT simulations required roughly 0.025 processor-seconds per timestep, whereas the explicit solvent simulations in this chapter required roughly 0.9 processor-seconds per timestep – a 36-fold increase in effort. Using these back-of-the-envelope figures, a 5 ns replica exchange simulation of the DMPC/EHPT/water system with a reasonable temperature span would take approximately 62,500 processor-hours, and the DMPC/HPT/water system would need about the same. The total substantially exceeds the computational resources available for this project or most other MD investigations of biomolecular phenomena. Only one replica exchange simulation of peptide insertion into a bilayer is known to have been completed, using one of the fastest supercomputers in the world [214]. There do exist methods to improve the scaling of the replica exchange approach by changing what entity is exchanged. These can be coarsely divided into methods that exchange the Hamiltonian potential energy operator [241–243] and methods in which exchange occurs on only part of the system [270–272]. Although both methods are interesting, neither is effectively applicable to the problem at hand. The Hamiltonian-altering methods require great care and the use of a priori knowledge in selecting the alterations [241, 243]. Moreover, they are not readily implemented with off-the-shelf MD software and are not amenable to standard weighted histogram analysis techniques (meaning that they may increase the amount of simulation time required to obtain a good potential of mean force). Methods in which only part of the system is exchanged are not appropriate for the system under study. The effect of interest is the interaction of the PT with a large fraction of the water and lipid solvent systems. Therefore, one must either exchange the majority of the system (gaining little to no computational savings) or introduce substantial inaccuracy by treating too much of the system with a mean-field approximation [270]. Neither result is an acceptable trade-off. Steered molecular dynamics is one of multiple reasonable alternatives to replica exchange for studying this large and complex system. In an SMD simulation, a molecule of interest is “steered” through a reaction by artificially restraining the molecule’s position to the vicinity of a point that moves along a pre-determined path. SMD is useful for studying a variety of binding and transit phenomena [273], including the movement of peptides through translocons [274], the transit of small molecules through nonspecific porins [275], and the effect of lipophilic anesthetics like halothane on sodium channels [276]. It has specifically been used to study the energy of interactions between hydrophobic molecules and lipid bilayers, including the energy of halothane partitioning from water into lipid [277], unfolding of a transmembrane peptide during AFM extraction [278], and bilayer insertion/removal of a monotopic protein [279]. By integrating the force applied to the molecule at each timestep to obtain the total work, the energy of the reaction under study can be estimated. The principal difficulty of SMD techniques is the need to pull the moving molecule very slowly. If the restraint point moves too quickly, excess work is done due to the “friction” of collisions with solvent molecules, resulting in a severe overestimation of the free energy change. Many techniques exist to overcome this limitation, all centered around the concept of performing multiple pulls of the same molecule [280–284]. SMD was applied to the system under study by pulling the center of mass of the rings of an EHPT or HPT decamer from a Z-axis distance of 52.5 A˚ away from the membrane center (an aqueous configuration parallel to the bilayer normal with the N-terminal residue 12 A˚ away from the membrane head groups) to 0 A˚ from the bilayer center (fully inserted). Since one long simulation

101 is informationally equivalent to or better than several short simulations [280, 283], the longest simulation feasible within the available computational resources was performed. Polymers were pulled into the bilayer at a rate of 0.01 A/ps˚ with a force constant of 10 kcal/mol/A.˚ This was expected to be slow enough to accurately capture the reaction pathway, based on others’ observations with peptides in a translocon [274]. The initial conﬁguration was heated as described in Subsection 9.2.1, but no temperature control was used during the SMD pull in order to more accurately estimate the energy changes. To better estimate the free energy of insertion, the reverse reaction was also simulated by pulling EHPT and HPT decamers from the bilayer center to a point 52.5 A˚ from the center. Following the method of [284], this should allow removal of the frictional work component by averaging of the forward and reverse energies. The initial conﬁguration for the reverse pull was generated by superimposing the polymer and bilayer center of masses, minimizing, and then running dynamics for at least 1 ns with the polymer restrained to remain at 0 A˚ from the center. All SMD runs output the current applied force every 20 fs, and saved system coordinates every 1 ps. All four SMD runs taken together (21 ns of simulation time) required approximately 3,500 processor-hours on the Cray XT3 machine of the Pittsburgh Supercomputing Center.

9.2.3 Umbrella Sampling

As described in Subsection 9.3.1 below, the SMD pulls provided a good view of the reaction pathway for EHPT and HPT insertion, but did not produce a reliable estimate of the free energy, largely due to sampling a questionably physical trajectory for the reverse pull. This could be alleviated by pulling at slower speeds that approach the fully reversible regime, but those speeds are estimated to be about 0.0001 A/ps˚ [282]. Such a simulation would require over 500 ns of total simulated time, and is not guaranteed to achieve a good estimate of the PMF for insertion. Given the availability of the reaction path from the forward SMD runs, it is possible to apply umbrella sampling techniques to extract the PMF without expending the computational effort on multiple SMD pulls. Umbrella sampling, briefly, is an MD simulation technique in which a moving molecule of interest is restrained at a variety of different positions (“windows”) using harmonic restraining potentials (“umbrella potentials”). The coordinate and energy histograms obtained from these simulations can be combined with essentially the same Weighted Histogram Analysis Method (WHAM) used for replica exchange data. Umbrella sampling has been suggested to be equivalent to SMD on an information-gained-per-simulation-unit basis, but may be easier to set up for a given problem if the reaction path is known [282]. It should be noted that the umbrella sampling process depends on modeling the interaction of the PT chain with a large number of lipid bilayer configurations. While the polymer is restrained near a particular Z coordinate (as explained below), it is free to diffuse in X and Y, and the individual phospholipids are similarly free. As a result, over the course of nanoseconds of simulated time, the energy of thousands of possible interaction configurations will be measured. This is essential, because the free energy of PT insertion into a bilayer is ultimately a statistical ensemble property that represents the average over a range of trajectories. As long as analysis snapshots are taken with sufficient spacing, the “ergodic assumption” of MD can be employed. Under this assumption, the set of configurations visited during a long simulation constitutes a sufficient sample to estimate ensemble quantities. This assumption has long been believed true for biomolecular systems, and there is no reason to believe that it fails in this particular case. The reaction paths obtained from the forward SMD pulls of EHPT and HPT were used to seed a set of umbrella sampling simulations. A set of distances along the Z-axis between polymer center and bilayer center were selected, with these umbrella centers being spaced more closely within the bilayer and more sparsely in the aqueous phase (due to greater viscosity in the bilayer). Centers ranged from 1 A˚ below the bilayer center (a coordinate of -1 A)˚ to 45 A˚ above the center (+45 A).˚ The SMD trajectory frame closest to the specified Z-distance was chosen as the starting point for each window. These frames were re-equilibrated for 50 ps (with the umbrella potential in place) to relax distortion caused by the SMD pull, then run for 3 ns each to collect data. All umbrella runs used a Langevin thermostat for temperature control, with a target temperature of 310 K and a coupling constant of 0.5 ps−1. Additional windows were added as necessary to ensure sufficient overlap between all

102 windows. The EHPT and HPT systems each required 28 windows, for a total of 168 ns simulated time, taking approximately 22,000 processor-hours on the Cray XT3 and Terascale Computing System of the Pittsburgh Supercomputing Center. For EHPT, 21 windows used a force constant of 0.6162 kcal/mol/A˚ (1 kBT at 310 K) and 7 used a force constant of 3 kcal/mol/A.˚ For HPT, 9 windows used a force constant of 0.6162 kcal/mol/A˚ and 19 used 3 kcal/mol/A.˚ After simulation, the processed trajectories and energies were combined using the WHAM method of [258], modiﬁed to properly handle using umbrella potentials using the direct-sum-over-data method of [285]. Un- certainties of the PMF were estimated as per [258], with all statistical ineﬃciency terms g−1 set to 1.

9.3 Results

9.3.1 Steered Molecular Dynamics: Reaction Path and Initial Energy Estimates

The forward and reverse SMD pulls were successful at revealing a likely reaction path for insertion of EHPT or HPT into a DMPC bilayer. RMS errors between commanded and actual polymer-to- bilayer distance (as measured using centers of mass) are 0.33 A˚ and 0.42 A˚ for the EHPT and HPT forward pulls, respectively. Both forward pulls showed the polymer rotating and reorienting itself as it inserted into the bilayer, suggesting that the pull rate of 0.01 A/ps˚ is slow enough to relax onto the true reaction path. The results for the reverse pull are less clear. The RMS command-to-actual errors for the reverse pulls are 0.81 A˚ and 0.71 A˚ for EHPT and HPT. These errors are twice those on the forward pulls, and are in a range that suggests the pull may have been too fast (or the spring constant too weak) to fully capture the physical trajectory for the removal reaction. Before turning to the details of that path, it is useful to consider the differences in work done along the EHPT and HPT trajectories. The work done during forward and reverse pulls for both polymers is shown in Figures 9.1 and 9.2. By averaging the forward and reverse curves at each point along the trajectory and appropriately translating the resulting sum so that the origin (at 52.5 A)˚ corresponds to zero energy, irreversible frictional work can be discounted and the true PMF of insertion revealed [284]; this average is also shown in Figures 9.1 and 9.2. As expected, both EHPT and HPT show a positive energy (work) of insertion, and the effect of frictional work is clearly visible in the continual upward slope of the curve. There is an inflection point in both forward curves just after 30 A,˚ as the polymers begin to push against and enter the bilayer. The post-entry work for HPT is much greater than that of EHPT. EHPT shows a brief period of decreasing work just before 20 A,˚ indicating that the polymer begins to “slide down the potential” on entering. The HPT forward work curve of Figure 9.2 does not show the same decrease. Although neither forward work curve can be considered an accurate estimate of the free energy of insertion, especially since these are single samples from two distributions, it is noteworthy that the total work of inserting HPT (137 kcal/mol) is nearly three times that of inserting EHPT (48 kcal/mol). The work curves for the reverse pull show less of a difference between EHPT and HPT and more concordance in shape between the two polymer pulls. The work of EHPT removal from the bilayer is higher than that of HPT removal (327 kcal/mol vs. 291 kcal/mol), but this cannot conclusively be said to be due to differences between the two polymers; these values are close enough that the difference might reverse sign if both pulls were repeated. It should be noted that the work of removal was over 100 kcal/mol higher for both polymers than the work of insertion. This leads to the reverse pull dominating the averaged curves seen in Figures 9.1 and 9.2. These averaged curves imply that the work of insertion for both polymers should be negative, with EHPT insertion being substantially more favorable than HPT insertion (145 kcal/mol vs. 82 kcal/mol). However, as will be illustrated further in the following paragraphs, the reverse SMD pull appears to sample a non-physical trajectory. This implies that the SMD work curves may indicate the correct trend (since the forward pulls appear to be accurate within the limits of frictional error), but cannot be considered a reliable quantitative estimate of the energy of insertion for EHPT or HPT.

103 Work of EHPT Insertion/Removal from Lipid Bilayer 150

100

-50

-100

Energy (kcal/mol) -150

-200

-250

EHPT Insertion -300 EHPT Removal EHPT Combined

-350 55 50 45 40 35 30 25 20 15 10 5 0 Polymer Distance from Bilayer Center (Angstroms)

Figure 9.1: Work curves for steered MD pulls of EHPT

Work of HPT Insertion/Removal from Lipid Bilayer 150

100

-50

-100

Energy (kcal/mol) -150

-200

-250

HPT Insertion -300 HPT Removal HPT Combined

-350 55 50 45 40 35 30 25 20 15 10 5 0 Polymer Distance from Bilayer Center (Angstroms)

Figure 9.2: Work curves for steered MD pulls of HPT

Forward, reverse, and summed work curves for the steered molecular dynamics pulling of EHPT (Figure 9.1) and HPT (Figure 9.2) into and out of the lipid bilayer. HPT requires substantially more work than EHPT for insertion (137 kcal/mol vs. 48 kcal/mol). HPT also requires less work for removal, although the diﬀerence is not as large as for insertion (291 kcal/mol vs. 327 kcal/mol). Solid conclusions cannot be drawn from single pulls, but these results point towards EHPT insertion being more energetically favorable than HPT insertion.

104 The force curves for the forward and reverse EHPT and HPT pulls, as smoothed by a Gaussian window of 100 ps width, are shown in Figure 9.3 and 9.4. These curves corroborate the trend seen in the SMD work results. HPT insertion forces are roughly equal to EHPT insertion forces until the polymers come in contact with the bilayer, at which point HPT forces rise and remain higher throughout most of the run. The HPT peak insertion force of 8 kcal/mol/A˚ is double the EHPT peak insertion force of 4 kcal/mol/A.˚ As with the work results, the reverse pull (removal) forces follow almost identical trajectories if overlaid, with the difference being that the EHPT force peaks slightly later (around 16 A˚ instead of 13 A)˚ and higher (18 kcal/mol/A˚ vs. 15 kcal/mol/A)˚ than the HPT force. The insertion trajectories of EHPT and HPT (selected frames of which are shown in Figure 9.5) reveal the reaction path and provide an explanation for the observed difference in the work curves. Both polymers begin similarly, by “docking” with the membrane in a horizontal configuration that maximizes the contact between the hydrophobic side chains and the bilayer surface. From this configuration, the polymer breaches a hole in the polar head group layer, then inserts through this hole into the bilayer interior. During the course of these runs, the polymers adopted a relatively vertical configuration within the lipid phase, but this may be due to insufficient relaxation time; we saw different configurations in the umbrella sampling runs reported in Subsection 9.3.2. The “sideways docking, interface destabilization, and bilayer insertion” path followed by these polymers is similar to the mechanism believed to be used by membrane-inserting peptides [286]. This path has also been observed in other simulations of transmembrane peptides [213, 214, 216] and in a coarse-grained MD study of channel protein insertion [215]. The difference in the insertion trajectories of EHPT and HPT lies in the way they (hypothetically) destabilize the head groups and insert. EHPT creates a hole that admits a single ring, then threads itself through in a ring-by-ring fashion, bending in the middle as it does so. This process is most clearly illustrated in the 3,659 ps panel of the EHPT series in Figure 9.5. By contrast, HPT destabilizes the head group interactions at multiple points simultaneously, then inserts in a largely horizontal configuration, with multiple rings passing through the polar region at once. That process is visible from 3,509 to 3,970 ps in the HPT series of Figure 9.5. The difference in the configuration of EHPT and HPT during insertion is further illustrated in Figure 9.6, which shows the polymers’ progress through a reduced configuration space of end-to-end distance and tilt. These trajectories can explain the previously-presented work and force curves. The EHPT work curve in Figure 9.1 remains relatively flat during the docking phase (roughly 25-30 A˚ distance), and does not change slope until nearly 20 A˚ distance, when EHPT is breaching the head group layer. The HPT work curve of Figure 9.2, on the other hand, changes slope and begins to climb between 25 and 30 A˚ distance, demonstrating that HPT is already displacing bilayer molecules as it docks. The more disruptive nature of HPT’s en bloc insertion also explains the continually high insertion forces seen in Figure 9.4. Samples from the reverse pull trajectories of EHPT and HPT are shown in Figure 9.7. As would be expected from the similarity of their work and force curves, both EHPT and HPT showed similar trajectories when pulled from the center of the bilayer to the outside. Both polymers encounter the polar region (starting at the carbonyl oxygens) as a barrier, as evidenced by the upturn of the work curves around 5-10 A.˚ Both EHPT and HPT then make a hole and begin to exit in a stretched vertical configuration. Unlike the forward SMD pull, no peak is seen on either work curve corresponding to this hole formation. It may be that this peak exists but is overwhelmed by the work required to continue dragging the polymer out of the membrane. As can be seen in the later panels of Figure 9.7, the polymers experience large amounts of force resisting their exit. They are continually in an extended conformation, implying that the entire polymer chain is experiencing tension and stretching. The reason for this high tension is illustrated in the fourth panel of each reverse trajectory. The DMPC tails bind strongly (presumably through VDW attractions) to the side chains of EHPT and HPT, with the attraction being so strong that one tail is “dragged” into the aqueous phase once the polymer has exited. The explanation for the high forces and resulting high work on both removal pulls is thus revealed: a continual forming and breaking of nonbonded attractions as the polymer moves out of the lipid phase.

105 Force of EHPT Insertion/Removal from Lipid Bilayer, 100 ps Window 10

EHPT Forward EHPT Reverse

-5 Force (kcal/mol/A)Force

-10

-15

-20 55 50 45 40 35 30 25 20 15 10 5 0 Polymer Distance from Bilayer Center (Angstroms)

Figure 9.3: Force curves for steered MD pull of EHPT

Force of HPT Insertion/Removal from Lipid Bilayer, 100 ps Window 10 HPT Forward HPT Reverse

-5 Force (kcal/mol/A)Force

-10

-15

-20 55 50 45 40 35 30 25 20 15 10 5 0 Polymer Distance from Bilayer Center (Angstroms)

Figure 9.4: Force curves for steered MD pull of HPT

Force curves for steered molecular dynamics pulling of EHPT (Figure 9.3) and HPT (Figure 9.4) into and out of the lipid bilayer, smoothed with a 100 ps Gaussian window. Forces are equal in the water phase, but HPT requires higher forces to achieve initial bilayer entry (8 kcal/mol/A˚ peak vs. 4 kcal/mol/A˚ peak for EHPT). The HPT force curve remains above the EHPT force curve for most of the forward pull through the lipid phase. Forces on the reverse pull are better-matched, but EHPT has a slightly higher peak force as it begins to leave the bilayer (18 kcal/mol/A˚ vs. 15 kcal/mol/A˚ for HPT). The removal forces for both polymers are substantially higher than the insertion forces. 106 EHPT HPT

967 ps, 42.9 A:˚ Free diﬀusion. 952 ps, 42.96 A:˚ Free diﬀusion.

2766 ps, 24.7 A:˚ Docking at bilayer. 2662 ps, 26.14 A:˚ Docking at bilayer.

3320 ps, 19.7 A:˚ Breaching headgroups. 3509 ps, 17.87 A:˚ Multi-point breach.

3659 ps, 15.6 A:˚ Vertical insertion. 3970 ps, 13.24 A:˚ Horizontal insertion.

5149 ps, 1.37 A:˚ Fully inserted. 5100 ps, 1.89 A:˚ Fully inserted. Figure 9.5: SMD insertion trajectories for EHPT (left) and HPT (right). Blue, tan, and red spheres represent cholines, phosphates, and oxygen. See text for analysis. 107 55

15 PT to Bilayer to Distance (A) PT 10

0 40

10 80 90 60 70 PT End-to-End Distance (A) 40 50 0 20 30 0 10 Angle vs Bilayer Normal (deg)

EHPT HPT

PT to Bilayer to Distance (A) PT 15

5 0 0 10 80 60 20 40 30 Angle vs Bilayer Normal (deg) 20 0 40 PT End-to-End Distance (A)

Figure 9.6: Three-dimensional plot of the insertion pathways for EHPT and HPT during SMD pulls, viewed from two different angles. This plot tracks polymer progress through a configuration space of end-to-end distance and tilt relative to the bilayer normal. Both polymers travel roughly the same path in configuration space until they reach the membrane, but after that, their paths diverge. EHPT adopts a smaller angle relative to the bilayer normal (more vertical), and shows a slightly smaller end-to-end distance (more flexed configuration).

108 Dragging of lipids alongside a hydrophobic molecule has also been observed in SMD studies of removal of a monotopic protein from a bilayer [279], but was not seen in a study of lipids being directly pulled out from a DPPC bilayer [287]. Dragging therefore cannot be taken to be an obligate feature of SMD pulls in a bilayer environment. More importantly, the lipid dragging combined with other features of the reverse SMD pulls suggest that these runs sampled a non-physical trajectory. As noted above, both reverse pulls showed substantial deviation between commanded and actual center-to-center distance, calling into question the applicability of the stiff-spring assumption that underlies most SMD analysis. The replica exchange runs of Chapter 8 predict that both EHPT and HPT are likely to prefer a slightly twisted/coiled conformation in aqueous solution, but the polymers in these reverse SMD runs show an almost-fully extended conformation until they break free of their attached lipid tails. In short, it appears that the reverse SMD is overly dominated by frictional forces, and that multiple slower pulls in both directions would be necessary to effectively recover the PMF for polymer insertion. However, the forward runs still appear reliable, and these can be used to inform umbrella sampling and obtain a PMF with less computational effort.

9.3.2 Umbrella Sampling: Accurate Potential of Mean Force

The PMFs for EHPT and HPT insertion into the DMPC bilayer, as computed from umbrella sampling, are shown in Figure 9.8. Comparing these to the SMD results in Figures 9.1 and 9.2, one sees that the SMD curves predict the correct trend, but miss the subtleties of the potential. Both polymers show relatively flat potentials in the water region, which extends to approximately 30 A˚ center-to-center distance. At that point, both polymers show an energy barrier, with HPT facing a taller (1.53 kcal/mol higher than EHPT) but narrower barrier. Past the barrier, the potentials become quite different. EHPT descends a slope that ends at 9-10 A,˚ leaving a flat potential within the lipid portion of the bilayer. HPT, conversely, shows a deeper well at 12 A.˚ If this point is considered to be insertion into the membrane, HPT insertion is actually more favorable than EHPT insertion, at −16 kcal/mol compared to EHPT’s minimum of −10 kcal/mol. However, it is not a true insertion, since neither polymer fully spans the bilayer at a center-to-center depth of 12 A.˚ When HPT is placed further into the membrane (i.e., into the actual lipid phase), the potential once again rises, reaching −6 kcal/mol at 0 A˚ compared to EHPT’s −10 kcal/mol. To further illustrate the configurations corresponding to these energy minima, representative frames at selected distances are shown in Figure 9.9. At 30 A,˚ as both polymers are docking with the membrane, a difference can already be seen: HPT is beginning to insert its head into the polar layers of the membrane, whereas EHPT is seeking to maximize its area of contact with the membrane. At their respective PMF maxima, both polymers are seen to be starting insertion, following the different modalities seen in the forward SMD trajectories. The reason for these two different insertion techniques becomes more clear when the conformations at 12 A˚ are considered. EHPT has created a hole and is threading itself through. Meanwhile, HPT has already sunk to its optimal resting place, at approximately the level of the carbonyl oxygens. This is also roughly the level to which water penetrates, so that HPT lies at the boundary between polar and nonpolar solvent phases [288]. The en bloc horizontal insertion observed in the forward SMD pull makes sense in this context – if HPT’s optimal state is one of admixture with the head groups and glycerol backbones of DMPC, then during the SMD run, it will pass through that configuration when it is commanded to the proper distance. The applied SMD force will cause HPT to sink further past the optimal point at 12 A,˚ but it will prefer to retain its horizontal configuration rather than re-orienting to a more vertical, EHPT-like configuration. It is equally informative to consider the configurations of both polymers at 6 A˚ (sinking through the lipid phase) and 0 A˚ (the bilayer center). At 6 A,˚ EHPT is effectively completely inserted, and can be seen “reaching out to the next leaflet”. HPT, by contrast, is in a twisted conformation that appears to attempt to minimize its contact with the lipid tails. At 0 A,˚ EHPT forms a bridge between the two membrane leaflets. HPT at 0 A˚ is in a planar and extended conformation in the low-density space between leaflets, and is not close enough to either DMPC/water interface to form an effective electrical bridge. All of these behaviors can also be seen in the three-dimensional

109 EHPT HPT

952 ps, 9.85 A:˚ Entering polar region. 907 ps, 7.97 A:˚ Entering polar region.

1784 ps, 18.86 A:˚ Vertical breach and exit. 1844 ps, 17.38 A:˚ Vertical breach and exit.

2955 ps, 32.4 A:˚ Exit with lipid dragging. 3123 ps, 31.14 A:˚ Exit with lipid dragging.

4818 ps, 50.49 A:˚ Continued lipid dragging. 4639 ps, 46.47 A:˚ Continued lipid dragging.

5175 ps, 54.67 A:˚ Breaking free from lipid. 5041 ps, 50.69 A:˚ Breaking free from lipid. Figure 9.7: SMD removal trajectories for EHPT (left) and HPT (right). Blue, tan, and red spheres represent cholines, phosphates, and oxygen. One lipid colored orange to highlight. See text for analysis. 110 PMF of EHPT and HPT Bilayer Insertion, Umbrella Sampling 6

4 EHPT HPT

-2

-4

-6

Energy (kcal/mol -8

-10

-12

-14

-16

45 40 35 30 25 20 15 10 5 0 Polymer Distance from Bilayer Center (Angstroms)

Figure 9.8: Potential of mean force for EHPT and HPT insertion into the lipid bilayer, as calculated from 28-window umbrella sampling. The potentials remain somewhat rugged due to the limits of the sampling, and should not be considered to be fully and quantitatively accurate. However, as can be seen from the small size of the error bars, they are close to correct. EHPT shows a steep descent to a broad, ﬂat-bottomed well within the bilayer, suggesting that it should be able to move freely within the lipid phase. HPT has an even steeper descent, but is not free to diﬀuse within the lipid tails. The HPT PMF shows a deep well around 12 A,˚ where HPT is likely to become trapped without fully inserting.

111 “configuration space” plot of Figure 9.10, where EHPT and HPT are seen to segregate into different configurations just as they did during SMD. Taken together, these features of the PMF and the associated polymer configurations highlight two key differences between EHPT and HPT. First, it is not energetically favorable for HPT to fully insert into the membrane. The minimum point of the HPT insertion PMF is at 12 A,˚ and the polymer is likely to become “stuck” at this point. This prevents HPT from entering the hydrocarbon layer, whereas the free energy landscape for EHPT permits free movement within all regions of the bilayer. Second, if HPT does escape the well at 12 Aand˚ move to the bilayer center, it will be forced into the horizontal orientation seen in the 0 A˚ panel of Figure 9.9. This configuration does not span the membrane, and thus seems unlikely to effectively conduct current across it. EHPT, by contrast, is seen to adopt a bilayer-spanning conformation. The replica exchange simulations of Chapter 8 made predictions regarding EHPT and HPT conformations in lipid, and Figures 9.11 and 9.12 compare the backbone angle distributions predicted by implicit lipid replica exchange to those actually observed as the polymers pass into and through the tails of the DMPC bilayer. The overall agreement is much better for HPT than for EHPT, suggesting that there is some property of EHPT (likely related to the branched side chain) that cannot be adequately captured by the electrostatic-force-focused generalized Born method. Where the replica exchange simulations predict that EHPT will adopt primarily anti angles, EHPT with DMPC shows primarily syn angles both at the entry to the lipid tails (12 A)˚ and in the low-density zone at the center of the bilayer (0 A).˚ Only when held in the high-density alkyl zone at 6 A˚ are anti-like angles favored, and then only weakly. Broadening of the anti peak (and of the distribution in general) is also seen at all three explicit lipid sample points, suggesting that EHPT is exploring more conformations in explicit lipid than was predicted in implicit solvent. The sharp syn peak at 12 A˚ is not entirely unexpected, since EHPT is still inserting into the bilayer and is likely to be forced away from an equilibrium configuration. It is less clear why the same peak should reappear at 0 A˚ where the polymer should have an opportunity to fully relax. HPT shows generally better implicit/explicit solvent agreement than does EHPT. Looking at the PMF, the distribution at 12 A˚ represents the closest thing to a “good solvent” that can be observed in these data, whereas 6 A˚ and 0 A˚ should be considered as marginal to poor solvents. In this light, it is interesting to see that the fraction of anti-like angles increases as HPT moves deeper into the lipid tails; the replica exchange results would point towards an expectation of fewer anti angles as HPT tries to form a helical aggregate. The observations here likely reflect once again the fact that generalized Born models cannot capture the effects of solvent van der Waals forces and viscosity on the solute. This explanation is supported by the fact that the best match between implicit and explicit solvent is at the center of the bilayer (0 A˚ curve in Figure 9.12), where the density and viscosity are lowest and HPT is most free to rotate. The values of the dihedral angle distribution peaks, the distributions of end-to-end distance, and the corresponding persistence lengths are all given in Table 9.1. EHPT is significantly more compressed than was predicted by the implicit solvent simulations; none of the explicit lipid end- to-end distance distributions include the replica exchange PMF minimum within their confidence interval. This is also reflected in the persistence length, which is much lower than that predicted from the 300 K implicit solvent replica, and which also shows a lower variance (likely due to the greater viscosity of explicit solvent systems). Most of the difference can be explained by the membrane- bridging conformation that EHPT is seen to adopt in the top panels of Figure 9.9. It was noted in Subsection 9.2.1 that the hydrocarbon thickness of this DMPC bilayer is approximately 12.58 A.˚ EHPT is trying to compress itself into that space, which forces a coiled conformation with lower persistence length. Since HPT’s dihedral angle distributions generally agree between implicit and explicit lipid, it is no surprise to find that the end-to-end distance and persistence length data also show qualitative agreement, with the values at 0 A˚ again being closest to the implicit solvent results. The principal finding is that at all points, HPT is in a more extended conformation with a higher persistence length than EHPT. This is consistent with the known solvatochromic properties of HPT; in a poor

112 EHPT HPT

30 A:˚ PMF slope increase, docking. 30 A:˚ PMF slope increase, head approach.

22.5 A:˚ PMF maximum, beginning insertion. 26 A:˚ PMF maximum, beginning insertion.

12 A:˚ PMF decreasing, partly inserted. 12 A:˚ PMF minimum, ﬂat in polar layer.

6 A:˚ PMF well, membrane spanning. 6 A:˚ PMF upslope, partially inserted.

0 A:˚ Membrane spanning. 0 A:˚ Flattened in membrane center. Figure 9.9: Umbrella sampling conﬁgurations for EHPT (left) and HPT (right). Blue, tan, and red spheres represent cholines, phosphates, and oxygens. See text for analysis. 113 55

PT to Bilayer to Distance (A) PT 15

0 40 30 80 20 60 10 40 PT End-to-End Distance (A) 20 0 0 Angle vs Bilayer Normal (deg)

EHPT HPT

PT to Bilayer to Distance (A) PT 15

0 0 80 10 60 20 40 30 20 PT End-to-End Distance (A) Angle vs Bilayer Normal (deg) 0 40

Figure 9.10: Three-dimensional plot of the distribution of EHPT and HPT during umbrella sampling, using the same configuration space as Figure 9.6. As in that figure, EHPT and HPT are shown to overlap in configuration space while outside the bilayer, but to occupy different regions once inside the bilayer. EHPT can be clearly seen to prefer more vertical positions (more aligned with the bilayer normal) compared to HPT. EHPT’s end-to-end distance distribution is seen to explore shorter distances, while only HPT reaches the most extended conformations. The difference is most obvious towards the bottom of the plot (near the center of the bilayer), consistent with the configurations shown in Figure 9.9. 114 Distribution of EHPT Backbone Dihedral Angles, Implicit vs. Explicit Solvent 0.045

EHPT Implicit Lipid EHPT Explicit Lipid, 0A 0.04 EHPT Explicit Lipid, 6A EHPT Explicit Lipid, 12A

0.035

0.03

0.025

Probability 0.02

0.015

0.01

0.005

0 0 20 40 60 80 100 120 140 160 180 Angle (degrees)

Figure 9.11: Backbone dihedral angles for EHPT in implicit and explicit lipid.

Distribution of HPT Backbone Dihedral Angles, Implicit vs. Explicit Solvent 0.045

HPT Implicit Lipid 0.04 HPT Explicit Lipid, 0A HPT Explicit Lipid, 6A HPT Explicit Lipid, 12A 0.035

0.03

0.025

Probability 0.02

0.015

0.01

0.005

0 0 20 40 60 80 100 120 140 160 180 Angle (degrees)

Figure 9.12: Backbone dihedral angles for HPT in implicit and explicit lipid.

Distribution of dihedral angles between the individual rings of the EHPT and HPT backbones, comparing the distribution in implicit lipid to the results at points within the explicit lipid bilayer during the umbrella sampling runs.

115 EHPT Implicit EHPT 0 A˚ EHPT 6 A˚ EHPT 12 A˚ Primary Peak (◦) 115-125 53-60 105-117 46-54 Secondary Peak (◦) 57-65 119-127 55-64 131-142 End-to-End (A)˚ 35.0-36.5 28.42±2.59 30.08±3.15 24.02±3.28 Persistence Length 3.36±1.59 2.66±0.48 2.98±0.62 1.91±0.5

HPT Implicit HPT 0 A˚ HPT 6 A˚ HPT 12 A˚ Primary Peak (◦) 56-61* 51-54* 45-55 52-60 Secondary Peak (◦) 117-124* 116-123* 118-132 112-128 End-to-End (A)˚ 35.0-36.0 36.37±2.39 31.55±4.92 30.20±4.56 Persistence Length 3.68±1.30 4.35±0.56 3.29±0.99 3.01±0.90

Table 9.1: Dihedral angle distribution peaks, end-to-end distances, and persistence lengths of EHPT and HPT in implicit lipid vs. explicit lipid. (*) in the dihedral distribution rows indicates a distribution where the primary and secondary peaks have equal magnitude. The persistence length is reported in repeat units. Persistence lengths and end-to-end distances in explicit solvent are given with a range of two standard deviations. solvent (such as the center of the bilayer), HPT should adopt more rigid conformations with greater persistence lengths, corresponding to the shift from red to purple colors. That trend is seen in the final row of Table 9.1, where HPT becomes more extended and reaches greater persistence lengths as it is pushed deeper into the lipid phase. This is accompanied by a shift of the syn-like peaks in the dihedral angle distribution. Those peaks are closer to planarity than was suggested by the implicit solvent results, likely reflecting the effect of VDW interactions with the DMPC tails.

9.4 Discussion

The results of the work described in this chapter illuminate the mechanisms and thermodynamics of PT insertion into a model lipid bilayer, and thus explain the BLM observations of Chapter 7. The SMD pulls show that the insertion of EHPT and HPT follows two different paths, with the path for HPT being less energetically favorable. Although SMD was not able to provide a good estimate of the actual energy of insertion due to the reverse pulls being dominated by irreversible work, the potential of mean force was still obtainable through umbrella sampling. The first question posed in the Introduction to this chapter, the first (the energy profile of insertion) is well-answered by the umbrella PMF of Subsection 9.3.2. The second (configuration of the polymer during and after insertion) is answered by the SMD trajectory and the configurations adopted at the end of umbrella sampling. Taken together, these data present two possible explanations for the observation that EHPT raises BLM electrical conductance and HPT does not:

1. Because HPT’s PMF minimum is far from the bilayer center and relatively close to the bilayer surface, HPT becomes “stuck” at the glycerol/carbonyl level with its long axis perpendicular to the bilayer normal. It never comes close to crossing the bilayer, and thus cannot electrically bridge it. 2. HPT does reach the bilayer center, and is highly conductive in this medium, but as seen in Figure 9.9, is again locked into a configuration with its long axis perpendicular to the membrane normal. In this configuration, it is too isolated from the potentials on either side of the bilayer to effectively conduct current.

Both explanations are probably true, and they most likely describe simultaneously-occurring

116 effects. However, the first explanation should be the dominant effect, if only because it corresponds to the lowest-energy configuration available for HPT. The deeper question, not yet answered by these simulations, is why EHPT is able to dissolve well in the lipid phase of the DMPC membrane and adopt the bridging conformations observed in these simulations. The answer almost certainly has to do with the branching of the 2-ethylhexyl side chain, since this is the only true difference between EHPT and HPT. However, there are many mechanisms through which the side chain branching could have an effect. One is through simple steric hindrance and the resultant disorder of the system; EHPT’s branched side chain may simply be more able to disrupt the packing of neighboring molecules, allowing the polymer to maneuver itself between them. It is known that polythiophenes whose side chains do not pack well are not able to form three-dimensionally ordered materials [252]. The branching also permits an interesting configuration in which the branches rotate so that the polymer backbone is entirely wrapped by a “solvation shell” of alkanes; this same conformation is not available in unbranched HPT. This mechanism may matter, because long-chain alkanes like the tails of a bilayer are believed to be good solvents for PT side chains but poor solvents for the PT backbone. If shielding of the backbone is the key, then side chains with a longer branch (e.g., 2-hexylhexyl) should be even more effective. Since branched-side-chain PTs (and PTs with bulky side chains in general) are hard to synthesize, testing of these hypotheses would likely start with further simulations. Insertion is important, but is only one part of the process. To create the proposed transmembrane electrode that lies at the heart of this thesis, the inserted polymer must also be conductive. The persistence length data shown above indicate that EHPT inserted into a lipid bilayer has a very short persistence length, and thus will have lower conductivity than most alkyl-PTs exhibit in their crystalline forms. At the same time, the BLM results indicate that some kind of electrical conduction is occurring across the bilayer, and there is every reason to believe that EHPT is mediating this conduction. The bilayer used here was composed of DMPC, in order to leverage that lipid’s greater fluidity. The BLM experiments used diphytanoylphosphatidylcholine, whose bilayers have a thicker hydrocarbon layer that may allow EHPT to adopt more extended conformations. Moreover, this chapter presents the study of PTs of a very specific length. The samples applied in the BLM experiments were certainly polydisperse enough to include shorter oligomers which would also be able to achieve a more extended conformation in the bilayer. If a large number of either coiled- conformation long chains or extended-conformation short chains inserted into the BLM, this could easily explain the tenfold increase in conductance that was observed in Chapter 7. At any rate, it must be kept in mind that even in a coiled and marginally-conductive configuration, EHPT will still be a better conductor than the bulk lipid it is displacing. Beyond providing a mechanism and further hypotheses for the results of Chapter 7, the simulations in this chapter also tested whether implicit solvent models can accurately describe the conformation of PTs in various media. While there is qualitative agreement between implicit and explicit solvent results for HPT, there is no such agreement for EHPT. On one hand, some of this disagreement can be explained by effects such as the “squeezing” of EHPT into the DMPC tail domain, which forces it to adopt conformations with shorter persistence lengths and more syn-like angles. On the other hand, the explicit solvent results are able to demonstrate evidence of HPT solvatochromism, which is not achievable with implicit solvent simulations. The most likely conclu- sion, however unfortunate, is that the simplified generalized Born model is not able to adequately capture all the interactions that govern PT conformation. It would be desirable to have a method for assessing membrane compatibility of polythiophenes (or other polymers) without having to un- dertake the computational effort of a full umbrella sampling run, but this method will have to await future developments in implicit solvent theories. In summary, the results of this chapter, combined with the preceding two, have demonstrated the feasibility of the self-assembled transmembrane polythiophene electrode concept. They have also demonstrated a set of in silico and in vitro models that can be used to validate candidate polymers for such electrodes before these candidates are used to record or stimulate actual cells. It is certainly true that even the well-tested phospholipid forcefield used in this chapter has limitations

117 on its accuracy [289], but the agreement between theoretical predictions and experimental data suggests that these limitations do not materially affect the usefulness of the simulation methods. Even with this successful proof of concept, substantial work remains to further understand the properties that will differentiate a useful from non-useful polymer, and to further validate the concept with in vitro and in vivo electrophysiological studies. The further details of these future studies and their possible findings will be laid out in the following, and final, chapter.

118 Chapter 10

Conclusions and Future Work

10.1 Summary of Results

The results of the preceding chapters, taken together, form a proof of concept for the new polymer- based neuro-robotic interface described in Chapter 3. This new electrode, with its core concept of lipophilic polythiophene self-assembled monolayers, has the following properties that combine to make it competitive with other known conductive polymer methods:

1. Biocompatibility. As shown in Chapter 4, the mixed SAMs formed from EHPT and 16-MHA can be functionalized to contain a large amount of coupled protein. Chapter 5 showed that these functionalized SAMs are biocompatible with primary mouse cortical neurons, a stringent in vitro test that is expected to predict in vivo biocompatibility. Even greater biocompatibility is likely possible by coupling protein blends or patterning the coupled protein in a manner favored by neurons. 2. Low impedance. Chapter 4 also demonstrated a roughening of metal surfaces by application of an EHPT SAM, and Chapter 6 verified that this increased surface area lowers the impedance of a stimulating/recording electrode. More importantly, this impedance lowering can be achieved even when the SAM contains sufficient MHA and protein to produce good biocompatibility. The impedance drop is not as large as that produced by thicker electrodeposited films, but it is significant. Furthermore, the PT SAMs’ other advantages, listed below, combine to make them a competitive technology. 3. Robust covalent binding. Chapter 2 laid out the difficulties that can arise from coatings that are not firmly attached to the electrode. PT SAMs are, by definition, bound more firmly and should not show as much delamination or degradation as electrodeposited films. As noted in Chapter 3, the binding between sulfur and gold is stronger than the binding between individual gold atoms. Some evidence for that robustness is presented in Chapter 4, and other groups have reported strong adhesion of silane-based PT SAMs [182]. It is also seen in Chapter 6 that these SAMs can be removed and reapplied, which should allow re-use of coated electrodes. 4. Patternable. Although the SAMs in this work were not subjected to patterning, there is a wide body of literature (referenced in Chapter 2) regarding the patterning of thiol- based SAMs. At least one of these techniques, microcontact printing, has been verified to work with polythiophenes [290]. This patternability enables the construction of neural electrodes presenting multiple proteins or polymers in a defined spatial geometry. While that is principally useful for in vitro network experiments, it may also be valuable for in vivo work if such patterns are shown to improve neural adhesion to the probe or otherwise enhance recording.

119 5. Controllable composition. Although the experiments of Chapter 5 were not able to detect differences in neurite outgrowth between different mixed SAMs, this appears to largely be a problem of the study’s power being limited by the amount of EHPT available. Morphological differences are visible between the neurons on the three different mixed SAMs, indicating that the underlying films are not functionally identical. As the relationship between solution concentration and SAM concentration of components is further delineated, it will become possible to create PT-containing SAMs with any relative concentration of two or even more molecules, thus tailoring the tradeoffs between biocompatibility and other film functions. This is an advantage that is essentially unavailable with electrodeposited films. 6. Membrane-inserted and specific. While last in the list, this may be the most important feature. Chapters 7-9 show that EHPT is able to integrate into a lipid bilayer and conduct current across it. It is unlikely that this is a unique property of EHPT; there undoubtedly exist many polythiophenes that can perform the same feat. Once this proof of concept is confirmed with live cells, the intracellular access alone is sufficient to establish polythiophene SAMs as a viable and desirable neuro-robotic interface technology. Intracellular access will allow highly specific stimulating and recording, and will provide access to neural computations that are currently hidden from view.

In summary, the experiments described in this thesis have proven that polythiophene SAMs, and EHPT SAMs in particular, are a viable technology for neuro-robotic interfaces. Through the properties just listed, these conductive polymer coatings can improve the performance of existing electrodes and compete eﬀectively with the current “gold standard”, electrodeposited coatings. Much study is still needed to expand these ﬁndings to in vivo experimentation and/or clinical applications, but the work described here has laid the foundation.

10.2 Future Directions

This thesis has explored a new type of neuro-robotic interface electrode, and as a result, has opened up two new frontiers that are rich with possibilities: the use of conductive polymer SAMs for in vivo and in vitro electrodes, and the concept of polymer-based intracellular electrophysiology. The many future experiments that follow up on this thesis can be divided into two general categories, those that are on the “critical path” towards in vivo use and those that would further elucidate mechanisms or expand the reach of the technologies described herein. These two categories will be treated separately.

10.2.1 Critical Path for Technology Development

The PT SAM approach has been conceptually proven, but the pieces must still be assembled into a fully functional and demonstrated device. The next step would most likely be a full characterization of EHPT SAM electrodes for in vitro electrophysiology. A schematic of such an experiment is shown in Figure 10.1. The essential component is head-to-head electrophysiological comparison between neurons on EHPT and neurons on a non-EHPT biocompatible coating. (An MHA SAM would be the best choice for the latter, to minimize cell-electrode separation.) On recording, neurons on EHPT should show much larger action potential amplitudes, since the signal will not be diminished by passage through the extracellular space. It may also be possible to observe an inversion in the sign of the action potential trace; this would occur if the non-EHPT electrodes are so close to the cell that they are capacitively coupling to the membrane potential. The EHPT SAMs should also be tested in stimulation, to verify that they substantially lower the threshold current required to depolarize a neuron. Assessment of this stimulation threshold can probably be done by recording with metal electrodes, since a single stimulus should trigger a train of spikes (and spikes on multiple MEA electrodes, through synaptic connections of the original cell) that will not be masked by stimulus

120 Figure 10.1: Schematic of a possible experiment to follow up the findings of this thesis with in vitro electrophysiology. (A), neuron growing on a planar multielectrode array. This could be an MEA with a biocompatible SAM on it (such as the MHA/protein SAM of Chapter 5), or the “industry standard” adsorbed polylysine/laminin coating. (B), neuron growing on a similar MEA, but functionalized with a SAM of EHPT or other lipophilic PT. Based on the results in this thesis, the cells on EHPT should show lower thresholds for stimulation, higher-amplitude recordings, and inversion of the action potential waveform compared to the cells not on EHPT. artifact. However, if stimulus artifact is indeed a problem, recording (and threshold assessment) could also occur through loading of the neurons with voltage-sensitive fluorescent dyes. This is a non-trivial technique, but the tools to perform it are commercially available if needed. Performing the in vitro electrophysiology is not a simple matter. One major question is whether the existing EHPT wires will reach the neuron when tethered in a SAM. A molecular weight of 3,000 g/mol (as seen for the polymers in these studies) corresponds to a hexadecamer, roughly 6.35 nm long when fully extended. This is sufficient to bridge the cell membrane, but it may not be sufficient to reach the membrane through the glycocalyx. Furthermore, if the presence of adhesion molecules in the SAM is not sufficient to keep neurons from moving relative to the probe, longer PTs will be needed in order to remain in contact with the membrane during micromotion. Higher molecular weight polymers can certainly be prepared; PTs are known to be able to reach weights ten times higher than those used here [177]. However, this may require further advances in synthesis techniques, including a fusion of the SAM and electrodeposition methods (see below in Subsection 10.2.2). It will be difficult to predict in advance whether a given PT sample will be able to reach the membrane without testing it. If this problem does occur (i.e., the PT used is of relatively low MW and no effect is seen), it may be necessary to apply the tethered PT to the cell membrane through a method other than cell culture, such as moving an EHPT-coated metal probe into membrane contact with a micromanipulator or AFM cantilever. These numerous methodological issues and potential confounders highlight the wisdom of performing the initial proof-of-concept work in simplified model systems. Another question to consider is whether the array used will be able to effectively connect with neurons at all. The EcoMEAs used in Chapter 6 were chosen primarily because they are fully solvent-compatible. Since there are few to no alternative solvent-compatible MEAs currently on the market, the experiment of Figure 10.1 may well need to be conducted using EcoMEAs and their extremely large electrodes. It is difficult to record single-unit signals using large electrodes, because they tend to spatially integrate and attenuate the action potential. However, recent in vitro experiments using arrays with large platinum spikes in dissociated culture showed that these arrays can record single units, mainly because the neurons are able to achieve tight contact over the large surface area [291]. The same phenomenon may be true of EcoMEA electrodes, particularly if there is a highly biocompatible SAM inducing neurons to closely approach the metal. Spatial integration could be a particular concern for successful recording of intracellular potentials, because a large part of the conductive surface of a given electrode might not be in contact with the neuronal membrane. This extracellular portion of the SAM would be picking up extracellular potentials, which would be inverted compared to the desired intracellular signal and would thus reduce the recording magnitude. Furthermore, the intracellular potential could become “shorted” to the extracellular space through the metal electrode, thus abolishing the resting potential. Both of these problems (as well as the issue raised in the preceding paragraph) can be solved by careful

121 re-engineering of the electrode geometry if they do arise. The metal electrode sites themselves can be made smaller, or the electrode can be patterned to only have PT-containing SAM over a small portion of its surface area (with the rest being covered by an insulating MHA layer). Nevertheless, if such interventions are needed, converging on the optimal electrode structure will require several iterations of the design process. In a more positive vein, it should be noted that shorting of the membrane potential would presumably interfere with the normal electrical activity required for healthy development of embryonic/neonatal neurons. The neurons observed growing on mixed EHPT:MHA SAMs in Chapter 5 were mostly normal in morphology, suggesting that such shorting did not occur. All of the above concerns must be addressed during in vitro followup of this work, but that will not be the end. Whether in parallel with or subsequent to the in vitro electrophysiology, both the biocompatibility and the electrical properties of the EHPT (or other PT) SAMs will need to be re-proven in vivo. This will require further sophistication in experimental techniques. N-CAM13 works well as a cell culture compatibility protein (and is relatively easy to obtain in large quantities), but in vivo work will likely require actual brain proteins and peptides, which are less stable and harder to obtain. Multiple studies have suggested that the best results are likely to be obtained with a ﬁlm that presents a mixture of adhesion molecules or their key domains [292, 293], and the best mixture choice will need to be optimized in cell culture. Once those issues are solved, the existing SAM techniques will need to be adapted to work with implantable MEAs, which have highly varied materials of construction and are much more fragile than in vitro MEAs. None of these is an insurmountable obstacle, but all will require months to years of careful experimentation. Biocompatibility and electrical performance assessment in vivo will then each require a further one to two years. Only once the in vivo performance is proven will the PT SAM technology be ready for substantive adoption by other investigators and possible human trials.

10.2.2 Mechanisms, Tools, and Expanded Techniques

Further Deﬁnition of SAM Composition and Nanostructure

The results of Chapter 4 verify that SAMs can be formed, but there is much more that can and should be learned, especially if the SAMs are to be further optimized for the work described in Subsec- tion 10.2.1. Scanning tunneling microscopy to visualize individual molecules (using the appropriate equipment) is an obvious choice, as is testing of more SAM compositions. However, scanning probe methods can only reveal topography. In order to truly probe the fractional composition of the mixed SAMs, other methods such as x-ray photoelectron spectroscopy (XPS) or Fourier transform infrared spectroscopy (FTIR) will likely be needed. XPS in particular is very sensitive to the presence of extraneous contaminants. Making glassware and solvents free of such contaminants is simple; ensuring that the polymer sample is equally pure will be harder. PT samples often contain residual catalyst and other impurities, and thiolated PTs can contain elemental sulfur even after vigorous washing. This need for highly pure and clean samples has, to date, prevented XPS from being applied to the EHPT SAMs described in this work. Another area ripe for further followup is modeling of the impedance spectra observed in Chap- ter 6. Equivalent circuit modeling could confirm the proposed explanations and provide insight into how the SAMs can be manipulated to show even better impedance. This modeling may be feasible from the existing spectral data, but accurate models would more likely require re-collecting the impedance spectra with more points per decade and a wider frequency range. As noted in Chap- ter 6, starting points for such models would be existing proposals for models of polythiophene thick films [192, 193], alkanethiol SAMs [194], immobilized protein [191], and electrodeposited polymer films [111, 112]. It is not yet clear how best to combine the known models to find an equivalent circuit for PT SAMs. Further modeling and impedance studies (possibly in combination with further quantum calculations of membrane-inserted PT structures) could also shed more light on the mechanism by which EHPT conducts current across a membrane. Throughout this thesis, it has been implicitly assumed

122 that a Helmholtz double layer forms at the interface between the PT and the aqueous electrolyte, such that current is translated between the electronic and ionic regimes by a purely capacitive mechanism. However, nothing presented in this work conclusively rules out the presence of supplemental redox (Faradaic current) mechanisms taking place, particularly within a mixed SAM. Such reactions would likely not be deleterious (being analogous to the “pseudocapacitive” hydride-formation current of a platinum electrode), but detecting their presence through modeling and/or cyclic voltammetry would improve understanding of the behavior and applications of PT SAM electrodes.

Eﬀect of Coupled Biomolecules on PT Impedance

One particularly interesting hypothesis presented in Chapter 6 is that the presence of coupled protein and/or the charged end groups of the alkanethiol SAM component may be doping the EHPT component and increasing its conductivity. To follow up on this hypothesis, it would be interesting to test the impedance of a series of mixed SAMs that present different charges. A reasonable approach would be to test a PT in SAMs whose alkane component bore positively and negatively charged groups, e.g. amine and carboxylate (both of which can be used to couple proteins). From there, each SAM could be coupled with a positively charged, negatively charged, or neutral peptide (which should neutralize the charge on the alkanethiol end group). Since PT doping occurs when a PT sample contains a high anion concentration, the anionic films should have lower impedance. There might also be a particularly large effect in the case of coupling an anionic peptide to a cationic alkanethiol, since the opposite charges should increase the concentration of peptide near the SAM and thus improve the coupling efficiency. A related question is whether the alkanethiol component is necessary at all. It is known that carboxylate side chains can be placed directly on the PT itself [179]. If a small fraction of PT molecules in the SAM bear such side chains, or if each molecule bears only one or two (i.e., through a block copolymerization process), coupling could be achieved without introducing insulating molecules into the SAM. On the other hand, this method may not achieve high biocompatibility – the coupling reaction is probably not sterically favored if the PT chains are packed closely together and their side chains made less available to proteins in solution.

Further Biocompatibility Improvements

One method for improving biocompatibility was presented in Subsection 10.2.1: use a mixture of proteins, which has been suggested to produce a supralinear response from neurons [293,294]. However, the coupling process itself could stand to be improved. Results in the literature suggest that coupling an antibody (with a negative Fc tail) to a COOH-bearing SAM will tend to bind a large amount of the protein with its active FAb site facing the metal, not the cells [295]. This inefficient coupling will require more protein to achieve biocompatibility. By altering the charge on the alkanethiol SAM component and/or changing the coupling method to one that favors a better orientation, it may be possible to achieve more biocompatibility for the same EHPT:MHA ratio. Examples of appropriate techniques are given in [295] and [296]. The resulting improvement in biocompatibility could also be traded off for better impedance by lowering the MHA fraction. Such results are valuable even when using a true adhesion molecule instead of an antibody such as N-CAM13, because many neuronal adhesion molecules have substantial homology with the immunoglobulins. If further in vitro biocompatibility work is needed to test a new coupling method or a multi- protein blend, it may also be worth improving the biocompatibility regime. It has recently been suggested that biocompatibility for neuro-robotic interfaces be assessed not in standard neuron culture, but in specifically designed neuron-astrocyte cocultures [297]. The suggested paradigm in this model is to coat a metal wire with the SAM of interest, then simply drop that wire onto an established culture. In theory, this should more closely mimic the stab wound and micromotion effects of real implants, as well as modeling the effects of introducing a foreign body into an established system of neurons and glia. It would become harder to visualize the responses of individual cells,

123 particularly if the astrocytes or microglia become reactive, but this test would be an even better predictor of how the SAM would fare in vivo.

Electrodeposition Methods for Polythiophenes

As noted in Subsection 10.2.1, there is some concern that EHPT and related thiolated PTs cannot be chemically polymerized at molecular weights high enough to successfully reach the cell membrane. Moreover, the strong solvents needed to dissolve PTs for SAM formation are not compatible with a wide variety of materials (most notably silicones) used in in vitro and in vivo electrode systems. It would be desirable to have a method that used more gentle solvents while providing higher molecular weights, as electropolymerization does. Electropolymerization has not gained favor for most PTs because it is difficult to ensure regioregular head-to-tail coupling. PTs also show a strong tendency to “over-oxidize” into a non-conductive state at potentials only slightly greater than those needed to polymerize them. It has recently been demonstrated that, with careful tuning of the electrolyte, it is possible to electropolymerize a mostly defect-free PT without over-oxidation [168]. Other groups have demonstrated the formation of SAMs capped with thiophene monomers [169] and electropolymerization from those monomers [182]. Com- bining these two techniques would be highly nontrivial, but would allow PTs to match the molecular weight and impedance of current electrodeposited films, while still keeping the advantage of being a SAM. Patterning techniques could also be applied to separate out particular nanodomains where PT could be grown, so that there would be sufficient SAM area available for protein coupling [298]. The prospect of electrochemically grown PT is particularly exciting in light of recent reports (not yet published) that conductive polymers can be grown in the presence of living neurons without harming the cells. It may be possible to grow a “cage” of EHPT around a neuron, thus ensuring access to its membrane and the intracellular potential from a variety of points. Even if electrodeposition of covalently bound alkylated PTs proves to be difficult, electrical techniques can also help SAM formation and thus may help preserve delicate electrodes. Application of a potential to the metal surface has been shown to speed up SAM formation for both conductive polymers [299] and alkanethiols [300]. This could permit formation of good SAMs even from solvents that can dissolve only small amounts of PT, and would reduce the time that MEAs need to be exposed to harsh solvents. The same technique may also be used for cleaning SAMs off electrodes, altering film composition after a SAM has formed, and so on.

AFM Veriﬁcation of PT Insertion/Removal Energies

The SMD and umbrella sampling experiments of Chapter 9 provide an accurate qualitative picture of the energetics of EHPT and HPT insertion into a lipid bilayer. However, due to the approximations inherent in molecular dynamics techniques, the quantitative results of those simulations may not match reality. In order to verify and better calibrate the simulations, it would be valuable to empirically measure the energy of EHPT and HPT insertion into a lipid bilayer. One possible technique for such a measurement would be to bring a PT-coated AFM probe in proximity to a lipid bilayer. As the probe approaches the bilayer, the PT coating will be inserted into the lipid phase, and the force required to drive that insertion should change depending on the relative insertability of the specific PT. Integrating the applied force should give the relative free energy change. However, actually performing this experiment is highly non-trivial. It is not clear that the bilayer would remain intact in the presence of an AFM probe; if the probe applies too much force, the membrane will simply develop a large hole around the entire probe tip. Even if that does not occur, the relative difference in force between two PTs may be smaller than the noise level of the AFM scanner stack. In short, while the prospective data from this line of research would be interesting, significant investigation is necessary to determine whether it is feasible.

124 Further Molecular Simulations

The molecular dynamics work of Chapters 8 and 9 does not only represent validation of the hypothesized EHPT insertion mechanism. It also can be a powerful tool for seeking out better polymers and understanding the behavior of PTs in SAMs. It is well-known that regioregular PTs readily aggregate, even in dilute solutions [183,259]. The grains seen in EHPT SAMs in Chapter 4 are hypothesized to be aggregates, and the solutions used in Chapter 7 likely contained aggregates due to their high concentration. This suggests that what would be inserted into the membrane in a real device is not a single molecule of EHPT, but an aggregate containing two, three, or even several EHPT chains. The flexibility of such an aggregate could be quite different from that of the single molecule, and there is evidence that more flexible PTs will have more flexible aggregates [198]. Because EHPT is not solvatochromic (i.e., it retains its random-chain behavior even in poor solvent), its aggregates may remain relatively flexible and disordered and thus be able to insert. HPT aggregates may be stiff and rod-like and may not be able to use the mechanism seen for single molecules in Chapter 9. It would therefore be highly valuable to perform the MD simulations of Chapter 9 using multi-chain aggregates. Developing good structures for such aggregates would be tricky, and may require a return to the replica exchange methods of Chapter 8. A simpler, albeit less accurate, approach would be to continue simulating single decamer chains while decreasing the X and Y simulation box dimensions. This would cause the simulated PT to interact with its periodic replicas, which could shed light on the behavior of PT molecules in a tightly-packed SAM. Another desirable path of simulation is the testing of side chains beyond the 2-ethylhexyl of EHPT. PTs with side chains that are not straight alkanes are often difficult to synthesize, mainly due to the bulky side chains hindering the reaction between monomers. Before expending effort on syntheses, numerous different side chains could be tested in silico, again using methods similar to those seen in Chapter 9. Developing parameters for these different side chains would be fairly simple using the process of Chapter 8, taking only a few days of effort per new side chain. The process could most likely be automated, and would go faster if more computational resources were made available to the ab initio partial charge step. At the same time, because the branched side chain of EHPT is believed to be the mechanism behind its solubility and membrane insertion, it is particularly important to test a variety of branched side chains. Would it be better to have two short but equal branches, reducing the polymer’s overall size? Would two longer branches be better, so that they can mix more with the alkyl tails? Will PTs with longer branched side chains follow the pattern of longer straight side chains and be more likely to adopt extended conformations [301]? What would be the effect of polymers with side chains that branch at more than one point? All of these questions would be extremely difficult to answer by synthesizing the polymers of interest, whereas the only limitation to answering them in simulation is the available computer time. Given the limitations identified in Chapter 9, it is na¨ıve to think that the simulations can fully model reality. However, they can provide heuristics that will guide exploration to a small handful of candidate polymers that can then be tested in BLM and cell culture models. Beyond the specific optimization of the side chains, simulation using the existing PT parameters can answer some of the other questions raised above. PTs of various lengths could be simulated, with one end tethered in space, to determine how long the chain must be to reach and bridge the membrane. If quantum mechanics is used to treat all or part of the system, it should be possible to predict the conductivity of a polymer inserted into the membrane, improving on the limited precision of the persistence length calculations. These same techniques could also be a theoretical companion to the above-mentioned examination of how neighboring SAM molecules might dope a tethered PT. Finally, MD simulations could help predict the behavior of PTs with a combination of hydrophobic and hydrophilic side chains. Membrane channels have been shown in coarse-grained MD models to insert better if they have hydrophilic end caps [215]. The same may be true of block copolymerized amphiphilic PTs, whose synthesis has previously been demonstrated [178,207]. All of these simulations would once again be heuristics, and would need to be combined with experimental

125 data, but they would provide valuable guidance on where to look. It may be possible to improve the accuracy of PT behavior predictions by placing more effort into deriving PT simulation parameters. The inter-ring torsion potential used in this work is an accurate representation of the quantum rigid rotor potential, but no specific effort was made to optimize other parts of the system such as the VDW interaction between PT atoms and those of the lipid bilayer. Given that the parameters used here are from “generic” force fields that are meant to be widely applicable, it is not likely that small changes in any parameters would produce substantially different results. However, if further simulations are conducted and are found to make inaccurate predictions about the insertability of various PT formulations, parameter improvement would be the first logical place to start.

10.3 Final Thoughts

The data presented over the course of this thesis show that EHPT SAMs are a promising technology for neuro-robotic interfaces, but the preceding section makes it equally clear that there is a great deal still to be done before that promise becomes a reality. Moving lipophilic PT SAMs into production use will most likely require advances in PT synthesis technology, and will certainly involve years of exploration of the chemical space surrounding the speciﬁc SAMs tested in this thesis. Nevertheless, such exploration has not previously been undertaken because there was no clear motivation or application for the results. With the proof of concept presented here, chemists and bioengineers now have a reason to follow the lines of research presented above. It is my hope that these small ﬁrst steps will light the way for many fruitful investigations and valuable results from this new technology.

126 Glossary

ab initio calculation A quantum chemical computation in which the structure and internal forces of a molecule are calculated by optimizing the position of simulated electron orbitals, with the orbitals being represented by functions from a chosen basis set (q.v.). ab initio calculations tend to be highly accurate (due to the relatively small number of assumptions and approximations involved), but are also extremely computationally expensive. Examples of ab initio methods include Hartree-Fock (HF)/Self Consistent Field (used in this thesis) and M¨oeller- Plesset Perturbation (MPn) theories.

Basis set A set of functions, usually Gaussians, that are linearly combined to describe atomic orbitals during a quantum chemical calculation. Many varieties of basis set exist, diﬀering in which orbitals are represented by their functions, how many Gaussians are used per orbital, whether they include supplemental functions for eﬀects such as polarization, etc. The basis set most commonly referred to in this thesis is 6-31G*, which uses a 6-Gaussian function for each core orbital and two functions for each valence orbital (one with 3 Gaussians and one with 1 Gaussian). The “*” in 6-31G* denotes the presence of polarization functions.

BLM Black Lipid Membrane or Bilayer Lipid Membrane. An experimental technique involving the creation of an artiﬁcial unsupported phospholipid bilayer that separates two compartments of saline. BLMs are used as an in vitro model for cell membranes. Their principal limitation is their fragility, which is at least partially due to their single-lipid composition and lack of strengthening components such as cholesterol and protein.

DMPC DiMyristoylPhosphatidylCholine, a phospholipid with a choline (three methyl groups on a nitrogen) head group and two saturated 14-carbon tails. DMPC has a phase transition temperature of 296 K.

EHPT Poly(3-(2-ethylhexyl)-thiophene), a conductive polymer. EHPT has a branched side chain, which leads to greater disorder and less regular packing; solid-state EHPT is soft and waxy, compared to the hard crystalline powders obtained with most PTs. Presumably due to this branched side chain, EHPT is also soluble in solvents that are normally poor solvents for PTs, including hexane and other non-polar hydrocarbons. This suggests that EHPT should be able to enter the lipid phase of a cell membrane, and it is used in this work as a model lipophilic polymer.

Faradaic current Electrical current that ﬂows through an electrode by driving oxidation-reduction (redox) reactions at the electrode surface. These reactions can be reversible or irreversible. Irreversible reactions caused through Faradaic current can cause tissue damage by creating toxic chemical species in the vicinity of the electrode.

Fascicle A small bundle of axons, wrapped by its own connective tissue sheath, that is found within a peripheral nerve. Each fascicle contains axons that all innervate the same muscle or small region of a target organ; these axons are usually a mix of sensory and motor ﬁbers.

127 Forcefield In a molecular dynamics (q.v.) simulation, the forcefield is a set of constants that collectively specify all possible interactions between the simulated atoms. Most modern forcefields specify spring constants for bond stretching, angle bending, and bond torsion, along with charges for each atom and van der Waals parameters (radius and potential strength) based on element type. Commonly-used MD programs such as AMBER, CHARMM, and GROMACS include “generic” force fields that are meant to provide reasonable starting values for most parameters, but refinement of those parameters is often required to achieve accurate results for non-biological molecules.

Functional electrical stimulation (FES) The delivery of electrical stimuli to paralyzed muscles or to the nerves innervating those muscles, with the goal of bringing a limb or organ back under voluntary control. Control strategies can be as simple as a hand-held controller or can attempt to automatically detect and respond to the patient’s desires.

Generalized Born model (GB) An implicit solvent model (solvent model without explicit representation of individual solvent atoms/molecules) in which the solvent is considered as a uniform dielectric medium surrounding the simulated molecule (solute). Coulombic interactions between solute atoms are then scaled according to the strength of the dielectric, with further corrections introduced to compensate for effects such as atoms being “buried” in the molecule’s interior (and thus screened from interaction with the solvent and most other atoms). GB and other implicit solvent models are much more computationally efficient than explicit solvent, but they cannot capture effects such as direct Coulomb or VDW interactions between the solute and solvent atoms.

Hartree-Fock (HF) A type of ab initio (q.v.) quantum chemical calculation.

HPT Poly(3-hexylthiophene), a conductive polymer. HPT is a standard model for the regioregular, straight-side-chain PTs. Compared to EHPT (q.v.), HPT is much more ordered and crystalline in the solid state, and HPT is poorly soluble in hydrocarbons. HPT is therefore used in this work as a model non-lipophilic polymer.

M¨oeller-Plesset (MPn) A type of ab initio (q.v.) quantum chemical calculation, believed to have higher accuracy than Hartree-Fock (q.v.) methods on some systems. This accuracy is achieved by taking into account the repulsion between pairs of electrons that is neglected in HF calculations. The “2” of the commonly-seen “MP2” notation indicates that second-order repulsion terms are used; higher orders of MP theory are available, but are not referenced in this document.

MD Molecular dynamics (q.v.).

MHA 16-mercaptohexadecanoic acid, a long-chain hydrocarbon with a thiol (SH) group on one end and a carboxylic acid (COOH) group on the other. The thiol group allows MHA to form self-assembled monolayers on noble metal substrates, while the carboxylate group serves as a “hook” for protein coupling through reactive esteriﬁcation.

Molecular dynamics (MD) A chemical simulation technique in which molecules are treated as spherical atoms connected by spring-like bonds and interacting through repulsive and attractive forces. MD is computationally eﬃcient compared to, e.g., ab initio techniques, and can produce an accurate model of physical phenomena as long as the underlying parameters (bond spring constants and atomic charges) are well-selected.

MP2 Second order M¨oeller-Plesset (q.v.).

Neuro-robotic interface A technology that uses the electrical activity of neurons to control a device and/or stimulates neurons to alter the function of the nervous system.

128 PBS Phosphate Buffered Saline, a physiologic salt solution. PBS is used in place of water for most applications involving proteins or other delicate biomolecules; the use of a solution of near-physiologic osmolarity and pH prevents these molecules from denaturing as they would in pure water. PBS is also free of any primary amines (as opposed to the equally common TBS, Tris-buffered saline). This ensures that it will not interfere with amine-based protein coupling reactions. PEDOT Poly(3,4-ethylenedioxythiophene), a conductive polymer. Persistence length Half the Kuhn length of a polymer. If the polymer is considered as a set of rigid links connected by flexible joints, the Kuhn length is the average length of one rigid segment. In other words, the persistence length is half the length over which the polymer effectively does not bend. For conductive polymers such as polythiophene, bends in the chain interrupt electronic conjugation and decrease conductivity. Therefore, conductivity is proportional to the persistence length, and the persistence length provides a method for assessing the relative conductivity of two polymer conformations. Potential of Mean Force (PMF) A plot of the free energy of a system as a function of one or more selected coordinates. The system is expected to spend more time in lower energy coordinate bins, with the transition time between bins governed by the height of the barriers between those bins. PPY Polypyrrole, a conductive polymer. PT Polythiophene. In this work, “PT” refers to a poly(alkylthiophene), i.e., a polythiophene whose rings are decorated with alkyl side chains. Other works sometimes use “PT” for polythiophenes of all side chains, and reserve “PAT” to indicate poly(alkylthiophene)s.

Replica exchange A molecular dynamics (q.v.) simulation technique in which multiple replicas of the system of interest are simulated in parallel at diﬀerent temperatures. Regular attempts are then made to exchange the copies, with the probability of success determined by the potential energy and temperature gap between the two systems. The exchange process makes it easier for replicas to leave local energy minima and more thoroughly explore conformation space in a shorter amount of time. The principal drawback of replica exchange is that the number of replicas needed scales roughly as the square root of the number of atoms in the system. Thus, replica exchange can readily be applied to small implicit-solvent systems, but large explicit- solvent systems such as the membrane-polymer interactions of this thesis become intractable to this technique unless one has extraordinary computational resources.

SAM Self-assembled monolayer. Steered molecular dynamics (SMD) A molecular dynamics (q.v.) technique in which one part of the simulated system (usually a small molecule of interest) is “steered” by a gradually- shifting applied force into interacting with another part of the system. If the steering force changes suﬃciently slowly, SMD simulations can reveal the conﬁguration changes that take place during the reaction and may allow determination of the free energy of the interaction. However, most SMD simulations include some work done due to the “friction” of moving the molecule too quickly, and multiple iterations of the same pull are generally required to obtain an accurate free energy estimate.

Umbrella sampling A molecular dynamics (q.v.) technique in which a molecule of interest is restrained at a series of points along a reaction path using a set of harmonic restraining potentials with different centers and/or strengths (“umbrella potentials”). Each restrained simulation then produces a histogram of system energies and values of the reaction coordinate. These histograms can be combined and the effect of the biases subtracted out, and if there is sufficient overlap between the histogram windows, the free energy change along the reaction coordinate can be estimated.

129 130 References Cited

[1] F. Spelman, “Cochlear electrode arrays: past, present and future,” Audiology & Neuro-otology, vol. 11, no. 2, pp. 77–85, 2006.

[2] P. H. Peckham and J. S. Knutson, “Functional electrical stimulation for neuromuscular applications,” Annual Review of Biomedical Engineering, vol. 7, pp. 327–60, 2005.

[3] L. R. Hochberg, M. D. Serruya, G. M. Friehs, J. A. Mukand, M. Saleh, A. H. Caplan, A. Bran- ner, D. Chen, R. D. Penn, and J. P. Donoghue, “Neuronal ensemble control of prosthetic devices by a human with tetraplegia,” Nature, vol. 442, pp. 164–70, 2006.

[4] J. D. Weiland, W. Liu, and M. S. Humayun, “Retinal prosthesis,” Annual Review of Biomedical Engineering, vol. 7, pp. 361–401, 2005.

[5] P. Troyk, M. Bak, J. Berg, D. Bradley, S. Cogan, R. Erickson, C. Kufta, D. McCreery, E. Schmidt, and V. Towle, “A model for intracortical visual prosthesis research,” Artiﬁcial Organs, vol. 27, no. 11, pp. 1005–1015, 2003.

[6] T. M. Brushart, R. Jari, V. Verge, C. Rohde, and T. Gordon, “Electrical stimulation restores the speciﬁcity of sensory axon regeneration,” Experimental Neurology, vol. 194, pp. 221–229, 2005.

[7] D. M. Taylor, S. I. Helms Tillery, and A. B. Schwartz, “Direct cortical control of 3D neuroprosthetic devices,” Science, vol. 296, no. 5574, pp. 1829–32, 2002.

[8] J. M. Carmena, M. A. Lebedev, R. E. Crist, J. E. O’Doherty, D. M. Santucci, D. F. Dimitrov, P. G. Patil, C. S. Henriquez, and M. A. L. Nicolelis, “Learning to control a brain-machine interface for reaching and grasping by primates,” PLoS Biology, vol. 1, no. 2, pp. 193–208, 2003.

[9] H. K. Kim, S. J. Biggs, D. W. Schloerb, J. M. Carmena, M. A. Lebedev, M. A. L. Nicolelis, and M. A. Srinivasan, “Continuous shared control for stabilizing reaching and grasping with brain-machine interfaces,” IEEE Transactions on Biomedical Engineering, vol. 53, no. 6, pp. 1164–73, 2006.

[10] S. Musallam, B. Corneil, B. Greger, H. Scherberger, and R. Andersen, “Cognitive control signals for neural prosthetics,” Science, vol. 305, no. 5681, pp. 258–62, 2004.

[11] S. Takeuchi and I. Shimoyama, “A radio-telemetry system with a shape memory alloy microelectrode for neural recording of freely moving insect,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 1, pp. 133–7, 2004.

[12] S. K. Talwar, S. Xu, E. S. Hawley, S. A. Weiss, K. A. Moxon, and J. K. Chapin, “Behavioural neuroscience: rat navigation guided by remote control,” Nature, vol. 417, no. 6884, pp. 37–8, 2002.

131 [13] M. Kositsky, A. Karniel, S. Alford, K. M. Fleming, and F. A. Mussa-Ivaldi, “Dynamical dimension of a hybrid neurorobotic system,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 11, no. 2, pp. 155–9, 2003. [14] M. E. Ruaro, P. Bonifazi, and V. Torre, “Toward the neurocomputer: image processing and pattern recognition with neuronal cultures,” IEEE Transactions on Biomedical engineering, vol. 52, no. 3, pp. 371–383, 2005. [15] S. Prasad, X. Zhang, M. Yang, C. S. Ozkan, and M. Ozkan, “Neurons as sensors: individual and cascaded chemical sensing,” Biosensors and Bioelectronics, vol. 19, pp. 1599–610, 2004. [16] J. Wolpaw, W. Birbaumer, N.and Heetderks, D. McFarland, P. Peckham, G. Schalk, E. Donchin, L. Quatrano, C. Robinson, and T. Vaughan, “Brain-computer interface technology: a review of the first international meeting,” IEEE Transactions on Rehabilitation Engineering, vol. 8, no. 2, pp. 164–73, Jun 2000. [17] D. M. Taylor, S. I. H. Tillery, and A. B. Schwartz, “Information conveyed through brain- control: Cursor versus robot,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 11, no. 2, pp. 195–9, 2003. [18] G. Santhanam, S. I. Ryu, B. M. Yu, A. Afshar, and K. V. Shenoy, “A high-performance brain-computer interface,” Nature, vol. 442, pp. 195–8, 2006. [19] J. del R. Millán, F. Renkens, J. Mouriño, and W. Gerstner, “Noninvasive brain-actuated control of a mobile robot by human EEG,” IEEE Transactions on Biomedical Engineering,, vol. 51, no. 6, pp. 1036–33, 2004. [20] P. Heiduschka and S. Thanos, “Implantable bioelectric interfaces for lost nerve functions,” Progress in Neurobiology, vol. 55, no. 5, pp. 433–6, Aug 1998. [21] T. Sinkjaer, “Integrating sensory nerve signals into neural prosthesis devices,” Neuromodulation, vol. 3, no. 1, pp. 35–41, Jan 2000. [22] K. Singh, F. Richmong, and G. Loeb, “Recruitment properties of intramuscular and nerve- trunk stimulating electrodes,” IEEE Transactions on Rehabilitation Engineering, vol. 83, no. 3, pp. 276–85, 2000. [23] M. D. Tarler and J. T. Mortimer, “Selective and independent activation of four motor fascicles using a four contact nerve-cuff electrode,” IEEE Transactions on Neural Systems and Rehabitation Engineering, vol. 12, no. 2, pp. 251–257, 2004. [24] P. B. Yoo and D. M. Durand, “Selective recording of the canine hypoglossal nerve using a multicontact flat interface nerve electrode,” IEEE Transactions on Biomedical Engineering, vol. 52, no. 8, pp. 1461–1469, 2005. [25] D. Tyler and D. Durand, “Functionally selective peripheral nerve stimulation with a flat interface nerve electrode,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 10, no. 4, pp. 294–303, 2002. [26] D. K. Leventhal, M. Cohen, and D. M. Durand, “Chronic histological effects of the flat nerve interface electrode,” Journal of Neural Engineering, vol. 3, pp. 102–13, 2006. [27] L. Wallman, A. Levinsson, J. Schouenborg, H. Holmberg, L. Montelius, N. Danielsen, and T. Laurell, “Perforated silicon nerve chips with doped registration electrodes: in vitro performance and in vivo operation,” IEEE Transactions on Biomedical Engineering, vol. 46, no. 9, pp. 1065–73, Sep 1999. [28] T. Stieglitz, H. Ruf, M. Gross, M. Schuettler, and J.-U. Meyer, “A biohybrid system to interface peripheral nerves after traumatic lesions: design of a high channel sieve electrode,” Biosensors and Biolelectronics, vol. 17, pp. 685–96, 2002.

132 [29] P. Heiduschka, I. Romann, T. Stieglitz, and S. Thanos, “Perforated microelectrode arrays implanted in the regenerating adult central nervous system,” Experimental Neurology, vol. 171, pp. 1–10, 2001.

[30] N. Lago, D. Ceballos, F. J. Rodriguez, T. Stieglitz, and X. Navarro, “Long term assessment of axonal regeneration through polyimide regenerative electrodes to interface the peripheral nerve,” Biomaterials, vol. 26, pp. 2021–2031, 2005.

[31] L. Wallman, Y. Zhang, T. Laurell, and N. Danielsen, “The geometric design of micromachined silicon sieve electrodes inﬂuences functional neve regeneration,” Biomaterials, vol. 22, pp. 1187–93, 2001.

[32] A. Ramachandran, M. Schuettler, N. Lago, T. Doerge, K. P. Koch, X. Navarro, K.-P. Hoﬀ- mann, and T. Stieglitz, “Design, in vitro and in vivo assessment of a multi-channel sieve electrode with integrated multiplexer,” Journal of Neural Engineering, vol. 3, pp. 114–24, 2006.

[33] P. Dario, P. Garzella, M. Toro, S. Micera, M. Alavi, U. Meyer, E. Valderrama, L. Sebastiani, B. Ghelarducci, C. Mazzoni, and P. Pastacaldi, “Neural interfaces for regenerated nerve stimulation and recording,” IEEE Transactions on Rehabilitation Engineering, vol. 6, no. 4, pp. 353–63, Dec 1998.

[34] E. M. Maynard, C. T. Nordhausen, and R. A. Normann, “The Utah Intracortical Electrode Array: a recording structure for potential brain-computer interfaces,” Electroencephalography and Clinical Neurophysiology, vol. 102, no. 3, pp. 228–39, Mar 1997.

[35] P. J. Rousche and R. A. Normann, “Chronic recording capability of the Utah Intracortical Electrode Array in cat sensory cortex,” Journal of Neuroscience Methods, vol. 82, pp. 1–15, 1998.

[36] D. R. Kipke, R. J. Vetter, J. C. Williams, and J. F. Hetke, “Silicon-substrate intracortical microelectrode arrays for long-term recording of neuronal spike activity in cerebral cortex,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 11, no. 2, pp. 151–5, 2003.

[37] R. J. Vetter, J. C. Williams, J. F. Hetke, E. A. Nunamaker, and D. R. Kipke, “Chronic neural recording using silicon-substrate microelectrode arrays implanted in cerebral cortex,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 6, pp. 896–904, 2004.

[38] D. McCreery, V. Pikov, A. Lossinsky, L. Bullara, and W. Agnew, “Arrays for chronic functional microstimulation of the lumbosacral spinal cord,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 12, no. 2, pp. 195–207, 2004.

[39] A. Branner, R. B. Stein, and R. A. Normann, “Selective stimulation of cat sciatic nerve using an array of varying-length microelectrodes,” Journal of Neurophysiology, vol. 85, no. 4, pp. 1585–94, 2001.

[40] A. Branner, R. B. Stein, E. Fernandez, Y. Aoyagi, and R. A. Normann, “Long-term stimulation and recording with a penetrating microelectrode array in cat sciatic nerve,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 1, pp. 146–57, 2004.

[41] C. A. Miller, B. K. Robinson, J. F. Hetke, P. J. Abbas, and K. V. Nourski, “Feasibility of using silicon-substrate recording electrodes within the auditory nerve,” Hearing Research, vol. 198, pp. 48–58, 2004.

[42] D. McDonnall, G. A. Clark, and R. A. Normann, “Interleaved, multisite electrical stimulation of cat sciatic nerve produces fatigue-resistant, ripple-free motor responses,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 12, no. 2, pp. 208–215, 2004.

133 [43] J. Malmstrom, T. McNaughton, and K. Horch, “Recording properties and biocompatibility of chronically implanted polymer-based intrafascicular electrodes,” Annals of Biomedical Engineering, vol. 26, no. 6, pp. 1055–64, Nov-Dec 1998.

[44] S. M. Lawrence, G. S. Dhillon, and K. W. Horch, “Fabrication and characteristics of an implantable, polymer-based, intrafascicular electrode,” Journal of Neuroscience Methods, vol. 131, pp. 9–26, 2003.

[45] S. M. Lawrence, G. Dhillon, W. Jensen, K. Yoshida, and K. W. Horch, “Acute peripheral nerve recording characteristics of polymer-based longitudinal intrafascicular electrodes,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 12, no. 3, pp. 345–348, 2004.

[46] G. Dhillon, T. Kr¨uger,J. Sandhu, and K. Horch, “Eﬀects of short-term training on sensory and motor function in severed nerves of long-term human amputees,” Journal of Neurophysiology, vol. 93, pp. 2625–2633, 2005.

[47] G. S. Dhillon and K. W. Horch, “Direct neural sensory feedback and control of a prosthetic arm,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 13, no. 4, pp. 468–472, 2005.

[48] D. P. Papageorgiou, S. E. Shore, J. Sanford C Bledsoe, and K. D. Wise, “A shut- tered neural probe with on-chip ﬂowmeters for chronic in vivo drug delivery,” Journal of Microelectromechanical Systems, vol. 15, no. 4, pp. 1025–33, 2006.

[49] U. G. Hoﬀman, A. Folkers, F. M¨osch, T. Malina, K. M. Menne, G. Biella, P. Fagerstedt, E. De Schutter, W. Jensen, K. Yoshida, D. Hoehl, U. Thomas, M. G. Kindlundh, P. Norlin, and M. de Curtis, “A novel high channel-count system for acute multisite neuronal recordings,” IEEE Transactions on Biomedical Engineering, vol. 53, no. 8, pp. 1672–7, 2006.

[50] K. A. Moxon, S. C. Leiser, G. A. Gerhardt, K. A. Barbee, and J. K. Chapin, “Ceramic- based multisite electrode arrays for chronic single-neuron recording,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 4, pp. 647–56, 2004.

[51] K. A. Moxon, N. M. Kalkhoran, M. Markert, M. A. Sambito, J. McKenzie, and J. T. Webster, “Nanostructured surface modiﬁcation of ceramic-based microelectrodes to enhance biocompatibility for a direct brain-machine interface,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 6, pp. 881–9, 2005.

[52] P. Rousche, D. Pellinen, D. Pivin, J. Williams, R. Vetter, and D. Kipke, “Flexible polyimide-based intracortical electrode arrays with bioactive capability,” IEEE Transactions on Biomedical Engineering, vol. 48, no. 3, pp. 361–71, 2001.

[53] S. Metz, A. Bertsch, D. Bertrand, and P. Renaud, “Flexible polyimide probes with microelectrodes and embedded microﬂuidic channels for simultaneous drug delivery and multi-channel monitoring of bioelectric activity,” Biosensors and Bioelectronics, vol. 19, pp. 1309–18, 2004.

[54] S. Snow, S. C. Jacobsen, D. L. Wells, and K. W. Horch, “Microfabricated cylindrical mul- tielectrodes for neural stimulation,” IEEE Transactions on Biomedical Engineering, vol. 53, no. 2, pp. 320–326, 2006.

[55] S. Snow, K. W. Horch, and V. K. Mushahwar, “Intraspinal microstimulation using cylindrical microelectrodes,” IEEE Transactions on Biomedical Engineering, vol. 53, no. 2, pp. 311–319, 2006.

[56] P. S. Motta and J. W. Judy, “Multielectrode microprobes for deep-brain stimulation fabricated with a customizable 3-D electroplating process,” IEEE Transactions on Biomedical Engineering, vol. 52, no. 5, pp. 923–933, 2005.

134 [57] T. A. Fofonoﬀ, S. M. Martel, N. G. Hatsopoulos, J. P. Donoghue, and I. W. Hunter, “Mi- croelectrode array fabrication by electrical discharge machining and chemical etching,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 6, pp. 890–5, 2004.

[58] K. Wise, D. Anderson, J. Hetke, D. Kipke, and K. Najaﬁ, “Wireless implantable microsystems: high-density electronic interfaces to the nervous system,” Proceedings of the IEEE, vol. 92, no. 1, pp. 76–97, 2004.

[59] M. Mojarradi, D. Binkley, B. Blalock, R. Andersen, N. Ulshoefer, T. Johnson, and L. D. Castillo, “A miniaturized neuroprosthesis suitable for implantation into the brain,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 11, no. 1, pp. 38–42, 2003.

[60] W. R. Patterson, Y.-K. Song, C. W. Bull, I. Ozden, A. P. Deangellis, C. Lay, J. L. McKay, A. V. Nurmikko, J. D. Donoghue, and B. W. Connors, “A microelectrode/microelectronic hybrid device for brain implantable neuroprosthesis applications,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 10, pp. 1845–53, 2004.

[61] Y.-K. Song, W. R. Patterson, C. W. Bull, J. Beals, N. Hwang, A. P. DeAngelis, C. Lay, J. L. McKay, A. V. Nurmikko, M. R. Fellows, J. D. Simeral, J. P. Donogue, and B. W. Connors, “Development of a chipscale integrated microelectrode/microelectronic device for brain implantable neuroengineering applications,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 13, no. 2, pp. 220–226, 2005.

[62] A. Grumet, J. W. Jr, and J. R. 3rd, “Multi-electrode stimulation and recording in the isolated retina,” Journal of Neuroscience Methods, vol. 101, no. 1, pp. 31–42, Aug 2000.

[63] U. Egert, B. Schlosshauer, S. Fennrich, W. Nisch, M. Fejtl, T. Knott, T. M¨uller, and H. H¨ammerle, “A novel organotypic long-term culture of the rat hippocampus on substrate- integrated multielectrode arrays,” Brain Research Protocols, vol. 2, no. 4, pp. 229–42, 1998.

[64] S. K. Mistry, E. W. Keefer, B. A. Cunningham, G. M. Edelman, and K. L. Crossin, “Cul- tured rat hippocampal neural progenitors generate spontaneously active neural networks,” Proceedings of the National Academy of Sciences, vol. 99, no. 3, pp. 1621–6, 2002.

[65] M. O. Heuschkel, M. Fejtl, M. Raggenbass, D. Bertrand, and P. Renaud, “A three-dimensional multi-electrode array for multi-site stimulation and recording in acute brain slices,” Journal of Neuroscience Methods, vol. 114, no. 2, pp. 135–48, 2002.

[66] G. Krause, S. Lehmann, M. Lehmann, I. Freund, E. Schreiber, and W. Baumann, “Mea- surement of electrical activity of long-term mammalian neuronal networks on semiconductor neurosensor chips and comparison with conventional microelectrode arrays,” Biosensors and Bioelectronics, vol. 21, pp. 1272–1282, 2006.

[67] G. Gholmieh, W. Soussou, M. Han, A. Ahuja, M.-C. Hsiao, D. Song, J. Armand R. Tanguay, and T. W. Berger, “Custom-designed high-density conformal planar multielectrode arrays for brain slice electrophysiology,” Journal of Neuroscience Methods, vol. 152, no. 1-2, pp. 116–29, 2006.

[68] S. L. Smith, J. W. Judy, and T. S. Otis, “An ultra small array of electrodes for stimulating multiple inputs into a single neuron,” Journal of Neuroscience Methods, vol. 133, pp. 109–114, 2004.

[69] A. van Bergen, T. Papanikolaou, A. Schuker, A. M¨oller, and B. Schlosshauer, “Long-term stimulation of mouse hippocampal slice culture on microelectrode array,” Brain Research Protocols, vol. 11, no. 2, pp. 123–33, 2003.

135 [70] M. Sorensen, S. DeWeerth, G. Cymbaluk, and R. L. Calabrese, “Using a hybrid neural system to reveal regulation of neuronal network activity by an intrinsic current,” Journal of Neuroscience, vol. 24, no. 23, pp. 5427–38, 2004.

[71] D. A. Stenger, J. J. Hickman, K. E. Bateman, M. S. Ravenscroft, W. Ma, J. J. Pancrazio, K. Shaﬀer, A. E. Schaﬀner, D. H. Cribbs, and C. Cotman, “Microlithographic determination of axonal/dendritic polarity in cultured hippocampal neurons,” J Neuroscience Methods, vol. 82, pp. 167–73, 1998.

[72] C. Wyart, C. Ybert, L. Bourdieu, C. Herr, C. Prinz, and D. Chatenay, “Constrained synaptic connectivity in functional mammalian neuronal networks grown on patterned surfaces,” Journal of Neuroscience Methods, vol. 117, no. 2, pp. 123–31, 2002.

[73] T. Cornish, D. W. Branch, B. C. Wheeler, and J. T. Campbell, “Microcontact printing: A versatile technique for the study of synaptogenic molecules,” Molecular and Cellular Neuroscience, vol. 20, no. 1, pp. 140–53, 2002.

[74] B. Wheeler, J. Corey, G. Brewer, and D. Branch, “Microcontact printing for precise control of nerve cell growth in culture,” Journal of Biomechanical Engineering, vol. 121, pp. 73–8, 1999.

[75] S. Martinoia, M. Bove, M. Tedesco, B. Margesin, and M. Grattarola, “A simple microﬂuidic system for patterning populations of neurons on silicon micromachined substrates,” Journal of Neuroscience Methods, vol. 87, no. 1, pp. 35–44, 1999.

[76] A. B. Schwartz, “Cortical neural prosthetics,” Annual Review of Neuroscience, vol. 27, pp. 487–507, July 2004.

[77] S. Suner, M. R. Fellows, C. Vargas-Irwin, G. K. Nakata, and J. P. Donoghue, “Reliability of signals from a chronically implanted, silicon-based electrode array in non-human primate primary motor cortex,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 13, no. 4, pp. 524–41, 2005.

[78] W. Shain, L. Spataro, J. Dilgen, K. Haverstick, S. Retterer, M. Isaacson, M. Saltzman, and J. Turner, “Controlling cellular reactive responses around neural prosthetic devices using peripheral and local intervention strategies,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 11, no. 2, pp. 186–8, 2003.

[79] M. D. Johnson, K. J. Otto, and D. R. Kipke, “Repeated voltage biasing improves unit recordings by reducing resistive tissue impedances,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 13, no. 2, pp. 160–5, 2005.

[80] R. Biran, D. C. Martin, and P. A. Tresco, “Neuronal cell loss accompanies the brain tissue response to chronically implanted silicon microelectrode arrays,” Experimental Neurology, vol. 195, pp. 115–26, 2005.

[81] C. S. Bjornsson, S. J. Oh, Y. A. Al-Kofahi, Y. J. Lim, K. L. Smith, J. N. Turner, S. De, B. Roysam, W. Shain, and S. J. Kim, “Eﬀects of insertion conditions on tissue strain and vascular damage during neuroprosthetic device insertion,” Journal of Neural Engineering, vol. 3, pp. 196–207, 2006.

[82] H. Lee, R. V. Bellamkonda, W. Sun, and M. E. Levenston, “Biomechanical analysis of silicon microelectrode-induced strain in the brain,” Journal of Neural Engineering, vol. 2, pp. 81–9, 2005.

[83] J. Subbaroyan, D. C. Martin, and D. R. Kipke, “A ﬁnite-element model of the mechanical eﬀects of implantable microelectrodes in the cerebral cortex,” Journal of Neural Engineering, vol. 2, pp. 103–13, 2005.

136 [84] D. R. Merrill, M. Bikson, and J. G. Jeﬀerys, “Electrical stimulation of excitable tissue: design of eﬃcacious and safe protocols,” Journal of Neuroscience Methods, vol. 141, pp. 171–198, 2005.

[85] P. J. Rousche and R. A. Normann, “Chronic intracortical microstimulation (ICMS) of cat sensory cortex using the Utah Intracortical Microelectrode Array,” IEEE Transacations on Rehabilitation Engineering, vol. 7, no. 1, pp. 56–68, 1999.

[86] S. F. Cogan, A. A. Guzelian, W. F. Agner, T. G. Yuen, and D. B. McCreery, “Over-pulsing degrades activated iridum oxide ﬁlms used for intracortical neural stimulation,” Journal of Neuroscience Methods, vol. 137, pp. 141–150, 2004.

[87] D. McCreery, T. Yuen, W. Agnew, and L. Bullara, “A characterization of the eﬀects on neuronal excitability due to prolonged microstimulation with chronically implanted microelectrodes,” IEEE Transactions on Biomedical Engineering, vol. 44, no. 10, pp. 931–9, Oct 1997.

[88] D. McCreery, T. Yuen, and L. Bullara, “Chronic microstimulation in the feline ventral cochlear nucleus: physiologic and histologic eﬀects,” Hearing Research, vol. 149, pp. 233–8, 2000.

[89] D. B. McCreery, W. F. Agnew, and L. A. Bullara, “The eﬀects of prolonged intracortical microstimulation on the excitability of pyramidal tract neurons in the cat,” Annals of Biomedical Engineering, vol. 30, pp. 107–19, 2002.

[90] D. Warren, E. Fernandez, and R. Normann, “High-resolution two-dimensional spatial mapping of cat striate cortex using a 100-microelectrode array,” Neuroscience, vol. 105, no. 1, pp. 19–31, 2001.

[91] A. Jackson, C. T. Moritz, J. Mavoori, T. H. Lucas, and E. E. Fetz, “The Neurochip BCI: Towards a neural prosthesis for upper limb function,” IEEE Transactions on Neural Systems & Rehabilitation Engineering, vol. 14, no. 2, pp. 187–90, 2006.

[92] A. Jackson, J. Mavoori, and E. E. Fetz, “Long-term motor cortex plasticity induced by an electronic neural implant,” Nature, vol. 44, pp. 56–60, 2006.

[93] D. J. Warren and R. A. Normann, “Functional reorganization of primary visual cortex induced by electrical stimulation in the cat,” Vision Research, vol. 45, pp. 551–565, 2005.

[94] D. Ryugo, E. Kretzmer, and J. Niparko, “Restoration of auditory nerve synapses in cats by cochlear implants,” Science, vol. 310, no. 5753, pp. 1490–2, 2005.

[95] K.-K. Lee, J. He, A. Singh, S. Massia, G. Ehteshami, B. Kim, and G. Raupp, “Polyimide-based intracortical neural implant with improved structural stiﬀness,” Journal of Micromechanics & Microengineering, vol. 14, pp. 32–37, 2004.

[96] A. Singh, H. Zhu, and J. He, “Improving mechanical stiﬀness of coated benzocyclobutene (BCB) based neural implant,” in Proceedings of the 26th Annual International Conference of the Engineering in Medicine and Biology Society (EMBC 2004), no. 2, 2004, pp. 4298–301.

[97] D.-H. Kim, M. Abidian, and D. C. Martin, “Conducting polymers grown in hydrogel scaﬀolds coated on neural prosthetic devices,” Journal of Biomedical Materials Research, vol. 71A, pp. 575–85, 2004.

[98] J. C. Williams, M. M. Holecko II, S. P. Massia, P. Rousche, and D. R. Kipke, “Multi- site incorporation of bioactive matrices into MEMS-based neural probes,” Journal of Neural Engineering, vol. 2, pp. L23–L28, 2005.

[99] D.-H. Kim and D. C. Martin, “Sustained release of dexamethasone from hydrophilic matrices using PLGA nanoparticles for neural drug delivery,” Biomaterials, vol. 27, no. 15, pp. 3031– 3037, 2006.

137 [100] L. Spataro, J. Dilgen, S. Retterer, A. Spence, M. Isaacson, J. Turner, and W. Shain, “Dex- amethasone treatment reduces astroglia responses to inserted neuroprosthetic devices in rat cortex,” Experimental Neurology, vol. 194, pp. 289–300, 2005.

[101] V. S. Polikov, P. A. Tresco, and W. M. Reichert, “Response of brain tissue to chronically implanted neural electrodes,” Journal of Neuroscience Methods, vol. 148, pp. 1–18, 2005.

[102] W. He and R. V. Bellamkonda, “Nanoscale neuro-integrative coatings for neural implants,” Biomaterials, vol. 26, pp. 2983–90, 2005.

[103] C. Vodouhe, M. Schmittbuhl, F. Boulmedais, D. Bagnard, D. Vautier, P. Schaaf, C. Egles, J.-C. Voegel, and J. Ogier, “Eﬀect of functionalization of multilayered polyelectrolyte ﬁlms on motoneuron growth,” Biomaterials, vol. 26, pp. 545–54, 2005.

[104] K. C. Cheung, K. Djupsund, Y. Dan, and L. P. Lee, “Implantable multichannel electrode array based on SOI technology,” IEEE Journal of Microelectromechanical Systems, vol. 12, no. 2, pp. 179–84, 2003.

[105] S. T. Retterer, K. L. Smith, C. S. Bjornsson, K. B. Neeves, A. J. Spence, J. N. Turner, W. Shain, and M. S. Isaacson, “Model neural prostheses with integrated microﬂuidics: a potential intervention strategy for controlling reactive cell and tissue responses,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 11, pp. 2063–2073, 2004.

[106] R. Rathnasingham, D. R. Kipke, S. C. B. Jr., and J. D. McLaren, “Characterization of implantable microfabricated ﬂuid delivery devices,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 1, pp. 138–145, 2004.

[107] X. Cui, V. A. Lee, Y. Raphael, J. A. Wiler, J. F. Hetke, D. J. Anderson, and D. C. Martin, “Surface modiﬁcation of neural recording electrodes with conducting polymer/biomolecule blends,” Journal of Biomedical Materials Research, vol. 56, no. 2, pp. 261–72, 2001.

[108] X. Cui, J. F. Hetke, J. A. Wiler, D. J. Anderson, and D. C. Martin, “Electrochemical deposition and characterization of conducting polymer polypyrrole/PSS on multichannel neural probes,” Sensors and Actuators A, vol. 93, pp. 8–18, 2001.

[109] X. Cui, J. Wiler, M. Dzaman, R. A. Altschuler, and D. C. Martin, “In vivo studies of polypyrrole/peptide coated neural probes,” Biomaterials, vol. 24, no. 5, pp. 777–87, 2003.

[110] J. Yang and D. C. Martin, “Microporous conducting polymers on neural microelectrode arrays. I: Electrochemical deposition,” Sensors & Actuators B, vol. 101, pp. 133–42, 2004.

[111] ——, “Microporous conducting polymers on neural microelectrode arrays. II: Physical characterization,” Sensors & Actuators A, vol. 113, pp. 204–11, 2004.

[112] W. R. Stauﬀer and X. T. Cui, “Polypyrrole doped with 2 peptide sequences from laminin,” Biomaterials, vol. 27, pp. 2405–13, 2006.

[113] X. Cui and D. C. Martin, “Fuzzy gold electrodes for lowering impedance and improving adhesion with electrodeposited conducting polymer ﬁlms,” Sensors and Actuators A, vol. 103, pp. 384–94, 2003.

[114] ——, “Electrochemical deposition and characterization of poly(3,4-ethylenedioxythiophene) on neural microelectrode arrays,” Sensors and Actuators B, vol. 89, pp. 92–102, 2003.

[115] Y. Xiao, X. Cui, and D. C. Martin, “Electrochemical polymerization and properties of PEDOT/S-EDOT on neural microelectrode arrays,” Journal of Electroanalytical Chemistry, vol. 573, pp. 43–8, 2004.

138 [116] K. A. Ludwig, J. D. Uram, J. Yang, D. C. Martin, and D. R. Kipke, “Chronic neural recordings using silicon microelectrode arrays electrochemically deposited with a poly(3,4- ethylenedioxythiophene) (PEDOT) ﬁlm,” Journal of Neural Engineering, vol. 3, no. 1, pp. 59–70, 2006.

[117] K. Wang, H. A. Fishman, H. Dai, and J. S. Harris, “Neural stimulation with a carbon nanotube microelectrode array,” Nano Letters, vol. 6, no. 9, pp. 2043–8, 2006.

[118] T. J. Webster, M. C. Waid, J. L. McKenzie, , R. L. Price, and J. U. Ejiofor, “Nano-biotechnology: carbon nanoﬁbres as improved neural and orthopaedic implants,” Nanotechnology, vol. 15, pp. 48–54, 2004.

[119] H. Hu, Y. Ni, V. Montana, R. C. Haddon, and V. Parpura, “Chemically functionalized carbon nanotubes as substrates for neuronal growth,” Nano Letters, vol. 4, no. 3, pp. 507–11, 2004.

[120] C. D. Bain, E. B. Troughton, Y.-T. Tao, J. Evall, G. M. Whitesides, and R. G. Nuzzo, “Formation of monolayer ﬁlms by the spontaneous assembly of organic thiols from solution onto gold,” Journal of the American Chemical Society, vol. 111, pp. 321–35, 1989.

[121] N. Patel, M. C. Davies, M. Hartshorne, R. J. Heaton, C. J. Roberts, S. J. B. Tendler, and P. M. Williams, “Immobilization of protein molecules onto homogeneous and mixed carboxylate- terminated self-assembled monolayers,” Langmuir, vol. 13, no. 24, pp. 6485–90, 1997.

[122] G. E. Slaughter, E. Bieberich, G. E. Wnek, K. J. Wynne, and A. Guiseppi-Elie, “Improving neuron-to-electrode surface attachment via alkanethiol self-assembly: an alternating-current impedance study,” Langmuir, vol. 20, pp. 7189–7200, 2004.

[123] D. Ryan, B. A. Parviz, V. Linder, V. Semetey, S. K. Sia, J. Su, M. Mrksich, and G. M. Whitesides, “Patterning multiple aligned self-assembled monolayers using light,” Langmuir, vol. 20, no. 21, pp. 9080–8, 2004.

[124] J. Lahiri, E. Ostuni, and G. M. Whitesides, “Patterning ligands on reactive SAMs by microcontact printing,” Langmuir, vol. 15, pp. 2055–60, 1999.

[125] J. L. Tan, J. Tien, and C. S. Chen, “Microcontact printing of proteins on mixed self-assembled monolayers,” Langmuir, vol. 18, pp. 519–23, 2002.

[126] Y. Nam, J. C. Chang, B. C. Wheeler, and G. J. Brewer, “Gold-coated microelectrode array with thiol linked self-assembled monolayers for engineering neuronal cultures,” IEEE Transactions on Biomedical Engineering, vol. 51, no. 1, pp. 158–65, 2004.

[127] J. C. Chang, G. J. Brewer, and B. C. Wheeler, “Modulation of neural network activity by patterning,” Biosensors & Bioelectronics, vol. 16, pp. 527–33, 2001.

[128] ——, “Neuronal network structuring induces greater neuronal activity through enhanced as- troglial development,” Journal of Neural Engineering, vol. 3, pp. 217–26, 2006.

[129] P. Kennedy and R. Bakay, “Restoration of neural output from a paralyzed patient by a direct brain connection,” Neuroreport, vol. 9, no. 8, pp. 1707–11, 1998.

[130] P. Kennedy, R. Bakay, M. Moore, K. Adams, and J. Goldwaithe, “Direct control of a computer from the human central nervous system,” IEEE Transactions on Rehabilitation Engineering, vol. 8, no. 2, pp. 198–202, Jun 2000.

[131] P. R. Kennedy, M. T. Kirby, M. M. Moore, B. King, and A. Mallory, “Computer control using human intracortical local ﬁeld potentials,” IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 12, no. 3, pp. 339–44, 2004.

139 [132] “Increasing spectral channels through current steering in HiResolution Bionic Ear(r) users,” Advanced Bionics Corporation, February 2005, accessible online at http://www.bionicear.com/printables/spectral channels.pdf.

[133] M. Brecht, M. Schneider, B. Sakmann, and T. W. Margrie, “Whisker movements evoked by stimulation of single pyramidal cells in rat motor cortex,” Nature, vol. 427, no. 6976, pp. 704–10, 2004.

[134] N. Fertig, R. H. Blick, and J. C. Behrends, “Whole cell patch clamp recording performed on a planar glass chip,” Biophysical Journal, vol. 82, no. 6, pp. 3056–62, 2002.

[135] X. Li, K. G. Klemic, M. A. Reed, and F. J. Sigworth, “Microﬂuidic system for planar patch clamp electrode arrays,” Nano Letters, vol. 6, no. 4, pp. 815–9, 2006.

[136] B. Matthews and J. Judy, “Design and fabrication of a micromachined planar patch- clamp substrate with integrated microﬂuidics for single-cell measurements,” IEEE Journal of Microelectromechanical Systems, vol. 15, no. 1, pp. 214–222, 2006.

[137] W. L. C. Rutten, “Selective electrical interfaces with the nervous system,” Annual Reviews: Biomedical Engineering, vol. 4, pp. 407–52, 2002.

[138] R. Sch¨atzhauer and P. Fromherz, “Neuron-silicon junction with voltage-gated ionic currents,” European Journal of Neuroscience, vol. 10, no. 6, pp. 1963–7, Jun 1998.

[139] A. Stett, B. Muller, and P. Fromherz, “Two-way silicon neuron interface by electrical induc- tion,” Physical Review E, vol. 55, no. 2, pp. 1779–82, 1997.

[140] B. Eversmann, M. Jenker, F. Hofmann, C. Paulus, R. Brederlow, B. Holzapﬂ, P. Fromherz, M. Merz, M. Brenner, M. Schreiter, R. Gabl, K. Plehnert, M. Steinhauser, G. Eckstein, D. Schmitt-Landsiedel, and R. Thewes, “A 128x128 CMOS biosensor array for extracellular recording of neural activity,” IEEE Journal of Solid-State Circuits, vol. 38, no. 12, pp. 2306–2317, 2003.

[141] M. Jenkner, B. Muller, and P. Fromherz, “Interfacing a silicon chip to pairs of snail neurons connected by electrical synapses,” Biological Cybernetics, vol. 84, no. 4, pp. 239–49, Apr 2001.

[142] P. Bonifazi and P. Fromherz, “Silicon chip for electronic communication between nerve cells by non-invasive interfacing and analog-digital processing,” Advanced Materials, vol. 14, no. 17, pp. 1190–1193, 2002.

[143] P. Fromherz, “Semiconductor chips with ion channels, nerve cells and brain,” Physica E, vol. 16, pp. 24–34, 2003.

[144] M. Hutzler and P. Fromherz, “Silicon chip with capacitors and transistors for interfacing organotypic brain slice of rat hippocampus,” European Journal of Neuroscience, vol. 19, pp. 2231–2238, 2004.

[145] J. Muthuswamy, M. Okandan, T. Jain, and A. Gilletti, “Electrostatic microactuators for precise positioning of neural microelectrodes,” IEEE Transactions on Biomedical Engineering, vol. 52, no. 10, pp. 1748–1755, 2005.

[146] J. Muthuswamy, M. Okandan, A. Gilletti, M. S. Baker, and T. Jain, “An array of microactu- ated microelectrodes for monitoring single-neuronal activity in rodents,” IEEE Transactions on Biomedical Engineering, vol. 52, no. 8, pp. 1470–1477, 2005.

[147] M. J. Tarlov, J. Donald R.F. Burgess, and G. Gillen, “UV photopatterning of alkanethiolate monolayers self-assembled on gold and silver,” Journal of the American Chemical Society, vol. 115, pp. 5305–6, 1993.

140 [148] D. R. Jung, R. Kapur, T. Adams, K. A. Giuliano, M. Mrksich, H. G. Craighead, and D. L. Tay- lor, “Topographical and physicochemical modiﬁcation of material surface to enable patterning of living cells,” Critical Reviews in Biotechnology, vol. 21, no. 2, pp. 111–54, 2001.

[149] G. Yang, N. A. Amro, Z. B. Starkewolfe, and G.-Y. Liu, “Molecular-level approach to inhbit degradations of alkanethiol self-assembled monolayers in aqueous media,” Langmuir, vol. 20, no. 10, pp. 3995–4003, 2004.

[150] N. T. Flynn, T. N. T. Tran, M. J. Cima, and R. Langer, “Long-term stability of self-assembled monolayers in biological media,” Langmuir, vol. 19, no. 26, pp. 10 909–15, 2003.

[151] S. Sharma, R. W. Johnson, and T. A. Desai, “Evaluation of the stability of nonfouling ultrathin poly(ethylene glycol) ﬁlms for silicon-based microdevices,” Langmuir, vol. 20, no. 2, pp. 348– 56, 2004.

[152] A. Ulman, “Formation and structure of self-assembled monolayers,” Chem. Rev, vol. 96, no. 4, pp. 1533–54, 1996.

[153] D. Richardson and J. Mahanty, “Van der Waals contribution to the binding energy of noble metals,” Journal of Physics C (Solid State Physics), vol. 10, no. 20, pp. 3971–6, 1977.

[154] D.-F. Yang, C. Wilde, and M. Morin, “Electrochemical desorption and adsorption of nonyl mercaptan at gold single crystal electrode surfaces,” Langmuir, vol. 12, no. 26, pp. 6570–7, 1996.

[155] T. W. Schneider and D. A. Buttry, “Electrochemical quartz crystal microbalance studies of adsorption and desorption of self-assembled monolayers of alkyl thiols on gold,” Journal of the American Chemical Society, vol. 115, no. 26, pp. 12 391–7, 1993.

[156] D.-F. Yang, C. P. Wilde, and M. Morin, “Studies of the electrochemical removal and eﬃcient re-formation of a monolayer of hexadecanethiol self-assembled at an Au(111) single crystal in aqueous solutions,” Langmuir, vol. 13, no. 2, pp. 243–9, 1997.

[157] X.-L. Su and Y. Li, “A self-assembled monolayer-based piezoelectric immunosensor for rapid detection of Escheria coli O157:H7,” Biosensors and Bioelectronics, vol. 19, pp. 563–74, 2004.

[158] M. Veiseh, B. T. Wickes, D. G. Castner, and M. Zhang, “Guided cell patterning on gold-silicon dioxide substrates by surface molecular engineering,” Biomaterials, vol. 25, pp. 3315–24, 2004.

[159] L. Shriver-Lake, B. Donner, R. Edelstein, K. Breslin, S. Bhatia, and F. Ligler, “Antibody immobilization using heterobifunctional crosslinkers,” Biosensors and Biolectronics, vol. 12, no. 11, pp. 1101–6, 1997.

[160] H. Lu, C. Campbell, and D. Castner, “Attachment of functionalized poly(ethylene glycol) ﬁlms to gold surfaces,” Langmuir, vol. 16, no. 4, pp. 1711–8, Feb 2000.

[161] Z. Zhang, R. Yoo, M. Wells, J. Thomas P. Beebe, R. Biran, and P. Tresco, “Neurite outgrowth on well-characterized surfaces:preparation and characterization of chemically and spatially controlled ﬁbronectin and RGD substrates with good bioactivity,” Biomaterials, vol. 26, pp. 47–61, 2005.

[162] B. T. Houseman, E. S. Gawalt, and M. Mrksich, “Maleimide-functionalized self-assembled monolayers for the preparation of peptide and carbohydrate biochips,” Langmuir, vol. 19, no. 5, pp. 1522–31, 2003.

[163] J. J. S. P. Harder, T. A. Daniel, M. D. Reinard, Y. Yao, D. W. Price, J. M. Tour, and D. L. Allara, “Self-assembled oligo(phenylene-ethynylene) molecular electronic switch monolayers on gold: Structures and chemical stability,” Langmuir, vol. 19, no. 20, pp. 8245–55, 2003.

141 [164] C. A. Hacker, J. D. Batteas, J. C. Garno, M. Marquez, C. A. Richter, L. J. Richter, R. D. van Zee, and C. D. Zangmeister, “Structural and chemical characteristics of monoﬂuoro- substitued oligo(phenylene-ethylene) thiolate self-assembled monolayers on gold,” Langmuir, vol. 20, no. 15, pp. 6195–205, 2004.

[165] M. Jeﬀries-El, G. Sauv´e, and R. D. McCullough, “In-situ end group functionalization of regioregular poly-3-alkylthiophenes using the Grignard metathesis polymerization reaction,” Advanced Materials, vol. 16, no. 2, pp. 1017–9, 2004.

[166] ——, “Facile synthesis of end-functionalized regioregular poly(3-alkylthiophene)s via modiﬁed Grignard metathesis reaction,” Macromolecules, vol. 38, no. 25, pp. 10 346–52, 2005.

[167] J. G. Kushmerick, J. Naciri, J. C. Yang, and R. Shashidhar, “Conductance scaling of molecular wires in parallel,” Nano Letters, vol. 3, no. 7, pp. 897–900, 2003.

[168] S. Jin, S. Cong, G. Xue, H. Xiong, B. Mansdorf, and S. Z. Cheng, “Anisotropic polythiophene ﬁlm with high conductivity and good mechanical properties via a new electrochemical synthesis,” Advanced Materials, vol. 14, no. 20, pp. 1492–6, 2002.

[169] K. E. Harrison, J. F. Kang, R. T. Haasch, and S. M. Kilbey II, “Surface structure and composition of thiophene-bearing monolayers,” Langmuir, vol. 17, no. 21, pp. 6560–8, 2001.

[170] A. Blau, C. Weinl, J. Mack, S. Kienle, G. Jungc, and C. Ziegler, “Promotion of neural cell adhesion by electrochemically generated and functionalized polymer ﬁlms,” Journal of Neuroscience Methods, vol. 112, no. 1, pp. 65–73, 2002.

[171] H.-K. Song, B. Toste, K. Ahmann, D. Hoﬀman-Kim, and G. Palmore, “Micropatterns of positive guidance cues anchored to polypyrrole doped with polyglutamic acid: A new platform for characterizing neurite extension in complex environments,” Biomaterials, vol. 27, pp. 473– 84, 2006.

[172] J. L. Reddinger and J. R. Reynolds, “Molecular engineering of π-conjugated polymers,” Advances in Polymer Science, vol. 145, pp. 57–122, 1999.

[173] D. Braun and P. Fromherz, “Fluorescence interferometry of neuronal cell adhesion on mi- crostructured silicon,” Physical Review Letters, vol. 81, no. 23, pp. 5241–4, Dec 1998.

[174] G. Zeck and P. Fromherz, “Noninvasive neuroelectronic interfacing with synaptically connected snail neurons immobilized on a semiconductor chip,” Proceedings of the National Academy of Sciences, vol. 98, no. 18, pp. 10 457–62, Aug 2001.

[175] S. A. Maskarinec and K. Y. C. Lee, “Comparative study of poloxamer insertion into lipid monolayers,” Langmuir, vol. 19, no. 5, pp. 1809–15, 2003.

[176] R. D. McCullough and P. C. Ewbank, “Regioregular, head-to-tail coupled poly(3- alkylthiophene) and its derivatives,” in Handbook of Conducting Polymers, T. A. Skotheim, R. L. Elsenbaumer, and J. R. Reynolds, Eds. New York, NY: Marcel Dekker, 1998, ch. 9, pp. 225–58.

[177] R. D. McCullough, “The chemistry of conducting polythiophenes,” Advanced Materials, vol. 10, no. 2, pp. 93–116, 1998.

[178] T. Bjornholm, D. R. Greve, N. Reitzel, et al., “Self-assembly of regioregular amphiphilic polythiophenes into highly ordered π-stacked conjugated polymer thin ﬁlms and nanocircuits,” Journal of the American Chemical Society, vol. 120, no. 30, pp. 7643–4, 1998.

[179] D. T. McQuade, A. E. Pullen, and T. M. Swager, “Conjugated polymer-based chemical sensors,” Chemical Reviews, vol. 100, no. 7, pp. 2537–74, 2000.

142 [180] P. Wagner, M. Hegner, H.-J. Güntherodt, and G. Semenza, “Formation and in situ modification of monolayers chemisorbed on ultraflat template-stripped gold surfaces,” Langmuir, vol. 11, pp. 3867–75, 1995. [181] H. Choo, E. Cutler, and Y.-S. Shon, “Synthesis of mixed monolayer-protected gold clusters from thiol mixtures: Variation in the tail group, chain length, and solvent,” Langmuir, vol. 19, no. 20, pp. 8555–9, 2003. [182] J. F. Kang, J. D. Perry, P. Tian, and S. M. Kilbey II, “Growth and morphology of polythiophene on thiophene-capped monolayers: 1. Single component monolayers,” Langmuir, vol. 18, no. 26, pp. 10 196–201, 2002. [183] S. Yue, G. Berry, and R. McCullough, “Intermolecular association and supramolecular organization in dilute solution. 1. Regioregular poly(3-dodecylthiophene),” Macromolecules, vol. 29, pp. 933–9, 1996. [184] N. Kiriy, E. Jähne, H.-J. Adler, M. Schneider, A. Kiriy, S. Minko, D. Jehnichen, P. Simon, A. A. Fokin, and M. Stamm, “One-dimensional aggregation of regioregular polyalkylthiophenes,” Nano Letters, vol. 3, no. 6, pp. 707–12, 2003. [185] N. Kiriy, E. Jähne, A. Kiriy, and H.-J. Adler, “Conformational transitions and aggregations of regioregular polythiophenes,” Macromolecular Symposia, vol. 210, pp. 359–67, 2004. [186] J. Schnitzer and M. Schachner, “Expression of Thy-1, H-2, and NS-4 cell surface antigen and tetanus toxin receptors in early postnatal and adult mouse cerebellum,” Journal of Neuroimmunology, vol. 1, pp. 429–56, 1981. [187] G. Brewer and P. Price, “Viable cultured neurons in ambient carbon dioxide and hibernation storage for a month,” Neuroreport, vol. 7, no. 9, pp. 1509–12, 1996. [188] V. Mirsky, M. Riepl, and O. Wolfbeis, “Capacitive monitoring of protein immobilization and antigen-antibody reactions on monomolecular alkylthiol films on gold electrodes,” Biosensors and Biolectronics, vol. 12, no. 9-10, pp. 977–89, 1997. [189] D. E. Weisshaar, B. D. Lamp, and M. D. Porter, “Thermodynamically controlled electrochemical formation of thiolate monolayers at gold: characterization and comparison to self-assembled analogs,” Journal of the American Chemical Society, vol. 114, no. 14, pp. 5860–2, 1992. [190] C. A. Widrig, C. Chung, and M. D. Porter, “The electrochemical desorption of n-alkanethiol monolayers from polycrystalline Au and Ag electrodes,” Journal of Electroanalytical Chemistry, vol. 310, pp. 335–59, 1991. [191] F. Darain, D.-S. Park, J.-S. Park, and Y.-B. Shim, “Development of an immunosensor for the detection of vitellogenin using impedance spectroscopy,” Biosensors and Bioelectronics, vol. 19, no. 10, pp. 1245–52, 2004. [192] M. Grzeszczuk, J. Bobacka, and A. Ivaska, “Ion transfer at a poly(3-octylthiophene) film electrode,” Journal of Electroanalytical Chemistry, vol. 362, pp. 287–9, 1993. [193] J. Bobacka, M. Grzeszczuk, and A. Ivaska, “Electron transfer at conducting polymer film electrodes: mechanism and kinetics of ferrocene oxidation at poly(3-octylthiophene),” Journal of Electroanalytical Chemistry, vol. 427, pp. 63–9, 1997. [194] R. P. Janek, W. R. Fawcett, and A. Ulman, “Impedance spectroscopy of self-assembled monolayers on Au(111): Evidence for complex double-layer structure in aqueous NaClO4 at the potential of zero charge,” Journal of Physical Chemistry B, vol. 101, no. 42, pp. 8550–8, 1997. [195] S. M. Bezrukov and J. J. Kasianowicz, “Current noise reveals protonation kinetics and number of ionizable sites in an open protein ion channel,” Physical Review Letters, vol. 70, no. 15, pp. 2352–5, 1993.

143 [196] M. Montal and P. Mueller, “Formation of bimolecular membranes from lipid monolayers and a study of their electrical properties,” Proceedings of the National Academy of Sciences, vol. 69, pp. 3561–6, 1972. [197] J. McCammon, B. Gelin, and M. Karplus, “Dynamics of folded proteins,” Nature, vol. 267, pp. 585–90, 1977. [198] P. Shibaev, K. Schaumburg, T. Bjornholm, and K. Norgaard, “Conformation of polythiophene derivatives in solution,” Synthetic Metals, vol. 97, pp. 97–104, 1998. [199] P. Shibaev, M. Arzhakov, K. Schaumburg, R. Vinokur, T. Bjornholm, D. Greve, A. Komolov, and K. Norgaard, “The conformation of poly (3-dodecyl thiophene) within methacrylic polymer matrix,” Journal of Polymer Science Part B Polymer Physics, vol. 37, no. 21, pp. 2909–17, 1999. [200] P. Shibaev and K. Schaumburg, “Conformational transitions in chiral polythiophenes,” Synthetic Metals, vol. 124, no. 2, pp. 291–4, 2001. [201] L. Vouyovitch, D. Brown, S. Neyertz, and B. Gallot, “Prediction of the crystalline structure of a novel polythiophene using molecular dynamics simulations,” Soft Materials, vol. 1, pp. 93–114, 2003. [202] K. Veluri, J. Corish, D. Morton-Blake, and F. Beniere, “A lattice simulation of the migration − of the BF4 ion in polythiophene and polypyrrole lattice,” Journal of Molecular Structure (THEOCHEM), vol. 365, no. 1, pp. 13–19, 1996. [203] J. Corish, D. Feeley, D. Morton-Blake, F. Beniere, and M. Marchetti, “Atomistic investigation of thermochromism in a poly (3-alkylthiophene),” The Journal of Physical Chemistry B, vol. 101, no. 48, pp. 10 075–85, 1997. [204] S. O’Dwyer, H. Xie, J. Corish, and D. Morton-Blake, “An atomistic simulation of the effect of pressure on conductive polymers,” Journal of Physics: Condensed Matter, vol. 13, no. 10, pp. 2395–410, 2001. [205] H. Xie, S. O’Dwyer, J. Corish, and D. Morton-Blake, “The thermochromism of poly(3- alkylthiophene)s: the role of the side chains,” Synthetic Metals, vol. 122, no. 2, pp. 287–96, 2001. [206] D. Morton-Blake and E. Yurtsever, “The entry of molecular species into the lattice of an electroactive polymer during its dissolution,” Journal of Molecular Liquids, vol. 124, no. 1-3, pp. 106–13, 2006. [207] D. Leith and D. Morton-Blake, “A molecular dynamics simulation of the structure and properties of a self assembled monolayer formed from an amphiphilic polymer on a water surface,” Molecular Simulation, vol. 31, no. 14, pp. 987–97, 2005. [208] J. Cirák, P. Tomˇcik,D. Baranˇcok, A. Bolognesi, and M. Ragazzi, “Dipole moment of a modified poly(3-alkylthiophene) at the air/water interface,” Thin Solid Films, vol. 402, pp. 190–4, 2002. [209] A. Gesquière, M. Abdel-Mottaleb, S. De Feyter, F. De Schryver, F. Schoonbeek, J. van Esch, R. Kellogg, B. Feringa, A. Calderone, R. Lazzaroni, and J. Breédas,“Molecular organization of bis-urea substituted thiophene derivatives at the liquid/solid interface studied by scanning tunneling microscopy,” Langmuir, vol. 16, no. 26, pp. 10 385–91, 2000. [210] M. Bachar and O. M. Becker, “Protein-induced membrane disorder: a molecular dynamics study of mellitin in a dipalmitoylphosphatidylcholine bilayer,” Biophysical Journal, vol. 78, no. 3, pp. 1359–75, 2000. [211] J.-H. Lin and A. Baumgaertner, “Stability of a mellitin pore in a lipid bilayer: a molecular dynamics study,” Biophysical Journal, vol. 78, no. 4, pp. 1714–24, 2000.

144 [212] S. K. Kandasamy and R. G. Larson, “Binding and insertion of α-helical anti-microbial peptides in POPC bilayers studied by molecular dynamics simulations,” Chemistry and Physics of Lipids, vol. 132, no. 1, pp. 113–32, 2004.

[213] D. P. Tieleman, H. J. C. Berendsen, and M. S. P. Sansom, “Voltage-dependent insertion of alamethicin at phospholipid/water and octane/water interfaces,” Biophysical Journal, vol. 80, no. 1, pp. 331–46, 2001.

[214] H. Nymeyer, T. B. Woolf, and A. E. Garcia, “Folding is not required for bilayer insertion: replica exchange simulations of an α-helical peptide with an explicit lipid bilayer,” PROTEINS: Structure, Function, and Bioinformatics, vol. 59, pp. 783–90, 2005.

[215] C. F. Lopez, S. O. Nielsen, B. Ensing, P. B. Moore, and M. L. Klein, “Structure and dynamics of model pore insertion into a membrane,” Biophysical Journal, vol. 88, no. 5, pp. 3083–94, 2005.

[216] P. Bond and M. Sansom, “Insertion and assembly of membrane proteins via simulation,” Journal of the American Chemical Society, vol. 128, no. 8, pp. 2697–704, 2006.

[217] M. O. Jensen, O. G. Mouritsen, and G. H. Peters, “Simulations of a membrane-anchored peptide: Structure, dynamics, and inﬂuence on bilayer properties,” Biophysical Journal, vol. 86, no. 6, pp. 3556–3575, 2004.

[218] S. Stepaniants, S. Izrailev, and K. Schulten, “Extraction of lipids from phospholipid membranes by steered molecular dynamics,” Journal of Molecular Modeling, vol. 3, pp. 473–5, 1997.

[219] D. P. Tieleman and H. J. C. Berendsen, “A molecular dynamics study of the pores formed by Escherichia coli OmpF porin in a fully hydrated palmitoyloleoylphosphatidylcholine bilayer,” Biophysical Journal, vol. 74, no. 6, pp. 2786–801, 1998.

[220] T. Bastug and S. Kuyucak, “Test of molecular dynamics force ﬁelds in gramicidin A,” European Biophysics Journal, vol. 34, pp. 377–82, 2005.

[221] D. Case, D. Pearlman, J. Caldwell, I. T.E. Cheatham, J. Wang, W. Ross, C. Simmerling, T. Darden, K. Merz, R. Stanton, A. Cheng, J. Vincent, M. Crowley, V. Tsui, H. Gohlke, R. Radmer, Y. Duan, J. Pitera, I. Massova, G. Seibel, U. Singh, P. Weiner, and P. Kollman, “AMBER 7,” 2002, University of California, San Francisco.

[222] M. Kofranek, T. Kov´a˘r,H. Lischka, and A. Karpfen, “Ab initio studies on heterocyclic conductive polymers: structure and vibrational spectra of thiophene, oligothiophenes, and polythiophene,” Journal of Molecular Structure (THEOCHEM), vol. 259, pp. 181–98, 1992.

[223] M. Ciofalo and G. La Manna, “Ab initio conformational study of 2,2’:5’,2”-terthiophene,” Chemical Physics Letters, vol. 263, pp. 73–8, 1996.

[224] M. J. Frisch, G. W. Trucks, H. B. Schlegel, G. E. Scuseria, M. A. Robb, J. R. Cheeseman, J. A. Montgomery, Jr., T. Vreven, K. N. Kudin, J. C. Burant, J. M. Millam, S. S. Iyengar, J. Tomasi, V. Barone, B. Mennucci, M. Cossi, G. Scalmani, N. Rega, G. A. Petersson, H. Nakatsuji, M. Hada, M. Ehara, K. Toyota, R. Fukuda, J. Hasegawa, M. Ishida, T. Nakajima, Y. Honda, O. Kitao, H. Nakai, M. Klene, X. Li, J. E. Knox, H. P. Hratchian, J. B. Cross, V. Bakken, C. Adamo, J. Jaramillo, R. Gomperts, R. E. Stratmann, O. Yazyev, A. J. Austin, R. Cammi, C. Pomelli, J. W. Ochterski, P. Y. Ayala, K. Morokuma, G. A. Voth, P. Salvador, J. J. Dannenberg, V. G. Zakrzewski, S. Dapprich, A. D. Daniels, M. C. Strain, O. Farkas, D. K. Malick, A. D. Rabuck, K. Raghavachari, J. B. Foresman, J. V. Ortiz, Q. Cui, A. G. Baboul, S. Cliﬀord, J. Cioslowski, B. B. Stefanov, G. Liu, A. Liashenko, P. Piskorz, I. Komaromi, R. L. Martin, D. J. Fox, T. Keith, M. A. Al-Laham, C. Y. Peng, A. Nanayakkara, M. Challacombe, P. M. W. Gill, B. Johnson, W. Chen, M. W. Wong, C. Gonzalez, and J. A. Pople, “Gaussian 03, Revision B.05,” Gaussian, Inc., Wallingford, CT, 2004.

145 [225] J. Wang, W. Wang, and P. Kollman, “Antechamber, an accessory software package for molecular mechanical calculations,” Journal of Computational Chemistry, vol. 25, pp. 1157–74, 2005.

[226] F. Ramirez, V. Hernandez, and J. Navarrete, “Transferable semiempirical quadratic force ﬁelds: The case of polythiophene and shorter oligomers,” Journal of Computational Chemistry, vol. 15, no. 4, pp. 405–23, 1994.

[227] M. Belletˆete, M. Leclerc, and G. Durocher, “Potentialities of semiempirical calculations (AMPAC and INDO/S) in determining the conformation and electronic properties of 2,2’- bithiophene: a new joint experimental and theoretical approach,” Journal of Physical Chemistry, vol. 98, no. 38, pp. 9450–6, 1994.

[228] T. Porter, D. Minore, and D. Zhang, “Semiempirical calculations and scanning probe microscopy studies of polythiophene thin ﬁlms,” The Journal of Physical Chemistry, vol. 99, no. 35, pp. 13 213–13 216, 1995.

[229] N. Di C´esare, M. Belletˆete, M. Leclerc, and G. Durocher, “A conformational study of ethyl- substituted bithiophenes. Semi-empirical versus ab initio methods,” Synthetic Metals, vol. 94, pp. 291–8, 1998.

[230] F. van Bolhuis, H. Wynberg, E. Havinga, E. Meijer, and E. G. Staring, “The X-ray structure and MNDO calculations of α-terthienyl: a model for polythiophenes,” Synthetic Metals, vol. 30, no. 3, pp. 381–9, 1989.

[231] A. Karpfen, C. H. Choi, and M. Kertesz, “Single-bond torsional potentials in conjugated systems: a comparison of ab initio and density functional results,” Journal of Physical Chemistry A, vol. 101, no. 40, pp. 7426–33, 1997.

[232] S. Samdal, E. Samuelsen, and H. Volden, “Molecular conformation of 2, 2’-bithiophene determined by gas phase electron diﬀraction and ab initio calculations,” Synthetic Metals, vol. 59, no. 2, pp. 259–65, 1993.

[233] E. Ort´ı, P. M. Viruela, J. S´anches-Mar´ın, and F. Tom´as, “Ab initio determination of the geometric structure and internal rotation potential of 2,2’-bithiophene,” Journal of Physical Chemistry, vol. 99, no. 14, pp. 4955–63, 1995.

[234] V. Hernandez and J. Navarrete, “Ab initio study of torsional potentials in 2, 2’-bithiophene and 3, 4’- and 3, 3’-dimethyl-2, 2’-bithiophene as models of the backbone ﬂexibility in polythiophene and poly (3-methylthiophene),” The Journal of Chemical Physics, vol. 101, p. 1369, 1994.

[235] C. Quattrocchi, R. Lazzaroni, and J. Br´edas,“Theoretical investigation of the conformational behavior of 2,2’-bithiophene,” Chemical Physics Letters, vol. 208, no. 1-2, pp. 120–4, 1993.

[236] A. Onufriev, D. Bashford, and D. Case, “Exploring protein native states and large-scale conformational changes with a modiﬁed generalized Born model,” Proteins, vol. 55, no. 2, pp. 383–94, 2004.

[237] Y. Sugita and Y. Okamoto, “Replica-exchange molecular dynamics method for protein folding,” Chemical Physics Letters, vol. 314, pp. 141–51, 1999.

[238] R. Faller, Q. Yan, and J. J. de Pablo, “Multicanonical parallel tempering,” Journal of Chemical Physics, vol. 116, no. 13, pp. 5419–23, 2002.

[239] A. Mitsutake, Y. Sugita, and Y. Okamoto, “Generalized-ensemble algorithms for molecular simulations of biopolymers,” Biopolymers (Peptide Science), vol. 60, no. 96, pp. 96–123, 2001.

[240] A. Mitsutake and Y. Okamoto, “Replica-exchange extensions of simulated tempering method,” Journal of Chemical Physics, vol. 121, no. 6, pp. 2491–504, 2004.

146 [241] Y. Pak, S. Jang, and S. Shin, “Prediction of helical peptide folding in an implicit water by a new molecular dynamics scheme with generalized effective potential,” Journal of Chemical Physics, vol. 116, no. 15, pp. 6831–5, 2002. [242] S. Jang, S. Shin, and Y. Pak, “Replica-exchange method using the generalized effective potential,” Physical Review Letters, vol. 91, no. 5, p. 58305, 2003. [243] H. Fukunishi, O. Watanabe, and S. Takada, “On the Hamiltonian replica exchange method for efficient sampling of biomolecular systems: application to protein structure prediction,” Journal of Chemical Physics, vol. 116, no. 20, pp. 9058–67, 2002. [244] S. Gnanakaran, H. Nymyer, J. Portman, K. Y. Sanbonmatsu, and A. E. Garcia, “Peptide folding simulations,” Current Opinion on Structural Biology, vol. 13, pp. 168–74, 2003. [245] D. J. Earl and M. W. Deem, “Parallel tempering: theory, applications, and new perspectives,” Physical Chemistry Chemical Physics, vol. 7, pp. 3910–16, 2005. [246] N. Rathore, M. Chopra, and J. J. DePablo, “Optimal allocation of replicas in parallel tempering simulations,” Journal of Chemical Physics, vol. 122, p. 02411, 2005. [247] A. Kone and D. A. Kofke, “Selection of temperature intervals for parallel-tempering simulations,” Journal of Chemical Physics, vol. 122, p. 206101, 2005. [248] D. A. Case, T. Darden, I. T.E. Cheatham, C. Simmerling, J. Wang, R. Duke, R. Luo, K. Merz, D. Pearlman, M. Crowley, R. Walker, W. Zhang, B. Wang, S. Hayik, A. Roitberg, G. Seabra, K. Wong, F. Paesani, X. Wu, S. Brozell, V. Tsui, H. Gohlke, L. Yang, C. Tan, J. Mongan, V. Hornak, G. Cui, P. Beroza, D. Mathews, C. Schafmeister, W. Ross, and P. Kollman, “AMBER 9,” 2006, University of California, San Francisco. [249] B. Bak, D. Christensen, L. Hansen-Nygaard, and J. Rastrup-Andersen, “The structure of thiophene,” Journal of Molecular Spectroscopy, vol. 7, no. 1-6, pp. 58–63, 1961. [250] A. Almenningen, O. Bastiansen, and P. Svendsas, p. 1671, 1958. [251] D. H. Mosley, J. G. Fripiat, B. Champagne, and J.-M. André, “Ab initio investigation of the static polarizability of planar and twisted infinite polythiophene chains,” International Journal of Quantum Chemistry, vol. 28, pp. 451–67, 1994. [252] T. Prosa, M. Winokur, J. Moulton, P. Smith, and A. Heeger, “X-ray structural studies of poly(3-alkylthiophenes): an example of an inverse comb,” Macromolecules, vol. 25, no. 17, pp. 4364–72, 1992. [253] N. Di Césare, M. Belletête, G. Durocher, and M. Leclerc, “Towards a theoretical design of thermochromic polythiophenes,” Chemical Physics Letters, vol. 275, pp. 533–9, 1997. [254] G. Barbarella, M. Zambianchi, A. Bongini, and L. Antolini, “Conformational chirality of oligothiophenes in the solid state. x-ray structure of 3,4’,4”-trimethyl-2,2’:5’,2”-terthiophene,” Advanced Materials, vol. 6, no. 7/8, pp. 561–4, 1994. [255] C. Khetrapal and A. Kunwar, “The conformation of 2, 2’-bithiophene as determined from PMR studies in a nematic solvent,” Molecular Physics, vol. 28, no. 2, p. 4416, 1974. [256] P. Bucci, M. Longeri, C. Veracini, and L. Lunazzi, “Nematic phase nuclear magnetic resonance investigations of rotational isomerism. III. conformational preferences and interconversion barrier of 2, 2’-bithienyl,” Journal of the American Chemical Society, vol. 96, no. 5, p. 13059, 1974. [257] J. Chadwick and B. Kohler, “Optical spectra of isolated s-cis-and s-trans-bithiophene: Tor- sional potential in the ground and excited states,” The Journal of Physical Chemistry, vol. 98, no. 14, pp. 3631–7, 1994.

147 [258] J. J. Chodera, W. C. Swope, J. W. Pitera, C. Seok, and K. W. Dill, “Use of the weighted histogram analysis method for the analysis of simulated and parallel tempering simulations,” Journal of Chemical Theory and Computation, accepted; in press. [259] G. W. Heffner, D. S. Pearson, and C. Gettinger, “Characterization of poly(3-octylthiophene). I: Molecular characterization in dilute solution,” Polymer Engineering and Science, vol. 35, no. 10, pp. 860–7, 1995. [260] C. Sandstedt, R. Rieke, and C. Eckhardt, “Solid-state and solvatochromic spectra of a highly regular polythiophene,” Chemical Materials, vol. 7, no. 6, pp. 1057–9, 1995. [261] G. Bidan, A. De Nicola, V. Enée, and S. Guillerez, “Synthesis and UV-visible properties of soluble regioregular oligo(3-octylthiophenes), monomer to hexamer,” Chemical Materials, vol. 10, no. 4, pp. 1052–8, 1998. [262] B. de Groot, D. Tieleman, P. Pohl, and H. Grubmuller, “Water permeation through gramicidin A: desformylation and the double helix: a molecular dynamics study,” Biophysical Journal, vol. 82, no. 6, pp. 2934–42, 2002. [263] S. K. Kandasamy and R. G. Larson, “Molecular dynamics simulations of model transmembrane peptides in lipid bilayers: A systematic investigation of hydrophobic mismatch,” Biophysical Journal, vol. 90, no. 7, pp. 2326–43, 2006. [264] E. Lindahl, B. Hess, and D. van der Spoel, “GROMACS 3.0: a package for molecular simulation and trajectory analysis,” Journal of Molecular Modeling, vol. 7, no. 8, pp. 306–17, 2001. [265] Y. Takaoka, M. Pasenkiewicz-Gierula, H. Miyagawa, K. Kitamura, Y. Tamura, and A. Kusumi, “Molecular dynamics generation of nonarbitrary membrane models reveals lipid orientational correlations,” Biophysical Journal, vol. 79, no. 6, pp. 3118–38, 2000. [266] J. Ryckaert and A. Bellemans, “Molecular dynamics of liquid alkanes,” Faraday Discussions of the Chemical Society, vol. 66, pp. 95–106, 1978. [267] H. A. Petrache, S. W. Dodd, and M. F. Brown, “Area per lipid and acyl length distributions in fluid phosphatidylcholines determined by 2H NMR spectroscopy,” Biophysical Journal, vol. 79, no. 6, pp. 3172–92, 2000. [268] W. Humphrey, A. Dalke, and K. Schulten, “VMD: Visual Molecular Dynamics,” Journal of Molecular Graphics, vol. 14, no. 1, pp. 33–8, 1996. [269] J. Phillips, R. Braun, W. Wang, J. Gumbart, E. Tajkhorshid, E. Villa, C. Chipot, R. Skeel, L. Kale, and K. Schulten, “Scalable molecular dynamics with NAMD,” Journal of Computational Chemistry, vol. 26, pp. 1781–802, 2005. [270] X. Cheng, G. Cui, V. Hornak, and C. Simmerling, “Modified replica exchange simulation methods for local structure refinement,” Journal of Physical Chemistry B, vol. 109, no. 16, pp. 8820–30, 2005. [271] P. Liu, B. Kim, R. A. Friesner, and B. Berne, “Replica exchange with solute tempering: A method for sampling biological systems in explicit water,” Proceedings of the National Academy of Sciences, vol. 102, no. 39, pp. 13 749–54, 2005. [272] A. Okur, L. Wickstrom, M. Layten, R. Geney, K. Song, V. Hornak, and C. Simmerling, “Improved efficiency of replica exchange simulations through use of a hybrid explicit/implicit solvent model,” Journal of Chemical Theory and Computation, vol. 2, no. 2, pp. 420–33, 2006. [273] S. Izrailev, S. Stepaniants, B. Isralewitz, D. Kosztin, H. Lu, F. Molnar, W. Wriggers, and K. Schulten, “Steered molecular dynamics,” in Computational Molecular Dynamics: Challenges, Methods, Ideas, ser. Lecture Notes in Computational Science and Engineering, P. Deuflhard, J. Hermans, B. Leimkuhler, A. Mark, S. Reich, and R. Skeel, Eds. Berlin: Springer-Verlag, 1998, no. 4, pp. 39–65.

148 [274] J. Gumbart and K. Schulten, “Molecular dynamics studies of the archaeal translocon,” Biophysical Journal, vol. 90, no. 7, pp. 2356–67, 2006.

[275] K. M. Robertson and D. P. Tieleman, “Orientation and interactions of dipolar molecules during transport through OmpF porin,” Federation of European Biochemical Societies Letters, vol. 528, pp. 53–7, 2002.

[276] Z. Liu, Y. Xu, and P. Tang, “Steered molecular dynamics simulations of Na+ permeation across the gramicidin A channel,” Journal of Physical Chemistry B, vol. 110, no. 25, pp. 12 789–95, 2006.

[277] S. Vemparala, L. Saiz, R. G. Eckenhoﬀ, and M. L. Klein, “Partitioning of anesthetics into a lipid bilayer and their interaction with membrane-bound peptide bundles,” Biophysical Journal, vol. 91, no. 8, pp. 2815–25, 2006.

[278] S. A. Contera, V. Lemaitre, M. R. R. de Planque, A. Watts, and J. F. Ryan, “Unfolding and extraction of a transmembrane α-helical peptide: Dynamic force spectroscopy and molecular dynamics simulations,” Biophysical Journal, vol. 89, no. 5, pp. 3129–40, 2005.

[279] P. W. Fowler and P. V. Coveney, “A computational protocol for the integration of the monotopic protein prostaglandin H2 synthase into a phospholipid bilayer,” Biophysical Journal, vol. 91, no. 2, pp. 401–10, 2006.

[280] M. Balsera, S. Stepaniants, S. Izrailev, Y. Oono, and K. Schulten, “Reconstructing potential energy functions from simulated force-induced unbinding processes,” Biophysical Journal, vol. 73, pp. 1281–7, 1997.

[281] J. R. Gullingsrud, R. Braun, and K. Schulten, “Reconstructing potentials of mean force through time series analysis of steered molecular dynamics simulations,” Journal of Computational Physics, vol. 151, pp. 190–211, 1999.

[282] S. Park, F. Khalili-Araghi, E. Tajkhorsid, and K. Schulten, “Free energy calculation from steered molecular dynamics simulations using Jarzynski’s equality,” Journal of Chemical Physics, vol. 119, no. 6, pp. 3559–66, 2003.

[283] S. Park and K. Schulten, “Calculating potentials of mean force from steered molecular dynamics simulations,” Journal of Chemical Physics, vol. 120, no. 13, pp. 5946–61, 2003.

[284] I. Kosztin, B. Barz, and L. Janosi, “Calculating potentials of mean force and diﬀusion co- eﬃcients from nonequilibrium processes without Jarzynski’s equality,” Journal of Chemical Physics, vol. 124, p. 064106, 2006.

[285] S. Kumar, D. Bouzida, R. H. Swendsen, P. A. Kollman, and J. M. Rosenberg, “The Weighted Histogram Analysis Method for free-energy calculations on biomolecules. I. The method,” Journal of Computational Chemistry, vol. 13, no. 8, pp. 1011–21, 1992.

[286] Y. Shai, “Mechanism of the binding, insertion, and destabilization of phospholipid bilayer membranes by α-helical antimicrobial and cell non-selective membrane-lytic peptides,” Biochimica et Biophysica Acta, vol. 1462, no. 1-2, pp. 55–70, 1999.

[287] S.-J. Marrink, O. Berger, D. P. Tieleman, and F. J¨ahnig, “Adhesion forces of lipids in a phospholipid membrane studied by molecular dynamics simulations,” Biophysical Journal, vol. 74, pp. 931–43, 1998.

[288] D. Tieleman, S. Marrink, and H. Berendsen, “A computer perspective of membranes: molecular dynamics studies of lipid bilayer systems,” Biochimica et Biophysica Acta, vol. 1331, pp. 235–70, 1997.

149 [289] R. W. Benz, F. Castro-Román, D. J. Tobias, and S. H. White, “Experimental validation of molecular dynamics simulations of lipid bilayers: a new approach,” Biophysical Journal, vol. 88, pp. 805–17, 2005. [290] L. Zhai, D. W. Laird, and R. D. McCullough, “Soft-lithography patterning of functionalized regioregular polythiophenes,” Langmuir, vol. 19, no. 16, pp. 6492–7, 2003. [291] Y. Nam, B. C. Wheeler, and M. O. Heuschkel, “Neural recording and stimulation of dissociated hippocampal cultures using microfabricated three-dimensional tip electrode array,” Journal of Neuroscience Methods, vol. 155, pp. 296–9, 2006. [292] M. Niere, B. Braun, R. Gass, S. Sturany, and H. Volkmer, “Combination of engineered neural cell adhesion molecules and GDF-5 for improved neurite extension in nerve guide concepts,” Biomaterials, vol. 27, no. 18, pp. 3432–40, 2006. [293] K. Thelen, V. Kedar, A. K. Panicker, R.-S. Schmid, B. R. Midkiff, and P. F. Maness, “The neural cell adhesion molecule L1 potentiates integrin-dependent cell migration to extracellular matrix proteins,” Journal of Neuroscience, vol. 22, no. 12, pp. 4918–4931, 2002. [294] C. Deister and C. Schmidt, “Optimizing neurotrophic factor combinations for neurite outgrowth,” Journal of Neural Engineering, vol. 3, pp. 172–9, 2006. [295] H. Wang, D. G. Castner, B. D. Ratner, and S. Jiang, “Probing the orientation of surface- immobilized immunoglobulin G by time-of-flight secondary ion mass spectrometry,” Langmuir, vol. 20, no. 5, pp. 1877–87, 2004. [296] C. D. Hodneland, Y.-S. Lee, D.-H. Min, and M. Mrksich, “Selective immobilization of proteins to self-assembled monolayers presenting active site-directed capture ligands,” Proceedings of the National Academy of Sciences USA, vol. 99, no. 9, pp. 5048–52, 2002. [297] V. S. Polikov, M. L. Block, J.-M. Fellous, J.-S. Hong, and W. M. Reichert, “In vitro model of glial scarring around neuroelectrodes chronically implanted in the cns,” Biomaterials, vol. 27, pp. 5368–76, 2006. [298] S. Holdcroft, “Patterning π-conjugated polymers,” Advanced Materials, vol. 13, no. 3, pp. 1753–65, 2001. [299] L. Cheng, J. Yang, Y. Yao, D. W. P. Jr., S. M. Dirk, and J. M. Tour, “Comparative study of electrochemically directed assembly versus conventional self-assembly of thioacetyl-terminated oligo(phenylene ethylene)s on gold and platinum surfaces,” Langmuir, vol. 20, no. 4, pp. 1335– 41, 2004. [300] J. Wang, M. Jiang, A. M. Kawde, and R. Polsky, “Electrochemically induced deposition of thiol-based monolayers onto closely spaced microelectrodes,” Langmuir, vol. 16, no. 25, pp. 9687–9, 2000. [301] P. Buvat and P. Hourquebie, “Enhanced infrared properties of regioregular poly(3- alkylthiophenes),” Macromolecules, vol. 30, no. 9, pp. 2685–92, 1997.

150