Investigating Protein Properties via Microfluidic Techniques Yuewen Zhang Department of Chemistry University of Cambridge Supervised by Professor Tuomas P. J. Knowles This dissertation is submitted for the degree of Doctor of Philosophy Downing College September 2018 Declaration This dissertation is the result of my own work and contains nothing that is the outcome of work done in collaboration, expect where indicated specifically in the text. It is not sub- stantially the same as any that I have submitted for a degree, diploma, or other qualification at the University of Cambridge or other university or similar institution. This dissertation contains less than 60,000 words including the abstract, tables and footnotes, which does not exceed the word limit set by the Degree Committee of Physics and Chemistry. Signed: Yuewen Zhang University of Cambridge Acknowledgements First of all, I would like to thank my supervisor Prof. Tuomas P. J. Knowles for providing me the opportunity to work in such an excellent research environment. His insight, enthusiasm, support and advice have been invaluable for me. I would also like to thank Prof. Christopher M. Dobson for his helpful guidance and support during my PhD studies. Secondly, I would like to thank my collaborators, Prof. Sara Linse and Dr. Stefan Kreida (Lund University), for providing me those precious protein samples and useful scientific discussions. I would also like to thank Prof. John Christodoulou and Dr. Lisa Cabrita (University College London) for their discussions as well as for the great collaboration. During my PhD, I am really grateful for many people to teach me new techniques and skills. Especially, Dr. Emma Yates introduced me microfluidics and Dr. Therese Herling taught me electrophoresis microfluidics and simulations. In addition, thanks to the following colleagues for their great support to my research: Dr. Maya Wright, Dr. Liu Hong, Dr. Pavan K. Challa, Ms. Kadi L. Saar, Mr. Quentin Peter, Dr. Thomas Mu¨ller, Dr. Sean Devenish, Dr. Georg Meisl, Dr. Je´roˆme Charmet and all the other members of the Knowles’ group and the Centre for Misfolding Diseases. Finally, I would like to thank my parents and husband for their support, encouragement and accompany. Also thanks to my friends, Maya Wright, Urszula Lapinska, Emma Yates, Xiaoting Yang, Holly Sun, Junzhan Wang, Yiwei Guan, Annie Chiu and Yuanhui Sun for many great moments together in Cambridge. Abstract Proteins are one of the most abundant biomacromolecules and play important roles in bioac- tivities within organisms, including key functions in DNA replication, immune system re- sponse, metabolic reactions and molecular transportation. Proteins fold into their unique compact three-dimensional structures to precisely perform their functions. Knowledge of proteins on their structures and interactions with other molecules or biomacromolecules is fundamental to understand the mechanism of many diseases. Thus, in this thesis, the focus is on studying the physical properties of proteins, and protein interactions using microfluidic techniques. Chapter 1 provides a brief introduction to protein stability and identification as well as their interactions. Then, conventional techniques for studying protein systems are re- viewed. Moreover, the principles, designs and applications of microfluidic techniques are introduced. Chapter 2 demonstrates that a microfluidic diffusional sizing device can be employed to study protein interactions with small molecules by measuring the variation of hydrody- namic radius of bovine serum albumin (BSA) in aqueous solution as a function of pH. By simulating the behaviour of folded and unfolded BSA, the relative population of BSA in dif- ferent states can be calculated. In addition, the key residues that regulate the BSA unfolding process are predicted. Furthermore, I utilize a modified microfluidic diffusional sizing de- vice to measure the hydrodynamic radius of Escherichia coli 70S ribosome and study its interactions with antibiotics, such as chloramphenicol. vIn Chapter 3, a microfluidic device for protein identification is designed to measure the physical parameters of proteins, including the fluorescence intensities of specific amino acids (tryptophan, tyrosine, lysine) and the hydrodynamic radius of proteins. Thus, proteins can be correctly identified in a single microfluidic device. The results have significant impli- cations in development of effective and efficient techniques for precise protein identification in their native state. In Chapter 4, the interactions between a membrane protein (aquaporins) and a small reg- ulator protein (calmodulin) are investigated by detecting the changes of their electrophoretic mobility and diffusivity in microfluidic devices. Then the effective charges of aquaporin0 and its complex with calmodulin are calculated. Finally, the selective binding between calmodulin and different types of aquaporins is investigated based on their relative binding affinities. Publications The work described in this thesis has resulted in a number of manuscripts, which have been published or are in preparation for submission. In particular, the following manuscripts form the basis of the chapters in this thesis: 1. Yuewen Zhang, Emma V. Yates, Liu Hong, Kadi L. Saar, Georg Meisl, Christo- pher M. Dobson and Tuomas P. J. Knowles. On-chip measurements of protein unfolding from direct observations of micron-scale diffusion. Chemical Sci- ence, 14, 13503-3507 (2018). 2. Challa P. Kumar, Quentin Peter, Maya A. Wright, Yuewen Zhang, Kadi L. Saar, Jacqueline A. Carozza, Justin L. P. Benesch, and Tuomas P. J. Knowles. Real-Time Intrinsic Fluorescence Visualization and Sizing of Proteins and Pro- tein Complexes in Microfluidic Devices. Analytical Chemistry, 6, 3849-3855, (2018). 3. Yuewen Zhang, Maya A. Wright, Challa P. Kumar, Kadi L. Saar, Sean De- venish, Emma V. Yates, Quentin Peter, Christopher M. Dobson and Tuomas P. J. Knowles. Multidimensional Top-down Protein Identification. 2018, In preparation. 4. Yuewen Zhang, Therese W. Herling, Stefan Kreida, Quentin Peter, Tadas Kar- tanas, Susanna To¨rnroth-Horsefiel, Sara Linse and Tuomas P. J. Knowles. On- chip study of membrane protein interactions. 2018, In preparation. vii 5. Christian Bortolini, Tadas Kartanas, Davor Copic, Itzel C. Morales, Yuewen Zhang, Challa P. Kumar, Quentin Peter, Tamas Ja´vorfi, Rohanah Hussain, Mingdong Dong, MingGiuliano Siligardi, Tuomas P. J. Knowles and Je´roˆme Charmet. Resolving protein mixtures using microfluidic diffusional sizing com- bined with synchrotron radiation circular dichroism. Lab on a Chip, (2018), Accepted. Contents Contents viii List of Figures xii List of Tables xv 1 Introduction 1 1.1 Proteins . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1.1 Protein stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.1.2 Protein-protein interactions . . . . . . . . . . . . . . . . . . . . . . 3 1.1.3 Conventional techniques for studying proteins and protein-protein interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 1.2 Microfluidics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.2.1 Types of microfluidic devices . . . . . . . . . . . . . . . . . . . . 10 1.2.2 Fabrication of PDMS microfluidic devices . . . . . . . . . . . . . . 11 1.2.3 Applications of microfluidic techniques in biological areas . . . . . 13 1.3 Outline of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2 On-chip measurements of proteins reacting with small molecules 19 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 2.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 Contents ix 2.2.1 Microfluidic diffusional sizing results . . . . . . . . . . . . . . . . 21 2.2.2 Prediction of pKa values for His, Asp and Glu . . . . . . . . . . . . 28 2.2.3 Determination of the fraction of α-helix . . . . . . . . . . . . . . . 30 2.2.4 On-chip measurements of ribosome-antibiotic interactions . . . . . 33 2.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 2.4 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 2.4.1 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 2.4.2 Measurement of protein concentration . . . . . . . . . . . . . . . . 40 2.4.3 Protein labelling . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 2.4.4 Escherichia coli 70S ribosome preparation and purification . . . . . 41 2.4.5 Imaging set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 2.4.6 Fabrication of black PDMS microfluidic devices . . . . . . . . . . 42 2.4.7 Microfluidic diffusional sizing experiments . . . . . . . . . . . . . 43 2.4.8 Image analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 2.4.9 Circular Dichroism experiment and data analysis . . . . . . . . . . 45 2.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 3 Top-down protein identification with a microfluidic device 51 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 3.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 3.2.1 Feasibility test of a home-built intrinsic fluorescence microscope . . 54 3.2.2 Design of microfluidic devices for top-down protein identification . 57 3.2.3 Top-down measurements of protein physical characteristics . . . . . 60 3.2.4 Protein identification in a single microfluidic device . . . . . . . . 62 3.2.5 Application of top-down protein identification technique . . . . . . 65 3.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 3.4 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 x Contents 3.4.1 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 3.4.2 Sample preparation . . . . . . . . . . . . . . . . . . . . . . . . . . 69 3.4.3 Fabrication of microfluidic devices . . . . . . . . . . . . . . . . . . 69 3.4.4 Fluorescence intensity measurements by plate reader . . . . . . . . 70 3.4.5 Top-down identification measurements . . . . . . . . . . . . . . . 70 3.4.6 Image analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 3.4.7 Diffusional sizing data analysis . . . . . . . . . . . . . . . . . . . 72 3.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 4 Investigation of protein-protein interactions using microfluidic techniques 78 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 4.2 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81 4.2.1 Hydrodynamic radius and interactions of CaM and full-length AQP0 81 4.2.2 Electrophoretic mobility and interactions of labelled CaM and full- length AQP0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 4.2.3 Electrophoretic mobility and interactions of AQP0 and CaM by de- tecting intrinsic fluorescence intensity . . . . . . . . . . . . . . . . 88 4.2.4 Comparing the equilibrium dissociation constants . . . . . . . . . . 91 4.2.5 Calculation of protein charge . . . . . . . . . . . . . . . . . . . . . 93 4.2.6 Detection of interactions between CaM and AQP2 . . . . . . . . . 95 4.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 4.4 Materials and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 4.4.1 Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 4.4.2 Sample preparation . . . . . . . . . . . . . . . . . . . . . . . . . . 97 4.4.3 Imaging set-ups . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 4.4.4 Protein sizing measurements . . . . . . . . . . . . . . . . . . . . . 98 4.4.5 Electrophoresis microfluidic device design . . . . . . . . . . . . . 98 Contents xi 4.4.6 Fabrication of electrophoresis microfluidic device . . . . . . . . . . 99 4.4.7 Microfluidic free-flow electrophoresis experiments . . . . . . . . . 99 4.4.8 Electrophoresis data analysis . . . . . . . . . . . . . . . . . . . . . 100 4.5 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 5 Conclusions 106 5.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 5.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Bibliography 110 List of Figures 1.1 Gibbs free energy landscape of proteins . . . . . . . . . . . . . . . . . . . 3 1.2 The central roles of calmodulin (CaM) . . . . . . . . . . . . . . . . . . . . 4 1.3 Schematic of the glass capillary microfluidic device . . . . . . . . . . . . . 10 1.4 Typical PDMS microfluidic device. . . . . . . . . . . . . . . . . . . . . . . 11 1.5 Scheme of preparation of PDMS microfluidic device. . . . . . . . . . . . . 12 1.6 Schematic of the diffusional sizing device . . . . . . . . . . . . . . . . . . 14 1.7 Schematic of the diffusional sizing device . . . . . . . . . . . . . . . . . . 16 1.8 Schematic of a droplet-maker microfluidic device. . . . . . . . . . . . . . . 17 2.1 The schematic of microfluidic diffusional sizing device . . . . . . . . . . . 22 2.2 Images for proteins with different denaturants in the nµ-size device with clogging issues. A: The Y-junction where protein mixing with the fluores- cent label was clogged with crystal formation. B: The T-junction where labelled protein and unlabelled protein contaction was clogged with the formation of crystals. C: Because both Y-junction and T-junction were clogged, there is no fluorescence signal in the observation region. . . . . . . 23 2.3 Interactions between OPA, BSA and different concentration of urea. A: OPA was added to 3 µM BSA without urea. B: OPA was added to 3 µM BSA with 1.5 M urea. C: OPA was added to 3 µM BSA with 7.5 M urea. . 24 List of Figures xiii 2.4 The condensation interactions between OPA and denaturants. A: OPA in- teracts with urea to form polymer. B: OPA interacts with GdnHCl to form polymer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.5 The average Rh of BSA is measured by the microfluidic diffusional sizing device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.6 Plots of the average Rh versus the number of residues in a polypeptide chain. 26 2.7 The normalised fraction of folded BSA derived from the measured Rh. . . . 27 2.8 Simulation of the diffusion process of folded and unfolded BSA. . . . . . . 28 2.9 Sequence and structure of BSA . . . . . . . . . . . . . . . . . . . . . . . . 29 2.10 Effect of the pH value on the secondary structure of BSA . . . . . . . . . . 31 2.11 Plots of the average Rh versus the normalised fraction of α-helix shows clustering in different states. . . . . . . . . . . . . . . . . . . . . . . . . . 32 2.12 The structure of the 70S ribosome . . . . . . . . . . . . . . . . . . . . . . 33 2.13 Schematic of the ribosome-sizing device . . . . . . . . . . . . . . . . . . . 35 2.14 Simulated diffusion profiles of large particles . . . . . . . . . . . . . . . . 36 2.15 Hydrodynamic radius of 70S ribosome and the mixture of the 70S ribosome and chloramphenicol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 2.16 The process of fluorogenic labelling. . . . . . . . . . . . . . . . . . . . . . 41 2.17 Data fitting of measured Rh and fraction of α-helix of BSA . . . . . . . . . 50 3.1 Top-down protein identification within a single microfluidic device . . . . . 53 3.2 Intrinsic fluorescence microscope . . . . . . . . . . . . . . . . . . . . . . . 55 3.3 Different designs of microfluidic devices for top-down protein identification. 58 3.4 Final design of microfluidic device for top-down protein identification . . . 59 3.5 Diffusion sizing measurements of different proteins . . . . . . . . . . . . . 60 3.6 Fluorescence intensities of lysine residues from each protein measured by microfluidic device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 xiv List of Figures 3.7 Multidimensional data analysis. . . . . . . . . . . . . . . . . . . . . . . . 63 3.8 The probability of protein identification. . . . . . . . . . . . . . . . . . . . 64 3.9 Identification of αB-crystallin and titin . . . . . . . . . . . . . . . . . . . . 66 3.10 Simulation of protein identification . . . . . . . . . . . . . . . . . . . . . . 67 3.11 Fluorescence intensity of tryptophan, tyrosine and lysine . . . . . . . . . . 71 3.12 Measured fluorescence intensities of different proteins . . . . . . . . . . . 75 4.1 Schematic steps of CaM binding to the AQP0 tetramer. . . . . . . . . . . . 80 4.2 Schematic of the protein-sizing microfluidic device . . . . . . . . . . . . . 82 4.3 Rh measurements of CaM under different flow rates. . . . . . . . . . . . . 83 4.4 The measured Rh of CaM and AQP0 . . . . . . . . . . . . . . . . . . . . . 84 4.5 Electrophoresis microfluidic measurements . . . . . . . . . . . . . . . . . 86 4.6 The measured electrophoretic mobility of CaM and AQP0 . . . . . . . . . 87 4.7 The electrophoretic mobility of AQP0 and its complex with titrated CaM . . 89 4.8 Measured and predicted charge of CaM, AQP0, CaM-AQP0 and 2CaM- AQP0. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 4.9 The measured Rh and electrophoretic mobility of CaM and AQP2 . . . . . 95 4.10 The free-flow electrophoresis microfluidic device. . . . . . . . . . . . . . . 99 List of Tables 1.1 Biophysical techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 2.1 The Rh of BSA in buffer with different pH. . . . . . . . . . . . . . . . . . 46 2.2 The Rh and the number of residues of folded and unfolded proteins. . . . . 47 3.1 Summary of tested proteins under the intrinsic fluorescence microscope . . 56 3.2 Absorbance of each protein at 280 nm. . . . . . . . . . . . . . . . . . . . . 69 3.3 Detected proteins in this study . . . . . . . . . . . . . . . . . . . . . . . . 73 3.4 Effective brightness of the aromatic amino acids . . . . . . . . . . . . . . . 73 3.5 Comparison of microfluidic measured Rh of different proteins with literature values. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 3.6 Measured fluorescence intensity ratio and Rh for each protein . . . . . . . . 76 3.7 The protein identification probabilities . . . . . . . . . . . . . . . . . . . . 77 4.1 Dissociation constants (Kd in µM) for full-length or peptide AQP0 interac- tions with CaM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 4.2 Dissociation constants (Kd) for full-length AQP0 interactions with CaM are measured by microfluidic methods . . . . . . . . . . . . . . . . . . . . . . 92 4.3 Physical parameters of AQP0, CaM and its complex . . . . . . . . . . . . . 93 xvi List of Tables Abbreviations HD Huntington Disease PD Parkinson disease AD Alzheimer disease MS Mass spectrometry PPIs Protein-protein interactions 2DE Two dimensional gel electrophoresis NMR Nuclear magnetic resonance cryo-EM Cryo-electron microscopy AUC Analytical ultracentrifugation SEC Size exclusion chromatography HPLC High-performance liquid chromatography ITC Isothermal titration calorimetry SPR Surface plasmon resonance DLS Dynamic light scattering XRC X-ray crystallography MST Microscale thermophoresis Re Reynolds numbers Rh Hydrodynamic radius PDMS Polydimethylsiloxane PMMA Polymethyl methacrylate PGMEA Propylene glycol monomethyl ether acetate IPA Isopropanol pI Isoelectric point PCR Polymerase chain reaction ε Extinction coefficients β -lac β -lactoglobulin α-lac α-lactalbumin HEPES 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid OPA Ortho-phthalaldehyde BME β -Mercaptoethanol SDS Sodium dodecyl sulfate Tris 2-Amino-2-(hydroxymethyl)-1,3-propanediol EDTA Ethylenediaminetetraacetic acid DDM n-Dodecyl β -D-maltoside CaM Calmodulin AQP0 Aquaporin-0 AQP2 Aquaporin-2 CD Circular Dichroism LED Light emitting diode DPI Dual polarisation interferometry MDS Microfluidic diffusional sizing BSA Bovine serum albumin Tyr Tyrosine Trp Tryptophan 4MU 4-methylumbelliferone Chapter 1 Introduction 1.1 Proteins Proteins are the fundamental composition of most biosystems, and consist of amino acid residues. Based on that, proteins fold into three-dimensional structures to determine their specific functions. Proteins are crucial units of cells, tissues and organs, and each pro- tein plays unique functions in different bioactivities, such as DNA replication, catalysing metabolic reactions, responding to stimuli and transportation. Some proteins work to- gether to form protein complexes and play coordinative functions. Over 20,000 different types of proteins exist in the human body and are strongly related to health and disease [127, 167, 178]. Around 80% of proteins are involved in non-covalent and transient protein-protein inter- actions, which regulate cellular functions [21]. However, these interactions are difficult to study by conventional methods due to long experimental time or weak protein interactions. This thesis focuses on investigating the physical properties of proteins to better understand protein unfolding process, protein identification and protein-protein interactions using novel microfluidic techniques. 2 Introduction 1.1.1 Protein stability Proteins exist in different structural states, determined by free energy barriers, thermody- namic stabilities and chemical regulations. The whole range of cellular processes is regu- lated by protein folding and unfolding. Therefore, over the last few decades, scientists have tried to probe the fundamental aspects of protein stability [139]. Proteins in their native states are thermodynamically stable under a range of physiolog- ical conditions [60]. The native structure of proteins are defined by the balance of inter- molecular interactions, including van der waals forces, hydrophobic interactions, hydrogen bonding and electrostatic forces [205]. In general, interactions between proteins in their native states are more stable than in non-native states, because the polypeptide chains are able to find their lowest energy structure. During the protein folding process, the lowest free energy of a protein is determined by its ‘funnel-shaped’ Gibbs-free-energy landscape (Equa- tion 1.1). The width of the ‘funnel-shaped’ energy landscape indicates the chain entropy, therefore, a broad top represents that there are many denatured (unfolded) proteins. The protein folding process is described as going down to the bottom of the funnel landscape, such that the proteins are able to fold rapidly and efficiently (Figure 1.1) [36, 58, 148, 256]. ∆G = ∆H−T∆S (1.1) where the ∆G, ∆H and ∆S are the changes in the total Gibbs free energy, enthalpy and entropy, respectively. And T is the temperature. Protein folding and unfolding is a process in which the secondary and tertiary struc- tures of a protein are altered. The unfolding process is normally caused by external stimuli, such as variations of temperature or pH, the presence of an electric field [165] or the addi- tion of denaturants [54], which can destroy or change the balance of forces within native proteins. Protein misfolding relates to a variety of human diseases, including cystic fibro- sis and some types of familial emphysema. In addition, misfolded proteins form insoluble 1.1 Proteins 3 fibrils or plaques, which accumulate in organs, such as in brain, liver, kidney and heart. These can cause a variety of human diseases, including neurodegenerative diseases such as Alzheimer’s and Parkinson’s diseases [45, 56, 58, 214]. Entropy En er gy Unfolded State Figure 1.1 Gibbs free energy landscape of proteins. Figure reproduced from reference [148]. 1.1.2 Protein-protein interactions Protein-protein interactions (PPIs) are essential for almost all biological processes, includ- ing signal transduction, cell proliferation, etc. [79, 173]. The dissociation constant is com- monly used to describe how tightly a protein binds to another. The binding affinities are influenced by non-covalent intermolecular interactions, including electrostatic interactions, hydrophobic interactions, hydrogen bonding and van der Waals forces [26, 264]. The network of PPIs regulates many bioactivities within organisms. Thus, studying PPIs can help us to understand how organisms fulfill their functions. For example, CaM is the most ubiquitous Ca2+ sensor in cells. It has been shown that CaM interacts with various pro- teins, such as membrane transporters, receptors, enzymes and cytoskeleton proteins, playing important roles in cellular pathways (Figure 1.2). Those studies of PPIs of CaM facili- 4 Introduction tate our understanding of how CaM regulates signalling pathways within inflammatory pro- cesses, immune response, gene expression and cell proliferation, etc. [5, 19, 47, 111, 253]. Figure 1.2 The central roles of calmodulin (CaM). Abbreviations: hnRNPs: Hetero- geneous nuclear ribonucleoproteins; MARCKS:Myristoylated alanine-rich C-kinase sub- strate; MLCK: Myosin light-chain kinase. Figure reproduced from reference [5]. Investigation of PPIs is also of significant importance for pharmaceuticals. One direct application is on development of antibiotics. For example, ribosomes, which are composed of RNAs and proteins, play an essential role in the synthesis of proteins under regulation of PPIs [14]. Thus, ribosomes are often chosen as anti-microbial targets for antibiotics, which are capable of inhibiting the PPIs during translation of bacterial ribosomes, so as to prevent protein synthesis for regeneration of bacteria [246]. Moreover, perturbation of the networks of PPIs or a malfunction in the pathway of PPIs is often related to human diseases. For instance, protein-aggregation related diseases, such as Alzheimer’s disease and creutzfeldt–Jakob, are caused by the aberrant PPIs [109, 120]. Therefore, investigation of PPIs is critical for understanding the physiological properties of both healthy and diseased organisms, offering important insights for drug discovery. 1.1 Proteins 5 1.1.3 Conventional techniques for studying proteins and protein-protein interactions There are a variety of well-established techniques for detecting and studying physical prop- erties of proteins and protein-protein interactions, which are summarised in Table 1.1 and reviewed as below. Conventional techniques for protein characterisation The structural information of proteins are the key parameters for understanding protein func- tions. X-ray crystallography is one of the widely used techniques to study complicated protein structures at atomic resolution [7, 8, 13, 250]. However, it is challenging to study complicated proteins using x-ray crystallography because high quality crystals of protein are required. The heterogeneity and weak interactions make crystallisation of protein ex- tremely difficult. Therefore, scientists always use other techniques together with x-ray crys- tallography to study the structure of proteins [191]. For instance, cryo-electron microscopy (cryo-EM), another important technique for studying protein structures, is often adopted to study the structures of protein together with x-ray crystallography. High-resolution struc- tural data of proteins and their complex at atomic level can be obtained in this manner [22, 119, 122, 239, 262]. Limited by the way of sample preparation both x-ray crystallog- raphy and cryo-EM are not suitable for studying transient protein-protein interactions. Nuclear magnetic resonance (NMR) spectrometry is another important tool for provid- ing structural information of proteins at atomic resolution [206]. Moreover, it has advan- tages on studying dynamic protein interactions. NMR spectra are well resolved, analyti- cally tractable and highly predictable for measuring proteins, with their interactions taking place in solution phase. However, NMR spectrometry is relatively expensive and requires high solubility of protein and protein complexes [232]. In addition, due to long experi- mental time, it is also unsuitable to study less-stable proteins and transient protein-protein 6 Introduction interactions. Moreover, it consumes a high quantity of proteins, at generally hundreds of micromolar level [11, 164]. Molecular weight provides critical information to study proteins and protein-protein in- teractions. Mass spectrometry (MS) is a powerful and higher throughput tool to obtain molecular weight from a few to thousands of Daltons [21, 25]. However, one disadvantage of MS is that proteins are studied in gas phase, which might disrupt protein conformations and protein-protein interactions [155, 215]. Moreover, it requires preliminary separation and purification of proteins prior the measurements. There are several methods widely used for separation of proteins. Two-dimensional polyacrylamide gel electrophoresis (2D-PAGE) is commonly used. It separates proteins based on protein molecular weight and isoelectric point, generating 2D protein maps. Then, separated proteins can be extracted and used for protein characterisation [86]. Size exclusion chromatography (SEC) and high-performance liquid chromatography (HPLC) are also commonly used to separate proteins and protein complexes [126]. SEC and HPLC can also provide some information on molecular weight and are usually coupled with mass spectrometry (MS) to study proteins and protein-protein interactions. Hydrodynamic radius is also an important factor of proteins. Dynamic light scattering (DLS) is a technique used to determine the size and size distribution profile of particles in solution phase. The main disadvantage of DLS is that it requires large amount of sample, which is not feasible for a lot of proteins. Moreover, it is limited by optical transparency of the sample to give reliable signals, and it has a lack of selectivity and has relative low signal strength. Furthermore, it is not generally suitable for heterogeneous mixtures, as the average hydrodynamic radius is dominated by large species with strong dependence of scattering intensity [23, 91, 115]. 1.1 Proteins 7 Conventional techniques for studying protein-protein interactions Protein-protein interactions (PPIs) are essential for almost all bioactivities. Isothermal titra- tion calorimetry (ITC) is a common method for studying PPIs, providing the enthalpy and dissociation constant of protein binding [176]. However, it only detects protein interac- tions that are enthalpically driven, rather than entropically driven [30]. It also requires high concentrations and volumes of protein samples. Surface plasmon resonance (SPR) is a popular optical technique in biology and bio- chemistry for detecting PPIs. With this approach, proteins are coupled onto a planar-coated surface with another protein introduced onto the surface. The change of adsorption is then detected with the increased mass of bound protein partner [141]. There are several advan- tages of SPR, including small amounts of samples are required, and it allows label-free and real-time detection [141]. However, the cost of commercial devices is high, and it requires protein immobilisation which might change the structures or activities of proteins [21]. Protein microarray is another high-throughput method to identify proteins and study PPIs [156, 248, 251]. Generally, it uses antibodies that are functionalised on a surface to target and label specific proteins, then the resulting complexes of antibodies and proteins are read by laser scanners to give information of proteins and PPIs [222, 225]. Protein microar- rays are rapid, automated, economical, highly sensitive and consume small quantities of sample. However, it requires a tailored design of specific antibody probes to target proteins, and the immobilisation of proteins can potentially disrupt protein structures [138]. Analytical ultracentrifugation (AUC) can be used to study protein-protein interactions based on sedimentation of proteins and protein complexes, which depends on their mass, density and sedimentation coefficient. However, due to the long experiment time, it is not applicable for studying protein systems that are not at equilibrium. In addition, centrifugal force may influence the weak PPIs [136]. 8 Introduction Table 1.1 Biophysical techniques, used to detect protein physical characteristics and study protein-protein interactions, -including cryo-electron microscopy (cryo-EM), nuclear mag- netic resonance (NMR), 2-dimensional polyacrylamide gel electrophoresis (2D-PAGE), size exclusion chromatography (SEC), high-performance liquid chromatography (HPLC), mass spectrometry (MS), isothermal titration calorimetry (ITC), surface plasmon resonance (SPR), analytical ultracentrifugation (AUC), dynamic light scattering (DLS) Techniques Experimental information References X-ray crystallography Structure data of proteins [7, 8] cryo-EM Structure data of proteins [22, 119, 122, 262] NMR Structural and dynamic infor- mation of proteins [206] 2D-PAGE Separation of proteins [86] SEC Separation of proteins [126] HPLC Separation of proteins [126] MS Molecular weight [21, 25] Protein microarrays Identification of proteins [156, 225, 248, 251] ITC Protein-protein interaction [30, 176] SPR Protein-protein interaction [21, 141] AUC Protein-protein interactions and protein separation [136] DLS Hydrodynamic radius [91, 115] 1.2 Microfluidics Microfluidics is a technique that processes small amounts of fluids, using channels with dimensions of tens to hundreds of micrometers [243]. It was first developed at the beginning of the 1980s [102], and used for the development of lab-on-a-chip technology [197, 243]. By miniaturising the system to the micron scale, fluid behaviours such as fluidic resistance, 1.2 Microfluidics 9 energy consumption, diffusion and surface tension [17, 212, 217, 243] differ to those on the macroscopic level. On the microfluidic length scale, viscous forces control the fluid flow [243], which is characterised by Reynolds numbers (Re). Re is a dimensionless quantity that defines fluid as being turbulent or laminar. It is calculated by the ratio of inertial forces to viscous forces (Equation 1.2) [87]. Re = ρνL η (1.2) where ρ is the density of fluid (kg/m3), ν is the fluid velocity (m/s), L is the characteristic length (µm) and η is the dynamic viscosity of fluid (kg/(m·s)). At low Reynolds numbers [212, 218] the inertial effects are negligible and laminar flow occurs as the most important feature. Another important parameter for describing fluid in channels is Pe`clet number (Pe), which compares the relative effects of advective to diffusive transport [213], Pe = µw D (1.3) where µ is flow velocity, w is the channel width, and D is the diffusion coefficient. For standard microfluidic applications, the fluids are laminar rather than turbulent when flow rates are lower than 10 m/s, offering unique opportunities for spatial separation instead of temporal separation. The key characteristic of laminar flow is that two or more streams do not mix except during diffusion [218], which makes it possible to extract analytes’ physical characteristic parameters, such as the diffusion coefficient. Thus, the hydrodynamic radius (Rh) can be calculated using the Stokes-Einstein relation (Equation1.4). Rh = κBT 6πηD (1.4) where κB is the Boltzmann constant, T is temperature (K), η is the solution viscosity and D 10 Introduction is the diffusion coefficient of the species. 1.2.1 Types of microfluidic devices Microfluidic devices can be classified into two major categories: glass capillary microfluidic devices and lithographically fabricated microfluidic devices. Glass capillary microfluidic devices Since 2005, glass capillary microfluidic devices have been widely used due to their high chemical resistance [235]. Typically, glass capillary microfluidic devices consist of simple assemblies of basic capillary modules, such as injection tube, transition tube and collection tube, which is generally used for formation of multiple emulsions droplets (Figure 1.3) [46]. However, it is difficult and time-consuming to align all the basic capillary modules, which limits the reproducibility of glass capillary microfluidic devices, and impedes the fabrication of microfluidic devices with hierarchical constructions. Figure 1.3 Schematic of the glass capillary microfluidic device that used for generation of double emulsions from coaxial jets. Figure reproduced from reference [235]. Lithographically fabricated microfluidic devices A widely used alternative method for microfluidic devices is lithography based fabrication, introduced in the late 1990s [62], which allows rapid prototyping and low-cost fabrication. It leads to the popularisation and wide use of microfluidic studies [234]. 1.2 Microfluidics 11 A typical lithography based method for microfluidic devices is soft-lithography of poly- dimethylsiloxane (PDMS) with microchannels followed by its attachment onto substrates (Figure. 1.4). A detailed introduction of soft-lithographical fabrication of PDMS microflu- idic devices is described in the next subsection. PDMS is one of the most common materials for microfluidic devices on account of its biocompatibility, deformability and low toxic- ity [154]. PDMS is also optically transparent to ensure its compatibility with a standard optical microscope [243]. In addition, the accuracy of lithography enables fabrication of PDMS microfluidic devices with advanced and hierarchical constructions [154, 229, 243]. Figure 1.4 Typical PDMS microfluidic device, figure reproduced from reference [124]. Apart from PDMS, there are also various materials that are used for lithographic fabrica- tion of microfluidic devices, including glass [93, 254], polymethyl methacrylate (PMMA) [34, 158], perfluoropolyether (PFPE) [123], fluoropolymer [192], polytetrafluoroethylene (PTFE) and ethylene-propylene (FEP) [89, 189]. Microfluidic devices with different me- chanical properties and chemical compatibilities are prepared using those materials. 1.2.2 Fabrication of PDMS microfluidic devices Lithography based fabrication allows microfluidic devices with complex and advanced de- sign. In this research, PDMS microfluidic devices fabricated based on soft-lithography technique are employed [62]. Therefore, the fabrication of PDMS microfluidic devices is introduced in this section. 12 Introduction Typically, a microfluidic device is firstly designed using AutoCAD software (Autodesk, Inc.). Then the design is printed onto an opaque acetate film, forming the photomask with clear regions for the microchannels, and black regions for the background. Silicon wafer SU-8 Photoresist(a) (b) (c) (d) Silicon wafer PDMS (e) (f) Glass slide Microfluidic channel Silicon wafer Pattern master Photomask Silicon wafer UV light PDMS PDMS Figure 1.5 Scheme of preparation of PDMS microfluidic device. (a): Silicon wafer is coated with SU-8 photoresist. (b): The photoresist is cross-linked by exposure under UV light. (c): Uncross-linked photoresist is removed by developer. (d): PDMS device is cast on the master. (e): The cured PDMS is peeled out. (f): PDMS is bounded to a glass slide to make a microfluidic device. Then, a pattern master is fabricated via soft-lithography (Figure 1.5a, b and c). A layer of SU-8 photoresist with certain height is spin-coated onto a silicon wafer. It is solidified at 96 ◦C for 15 minutes (Figure 1.5a). The silicon master with positive feature is prepared by blocking part of the SU-8 photoresist with the acetate mask and cross-linking the exposed region under UV light [40] (Figure 1.5b). After that, propylene glycol monomethyl ether 1.2 Microfluidics 13 acetate (PGMEA) developer and isopropanol (IPA) are used to remove uncross-linked SU-8 photoresist, leading to formation of the pattern master (Figure 1.5c). The pattern master is then cast with PDMS, which is composed with 10:1 weight ra- tio between PDMS elastomer and curing agent. After removal of trapped bubbles in the PDMS in vacuum, the PDMS is solidified at 65 ◦C for around 2.5 hours (Figure 1.5d). The PDMS is then peeled off (Figure 1.5e) and bonded to either glass slides or quartz slides us- ing an Electronic Diener Femto Plasma bonder, forming PDMS based microfluidic devices (Figure 1.5f). In order to increase the hydrophilicity of the PDMS microfluidic device, the device can be treated by oxygen plasma for 500 s at 80 mW [224]. The device is immediately filled with water afterwards with inlets and outlets blocked with water-filled gel-loading tips. The hydrophilicity can be maintained for at least one week in this manner. 1.2.3 Applications of microfluidic techniques in biological areas Because of the small size of microfluidic devices, it has many advantages compared to con- ventional methods, namely, delivering small quantities of samples and reagents, shortening analysis times, increasing sensitivity for separation and detection of biomolecules down to nanomolar concentrations [151]. Microfluidic techniques are widely used in biological areas, such as disease diagnostics, drug delivery and tissue engineering [16, 140, 252]. Many microfluidic-based diagnos- tic devices have been developed [24, 43, 44, 197, 255, 255]. For example, one research group reported a microfluidic device, which can be used to quantify proteinaceous disease biomarkers. Such diagnostic microfluidic device has the potential applications in clinical studies for rapid detection [101]. 14 Introduction Diffusional microfluidic device When flow of fluids is being miniaturised to microscale in a microfluidic device, laminar flow can be realised at low Reynold numbers. Thus, diffusional mixing of molecules can only happened at convective direction in the laminar flow. This phenomenon has been used for detecting, sizing and sorting particles [94]. Recently, my colleagues, Dr. Yates et al. [257], developed a native microfluidic diffu- sional sizing system to measure proteins in their native states (Figure 1.6). The design of the microfluidic device is based on different diffusivity of particles in the convective direc- tion of laminar flow. Another advantage of the microfluidic device is that it allows latent labelling of proteins just after the diffusion process. i ii Figure 1.6 The schematic of the diffusional sizing device. Reproduced from reference [257]. In the microfluidic diffusional sizing device, protein and buffer meet at t0 (green re- gion), where particles have the same starting spatial distribution across the diffusion chan- nel. At position tD, proteins’ spatial distributions are determined by their hydrodynamic radius (Rh). A well-defined fraction of the protein stream (yellow region) is mixed with the latent labelling dye, and then fluorescence intensity is detected in the observation region. 1.2 Microfluidics 15 This sensitive method can accurately measure the sizes of proteins with over three orders of magnitude of molecular weight while only consuming a few microliters of sample of mi- cromolar concentration. It can also be used to study heterogeneous mixtures, characterise α-synuclein immune complex related to Parkinson’s disease [257], as well as investigate the protein unfolding process [261]. Electrophoresis microfluidic device Almost all proteins have charged amino acids, many of which are located at the active site and play critical roles [83]. The charge of proteins is crucial for us to understand protein stability, solubility, and interactions with other biomolecules. Electric fields are widely used in microfluidic techniques, due to the large surface to volume ratios, joule heat can be dissi- pated efficiently and thus it is negligible [108, 130, 171]. The combination of electronic and microfluidic techniques facilitates the development of microfluidic free-flow electrophore- sis approach that allows researchers to perform electrophoresis measurements in microflu- idic devices [78, 183, 259]. The electrophoresis microfluidic techniques can be used for a wide range of applications, including protein sorting [3, 4], immunoassay development [95, 97, 106], proteins isoelectric point detection [135], measurement of protein charges [78, 183, 259] and protein purification [84]. Recently, my colleagues, Herling et al. [99] developed a free-flow electrophoresis mi- crofluidic technique to firstly detect biomolecule deflections with an applied electric field, which consequently allows the calculation of electrophoretic mobility (Figure 1.7). [98, 99]. In the end, the net charge of biomolecules can be determined based on the electrophoretic mobility. In addition, since the net charge of protein-protein complex is different from the charge of isolated proteins, the microfluidic free-flow electrophoresis approach can be further applied to study protein-protein interactions [100]. This approach requires only mi- croliter scale of analyte at micromolar concentration. 16 Introduction Figure 1.7 The schematic of the electrophoresis device. Reproduced from reference [99]. Flow-focusing microfluidic device In addition to the above microfluidic techniques, microdroplets present an alternative ap- proach to study chemical and biological events, especially for studying single cells and molecules, with a quick and efficient way [195, 226]. The early stages of amlyoid growth are critical in relation to protein aggregation which is closely related to many pathologies. The protein aggregation process is difficult to study using conventional techniques. The flow-focusing microfluidic technique is then developed (Figure 1.8) to separate single nu- cleation sites into microdroplets (Figure 1.8). Taking advantage of the physical separation, the event of primary and secondary nucleation can be observed and quantified conveniently [128]. Moreover, the Flow-focusing microfluidic techniques are widely used to generate emulsion droplets for applications in the controlled loading and releasing protein-type of drugs [96, 208]. In addition, microdroplets generated by the flow-focusing microfluidic technique can be used to screen protein crystallization process and this approach is more effective than other conventional techniques which requires either expensive equipment or a long experiment time [209, 263]. 1.3 Outline of the thesis 17 Figure 1.8 A: Schematic of a microfluidic device that combines a microdroplet maker (B) and a storage array (C). Figure reproduced from reference [128]. Furthermore, there are a variety of applications of microdroplet techniques, including single-cell detection [128, 153, 242, 243], high throughput screening [3, 33, 221], on-chip polymerase chain reaction (PCR) studies [121, 134, 258] and detection of the spatial propa- gation of amyloid formations [128]. Droplet-based microfluidic systems are also used in the biomimetics areas to mimic in vivo functions of human organs [197]. These studies could be used for drug assay before animal testing and clinical human trials. 1.3 Outline of the thesis This thesis describes how to detect unfolding process and some key physical parameters of proteins as well as protein-protein interactions using novel microfluidic techniques. Chapter 2 describes a microfluidic diffusional sizing device and its application in in- vestigation of the processes of pH-induced unfolding of proteins. The protein unfloding processes are revealed by measuring the hydrodynamic radii of proteins at various pH con- ditions and simulating the behaviour of the protein in a systemic manner. The folded and unfolded states of the protein can then be determined by combining results from hydro- dynamic radius measurements with that obtained by circular dichroism. In addition, the key residue that affects the stability of the protein is identified. Followed by that, similar microfluidic diffusional sizing device is developed to measure the hydrodynamic radius of 18 Introduction the 70S ribosome and its subunits (50S ribosome and 30S ribosome?). Consequently, in- teractions between 70S ribosome and antibiotics is investigated using the ribosome-sizing microfluidic device. Based on the microfluidic devices outlined in Chapter 2, a novel microfluidic device equiped with a diffusion channel is developed in Chapter 3 for protein identification. In a single microfluidic device, the intrinsic fluorescence intensity, hydrodynamic radius and relative lysine residue content of the protein can be detected. A database is then generated based on these unique physical parameters that are only related to specific protein. Conse- quently, It demonstrates that the microfluidic device can be used for separation and identifi- cation of protein. In addition, a simulation study on the accuracy of protein identification is illustrated. Chapter 4 describes the measurement of the electrophoretic mobility as well as the sizes of a membrane protein (aquaporin) and its regulatory protein, calmodulin, in the solution phase using an electrophoresis microfluidic device. The state of each binding complex of aquaporin and calmodulin is identified and the effective charge of each component of the complex is calculated based on the measured electrophoretic mobility and sizes. Conse- quently, the binding affinity of the membrane protein aquaporin and calmodulin is calcu- lated. In addition, the selective binding of calmodulin to membrane protein is identified. Chapter 2 On-chip measurements of proteins reacting with small molecules Part of this chapter is based on the publication: Yuewen Zhang; Emma V. Yates; Liu Hong; Kadi L. Saar; Georg Meisl; Christopher M. Dobson; Tuomas P. J. Knowles; On-chip mea- surements of protein unfolding from direct observations of micron-scale diffusion. Chemical Science, 14 (2018): 3503-3507. Abstract Investigations of protein folding, unfolding and stability are critical for understanding of the molecular basis of biological structure and function. A microfluidic approach is described to probe the unfolding of unlabelled protein molecules in microliter volumes. This objec- tive is achieved by using a microfluidic platform, which allows the changes in molecular diffusivity upon folding and unfolding to be detected directly. This approach is illustrated by monitoring the unfolding of bovine serum albumin in solution as a function of pH. These results show the viability of probing protein stability on-chip in small volumes. In addition, the microfluidic approach is further employed to study ribosome/antibiotic binding interac- 20 On-chip measurements of proteins reacting with small molecules tions. The hydrodynamic radius of the Escherichia coli 70S ribosome and its interactions with chloramphenicol are studied. 2.1 Introduction Biomolecular stability plays an important role in virtually every biological process tak- ing place within living systems. Specifically, most proteins must fold precisely into their unique three-dimensional structures to perform their diverse biological functions. Incorrect protein folding often causes malfunction, and can give rise to a range of human diseases [45, 57, 59, 61, 129, 227]. Indeed, a particularly prevalent class of disorders associated with the aberrant folding of proteins involves amyloid formation and is connected to neurode- generative diseases, such as Alzheimer’s disease and Parkinson’s disease [45, 68, 241]. Proteins can be denatured by changing their chemical or physical environment, such as by adding chemical denaturants, changing the solution pH, heating or applying pressure. The thermodynamic stability of the folded-state proteins, quantified as the Gibbs free energy difference between the folded and unfolded states, is commonly probed through denatura- tion experiments, which promote unfolding [70]. A number of methods have been estab- lished for studying the unfolding of protein structures, including circular dichroism (CD) [88], nuclear magnetic resonance (NMR) spectroscopy [63, 179, 244], dual polarisation in- terferometry (DPI) [228] and fluorescence-based optical techniques [266]. These methods have advanced very significantly our understanding of the nature of protein structure and stability. Generally, however, these approaches require high concentrations of protein, need long processing times of several hours, and may cause changes in the native folded protein structure due to the installation of labels which are often used to enhance optical or magnetic signals [201]. Micron-scale measurements of molecular diffusivity have been shown to be a highly sensitive approach to define the sizes of proteins and to bring together the benefits of label- 2.2 Results and Discussion 21 based and label-free methods [10, 257, 260]. The ability to assess rapidly the folding state of a protein, using small volumes of unlabelled analytes, could have applications for laboratory scale protein science, where stability is a key parameter of interest, as well as for person- alised medicine and diagnostics. Indeed, a commonly used modality to detect the binding of small molecule drugs to protein targets is to follow the resultant increase in the stability of the native state, a process that could be miniaturised using platforms of the type described in this chapter. Microfluidic systems are highly portable, cost effective, and can easily be integrated into sensing platforms with potential applications in personalised medicine. Re- cently, I reported a microfluidic approach for measuring the sizes of proteins [257] with the key characteristics that the proteins of interest are labelled on-chip with a fluorogenic dye immediately prior to an optical detection step (Figure 1.6). This approach has an additional advantage of allowing the study of proteins with high sensitivity. 2.2 Results and Discussion 2.2.1 Microfluidic diffusional sizing results In this study, I set out to explore how the microfluidic diffusional sizing approach [257] can be used to study the changes in protein size induced by folding and unfolding. In particular, this approach was employed to study in detail the denaturation process of bovine serum albumin (BSA) induced with different denaturants (urea and guanidine hydrochloride (GdnHCl) ), as we as with varying the pH. Microfluidic diffusional sizing device The schematic design of the microfluidic diffusional sizing device [257] is shown in Fig- ure 2.1, which contains one outlet and three inlets for protein, buffer and dye, respectively. In brief, the protein and phosphate buffer streams, with equal volumetric flow rates, meet 22 On-chip measurements of proteins reacting with small molecules at the position labelled t0. At this point, protein molecules have not diffused into the buffer stream, and they have the same initial distribution in the protein flow, irrespective of molec- ular weight or structures. Also, each of the protein and buffer stream spans half the width of the diffusional channel (Figure 2.1 and Figure 2.8a), existing as laminar flow in the dif- fusional channel. edb Y-junction Labelling region T-junctionObservation region c a Protein Inlet Buffer Inlet Dye Inlet t0 b c d e Diffusional channel tD Figure 2.1 (a): 3D schematic of the microfluidic diffusional sizing device used in this study. (b): The Y-junction showing the protein mixing with fluorogenic labelling solution (c): The labelling region for o-Phthalaldehyde (OPA) react with primary amine containing residues on the protein. (d): The observation region for monitoring the fluorescence intensity of the labelled protein. (e): The T-junction showing the flow of labelled and unlabelled protein solution. Subsequently, protein diffuses laterally into the buffer stream along with the flow di- rection in the diffusional channel. At the end of the diffusional channel (tD), the proteins of smallest hydrodynamic radius (Rh) have diffused furthest into the buffer stream. Subse- quently, a third of the total stream (Figure 2.1 and Figure 2.8a) is diverted into the latent labelling region (Figure 2.1b) where the diffused protein molecules are quantitatively la- belled with o-Phthalaldehyde (OPA) [20, 194]. Note that, in the diffusional channel, the mixing process proceeds exclusively via diffusion as convective mixing is suppressed in 2.2 Results and Discussion 23 small volumes at low Reynold numbers [35]. The total concentration of protein molecules diverted for labelling is therefore determined by diffusivity alone and hence by the protein Rh [257]. In other words, measuring the fluorescence intensity in the observation region (Figure 2.1d) defines the total concentration of protein diverted for labelling at position tD, which in turn reveals the protein distribution at position tD, allowing determination of Rh by comparison with values simulated for particles of known Rh values [159, 257]. Exploring the process of BSA folding and unfolding induced by denaturants The microfluidic diffusional device is initially employed to investigate the folding and un- folding of BSA initiated by denaturants of urea and/or GdnHCl based on the determination of the variation of the hydrodynamic radius of BSA. During the experiment, we’ve found that crystals emerge inside the channel of the microfluidic device as shown in Figure 2.2. A B C Figure 2.2 Images for proteins with different denaturants in the nµ-size device with clogging issues. A: The Y-junction where protein mixing with the fluorescent label was clogged with crystal formation. B: The T-junction where labelled protein and unlabelled protein contaction was clogged with the formation of crystals. C: Because both Y-junction and T-junction were clogged, there is no fluorescence signal in the observation region. This may be caused by the interactions between BSA, urea/GdnHCl and OPA mixture, or any of the component with the PDMS wall of the microfluidic device. In order to un- derstand the origin of these crystals, the reactions between proteins, denaturants and OPA labelling mixtures were examined macroscopically in eppendorf tubes (Figure 2.3). When OPA was added to BSA solution, the solution remains clear and no precipitates can be 24 On-chip measurements of proteins reacting with small molecules observed. Yet, precipitates are found when OPA was added to mixtures of BSA and denat- urants (urea or GdnHCl, Figure 2.3 B and C, respectively). Hence, the crystals observed in microfluidic channels are resulted from the reaction interactions between the denaturants and OPA. Actually, it has been known that urea or GdnHCl can react with OPA, forming condensed polymers at the conditions employed in the experiment (Figure 2.4). A B C Figure 2.3 Interactions between OPA, BSA and different concentration of urea. A: OPA was added to 3 µM BSA without urea. B: OPA was added to 3 µM BSA with 1.5 M urea. C: OPA was added to 3 µM BSA with 7.5 M urea. O O H H H2N NH2 NN O n OPA Urea O O O H H H2N NH2 NN NH n NH OPA OPA urea GdnHCl A B Figure 2.4 The condensation interactions between OPA and denaturants. A: OPA interacts with urea to form polymer. B: OPA interacts with GdnHCl to form polymer. Due to the side reaction between OPA and the denaturants of urea or GdnHCl, we thus switched to investigate the BSA unfolding process induced by variation of pH using the microfluidic diffusional sizing device. 2.2 Results and Discussion 25 Measurement of Rh of BSA at different pH values Changing the pH is a common way to achieve protein unfolding. The reason for protein unfolding at certain pH is that buried ionisable groups on amino acids within the polypeptide sequence have a highly perturbed pKa. Typically, the buried groups of proteins have lower pKa values in the native state than in the unfolded state [70]. This difference creates a thermodynamic driving force increasingly favouring the unfolded state when the pH of the solution is lowered. Folded and unfolded proteins have different sizes. In order to probe the unfolding pro- cess of proteins, BSA has been chosen as a model protein in this study. The average Rh values of BSA under different pH conditions is measured using the microfluidic diffusional sizing device. 0 1 2 3 4 5 6 7 8 9 pH Value 10 11 A ve ra ge R h ( nm ) 3 4 5 6 7 8 9 Microfluidic data Figure 2.5 The average Rh of BSA is measured by the microfluidic diffusional sizing device in buffer solutions at varying pH. The results are shown in Figure 2.5 and Table 2.1 (in the appendix section). The average Rh of BSA is observed to be almost constant when the pH is between 4.3 and 10.2; at pH 7.0, the average Rh for BSA is found to be 3.60 ± 0.41 nm, which is consistent with the reported value of 3.39 ± 0.27 nm [74]. Therefore, all BSA molecules in the solution are in their folded state when the pH is between 4.3 and 10.2. As the pH is reduced below 4.3, the 26 On-chip measurements of proteins reacting with small molecules average Rh value is observed to increase progressively (Figure 2.5), indicating the unfolding of BSA. At this stage, folded and unfolded BSA coexists in the solution. When the pH is further reduced to 1.2, the average Rh value of BSA reaches to 8.4 ± 0.16 nm, suggesting completely unfolding of BSA. Moreover, the average Rh values of BSA in folded and unfolded state measured using the microfluidic diffusional sizing device fit well to a polymer scaling law [104] between hydrodynamic radius and number of residues of reported proteins at folded or unfolded states [244] (Rh ∝ Nα , Figure 2.6 and Table 2.2 in the appendix section). Number of residues 50 100 200 400 600 H yd ro dy na m ic ra di us (n m ) 1 2 4 6 9 folded unfolded N0.43 ±0.09 N0.52 ±0.07 Unfolded BSA Folded BSA Figure 2.6 Plots of the average Rh versus the number of residues in a polypeptide chain. The values for folded and unfolded BSA (shown in green) are measured using the microfluidic diffusional sizing device. Literature values are shown as blue diamonds and grey circles for a range of folded and unfolded proteins respectively [53, 73, 116, 117, 168, 184, 204, 244] (Table 2.2). Investigation of folded and unfolded BSA The microfluidic approach can be used not only to obtain the average Rh value of folded and unfolded BSA, but also to derive the relative populations of the two forms in a given solution. The fraction of folded BSA at different pH values is calculated based on the experimentally measured average Rh of BSA using the microfluidic diffusional sizing device and linear interpolation of Rh = (Rmaxh −Rminh ) · (1− fN)+Rminh (Figure 2.7). 2.2 Results and Discussion 27 pH value 11 N or m al ise d fra ct io n of fo ld ed B SA 0 0.5 1 0 1 2 3 4 5 6 7 8 9 10 Eq. 2.1 Figure 2.7 The normalised fraction of folded BSA derived from the measured Rh. Together with Kadi, we further validate the data by simulating of a mixture of BSA with different sizes flowing through the diffusional channel from position t0 to tD (Fig- ure 2.8a). In the simulation, the movement of 106 particles with various compositions of folded (3.4 nm) and unfolded (8.4 nm) BSA is monitored in a rectangular channel of 200 µm width, 25 µm height and 17000 µm length at a flow rate of 25 µL/h [159]. The simulations are based on solving the Langevin equation describing diffusional behaviour in advection [32, 159, 219, 240]. For each ratio of folded and unfolded BSA, the diffusivity is simulated as follows: the first round with half of the channel filled with protein molecules and the second round with full channel filled with protein molecules. C++ was used for the simulations, and details could be found in reference [159]. By integrating the particle distri- bution across the rectangular channel at the end of the diffusion channel (tD, Figure 2.8b), I obtain the number of molecules that have diffused far enough to enter the fluid stream that subsequently flows into the labelling region of the device for each of the simulations (Figure 2.8b dotted line). By comparing the relative intensities of the two simulations, a calibration curve is constructed, which linked the recorded fluorescence intensity ratios to the fraction of folded and unfolded protein molecules (Figure 2.8c). This constructed curve is then used to relate the observed fluorescence intensities at each of the pH values and 28 On-chip measurements of proteins reacting with small molecules the average Rh to the relative population of folded and unfolded proteins in the mixture. The obtained unfolding curves agree well with previously published results of acid-induced unfolding measured by different techniques [66]. a b c unfolded BSA folded BSA Buffer Protein Labelling 0 50 100 1500 0.2 0.4 0.6 0.8 200 fN = 1 fN = 0.5 fN = 0 (µm) to labelling Fl uo re sc en t I nt en si ty R at io Sp ec ie s c on ce nt ra tio n (a .u .) Fraction of folded moleculesChannel position 1.0 0.80.60.40.20 1.0 0.06 0.05 0.04 0.03 0.02 0.01 0 µm 200 µm ch an ne l p os iti on 50 µm 100 µm t0 tD Figure 2.8 Simulation of the diffusion process of folded and unfolded BSA. (a): Schematic illustration of BSA diffusive process within the diffusional channel of the microfluidic dif- fusional sizing device. (b): Diffusive behaviour of a mixture of folded and unfolded BSA at the end of the diffusional channel (tD). Completely folded and unfolded BSA correspond to fN = 1 and fN = 0 respectively. 50% folded and 50% unfolded BSA corresponds to fN = 0.5. (c): Calibration curve of the fraction of folded and unfolded BSA against the fluorescence intensity ratio. 2.2.2 Prediction of pKa values for His, Asp and Glu To obtain an estimate for the unfolding free energy, together with Dr. Liu Hong, we explored whether a single ionisable group can act as a key titration site during the unfolding process (m=1) (Figure 2.17, in the appendix section). Indeed, analysis of the pKa values of the ionisable residues of the BSA sequence reveals that one particular residue, His241, has large difference between its pKa values in the folded and unfolded states (Figure 2.9a). 2.2 Results and Discussion 29 Asp Glu His b a His-241 1-4 0 -4 0 6 residue number 6 3 3 3 6 9 13 16 17 18 37 38 39 45 48 56 57 59 63 67 72 73 82 16 6 17 1 17 2 18 2 86 89 92 95 97 10 0 10 5 10 7 10 8 11 1 11 8 12 4 12 5 12 9 13 0 14 0 14 5 15 2 18 6 20 7 22 6 22 9 23 6 24 1 24 3 24 6 24 8 25 1 25 4 25 5 25 8 26 5 26 8 27 6 27 9 28 4 28 7 29 1 29 3 29 5 29 9 30 7 31 0 31 1 31 3 32 0 32 3 33 2 33 7 33 9 35 1 35 3 35 7 pK a - p Ka -4 0 6 3 35 8 36 3 36 4 36 6 37 4 37 8 38 1 38 2 39 2 39 5 39 9 42 4 44 1 44 3 44 9 45 0 46 3 46 4 47 0 47 8 49 3 49 4 50 3 50 9 51 1 51 7 51 9 53 0 53 4 54 0 54 1 54 8 55 5 56 1 56 2 57 0 56 4 54 8 pK a - p Ka pK a - p Ka Figure 2.9 (a): The difference in pKa values between folded and unfolded states for the Asp, Glu and His residues of BSA along the sequence. The key titration site His241 is indicated by an arrow. (b): The crystal structure of BSA (PDB ID: 4F5S) is shown in ribbon structure. His241 (red sphere) is highlighted and shown in more detail in the inset. More specifically, since the pH induced unfolding transition observed in Figure 2.5 oc- curs between pH = 1.2 and pH = 4.3, the pKa values for the potential titration sites must span this interval. This condition guarantees that over the pH range (from pH 1.2 to 4.3), where BSA is observed to unfold, the ionisable groups in the unfolded state is increasingly protonated, while the folded state remains unchanged, thus driving the equilibrium towards the unfolded state. In particular, for a single titration site m = 1 and given that the unfolding occurs over the pH range [1.2, 4.3], we require pKa < 1.2, pK˜a > 4.3 and hence pK˜a - pKa > 3.1. The pKa values of each ionisable group in the folded state are predicted by computational analysis (DEPTH server [223], Figure 2.9a), leading to the identification of one key titration site, His241, satisfying the above criterion with pK˜a = 6.04 and pKa = 0.6. The pKa values for other residues (such as glutamic acid (Glu) and aspartic acid (Asp)) do 30 On-chip measurements of proteins reacting with small molecules not meet the requirement (Figure 2.9a). For a single titration site, thermodynamic arguments (Equation 2.12) yield a simple expression for the fraction of folded protein. fN = KN KN+10pK˜a−pH (2.1) where fN is fraction of folded protein; KN is the equilibrium constant for BSA folding at the given pH. Based on the microfluidic results, KN = 2213 ± 272, corresponding to a standard free energy ∆G⊖ = -4.55 ± 0.08 kcal/mol. 2.2.3 Determination of the fraction of α-helix I also investigated the extent of secondary structure change under different pH conditions by circular dichroism (CD) spectroscopy. The normalised fractions of folded BSA derived from the molar ellipticity at 222 nm, 208 nm and total integrated area between 200 - 250 nm were calculated using Equation 2.2 and Equation 2.3 (Figure 2.10). From the CD spectra, the folding free energy of the BSA is determined to be ∆G⊖ = -4.24± 0.03 kcal/mol, which agrees well with the value of -4.55 ± 0.08 kcal/mol based on our microfluidic results (see details later), as well as -4.04 kcal/mol and -4.60 kcal/mol reported from previous studies [90]. This close agreement of the free energies obtained in our study by unfolding at neutral pH via denaturant, suggests that any protonation of residues other than His241 at pHs above 4.3 does not significantly affect the relative stabilities of the folded and unfolded states (see details later). 2.2 Results and Discussion 31 Phosphate buffer with different pH 10 11E llip tic ut y [ θ] (d eg · c m 2 · d m ol- 1 ) -2300 -2100 -1900 -1700 -1500 c 208nm model Phosphate buffer with different pH 10 11E llip tic ut y [ θ] (d eg · c m 2 · d m ol- 1 ) -2300 -2100 -1900 -1700 -1500 -1300 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 222nm model 10 11987654321 Phosphate buffer with different pH El lip tic ut y [ θ] (d eg · c m 2 · d m ol- 1 ) × 104 4 4.5 5 5.5 6 6.5 area model d ba M ola r E llip tic ity Wavelength (nm) 200 210 220 230 240 250 -3000 -2000 -1000 0 1000 2000 pH=1.2 pH=2.0 pH=2.5 pH=2.7 pH=3.3 pH=3.7 pH=4.3 pH=5.0 pH=6.0 pH=7.0 pH=7.9 pH=9.0 pH=10.2 Figure 2.10 Effect of the pH value on the secondary structure of BSA (3 µM) in sodium phosphate (5 mM) buffer using far-UV CD (200 nm - 250 nm). Based on the far-UV CD spectra. (a): the molar ellipticity of BSA at 222 nm (b): 208 nm (c) and total area between 200 - 250 nm (d) are calculated. All the measurements are done three times and error bars represent the standard deviation among independent replicates. The normalised fraction of α-helices (with respect to maximal amount of α-helices of BSA) can be extracted from the mean residue ellipticity measured by far-UV CD at 222 nm, as fα = ( Θobs222−Θmin222 Θmax222 −Θmin222 ) × (fNα − fUα)+ fUα (2.2) where Θobs222, Θ max 222 and Θ min 222 are the observed, maximal and minimal mean residue ellipticity at 222 nm respectively. fUα and f N α are the fraction of α-helix in the unfolded and folded proteins. A hypothesis is proposed that there is a linear relationship between the ellipticity and secondary structure content. To explain the dependence of α-helix content on different pH conditions, we noticed the fact that, due to the electrostatic repulsion, the α-helix structure becomes unstable in the presence of a high concentration of hydrogen ions. The unfolding of BSA through residue 32 On-chip measurements of proteins reacting with small molecules protonation at low pH values is a consequence of this mechanism. Since on average 73% of residues are in the helical form [181], a good correlation could be assumed between the fraction of α-helix and the degree of protein folding fN ∈ [0,1], fα = fNfNα +(1− fN)fUα (2.3) The availability of both microfluidic and CD measurements allow us to carry out multi- dimensional cluster analyses of the folding and unfolding process (Figure 2.11). Corre- sponding to the folded and unfolded state, two major clusters are observed in agreement with the reported two-state folding behaviour of BSA [90]. Interestingly, four additional data sets with intermediate average Rh values and α-helix contents are also observed that correspond to mixtures of folded and unfolded BSA, identifying the trajectory from the folded to unfolded state (Figure 2.11). The fact that the folding transition occurs in a similar manner along a coordinate measuring global structure (Rh, horizontal axis, Figure 2.11) and along a coordinate sensitive to local structure (fraction of α-helix, vertical axis, Figure 2.11) supports the two-state nature of this transition. 0 0.2 0.4 0.6 0.8 1.0 unfolded folded 3 4 5 6 7 8 9 222nm 208nm area Average hydrodynamic radius (nm) N or m al is ed fr ac tio n of α -h el ix Figure 2.11 Plots of the average Rh versus the normalised fraction of α-helix shows cluster- ing in different states. The normalised fraction of α-helix derived from the molar ellipticity at 222 nm, 208 nm and total area between 200 - 250 nm are calculated using Equation 2.2. 2.2 Results and Discussion 33 2.2.4 On-chip measurements of ribosome-antibiotic interactions The ribosome is a large and complex macromolecular machine that exists in all living cells, and it plays an essential role in synthesis of polypeptide chains. It is also a platform for regulation of enzymes and chaperones, which control nascent polypeptides emerging from the exit tunnel. The ribosome complex in Escherichia coli is used as a model for this study. The Es- cherichia coli ribosome, known as ‘70S’, is composed of a large subunit 50S and small subunit 30S as shown in Figure 2.12. 30S 50S A P E Figure 2.12 The complex of the 70S ribosome was generated using Pymol. Subunits of the 70S are magnified and represented in cartoon, 50S (purple, PDB: 2AW4 [203]) and the 30S (grey, PDB: 2AVY [203]). The A, P and E sites of tRNA are shown on the surface representation in red, yellow and green, respectively. Different subunits of the ribosome are associated with different functions in protein synthesis. The 30S is typically responsible for correctly decoding mRNA, while the function of 50S is to catalyse peptide bond formation and also to promote protein folding during synthesis [200]. The compositions: the 30S are 16S RNA and 21 proteins [249], while the 50S has three components of 5S rRNA subunit, the 23S rRNA subunit and 31 proteins [14]. RNA accounts for 2/3 of the total mass of the ribosome and also acts as the main 34 On-chip measurements of proteins reacting with small molecules functional core of ribosome. The 23S rRNA in 50S has 6 large domains, including domain V, which constitutes the peptidyl transferase center (PTC) and also holds the A, P and E sites (Figure 2.12) [198], the most important region for peptide bond formation. The PTC is the place for polypeptide formation, therefore, it is also a prime target binding site for a number of antibiotics, such as chloramphenicol and puromycin [92]. On the small subunit 30S, some antibiotics bind along the process of mRNA to tRNAs, while in contrast, for the 50S subunit, most antibiotics bind at or near the PTC [247]. Dif- ferent binding sites affect the ribosome’s function differently, including peptide transfer, GTPase activity, nascent peptide chain transit and physiological changes. Many antibiotics have been used in clinical applications, for example, chloramphenicol and erythromycin have been used for treatment of ocular infections and respiratory tract infections [177]. Thus, a better understanding on ribosomes and their interactions with antibiotics is impor- tant in drug design. The microfluidic diffusional sizing approach is further employed here to study 70S and its interactions with an antibiotic. Modification of the microfluidic diffusional sizing device The microfluidic diffusional sizing device (Figure 2.1) for studying unfolding of BSA can only measure particles up to 10 nm in radius [257]. Yet, the 70S ribosome is reported to be around 12 nm in radius [77]. Therefore, a modified microfluidic diffusional sizing device (ribosome-sizing device) is required to measure the hydrodynamic radius of 70S ribosome. The design principles for the ribosome-sizing device are that: 1), there should be no convective mixing of the flows from the two inlets in the microfluidic diffusional channel. 2), a long enough diffusional channel is required for larger ribosome particles to diffuse into the observation channel. 2.2 Results and Discussion 35 Labelling O bs er va tio n Y-junction Dye Inlet Ribosome Inlet Buffer Inlet Outlet T-junctionT0 Td Diffusion Channel Figure 2.13 The schematic of ribosome-sizing device. The Y-junction, labelling region, ob- servation region and T-junction are highlighted in yellow, pink, blue and green, respectively. The channel height, width and length for the microfluidic diffusional sizing device (Fig- ure 2.1) are 25 µm, 200 µm and 17000 µm, respectively. Based on equations 2.4 and 2.5, the channel length for the ribosome-sizing device was increased to 70000 µm, while keeping the height and width the same as the microfluidic diffusional sizing device. The modified microfluidic diffusional sizing device is shown in Figure 2.13. x = √ 2Dt (2.4) where x is the diffusion distance, t is the diffusion time and D is diffusion coefficient. Rh = κBT 6πηD (2.5) where Rh is hydrodynamic radius, κB is the Boltzmann constant, T is temperature (K), η is the solution viscosity and D is the diffusion coefficient of species. Hydrodynamic radius of the 70S ribosome To determine the Rh of the 70S ribosome, a low concentration of this species (3.74 nM) was used. In the ribosome-sizing device (Figure 2.13), ribosome contact with buffer oc- 36 On-chip measurements of proteins reacting with small molecules curs at the T0 position, before diffusion takes place. At this stage, the stream of the 70S ribosome and the stream of the buffer each occupy half of the diffusional channel. As the streams flow from the point T0 to the point Td , the 70S ribosome diffuses from the 70S ribosome-containing stream to the buffer-containing stream. At the point Td , one third of the stream in the diffusional channel (which contains diffused 70S ribosomes) flows into the labelling region (Figure 2.13, highlight in pink) along with the labelling solution containing fluorescent dye. After labelling is complete, the fluorescence intensity is then detected in the observation region (Figure 2.13, highlighted in blue). Fluorescence intensities in the observation region were analysed using ImageJ, details are described in Section 2.4.8. 0 20 40 60 80 100 120 140 160 180 0.4 1 3.0 4.0 5.3 7.1 9.5 12.6 16.8 22.4 29.8 39.7 50.0 0 0.2 0.6 0.8 Channel width (µm) Pa rti cl e C on ce nt ra tio n (A U ) 10-2 10-1 5 10 25 Observed intensity ratio Hy dr od yn am ic R ad iu s ( nm ) a b Figure 2.14 Simulated diffusion profiles of large particles. (a): Basis function is defined by the concentration gradients simulated for reference particles with already-known hydrody- namic radii. Particles at T0 (Figure 2.13) have the same distribution (dashed line). After diffusion through the channel, at Td position (Figure 2.13), one third of stream in the diffu- sional channel with diffused particles is selected and fluorescently labelled (gray rectangle). The fluorescence intensity ratio was calculated based on the ratio of observation region flu- orescence intensity for the homogeneously distributed species (black line). Hydrodynamic radius of the protein was calculated by fitting the experimental fluorescence intensity ratios as a calibration curve (b). Then, the Rh of the 70S ribosome was calculated by fitting the experimental fluorescence intensity ratios to a calibration curve (Figure 2.14b), which was obtained by plotting refer- ence particles with already-known hydrodynamic radii against their observed fluorescence 2.2 Results and Discussion 37 intensity ratio (Figure 2.14a). The fluorescence intensity of the 70S ribosome was detected, and its Rh was calculated as 12.02 ± 0.87 nm. This measured result is in agreement with a literature value where the hydrodynamic radius of the 70S ribosome was obtained by atomic force microscopy [149]. Measurement of interactions of 70S ribosome and chloramphenicol The interactions between the 70S ribosome and chloramphenicol was then detected by titrat- ing the different concentrations of chloramphenicol against 3.74 nM 70S ribosome. Chlo- ramphenicol is a well-characterised antibiotic that binds to two different sites on the 50S subunit of the ribosome to block peptidyl transferase and prevent protein synthesis How- ever, I did not detect any interactions between 70S ribosome and chloramphenicol because there was no difference in hydrodynamic radius between the 70S ribosome and the mix- ture of the 70S ribosome and chloramphenicol (Figure 2.15). The reason might be that the hydrodynamic radius of chloramphenicol is too small to be detected. Figure 2.15 Hydrodynamic radius of the 70S ribosome and the mixture of 70S ribosome and chloramphenicol. Alternatively, in the future, the binding properties of the ribosome with antibiotics of interest can be studied by measuring the diffusion coefficient or electrophoretic mobility. 38 On-chip measurements of proteins reacting with small molecules The antibiotics binding efficiency of ribosome will be measured by calculating the dissoci- ation constants (KD) of ribosome-antibiotics interactions, which will have implications for improved drug design. Having validated the capability of the ribosome sizing device to accurately determine the dimension of intact ribosome by consuming several nanomolar sample. Further work may involve kinetic studies of the dissociation/association of ribosomal subunits by adding reagents, such as Mg2+, by following changes in the hydrodynamic radius variation of the ribosome. Second, the technique will be used to study peptide binding to or cleavage from the ribosome. In addition, the microfluidic technique can be employed to explore the mech- anism of nascent chain release from the ribosome, such as the time required for nascent chain release from the 70S ribosomal subunit and nascent chain interactions with trigger factors. 2.3 Conclusions In this work, I have shown that the microfluidic diffusional sizing approach can be used to investigate the interactions of proteins with small molecules. In particular, by measuring the average Rh of BSA, I found BSA is unfolded under acidic conditions (between pH 1.2 and 4.3). The average Rh of folded and unfolded BSA were measured to be 3.6 ± 0.41 nm and 8.4 ± 0.16 nm, respectively. During the unfolding process, the relative fractions of folded and unfolded BSA were calculated based on the measured average Rh. By combining the measured secondary structure of BSA with circular dichroism, a two-state model was ob- served. The key residue affecting the relative stabilities of the folded and unfolded states was identified. In addition, based on the microfluidic diffusional sizing device, a ribosome- sizing device was constructed with a longer diffusional channel, which could be used for measuring the Rh of the 70S ribosome. The Rh of the 70S ribosome was 12.02 ± 0.87 nm, in a agreement with the literature reported value of 12.5 nm. The interaction between the 2.3 Conclusions 39 70S ribosome and chloramphenicol was then detected by titrating chloramphenicol with 70S ribosome. However, the Rh of the mixture was 13.25± 0.61 nm, which is similar to the Rh of the 70S ribosome. Thus, no conclusive interactions between the 70S ribosome and chloramphenicol could be detected using the ribosome-sizing device. Compared to conven- tional techniques, the microfluidic technique only requires microliters of sample solution. The residence time is of the order of a few seconds for each measurement. I therefore an- ticipate that this microfluidic approach will open up new possibilities for the study of the structural stability of proteins and other biomolecules under a variety of conditions. 40 On-chip measurements of proteins reacting with small molecules 2.4 Materials and Methods 2.4.1 Materials Bovine serum albumin (BSA), sodium bicarbonate, sodium carbonate, ortho-phthalaldehyde (OPA), sodium phosphate dibasic) and sodium phosphate monobasic (#S5011), sodium do- decyl sulfate (SDS) were obtained from Sigma-Aldrich. β -mercaptoethanol (BME) was obtained from Thermo Scientific (Leicestershire, UK). All solutions were prepared using ultrapure water and filtered. The standard labelling solution was 12 mM OPA, 18 mM BME, and 20% w/v SDS in 200 mM sodium carbonate buffer, pH 10.5 [257]. The labelling solution was protected from light, and sonicated for 15 mins to completely dissolved. Then the labelling dyes was filtered with Millex-GP 0.22 µm filter (Merck Millipore). It can be stored at room temperature and used within 3 days of preparation. The buffer used for detecting 70S sibosome contains 10 mM Hepes ((4-(2-hydroxyethyl)- 1-piperazineethanesulfonic acid), 2 mM EDTA (Ethylenediaminetetraacetic acid), 30 mM NH4Cl, 12 mM MgCl2, 1 mM BME (2-Mercaptoethanol), 0.1% protease inhibitor tablet, pH 7.5. 2.4.2 Measurement of protein concentration BSA was dissolved in 5 mM sodium phosphate buffer with different pH values. The pH value was observed not to change substantially after addition of the small concen- tration of BSA. The concentrations of BSA were measured using standard protein A280 settings on Nanodrop spectrophotometer. The molar extension coefficient (ε) for BSA is 43,824 M−1cm−1. 2.4 Materials and Methods 41 2.4.3 Protein labelling Fluorescence is generated by a reaction between OPA and primary amines. Primary amine groups, such as lysine groups and the protein N-terminus, react with the OPA dialdehyde in the presence of the thiol group in BME to forms a conjugated pyrrole ring, resulting in the formation of a substituted isoindole, and fluorescence in the blue region of the spectrum was generated [211, 216] (Figure 2.16). High concentrations of SDS were added to ensure all primary amines are exposed and available for reaction [169, 257]. OPA BME Native State Protein Fluorescence H H O O NH2 SDS H NH O O H H NH O HO S H NH O S OH N OH SH OH N S OH - H2O - H2O Figure 2.16 Native proteins are denatured with SDS to expose all the primary amine groups. OPA is not fluorescent until it reacts with a primary amine to form a substituted isoindole. The fluorescence was formed in the blue region of the spectrum. 2.4.4 Escherichia coli 70S ribosome preparation and purification The 70S ribosome was produced from Escherichia coli as previous described [38]. Expres- sion and purification of 70S ribosome was provided by Dr. Lisa Cabrita, UCL. The 70S ribosome containing fractions was flash-frozen in liquid nitrogen and stored at -80 ◦C for microfluidic measurements. 42 On-chip measurements of proteins reacting with small molecules 2.4.5 Imaging set-up In order to detect proteins in the microfluidic devices, a fluorescence microscope set-up was used. Imaging of proteins labelled with OPA was done using an inverted fluorescence microscope (Zeiss Axio Observer) fitted with a 49000 DAPI filter. The photons emitted from the sample were acquired with an Evolve 512 EMCCD camera (Photometrics, Arizona, USA) coupled with a 10X objective. This gives a field of view of 825 × 825 µm. 2.4.6 Fabrication of black PDMS microfluidic devices A layer of SU-8 photoresist (3 mL, MicroChem Corp) was spin coated onto a silicon wafer (3 inch diameter, MicroChemicals) and baked on hotplate at 96 ◦C for 15 minutes. The device height was set by SU-8 3025 photoresist for 25 µm. It was solidified at 96 ◦C for 15 minutes (Figure 1.5a). The silicon master with positive feature was prepared by block- ing part of the photoresist with the acetate mask and cross-linking the exposed region with the exposed UV light with 30 s, 100 mV [40] (Figure 1.5b). The silicon wafer was then post-baked on 96 ◦C hotplate for 5 minutes. After that, propylene glycol monomethyl ether acetate (PGMEA) developer (Sigma) and isopropanol (IPA) were used to remove uncross- linked SU-8 photoresist, leading to formation the pattern master (Figure 1.5c). All the microfluidic channels heights were measured by scanning the soft-lithography master with a profilometer (DektakXT, Bruker). Height variations between devices were taken into ac- count to eliminate the influence of device height variations of the data analysis. The pattern master was then cast with PDMS, which was composed with the 10:1 weight ratio between PDMS elastomer and curing agent (Dow Corning, product 184). Black PDMS microfluidic devices were used, which can reduce the noise during image acquisition, around 1 mg/mL of carbon nanopowder (Sigma, product 633100) should be added to the elastomer/curing agent, and mixed thoroughly with centrifuge at 5000 rpm for 10 mins. The mixture was then poured onto the master and bubbles are removed by vacuum desiccation 2.4 Materials and Methods 43 for 30 minutes. Finally the device was baked at 65 ◦C for around 2.5 hours (Figure 1.5d). After the PDMS was cooled, it was peeled off from silicon wafer with microchannels, and holes were punched with 0.75 mm diameter Harris Uni-Core punchers (Figure 1.5e). All debris were removed with sellotape and sonicated with IPA. It is important to remove all residual IPA before the bonding step, thus the device was dried by nitrogen gun and baked for 10 minutes at 65 ◦C. The PDMS was then bonded to Thermo Scientific 76 x 26 mm glass slides (catalog 8037) using an Electronic Diener Femto Plasma bonder, forming PDMS based microfluidic devices (Figure 1.5f). In order to get hydrophilic device, the bonding was firstly involved a 30 s at 40 mW generation of oxygen plasma. After the device was put in contact with the glass slide, it was baked for 10 minutes at 65 ◦C to form a completely sealed device. Then, the sealed device was exposed to oxygen plasma for 500 s at 80 mW to prevent hydrophobic recovery [224], because oxygen plasma interacts with PDMS surface to induce radical formation. Radicals then interact with plasma to form functionalised group that has hydrophilic properties [28]. The device was immediately filled with water after bonding step by using portex 0.38 mm internal 16 diameter, 1.09 mm external diameter tubing, connected to 1 mL plastic Air-Tite syringes. The inlets and outlets were blocked with water-filled gel loading tips. The device was successfully made and can be used for at least one week. 2.4.7 Microfluidic diffusional sizing experiments The devices were filled with buffer using a 1 mL glass syringe (Hamilton, Switzerland) connected with 27 gauge needle using portex tubing. Proteins and buffer were filtered by 0.22 µm syringe filter (Millipore) immediately before experiments, which can reduce the chance for device clogging. The flow was withdrawal through the outlet using a glass syringe (Hamilton, Switzer- land) and neMESYS syringe pumps (Cetoni GmbH,Korbussen, Germany) to actually con- 44 On-chip measurements of proteins reacting with small molecules trol flow rate through the channels. Different flows rates were used depend on device height and design. At the beginning, 20 µL fluid with the 300 µL/h flow rate was withdrawn to eliminate potential inlet cross flow from the loading step. After that, a flow rate of 33.3 µL/h was used and allowed to equilibrate for approximately 18 minutes before image acquisition. Images are illuminated with 365 nm Cairn OptoLED (Photometrics) equipped with a Chromo 49000 DAPI filter and an Evolve 512 EMCCD camera for the fluorescence images. The range of OPA-labelled (49000 filter) excitation and emission wavelength was 433 - 485 nm. The 10X objectives was used, exposure times of images between 500 ms to 2 s are selected depending on the sample fluorescence intensity, and 60 images are averaged during each acquisition. For each set of measurements, a background image was taken into account for the minimal fluorescence of the unreacted dye, and a flatfield background image was also acquired. Measurements are taken in dark environment and temperature was controlled at 25 ◦C. 2.4.8 Image analysis Images at different positions of the microfluidic sizing device were acquired under fluores- cent microscope (Figure 2.1 b-e). The Y-junction is the region where the diffused proteins is labelled with OPA, along the labelling channel. OPA interacts with protein primary amine to generate substituted isoindole, which can be detected in the blue region of the spectrum. The T-junction indicates labelled protein molecules (i.e. fluorescence intensity) in contact with unlabelled protein molecule (i.e. no fluorescence intensity). The interface is at the center of the channel. If the flow profile has not shown as Figure 2.1e, it might be caused by clogging or fabrication abnormalities. If the images looks like Figure 2.1b, c and e, then image analysis can proceed (Figure 2.1d). Images of the observation region were analysed using ImageJ. The background was subtracted from each acquired image, which can reduce the effect of sample absorption 2.4 Materials and Methods 45 onto the PDMS. Quantitative labelling ensures that absolute protein concentration can be determined from measurement of fluorescence intensity within the observation region. The fluorescence intensity ratio is calculated based on Equation 2.6. [φ ] = γ1 − γd γ2 − γd (2.6) where φ is the fluorescence intensity ratio, γ1 is the fluorescence intensity observed in the observation region in the case of the protein and buffer loaded into the protein inlet and buffer inlet respectively. γ2 is the fluorescence intensity observed in the observation region in the case of the protein loaded into both protein inlet and buffer inlet. γd is the background corrected fluorescence intensity observed in the observation region in the case of buffer loaded into both protein inlet and buffer inlet. 2.4.9 Circular Dichroism experiment and data analysis Samples for the CD experiment were prepared by incubating 3 µM BSA in 5 mM sodium phosphate buffer with different pH values at room temperature for at least 2 h. Far-UV (spectra between 250 nm to 200 nm) were recorded on JASCO J-810 equipped with a Peltier thermally controlled cuvette holder at 25 ◦C. A 1 mm path length quartz cuvette was used, and CD spectra were collected by averaging five individual recordings with the data pitch 0.5 nm, bandwidth 1 nm, scanning speed 50 nm/min and response time 1 s. Each CD measurement was repeated three times. The data is converted to the molar ellipticity ([θ ]) in deg cm2dmol−1 by using the formula [88] [θ ] = millidegrees×mean residue weight pathlength in millimeters× concentration in mg/ml (2.7) 46 On-chip measurements of proteins reacting with small molecules 2.5 Appendix Table 2.1 The Rh of BSA in buffer with different pH. Buffer pH R h of BSA (nm) 1.2 8.40 ± 0.16 2.0 7.73 ± 0.21 2.5 6.25 ± 0.20 2.7 5.93 ± 0.31 3.3 5.17 ± 0.90 3.7 4.66 ± 0.40 4.3 3.30 ± 0.52 5.0 3.31± 0.62 6.0 3.78 ± 0.18 7.0 3.60 ± 0.41 8.0 3.80 ± 0.26 9.0 3.50 ± 0.37 10.2 3.34 ± 0.42 2.5 Appendix 47 Table 2.2 The Rh and the number of residues of folded and unfolded proteins. Proteins R h (Å) of folded proteins R h (Å) of unfolded proteins Number of residues Desaturation conditions BSA 36 ± 4.1 84 ± 1.6 589 pH 1.2 bovine ubiquitin [112] 13.2 26.3 76 5M urea horse ferricytochrome c [116] 13.5 ± 0.1 32.4 ± 1.6 104 4M GuHCl horse ferricytochrome c [53] 13.5 ± 0.1 46 ± 0.5 104 pH 2.3 staphylococcal nuclease [73] 15.9 ± 0.2 35 149 5M urea horse myoglobin [117] 17.5 ± 0.1 35.8 ± 1 153 in GuHCl bovine carbonic anhydrase B [204] 19 ± 2 59 ± 2 259 4.5M GuHCl yeast phosphoglycerate kinase [184] 23.4± 0.2 84± 1.6 589 4M GuHCl 48 On-chip measurements of proteins reacting with small molecules Model for pH-induced protein unfolding Dr. Liu Hong kindly helped to write the model and fit the data. The acid induced unfolding of a protein molecule driven by the protonation of a single residue with an ionisable side group can be captured through the thermodynamic cycle [6, 238] N Ka↼−−−⇁ +H NH KN ↿⇂ K˜N ↿⇂ U K˜a↼−−−⇁ +H UH (2.8) in which H, U, N, UH and NH represent protons, unfolded and folded states, the protonated unfolded and folded states of BSA respectively. KN , K˜N are the equilibrium constants for protein folding and unfolding, and pKa, pK˜a are the logarithm acid dissociation constants for folded and unfolded BSA. In the equilibrium state, we have KN = [N] [U] , Ka = [H+][N] [NH] , K˜a = [H+][U] [UH] , K˜N = [NH] [UH] (2.9) fN 1− fN = [N]+ [NH] [U]+ [UH] = [N] [U] ( 1+[H+]/Ka 1+[H+]/K˜a ) = KN ( 1+10pKa−pH 1+10pK˜a−pH ) (2.10) where we have used the definition of pH and pKa values, pH=− log10[H+], pKa =− log10 Ka and pK˜a =− log10 K˜a. Assuming the titration sites are independent, this model can be generalised to include 2.5 Appendix 49 multiple titration sites: fN 1− fN = ∑mi=0[[NH]i] ∑mi=0[[UH]i] (2.11) = [N] [U] ( 1+[H+]/K(1)a 1+[H+]/K˜(1)a )( 1+[H+]/K(2)a 1+[H+]/K˜(2)a ) · · · ( 1+[H+]/K(m)a 1+[H+]/K˜(m)a ) = KN ( 1+10pK (1) a −pH 1+10pK˜ (1) a −pH )( 1+10pK (2) a −pH 1+10pK˜ (2) a −pH ) · · · ( 1+10pK (m) a −pH 1+10pK˜ (m) a −pH ) (2.12) where UHi and NHi denote the concentrations of protonated unfolded and folded BSA with i protons. pKia and pK˜ i a are the logarithm equilibrium constants for the deprotonation of titra- tion site i in folded and unfolded BSA respectively. pKia =− log10 Kia and pK˜ia =− log10 K˜ia are defined accordingly. As a consequence, the difference of free energy between folded and unfolded states under different pH conditions is △G(pH) =△G⊖− kBT m ∑ i=1 ln ( 1+10pK (i) a −pH 1+10pK˜ (i) a −pH ) , (2.13) where △G⊖ =−kBTlnKN is the free energy of folding at pH=7. 50 On-chip measurements of proteins reacting with small molecules a 0 1 2 3 4 5 6 7 8 9 pH value 10 11A ve ra ge h yd ro dy na m ic ra di us (n m ) 3 4 5 6 7 8 9 Microfluidic data Eq.2.10 (m=1) Eq.2.10 (m=2) b pH value 10 11 N or m al is ed fr ac tio n of α -h el ix fa U (f +f aN ) /2 fa N 0 1 2 3 4 5 6 7 8 9 aU 222nm 208nm area Eq.2.10 (m=1) Eq.2.10 (m=2) Figure 2.17 Data fitting of measured Rh and fraction of α-helix of BSA. (a): Rh value of BSA in buffer solutions with varying pH values, which were measured by the microfluidic diffusional sizing device. Fitting was performed with Equation 2.10 and linear interpolation Rh = (Rmaxh −Rminh ) · (1− fN)+Rminh , values of Rmaxh = 8.4 nm and Rminh = 3.5 nm were determined directly from the data. (b): The renormalised fraction of α-helical derived from the molar ellipticity at 222 nm, 208 nm and total area were calculated using Equation 2.10. Predictions based on one titration site, Histidine241, (His241) and two other titration sites (here we tried any two combinations of Asp, Glu and His) are performed according to Equation 2.10 and fα=fUα+(f N α -f U α ) ·fN, fUα and fNα are the fraction of α-helix in the unfolded and native proteins. Chapter 3 Top-down protein identification with a microfluidic device This chapter is based on the publication: Pavan K. Challa; Quentin Peter; Maya A. Wright; Yuewen Zhang; Kadi L. Saar; Jacqueline A. Carozza; Justin L. P. Benesch. and Tuomas P. J. Knowles; Real-Time Intrinsic Fluorescence Visualization and Sizing of Proteins and Protein Complexes in Microfluidic Devices. Analytical Chemistry, 90.6 (2018), 3849-3855. and the manuscript: Yuewen Zhang; Maya A. Wright; Pavan K. Challa; Kadi L. Saar; Sean Devenish; Emma V. Yates; Quentin Peter; Christopher M. Dobson and Tuomas P. J. Knowles; Multidimensional Top-down Protein Identification. (2018), In preparation. Abstract Protein identification and profiling are critical for the advancement of cell and molecular bi- ology as well as medical diagnostics. Although mass spectrometry and protein microarrays are commonly used for protein identification, both methods require extensive experimental steps and long data analysis time. Herein, I present a microfluidic top-down protein identifi- cation platform, which gives direct read-outs of the key amino acids of proteins within min- 52 Top-down protein identification with a microfluidic device utes. Multidimensional physical parameters of proteins, including hydrodynamic radius, fluorescence intensities of tryptophan (Trp), tyrosine (Tyr) and lysine (Lys) residues, are obtained using a combined design of microfluidic diffusional sizing, detection of intrinsic fluorescence intensities and on-chip fluorescence labelling. I thereby achieved identification of proteins on a single microfluidic device. The results have show significant implications for the development of easy and rapid platforms to use for native protein identification. 3.1 Introduction Protein identification is fundamental for biological and medical research [48]. It provides key information in identifying the specific species responsible for human disease, thus helps to develop novel disease biomarkers, treatments in personalised medicine and targeted ther- apeutics, leading to the rise of new generations of pharmaceuticals [42, 144, 175, 265]. It is, therefore, of critical importance to develop protein identification methods that al- low proteins to be characterised in their native states with high accuracy and efficiency [51, 81, 114, 166, 180, 220]. Protein can be identified via a ‘bottom-up’ approach, in which proteins are digested into fragments before analysis [110]. Typically, the progress of ‘bottom-up’ identification in- volves solution proteolysis of a complex mixture of proteins, followed by chromatographic separation of peptides prior to tandem mass spectrometry sequencing. Proteins are then identified in the gas phase based on the mass of the fragments, and subsequent comparison to predetermined sequence databases [2, 75, 150, 188, 237]. However, due to the complexity of data collection, only 8% to 25% useful peptide information can be obtained. A ‘top-down’approach is another way to achieve protein identification where intact pro- tein is analysed [110]. A widely-used technique for ‘top-down’ protein identification is pro- tein microarrays, which target proteins using specific analyte reagents such as antibodies, allowing quantitative information on the species present to be obtained [133, 147]. Pro- 3.1 Introduction 53 teins can then be detected based on fluorescent labelling, chemiluminescence or radioac- tivity [225]. This technique, however, requires surface immobilisation and fluorescent la- belling [118, 182]. Another technique for ‘top-down’ protein identification is based on two-dimensional gel electrophoresis (2DE), in which proteins can be separated due to dif- ferent mass and isoelectric point. Then targeted proteins are extracted and analysed by mass spectrometry [75, 150, 188, 237]. However, 2DE is not applicable to separate proteins with molecular weight larger than 120 kDa or those with low solubility in buffer. Due to there limitations, less than 10% of the mammalian proteome is accessible [190]. In addition, currently available ‘top-down’ protein identification techniques generally require extensive sample preparation steps and long experimental analysis time.a BSAH. Transferrin Ubiquitin Al. DehydrogenaseGlucose Oxidase α-lac β-lac Ovalbumin Thyroglobulinβ-caseine Proteins ? Rh Lys Trp Tyr b On-chip multidimensional measurements Figure 3.1 Top-down protein identification within a single microfluidic device (a): The crys- tal structure of selected proteins shown in ribbon structures. The protein sequences were downloaded from Protein Data Bank [1], and the 3D structures were reproduced by Pymol. (b): Four sets of physical parameters of selected proteins (Rh, intrinsic fluorescence inten- sities of Trp and Tyr, as well as Lys fluorescence intensity via latent labelling) are detected in a single microfluidic device. In this chapter, I propose a proof-of-principle microfluidic protein-identification tech- nique that combines multidimensional data from measuring the physical parameters of pro- teins in a single microfluidic device. To achieve protein identification, four sets of infor- mation on physical parameters of selected proteins are detected. Hydrodynamic radius (Rh) 54 Top-down protein identification with a microfluidic device of the proteins is firstly measured based on the microfluidic diffusional sizing technique. Then, two-sets information on the intrinsic fluorescence intensities of tryptophan (Trp) and tyrosine (Tyr) residues within the proteins is detected. Finally the fluorescence intensity of lysine (Lys) residues labelled with a latent fluorophore is measured (Figure 3.1b). In this study, each selected protein (Figure 3.1a and Table 3.3 in the appendix section) has a unique combination of these four parameters (Rh, fluorescence intensities of tryptophan, tyrosine and lysine), allowing us to identify proteins by mapping the ratio of the obtained parameters in multidimensional space. 3.2 Results and Discussion 3.2.1 Feasibility test of a home-built intrinsic fluorescence microscope The intrinsic fluorescence microscope used in this study was built by my colleague, Dr. Pavan K. Challa [41]. I helped to test the feasibility of the intrinsic fluorescence micro- scope by measuring hydrodynamic radii of various proteins that have different numbers of tryptophan and tyrosine residues. The photograph of the intrinsic fluorescence microscope is shown in Figure 3.2a and a schematic of the optical layout is shown in Figure 3.2c. Light from a 280 nm LED (Thor- labs M280L3) and 365 nm LED (Thorlabs M365L2) are individually selected using a flip mirror depending upon the type of amino acid residues being imaged. The light is passed through an aspherical lens of focal length 20 mm to get a nearly collimated output beam. The beam is then directed onto a quartz cube lens, which consists of an excitation filter, dichroic mirror and an emission filter. The deep UV LED is able to excite protein intrinsic fluorescence intensities based on aromatic amino acids, including tyrosine (Tyr) and tryp- tophan (Trp), which allow proteins to be measured in their native states without adding extra labelling dye (Table 3.4, in the appendix section). Consequently, the emission filters 3.2 Results and Discussion 55 can be applied to selectivly detect fluorescence intensity either from Tyr (305 nm) or Trp (350 nm) (Figure 3.2). Images of the relative fluorescence intensities of Tyr and Trp in 30 µM BSA are shown in Figure 3.2 d and e, respectively. The effective brightness of the intrinsic fluorescence intensities of tryptophan, tyrosine and phenylalanine are determined by their excitation coefficients and quantum yield, and detected as the number of emitted photons under known illumination power at 280 nm. Comparing the effective brightness of tryptophan and tyrosine, the brightness of phenylalanine is negligible (Table 3.4, in the appendix section) [193]. ba Camera Quartz tube lens Quartz Dichroic mirror 280 nm LED365 nm LED Quartz objective c Trp Tyr d e f Quartz Slide Mirror Quartz Flip Mirror Quartz Lens Quartz Lens Microfluidic Device Quartz Objective EMCCD Camera 280 nm LED 365 nm LED Wavelength / nm Wavelength / nm N or m al is ed E m is si on /A U E xt in ct io n C oe ffi ci en t / M -1 c m -1 Figure 3.2 (a): Photograph of the home-built intrinsic fluorescence microscope used for vi- sualisation of proteins in this work. Two light sources (280 nm and 365 nm) and emission filters built into the microscope can be easily switched. (b): Image of black PDMS microflu- idic device bonded to a quartz slide. Buffer and sample are filled in gel loading tips and the outlet is connected with polythene tubing to a syringe pump. (c): Schematic illustration of the intrinsic fluorescence microscope. (d) and (e): Tyr and Trp intrinsic fluorescence intensity image of 30 µM BSA. (f): Absorption (top) and emission (bottom) spectra of tryptophan (blue), tyrosine (red) and phenylalanine (green). Images of Tyr and Trp acquired under intrinsic fluorescence microscope can be achieved by switching the emission filters between 305 nm and 350 nm. The figure (f) is reproduced from [193]. It is reported that UV lights may induce photochemical reactions, and cause protein structure change, as well as malfunction of proteins. However, most of the experiments in 56 Top-down protein identification with a microfluidic device the literature are done with high-power UV lamps (100 W) or strong UV lasers [49, 50, 199]. The LED used in this study is a Thorlabs 280 nm LED with a power of 25 mW. Moreover, the LED lamp is only turned on during image acquisition, which is around 5 -10 s. Thus, the effect of UV irradiation from our LED lamp on proteins is negligible. The intrinsic fluorescence microscope (Figure 3.2) was firstly tested with a microfluidic diffusion sizing device reported previously by my colleague, Dr. Paolo Arosio [10] on mea- surement of proteins with different numbers of Trp and Tyr (Table 3.1). The diffusion pro- files of each protein are detected based on their intrinsic fluorescence intensity. By analysing the diffusion profiles (see details later), the diffusion coefficient of analyte molecules can be calculated, and thus the Rh of each protein is obtained based on the Stokes-Einstein equa- tion. Measured Rh values for selected proteins are shown in the Table 3.1, which are in good agreement with literature values. Table 3.1 Summary of tested proteins under the intrinsic fluorescence microscope Proteins No. of Trp No. of Tyr Rh ∗ Literature value BSA 2 21 3.52 ± 0.3 nm 3.39 ± 0.27 nm [74] α-synuclein 0 4 3.05 ± 0.2 nm 2.66 ± 0.05 nm [157] Lysozyme 6 3 1.49 ± 0.11 nm 2.05 nm [245] ∗ Rh is measured with the diffusional microfluidic device. All the measurements were done three times and error bars represent the standard deviation among independent replicates. The home-built intrinsic fluorescence microscope proved accurate and effective to mea- sure protein properties in their native state. This unique advantage of the home-built mi- croscope can be utilised as a detecting technique for protein identification. In the following sections, microfluidic techniques are introduced for top-down protein identification together with the home-built intrinsic fluorescence microscope. 3.2 Results and Discussion 57 3.2.2 Design of microfluidic devices for top-down protein identification Original designs of top-down protein identification devices A microfluidic device was designed to identify proteins as shown in Figure 3.3a. The mi- crofluidic device consists of a traditional diffusional sizing module with 12 detection re- gions [10] in parallel with a dye inlet and mixing chamber. The dye solution consists of ortho-phthalaldehyde (OPA), reducing regent β -mercaptoethanol (BME), and sodium do- decyl sulfate (SDS) at pH 10.5 [257], which interacts with primary amine groups within the protein’s lysine residues to form isoindole rings with fluorescence at 365 nm (details in Sec- tion 2.4.3) [20, 194, 211, 257]. The labelling dye mixed with protein via lateral diffusion from one side, with a mixing time of at least 3 s at a flow rate of 200 µL/h and channel height of 50 µm. Therefore, the fluorescence intensity of proteins with lysine residues can be detected. I tested this device under flow rate at 200 µL/h and found that air bubbles formed at the outlet. The reason is that the pressure drop across the device inlets and outlet is too high for stable operation. The device was then adjusted to have wider labelling channels, and the OPA mixing chamber was re-designed (Figure 3.3b), so that dye would diffuse into the protein sample from both sides allowing a decrease in the length of the mixing channels. This design decreased the pressure drop across the device inlets and outlet significantly and allowed stable operation of the device at 200 µL/h for at least an hour and was then even applicable at higher flow rates of 400 or 500 µL/h. Standard solutions of 10 µM L-Tryptophan and 1 µM 4-methylumbelliferone (4MU) were chosen to calibrate the light wavelengths of 280 nm and 365 nm, respectively. Before the measurement of tryptophan, tyrosine and lysine-OPA fluorescence intensities, standard solutions were imaged in the PDMS channels (Figure 3.3b, calibration channel) to calibrate fluctuations in the LED brightness from day to day. Fluorescence intensities of the stan- dard calibration solutions were imaged on each single microfluidic device in the detection 58 Top-down protein identification with a microfluidic device regions of the calibration channels, and later used to normalise the measured fluorescence intensities. a b Figure 3.3 Different designs of microfluidic devices for top-down protein identification. (a): Initial design with OPA dye mixing from one side. (b): Second design with OPA mixing from both sides allowing for a shorter mixing channel and lower resistance. Diffusion pro- files are imaged at 12 points along the diffusion channels labelled in red and the relative fluorescence intensities are imaged in their respective detection regions by changing light sources or emission filters. Final design of top-down protein identification device The final design of top-down protein identification device is shown in Figure 3.4, containing three major regions. The region labelled with yellow is for detection of the intrinsic fluores- cence intensity of proteins with Trp and Tyr. The blue region contains a diffusional sizing module for measurement of hydrodynamic radius of proteins. The pink area is to obtain the fluorescence intensity of Lys after latent labelling. 3.2 Results and Discussion 59 Tyr & Trp Detection Region Sizing Region Lys Detection Region Buffer Inlet Sample Inlet OPA Inlet Outlet Tyr Intensity Lys IntensityTrp Intensity Sizing a b Figure 3.4 (a): Final design of microfluidic device for top-down protein identification. Rh of proteins is measured in the sizing region (highlighted in blue) and the related intrinsic fluorescence intensities of tryptophan (Trp) and tyrosine (Tyr) are imaged under the intrinsic fluorescence microscope excitation at 280 nm with different filters (Trp: 350 nm and Tyr: 305 nm) (highlighted in yellow). The light source is then switched to 365 nm excitation and the OPA fluorescence intensity coming from the protein lysine residues (Lys) conjugated to OPA dye is imaged in the Lys detection region (highlighted in pink). (b): Fluorescent images of Tyr, Trp and Lys, and diffusional sizing image were acquired under the home- built intrinsic fluorescence microscope. One advantage of this design compared to the two original designs described in the above section is that only one image rather than 12 images is required to be taken in the sizing region (Figure 3.4, highlight in blue). This change is extremely important as it is time-efficient and avoids changing of the focus of images when the stage of the microscope is manually moved. Moreover, the calibration channel is removed in this design, because a background subtraction algorithm is applied. The algorithm is based on a recently pub- 60 Top-down protein identification with a microfluidic device lished paper in our group [41], in which the LED brightness, inhomogeneities of the black carbon nanopowder, as well as the different of angles and scale between the background and detected protein fluorescence intensity are taken into account. 3.2.3 Top-down measurements of protein physical characteristics The protein identification device designed in this study consists of a diffusional sizing mod- ule, allowing the protein Rh to be determined in the solution phase on a single chip (Fig- ure 3.4a). Protein, buffer and OPA labelling dye are loaded to their respective inlets, and the flow is controlled by the application of negative pressure at the outlet with a syringe pump. The Rh of proteins and intrinsic fluorescence intensities of Trp and Tyr are detected using the intrinsic fluorescence microscope (Figure 3.2c) fitted with a deep UV-LED at excitation of 280 nm, as described in the previous sections [41]. a b c Position (µm) N or m al is ed a m pl itu de Fits Profiles 0 50 100 150 200 250 300 350 400 0.00 0.01 0.02 0.03 0.04 Figure 3.5 Diffusion sizing measurements of different proteins. (a): Diffusion profiles of thyroglobulin at different positions along the diffusional channel. (b): Diffusion profiles (blue) are processed from images and fitted to simulated basis functions (orange). (c): Plot of measured Rh by microfluidic device and reported Rh from literature of different proteins with various molecular weight. 3.2 Results and Discussion 61 I firstly used the intrinsic fluorescence microscope fitted with a tryptophan emission filter at 350 nm to size the native proteins by imaging the extent of their lateral diffusion into auxiliary buffer streams under steady laminar flow (sizing region, Figure 3.4a). The diffusion image (Figure 3.5a) was processed to the diffusion profile and fitted to a linear combination of simulated basis functions for particles of known radii [159] (Figure 3.5b), yielding the diffusion coefficient of the selected protein. Thus, the Rh of each protein is calculated based on the Stokes-Einstein equation. The measured Rh values of the tested proteins in this study, varying by three orders of magnitude in molecular weight, are con- sistent with the values reported in the literature (Figure 3.5c and Table 3.5 in the appendix section). The final design of the top-down protein identification microfluidic device also includes a Trp and Tyr detection region, allowing the detection of intrinsic fluorescence intensities of proteins (Figure 3.4a). The characteristic Trp fluorescence intensity (emission filter at 350 nm) of each protein at the resistance channel (Trp detection region, Figure 3.4a) is imaged. After that, the emission filter is changed to 305 nm to measure the related Tyr fluorescence intensity for proteins in their native state (Tyr detection region, Figure 3.4a). The intensities were normalised using a local background correction algorithm, which is described in detail in Section 3.4.6. Moreover, the microfluidic device has a latent labelling module, allowing measurement of lysines after OPA labelling (Figure 3.4a). At the downstream of the sizing region, an on- chip latent labelling strategy is used, described previously to conjugate the protein’s lysine residues to OPA dye molecules (Lys detection region, Figure 3.4a) [257, 261]. Conse- quently, the characteristic fluorescence intensity from lysine residues is imaged by switch- ing the UV-LED light source at 280 nm wavelength to an OPA-LED light source at 365 nm wavelength. I verified that the relative fluorescence intensities of the OPA labelled lysine residues measured on the microfluidic device are in a linear relationship with the concentra- 62 Top-down protein identification with a microfluidic device tion of lysine residues across the range of different proteins tested in this study (Figure 3.6). Figure 3.6 Fluorescence intensities of lysine residues from each protein measured by mi- crofluidic device: plot of lysine fluorescence intensities versus the relative lysine concentra- tion of different proteins. 3.2.4 Protein identification in a single microfluidic device The images acquired for each protein under the intrinsic fluorescence microscope were anal- ysed by using the background subtraction method described in the previous sections. The Python script for background subtraction was written by my colleague, Quentin Peter. In brief, I firstly took a background image of all the detection regions at the same position under the intrinsic fluorescence microscope with only buffer filled in the microfluidic de- vice. Then, the fluorescence images were taken after flowing sample and buffer into the device. The Python script was then used to subtract background and calculate the relative fluorescence intensity, as described in Section 3.4.7. To perform multidimensional data analysis for protein identification, variations of pro- tein concentrations across the different proteins were taken into account. In order to do so, 3.2 Results and Discussion 63 the raw fluorescence intensities obtained from the background subtraction algorithm were converted into two fluorescence intensity ratios: Trp/Lys and Tyr/Lys to normalise of dif- ferent protein concentrations and to define the identification axes. The ratios were chosen as the identification axes arbitrarily, and any combinations of the intensity ratio showed the same power for protein identification. The protein concentration independent parameter Rh was used as another axis to map the position of each of the ten proteins in 3D space. The results are summarised in Table 3.6, in the appendix section. c d Ovalbumin Al. dehydrogenase H. transferrin Thyrogloblin β-casine Ubiqutin BSA β-lac Glucose Oxidase ɑ-lac Ty r/L ys 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Rh (n m)21 3 4 5 6 7 8 0 1 2 3 4 Trp/Lys Trp/Lys 0.0 0.5 1.0 Ty r/L ys a Tr p/ Ly s Rh (nm) 2 3 4 5 6 7 8 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 b Rh (nm) 2 3 4 5 6 7 8 Ty r/L ys 0.0 0.2 0.4 0.6 0.8 1.0 Figure 3.7 Multidimensional data analysis: Protein identification based on the intensity ratio of Trp/Lys versus Rh (a), Tyr/Lys versus Rh (b) and Tyr/Lys versus Trp/Lys (c). (d), Protein identification in three dimensional space via Tyr/Lys, Trp/Lys versus Rh. Generally, each protein was measured with four independent repeats. Three out of four data points were used to define the protein area. For each of the protein areas along each of the axis (Rh, Trp/Lys and Tyr/Lys), the center of the area was plotted as the average value (µ) of three out of four measurements and the standard deviations (σ ) along each of the three axes were calculated to plot the ellipse area (Figure 3.7). Then, the fourth data point as a test point was used to calculate the probability of protein identification. The data point 64 Top-down protein identification with a microfluidic device was randomly chosen by bootstrapping analysis. The data analysis is based on the model of each protein as a standard normal distribution and assigned the test point as a z-score. The z-scores for each test point relative to each protein along each axis was calculated based on standard normal distributions N (µ , σ ). z = x−µ σ (3.1) where x is the test point, µ is the mean of the population and σ is the standard deviation of the population. Glucose Oxidase BSA β-la c β-ca sein α-la c Ova lbum in H. t rans ferr in Al. deh ydr oge nas e Thy rog lobu lin Ubi quit in Glu cos e O xida se BSA β-lac β-casein α-lac Ovalbumin H. transferrin Al. dehydrogenase Thyroglobulin Ubiquitin 0.0 0.2 0.4 0.6 0.8 1.0 Figure 3.8 The identification probability of 10 proteins with biological interest is indicated colorimetrically, with yellow corresponding to a high probability and blue corresponding to a low probability. The proteins in the database are plotted on the y-axis, and the x-axis shows the proteins tested against the database. Then, the z-score for each axis was converted to probability, and these probabilities multiplied together to obtain the overall probability of each test point falling within the defined protein ellipse area. The probabilities are shown in Table 3.7 in the appendix section, and the highest probabilities are highlighted with bold blue font. Finally, a heat map is plotted to illustrate the protein identification probability, which is indicated colorimetrically, 3.2 Results and Discussion 65 ranging from lowest (0.0, dark purple) to highest (1.0, bright yellow) (Figure 3.8). It clearly indicates that all the proteins have higher than 70% accuracy falling within the accurate protein area. Thus, based on the identification probability, all the ten proteins can be correctly identified. In other words, proteins can be identified based on their physical characteristics, including Rh, intrinsic fluorescence intensities of tryptophan and tyrosine, as well as lysine content, in a single step using a microfluidic device in a top-down manner. 3.2.5 Application of top-down protein identification technique The protein identification microfluidic device demonstrates that ten proteins can be suc- cessfully identified by their unique combination of Rh and relative amino acid fluorescence intensities. Preliminary experiments have performed by exploring the application of this technique. αB-crystallin is measured of a small heat shock protein family and functions as a molecular chaperone that primarily binds to misfolded proteins to prevent protein aggre- gation. Defects in this protein have been associated with neurodegenerative diseases such as Alzheimer’s disease and Parkinson’s disease [29, 64, 103]. Titin, as a client protein can interact with αB-crystallin in its native state and regulate its functions [233]. The Rh and relative fluorescence intensities of αB-crystallin and titin were measured by the top-down protein identification microfluidic device individually. The value of Rh and fluorescence intensities of Trp, Try and Lys were significantly different between αB-crystallin and titin (Figure 3.9). An Akta was then coupled to our microfluidic platform, which allows protein peaks eluted from size exclusion columns to be loaded directly into the top-down protein identi- fication microfluidic device, with the aim of revealing the identity of each peak. However, there are some challenges, such as quantifying the number of measured amino acid residues of each protein on the microfluidic device. It could significantly improve the power and res- 66 Top-down protein identification with a microfluidic device olution of the microfluidic platform. Based on our dataset, only the lysine fluorescence in- tensity showed a linear relationship with the entire database of tested proteins (Figure 3.6c). The fluorescence intensity of tryptophan and tyrosine residues varies across different pro- teins, due to the strong dependence on local environment surrounding these residues. Thus, we cannot predict the content of tryptophan and tyrosine based on its intrinsic fluorescence intensity. R h ( nm ) Ty r/L ys Trp/Lys Titin αB-crystallin Figure 3.9 Identification of αB-crystallin and titin with the top-down protein identification microfluidic device. To address this challenge, together with Dr. Emma Yates, we conducted a simulation to calculate the accuracy of protein identification based on the measured parameters (flu- orescence intensities of Trp, Try, Lys, and Rh) using the top-down microfluidic platform. Firstly, the human proteome (20k protein) were downloaded and a database built. Then, the number of lysine, tryptophan, and tyrosine residues of each protein were determined, the molecule weight of each protein was calculated and the diffusion coefficient was mod- elled by assuming the proteins are spherical. The results for the theoretical calculations show that accuracy for protein identification in human proteome is 83 % by measuring the fluorescence intensity of Trp, Tyr, Lys, and Rh (Figure 3.10). 3.3 Conclusions 67 2 4 6 8 10 12 degeneracy 0 2000 4000 6000 8000 10000 12000 14000 # of o cc ur en ce s 83% of time, set of K-W-Y-Rh(0.1) refers to single protein in human proteome (20k), only 5% of time does it refer to 3+ proteins Figure 3.10 Analyse of how many sets of Lys-Trp-Tyr-Rh (K-W-Y-Rh) counts occurred 1, 2, 3, or n times within the human proteome. 83 % of counts occur only 1 time, meaning that they uniquely identify a human protein sequence. This result is promising because in practice, the practical identification frequency would actually be significantly lower than the theoretical one, when error is included in all dimen- sions. In addition, the protein will be identified from smaller subset (such as human blood serum) rather than the whole human genome. Thus, we should be able to get higher accu- racy for protein identification by employing this strategy. In the future, we will develop a quantitative method combined with the protein identification microfluidic device to predict the tryptophan and tyrosine content of each protein, and thus identify proteins in human genome by a single measurement on the protein identification microfluidic device. 3.3 Conclusions In summary, we have developed a novel top-down protein identification microfluidic device that includes Rh determination, intrinsic fluorescence detection of tryptophan and tyrosine 68 Top-down protein identification with a microfluidic device residues and latent labelling of lysine residues. This allows the protein to be unambigu- ously identified. Using this method, multidimensional information for a given protein is obtained in a single experiment. Ten proteins with varying molecular weights and amino acid content were correctly identified in their native states. In addition, physical parame- ters for αB-crystallin and its native client titin were measured to explore probing molecular chaperone-substrate interactions. In order to assess the potential of ‘blindly’ identifying pro- teins, simulations show that 83 % of proteins can be identified by measuring the absolute fluorescence intensities of Trp, Try, Lys, and Rh. In the future, a quantitative method will be developed that combines this approach with the protein identification microfluidic device to detect the content of Trp and Tyr. Thus, proteins or protein mixtures will be identified with a single measurement by measured content of Trp, Try and Lys, as well as hydrodynamic radius. This microfluidic based top-down protein identification has advantages of simplicity, low sample consumption, and identification on a single microfluidic chip in solution phase. It shows great potential to develop clinical diagnostic assays and to extend protein profiling for patient care. 3.4 Materials and Methods 3.4.1 Materials Proteins used in this chapter: bovine serum albumin (BSA), β -lactoglobulin (β -lac), glu- cose oxidase, α-lactalbumin (α-lac), ovalbumin, human transferrin, thyroglobulin, β -casein and ubiquitin were obtained from Sigma-Aldrich, and alcohol dehydrogenase was obtained from Alfa Aesar. Chemicals used in this chapter: sodium bicarbonate, sodium carbonate, sodium phosphate, β -Mercaptoethanol (BME), ortho-phthalaldehyde (OPA), sodium dode- cyl sulfate (SDS) and sodium carbonate were obtained from Sigma-Aldrich. All solutions were prepared using MilliQ water and filtered. 3.4 Materials and Methods 69 3.4.2 Sample preparation All the proteins with different concentrations were dissolved in 25 mM sodium phosphate buffer (pH 7). Each protein concentration was measured by NanoDrop UV-Vis spectropho- tometer (Thermo Scientific, Wilmington, DE, USA). The extinction coefficients (ε) of each protein are summarised in Table 3.2. Table 3.2 Absorbance of each protein at 280 nm. Proteins Absorbance at 280 nm (M−1cm−1) BSA 43,824 β -lac 17,335 β -casein 25,900 Glucose oxidase 96,845 α-lac 16,246 Ovalbumin 31,775 Alcohol dehydrogenase 31,815 Human transferrin 102,720 Thyroglobulin 588,235 Ubiquitin 1280 3.4.3 Fabrication of microfluidic devices The details of fabrication of microfluidic devices are in Section 2.4.6. In this chapter, 50 µm height microfluidic devices were fabricated, instead of the standard 25 µm height, to in- crease signal-to-noise ratio. The microfluidic devices were baked in PDMS with black carbon nanopowder and bonded to quartz slides to eliminate background fluorescence from PDMS and glass at deep-UV wavelength. 70 Top-down protein identification with a microfluidic device 3.4.4 Fluorescence intensity measurements by plate reader The fluorescence intensities were detected in a plate reader (Fluostar Omega, BMG Labtech, Germany). Each protein was dissolved in 25 mM phosphate buffer (pH 7) with a series of dilutions. The 96-well half-area, low binding, clear bottom and PEG coated plate (Corning 3881) was used 80 µL of each sample was pipetted into each well. Fluorescence spectral scans are measured with emission and excitation filter at 355 nm and 280 nm, respectively. 3.4.5 Top-down identification measurements The sample, buffer and labelling solution were loaded in inlets using gel loading pipette tips. Fluid flow through the channels was controlled by neMESYS syringe pumps accurately and precisely at 200 µL/h through the channels. After the flows stabilised, fluorescence intensi- ties were acquired along the diffusion channels at 0.1, 1, 2 and 4 cm positions from nozzle. Fluorescence intensities of tryptophan and tyrosine, as well as background were detected using home-build intrinsic fluorescence microscope (Figure 3.4), fitted with the appropri- ate emission filter (Tryptophan: 350 nm; Tyrosine: 305 nm), and excitation wavelength at 280 nm. In order to get the fluorescence intensity related to the protein’s lysine content, OPA dye solution was mixed with the protein on-chip via lateral diffusion from both sides, with a mixing time of at least 3 s at a flow rate of 200 µL/h to ensure complete labelling. The OPA fluorescence intensity and background were acquired under the OPA detection region using the home-build fluorescence microscope fitted with a 365 nm LED and corresponding dichroic filter set. 3.4.6 Image analysis All the images were acquired with the EMCCD camera fitted with the UV-LED light or OPA-LED light at wavelength of 280 nm or 365 nm, respectively. Different filters were fitted for different intrinsic fluorescence intensities (tryptophan or tyrosine) measurements. 3.4 Materials and Methods 71 For each set of measurements, backgrounds were taken to minimise fluorescence intensity from black PDMS. Background correction for each images were done using a sensitive background subtraction algorithm, published by our group recently [41]. a b c d e f Figure 3.11 Fluorescence intensity of tryptophan (a, b), tyrosine (c, d) and lysine (e, f) of 30 µM BSA before (top) and after (bottom) background subtraction. The examples in figure 3.11 show image background subtraction under different filters with different LED lights. Images were analysed by Python script, written by Quentin Peter. In brief, intensity was firstly fitted with a polynomial and the images were flattened. Then, different scale, angle and x-y offset between fluorescence image and background were detected and matched to get correct superposition. After that, background was subtracted from sample image to eliminate inhomogeneities from black carbon powder and variation of LED brightness. Finally, relevant useful data, fluorescence intensity can be extracted [41]. 72 Top-down protein identification with a microfluidic device 3.4.7 Diffusional sizing data analysis The diffusion coefficient of each sample was measured with the microfluidic diffusional module. The background and fluorescence image were acquired. After subtraction, fluo- rescence intensity was fitted to the simulated basis functions, published by our group [159] (Figure 3.5 a and b). Particles between 0.5 nm and 10 nm were simulated, which cor- responds to the diffusion coefficients. The simulation is based on the laminar poiseuille flow through the diffusion channel [159]. The diffusion profiles were processed to a 2- dimensional lateral fluorescence intensity scan, and fitted to a linear combination of simu- lated basis functions calculated for ideal monodisperse solutions [159]. The data analysis were performed with calibrated flow rate, device height, temperature and samples viscosity. All the device heights were measured by scanning the soft-lithography master with a pro- filometer (DektakXT, Bruker). Based on the simulation, the diffusion coefficient of proteins in the solution phase can be acquired [10, 52, 113]. Thus,the Rh of proteins were calculated based on the Stoke-Einstein equation. All the data analysis were performed by Python. 3.5 Appendix 73 3.5 Appendix Table 3.3 Detected proteins in this study Proteins Molecular weight (kDa) N. of Trp. N. of Tyr. N. of Lys. BSA 66 2 21 60 β -lactoglobulin 18 2 4 16 β -casein 24 1 4 12 Glucose oxidase 64 12 28 27 α-lactalbumin 16 4 4 12 Ovalbumin 42.8 4 10 20 Alcohol dehydrogenase 37 5 14 24 Transferrin 75 8 26 58 Thyroglobulin 660 45 66 91 Ubiquitin 8.5 2 4 1 Table 3.4 Effective brightness of the aromatic amino acids (table reproduced from [193]) Amino acid Extinction coeffi- cient (M−1 cm−1) Effective bright- ness at 280 nm Effective bright- ness at λ abs,max Quantum yield Tryptophan 5579 722 725 0.13 Tyrosine 1405 170 197 0.14 Phenylalanine 195 0.11 5 0.24 74 Top-down protein identification with a microfluidic device Table 3.5 Comparison of microfluidic measured Rh of different proteins with literature val- ues. Proteins Measured Rh (nm) Literature Rh (nm) References BSA 3.52 ± 0.30 3.39 [74] β -lactoglobulin 2.46 ± 0.11 2.58 [230] β -casein 3.88 ± 0.29 3.7 [152] Glucose oxidase 3.49 ± 0.25 4.3 [207] α-lactalbumin 1.74 ± 0.21 2.1 [245] Ovalbumin 2.94 ± 0.18 3.10 [107] Alcohol dehydrogenase 3.39 ± 0.07 4.55 [210] Transferrin 3.22 ± 0.29 3.72 [69] Thyroglobulin 8.12 ± 0.56 8.58 [65] Ubiquitin 2.06 ± 0.20 1.4 [76] 3.5 Appendix 75 Plate reader measurements of fluorescence intensity measurements Each protein has different intrinsic fluorescence intensities at a given concentration, be- cause their aromatic amino acid residues are contained in different physical environments [18, 82, 196]. To detect the relationship between protein concentration and fluorescence intensities, the fluorescence intensities of all the ten proteins under different concentrations were checked with a plate reader. Fluorescence spectral scans were measured with emission and excitation filter at 355 nm and 280 nm, respectively. From these measurements, the fluorescence intensity of each protein is in a linear relationship with protein concentration (Figure 3.12). Thus, the protein concentrations used for microfluidic measurements were chosen based on the obtained calibration curves. Proteins with different concentrations were measured in the microfluidic device in order to get sufficient fluorescence intensity within the linear limits determined by the plate reader measurements. Figure 3.12 The relationship between different protein concentrations and fluorescence in- tensities. Fluorescence intensity was detected by plate reader with the excitation and emis- sion at 280 nm and 355 nm respectively. 76 Top-down protein identification with a microfluidic device Table 3.6 Measured fluorescence intensity ratio and Rh for each protein Proteins Tyr/Lys Trp/Lys Rh (nm) BSA 0.424 0.081 3.48 0.522 0.113 3.60 0.341 0.08 3.13 0.422 0.09 3.60 β -lac 0.343 1.16 2.34 0.262 0.912 2.48 0.174 0.678 2.56 0.184 0.441 2.36 Glucose oxidase 0.631 3.329 3.74 0.669 2.750 3.24 0.420 4.288 3.50 0.551 2.666 3.98 α-lac 0.541 0.696 1.63 0.462 0.701 1.98 0.422 0.598 1.60 0.407 0.502 1.93 Ovalbumin 0.655 0.509 3.14 0.486 0.401 2.78 0.443 0.504 2.90 0.535 0.601 3.10 Al. dehydrogenase 0.953 1.546 3.40 1.010 0.998 3.32 0.801 1.278 3.45 0.907 1.280 3.36 Transferrin 1.077 0.256 3.20 0.687 0.185 3.24 0.809 0.233 2.92 0.965 0.202 2.84 Thyroglobulin 0.083 2.823 8.02 0.141 1.480 7.63 0.054 2.335 8.74 0.098 2.151 8.52 β -casein 0.719 0.776 3.77 0.790 0.449 4.22 1.235 0.970 3.66 1.10 0.926 3.82 Ubiquitin 0.711 2.365 3.25 0.546 1.975 2.85 0.771 1.485 3.09 0.626 1.496 2.88 3.5 Appendix 77 Ta bl e 3. 7 T he pr ot ei n id en tifi ca tio n pr ob ab ili tie s∗ Pr ob ab ili tie s m ul tip lie d (3 D ) B SA β -l ac G lu co se O xi da se α -l ac O va lb um in A l. de hy - dr og en as e Tr an sf er ri n T hy ro gl ob ul in β -c as in e U bi qu iti n Po in t1 1. 2 2. 6 14 .3 2. 4 1. 6 6. 4 4. 0 12 .3 5. 2 6. 1 Po in t2 2. 4 1. 5 14 2. 5 3. 0 10 .0 7. 9 12 .1 9. 4 7. 3 Po in t3 9. 0 6. 4 2. 0 6. 9 6. 9 5. 4 9. 4 8. 2 7. 3 2. 2 Po in t4 1. 9 1. 8 12 .0 1. 3 1. 6 6. 1 4. 6 14 .3 5. 8 4. 9 Po in t5 2. 1 3. 4 10 .5 2. 0 1. 4 2. 9 2. 1 13 .8 2. 5 3. 2 Po in t6 6. 1 7. 5 8. 2 4. 9 4. 0 1. 2 2. 5 16 .6 1. 6 2. 5 Po in t7 5. 4 8. 9 15 .5 5. 1 4. 0 2. 6 1. 4 22 .8 1. 8 5. 8 Po in t8 13 .3 11 .3 11 .9 15 .8 13 .8 17 .5 19 .8 1. 2 17 .7 12 .5 Po in t9 8. 4 11 .6 12 .4 7. 7 6. 2 1. 8 2. 6 21 .5 1. 7 5. 0 Po in t1 0 4. 1 3. 5 5. 7 2. 6 2. 5 2. 6 3. 8 11 .9 3. 3 1. 5 ∗ th e da ta is ca lc ul at ed ba se d on st an da rd no rm al di st ri bu tio n. A ll th e da ta ar e ca lc ul at ed in lo ga ri th m . T he da ta re pr es en ts th e pr ob ab ili tie s of pr ot ei n id en tifi ca tio n. T he hi gh es tp ro ba bi lit y is hi gh lig ht ed w ith bo ld bl ue fo nt . Chapter 4 Investigation of protein-protein interactions using microfluidic techniques This chapter is based on the manuscript: Yuewen Zhang; Therese W. Herling; Stefan Kreida; Quentin Peter; Tadas Kartanas; Susanna To¨rnroth-Horsefield; Sara Linse; Tuomas P. J. Knowles; On-chip study of membrane protein interactions. (2018), In preparation. Abstract Protein-protein interactions play a critical role in regulating human aquaporins (AQPs), which are the key membrane protein in the regulation of water homeostasis in cells. The structure and function of AQPs have been well studied. However, the interactions between full-length AQPs and small regulatory proteins are poorly understood. Here I demonstrate a rapid and quantitative microfluidic approach for studying the interactions between full- length AQPs and the regulatory protein, calmodulin (CaM), which mediates AQP0 gat- ing. By measuring the diffusion coefficients and electrophoretic mobilities of the individual 4.1 Introduction 79 components and the resulted complex, the binding equilibrium and effective charge of each component can be characterised. Our studies highlight that CaM binds to AQP0 in a non- cooperative manner. Moreover, the results show that CaM selectively binds to AQP0 rather than AQP2. 4.1 Introduction Membrane proteins play many critical roles in living cells by controlling signalling, trans- portation and molecular recognition. Membrane proteins constitute more than 60% of cur- rent drug targets, which are of particular interest because of their role in diseases [67, 105, 146, 170]. Therefore, membrane protein binding assays are becoming increasingly im- portant for understanding biomolecular functions, both in fundamental research and drug discovery applications. However, membrane proteins are particularly difficult to study be- cause they have large hydrophobic surfaces that can result in misfolding or aggregation in aqueous solutions [85]. Thus, detergents or lipids are widely used to solubilise membrane proteins for many applications [85]. Water transport is fundamental for organism physiology, which is achieved by aquapor- ins (AQPs), regulating membrane proteins. AQPs control water transport across cellular membranes along osmotic gradients. AQPs are a family of membrane proteins that conduct water and other small solute transport across cell membranes [187]. Structurally, AQP is a homo-tetramer with each monomer comprise six transmembrane α-helices surrounding a narrow water-conducting channel [231]. All cells depend on their ability to maintain water homeostasis, which is achieved via regulation by AQPs. There are multiple AQP isoforms and these can be found in most species, from bacteria to higher eukaryotes [125]. Thirteen AQPs (AQP0 - AQP12) have been identified in the human proteome, and each of them is expressed in different tissues. The AQP isoforms possess unique substrate permeability, responding to a diverse range of bioactivities, including urine concentration, cell migration, 80 Investigation of protein-protein interactions using microfluidic techniques brain oedema and adipocyte metabolism [132, 186]. AQP0 is expressed in the eye lens, where it constitutes more than 60% of the total mem- brane protein content in the fiber cells [27]. It is reported that AQP0 permeability is regu- lated by three independent mechanisms, including C-terminal cleavage [85], pH mediation and calmodulin (CaM) regulation [160, 161, 187]. CaM, small calcium regulated protein (17 kDa), functions as a ubiquitous messenger in several Ca2+ signalling pathways. CaM binds two Ca2+ ions at each of its N-terminal and C-terminal domains via EF-hands, thus exposing hydrophobic binding pockets, that can be used to bind a variety of target proteins, such as AQP0. Previous studies showed that CaM interacts with AQP0 in a Ca2+ dependent manner [187]. The treatment of retina disease is regulated by CaM inhibitors, which reduce the Ca2+ effects on AQP0 water permeability [160, 236]. 1st Calmodulin 2nd CaM Step 1 Step 2 AQP0 tetramer Extracellular Cytoplasm ? Figure 4.1 The crystal structure of AQP0 tetramer and CaM (PDB ID: 3J41) is shown as a ribbon structure in purple and blue respectively. Limited by the solubility of full-length AQP0, previous studies on interactions between 4.2 Results and Discussion 81 AQP0 and CaM are mainly fulfilled using AQP0 peptide, which provides useful informa- tion of binding locations and specific interacting residues, yet, has higher solubility in aque- ous solution. However, the AQP0 peptide does not provide the complete biological con- text, which may miss some information that affects binding affinity [71, 161, 186, 187]. Therefore, the interactions between full-length AQP0 and CaM are still poorly understood (Figure 4.1). Hence, a full understanding of how CaM interacts with full-length AQP0 is important. In this study, I demonstrate microfluidic free-flow assays to investigate the interactions between full-length AQPs and CaM. To quantify the interactions between full-length AQPs and CaM, the microfluidic diffusional and electrophoresis techniques are adopted to mea- sure the key physical parameters of individual components and the resulting complex, in- cluding diffusion coefficients and electrophoretic mobilities. The effective charge and bind- ing equilibrium can be characterised based on the measurement of these parameters. The results show that CaM binds to full-length AQP0 in a non-cooperative manner. In addition, it indicates that CaM is selectively binding to AQP0 rather than AQP2. 4.2 Results and Discussion 4.2.1 Hydrodynamic radius and interactions of CaM and full-length AQP0 In order to study full-length AQP interaction with CaM in solution phase in a non-disruption manner, the variation of diffusive coefficient was measured based on a microfluidic method (Figure 4.2). First, I investigated the protein binding affinity of AlexaFluor488-labelled calmodulin (CaM-Alexa488) as a function of the increased concentration of unlabelled tar- get protein (AQPs). 82 Investigation of protein-protein interactions using microfluidic techniques Detection Region Sample Inlet Buffer Inlet Outlet a) b) c) 0 200 400 800600 1000 1200 Position (μm) 0.000.02 0.040.06 0.080.10 0.12 1uM CaM-Alexa 488 N or m al is ed A m pl itu de Figure 4.2 (a): Schematic of the protein-sizing microfluidic device. The channel is 300 µm wide, 40,000 µm length and either 25 µm or 50 µm in height depending on the experiment. (b): Experimental fluorescence image of diffused CaM. (c): Fitting the simulated profiles (orange line) to the observed CaM diffusion profile (blue line, which is behind the orange line). Testing the protein-sizing microfluidic device Firstly, the Rh of 1 µM CaM-Alexa488 was measured in the solution phase under steady laminar flow conditions in the protein-sizing microfluidic device (Figure 4.2a). Different flow rates were set to detect the dependence of the Rh and flow rates (60 µL/h to 200 µL/h). AlexaFluor488 was used to label CaM for visualisation of the sample diffusion profiles un- der the microscope. Fluorescence images of CaM were taken in the diffusion region (Fig- ure 4.2b). The diffusion profiles (Figure 4.2c) were fitted to the simulated basic functions using known Rh particles [159]. The average diffusion coefficient of CaM was calculated, and then based on the Stokes-Einstein equation, the Rh of CaM can be calculated. The results show that the Rh of CaM is stable at 2.56 ± 0.18 nm, independent of flow rates (Fig- ure 4.3). The measured Rh of CaM is in agreement with literature values [172]. After that, the flow rate of 200 µL/h was used for all the microfluidic diffusional sizing measurements. 4.2 Results and Discussion 83 Figure 4.3 Rh measurements of CaM under different flow rates. Hydrodynamic radius of CaM and full-length AQP0 The diffusion coefficient of AlexaFluor488 labelled full-length AQP0 is then measured us- ing the protein-sizing microfluidic device (Figure 4.2a). The fluorescence images of the 1 µM full-length AQP0 were taken in the diffusion region, and the diffusion profiles were fitted to the simulated basis functions [159]. The Rh of full-length AQP0 was detected as 6.45 ± 0.34 nm, which is consistent with the literature value [72]. The Rh of unlabelled AQP0 was also measured using the intrinsic fluorescence microscope described in Chap- ter 3 [41]. The obtained Rh of 6.47 ± 0.28 nm agrees well with that measured using the protein-sizing microfluidic device (Figure 4.2a). As it is expected that CaM will have the largest relative change in size upon complex for- mation with AQP0, unlabelled AQP0 was titrated against a fixed concentration of Alexa488 labelled CaM (1 µM) in the buffer (10 mM Tris, 10 mM NaCl, 0.1 mM CaCl2, 0.03% DDM (n-Dodecyl β -D-maltoside) at pH 7.5). The Rh of the complexes were detected using the protein-sizing microfluidic device. As shown in Figure 4.4, the observed Rh of CaM and AQP0 complex gradually increased to a plateau value of 6.7 ± 1.13 nm upon addition of unlabelled AQP0. 84 Investigation of protein-protein interactions using microfluidic techniques Figure 4.4 The measured Rh of CaM with the increased concentration of AQP0 by the protein-sizing microfluidic device. The dashed line indicates a fit of Kd for the 2:1 non- cooperative interaction between CaM and AQP0. The shaded area covers a variation in Kd between three independent measurements. Kd of CaM and full-length AQP0 based on Rh The expression for the dissociation constant (Kd) for the interaction between full-length AQP0 and CaM is written as a function of concentration of isolated CaM (free CaM in solution, noted as C), single bound CaM (one CaM bound with AQP0, noted as S), and double bound CaM (two CaM bound with AQP0, noted as D). Then, combining the solved C, S or D from Equation 4.1, the Kd can be calculated according to three models described in the appendix section. Rhobs = Rhc C Ct +Rhs S Ct +Rhd D Ct (4.1) where Rhobs is the measured Rh; Rhc is the measured Rh of CaM alone; C is the concentration of isolated CaM; Rhs is the measured plateau value Rh of 1 µM CaM with excess AQP0; S 4.2 Results and Discussion 85 is the concentration of single bound CaM; Rhd is the measured plateau value Rh of 1 µM AQP0 with excess CaM; D is the concentration of double bound CaM; and Ct is the sum of C, S and D; For the 1:1 binding model, the Kd was calculated as 0.78 ± 0.1 µM. For the 2:1 non- cooperative binding model, the Kd was calculated as 2.4 ± 0.13 µM. Moreover, for the 2:1 sequential binding model, the first and the second Kd were calculated as 2 ± 0.14µM and 1.8 ± 0.21 M, respectively (Table 4.2, Entry 1). The high Kd of second binding indicates that the 2:1 sequential binding model is not applicable for full-length AQP0 and CaM inter- actions. Therefore, it suggests a non-cooperative binding manner between CaM and AQP0, if there is a 2:1 binding. 4.2.2 Electrophoretic mobility and interactions of labelled CaM and full-length AQP0 Electrophoretic mobility of labelled CaM and full-length AQP0 After detecting the Rh of the Alexa488 labelled CaM, AQP0 and their complex, the elec- trophoretic mobilities of the same sample sets were also measured in a free-flow elec- trophoresis microfluidic device to further understand the interactions between CaM and full-length AQP0. Flow through the electrophoresis devices was controlled by applying negative pressure at the outlet with a syringe pump, the flow rate for electrophoresis mea- surements was set to 500 µL/h. Proteins migrate perpendicularly to the direction of flow upon the application of a voltage (0 - 2 V, with 0.2 V intervals) between integrated InBiSn electrodes (Figure 4.5a). Fluorescence image of 1 µM CaM under 0 V and 2 V were ac- quired under fluorescence microscope (Figure 4.5a). The fluorescence signal of labelled proteins was recorded with applied voltage. Sample deflection (δ ) and current (I) were measured as a function of applied voltage (Figure 4.5b). By combining the buffer conduc- tance in a given device and the measured current, effective voltage drop can be calculated. 86 Investigation of protein-protein interactions using microfluidic techniques Then the electric field value is calculated based on the effective voltage drop and device ge- ometry. In addition, electrophoretic velocities were calculated based on the residence time (Figure 4.5c). By linear fitting to the slop of the sample drift velocity against the electric field across the solution, the electrophoretic mobilities were obtained. (details for the data analysis were described in Section 4.4.8.) Buffer Inlet Sample Inlet Electrodes 0V 2VDetectionRegion Outlet Fllow Direction Electri c Field Directi on a) b) c) Electric Field / V cm-1 Velocit y / μm/ s End De �l ec ti on / μ m Vlotage / V Figure 4.5 Electrophoresis microfluidic measurements. (a): Schematic of the free-flow elec- trophoresis microfluidic device [98]. Samples deflected perpendicularly to the flow direc- tion, and detection region is shown in blue. The electrophoresis channel is 2150 µm wide, 10,000 µm in length and either 25 µm or 50 µm in height depending on the experiment. Relative fluorescent images of 1 µM CaM at 0 V and 2 V were acquired. (b): With applied voltage, deflection of samples was measured. (c): Sample velocity against the electric field is plotted for the same data sets as in (b). By measuring the change of electrophoretic mobility (µobs) of CaM in the presence of AQP0, the binding equilibrium can be identified (Figure 4.6). Firstly, the µobs of CaM was measured with three independent repeats, -3.18 ± 0.03 ×10−8 m2 V−1 s−1, which is consistent with the literature value [100]. Also, based on the amino acid sequence of CaM, it 4.2 Results and Discussion 87 has 37 negatively-charged residues (aspartic acid and glutamic acid) and only 14 positively- charged residues (arginine and lysine) [80], indicating that CaM has a net negative charge. Then, the electrophoretic mobility of CaM-Alexa488 with titrated AQP0 was measured. The results show that it increases from - 3.2 ×10−8 to - 0.2 ×10−8 m2V−1s−1 and reaches a plateau (Figure 4.6). The two measured parameters (Rh and µobs) of AQP0 and CaM complex reach a plateau at the same concentration of AQP0 binding sites (8 µM). Figure 4.6 The measured electrophoretic mobility of CaM with titrated AQP0. The dashed line indicates a fit of Kd for the 2:1 non-cooperative interaction between CaM and AQP0. The shaded area covers a variation in Kd between three independent measurements. Kd of labelled CaM and full-length AQP0 based on electrophoretic mobility The expression for the dissociation constant (Kd) for the interaction between full-length AQP0 and labelled CaM is written as a function of concentration of isolated CaM (free CaM in solution, noted as C), single bound CaM (one CaM bound with AQP0, noted as S), and double bound CaM (two CaM bound with AQP0, noted as D). Then, combing the solved C, S or D from Equation 4.2, the Kd can be calculated according to three models 88 Investigation of protein-protein interactions using microfluidic techniques described in the appendix section. µobs = µC C Ct +µS S Ct +µD D Ct (4.2) where µobs is the measured electrophoretic mobility; µC is the measured electrophoretic mobility of CaM alone; C is the concentration of isolated CaM; µS is the measured plateau value electrophoretic mobility of 1 µM CaM with excess AQP0; S is the concentration of single bound CaM; µD is the measured plateau value electrophoretic mobility of 1 µM AQP0 with excess CaM; D is the concentration of double bound CaM; and Ct is the sum of C, S and D. For the 1:1 binding model, the Kd was calculated as 0.36 ± 0.06 µM. For the 2:1 non- cooperative binding model, the Kd was calculated as 1.1 ± 0.16 µM. Moreover, for the 2:1 sequential binding model, the first and the second Kd were calculated as 1.2 ± 0.11 µM and 4 ± 0.09M, respectively (Table 4.2, Entry 2). The high Kd of second binding indicates that the 2:1 sequential binding model is not applicable for full-length AQP0 and CaM in- teractions. Therefore, it also suggests a non-cooperative binding manner between CaM and AQP0, if there is a 2:1 binding. 4.2.3 Electrophoretic mobility and interactions of AQP0 and CaM by detecting intrinsic fluorescence intensity Electrophoretic mobility measurements by detecting intrinsic fluorescence intensity Recently, we reported a new approach for detecting intrinsic fluorescence intensities of pro- teins under a home-built intrinsic fluorescence microscope as described in Chapter 3 (Fig- ure 3.2) [41]. The deep UV LED (280 nm) is able to excite aromatic amino acid residues within proteins, including tryptophan and tyrosine, which allow detecting proteins in their native states without labelling proteins. The emission filters are able to switch between 4.2 Results and Discussion 89 350 nm and 305 nm to image fluorescence intensity from tryptophan and tyrosine, respec- tively (details are described in Chapter 3). Because the effective brightness of tryptophan is higher than tyrosine [193], and CaM does not have any tryptophan residues in its amino acid sequence. I decided to invert the measurement and monitor the electrophoretic mobility of AQP0 as a function of increased CaM concentration by measuring the intrinsic fluorescence intensity of AQP0. The excess amounts of CaM allow a plateau to be reached with mostly CaM bound to AQP0. Figure 4.7 The electrophoretic mobility of AQP0 and its complex with titrated CaM were detected under the intrinsic fluorescence microscope. The dashed line indicates a fit of Kd for the 2:1 non-cooperative interaction between CaM and AQP0. The shaded area covers a variation in Kd between three independent measurements. The electrophoretic mobility of AQP0 was measured as 0.21 ± 0.02 ×10−8 m2V−1s−1. The result was reasonable on account of the positive nature of AQP0 where positively- charged lysine and arginine residues exist in the amino acid sequence [80]. By titrating CaM (0 - 15 µM) to 5 µM AQP0, the electrophoretic mobility of the complex decreases and then reaches a plateau at - 0.28± 0.06×10−8 m2V−1s−1 (Figure 4.7), which is in good agreement with previous results for the electrophoretic mobility of the AQP0 and CaM 90 Investigation of protein-protein interactions using microfluidic techniques complex of - 0.2 ×10−8 m2V−1s−1 (Figure 4.6). Both experimental results show that the electrophoretic mobility of the AQP0 and CaM complex reaches a plateau at approximately - 0.3×10−8 m2V−1s−1, which suggests the complex of AQP0 and CaM is slightly negatively charged. Kd of CaM and full-length AQP0 based on electrophoretic mobility The expression for dissociation constant (Kd) for the interactions between CaM and full- length AQP0 is written as a function of concentration of isolated AQP0 (free AQP0 in solu- tion, noted as A), single bound AQP0 (one CaM bound with AQP0, noted as S), and double bound AQP0 (two CaM bound with AQP0, noted as D). Then, combing the solved A, S or D from Equation 4.3, the Kd can be calculated according to three models described in the appendix section. µobs = µA A At +µS S At +µD D At (4.3) where µobs is the measured electrophoretic mobility; µA is the measured electrophoretic mobility of AQP0 alone; A is the concentration of isolated AQP0; µS is the measured plateau value of electrophoretic mobility of 1 µM CaM with excess AQP0; S is the concentration of single bound AQP0; µD is the measured plateau value electrophoretic mobility of 5 µM AQP0 with excess CaM; D is the concentration of double bound AQP0; and At is the sum of A, S and D; For the 1:1 binding model, the Kd was calculated as 1.2 ± 0.14 µM. For the 2:1 non- cooperative binding model, the Kd was calculated as 3.97 ± 0.24 µM. Moreover, for the 2:1 sequential binding model, the first and the second Kd were calculated as 4.2 ± 0.11 µM and 7 ± 0.21 M, respectively (Table 4.2, Entry 3). The high Kd of second binding indicates that the 2:1 sequential binding model is not applicable for full-length AQP0 and CaM inter- actions. Therefore, it further suggests a non-cooperative binding manner between CaM and 4.2 Results and Discussion 91 AQP0, if there is a 2:1 binding. 4.2.4 Comparing the equilibrium dissociation constants Isothermal titration calorimetry (ITC) studies have suggested that CaM binds to the AQP0 C-terminal peptide with a two step binding process, which is indicated by first high binding affinity (Kd=71 nM) to the C-terminal helix, and then followed by a lower binding affinity (Kd=13 µM) [185]. Nuclear magnetic resonance spectroscopy (NMR) studies illustrated that two CaM molecules bind to the AQP0 peptide with a stepwise binding mechanism.Studies of the interactions be- tween full-length AQP0 and CaM may provide additional knowledge in biological context. Such knowledge could be helpful for drug design and development of artificial water chan- nels [15, 137]. However, NMR technique cannot study full-length AQP0 and CaM complex, because of the low solubility and large size of the complex. Thus, it is not certain if CaM follows a similar stepwise binding mechanism to the full-length AQP0 as for the AQP0 peptide [187]. Table 4.1 Dissociation constants (Kd in µM) for full-length or peptide AQP0 interactions with CaM Techniques Full-length Peptide Reference MST 2.5 40 5.0 ± 2.0 [131] Fluorescence Anisotropy n.d. 5.0 ± 2.0 1000 ± 2000 [131] ITC n.d. 0.071 13 [185] Fluorescence-based method n.d. 0.5 [142] 92 Investigation of protein-protein interactions using microfluidic techniques Recently, the binding affinity between AQP0 and CaM was studied using thermophore- sis (MST) and fluorescence anisotropy. MST studies showed that CaM binds to full-length AQP0 with a strong positive cooperative binding manner, while fluorescence anisotropy could not detect the same interactions. For the interactions between CaM and AQP0 pep- tide, both methods (MST and fluorescence anisotropy) indicate a non-cooperative interac- tion [131]. The studies of Kd between CaM and full-length AQP0 or AQP0 peptide with different techniques are summarised in Table 4.1. Table 4.2 Dissociation constants (Kd) for full-length AQP0 interactions with CaM are mea- sured by microfluidic methods Entry 1:1 binding 2:1 non-cooperative binding 2:1 sequential binding 1∗ 0.78 ± 0.1µM 2.4 ± 0.13 µM 2 ± 0.14 µM 1.8 ± 0.21 M 2∗ 0.36 ± 0.06 µM 1.1 ± 0.16 µM 1.2 ± 0.11 µM 4 ± 0.09M 3∗ 1.2 ± 0.14 µM 3.97 ± 0.24 µM 4.2 ± 0.11 µM 7 ± 0.21M 1∗ Based on Rh results; 2∗ Based on electrophoretic mobility measured with labelled CaM; 3∗ Based on electrophoretic mobility measured by intrinsic fluorescence microscope However, previous studies are generally based on the interactions between AQP0 peptide and CaM. It may provide essential information about binding site, yet, it might not provide the complete biological context [142, 185, 187]. For our microfluidic-based techniques, full- length AQP0 is adopted to study interactions with CaM. In general, as shown in Table 4.2, all sets of calculated Kd values of the 2:1 non-cooperative binding model is similar to that of the first Kd in the 2:1 sequential binding model. Moreover, all the second Kd of the 2:1 sequential binding model are too high, indicating the 2:1 sequential binding model is 4.2 Results and Discussion 93 not applicable for CaM and full-length AQP0 interaction. In other words, it indicates a non-cooperative binding manner if two CaM bind to one full-length AQP0 in solution. 4.2.5 Calculation of protein charge Biomolecule charge is the key physical parameter that regulates protein interactions. How- ever, the effective charge of a protein in solution can be affected by a wide range of factors. Also, protein charge comes from ionised amino acid residues and the protein termini, and these vary with pH and other solution conditions. Thus, the effective charge of proteins in the solution phase is difficult to predict [145]. In order to calculate effective protein charges, we combined the measurements of the diffusion coefficient and electrophoretic mobility of proteins. Based on these two parameters, the effective charge of molecules can be calcu- lated. The electrophoretic mobility (µ) of sample is related with diffusion coefficient (D) and effective charge (q), which is described by the Nernst-Einstein relation (Equation 4.4). µe = qD KBT (4.4) As a result, the effective charges of AQP0, CaM and its complex were calculated based on the well-defined equation 4.4. The key feature for this method is that the effective charge is determined without assumption of any specific shape or hydrodynamic radius of analytes, as the measured diffusion coefficient is used directly. Table 4.3 Physical parameters of AQP0, CaM and its complex Different components Rh (nm) Mobility (x 10−8 m2V−1s−1) Effective charge (e) CaM 2.6 ± 0.18 -3.2 ± 0.03 -8.7 ± 0.19 AQP0 6.4 ± 0.34 +0.21 ± 0.02 +1.4 ± 0.35 CaM-AQP0 6.4 ± 0.39 -0.2 ± 0.06 -1.3 ± 0.32 2CaM-AQP0 6.7 ± 1.13 -0.41 ± 0.12 -2.9 ± 0.15 94 Investigation of protein-protein interactions using microfluidic techniques The effective charges of CaM, AQP0 and its complex were obtained, where single bound (CaM-AQP0) and double bound (2CaM-AQP0) complex were achieved by having an excess of AQP0 and CaM, respectively. The effective charge of CaM and AQP0 was calculated as -8.5 e and +1.4 e, respectively. In addition, the effective charge of CaM-AQP0 and 2CaM- AQP0 is -2.1 e and -2.7 e, respectively (Figure 4.8 and Table 4.3). CaM AQP0 CaM-AQP0 2CaM-AQP0 2.6 nm CaM CaM-AQP0 6.4 nm 6.7 nm 2CaM_AQP0 6.4 nm AQP0 C ha rg e/ e 0 -5 -10 -15 -20 -25 -30 Figure 4.8 Measured and predicted charge of CaM, AQP0, CaM-AQP0 and 2CaM-AQP0. The charge of each component was also predicted based on its sequence alone using the Prot pi online bioinformatic tool. The charge of a protein arises from each charged ionis- able amino acid residue, and the charge of these amino acids carry at a given pH depends on the percentage of ionised residues, which is based on the pKa values of the side chains. Thus, the protein charge is predicted by a sum of the charged residues [9]. However, the charge predicted using this method is for unfolded proteins and does not necessarily corre- late with the solvated charge of the native protein structure, which might explain the differ- ence between predicted and microfluidic measured protein charges (Figure 4.8). Our data 4.2 Results and Discussion 95 (Table 4.3) indicates that the complex of 2CaM-AQP0 is slightly larger and more negatively charged than CaM-AQP0 (Figure 4.8). 4.2.6 Detection of interactions between CaM and AQP2 AQP2 is used as a negative control protein in this study. It is located in kidney collecting duct, which is responsible for water reabsorption and essentially regulates urine volume [174]. Several diseases, such as preeclampsia, nephrogenic diabetes insipidus and liver cirrhosis, are related to AQP2 malfunction [55, 202]. The Rh of CaM (1 µM) with titrated AQP2 were measured based on their diffusivity, Rh of CaM with titrated AQP2 is stable at 2.97 ± 0.19 nm (Figure 4.9, green dots), which is consistent with the Rh of CaM alone. I also measured electrophoretic mobilities of CaM with titrated AQP2. The electrophoretic mobility remains at - 3.2 ×10−8 m2V−1s−1 (Fig- ure 4.9, cyan dots), the same as the measured electrophoretic mobility of CaM itself. The result clearly indicates that there is no binding between CaM and AQP2. Figure 4.9 The measured Rh and electrophoretic mobility of CaM with increased AQP2 concentration. 96 Investigation of protein-protein interactions using microfluidic techniques 4.3 Conclusions Directly studying membrane protein interactions is often technically challenging due to low sample availability and high concentration requirements. To overcome this, membrane protein peptides have been used to investigate membrane protein interactions, which can provide information about binding sites and interacting residues. However, by using this approach it is possible to miss important information only present in full-length proteins. In this chapter, I demonstrate the marriage of microfluidic diffusion and electrophore- sis platforms employed for probing the interactions between AQPs (membrane protein) and CaM (regulator protein). These microfluidic techniques do not require any specific solu- tion conditions and only consume few microliters of sample. By detecting the diffusion coefficients and electrophoretic mobility of AQPs, CaM and their complexes in the solu- tion phase, the binding equilibria were characterised. A low µM equilibrium dissociation constant (Kd) was determined. Both diffusional sizing results and electrophoresis data sug- gest that the interactions between AQP0 and CaM follow a 2:1 non-cooperative binding model. In addition, the effective charge of AQPs, CaM and their complexes were calculated based on the measured diffusion coefficients and electrophoretic mobility. Furthermore, I confirmed that CaM specifically binds to AQP0, but there are no binding-related signals ob- served between CaM and AQP2. This study highlights that studying full-length membrane proteins in the solution phase can provide more reliable estimations of binding affinities and provide additional insights into the binding events, which is important for understanding the membrane protein interactions. 4.4 Materials and Methods 97 4.4 Materials and Methods 4.4.1 Materials All chemicals used in this Chapter were obtained from Sigma-Aldrich, and used directly without purification. Buffer used in these studies was 10 mM 2-Amino-2-(hydroxymethyl)- 1,3-propanediol, 10 mM NaCl, 0.03% detergent, N-Dodecyl β -D-maltoside (DDM), at pH 7.5, with 0.1 mM CaCl2. 4.4.2 Sample preparation Expression and purification of calmodulin (CaM), aquaporin-0 (AQP0) and aquaporin-2 (AQP2) were performed by Prof. Sara Linse and Stefan Kreida at Lund University, Sweden. Full-length AQP0 and AQP2 were expressed from Pichia pastoris as previous described [162]. AQP0 and AQP2 were purified and labelled with cysteine-reactive dye C5 Maleimide- Alexa 488 (Thermo Fisher) [131]. CaM was expressed from Escherichia coli and purified as previous described [31, 143, 163]. Serine 17 of CaM was replaced by cysteine, which allows for labelling cysteine- relative dyes [131]. PD 10 desalting column was used to remove excess dye. AQP0, AQP2 and CaM containing fractions were flash-frozen in liquid nitrogen and stored at -80 ◦C for microfluidic measurements. The concentrations of Alexa488 labelled CaM and AQP0 were detected by a CLAR- IOstar microplate reader (BMG LabTech), with half-area 96-well microplates (Corning, product #3679). The extinction coefficient (ε) of calmodulin and aquaporin is 2890 (M−1cm−1) and 39545 (M−1cm−1), respectively. The equation for calculating Alexa488 labelled protein concentration is: concentration(M) = A280−0.11×A494 ε (4.5) 98 Investigation of protein-protein interactions using microfluidic techniques 4.4.3 Imaging set-ups In order to detecting proteins in the microfluidic devices, a number of different fluorescence microscope set-ups were used. Imaging of proteins labelled with Alexa Fluor-488 was done using an inverted fluorescence microscope (Zeiss Axio Observer) fitted with a 49002 GFP filter (Chroma Technology, Vermont, USA). The photons emitted from the sample were acquired with an Evolve 512 EMCCD camera (Photometrics, Arizona, USA) coupled with a 10X objective. This give a field of view of 825 × 825 µm. Intrinsic fluorescence intensities of proteins were detected on a home-built intensities fluorescence microscope [41] (details as described in Chapter 3). 4.4.4 Protein sizing measurements Different flow rate between 60 µL/h to 200 µL/h was applied for the protein sizing measure- ments, which was controlled by neMESYS syringe pumps. After the system equilibrated, images of diffusion profiles were acquired by a fluorescence microscope (Zeiss Axio Ob- server) fitted with a white LED (Cairn Research, Kent, UK) and a 49002 GFP filter (Chroma Technology, Vermont, USA). The fluorescent diffusion profiles of Alexa488 labelled pro- teins along the channel were imaged at the detection position using an Evolve 512 CCD camera with a 5X objective and exposure times ranging from 0.5 s to 2 s depending on the sample concentration. Each diffusion experiment consumed 5 µL of sample. Details of data analysis are described in Section 3.4.5. 4.4.5 Electrophoresis microfluidic device design Microfluidic electrophoresis devices used in this chapter was designed by my colleague, Dr. Therese Herling [98]. The free-flow electrophoresis was designed with a central elec- trophoresis chamber with flanked metal alloy electrodes (Figure 4.5) [98]. 4.4 Materials and Methods 99 4.4.6 Fabrication of electrophoresis microfluidic device The details of fabrication of microfluidic devices are described in the Section 2.4.6. After bonding PDMS onto the glass, the devices were put on the hot plate at 82 ◦C to incorporate electrodes by inserting a low melting point InBiSn alloy (51% In, 32.5% Bi, 16.5% Sn, Conro Electronics). The solder wire was inserted through the electrode inlet, which melted and filled through the electrode channel with light pressure. Electrods PDMS Figure 4.10 The free-flow electrophoresis microfluidic device. Pillars were designed to prevent alloy wires entering the main channel. The diameter of pillars is 25 µm and the distances between them is also 25 µm. The inserted electrodes then solidify at room temperature (Figure 4.10). After that, the main channel of each device was filled with water to make device surface hydrophilicity. Before electrophoresis exper- iments, alloy metal was soldered with silver wires, thus voltage could be generated across electrodes. These wires were connected with an external voltage supply and a lock-in am- plifier. 4.4.7 Microfluidic free-flow electrophoresis experiments After prepare electrophoresis microfluidic devices, silver wire was soldered to the alloy metal for applying voltage. These wires were connected to an external voltage supply and a lock-in amplifier. In order to consider the variability of each devices, before each microflu- idic experiments, the cell constants and buffer conductivities of each device were measured 100 Investigation of protein-protein interactions using microfluidic techniques as previously described [98]. The measurements were then performed by loading buffers and samples to buffer inlet and sample inlet, respectively, and flow rate was controlled by neMESYS pump. The fluid withdrawal rate was 500 µL/h and device was equilibrated for 10 mintues. Voltage was ap- plied across device electrodes and the current was measured with a lock-in amplifier. Sam- ple stream migrated perpendicular to the direction of flow according to their electrophoretic mobility. Four repeats of a voltage ramp of 0 – 2 V at 0.2 V intervals were applied for each sample, and at each voltage three images with 500 ms to 1000 ms exposure time were acquired. Each electrophoretic mobility measurement consumed 10 µL of sample. 4.4.8 Electrophoresis data analysis As previous described [98], each device was filled with conductivity standard solution (KCl, 500 µS/cm) and 10 mV AC current was applied across the electrodes at variable frequencies of up to around 100 kHz. The frequency range was repeated five times for each solution. From the measured conductance (G) of the conductivity standard solution, the cell constant can be determined (Equation 4.7). We were able to take into account variation between individual microfluidic devices and buffer solutions that we used. During the voltage drop across an external 220 Ω resistor, the current (I) through the circuit was measured. The resistance of filled standard solution microfluidic device can be calculated, Rdevice+Rresistor = V I (4.6) and cell constant (K) of each device can be determined, K = σ G (4.7) where G = 1/Rdevice is solution conductance and σ is conductivity. By combing cell constant 4.4 Materials and Methods 101 and buffer conductivity, the buffer conductance (G) can be obtained. Thus, by combining the buffer conductance (G) in a given device and the measured current, the effective voltage drop, Ve f f ective, can be determined based on Ohm’s law (I=V/R, R=1/G). Then, an absolute value for the electric field strength, E, is able to obtained, across the solution from division of Ve f f ective by the distance between the electrodes (E=V/w, with the width of the channel w = 2150 µm). The electrophoretic velocities, vd , are calculated based on the deflection (δ ) divided by the residence time between the electrodes, which is known from the flow rate through the device and the channel dimensions. Linear fits to the slope of the plots of velocity against electric field, which are then performed to determine the electrophoretic mobility (µe = vd/E) of each sample. 102 Investigation of protein-protein interactions using microfluidic techniques 4.5 Appendix Calculation of equilibrium dissociation constants (Kd) There are several different models that we tried to fit our microfluidic diffusional sizing and electrophoretic mobility data. Generally, we assumed that CaM could migrate as: isolated CaM (C); a single CaM bound to AQP0 tetramer (S); and two CaM molecules bound to AQP0 tetramer (D). The free concentration AQP0 is represented by A. The total concentra- tion of CaM and AQP0 are noted as Ct and At , respectively. The equilibrium dissociation constants for the binding of a single and a second CaM to AQP0 is defined as KdA and KdB respectively. The observed hydrodynamic radius (Rh) or electrophoretic mobility (µobs) are present as a function of the different forms of CaM. 1:1 binding model with single Kd In this model, I assumed that there is one binding site of AQP0. Non-cooperative is assumed in the first instance. In this model, the total concentration of CaM (Ct , 1 µM) and AQP0 (At) are: Ct =C+bound protein (4.8) At = A+bound protein (4.9) Reorganised the Equation 4.8 and 4.9, I can get, A = At −Ct +C (4.10) Thus, Kd can be expressed as, Kd = C×A bound protein = C× (At −Ct +C) Ct −C (4.11) 4.5 Appendix 103 By rearranging the Equation 4.12 to get equation for this signal Kd model: 0 =C2+C× (At −Ct +Kd)−Ct ×Kd (4.12) 2:1 non-cooperative binding model In this model, I assumed that there are two independent binding sites on AQP0 (A and B) with the same Kd value. Non-cooperative is assumed in the first instance. As we assumed two CaM bind to one AQP0, so the concentration of binding sites is 0.5 µM. Same as 1:1 binding model, C is solved based on: 0 =C2+C× (At −Ct +Kd)−Ct ×Kd (4.13) 2:1 sequential binding model The sequential binding model describes that two CaMs are dependently binding to AQP0. During binding process, conformations of proteins are changed with each bound ligand, resulting in changing binding affinity. In this model, I supposed that two CaM sequentially binds to AQP0 tetramer. I assumed that the binding process is CaM + AQP0 → S, and then CaM + S → D. Ct and At can be expressed: Ct =C+S+2D (4.14) And, At = A+S+D (4.15) 104 Investigation of protein-protein interactions using microfluidic techniques KdA and KdB can be expressed: KdA = C×A S (4.16) Reorganised to give, S = C×A KdA (4.17) and, KdB = C×S D (4.18) Reorganised to give, D = C×S KdB = C2×A KdA×KdB (4.19) Because of Equation 4.14, Ct =C+S+ 2×C×S KdB (4.20) Reorganised to give, S = Ct −C 1+ 2×CKdB (4.21) Combine Equation 4.17 and Equation 4.19 with Equation 4.15, I can get S = At ×C KdA+C+C2/KdB (4.22) Then, set the two expressions for S equal to each other and reorganise to get a cubic equation for C. S = Ct −C 1+2×C/KdB = At ×C KdA+C+C2/KdB (4.23) 4.5 Appendix 105 Thus, I get the equation for sequential binding model: 0 = −1 KdB ×C3+( Ct KdB −1− 2×At KdB )×C2+(Ct −KdA−At)×C+Ct ×KdA (4.24) Chapter 5 Conclusions 5.1 Conclusions The work described in this dissertation illustrates the power of microfluidic techniques in performing quantitative studies of proteins and their interactions. The microfluidic-based measurements are effective and efficient, requiring low amount of samples. Moreover, all the measurements are performed in solution. These advantages allow investigation of pro- teins and their interactions under native conditions. In Chapter 2, I demonstrate that microfluidic techniques can be used to investigate pro- teins reacting with small molecules. Firstly, the BSA unfolding process induced by variation of pH was investigated by using a microfluidic diffusional sizing device, in which proteins were latently labelled and analysed in their native states. By detecting the fluorescence intensity, hydrodynamic radius (Rh) of BSA can be calculated. Based on that, the Rh of BSA as a function of pH was measured. In addition, the relative populations of folded and unfolded BSA under different pH conditions were simulated. Thus, the relative fractions of folded and unfolded BSA were calculated. Moreover, secondary structure of BSA was detected by circular dichroism under various pH conditions, to determine the fraction of α-helix. By combining microfluidic and circular dichroism measurements, the two-state 5.1 Conclusions 107 folding behaviour of BSA was observed. Furthermore, pKa values of each ionisable group in the folded state were predicted, which indicated that His241 was the key residue affect- ing stabilities of the folded and unfolded BSA. After that, the microfluidic diffusional sizing device was modified by increasing the length of diffusional channel to measure large par- ticles, such as the 70S ribosome. The Rh of 70S ribosome was measured. Moreover, the interactions between 70S ribosome and chloramphenicol (antibiotic) were detected based by the variation of Rh. The results show that is not clear whether chloramphenicol binds to 70S or not. I then incorporated the on-chip latent labelling approach with diffusional channel to design a top-down protein identification microfluidic device (Chapter 3). This microfluidic device was used to screen the physical parameters of proteins, including hydrodynamic radius, fluorescence intensity of related tryptophan, tyrosine and lysine residues, in their native state. Ten proteins with different numbers of amino acids and molecular weight were measured, with their unique combination of each measured parameter, each protein can be identified with high confidence in a single measurement. Chaperone-client clusters (αB- crystallin-titin) were then detected with the identification microfluidic device, the significant values of hydrodynamic radius, fluorescence intensity of related tryptophan, tyrosine and lysine radius of αB-crystallin and its client, titin, were measured. Finally, in this dissertation I explored membrane protein interactions by combining a microfluidic free-flow electrophoresis method and diffusional sizing approach (Chapter 4). I applied these methods to measure the changes of electrophoretic mobility and diffusion coefficients of membrane protein (aquaporins) and its regulatory molecule (calmodulin). The equilibrium dissociation constant was determined. In addition, with the measurements, I was able to identify calmodulin selectively binding to aquaporin-0 instead of aquaporin-2. Moreover, different binding complexes were identified and effective charge are calculated. Overall, I have demonstrated microfluidic approaches can be widely used for quantita- 108 Conclusions tive measurements of protein physical parameters and their interactions, which are difficult to access by conventional techniques. 5.2 Outlook Additional research includes studying protein aggregation processes and reactions, as well as their interactions with other proteins. In Chapter 3, the microfluidic diffusional sizing device was re-designed to measure the intact 70S ribosomal subunit that exceeded the measuring range of the original device. The modified microfluidic diffusional sizing device can be employed to explore the kinetics of the assembly of 70S ribosome subunits in the presence of magnesium [12, 267]. Moreover, this approach can be used to study the co-translational protein folding process [37, 39]. In the future, we will integrate approaches for quantitatively measuring tryptophan and tyrosine using our protein identification microfluidic device, with the aim of predicting their tryptophan and tyrosine content of each protein. This will allow us to identify proteins in the human genome via a single measurement. Due to its simplicity, low sample con- sumption and the advantages of protein identification on a single microfluidic chip in the solution phase, this method has significant potential in clinical diagnostic applications and for extending the platform of protein profiling for patient care. In addition, we will couple AKTA with our protein identification microfluidic platform. This will allow protein peaks eluted from size exclusion columns to be loaded directly into our protein identification microfluidic device, with the aim of revealing the identity of each peak. However, this may present several challenges, such as quantifying the exact number of measured amino acid residues within each protein on-chip, which we are currently seek- ing to overcome. In order to assess the potential of correctly identifying proteins without prior knowledge, we will integrate approaches for quantitatively measuring tryptophan and tyrosine together with our protein identification microfluidic device to predict the trypto- 5.2 Outlook 109 phan and tyrosine content of each protein. This will allow us to identify proteins (83%) in the human genome via a single measurement on our protein identification microfluidic device. In addition, incorporating the latent labelling of cysteine residues, the accuracy of protein identification in the human genome can be further increased to 93%. Furthermore, we believe that the electrophoresis microfluidic developed here can be provided a platform for a wide range of biophysical applications, such as studying pro- tein isoelectric points, chaperone-ligand binding, antibody-antigen binding, solutions effect of protein charge and their interactions. In addition, electrophoresis microfluidic methods can be used to separate the mixture of proteins and can be further coupled with the top- down protein identification method described in this thesis to screen and identify individual molecules. Bibliography [1] Abola, E. E., Bernstein, F. C., and Koetzle, T. F. (1984). The protein data bank. In Neutrons in Biology, pages 441–441. Springer. [2] Aebersold, R., , and Goodlett, D. R. (2001). Mass spectrometry in proteomics. Chemi- cal Reviews, 101(2):269–296. [3] Agresti, J. J., Antipov, E., Abate, A. R., Ahn, K., Rowat, A. C., Baret, J.-C., Marquez, M., Klibanov, A. M., Griffiths, A. D., and Weitz, D. A. (2010). Ultrahigh-throughput screening in drop-based microfluidics for directed evolution. Proceedings of the National Academy of Sciences, 107(9):4004–4009. [4] Ahn, K., Kerbage, C., Hunt, T. P., Westervelt, R., Link, D. R., and Weitz, D. A. (2006). Dielectrophoretic manipulation of drops for high-speed microfluidic sorting devices. Ap- plied Physics Letters, 88(2):024104. [5] Alaimo, A. and Villarroel, A. (2018). Calmodulin: A multitasking protein in kv7. 2 potassium channel functions. Biomolecules, 8(3):57. [6] Anderson, D. E., Becktel, W. J., and Dahlquist, F. W. (1990). ph-induced denaturation of proteins: a single salt bridge contributes 3-5 kcal/mol to the free energy of folding of t4 lysozyme. Biochemistry, 29(9):2403–2408. [7] Anfinsen, C. B. (1973). Principles that govern the folding of protein chains. Science, 181(4096):223–230. [8] Anfinsen, C. B., Haber, E., Sela, M., and White, F. (1961). The kinetics of formation of native ribonuclease during oxidation of the reduced polypeptide chain. Proceedings of the National Academy of Sciences, 47(9):1309–1314. [9] Antosiewicz, J., McCammon, J. A., and Gilson, M. K. (1994). Prediction of ph- dependent properties of proteins. Journal of molecular biology, 238(3):415–436. [10] Arosio, P., Müller, T., Rajah, L., Yates, E. V., Aprile, F. A., Zhang, Y., Cohen, S. I., White, D. A., Herling, T. W., De Genst, E. J., Linse, S., Vendruscolo, M., Dobson, C. M., and Tuomas, K. P. J. (2015). Microfluidic diffusion analysis of the sizes and interactions of proteins under native solution conditions. ACS nano, 10(1):333–341. [11] Aue, W., Bartholdi, E., and Ernst, R. R. (1976). Two-dimensional spectroscopy. ap- plication to nuclear magnetic resonance. The Journal of Chemical Physics, 64(5):2229– 2246. Bibliography 111 [12] Ball, L. A., Johnson, P. M., and Walker, I. O. (1973). The dissociation of escherichia coli ribosomes. European journal of biochemistry, 37(1):12–20. [13] Ban, N., Nissen, P., Hansen, J., Moore, P. B., and Steitz, T. A. (2000a). The com- plete atomic structure of the large ribosomal subunit at 2.4 å resolution. Science, 289(5481):905–920. [14] Ban, N., Nissen, P., Hansen, J., Moore, P. B., and Steitz, T. A. (2000b). The complete atomic structure of the large ribosomal subunit at 2.4 A resolution. Science, 289:905– 920. [15] Barboiu, M. and Gilles, A. (2013). From natural to bioassisted and biomimetic artifi- cial water channel systems. Accounts of chemical research, 46(12):2814–2823. [16] Barbulovic-Nad, I., Lucente, M., Sun, Y., Zhang, M., Wheeler, A. R., and Bussmann, M. (2006). Bio-microarray fabrication techniques—a review. Critical reviews in biotech- nology, 26(4):237–259. [17] Beebe, D. J., Mensing, G. A., and Walker, G. M. (2002). Physics and applications of microfluidics in biology. Annual Review of Biomedical Engineering, 4:261–286. [18] Beechem, J. M. and Brand, L. (1985). Time-resolved fluorescence of proteins. Annual review of biochemistry, 54(1):43–71. [19] Behnen, P., Davis, E., Delaney, E., Frohm, B., Bauer, M., Cedervall, T., O’Connell, D., Åkerfeldt, K. S., and Linse, S. (2012). Calcium-dependent interaction of calmodulin with human 80s ribosomes and polyribosomes. Biochemistry, 51(34):6718–6727. [20] Benson, J. R. and Hare, P. (1975). O-phthalaldehyde: fluorogenic detection of primary amines in the picomole range. comparison with fluorescamine and ninhydrin. Proceed- ings of the National Academy of Sciences, 72(2):619–622. [21] Berggård, T., Linse, S., and James, P. (2007). Methods for the detection and analysis of protein–protein interactions. Proteomics, 7(16):2833–2842. [22] Berman, H. M., Westbrook, J., Feng, Z., Gilliland, G., Bhat, T. N., Weissig, H., Shindyalov, I. N., and Bourne, P. E. (2000). The protein data bank. Nucleic acids re- search, 28(1):235–242. [23] Berne, B. J. and Pecora, R. (2000). Dynamic Light Scattering: With Applications to Chemistry, Biology, and Physics. Dover Publications, New York, dover edition. [24] Bhandari, P., Narahari, T., and Dendukuri, D. (2011). ‘fab-chips’: a versatile, fabric-based platform for low-cost, rapid and multiplexed diagnostics. Lab on a Chip, 11(15):2493–2499. [25] Biemann, K. and Martin, S. A. (1987). Mass spectrometric determination of the amino acid sequence of peptides and proteins. Mass Spectrometry Reviews, 6(1):1–75. [26] Bisswanger, H. (2017). Enzyme kinetics: principles and methods. John Wiley & Sons. 112 Bibliography [27] Bloemendal, H., Zweers, A., Vermorken, F., Dunia, I., and Benedetti, E. (1972). The plasma membranes of eye lens fibres. biochemical and structural characterization. Cell differentiation, 1(2):91–106. [28] Bodas, D. and Khan-Malek, C. (2007). Hydrophilization and hydrophobic recovery of pdms by oxygen plasma and chemical treatment—an sem investigation. Sensors and Actuators B: Chemical, 123(1):368–373. [29] Boncoraglio, A., Minoia, M., and Carra, S. (2012). The family of mammalian small heat shock proteins (hspbs): implications in protein deposit diseases and motor neu- ropathies. The international journal of biochemistry & cell biology, 44(10):1657–1669. [30] Bornhop, D. J., Latham, J. C., Kussrow, A., Markov, D. A., Jones, R. D., and Sørensen, H. S. (2007). Free-solution, label-free molecular interactions studied by back-scattering interferometry. science, 317(5845):1732–1736. [31] Brodin, P., Johansson, C., Forsén, S., Drakenberg, T., and Grundström, T. (1990). Functional properties of calbindin d9k mutants with exchanged ca2+ binding sites. Jour- nal of Biological Chemistry, 265(19):11125–11130. [32] Brody, J. P. and Yager, P. (1997). Diffusion-based extraction in a microfabricated device. Sensors and Actuators-A-Physical Sensors, 58(1):13–18. [33] Brouzes, E., Medkova, M., Savenelli, N., Marran, D., Twardowski, M., Hutchison, J. B., Rothberg, J. M., Link, D. R., Perrimon, N., and Samuels, M. L. (2009). Droplet microfluidic technology for single-cell high-throughput screening. Proceedings of the National Academy of Sciences, 106(34):14195–14200. [34] Brown, L., Koerner, T., Horton, J. H., and Oleschuk, R. D. (2006). Fabrication and characterization of poly (methylmethacrylate) microfluidic devices bonded using surface modifications and solvents. Lab on a Chip, 6(1):66–73. [35] Bruus, H. (2008). Theoretical microfluidics. oxford master series in condensed matter physics. [36] Bryngelson, J. D., Onuchic, J. N., Socci, N. D., and Wolynes, P. G. (1995). Funnels, pathways, and the energy landscape of protein folding: a synthesis. Proteins: Structure, Function, and Bioinformatics, 21(3):167–195. [37] Cabrita, L. D., Cassaignau, A. M., Launay, H. M., Waudby, C. A., Wlodarski, T., Camilloni, C., Karyadi, M.-E., Robertson, A. L., Wang, X., and Wentink, A. S. (2016). A structural ensemble of a ribosome–nascent chain complex during cotranslational protein folding. Nature structural & molecular biology, 23(4):278. [38] Cabrita, L. D., Hsu, S.-T. D., Launay, H., Dobson, C. M., and Christodoulou, J. (2009). Probing ribosome-nascent chain complexes produced in vivo by nmr spectroscopy. Pro- ceedings of the National Academy of Sciences, 106(52):22239–22244. [39] Cassaignau, A. M., Launay, H. M., Karyadi, M.-E., Wang, X., Waudby, C. A., Deck- ert, A., Robertson, A. L., Christodoulou, J., and Cabrita, L. D. (2016). A strategy for co-translational folding studies of ribosome-bound nascent chain complexes using nmr spectroscopy. Nature protocols, 11(8):1492. Bibliography 113 [40] Challa, P. K., Kartanas, T., Charmet, J., and Knowles, T. P. (2017). Microfluidic devices fabricated using fast wafer-scale led-lithography patterning. Biomicrofluidics, 11(1):014113. [41] Challa, P. K., Peter, Q., Wright, M. A., Zhang, Y., Saar, K. L., Carozza, J. A., Benesch, J. L., and Knowles, T. P. (2018). Real-time intrinsic fluorescence visualization and siz- ing of proteins and protein complexes in microfluidic devices. Analytical chemistry, 90(6):3849–3855. [42] Chen, R. and Snyder, M. (2013). Promise of personalized omics to precision medicine. Wiley Interdisciplinary Reviews: Systems Biology and Medicine, 5(1):73–82. [43] Chin, C. D., Laksanasopin, T., Cheung, Y. K., Steinmiller, D., Linder, V., Parsa, H., Wang, J., Moore, H., Rouse, R., Umviligihozo, G., et al. (2011). Microfluidics-based diagnostics of infectious diseases in the developing world. Nature medicine, 17(8):1015. [44] Chin, C. D., Linder, V., and Sia, S. K. (2012). Commercialization of microfluidic point-of-care diagnostic devices. Lab on a Chip, 12(12):2118–2134. [45] Chiti, F. and Dobson, C. M. (2006). Protein misfolding, functional amyloid, and hu- man disease. Annual Review of Biochemistry, 75:333–366. [46] Chu, L.-Y., Utada, A. S., Shah, R. K., Kim, J.-W., and Weitz, D. A. (2007). Control- lable monodisperse multiple emulsions. Angewandte Chemie, 119(47):9128–9132. [47] Clapham, D. E. (2007). Calcium signaling. Cell, 131(6):1047–1058. [48] Collins, F. S., Green, E. D., Guttmacher, A. E., and Guyer, M. S. (2003). A vision for the future of genomics research. Nature, 422(6934):835. [49] Correia, M., Neves-Petersen, M. T., Jeppesen, P. B., Gregersen, S., and Petersen, S. B. (2012). Uv-light exposure of insulin: pharmaceutical implications upon covalent insulin dityrosine dimerization and disulphide bond photolysis. PLoS One, 7(12):e50733. [50] Correia, M., Thiagarajan, V., Coutinho, I., Gajula, G. P., Petersen, S. B., and Neves- Petersen, M. T. (2014). Modulating the structure of egfr with uv light: new possibilities in cancer therapy. PLoS One, 9(11):e111617. [51] Coscia, F., Watters, K. M., Curtis, M., Eckert, M. A., Chiang, C. Y., Tyanova, S., Montag, A., Lastra, R. R., Lengyel, E., and Mann, M. (2016). Integrative proteomic pro- filing of ovarian cancer cell lines reveals precursor cell associated proteins and functional status. Nature Communications, 7:12645. [52] Culbertson, C. T., Jacobson, S. C., and Ramsey, J. M. (2002). Diffusion coefficient measurements in microfluidic devices. Talanta, 56(2):365–373. [53] Damaschun, G., Damaschun, H., Gast, K., Gernat, C., and Zirwer, D. (1991). Acid denatured apo-cytochrome c is a random coil: evidence from small-angle x-ray scatter- ing and dynamic light scattering. Biochimica et Biophysica Acta (BBA)-Proteins and Proteomics, 1078(2):289–295. 114 Bibliography [54] De Genst, E., Messer, A., and Dobson, C. M. (2014). Antibodies and protein misfold- ing: from structural research tools to therapeutic strategies. Biochimica et Biophysica Acta (BBA)-Proteins and Proteomics, 1844(11):1907–1919. [55] Deen, P., Verdijk, M., Knoers, N., Wieringa, B., Monnens, L., van Os, C. H., and van Oost, B. A. (1994). Requirement of human renal water channel aquaporin-2 for vasopressin-dependent concentration of urine. Science, 264(5155):92–95. [56] Dobson, C. M. (1999). Protein misfolding, evolution and disease. Trends in biochem- ical sciences, 24(9):329–332. [57] Dobson, C. M. (2001). The structural basis of protein folding and its links with human disease. Philosophical Transactions of the Royal Society B, 356(1406):133–145. [58] Dobson, C. M. (2003a). Protein folding and misfolding. Nature, 426(6968):884. [59] Dobson, C. M. (2003b). Protein folding and misfolding. Nature, 426(December):884– 890. [60] Dobson, C. M., Šali, A., and Karplus, M. (1998a). Protein folding: a perspective from theory and experiment. Angewandte Chemie International Edition, 37(7):868–893. [61] Dobson, C. M., Sali, A., and Karplus, M. (1998b). Protein Folding: A Perspective from Theory and Experiment. Angewandte Chemie, 37(7):868–893. [62] Duffy, D. C., McDonald, J. C., Schueller, O. J., and Whitesides, G. M. (1998). Rapid prototyping of microfluidic systems in poly (dimethylsiloxane). Analytical Chemistry, 70(23):4974–4984. [63] Dyson, H. J. and Wright, P. E. (2004). Unfolded proteins and protein folding studied by nmr. Chemical Reviews, 104(8):3607–3622. [64] Ecroyd, H. and Carver, J. A. (2009). Crystallin proteins and amyloid fibrils. Cellular and Molecular Life Sciences, 66(1):62. [65] Edelhoch, H. (1960). The properties of thyroglobulin i. the effects of alkali. Journal of Biological Chemistry, 235(5):1326–1334. [66] El Kadi, N., Taulier, N., Le Huérou, J. Y., Gindre, M., Urbach, W., Nwigwe, I., Kahn, P. C., and Waks, M. (2006). Unfolding and refolding of bovine serum albumin at acid pH: ultrasound and structural studies. Biophysical Journal, 91(9):3397–3404. [67] Engel, A. and Gaub, H. E. (2008). Structure and mechanics of membrane proteins. Annual Review of Biochemistry, 77:127–148. [68] Erwin, D. G., Messer, A., and Dobson, C. M. (2014). Antibodies and protein misfold- ing: From structural research tools to therapeutic strategies. Biochimica et Biophysica Acta (BBA)-Proteins and Proteomics, 1844(11):1907–1919. [69] Felgenhauer, K. (1974). Protein size and cerebrospinal fluid composition. Klinische Wochenschrift, 52(24):1158–1164. Bibliography 115 [70] Fersht, A. (1999). Structure and mechanism in protein science. W. H. Freeman. [71] Fetter, K., Van Wilder, V., Moshelion, M., and Chaumont, F. (2004). Interactions between plasma membrane aquaporins modulate their water channel activity. The Plant Cell, 16(1):215–228. [72] Fields, J. B., Németh-Cahalan, K. L., Freites, J. A., Vorontsova, I., Hall, J. E., and Tobias, D. J. (2017). Calmodulin gates aquaporin 0 permeability through a positively charged cytoplasmic loop. Journal of Biological Chemistry, 292(1):185–195. [73] Flanagan, J. M., Kataoka, M., Shortle, D., and Engelman, D. M. (1992). Truncated staphylococcal nuclease is compact but disordered. Proceedings of the National Academy of Sciences, 89(2):748–752. [74] Flecha, F. L. G. and Levi, V. (2003). Determination of the molecular size of bsa by fluorescence anisotropy. Biochemistry and molecular biology education, 31(5):319–322. [75] Frank, A. M., Savitski, M. M., Nielsen, M. L., Zubarev, R. A., , and Pevzner, P. A. (2007). De novo peptide sequencing and identification with precision mass spectrometry. Journal of Proteome Research, 6(1):114–123. [76] Gabel, F., Jensen, M. R., Zaccai, G., and Blackledge, M. (2009). Quantitative mod- elfree analysis of urea binding to unfolded ubiquitin using a combination of small angle x-ray and neutron scattering. Journal of the American Chemical Society, 131(25):8769– 8771. [77] Gabler, R., Westhead, E., and Ford, N. (1974). Studies of ribosomal diffusion coeffi- cients using laser light-scattering spectroscopy. Biophysical journal, 14(7):528–545. [78] Gao, J., Gomez, F. A., Härter, R., and Whitesides, G. M. (1994). Determination of the effective charge of a protein in solution by capillary electrophoresis. Proceedings of the National Academy of Sciences, 91(25):12027–12030. [79] Gao, J., Li, W.-X., Feng, S.-Q., Yuan, Y.-S., Wan, D.-F., Han, W., and Yu, Y. (2008). A protein–protein interaction network of transcription factors acting during liver cell proliferation. Genomics, 91(4):347–355. [80] Gasteiger, E., Hoogland, C., Gattiker, A., Wilkins, M. R., Appel, R. D., and Bairoch, A. (2005). Protein identification and analysis tools on the expasy server. In The pro- teomics protocols handbook, pages 571–607. Springer. [81] Genovese, F., Gualandi, A., Taddia, L., Marverti, G., Pirondi, S., Marraccini, C., Perco, P., Pela, M., Guerrini, R., Amoroso, M. R., Esposito, F., Martello, A., Ponterini, G., Arca, D. D., and Costi, M. P. (2014). Mass Spectrometric/Bioinformatic Identi fi cation of a Protein Subset That Characterizes the Cellular Activity of Anticancer Peptides. Journal of Proteome Research, 13:5250–5261. [82] Ghisaidoobe, A. B. and Chung, S. J. (2014). Intrinsic tryptophan fluorescence in the detection and analysis of proteins: a focus on förster resonance energy transfer tech- niques. International journal of molecular sciences, 15(12):22518–22538. 116 Bibliography [83] Gitlin, I., Carbeck, J. D., and Whitesides, G. M. (2006). Why are proteins charged? networks of charge–charge interactions in proteins measured by charge ladders and cap- illary electrophoresis. Angewandte Chemie International Edition, 45(19):3022–3060. [84] Goldman, W. F. and Baptist, J. N. (1979). Isoelectric point electrophoresis: A new technique for protein purification. Journal of Chromatography A, 179(2):330–332. [85] Gonen, T., Cheng, Y., Kistler, J., and Walz, T. (2004). Aquaporin-0 membrane junc- tions form upon proteolytic cleavage. Journal of molecular biology, 342(4):1337–1345. [86] Görg, A., Weiss, W., and Dunn, M. J. (2004). Current two-dimensional electrophoresis technology for proteomics. Proteomics, 4(12):3665–3685. [87] Greene, R. F. and Pace, C. N. (1974). Urea and Guanidine Hydrochloride Denaturation of Ribonuclease, Lysozyme, alpha-Chymotrypsin and beta-Lactoglobulin. The Journal of Biological Chemistry, 249(17):5388–5393. [88] Greenfield, N. J. (2006). Using circular dichroism spectra to estimate protein sec- ondary structure. Nature protocols, 1(6):2876–2890. [89] Grover, W. H., von Muhlen, M. G., and Manalis, S. R. (2008). Teflon films for chemically-inert microfluidic valves and pumps. Lab on a Chip, 8(6):913–918. [90] Halim, A. A. A., Kadir, H. A., and Tayyab, S. (2008). Bromophenol blue binding as a probe to study urea and guanidine hydrochloride denaturation of bovine serum albumin. The Journal of Biochemistry, 144(1):33–38. [91] Hanlon, A. D., Larkin, M. I., and Reddick, R. M. (2010). Free-solution, label-free protein-protein interactions characterized by dynamic light scattering. Biophysical jour- nal, 98(2):297–304. [92] Hansen, J. L., Moore, P. B., and Steitz, T. a. (2003). Structures of Five Antibiotics Bound at the Peptidyl Transferase Center of the Large Ribosomal Subunit. Journal of Molecular Biology, 330(5):1061–1075. [93] Harrison, D. J., Fluri, K., Seiler, K., Fan, Z., Effenhauser, C. S., and Manz, A. (1993). Micromachining a miniaturized capillary electrophoresis-based chemical analysis system on a chip. Science, 261(5123):895–897. [94] Hatch, A., Kamholz, A. E., Hawkins, K. R., Munson, M. S., Schilling, E. A., Weigl, B. H., and Yager, P. (2001a). A rapid diffusion immunoassay in a T-sensor. Nature biotechnology, 19(May):461–465. [95] Hatch, A., Kamholz, A. E., Hawkins, K. R., Munson, M. S., Schilling, E. A., Weigl, B. H., and Yager, P. (2001b). A rapid diffusion immunoassay in a t-sensor. Nature biotechnology, 19(5):461. [96] Haynes, D. H. (1988). Microdroplets of water-insoluble drugs and injectable formula- tions containing same. US Patent 4,725,442. [97] He, M. and Herr, A. E. (2009). Microfluidic polyacrylamide gel electrophoresis with in situ immunoblotting for native protein analysis. Analytical chemistry, 81(19):8177– 8184. Bibliography 117 [98] Herling, T., Müller, T., Rajah, L., Skepper, J., Vendruscolo, M., and Knowles, T. (2013). Integration and characterization of solid wall electrodes in microfluidic devices fabricated in a single photolithography step. Applied physics letters, 102(18):184102. [99] Herling, T. W., Arosio, P., Müller, T., Linse, S., and Knowles, T. P. (2015). A microflu- idic platform for quantitative measurements of effective protein charges and single ion binding in solution. Physical Chemistry Chemical Physics, 17(18):12161–12167. [100] Herling, T. W., O’Connell, D. J., Bauer, M. C., Persson, J., Weininger, U., Knowles, T. P., and Linse, S. (2016). A microfluidic platform for real-time detection and quantifi- cation of protein-ligand interactions. Biophysical journal, 110(9):1957–1966. [101] Herr, A. E., Hatch, A. V., Throckmorton, D. J., Tran, H. M., Brennan, J. S., Gianno- bile, W. V., and Singh, A. K. (2007). Microfluidic immunoassays as rapid saliva-based clinical diagnostics. Proceedings of the National Academy of Sciences, 104(13):5268– 5273. [102] Hjerten, S., L. J. L. and Zhang, R. (1989). High-performance liquid chromatography on continuous polymer beds. Journal of Chromatography A, 473:273–275. [103] Hochberg, G. K. and Benesch, J. L. (2014). Dynamical structure of αb-crystallin. Progress in biophysics and molecular biology, 115(1):11–20. [104] Hong, L. and Lei, J. (2009). Scaling law for the radius of gyration of proteins and its dependence on hydrophobicity. Journal of Polymer Science Part B: Polymer Physics, 47(2):207–214. [105] Hsia, C.-Y., Richards, M. J., and Daniel, S. (2015). A review of traditional and emerging methods to characterize lipid–protein interactions in biological membranes. Analytical Methods, 7(17):7076–7094. [106] Huang, Y., Zhao, S., Shi, M., Liu, J., and Liang, H. (2012). Microchip electrophore- sis coupled with on-line magnetic separation and chemiluminescence detection for mul- tiplexed immunoassay. Electrophoresis, 33(7):1198–1204. [107] Hulse, W. L. and Forbes, R. T. (2011). A nanolitre method to determine the hydrody- namic radius of proteins and small molecules by taylor dispersion analysis. International journal of pharmaceutics, 411(1-2):64–68. [108] Janasek, D., Schilling, M., Franzke, J., and Manz, A. (2006). Isotachophoresis in free-flow using a miniaturized device. Analytical chemistry, 78(11):3815–3819. [109] Jellinger, K. A. (2012). Interaction between pathogenic proteins in neurodegenerative disorders. Journal of cellular and molecular medicine, 16(6):1166–1183. [110] Jiang, H. (2010). Mass spectrometric analysis of biopharmaceutical and follow-on protein drugs. [111] Kahl, C. R. and Means, A. R. (2003). Regulation of cell cycle progression by calcium/calmodulin-dependent pathways. Endocrine reviews, 24(6):719–736. 118 Bibliography [112] Kamatari, Y. O., Ohji, S., Konno, T., Seki, Y., Soda, K., Kataoka, M., and Akasaka, K. (1999). The compact and expanded denatured conformations of apomyoglobin in the methanol-water solvent. Protein Science, 8(4):873–882. [113] Kamholz, A. E. and Yager, P. (2001). Theoretical analysis of molecular diffusion in pressure-driven laminar flow in microfluidic channels. Biophysical journal, 80(1):155– 160. [114] Kang, S., Maeng, H., Kim, B. G., Qing, G. M., Choi, Y. P., Kim, H. Y., Kim, P. S., Kim, Y., Kim, Y. H., Choi, Y. D., and Cho, N. H. (2012). In situ Identi fi cation and Localization of IGHA2 in the Breast Tumor Microenvironment by Mass Spectrometry. Journal of Proteome Research, 11:4567–4574. [115] Karow, A. R., Götzl, J., and Garidel, P. (2015). Resolving power of dynamic light scattering for protein and polystyrene nanoparticles. Pharmaceutical development and technology, 20(1):84–89. [116] Kataoka, M., Hagihara, Y., Mihara, K., and Goto, Y. (1993). Molten globule of cytochrome c studied by small angle x-ray scattering. Journal of Molecular Biology, 229(3):591–596. [117] Kataoka, M., Nishii, I., Fujisawa, T., Ueki, T., Tokunaga, F., and Goto, Y. (1995). Structural characterization of the molten globule and native states of apomyoglobin by solution x-ray scattering. Journal of Molecular Biology, 249(1):215–228. [118] Kaushansky, A., Allen, J. E., Gordus, A., Stiffler, M. A., Karp, E. S., Chang, B. H., and MacBeath, G. (2010). Quantifying protein–protein interactions in high throughput using protein domain microarrays. Nature protocols, 5(4):773. [119] Kendrew, J. C., Bodo, G., Dintzis, H. M., Parrish, R., Wyckoff, H., and Phillips, D. C. (1958). A three-dimensional model of the myoglobin molecule obtained by x-ray analysis. Nature, 181(4610):662–666. [120] Keskin, O., Tuncbag, N., and Gursoy, A. (2016). Predicting protein–protein interac- tions from the molecular to the proteome level. Chemical reviews, 116(8):4884–4909. [121] Khandurina, J., McKnight, T. E., Jacobson, S. C., Waters, L. C., Foote, R. S., and Ramsey, J. M. (2000). Integrated system for rapid pcr-based dna analysis in microfluidic devices. Analytical Chemistry, 72(13):2995–3000. [122] Khatter, H., Myasnikov, A. G., Natchiar, S. K., and Klaholz, B. P. (2015). Structure of the human 80s ribosome. Nature, 520(7549):640. [123] Kim, J.-O., Kim, H., Ko, D.-H., Min, K.-I., Park, S.-Y., and Kim, D.-P. (2014). A monolithic and flexible fluoropolymer film microreactor for organic synthesis applica- tions. Lab on a Chip, 14(21):4270–4276. [124] Kim, P., Kwon, K. W., Park, M. C., Lee, S. H., Kim, S. M., and Suh, K. Y. (2008). Soft lithography for microfluidics: a review. [125] King, L. S., Kozono, D., and Agre, P. (2004). From structure to disease: the evolving tale of aquaporin biology. Nature reviews Molecular cell biology, 5(9):687. Bibliography 119 [126] Kluger, R. and Alagic, A. (2004). Chemical cross-linking and protein–protein inter- actions—a review with illustrative protocols. Bioorganic chemistry, 32(6):451–472. [127] Knowles, T. P., Vendruscolo, M., and Dobson, C. M. (2014a). The amyloid state and its association with protein misfolding diseases. Nature reviews Molecular cell biology, 15(6):384. [128] Knowles, T. P., White, D. A., Abate, A. R., Agresti, J. J., Cohen, S. I., Sperling, R. A., De Genst, E. J., Dobson, C. M., and Weitz, D. A. (2011). Observation of spatial propa- gation of amyloid assembly from single nuclei. Proceedings of the National Academy of Sciences, 108(36):14746–14751. [129] Knowles, T. P. J., Vendruscolo, M., and Dobson, C. M. (2014b). The amyloid state and its association with protein misfolding diseases. Nature Reviews Molecular Cell Biology, 15(6):384–96. [130] Köhler, S., Weilbeer, C., Howitz, S., Becker, H., Beushausen, V., and Belder, D. (2011). Pdms free-flow electrophoresis chips with integrated partitioning bars for bubble segregation. Lab on a Chip, 11(2):309–314. [131] Kreida, S., Roche, J. V., Olsson, C., Linse, S., and Törnroth-Horsefield, S. (2018). Protein-protein interactions in aqp regulation–biophysical characterization of aqp0-cam and aqp2-lip5 complex formation. Faraday Discussions. [132] Kreida, S. and Törnroth-Horsefield, S. (2015). Structural insights into aquaporin selectivity and regulation. Current opinion in structural biology, 33:126–134. [133] LaBaer, J. and Ramachandran, N. (2005). Protein microarrays as tools for functional proteomics. Current Opinion in Chemical Biology, 9(1):14–19. [134] Lagally, E., Medintz, I., and Mathies, R. (2001). Single-molecule dna amplification and analysis in an integrated microfluidic device. Analytical chemistry, 73(3):565–570. [135] Łapin´ska, U., Saar, K. L., Yates, E. V., Herling, T. W., Müller, T., Challa, P. K., Dobson, C. M., and Knowles, T. P. (2017). Gradient-free determination of isoelectric points of proteins on chip. Physical Chemistry Chemical Physics, 19(34):23060–23067. [136] Lebowitz, J., Lewis, M. S., and Schuck, P. (2002). Modern analytical ultracentrifu- gation in protein science: a tutorial review. Protein Science, 11(9):2067–2079. [137] Lee, C. S., Choi, M.-k., Hwang, Y. Y., Kim, H., Kim, M. K., and Lee, Y. J. (2018). Facilitated water transport through graphene oxide membranes functionalized with aquaporin-mimicking peptides. Advanced Materials, 30(14):1705944. [138] Lee, H. J., Wark, A. W., and Corn, R. M. (2008). Microarray methods for protein biomarker detection. Analyst, 133(8):975–983. [139] Levinthal, C. (1968). Are there pathways for protein folding? Journal de chimie physique, 65:44–45. [140] Li, W., Zhang, L., Ge, X., Xu, B., Zhang, W., Qu, L., Choi, C.-H., Xu, J., Zhang, A., and Lee, H. (2018). Microfluidic fabrication of microparticles for biomedical applica- tions. Chemical Society Reviews, 47(15):5646–5683. 120 Bibliography [141] Liedberg, B., Nylander, C., and Lunström, I. (1983). Surface plasmon resonance for gas detection and biosensing. Sensors and actuators, 4:299–304. [142] Lindsey Rose, K., Wang, Z., Magrath, G., Hazard, E., Hildebrandt, J., and Schey, K. (2008). Aquaporin 0- calmodulin interaction and the effect of aquaporin 0 phosphoryla- tion. Biochemistry, 47(1):339–347. [143] Linse, S., Johansson, C., Brodin, P., Grundstroem, T., Drakenberg, T., and Forsen, S. (1991). Electrostatic contributions to the binding of calcium in calbindin d9k. Biochem- istry, 30(1):154–162. [144] Liotta, L. A., Kohn, E. C., and Petricoin, E. F. (2001). Clinical proteomics: person- alized molecular medicine. Jama, 286(18):2211–2214. [145] Lund, M. and Jönsson, B. (2005). On the charge regulation of proteins. Biochemistry, 44(15):5722–5727. [146] Lundstrom, K. (2004). Structural genomics on membrane proteins: mini review. Combinatorial chemistry & high throughput screening, 7(5):431–439. [147] MacBeath, G. (2002). Protein microarrays and proteomics. Nature Genetics, 32:526– 32. [148] Mallam, A. L. (2009). How does a knotted protein fold? The FEBS journal, 276(2):365–375. [149] Malyuchenko, N., Tonevitsky, E., Agapov, I., Pevzner, I., Bykov, V., Kirpichnikov, M., and Tonevitsky, A. (2006). A study of e. coli and t. maritima ribosomes by atomic force microscopy. Biophysics, 51(3):385–390. [150] Mann, M., Hendrickson, R. C., and Pandey, A. (2001). Analysis of proteins and proteomes by mass spectrometry. Annual Review of Physical Chemistry, 70(1):437–473. [151] Manz, A., Harrison, D. J., Verpoorte, E. M. J., Fettinger, J. C., Paulus, A., Li, H., and Widmer, H. M. (1992). Planar chips technology for miniaturization and integration of separation techniques into monitoring systems Capillary electrophoresis on a chip. J. Chromatogr. A, 593:253–258. [152] Marchesseau, S., Mani, J., Martineau, P., Roquet, F., Cuq, J., and Pugniere, M. (2002). Casein interactions studied by the surface plasmon resonance technique. Journal of dairy science, 85(11):2711–2721. [153] Mazutis, L., Gilbert, J., Ung, W. L., Weitz, D. A., Griffiths, A. D., and Heyman, J. A. (2013). Single-cell analysis and sorting using droplet-based microfluidics. Nature protocols, 8(5):870. [154] Mcdonald, J. C., Duffy, D. C., Anderson, J. R., and Chiu, D. T. (2000). Fabrication of microfluidic systems in poly (dimethylsiloxane). Electrophoresis, 21:27–40. [155] Mehmood, S., Allison, T. M., and Robinson, C. V. (2015). Mass spectrometry of protein complexes: from origins to applications. Annual review of physical chemistry, 66:453–474. Bibliography 121 [156] Mitchell, P. (2002). A perspective on protein microarrays. Nature biotechnology, 20(3):225. [157] Morar, A. S., Olteanu, A., Young, G. B., and Pielak, G. J. (2001). Solvent-induced collapse of α-synuclein and acid-denatured cytochrome c. Protein Science, 10(11):2195– 2199. [158] Muck, A., Wang, J., Jacobs, M., Chen, G., Chatrathi, M. P., Jurka, V., Vy`borny`, Z., Spillman, S. D., Sridharan, G., and Schöning, M. J. (2004). Fabrication of poly (methyl methacrylate) microfluidic chips by atmospheric molding. Analytical chemistry, 76(8):2290–2297. [159] Müller, T., Arosio, P., Rajah, L., Cohen, S. I., Yates, E. V., Vendruscolo, M., Dobson, C. M., and Knowles, T. P. (2016). Particle-based monte-carlo simulations of steady- state mass transport at intermediate péclet numbers. International Journal of Nonlinear Sciences and Numerical Simulation, 17(3-4):175–183. [160] Németh-Cahalan, K. L. and Hall, J. E. (2000). ph and calcium regulate the water permeability of aquaporin 0. Journal of Biological Chemistry, 275(10):6777–6782. [161] Németh-Cahalan, K. L., Kalman, K., and Hall, J. E. (2004). Molecular basis of ph and ca2+ regulation of aquaporin water permeability. The Journal of general physiology, 123(5):573–580. [162] Öberg, F., Ekvall, M., Nyblom, M., Öberg, F., Ekvall, M., Nyblom, M., Backmark, A., Neutze, R., and Hedfalk, K. (2009). Insight into factors directing high production of eukaryotic membrane proteins; production of 13 human aqps in pichia pastoris. Molecu- lar membrane biology, 26(4):215–227. [163] O’Connell, D. J., Bauer, M. C., O’Brien, J., Johnson, W. M., Divizio, C. A., O’Kane, S. L., Berggård, T., Merino, A., Åkerfeldt, K. S., Linse, S., and Dolores, J. C. (2010). Integrated protein array screening and high throughput validation of 70 novel neural calmodulin binding proteins. Molecular & cellular proteomics. [164] O’Connell, M. R., Gamsjaeger, R., and Mackay, J. P. (2009). The structural analysis of protein–protein interactions by nmr spectroscopy. Proteomics, 9(23):5224–5232. [165] Ojeda-May, P. and Garcia, M. E. (2010). Electric field-driven disruption of a native β -sheet protein conformation and generation of a helix-structure. Biophysical journal, 99(2):595–599. [166] Oláh, Z., Kálmán, J., Tóth, M. E., Zvara, Á., Sántha, M., Ivitz, E., Janka, Z., and Pákáski, M. (2015). Proteomic analysis of cerebrospinal fluid in Alzheimer’s disease: wanted dead or alive. Journal of Alzheimer’s Disease, 44(4):1303–1312. [167] Omenn, G. S., Lane, L., Lundberg, E. K., Beavis, R. C., Overall, C. M., and Deutsch, E. W. (2016). Metrics for the human proteome project 2016: progress on identifying and characterizing the human proteome, including post-translational modifications. Journal of proteome research, 15(11):3951–3960. 122 Bibliography [168] Osváth, S., Köhler, G., Závodszky, P., and Fidy, J. (2005). Asymmetric effect of domain interactions on the kinetics of folding in yeast phosphoglycerate kinase. Protein Science, 14(6):1609–1616. [169] Otzen, D. (2011). Protein-surfactant interactions: a tale of many states. Biochimica et Biophysica Acta (BBA)-Proteins and Proteomics, 1814(5):562–591. [170] Overington, J. P., Al-Lazikani, B., and Hopkins, A. L. (2006). How many drug targets are there? Nature reviews Drug discovery, 5(12):993. [171] Pamme, N. (2007). Continuous flow separations in microfluidic devices. Lab on a Chip, 7(12):1644–1659. [172] Papish, A. L., Tari, L. W., and Vogel, H. J. (2002). Dynamic light scattering study of calmodulin–target peptide complexes. Biophysical journal, 83(3):1455–1464. [173] Pawson, T. and Nash, P. (2000). Protein–protein interactions define specificity in signal transduction. Genes & development, 14(9):1027–1047. [174] Pearce, D., Soundararajan, R., Trimpert, C., Kashlan, O. B., Deen, P. M., and Kohan, D. E. (2015). Collecting duct principal cell transport processes and their regulation. Clinical Journal of the American Society of Nephrology, 10(1):135–146. [175] Petricoin, E., Wulfkuhle, J., Espina, V., and Liotta, L. A. (2004). Clinical Proteomics: Revolutionizing Disease Detection and Patient Tailoring Therapy. Journal of Proteome Research, 3(2):209–217. [176] Pierce, M. M., Raman, C., and Nall, B. T. (1999). Isothermal titration calorimetry of protein–protein interactions. Methods, 19(2):213–221. [177] Poehlsgaard, J. and Douthwaite, S. (2005). The bacterial ribosome as a target for antibiotics. Nature Reviews Microbiology, 3(11):870. [178] Ponomarenko, E. A., Poverennaya, E. V., Ilgisonis, E. V., Pyatnitskiy, M. A., Kopy- lov, A. T., Zgoda, V. G., Lisitsa, A. V., and Archakov, A. I. (2016). The size of the human proteome: the width and depth. International journal of analytical chemistry, 2016. [179] Price, W. S. (1997). Pulsed-field gradient nuclear magnetic resonance as a tool for studying translational diffusion: Part 1. basic theory. Concepts in Magnetic Resonance: An Educational Journal, 9(5):299–336. [180] Puchades, M., Hansson, S. F., Nilsson, C. L., Andreasen, N., Blennow, K., and Davidsson, P. (2003). Proteomic studies of potential cerebrospinal fluid protein mark- ers for Alzheimer’s disease. Molecular brain research, 118(1-2):140–146. [181] Putnam, F. (2012). The Plasma Proteins V4: Structure, Function, and Genetic Con- trol. Elsevier. [182] Ramachandran, N., Hainsworth, E., Bhullar, B., Eisenstein, S., Rosen, B., Lau, A. Y., Walter, J. C., and LaBaer, J. (2004). Self-assembling protein microarrays. Science, 305(5680):86–90. Bibliography 123 [183] Raymond, D. E., Manz, A., and Widmer, H. M. (1994). Continuous sample pretreat- ment using a free-flow electrophoresis device integrated onto a silicon chip. Analytical Chemistry, 66(18):2858–2865. [184] Receveur, V., Durand, D., Desmadril, M., and Calmettes, P. (1998). Repulsive in- terparticle interactions in a denatured protein solution revealed by small angle neutron scattering. FEBS Letters, 426(1):57–61. [185] Reichow, S. L., Clemens, D. M., Freites, J. A., Németh-Cahalan, K. L., Heyden, M., Tobias, D. J., Hall, J. E., and Gonen, T. (2013a). Allosteric mechanism of water-channel gating by ca 2+–calmodulin. Nature Structural and Molecular Biology, 20(9):1085. [186] Reichow, S. L., Clemens, D. M., Freites, J. A., Németh-Cahalan, K. L., Heyden, M., Tobias, D. J., Hall, J. E., and Gonen, T. (2013b). Allosteric mechanism of water-channel gating by ca2+–calmodulin. Nature structural & molecular biology, 20(9):1085–1092. [187] Reichow, S. L. and Gonen, T. (2008). Noncanonical binding of calmodulin to aquaporin-0: implications for channel regulation. Structure, 16(9):1389–1398. [188] Reid, G. E. and McLuckey, S. A. (2002). ’Top down’ protein characterization via tandem mass spectrometry. J. Mass Spectrom., 37(7):663–675. [189] Ren, K., Dai, W., Zhou, J., Su, J., and Wu, H. (2011). Whole-teflon microfluidic chips. Proceedings of the National Academy of Sciences, 108(20):8162–8166. [190] Resing, K. A. and Ahn, N. G. (2005). Proteomics strategies for protein identification. FEBS letters, 579(4):885–889. [191] Robinson, C. V., Sali, A., and Baumeister, W. (2007). The molecular sociology of the cell. Nature, 450(7172):973. [192] Rolland, J. P., Van Dam, R. M., Schorzman, D. A., Quake, S. R., and DeSimone, J. M. (2004). Solvent-resistant photocurable “liquid teflon” for microfluidic device fabrication. Journal of the american chemical society, 126(8):2322–2323. [193] Ross, J. A. and Jameson, D. M. (2008). Time-resolved methods in biophysics. 8. frequency domain fluorometry: applications to intrinsic protein fluorescence. Photo- chemical & Photobiological Sciences, 7(11):1301–1312. [194] Roth, M. (1971). Fluorescence reaction for amino acids. Analytical Chemistry, 43(7):880–882. [195] Rotman, B. (1961). Measurement of activity of single molecules of β -d- galactosidase. Proceedings of the National Academy of Sciences, 47(12):1981–1991. [196] Royer, C. A. (2006). Probing protein folding and conformational transitions with fluorescence. Chemical reviews, 106(5):1769–1784. [197] Sackmann, E. K., Fulton, A. L., and Beebe, D. J. (2014). The present and future role of microfluidics in biomedical research. Nature, 507(7491):181. [198] Samaha, R. R., Green, R., and Noller, H. F. (1995). A base pair between tRNA and 23S rRNA in the peptidyl transferase centre of the ribosome. Nature, 377:309–314. 124 Bibliography [199] Santos, A. L., Moreirinha, C., Lopes, D., Esteves, A. C., Henriques, I., Almeida, A., Domingues, M. R. M., Delgadillo, I., Correia, A., and Cunha, A. (2013). Effects of uv radiation on the lipids and proteins of bacteria studied by mid-infrared spectroscopy. Environmental science & technology, 47(12):6306–6315. [200] Schluenzen, F., Tocilj, A., Zarivach, R., Harms, J., Gluehmann, M., Janell, D., Bashan, A., Bartels, H., Agmon, I., Franceschi, F., and Yonath, A. (2000). Structure of functionally activated small ribosomal subunit at 3.3 angstroms resolution. Cell, 102:615–623. [201] Schmid, A., Dordick, J. S., Hauer, B., Kiener, A., Wubbolts, M., and Witholt, B. (2001). Industrial biocatalysis today and tomorrow. Nature, 409(January):258–268. [202] Schrier, R., Fassett, R., Ohara, M., and Martin, P. (1998). Vasopressin release, wa- ter channels, and vasopressin antagonism in cardiac failure, cirrhosis, and pregnancy. Proceedings of the Association of American Physicians, 110(5):407–411. [203] Schuwirth, B. S., Borovinskaya, M. A., Hau, C. W., Zhang, W., Holton, J. M., and Cate, J. H. D. (2005). Structures of the Bacterial Ribosome at 3.5 A Resolution. Science, 310(November):827–834. [204] Semisotnov, G. V., Kihara, H., Kotova, N. V., Kimura, K., Amemiya, Y., Wak- abayashi, K., Serdyuk, I. N., Timchenko, A. A., Chiba, K., and Nikaido, K. (1996). Protein globularization during folding. a study by synchrotron small-angle x-ray scatter- ing. Journal of Molecular Biology, 262(4):559–574. [205] Serdakowski, A. L. and Dordick, J. S. (2008). Enzyme activation for organic solvents made easy. Trends in biotechnology, 26(1):48–54. [206] Shah, N., Sattar, A., Benanti, M., Hollander, S., and Cheuck, L. (2006). Magnetic resonance spectroscopy as an imaging tool for cancer: a review of the literature. The Journal of the American Osteopathic Association, 106(1):23–27. [207] Shenoy, R. and Bowman, C. N. (2012). Kinetics of interfacial radical polymerization initiated by a glucose-oxidase mediated redox system. Biomaterials, 33(29):6909–6914. [208] Shimanovich, U., Michaels, T. C., De Genst, E., Matak-Vinkovic, D., Dobson, C. M., and Knowles, T. P. (2017a). Sequential release of proteins from structured multishell microcapsules. Biomacromolecules, 18(10):3052–3059. [209] Shimanovich, U., Ruggeri, F. S., De Genst, E., Adamcik, J., Barros, T. P., Porter, D., Müller, T., Mezzenga, R., Dobson, C. M., Vollrath, F., et al. (2017b). Silk micrococoons for protein stabilisation and molecular encapsulation. Nature communications, 8:15902. [210] SIM, E. and SIM, R. B. (1979). Hydrodynamic parameters of the detergent- solubilised hydrogenase from paracoccus denitrificans. The FEBS Journal, 97(1):119– 126. [211] Simons, S. S. and Johnson, D. F. (1976). The Structure of the Fluorescent Adduct Formed in the Reaction of o-Phthalaldehyde and Thiols with Amines. Journal of the American Chemical Society, 98:7098–7099. Bibliography 125 [212] Squires, T. M. (2005). Microfluidics: Fluid physics at the nanoliter scale. Reviews of Modern Physics, 77(July):977–1026. [213] Squires, T. M. and Quake, S. R. (2005). Microfluidics: Fluid physics at the nanoliter scale. Reviews of modern physics, 77(3):977. [214] Stefani, M. and Dobson, C. M. (2003). Protein aggregation and aggregate toxicity: new insights into protein folding, misfolding diseases and biological evolution. Journal of molecular medicine, 81(11):678–699. [215] Sterling, H. J., Batchelor, J. D., Wemmer, D. E., and Williams, E. R. (2010). Effects of buffer loading for electrospray ionization mass spectrometry of a noncovalent protein complex that requires high concentrations of essential salts. Journal of the American Society for Mass Spectrometry, 21(6):1045–1049. [216] Sternson, L. A., Stobaugh, J. F., and Repta, A. J. (1985). Rational design and evalu- ation of improved o-phthalaldehyde-like fluorogenic reagents. Analytical Biochemistry, 144:233–246. [217] Stone, H., Stroock, A., and Ajdari, A. (2004). Engineering flows in small devices: microfluidics toward a lab-on-a-chip. Annual Review of Fluid Mechanics, 36(1):381– 411. [218] Stone, H. A. and Kim, S. (2001). Microfluidics: basic issues, applications, and chal- lenges. Aiche Journal, 47(6):1250–1254. [219] Su, T., Lu, J., Thomas, R., Cui, Z., and Penfold, J. (1998). The conformational structure of bovine serum albumin layers adsorbed at the silica-water interface. The Journal of Physical Chemistry B, 102(41):8100–8108. [220] Sultana, R., Boyd-Kimball, D., Cai, J., Pierce, W. M., Klein, J. B., Merchant, M., and Butterfield, D. A. (2007). Proteomics analysis of the Alzheimer’s disease hippocampal proteome. Journal of Alzheimer’s Disease, 11(2):153–64. [221] Sundberg, S. A. (2000). High-throughput and ultra-high-throughput screening: solution-and cell-based approaches. Current opinion in biotechnology, 11(1):47–53. [222] Sutandy, F. R., Qian, J., Chen, C.-S., and Zhu, H. (2013). Overview of protein mi- croarrays. Current protocols in protein science, 72(1):27–1. [223] Tan, K. P., Nguyen, T. B., Patel, S., Varadarajan, R., and Madhusudhan, M. S. (2013). Depth: a web server to compute depth, cavity sizes, detect potential small-molecule ligand-binding cavities and predict the pka of ionizable residues in proteins. Nucleic Acids Research, 41:W314–W321. [224] Tan, S. H., Nguyen, N.-T., Chua, Y. C., and Kang, T. G. (2010). Oxygen plasma treatment for reducing hydrophobicity of a sealed polydimethylsiloxane microchannel. Biomicrofluidics, 4(3):32204. [225] Templin, M. F., Stoll, D., Schrenk, M., Traub, P. C., Vöhringer, C. F., and Joos, T. O. (2002). Protein microarray technology. Drug discovery today, 7(15):815–822. 126 Bibliography [226] Theberge, A. B., Courtois, F., Schaerli, Y., Fischlechner, M., Abell, C., Hollfelder, F., and Huck, W. T. (2010). Microdroplets in microfluidics: an evolving platform for discov- eries in chemistry and biology. Angewandte Chemie International Edition, 49(34):5846– 5868. [227] Thomas, P. J., Qu, B., and Pedersen, P. L. (1995). Defective protein folding as a basis of human disease. Trends in biochemical sciences, 20(11):456–459. [228] Thompsett, A. R. and Brown, D. R. (2007). Dual polarisation interferometry analysis of copper binding to the prion protein: evidence for two folding states. BBA-Proteins Proteomics, 1774(7):920–927. [229] Thorsen, T., Maerkl, S. J., and Quake, S. R. (2002). Microfluidic large-scale integra- tion. Science, 298(5593):580–584. [230] Timasheff, S. N., Mescanti, L., Basch, J. J., and Townend, R. (1966). Conforma- tional transitions of bovine β -lactoglobulins a, b, and c. Journal of Biological Chemistry, 241(11):2496–2501. [231] Törnroth-Horsefield, S., Hedfalk, K., Fischer, G., Lindkvist-Petersson, K., and Neutze, R. (2010). Structural insights into eukaryotic aquaporin regulation. FEBS letters, 584(12):2580–2588. [232] Tugarinov, V., Choy, W.-Y., Orekhov, V. Y., and Kay, L. E. (2005). Solution nmr-derived global fold of a monomeric 82-kda enzyme. Proceedings of the National Academy of Sciences of the United States of America, 102(3):622–627. [233] Unger, A., Beckendorf, L., Böhme, P., Kley, R., von Frieling-Salewsky, M., Lochmüller, H., Schröder, R., Fürst, D. O., Vorgerd, M., and Linke, W. A. (2017). Translocation of molecular chaperones to the titin springs is common in skeletal my- opathy patients and affects sarcomere function. Acta neuropathologica communications, 5(1):72. [234] Unger, M. A., Chou, H.-P., Thorsen, T., Scherer, A., and Quake, S. R. (2000). Monolithic microfabricated valves and pumps by multilayer soft lithography. Science, 288(5463):113–116. [235] Utada, A., Lorenceau, E., Link, D., Kaplan, P., Stone, H., and Weitz, D. (2005). Monodisperse double emulsions generated from a microcapillary device. Science, 308(5721):537–541. [236] Varadaraj, K., Kumari, S., Shiels, A., and Mathias, R. T. (2005). Regulation of aquaporin water permeability in the lens. Investigative ophthalmology & visual science, 46(4):1393–1402. [237] Vyatkina, K., Wu, S., Dekker, L. J. M., Vanduijn, M. M., Liu, X., Tolic´, N., Dvorkin, M., Alexandrova, S., Luider, T. M., Paša-Tolic´, L., and Pevzner, P. A. (2015). De novo sequencing of peptides from top-down tandem mass spectra. Journal of Proteome Re- search, 14(11):4450–4462. Bibliography 127 [238] Walter, S., Hubner, B., Hahn, U., and Schmid, F. X. (1995). Destabilization of a protein helix by electrostatic interactions. Journal of Molecular Biology, 252(1):133– 143. [239] Wang, H.-W. and Wang, J.-W. (2017). How cryo-electron microscopy and x-ray crystallography complement each other. Protein Science, 26(1):32–39. [240] Weigl, B. H. and Yager, P. (1999). Microfluidic diffusion-based separation and de- tection. Science, 283(5400):346–347. [241] Weinreb, P. H., Zhen, W., Poon, A. W., Conway, K. A., and Lansbury, P. T. (1996). Nacp, a protein implicated in alzheimer’s disease and learning, is natively unfolded. Bio- chemistry, 35(43):13709–13715. [242] Wheeler, A. R., Throndset, W. R., Whelan, R. J., Leach, A. M., Zare, R. N., Liao, Y. H., Farrell, K., Manger, I. D., and Daridon, A. (2003). Microfluidic device for single- cell analysis. Analytical chemistry, 75(14):3581–3586. [243] Whitesides, G. M. (2006). The origins and the future of microfluidics. Nature, 442(7101):368. [244] Wilkins, D. K., Grimshaw, S. B., Receveur, V., Dobson, C. M., Jones, J. A., and Smith, L. J. (1999a). Hydrodynamic radii of native and denatured proteins measured by pulse field gradient nmr techniques. Biochemistry, 38(50):16424–16431. [245] Wilkins, D. K., Grimshaw, S. B., Receveur, V., Dobson, C. M., Jones, J. A., and Smith, L. J. (1999b). Hydrodynamic radii of native and denatured proteins measured by pulse field gradient nmr techniques. Biochemistry, 38(50):16424–16431. [246] Wilson, D. N. (2009). The A-Z of bacterial translation inhibitors. Critical Reviews in Biochemistry and Molecular Biology, 44(6):393–433. [247] Wilson, D. N. (2014). Ribosome-targeting antibiotics and mechanisms of bacterial resistance. Nature Reviews Microbiology, 12(1):35–48. [248] Wilson, D. S. and Nock, S. (2003). Recent developments in protein microarray tech- nology. Angewandte Chemie International Edition, 42(5):494–500. [249] Wimberly, B. T., Brodersen, D. E., Clemons, W. M., Morgan-Warren, R. J., Carter, A. P., Vonrhein, C., Hartsch, T., and Ramakrishnan, V. (2000a). Structure of the 30S ribosomal subunit. Nature, 407:327–339. [250] Wimberly, B. T., Brodersen, D. E., Clemons Jr, W. M., Morgan-Warren, R. J., Carter, A. P., Vonrhein, C., Hartsch, T., and Ramakrishnan, V. (2000b). Structure of the 30s ribosomal subunit. Nature, 407(6802):327. [251] Wu, P., Castner, D. G., and Grainger, D. W. (2008). Diagnostic devices as biomateri- als: a review of nucleic acid and protein microarray surface performance issues. Journal of Biomaterials Science, Polymer Edition, 19(6):725–753. [252] Xia, Y. and Whitesides, G. M. (1998). Soft lithography. Angewandte Chemie Inter- national Edition, 37(5):550–575. 128 Bibliography [253] Xia, Z. and Storm, D. R. (2005). The role of calmodulin as a signal integrator for synaptic plasticity. Nature Reviews Neuroscience, 6(4):267. [254] Yadavali, S., Jeong, H.-H., Lee, D., and Issadore, D. (2018). Silicon and glass very large scale microfluidic droplet integration for terascale generation of polymer micropar- ticles. Nature communications, 9(1):1222. [255] Yager, P., Edwards, T., Fu, E., Helton, K., Nelson, K., Tam, M. R., and Weigl, B. H. (2006). Microfluidic diagnostic technologies for global public health. Nature, 442(7101):412. [256] Yang, L.-Q., Ji, X.-L., and Liu, S.-Q. (2013). The free energy landscape of protein folding and dynamics: A global view. Journal of Biomolecular Structure and Dynamics, 31(9):982–992. [257] Yates, E. V., Müller, T., Rajah, L., De Genst, E. J., Arosio, P., Linse, S., Vendrus- colo, M., Dobson, C. M., and Knowles, T. P. (2015). Latent analysis of unmodified biomolecules and their complexes in solution with attomole detection sensitivity. Nature chemistry, 7(10):802. [258] Zhang, C., Xu, J., Ma, W., and Zheng, W. (2006). Pcr microfluidic devices for dna amplification. Biotechnology advances, 24(3):243–284. [259] Zhang, C.-X. and Manz, A. (2003). High-speed free-flow electrophoresis on chip. Analytical chemistry, 75(21):5759–5766. [260] Zhang, Y., Buell, A. K., Müller, T., De Genst, E., Benesch, J., Dobson, C. M., and Knowles, T. P. J. (2016). Protein aggregate-ligand binding assays based on microfluidic diffusional separation. ChemBioChem, 17(20):1920–1924. [261] Zhang, Y., Yates, E. V., Hong, L., Saar, K. L., Meisl, G., Dobson, C. M., and Knowles, T. P. (2018). On-chip measurements of protein unfolding from direct observations of micron-scale diffusion. Chemical Science, 9(14):3503–3507. [262] Zhao, G., Perilla, J. R., Yufenyuy, E. L., Meng, X., Chen, B., Ning, J., Ahn, J., Gronenborn, A. M., Schulten, K., and Aiken, C. (2013). Mature hiv-1 capsid structure by cryo-electron microscopy and all-atom molecular dynamics. Nature, 497(7451):643. [263] Zheng, B., Roach, L. S., and Ismagilov, R. F. (2003). Screening of protein crystal- lization conditions on a microfluidic chip using nanoliter-size droplets. Journal of the American chemical society, 125(37):11170–11171. [264] Zhou, H.-X., Rivas, G., and Minton, A. P. (2008). Macromolecular crowding and confinement: biochemical, biophysical, and potential physiological consequences. An- nual Review of Biophysics, 37:375–397. [265] Zhu, H., Klemic, J. F., Chang, S., Bertone, P., Casamayor, A., Klemic, K. G., Smith, D., Gerstein, M., Reed, M. A., and Snyder, M. (2000). Analysis of yeast protein kinases using protein chips analysis of yeast protein kinases using protein chips analysis of yeast protein kinases using protein chips. Nature genetics. Bibliography 129 [266] Zhuang, X., Ha, T., Kim, H. D., Centner, T., Labeit, S., and Chu, S. (2000). Fluores- cence quenching: A tool for single-molecule protein-folding study. Proceedings of the National Academy of Sciences, 97(26):14241–14244. [267] Zitomer, R. S. and Flaks, J. G. (1972). Magnesium dependence and equilibrium of the escherichia coli ribosomal subunit association. Journal of molecular biology, 71(2):263– 279.