Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Custom affinity chromatography : development of a novel platform for rapid creation and validation of… Ang Kian Meng, Aaron 2018

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2018_may_ang_aaron.pdf [ 4.84MB ]
Metadata
JSON: 24-1.0364031.json
JSON-LD: 24-1.0364031-ld.json
RDF/XML (Pretty): 24-1.0364031-rdf.xml
RDF/JSON: 24-1.0364031-rdf.json
Turtle: 24-1.0364031-turtle.txt
N-Triples: 24-1.0364031-rdf-ntriples.txt
Original Record: 24-1.0364031-source.json
Full Text
24-1.0364031-fulltext.txt
Citation
24-1.0364031.ris

Full Text

   CUSTOM AFFINITY CHROMATOGRAPHY— DEVELOPMENT OF A NOVEL PLATFORM FOR RAPID CREATION AND VALIDATION OF AFFINITY MEDIA USING DNA APTAMER BASED LIGANDS  by  AARON ANG KIAN MENG B.A.Sc, The University of British Columbia, 2015  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF   MASTER OF APPLIED SCIENCE in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Chemical and Biological Engineering)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  February 2018  © Aaron Ang, 2018  ii  Abstract Biotechnology companies are now well skilled in the technologies and operations needed to manufacture biologic drugs safely for the treatment of major diseases. Those technologies have enabled development of highly efficient and cost-effective platforms for downstream processing of monoclonal antibody (mAb) based drugs. But translating those advances to create cost-effective DSP platforms for non-mAb protein therapeutics having relatively low annual production rates, often termed “orphan” drugs, has proven difficult.  A key driver of the efficiency of mAb DSP platforms is the use of protein A affinity chromatography to capture and purify the product directly from clarified culture supernatants. Unfortunately, effective ligands for affinity capture of non-mAb biologics are generally not available. But this could change through the development of a technology that rapidly discovers and validates cost-effective affinity ligands against non-antibody protein targets. This project describes the development of a new technology pipeline to accelerate the discovery, optimization, and validation of affinity chromatography media that is specifically tailored to provide for robust economical capture of non-mAb biologic drugs from complex cell cultures. It is based on the use of DNA aptamers as affinity ligands discovered using an advanced aptamer screening technology we call High-Fidelity Systematic Evolution of Ligand by Exponential Enrichment (Hi-Fi SELEX). The refined, truly robust Hi-Fi SELEX technology described in this thesis greatly improves upon a proof-of-concept version of that method we recently described.    This second-generation Hi-Fi SELEX method was used to successfully select high-affinity ligands against two non-mAb target proteins, human complement Factor D and human mesothelin. Anti-Factor D (aFD-30) aptamer against Factor D was then used as ligand in preparative affinity chromatography columns. The chemically modified aFD-30 with 3’ inverted dT nucleotide cap was immobilized on preparative affinity chromatography matrix for the capture and purification of Factor D from CHO cell supernatant. Standard column performance data were collected, including static and dynamic binding capacities, purities, concentration factors, and yields, which showed excellent separation performance. These results therefore demonstrate the potential of the proposed technology for custom design and validation of preparative chromatography media that can benefit the growing orphan drugs market by reducing manufacturing costs. iii  Lay Summary Biologic therapeutics, most notably antibodies, for the treatment of major diseases have been successfully manufactured at large scale using highly efficient and cost-effective technologies. However, translating those advances to create cost-effective platforms for non-antibodies based drugs, often termed “orphan drugs” with relatively low annual production rates have proven difficult.   This project describes the development of a new technology pipeline to accelerate the discovery, optimization, and validation of affinity media using a cheap and stable genetic material based binding agent that is specifically tailored to provide for robust economical capture of non-antibodies biologic drugs from complex cell cultures. The results from this project demonstrate the potential of the proposed technology for custom affinity column using this binding agent that can benefit the growing orphan drugs market by reducing manufacturing costs and the associated costs of treatments.  iv  Preface A version of Chapter 2 from this thesis has been published as:  Ang A, Ouellet E, Cheung KC, Haynes C. Highly Efficient  and Reliable DNA Aptamer Selection Using the Partitioning Capabilities of ddPCR — The Hi-Fi SELEX Method. Methods in Molecular Biology. in press (Dec. 2017).  As a first author, I drafted the manuscript and further expert recommendations and contributions were made by my principle supervisor Dr. Charles Haynes. Some elements of the second-generation Hi-Fi SELEX method described in this chapter were completed in collaboration with Dr. Eric Ouellet, a former student in the Haynes Lab.    The work in Chapter 3 is part of a collaboration with BioRad Inc. (Hercules, CA) to discover DNA aptamers by Hi-Fi SELEX and evaluate their performance as ligands on preparative affinity chromatography. My contribution to this work was to isolate high affinity aptamers against two different target proteins, human Factor D and human mesothelin, using the second-generation Hi-Fi SELEX method. Mr. Nathan Chan, an undergraduate student who worked in the Haynes Lab under my supervision, provided Figure 3.3 using a new algorithm for defining multicomponent binding of DNA aptamers during their selection by Hi-Fi SELEX (Appendix). BioRad Inc. provided data on the column performance using the chosen ligand, as shown in Figures 3.7, 3.8, 3.9, and 3.10. BioRad Inc. also provided Tables 3.5 and 3.6. As with Chapter 1, I wrote the initial  manuscript draft for this chapter and it was improved with contributions from Dr. Charles Haynes.      v  Table of Contents Abstract .......................................................................................................................................... ii Lay Summary ............................................................................................................................... iii Preface ........................................................................................................................................... iv Table of Contents .......................................................................................................................... v List of Tables ............................................................................................................................... vii List of Figures ............................................................................................................................. viii Nomenclature ............................................................................................................................. xiii Chapter 1: Thesis Motivation, Objectives and Background ..................................................... 1 1.1 The Challenge to Downstream Processing an Ever-Expanding Biologics Portfolio ............ 1 1.2 Thesis Objectives .................................................................................................................. 5 1.3 Review of Relevant Literature .............................................................................................. 6 1.3.1 Proposed Ligands and Technologies to Enable Custom Affinity Chromatography ...... 6 1.3.2 Aptamers ........................................................................................................................ 9 1.3.3 SELEX and Related Technologies for Selecting DNA Aptamers ............................... 11 1.3.4 Technical (Non-Therapeutic) Applications of Aptamers ............................................ 16 1.3.5 Affinity Chromatography – Uses and Underpinning Concepts ................................... 18 1.4 The Aptamer-Based Custom Chromatography Development Pipeline - Basic Concept ... 22 1.5 Anticipated Impact of Thesis Research .............................................................................. 24 Chapter 2: Highly Efficient and Reliable DNA Aptamer Selection Using the Partitioning Capabilities of ddPCR – The Hi-Fi SELEX Method ............................................................... 26 2.1 Introduction ......................................................................................................................... 26 2.2 Materials ............................................................................................................................. 33 2.2.1 Oligonucleotides .......................................................................................................... 33 2.2.2 Reagents and Consumables.......................................................................................... 35 2.2.3 Droplet Digital PCR (ddPCR) and qPCR Reagents .................................................... 36 2.3 The (Second-Generation) Hi-Fi SELEX Protocol .............................................................. 36 2.3.1 Library Design and Synthesis ...................................................................................... 37 2.3.2 Target Immobilization ................................................................................................. 38 2.3.3 Partitioning and Retained Fraction Recovery .............................................................. 41 2.3.4 Retained Pool Amplification by ddPCR ...................................................................... 44 2.3.4.1 Determination of the Concentration of the Retained Library CLibrary .................... 45 2.3.4.2 Droplet-Based Amplification of the Retained Pool ............................................... 47 vi  2.3.5 Measuring the Sequence Diversity of the Amplified Retained Pool ........................... 50 2.3.6 Enzymatic Regeneration of the Single-Stranded Library ............................................ 54 2.3.7 Purification and Recovery of the Regenerated ssDNA Library ................................... 55 2.3.8 Measuring the Mean Kd of the Retained and Regenerated Library ............................. 56 Chapter 3: Discovery of DNA Aptamers by Hi-Fi SELEX and Their Evaluation as Ligands for Preparative Affinity Chromatography ............................................................................... 61 3.1 Introduction ......................................................................................................................... 61 3.1.1 Human Factor D ........................................................................................................... 63 3.1.2 Human mesothelin ....................................................................................................... 63 3.2 Methods and Results ........................................................................................................... 64 3.2.1 DNA Aptamer Discovery by Hi-Fi SELEX ................................................................ 64 3.2.2 Sequencing and Characterization of Final Retained Pools .......................................... 68 3.2.3 Preliminary Evaluation of Candidate Ligands ............................................................. 72 3.2.4 Candidate Aptamer Stabilization Against (exo)Nuclease Catalyzed Degradation ...... 78 3.2.5 Proof-of-Concept Testing of Stabilized Aptamer Columns on CHO Feedstocks ....... 80 Chapter 4: Conclusions and Proposed Future Work .......................................81_Toc507009849 4.1 Summary of Thesis Work ................................................................................................... 81 4.2 Proposed Future Work ........................................................................................................ 83 4.2.1 Rapid Screening and Discovery of Effective Eluents .................................................. 83 4.2.2 High-Throughput Methods to Evaluate and Improve the Stability of Aptamers......... 85 4.2.3 A Method for Determining Ligand Specificity ............................................................ 86 References .................................................................................................................................... 88 Appendix: A New Algorithm for Defining Multicomponent Binding of DNA Aptamers During their Selection by Hi-Fi SELEX ................................................................................. 100 A.1 Overview .......................................................................................................................... 100 A.2 Model Development and Underlying Theory .................................................................. 102 A.3 Experimental Methods ..................................................................................................... 111 A.4 Application of the Algorithm to Adsorption Isotherm Data for Human Thrombin......... 117 A.5  Raw MATLAB Code ...................................................................................................... 123    vii  List of Tables Table 3.1: Sequences of ssDNA Hi-Fi SELEX library and reverse complement sequences ....... 65 Table 3.2: Round-by-round mean dissociation constants (Kd) and standard deviations measured by duplicate qPCR experiments for enriched pools selected against various therapeutic targets using Hi-Fi SELEX. ...................................................................................................................... 65 Table 3.3: Core random-region sequence and binding affinity (Kd) of some of the highest frequency members recovered from the retained pool after Round 3 of Hi-Fi SELEX applied to DNA aptamer selection against human Factor D. ........................................................................ 69 Table 3.4: Core random-region sequence and binding affinity (Kd) of some of the highest frequency members recovered from the retained pool after Round 3 of Hi-Fi SELEX applied to DNA aptamer selection against human mesothelin. ..................................................................... 70 Table 3.5: Average separation performance observed for the first cycle of a pre-packed 5-mL Profinity™ Epoxide mini-column bearing immobilized 5’-amino-aFD-30 aptamer.  Data are for n = 4 replicates, and basic column operating conditions are as described in Figure 3.9. ............ 78 Table 3.6:  Average separation performance observed for the first cycle of a pre-packed 5-mL Profinity™ Epoxide mini-column bearing immobilized modified 5’-amino-aFD-30-inverted-dT-3’ aptamer.  Data are for n = 2 replicates (errors are average errors), and basic column operating conditions are as described in Figure 3.9. .................................................................................... 80      viii  List of Figures Figure 1.1: A DNA aptamer – protein complex (Adapted from Thiel, 2004) ............................. 10 Figure 1.2: Conventional SELEX consists of four steps: binding, partitioning and elution, amplification, and regeneration (Schütze et al., 2011). ................................................................ 12 Figure 1.3: A typical separation scheme and chromatogram for affinity chromatography (GE Healthcare Bio-Sciences AB, 2007). ............................................................................................ 20 Figure 1.4: Schema of the Proposed Custom Chromatography Development Pipeline .............. 23 Figure 2.1: Schematic showing the sequence of operations comprising a single round of Hi-Fi SELEX. ......................................................................................................................................... 28 Figure 2.2: Basic structures of the semi-combinatorial DNA libraries used in standard SELEX (standard library) and Hi-Fi SELEX (competent library): Hi-Fi SELEX libraries differ from standard SELEX libraries through their use of novel fixed-region complements (blocking elements) to improve the functional diversity of the starting semi-combinatorial library.  Each member of the Hi-Fi SELEX ssDNA library is therefore comprised of an 80 nt library sequence containing a 40-mer random core region (N40) flanked by a 5’ universal 20-mer flanking sequence and a 3’ universal 20-mer primer binding sequence, with each flanking sequence hybridized to its complement, which are hereafter denoted as 5’-Comp and 3’-Comp, respectively.  By eliminating single-strand structures within the fixed regions, the competent Hi-Fi SELEX library isolates aptamer fold and function to within the variable core region of the library, while reducing artifacts that might compromise or eliminate the discovery of a tight-binding sequence within that region. Unblocked, the fixed-region sequences within a given selection library can interfere with aptamer fold and function through their potential to adopt stable secondary structures created through either 1) self-association, or 2) association with complementary nucleotides within the variable core region, the opposing fixed region, or both.  These types of unwanted structures can occur either within an individual library member or between complementary regions of different members of the library, and the net effect is to significantly reduce the total functional diversity of the library. .................................................. 34 Figure 2.3: Schematic of chemistries used to immobilize target protein, neutralize reactive groups, and passivate non-specific binding sites on the surface of a Nunc Immobilizer Amino plate. .............................................................................................................................................. 40 ix  Figure 2.4: Comparison of the percentage of a starting library that is non-specifically adsorbed to a standard SELEX library screening surface (“Conventional SELEX”; MyOne magnetic beads (ThermoFisher Inc.)) and to our passivated form (“Proposed Method”) of the Nunc Immobilizer surface.  In both cases, no target is displayed on the surface and reactive groups on the surface have been neutralized.  The data show that more than 2% of the starting library is non-specifically retained on the MyOne beads, while undetectable amounts (by qPCR analysis) are recorded for our system when 0.005% Tween 20 is present. ....................................................... 42 Figure 2.5: Comparison of amplification of a retained pool (105) of library members by conventional PCR (standard SELEX; upper panel) and by ddPCR (Hi-Fi SELEX; lower panel): Although the desired 80-bp dsDNA amplicon is created in conventional PCR, a maximum in its total abundance is typically reached after a limited number of cycles.  Beyond that cycle, various artifacts, including formation of 80-mer hetero-duplexes hybridized together through only their common flanking sequences, oligonucleotide stretches within the universal primer regions mis-priming certain variable core region sequences, and improperly extended products acting as spurious primers on heterologous sequences, promote conversion of the library to increasingly aberrant high molecular weight (HMW) by-products.  A small-scale “pilot” PCR reaction (Lou et al., 2009; Nieuwlandt, 2000) is therefore generally performed to determine the maximum number of PCR cycles that can be conducted before accumulating unacceptable amounts of by-products, which are known to adversely affect selection and must therefore be removed by gel electrophoresis or other means (Musheev and Krylov, 2006). That pilot reaction typically shows that standard PCR amplification of the retained pool must be stopped at ca.  22 to 25 cycles since HMW by-products generally start accumulating when the amplicon concentration reaches ca. 20-50 nM. Termination of amplification at this relatively low cycle number generally yields ~1010 to 1012 80-bp amplicons, or ≤ 1% of that needed to initiate the next selection round.  As a result, in SELEX, the PCR step must be multiplexed across 100 or more parallel reactions, each amplifying between ~ 105 to 106 library members to create 1011 to 1012 amplicons per well. The products of the parallel reactions are then pooled and concentrated to reach the concentration (i.e. 1014 amplicons in 100 µl) required for downstream processing and the next round of SELEX. The use of emulsions in ddPCR to isolate and amplify single templates by PCR is well established, and it is known that the resulting partitioning of single templates into individual droplets reduces formation of unwanted by-products when co-amplifying mixtures of templates x  (e.g. multiple genes) (Nakano et al., 2003; Williams et al., 2006). Spurious priming events are greatly reduced within each droplet, in part because competition between different templates and biases resulting from differences in amplification efficiencies are avoided (Margulies et al., 2005). Moreover, post amplification, the emulsions can be broken to recover the full set of amplicons in an aqueous phase suitable for downstream processing. In Hi-Fi SELEX, the pool (~108) of competent 80-nt ssDNA library members retained after a selection round is therefore partitioned among a similar number of nL-sized droplets. ddPCR partitions ca. 20,000 droplets per well, which means 100 wells are required to accommodate 50 templates into each droplet (CPD=50). As a result of the low sequence heterogeneity per droplet, minimal HMW by-products formation is observed over 40 or more ddPCR cycles, permitting high-fidelity end-point amplification of all retained library members into more than 1014 total copies of the desired 80-bp dsDNA amplicon products. ...................................................................................................... 46 Figure 2.6: Representative standard curve for determining sequence diversity.  Quantitation cycle (Cq) data for diversity standards from 109 unique sequences to 105 unique sequences plotted as green circles. The E-value reported represents the efficiency of the qPCR amplification, which should be between 95% and 105% for the standard curve to be accepted. CLibrary, the starting library concentration, may also be determined from the standard curve by taking the starting quantity (e.g. ~107; green X in figure) estimated using the measured Cq (qPCR quantitation cycle) and dividing by 5 µl and then multiplying by the dilution factor used to prepare the sample. ....................................................................................................................... 47 Figure 2.7: Determination as a function of average copies per droplet (CPD) of the maximum cycle number that may be employed to amplify the retained library by ddPCR without formation of unwanted HMW by-products. .................................................................................................. 48 Figure 2.8: Fluorescence (SYBR green) based melt analysis for serial reductions in the sequence diversity of dsDNA-amplicon representations of an 80-nt Hi-Fi SELEX library: The double stranded amplicons of retained library members are homogeneous in terms of their two flanking sequences, while presenting a highly diverse ensemble of variable core-region sequences.  Denaturation of the amplified library followed by cooling to 55 °C therefore results in two distinct dsDNA populations: fully homo-duplexed amplicons characterized by a Gaussian melting envelope centered at a relatively high melting temperature (Tm ~ 81 °C), and hetero-duplexes exhibiting only partial complementarity (typically through only their common flanking xi  sequences). The hetero-duplexed pool of amplicons collectively exhibits a Gaussian melting peak characterized by a much lower Tm (~ 67 °C). ....................................................................... 51 Figure 2.9: Standard curve for qPCR-based sequence diversity determination: normalized melt peak areas (A) and fHetDNA values (B) as a function library diversity. .......................................... 52 Figure 2.10: Efficient stoichiometric regeneration of the ssDNA library in Hi-Fi SELEX: (A) ddPCR amplification of retained members results in formation of fully complimentary homo-duplexed amplicons, while conventional bulk PCR used in SELEX yields a mixture of homo- and hetero-duplexed amplicons; (B) λ-exonuclease processing of the homo-duplexed amplicons produced by Hi-Fi SELEX results in stoichiometric recovery of the ssDNA library in 60 min, while relatively little ssDNA product is recovered from the bulk PCR product. ......................... 56 Figure 2.11: Comparison of yields (expressed as % recovery) of regenerated 80-mer ssDNA library material recovered by the standard method and the new protocol post phenol/chloroform extraction....................................................................................................................................... 57 Figure 2.12: Representative adsorption isotherm and regressed mean Kd value.  Isotherm data determined using the qPCR-based binding analysis method and then fit to equation 2.5: data shown are for a retained pool of binding library members (after 3rd round of Hi-Fi SELEX) with variable regions having high affinity for human mesothelin. ....................................................... 60 Figure 3.1: Mean binding isotherm and standard deviations measured by duplicate qPCR experiments for complexation of the retained library to Factor D after 3 rounds of Hi-Fi SELEX........................................................................................................................................................ 66 Figure 3.2: Mean binding isotherm and standard deviations measured by duplicate qPCR experiments for complexation of the retained library to mesothelin after 3 rounds of Hi-Fi SELEX…………………………………………………………………………………………...67 Figure 3.3: Pseudo-component histograms for Round 3 of Hi-Fi SELEX applied to DNA aptamer selection against human Factor D.  In round 3, the isotherm was best fit using two pseudo-components with Kdi values of Kd2 = 10-7 M and Kd1 = 10-9 M. ....................................... 67 Figure 3.4: SPR sensorgrams for binding of Factor D to immobilized biotinylated aFD-30 aptamer (Kd = 0.71 ± 0.25 nM).  Experiments were performed on a Biacore 3000 using an SA chip. Several concentrations of Factor D were injected, and the results fitted to a 1:1 Langmuir isotherm…………………………………………………………………………………………..70 xii  Figure 3.5: M-fold predicted secondary structures for the full-length blocked aptamer sequences against Factor D whose core random-region sequences are reported in Table 3.3.  Several of the candidate aptamers fold into bulge-hairpin motif (e.g. aFD-14), while others assume a bifurcated bulge structure (e.g. aFD-30). ....................................................................................................... 71 Figure 3.6: M-fold predicted secondary structures for the full-length blocked aptamer sequences against mesothelin whose core random-region sequences are reported in Table 3.4. ................. 72 Figure 3.7: Time course of the viable cell density and recombinant human Factor D concentration in the culture supernatant during batch cultures of the 13D CHO cells used to produce recombinant human Factor D. ......................................................................................... 74 Figure 3.8: Equilibrium adsorption isotherm at 20°C for binding of pure Factor D to the aFD-30 bearing stationary phase (AF buffer was the solvent for these binding experiments).  The data indicate a static binding capacity of 35 (± 3) mg mL-1. ................................................................ 76 Figure 3.9: Measured chromatogram for affinity capture and purification of recombinant human Factor D from clarified supernatant of a 13D CHO cell supernatant.  The pre-packed 5-mL Profinity™ Epoxide mini-column bore immobilized 5’-amino-aFD-30 aptamer at a density of 53 mg mL-1. Clarified supernatant containing 0.48 g L-1 Factor D that had been conditioned for affinity separation by diafiltering into 1X AF buffer (pH 7.4) was frontally loaded at                0.4 CV min-1 (CV = column volumes) to approximately 15% breakthrough.  The loaded column was washed with 5 CV of AF buffer and the captured protein was then eluted by an isocratic step gradient into DE buffer. The associated sodium dodecyl-sulfate polyacrylamide gel electrophoresis (SDS-PAGE) documentation of the separation is also shown, with lane 1 showing the MW standards, lane 2 the clarified diafiltered supernatant (SN), lane 3 the column flow through (FT), and lane 4 the pooled eluent peak fraction (EL).  Data are for a previously unused column. ............................................................................................................................. 77 Figure 3.10: Dynamic binding capacity (DBC) as a function of column cycle number (n = 2). 79   xiii  Nomenclature A A280 nm  A67°C  A81°C  AF  aFD  aMT  bp  C  CE-SELEX  CF  Cfree   CHO  CIP  CLibrary  CPD  Cq  CV  DBC  ddPCR  DE  DHFR  DNA  DSP  E. coli  EL  ELISA  EMA  FDA  Adenine Absorbance at 280 nm; Au Area beneath melt curve centred at 67 °C Area beneath melt curve centred at 81 °C Aptamer folding buffer Anti Factor D aptamer Anti mesothelin aptamer Base pairs Cytosine Capillary electrophoresis SELEX Concentration factor Concentration of library remaining in the solution phase Chinese Hamster Ovarian cells Clean-in-place Library concentration; sequences/µL Copies per droplet Quantitation cycle Column volumes Dynamic binding capacity Droplet digital PCR Denaturing elution buffer Dihydrofolate reductase  Deoxyribonucleic acid Downstream processes Escherichia coli Elution peak fraction Enzyme-linked immunosorbent assay European Medicines Agency US Food and Drug Administration  xiv  fhetDNA  FP FT  g  G  GMP  Hi-Fi SELEX  His-tagged  HMW  HPLC  HT  Is   ITC Kd   mAb  M-SELEX  MW  MWCO  N40   NMR  nt  NTU  PCR  PE  PEG  qPCR  QSAR  RNA  RP  RT-PCR  SBC  Heteroduplex fraction  Forward primer Flow-through  Gravitational force Guanine Good Manufacturing Practices High-Fidelity SELEX Histidine-tagged High molecular weight High-performance liquid chromatography High throughput Specificity Index  Isothermal titration calorimetry Equilibrium dissociation constant Monoclonal antibodies Microfluidic SELEX Molecular weight Molecular weight cut-off 40-mer random core region Nuclease magnetic resonance spectroscopy Nucleotide Nephelometric Turbidity Unit Polymerase chain reaction Partition efficiency Polyethylene glycol Real-time quantitative PCR Quantitative structure-activity relationships Ribonucleic acid Phosphorylated reverse primer Reverse transcriptase PCR Static binding capacity  xv  SDS-PAGE  SELEX SN  SP  SPR  SPRi  ssDNA  ssRNA  SW  T  TE  TI  Tm  USD  USP  VH  VL  θ  µFFE  10K  3’ dT  3’-Comp  3K  5’-Comp  λ-exonuclease  2      Sodium dodecyl sulfate polyacrylamide gel electrophoresis Systematic Evolution of Ligands by Exponential Enrichment Diafiltered supernatant Surface passivation buffer Surface plasmon resonance Surface plasmon resonance imaging Single-stranded DNA Single-stranded RNA Stringent wash buffer Thymine Tris-EDTA buffer Target immobilization buffer Melting temperature; °C Ultra-scale down Upstream processes Variable domain of the heavy chain  Variable domain of the light chain Bound fraction  Micro free-flow electrophoresis 10 kiloDalton 3’ inverted deoxythymine nucleotide 3’complementary blocker 3 kiloDalton 5’ complementary blocker Lambda exonuclease Chi-squared error      1  Chapter 1 Thesis Motivation, Objectives and Background 1.1 The Challenge to Downstream Processing an Ever-Expanding Biologics Portfolio Therapeutic proteins either recovered from natural sources or produced using recombinant DNA and cell culture technologies now represent a more than $400 billion CAD (Lindsley, 2017) per annum market in North America alone.  Human monoclonal antibodies (mAbs) produced by mammalian cell culture are the largest class of therapeutic proteins, and highly efficient and robust platforms for purifying mAbs from culture supernatants are available and used within most if not all major biotechnology and pharmaceutical companies engaged in mAb production (Gronemeyer et al., 2014; Jacquemart et al., 2016; Kunert and Reinhart, 2016).  Those platforms typically capture mAb products from clarified culture supernatants using affinity chromatography with immobilized protein A, or an engineered variant of protein A, serving as the affinity ligand (Arora et al., 2017; Koguma et al., 2013; Pabst et al., 2014; Tsukamoto et al., 2014).  Protein A binds mAbs with both high affinity and high specificity, enabling rapid purification and concentration of the desired product such that only a modest amount of further downstream processing of the product is required. However, the diversity of non-mAb therapeutic proteins is growing, as is the diversity of host organisms used to produce those proteins.  One such example is provided by the growing range of non-mAb biologic orphan drugs either under development or currently being used for the treatment of rare diseases. In the US, more than 25 million people are afflicted with one of more than 7,000 diseases that are considered rare (Hall and Carlson, 2014); numbers of rare disease cases within the European Union are similar (Hughes-Wilson et al., 2012). The relatively low frequencies of these often life threatening health conditions in both jurisdictions present enormous challenges to the development and economical production of protein therapeutics to treat them (Hughes-Wilson et al., 2012).  Several factors contribute to this challenge, beginning with the fact that current pipelines used within industry to discover and develop non-mAb therapeutics, and then to design and validate the processes used to manufacture those complex molecules often take one to two years to complete, delaying product launch and reducing return on investment.    2  Fueled by major initiatives such as the Precision Medicine Initiative (Collins, 2015), ‘omics science (genomics, proteomics, metabolomics, etc.) is rapidly discovering genes harboring polymorphic or somatic variations, and altered gene products arising from them, that contribute to human disease. This is enabling clinicians to more accurately sub-classify major diseases, including many forms of cancer, on the basis of underlying genetic and gene-product variations, and to then create tailored therapeutics, including protein therapeutics, that specifically address the root cause of disease sub-classes. Precision medicine therefore offers tremendous opportunities to improve health and health outcomes, but the challenges to its effective implementation are significant as well. Adoption of precision medicine is currently restricted by lack of a comprehensive strategy to deliver its benefits to patients. That strategy must include ways to rapidly and cost-effectively develop and manufacture new targeted therapies that improve treatment efficacy, and to optimize that treatment by steering patients to the right drug at the right dose at the right time.  Patient populations for many disease sub-classes are likely to be relatively small, again presenting the challenge of how to effectively and economically produce protein therapeutics specifically designed to treat those afflictions. As evidenced by these two examples, production levels and profit margins on many future protein therapeutics are likely to be lower than for the so-called “blockbuster” biological drugs (Sharma et al., 2010) currently approved to treat major diseases. As industry expands their operations toward lower-demand products, it now seeks, with some urgency, new strategies to address time-to-market pressures, including ways to rapidly design and optimize non-mAb production and purification trains that are economical and robust. Progress toward this ambitious goal is being realized, most notably through advances in upstream (cell culture) processing to produce protein therapeutics at ever higher titres.  Mechanical considerations, such as the design of an impeller, have been addressed to greatly improve the quality and yield of biologic products produced by cell culture. For example, advanced bioreactor modeling and simulation tools have enabled industry to optimize impeller designs to achieve improved gas hold-up and oxygen mass transfer in large-scale cultures (Rios, 2012).  Likewise, advances in perfusion culture processing, which involves constant feeding of fresh media and removal of spent media and products in growing cultures of viable host cells (Challener, 2016), have enabled industry to achieve high cell densities and product titres in bioreactors of relatively small volume and capital cost (Bonham-Carter and Shevitz, 2011; Rios, 2012).  3  In addition, an ever-widening range of organisms are being genetically tailored and used as hosts for cell-culture based production of biologics.  This includes highly engineered strains of Escherichia coli. Many therapeutic proteins are produced in E. coli, including human insulin, a hormone used to treat Type I and Type II diabetes (Jozala et al., 2016), and human interferon α2-beta (IFN- α2b), used to treat melanoma (Tarhini et al., 2015). Past strains of E. coli employed by industry were limited in their ability to produce proteins of high molecular weight or requiring complex post-translational modifications, as natural strains of E. coli cannot glycosylate proteins and are poor at properly folding proteins of higher molecular weight or containing multiple disulfide bonds.  But successes in re-engineering strains of E. coli to overcome these production limitations are being realized, as are improvements to many other prokaryotic and eukaryotic hosts (Jostock and Li, 2010; Khan, 2013; Li et al., 2010; Scott, 2011) (Khan, 2013). Indeed, production yields of therapeutic proteins have improved roughly 100-fold over the past three decades (Jostock and Li, 2010) as a result of a better understanding of gene expression, metabolism, growth, and apoptosis delay in various host cells. The exceptionally high cell densities that can now be realized in production-scale cell cultures are understandably motivating innovations in many areas of downstream processing, beginning with technologies to rapidly and reliably separate (and possibly recycle to the bioreactor) host cells from culture supernatant so as to recover the product-rich component requiring further processing. For example, if a protein product is secreted into the culture supernatant, clarification to remove all cells, cell debris and other suspended materials is required.  However, the performance of standard operations used to do this (e.g., centrifugation, various modalities of tangential flow filtration such as alternating tangential flow filtration) is often compromised when applied to feed streams bearing very high cell densities (Jacquemart et al., 2016; Le Merdy, 2014; Minow et al., 2014).     Following cell and crude product harvesting and clarification operations, platforms for producing mAb-based therapeutics typically capture the product using Protein A affinity chromatography.  Protein A columns, in use since the 1960s, have proven tremendously effective in capturing mAb therapeutics at purities often above 99% (Curling, 2007). Their success recognizes the intrinsic value in employing chromatography operations specifically designed to provide optimal purification factors and yields for a particular protein product (in this case a mAb), as opposed to using more versatile modes of chromatography (e.g., ion exchange,  4  hydrophobic interaction) designed to provide acceptable purification performance for a wide range of products and feedstocks. The platform-based production of mAbs using CHO cell cultures and Protein A-based affinity capture has therefore proven to be a tremendously effective combination that assures process developers a robust and scalable manufacturing process that can be validated and approved by the US Food and Drug Administration (FDA) and/or the European Medicines Agency (EMA).  However, as noted, many protein therapeutics in development pipelines are not mAbs. These include various protein-based vaccines, hormones, enzymes, cytokines, growth factors, and recombinant plasma proteins (Scott, 2011). Within current industrial business models, the production of these biomolecules at large scale is often deemed to be risky, due in part to uncertainties over regulatory approval coupled with anticipated high operating costs and low return on investment. Growth of these product classes could therefore benefit from leveraging, where appropriate, common and accepted unit operations and processing steps used in mAb downstream processing platforms. Crucial to the success of mAb downstream processing platforms is the recognized importance of early capture and concentration of the product using affinity chromatography.   Establishing an integrated set of products and decision-making tools that enable industry to rapidly create an affinity capture step that efficiently purifies and concentrates the target, while preserving product yield and integrity, could enable one to translate many of the benefits of mAb production platforms to the purification of non-mAb therapeutics. This concept is challenged by the fact that affinity capture columns used in preparative manufacturing are subjected to harsh elution conditions (e.g. high salt concentration, low pH) and clean-in-place (CIP) procedures, in which some ligands and protein product can be denatured or destroyed (Peyrin, 2006). Moreover, the use of affinity chromatography ligands at large scales is, in general, an expensive proposition due in part to the challenge to their discovery, the often significant cost of the affinity ligands so discovered, and the generally limited lifetime of the column. This problem is especially acute for biologic orphan drugs and other therapeutic proteins in development pipelines for which production per annum is less than a few hundred kg, as the cost of developing and using the affinity capture column is then distributed across a smaller number of therapeutic units sold.  Progress in this area will therefore require the development of technologies that enable the rapid discovery and implementation of relatively inexpensive  5  custom (i.e. product-specific and selective) affinity ligands that prove robust and stable in capturing, concentrating and purifying the desired target over multiple column cycles.   1.2 Thesis Objectives  Though the concept of protein purification by customized affinity chromatography has been considered by a small collection of researchers over the past couple of decades, to date Protein A media remains the only widely employed example, having served as the primary means of purifying mAb products for over 30 years (Curling, 2004b).  This gives evidence to the current view within industry that a new approach to operationalizing the custom-chromatography concept is required if its benefits and widespread implementation are to be realized. The primary objective of this thesis work therefore was to design a new technology and associated pipeline for the discovery, development and validation of product-specific affinity media using DNA aptamers as ligands. The feasibility and performance of the aptamer discovery method and the pipeline will be tested through application to the development of effective affinity chromatography media against one or more model proteins. The following research goals were accomplished in order to achieve the main objective of the thesis: • Identify and address technical limitations to current DNA aptamer selection methods to establish a new technology that enables the rapid and robust discovery of aptamers suitable for use as affinity ligands  • Validate that new aptamer ligand discovery technology by applying it to the discovery of DNA aptamers against one or more protein target(s) • Sequence, produce and characterize candidate aptamers offering a mean equilibrium dissociation constant (Kd) in the range of 0.1 nM to 10 nM. • Couple candidate DNA aptamers to preparative chromatography media widely used in industry (e.g. BioRad UNOsphere epoxide and/or UNOsphere diol media) • Perform column screening studies to measure dynamic binding capacities, purities, yields, concentration factors when applied to product containing feedstocks • Use the collective data set to evaluate at the proof-of-principle level the merits of the proposed DNA aptamer based custom affinity chromatography platform.   6  1.3 Review of Relevant Literature 1.3.1 Proposed Ligands and Technologies to Enable Custom Affinity Chromatography Screening of semi-combinatorial peptide and synthetic small-molecule libraries has been pursued both by academic researchers (Camperi et al., 2014; Fields et al., 2016; Huang and Carbonell, 1999; Menegatti et al., 2013; Sousa and Taipa, 2014; Wang et al., 2005) and industry (Williams et al., 2014) as a strategy for discovering ligands against non-antibody protein targets.  Peptide-based ligands are a class of affinity ligands that may be discovered using high-throughput semi-combinatorial techniques, including screening of immobilized synthetic peptide libraries, as well as peptide libraries generated by phage display or ribosomal display. In the phage display screening method, a suitable host organism is infected with a bacteriophage, most often E. coli-specific bacteriophage M13, that is engineered so that each phage within the population generates a unique exogenous DNA sequence and surface displays the translated peptide sequence to permit a large library of sequences to be presented across the entire infected cell population (Smith, 1985). The resulting library of peptides are screened by equilibrating with the desired target, removing those phage that do not bind the target, and enriching the pool of peptide ligands showing affinity for the target (John Maclennan, 1995). Though it can be effective, screening of peptide libraries generated through phage display is labor intensive.  Moreover, the diversity of starting library is often fairly limited, often lying between 107 and 1010 unique peptide sequence, due in part to the inefficiency of vector transformation into the host cells (Schaffitzel et al., 1999).  Significantly higher library diversity can be realized using an alternative biological method, ribosomal display. By leveraging cell-free E. coli S30 coupled transcription/translation systems, ribosomal display technology can be used to generate and screen libraries comprised of up to 1012 unique sequences (Coia et al., 2001).  It is a powerful approach, but is operationally challenged as a means to rapidly discovery affinity ligands for custom chromatography applications by the fact that the enrichment process is very slow (weeks to months) and the final pool of enriched candidate peptide ligands generally retains high sequence diversity (usually higher than 10,000). A relatively large number of individual peptides ligands within that population must therefore be synthesized for further evaluation to identify those suitable for the intended application (Mattheakis et al., 1994).  7  Despite their recognized limitations, both phage display and ribosomal display methods have proven useful as general methods for discovering high-affinity peptide ligands against various protein targets.  Many high affinity peptides have been discovered by each method.  The same is true for various in vitro combinatorial methods that leverage the ability to assemble diverse libraries of short peptides. The most popular way of generating such libraries is through the ‘one-bead, one peptide’ approach (Lam et al., 1991) in which a unique peptide is synthesized on a given bead by surface-initiated polymerization. Once the peptide synthesis process on the beads is complete, the resulting library of peptide-displaying beads is pooled.  Beaded libraries displaying peptides 5 to 6 amino acids in length can be achieved using this approach to generate a diversity often around 106 unique members (Houghten et al., 1991). Libraries comprised of longer peptides have proven more difficult to create due to inefficiencies in the peptide-synthesis reactions used, which can increase for peptides longer than ~ 8 amino acids because folding may then occur and interfere with further amino-acid addition (Perret et al., 2015). But despite these limitations, peptide ligands have been used for the purification of immunoglobulin, blood factors, and therapeutic enzymes with purity exceeding 90% (Menegatti et al., 2013; Yang et al., 2009), albeit only at laboratory scale and never for actual production.  It is also important to note that, on occasion, peptide-based ligands have proven highly selective and show high chemical stability and structural diversity, but they generally suffer from low binding affinities/capacities, limited life cycles, and low scale-up potential (Lowe, 2001).  Synthetic libraries of small molecules have also been explored as a potential platform for affinity ligand discovery.  This approach was inspired, at least in part, by the high affinity of the Cibacron Blue 3GA dye ligand (discovered in 1973) for many proteins and enzymes.  Given the triazine dye like structure of Cibacron Blue, this led to the hypothesis that screening of large synthetic libraries of triazine dye variants may yield ligands with high affinity and specificity for products of interest (Curling, 2004a). Initial efforts were only modestly successful, with the ligands discovered exhibiting relatively poor performance when applied to protein purification by affinity chromatography (Dean and Watson, 1979).  This has led to the development more diverse synthetic ligand libraries based on a generalized triazine scaffold that offers two substitution positions for the attachment of a combination of functional groups such as amino acid/peptide, mimetic, short aliphatic, and cyclic moieties (Clonis, 2006). While evidence  8  suggests this has resulted in discovery of some ligands offering higher specificity or affinity, it regrettably has not resulted in the adoption of the platform by industry (Menegatti et al., 2013).   Motivated in part by the limited success of these approaches, researchers have begun exploring the potential value in creating tools and methods that adopt a hybrid approach that uses theoretical considerations to define the type of semi-combinatorial library to be synthesized and screened. There are three methods commonly used for rational design and modelling of synthetic ligands. The first involves exploiting pre-existing knowledge of the interaction of the target protein for its natural ligand(s) to qualitatively define library chemistries that might work. This knowledge may include X-ray crystallography, NMR, and homology structural data of a natural protein-ligand interaction (Dias and Ciulli, 2014; Turnbull and Emsley, 2013). The second approach is to exploit molecular mechanics or QSAR (quantitative structure-activity relationships) simulation software to design putative ligands that show complementarity to a selected cluster of exposed residues on the target protein (Vilar and Costanzi, 2012). This method, which at present is largely an open research question, could prove useful in cases where there are insufficient structural data describing the protein - natural ligand interaction.  In both of these approaches, the results are used to refine (bias) the type of semi-combinatorial library produced and screened on the solid phase.  In the last approach, the ligand is designed in silico by direct mimicking of natural biological recognition interactions (Lowe et al., 2001). Screening of the biased ligand library is conducted to isolate ‘lead ligands’, which may then be synthesized and immobilized on affinity adsorbents for further evaluation. If necessary, the rational design is refined and the selection repeated using the isolated ‘lead ligands’ (Clonis, 2006). Small synthetic ligands against some important human proteins (e.g. immunoglobulin G and insulin precursor) have been discovered using these various approaches (Sproule et al., 2000; Teng et al., 1999). Synthetic ligands have also been used for product polishing after Protein A-chromatography to capture leaked Protein A (Lund et al., 2012), as the presence of Protein A can be immunogenic when mAb is administered as a human therapeutic.  Synthetic ligands are significantly cheaper to produce and can offer good resistance to chemical and biological degradation, but the chemical diversity of those libraries is quite limited, and no ligands discovered by this approach have been used for product manufacturing (Clonis et al., 2000; Lund et al., 2012). Selectivity remains a problem. Additionally, peptide and synthetic ligands are prone to be toxic and to trigger an immunogenic response in humans (Menegatti et  9  al., 2013). Additional unit operations post affinity chromatography must therefore be introduced in the purification train to ensure any and all leaked ligands are removed. In downstream processing, each additional unit operation translates to additional operating costs and reduced purity and yield of the target of interest.  A need therefore exists for a new class of non-toxic non-immunogenic ligands that offer large chemical diversity, can be rapidly screened to discover members offering high affinity and specificity for a protein target, can be chemically modified to ensure ligand stability during harsh bioprocessing operations, and can be immobilized on proven preparative chromatography media to create robust and economical capture columns for therapeutic protein manufacturing.  The central aim of this project is to design a new pipeline to rapidly develop and validate affinity capture columns against specific orphan drugs that uses DNA aptamer based ligands for the purification.   1.3.2 Aptamers Aptamers are single-stranded RNA or DNA oligonucleotides, typically 7 – 30 nucleotides in length, that bind a target molecule (e.g. a protein) with very high affinity and specificity.  Due to their single stranded form, RNA or DNA aptamers are able to fold into specific and stable three-dimensional (tertiary) structures based on their nucleotide sequence and the resulting set of intramolecular interactions formed between the nucleotide bases within the strand (Gold, 2015). Through their tertiary structures, aptamers can interact with and bind a target.  As with protein-protein interactions, the complex formed between a target protein and its aptamer ligand can include electrostatic interactions, hydrogen bonding, van der Waals forces of attraction, and/or hydrophobic interactions.  One example of a protein - DNA aptamer complex is shown in Figure 1.1.   10   Figure 1.1: A DNA aptamer – protein complex (Adapted from Thiel, 2004)  Aptamer identification began in 1990 through the creation of an iterative selection scheme known as Systematic Evolution of Ligand by Exponential Enrichment, or SELEX (Ellington and Szostak, 1990; Tuerk and Gold, 1990).  SELEX operates as an in vitro selection method based on immobilizing the target of interest to a suitable solid support and then repeatedly screening a semi-combinatorial library of candidate aptamer sequences against the target.  Only those oligonucleotide sequences within the library that bind the target are retained and enriched in each selection round.  The first SELEX experiment isolated two 8-mer oligonucleotide hairpins showing affinity for a selected target from a semi-combinatorial pool of ~65,000 random 8-mer sequences. The length of the randomized domain has since been expanded to 40 (typically) or even 50 nucleotides to greatly increase the diversity of the starting library. These larger systems and associated studies on them have shown that random libraries at least 30 nucleotides in length are sufficient to generate a wide diversity of stable tertiary structures, including hairpins, G-quartets, bulges, and pseudo-knots (Gold et al., 2012). The lengthening of the random region can therefore greatly increase the likelihood of isolating a highly effective aptamer against a desired target. In theory, a library of 40 randomized nucleotides should create 1024 unique sequences with a total mass of ~50 kg of oligonucleotides. In practice, however, the production of a full combinatorial library of 1024 members is not straightforward using commercially available DNA synthesis machines (which can generally make up to at most 1020 oligonucleotides). Fortunately, a full combinatorial library is not necessary for a SELEX experiment because an effective ligand is generally found at a frequency of between 1 in 109 to 1 in 1013 sequences within a starting 40-mer library (Gold, 1995). Thus, a  11  semi-combinatorial library comprised of 1014 to 1015 unique 40-mer oligonucleotides is sufficient.   1.3.3 SELEX and Related Technologies for Selecting DNA Aptamers  Aptamer discovery by SELEX was first described and performed in 1990 (Ellington and Szostak, 1990; Tuerk and Gold, 1990). The general principles of SELEX can be defined by dividing the procedure into four steps (Figure 1.2): binding, partitioning and elution, amplification, and regeneration. In the binding step, the target of interest is immobilized onto a solid support and a library of 1014 to 1015 unique oligonucleotide sequences (the starting library) is introduced for incubation. The quantity of immobilized target is usually set about 100-fold below the quantity of the starting library. Following library equilibration with the immobilized target, a partitioning and elution step is conducted, whereby weakly or unbound library members are washed away with controlled stringency, and bound members are then eluted and recovered from the aptamer-target complex. The retained members are amplified by subjecting them to a polymerase chain reaction (PCR) that serves to restore the starting quantity, albeit with a smaller diversity, resulting in ~ 105 to 106-fold enrichment of each retained sequence. The PCR product is in double stranded form, making regeneration of single stranded library necessary prior to the next round of selection. Several different strategies have been used for this purpose (Citartan et al., 2012; Civit et al., 2012; Liang et al., 2015). When it works, this cyclical screening process is repeated a sufficient number of times to isolate a small pool (often several thousand) of enriched candidate aptamer sequences.  In general, the mean binding affinity of the pool of retained aptamers to the target is too weak to be measured during the first 3 or 4 rounds of selection.  In later rounds, the mean binding dissociation constant, Kd, can usually be determined using an appropriate method such as surface plasmon resonance (SPR) or possibly real-time PCR (qPCR). A mean Kd in the nanomolar range is sufficient to terminate the selection process.    12    Figure 1.2: Conventional SELEX consists of four steps: binding, partitioning and elution, amplification, and regeneration (Schütze et al., 2011).  Although conventional SELEX is a relatively straightforward method, it often takes up to 20 rounds of selection to isolate a pool of enriched aptamers having a nanomolar mean Kd.  When combined with cloning, sequencing and characterization of the individual sequences within the retained pool, a typical SELEX experiment often takes approximately 3 to 6 months to complete (Jayasena, 1999).  This may be acceptable in terms of a basic research method, but it clearly does not provide the throughput (aptamer discovery in less than a month) required by industry for affinity ligand discovery. But more important, and damaging, is the fact that the SELEX often does not result in the discovery of a useful (high affinity, high specificity) ligand.  Failure rates of 50% or more are generally observed (Famulok and Mayer, 2014; Ozer et al., 2014; Zhuo et al., 2017). Understandably, improvements to the SELEX procedure have therefore been proposed.  The most commonly employed will be briefly described in this section. They include capillary electrophoresis SELEX, magnetic-bead-based SELEX, Cell-SELEX, and in vivo SELEX.  13   Capillary electrophoresis SELEX (CE-SELEX) works in a similar manner as conventional SELEX, with the key changes being that the target is not immobilized and the partitioning step is achieved via CE. Specifically, following equilibration of the library with the target in solution, CE-SELEX separates the resulting aptamer-target complexes from unbound library members based on differences in their electrophoretic mobility (Mendonsa and Bowser, 2004). Oligonucleotides that bind to the target have lower mobility due to their larger size and lower charge density.  Unbound library members therefore elute from the capillary first and may be directed into a waste stream. CE-SELEX eliminates the usage of immobilization and associated linkers required in conventional SELEX. In theory, this might serve to improve the interaction between library members and the target. However, the true benefits, if any, of CE-SELEX remain unclear, as the method has not proven generally capable of delivering high partitioning efficiencies, possibly because the migration of oligonucleotides in the capillary is difficult to control.  A modified version of CE-SELEX aimed at improving the partitioning efficiency (PE) has therefore been proposed which uses micro free-flow electrophoresis (µFFE) (Jing and Bowser, 2011). Increases in PE (which is given by the ratio of the mass of library members removed over the mass of library members retained) have been reported, but the method requires fabrication of a specialized microfluidic µFFE device that is not commercially available, thus limiting its applicability in industrial settings. In addition, when large targets are used, a large number of selection rounds are required due to the poorer efficiency of the technique under these conditions.  Magnetic bead-based SELEX is also another variation, arguably the most popular, of SELEX. Here, the target is immobilized on magnetic beads instead of solid support (Bruno, 1997), with the idea that it should permit more stringent washing (without loss) of the bound material following equilibration. After incubating the oligonucleotide library with the target, a magnet is used to separate the aptamer-target complex from the unbound library members. The PCR amplification is then conducted while the washed retained library is still bound to the target-loaded beads.  Although high affinity aptamers have been successfully isolated using this method, there are challenges that limit the efficiency of this platform. In part because the retained library is highly heterogeneous in term of sequence, significant amounts of unwanted amplification by-products are generated with increasing number of PCR amplification cycles, resulting in the necessity to stop the amplification at a low cycle number. Only about 1% of the  14  required number of amplicons is therefore produced per PCR reaction. As a result, hundreds of parallel PCR reactions are required to expand the retained library to the quantity required for the next round of selection.  Improvements to magnetic-bead SELEX have been realized through the creation of microfluidic chips that allow for retention and stringent washing of the magnetic beads.   Known as microfluidic SELEX or M-SELEX (Lou et al., 2009), the method is able to achieve high partition efficiencies using the continuous-flow magnet-activated microfluidic-chip-based separation (CMACS) device. However, the CMACS device is not a commercial product and is difficult to construct and reliably operate. It demands considerable expertise in order to minimize aggregation of magnetic beads within the microchannel that would subsequently result in low purity and recovery of aptamers. At present, it is also expensive, requiring ~$10,000 per selection run.  But it is a potentially promising approach, as M-SELEX, when coupled with next generation sequencing (NGS) (Cho et al., 2013), is able to isolate high quality aptamers within 4 selection rounds. It is, arguably, the most advanced SELEX technology described to date, with its key limitations being the specialized nature of the required equipment, the expertise required to fabricate and operate that equipment, and the high cost associated with using the platform.   Other (and far less frequently employed) variations of SELEX include Cell-SELEX and in vivo SELEX. Cell-SELEX is not really an improvement of SELEX, but rather an extension of the technology.  It uses eukaryotic or prokaryotic cells to display the target on the cell surface. It is particularly well suited to the discovery of aptamers against membrane-bound proteins (receptors, etc.) since all post-translational modifications are left intact (Sefah et al., 2010). Cell-SELEX was first performed in 1999 to isolate RNA aptamers that function as biomarkers of pathogenic parasites (Homann and Göringer, 1999). It is likewise attractive in the field of cancer research and therapy, as the isolated aptamers can be used as probes capable of differentiating between tumour cells and normal cells. But Cell-SELEX is not an ideal platform for the general selection of aptamers. Non-specific binding (and thus unwanted retention) of library members is common due to the complexity/heterogeneity of the cell surface. Screening using a complex cell surface therefore requires additional counter selection steps using control cells that do not display the target (Mayer et al., 2010). The process is therefore inefficient and prone to failure.  Finally, in vivo SELEX isolates aptamers against a desired target presented within a living organism. This platform is also attractive in the field of cancer research and therapy. Its  15  major advantage is the exposure of a library to targets in their natural physiological state (Cheng et al., 2013). The method has resulted in in vivo selection of aptamers, although many negative selection rounds have to be performed in order to achieve this outcome. In vivo SELEX typically require more than 10 rounds of selection to obtain high affinity aptamers (Cheng et al., 2013; Mi et al., 2010), and loss of functional aptamers through non-specific binding is a constant problem. The use of animals requires proper expertise as aptamers have to be extracted from the harvested organs, thus limiting the applicability of in vivo SELEX in ordinary laboratories.  Currently available SELEX methods therefore share a few common limitations. One is non-specific binding of library members to the display surface, which subsequently results in poor partitioning efficiencies. A low PE leads to the necessity of a high number of selection rounds, and all of these selection methods often fail if more than 10 rounds of selection are required. Second, upon elution, retained members must be amplified using PCR. PCR is a highly efficient enzymatic reaction when used to amplify a single oligonucleotide sequence using a set of primers specific to that sequence. However, it can behave quite differently when amplifying a highly heterogeneous population of retained sequences using a common pair of amplification primers (Ouellet et al., 2014). While the PCR amplification proceeds as desired for the first few (10 or so) amplification cycles, continuing the amplification for a larger number of cycles results in the formation of unwanted reaction by-products (shown in this thesis) which can destroy the integrity of the library. This is due to partial complementarity between amplified members that enables creation of high molecular weight by-products at the expense of the desired enriched aptamer product (Musheev and Krylov, 2006). All the current SELEX platforms therefore stop the PCR prior to the unwanted conversion event, with the trade-off being the recovery of a low quantity of amplified library members. In order to proceed to the next round of selection, hundreds of parallel reactions need to be done to gather enough material.   This highlights the need for a new DNA-aptamer selection platform that is cheap, reliable, and can rapidly isolate high quality aptamers using technologies that are readily available in an industrial laboratory.     16  1.3.4 Technical (Non-Therapeutic) Applications of Aptamers In principle, aptamers and their discovery by SELEX comprise a promising platform for general discovery of affinity ligands suitable for a variety of technical applications, including diagnostic systems and affinity purification of proteins at manufacturing scales.  As an in vitro selection process, SELEX can be operated under non-physiological conditions, creating the flexibility to optimize the library-to-target ratio during rounds of selection and to screen for aptamers at any desired buffer composition, pH, and/or temperature (Kökpinar et al., 2011). In comparison to peptide-based ligands and synthetic ligands, aptamer based ligands can also be easily modified to enhance their stabilities, including against the harsh chemical reagents (e.g. 1M NaOH) used in column CIP procedures preferred by industry. Once discovered in the laboratory, aptamers may be produced at required scales by robust and inexpensive chemical synthesis processes that show little to no batch-to-batch variation. Finally, because they are short sequences of natural DNA or RNA, aptamers tend to be non-toxic and non-immunogenic, and readily re-fold following their denaturation. Functional aptamers can be regenerated within minutes by simply changing the buffer composition, pH, and temperature back to native conditions (Jayasena, 1999). Aptamers may be composed of single stranded RNA, DNA or (in principle) a mixture of those nucleotides. Aptamers containing RNA tend to be less chemically stable.  They are also more susceptible to nuclease degradation (Jayasena, 1999).  For manufacturing applications, DNA aptamers represent a more attractive option since the deoxyribose sugar in DNA has C-H bonds, which are less reactive than the corresponding C-OH bonds found in the ribose sugar of RNA. Moreover, starting libraries used for SELEX are generally prepared first in the form of single-stranded DNA (ssDNA). In order to perform SELEX using RNA, an extra reverse transcription PCR (RT-PCR) step is required; ssDNA is first converted to cDNA and an in vitro transcription reaction catalyzed by T7 RNA polymerase is then used to generate the single-stranded RNA library (Stoltenburg et al., 2007). Single-stranded DNA libraries, on the other hand, can be applied directly in SELEX without any further preparation steps. Aptamers are obviously chemically distinct from antibodies, yet they mimic their high-affinity high-selectivity binding behavior closely(Jayasena, 1999). However, unlike antibodies, which are typically too expensive for use as an affinity ligand for large-scale purification operations (Curling, 2004a), aptamers against a desired target can be selected and manufactured  17  rapidly and cheaply. The affinities of aptamers are comparable to those of antibodies, in some cases registering tighter binding. In terms of specificity, aptamers can discriminate targets based on chemical differences such as the presence of methyl or hydroxyl groups, or structural differences such as different enantiomeric forms of the target. Additionally, aptamers are small in size when compared to classic ligands used for affinity chromatography such as Protein A and antibodies. This could potentially allow for a higher density of ligand loading on the stationary phase, and thus possible improvements in binding capacities. Together, these attributes suggest that aptamers might be a useful new class of ligands to purify biomolecules at manufacturing scales with high purity and yield, and at a relatively low cost.  Surprisingly, however, little attention has been devoted to the idea of DNA aptamers as ligands for large-scale affinity chromatography. What is available is a smattering of studies toward use of aptamers, namely RNA aptamers, in analytical-scale diagnostic or HPLC systems.  While not specific to the goals of this thesis, those studies are nonetheless informative.  Analytical columns for designed for use in diagnostic assays have been described that successfully separate chiral molecules (Michaud et al., 2003), proteins (Romig et al., 1999) or small molecules (Deng et al., 2001). RNA aptamers have also been used in analytical-scale columns to isolate proteins, including Thermus aquaticus DNA polymerase (Oktem et al., 2007), various His-tagged proteins using an anti-His aptamer (Bartnicki et al., 2015), and concanavalin A (Ahirwar and Nahar, 2015). However, DNA aptamer ligands, which are the focus of this work, have not been explored as ligands for affinity chromatography or, more importantly, as a general platform for ligand development for preparative-scale separations. Aptamers against human plasma proteins (e.g. Factor VII, Factor H, Factor IX) have been described and used in analytical scale columns (Forier et al., 2017). Good separation was achieved, but this result must be tempered by the fact that the target of interest was first partially purified from human plasma using cation and anion exchange chromatography before the affinity purification was conducted. This would not be the case for a preparative affinity chromatography column used to capture and concentrate a protein product from a cell-culture feedstock, which is the focus of this work as it is the manner in which affinity chromatography is most often applied during bioprocessing of human therapeutics.  Nevertheless, this recent study does illustrate that aptamers at least have the potential to capture therapeutic proteins with high purity when used as ligands in affinity chromatography.  18  1.3.5 Affinity Chromatography – Uses and Underpinning Concepts Affinity chromatography is a mode of liquid chromatography that leverages the ability of a natural or carefully engineered synthetic molecule, known as the ligand, to specifically recognize and reversibly bind with high affinity a biologic target molecule (e.g. a protein therapeutic).  In affinity chromatography, the ligand is bound to the stationary phase of a column, permitting the target to be captured and thereby extracted from the complex feedstock. The term “affinity chromatography” was first introduced in 1968, although the earliest work on affinity chromatography dates back to the early 1900s (Cuatrecasas et al., 1968), when the botanist Michael Tswett used the method for a simple separation of plant pigments in 1903. In 1910, Emil Starkenstein studied the natural affinity between starch and α-amylase by immobilizing insoluble starch in a packed column to successfully purify the enzyme (Hage and Matsuda, 2015). Since then, many proteins, enzymes, hormones and nucleic acids have been successfully purified by using a natural or synthetic ligand as the immobilized capture agent (Burnouf and Radosevich, 2001; Clonis et al., 2000; Jozala et al., 2016; Murphy et al., 2003; Sakamoto et al., 2012).  Due to the exquisite specificity of a natural ligand for its cognate biologic target, affinity chromatography has become the mode of chromatography most preferred by industry for capturing and concentrating a protein therapeutic product from a recombinant cell culture fluid (e.g. cell culture supernatant) or a natural human tissue (e.g. blood plasma).  When used for that purpose, a well-designed affinity chromatography step can be expected to isolate the product to higher purity, yield and concentration factor when compared to other commonly employed modes of preparative chromatography, including ion-exchange, hydrophobic-interaction, mixed mode, or size-exclusion (Curling, 2007). But this is only true if an appropriate ligand is available and if the cost of using that ligand in a column format is favorable when applied at manufacturing scales.  Although exceptions can be identified, industry strives to keep the total cost of goods (including chromatographic resins) for producing and purifying a protein therapeutic below $1000 per gram of product manufactured; ideally, below $300/g. This sets a cap on the cost of using affinity chromatography as the initial capture step at no more than about $50 per gram of product manufactured.  Satisfying that constraint is not easy, as the ligand itself, which often must be produced biologically and then purified, is often expensive.  As a result, affinity chromatography is currently applied industrially to only a few select classes of biologic products - typically those for which the total market value is sufficiently large to justify a major  19  research and development effort into creating a robust low-cost ligand and associated affinity column.  The best known of these are monoclonal antibodies (mAbs), whose collective value currently represent roughly 1/3 of that of all proteins approved for therapeutic use.   The ligand used to capture mAbs via affinity chromatography is Protein A or a variant of Protein A that has been engineered to offer improved specificity, affinity or stability. Protein A is a complex biologic molecule, and industry has devoted decades of intense research to develop cost-effective ways to produce and utilize that complex ligand. That tremendous multi-company effort has paid enormous dividends, as essentially all FDA-approved mAbs are captured and concentrated by (variant) Protein A affinity chromtography (Curling, 2007), which delivers the product at > 99% purity and at > 95% yield when challenged with clarified CHO-cell culture supernatants containing the product.  But realizing this tremendous success story was achieved in part because mAb-based human therapeutics are collectively a ~ $100 billion per annum industry, which justified the significant costs associated with developing state-of-the-art Protein A columns. How one might exploit the tremendous separation advantages of affinity chromatography in the downstream processing of far less valuable, but nonetheless life-saving, protein therapeutics remains an open challenge (and goal) within industry. Purification by affinity chromatography, a form of frontal adsorption chromatography, consists of four main steps as illustrated in the chromatogram shown in Figure 1.3.   20    Figure 1.3: A typical separation scheme and chromatogram for affinity chromatography (GE Healthcare Bio-Sciences AB, 2007).  First, the crude sample containing the target of interest is pumped into an affinity column that has been pre-equilibrated in a mobile phase buffer with a composition (salt concentration, pH) that promotes high-affinity binding of the target protein to the immobilized ligand. In this loading stage, the sample is pumped onto the column until product breakthrough is detected at the column exit.  This process insures that the percentage of immobilized ligands bearing a bound target protein is maximized while the loss of product during purification is minimized. The highly selective high-affinity interaction between the target and ligand is often a result of a complex ensemble of electrostatic interactions, hydrophobic interactions, van der Waals’ forces and/or hydrogen bonding. The next step is the washing of unbound molecules from the column, with the bound target of interest retained on the affinity support. The target molecule is then eluted from the immobilized ligand either through biospecific elution or nonspecific elution. Biospecific elution uses competitive ligands to displace the target from the immobilized ligand, while nonspecific elution involves alteration of the mobile-phase pH, ionic strength, and/or polarity (Magdeldin and Moser, 2012). Biospecific elution is a gentler elution method but it takes longer, is generally more expensive, and creates a broader elution peak, which dilutes the  21  concentration of the target. Nonspecific elution, which is far more often used in industry, is faster and creates a much sharper elution peak (Hage et al., 2006). Nonspecific elution conditions ideally are set such that the yield of purified product is high while changes to the integrity and activity of the product are minimized.  Extreme pH and/or non-aqueous solvents are therefore rarely used by industry for eluting therapeutic proteins from affinity columns.  The final step is the regeneration of the column for the next round of purification. Typically, the column is regenerated using application buffer, which is the mobile phase used to equilibrate the column prior to the injection of the crude sample. However, after a specified number of purification cycles, the column may first be subjected to sterilization (to remove any unwanted trapped or retained contaminants) using an appropriate CIP procedure.  An affinity column is an orchestration of a few components, which include the supporting base matrix packed within the column and the affinity ligands that have been immobilized onto that base matrix. The immobilization of ligand to the chromatography matrix is typically attached covalently to ensure ligand stability. Upon immobilization, the coupled ligand should retain its specific binding to the target molecules. Flexible water-soluble spacer arms (linkers) are often used to display the immobilized ligands in a manner that provides for efficient target binding (steric interferences are minimized) and a more efficient purification (Cuatrecasas, 1970; Hage et al., 2006).   The stationary phase (media) of affinity chromatography columns used at manufacturing scales is usually in the form of a spherical porous particle, often 30 to 80 µm in diameter, onto which the ligand has been immobilized.  Elements that must be carefully considered during the design of that chromatography media include its chemical inertness, chemical and mechanical stability, particle size, and pore size (Hage et al., 2006). The matrix must be chemically inert (passive) so as to limit non-specific interactions with other molecules during the loading phase. Changes in column chemistry, including loss of ligand, during use must also be avoided, as such perturbations limit the lifetime of the column and can introduce new and unwanted contaminants in the product-bearing process stream. The matrix should also exhibit good mechanical stability to prevent collapse of matrix when pressure is applied. Porous matrices that have proven able to satisfy these stringent performance criteria include various forms of beaded agarose, polymethacrylate, or silica, but other supports are also used.   22  As noted, the diameter of the spherical particles comprising the media is generally 30 µm or higher for preparative work so as to provide satisfactory mass transfer while avoiding excessive pressure drops that might collapse the required packed-bed structure of affinity columns that can be up to several 100 L in volume.  Pore sizes must also be optimized to balance the need for good rates of product mass transfer within the pores, which favors larger pores, with the desire to maximize the loading density of ligand (and by extension bound target), which generally favors a dense network of smaller pores. Based on these considerations, a pore diameter that is about 5 times larger than the diameter of the target molecule is often preferred. A larger pore size is disadvantageous as the surface area per volume will be reduced and thus limit binding capacity.  1.4 The Aptamer-Based Custom Chromatography Development Pipeline - Basic Concept As illustrated in Figure 1.4, the proposed operational flow of the pipeline developed and presented in this thesis begins with high-throughput selection of a panel of candidate DNA-aptamer ligands for product capture from a highly diverse starting library comprised of more than 1014 unique 40-mer oligonucleotide sequences. The selection method that has been developed for this purpose is described in Chapter 2 of this thesis.  Specific end-group chemistries are then applied to the candidate aptamers to achieve orientation-specific immobilization of each candidate ligand to small aliquots of Bio-Rad UNOsphere Epoxide and/or UNOsphere diol preparative chromatography media, respectively.  These media were chosen based on their widespread use in industry as a base matrix for preparative-scale chromatography. These matrices are therefore already proven to offer excellent chemical and mechanical stability, to minimize non-specific adsorption, and to present pores with diameters optimized for protein purification.   A specific set of performance metrics for each prototype column is then collected and evaluated to narrow the list of candidate ligands and define the more suitable column chemistry. As this thesis work is supported by an industrial partner, BioRad Inc., one or possibly more BioRad Bio-Scale™ mini-columns, each bearing one of the candidate ligands, will be prepared and then subjected to an ultra-scale-down column analysis in which key performance metrics (such as static and dynamic binding capacities, purification factors, yields, etc.) are collected and  23  used to evaluate the performance and general potential of the proposed pipeline. The long-term chemical stability of the ligand and its performance as a function of column cycle number will also be assessed and results used to determine if modification of the aptamer is required to improve its stability and performance over time both in the presence of complex feedstock and when subjected to repeated column cycling. Results pertaining to evaluation of preparative affinity chromatography columns bearing DNA aptamer ligands are reported in Chapter 3 of this thesis.    Figure 1.4: Schema of the Proposed Custom Chromatography Development Pipeline     24  1.5 Anticipated Impact of Thesis Research  The biotech industry has become incredibly skilled at producing protein therapeutics to high concentrations in biological reactors. As a result, more than ever the primary cost and challenges associated with producing life-saving biologic therapeutics reside in the downstream processes (DSPs).  Monoclonal antibodies, to date, have maintained their dominance as the major class of biologic therapeutics because they are well tolerated and highly specific with minimal safety issues in human clinical trials (Ecker et al., 2015). To meet the growing demand for mAbs, the required DSPs have been heavily researched and optimized to provide excellent separation performance with a low cost of goods.  Antibodies represent about a third of biologic therapeutics sales, with a wide range of other protein-based biologics making up much of the remaining two-thirds. DSPs for these important molecules are largely product specific and developed on a more empirical basis. As a result, they are generally more expensive. In particular, safety, efficacy, and product quality take precedence over manufacturing costs (Scott, 2011). But the need to change this situation has become clear and urgent.  If we hope to cope with the rising existence of rare diseases in the 21st century, the costs to develop and manufacture therapeutics to effectively treat them must be reduced to levels that abate the rising costs of healthcare and bring the products of what is now a $400 billion US per annum industry in tune with the Affordable Care Act (Rosenbaum, 2011). The objective of my M.A.Sc. research and thesis was to provide one of those required capabilities; namely, a set of technologies that enable the efficient design and validation of an affinity capture step for any protein therapeutic of interest that preserves product yield and integrity while limiting further DSP demands, and thus cost.  The anticipated deliverable is a new pipeline for efficient creation of affinity capture media displaying DNA-aptamers custom designed to achieve specific performance metrics when applied to a particular feedstock. That pipeline, when fully realized, could be of considerable benefit to the growing orphan drugs market. The annual production of a biologic orphan drug used to treat a rare disorder is typically a few kg to a few tens of kg. If current manufacturing processes are applied to the production of these low-volume products, their cost to patients becomes unsustainably high (Hughes-Wilson et al., 2012). The product capture and concentration step is often the single largest cost centre in downstream processing (Curling, 2007). Previous approaches to addressing this issue, such as through the use of peptide (Lowe, 2001) or synthetic (Curling, 2004a) ligand  25  libraries, have not proven successful. However, relative to those approaches, the use of DNA aptamers as affinity ligands may offer increased quality, speed, and return of investment in protein purification. Through the technology described in this thesis, high affinity aptamers can be selected rapidly, and then synthesized in large quantities in a cost-effective manner. Importantly, though not the specific focus of this work, aptamers can be modified to increase their chemical and thermal stabilities, and robust linkage chemistries are available that can limit the possibility of ligand leakage during elution and harsh clean-in-place procedure.  The method is therefore designed to offer the many DSP advantages of using protein A affinity chromatography to purify antibodies to the many important therapeutic biomolecules for which a comparably effective and inexpensive affinity ligand is not currently available.  The new method developed here may thus provide a reliable means to rapidly create customized affinity chromatography media that lowers manufacturing costs and the associated cost of treating patients with a rare disease (Hall and Carlson, 2014).     26  Chapter 2 Highly Efficient and Reliable DNA Aptamer Selection Using the Partitioning Capabilities of ddPCR – The Hi-Fi SELEX Methoda,b  2.1 Introduction Biological reagents that recognize target molecules with both high affinity and specificity are routinely required in the life sciences and clinics, where in the latter case they often serve as capture, diagnostic or therapeutic agents. Amongst the available classes of affinity reagents, antibodies currently are the gold standard. Antibodies, including monoclonal antibodies (mAbs), bind their antigen with high affinity; nM or tighter equilibrium dissociation constants (Kds) are frequently reported (Edwards et al., 2003). Moreover, binding is generally highly selective, as demonstrated by the ability of mAbs to discriminate post-translational modifications to proteins, as well as among more subtle protein isoforms (Raska et al., 2014; Xing et al., 1994). These exquisite binding properties can serve to maximize therapeutic mAb potency and minimize off-target effects, while the relatively large size of mAbs enables long circulation half-lives (Chapman et al., 1999; Mould and Sweeney, 2007).  However, antibodies are limited in certain important aspects.  As large complex multi-sub-unit proteins, they are sensitive to environmental conditions, and can rapidly be inactivated under acidic conditions or at elevated temperatures. Though significant advances have been made toward their manufacture, mAbs remain relatively expensive to produce and purify at larger scales.  Most mAbs are incapable of permeating cells  or blood-brain barrier efficiently, which effectively restricts their application to extracellular antigens. Moreover, though humanized-antibody technology has greatly reduced immune responses, therapeutic mAbs often do not escape immune surveillance completely, further challenging their long-term efficacy (Chames et al., 2009).                                                            a The contents of this chapter are taken from a recent publication: Ang A, et al. “Highly Efficient and Reliable DNA Aptamer Selection Using the Partitioning Capabilities of ddPCR – The Hi-Fi SELEX Method”, Methods in Molecular Biology, in press (Dec. 2017). b Some elements of the second-generation Hi-Fi SELEX method described in this Chapter were completed in collaboration with Eric Ouellet, a former student in the Haynes Lab  27  As a result, the development of mAb alternatives for use as research and diagnostic affinity reagents, as well as therapeutic agents, has gained considerable interest in recent years (Ruigrok et al., 2011). A number of simplified antibody forms including nanobodies, VH and VL antibody domains, and single-chain variable fragments have proven effective as mAb surrogates (Skerra, 2007). In addition, a particular class of nucleic acids, known as aptamers, has emerged as a potent option (Jayasena, 1999). Each comprised of a short single-stranded (ss) oligonucleotide, aptamers can be produced relatively inexpensively at large scale with high precision. The discovery of useful aptamers is likewise facilitated by the ability to easily synthesize large semi-combinatorial libraries of ssDNA or, with a bit more effort, ssRNA that can be subjected to in vitro Darwinian-type selection strategies to enrich and select sequences that exhibit high affinity and specificity for a target as a result of their unique folds. However, as described in Chapter 1, standard aptamer selection methods, which typically employ a recursive selection process known as systematic evolution of ligands by exponential enrichment (SELEX), are not sufficiently robust to ensure the timely and cost-effective discovery of aptamers suitable for further development (Ellington and Szostak, 1990; Ozer et al., 2014; Robertson and Joyce, 1990; Tuerk and Gold, 1990).  This chapter therefore reports on a new method, and enhancements to it, that greatly improves the speed and robustness of aptamer selection. The method, which we call High Fidelity (Hi-Fi) SELEX, uses consumables and equipment that are available in essentially all standard molecular biology laboratories. A description of the initial “proof-of-principle” version of Hi-Fi SELEX was recently published (Ouellet et al., 2015). Many important elements of the Hi-Fi SELEX platform had not been optimized in that preliminary work. As a result, that initial technology was not robust, with the limited results obtained and reported primarily serving to demonstrate the potential of the technology. This thesis presents work on a new significantly more advanced and robust second-generation version of Hi-Fi SELEX, hereafter referred to as Hi-Fi SELEX for brevity, that exploits the partitioning capabilities of droplet digital PCR (ddPCR) to preserve library integrity during regeneration. The method therefore requires either a commercial droplet generation system or the ability to create emulsions suitable for PCR. It does not require droplet -reading capabilities, making the method cost effective since the remaining elements of  the method  28  can be carried out using equipment present in essentially all standard molecular biology laboratories.  Here, I describe in detail the Hi-Fi SELEX protocol, for which the basic processing scheme is shown in Figure 2.1, in sufficient detail to enable a user to reliably apply it to discover useful DNA aptamers against a target of interest.  I begin with a basic description of the key elements of the technology, then provide step-by-step instructions on how to conduct the screening technology in the lab.    Figure 2.1: Schematic showing the sequence of operations comprising a single round of Hi-Fi SELEX.  Hi-Fi SELEX provides a significant improvement over conventional SELEX by addressing a number of well-known issues that limit that technology, including (i) interferences from the fixed regions flanking each random sequence within the library (Ouellet et al., 2014), (ii) poor partitioning efficiencies due to the retention of non-specific sequences during screening  29  rounds (Berezovski et al., 2006), (iii) excessive accumulation of amplification artifacts (Musheev and Krylov, 2006), and (iv) significant loss of retained library material during the conversion and isolation of double-stranded DNA amplicons into the single-stranded DNA libraries required for the next selection cycle. By alleviating these challenges, Hi-Fi SELEX provides a rapid and highly robust platform for isolating high affinity aptamers. Hi-Fi SELEX begins with the preparation of competent aptamer library using complementary sequences that anneal to the fixed regions of the sequence. This step is required because the fixed regions can fold into structures themselves and sterically hinder the aptamers from binding to the target. At times, the fixed regions can also hybridize to the random region and thereby disrupt the aptamer fold required for target binding. In either case, the affinity of the sequence to the target is lost and the potentially desirable sequence will be removed from the retained library during screening. Some strategies proposed to solve this problem include minimizing the length of the fixed region (Pan et al., 2008) or completely removing and regenerating the fixed regions before and after each selection step (Jarosch et al., 2006). These methods may be effective in some cases, but they come with limitations such as promoting inefficiencies in PCR when short priming sites are used, as well as a loss of library members during the truncation and ligation of fixed regions between selection rounds. Hi-Fi SELEX therefore leverages a new strategy of blocking the fixed regions with complementary sequences.  Interferences to the folding process of the random region are thereby prevented and the binding affinity and structural diversity of the library members can be fully restored. The result is a more diverse library that can dramatically increase the chances of discovering useful aptamers.  In Hi-Fi SELEX, the target is immobilized onto the surface of microtitre plates that have been carefully engineered to effectively eliminate non-specific adsorption of library members.  Non-specific retention of aptamers on the surfaces on which the target is displayed is a major issue that is known to severely compromise aptamer selection. Reactive electrophiles on the selection surface can further serve as unwanted non-specific binding sites for aptamers. Without properly neutralizing and passivating the target-displaying selection surface, non-specific retention of aptamers can therefore significantly decrease the partition efficiency, PE. A low PE indicates a poor selection round.   30  Hi-Fi SELEX uses tris-HCl supplemented with a surfactant, Tween 20.  The tris group serves to neutralize all unreacted electrophiles following target immobilization, while the Tween 20 passivates the target-loaded surface against non-specific adsorption of library members. Unwanted non-specific retention of library members is thereby greatly reduced. Moreover, by combining this target-display strategy with a series of increasingly stringent washes of the retained library, library members whose binding is dominated by electrostatics can be selectively removed to further improve the PE value. When the competent library of ~1014 members is allowed to incubate with and interact with the target; only a very small fraction (~ 0.000001%) of the library is therefore retained and eluted.  In order to proceed to the next round of selection, the retained library members must be amplified through PCR to create a large number of copies of each sequence and restore the total quantity of library members back to ~1014.  To achieve this, roughly 1 million copies of each retained library member must be produced. PCR is highly efficient in amplifying with high fidelity a single template a million-fold or more. For a heterogeneous pool of eluted library members, however, PCR can only proceed to a certain cycle number of amplification.  Beyond that cycle number, the decreasing primer-to-template ratio will no longer favour the annealing of universal primers to the common fixed regions of each template. Instead of the desired formation of homo-duplexed amplicons, hetero-duplexed amplicons arise through cross hybridization of the amplified library members between the common fixed regions (Jensen and Straus, 1993). Hetero-duplexes are known to promote spurious priming events and subsequently generate high molecular weight by-products. A common method of solving this challenge is by stopping the PCR before the formation of the unwanted by-products and running hundreds of PCR reactions in parallel. Even so, the restoration of ~1014 sequences is difficult to achieve. Hi-Fi SELEX resolves this issue by using the partitioning effect of droplet digital PCR (ddPCR). In a single reaction, ddPCR can generate ca. 20,000 droplets, and into each can be partitioned a defined number of library members along with the enzymes and the reagents needed to amplify them by PCR (Hindson et al., 2011). The droplets so formed are heat stabilized to prevent the contents of one droplet from interacting with the contents of other droplets. In Hi-Fi SELEX, each droplet can therefore be filled with an average number of templates (i.e. copies per droplet, CPD) and amplification can proceed to end-point to generate ~1014 homo-duplexed amplicons.   31  I note that in the preliminary form of Hi-Fi SELEX (Ouellet et al., 2015), retained library members were partitioned into droplets at a maximum CPD of 50 templates per droplet and then amplified.  As a result, at least 108 retained library members needed to be recovered after each selection round to generate the 1014 sequences needed for the next round of selection. However, due to a combination of many influencing factors including the type of target, the stringency of the sequential washes, etc., far less or more than 108 library members are often recovered in each of the initial rounds of selection. Moreover, applying too stringent of a wash in the first round such that 108 or fewer library members are retained creates the real risk of losing potentially useful ligands, since the starting pool in theory contains no more than 6 copies of each unique member. In order to safeguard the library diversity and expand the applicability of Hi-Fi SELEX to a wide range of targets, an improved ddPCR protocol was developed to be able to sufficiently amplify any amount of retained library members up to a limit of 109 sequences.  At the end of amplification, the amplicons are pooled together and must then be converted to single-stranded library prior to the next round of selection. The most common method for doing this uses streptavidin-coated magnetic beads and alkaline treatment to strip away the unwanted strand.  Alternatively, -exonuclease catalyzed digestion may be used to remove the antisense strand and thereby regenerate the single-stranded library (Avci-Adali et al., 2010). In Hi-Fi SELEX, we have optimized the -exonuclease catalyzed digestion reaction to enable the complete conversion of dsDNA to ssDNA with 100% yield.  Complete removal of the 5’-phosphorylated antisense strand is achieved in a 30-min to 1-hour reaction.   Following -exonuclease processing, the enzyme and nucleotides produced by hydrolysis must be removed to recover the new ssDNA library in pure form. Phenol/chloroform extraction followed by ethanol precipitation of the library is typically used (Buckingham and Flaws, 2007; Chomczynski and Sacchi, 2006; Tan and Yiap, 2009).  Ethanol precipitation is known to work well for the nucleic acids which are kilobase pairs in length (Fuhrman et al., 1988; Strauss, 1998; Wilson, 1997). But the ssDNA of the Hi-Fi SELEX library is only 80-nt in length. As a result, we found that recovery of the regenerated library members using this method therefore results in poor yields, often below 20%. An alternative method was therefore developed to remove traces of phenol from the purified pool of ssDNA that uses membrane-assisted buffer exchange. The buffer-exchange operating conditions can be easily optimized in terms of spin speed and  32  membrane passivation to obtain a good recovery of ssDNA in the appropriate buffer for the next round of selection. Finally, establishing that a SELEX or a Hi-Fi SELEX selection process is effectively reducing (round-by-round) the size of the library so as to enrich for library members having a high affinity for the target requires a means to quantify both the sequence diversity and the mean dissociation binding constant Kd of the retained pool following each selection round.  Traditionally, the mean Kd of a retained pool has been measured using either isothermal titration calorimetry (ITC) or, more commonly, surface plasmon resonance (SPR) (Leavitt and Freire, 2001). Both methods are expensive and time-consuming, and more importantly neither has the sensitivity to measure Kd during the first few rounds of selection. We therefore developed a simple and inexpensive qPCR method that uses a small aliquot of the regenerated ssDNA library to generate the isotherm for binding of the retained pool to the target.  Isotherms for each retained pool can be generated, and the resulting data can be fit either directly with a Langmuir isotherm equation to estimate the mean Kd, or with a more sophisticated theory, detailed in the Appendix, that segregates the retained population into high-affinity and low-affinity fractions and estimates the mean Kd for each fraction. Similarly, the diversity of the retained library can be estimated using a novel qPCR-based melt-curve analysis method described here when the number of unique sequences exceeds about 1,000,000. Once that diversity decreases below ~106 unique sequences, an alternative simple qPCR-based fluorescence assay may then be used to accurately monitor sequence diversity.  I will show in this chapter and the next that Hi-Fi SELEX provides an efficient and highly robust means to isolate a panel of functional aptamers within a few days.  Each candidate aptamer within that panel can then be synthesized, sequenced and characterized by SPR and/or ITC to generate accurate binding affinity and binding thermodynamic data (Leavitt and Freire, 2001). From those studies, a single or small set of candidate aptamers can be chosen and synthesized in larger quantity to permit its evaluation as a ligand for affinity chromatography.  The complete set of reagents and experimental operations comprising the Hi-Fi SELEX method are described in detail in the remainder of this chapter.  Results from applying the method to discover new aptamers are presented in Chapter 3.   33  2.2 Materials 2.2.1 Oligonucleotides The advanced Hi-Fi SELEX protocol selects aptamers from a semi-combinatorial library of members in which each member is comprised of three oligonucleotide sequences hybridized together.  As shown in Figure 2.2, these oligonucleotides include an 80-nt ssDNA “library” sequence, a 20-nt ssDNA DNA sequence (5’-Comp) that is complementary to the 5’-end of each 80-nt member of the library, and a 20-nt ssDNA sequence (3’-Comp) that is 5’-phosphorylated and complementary to the 3’-end of each library member (Ouellet et al., 2014).   Each of the three required components (80-nt ssDNA library, 3’-Comp, 5’-Comp) is synthesized chemically by a reputable vendor (we use IDT Inc.; Coralville, IA) capable of delivering it in the quantities (typically a few µmoles), uniformities (ssDNA library and Comp sequences are independently processed by HPLC (strong anion exchange column) to eliminate truncated members and isolate a tight band of the desired molecular weight), and purities needed to satisfy the DNA quality and diversity requirements of the Hi-Fi SELEX process. Each component is then reconstituted in 1X AF buffer to create the required stock solutions. The initial library must be synthesized at a 1 µmole scale in order to generate enough material for conducting Hi-Fi SELEX. Following synthesis, the library is purified as described and then lyophilized. When required, the lyophilized product is reconstituted at a 100 µM concentration in 1X AF buffer and stored at -20°C away from all other reagents used in the Hi-Fi-SELEX protocol. A ssDNA library structure and corresponding set of 5’-Comp and 3’-Comp sequences appropriate for conducting Hi-Fi SELEX are provided below.     34    Figure 2.2: Basic structures of the semi-combinatorial DNA libraries used in standard SELEX (standard library) and Hi-Fi SELEX (competent library): Hi-Fi SELEX libraries differ from standard SELEX libraries through their use of novel fixed-region complements (blocking elements) to improve the functional diversity of the starting semi-combinatorial library.  Each member of the Hi-Fi SELEX ssDNA library is therefore comprised of an 80 nt library sequence containing a 40-mer random core region (N40) flanked by a 5’ universal 20-mer flanking sequence and a 3’ universal 20-mer primer binding sequence, with each flanking sequence hybridized to its complement, which are hereafter denoted as 5’-Comp and 3’-Comp, respectively.  By eliminating single-strand structures within the fixed regions, the competent Hi-Fi SELEX library isolates aptamer fold and function to within the variable core region of the library, while reducing artifacts that might compromise or eliminate the discovery of a tight-binding sequence within that region. Unblocked, the fixed-region sequences within a given selection library can interfere with aptamer fold and function through their potential to adopt stable secondary structures created through either 1) self-association, or 2) association with complementary nucleotides within the variable core region, the opposing fixed region, or both.  These types of unwanted structures can occur either within an individual library member or between complementary regions of different members of the library, and the net effect is to significantly reduce the total functional diversity of the library.    35  2.2.2 Reagents and Consumables The following solutions and chemicals must be prepared and available to complete the Hi -Fi SELEX DNA aptamer selection process:  1. Aptamer Folding (AF) buffer: 20 mM Tris-HCl (pH 7.4), 140 mM NaCl, 5 mM KCl, 1 mM MgCl2, 1 mM CaCl2. 2. ssDNA Library Stock Solution:  1 µmole purified 80-nt ssDNA in AF buffer (10 µl of a 100 µM stock required for complete (3 rounds) Hi-Fi SELEX selection).  The general structure of each library member is 5’-flanking sequence (20 nt) –N40 –flanking sequence (20 nt)-3’, with one suitable library sequence being library [5’-TCGCACATTCCGCTTCTACC–N40 –CGTAAGTCCGTGTGTGCGAA-3’]. 3. 5’ Complementary Blocker (5’-Comp): for the library sequence shown above, 5’-GGTAGAAGCGGAATGTGCGA-3’ resuspended at 100 µM stock in AF buffer (10 µl each needed to complete each round of Hi-Fi SELEX). 4. 3’ Complementary Blocker (3’-Comp): for the library sequence shown above, 5’-TTCGCACACACGGACTTACG-3’ resuspended at 100 µM stock in AF buffer (10 µl each needed to complete each round of Hi-Fi SELEX). 5. Stringent Wash (SW) buffer: 20 mM Tris-HCl (pH 7.4), 4 M NaCl, 5 mM KCl, 1 mM MgCl2, 1 mM CaCl2, 0.005% Tween 20. 6. Denaturing Elution (DE) buffer: 50 mM NaOH (followed by neutralization with 50 mM HCl). 7. Tris-EDTA (TE) buffer: 10 mM Tris, 1 mM EDTA, adjusted to pH 8.0 with HCl.  8. Target Immobilization (TI) buffer: 100 l of 100 mM sodium phosphate buffer (pH 7.5). 9. Surface Passivation (SP) buffer: AF buffer supplemented with 0.005% Tween 20. 10. λ-exonuclease; lambda exonuclease (New England Biolabs). 11. Nunc Immobilizer Amino Plates (ThermoFisher Scientific). 12. Thermomixer fitted with a plate adapter (e.g. Eppendorf Thermomixer C). 13. NanoSep® Omega 3K and 10K MWCO centrifugal filter unit (Pall Corporation).  36  14. Standard PCR thermal cycler (e.g. Bio-Rad C1000 or Eppendorf Mastercycler Thermal Cycler).  2.2.3 Droplet Digital PCR (ddPCR) and qPCR Reagents 1. Forward Primer (FP): 5’-TCGCACATTCCGCTTCTACC-3’ resuspended in AF at 100 µM stock concentration (for library sequence shown). 2. Phosphorylated Reverse Primer (RP): 5’-p-TTCGCACACACGGACTTACG-3’ resuspended in AF at 100 µM stock concentration (library sequence shown). 3. SYBR Green qPCR master mix: 2X iQ SYBR Supermix (Bio-Rad). 4. ddPCR master mix: ddPCR Supermix for probes without dUTP (Bio-Rad). 5. Thermal Cycler capable of real-time detection (e.g. Bio-Rad CFX96). 6. DG8 Droplet-Generation Cartridge (Bio-Rad). 7. Droplet Digital PCR (ddPCR) Droplet Generator (Bio-Rad QX100 or QX200).  2.3 The (Second-Generation) Hi-Fi SELEX Protocol Prior to the initiation of any aptamer selection work, it is essential that proper steps be taken to prevent cross-contamination of material recovered between selection rounds. Regular decontamination of all work surfaces and regular changes of gloves during the process is necessary to prevent such occurrences. Moreover, stock reagents used for PCR amplification must be kept separate from the initial library as well as from any recovered pools of library members during selection. Aliquots of each stock reagent should be made in designated nucleic acid template-free areas. If aptamer selection is to become routine in the lab, a dedicated separate environment for conducting Hi-Fi SELEX must be created, with regular maintenance and decontamination of all surfaces and instruments prior to and at the completion of each round of selection.     37  2.3.1 Library Design and Synthesis In its standard format (Ellington and Szostak, 1990; Robertson and Joyce, 1990; Tuerk and Gold, 1990), SELEX enriches a subset of short single-stranded oligonucleotide sequences from a synthetic library comprised of a semi-combinatorial population whose total diversity is similar to the body’s own antibody repertoire. The process is typically performed in vitro, creating the capacity to modulate the nature and stringency of the screening conditions to more efficiently select library members showing high affinity or specificity to a target. SELEX can operate on either a ssDNA or ssRNA library, with the choice typically made by carefully examining the intended application of the aptamer. RNA-based aptamers have a more flexible backbone that allows them to adopt a wider range of secondary and tertiary structures.  Though built on a less flexible backbone, ssDNA aptamers have higher chemical stability, and screening of DNA libraries requires fewer processing steps. In this chapter, the technological advances offered by Hi-Fi SELEX, including its use of the partitioning step of ddPCR, are specifically applied to ssDNA aptamer selections.  DNA libraries used for SELEX typically contain ~ 1014 unique members of equivalent length, each comprised of a random oligonucleotide sequence within a variable core region that is flanked by universal fixed sequences at the 5’ and 3’ ends.  Library diversity is largely encoded in the variable core region.  That region can be created utilizing different randomization strategies and nucleotide chemistries.  Most often, chemical syntheses that equally weight the frequency of each naturally occurring nucleotide are employed, but variable core regions comprised of partially-randomized sequences (Bartel et al., 1991), genomic DNA inserts (Lorenz et al., 2006; Zimmermann et al., 2010) and various chemically-modified nucleotides (Keefe and Cload, 2008; Klußmann et al., 1996; Kuwahara and Obika, 2013) have also been used with success. Hi-Fi SELEX employs a variable core region that is 40 nt in length (Figure 2.2) so as to generate reasonable diversity while keeping the mass of the starting library suitable for the screening process.  However, smaller or longer randomizations can be used.  The sequences of the universal fixed regions (typically 20 nt each) flanking the 5’ and 3’ ends of each variable core sequence are engineered to achieve high-fidelity amplification of retained library members  38  by eliminating, or at least minimizing, self-association and primer-dimer pairing reactions that can promote formation of unwanted by-products during PCR. To achieve this, both the flanking sequences and the associated forward (FP) and reverse (RP) primers are designed using Primer3 software (http://biotools.umassmed.edu/bioapps/primer3_www.cgi), and then scored for self-complimentary and primer-dimer formation using the OligoAnalyzer tool (https://www.exiqon.com/ls/Pages/ExiqonOligoOptimizerTool.aspx) of Exiqon, Inc. (Hall et al., 2009; Jiménez et al., 2012; Sefah et al., 2010). The random core region of the 80-mer ssDNA library is created combinatorially by dosing the required nucleotides at an A:C:G:T molar ratio of 3:3:2:2.4 in order to achieve equal probability incorporation of each nucleotide within the variable core region (Gopinath et al., 2006). The 5’-phosphorylated 3’-Comp sequence is used as the RP for PCR amplification.  2.3.2 Target Immobilization Levels of non-specifically retained library members are known to dictate the degree of enrichment of useful aptamers achieved in a given round of selection (Weng et al., 2012).   Practitioners of SELEX have therefore attempted to quantify the overall quality of each round of selection through a parameter PE known as the “partition efficiency”: 𝑃𝐸 =[𝐴][𝐴𝑃] (2.1) where [𝐴] = ∑[𝐴𝑖]𝑛𝑖=1 (2.2) Here, [A] is the total molar concentration of all unbound library members, given by the sum of the concentration [Ai] of each unbound library sequence i; likewise [AP] is the molar concentration of all bound library members recovered in the fraction eluted in DE buffer. In standard SELEX, per round PE values generally range from 10 to 1000, due in part to non-specific retention of library members. Reducing the starting population of ~1014 members to a manageable number of candidate aptamers, say ~104, then requires a  39  relatively large number of successful selection rounds (generally identified by an improvement in the mean Kd of the retained pool). In contrast, Hi-Fi SELEX generates per round PE values greater than 105 (usually close to 106) by displaying the target on a tailored substrate and in conditioned solvents that together greatly inhibit non-specific retention. Nunc Immobilizer plates, which present on the surface of each well a layer of end-grafted hyper-branched polyglycol chains at densities near or above those needed for soft brush formation, display good passivation against non-specific adsorption of proteins (Gong and Grainger, 2007) or oligonucleotides (Cattani-scholz et al., 2009; Schlapak et al., 2006), even in cases where the zeta potentials of the underlying base surface and sorbate are opposite in sign (Gong and Grainger, 2007; Poncin-Epaillard et al., 2012). This capability is integral to the Target Immobilization procedure described below:  1. Immobilize the target protein (50 – 100 nM target in TI buffer) to the Nunc Immobilizer Amino C8 strips by incubating the filled well overnight at 4 °C (see Figure 2.3).  Covalent immobilization of the target onto the well surface proceeds through reaction of amino groups (or other nucleophiles) on the protein with the electrophilic coupling agents displayed on polyglycol chain ends of the Nunc Immobilizer well. The manufacturer recommends conducting this reaction at pH 9.6, but we have found that the use of a basic pH creates possibilities for chemical modifications (e.g., de-amidation) that alter target protein structure, chemistry and/or activity.  Hi-Fi SELEX therefore uses the alternate but equally efficient coupling conditions described in steps 1 – 7 of the Hi-Fi SELEX protocol. Under these milder conditions, the reaction proceeds more slowly and, in coordination with selection of an appropriate target solution concentration, permits facile control of the surface density of immobilized target.    40   Figure 2.3: Schematic of chemistries used to immobilize target protein, neutralize reactive groups, and passivate non-specific binding sites on the surface of a Nunc Immobilizer Amino plate.  The hydrodynamic diameter of a random 80-mer ssDNA aptamer is ca. 8 nm.  We therefore set the solution concentration (50 – 100 nM) of the target protein in the coupling reaction so as to achieve a mean distance of separation of immobilized target molecules that is a bit greater than 8 nm (i.e., 9 – 10 nm).  As an example, for thrombin (MW = 37 kDa) this is achieved using a solution concentration of 80 nM. A lower concentration would be required for targets of higher molecular weight, and vice versa. To determine the optimal solution concentration, the bound moles of target are measured as a function of solution concentration using a mass balance and A280 nm values for the starting and final solution. The contact area of the well (0.95 cm2) is then used to compute the average distance of separation of immobilized target as a function of the starting solution concentration used.  2. Following target immobilization, the wells are washed 3 times with 300 µl of SP buffer.  During this wash sequence, the amines (Tris (hydroxyl-methyl-aminomethane)) present in SP buffer neutralize unreacted electrophiles  41  displayed on the end-grafted polyglycol surface. Supplementing the SP buffer used in both the 3X wash and 1X incubation steps with 0.005% Tween 20 is absolutely essential to fully passivate the surface against non-specific retention. 3. Fill the wells once more with 300 µl of SP buffer. 4. Incubate for 1 hr under gentle agitation (we set a thermomixer fitted with a plate adapter at 500 rpm) at room temperature to complete the neutralization reaction. 5. Aspirate the SP buffer out of each well at the end of the neutralization/ passivation process. 6. Perform a final wash (3X of 300 l of SP buffer) before sealing the wells with optical film. 7. Store the processed and sealed C8 strip at 4 °C until further use.  2.3.3 Partitioning and Retained Fraction Recovery Prior to Hi-Fi SELEX screening, the library is subjected to a thermal denaturation/ renaturation cycle designed to promote annealing of the 5’-Comp and 3’-Comp strands to each 80-mer library member, as well as folding of each member into its thermodynamically favourable conformation at screening conditions (SantaLucia and Hicks, 2004).  In conventional SELEX, one round of screening in the absence of the molecular target is often first performed in an attempt to reduce the number of members of the library retained through mechanisms unrelated to the target. This step is not required in Hi -Fi SELEX, as the unique surface passivation methods used reduces non-specific retention to undetectable levels as shown in Figure 2.4.  42   Figure 2.4: Comparison of the percentage of a starting library that is non-specifically adsorbed to a standard SELEX library screening surface (“Conventional SELEX”; MyOne magnetic beads (ThermoFisher Inc.)) and to our passivated form (“Proposed Method”) of the Nunc Immobilizer surface.  In both cases, no target is displayed on the surface and reactive groups on the surface have been neutralized.  The data show that more than 2% of the starting library is non-specifically retained on the MyOne beads, while undetectable amounts (by qPCR analysis) are recorded for our system when 0.005% Tween 20 is present.  Approximately 108 unique library sequences are generally retained and recovered in the first partitioning step of Hi-Fi SELEX – a number optimal for the remaining steps of the selection process. The library partitioning and retained fraction recovery protocol is as follow: 8. Mix 1 nmole (~1014 sequences) of the ssDNA library from the prepared stock solution with equimolar amounts of 5’-Comp and 3’-Comp in 100 l of SP buffer. 9. Heat that mixture to 95C for 5 min (denaturation) before slowly cooling it down to 25C at a rate of 0.5C min-1 (renaturation) in a standard thermal cycler.  43  10. Add the resulting 10 µM of competent library to a neutralized and passivated Nunc Immobilizer well displaying the target.  11. Equilibrate that system for 1 hr at 25C under gentle agitation (500 rpm) in a thermomixer equipped with a plate reader.   The competent library concentration used (10 µM), in combination with setting the surface density of immobilized target to achieve a mean separation distance of ~ 9 nm, results in library screening within the theoretically preferred 100:1 to 1000:1 aptamer-to-target range (Irvine et al., 1991; Vant-Hull et al., 1998), while also eliminating the possibility of bridging of bound library members between proximal targets, an effect that has been shown to confound aptamer selection (Ozer et al., 2013).    12. Remove unbound and weakly bound members by washing 3 times with 300 l of SP buffer.  13. Apply a second more stringent pair of 3X washes with 300 l of SW buffer. These washing steps do not require agitation.  The success of Hi-Fi SELEX in reliably and efficiently discovering useful aptamers requires strict adherence to this bind and wash protocol during each selection round.  The combined use of an aptamer-to-target ratio of 100:1 to 1000:1, conditioned solvents, and high-salt wash steps creates the stringent partitioning conditions required to reduce the sequence diversity 105 to 106 orders of magnitude in each of the first few selection rounds.  It should be recognized that the polyanionic form of aptamers naturally creates the unwanted potential to over-select for sequences whose binding is dominated by coulombic (electrostatic) interactions.  The second set of high-salt washes employed in Hi-Fi SELEX are designed to remove library members that bind the target through ion-exchange type mechanisms that generally lack sufficient specificity for the target.  During these processes, it is important to ensure that the immobilized target and aptamers are always kept hydrated. When done properly, the passivation step performed after target immobilization will significantly increase the hydrophilicity of the Nunc Immobilizer Amino well surface, in part due to the presence of Tween-20. The surface of the wells thereby remains hydrated during various required solution exchanges provided vacuum  44  aspirators are not used. A single or multichannel pipette should instead be used to add and remove solutions. 14. Elute the retained pool of specifically bound library members from the Nunc wells by incubating with 100 µl of DE buffer for 10 min at 70 °C and an agitation rate of 600 rpm.  15. After agitation, neutralize the pH by adding 100 µl of 50 mM HCl. 16. Repeat steps 14 and 15. 17. Combine the eluted aptamer pools.  18. Desalt that pool by exchange into AF buffer and concentrating to 25 µl (final) using a centrifugal filter unit operating at 14,000g for 15 min. Begin that centrifugation without addition of AF buffer so as to remove as much DE buffer as possible. Then add AF buffer and begin buffer exchange.    The molecular weight of the 80-nt aptamer is 24 kDa. We therefore use a 10 kDa molecular weight cut-off membrane for this buffer-exchange step (step 18) to ensure good aptamer recovery yields. Buffer exchange should proceed until the final concentration of salt in the aptamer pool falls to within the mM range.   2.3.4 Retained Pool Amplification by ddPCR Following each selection round, the polymerase chain reaction (PCR) is generally used to amplify the pool of retained library sequences. In SELEX, this step is conducted as a conventional bulk PCR reaction using a standard thermal cycler and universal forward and reverse primers targeting the fixed flanking sequences of each library member.   That approach tends to be problematic, as illustrated in Figure 2.5 for a relatively simple pool of 105 unique library members.  Though the desired 80-bp dsDNA amplicon is created, a maximum in its total abundance is typically reached after a limited number of cycles.  Beyond that cycle, various artifacts, including formation of 80-mer hetero-duplexes hybridized together through only their common flanking sequences, oligonucleotide stretches within the universal primer regions mis-priming certain variable core region sequences, and improperly extended products acting as spurious primers on  45  heterologous sequences, promote conversion of the library to increasingly aberrant high molecular weight (HMW) by-products. To avoid these complications, Hi-Fi SELEX replaces traditional PCR with the partitioning capabilities of droplet digital PCR (ddPCR). When as few as 105 retained-library members are recovered in a given Hi-Fi SELEX round and then amplified by the ddPCR-based protocol described, an 80-bp amplicon concentration of greater than 1 M is generally realized in the final ~25 µl sample. Moreover, all amplicons produced are in their fully complementary (homo-duplexed) dsDNA state. As a result, a regenerated ssDNA library can be created from as little as 105 retained members in quantities sufficient to not only proceed to the next selection round, but also to determine the mean binding affinity of the enriched pool after each selection round, providing a metric of how the overall selection is proceeding.  The amplification protocol used in Hi-Fi SELEX is conducted in two steps as follows:  2.3.4.1 Determination of the Concentration of the Retained Library CLibrary 19. Dilute 0.5 µl of the desalted concentrated pool of retained 80-nt library members in 9.5 µl nanopure water. 20. To a qPCR well, add 5 µl of the solution created in step 19, along with SYBR green mastermix (to 1X final), FP and RP (250 nM final each). Top up the final volume in the well to 20 µl by adding nanopure water.  21. Begin cycling with an initial activation step at 95°C for 3 min followed by 39 cycles of amplification, each comprised of denaturation at 95°C for 30 s and annealing/extension at 60°C for 30 s. Set the heating and cooling rates at 2.5°C s-1 for even heat distribution in the well. 22. Determine CLibrary by comparing the recorded quantitation cycle Cq to corresponding Cq data from a standard curve derived from a dilution series of the initial 100 µM library stock with sequence diversity from 109 down to 105 unique library members.  An example standard curve is shown in Figure 2.6.  46   Figure 2.5: Comparison of amplification of a retained pool (105) of library members by conventional PCR (standard SELEX; upper panel) and by ddPCR (Hi-Fi SELEX; lower panel): Although the desired 80-bp dsDNA amplicon is created in conventional PCR, a maximum in its total abundance is typically reached after a limited number of cycles.  Beyond that cycle, various artifacts, including formation of 80-mer hetero-duplexes hybridized together through only their common flanking sequences, oligonucleotide stretches within the universal primer regions mis-priming certain variable core region sequences, and improperly extended products acting as spurious primers on heterologous sequences, promote conversion of the library to increasingly aberrant high molecular weight (HMW) by-products.  A small-scale “pilot” PCR reaction (Lou et al., 2009; Nieuwlandt, 2000) is therefore generally performed to determine the maximum number of PCR cycles that can be conducted before accumulating unacceptable amounts of by-products, which are known to adversely affect selection and must therefore be removed by gel electrophoresis or other means (Musheev and Krylov, 2006). That pilot reaction typically shows that standard PCR amplification of the retained pool must be stopped at ca.  22 to 25 cycles since HMW by-products generally start accumulating when the amplicon concentration reaches ca. 20-50 nM. Termination of amplification at this relatively low cycle number generally yields ~1010 to 1012 80-bp amplicons, or ≤ 1% of that needed to initiate the next selection round.  As a result, in SELEX, the PCR step must be multiplexed across 100 or more parallel reactions, each amplifying between ~ 105 to 106 library members to create 1011 to 1012 amplicons per well. The products of the parallel reactions are then pooled and concentrated to reach the concentration (i.e. 1014 amplicons in 100 µl) required for downstream processing and the next round of SELEX. The use of emulsions in ddPCR to isolate and amplify single templates by PCR is well established, and it is known that the resulting partitioning of single templates into individual droplets reduces formation of unwanted by-products when co-amplifying mixtures of templates (e.g. multiple genes) (Nakano et al., 2003; Williams et al., 2006). Spurious priming events are greatly reduced within each droplet, in part because competition between different templates and biases resulting from differences in amplification efficiencies are avoided (Margulies et al., 2005). Moreover, post amplification, the emulsions can be broken to recover the full set of amplicons in an aqueous phase suitable for downstream processing. In Hi-Fi SELEX, the pool (~108) of competent 80-nt ssDNA library members retained after a selection round is therefore partitioned among a similar number of nL-sized droplets. ddPCR partitions ca. 20,000 droplets per well, which means 100 wells are required to accommodate 50 templates into each droplet (CPD=50). As a result of the low sequence heterogeneity per droplet, minimal HMW by-products formation is observed over 40 or more ddPCR cycles, permitting high-fidelity end-point amplification of all retained library members into more than 1014 total copies of the desired 80-bp dsDNA amplicon products.  47    Figure 2.6: Representative standard curve for determining sequence diversity.  Quantitation cycle (Cq) data for diversity standards from 109 unique sequences to 105 unique sequences plotted as green circles. The E-value reported represents the efficiency of the qPCR amplification, which should be between 95% and 105% for the standard curve to be accepted. CLibrary, the starting library concentration, may also be determined from the standard curve by taking the starting quantity (e.g. ~107; green X in figure) estimated using the measured Cq (qPCR quantitation cycle) and dividing by 5 µl and then multiplying by the dilution factor used to prepare the sample.  2.3.4.2 Droplet-Based Amplification of the Retained Pool As described in Section 2.1, the number of reaction wells required (and thus total droplets generated) is defined both by the size of the retained sample and by the average number of library members per droplet, also known as the copies per droplet (CPD = –ln(empty droplets/total read droplets)), employed. In this thesis work, we observed that the maximum number of cycles over which end-point amplification of the library is achieved with minimal accumulation of HMW by-products depends on the CPD employed, as shown in Figure 2.7.  For example, ddPCR at a CPD of 500 results in by-product formation and a concomitant loss of desired 80-bp amplicon product (and product quality) after 24  48  amplification cycles.  The results shown allow one to optimize the chosen CPD used based on the amount of purified retained library recovered in a given round of screening.  In general, anywhere from 106 library members to 1010 library members may be recovered due to a variety of factors, including the selected target, the quality of the partition step, the stringency of the wash steps, etc.    Figure 2.7: Determination as a function of average copies per droplet (CPD) of the maximum cycle number that may be employed to amplify the retained library by ddPCR without formation of unwanted HMW by-products.   In Hi-Fi SELEX, the results shown in Figure 2.7 are specifically applied to libraries retained in the first 2 rounds of selection, where each of the templates within each droplet is almost certain to harbour a unique sequence within its variable core region.  In later rounds of selection, duplicate library sequences will increasingly partition into any given drop, diminishing template heterogeneity per drop and the probability of forming by-products during amplification.  Higher CPDs may then be used, up to a CPD of 5000.  y = -3.877ln(x) + 47.522R² = 0.98421015202530354010 100 1000 10000Optimum cycle numberCopies per droplet (CPD) 49  23. From CLibrary and the retained library volume (~ 25 µl) the number of ddPCR wells required to conduct the retained library amplification is computed by assuming 17,000 readable drops are created per ddPCR well.  The number of retained library members is then used in combination with Figure 2.7 to set the CPD per droplet and the associated number of required reaction wells.  If, for example, the retained library contains 108 members (typical for 1st and 2nd rounds of Hi-Fi SELEX), a mean CPD of 62 is required to conduct the entire amplification reaction on a single 96-well plate (i.e., 96 wells*17000 droplets*62 copies/droplet = 108).   24. From the remaining 24 to 25 µl of desalted retained library, prepare an appropriate volume (= 20 µl x # of required wells (= ~ 2 to 2.4 mL)) of ddPCR sample mixture by adding ddPCR master mix (1X final), 900 nM each of FP and phosphorylated RP (final) and nanopure water. 25. Load a 20 µl aliquot of this sample mixture into each well of a droplet-generation cartridge.  26. Add 70 µl of fluorinated oil to each of the corresponding oil wells of the cartridge.  27. Insert and process the cartridge in the droplet generator.  28. Transfer the stable emulsion (containing ca. 17,000 readable droplets) formed in each well into a well of a standard 96-well PCR plate for thermal cycling. 29. Repeat steps 25 to 28 until the sample is fully processed. 30. Begin cycling with an initial activation step at 95C for 5 min, followed by 35 amplification cycles (note that this is fewer cycles than typically used for ddPCR quantitation), each comprised of denaturation at 95C for 30 s then annealing/extension at 64C for 30 s. Set the heating and cooling rates at 2.5C s-1 to ensure even heating of all droplets in each well. 31. Immediately after amplification, pool all of the wells and spin at 5000g to separate the reacted droplets from the continuous oil (bottom) phase.  32. Discard the continuous oil phase.  33. Recover the double-stranded DNA by subjecting the droplet phase to a freeze (80C for 15 minutes)/thaw cycle.  34. Immediately spin down the frozen droplets at 14,000g for 5 min to create sufficient force to burst them.   50  35. Repeat steps 33 and 34 two more times to generate and recover a clear aqueous (top) phase containing the soluble amplified material. Set aside 5 µl of the clarified aqueous phase. 36. Process the remaining clarified aqueous phase in a centrifugal filter unit to concentrate the amplified library to ~25 µl. 37. Analyze the 5 µl aliquot from step 36 on a 1.5% agarose gel alongside a properly sized molecular weight ladder to confirm proper amplification of the library.    2.3.5 Measuring the Sequence Diversity of the Amplified Retained Pool In general, a proper balancing of retained library sequence diversity and mean binding affinity must be maintained across selection rounds for a Hi-Fi SELEX screening to prove successful in discovering a useful set of candidate aptamers.  Maintaining that balance requires a method to quantify both the mean Kd and the total sequence diversity of the retained pool after each selection round. Retained library diversities can be determined using next-generation sequencing, but that approach is neither fast nor inexpensive (Hoon et al., 2011; Schütze et al., 2011).  In Hi-Fi SELEX, the sequence diversity of a retained pool of library members is therefore estimated using a novel qPCR assay that is simple and inexpensive, requiring equipment common to all molecular biology labs. The area of the melting peak for each population is recorded and the two peak areas are used to compute the fraction fhetDNA of amplicons in the hetero-duplexed state:  𝑓ℎ𝑒𝑡𝐷𝑁𝐴 =𝐴67℃(𝐴67℃ + 𝐴81℃) (2.3)  This melt analysis is repeated for serial reductions in the sequence diversity of the dsDNA-amplicon representations of the starting library (see above) to generate data as shown in Figure 2.8. The corresponding peak areas for the set of serial reductions are used to create a standard curve relating fhetDNA to the known sequence diversity (Figure 2.9).    51   Figure 2.8: Fluorescence (SYBR green) based melt analysis for serial reductions in the sequence diversity of dsDNA-amplicon representations of an 80-nt Hi-Fi SELEX library: The double stranded amplicons of retained library members are homogeneous in terms of their two flanking sequences, while presenting a highly diverse ensemble of variable core-region sequences.  Denaturation of the amplified library followed by cooling to 55 °C therefore results in two distinct dsDNA populations: fully homo-duplexed amplicons characterized by a Gaussian melting envelope centered at a relatively high melting temperature (Tm ~ 81 °C), and hetero-duplexes exhibiting only partial complementarity (typically through only their common flanking sequences). The hetero-duplexed pool of amplicons collectively exhibits a Gaussian melting peak characterized by a much lower Tm (~ 67 °C).     52   Figure 2.9: Standard curve for qPCR-based sequence diversity determination: normalized melt peak areas (A) and fHetDNA values (B) as a function library diversity.   For those standards, the area of the Tm = 67 °C melting peak decreases with decreasing sequence diversity, reflecting the higher probability of forming fully complementary homo-duplexes (which melt at Tm = 81°C) in such systems.  The value of fhetDNA measured for a  53  retained pool thereby permits estimation of its sequence diversity.  I note that this qPCR-based diversity assay only provides reliable estimates of sequence diversity for retained libraries comprising ≥ 1000 unique variable-region sequences. Hi-Fi SELEX screening is typically stopped before the retained library diversity falls to that limit, as the target mean Kd of the retained population has generally been met.   The complete protocol for measuring library sequence diversity is as follows:  38. Amplify the remaining 5 µl volume of the diluted working aliquot prepared in step 19 by qPCR using SYBR Green master mix (1X final) and qPCR thermal cycling conditions comprised of initial activation at 95C for 5 min followed by 13-15 amplification cycles: denaturation at 95C for 30 s, annealing at 64C for 30 s and extension at 72C for 30 s, with the heating and cooling ramp rate set at 3C s-1.    The number of cycles is set so as to generate sufficient amplified material from the amount of retained library present in the previously quantified starting aliquot. Typically, ca. 13 to 15 cycles is sufficient to generate the ≥ 10 ng of 80-bp amplicon required for the diversity assay. Note that cycling should be stopped (~ 16 to 20 cycles) before production of HMW amplification by-products is observed in gel-based visualization of the reaction product. The pilot PCR reaction is performed first to determine the optimal cycle number, where a small amount of the pool is amplified using standard 50 µl PCR. A sample is taken out every 3 cycles and a gel-based visualization is performed similar to that in Figure 2.5. The optimal cycle number should yield enough material to proceed to the next step. The amount of product can be quantified using NanoDrop™ ND2000 spectrophotometer. The remaining steps of the protocol are as follows:  39. Following cycling, cool the SYBR dye-bearing amplicons to 55°C.  40. Then, melt by heating from 55C to 95C in 0.5C increments in a real-time PCR thermal cycler and record the required A67°C and A81°C values.  54  41. Determine the sequence diversity from the standard curve relating fhetDNA to the known library diversity (e.g. Figure 2.9).    2.3.6 Enzymatic Regeneration of the Single-Stranded Library The ddPCR amplification step generates 80-bp dsDNA products from which the sense strands must be recovered in order to continue the Hi-Fi SELEX process. A number of purification methods have been established for stoichiometric removal of the antisense strand from the sense (aptamer library) strand of amplicons, including alkaline denaturation followed by streptavidin capture of biotinylated antisense strands (generated by chemically modifying the RP) (Kai et al., 1998) or electrophoretic separation of poly-T-labeled antisense strands (Pagratis, 1996). The relative performance of these various methods has been studied by Civit et al. (Civit et al., 2012) and others (Avci-Adali et al., 2010; Svobodová et al., 2012), who have collectively shown that enzymatic digestion of 5’-phosphorylated (PO4) antisense strands with -exonuclease generally works best.   However, the oligonucleotide hydrolysis activity of λ-exonuclease is appreciably lower on ssDNA than on complementary dsDNA (Lee et al., 2011). As a result, enzymatic regeneration of a ssDNA library stalls when λ-exonuclease acts on hetero-duplexed amplicons produced in the bulk PCR amplification step employed in standard SELEX.  A partially regenerated library comprised of a mixture of desired 80-nt ssDNA and partially processed dsDNA material is therefore created (Figure 2.10, right half), compromising the next round of selection. In Hi-Fi SELEX, in addition to mitigating formation of unwanted HMW products, the ddPCR based amplification of retained members described above provides a means to eliminate formation of hetero-duplexes during amplification and thereby yield a pool of fully homo-duplexed amplicons (as evidenced by melt analyses of those pools (Figure 2.10)).  Complete stoichiometric regeneration of the required 80-nt ssDNA library by λ-exonuclease processing can then be achieved (Figure 2.10, left half) using the following protocol:  42. Generate the sense-strand library from the 25 µl pool of concentrated amplicons prepared in step 36 by reacting with 5 U of λ-exonuclease at 37C for 1 hr, followed by heat inactivation at 75C for 10 min.  55  2.3.7 Purification and Recovery of the Regenerated ssDNA Library Following processing of the ddPCR-amplified product with -exonuclease, the ssDNA library must be separated from the spent enzyme and the hydrolysis products.  Traditionally, this is done in SELEX technology using a combination of phenol-chloroform extraction to remove the enzyme and isolate to 80-nt library, and ethanol precipitation to recover that library in a concentrated form that can then be resuspended in an appropriate buffer.  This approach is inspired by the proven effectiveness of the technique in purifying chromosomal DNA and other large DNA fragments (Buckingham and Flaws, 2007; Tan and Yiap, 2009).  However, as shown in Figure 2.11, while the method can be used to also purify short ssDNA such as the 80-nt library of interest, yields tend to be quite poor, which can result in a mass of purified library that is insufficient for conducting the next round of Hi-Fi SELEX. We therefore developed a new purification protocol that uses phenol-chloroform extraction in combination with membrane-assisted (spin columns) buffer exchange to purify and recover the regenerated library.  Significantly higher yields (~ 50%) are obtained (Figure 2.11), assuring that sufficient material will be available for subsequent rounds of selection. That purification protocol is as follows:   43. Purify the digested product using standard phenol/chloroform extraction by adding to the sense-strand library a 2X volume (i.e. ~ 50 µl) of phenol:chloroform:isoamyl alcohol (25:24:1). 44. Vortex to create an emulsion and centrifuge at 14,000g for 5 min at room temperature. 45. Recover the top aqueous phase and repeat steps 43 and 44 once more. 46. Add 500 µl SP buffer into a NanoSep® Omega 3K MWCO centrifugal filter unit. 47. Centrifuge at 2,500g for 15 min at room temperature. 48. The solubility of phenol in water is ~6.5% (ScienceLab.com, 2013). Dilute the recovered top aqueous phase containing traces of phenol with an appropriate amount of SP buffer to bring down the concentration of phenol from ~6.5% to ≤0.2% (Scientific, 2017), a concentration which is appropriate for PCR.  56  49. Centrifuge the diluted sample using the rinsed NanoSep® Omega 3K MWCO centrifugal filter unit at 2,500g to a volume of ~25 µl. Following buffer exchange, the retained library is typically at a concentration of ca. 1 µM, which is suitable for the next round of selection.   Figure 2.10: Efficient stoichiometric regeneration of the ssDNA library in Hi-Fi SELEX: (A) ddPCR amplification of retained members results in formation of fully complimentary homo-duplexed amplicons, while conventional bulk PCR used in SELEX yields a mixture of homo- and hetero-duplexed amplicons; (B) -exonuclease processing of the homo-duplexed amplicons produced by Hi-Fi SELEX results in stoichiometric recovery of the ssDNA library in 60 min, while relatively little ssDNA product is recovered from the bulk PCR product.  2.3.8 Measuring the Mean Kd of the Retained and Regenerated Library In conventional SELEX, the selection process is typically monitored by measuring the mean Kd of retained pools after blindly conducting a few selection rounds to enrich sufficient amounts of high-affinity binders to make Kd determination possible (typically by surface plasmon resonance (SPR) or fluorescence spectroscopy).  Parallel PCR amplifications followed by antisense strand removal are required to produce the required  57  quantities of regenerated ssDNA library.  If the mean Kd is to be determined by spectroscopy, that process must include fluorescent labeling or biotinylation of the regenerated library; no modification is required if SPR is to be used.  Either process is time consuming, fails to provide information in the critical early rounds of selection, and is costly if SPR is used to measure mean Kd values. Moreover, a risk of altered binding properties that bias selection is created if modification of library members is required.     Figure 2.11: Comparison of yields (expressed as % recovery) of regenerated 80-mer ssDNA library material recovered by the standard method and the new protocol post phenol/chloroform extraction. • Standard Method: Ethanol precipitation • New Method: ssDNA retained and buffer exchanged on a membrane that was pre-rinsed with SP buffer. Spin speed for ssDNA isolation and buffer exchange set at 2,500g.  To reduce costs and eliminate need for specialized SPR or fluorescence spectroscopy equipment, Hi-Fi SELEX incorporates a novel qPCR-based method to quantify the mean Kd. The method is based on creating a standard relation to quantify the concentration of library CLibrary in a given sample before and after incubation with an immobilized target, and then using that  58  knowledge to measure the adsorption isotherm for binding of the regenerated library to protein target presented on Nunc Immobilizer Amino plates as described above. This simple qPCR method has been used to show that Hi-Fi SELEX typically delivers a retained pool offering a mean Kd ≤ µM after the first round of selection, and a mean Kd of order nM after 3 rounds of selection.  It may be used to monitor the mean Kd after each round of selection and to terminate selection when the mean bulk affinity of retained members either remains unchanged for consecutive rounds or has reached a suitable value (often a mean Kd of order nM) for a particular application.  It requires accurate determination of CLibrary values, necessitating a statistically more robust measurement protocol. The protocol is as follow:  50. 2X dilute in SP the retained 25 µl of 80-nt ssDNA library.  51. Take 4 µL of that diluted library and dilute it 1000-fold in SP.  52. From step 51, prepare a serial dilution set (2X, 4X, 8X, 16X, 32X, 64X, 128X, 256X) to a final volume of 500 µL each. 53. Subject two 5 µl aliquots of each dilution to qPCR-based determination using steps 20 and 21 described above.  54. Record the mean Cq value (for a given RFU threshold value T) and standard error computed from the duplicates for each dilution. 55. In a base-10 semi-logarithmic graph, plot the threshold cycle versus the dilution factor and fit the data to a straight line and record the slope and error in it.  56. Compute the amplification efficiency E from that slope as 10-1/slope -1.   57. Determine CLibrary from the Cq value for each dilution using the standard relation  𝑙𝑜𝑔(𝐶𝐿𝑖𝑏𝑟𝑎𝑟𝑦) = 𝑙𝑜𝑔(𝑇) − 𝐶𝑞𝑙𝑜𝑔(𝐸) (2.4)  Determination of library concentrations can of course be done using other methods.  In my experience, the popular NanoDrop™ ND2000 spectrophotometer is not sufficiently accurate in measuring dsDNA at concentrations below ca. 50 µg/mL, typically overestimating the true  59  concentration. The Qubit™ Fluorescence Monitor is generally accurate at these concentrations, and one can use it as a secondary check of CLibrary values.  58. In duplicate, incubate each of the 8 serial dilutions (200 µl each) in a  target-presenting Nunc well for 1 hr at 25°C with gentle mixing at 300 rpm.  59. Remove the solution and then successively wash the retained pools in the equilibrated wells 3X with 300 l of SP.  60. Remove the final wash from each well. 61. Elute the retained members in 100 µl of 50 mM NaOH at 70°C.   62. Immediately neutralize the eluted pool from each well with 100 µl of 50 mM HCl.  63. Add 100 µL of 10 mM TE buffer. 64. Dilute 5 µl of each neutralized eluted pool by 50-fold to make the library concentration appropriate for analysis.   65. In duplicate, quantify the CLibrary in each neutralized eluent by qPCR according to steps 20 and 21. 66. For each dilution, determine the fraction  of the library that is bound to the target as the ratio of the dilution-corrected values of CLibrary for the eluted and initial samples.  The concentration of library remaining in the solution phase (Cfree) after equilibration with the immobilized target is given by the difference in dilution-corrected CLibrary values for the initial and eluted libraries, respectively.   67. From the  and Cfree values, construct the binding isotherm as illustrated in Figure 2.12. Nonlinear fitting of the Langmuir isotherm relation   𝜃 =𝐶𝑓𝑟𝑒𝑒𝐾𝑑 + 𝐶𝑓𝑟𝑒𝑒  (2.5)  to the binding isotherm data may then be used to estimate a value for the mean Kd. A more advanced method, detailed in the Appendix, that segregates the retained population into high-affinity and low-affinity fractions may instead be used to estimate the mean Kd for each of these fractions.   60   Figure 2.12: Representative adsorption isotherm and regressed mean Kd value.  Isotherm data determined using the qPCR-based binding analysis method and then fit to equation 2.5: data shown are for a retained pool of binding library members (after 3rd round of Hi-Fi SELEX) with variable regions having high affinity for human mesothelin.            00.20.40.60.811.20 5 10 15 20 25 30 35 40Bound Fraction, θConcentration, Cfree (nM) 61  Chapter 3 Discovery of DNA Aptamers by Hi-Fi SELEX and Their Evaluation as Ligands for Preparative Affinity Chromatography 3.1 Introduction The purification of therapeutic proteins produced by either natural tissues or recombinant cell culture is essential to the delivery of biologic drugs that provide a required therapeutic activity while also meeting the more general specifications (purity, stability, etc.) for safety and clinical use mandated by regulatory agencies.  All steps in the production of therapeutic proteins must therefore be in compliance with Good Manufacturing Practices (GMP) as regulated by the US Food and Drug Administration (FDA), or by similar agencies in other jurisdictions. They must also ensure that the purified biologic product is fully active and unaltered chemically. Thus, purified biologic drugs in their administered formulation must have acceptably low levels of known deleterious contaminants, including viruses, pyrogens (< 300 endotoxin units per dose), host DNA (< 10 pg per dose) and other macromolecules, or any foreign agents (e.g. leachates) introduced by the downstream process. The unit operations used, which typically include various forms of chromatography and membrane-based separations, must deliver these required product specifications while acting gently on the product itself, as protein structures and chemical compositions can be irreversibly changed by exposure to conditions that diverge from those comprising the protein’s natural in vivo environment. Affinity chromatography exploits bio-molecular recognition events ubiquitously found in nature to capture and purify a biologic molecule of interest from a complex feedstock.  Many proteins are known to have natural ligands to which they bind tightly and specifically.  The most common example is antibodies raised against the protein (the antigen) of interest, but natural ligands also include inhibitors, substrates, and effectors of the target. In affinity chromatography, the ligand is immobilized onto a base stationary phase media that provides good mechanical support, large surface areas, and excellent mass transfer.  Columns packed with a base media to which an affinity ligand has been immobilized thereby permit facile capture of the target protein, extensive washing to remove all unbound contaminants, and then elution of the purified and concentrated product through the introduction of a solvent that serves to disrupt the target-ligand complex. Because of the natural specificity of the ligands employed, affinity chromatography is  62  often preferred as the first chromatographic step in industrial downstream processes, particularly in systems where the immobilized ligand to be used is readily available, robust, cost effective and highly specific.  As described in Chapter 1 of this thesis, protein A and variants of it meet these criteria, and as a result affinity chromatography employing an immobilized protein A type ligand is almost universally used by industry to capture and purify monoclonal antibody (mAb) therapeutics directly from clarified cell culture supernatants (Arora et al., 2017; Koguma et al., 2013; Pabst et al., 2014; Tsukamoto et al., 2014).  Since the protein A ligand interacts specifically with the target to be purified, mAb separations by affinity chromatography tend to be robust and predictable/repeatable, qualities that are essential to a sound downstream process. Indeed, a simplified overall process design is often realized because the capture performance of the affinity ligand is generally less sensitive to the composition of the feedstock. However, other examples of the industrial use of affinity chromatography in production-scale downstream processes approved by regulatory agencies are rare, not because industry deems the technology unattractive for non-mAb products, but rather because the required ligands are either not available, too labile (fragile) to withstand column sterilization and sanitization processes, or are too costly for large-scale (i.e. preparative) use.  Due to its speed and robustness, the Hi-Fi SELEX technology offers the potential to rapidly discover DNA aptamers that might prove effective as “designer” affinity ligands against non-mAb biologic targets of interest.  In this chapter, I report data and analyses collectively designed to evaluate the potential of this “custom affinity chromatography” concept. The creation and validation of the proposed custom-chromatography development pipeline requires successful demonstration of five integral components, namely (i) discovery of candidate DNA aptamer ligands using the Hi-Fi SELEX method, (ii) sequencing, synthesis and characterization of candidate ligands, (iii) construction and basic performance evaluation of mini-columns bearing a candidate aptamer ligand, (iv) when required, chemical modification of candidate ligands to improve their stability, and (v) proof-of-concept testing of the performance of mini-columns bearing chemically stabilized ligands on actual cell-culture feedstocks.  I begin by reporting results on the application of Hi-Fi SELEX to the discovery of DNA aptamer ligands against two different target proteins, human mesothelin and human Factor D.  One of the aptamers discovered against Factor D is then evaluated in terms of its performance as  63  an affinity ligand for capturing and purifying recombinant human Factor D from a clarified host-cell (Chinese hamster ovarian cells) supernatant.  3.1.1 Human Factor D The 24-kDa serine protease complement Factor D (Factor D) has a critical function in the human complement system. The complement system, often simply called complement, is a component of normal blood plasma that facilitates the opsonization (an immune process where bacteria are targeted for destruction by an immune cell known as a phagocyte) of pathogenic bacteria. This activity is said to “complement” the natural antibacterial activity of antibodies, hence the name.  Complement is therefore an intrinsic part of the human innate immune system. The complement system is made up of a large number of different plasma proteins that interact and/or react with one another to opsonize pathogens and induce a series of inflammatory responses that help fight infection. A number of components of complement are proteases (including human Factor D) that are themselves activated by proteolytic cleavage. There are three distinct pathways through which complement can be activated on the surface of pathogens.  One of these, known as the alternative pathway, can be initiated when a spontaneously activated component of complement binds to the surface of the pathogen. It is of therapeutic interest because the pathway can proceed on many pathogen surfaces in the absence of a specific antibody against that pathogen. Circulating levels of human Factor D are known to modulate the activity of the alternative pathway. As a result, Factor D itself, as well as inhibitors of Factor D, are now being investigated as potential therapeutics for controlling innate immunity in humans (see, for instance, (Lorthiois et al., 2017)).   3.1.2 Human mesothelin Human mesothelin is a 40 kDa membrane protein present at low levels on normal peritoneal, pleural, and pericardial mesothelial cell surfaces.  It is over-expressed in many cancer types, most notably in pancreatic and ovarian cancers as well as lung adenocarcinomas.  The stark difference in mesothelin’s expression in normal and cancerous tissues have made it an attractive target for cancer therapy. Immunotoxins and mAbs targeting mesothelin are now undergoing clinical trials as potential cancer therapies (Hassan et al., 2016).  Mesothelin-targeted  64  CAR-T cells are also under development as a potential treatment for pancreatic cancer (Jiang et al., 2017).  However, DNA aptamers against mesothelin have not been considered or discovered, despite the fact that, if effective, they would potentially offer a much cheaper therapeutic option.   In collaboration with Professor Urs Häfeli (UBC Dept. of Pharmaceutical Sciences), who has interest in developing new cancer therapies (see, for example, (Bleul et al., 2013)), I therefore applied the Hi-Fi SELEX platform to the discovery of anti-mesothelin DNA aptamers.  3.2 Methods and Results 3.2.1 DNA Aptamer Discovery by Hi-Fi SELEX The second-generation High-Fidelity Systematic Evolution of Ligand by Exponential Enrichment (Hi-Fi SELEX) method described in Chapter 2 of this thesis was used to discover a panel of candidate DNA aptamers, first against human Factor D and then against human mesothelin.  In both cases, the protocol used followed the detailed step-by-step procedure presented in Chapter 2 of this thesis. Human mesothelin and human Factor D were obtained from G&P Biosciences (Santa Clara, CA) and Complement Technology Inc. (Tyler, TX) respectively. Human -thrombin, which was used in the model development studies reported in the Appendix, was purchased from Haematologic Technologies (Essex Junction, VT). Nunc Amino Immobilizer plates (Lock Well C8) were purchased from Thermo Fisher (Edmonton, AB). The ssDNA Hi-Fi SELEX library used in these studies was composed of an internal variable (random) region 40 nucleotides (N40) in length, with each member flanked by a 20-mer universal 3’ primer binding sequence and by a common 20-mer 5’ sequence. The resulting semi-combinatorial 80-mer library was synthesized by mixing the DNA nucleotides C:G:A:T at a molar ratio of 3:2:3:2.4 to achieve equal probability incorporation of each nucleotide in all sequences. The reverse complement to each flanking sequence was synthesized. Those complementary sequences, denoted as 5’-Comp and 3’-Comp, respectively, are provided in Table 3.1. As shown in that table, the 3’-Comp sequence was 5’-phosphorylated and also served as the reverse primer (REV) for PCR amplification.  All oligonucleotides were synthesized and HPLC purified by Integrated DNA Technologies, Inc. (IDT, Inc.; Coralville, IA) and  65  reconstituted in 1X Aptamer Folding (AF) buffer (20 mM Tris-HCl pH 7.4, 140 mM NaCl, 5 mM KCl, 1 mM MgCl2, 1 mM CaCl2).  Table 3.1: Sequences of ssDNA Hi-Fi SELEX library and reverse complement sequences Name Sequence (5’ – 3’) Comments Library Member TCGCACATTCCGCTTCTACC– N40 –CGTAAGTCCGTGTGTGCGAA N40 = random nucleotide region of the Hi-Fi SELEX library FOR TCGCACATTCCGCTTCTACC 5’ amplification primer REV or 3’-Comp p-TTCGCACACACGGACTTACG 3’ phosphorylated amplification primer / complementary blocker 5’-Comp GGTAGAAGCGGAATGTGCGA 5’ complementary blocker  For each target (Factor D or mesothelin), per-round mean Kd values obtained during Hi-Fi SELEX screening of the competent semi-combinatorial library are reported in Table 3.2. For both targets, the mean Kd value for the retained aptamer pools is of order nM after 3 selection rounds, indicating the successful enrichment and isolation of a pool of very high-affinity candidate aptamers. This is achieved due to the fact that Hi-Fi SELEX screenings against each target yield per-round PE values in excess of 8 x 105.  Table 3.2: Round-by-round mean dissociation constants (Kd) and standard deviations measured by duplicate qPCR experiments for enriched pools selected against various therapeutic targets using Hi-Fi SELEX.  Hi-Fi SELEX Round Mean Kd (nM) Factor D Mesothelin Round 1 74 ± 22 Undetectable Round 2 17 ± 3 28 ± 10 Round 3 7.2 ± 2.4 2.0 ± 0.9   66  The measured binding isotherm for the retained pool of library members after 3 rounds of Hi-Fi SELEX is shown in Figure 3.1 for Factor D and in Figure 3.2 for mesothelin.  In each case, over 75% of the retained library members are characterized by a mean Kd of 10-9 M, as exemplified in the pseudo-component binding-affinity histogram for Factor D (Figure 3.3).  That histogram was obtained by analyzing the round-3 binding isotherm for Factor D (Figure 3.1) using the advanced algorithm described in the Appendix, which partitions the population of retained members into a set of two “pseudo-components”, with each pseudo-component characterized by a unique Kd (in this case, a high-affinity pseudo-component of average Kd = 10-9 M, and a lower-affinity pseudo-component of average Kd = 10-7 M).  The algorithm developed and described in the Appendix then minimizes the chi-squared error (2) of the data-fit of the resulting model, which is effectively the multicomponent Langmuir Isotherm model, using the Levenberg-Marquardt numerical method.  From this analysis, the fraction of all retained library members that are high-affinity binders may be estimated after each round of selection.  Figure 3.1: Mean binding isotherm and standard deviations measured by duplicate qPCR experiments for complexation of the retained library to Factor D after 3 rounds of Hi-Fi SELEX.   67   Figure 3.2: Mean binding isotherm and standard deviations measured by duplicate qPCR experiments for complexation of the retained library to mesothelin after 3 rounds of Hi-Fi SELEX.   Figure 3.3: Pseudo-component histograms for Round 3 of Hi-Fi SELEX applied to DNA aptamer selection against human Factor D.  In round 3, the isotherm was best fit using two pseudo-components with Kdi values of Kd2 = 10-7 M and Kd1 = 10-9 M.  68  3.2.2 Sequencing and Characterization of Final Retained Pools Library members in final retained Hi-Fi SELEX pools were sequenced (as dsDNA) following ddPCR amplification and purification by the UBC Sequencing and Bioinformatics Consortium (SBC) using an Ion Torrent™ Ion PGM next generation sequencing system (ThermoFisher).  The required Ion PGM Template 200 and Ion PGM Sequencing 200 kits were purchased from Life Technologies, Inc. (Carlsbad, CA). Primers required for Ion PGM sequencing of library members were purchased from IDT (Coralville, IA), and an NEBNext Fast Library Prep Set kit (New England Bio-Labs) was used to process samples for sequencing using the standard Prep Set kit reagents supplemented with 1 nM DNA PCR reaction mix, 10 μM each of IDT Ion Torrent A primer, a truncated version of the FOR primer, and REV primer. Amplified samples were purified using an AMPure XP kit (Agencourt, Inc.), and then analyzed on a 2100 Bioanalyzer (Agilent, Inc.) to confirm size and concentration. Sequence frequency was analyzed using Genomics Workbench™ software (CLC Bio, Inc.) with the conserved flanking sequences digitally removed prior to analysis. Core (random region) sequences of some of the top (most frequent) anti-factor D DNA aptamers identified by next-generation sequencing within the final retention pool after 3 rounds of Hi-Fi SELEX are reported in Table 3.3 along with their binding affinity (Kd) for human Factor D as measured by surface plasmon resonance (SPR) on a Biacore 3000 SPR instrument (GE Healthcare).  A representative SPR sensorgram for one of the anti-Factor D aptamers (aFD-30) is shown in Figure 3.4, and the secondary structure of candidate anti-Factor D aptamers as predicted by M-Fold freeware (http://unafold.rna.albany.edu/?q=mfold) is shown in Figure 3.5.         69  Table 3.3: Core random-region sequence and binding affinity (Kd) of some of the highest frequency members recovered from the retained pool after Round 3 of Hi-Fi SELEX applied to DNA aptamer selection against human Factor D. Aptamer ID Core random-region sequence (5’ – 3’) Kd (nM) aFD-14 GTAACCACGTTGCCAGACCGAGTCTACCAGCGATCCTCAG      4.6 (± 0.5) aFD-2 ACGGAGAAAGAGAGAGTGTAATTGCTAGCATAACCGCTGC   15.2 (± 3.2) aFD-30 TATGCCCAAATCCCTCAAGTCGGCCAGGATACACCACCGT 0.71 (± 0.25) aFD-35 TCGGCCTTCCCAGACCACCGCAATCCCCAGGGAACAGGCA    19.4 (± 3.5) aFD-57 AATCAAAAGGCTCACGCGCGGATTGGTCAACCTTACAACC     8.1 (± 2.7) aFD-12 ACCAGGCACCCGACGGACTAACTCATCACTCAGGCGAGGG      4.8 (± 2.0) aFD-36 AACCCGCATGCCGATCGATGTCGTGCCTCGCTCCACGCTC     9.5 (± 3.1) aFD-13 CATCACACTGTCAACATACCCAGCCTGGGGAAAGACGAAC     4.8 (± 1.6)  * S.D. values computed from n = 3 independent SPR experiments    Each of the candidate anti-Factor D aptamers discovered by Hi-Fi SELEX is predicted to fold into a unique secondary structure that enables it to bind human Factor D with very high affinity.  Measured Kd values for the sequenced pool range from ~20 nM to 0.7 nM, and these affinities are comparable to the average strength of an antibody-antigen complex. The predicted folds of the candidate aptamers differ, suggesting that some or all of them may target a unique epitope on Factor D.  But given their high affinities, the DNA aptamers discovered by Hi-Fi SELEX and then sequenced are clearly suitable for further study as possible ligands for affinity capture of recombinant human Factor D from cell culture supernatants. Results were similar for Hi-Fi SELEX based screening of DNA aptamers against human mesothelin.  In this case, next generation sequencing and SPR characterization of sequences recovered from the final (round 3) pool of retained library members were used to identify 3 candidate anti-mesothelin (aMT) aptamers showing high affinity for the target (Table 3.4). M-Fold predicted secondary structures for those candidate aptamers are reported in Figure 3.6.   70   Figure 3.4: SPR sensorgrams for binding of Factor D to immobilized biotinylated aFD-30 aptamer (Kd = 0.71 ± 0.25 nM).  Experiments were performed on a Biacore 3000 using an SA chip. Several concentrations of Factor D were injected, and the results fitted to a 1:1 Langmuir isotherm.  Table 3.4: Core random-region sequence and binding affinity (Kd) of some of the highest frequency members recovered from the retained pool after Round 3 of Hi-Fi SELEX applied to DNA aptamer selection against human mesothelin. Aptamer ID Core random-region sequence (5’ – 3’) Kd (nM) aMT-1 TGCAGGGATCGGTCAGCTTATTCAACATCAAGTCTTATGC 67.0 (± 24.9) aMT-4 ACGCAACGAGGCCGCACCAAACGCTTATGTCTTAGCGAAA 21.7 (± 4.8) aMT-6 GGCCATTGGGATATAGCCGATCCAACCCAAACTTCCCTAG 59.7 (±19.7)  * S.D. values computed from n = 3 independent SPR experiments                                Response (RU)  	 71    Figure 3.5: M-fold predicted secondary structures for the full-length blocked aptamer sequences against Factor D whose core random-region sequences are reported in Table 3.3.  Several of the candidate aptamers fold into bulge-hairpin motif (e.g. aFD-14), while others assume a bifurcated bulge structure (e.g. aFD-30).                  aFD-2  aFD-14  aFD-30  aFD-57               aFD-35  aFD-13    aFD-36       aFD-12 	 72   Figure 3.6: M-fold predicted secondary structures for the full-length blocked aptamer sequences against mesothelin whose core random-region sequences are reported in Table 3.4.  3.2.3 Preliminary Evaluation of Candidate Ligands To demonstrate the applicability of DNA aptamers as immobilized ligands for preparative affinity chromatography of complex cell-culture feedstocks, the aFD-30 anti-Factor D aptamer (Table 3.3) was selected for further study.  The decision to focus on an anti-Factor D aptamer discovered by Hi-Fi SELEX was made at the request of the industrial sponsor of this thesis work, Bio-Rad Laboratories Inc. (Bio-Rad). Their request was motivated in part by the availability of a Chinese hamster ovarian (CHO) cell line engineered to produce recombinant human Factor D and secrete that target protein into the culture supernatant. The available ATF4 (activating transcription factor 4) overexpressing 13D CHO cell line producing recombinant human Factor D was established by dihydrofolate reductase (DHFR) mediated gene amplification. Cells were maintained in adherent culture at 37°C in the minimum essential medium α-MEM (STEMCELL Technologies, Inc.) without glucose or L-glutamate. The medium was supplemented with 2.2 g/L sodium carbonate, 0.292 g/L L-glutamine, 2 g/L glucose, 100 mg/L streptomycin sulfate, 58.8 mg/L penicillin G, 5 μM methotrexate, 1 mg/L insulin, 0.5% soy peptone solution, 330 μl/L trace metal solution, and 0.5% fetal bovine serum (SAFC Biosciences, Lenexa, KS, USA).  73  The 13D CHO cells were grown in batch culture for 120 hours using -MEM medium supplemented with 0.3 g/L L-glutamine, 2.5 g/L glucose, and 250 μg/mL G418. The viable cell density was determined by the trypan blue exclusion assay. The Factor D concentration in the culture supernatant was measured by a sandwich enzyme-linked immunosorbent assay (ELISA) using a rabbit anti-human Factor D antibody (DAKO, Glostrup, Denmark) and a peroxidase-conjugated sheep anti-human Factor D antibody (Cedarlane, Hornby, Ontario, Canada) as the first- and second-step antibodies, respectively. Pure plasma-derived human Factor D (Complement Technology Inc.) was used as a standard protein both for calibrating the ELISA assay and for determining static and dynamic binding capacities within aFD-30 immobilized affinity chromatography columns.  A representative time course of the viable cell density and recombinant human Factor D concentration in the culture supernatant during batch cultures of the 13D CHO cells is shown in Figure 3.7. Both the final cell density (8.1 (±0.2) x 105 viable cells mL-1 and the final concentration (17 (±2) mg L-1) of recombinant human Factor D are relatively low, but nevertheless sufficient for the proof-of-concept studies reported here.  From a 500-mL 120-hr batch culture, 427 mL of clarified culture supernatant were recovered by first processing the final culture in a Beckman Coulter Allegra X-12™ centrifuge (6000 rpm for 10 min), and then passing the decanted supernatant through a Millistak B1HC disposable lab-scale depth filter pod (MilliporeSigma) to further reduce turbidity to a desired level (<15 NTU) based on measurements using a LaMotte turbidity meter.   74   Figure 3.7: Time course of the viable cell density and recombinant human Factor D concentration in the culture supernatant during batch cultures of the 13D CHO cells used to produce recombinant human Factor D.   For candidate aptamer testing as an affinity ligand, the aFD-30 aptamer sequence was synthesized (IDT Inc.) with 5’ amine group added.  That group enables facile coupling of the candidate ligand to Profinity™ Epoxide preparative chromatography media (Bio-Rad Inc.).  The reaction between the epoxy group and the 5’-amino-ssDNA aptamer proceeds via a standard nucleophilic ring opening of the epoxide by the terminal amine. The 5′ amine group can make a covalent bond with epoxide groups, enabling end coupling of the aptamer to a proven preparative chromatography support offering good mechanical and flow/mass-transfer properties.    Pre-packed 5-mL Profinity™ Epoxide mini-columns (Bio-Rad Laboratories, Inc.) were loaded and incubated at 37°C for 6 hours with a solution containing a 100 mg L-1 5’-amino aFD-30 that was pre-adjusted to pH 9.5 by titration with 1M NaOH.  After the DNA immobilization, excess epoxide groups were masked through equilibration for 2 additional hours with a capping buffer (pH 9) containing 0.1 M Tris with 50 mM ethanolamine.  The resulting ligand-displaying  75  mini-columns were then subjected to a 50 CV (column volume) wash with AF buffer (pH 7.4) to remove excess reactants and equilibrate the column for sample loading. For use in determining the static binding capacity (SBC), aFD-30-loaded Profinity™ Epoxide media was also prepared in bulk (i.e. in free unpacked form) following the synthesis procedure described above.  The mass of aFD-30 aptamer loaded into the 20-mL reaction was used along with the mass of aFD-30 remaining in solution following the immobilization reaction to determine by mass balance the concentration of aFD-30 ligand loaded onto the stationary phase (52.6 mg mL-1). This mass-balance driven method is known as the depletion method, as determination is based on depletion of analyte within the solution phase due to immobilization on a surface phase.  The equilibrium isotherm for binding of pure Factor D to the aFD-30 bearing stationary phase, again measured by the depletion method over a range of Factor D concentrations, is shown in Figure 3.8.  AF buffer at 20 °C was used as the solvent for these binding experiments.  Under these favorable loading conditions, the affinity media is capable of specifically binding and retaining 35 ± 3 mg of Factor D per mL of stationary phase (SBC = 35 (± 3) mg mL-1).  This SBC is consistent with those generally observed in preparative affinity chromatography columns, including protein A columns, for which the SBC usually lies between 10 and 70 mg mL-1, with the larger SBC values reported when the target protein binding to the column is of low molecular weight.  Moreover, the data show that ~ 2/3 of the aFD-30 ligand immobilized onto the column is oriented or displayed such that it is capable of binding the target protein.  Breakthrough curve data for the pure protein were then measured as a function of flowrate by frontal loading of 0.52 g L-1 human Factor D (mobile phase: AF buffer, pH 7.4, 20°C) onto the pre-packed 5-mL Profinity™ Epoxide mini-column bearing immobilized aFD-30 aptamer. From those data, the dynamic binding capacity (DBC) at each flowrate (Table 3.5) was determined from the area behind the breakthrough curve up to the point where ~15% breakthrough of Factor D was observed. The results show that, due to the favorable hydrodynamic and mass transfer properties provided by the base Profinity™ Epoxide media, the affinity column maintains good Factor D capture potential under vendor-recommended flowrates.   76   Figure 3.8: Equilibrium adsorption isotherm at 20°C for binding of pure Factor D to the aFD-30 bearing stationary phase (AF buffer was the solvent for these binding experiments).  The data indicate a static binding capacity of 35 (± 3) mg mL-1.   Prepared affinity mini-columns were then challenged with a clarified and concentrated (10-kDa MWCO membrane used to diafilter supernatant into 1X AF buffer) 13D CHO batch culture supernatant containing 0.48 g L-1 recombinant human Factor D.  Figure 3.9 reports a representative chromatogram for the separation when the loaded column was washed with 5 CV of AF buffer and the captured protein then eluted isocratically in DE buffer. Densitometry analysis of the associated sodium dodecyl-sulfate polyacrylamide gel electrophoresis (SDS-PAGE) documentation of the separation (Figure 3.9) on a Typhoon™ 9500 scanner (GE Healthcare, Inc.) was used to quantify separation performance (Table 3.5).  For samples loaded onto a previously unused column (i.e. first column cycle), purities of recovered Factor D (n = 4 replicates) were in excess of 96%, while yields were somewhat lower (86 (± 3) %) due to the fact that column loading conditions were un-optimized. As a result, the column oversaturated, allowing some of the Factor D introduced onto the column to remain in the flow-through (FT).   77  Nevertheless, the SBC, DBC, purity and yield obtained for the first column cycle represent very good column performance at levels that are in line with industry standards.     Figure 3.9: Measured chromatogram for affinity capture and purification of recombinant human Factor D from clarified supernatant of a 13D CHO cell supernatant.  The pre-packed 5-mL Profinity™ Epoxide mini-column bore immobilized 5’-amino-aFD-30 aptamer at a density of 53 mg mL-1. Clarified supernatant containing 0.48 g L-1 Factor D that had been conditioned for affinity separation by diafiltering into 1X AF buffer (pH 7.4) was frontally loaded at 0.4 CV min-1 (CV = column volumes) to approximately 15% breakthrough.  The loaded column was washed with 5 CV of AF buffer and the captured protein was then eluted by an isocratic step gradient into DE buffer. The associated sodium dodecyl-sulfate polyacrylamide gel electrophoresis (SDS-PAGE) documentation of the separation is also shown, with lane 1 showing the MW standards, lane 2 the clarified diafiltered supernatant (SN), lane 3 the column flow through (FT), and lane 4 the pooled eluent peak fraction (EL).  Data are for a previously unused column.     A280(mAU)DCVs						CVsMW		SN		FT		EL  SN   FT   EL  78  Table 3.5: Average separation performance observed for the first cycle of a pre-packed 5-mL Profinity™ Epoxide mini-column bearing immobilized 5’-amino-aFD-30 aptamer.  Data are for n = 4 replicates, and basic column operating conditions are as described in Figure 3.9.  Separation Quality Indicator  Value Static Binding Capacity (SBC) Dynamic Binding Capacity (DBC)           0.4 CV min-1           0.6 CV min-1           0.8 CV min-1 Factor D Yield (@ 0.4 CV min-1) Factor D Purity (@ 0.4 CV min-1) Concentration Factor (@ 0.4 CV min-1) 35 ± 3         mg mL-1  28.2 ± 1.4   mg mL-1 28.0 ± 0.9   mg mL-1 27.1 ± 1.2   mg mL-1 86 (± 3) % 96 (± 2) % 8.6 ± 0.3   However, column performance declined with repeated cycling, as evidence in the measured losses in yield and DBC with increasing cycle number shown in Figure 3.10.  Specifically, the observed reduction in DBC suggests loss of immobilized ligand function with repeated exposure to CHO culture supernatant. As those supernatants are known to possess nuclease activity (Yuk et al., 2015), most notably exonuclease activity that could catalyze the hydrolysis of the immobilized aptamer, efforts to stabilize the aptamer via chemical modification were explored.  3.2.4 Candidate Aptamer Stabilization Against (exo)Nuclease Catalyzed Degradation When engineered carefully, specific chemical modifications may be used to create aptamers exhibiting remarkable chemical and thermal stabilities (Ni et al., 2017; Verma and Eckstein, 1998). These include modifying the chirality of the aptamer though use of Spiegelmers (non-natural l-nucleotides) in place of standard nucleotides (Vater and Klussmann, 2015).  The base and sugar structure of terminal or internal nucleotides may likewise be changed through a  79  variety of available chemical modifications, including locked nucleic acids, non-natural isodG/isodC base pairs, or 2’-amino purines.     Figure 3.10: Dynamic binding capacity (DBC) as a function of column cycle number (n = 2).   However, as CHO culture supernatants are known to display exonuclease activity, modification of the 3’-end of the 5’-amino-aFD-30 aptamer was explored as a means to stabilize the ligand, once immobilized, against exonuclease-catalyzed hydrolysis. A modified 5’-amino-aFD-30 aptamer bearing a nuclease-resistant 3’ inverted dT (deoxythymine) nucleotide was therefore synthesized (IDT, Inc.) and then immobilized onto pre-packed 5-mL Profinity™ Epoxide mini-columns as described above.  The mass of the resulting 5’-amino-aFD-30-inverted-dT-3’ aptamer that could be loaded onto the stationary phase (50.3 (±0.8) mg mL-1) was comparable to that achieved with the unmodified ligand.    80  3.2.5 Proof-of-Concept Testing of Stabilized Aptamer Columns on CHO Feedstocks In order to assess variability in affinity media performance, three independent mini-columns bearing the 3’-stabilized form of the aFD-30 DNA aptamer ligand were again subjected to repeated loading and elution cycles of clarified 13D CHO supernatant.   Each column was also subjected to prolonged exposure to a 1N NaOH solution to simulate repeated column sterilization steps.  Average DBC, purity and yield data as a function of column cycle number or following prolonged exposure to 1N NaOH are reported in Table 3.6.  Those data show that end-protection of the 5’-immobilized aptamer with a terminal non-natural inverted dT is highly effective in stabilizing the ligand, and that the basal exonuclease activity within CHO supernatants was causative of the loss in DBC observed with the unmodified aFD-30 ligand.  The modified affinity columns are seen to purify the recombinant Factor D product from clarified CHO supernatant at a purity in excess of 95% and a yield in excess of 85% over 30 column cycles, as well as following prolonged sanitation.  Together these data therefore provide a meaningful and promising demonstration of the potential of the proposed technology for rapid, custom design and validation of preparative chromatography media.   Table 3.6: Average separation performance observed for the first cycle of a pre-packed 5-mL Profinity™ Epoxide mini-column bearing immobilized modified 5’-amino-aFD-30-inverted-dT-3’ aptamer.  Data are for n = 2 replicates (errors are average errors), and basic column operating conditions are as described in Figure 3.9. Step DBC (mg/mL) Purity (%) Yield (%)  Concentration Factor  Unmodified Ligand, cycle 1 Modified, cycle 1 Modified, cycle 10 Modified, cycle 30 Modified, 30hr NaOH hold 28 ± 2 26 ± 2 28 ± 1 27 ± 2 26 ± 2 96 ± 1 95 ± 2 95 ± 1 96 ± 0 95 ± 2 86 ± 3 88 ± 2 87 ± 1 86 ± 2 85 ± 3 8.6 ± 0.2 8.8 ± 0.1 8.7 ± 0.1 8.4 ± 0.2 8.6 ± 0.4      81  Chapter 4 Conclusions and Proposed Future Work 4.1 Summary of Thesis Work Biotechnology companies are now highly advanced in their technical ability to produce biologic therapeutics, most notably monoclonal antibodies, for the treatment of major diseases. Both those technologies remain costly when applied to biologics for which a standard platform purification process is not applicable or whose annual demand does not require production at large scales.  Both conditions apply to orphan drugs used to treat rare diseases.  Biologics with potential to be classified as an orphan drug, are typically not monoclonal antibodies, and are often produced at scales of less than a few 100 kg per year.  But their health benefit is potentially large, as more than 25 million people in the US alone are afflicted with rare diseases whose treatment is possible with biologic therapeutics. As most orphan drugs used to treat rare diseases are not mAb-based, there exists a need to develop new methods that offer improved production-process economics and performance at smaller scales. Advances in upstream processing over the past decade have enabled the cost-effective production of protein therapeutics at high yield and titres.  Downstream processing (DSP) technology must now likewise improve to make orphan drugs affordable. In DSP, the single highest cost of operation most often lies in the initial product capture step, which can represent up to 50% of the total manufacturing costs. Therefore, development of technologies that enable rapid discovery and implementation of relatively inexpensive custom capture media could serve to greatly improve the overall economics to produce orphan drugs.  Protein A is the gold standard of this concept, as protein A affinity chromatography columns are almost universally used for the purification of mAbs because they deliver purities above 98% and offer a fast return on investment. The success of that capture technology has motivated academic researchers and industry to develop robust and cost-effective affinity ligands against non-antibody protein targets. Peptide-based ligands and synthetic small-molecule ligands are both being actively pursued, for example.   But published data suggest that peptide ligands suffer from low binding affinities/capacities, limited life cycles, and low scale-up potential, while effective synthetic ligands are difficult and expensive to discover, in part because the libraries used to screen for them generally exhibit low chemical diversity. Moreover, both of these ligand  82  classes are prone to be toxic and can trigger an immunogenic response in humans in the event of ligand leakage.  In contrast, aptamers are a relatively new class of non-toxic, non-immunogenic molecules that have received no significant attention to date as a potential source of ligands for preparative affinity chromatography.  Aptamer libraries of very large diversity can be screened to discover members offering high affinity and specificity for a protein target. They can also be chemically modified to increase ligand stability to withstand harsh bioprocessing operations, giving them a significant potential for use at industrial scales.  Aptamer discovery by Systematic Evolution of Ligands by Exponential Enrichment (SELEX) was first described and performed in 1990. Since then, variations of SELEX designed to improve the selection process have been described. However, those variations share a few common limitations, including non-specific retention of library members on the display matrix and inefficiencies and by-products formation during amplification of retained members by PCR.  This thesis therefore describes a greatly improved DNA aptamer selection method, High-Fidelity SELEX (Hi-Fi SELEX), that enables cheap, reliable, and rapid isolation of high quality aptamers. In its second-generation form described in this thesis, Hi-Fi SELEX begins with the incubation of ~1014 competent aptamer library members with a target displayed on the surface of a micro-titre plate that has been neutralized and passivated against non-specific adsorption. Upon equilibration, the bound members are eluted from the target and then amplified using droplet digital PCR to restore the total quantity of library members back to ~1014. The improved ddPCR protocol can amplify any amount of retained library up to a limit of 109 unique sequences. After regenerating the ssDNA from dsDNA amplification product using λ-exonuclease catalyzed digestion, phenol/chloroform extraction is combined with membrane-assisted buffer exchange to remove enzyme and digested nucleotides to recover the desired pure ssDNA library.  The second half of this thesis explored, at the proof-of-concept level, the potential to exploit Hi-Fi SELEX technology within a pipeline aimed at rapidly discovering and testing (validating) DNA aptamers that could be used as custom affinity chromatography ligands for the capture and initial purification of non-mAb proteins. Hi-Fi SELEX was therefore applied to the discovery of DNA aptamer ligands against human mesothelin and human Factor D. An anti-Factor D aptamer (aFD-30) with binding affinity 0.71 ± 0.25 nM was selected for preliminary evaluation of the custom affinity chromatography pipeline concept.  The aFD-30 aptamer was immobilized on  83  Profinity™ Epoxide preparative chromatography media (Bio-Rad Laboratories, Inc.). The resulting media showed high-affinity for human Factor D, exhibiting a static binding capacity (SBC) of 35 ± 3 mg mL-1. Mini-columns packed with the affinity media exhibited a dynamic binding capacity (DBC) of 28.2 ± 1.4 mg mL-1 when operated at the vendor recommended flowrate of 0.4 CV min-1 (CV = column volumes), delivering recombinant Factor D from a frontally loaded clarified CHO culture supernatant with a yield and purity of 86 ± 3% and 96 ± 2%, respectively, for the first purification cycle, results which are in line with industry standards. However, both DBC and yield were seen to decrease with increasing cycle number, suggesting loss of ligand function with repeated exposure to culture fluids. Efforts to improve the stability of the ligand were therefore made by modifying the 5’-amino-aFD-30 aptamer to bear a 3’terminal nuclease-resistant inverted dT (deoxythymine) nucleotide. The modified affinity ligand exhibited excellent stability against nuclease activity, and could be used to purify Factor D to > 95% at a yield > 85% for over 30 column cycles.  The immobilized modified ligand could also withstand prolonged exposure to 1N NaOH, which is frequently used for column sterilization between cycles.  Though limited, these results therefore demonstrate the potential of the proposed technology for custom design and validation of preparative chromatography media that can benefit the growing orphan drugs market by reducing manufacturing costs and thus the associated costs of treatments.   4.2 Proposed Future Work Though the results presented in this thesis work support the potential of DNA aptamers as a general source of useful ligands for custom affinity chromatography, a number of additional issues need to be addressed if industry is to widely adopt the pipeline.  These include:  4.2.1 Rapid Screening and Discovery of Effective Eluents  Affinity chromatography requires loading of crude sample, selective binding and retention of the target on the affinity matrix, washing of unbound contaminants out of the column, and finally elution of the desired protein target. All of these steps must be conducted in ways that do not damage the integrity and therapeutic activity of the target. Proteins are sensitive towards chemical changes, which often lead to irreversible denaturation and subsequently lower  84  product yield. However, due to the high affinity of the ligand-target complex employed in affinity chromatography separations, identifying an eluent that stoichiometrically releases the target protein at conditions that do not alter its structure or chemistry can be difficult.  Biospecific elution and nonspecific elution are the two methods used in industry to elute protein targets from a column. Nonspecific elution, which generally involves alteration of the mobile-phase pH, ionic strength and/or polarity, is preferred over biospecific elution (i.e. using competitive ligands for target displacement) because it is cheaper, faster, and often provides a sharper elution peak.  The development of an elution strategy is very much dependent on the interaction between the target and ligand. As a result, no strategy can be universally leveraged to find an appropriate elution buffer that is mild and does not extensively denature the product. The most common approach is essentially empirical (trial and error) screening, which is time and reagent consuming (Firer, 2001). Enzyme-linked immunosorbent assay (ELISA) elution screening has been proposed as a more efficient and cheaper method of screening buffers of different compositions and pH at small scale before selecting the appropriate buffer for full scale affinity column (Kummer and Li-Chan, 1998). ELISA, however, requires the use of antibodies to detect the presence of protein target. This can be time-consuming and expensive.  Surface plasmon resonance (SPR) is an alternative to ELISA for screening elution conditions, and is appealing in that it is a label-free method that does not require the availability of antibodies. When coupled with statistical techniques such as factorial design or response surface methodology, preliminary elution trends can be established using SPR experiments and promising elution buffers then analyzed further on small-scale columns to assess protein purity and yield using SDS-PAGE.  But SPR is relatively low throughput and time consuming.  That bottleneck could in principle be overcome using a new high-throughput form of SPR known as SPR imaging, or SPRi.  Novel microfluidic SPRi arrays recently described (Ouellet et al., 2010) could permit multiplexed detection of lead candidates by quantifying equilibrium binding of target to immobilized candidate aptamers as a function of solvent composition in an element-addressable fashion. With an ability to examine hundreds of candidate aptamer ligands in parallel, the technology could provide the ability to independently interrogate candidate aptamers discovered by Hi-Fi SELEX both for their selectivity for the target protein (measured by using  85  SPRi to show no binding of CHO supernatant components to the immobilized ligand (see section 4.2.3)) and for their ability to release the target under non-damaging conditions.  4.2.2 High-Throughput Methods to Evaluate and Improve the Stability of Aptamers  Like many biomolecules, unmodified aptamers are susceptible to physical and chemical degradation, particularly when exposed to culture fluids. An aptamer may degrade through a drastic change in solvent (chemical) composition or via digestion by endogenous and/or exogenous nucleases. In affinity chromatography, nucleases may be present in the crude samples loaded onto the column and the harsh clean-in-place procedures periodically used for column sanitization can likewise lead to ligand changes or losses.  After selection and sequencing of a panel of candidate ligands by Hi-Fi SELEX, what is required (or at least desired) is an in silico method that allows one to quickly screen each to assess 1) its chemical (resistance to nuclease degradation) and thermal (resistance to denaturation) stability, and 2) its potential for simple modifications that improve its stability.  This may involve a hybrid computational/brute-force screening method that includes immobilizing the candidate ligands on an appropriate solid support (e.g: Nunc wells, SPRi sensorchip array, small-scale chromatography column, etc.), and then exposing each to clarified CHO cell supernatant in the absence of protein target for an extended period. The quantity of functional ligands on the solid support can thereby be quantified before and after the exposure to determine losses and, thus, susceptibility to degradation. A loss in ligand function can then be attributed to the presence of nucleases in the supernatant using available nuclease activity assays (e.g., change in solution viscosity associated with hydrolysis of sperm-whale DNA), and tandem mass spectrometry of the aptamer fragments produced can be used to identify the sites of nuclease-catalyzed cleavage. The most promising (stable) ligands could thereby be identified. There are many known ways to chemically modify an oligonucleotide to improve its stability (Verma and Eckstein, 1998).  However, with aptamers, those chemical modifications can serve to reduce or eliminate its affinity for its target due to resulting changes in structure.  It is crucial to understand, and ideally be able to predict, the effect of a given modification on the structure of an aptamer ligand. Many bioinformatics tools are available to predict the 3D structure of ssRNA, but software for predicting the 3D structure of ssDNA is limited. However, a new study reports an algorithm that can predict the hairpin structures often observed in ssDNA  86  aptamers (Jeddi and Saiz, 2017). Nucleotides that are vulnerable to nucleases can thereby be screened for their location in the folded structure, allowing partitioning of candidate ligands into those that are likely amenable to chemical modification at the 2’-position of the deoxyribose-phosphate backbone from those that are not. This modification is essential to inhibit endonuclease activity, while exonuclease activity can be eliminated at the 5’ and 3’ ends of the ligand through a variety of strategies, including addition of PEG tails or an inverted dT (deoxythymine) nucleotide.   4.2.3 A Method for Determining Ligand Specificity   Aptamers have been shown to distinguish targets based on their unique chemical and structural features. But not all are highly specific, as the polyphosphate backbone of ssDNA can non-specifically bind cationic contaminants.  Strategies for selecting highly specific aptamers can be embedded into the Hi-Fi SELEX screening process itself, for instance by incorporating a counter-selection step in the second round onwards against non-targets.  Much like immobilizing targets on a Nunc well, non-targets can be immobilized and presented to the library. Members that bind to the non-targets will be removed, allowing unbound members to be collected after successive washing.  Conversely, members within the pool of candidate aptamers isolated by Hi-Fi SELEX as described in Chapter 2 can be screened for specificity using an independent high-throughput method such as SPRi. The quantitative measurement or ranking of the specificity of an aptamer relative to other candidate ligands is a useful parameter to gauge expected overall ligand performance. Indeed, while the selection of high affinity aptamers is claimed to automatically lead to the isolation of highly specific aptamers (Eaton et al., 1995), this does not always appear true. It may therefore be of value in selecting a final aptamer by scoring it and others under consideration in terms of a Specificity Index, Is. This index value would be determined after sequencing of the final retained aptamer pool by synthesizing and then analyzing the top-ranking aptamers (in terms of sequence frequency) for specificity. Using either SPR or SPRi, high-affinity binding of the target may first be demonstrated.  The quantity of non-target host material from the parental CHO cell line supernatant that binds each candidate aptamer can also be determined. From these results, Is can be computed as   87  𝐼𝑠 =𝛼𝛼 + 𝛽 (4.1) where α is the quantity (captured mass) of the target complexed to the immobilized aptamer, and β is the quantity of the non-target complexed.  The Is value ranges from 0 to 1. For an aptamer that is highly specific, the Is will approach 1, while an aptamer having an Is of 0.5 can be considered as exhibiting moderate specificity. (A more qualitative method of determining specificity may be gained by simply observing differences in the SPR sensorgram during the ligand and non-targets interactions (Narayan and Lemmon, 2006)). The combination of Kd and Is should allow one to objectively consider and choose the best lead ligand for use in affinity chromatography.                        88  References Ahirwar R, Nahar P. 2015. Development of an aptamer-affinity chromatography for efficient single step purification of Concanavalin A from Canavalia ensiformis. J. Chromatogr. B 997:105–109. Arora S, Saxena V, Ayyar BV. 2017. Affinity chromatography: A versatile technique for antibody purification. Methods 116:84–94. Avci-Adali M, Paul A, Wilhelm N, Ziemer G, Wendel HP. 2010. Upgrading SELEX Technology by Using Lambda Exonuclease Digestion for Single-Stranded DNA Generation. Molecules 15:1–11. Bartel DP, Zapp ML, Green MR, Szostak JW. 1991. HIV-1 Rev Regulation Involves Recognition of Non-Watson-Crick Base Pairs in Viral RNA. Cell 67:529–536. Bartnicki F, Kowalska E, Pels K, Strzalka W. 2015. Imidazole-free purification of His3-tagged recombinant proteins using ssDNA aptamer-based affinity chromatography. J. Chromatogr. A 1418:130–139. Berezovski M, Musheev M, Drabovich A, Krylov SN. 2006. Non-SELEX: selection of aptamers without intermediate amplification of candidate oligonucleotides. J. Am. Chem. Soc. 128:1410–1411. Bleul R, Thiermann R, Marten GU, House MJ, Pierre TG St., Häfeli UO, Maskos M. 2013. Continuously manufactured magnetic polymersomes – a versatile tool (not only) for targeted cancer therapy. Nanoscale 5:11385–11393. Bonham-Carter J, Shevitz J. 2011. A Brief History of Perfusion Biomanufacturing. Bioprocess Int. 9:24–31. Bruno JG. 1997. In vitro Selection of DNA to Chloroaromatics Using Magnetic Microbead-Based Affinity Separation and Fluorescence Detection. Biochem. Biophys. Res. Commun. 234:117–20. Buckingham L, Flaws M. 2007. Molecular Diagnostics: Fundamentals, Methods, and Clinical Applications. Philadelphia: F.A. Davis Company 225-262 p. Burnouf T, Radosevich M. 2001. Affinity chromatography in the industrial purification of plasma proteins for therapeutic use. J. Biochem. Biophys. Methods 49:575–586. Camperi SA, Martínez-ceron MC, Giudicessi SL, Marani MM, Albericio F, Cascone O. 2014. Peptide Affinity Chromatography Based on Combinatorial Strategies for Protein Purification. Methods Mol. Biol. 1129:277–302. Cattani-scholz A, Pedone D, Blobner F, Abstreiter G, Schwartz J, Tornow M, Andruzzi L. 2009. PNA-PEG Modified Silicon Platforms as Functional Bio-Interfaces for Applications in DNA Microarrays and Biosensors. Biomacromolecules 10:489–496. Challener C. 2016. A Look At Perfusion. Bioprocess Int. 14:44–45.  89  Chames P, Van Regenmortel M, Weiss E, Baty D. 2009. Therapeutic antibodies: Successes, limitations and hopes for the future. Br. J. Pharmacol. 157:220–233. Chapman AP, Antoniw P, Spitali M, West S, Stephens S, King DJ. 1999. Therapeutic antibody fragments with prolonged in vivo half-lives. Nat. Biotechnol. 17:780–783. Cheng C, Chen YH, Lennox KA, Behlke MA, Davidson BL. 2013. In vivo SELEX for Identification of Brain-penetrating Aptamers. Mol. Ther. — Nucleic Acids 2:1–9. Cho M, Soo Oh S, Nie J, Stewart R, Eisenstein M, Chambers J, Marth JD, Walker F, Thomson JA, Soh HT. 2013. Quantitative selection and parallel characterization of aptamers. Proc. Natl. Acad. Sci. 110:18460–18465. Chomczynski P, Sacchi N. 2006. The single-step method of RNA isolation by acid guanidinium thiocyanate-phenol-chloroform extraction: Twenty-something years on. Nat. Protoc. 1:581–585. Citartan M, Tang TH, Tan SC, Hoe CH, Saini R, Tominaga J, Gopinath SCB. 2012. Asymmetric PCR for good quality ssDNA generation towards DNA aptamer production. Songklanakarin J. Sci. Technol. 34:125–131. Civit L, Fragoso A, O’Sullivan CK. 2012. Evaluation of techniques for generation of single-stranded DNA for quantitative detection. Anal. Biochem. 431:132–138. Clonis YD, Labrou NE, Kotsira VP, Mazitsos C, Melissis S, Gogolas G. 2000. Biomimetic dyes as affinity chromatography tools in enzyme purification. J. Chromatogr. A 891:33–44. Clonis YD. 2006. Affinity chromatography matures as bioinformatic and combinatorial tools develop. J. Chromatogr. A 1101:1–24. Coia G, Pontes-Braz L, Nuttall SD, Hudson PJ, Irving RA. 2001. Panning and selection of proteins using ribosome display. J. Immunol. Methods 254:191–197. Collins FS. 2015. A New Initiative of Precision Medicine. N. Engl. J. Med. 372:793–795. Cuatrecasas P. 1970. Protein Purification by Affinity Chromatography. J. Biol. Chem. 245:3059–3065. Cuatrecasas P, Wilchek M, Anfinsen CB. 1968. SELECTIVE ENZYME PURIFICATION BY AFFINITY CHROMATOGRAPHY. Proc. Natl. Acad. Sci. 61:636–643. Curling J. 2004a. Affinity Chromatography – from Textile Dyes to Synthetic Ligands by Design, Part 2. BioPharm Int. 17:60–66. Curling J. 2004b. Affinity Chromatography – from Textile Dyes to Synthetic Ligands by Design, Part 1. BioPharm Int. 17:34–42. Curling J. 2007. Process Chromatography: Five Decades of Innovation. BioPharm Int.:10–18. Dean PDG, Watson DH. 1979. PROTEIN PURIFICATION USING IMMOBILISED TRIAZINE DYES. J. Chromatogr. 165:301–319.  90  Deng Q, German I, Buchanan D, Kennedy RT. 2001. Retention and Separation of Adenosine and Analogues by Affinity Chromatography with an Aptamer Stationary Phase. Anal. Chem. 73:5415–5421. Dias DM, Ciulli A. 2014. NMR approaches in structure-based lead discovery: Recent developments and new frontiers for targeting multi-protein complexes. Prog. Biophys. Mol. Biol. 116:101–112. Eaton BE, Gold L, Zichi DA. 1995. Let’s get specific: the relationship between specificity and affinity. Chem. Biol. 2:633–638. Ecker DM, Jones SD, Levine HL. 2015. The therapeutic monoclonal antibody market. MAbs 7:9–14. Edwards BM, Barash SC, Main SH, Choi GH, Minter R, Ullrich S, Williams E, Fou L Du, Wilton J, Albert VR, Ruben SM, Vaughan TJ. 2003. The Remarkable Flexibility of the Human Antibody Repertoire; Isolation of Over One Thousand Different Antibodies to a Single Protein, BLyS. J. Mol. Biol. 334:103–118. Ellington AD, Szostak JW. 1990. In vitro selection of RNA molecules that bind specific ligands. Nature 346:818–822. Famulok M, Mayer G. 2014. Aptamers and SELEX in Chemistry & Biology. Chem. Biol. 21:1055–1058. Fields C, Li P, O’Mahony JJ, Lee GU. 2016. Advances in Affinity Ligand-Functionalized Nanomaterials for Biomagnetic Separation. Biotechnol. Bioeng. 113:11–25. Firer MA. 2001. Efficient elution of functional proteins in affinity chromatography. J. Biochem. Biophys. Methods 49:433–442. Forier C, Boschetti E, Ouhammouch M, Cibiel A, Ducongé F, Nogré M, Tellier M, Bataille D, Bihoreau N, Santambien P, Chtourou S, Perret G. 2017. DNA aptamer affinity ligands for highly selective purification of human plasma-related proteins from multiple sources. J. Chromatogr. A 1489:39–50. Fuhrman JA, Comeau DE, Hagström A, Chan AM. 1988. Extraction from Natural Planktonic Microorganisms of DNA Suitable for Molecular Biological Studies. Appl. Environ. Microbiol. 54:1426–1429. GE Healthcare Bio-Sciences AB. 2007. Affinity Chromatography: Principles and Methods. Gold L. 1995. Oligonucleotides as Research, Diagnostic, and Therapeutic agents. J. Biol. Chem. Gold L. 2015. SELEX: How It Happened and Where It will Go. J. Mol. Evol. 81:140–143. Gold L, Janjic N, Jarvis T, Schneider D, Walker JJ. 2012. Aptamers and the RNA World , Past and Present Aptamers and the RNA World , Past and Present. Cold Spring Harb. Lab. Press:1–11.   91  Gong P, Grainger DW. 2007. Nonfouling Surfaces: A Review of Principles and Applications for Microarray Capture Assay Designs. Methods Mol. Biol. 381:59–92. Gopinath SCB, Sakamaki Y, Kawasaki K, Kumar PKR. 2006. An Efficient RNA Aptamer against Human Influenza B Virus Hemagglutinin. J. Biochem. 139:837–846. Gronemeyer P, Ditz R, Strube J. 2014. Trends in Upstream and Downstream Process Development for Antibody Manufacturing. Bioengineering 1:188–212. Hage DS, Bailon P, Bian M, Briscoe CJ, Burks R, Chen J, Clarke W, Clonis YD, Ehrlich GK, Friedberg F, Gadgil H, Gustavsson P-E, Haupt K, Heegaard N, Jarrett HW, Jordan N, Jurado LA, Karle E, Kim HS, Krull IS, Labrou NE, Larsson P-O, Liu X-C, Long SD, Lough WJ, Mazitsos K, Moser AC, Moxley RA, Myszka DG, Nachman-Clewner M, Nelson MA, Oak S, Ohnmacht C, Patel S, Phillips TM, Rhoads AR, Ruhn PF, Scouten WH, Schou C, Spence CL, Subramanian A, Todorova D, Vijayalakshmi MA, Wa C, Wainer IW, Winzor DJ, Wolfe CAC, Xuan H. 2006. Handbook of Affinity Chromatography. Ed. David S Hage 2nd ed. Florida: CRC Press Taylor & Francis Group 1-856 p. Hage DS, Matsuda R. 2015. Affinity Chromatography: A Historical Perspective. Methods Mol. Biol. 1286:1–19. Hall AK, Carlson MR. 2014. The current status of orphan drug development in Europe and the US. Intractable Rare Dis. Res. 3:1–7. Hall B, Micheletti JM, Satya P, Ogle K, Pollard J, Ellington AD. 2009. Design, synthesis, and amplification of DNA pools for in vitro selection. Curr. Protoc. Mol. Biol.:24.2.1-24.2.27. Hassan R, Thomas A, Alewine C, Le DT, Jaffee EM, Pastan I. 2016. Mesothelin Immunotherapy for Cancer: Ready for Prime Time? J. Clin. Oncol. 34:4171–4179. Hindson BJ, Ness KD, Masquelier D a., Belgrader P, Heredia NJ, Makarewicz AJ, Bright IJ, Lucero MY, Hiddessen AL, Legler TC, Kitano TK, Hodel MR, Petersen JF, Wyatt PW, Steenblock ER, Shah PH, Bousse LJ, Troup CB, Mellen JC, Wittmann DK, Erndt NG, Cauley TH, Koehler RT, So AP, Dube S, Rose K a., Montesclaros L, Wang S, Stumbo DP, Hodges SP, Romine S, Milanovich FP, White HE, Regan JF, Karlin-Neumann G a., Hindson CM, Saxonov S, Colston BW. 2011. High-throughput Droplet Digital PCR System for Absolute Quantitation of DNA Copy Number. Anal. Chem. 83:8604–8610. Homann M, Göringer HU. 1999. Combinatorial selection of high affinity RNA ligands to live African trypanosomes. Nucleic Acids Res. 27:2006–2014. Hoon S, Zhou B, Janda KD, Brenner S, Scolnick J. 2011. Aptamer selection by high-throughput sequencing and informatic analysis. Biotechniques 51:413–416. Houghten R a, Pinilla C, Blondelle SE, Appel JR, Dooley CT, Cuervo JH. 1991. Generation and use of synthetic peptide combinatorial libraries for basic research and drug discovery. Nature 354:84–86. Huang PY, Carbonell RG. 1999. Affinity Chromatographic Screening of Soluble Combinatorial Peptide Libraries. Biotechnol. Bioeng. 63:633–641.  92  Hughes-Wilson W, Palma A, Schuurman A, Simoens S. 2012. Paying for the Orphan Drug System: break or bend? Is it time for a new evaluation system for payers in Europe to take account of new rare disease treatments? Orphanet J. Rare Dis. 7:1–8. Irvine D, Tuerk C, Gold L. 1991. SELEXION: Systematic Evolution of Ligands by Exponential Enrichment with Integrated Optimization by Non-linear Analysis. J. Mol. Biol. 222:739–761. Jacquemart R, Vandersluis M, Zhao M, Sukhija K, Sidhu N, Stout J. 2016. A Single-use Strategy to Enable Manufacturing of Affordable Biologics. Comput. Struct. Biotechnol. J. 14:309–318. Jarosch F, Buchner K, Klussmann S. 2006. In vitro selection using a dual RNA library that allows primerless selection. Nucleic Acids Res. 34. Jayasena SD. 1999. Aptamers: An Emerging Class of Molecules That Rival Antibodies in Diagnostics. Clin. Chem. 45:1628–1650. Jeddi I, Saiz L. 2017. Three-dimensional modeling of single stranded DNA hairpins for aptamer-based biosensors. Sci. Rep. 7:1–13. Jensen MA, Straus N. 1993. Effect of PCR conditions on the formation of heteroduplex and single-stranded DNA products in the amplification of bacterial ribosomal DNA spacer regions. Genome Res. 3:186–194. Jiang H, Song B, Wang P, Shi B, Li Q, Fan M, Di S, Yang J, Li Z. 2017. Efficient growth suppression in pancreatic cancer PDX model by fully human anti-mesothelin CAR-T cells. Protein Cell 8:926–931. Jiménez E, Sefah K, López-Colón D, van Simaeys D, Chen HW, Tockman MS, Tan W. 2012. Generation of Lung Adenocarcinoma DNA Aptamers for Cancer Studies. PLoS One 7:1–7. Jing M, Bowser MT. 2011. Isolation of DNA aptamers using micro free flow electrophoresis. Lab Chip 11:3703–3709. John Maclennan. 1995. Engineering Microprotein Ligands for Large-Scale Affinity Purification. Nature 13:565–576. Jostock T, Li J. 2010. Expression of IgG Antibodies in Mammalian Cells. Antib. Eng. 1:517–529. Jozala AF, Geraldes DC, Tundisi LL, Feitosa V de A, Breyer CA, Cardoso SL, Mazzola PG, de Oliveira-Nascimento L, Rangel-Yagui C de O, Magalhães P de O, de Oliveira MA, Pessoa A. 2016. Biopharmaceuticals from microorganisms: from production to purification. Brazilian J. Microbiol. 47:51–63. Kai E, Sumikura K, Ikebukuro K, Karube I. 1998. Purification of single stranded DNA from asymmetric PCR product using the SMART system. Biotechnol. Tech. 12:935–939.   93  Keefe AD, Cload ST. 2008. SELEX with modified nucleotides. Curr. Opin. Chem. Biol. 12:448–456. Khan KH. 2013. Gene Expression in Mammalian Cells and its Applications. Adv. Pharm. Bull. 3:257–263. Klußmann S, Nolte A, Bald R, Erdmann VA, Fürste JP. 1996. Mirror-image RNA that binds D-Adenosine. Nat. Biotechnol. 14:1112–1115. Koguma I, Yamashita S, Sato S, Okuyama K, Katakura Y. 2013. Novel purification method of human immunoglobulin by using a thermo-responsive protein A. J. Chromatogr. A 1305:149–153. Kökpinar Ö, Walter JG, Shoham Y, Stahl F, Scheper T. 2011. Aptamer-Based Downstream Processing of His-Tagged Proteins Utilizing Magnetic Beads. Biotechnol. Bioeng. 108:2371–2379. Kummer A, Li-Chan ECY. 1998. Application of an ELISA-elution assay as a screening tool for dissociation of yolk antibody-antigen complexes. J. Immunol. Methods:125–137. Kunert R, Reinhart D. 2016. Advances in recombinant antibody manufacturing. Appl. Microbiol. Biotechnol. 100:3451–3461. Kuwahara M, Obika S. 2013. In vitro selection of BNA (LNA) aptamers. Artif. DNA PNA XNA 4:39–48. Lam KS, Salmon SE, Hersh EM, Hruby VJ, Kazmierski WM, Knapp RJ. 1991. A new type of synthetic peptide library for identifying ligand-binding activity. Nature 354:82–84. Leavitt S, Freire E. 2001. Direct measurement of protein binding energetics by isothermal titration calorimetry. Curr. Opin. Struct. Biol. 11:560–566. Lee G, Yoo J, Leslie BJ, Ha T. 2011. Single-molecule analysis reveals three phases of DNA degradation by an exonuclease. Nat. Chem. Biol. 7:367–374. Li F, Vijayasankaran N, Shen A, Kiss R, Amanullah A. 2010. Cell culture processes for monoclonal antibody production. MAbs 2:466–479. Liang C, Li D, Zhang G, Li H, Shao N, Liang Z, Zhang L, Lu A, Zhang G. 2015. Comparison of the methods for generating single-stranded DNA in SELEX. Analyst 140:3439–3444. Lindsley CW. 2017. New 2016 Data and Statistics for Global Pharmaceutical Products and Projections through 2017. ACS Chem. Neurosci. 8:1635–1636. Lorenz C, von Pelchrzim F, Schroeder R. 2006. Genomic systematic evolution of ligands by exponential enrichment (Genomic SELEX) for the identification of protein-binding RNAs independent of their expression levels. Nat. Protoc. 1:2204–2212. Lorthiois E, Anderson K, Vulpetti A, Rogel O, Cumin F, Ostermann N, Steinbacher S, Mac Sweeney A, Delgado O, Liao SM, Randl S, Rüdisser S, Dussauge S, Fettis K, Kieffer L, De Erkenez A, Yang L, Hartwieg C, Argikar UA, La Bonte LR, Newton R, Kansara V, Flohr S,  94  Hommel U, Jaffee B, Maibaum J. 2017. Discovery of Highly Potent and Selective Small-Molecule Reversible Factor D Inhibitors Demonstrating Alternative Complement Pathway Inhibition in Vivo. J. Med. Chem. 60:5717–5735. Lou X, Qian J, Xiao Y, Viel L, Gerdon AE, Lagally ET, Atzberger P, Tarasow TM, Heeger AJ, Soh HT. 2009. Micromagnetic selection of aptamers in microfluidic channels. Proc. Natl. Acad. Sci. 106:2989–2994. Lowe CR. 2001. Combinatorial approaches to affinity chromatography. Curr. Opin. Chem. Biol. 5:248–256. Lowe CR, Lowe AR, Gupta G. 2001. New developments in affinity chromatography with potential application in the production of biopharmaceuticals. J. Biochem. Biophys. Methods 49:561–574. Lund LN, Gustavsson PE, Michael R, Lindgren J, Nørskov-Lauritsen L, Lund M, Houen G, Staby A, St. Hilaire PM. 2012. Novel peptide ligand with high binding capacity for antibody purification. J. Chromatogr. A 1225:158–167. Magdeldin S, Moser A. 2012. Affinity Chromatography: Principles and Applications. Affin. Chromatogr.:3–29. Margulies M, Egholm M, Altman WE, Attiya S, Bader JS, Bemben LA, Berka J, Braverman MS, Chen YJ, Chen Z, Dewell SB, Du L, Fierro JM, Gomes X V., Godwin BC, He W, Helgesen S, Ho CH, Irzyk GP, Jando SC, Alenquer MLI, Jarvie TP, Jirage KB, Kim JB, Knight JR, Lanza JR, Leamon JH, Lefkowitz SM, Lei M, Li J, Lohman KL, Lu H, Makhijani VB, McDade KE, McKenna MP, Myers EW, Nickerson E, Nobile JR, Plant R, Puc BP, Ronan MT, Roth GT, Sarkis GJ, Simons JF, Simpson JW, Srinivasan M, Tartaro KR, Tomasz A, Vogt KA, Volkmer GA, Wang SH, Wang Y, Weiner MP, Yu P, Begley RF, Rothberg JM. 2005. Genome sequencing in microfabricated high-density picolitre reactors. Nature 437:376–380. Mattheakis LC, Bhatt RR, Dower WJ. 1994. An in vitro polysome display system for identifying ligands from very large peptide libraries. PNAS 91:9022–9026. Mayer G, Ahmed M-SLM, Dolf A, Endl E, Knolle P a, Famulok M. 2010. Fluorescence-activated cell sorting for aptamer SELEX with cell mixtures. Nat. Protoc. 5:1993–2004. Mendonsa SD, Bowser MT. 2004. In vitro evolution of functional DNA using capillary electrophoresis. J. Am. Chem. Soc. 126:20–21. Menegatti S, Naik AD, Carbonell RG. 2013. The hidden potential of small synthetic molecules and peptides as affinity ligands for bioseparations. Pharm. Bioprocess. 1:467–485. Le Merdy S. 2014. Evolving Clarification Strategies to Meet New Challenges. Bioprocess Int. 12:2–4. Mi J, Liu Y, Rabbani ZN, Yang Z, Urban JH, Sullenger BA, Clary BM. 2010. In vivo selection of tumor-targeting RNA motifs. Nat. Chem. Biol. 6:22–24.  95  Michaud M, Jourdan E, Villet A, Ravel A, Grosset C, Peyrin E. 2003. A DNA Aptamer as a New Target-Specific Chiral Selector for HPLC. J. Am. Chem. Soc. 125:8672–8679. Minow B, Egner F, Jonas F, Lagrange B. 2014. High-Cell-Density Clarification By Single-Use Diatomaceous Earth Filtration. Bioprocess Int. 12:2–9. Mould DR, Sweeney KRD. 2007. The pharmacokinetics and pharmacodynamics of monoclonal antibodies - Mechanistic modeling applied to drug development. Curr. Opin. Drug Discov. Dev. 10:84–96. Murphy JC, Jewell DL, White KI, Fox GE, Willson RC. 2003. Nucleic Acid separations utilizing Immobilized Metal Affinity Chromatography. Biotechnol. Prog. 19:982–986. Musheev MU, Krylov SN. 2006. Selection of aptamers by systematic evolution of ligands by exponential enrichment: Addressing the polymerase chain reaction issue. Anal. Chim. Acta 564:91–96. Nakano M, Komatsu J, Matsuura SI, Takashima K, Katsura S, Mizuno A. 2003. Single-molecule PCR using water-in-oil emulsion. J. Biotechnol. 102:117–124. Narayan K, Lemmon MA. 2006. Determining Selectivity of Phosphoinositide-Binding Domains. Methods 39:122–133. Ni S, Yao H, Wang L, Lu J, Jiang F, Lu A, Zhang G. 2017. Chemical Modifications of Nucleic Acid Aptamers for Therapeutic Purposes. Int. J. Mol. Sci. 18:1–21. Nieuwlandt D. 2000. In Vitro Selection of Functional Nucleic Acid Sequences. Curr. Issues Mol. Biol. 2:9–16. Oktem HA, Bayramoglu G, Ozalp VC, Arica MY. 2007. Single-Step Purification of Recombinant Thermus aquaticus DNA polymerase Using DNA-Aptamer Immobilized Novel Affinity Magnetic Beads. Biotechnol. Prog. 23:146–154. Ouellet E, Foley JH, Conway EM, Haynes C. 2015. Hi-Fi SELEX: A High-Fidelity Digital-PCR Based Therapeutic Aptamer Discovery Platform. Biotechnol. Bioeng. 112:1506–1522. Ouellet E, Lagally ET, Cheung KC, Haynes C a. 2014. A Simple Method for Eliminating Fixed-Region Interference of Aptamer Binding During SELEX. Biotechnol. Bioeng. 111:2265–2279. Ouellet E, Lausted C, Lin T, Yang CWT, Hood L, Lagally ET. 2010. Parallel microfluidic surface plasmon resonance imaging arrays. Lab Chip 10:581–588. Ozer A, Pagano JM, Lis JT. 2014. New Technologies Provide Quantum Changes in the Scale, Speed, and Success of SELEX Methods and Aptamer Characterization. Mol. Ther. Acids 3:1–18. Ozer A, White BS, Lis JT, Shalloway D. 2013. Density-dependent cooperative non-specific binding in solid-phase SELEX affinity selection. Nucleic Acids Res. 41:7167–7175.   96  Pabst TM, Palmgren R, Forss A, Vasic J, Fonseca M, Thompson C, Wang WK, Wang X, Hunter AK. 2014. Engineering of novel Staphylococcal Protein A ligands to enable milder elution pH and high dynamic binding capacity. J. Chromatogr. A 1362:180–185. Pagratis NC. 1996. Rapid preparation of single stranded DNA from PCR products by streptavidin induced electrophoretic mobility shift. Nucleic Acids Res. 24:3645–3646. Pan W, Xin P, Clawson GA. 2008. Minimal primer and primer-free SELEX protocols for selection of aptamers from random DNA libraries. Biotechniques 44:351–360. Perret G, Santambien P, Boschetti E. 2015. The quest for affinity chromatography ligands: Are the molecular libraries the right source? J. Sep. Sci. 38:2559–2572. Peyrin E. 2006. Aptamers as Ligands for Affinity Chromatography and Capillary Electrophoresis Applications. Aptamer Handb. Funct. Oligonucleotides Their Appl.:324–342. Poncin-Epaillard F, Vrlinic T, Debarnot D, Mozetic M, Coudreuse A, Legeay G, El Moualij B, Zorzi W. 2012. Surface Treatment of Polymeric Materials Controlling the Adhesion of Biomolecules. J. Funct. Biomater. 3:528–543. Raska M, Czernekova L, Moldoveanu Z, Zachova K, Elliott MC, Novak Z, Hall S, Hoelscher M, Maboko L, Brown R, Smith PD, Mestecky J, Novak J. 2014. Differential glycosylation of envelope gp120 is associated with differential recognition of HIV-1 by virus-specific antibodies and cell infection. AIDS Res. Ther. 11:1–16. Rios M. 2012. A Decade of Bioreactor and Upstream Technologies. Bioprocess Int.:10–13. Robertson DL, Joyce GF. 1990. Selection in vitro of an RNA enzyme that specifically cleaves single-stranded DNA. Nature. Romig TS, Bell C, Drolet DW. 1999. Aptamer affinity chromatography: combinatorial chemistry applied to protein purification. J. Chromatogr. B 731:275–284. Rosenbaum S. 2011. Law and the Public’s Health. Public Health Rep. 126:130–135. Ruigrok VJB, Levisson M, Eppink MHM, Smidt H, van der Oost J. 2011. Alternative affinity tools: more attractive than antibodies? Biochem. J. 436:1–13. Sakamoto S, Hatakeyama M, Ito T, Handa H. 2012. Tools and methodologies capable of isolating and identifying a target molecule for a bioactive compound. Bioorganic Med. Chem. 20:1990–2001. SantaLucia J, Hicks D. 2004. The Thermodynamics of DNA Structural Motifs. Annu. Rev. Biophys. Biomol. Struct. 33:415–440. Schaffitzel C, Hanes J, Jermutus L, Plückthun A. 1999. Ribosome display: An in vitro method for selection and evolution of antibodies from libraries. J. Immunol. Methods 231:119–135. Schlapak R, Pammer P, Armitage D, Zhu K, Hinterdorfer P, Vaupel M, Frühwirth T, Howorka S. 2006. Glass Surfaces Grafted with High-Density Poly(ethylene glycol) as Substrates for DNA Oligonucleotide Microarrays. Langmuir 22:277–285.  97  Schütze T, Wilhelm B, Greiner N, Braun H, Peter F, Mörl M, Erdmann VA, Lehrach H, Konthur Z, Menger M, Arndt PF, Glökler J. 2011. Probing the SELEX Process with Next-Generation Sequencing. PLoS One 6:1–10. ScienceLab.com. 2013. Material Safety Data Sheet Phenol MSDS. MSDS 1-6 p. Scientific T. 2017. Poor Efficiency of PCR. ThermoFisher Sci. https://www.thermofisher.com/ca/en/home/life-science/pcr/real-time-pcr/qpcr-education/real-time-pcr-troubleshooting-tool/gene-expression-quantitation-troubleshooting/poor-pcr-efficiency.html. Scott C. 2011. Toward Nonantibody Platforms. Bioprocess Int. 10:31–42. Sefah K, Shangguan D, Xiong X, O’Donoghue MB, Tan W. 2010. Development of DNA aptamers using Cell-SELEX. Nat. Protoc. 5:1169–1185. Sharma A, Jacob A, Tandon M, Kumar D. 2010. Orphan drug: Development trends and strategies. J. Pharm. Bioallied Sci. 2:290–299. Skerra A. 2007. Alternative non-antibody scaffolds for molecular recognition. Curr. Opin. Biotechnol. 18:295–304. Smith GP. 1985. Filamentous Fusion Phage: Novel Expression Vectors that Display Cloned Antigens on the Virion Surface. Science (80-. ). 228:1315–1317. Sousa IT, Taipa MA. 2014. Biomimetic Affinity Ligands for Protein Purification. Methods Mol. Biol. 1129:231–262. Sproule K, Morrill P, Pearson JC, Burton SJ, Hejnæs KR, Valore H, Ludvigsen S, Lowe CR. 2000. New strategy for the design of ligands for the purification of pharmaceutical proteins by affinity chromatography. J. Chromatogr. B Biomed. Sci. Appl. 740:17–33. Stoltenburg R, Reinemann C, Strehlitz B. 2007. SELEX-A (r)evolutionary method to generate high-affinity nucleic acid ligands. Biomol. Eng. 24:381–403. Strauss WM. 1998. Preparation of Genomic DNA from Mammalian Tissue. Curr. Protoc. Mol. Biol.:2.2.1-2.2.3. Svobodová M, Pinto A, Nadal P, O’ Sullivan CK. 2012. Comparison of different methods for generation of single-stranded DNA for SELEX processes. Anal. Bioanal. Chem. 404:835–842. Tan SC, Yiap BC. 2009. DNA, RNA, and protein Extraction: The Past and The Present. J. Biomed. Biotechnol. 2009:1–10. Tarhini AA, Gogas H, Kirkwood JM. 2015. HHS Public Access 510:84–91. Tasset DM, Kubik MF, Steiner W. 1997. Oligonucleotide Inhibitors of Human Thrombin that Bind Distinct Epitopes. J. Mol. Biol. 272:688–698.   98  Teng SF, Sproule K, Hussain A, Lowe CR. 1999. A strategy for the generation of biomimetic ligands for affinity for affinity chromatography. Combinatorial synthesis and biological evaluation of an IgG binding ligand. J. Mol. Recognit. 12:67–75. Thiel K. 2004. Oligo oligarchy—the surprisingly small world of aptamers. Nat. Biotechnol. 22:649–651. Tsukamoto M, Watanabe H, Ooishi A, Honda S. 2014. Engineered protein A ligands, derived from a histidine-scanning library, facilitate the affinity purification of IgG under mild acidic conditions. J. Biol. Eng. 8:1–9. Tuerk C, Gold L. 1990. Systematic evolution of ligands by exponential enrichment: RNA ligands to bacteriophage T4 DNA polymerase. Science (80-. ). 249:505–510. Turnbull AP, Emsley P. 2013. Studying Protein-Ligand Interactions Using X-Ray Crystallography. Methods Mol. Biol. 1008:457–477. Vant-Hull B, Payano-Baez A, Davis RH, Gold L. 1998. The Mathematics of SELEX Against Complex Targets. J. Mol. Biol. 278:579–597. Vater A, Klussmann S. 2015. Turning mirror-image oligonucleotides into drugs: the evolution of Spiegelmer therapeutics. Drug Discov. Today 20:147–155. Verma S, Eckstein F. 1998. MODIFIED OLIGONUCLEOTIDES: Synthesis and Strategy for Users. Annu. Rev. Biochem. 67:99–134. Vilar S, Costanzi S. 2012. Predicting Biological Activities through QSAR Analysis and Docking-based Scoring. Methods Mol. Biol. 914:271–284. Wang G, Salm JR, Gurgel P V., Carbonell RG. 2005. Small Peptide Ligands for Affinity Separations of Biological Molecules. Chem. Eng. Trends Dev. 4:63–83. Weng C-H, Huang C-J, Lee G-B. 2012. Screening of Aptamers on Microfluidic Systems for Clinical Applications. Sensors 12:9514–9529. Williams R, Peisajovich SG, Miller OJ, Magdassi S, Tawfik DS, Griffiths AD. 2006. Amplification of complex gene libraries by emulsion PCR. Nat. Methods 3:545–550. Williams S, Morton P, Baines D. 2014. Synthetic Ligand Affinity Chromatography Purification of Human Serum Albumin and Related Fusion Proteins. Methods Mol. Biol. 1129:181–195. Wilson K. 1997. Preparation of Genomic DNA from Bacteria. Curr. Protoc. Mol. Biol.:2.4.1-2.4.5. Xing PX, Russell S, Prenzoska J, McKenzie I. 1994. Discrimination between alternatively spliced STP-A and -B isoforms of CD46. Immunology 83:122–127. Yang H, Gurgel P V., Carbonell RG. 2009. Purification of human immunoglobulin G via Fc-specific small peptide ligand affinity chromatography. J. Chromatogr. A 1216:910–918.   99  Yuk IH, Nishihara J, Walker D, Huang E, Gunawan F, Subramanian J, Pynn AFJ, Yu XC, Zhu-Shimoni J, Vanderlaan M, Krawitz DC. 2015. More Similar Than Different: Host Cell Protein Production Using Three Null CHO Cell Lines. Biotechnol. Bioeng. 112:2068–2083. Zhuo Z, Yu Y, Wang M, Li J, Zhang Z, Liu J, Wu X, Lu A, Zhang G, Zhang B. 2017. Recent advances in SELEX technology and aptamer applications in biomedicine. Int. J. Mol. Sci. 18:1–19. Zimmermann B, Bilusic I, Lorenz C, Schroeder R. 2010. Genomic SELEX: A discovery tool for genomic aptamers. Methods 52:125–132.     100  Appendix A New Algorithm for Defining Multicomponent Binding of DNA Aptamers During their Selection by Hi-Fi SELEXc A.1 Overview In Hi-Fi SELEX technology, the equilibrium dissociation constant Kd for a single aptamer to its aptigen (i.e. its protein target) is generally characterized by fitting the Langmuir binding relation (equation 2.5) to a measure binding isotherm. However, the starting material for Hi-Fi SELEX is a collection of 100 trillion different ssDNA sequences, making it difficult to evaluate the binding characteristics of a single member within this population during the selection process. What is typically therefore done is to measure the binding isotherm for the retained pool after each selection round, and then fit equation 2.5 to that data to regress an apparent mean Kd, 𝐾𝑑𝑎𝑣𝑔, for the pool. This screening process is repeated until no appreciable increase in 𝐾𝑑𝑎𝑣𝑔 can be detected. The final enriched pool is then sequenced, with individual members produced to create a bottom-up high-throughput approach to affinity ligand discovery. While useful, measurement of 𝐾𝑑𝑎𝑣𝑔 does not provide information on the distribution of Kdi values characterizing the binding affinity of each aptamer i in the retained pool after each round of selection. This hampers selection efficiency, as what is sought is to grow and ultimately isolate that pool of candidate aptamers having a Kd of 10-9 M or better for the target.  If such information could be obtained, it would allow one to construct histograms of the Kdi distribution after each selection round, as illustrated in Figure A.1, so as to evaluate that fraction of the total retained population after each selection round that exhibits a high affinity for the target. Knowledge of this high-affinity fraction provides valuable information regarding the quality of the selection process, as well as the ability to monitor the response of Hi-Fi SELEX to changes in screening conditions. To enable the binning of retained library members after each round of selection in terms of their Kd, a method was first developed to fit the binding isotherm of a retained aptamer pool measured using the qPCR determination method described in Section 2.3.7 of this thesis. This                                                           c The method reported in this Appendix was developed in collaboration with Mr. Nathan Chan, an undergraduate student who worked in the Haynes Lab under my supervision.  101  new method, which allows one to estimate the percentage of high affinity binders after each selection round, was coded in MATLAB and operates on the assumption that the population of retained members may be partitioned into a set of “pseudo-components”, with each pseudo-component characterized by a unique Kd. The algorithm developed and described here minimizes the chi-squared error (2) of the data-fit of the model, which is effectively the multicomponent Langmuir Isotherm, using the Levenberg-Marquardt method. From this information, the population of high-affinity binders after each round of selection is estimated and characterized.   Figure A.1:  Illustrative Kai histograms (Ka = 1/Kd) for a three pseudo-component population (low-affinity, moderate affinity, high affinity) as a function of Hi-Fi SELEX selection round number.  The desire is to be able to analyze the fraction of members with a binding affinity Ka of ~ 109 M-1, shown in red, and the increase in the fractional population of this pool after each round of selection. That information would provide a more direct understanding of round-by-round progress towards isolating a small pool of candidate high-affinity ligands, and information from one round could potentially be used to set the conditions used for the next round to ensure the process continues to move toward isolation of the desired high-affinity pool. With a computational tool for determining such distributions, the quality of the selection process can be assessed and the response to any changes in the  102  screening conditions can be monitored to adapt and improve the high-throughput screening of high-affinity binding aptamers.   A.2 Model Development and Underlying Theory A.2.1 The Binding Partition Function and Competitive Binding Model Derivation   The binding of proteins to ligands is often described in terms of binding reactions.  Those reactions generally define the relation between the amount of ligand X bound to the macromolecule M at various activities/concentrations of the ligand, expressed in the form of a binding curve.  The interpretation of binding curves in terms of the underlying mass balances and equilibrium relations may then be used to obtain a quantitative analysis of the binding processes of interest.   The type and order of the binding process dictate the number and form of equilibrium relations required.  To fix ideas, I first derive those relations for the simplest binding process – binding of a single ligand X to a macromolecule M offering a single binding site per molecule. A.2.1.1 Binding of Ligand X to a Single Site: For binding of a single ligand molecule X to a single site on macromolecule M, the reaction equation may be written as      Ka1  𝑀 +   𝑋 ↔  𝑀𝑋        (A.1)  The law of mass action gives the equilibrium association constant Ka1 for this reaction in terms of the species activities aX, aM and aMX:   𝐾𝑎1 =𝑎𝑀𝑋𝑎𝑀𝑎𝑋         (A.2) We can then assume that the ratio of activities of M and MX is determined by the ratio of their concentrations, [MX]/[M], so that the equilibrium relation can be rewritten as   𝐾𝑎1 =[𝑀𝑋][𝑀]𝑥         (A.3) where for brevity ax has been replaced with x.    103  The fractional degree of binding saturation, 𝜃, is defined as the moles of X bound per mole of macromolecule, and is derived from the total mass balance for the macromolecule:  𝜃 =[𝑀𝑋1]𝑀𝑡𝑜𝑡=[𝑀𝑋][𝑀] + [𝑀𝑋]        (A.4) Substitution of [MX] in equation A.4 with equation A.3 gives  𝜃 =𝐾𝑎1𝑥1+ 𝐾𝑎1𝑥≅𝐾𝑎1[𝑋]1+ 𝐾𝑎1[𝑋]       (A.5) Equation A.5 is the well-known Langmuir equilibrium-binding model for a single independent binding site and single ligand species.  The form of equation A.5 is that of the positive branch of a rectangular hyperbola centered at the origin.   Equation A.5 can be rearranged into a number of forms useful for data analysis.  One such form is  𝜃1− 𝜃= 𝐾𝑎1[𝑋]         (A.6) where the ratio of occupied to unoccupied binding sites is given by the left-hand side and is shown to depend linearly on [X], the free ligand concentration at equilibrium.   Taking the logarithm of equation A.6 yields the Hill equation  𝑙𝑜𝑔 (𝜃1− 𝜃) = 𝑙𝑜𝑔(𝐾𝑎1)  +   𝑙𝑜𝑔([𝑋])      (A.7) This relation states that any binding reaction where ligand L binds to a single site on macromolecule M should yield a slope of unity and an intercept of log (Ka1) when plotted according to equation 7.  That plot is known as a Hill plot, and the observance of curvature in binding data plotted in that form indicates the binding reaction is complex and cannot be described by the simple Langmuir model embodied in equation A.5. A fundamental and useful representation of the binding curve is obtained by a plot of 𝜃 versus log ([X]) because the logarithm of the ligand concentration is proportional to the chemical potential of the ligand.  Presentation of binding data in this form is sometimes called the “titration binding curve” since the data are typically obtained by a titration experiment in which fraction of binding sites occupied is measured as ligand is titrated into the system to increasing concentrations.  The shape of the titration binding curve is exceptionally sensitive to the nature of the underlying binding reaction(s).  104   A.2.1.2 The Binding Partition Function: While the simple binding reaction described by equation A.5 may be derived by combining a mass balance on the macromolecule with a single reaction equilibrium constraint, more complex binding reaction are best described by means of the binding partition function Q or, equivalently, the binding polynomial P.  Q represents the sum of the concentrations of all possible macromolecular species found in the system relative to that of a reference species, typically given by the equilibrium concentration of macromolecule having no ligands bound to it [M].  For a macromolecule having n binding sites per molecule,  𝑄 =[𝑀] + [𝑀𝑋1] + [𝑀𝑋2] + [𝑀𝑋3] +⋯ [𝑀]      (A.8) As shown in section A.2.1.1, equilibrium binding of the first ligand to M is defined by equation A.3. Binding of the second ligand is then described by the following reaction equation        Ka2  𝑀𝑋 +   𝑋 ↔  𝑀𝑋2        (A.9)  and so forth.  The partition function may therefore be rewritten as  𝑄 = 1 + 𝐾𝑎1[𝑋] + 𝐾𝑎1𝐾𝑎2[𝑋]2 + ⋯ + ∏ 𝐾𝑎𝑖𝑛𝑖=1 [𝑋]𝑛    (A.10) For binding of a single type of ligand X to a macromolecule M offering a single binding site per molecule, the partition function reduces to  𝑄 = 1 + 𝐾𝑎1[𝑋]        (A.11) and for macromolecule offering two different binding sites per molecule  𝑃 ≡ 𝑄 = 1 + 𝐾𝑎1[𝑋] + 𝐾𝑎1𝐾𝑎2[𝑋]2       (A.12) These equations for Q are polynomials in [X].  When the partition function is expressed in this form, it is called the “Binding Polynomial” and Q is replaced by the symbol P to reflect this fact.  In the binding polynomial, the concentrations of the various species present relative to that of the fully unoccupied macromolecule are given by the successive terms on the right-hand side of the equality. The binding equilibrium relation (binding isotherm equation) is obtained from the derivative of the binding polynomial.  Specifically, the fractional occupancy 𝜃 is found by taking the derivative of ln P with respect to ln [X].  Rigorously, we are actually taking the partial  105  derivative since the various Kai will depend on other variables, including temperature and pressure.   But those other variables are fixed in the system we will study.  Then,   𝜃 =𝑑(ln 𝑃)𝑑(ln[𝑋])=[𝑋]𝑃𝑑𝑃𝑑[𝑋]        (A.13) Application of equation A.13 to the single site reaction described by the binding polynomial given in equation A.11 therefore yields  𝜃 =𝑑(ln(1+𝐾𝑎1[𝑋]))𝑑(ln[𝑋])=[𝑋]1+𝐾𝑎1[𝑋]𝑑(1+𝐾𝑎1[𝑋])𝑑[𝑋]=𝐾𝑎1[𝑋]1+ 𝐾𝑎1[𝑋]    (A.14) Thus, we recover the same result as given in equation A.5. Importantly for the model fitting work we hope to achieve, the reverse operation, namely the integration of 𝜃 with respect to ln [X] must give ln P.  This operation is graphically represented by the area under the titration binding curve since integration of equation A.13 from [X] = 0 to [X] gives ln P as follows:  ∫ 𝜃 𝑑(𝑙𝑛[𝑋]) = ∫ 𝑑(ln 𝑃)ln 𝑃0= ln 𝑃 = 𝐴ln [𝑋]−∞    (A.15) where A is the area under the titration curve bounded by the chosen limits of integration.  This exercise therefore shows that  𝑃 = 𝑒𝐴         (A.16) We now must define the form of the binding polynomial that best describes binding in our system of interest – that of multiple different ligands Xi competing for a single binding site presented on each macromolecule M.  A.2.1.3 Competitive Binding of Different Ligands to a Single Site: In the screening of a DNA aptamer library using Hi-Fi SELEX, up to 1014 different aptamer sequences are incubated and equilibrated with a target protein of similar size to each 80-nt aptamer library member. To a good first approximation, we can therefore assume that binding of one library member to the target protein will prevent binding of another due to the steric repulsion created.  Though each is in principle capable of binding to the macromolecule with a unique Kai, defining P for that situation is prohibitively complex as n in equation 10 is then of order ~ 1014.   A more granular description is therefore required.  To fix ideas, let us for the moment assume the binding characteristics of the library of aptamers can be described using two unique binding populations – a pseudo- 106  population 1 where each member binds the target protein with high affinity Ka1, and a pseudo-population 2 where each member binds the target with lower affinity Ka2.  Binding is competitive because the target offers a single site that binds either X1 or X2.  The binding reactions for this pseudo-population model are then given by  𝑀 +  𝑋1  ↔   𝑀𝑋1                            𝐾𝑎1 =[𝑀𝑋1][𝑀][𝑋1]      (A.17a)  𝑀 +  𝑋2  ↔   𝑀𝑋2                            𝐾𝑎1 =[𝑀𝑋2][𝑀][𝑋2]      (A.17b) In this reaction system, there is therefore a simple partition of populations 1 and 2 among available binding sites.  The binding polynomial is then given by  𝑃 = 1 + 𝐾𝑎1[𝑋1] + 𝐾𝑎2[𝑋2]         (A.18) where the successive terms on the right-hand side of the equality represent the relative concentrations of the three possible pseudo-species.  Generalizing equation A.13, we find  𝜃1 =𝑑(ln 𝑃)𝑑(ln[𝑋1])=[𝑋1]𝑃𝑑𝑃𝑑[𝑋1]=𝐾𝑎1[𝑋1]1+ 𝐾𝑎1[𝑋1] + 𝐾𝑎2[𝑋2]      (A.19a)  𝜃2 =𝑑(ln 𝑃)𝑑(ln[𝑋2])=[𝑋2]𝑃𝑑𝑃𝑑[𝑋2]=𝐾𝑎2[𝑋2]1+ 𝐾𝑎1[𝑋1] + 𝐾𝑎2[𝑋2]      (A.19b) The logarithm of P is the sum of areas under the 𝜃1 versus ln [X1] and 𝜃2 versus ln [X2] titration binding curves  𝑑(ln 𝑃) = (𝑑(ln 𝑃)𝑑(ln[𝑋1]))[𝑋2]𝑑(𝑙𝑛[𝑋1])  + (𝑑(ln 𝑃)𝑑(ln[𝑋2]))[𝑋1]𝑑(𝑙𝑛[𝑋2])    (A.20a)  𝑑(ln 𝑃) = 𝜃1 𝑑(𝑙𝑛[𝑋1])  +  𝜃2 𝑑(𝑙𝑛[𝑋2])      (A.20b) We can integrate equation A.20b from a reference state of no bound 1 or 2, where P = 1, to a state where the concentrations of bound 1 and 2 are [X1] and [X2], respectively  ln 𝑃 = ∫ 𝜃1 𝑑(𝑙𝑛[𝑋1])[𝑋1]0 + ∫ 𝜃2 𝑑(𝑙𝑛[𝑋2])[𝑋2]0     (A.21) Equation A.21 shows that the sum of the areas beneath the two titration binding curves defines the value of the partition function.   The effect of the presence of one ligand on the binding properties of another is known as a “linkage effect”.  For the two pseudo-component system considered here, that linkage effect can be assessed from the heterotrophic derivative properties of P. Specifically, the heterotrophic second derivatives of ln P with respect to ln [X1] and ln [X2] are equal since the order of differentiation is immaterial if ln P is well behaved and continuous.  That is  107   𝑑2(ln 𝑃)𝑑(ln[𝑋2])𝑑(ln[𝑋1])= −𝐾𝑎1[𝑋1]𝐾𝑎2[𝑋2](1+ 𝐾𝑎1[𝑋1] + 𝐾𝑎2[𝑋2])2 =  𝑑2(ln 𝑃)𝑑(ln[𝑋1])𝑑(ln[𝑋2]) = − 𝜃1𝜃2  (A.22) So the two linkage derivatives are equal, as they must be, and in this competitive binding system we see the derivative is negative since the product 𝜃1 𝜃2 is always a positive quantity.  This is termed negative heterotrophic cooperativity – the linkage derivative is everywhere negative with maximum values observed at [X1] = 1/Ka1 and [X2] = 1/Ka2. Though derived here for two pseudo-components, it should be clear that the model and method used may be easily generalized for n pseudo-components.  A.2.1.4 Application of Multicomponent Binding Polynomial Theory to Experimental Hi-Fi SELEX Isotherm Data: At the completion of each selection round of the Hi-Fi SELEX process, one expects to isolate from the original aptamer library pool a subset of those sequences having binding affinity for the target protein.  In principal, each unique sequence i within that retained subset will have a unique binding affinity Kai for the target.  Ideally, we would like to derive and fit the binding polynomial for that system to binding titration curve data to regress the complete set of Kai.  But that is not possible due both to the uncertainties associated with regressing that many parameters (e.g. ~ 108 different Kai values) to a limit data set presenting relatively simple features (a rectangular hyperbola) and to the fact that we have no direct experimental knowledge of the values of the large set of [Xi] as a function of the total concentrations of target and aptamer library present in the system. Rather, we are only able to measure the average titration binding curve that plots the total fraction of occupied sites 𝜃i,tot (= ∑ 𝜃𝑖𝑖 ) as a function of the concentration of total free ligand (aptamer) [Xi,tot] (= ∑ [𝑋𝑖]𝑖 ):  108   Figure A.2:  Representation of an average titration binding curve for a retained population after a round of Hi-Fi SELEX  Thus, the type of data that can be collected is limited and does not provide knowledge of the large set of individual [Xi] and 𝜃i values at equilibrium. However, following the concepts introduced in section A.2.1.3, we can partition the pool of library members retained after each selection round into a selected or optimal number of pseudo-components where each pseudo-component is distinguished by its Kai value. To illustrate, let’s first arbitrarily set that number at 3 and then partition retained library members into pseudo-ligand 1 (characterized by a fixed Ka1 = 109 M-1), 2 (Ka2 = 107 M-1), and 3 (Ka3 = 105 M-1).  The resulting binding polynomial for this pseudo-system is then given by:  𝑃 = 1 + 𝐾𝑎1[𝑋1] + 𝐾𝑎2[𝑋2] + 𝐾𝑎3[𝑋3]       (A.23) where the successive terms on the right-hand side of the equality again represent the relative concentrations of the possible species. As the set of Kai values has been fixed, we would like to fit equation A.23 to an average titration curve, as shown in Figure A.2, to regress the most likely values for [X1] to [X3], and from that construct the model-generated histogram (Figure A.3) predicting the relative abundance of each pseudo-ligand i within the retained pool.  00.10.20.30.40.50.60.70.80.910 5 10 15 20 25 30Bound Fraction, θi,totFree Concentration [Xi,tot] (nM) 109   Figure A.3:  Histogram representation of retained ligands binned into three pseudo-components (1, 2 and 3)  A.2.1.5 Setting the Pseudo-Component Granularity: For the system setup in section A.2.1.4, the accuracy of the developed algorithm is dependent on the granularity or the number of pseudo-components that are encompassed in the binding polynomial. As mentioned previously, regressing multiple parameters exhibits larger uncertainties so the optimization of the ideal number of pseudo-components must be investigated. In equation A.23, a three pseudo-component is demonstrated and a resulting Langmuir Binding Isotherm would be represented as shown in equation A.24.  𝜃𝑖,𝑡𝑜𝑡 =𝐾𝑎1[𝑋1]+𝐾𝑎2[𝑋2]+𝐾𝑎3[𝑋3]1+𝐾𝑎1[𝑋1]+𝐾𝑎2[𝑋2]+𝐾𝑎3[𝑋3]      (A.24) For the binding affinity of each component characterized by Ka1 = 109 M-1, Ka2 = 107 M-1, and Ka3 = 105 M-1, a thought experiment can be performed to determine contributing components in the total binding isotherm. Through observation of the binding isotherm, each i component is attributed by the product of both its binding affinity and its free concentration at equilibrium. At initial rounds of selection, the concentration of the high-affinity binding aptamers is small relative to the massive pool in the starting library. For the situation in which only one percent of the total population is characterized by the tight binders at a Ka value of 109 M-1, and the  110  majority of the other aptamers bind with a lower affinity of 105 M-1, Table A.1 summarizes this data using a three pseudo-component system.   Table A.1: Three pseudo-component granularity example i Kai  (M-1) [Xi] (M) Kai [Xi] 1 109 0.01 107 2 107 0 - 3 105 0.99 105   The last column of Table A.1 reports the product of the binding affinity Kai and the free aptamer concentration [Xi] for the first and third pseudo-components. Despite only representing only 1% of the total retained pool, the first component (the tight binders) contributes 99% of the slope and shape of the binding isotherm.  That rare pseudo-component therefore dominates the binding system due to its ability to bind the target protein with high affinity. From this observation, segregation of the large population of weaker binders into a set of pseudo-components of differing Kai values is not necessary for our analysis. In the case just described, the analysis method used tends to naturally converge on a two-pseudo-component system, one having a weak Ka and the other a tight Ka. Assuming a larger number of pseudo-components intrinsically carries more errors and uncertainty in the parameter estimation. As a result, it would limit the scope of systems (selection rounds) for which the algorithm is applicable. Therefore, we henceforth take a granular approach in which the retained population is modeled in terms of two pseudo-components and the Ka value and fraction of each is regressed.    111   A.3 Experimental Methods A.3.1 Binding Isotherm Measurement Both the total bound fraction and the total free concentration of the retained library is required to construct the binding isotherm. To select for the desired population of library members (aptamers) that bind to a target macromolecule with high affinity, a known concentration of the macromolecule (protein target), Mtot, is immobilized onto a surface in which the retined library, Xtot, can be incubated with the macromolecule. For the work reported in this Appendix, the target macromolecule used is human alpha thrombin. The system consisting of the library and the immobilized thrombin are allowed to reach equilibrium and at equilibrium, the concentration can be quantified in one of two ways – quantifying the bound faction or the free concentration of library members in solution. Based on a total mass balance, either quantification method is sufficient to generate the binding isotherm.  To recover the bound population, the free (unbound) members in solution are first washed away. The bound population is then recovered by denaturation of the protein via a strong alkaline solution, neutralization with a strong acidic solution, and stabilization with a buffering solution. The bound fraction or 𝜃i,tot  is then quantified using a qPCR determination technique discussed in the next section. For a given Xtot and Mtot, the basic process is shown in Figure A.4.    Figure A.4: Schematic of steps in the experimental determination of a binding isotherm   112  The aptamer TBA29 is known to have a strong binding affinity to thrombin (Tasset et al., 1997). The screening of TBA29, in addition to the Hi-Fi SELEX library, was therefore used as a control to ensure that the algorithm was functioning properly.  In each selection round of Hi-Fi SELEX, approximately 0.01% of the library screened is retained. With a starting library of initially 1014 unique ssDNA sequences, the first round of selection therefore yields approximately 108 retained sequences.  That population is then amplified by ddPCR and one strand of the resulting duplex DNA removed by -exonuclease processing to create a new ssDNA “retained” library for which we wish to analyze the binding properties.  The method to measure the bound fraction, 𝜃i,tot, when a small aliquot of the retained population is equilibrated with immobilized target utilizes a qPCR-based determination. Following equilibration, the amount of bound material is generally extremely low, and qPCR provides a non-direct method for determining the concentration by first amplifying the bound material to detectable levels. The methodology behind qPCR is based on traditional PCR techniques in which double stranded DNA is first denatured by an increase in temperature (usually to 95°C) to form two single stranded DNA components. With the addition of primers that bind to the 20-nucleotide fixed regions of each library member, as well as the addition of Taq polymerase and free nucleotides, two copies of each DNA sequence present can be made after one amplification cycle. This process is repeated with the concentration of the amplified DNA growing at an exponential rate cycle-by-cycle. The quantification portion of qPCR is provided by the addition to the reaction of a fluorophore that provides signal only when bound to double stranded DNA (the amplification product). Fluorescence is measured and quantified after each amplification cycle. The fluorescence intensity is proportional to the double stranded DNA concentration and a plot of the relative florescence can be plotted verses the cycle number of amplification. Initially, the concentration of recovered bound material is too low to record a fluorescence intensity reading, but as the amplification process proceeds sufficient double stranded material is produced to generate a measurable fluorescence. The amplification cycle in which the fluorescence intensity can first be reliably quantified is called the quantification or quantitation cycle, Cq.  113  From a Cq value for the bound material, the concentration of that material can be determined using a calibration curve constructed using known concentrations of library material. Those known concentrations are amplified and then plotted verses their associated Cq values to create a calibration curve from which concentration of the bound material can be determined from its corresponding Cq. Since the total amount of library material Xtot equilibrated with the target is known, the amount of bound aptamer 𝜃i,tot for a given macromolecule concentration Mtot can be used in combination with a total mass balance to determine the total free (unbound) library concentration [Xi,tot]. This process is repeated for a series of different Xtot to generate the binding isotherm.  A.3.2 Fitting of Model to Experimental Binding Isotherm Data The binding polynomial function, P, for the competitive multicomponent binding of library members to immobilized target is given by equation A.25, where [Xi] is the free concentration of library member i and Kai is the binding affinity of library member i. 𝑃 = 1 + ∑ 𝐾𝑎𝑖[𝑋𝑖]𝑖           (A.25) The general form of the binding isotherm for any library member i can then be derived by taking the derivative of ln P as shown in equation A.26. 𝜃𝑖 =𝑑(ln 𝑃)𝑑(ln[𝑋𝑖])=𝐾𝑎𝑖[𝑋𝑖]1+ ∑ 𝐾𝑎𝑗[𝑋𝑗]𝑗           (A.26) In general, each retained library member has a unique Kai. Experimentally, 𝜃i cannot be measured as a function of Xi for a given member within a retained library. Rather, the measured isotherm plots the total amount of all bound members 𝜃i,tot against the total free concentration [Xi,tot]. 𝜃i,tot is a summation of each the bound amounts of the individual components and, likewise, the total free concentration is a summation of the free concentrations of the components, as shown in equations A.27 and A.28. 𝜃𝑖,𝑡𝑜𝑡 = ∑ 𝜃𝑗 =∑ 𝐾𝑎𝑗[𝑋𝑗]𝑗1+∑ 𝐾𝑎𝑗[𝑋𝑗]𝑗𝑗         (A.27) [𝑋𝑖,𝑡𝑜𝑡] = ∑ [𝑋𝑗]𝑗          (A.28)  114  A mass balance on the total mass of retained library used Xtot can be expressed as a sum of the total free and bound concentrations at equilibrium, with the total macromolecule concentration Mtot used to normalize concentration units. 𝑋𝑡𝑜𝑡 = [𝑋𝑖,𝑡𝑜𝑡] + 𝑀𝑡𝑜𝑡𝜃𝑖,𝑡𝑜𝑡        (A.29) Equation A.29 can be rewritten and combined with equations A.27 and A.28 𝑋𝑡𝑜𝑡 = ∑ [𝑋𝑖]𝑖 + 𝑀𝑡𝑜𝑡 (∑ 𝐾𝑎𝑖[𝑋𝑖]𝑖1+∑ 𝐾𝑎𝑖[𝑋𝑖]𝑖)        (A.30) with i indexing the number of pseudo-components.   With i = 2, equations A.27 and A.30 may be fit to binding isotherm data to estimate the Kai and fractional representation (relative to all bound members) of each pseudo-component i.  For a two pseudo-component system, the equilibrium condition (equation A.27) may be combined with the mass balance (A.3 to give equations A.31 and A.32.  𝑋1𝑡𝑜𝑡 = [𝑋1] + 𝑀𝑡𝑜𝑡𝜃1 = [𝑋1] + 𝑀𝑡𝑜𝑡 (𝐾𝑎1[𝑋1]1+𝐾𝑎1[𝑋1]+𝐾𝑎2[𝑋2])    (A.31) 𝑋2𝑡𝑜𝑡 = [𝑋2] + 𝑀𝑡𝑜𝑡𝜃2 = [𝑋2] + 𝑀𝑡𝑜𝑡 (𝐾𝑎2[𝑋2]1+𝐾𝑎1[𝑋1]+𝐾𝑎2[𝑋2])    (A.32) These equations provide a way in which to predict the free concentration of each pseudo-component given the total amount of library in the system.  In a two-component system, the concentration of the first pseudo-component may be expressed as a fraction of the total concentration of members in the system. To achieve this, a parameter a1 is introduced and denoted as the ratio of X1 to Xtot, as shown in equation A.33, such that the second component concentration X2 may then be determined from a mass balance of Xtot and X1. Since multiple different concentrations are required to generate the binding isotherm, optimizing and regressing to solve a single parameter that remains consistent simplifies the task.   𝑎1 =𝑋1𝑋𝑡𝑜𝑡         (A.33) The parameter a1 therefore provides a way to relate Xtot to the free concentration of each component (X1 and X2), and this information may be used to then compute [X1] and [X2]. From those values, 𝜃𝑖,𝑡𝑜𝑡 can be predicted using equation A.27 and compared with the experimental 𝜃𝑖,𝑡𝑜𝑡.  115   A.3.3 Regression Algorithm:  The model fitting and parameter estimation process is summarized as follows: 1. Guess a1 and use to obtain estimates of the relative amounts of X1 and X2 for a chosen set of Ka1 and Ka2 values 2. Use numerical root finding (Newton Raphson method; built-in MATLAB fzero function) to then calculate [X1] and [X2] 3. Use estimated [X1] and [X2] to calculate [Xi,tot], 𝜃1 , 𝜃2 , and 𝜃𝑖,𝑡𝑜𝑡 • The function that relates a1 to 𝜃𝑖,𝑡𝑜𝑡 is our objective function 4. Minimize chi-squared (χ2) between the experimental data and the model-derived data • Use χ2 to update the a1 parameter and repeat until system converges To solve for the concentration of each pseudo population, a non-linear regression method is required that is able to accurately predict the binding affinity distribution after each round of Hi-Fi SELEX. Here, the Levenberg-Marquardt method for non-linear least squares curve fitting is used. The Levenberg-Marquardt method is a minimization technique that minimizes the chi-squared error (χ2), which is a goodness-of-fit measurement. This measurement relates the extent to which experimental values, 𝜃𝑖,𝑡𝑜𝑡, are able to be fitted with model-derived values, 𝜃𝑖,𝑡𝑜𝑡, that are themselves a nonlinear function of the model parameter, a1. 𝜒2 = ∑[𝜃𝑖,𝑡𝑜𝑡−?̂?𝑖,𝑡𝑜𝑡(𝑎1)]2𝜎𝑖2𝑛𝑖=1         (A.34) Using the Levenberg-Marquardt method, the parameter a1 is therefore adjusted to determine the fraction of all retained library members having a high-affinity for the target. That fraction is given by minimizing the residuals of the predicted and experimental values. The Levenberg-Marquardt method uses a dampening factor λ that applies a combination of the Gradient Descent method and the Gauss-Newton method depending on how close the method is to minimizing χ2. At the initiation of the regression, the dampening factor is set high to apply more of the gradient descent method in determining the parameter difference change. As more iterations are performed and the regression proceeds, the dampening factor value decreases so that the minimization is driven by the Gauss Newton method as a minimum is approached.   116  The Levenberg-Marquardt based regression therefore follows the flowchart outlined in Figure A.5. The regression algorithm is coded in MATLAB, with a separate code utilizing the built-in MATLAB root finding method fzero employed to accurately estimate the free concentrations of each pseudo-component from the adjusted parameter, a1.    Figure A.5: Computational Flowchart for model regression to isotherm data  Upon initialization of both the a1 parameter and the λ value, the chi squared is calculated and evaluated to determine if its value falls a below a pre-determined tolerance level indicative of a good model fit. If the tolerance criteria is not met, minimization of the model error is continued using a new a1 parameter calculated based on the current χ2 and  values. The λ value dictates the extent to which either the steepest decent or Gauss Newton method is applied.  If the new value of a1 is successful in reducing the chi squared, that value is accepted as the new parameter and λ is decreased. If the chi square value does not improve, the trial parameter is rejected and λ is increased in an attempt to improve the fit and provide a better trial parameter. This cyclic minimization of the chi squared is continued until the convergence tolerance is met.  117  The full algorithm code and the associated MATLAB files are provided at the end of the Appendix.  A.4 Application of the Algorithm to Adsorption Isotherm Data for Human Thrombin  A.4.1 Hi-Fi SELEX Both the conventional SELEX and the Hi-Fi SELEX aptamer selection protocol were applied to the discovery of DNA aptamers against human alpha thrombin.  After each round of selection, the mean binding isotherm for the pool of retained members was measured. Representative results for the Hi-Fi SELEX process are shown in Figure A.6.   Figure A.6:  Hi-Fi SELEX Binding Isotherm Data for the pool of library members retained after each selection round.  DNA aptamers are selected against human -thrombin in this system.  00.10.20.30.40.50.60.70.80.910 10 20 30 40 50Bound Fraction, θi,totConcentration of Total Free Aptamers (nM)Round 0 Round 1 Round 2 Round 3Mean Ka = NAMean Ka = 3.26 x 107 nM-1Mean Ka = 6.06 x 107 nM-1Mean Ka = 1.76 x 108 nM-1 118  Here, “round 0” represents the binding isotherm of the initial starting library, for which binding could not be detected. As the rounds of selection progress, the binding affinity of the retained pool of increases, resulting in an increase in slope of the linear region of the binding isotherm. This increase in mean binding affinity is verified by numerical values calculated using the traditional one-component Langmuir fit (see Chapter 2; equation 2.5), shown in Table A.2. Table A.2: Mean Ka for binding of each retained pool to human -thrombin Selection Round Mean Ka Round 0 N/A Round 1 3.26 x 107 nM-1 Round 2 6.06 x 107 nM-1 Round 3 1.76 x 108 nM-1  The algorithm described was then applied to the isotherm data to estimate the fraction of high binding affinity aptamers as a function of the selection round. Since the high binding affinity fraction is of interest, we set the Ka value of that pseudo-component to 109 M-1.  That of the second lower-affinity pseudo-component was set at Ka2 = 107 M-1. The results of the fitting are summarized in Table A.3.   Table A.3:  Estimated fraction (X1/Xtot) of high affinity binders after each round of Hi-Fi SELEX when applied to the discovery of DNA aptamers against human -thrombin.  Model fit was taken as satisfactory when χ2 fell below 1 x 10-2. Selection Round X1/Xtot (%) Lambda λ Chi Squared χ2 Round 1 2.55% 2.48 x 10-6 6.79 x 10-3 Round 2 5.13% 3.37 x 10-7 6.42 x 10-3 Round 3 80.1% 6.20 x 10-9 9.06 x 10-3  119   From the data, the percentage of high binding affinity aptamers in component 1 greatly increases from round two to round three, showing that the majority of the selected population contains very tight binding aptamers after just 3 rounds of Hi-Fi SELEX. The ability of the algorithm to predict the fraction of high binding affinity aptamers is confirmed by the value of chi squared, which shows the goodness of fit of the residuals in the predicted and experimental 𝜃𝑖,𝑡𝑜𝑡 values. In addition, the approach of the dampening factor, λ, to a value close to zero ensures the convergence of an accurate solution to the minimization problem initially stated.  The model-predicted binding isotherm, taking into account the relative size of the tight-binding pseudo-component, is seen to agree very well with experiment for each selection round. Those fits are shown in Figures A.7 to A.9.  Figure A.7: Model fit to Round 1 Isotherm  120   Figure A.8: Model fit to Round 2 Isotherm   Figure A.9: Model fit to Round 3 Isotherm   121  In the figures, the blue triangles represent the values obtained experimentally while the model-derived points are represented by the yellow diamonds. The fit of the experimental values is graphically represented by the red line.  A.4.2 SELEX vs. Hi-Fi SELEX Aptamer Selection for Human a-Thrombin The fact that more than 80% of the total aptamer population is characterized by high affinity binding of average 109 M-1 after three rounds of Hi-Fi SELEX shows the efficiency of our new selection process in growing the high affinity population in a relatively small number of selections rounds. To provide a benchmark, pseudo-component histogram data for application of traditional SELEX to aptamer selection for human alpha thrombin are shown in Figure A.10 for selection rounds 3 and 5 of that process.  The histogram for round 3 of Hi-Fi SELEX is shown in Figure A.11. In comparison to Hi-Fi SELEX, SELEX technology is seen to perform poorly. Following the third round of SELEX, the majority of the retained library members fall in a pseudo-component exhibiting a mean Ka of 105 M-1.  In Hi-Fi SELEX, the vast majority of retained library members are tight binding having a Ka ~ 109 M-1.  This shows a difference of five orders of magnitude in binding affinity between these two technologies.   122    Figure A.10:  Pseudo-component histograms for Rounds 3 and 5 of traditional SELEX applied to DNA Aptamer selection against human -thrombin.  In round 3, the isotherm was best fit using two pseudo-components with Kai values of Ka2 = 105 M-1 and Ka1 = 107 M-1.  In round 5, the pseudo-component Kai values were Ka2 = 107 M-1 and Ka1 = 109 M-1.                Kai          105      107       109         M-1        2         1        P seudo-component 	           Kai          105      107       109         M-1                 2         1          Pseudo-component 	 123   Figure A.11: Pseudo-component histograms for Round 3 of Hi-Fi SELEX applied to DNA Aptamer selection against human -thrombin.  In round 3, the isotherm was best fit using two pseudo-components with Kai values of Ka2 = 107 M-1 and Ka1 = 109 M-1.   A.5  Raw MATLAB Code aptamerLM.m function [params SD] = aptamerLM(data,Xitot,InitGuess,func,Ka,qmax) %the goal of this function is to return an iterative parameter regression %using the Levenberg Marqardt scheme that combines the Gauss Newton %method and the fastest descent method   % data should be input in the form [[xitot], thetaitot]; experimental data % Xitot is the total concentration of aptamers introduced into the system % InitGuess is an initial guess for the parameter value a % func must be a variable input function defined in an m file % Ka is a vector of the equilibrium association constants; length should be 2 for a 2-pseduocomponent system  	00.20.40.60.811 1.00E+04 1.00E+05 1.00E+07Round	3	HiFi SELEX1															104														105													107													109	Ka	Values     Ka alues 1               104              105              107            109  124  % qmax is the number of total available sites; experimentally is the amount of macromolecule immobilized   %% DEFINE CONSTANTS lambda=1E-3; % this is our initial value for the dampening parameter lambdaUP=11;% this will be used to change lambda up lambdaDW=9;% this will be used to change lambda down a=InitGuess;% is our initial guess of the parameter set; fraction of Xitot maxIT=500; %this is the maximum number of iterations count=1; % the first iteration; tol=1E-10; % set a tolerance for comparison of chi squared iterations thetaerr=1; %initially set as 1 for the variance value   %% DEFINE DATA (Y=thetaitot; X=[Xitot])   [m,n]=size(data); Xdata=data(1:m,1); %a vector; [X,ITOT], EXPERIMENTAL VALUES; x in column1 Ydata=data(1:m,n); % a vector; THETA,ITOT, EXPERIMENTAL VALUES; y in column2  Sigma=thetaerr*ones(m,1);% this can be manipulated to be a vector when data is provided W=1/(thetaerr).^2*eye(m); %EROR WEIGHTED Matrix  %% L_M METHOD STEPS    % open data file for collection  125  fileID = fopen('data_new.txt','w');   % check initial chi squared value with initial guess chiSquared=sum(((Ydata-func(Xitot, a, Ka, qmax, Xdata))./Sigma).^2);     % print out titles and initial guess values fprintf(fileID, '%10s %10s %20s\r\n','a','lambda','chiSquared'); fprintf(fileID,'%10.9f %15.10f %15.5E\r\n', a, lambda,chiSquared);     while chiSquared>tol && count <maxIT % if chi squared reaches zero we can stop    J=numJacobian(func, Xitot, a, Ka, qmax, Xdata);    alpha=J'*W*J +lambda*(diag(diag(J'*W*J),0));    beta=J'*W*(Ydata-func(Xitot, a, Ka, qmax, Xdata));    astep=alpha\beta;    aTest=a+astep;          %check new test chi squared value    chiSquaredTest=sum(((Ydata-func(Xitot, aTest, Ka, qmax, Xdata))./Sigma).^2);    % Need some criteria to stop if chi squared won't reach our tolerance     if abs(chiSquaredTest-chiSquared)<1E-10 % chi Squared is greater than the tolerance but is not changing anymore        break     end      126     if chiSquaredTest<chiSquared % if our guess improved        %update a        a=aTest;        %reduce lambda        lambda=lambda/lambdaUP;        chiSquared=chiSquaredTest; %        display('good guess') %        display(a) %        display(lambda)                      else % our guess did not improve        % no need to update a        % increase lambda though:        lambda=lambda*lambdaDW; %        display('bad guess') %      display(a) %        display(lambda)                           end    count=count+1; % count the number of iterations    fprintf(fileID,'%10.9f %15.10f %15.5E\r\n', a, lambda, chiSquared); %print out results after one LM iteration end % out of while loop, we have solved for parameters    127  params=a; % display(chiSquared) % display(chiSquaredTest)   fclose(fileID); % ERROR ANALYSIS % Now that we have the parameters we need to calculate the parameter SDs  %    J=numJacobian(func, Xitot, a, Ka, qmax, Xdata); %    alpha=J'*W*J +lambda*(diag(diag(J'*W*J),0)); %    V=alpha^-1; % the co variance matrix     % Want the asymptotic standard parameters: measure of how unexplained % variability in the data propagates to variability in the parameters % SD=sqrt(diag(V)); end     128  afunction.m function theta = afunction(Xitot, a, Ka, qmax, Xdata)  % this function relates a mole fraction for a two pseudo component system % to a predicted bound fraction or theta using user inputted values   % Xitot is the total aptamer concentration introduced into the system % a is the mole fraction of the first component in a two pseudo-component system % Ka is a vector of equilibrium association constants % qmax is the number of total available sites; experimentally is the amount of macromolecule immobilized % Xdata is the total free apatamer concentration or xitot   X = ones(length(Xitot),2); %creats a matrix with 2 columns - first column X1; second column X2 X_free=X; X(:,1) = Xitot*a; %first column X1; pass as mole fraction times total aptamer concentration X(:,2) = Xitot*(1-a); %second column X2; mass balance with X1 and total value   % calculate the roots using built-in MATLAB fzero function for datacount = 1:length(Xitot)       fun= @(x0) freeX1(Ka,qmax,X,Xdata,x0,datacount); %freeX1.m m-file with analytical solution; positive root of quadratic     X_free(datacount,1) = fzero(fun,a);  129      X_free = real(X_free); % takes real root values end   % calculate free aptamer concentration X_free(:,2) = Xdata-X_free(:,1); X1_free = X_free(:,1); % determine theta value theta = Thetaitot(Xdata,Ka,X1_free); end     130  freeX1.m function fun = freeX1(Ka,qmax,X,Xitot_free,x0,count)   % this function is the analytical solution to solving total concentration and free concentration of aptamers % positive root for positive answers   freeX2 = Xitot_free(count)-x0;   term1= (Ka(1)*X(count,1)-Ka(2)*freeX2-qmax*Ka(1)-1); % -b portion of quadratic term2= sqrt((1+Ka(2)*freeX2+qmax*Ka(1)-Ka(1)*X(count,1))^2+4*Ka(1)*X(count,1)*(1+Ka(2)*freeX2)); % sqrt of b^2-4ac term3= 2*Ka(1); % 2a   fun = (term1+term2)/term3-x0; % positive root of the quadratic formula    end      131  thetahat.m function thetahat = Thetaitot(Xdata,Ka,X_free)   % this function determines the predicted bound fraction (theta) given a vector of [xitot] and the [x1] % for a two pseudo-component system   % Xdata is the total free concentration [xitot] % Ka is the user defined binding affinity constant; a vector of size 2 % X_free is the calculated value for the free x1 aptamer concentration [X1]   %DEFINE VARIABLES   Xitot=Xdata; %vector of x variables (xitot) X1=X_free; %vector of concentrations of X2; passed as a guess   top= X1*(Ka(1)-Ka(2))+Xitot*Ka(2);   P=1+X1*(Ka(1)-Ka(2))+Xitot*Ka(2); %binding polynomial function for 2 components thetahat=top./P; %calculation of the bound fraction end     132  numJacobian.m function J = numJacobian(func, Xitot, a, Ka, qmax, Xdata)   % This function evaluates the numerical jacobian Ndatapoints x M parameters %ex: J (i,j)= dFunci/dxj   % func is a funciton handle % xitot, Ka, qmax is value used in func % Xdata is a vector input for [xitot] or free aptamer concentration % a is the parameter   %define constants da=1E-7; %this is our step size for our parameters    %Make the Jacobian matrix J=zeros(length(Xdata),1); % make an NxM matrix for J   for i =1:length(J(1,:)) % for every column in J          %every row we need to step only the parameter corresponding to that row     aFstep=a;     aRstep=a;     aFstep(:,i)=aFstep(:,i)+da; %forward step     aRstep(:,i)=aRstep(:,i)-da; %reverse step  133           %Using a central differencing approximation     J(:,i)= (func(Xitot, aFstep, Ka, qmax, Xdata)-func(Xitot, aRstep, Ka, qmax, Xdata))./(2*da); end   %J %pause   end      

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0364031/manifest

Comment

Related Items