INTERACTIONS BETWEEN THERAPEUTIC PROTEIN FORMULATIONS and SURFACES by BRANDON MICHAEL TESKA B.S., University of Colorado Boulder

INTERACTIONS BETWEEN THERAPEUTIC PROTEIN FORMULATIONS AND SURFACES by BRANDON MICHAEL TESKA B.S., University of Colorado Boulder, 2007 B.M., University of Colorado Boulder, 2007

A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulﬁllment of the requirements for the degree of Doctor of Philosophy Pharmaceutical Sciences Program 2015 This thesis for the Doctor of Philosophy degree by Brandon Michael Teska has been approved for the Pharmaceutical Sciences Program by

Thomas J. Anchordoquy, Chair John F. Carpenter, Advisor David L. Bain Theodore W. Randolph Krishna M. G. Mallela

Date: 07/30/2015

ii Teska, Brandon Michael (Ph.D., Pharmaceutical Sciences) Interactions Between Therapeutic Protein Formulations and Surfaces Thesis directed by Professor John F. Carpenter

ABSTRACT

Therapeutic protein formulations encounter a multitude of different surfaces in every part of their production, packaging, storage and administration to patients. These interfaces can be very different—chemically—from the formulation’s solution chemistry and can have unintended, negative effects on the formulation. Protein molecules can adsorb to these surfaces, which can induce structural perturbations in the therapeutic and promote aggregation. Components from the formulation can be absorbed into the surfaces, altering the formulation solution conditions, which can bring about additional stresses on the therapeutic. Formulations can even chemically modify surfaces they come in contact with, altering the surface properties. Understanding such interactions between formulations components and surfaces is critical to developing better storage and delivery devices and improved formulations. In this work, I examined the pharmaceutical compatibility of a novel syringe plungercoating—designedtobe used in silicone oil-free, pre-filled syringes—and found it to cause much less protein aggregation during agitation than was observed in a traditional siliconized syringe. Second, I found that plastics found in catheters absorbed phenolic compounds from insulin analog formulations and that this depletion had a profound impact on different insulin analogs’ assembly states and stabilities under thermal stress. Finally, I also found that zinc ions, found in insulin formulations as well, chemically damaged analytical size exclusion chromatography columns, which are used to monitor soluble aggregates of insulin in therapeutic formulations.

iii The form and content of this abstract are approved. I recommend its publication.

Approved: John F. Carpenter

iv DEDICATION

This work is dedicated to my wife, Johanne, who has been my inspiration and en- couragement throughout my graduate career. To my parents, Mike and Cindy, who encouraged and supported my aspirations to become a scientist from my early years and to my brothers and all my good friends who have been my biggest fans throughout the years.

v ACKNOWLEDGEMENTS

First and foremost, I would like to thank Dr. John Carpenter for his mentorship, guidance and support throughout my thesis work in his lab. John is passionate about ﬁghting for patients, one of the many things I’ve loved about working in his lab. Because of his diligence and persistence, we’re beginning to see pharmaceutical companies, regulators and clinicians work together to identify, track and mitigate sources of immunogenicity. I feel very honored to have been a part of this—we are literally changing the face of medicine for the betterment ofthepatients. I would also like to thank the members of my committee. Dr. DaveBainfor encouraging me to work towards a deeper, mechanistic understanding of my ﬁndings, Dr. Ted Randolph for unique way of thinking about the biophysics that helped drive me to a deeper understanding of pharmaceutics, Dr. Krishna Mallela for sharing his protein characterization expertise and Dr. Tom Anchordoquy for his insightful questions and for his uncanny ability to make everyone smile. Thank you all! I am eternally grateful for your guidance and mentorship during my thesis work. Secondly, I would like to thank all of my colleagues and friends at the University during my tenure in John’s lab (in no particular order): Dr. AjayThomas,Dr.Merry Christie, Dr. Tyson Smyth, Indu Persaud, Dr. Nicole Payton, Dr. Regina Bis, Dr. Tanya Clapp, Neha Pardeshi, Dr. Pinaki Basu, Dr. Jim Barnard, Dr. Wei Qi, Dr. Pradyot Nandi, Jan Jaap Verhoef, Shyam Mehta, Chen Zhou, Dr. Aaron Krueger, Aditya Gandhi and Dr. Luke Liu. For all your stimulating conversations, advice and laughs on all things science (and non-science) over the years, thank you! Third, I would like to thank all of the students that worked for me on various parts of the work in this thesis: Brad Butler, Rahie Talukder, Isabel Randolph, Mary Whitney, Sara Brown, Brian Adams, Lindsey Ross and Samantha Egger. Thank you for all your hard work and dedication contributing to thisimportanttopicin

vi pharmaceutical sciences. Fourth, I would like to that the organizations that have funded parts of my research. Becton Dickinson and Company, W. L. Gore and Associates, and the Phar- maceutical Research and Manufacturers of America. Lastly—but certainly not least—I would like the thank my wife, Johanne, for standing by me through thick and thin. I would also like to thank my parents, my brothers and my friends for their love and support.

vii TABLE OF CONTENTS

CHAPTER

I INTRODUCTION 1

II INVESTIGATION OF THE PHARMACEUTICAL COMPATIBILITY OF A NOVEL SURFACE FOR USE IN THERAPEUTIC PROTEIN DELIVERY DEVICES 17

Abstract...... 17

Introduction ...... 18

Materials and Methods ...... 22

Materials ...... 22

AgitationinVials...... 23

AgitationinSyringes ...... 24

Micro-FlowImaging(MFI)...... 26

SizeExclusionChromatography(SEC) ...... 27

Confocal Microscopy of the Adsorption of Fluorescently Labeled IVIGtotheFluoropolymerSurface ...... 27

Determination of the Second Osmotic Viral Coeﬃcient (B22)..... 29

IVIGUnfoldingwithUrea ...... 30

Results and Discussion ...... 30

IncreasedAgitationStressinVials ...... 33

Eﬀect of Buﬀer Type on Aggregation of IVIG During Agitation in Vials 34

Agitation of IVIG in Glycine Buﬀer ...... 35

Agitation of IVIG in PBS Buﬀer ...... 36

AgitationofAvastin ...... 38

AgitationinSyringes ...... 41

Agitation of IVIG in Glycine Buﬀer ...... 41

Agitation of IVIG in PBS Buﬀer ...... 42

viii Morphological Diﬀerences Between Protein Particles Formed From Agitation of IVIG in the Fluoropolymer Syringe and the Siliconized Syringe . 44

AgitationofAvastin ...... 47

Mechanism(s) for Inhibition of Protein Aggregation by Polysorbate 20DuringAgitation ...... 47

Eﬀects of pH and Buﬀer on the Conformational and Colloidal StabilityofIVIG ...... 53

The Utility of the Fluoropolymer Surface in Silicone Oil-Free Syringes 58

Acknowledgments...... 59

III EFFECTS OF PHENOL AND META-CRESOL DEPLETION ON INSULIN ANALOG STABILITY AT PHYSIOLOGICAL TEMPERATURE 60

Abstract...... 60

Introduction ...... 61

Materials and Methods ...... 64

Materials ...... 64

Stability Study ...... 64

SizeExclusionChromatography(SEC) ...... 65

Reversed Phase Chromatography (RP) ...... 66

Micro-FlowImaging(MFI)...... 66

AtomicAbsorptionSpectrometry(AAS) ...... 67

AnalyticalUltracentrifugation(AUC) ...... 67

Results ...... 68

AtomicAbsorptionSpectroscopy ...... 68

SizeExclusionChromatography ...... 69

Reversed-Phase Chromatography ...... 71

AnalyticalUltracentrifugation ...... 76

Micro-FlowImaging ...... 79

ix Discussion ...... 80

Acknowledgements ...... 84

IV ANALYZING INSULIN SAMPLES BY SIZE-EXCLUSION CHROMATOGRAPHY: A COLUMN DEGRADATION STUDY 93

Abstract...... 93

Introduction ...... 93

Materials and Methods ...... 95

Materials ...... 95

Silica Resin Incubation ...... 95

Liquid-Liquid Extraction ...... 96

Size-Exclusion Chromatography ...... 96

Nuclear Magnetic Resonance and Gas Chromatography-Mass Spectrometry ...... 97

Results and Discussion ...... 98

Conclusion ...... 103

Acknowledgements ...... 103

V CONCLUSIONS AND FUTURE DIRECTIONS 107

ChapterII: NewSurfacesin Pre-ﬁlled Syringes ...... 107

ChapterIII: InsulinAnalogAssembly...... 112

Chapter IV: Eﬀects of Formulations on Surfaces ...... 119

Concluding Remarks ...... 120

REFERENCES 146

APPENDIX 147

A SUPPLEMENTAL FIGURES FOR CHAPTER II 147

Methods ...... 147

x Resonant Mass Measurement (RMM) ...... 147

Results and Discussion ...... 148

Vial Agitation: IVIG in Glycine Buﬀer ...... 148

Vial Agitation: IVIG in PBS Buﬀer ...... 150

VialAgitation:Avastin...... 150

Syringe Agitation: IVIG in Glycine Buﬀer ...... 150

Syringe Agitation: IVIG in PBS Buﬀer ...... 153

SyringeAgitation:Avastin...... 153

B EXTENT OF ADSORPTION OF PHENOLIC PRESERVATIVES BY DIFFERENT BRANDS OF COMMONLY USED INSULIN PUMP CATHETERS 156

Abstract...... 156

Introduction ...... 156

Materials and Methods ...... 157

Preparation and Incubation of the Catheter Tubing ...... 158

UVAbsorbanceSpectroscopy ...... 160

Results and Discussion ...... 160

Limitations and Future Directions ...... 160

Acknowledgments...... 162

xi LIST OF TABLES

TABLE

1 Phenolic Preservative Concentration in Various Units ...... 85

2Mobilephasecompositionduringreversed-phasechromatography . . 86

3Theoreticalsedimentationcoeﬃcient limits for diﬀerent self-assemblies of insulin glulisine...... 117

4Cathetersetdimensions...... 159

xii LIST OF FIGURES

FIGURE

1 Agitationstudyvialstoppers...... 24

2Glass-onlyagitationaparatus...... 25

3 Agitationstudysyringesandplungers...... 26

4 Particle formation by MFI for IVIG in 0.2M glycine buﬀer after agitationinvials...... 35

5 Total soluble protein mass recovery by SEC for IVIG in 0.2M glycine buﬀer after agitation in vials...... 36

6 Particle formation by MFI for IVIG in PBS buﬀer after agitation invials...... 37

7 Total soluble protein mass recovery by SEC for IVIG in PBS buﬀer after agitation in vials...... 38

8 Particle formation by MFI for Avastin after agitation in vials. . . . . 40

9 Total soluble protein mass recovery by SEC for Avastin after agitationinvials...... 40

10 Particle formation by MFI for IVIG in 0.2 glycine buﬀer after agitationinsyringes...... 43

11 Total soluble protein mass recovery by SEC for IVIG 0.2M glycine buﬀer after agitation in syringes...... 43

12 Particle formation by MFI for IVIG in PBS buﬀer after agitation insyringes...... 45

13 Total soluble protein mass recovery by SEC for IVIG in PBS buﬀer after agitation in syringes...... 45

14 Representative MFI images for IVIG particles from day seven of agitationinsyringes...... 46

15 Particle formation by MFI for Avastin after agitation in syringes. . . 48

16 Total soluble protein mass recovery by SEC for Avastin after agitationinsyringes...... 48

17 Representative confocal microscopy images for IVIG adsorbed to the ﬂuoropolymer surface...... 50

xiii 18 Representative confocal microscopy images for IVIG adsorbed to the ﬂuoropolymer surface after washes...... 52

19 Representative confocal microscopy images for IVIG adsorbed to a siliconized surface...... 54

20 Representative confocal microscopy images for IVIG adsorbed to a siliconized surface after washes...... 55

21 Urea unfolding curves for IVIG in glycine and PBS buﬀers...... 56

22 Second virial coeﬃcient (B22) for IVIG in glycine and PBS buﬀers. . 57

23 Zinc content in insulin lispro, aspart and glulisine before and after processingasmeasuredbyAAS...... 69

24 Representative SEC chromatograms of insulin lispro elution...... 70

25 Insulinlispro degradation measuredbySEC...... 71

26 Insulin aspart degradation measured by SEC...... 72

27 Insulin glulisine degradation measured by SEC...... 73

28 Insulin lispro, aspart and glulisine drug product degradation measured by SEC...... 74

29 Representative RP chromatograms of insulin lispro elution...... 75

30 InsulinlisprodegradationmeasuredbyRP...... 77

31 Insulin aspart degradation measured by RP...... 87

32 Insulin glulisine degradation measured by RP...... 88

33 Insulin lispro, aspart and glulisine drug product degradation measuredbyRP...... 89

34 Representative c(s20,w) distributions for insulin lispro, aspart and glulisine...... 90

35 Total particle concentration data for insulin lispro, aspart and glulisineasmeasuredbyMFI...... 91

36 Total particle concentration data for insulin lispro, aspart and glulisinedrugproductsasmeasuredbyMFI...... 92

37 SEC columnn degradation during a previous study...... 100

38 SEC elution proﬁle of insulin lispro by SEC-MALS...... 101

xiv 39 NMRspectraofdegradedcolumnresinextract...... 104

40 GC/MS chromatograms of SEC column resin extracts after diﬀerent treatments...... 105

41 GC/MS chromatograms of SEC column resin extracts after treatmentwithZincandETDA...... 105

42 SEC column performance with and without EDTA in the mobile phase. 106

43 Particle formation by RMM for IVIG in 0.2M glycine buﬀer agitated in vials...... 149

44 Particle formation by RMM for IVIG in PBS buﬀer agitated in vials. 151

45 Particle formation by RMM for IVIG in 0.2M glycine buﬀer agitatedinsyringes...... 152

46 Particle formation by RMM for IVIG in PBS buﬀer agitated in syringes.154

47 Particle formation by RMM for Avastin agitated in syringes...... 155

48 Amount of phenolic preservative remaining after incubation in diﬀerent catheters...... 161

xv LIST OF ABBREVIATIONS

BSA: Bovine Serum Albumin

BSAU: Bovine Serum Albumin plus Uracil

CD: Circular Dichroism

CDCl3: Deuterated Chloroform

ECD: Equivalent Circular Diameter

EDTA: Ethylene-Diamine-Tetra-acetic Acid

EtOAc: Ethyl Acetate

FTIR: Fourier-Transform Infrared Spectroscopy

FTIR-ATR: Fourier-Transform Infrared Spectroscopy using Attenuated Total Reﬂectance

GC/MS: Gas Chromatography-Mass Spectrometry

GS: Gore Fluoropolymer Surface

IVIG: Intravenous Immunoglobulin (Gammagard⃝R )

MALS: Multi-Angle Light Scattering

MgSO4: Magnesium Sulfate

MFI: Microﬂow Imaging

NMR: Nuclear Magnetic Resonance

NTA: Nano-Tracking Analysis (NanoSight)

PBS: Phosphate Buﬀered Saline

PTFE: Poly-Tetra-Fluoro Ethylene

PS20: Polysorbate 20 (Tween 20)

RP, RPC: Reversed-Phase Chromatography

RT: Room Temperature (18-22◦C)

SiS: Siliconized Surface

SEC: Size-Exclusion Chromatography

SE-AUC: Sedimentation Equilibrium Analytical Ultracentrifugation

SV-AUC: Sedimentation Velocity Analytical Ultracentrifugation

xvi CHAPTER I

INTRODUCTION

The development and use of protein therapeutics has increased greatly in the past decade [1–3], rising from an approximately $50 billion dollar market in 2004 to a nearly $160 billion dollar market in 2014 [4]. Proteins are highly complex, specialized molecules and their use as pharmaceuticals has opened up new avenues of treatment for once intractable diseases such as diabetes [5], hemophilia [6] and idiopathic thrombocytopenic purpura [7–9]. To properly function, proteins must adopt and maintain a complex three-dimensional, folded structure, which ensures that the specific interactions sites on the surface of the protein—which allow the protein to perform its biological (or pharmaceutical) function—are correctly formed. Alterations to, or loss of, the protein’s structure can cause the protein to lose its ability to perform its pharmaceutical function. An unfolded protein also exposes new surfaces that would be otherwise hidden in the properly folded protein. These surfaces have the potential to interact with other protein molecules and to form aggregates in solution. In general, the surface sites responsible for the protein therapeutic’s pharmaceutical function are either malformed or inaccessible in aggregates, which leads to adecreaseinpharmaceuticalefficacy. Protein aggregation in therapeutic proteins is also a concern for patient safety, as aggregates have been found to contribute immune responses in patients to protein therapeutics [10–16]. Immune responses to protein therapeutics can range from relatively benign, such as is thecasewithinsulin—where roughly 30% of patients produce anti-insulin antibodies [17] but the presence of which doesn’t affect the overall clinical utility of insulin therapy—to generating anti-drug antibodies that neutralize the therapeutic protein [18, 19]; some anti-drug antibodies even have the potential to cross-react with endogenous proteins in the patient’s body [20, 21] which can have grave consequences for the patient.

1 To avoid the issues of immunogenicity and loss of efficacy of protein therapeutics, formulation scientists at pharmaceutical companies take great care in developing and optimizing the solution conditions in which a protein therapeutic will be produced, stored and delivered. The solution buffer, pH, ionic strength, and additional stabilizing excipients all affect the structural and chemical stability of the individual protein molecules in solution, and can also affect the colloidal stability of the entire population of proteins [22]. Optimizing solution pH is perhaps the most general and most effective means of stabilizing protein therapeutics in their pharmaceutical formulation [23]. In practice, anumberofstudieshavefoundthatsolutionpHgreatlyinfluenced the stability of protein therapeutics [24–26]. Increasing or decreasing thepHawayfromtheprotein’s isoelectric point results in an increase in the overall surface charge on the molecule. Localized regions of like charges in close proximity on the surface of the protein can then repel each other which can affect the protein’s overall conformational stability [27]. Increased surface charge can also result in attractiveelectrostaticforcesbetween molecules which can affect the protein’s overall colloidal stability. If the distribution of surface charges is such that the protein has an overall dipole, this can result in aggregation due to the electrostatic attractions between opposite charges on each molecule [28]. The ionic strength of the formulation also affects electrostatic interactions within and between protein molecules. The nature of salt effects on protein stability are complex, affecting both the conformational and colloidal stability of the protein and the protein’s overall solution solubility [29–33]. However, there are two mechanisms of salt effects on protein stability that are relevant to this work. First, salts can reduce electrostatic interactions between regions of localized charge on the surface of aproteinandbetweenproteinmoleculesbychargeshielding. In the case discussed above, where a change in pH resulted in a highly charged, conformationally unstable

2 molecule, charge shielding can help stabilize the protein’sconformationbydampen- ing the electrostatic repulsions between regions of localized charge on the protein’s surface. However, charge shielding can also dampen overall electrostatically repulsive forces between protein molecules, thereby lowering the barrier to aggregation. Sec- ond, salt ions can bind to specific locations on the protein and create salt bridges between charged amino acids. The creation of these salt bridges can stabilize the conformation of a protein. The interactions of salts with different proteins are dependent on the overall charge of the molecule. Therefore, theeffects of ionic strength and pH on protein conformational and colloidal stability areintertwined. In addition to formulation pH and ionic strength, additional chemicals—termed “excipients” in pharmaceutical sciences—can be added to formulations that can affect protein stability [23]. Two examples of this, relevant to this work, are the addition of metacresol and phenol, two antimicrobial preservatives, and polysorbate 20, a surfactant. Phenolic preservatives, such as phenol and metacresol, protect a therapeutic protein formulation from microbial contamination. When drug is repeatedly drawn from the same vial, each needle insertion has the potential tointroducemicrobes into the therapeutic protein formulation. For multi-dose vials, the FDA requires antimicrobial preservatives to be added to prevent contamination [34]. In many cases, antimicrobial preservatives are denaturing to proteins and enhance aggregation propensity [35–38]. Antimicrobial preservatives have been shown to cause proteins to partially unfold, which subsequently leads to higher rates of aggregation [39–43]. However, as will be discussed in depth in Chapter III, insulin (and insulin analogs) are a notable exception to this generality. Phenol and metacresol specifically bind to insulin causing a change in conformation that enhancesinsulin’sthermal stability. Surfactants are surface-active molecules that consist of a hydrophilic head group and a long hydrophobic tail. When surfactants are added to an aqueous solution,

3 they spontaneously adsorb to interfaces, such as the air-water interface and the water- container surface interface oriented such that their hydrophilic head faces the solution and their hydrophobic tail faces the more hydrophobic surface. Protein molecules also consist of hydrophilic and hydrophobic regions and therefore also spontaneously adsorb to interfaces such as the air-water interface [44, 45] and other interfaces found in pharmaceutical containers [46]. Protein adsorption to hydrophobic interfaces, in particular, has been shown to cause perturbations to the structure of adsorbed proteins [47–51] which, in turn, can result in increased protein aggregation rates [52–55]. Including a surfactant in a therapeutic protein formulation introduces competition for the hydrophobic interface and thereby helps prevent protein denaturation and subsequent aggregation on the surface. By adjusting solution pH, ionic strength and by adding additional excipients to a particular therapeutic protein formulation, one can createaformulationthatmaxi- mizes the conformational stability of the individual protein molecules and maximizes the colloidal stability of the entire population of proteins as a whole, thereby minimizing protein aggregation. Achieving maximal formulation stability is important not only during quiescent storage, but also when the protein therapeutic is subjected to common stresses that it will experience before ultimatelybeingdeliveredtothe patient. Stresses to the therapeutic protein can come from all stages of the protein’s production and use: the production process, the purification process [56], and final vial or syringe filling and packaging [57]. Even transportation ofthefinaldrugproduct can place stress on the protein therapeutic [58, 59]. All of these stresses can cause perturbations in the protein’s structure and/or cause it to aggregate [60]. Protein aggregates can exist in many sizes, ranging from “soluble” aggregates (10–100nm in size), to micron-sized, “sub-visible” protein particles[61].Themeasurementof sub-visible particles in particular (specifically particles between 1µmand∼150µm)

4 has emerged as one of the most sensitive techniques to detect and measure protein aggregation in solution [62, 63]. Particles of this size are also of particular interest to our lab as these particles have been implicated in cases of patient immunogenicity to protein therapeutics [64–67] (for an extensive review of protein particles and immunogenicity see ref. [68]). This implication, in turn, has led totheregulatoryrequirement [69] (and subsequent pharmaceutical industry business imperative) to understand the causes of protein aggregation and particle formation in protein therapeutics, and to develop control strategies to minimize them [70–73]. Concurrent with the focus on controlling aggregation and sub-visible particle formation in protein therapeutics, the industry has seen the development and rapid adoption of tailor-made delivery devices for individual protein therapeutics such as auto-injectors, infusion pumps and pre-filled syringes. These devices offer numerous advantages to manufacturers—less waste due to excessive overfill of vial containers— and offer patients better safety and more convenient and accurate dosing as well [74, 75]. Of particular interest—and the focus of this work—is the effects of surfaces in these new devices on therapeutic protein formulations. The interactions between surfaces and proteins are a well researched area [50, 76–82] as they encounter a multitude of different surfaces throughout the entire manufacturing process [83]. However, what is new and unique about these new, tailor-made delivery devices is that they occupy an interesting space—they must serve both as a storage vessel, and as a func- tional delivery device for the duration of their use which can range from 48–72 hours, as is the case with insulin pumps, to two years of shelf-life, in the case of a pre-filled syringe. In these delivery devices, a protein therapeutic might encounter a number of different surfaces that can interact with the protein. Examplesinclude:plastics,metals, different types of glasses, rubbers and lubricating oils. All of of these surfaces present the protein with a chemical environment that is significantlydifferent from the envi-

5 ronment it experiences in the bulk aqueous formulation—awayfromthesurface.For example, silicone oil, used as a lubricant in pre-filled syringes, presents a hydrophobic, fluid surface to the protein. In contrast, the glass wallsofavialpresentarigid, hydrophilic surface, that can be negatively charged. Flexible plastic catheters also present a hydrophilic surface, but less rigid than glass. Even the air-water interface presents a hydrophobic surface to the therapeutic protein formulation. All of these surfaces have the potential to interact with the protein in different ways. And, these interactions between the therapeutic protein formulation and the surface have the potential to destabilize the formulation, possibly resulting in protein aggregation and sub-visible particle formation. Protein interactions with surfaces are facilitated via the chemistry of the protein’s surface. The surface chemistry of any given protein is heterogeneous, consisting of areas with localized net electrostatic charge; uncharged, polar, hydrophilic areas; and hydrophobic areas. This heterogeneity allows proteins to interact with virtually any surface. For therapeutic proteins, some protein-surface interactions have been shown to have a negative effect on protein structure and/or aggregation propensity. In the context of protein-surface interactions between therapeutic protein formulations and delivery devices, there are two surfaces in particular that warrant introduction here: air and silicone oil. In all therapeutic protein storage and delivery devices—including glass vials—the most ubiquitous surface that a therapeutic protein will encounter is the air-water interface. When a protein formulation is exposed to an air-water interface, protein molecules spontaneously adsorb to the interface [44, 45]. Attheinterface,proteins orient themselves such that any hydrophobic regions on the protein’s surface orient towards the air side of the air-water interface. This specificorientationofproteins at the air-water interface occurs because solvation of the hydrophobic patches on the protein’s surface by water is entropically unfavorable. Water is unable to hydrogen-

6 bond with the hydrophobic surface and this limits the number of conformations nearby water molecules can assume to maximize their network of hydrogen bonds. Similarly, solvation of a hydrophobic surface (such as the air) by water is also unfavorable. Protein adsorption to the the hydrophobic surface, mediated through interactions between hydrophobic patches on the protein’s surface, reduces the overall hydrophobic surface area that is exposed to water, thus minimizing the overall entropic cost of solvation and driving protein adsorption to the interface [84, 85]. At the interface, proteins continue to adsorb and eventuallyforma“gel”layer. In a gelled protein layer at the air-water interface, the protein molecules are in close proximity with each other and are influenced by other protein molecules nearby. The process of adsorption alone, for some proteins, can be sufficient to induce structural perturbations in the protein. These structural perturbations, in combination with the close proximity of the protein molecules in the gel layer, can enhance protein aggregation. However, it has also been shown for the air-waterinterfacethatprotein aggregation measured in the solution in contact with the surface is dependent—in part—on the disruption of the protein gel layers adsorbed at the interface, either through mechanical rupture [86] or by compression and dilation of the interface [87]. Therefore, the overall effect of the air-water interface on protein aggregation in solution is dependent on the number times the adsorbed—and possibly aggregated and/or structurally perturbed—gel layer of proteins at the air-water interface is disrupted and released back into the bulk aqueous solution over a given amount of time [88]. The second surface common to therapeutic protein delivery devices is the silicone oil-water interface. Pre-filled syringe barrels are sprayed with a thin layer of silicone oil to facilitate smooth movement of the plunger during delivery of the therapeutic protein to the patient. The rubber plunger is also siliconized to provide lubrication during delivery and for machine handling during assembly and filling of the pre- filled syringe. Like the air-water interface, the silicone oil-water interface presents

7 a hydrophobic, fluid surface to the protein formulation. And, like the air-water interface, when exposed to a silicone oil surface, proteins spontaneously adsorb to the silicone oil-water interface, again driven by tendency to the minimize water solvated hydrophobic surfaces both on the protein and the silicone oil. Some studies have shown that adsorption of protein to the silicone oil-water interface was sufficient to cause structural perturbations to some proteins [84, 85, 89, 90], however, structural perturbations were not observed in all cases [91, 92]. Proteins adsorbed to the silicone oil-water interface also form a gel layer, similar to the air-water interface. Recent work by Mehta et al. [93] showed that, like the air-water interface, the gel layer of protein adsorbed to the silicone oil-water interface must be disrupted in order for there to be increased protein aggregation measured in the bulk formulation. In therapeutic protein delivery devices, specifically pre-filled syringes, the disruption of the protein gel layer adsorbed tothesiliconeoil-water interface is accomplished by agitating the delivery device. Pre-filled syringes must be filled with an air bubble1, which during agitation provides an air-water interface in addition to the silicone-oil water interface. Recent work by Gerhardt et al. [95] has proposed a mechanism toexplainhow the air-water interface is able to disrupt the gel layer of adsorbed protein on the silicone oil-water interface. First, proteins exposed to a silicone oil surface adsorb to the surface forming a gel layer at the silicone oil-water interface. When an air bubble comes in contact with the silicone oil-water interface, this creates an intersection between the air-water, silicone oil-water and the silicone oil-air interfaces. At the point of intersection, the surface tension of the air-water interface is imbalanced and creates a net force acting orthogonal to the silicone oil-water interface, towards the bulk solution. Based on estimates of the surface tensions of each of the three interfaces, Gerhardt et al. [95] estimated the magnitude of this force to be sufficient

1New syringe filling techniques now are capable of filling a pre-filled syringe without an air bubble[94], however, this technology is not extensively used as of yet.

8 to fragment and forcibly displace the gelled protein layer, releasing it back into the bulk formulation. Removal of the gel layer then exposes more of the silicone oil-water interface, thereby freeing space for the adsorption of more protein molecules from the bulk solution. The new protein layer forms a gel and the entireprocessisrepeated. As a consequence, agitating protein therapeutics in devices with siliconized surfaces with an included air bubble enhances protein aggregation measured in solution. Therefore, not only can the air-water interface have negative effects on a therapeutic protein formulation itself, but in combination with other interfaces, the air-water interface enhances the effects of the silicone oil-water interface on the therapeutic protein formulation measured in the bulk. This mechanism is supported—at least qualitatively—by a number of studies that have found that agitating a protein formulation in contact with a silicone oil surface with an air bubble caused significant increases in protein aggregation measured in solution [25, 26, 95, 96, 96]. Unfor- tunately, the stress of agitation is unavoidable as every protein therapeutic will in- evitably encounter some level of agitation stress during its production, shipping and possibly during patient/provider handling. To minimize the effects of the silicone oil-water interface on protein aggregation during agitation, surfactants are often added to the formulation [25, 26, 91, 97]. Like proteins, surfactants are surface-active molecules and also spontaneously adsorb to hydrophobic interfaces. Surfactants are thought to reduce the amount of protein adsorbed to the surface, presumably by out-competing the protein for adsorption sites on the silicone oil-water interface [98–100], however they have been found to directly interact with certain proteins in some cases [101–105]. It has been shown, however, that in some cases the presence of a surfactant is not sufficient to completely inhibit protein particle formation [25]. Furthermore, due to their surface-active nature, surfactants can also solubilize [106] and stabilize silicone oil droplets [97, 107] in the aqueous solution. Therefore, even in the absence of protein aggregation, formula-

9 tions including a surfactant can have elevated levels of silicone oil droplets that are injected with the therapeutic protein. This is undesirable, as there has been concern that injection of silicone oil droplets with therapeutic proteins may stimulate adverse immunogenicity in patients [67]. In addition to the inhibitory effects of surfactants on agitation-induced aggregation, a number of studies have shown that the therapeutic protein’s formulation also affects the rate of aggregation during agitation with siliconized surfaces [25, 26, 90]. Basu et al. [25] and Thirumangalathu et al. [26] found that, with the antibodies they studied, protein aggregation resulting from agitation in contact with a siliconized surface was much faster when the formulation was adjusted to pH 7.4 compared to pH 5.0. This is an interesting finding as it suggests that both conformational and/or colloidal instability of the protein—two properties measured in the bulk solution—can exacerbate protein aggregation during agitation. In addition, Basu et al. [25] and Thirumangalathu et al. [26] observed increased protein aggregation during agitation with siliconized surfaces in formulations with high ionic strength (150mM NaCl). Both found that the increased ionic strength of the formulation reduced the colloidal stability of their respective antibodies. Both studies also found that surface-induced aggregation could be reduced by maximizing the conformational and colloidal stability of the monoclonal antibody. In contrast, Gerheardt et al. [90] found that—for their specific antibody—increasing the ionic strength of their formulation resulted in decreased aggregation resulting from agitation in contact with a siliconized surface. Taken together, the results from these three studies show that formulation conditions have a profound effect on a given protein’s susceptibility to agitation-induced aggregation in contact with siliconized surfaces, and that each protein molecule’s behavior—even within classes of proteins, such as antibodies—is unique and distinct. In some cases, adding a surfactant and optimizing a therapeutic protein’s sta-

10 bility in formulation may still not be sufficient to stabilize the protein against the effects of the siliconized surface. Therefore, the next prudent step towards reducing the ill effects of siliconized surfaces on therapeutic proteins would be to eliminate silicone oil from delivery devices completely. To that end, two companies, Terumo Medical Corportation and Daikyo Seiko, have developed syringes with plastic barrels and complementary plungers that do not require additional silicone oil lubrication to function. However, despite the development of silicone oil-free, polymer-based syringe systems, plastic pre-filled syringes have not been eagerly adopted for use with therapeutic proteins in most of the world [108]. This may be due to the opacity of early plastics, which complicated visual inspection of the final filled syringe, although currently, new polymers are approaching glass-like transparency [108]. Or, this may be due to other issues such as: larger numbers and broader chemical diversity of leachable compounds from plastic syringes [109, 110], gas permeation issues [111] and fewer compendial standards for plastics [112] as compared toglass.Regardlessofthe reasons for the lack of adoption of plastic pre-filled syringes for protein therapeutics, siliconized, glass syringes still remain the standard technology for pre-filled syringes world-wide. In Chapter II, I examined therapeutic protein stability in a new type of glass, silicone oil-free, pre-filled syringe system. Neither the glass barrel of the syringe nor the syringe plunger in this system is siliconized. Instead, lubrication is provided by an experimental fluoropolymer-surfaced plunger. The fluoropolymer surface provides “solid phase” lubrication for the syringe plunger’s movement against the bare-glass syringe barrel. Given the negative effects siliconized surfaces can have on a therapeutic protein discussed above, the development and characterization of this novel pre-filled syringe system represents a significant step in thedevelopmentofalternative materials for use in delivery devices for therapeutic proteins. As I alluded to earlier, by the diverse nature of their surface chemistry, proteins

11 have the potential to interact with virtually any surface. Numerous previous studies have examined the effects of polytetrafluoroethylene (PTFE, better known as “Teflon”) surfaces on proteins—which is chemically similar to the experimental fluoropolymer surface I examined in Chapter II. Vermeer et al. [113, 114] and Mollmann et al. [115] found that absorption of an antibody and insulin molecules, respectively, to PTFE surfaces resulted in perturbation in the protein’s secondary structure. And, like siliconized surfaces, agitation of proteins in contact with PTFE surfaces resulted in increased protein aggregation [52–55]. The experimental fluoropolymer surface I examined in ChapterIIwasnotcom- pletely “inert” with respect to protein aggregation and particle formation during agitation of the antibodies I tested. However, based on my work, and building on the observations of others on protein interactions with PTFE surfaces [52, 116], I was able to present rational explanations as to why the surface had some damaging effects on the proteins I tested and what changes to the formulation could be made to minimize them. After the formulation was optimized, agitation of two therapeutic antibodies in contact with this fluoropolymer surface in a bare-glass syringe with the fluoropolymer-surfaced plunger generated far less protein particle formation than an equivalent siliconized syringe system. My results demonstrate the promise of new materials such as this fluoropolymer surface as a component in the next generation of pre-filled, silicone oil-free, glass syringe systems for therapeutic proteins. In addition to direct effects of a surface on the therapeutic protein, as I explored in Chapter II, surfaces can also indirectly affect a therapeutic protein by interacting with and altering the therapeutic protein’s formulation. Changes in the therapeutic protein’s formulation—mediated by surfaces the formulation comes in contact with in delivery devices—can have dramatic impacts on the protein’sstabilityinthedelivery device. In Chapter III, I explored the effects of surface-induced excipient depletion on

12 insulin analog stability. Insulin is an important therapeutic for millions of diabetics worldwide. It is a small, 5.8kDa, peptide hormone that regulates—among other things—glucose uptake by cells in the body. The condition where a person’s body either doesn’t produce enough insulin (termed “type 1”) or when the insulin produced doesn’t work well with insulin receptors (termed “type 2”), is called diabetes. Dia- betics are dependent on multiple subcutaneous injections ofinsulinperday,primarily to control postprandial spikes in blood glucose concentrations after meals [117]. Endogenous human insulin is stored in pancreas beta-islet cells before release into the blood. As part of this storage mechanism, insulin undergoes a complex self- assembly process from a single, fragile, monomeric insulin molecule to a hexameric assembly of six insulin molecules coordinated around two zinc ions [118]. In this assembled state, insulin is exceptionally stable, but is pharmacologically inactive. To exert its biological function, insulin must dissociate backintothemonomericformin order to bind its receptors. When used as protein therapeutic, the intrinsic process of dissociation for human insulin can complicate treatment. Human insulin dissociatesslowly,resultingina considerable, 30–60 minute delay between a therapeutic injection of human insulin and its pharmaceutical eﬀect [119]. Without insulin’s control of blood glucose levels, elevated blood glucose can eventually lead to diabetic ketoacidosis—a dangerous, potentially life-threatening condition [120]. To combat this, pharmaceutical companies have developed insulin analogs, protein therapeutics with one to three mutations in the human insulin amino acid sequence. These mutations destabilize the hexameric assembly of the insulin analog, allowing them to more rapidly dissociate into the pharmaceutically active monomer upon injection [121–123]—virtually eliminating the 30–60 minute lag time in the onset of pharmaceutical eﬀect seen with human insulin [119]. Pharmaceutical formulations of insulin analogs, however, need to remain stable

13 for their entire two year shelf-life. This means manufacturers have to walk a thin line between creating a stable formulation of insulin and ensuring a fast acting pharmaceutical eﬀect upon delivery to the patient. The latter is accomplished through the sequence mutations, the former through careful developmentofthepharmaceutical formulation. The phenolic preservatives, phenol and metacresol (m-cresol), are added to multi- dose insulin analog formulations to protect against potential bacterial contamination as a result of multiple needle insertions into the insulin analog product vial over the course of a typical diabetic’s daily treatment. However, as alluded to earlier, these phenolic compounds play a dual-role in insulin analog formulations, acting both as an antimicrobial agent and as a stabilizing excipient by promoting insulin assembly into the more stable hexameric form [124–126]. Since diabetics require both fast action of their insulin therapy and frequent dosing, wearable continuous infusion insulin pump devices—which deliver programmed amounts of insulin to the patient through a catheter and a cannula—can be a more convenient route of insulin analog administration for some patients. During use however, phenolic preservatives in the insulin analog formulation are absorbed into the plastics of the infusion pump catheter [127–129] and—in real-world conditions—can become nearly completely depleted from the formulation [127], as I observed during work detailed in Appendix B. When these stabilizing molecules are removed by surface-induced depletion, insulin analogs become less assembled and subsequently, less stable. My work in Chapter III elucidated the eﬀects on the overall assembly state and the thermal stability of three common insulin analogs when their formulations were depleted of phenolic preservatives. In Chapter IV, I explore the reverse case—damage done to a specialized surface by a pharmaceutical formulation. While tracking the degradation of the insulin analogs under conditions of depleted phenolic preservatives in Chapter III, I used

14 size-exclusion chromatography (SEC) to measure the formation of soluble protein aggregates. SEC separates molecules in solution by flowing them over a column packed with small, porous, silica beads. Smaller species can fit intothenooksandcrannies of the beads and, as a result, linger in the column longer than larger species which can only fit in the relatively large spaces between the beads. For a given sample flowed through the column, its size determines the volume accessible to it within the column. For polydisperse samples, species of different size elute at different times, allowing for the separation and quantification of different amounts of different sized molecules in the sample [130]. This type of separation is based on one fundamental assumption: that there are no interactions between the silica beads and the sample being analyzed. The silica surface consists of terminal hydroxyl groups, which under acidic pHs can hydrogen-bond with samples and, under basic conditions, can become charged by deprotonation. To mitigate this, manufacturers of SEC chromatography columns react various compounds with these hydroxyl groups to “end-cap” the silica resin with alternative chemistries that are more inert [131–133]. These end-caps are very resistant to reacting with the components in most samples, but as I found with therapeutic insulin analog samples, they are not compatible with all chemicals. During work done in Chapter III, I noticed changes in the elution patterns of control samples—used alongside experimental samples to verifycolumnperformance— over the course of the study. In consultation with several experts who had also observed this behavior, I initially hypothesized that the relatively high concentrations of the phenolic preservatives might be somehow accumulating on the surface of the silica beads, creating a much more hydrophobic surface than would otherwise be there. This might lead to additional hydrophobic interactions between the control samples and the silica, which would explain the changes I saw during elution. However, upon examining the silica beads directly, I found that their surface

15 chemistry had been chemically altered. Further investigation revealed that another component in some insulin analog formulations—zinc—reacted with the end-caps on the surface of the silica, permanently altering the surface chemistry and the column’s performance. This ﬁnding allowed me to propose a simple addition to the mobile phase that protected the surface from the zinc and completely inhibited its adverse eﬀects.

16 CHAPTER II

INVESTIGATION OF THE PHARMACEUTICAL COMPATIBILITY OF A NOVEL SURFACE FOR USE IN THERAPEUTIC PROTEIN DELIVERY DEVICES

Abstract

Pre-filled syringes are an important new device format for packaging and delivery of therapeutic proteins. Silicone oil lubrication is required for proper movement of the syringe plunger within the syringe. However, exposure of protein to the oil-water interface can enhance protein aggregation during agitation. Agitation stresses are inevitable in the transportation and use of a protein therapeutic over the course its shelf-life and it is therefore important to minimize agitation-induced aggregation. In this work, I examined the effects of agitating two protein therapeutics, intravenous immunoglobulin (IVIG) and Avastin, in contact with a novel fluoropolymer surface. The fluoropolymer surface provides “solid-phase” lubrication for the syringe plunger—obviating the need for silicone oil lubrication in pre-filled syringes. The aim of this work was to (1) examine the pharmaceutical compatibility of the fluoropolymer surface as a component in drug product primary container, and (2) to compare the effects of agitation stresses on two model proteins between the fluoropolymer surface and a typical siliconized surface. To test this, Gore provided vial stoppers and syringe plungers having a fluoropolymer surface for protein aggregation studies. I tested a vial stopper and a syringe plunger stopper using the fluoropolymer surface and agitated IVIG and Avastin in glass vials, and in pre-filled syringes in contact with either the fluoropolymer surfaced stopper or a siliconized stopper. I also examined the effects of two different buffers, phosphate buffered saline (PBS) and 0.2M glycine, both with and without the addition of the non-ionic surfactant, polysorbate

17 20, on agitation-induced aggregation of IVIG. Aggregation was monitored by tracking subvisible particle formation and soluble protein loss. Subvisible protein particle formation resulting from agitation in contact with the fluoropolymer surface was lower than the siliconized surface in both the vial and syringe configurations. The presence of polysorbate 20 in the formulation further reduced particle formation. No overall soluble protein loss was detected. Additionally, I measured adsorption of fluorescently labeled IVIG to the fluoropolymer surface in the presence and absence of polysorbate 20. With surfactant, adsorption of IVIG to the fluoropolymer surface was inhibited.

Introduction

Many protein-based therapeutic products are now being developed in pre-filled syringes [74, 75]. Pre-filled syringes offer a number of advantages to both patients and manufacturers, such as improved dosing accuracy and reduced drug waste due to overfilling [74, 75]. And, as a result, it is likely that more products will be developed in pre-filled syringes in the future [134, 135]. In order to properly function, glass, pre-filled syringes require lubrication to facilitate the smooth movement of the syringe plunger during the administration of the protein therapeutic to patients. Silicone oil—which provides the required lubrication in pre-filled syringes—has been shown to have deleterious effects on protein tertiary structure [89, 90]. And, adsorption to silicone oil (and to hydrophobic surfaces in general) alone may be sufficient to induce structural changes in the adsorbed protein [84, 85]. In some cases, the protein may undergo additional conformational changes after initial adsorption to a hydrophobic surface to maximize the amount of contact between the hydrophobic surface area within the molecule and the hydrophobic surface [51, 136]. Conformational perturbations of proteins adsorbed to the surface may also lead to increased protein-protein interactions [84], which in turn, can promote aggregation on the surface.

18 Once a protein is adsorbed to a hydrophobic surface, it has been previously suggested that simple desorption of the protein back into the solution is generally en- ergetically not possible [137]. However, it has been shown that proteins in solution can displace adsorbed proteins, exchanging them back into the solution [138–142]. In contrast, recent studies using single-molecule trackingtechniqueshaveshownthat adsorbed proteins dynamically and spontaneously desorb from surfaces [143], however the dynamics of desorption are complex and vary with surface-protein interactions [144], protein-protein interactions on the surface [145], and solution protein concentration [146]. Walder and Schwartz [146] found that BSA readily aggregated on the silicone oil-water interface and that increasing the bulk concentration of BSA significantly increased the surface aggregation rate. Walder and Schwartz [146] also found that larger aggregates had reduced rates of desorption compared to monomeric BSA. In pharmaceutical containers, agitation is thought accelerate the release of adsorbed protein aggregates back in to the bulk solution by physically disrupting the layer of absorbed protein that is possibly structurally perturbed and/or aggregated. Recently, Gerhardt et al. [95] have proposed a mechanism explaining how a protein layer adsorbed to the silicone oil-water interface can be disrupted by the air-water interface during agitation. When an air bubble comes in contact with a layer of proteins adsorbed to the silicone oil-water interface, it creates an intersection between the air-water, silicone oil-water and the silicone oil-air interfaces. This creates a force acting perpendicular and away from the surface that was estimated to be sufficient in magnitude to fragment the adsorbed protein layer leading to particle formation in the bulk solution [95]. Pre-filled syringes currently cannotbefilledwithoutanair bubble, so the mechanism proposed by Gerhardt et al. could potentially affect every protein formulation in a pre-filled syringe. Previous agitation studies have shown that agitation in concert with silicone oil interfaces further increases the production of aggregates in solution in protein thera-

19 peutics which is manifest both by sub-visible particle formation [25, 95, 96] and loss of soluble protein [26, 96]. Furthermore, Mehta et al. have showed that rupture of a protein layer adsorbed to the silicone oil-water interface markedly increased aggregation in the bulk solution [93]. Therefore agitation is an important and relevant stress to study in the development of protein therapeutics in pre-filled syringes. To minimize the effects of the silicone oil-water interface on protein aggregation during agitation, surfactants are often added to the formulation [25, 26, 91, 97]. Sur- factants, such as polysorbates, reduce the amount of protein adsorbed to the surface, presumably by out-competing the protein for adsorption sites on the silicone oil-water interface [98–100]. However, it has been shown in some cases that the presence of a surfactant is not sufficient to completely inhibit protein particle formation [25]. Fur- thermore, there is also concern that injection of silicone oil droplets in conjunction with therapeutic proteins may stimulate immunogenicity in patients [67, 147]. Another approach to avoid the problems caused by silicone oil is to use alternative materials and/or surfaces for the plunger and syringe that do not require an additional lubricant for the syringe to function—obviatingtheneedforsiliconeoil completely. Currently, there are two silicone oil-free, commercially available complete syringe systems for pharmaceuticals, both using polymer-based, plastic syringe barrels: PLAJEX⃝R syringes (Terumo Medical Corp., Somerset, NJ) and Crystal Zenith⃝R syringes (CZ-resin) using FluroTec⃝R plungers (Daikyo Seiko, Ltd., Tokyo, Japan). Additionally, there is another fluoropolymer coated plunger called Omniflex⃝R (Datwyler, Schattdorf, Switzerland), that is commerciallyavailable[148].Tomy knowledge, none of these materials are currently used in pre-filled syringes for a commercially marketed therapeutic protein product. In 2014, Krayukhina et al. compared the pharmaceutical compatibility of two model therapeutics with silicone oil-free glass syringes, PLAJEX plastic syringes, and siliconized syringes after agitation [96]. Using a Fc-fusion protein and a monoclonal

20 antibody as model proteins, they found that protein stability, as measured by soluble protein aggregate formation and subvisible particles formation, was improved with both the bare-glass syringes and plastic syringes as compared to siliconized syringes, both glass and PLAJEX plastic [96]. To my knowledge, there are nopublishedreports on the protein compatibility or agitation studies in pre-filled FluroTec⃝R /CZ-resin syringes with protein therapeutics. No compatibility nor agitation stress studies with protein therapeutics could be found in the literature for the Omniflex⃝R plunger either, further highlighting the need for more research on alternative materials for use in delivery devices for protein therapeutics. In this work, I examined a plunger developed by W. L. Gore and Associates having a novel fluoropolymer surface that eliminates the need for silicone oil lubrication. The fluoropolymer surface provides both solid-phaselubrication,tofacilitate the movement of the syringe plunger against a glass syringe barrel during delivery to the patient, and a barrier between the rubber stopper and the pharmaceutical formulation. My aim is to understand if the fluoropolymer surface in conjunction with a silicone oil-free glass syringe impacts protein aggregation as compared to traditional siliconized pre-filled syringe. For the initial investigation of the pharmaceutical compatibility of this new interface, I examined a vial closure system using the same fluoropolymer surface material as is used in syringe plungers. The vial configuration offered us two distinct experimental advantages over syringes. First, the vial configuration allowed for an increased rate of rotation in comparison to a syringe. This increased the agitation stress applied to the protein formulation, and subsequently, the rate of protein aggregation. Second, it allowed us to better normalize the surface-water contact area between the fluoropolymer surface and the siliconized surface. The fluoropolymer-surfaced vial stopper was designed to have a solution exposed surface area of 2.95cm2 which was similar to the surface area of the commercially available siliconized rubber stoppers

21 (3.08cm2). Matching surface areas between the fluoropolymer surface and siliconized surfaces is not possible in a typical siliconized syringe system because the entire barrel of the syringe—in addition to the plunger—is siliconized. In the syringe with the fluoropolymer coated plunger, the entire syringe barrel is unsiliconized, and only the front face of the fluoropolymer coated plunger is exposed to the protein formulation. Even with the elimination of silicone oil from the pre-filled syringe system, two other surfaces may facilitate protein aggregation: the fluoropolymer surface and the bare-glass syringe wall. Similar to a silicone oil surface, the fluoropolymer surface is hydrophobic. Protein adsorption to the fluoropolymer surface may result in aggregation, especially during agitation, as has been found with other fluoropolymer surfaces [52–55, 149]. In this study, I investigated a model protein, intravenous immunoglobulin (IVIG), in a vial system, a typical siliconized syringe system and a bare-glass syringe with the fluoropolymer surfaced plunger. The effects of buffer type, pH and the addition of polysorbate 20 were studied. My hypothesis was that aggregation occurring due to the agitation in the presence of the fluropolymer surface would be inhibited by polysorbate 20, due to the reduction of protein adsorption tothesurfaceinthepres- ence of surfactant [52, 116, 150]. I also compared the stability of a commercially available monoclonal antibody, Avastin (bevacizumab), in the bare-glass syringe with the fluoropolymer surfaced plunger to its stability in a traditional siliconized pre-filled syringe in an attempt to demonstrate real-world applicability of the fluoropolymer syringe system.

Materials and Methods

Materials

All laboratory chemicals used were of analytical or higher grade. Deionized water was puriﬁed through a Millipore Synergy UV (Billerica, MA, subsequently referred

22 to as “MilliQ” water) filtration unit and was used in mobile phases, formulations and buffers. Intravenous Immunoglobulin (Gammagard, Baxter International Inc., Deer- field, IL, referred to subsequently as IVIG) and Avastin (bevacizumab, Genentech, San Francisco, CA) were purchased from local pharmacies and used before their ex- piration dates. The fluoropolymer surfaced vial stoppers and syringe plungers were provided by W. L. Gore and Associates. Siliconized stoppers, siliconized syringe plungers and syringes were also provided by W. L. Gore and Associates.

Agitation in Vials

IVIG was diluted 100-fold from the commercial pharmaceutical formulation (100mg/mL, 0.25M glycine, pH 4.25) to 1 mg/mL using either 0.2M glycine buffer (pH 4.25) or phosphate buffered saline (PBS; pH 7.4) which consisted of 137mM sodium chloride, 3mM potassium chloride, 10mM anhydrous dibasic sodium phosphate and 2mM monopotassium phosphate. All buffers were filtered through a 0.22um bottle-top filter (EMD Millipore, Billerica, MA) before use. For all buffers, pH was adjusted using 6.25M hydrochloric acid or 10M sodium hydroxide as needed. Avastin was used directly from the product vial (25 mg/mL) and diluted to 1 mg/mL with its placebo formulation which consisted of 60 g/L α, α-trehalose dihy- drate, 5.8 g/L monobasic sodium phosphate monohydrate, 1.2 g/L anhydrous dibasic sodium phosphate, and 0.4 g/L polysorbate 20 (Tween 20), at pH6.2[151]. Fiolax (6mL size, Part VCDIN6R) clear glass vials (Schott AG, Mullheim, Ger- many) were filled with 3mL of sample, capped with either a siliconized vial stopper or a custom fluoropolymer-surfaced stopper (Figure 1), and secured with an aluminium crimp. Vials were placed in a cardboard freezer box which was rotated end-over-end on a Rotamix test tube rotator (Appropriate Technical Resources Inc., Laurel, MD) at 40 RPM. Individual sample vials were removed at 0, 2, 4 and 24hours.Triplicate vials were prepared for each sample condition and each time point unless noted.

23 Stoppers StoppersinVials

A B

Figure 1: Stoppers used in during vial agitation (A) and assembled vials with stoppers (B). In each panel A and B shows the the Gore ﬂuoropolymer stopper (left) and a typical siliconized stopper (right).

A glass-only vial system was constructed by placing a standard glass microscope slide over the vial opening. A rubber washer was then placed on top of the slide, and aligned with the vial opening underneath to evenly distribute pressure when clamped to the agitator. The microscope slide and the vial were then clamped together on a custom-designed attachment for the test tube rotator (Figure2).Asamplevolume of 3mL was pipetted into the vial, the vials were then sealed with the glass slide, and placed on the clamp/rotator setup and agitated end-over-end at 40 RPM. Individual sample vials were removed at 0, 2, 4 and 24 hours. Triplicate vials were prepared for each sample condition and each time point unless otherwise noted. All samples were protected from light during agitation by placing the agitator underneath a cardboard box.

Agitation in Syringes

Syringes were used in two conﬁgurations: bare-glass syringebarrelswiththeﬂuo- ropolymer plunger and siliconized syringe barrels with siliconized plungers. Both the bare-glass and the siliconized syringes were 1mL, “long-type,” type I borosilicate

24 Figure 2: Custom glass-only system clamp schematic (left) andphotoofvials clamped to the test tube rotator (right). glass syringes with staked needles, both syringes had identical overall dimensions (as per ISO 11040-4). Syringes were placed with the capped needlefacingdownand were filled with 1.2mL of protein solution. A fluoropolymer-surfaced plunger (for the bare-glass syringes) or a siliconized plunger (for the siliconized syringes) was inserted such that the base of the plunger was flush with the flange on the base of the syringe barrel (Figure 3), leaving an approximately 300-400uL air bubble. After insertion of the plunger, the syringes were oriented with the needle up and the needle cap was removed to relieve pressure built up by the insertion of the plunger. The syringe needle was then recapped and the syringe was clipped to the rotator at the approximate midpoint of the syringe’s overall length and agitated end-over-end at 20 RPM on a Rotamix test tube rotator (Appropriate Technical Resources Inc., Laurel, MD). Dur- ing one full revolution, the air bubble travelled the entire length of the syringe barrel. All syringe samples were protected from light during agitation in the same manner as was the vials. Individual sample syringes were removed at 0, 1, 3 and 7 days. Trip- licate syringes were prepared for each sample condition and each time point unless

25 otherwise noted. Samples were collected from the syringes byexpellingthecontents of the syringe through the needle into a 1.5mL, polypropylene microcentrifuge tube (Thermo Fisher Scientiﬁc Inc., Pittsburg, PA).

Plungers Plungers in Syringes

A B

Figure 3: Syringe plungers (A) and assembled, filled syringes (B). In panel A, the fluoropolymer plunger is picutred on the left and a standard siliconized plunger on the right. In panel B, a bare-glass syringe barrel with the fluoropolymer plunger (left) and a traditional siliconized syringe and syringe plunger (right).

Micro-Flow Imaging (MFI)

Total particle concentration was obtained for particles from 1 to 150 micron in size using a DPA4200 Micro-Flow Imager (MFI, ProteinSimple, Santa Clara, CA). A sample of 0.45 mL was flowed through the system, and the total analysis volume was 0.16mL. System suitability was ensured between each sample set by flowing particle- free water, purified through a Millipore Synergy UV (Billerica, MA) purifier, through the system. The system was deemed suitable if total particle counts in the particle- free water were less than 500 particles per mL. Between sampleruns,theflowcell was rinsed with the sample buffer, and if needed, the cells werecleanedwitha1% (v/v) Tergazyme solution (Alconox, White Plains, NY) or a 1% (v/v) Hellmanex II solution (Hellma Analytics, Mullheim, Germany). After cleaning, the flow cell was rinsed extensively with water and then with sample buffer. Data were exported from the MFI’s MVSS software and analyzed using custom computer scripts written in

26 Perl.

Size Exclusion Chromatography (SEC)

Total soluble protein and soluble protein aggregates were quantified by SEC using an Acquity UPLC BEH200 SEC column (4.6mm x 150mm, Waters Corporation, Mil- ford, MA) on an Agilent 1100/1200 HLPC (Agilent Technologies, Santa Clara, CA). The mobile phase consisted of 10mM sodium phosphate and 300mMsodiumchloride at a pH 7.4. The mobile phase was filtered through a 0.22um filter(EMDMilli- pore, Billerica, MA) and degassed before being used at a flow rate of 0.3mL/min. Absorbance was monitored at 280nm for the entire 12 minute per sample runtime. Samples were centrifuged at ∼ 9000g for 10 minutes to remove any large, insoluble aggregates before injection onto the column. The sample injection volume was 10µL. Data were exported from Agilent’s Chemstation software and baseline corrected, integrated and analyzed for high molecular weight species and total recovered protein mass with custom computer scripts written in Matlab (Mathworks, Natick, MA).

Confocal Microscopy of the Adsorption of Fluorescently Labeled IVIG to the Fluo- ropolymer Surface

IVIG was fluorescently labeled which was labeled with AlexaFluor488SDPfollowing the manufacturer’s protocol. IVIG (10 mg/mL) in PBS was incubated with 10mg/mL of AlexaFluor for one hour. The labeled protein was purified awayfromfreedyeby buffer exchange using Amicon 30kDa molecular weight cut off centrifugal filters (EMD Millipore, Billerica, MA) with at least 12 volume exchanges. The final dye-protein molar ratio was determined to be 1.9 by UV spectroscopy (data not shown). Fluoropolymer surface samples on standard microscope slideswereprovidedby W. L. Gore and Associates. Siliconized surface samples were prepared by applying SurfaSil to standard microscope slides using the manufacturer’s “wipe-on” method

27 [152]. Briefly, concentrated SurfaSil was pipetted on to the microscope slide, wiped to cover the entire slide and was then allowed to react with theglassslidefor30 seconds. Following the reaction time, the microscope slide was rinsed with excess hexanes and the reaction was then quenched by rinsing with excess methanol. The siliconized slides were allowed to dry for 24 hours at room temperature before use. Fluorescently labeled IVIG was prepared at a concentration of 185nM in PBS (pH 7.4), and in 0.2M glycine (pH 4.25), both with and without the inclusion of 0.02% (v/v) polysorbate 20. An aliquot of 20uL of labeled IVIG was pipetted onto the surface. The solution droplet on the surface was then covered with a coverslip to ensure proper spreading, and the system was incubated at room temperature for 60 minutes. The coverslip was removed, and the surface was rinsed three times with the corresponding buffer. Rinsing was accomplished by pipetting 50uL of buffer onto the position where the protein solution had been placed, placing a new coverslip over the drop and gently spreading the buffer to ensure coverage of the sample area. The coverslip was then removed and the process was repeated a total of three times. Finally, 20uL of buffer was pipetted onto the surface, and the drop was covered with acoverslip. For some protein samples that were applied to the surface in the absence of polysorbate 20, a rinse buffer containing polysorbate 20 was used. This approach was done to compare the effects of rinsing in the presence or absence of the surfactant. The rinse procedure was identical to the procedure above with the following deviations: labeled IVIG was pipetted onto the surface in buffer without polysorbate 20, and the rinse steps were completed using the corresponding buffer with polysorbate 20. Prior to examination on the microscope, the edges of the coverslip were sealed with clear nail polish to prevent coverslip movement and evaporation of the sample. Samples were imaged on a Zeiss LSM 780 confocal microscope (Carl Zeiss Microscopy

28 GmbH, Jena, Germany) using the 488nm laser for fluorescence excitation. The surface was located using a 1AU (900nm z-plane thickness) confocal pinhole opening. The pinhole was then completely opened (14.7um z-plane thickness) to account for variations on the order of 2–4um in the surface topology. Three areas of the sample region where the protein was applied were randomly selected and imaged. Images were exported from the instrument and total fluorescent density was quantified using the Fiji software package (ImageJ version 2.0.0-rc-24/1.49m) [153]. To determine if either the fluoropolymer or the siliconized surfaces themselves were fluorescent at my excitation wavelength, I prepared a sample using only PBS buffer loaded on the surface. There was no fluorescence from either surface alone (data not shown).

Determination of the Second Osmotic Viral Coeﬃcient (B22)

Second osmotic virial coeﬃcients were calculated from static light scattering data obtained with a Brookhaven Research Goniometer light scattering system (Brookhaven Instruments Corp., Holtsville, NY). Samples were measured at 0.1, 0.5, 1, 2, 3, 4 and 5 mg/mL using 90◦ light scattering. Data were ﬁt to the virial expansion of the osmotic pressure equation using the Brookhaven instrument software:

K 1 c = +2B c (1) R M 22

◦ Where Kc is the optical constant, R is the excess Rayleigh ratio at 90 , M is the molecular weight of the protein, c is the protein concentration and B22 is the second osmotic virial coeﬃcient.

29 IVIG Unfolding with Urea

IVIG was diluted to 1 mg/mL in PBS or glycine buffer made with ureaconcentrations ranging from zero to 8.9M. Final urea concentration in each buffer was determined by refractometry. Samples were allowed to equilibrate at room temperature overnight, protected from light before measurement. Intrinsic tryptophan fluorescence was measured on a PTI fluorescence spectrometer (Photon Technology International, Inc., Edison, NJ) using an excitation wavelength of 295nm and recording emission spectra from 305 to 400nm. Both excitation and emission slit widths were set to 2nm. Data were exported from the instrument and analyzed for emission center of spectral mass position [154] in Matlab (MathWorks, Natick, MA) using raw emission values. Sig- moidal fits of the raw center of mass emission values versus urea concentration were calculated by fitting the data to the equation

CM(x)=h(1 + exp [−f(x − xm)]) + b (2)

Where CM is the center of mass of the fluorescence emission spectrum as afunction of urea concentration x, f is a sigmoidal shape parameter, b is the center of mass of the fluorescence emission spectrum for the folded protein, h is the spectral distance between the center of mass of the fluorescence emission spectrum for the folded protein

and the unfolded protein and xm is the midpoint of the unfolding curve.

Results and Discussion

Many previous studies with hydrophobic surfaces have found that, in general, adsorption of protein molecules onto these surfaces results in perturbations to the structure of the protein [47–51] and results protein aggregation [52–55]. Of particular interest—in the context of this work—is protein interactions with polytetraﬂuoroethy- lene (PTFE), which is chemically similar to the proprietary ﬂuoropolymer surface

30 tested here. Vermeer et al. [113, 114] found that absorption of antibody molecules to PTFE surfaces resulted in perturbation of the antibody’s secondary structure. Similarly, work on insulin analogs adsorbed to PTFE surfaces also showed structural perturbations in both secondary and tertiary structure [115, 155]. Such alteration of structure of adsorbed protein molecules has been shown to be associated with increased rates of aggregation. With PTFE in particular, the structural perturbations in combination with agitation have been found to enhance aggregation detected in the solution phase that is in contact with the surface. Examples of increases in protein aggregation when agitated with PTFE surfaces include: the aggregation of a monoclonal antibody [52], the inactivation and aggregation of lysozyme [53], and increased amyloid fibril formation rate in recombinant insulin analogs[54,55]. As with silicone oil-water interfaces [25, 26, 89, 156], it hasbeenshownthat agitation-induced protein aggregation due to contact with PTFE surfaces can be inhibited by the inclusion of non-ionic surfactants [52, 116], such as polysorbate 20. A detailed discussion of the proposed mechanisms for how polysorbates inhibit surface- induced protein aggregation is presented later. However, theobservationthatsurfac- tants could inhibit agitation-induced aggregation of protein exposed to PTFE surfaces motivated us to examine effects of agitating IVIG in the presence and absence of polysorbate 20 when the protein solution was in contact with the fluoropolymer surface. In addition to the inhibitory effects of surfactants on agitation-induced aggregation, a number of studies have also found that agitation-induced protein aggregation in the presence of silicone oil surfaces could be affected by the formulation’s solution conditions [25, 26, 90]. Basu et al. [25] and Thirumangalathu et al. [26] found that, with the antibodies they studied, protein aggregation resulting from agitation in contact with a siliconized surface was much faster when the formulation was adjusted to pH 7.4 compared to pH 5.0. At pH 7.4, Basu et al. [25] found that the antibody they

31 studied was less colloidally stable than at pH 5.0. Basu et al. also observed increased aggregation during agitation in the presence of 150mM NaCl which, they found in their study, further reduced the overall colloidal stability of their antibody. Thiru- mangalathu et al. [26] also found that agitation of their antibody at pH 7.4 resulted in more rapid protein aggregation than at pH 5.0 and that the addition of salt accelerated monomer loss at pH 5.0. Both studies found that surface-induced aggregation was reduced when the colloidal stability of the monoclonal antibody in solution was greatest. For both Basu et al. and Thirumangalathu et al., theformulationwiththe greatest colloidal stability—and lowest rates of agitation-induced aggregation—for their respective proteins their studies occurred in the pH 5.0 buffer with low ionic strength. In contrast, Gerheardt et al. [90] found that—for their specific antibody— increasing the ionic strength of their formulation resulted in decreased aggregation resulting from agitation in contact with a siliconized surface. Results from these studies demonstrate that agitation-induced protein aggregation can be affected by the protein’s solution stability and that each protein is unique in its solution stability and susceptibility to agitation stresses. As with the effects of surfactants on the inhibition of agitation-induced protein aggregation, these observations motivated me to investigate the effects of agitation in the presence of the fluoropolymer surface with two different buffers for IVIG, glycine (pH 4.25, ionic strength 2.8mM) and PBS (pH 7.4, ionic strength 139.6mM). Given the results of the previous studies examining agitation-induced protein aggregation in contact with PTFE and silicone oil surfaces, my aims for this study were: (1) to assess differences in the agitation-induced aggregation of IVIG in contact with the fluoropolymer surface in different two buffers, glycine and PBS, (2) to test the hypothesis that the addition of polysorbate 20 to both of these buffers would inhibit agitation-induced aggregation of IVIG in contact with the fluoropolymer surface, (3) to compare the overall agitation-induced protein aggregation of IVIG between the

32 ﬂuoropolymer surface and a typical siliconized surface and, (4) to compare aggregation propensity of a commercial monoclonal antibody, Avastin, during agitation in contact with bare-glass syringe with a ﬂuoropolymer-surfaced syringe plunger versus agitation in contact with a typical siliconized syringe. Because I will be comparing agitation-induced protein aggregation in both vials and syringes in this manuscript, for clarity, I will refer to the vial closure as a “stopper” and the syringe plunger stopper as a “plunger.”

Increased Agitation Stress in Vials

Agitation is a significant stress that protein therapeutics encounter in the many stages of production, purification, transportation, storage delivery to patients [56–59]. Agi- tation alone has been shown to accelerate protein aggregation rates [99, 102, 116, 157]. Aggregation has been found to be further increased when proteins have been agitated in the presence of other surfaces such as PTFE [52–55] or silicone oil [25, 26, 89, 156]. Therefore agitation, even agitation more intense than that occurring in pre-filled syringes, was an important stress to investigate with the fluoropolymer surface used in the current study. To increase the intensity of the agitation stress by increasing the rate of interfacial exchange, I used a vial closure, shown in Figure 1, made of the same fluoropolymer surface used for the syringe plunger. As compared to agitation in syringes, vials could be rotated at a greater rate. Increasing the rate of rotation, in combination with the geometry of the vials, increased rate the air-water interface repeatedly contacted the stopper surface. Based on the mechanism proposed by Gerhardt et al. [95], this is expected to contribute to greater disruption of protein layers adsorbed to the silicone oil, fluoropolymer or glass surfaces and subsequently lead to accelerated protein aggregation in vials compared to syringes. Agitation in syringes, in contrast to agitation in vials, is limited in the rate of interfacial exchange by the mobility

33 of the air bubble within the syringe. Therefore, agitation in vials allowed us to substantially enhance the eﬀect of the interfaces on protein aggregation measured in the bulk solution.

Eﬀect of Buﬀer Type on Aggregation of IVIG During Agitation in Vials

With the vial system, I tested three different configurations: 1) stoppers with the fluoropolymer surface; 2) siliconized rubber stoppers; and 3) the glass-only system. In the absence of agitation, during quiescent incubation forupto24hoursatroom temperature, there was no detectable formation of protein particles nor loss of soluble protein (data not shown). Before discussing the results, a brief note on the particle counts as measured by the MFI. Several numerical techniques [158–160] have been proposed to mathematically differentiate silicone oil droplets from proteins particlesbasedonthemorphological parameters of each particle as reported by the MFI. I initiallyappliedthenumerical technique proposed by Weinbuch et al. [160] to my data, but found that, in some samples, a substantial fraction of particles were misclassified. In addition, Shomali et al. [67] and Chisholm et al. [147] have recently demonstrated that protein adsorbed to silicone oil droplets may have an adjuvant-like effect in eliciting immune responses to therapeutic proteins. Therefore, from the perspective of eliciting an immune response in patients, it appears that it may not matter whether the particles are purely proteinaceous or heterogeneic particle consisting of protein adsorbed to silicone oil droplets. In considering the findings of Shomali et al. and Chisholm et al., combined with the unknown accuracy of particle differentiation techniques, I decided to report the particle concentration directly from the MFI without any additional processing.

34 Agitation of IVIG in Glycine Buﬀer

MFI data for IVIG, agitated in glycine buffer, in vials with the fluoropolymer surface, a siliconized rubber stopper and the glass-only surface are shown in Figure 4. With the siliconized stopper, I observed considerable particle formation over the time course of agitation. In contrast, no significant particle formation was detected in the samples agitated in the presence of the fluoropolymer surface, and only minimal particle formation was observed for samples in the glass-only system. Particle formation in the vials agitated with the siliconized stoppers was greatly reduced with the inclusion of 0.02% polysorbate 20 in the formulation. This result is consistent with previous findings by others that have shown inhibition ofproteinaggregationby polysorbate—likely by preventing adsorption of the protein to the silicone oil-water interface [26, 156].

500,000 Gore, Gly Gore, Gly+PS20 400,000 Siliconized, Gly Siliconized, Gly+PS20 300,000 Glass, Gly Glass, Gly+PS20 200,000

100,000

Particle Concentration (#/mL) 0

0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 24 24 24 24 24 24 Agitation Time (hrs)

Figure 4: IVIG agitated in vials, in 0.2M Glycine buﬀer (pH 4.25), with the experimental ﬂuoropolymer-based surface (Gore), a traditional siliconized surface and a glass surface as measured by MFI. Data areplotted as mean ± standard deviation of three independent replicates except for siliconized samples on hour 24 both with and without surfactant, which are plotted as the mean and range of two independent replicates.

By SEC analysis (Figure 5), there were small and variable amounts of soluble protein loss detected in the samples agitated in the presenceoftheﬂuoropolymerand

35 in samples agitated in the presence of the siliconized surface. No soluble protein loss was detected in the glass-only system. No soluble protein aggregates were observed for any of these samples at any time point during the agitation study (data not shown). The presence of polysorbate 20 inhibited virtually all soluble protein loss in the samples agitated in the presence of the ﬂuoropolymer and the siliconized surfaces, again in agreement with previous studies [26].

100% Gore Gly Gore Gly+PS20 Siliconized Gly 75% Siliconized Gly+PS20 Glass Gly 50% Glass Gly+PS20

25% Percent Mass Recovery 0%

0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 24 24 24 24 24 24 Agitation Time (hrs)

Figure 5: Total soluble protein mass recovery as measured by SEC for IVIG agitated in vials with the experimental ﬂuoropolymer-based surface (Gore), a traditional siliconized surface and a glass surface in 0.2M glycine buﬀer (pH 4.25). Data are plotted as mean ± standard deviation of three independent replicates. No high molecular weight species were observed for any of the samples. Loss of total soluble protein and considerable vial- to-vial variability was observed in samples agitated without the inclusion of surfactant.

Agitation of IVIG in PBS Buﬀer

In investigating agitation-induced protein aggregation with silicone oil surfaces, Thiru- mangalathu et al. [26] and Basu et al. [25] both made an interesting observation that solution pH affected agitation-induced protein aggregation. In both cases, at pH 7.4—closer to the antibody’s isoelectric point—the aggregation rate for the antibody was much greater than that observed at pH 5.0. For the current study, this motivated the question: what is the effect of the fluoropolymer surface when agitation occurs in

36 suboptimal formulation buffer conditions? The commercial IVIG product is formulated at pH 4.25 in glycine buffer. For the suboptimal buffer condition, I studied the effects of agitation in PBS buffer, with and without the inclusion of polysorbate 20. By MFI particle counts (Figure 6) I observed much greater levelsofparticlefor- mation for IVIG agitated in the PBS buffer in comparison to agitation in the glycine buffer—especially note the differences in the ordinate axis scale between Figure 6 and Figure 4. The buffer effect was observed for IVIG samples agitated in the presence of all three surfaces. However, with all three surfaces—even when IVIG was formulated in this suboptimal buffer and pH—the addition of polysorbate 20tothePBSbuffer significantly reduced the amount of particle formation I observed.

2,500,000 Gore, PBS Gore, PBS+PS20 2,000,000 Siliconized, PBS Siliconized, PBS+PS20 1,500,000 Glass, PBS Glass, PBS+PS20 1,000,000

500,000

Particle Concentration (#/mL) 0

0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 24 24 24 24 24 24 Agitation Time (hrs)

Figure 6: IVIG agitated in vials, in PBS buﬀer (pH 7.4), with the experimental ﬂuoropolymer-based surface (Gore), a traditional siliconized surface and a glass surface as measured by MFI. Data are plotted as mean ± standard deviation of three independent replicates.

Analysis by SEC (Figure 7) showed considerable loss of soluble protein when agitated in contact with the ﬂuoropolymer surface. Agitation of IVIG in contact with the siliconized surface also showed loss of soluble protein, however, less than what was observed with the ﬂuoropolymer surface. For both surfaces, when polysorbate 20 was included in the formulation, no soluble protein loss was observed. In the glass-only system, I observed soluble protein loss that increased with increasing agitation time

37 and was unaﬀected by the addition of polysorbate to the formulation.

100% Gore PBS Gore PBS+PS20 75% Siliconized PBS Siliconized PBS+PS20 50% Glass PBS Glass PBS+PS20

25% Percent Mass Recovery 0%

0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 24 24 24 24 24 24 Agitation Time (hrs)

Figure 7: Total soluble protein mass recovery as measured by SEC for IVIG agitated in vials with the experimental ﬂuoropolymer-based surface (Gore), a traditional siliconized surface and a glass surface in PBS buﬀer (pH 7.4). Data are plotted as mean ± standard deviation of three independent replicates. No high molecular weight species were observed for any of the samples. The addition of the surfactant, polysorbate 20, to the formulation inhibited the loss of soluble protein.

Protein adsorption to the silicone oil-water interface has been shown to be inhibited by polysorbate [98–100]. Similarly, polysorbate has been shown to inhibit the adsorption of protein to glass surfaces [161]. Proteins havealsobeenshowntoadsorb to PTFE surfaces [52–55] suggesting that in my system, polysorbate could potentially be inhibiting the adsorption of IVIG to the fluoropolymer surface as well. This hypothesis is tested below. Also, I hypothesized that the buffer effect I observed may be due to differing conformational and/or colloidal stabilities of IVIG in the PBS buffer, as had been observed by Basu et al. [25] and Thirumangalathu etal.[26]during agitation of monoclonal antibodies with siliconized surfaces. This hypothesis too, is tested below.

Agitation of Avastin

During the agitation studies of IVIG in vials, I observed that pH, the contact surface and polysorbate 20 all had an eﬀect on IVIG’s susceptibility toagitation-induced

38 aggregation. If I was to test a well-formulated, commercially available therapeutic monoclonal antibody product, what effect would agitation in contact with the fluoropolymer surface have on protein aggregation? And, in the context of silicone oil-free delivery devices, how would the results compare to those obtained with Avastin agitated in a traditional siliconized syringe and the all-glass vial system? To address these questions, I studied the agitation of Avastin at a protein concentration of 1 mg/mL in its commercial formulation. By MFI (Figure 8), agitation of Avastin showed no significant particle formation over the course of the study when in contact with the fluoropolymer surface or the glass-only system. However, I did observe an increase in particle formation over the course of the study during agitation with the siliconized stopper. This could be a combination of protein particle formation and silicone oil droplets being shed from the siliconized stopper surface, similar to what I observed with IVIG agitated with the siliconized stopper. No soluble protein loss was detected by SEC for Avastin agitated in vials with any of the three surfaces either (Figure9). Avastin’s formulation includes polysorbate 20, which has been shown to prevent surface-induced aggregation by inhibiting protein adsorption to interfaces [98–100]. Presumably, the inclusion polysorbate in Avastin’s formulation also explains why Iobservedlittletonoparticleformationduringagitation. Additionally, Avastin’s isoelectric point is 8.6 [162]. At its formulation pH, 6.2, Avastin is positively charged. It is reasonable to assume that the level of charge at this pH provides for sufficient intermolecular electrostatic repulsion which, in part, provides Avastin with sufficient colloidal stability in formulation. Avastin’s formulation also contains trehalose—a potent, conformation-stabilizing sugar [163].

39 100,000 Gore Siliconized 80,000 Glass

60,000

40,000

20,000

Particle Concentration (#/mL) 0

0 2 4 0 2 4 0 2 4 24 24 24 Agitation Time (days)

Figure 8: Avastin agitated in vials with the experimental fluoropolymer- based surface (Gore), a traditional siliconized surface and a glass surface as measured by MFI. Data are plotted as mean ± standard deviation of three independent replicates. Particle formation was detected in the siliconized system, however, no significant particle formation was detected during the course of agitation for either the fluoropolymer surface or the glass-only system.

100% Gore Siliconized 75% Glass

50%

25% Percent Mass Recovery 0% 0 2 4 0 2 4 0 24 24 24 Agitation Time (days)

Figure 9: Total soluble protein mass recovery as measured by SEC for Avastin agitated in vials with the experimental ﬂuoropolymer-based surface (Gore), a traditional siliconized surface and a glass surface. Data are plotted as mean ± standard deviation of three independent replicates. No high molecular weight species were observed and no total soluble protein mass loss was detected during the duration of agitation.

40 Agitation in Syringes

The vial format used for the above experiments provided a convenient and efficient way to increase agitation stress in comparison to a syringe format. However, the ultimate application for the fluoropolymer surface is as a solid-phase lubricant and plunger barrier in a silicone oil-free syringe. Also, compared to a pre-filled syringe, only a minimal amount of silicone oil is required for siliconized vial stoppers—just enough to ensure lubrication for machine handling during processing and insertion into the vial [164]. Therefore, it was important to examine the effects of fluoropolymer surface on protein stability in a pre-filled syringe system and compare the results to those obtained during agitation in traditional siliconized syringes. My experimental set up used 1mL “long-type” glass syringes with staked needles. In the photo in Figure 3, panel B, the syringe on the left in is an unsiliconized, bare-glass syringe barrel with the fluoropolymer surfaced plunger. The traditional siliconized syringe on the right consists of the identical syringe barrel that hasbeensiliconizedanda siliconized rubber plunger.

Agitation of IVIG in Glycine Buﬀer

When agitated in contact with the fluoropolymer surface, in the syringe system IVIG in glycine buffer produced only a small number of particles as measured by MFI (Figure 10) on days 3 and 7. No particles were not observed for samples containing polysorbate when agitated with the fluoropolymer surface. However, when samples were agitated in the siliconized syringe, IVIG formed significant levels of protein particles during agitation as measured by MFI. In the presence of polysorbate 20, there was still substantial levels of particles observed (Fig- ures 10). Also noted was that samples agitated in siliconized syringes with polysorbate 20 in the formulation showed significant syringe-to-syringevariabilityondaythree. A substantially larger number of particles were observed in samples agitated in

41 the the siliconized syringe, again suggesting that, in addition to protein aggregation, silicone oil droplets may be shed from the syringe barrel and siliconized plunger. When polysorbate 20 was included in the formulation, I observed an increase in the particle concentration on day three in samples agitated in the siliconized syringe, but this increase was not statistically significant. However, polysorbate has been shown to affect the formation and stability of silicone oil emulsions. Felsovalyi et al. have shown that polysorbate 80 increased migration of silicone oil from the barrel walls of siliconized syringes into the bulk solution by decreasing the surface tension between the silicone oil and water layers [106]. Polysorbate 20 has also been shown to stabilize silicone oil emulsions [97] by steric inhibition of droplet coalescence [107]. Finally, I observed no soluble protein loss by SEC for IVIG agitated in either the siliconized syringe or the fluoropolymer syringe (Figure 11). The fact that I observed no soluble protein loss by SEC, but observed particle formation by MFI further highlights the importance of particle counting as an extremely sensitive measure of detecting protein aggregation [62].

Agitation of IVIG in PBS Buﬀer

Agitation of IVIG in the PBS buffer in contact with the fluoropolymer surface led to a small amount of particle formation on day three by MFI (Figure 12). However, the addition of polysorbate 20 virtually eliminated the formation of particles generated from agitation in contact with the fluoropolymer surface. In siliconized syringes, I observed the formation of particles by MFI (Figure 12). Particle formation for samples agitated in the siliconized syringe were not reduced with the addition of polysorbate 20. The concentration of silicone oil droplets was greater in samples including polysorbate 20 in the formulation. As discussed above, this is likely due to the solublizing and emulsion stabilizing effects of polysorbate. Particle concentrations for samples agitated in the siliconized syringes with and

42 1,500,000 Gore, Gly Gore, Gly+PS20 Siliconized, Gly 1,000,000 Siliconized, Gly+PS20

500,000

Particle Concentration (#/mL) 0 0 1 3 7 0 1 3 7 0 1 3 7 0 1 3 7 Agitation Time (days)

Figure 10: Particle formation measured by MFI for IVIG agitated in 0.2M glycine buﬀer (pH 4.25), in a bare-glass syringe with a ﬂuoropoly- mer coated plunger (Gore) and in a traditional siliconized syringe with asiliconizedrubberplunger.Dataareplottedasmean± standard deviation of three independent replicates except for the siliconized sample without polysorbate 20 on day 7, which is plotted as the mean and range of two independent replicates.

100% Gore Gly Gore Gly+PS20 75% Siliconized Gly Siliconized Gly+PS20 50%

25% Percent Mass Recovery 0% 0 1 3 7 0 1 3 7 0 1 3 7 0 1 3 7 Agitation Time (days)

Figure 11: Total soluble protein mass recovery as measured by SEC for IVIG agitated in in a bare-glass syringe with a ﬂuoropolymer surface coated plunger and in a traditional siliconized syringe with a siliconized rubber plunger in 0.2M glycine buﬀer (pH 4.25). Data are plotted as mean ± standard deviation of three independent replicates except for the siliconized sample without polysorbate 20 on day seven, which is plotted as the mean and range of two independent replicates.

43 without polysorbate 20 declined slightly from day three to day seven of agitation. However, neither decrease was statistically signiﬁcant. I did, however, observe visible particles in some of the siliconized syringe samples at this time point. This observation may help explain the apparent decline in particle concentration I observed from day three to day seven. Visible particles are generally particles greater than 100µm in size [61]. For illustrative purposes, let us assume that we have a 100µm by 100µmby100µm cubic particle. This single particle could be composed of one million,1µm cubic particles. Therefore, the formation of large particles can greatly decrease the population of smaller particles. If such large particle were to form between day three and day seven, we might expect a large drop in overall subvisible particle concentration. Finally, I observed no overall soluble protein loss by SEC for IVIG agitated in the PBS buﬀer (Figure 13). This observation again reinforces the utility of measuring particle formation when the aggregation of protein therapeutics.

Morphological Diﬀerences Between Protein Particles Formed From Agita- tion of IVIG in the Fluoropolymer Syringe and the Siliconized Syringe

Protein particles formed from agitating IVIG in PBS in the two different surfaces had some interesting morphological characteristics worth noting (Figure 14). Par- ticles generated in the siliconized syringes were much more opaque than particles formed in the fluoropolymer system suggesting more dense particles. Particles from both samples also resembled bits of torn up, twisted protein film. Qualitatively, this observation agrees well with the mechanism of protein particle formation resulting from disruption of a protein layer adsorbed to an interface asproposedbyaGerhardt et al. [95] and Mehta et al. [93].

44 2,500,000 Gore, PBS Gore, PBS+PS20 2,000,000 Siliconized, PBS Siliconized, PBS+PS20 1,500,000

1,000,000

500,000

Particle Concentration (#/mL) 0 0 1 3 7 0 1 3 7 0 1 3 7 0 1 3 7 Agitation Time (days)

Figure 12: Particle formation measured by MFI for IVIG agitated in PBS buﬀer (pH 7.4), in a bare-glass syringe with a ﬂuoropolymer coated plunger (Gore) and in a traditional siliconized syringe with a siliconized rubber plunger. Data are plotted as mean ± standard deviation of three independent replicates except for the siliconized sample without polysorbate 20 on days 1 and 7, which are plotted as the mean and range oftwo independent replicates.

Gore PBS 100% Gore PBS+PS20 Siliconized PBS 75% Siliconized PBS+PS20

50%

25% Percent Mass Recovery 0% 0 1 3 7 0 1 3 7 0 1 3 7 0 1 3 7 Agitation Time (days)

Figure 13: Total soluble protein mass recovery as measured by SEC for IVIG in PBS buﬀer, agitated in in a bare-glass syringe with a ﬂuoropoly- mer surface coated plunger and in a traditional siliconized syringe with asiliconizedrubberplunger.Dataareplottedasmean± standard deviation of three independent replicates except for the siliconized sample without polysorbate 20 on day seven, which is plotted as the mean and range of two independent replicates. No high molecular weightspecies were observed for any of the samples.

45 Fluoropolymer Surface Siliconized Surface

21-34µm 25-31µm

35-39µm 52-57µm 46

56-91µm 74-105µm

Figure 14: Representative flow microscopy images of particlesobservedondaysevenofthesyringeagitation in PBS buffer with the Gore fluoropolymer surface (left) and the siliconized surface (right). Under each set of particles is the range of equivalent circular diameters for the group. Agitation of Avastin

For Avastin I chose to agitate both at the full formulation protein concentration (25 mg/mL), to simulate a commercial pre-filled syringe of the product, and at 1 mg/mL to allow for comparisons with the vial agitation studies. By MFI, I observed no significant particle formation for Avastin when agitated in the presence of the fluoropolymer surface at 1 mg/mL or at the full formulation concentration (25 mg/mL) (Figure 15). This observation was expected given that polysorbate 20 is included in Avastin’s formulation. In the siliconized syringes, by MFI, I observed substantial particle formation, possibly consisting of both protein particles and silicone droplets (Figure 15). Re- sults were essentially equivalent in samples at the two protein concentrations. Thus, it appears that in a traditional siliconized syringe, the commercial formulation of Avastin—which was developed for use in a non-siliconized vial—does not provide sufficient protection. Yet, this same formulation inhibited particle formation in samples agitated in the presence of the fluoropolymer surface. These results suggest that stresses to Avastin in siliconized syringes during agitationaremoredamagingthan in unsiliconized syringes with the fluoropolymer plunger surface. Finally, no loss of soluble Avastin was observed by SEC (Figure 16) at either concentration agitated in contact in either the fluoropolymersyringeorthesiliconized syringe.

Mechanism(s) for Inhibition of Protein Aggregation by Polysorbate 20 During Agita- tion

The addition of polysorbate 20 to both PBS and glycine formulations of IVIG resulted in significant reduction in particle formation duringagitationbothinthevials and in syringes. These effects were observed in samples agitated in the presence of both siliconized surfaces and the fluoropolymer surface. There have been a number

47 Gore, 25mg/mL 1,500,000 Gore, 1mg/mL Siliconized, 25mg/mL Siliconized 1mg/mL 1,000,000

500,000

Particle Concentration (#/mL) 0 0 3 7 0 3 7 0 3 7 0 3 7 Agitation Time (days)

Figure 15: Avastin agitated in syringes with the experimental fluoropolymer-based surface plunger (Gore) in a bare glass syringe and a traditional siliconized plunger and siliconized glass surface as measured by MFI. Data are plotted as mean ± standard deviation of three independent replicates. Substantial particle formation was observed with samples agitated in the siliconized syringe. No significant particle formation was detected during the course of agitation for the fluoropolymersurface.

100% Gore, 25mg/mL Gore, 1mg/mL 75% Siliconized, 25mg/mL Siliconized, 1mg/mL 50%

25% Percent Mass Recovery 0% 0 7 0 7 0 7 0 7 Agitation Time (days)

Figure 16: Total soluble protein mass recovery as measured by SEC for Avastin agitated in syringes with the experimental ﬂuoropolymer-based surface coated plunger (Gore) and a bare glass syringe barrel, and a traditional siliconized plunger and syringe barrel. Data are plotted as mean ± standard deviation of three independent replicates. No high molecular weight species were observed and no total soluble protein mass loss was detected for the duration of the agitation.

48 of proposed mechanisms by which polysorbate 20 inhibits aggregation and particle formation: competition for surface adsorption sites [98–100], binding to specific hydrophobic binding pockets in the protein [101–103], and other non-specific binding that stabilizes the native structure of the protein in solution [104, 105]. However, these mechanisms may not be mutually exclusive [156]. For example, polysorbate 20 could bind to a hydrophobic pocket on the protein and, in doing so, prevent adsorption to asurface.Simultaneously,otherpolysorbatemoleculescould adsorb to the surface an prevent other proteins from adsorbing. To my knowledge, antibodies in general do not specifically bind polysorbate [165]. Also, it has been observed previously that polysorbate inhibits protein adsorption to similar, fluoropolymer-based surfaces [150], and therefore, for my system the most plausible mechanism fortheobservationthat polysorbate reduced aggregation resulting from agitation was inhibition of IVIG adsorption. To test this hypothesis, I analyzed the adsorption of fluorescently labeled IVIG to the fluoropolymer surface using confocal microscopy, both in the presence and absence of polysorbate. Representative images from these experiments are shown in Figure 17. The green regions show adsorbed, fluorescently-labeled IVIG. In both buffers without polysorbate in the formulation, I observed substantial adsorption to the fluoropolymer surface. When I included polysorbate 20 in the formulation and repeated the same incubation with the fluoropolymer surface, I observed virtually no protein adsorption to the fluoropolymer surface. These results demonstrated that with polysorbate 20 in the formulation, adsorption of IVIG to the fluoropolymer surface was inhibited. However, if IVIG was already adsorbed to the surface, would polysorbate be able to displace the protein molecules from the surface? To determine if this were the case, I allowed labeled IVIG to adsorb to the surface and then rinsed the surface with either buffer (analogous to the previous experiment) or bufferwith0.02%polysorbate 20. Representative images from this experiment are shown in Figure 18. Interest-

49 Glycine

Glycine+Polysorbate 20

PBS

PBS+Polysorbate 20

Figure 17: Representative confocal microscopy images for AlexaFluor- 488 labelled IVIG exposed to the Gore ﬂuoropolymer surface and then washed with buﬀer not containing polysorbate 20. Green regions indicate adsorbed IVIG. With polysorbate 20 in the formulation, adsorption was undetectable.

50 ingly, there was a significant visual difference between the amount of fluorescently labeled IVIG adsorbed to the fluoropolymer surface in the glycine buffer and in the PBS buffer. When washed with glycine buffer containing polysorbate, I observed asignificantreductioninthetotalamountoffluorescencewhereas washing with the polysorbate-containing PBS buffer, I did not. In fact, when adsorbed IVIG was washed with the PBS buffer containing polysorbate, I observed clumps of fluorescence and “holes” in an otherwise continuous fluorescent layer. Total integrated fluorescent density for these samples was quantified and is shown in Figure 20. Irepeatedthisexperimentwithachemicallysiliconizedglass surface (Figure 19). Total fluorescent density for these samples and the fluoropolymer surface samples are also shown in Figure 20. I observed essentially equivalent adsorption of IVIG to the siliconized surface and the fluoropolymer surface when IVIG was formulated in the glycine buffer. When IVIG was formulated in the PBS buffer, I saw much more variable adsorption, with no statistically significant reduction in fluorescent density when washed with PBS buffer containing polysorbate. High salt concentration has also been shown to promote protein adsorption to hydrophobic surfaces [166–169]. High ionic strength also dampens intermolecular repulsion between protein molecules and lowers barriers to aggregation between adsorbed protein molecules. In combination, these two factorslikelyexplainwhyIsaw higher different adsorption profiles between IVIG in the PBS buffer and in the glycine buffer. Once adsorbed, the charge shielding provided by the salt in the PBS buffer dampens the repulsive forces between adsorbed molecules resulting in a lower barrier to aggregation on the surface. In addition, the lower net charge on the protein at pH 7.4 may also contribute to lower intermolecular repulsion. This in turn, would promote the formation of larger aggregates, that in ensemble, are more tightly bound to the surface, owing to their larger collective area of hydrophobic interaction. The collective strength of the adsorption makes it more difficult for polysorbate to displace

51 Glycine Wash

Glycine+Polysorbate 20 Wash

PBS Wash

PBS+Polysorbate 20 Wash

Figure 18: Representative confocal microscopy images for AlexaFluor- 488 labelled IVIG exposed to the ﬂuoropolymer surface and subsequently washed with buﬀers with and without the surfactant, polysorbate 20. Green regions indicate adsorbed IVIG.

52 the protein aggregate from the surface.

Eﬀects of pH and Buﬀer on the Conformational and Colloidal Stability of IVIG

The solution aggregation propensity of different monoclonal antibodies has been shown to be variable and complex, depending on both pH, ionic strength and salt type [170, 171]. Generally, in acidic conditions, antibody conformational stability has been shown to be lower [172, 173], specifically due to the sensitivity of the highly con- served CH2 domain in the Fc region to acidic conditions [24, 174–177]. Decreases in conformational stability can promote protein aggregation [176]. However, the lower conformational stability in acidic pH solutions was balanced by increased colloidal stability [175, 178, 179]. However, the effect of ionic strength on aggregation at neutral pH appears to be protein dependent. Some studies have found lower ionic strength promoted aggregation [178–180] and others found that higher ionic strength enhanced aggregation [178, 181, 182]. In acidic pH solutions, aggregation was enhanced with increasing ionic strength [175, 178, 179]. Similar to previous studies, my results showed that the choice of buffer had a dramatic effect on IVIG aggregation propensity during agitation. When agitated in the PBS buffer, IVIG showed greater protein particle formation by MFI and RMM than when agitated in the glycine buffer. This observation wastrueforagitationin contact with both the siliconized surface and the fluoropolymer surface. My hypothe- ses for the mechanisms of these effects were: 1) in PBS at pH 7.4, the conformational stability of IVIG will be greater than in glycine buffer at pH 4.25; 2) the alkaline pH of PBS and the additional salt will reduce the overall colloidal stability of IVIG in solution; and 3) the colloidal stability will be the dominanteffect during the agitation studies. To compare IVIG’s conformational stability in both buffers, I generated protein

53 Glycine Wash

Glycine+Polysorbate 20 Wash

PBS Wash

PBS+Polysorbate 20 Wash

Figure 19: Representative confocal microscopy images for AlexaFluor- 488 labelled IVIG exposed to a chemically siliconized glass surface. Green regions indicate adsorbed IVIG.

54 8.0010 6

6.0010 6

4.0010 6

2.0010 6 Fluorescent Density

Gore, Gly Gore, PBS

Gore, Gly+PS20 Gore, PBS+PS20 Surface, Wash Buffer 8.0010 6

6.0010 6

4.0010 6

2.0010 6 Fluorescent Density

Siliconized, Gly Siliconized, PBS

Siliconized, Gly+PS20 Siliconized, PBS+PS20 Surface, Wash Buffer

Figure 20: Total remaining fluorescence for IVIG adsorbed to a siliconized surface and the fluoropolymer surface and washed withdifferent buffers.

55 unfolding curves, shown in Figure 21. Less denaturant was needed to unfold IVIG in glycine buffer than the PBS buffer (midpoints of transition, 3.9M and 6.2M respectively) indicating that IVIG is conformationally less stable in the glycine buffer at pH 4.25 than in PBS at pH 7.4, consistent with similar findings of conformational instability in antibodies at acidic pHs by others [24, 174–176, 179]. This result con- firmed that conformational instability was not the driving factor in IVIG’s increased susceptibility to agitation induced aggregation in PBS.

354 Glycine Buffer PBS Buffer 352

350

348

346

344 Center of Spectral (nm)Mass 342 0 1 2 3 4 5 6 7 8 9 Urea Concentration (M)

Figure 21: Urea unfolding curves for 1 mg/mL IVIG in 0.2M glycine (pH 4.25) and in PBS (pH 7.4), plotted as points. A sigmoidal ﬁtof the data to Equation 2 is plotted with dashed lines of corresponding color and was used to estimate the midpoint, xm, of the transition. The estimated midpoint of the transitions for glycine and PBS was 3.90M and 6.17M, respectively.

Since the conformational stability of IVIG in the PBS buffer and in the glycine buffer did not correlate with agitation-induced aggregation I observed for IVIG these buffers, it seemed likely then, that IVIG’s colloidal stability was lower in PBS compared to the glycine buffer. To determine is this was the case, Iestimatedthesecond osmotic virial coefficient by static light scattering2 for IVIG in PBS and glycine buffers

2It was discovered near the time of publication that the toluene solution used to calibrate the light scattering system was contaminated at the time these measurements were made. Therefore,

56 at both pHs, 4.25 and 7.4 (Figure 22).

0.004 0.2M Glycine )

2 0.003 PBS mol/g

3 0.002 (cm

22 0.001 B

0.000

7.4 7.4 4.25 4.25 pH

Figure 22: Second virial coeﬃcient (B22), measured by static light scattering at 90◦ for IVIG in 0.2M Glycine buﬀer and PBS at both pH 4.25 and 7.4. Data are plotted as mean ± standard deviation of three replicates.

Overall, B22 values in both buffers at both pHs were positive, indicating overall repulsion between molecules [183]. However, in glycine bufferatpH4.25,therepul- sion was much greater than in the same buffer at pH 7.4. The isoelectric point of IVIG is 8.1 [184], indicating that IVIG has a net positive chargeinbothformulations. However, at pH 7.4, the IVIG molecules have a much lower amount ofoverallcharge than at pH 4.25. It is plausible that the additional net chargeatpH4.25maycon- tribute to higher repulsive forces between molecules in the glycine buffer, consistent with findings on other antibodies [175, 178, 179] and other proteins [185]. In PBS, which has an overall ionic strength nearly 50-fold greater than the glycine buffer (140mM vs. 3mM, respectively), there was no effect of pH on B22 values. The apparent repulsive forces between the molecules remained the same at both pHs, despite the additional net charge on IVIG when the PBS buffer wastitratedtopH the absolute values reported here are not accurate and it was not possible to correct these values for the hard sphere contribution, as is commonly done when reporting values of the second virial coefficient. However, considering the relative differences between the measurements presented here is still valid as these measurements were all made using the same calibration—albeit an erroneous one.

57 4.25. This suggests that the additional ionic strength of this buffer provides a level of charge shielding which dampened repulsion between molecules [186] which, in turn, lowered the barrier to subsequent aggregation. Overall, these results suggest that the lower colloidal stability of IVIG in the PBS buffer may be, at least in part, responsible for the differencesIobservedinagitation- induced aggregation of IVIG in the glycine buffer and the PBS buffer.

The Utility of the Fluoropolymer Surface in Silicone Oil-Free Syringes

Results from this study showed that (1) IVIG adsorbed to the fluoropolymer surface, (2) IVIG’s colloidal stability had a significant impact on the rate of agitation-induced protein aggregation in contact with both the fluoropolymer surface and the siliconized surface, (3) IVIG (and presumably Avastin) adsorption to each surface could be inhibited with the addition of polysorbate 20 and (4) inhibition of protein adsorption by polysorbate 20 was correlated with lower amounts of particle formation in the bulk solution upon subsequent agitation. Given these results, the fluoropolymer surface shows potential to reduce protein aggregation resulting from agitation in drug delivery devices for three reasons. First, I observed less overall particle formation for samples agitated in contact with the fluoropolymer surface. Second, the addition of polysorbate 20, virtually eliminated all particle formation for samples agitated in contact with the fluoropolymer surface. Finally, the fluoropolymer surface did not shed particles into the solution as the silicone oil surfacesdid.Theresultsobtainedin this study are in good agreement with previous work examiningproteinaggregation and particle formation in other silicone oil-free syringes [96]. The fluoropolymer syringe system, with its bare-glass syringe barrel, offers some potential benefits over a silicone oil-free, polymer-based syringe system. First, glass has a long history of use in pharmaceuticals—although not without some issues of its own [187]. Most therapeutic proteins are formulated for storage in a bare-glass

58 vial as the primary container. Therefore, for pharmaceutical companies, moving to asiliconeoil-free,bare-glasssyringewithafluoropolymerplungerrepresentsamore incremental step into this new technology as opposed to moving to a polymer-based syringe. Second, as with any established technology, glass is much more standardized as compared to plastic materials [112], which presents the possibility that vial-to-vial or syringe-to-syringe variability may be greater with plastics. This is a particularly significant worry with the larger number and chemical diversity of extractable and leachable compounds from plastics [109, 110]. Finally, polymer-based vials have been shown to have higher gas permeation rates than glass [111] which may present an issue for oxygen sensitive proteins. In summary, the findings from my study, in combination with thefindingsof Krayukhina et al. and many others, further highlight the damaging effects of silicone oil on protein formulations—especially when compared to novel materials and systems designed specifically to replace siliconized syringe systems. The fluoropolymer surface studied in this work shows promise from a protein compatibility perspective as a component in a glass, silicone oil-free, pre-filled syringe system.

Acknowledgments

I would like to thank W. L. Gore and Associates, Inc. for their generous ﬁnancial support and for providing stopper and plunger samples for this work. I would like to thank Dr. Radu Moldovan and Greg Glazner of the University of Colorado’s Advanced Light Microscopy Core for their help with and stimulating conversations on confocal microscopy. Also, I would like to thank Shaun Bevins of the UCD Biophysics Core for our interesting conversations on SPR and surface adsorption.

59 CHAPTER III

EFFECTS OF PHENOL AND META-CRESOL DEPLETION ON INSULIN ANALOG STABILITY AT PHYSIOLOGICAL TEMPERATURE3

Abstract

The stability of three commercial “fast-acting” insulin analogs, insulin lispro, insulin aspart and insulin glulisine, was studied at various concentrations of phenolic preservatives (phenol and/or meta-cresol) during nine days of incubation at 37◦C. Anal- ysis by both size exclusion and reversed-phase chromatography showed degradation of lispro and aspart that was inversely dependent on the concentration of phenolic preservatives. Insulin glulisine was much more stable than the other analogs and showed minimal degradation even in the absence of phenolic preservatives. With sedimentation velocity ultracentrifugation, I determined thepreservatives’eﬀect on the insulins’ self-assembly. When depleted of preservatives, insulin glulisine dissociates from higher molecular weight species into a number of intermediate molecular weight species, in between monomer and hexamer, whereas insulin aspart and insulin lispro dissociate into monomers and dimers. Decreased stability ofinsulinlisproandinsulin aspart seems to be due the extent of dissociation when depleted of preservative. In- sulin glulisine’s dissociation to intermediate molecular weight species appears to help minimize its degradation during incubation at 37◦C.

3Previously Published in the Journal of Pharmaceutical Sciences, Volume 103, Issue 8, August 2014, Pages 2255–2267.

60 Introduction

Endogenous human insulin monomers assemble into dimers, which further assemble into hexamers coordinated by two zinc ions [188]. Both hexamers and dimers are biologically inactive in the context of regulating the uptake of glucose. However, they are much more conformationally stable than the insulin monomer [189]; resulting in greatly reduced aggregation, fibrillation and chemical degradation during storage compared to the monomer [190]. The hexamer is further stabilized by binding phenolic compounds, such as phenol and meta-cresol (m-cresol) to a hydrophobic pocket in the monomer-monomer interface within the hexamer. This induces a structural change in the B-chain residues B1-B8 from an open loop (T-state) to an α-helix (R- state) [191, 192] which has been shown to have increased stability [124–126, 193, 194]. The overall hexamer structure is designated as one of three states depending on the number of monomers in the T- or R-state: T6, T3R3 or R6. In addition, the transition from the T- to the R-state is favored by heterotropic cooperativity with the anion binding sites on the zinc ligand within the core of the hexamer [195–197]. This ligand-induced structural transition in the formation of the hexamer and the requisite process of dissociation to active insulin monomer delay the onset of biological action of human insulin [121], but help reduce degradation of the protein in drug product formulations [198–200]. Today, the vast majority of patients on insulin therapy use “fast-acting” recombinant human insulin analogs [122]. Point mutations in theseanalogs,designed specifically to disrupt the stabilizing intramolecular interactions of insulin dimers and hexamers, provide faster onset of biological action when compared to human insulin [121, 123, 201]. The three most common analogs used today are insulin lispro (HumalogTM, Eli Lilly, Indianapolis, IN), insulin aspart (NovologTM, Novartis, Basel, Switzerland) and insulin glulisine (ApidraTM, Sanofi Aventis, Bridgewater, NJ). In-

61 sulin lispro is modified by reversing the penultimate two amino acids in the natural human insulin B-chain (from ProB28-LysB29 to LysB28-ProB29) which weakens backbone hydrogen-bonding between monomers [202, 203]. As with human insulin, lispro forms stable hexamers in the presence of zinc and phenol [204]. However, the weak- ening of monomer-monomer interactions in the dimer allows for quicker dissociation from hexamer to the biologically active monomer, resulting in an earlier onset and shorter duration of action than human insulin [205, 206]. Lispro has been shown to be mostly hexameric in the pharmaceutical formulation which contains both zinc and phenol [207]. In the one published study on aspart and lispro pharmaceutical stability, Lougheed et al. found des-amido degradation products of lispro in infu- sions systems [208]. In insulin aspart, the proline at position 28 on the B-chain is replaced with a charged aspartic acid residue. The removal oftheB28prolineweak- ens monomer-monomer backbone hydrogen bonding [125] in much the same way that lispro self-association is destabilized. This decreases aspart’s propensity to dimerize, when compared to human insulin [198], and leads to quicker onset and shortened duration of action [205]. Jars et al. have shown that aspart forms predominantly Asp/IsoAsp degradation products at positions B3 (like lispro)andB28[209]. Insulin glulisine has two substitutions: asparagine to lysine at position B3 and lysine to glutamic acid at position B29. Unlike aspart or lispro dimers, both of which are primarily weakened by the removal of the B28 proline, insulin glulisine leaves the B28 proline intact and instead uses charge-charge repulsion between monomers to destabilize the hexamer [210]. Glulisine is formulated without zinc, and it is expected that glulisine in the drug product does not significantly self-associate into hexamers [210]. However, to my knowledge, there are no published reports on glulisine self- association, structure or stability in pharmaceutical formulations. However, recent work as shown that degradation by fibrillation was found to be comparable to that of lispro [211].

62 Therapeutic insulin analogs are often delivered by subcutaneous injection. With an average patient needing around three injections per day [117], continuous insulin infusion pumps, which were introduced in the early 1980’s, can provide a more convenient mode of treatment. Wearable, “online” blood glucosemonitors[212–214]are being incorporated into insulin pumps and, over the long term, it is expected that the enhanced control with these combined technologies will become the preferred mode of therapy for a large proportion of patients. Since the insulin pump was introduced, however, there have been numerous problems with insulin stability under the conditions within the pumps. For example, insulin has been found to be aggregated in solution in the pump reservoir and tubing [127] and in some cases to have precipitated [215–217]. These two degradation pathways are suspected to lead to catheter occlusion—a continuing clinical problem with analogs in insulin pumps [205, 218–227], which may occurinmorethanhalfof catheters using fast-acting insulin analogs longer than 48 hours [228]. A number of studies also have shown that levels of phenolic compounds, which are included in pharmaceutical formulations as antimicrobial preservatives as well as insulin stabilizers, are depleted by absorption into the insulin pump catheters [127– 129], with as little as 10% remaining after 24 hours [127]. Experiments performed in my lab with currently available catheter sets showed the complete loss of m-cresol upon incubation at 37◦C for 46.5 hours (see Appendix B). Because phenolic preservatives play such important roles in human insulin self-assembly, structure and storage stability [125, 211], I hypothesized that depletion of thesestabilizingligandswouldde- crease the extent of hexamer assembly in these insulin analogs as well. Subsequently, with more insulin analog molecules in the monomeric state, I expect reduced stability at physiological temperature. To test this hypothesis, I removed preservatives from marketed insulin analog formulations, leaving the remainder of the formulation intact. I then analyzed the assembly state of each of the analogs with and without

63 phenolic preservatives by analytical ultracentrifugation. Next, I added various frac- tions of the original amount of phenolic preservatives back into the formulation and incubated samples quiescently at 37◦Cforninedays.Atvarioustimepointsthrough- out the incubation, samples were analyzed for physical degradation by size-exclusion chromatography and chemical degradation by reversed-phasechromatographytode- termine the eﬀect of reduced phenolic preservative levels on stability.

Materials and Methods

Materials

Insulin analogs were purchased as commercial drug products from a local pharmacy and were used before their expiry dates. All insulin analogs were purchased in 10mL vials. Insulin lispro is formulated in a 1.88 mg/mL sodium phosphate buffer with 16 mg/mL glycerol, 3.15 mg/mL meta-cresol (m-cresol) and 0.0197 mg/mL zinc [229]. Insulin aspart is formulated in a 1.25 mg/mL sodium phosphate buffer, with 16 mg/mL glycerol, 0.58 mg/mL sodium chloride, 0.0196 mg/mL zinc, and uses a combination of 1.5 mg/mL phenol and 1.72 mg/mL meta-cresol as antimicrobial agents [230]. Insulin glulisine is formulated in 6 mg/mL tromethanebuffer, 5 mg/mL sodium chloride, 3.15 mg/mL m-cresol and 0.01 mg/mL Tween 20 [231]. All laboratory chemicals used were of analytical grade or higher and were used without further purification. Water used in mobile phases, formulations and buffers was purified through a Millipore Synergy UV (Billerica, MA) filtration unit.

Stability Study

Phenol and m-cresol were removed from insulin analogs using Zeba (Thermo Scien- tific, Rockford, IL) desalting columns. The desalting columns were washed three times with formulation buffer not containing phenolic preservatives, and analogs were eluted from the desalting column using the same buffer, as has been done previously [211].

64 A significant number of particles were shed from the Zeba columns during preparation. These particles were removed by preparative ultracentrifugation at ∼110,000g for 70 minutes at 4◦C. The supernatant from the ultracentrifugation step was concentrated to ∼8 mg/mL using Amicon Ultra 3,000 MWCO centrifugal filters (Millipore, Cork, Ireland) and then diluted to 6.94mg/mL (two times the marketed insulin concentration). The concentrated insulin analog solutions were then diluted to a final concentration of 3.47mg/mL and final phenol and/or m-cresol concentrations of 0%, 20%, 40%, 60%, 80% and 100% of those in the each of the respective insulin analog commercial formulation (Table 1). 200 uL samples were pipetted into 1.3 mL glass lyophilization vials (West Pharmaceuticals, Exton, PA). Vials were capped and incubated quiescently at 37◦C. Three separate vials for each sample type were tested at each of the 0, 1, 6 and 9 day time-points.

Size Exclusion Chromatography (SEC)

None of the insulin analogs I studied have any tryptophan residues. Therefore, the strongest UV absorbance comes from tyrosine residues that unfortunately have an absorbance spectrum that overlaps those of m-cresol and phenol, which are present in nearly 50-fold molar excess in the marketed formulations. In addition, HPLC SEC column performance degrades rapidly with these samples due to insulin aggregates, phenolic preservative and/or zinc interactions with the column [232, 233]. To overcome these limitations and avoid the pitfalls of the US Pharmacopoeia SEC method for insulin [234], I developed a UPLC-SEC method. Insulin monomer and soluble aggregates were detected and quantified by SEC using Waters Acquity UPLC BEH200 SEC column (4.6mm x 150mm) on an Agilent 1100/1200 HPLC. Toremove large insoluble aggregates, samples were centrifuged at ∼9650g for 5 minutes, and the supernatant was used for injection onto the SEC column. Degassed and 0.2um filtered 0.03M NaCl, 0.01M NaPO4, pH 7.4 mobile phase was used at aflowrate

65 of 0.208mL/min, with absorbance monitored at 280 nm. Run timewas34minutes per sample. Data were exported from Agilent’s Chemstation and baseline corrected, integrated and analyzed with custom scripts written in Matlab.

Reversed Phase Chromatography (RP)

RP chromatography was performed using an Agilent 1100 HPLC with UV absorbance detection at 220nm. Samples were separated on a Grace Vydac 218TP C18 5um column (150mm x 4.6mm) (Deerfield, IL), using a flow rate of 1 mL/min. Mobile phase A contained 5% acetonitrile, 95% water and 0.1% trifluoroacetic acid. Mobile phase B contained 95% acetonitrile, 5% water and 0.1% trifluoroacetic acid. Mobile phase composition during chromatography is shown in Table 2. Data were exported from Agilent’s Chemstation and baseline corrected, integrated and analyzed with custom scripts written in Matlab.

Micro-Flow Imaging (MFI)

Particle concentration and size data were obtained from a DPA4100 Micro-Flow Im- ager (MFI, ProteinSimple, Santa Clara, CA). A sample of 0.45 mL was flowed through the system, and the total analysis volume was 0.16mL. System suitability was checked before each run using Millipore Synergy UV (Billerica, MA) filtered deionized (DI) water. Results were deemed acceptable if total particle counts were less than 500 particles per mL. Flow cells were cleaned in place by flushing 1mL ofa1%Tergazyme (Alconox, White Plains, NY) or a 1% Hellmanex (Hellma Analytics, Mullheim, Ger- many) solution, and then rinsed with excess water as needed throughout the analysis. Formulation buffer made without phenolic preservatives was used to flush the system in between samples. MFI data were exported and analyzed using custom scripts written in Perl.

66 Atomic Absorption Spectrometry (AAS)

Zinc concentration was determined by atomic absorption spectrometry on an AAn- alyst 400 (PerkinElmer Instruments LLC, Shelton, CT), with adeuterium-arcback- ground correction. Zinc standards were prepared from a zinc chloride standard solution (Sigma-Aldrich, Saint Louis, MO), and a three-point calibration curve was created for zinc around the expected sample concentration. Sample solution was aspirated to the ﬂame and read four times at a wavelength of 213.9 nm.

Analytical Ultracentrifugation (AUC)

Sedimentation velocity AUC experiments were performed on a ProteomeLab XL- A/XL-I (Beckman Coulter, Indianapolis, IN) ultracentrifuge using interference optics, sapphire windows and interference window holders (Spin Analytical, Berwick, ME). The 0% and 100% phenolic samples were prepared identically tothesamplesprepared for the stability study (see above). All AUC samples were dialyzed overnight at 4◦C against their respective buffer (0% or 100% phenolic-containing formulation buffer) before analysis. Drug product was used straight from the product vial, but was dialyzed to ensure buffer match. Samples were run in triplicate at 50,000 RPM at 22◦Cfor12hours.Cellswerecleanedbyrinsingwitha5%Liquinox solution (Alconox Inc., White Plains, NY), excess MilliQ water and finally with 70% ethanol. The decision to run SV-AUC at 22◦C was to balance concerns about protein degradation during the AUC run with a desire to match the temperatureoftheincubation experiments. Because incubation at 37◦C caused significant degradation of the insulin analogs over the time course used for an AUC run, I chose to use the lower temperature of 22◦C. Data were collected with ProteomeLab and analyzed with both DCDT+ [235], to confirm my method’s comparability with previously published g(s∗)distributions on insulin lispro [207, 236], and with SEDFIT [237]. In SEDFIT, continuous c(s)distributionswerefitusingbuffer density, buffer viscosityandpartialspecific

67 volume estimates from Sednterp (20130813 Beta). All experimental S-values were converted to S20,w for comparison. Overall weight-averaged S-values are expressed as mean ± standard deviation for three independent replicates. Beckman-Coulter lists phenol and m-cresol both as chemicalstoavoidusingin combination with their charcoal/epon centerpieces [238, 239] due to issues with adsorption and potential subsequent deformation of the centerpieces. I investigated this potential incompatibility at insulin analog formulation concentrations of m-cresol and phenol and found no adverse effects on sedimentation velocityexperiments(datanot shown). I have also performed experiments looking at the concentration dependence of insulin analogs on the c(s)distributionsat4◦C(datanotshown).Insulinlisproand insulin aspart showed relatively slow association kineticsinrelationtothetimescale of a SV-AUC experiment and therefore I am more confident that the peaks observed represent distinct molecular species. With insulin glulisine, however, I observed relatively rapid self-assembly kinetics and as a result, I am unable to unambiguously propose an assembled state for each observed peak (manuscript in preparation). Be- cause the kinetics may differ at the experimental temperatureusedinthecurrent work, I would like to stress that the species highlighted in the analysis are only estimates based to molecular weight calculations from the c(s)distributions.

Results

Atomic Absorption Spectroscopy

As mentioned above, zinc is an important contributor to insulin self-assembly and stability [189, 240, 241]. To confirm that zinc levels were notalteredduringremoval of the phenolic preservatives, I quantified zinc concentration before and after treatment with the desalting columns by AAS. I observed no change in zinc concentration through the buffer exchange procedure (Figure 23).

68 20 Drug Product Desalted with Zeba Column 15

[Zinc] (ug/mL) [Zinc] 5

Insulin Lispro Insulin Aspart Insulin Glulisine

Figure 23: Atomic Absorption Spectrometry analysis of insulin lispro, aspart and glulisine. Samples were taken from directly from drug product vials (blue bars), and after buﬀer exchange through Zeba desalting columns (red bars). Data are plotted as mean ± standard deviation of three replicates.

Size Exclusion Chromatography

An overlay of example SEC chromatograms over the time course oftheincubation study is shown in Figure 24. The native insulin peak elutes at approximately 9 minutes, with high molecular weight species (HMWS) eluting from approximately 4 to 8 minutes. The m-cresol peak elutes from approximately 20 to 23 minutes and was found to elute earlier as a function of the number of sampleinjectionsontothe column, unrelated to insulin analog type or sample phenolic preservative level. The main peak area and HMWS peaks area were quantiﬁed for the three insulin analogs over all phenolic preservative concentrations studied (Figures 25, 26 and 27). For all three analogues over the time course of the study, there was no reduction in the overall protein mass recovery during SEC analysis, indicating no loss of soluble

69 Day 0 300 Day 1 Day 6 Day 9

200 mAU

100

0 0 5 10 15 20 25 30 35 Time (min)

Figure 24: Example chromatograms of lispro with 40% phenolic (m-cresol) over the course of the study. The insulin native peak elutes atapprox- imately 9 minutes, HMWS elute from approximately 4-8 minutes. The m-cresol peak elutes from approximately 20-23 minutes. Labels of “percent phenolic” in this figure refer to the amount of phenolic preservative in the sample with respect to the amount of phenolic preservative in the marketed formulation. See Table 1 for a list of phenolic concentrations is various units. protein during incubation (data not shown). Based on formation of HMWS, insulin lispro showed the greatest rate and extent of degradation at all preservative levels tested. With insulin lispro and aspart, in support of my hypothesis, degradation rate increased with reductions in preservative levels, although this effect was less pronounced with insulin aspart. During the incubation period studied, essentially no degradation was observed for insulin glulisine by SEC analysis, even in the complete absence of phenolic preservatives. Based on SEC analysis, there were no significant differences in the stability of the drug products of the three insulin analogues (Figure 28). Also, for all drug product samples tested, total protein mass recovery on SEC was unchanged over the course

70 Figure 25: Percent main peak (A) and HMWS (B) over various concentrations of phenolic preservatives as measured by SEC forinsulin lispro. All values are plotted as mean ± standard deviation for three replicates except the 40% phenolic sample which is plotted as the mean and range of two replicates. Labels of “percent phenolic” in this ﬁgure refer to the amount of phenolic preservative in the sample with respect to the amount of phenolic preservative in the marketed formulation. See Table 1 for a list of phenolic concentrations is various units. of the study (data not shown).

Reversed-Phase Chromatography

Example RP chromatograms observed during the time course of the incubation study are shown in Figure 29. The main insulin peak elutes at approximately 12.5 minutes.

71 Figure 26: Percent main peak (A) and HMWS (B) over various concentrations of phenolic preservatives as measured by SEC forinsulin aspart. All values are plotted as mean ± standard deviation for three replicates except day 9, which is plotted as the mean and rangeoftwo replicates. Axes are rescaled to better show detail in the inset ﬁgure in panel B. Labels of “percent phenolic” in this ﬁgure refer to the amount of phenolic preservative in the sample with respect to the amount of phenolic preservative in the marketed formulation. See Table 1 for a list of phenolic concentrations is various units.

The peak at 16.5 was determined to be HMWS by collecting material eluting from the SEC HMWS peak (from 4-8 minutes) and injecting that materialonRP(data not shown). All other peaks in the RP chormatogram, from approximately 13 to 16 minutes, were considered products of chemical degradation of the insulin analogues [208, 209, 242] and were collectively quantiﬁed as degradantpeaks.Forinsulinlispro

72 Figure 27: Percent main peak (A) and HMWS (B) over various concentrations of phenolic preservatives as measured by SEC forinsulin glulisine. All values are plotted as mean ± standard deviation for three replicates. Error bars are smaller than the symbols. Axes are rescaled to better show detail in the inset figure in panel B. Labels of “percent phenolic” in this figure refer to the amount of phenolic preservative in the sample with respect to the amount of phenolic preservative in the marketed formulation. See Table 1 for a list of phenolic concentrations is various units. and aspart, degradation products eluting at 13.2 and 13.5 minutes are likely products of deamidation of the B3 asparagine into iso-aspartic acid and aspartic acid [208, 209, 242]. Insulin glulisine lacks the B3 asparagine. Over the time course of the incubation, peak areas for native, HMWS and chemically degraded species were quantified for the three insulin analogs across the range of

73 Figure 28: Percent main peak (A) and HMWS (B) in drug product for all three analogs over the course of this study. All values are plotted as mean ± standard deviation of three replicates. Axes are rescaled to better show detail in the inset figure in panel B. Labels of “percent phenolic” in this figure refer to the amount of phenolic preservative in the sample with respect to the amount of phenolic preservative in the marketed formulation. See Table 1 for a list of phenolic concentrations is various units. preservative levels tested (Figures 30, 31 and 32). In general, the RP results followed closely with what was observed by SEC analysis. At all levels of phenolic preservative tested, insulin lispro formed HMWS and chemically degraded products more rapidly when the preservative level was decreased (Figure 30). Insulin aspart had much slower degradation rates, which were not significantly impacted by the level of preservative

74 Figure 29: Example chromatograms of insulin lispro with 40% of the concentration of phenolic preservative (m-cresol) in the commercial drug product over the course of the study. Solvent peaks eluted before 4 minutes, m-cresol eluted at approximately 4.5 minutes and the native insulin peak eluted at approximately 12.5 minutes. High Molecular Weight Species (HMWS) were determined to elute at ∼16.5 minutes by injection of SEC HMWS. All other peaks, eluting from approximately 13 to 16 minutes, were considered degradation products and are collectively referred to as degradant peaks.

75 (Figure 31). As observed with SEC analysis, RP chromatography documented that insulin glulisine did not degrade during the time course of the incubation, even in the absence of preservatives (Figure 32). No signiﬁcant diﬀerence between the drug products was noted by RP chromatography over the course of the study (Figure 33).

Analytical Ultracentrifugation

To my knowledge, the only published SV-AUC self-assembly analysis in formulation conditions for any of these three insulin analogs was done on insulin lispro using g(s∗) distributions [207, 236]. The g(s∗)distributionisarelatively“lowresolution”method when compared to more modern direct-boundary modeling methods implemented in SEDFIT. Given the complexity and number of species in insulin self-association, I decided to use direct-boundary modeling to better quantify and characterize diﬀer- ences in association state in the presence and absence of phenolics. However, direct- boundary modeling requires three key assumptions that may not be valid for my particular experimental set up; namely that each species hasthesameoverallshape, net association/dissociation is negligible over the time course of the experiment, and that all the species sediment ideally. The “model-independent” g(s∗)distributions allow us to avoid these assumptions—albeit at the expense of resolution. My data, when analyzed by g(s∗)distributions,showgoodagreementwithpreviouslypub- lished results on insulin lispro [207, 236], but I was not abletoidentifyandquantify all the species present in my samples by gaussian-lorentzian deconvolution of the g(s∗) distributions. Using direct-boundary modeling in SEDFIT, I was able to get much better data resolution. However, I again caution the over-interpretation of these results given the potential unknown validity of the aforementioned assumptions. Both in the commercial drug product and in the sample prepared with 100% phenolics, insulin lispro sedimented predominantly (∼81%) with a peak value of ∼ 3S, corresponding to an estimated molecular weight consistent with a hexamer (Fig-

76 Figure 30: Percent main peak (A), degradant species (B) and HMWS (C) over various concentrations of phenolic preservative as measured by RP for insulin lispro. All values are plotted as mean ± standard deviation of three replicates. Labels of “percent phenolic” in this ﬁgure refer to the amount of phenolic preservative in the sample with respect to the amount of phenolic preservative in the marketed formulation. See Table 1 for a list of phenolic concentrations is various units. ure 34A). In both of these samples, only ∼5% sedimented with a peak value less than 1.5S (an estimated molecular weight between monomer and dimer). Interest-

77 ingly, both preparations also showed peaks at sedimentation coefficients of 4.5S and 6S, which I suspect may be 12-mer (∼12%) and 18-mer (∼3%), respectively. The overall, weight-averaged sedimentation coefficients for drug product and the 100% preservative sample were 2.94 ± 0.03S and 3.09 ± 0.03S, respectively. With the preservative removed, insulin lispro dissociated, and ∼30% sedimented slower than 1.5S. The remaining sedimentation values ranged from ∼2.5S to 5S. Peaks in this range were not well resolved in the analysis leading us to suspect they may have been due to aggregated degradation products, as suggested by my SEC data. The overall weight-averaged sedimentation coefficient, including the likely degradation products was 2.95 ± 0.02S. The value was 1.17 ± 0.04S if the peaks for degradation products are excluded. For insulin aspart, samples prepared with 100% phenolic preservative and the commercial drug product sedimented almost entirely (>97%) with a peak value of ∼3S, also consistent with a hexamer. The remaining ∼3% sedimented slower than 1.5S (an estimated molecular weight between monomer and dimer). Also, a single peak was observed at a sedimentation coefficient of approximately 6S, however, this species accounted for less than 1% of the total population. The overall weight- averaged sedimentation coefficients for the drug product and 100% phenolic sample were 2.87 ± 0.02S and 2.90 ± 0.00S, respectively. Upon complete depletion of the phenolic preservatives, insulin aspart showed a slightly lesser degree of dissociation compared to insulin lispro, with ∼19% sedimenting slower than 1.5S. The overall weight-averaged sedimentation coefficient without the phenolics was 2.63 ± 0.01S. Insulin glulisine showed a unique and surprising sedimentation profile with more than four unique peaks. With the 100% phenolics sample and in the drug product, insulin glulisine sedimented with peaks at ∼ 3S and ∼ 4S, along with about 10% sedimenting below 2S. This observation was surprising given the lack of zinc in the formulation. Based on my observations with insulin lispro and insulin aspart and

78 the molecular weight estimates from their c(s)fits,Iwouldexpectthesepeaksto be roughly hexameric or larger, on average. Overall weight-average sedimentation coefficients for drug product and the 100% phenolic samples were 3.01 ± 0.01S and 3.09 ± 0.03S, respectively. Upon depletion of the phenolics, glulisine also dissociated with approximately 33% of the population sedimenting below 2S and the overall weight-average sedimentation coefficient dropped to 2.21 ± 0.02S.

Micro-Flow Imaging

Initially for analysis of the results from MFI, particles dataweredividedintothree size classes: greater than or equal to 1um and less than or equal to 10um; greater than 10um but less than or equal to 25um; and greater than 25um (large particles). However, for all of insulin analogues, no discernible trends were noted with the results for the different size classes with respect to either incubation time or phenolic preservative level (data not shown). Therefore, the total particle concentration for particles greater than or equal to 1um was used to compare results (Figure 35). The total particle concentration for insulin lispro sampleswasessentiallyun- changed throughout the course of the study, and in general lispro had more particles than either insulin aspart or glulisine samples. In samples with high levels of phenolic preservatives (80% and 100%) at 6 and 9 days incubation, insulin aspart samples showed nucleation-like increase in total particle concentration. Insulin glulisine exhibited an initial growth of total particles at day 1 at high preservative concentration similar to aspart, however, over time the total particle counts returned to day 0 levels. It is interesting to note the considerable differences in particle concentrations between the three commercial drug products, analyzed directly out of the product vials (day 0) and with incubation at elevated temperature (Figure 36). Insulin lispro drug product initially had significantly higher total particle concentration (22, 192 ± 1, 445 particles per mL, mean ± standard deviation), compared with either glulisine

79 (5, 780 ± 4, 612) or aspart (685 ± 560). During incubation at 37◦C, insulin glulisine drug product also exhibited nucleation-like increase in total particle concentration and subsequent reduction to day 0 levels, similar to the observation with the sample prepared with 80% and 100% preservative. With insulin lispro and insulin aspart drug products, particle levels were essentially unchanged during the incubation study.

Discussion

The intrinsic self-assembly process of insulin lispro and insulin aspart are both destabilized primarily by the removal of the B28 proline [193, 199, 200], however, both of these insulin analogs are formulated with both phenolic preservatives and zinc in the drug products. Insulin glulisine, on the other hand, retains the B28 proline and includes m-cresol, but lacks zinc in the commercial drug product. Both insulin lispro and insulin aspart are reported to require zinc and m-cresol (and/or phenol) to form hexamers [203, 204, 236]. I found with SV-AUC that with full formulation levels of phenolic preservatives and in the respective drug products, both lispro insulin and aspart insulin were indeed primarilyassembledintohexamers. When assembled, both insulin analogs were less susceptible todegradationduring incubation at 37◦Cthanwhenincubatedintheabsenceofpreservatives. Surprisingly, my results from SV-AUC showed that in samples of insulin aspart fully depleted of phenolic preservatives, there was still a signiﬁcant fraction assembled, approximately 81%, and the weight-averaged sedimentation coeﬃcient remained relatively high at ∼ 2.6S. In contrast, insulin lispro appears to dissociate more readily. When the phenolic preservatives were removed from insulin lispro, I observed a large fraction of smaller species and a polydisperse population of aggregated species, consistent with my SEC results and similar to what has been seen before [236]. The higher fraction of the molecules in an assembled state for insulin aspart most likely accounts for its slower rate of degradation during incubation at 37◦Ccomparedto

80 that of insulin lispro. Insulin lispro and insulin aspart’s formulations differ mainly in the choice of phenolic preservative and the addition of sodium chloride. Insulin aspart uses a combination of phenol and m-cresol whereas insulin lispro uses only m-cresol. Previous observations with human insulin have showed that phenol is a more potent stabilizer of the R-state hexamer than m-cresol [200]. One explanation for thedifferences in stability and assembly between insulin lispro and insulin aspart may bethatatraceamountof phenol could have been retained during my preservative depletion process (see Meth- ods). This level of phenol may have been enough to stabilize more aspart hexamers compared to the number of lispro hexamers stabilized by a trace amount of m-cresol. Iwasabletoruleoutthispossibilitybydeterminingthelevels of preservatives remaining in the samples after the buffer exchange process. First, I took insulin lispro and insulin aspart samples that had been processed through the desalting columns and denatured the proteins in 6M guanidine hydrochloride. This should dissociate any residual preservative bound to the proteins. The denatured samples were then analyzed by SEC. Based on the baseline noise on my HPLC, I determined that the limit of detection of my assay for phenol or m-cresol was approximately 175 fg/mL. In denatured samples of insulins that should have been completely depleted of phenolic preservatives by buffer exchange, I found no signal in the SEC chromatogram for residual phenol or m-cresol. Therefore, the stability differences between insulin lispro and insulin aspart were not due to retention of residual preservatives and were most likely due to the addition of the chloride anion in insulin aspart and intrinsic differences in self-assembly properties between insulin lispro and insulin aspart. Like phenolic preservatives, anion binding helps drive the assembly of insulin hexamers [195–197]. The sodium chloride in insulin aspart’s formulation, a component that insulin lispro lacks, appears to help insulin aspart remain predominantly assembled—even when fully depleted of phenolic preservatives. Anion binding also increases

81 the hexamer’s affinity for phenolic ligands, meaning that the relative distributions of T6, T3R3 and R6 hexamers between insulin lispro and insulin aspart may differ significantly as the phenolic preservatives are depleted. I would expect insulin aspart to prefer the phenolic-bound, more stable R-state to a greater degree than insulin lispro due to the inclusion of sodium chloride in the formulation. In addition, there may be intrinsic differences in stability between the insulin lispro and insulin aspart T-state hexamers themselves that may contribute to degradation. When fully depleted of phenolic preservatives, I expect both insulin lispro and insulin aspart hexamers to be in the T-state. Solution NMR studies have shown that the T-state is essentially analogous to the free monomer structure [243, 244], and raises the possibility that differences in degradation could also be due to intrinsic differences in T6 hexamer stability between insulin lispro and insulin aspart in combination with changes in assembly state. Incubation of insulin glulisine at 37◦Cresultedinalmostnodetectabledegradation over the 9 days of the experiment by SEC and RP. This observation was made with all levels of preservative tested and, remarkably, even in samples incubated without any preservative. With SV-AUC, I found that in the presence of 100% of the drug product preservative level, insulin glulisine was assembled into higher order species, contrary to previous expectations [210], although I was unable to unambiguously determine the identity of the assembled species. This assembly could account for the lack of degradation of these samples during incubation at37◦Cinthepresence of 100% preservative, as was observed with insulin lispro and insulin aspart in the presence of preservative. Interestingly, when completely depleted of phenolic preservatives, a substantial fraction of insulin glulisine became dissociated, and the overall assemblies shifted toward smaller species between S-values that would be consistent with hexamer and monomer for insulin lispro and insulin aspart. It appears that assembly into these

82 intermediate species protected glulisine from degradation during incubation at 37◦C without preservatives. Insulin lispro and insulin aspart showed no discernible trends in formation of subvisible particles during the course of the study, in contrasttothedegradationprofiles observed by SEC. Therefore, it appears that the formation of soluble aggregates was the dominant path of physical degradation. However, as evidenced by results with insulin glulisine, particle concentration is an important parameter to monitor during stability studies. Even though I did not observe any significant degradation of this analog by SEC or RP chromatography, insulin glulisine samples still showed formation of subvisible particles. These results further underscore the importance of measuring subvisible particles during pharmaceutical stability testing [245] and that particle formation is protein and formulation specific [246]. One aspect of insulin assembly that was not examined in the study was the cooperativity of heterotropic anion binding with the phenolic binding in each of the three analogs. Given the strength of this effect with human insulin [195–197], I suspect that the insulin manufacturers control these levels during commercial production. However, specific anion concentration information is not listed on the product insert, making the exact formulation difficult to duplicate. This difference may also explain why some of my 100% preservative samples deviated from the behavior of the drug product. Additionally, I did not quantify any reduction in anions through the buffer exchange step, which could potentially have contributed to deviations from drug product behavior. However, I expect that these changes would be uniform over all samples and that the overall conclusions of this study would not be affected. But as has been seen with other studies where protein sample material was derived from drug products [247], caution should be used when interpreting these results in the context of the commercial drug product. In summary, results from my study document that upon depletion of preservatives,

83 insulin lispro and insulin aspart degrade more rapidly during incubation at 37◦Cwhich appears to correlate with decreased overall self-assembly. Such degradation suggests this mechanism of phenolic depletion could potentially contribute to the occlusion of catheters used in insulin pump systems—although more research is required to investigate this potential link directly. My results clearly show, especially for insulin lispro, that in order to maximize stability, preservative depletion should be minimized. Finally, even in the absence of preservatives, insulin glulisine was more resistant to degradation than insulin lispro or insulin aspart during incubation at 37◦C.

Acknowledgements

This work was supported by a generous grant from BD Technologies and by a PhRMA Pre-Doctoral Fellowship. I would like to thank Amiee Howard, Dr. Lei Sian and Dr. Nancy Krebs of the Pediatric Nutrition Lab at the University of Colorado Denver for their help in running the AAS instrument. I would also like to thank Dr. Brooke Hirsch at the University of Colorado Biophysics Core, Dr. KeithConnaghanandDr. David Bain for their stimulating conversations about AUC. Additionally, I would like to thank the reviewers for their helpful comments, insights and suggestions on this manuscript.

84 Table 1: Phenolic Preservative Concentration in Various Units

Insulin Glulisine/Lispro Insulin Aspart

m-Cresol m-Cresol Phenol

Mass Molar Mass Molar Mass Molar Percent Concentration Concentration Concentration Concentration Concentration Concentration Phenolic

85 (g/L) (mM) (g/L) (mM) (g/L) (mM)

0% 0.00 0.00 0.00 0.00 0.00 0.00

20% 0.63 5.83 0.34 3.18 0.30 3.19

40% 1.26 11.65 0.69 6.36 0.60 6.38

60% 1.89 17.48 1.03 9.54 0.90 9.56

80% 2.52 23.30 1.38 12.72 1.20 12.75

100% 3.15 29.13 1.72 15.91 1.50 15.94 Table 2: Mobile phase composition during reversed-phase chromatography Time (min) Mobile Phase A (%) Mobile Phase B (%) 08020 14 65 35 15 10 90 18 10 90 20 80 20 25 80 20

86 Figure 31: Percent main peak (A), degradant species (B) and HMWS (C) over various concentrations of phenolic preservative as measured by RP for insulin aspart. All values are plotted as mean ± standard deviation of three replicates except day 0, which is shown as the mean and range of two replicates. Axes are rescaled to better show detail in the inset ﬁgure in panel C. Labels of “percent phenolic” in this ﬁgure refer to the amount of phenolic preservative in the sample with respect to the amount of phenolic preservative in the marketed formulation. See Table 1 for a list of phenolic concentrations is various units.

87 Figure 32: Percent main peak (A), degradant species (B) and HMWS (C) over various concentrations of phenolic preservative as measured by RP for insulin glulisine. All values are plotted as mean ± standard deviation of three replicates except day 6, 60% preservative, which is plotted as the mean and range of two replicates. Axes are rescaled to better show detail in the inset ﬁgure in panel C. Labels of “percent phenolic” in this ﬁgure refer to the amount of phenolic preservative in the sample with respect to the amount of phenolic preservative in the marketed formulation. See Table 1 for a list of phenolic concentrations is various units.

88 Figure 33: Percent main peak (A), degradant species (B) and HMWS (C) for drug product as measured by RP. All values are plotted as mean ± standard deviation for three replicates. Axes are rescaled to better show detail in the inset ﬁgure in panels BandC.

89 0% Preservative A 100% Preservative Drug Product c(s) (fringes/Svedberg) 1 2 3 4 5 6 7

Sedimentation Coefficient (s 20,w, Svedbergs)

B c(s) (fringes/Svedberg) 1 2 3 4 5 6 7

Sedimentation Coefficient (s 20,w , Svedbergs)

C c(s) (fringes/Svedberg) 1 2 3 4 5 6 7

Sedimentation Coefficient (s 20,w , Svedbergs)

Figure 34: Representative c(s20,w) distributions for insulin lispro (A), insulin aspart (B) and insulin glulisine (C) from SV-AUC. Red traces represent formulations with 0% phenolic preservative; bluetraces,100% phenolic preservative; black traces, drug product. Labels of “percent phenolic” in this ﬁgure refer to the amount of phenolic preservative in the sample with respect to the amount of phenolic preservative in the marketed formulation. See Table 1 for a list of phenolic concentrations is various units.

90 Figure 35: Total particle concentration data for insulin lispro (A), aspart (B) and glulisine (C) as measured by MFI. Data are expressed as mean ± standard deviation for three replicates except: aspart 60% and 80% preservativeday1,aspart60% preservative day 6, glulisine all levels day 9, and glulisine40%preservativeday6 which are all plotted as the mean and range of two replicates. Note that the scales diﬀer between the three plots. Labels of “percent phenolic” in this ﬁgure refer to the amount of phenolic preservative in the sample with respect to the amount of phenolic preservative in the marketed formulation. See Table 1 for a list of phenolic concentrations is various units.

91 Figure 36: Particle concentration data for insulin lispro, aspart and glulisine drug products as measured by MFI over the course of the study. All values are plotted as mean ± standard deviation for three replicates except glulisine, which is plotted as the mean andrangeof two replicates on day 1, 6 and 9.

92 CHAPTER IV

ANALYZING INSULIN SAMPLES BY SIZE-EXCLUSION CHROMATOGRAPHY: A COLUMN DEGRADATION STUDY4

Abstract

Investigating insulin analogs and probing their intrinsic stability at physiological temperature, I observed signiﬁcant degradation in the size-exclusion chromatography (SEC) signal over a moderate number of insulin sample injections, which generated concerns about the quality of the separations. Therefore, myresearchgoalwasto identify the cause(s) for the observed signal degradation and attempt to mitigate the degradation in order to extend SEC column life-span. In these studies, I used multi-angle light scattering (MALS), nuclear magnetic resonance (NMR) and gas chromatography-mass spectrometry (GC/MS) methods to evaluate column degradation. The results from these studies illustrate: i) that zinc ions introduced by the insulin product produced the observed column performance issues; and ii) that including EDTA, a zinc chelator, in the mobile phase helped to maintain column performance.

Introduction

A recent study using size-exclusion chromatography (SEC) toassessthestability of three diﬀerent insulin analogs containing various phenolic preservatives was reported [248]. In these studies, Teska et al. observed a rapid decrease in the SEC column performance; in fact, it only required a moderate number of sample injections (approximately 80-100) to produce this observation [248]. These observations were

4Previously Published in the Journal of Pharmaceutical Sciences, Volume 104, Issue 4, April 2015, Pages 1555–1560.

93 corroborated via conversations with other researchers who extensively perform insulin research and analysis [232, 249]. Given the large amount of insulin and insulin analogs produced worldwide, and the considerable cost of the SEC columns, I felt that these column performance issues warranted further investigation and are the scope of the current work. Since many diabetics require multiple injections per day in order to maintain an acceptable blood glucose level, marketed insulin and insulin analogs are formulated in multi-dose vials [117]. Furthermore, multiple needle insertions into a sterile drug product vial inherently increase the possibility of bacterial contamination. The FDA (Food and Drug Administration) requires drug products in multi-dose vials to include an antimicrobial preservative [34]. In the case of insulin formulations, phenol and/or meta-cresol are commonly employed. Insulin is somewhat of a special case, as it is well established that the presence of phenol and/or meta-cresol promotes favorable conformational changes in the insulin hexamer form [191, 192] and provides added stability to the drug product [124, 194]. It is also well known that insulin, and insulin analogs, exhibit a complex self-assembly process to produce hexamers coordinated by two zinc ions [188]. This assembly confers additional stability [189, 198–200, 248] providing a more robust shelf-life, but it is important to note that the monomer is the pharmacologically active unit [121]. When I inject an insulin sample onto a SEC column, I assume—due to the concentration dependence of insulin self-association [250]—that the dilution into the column’s ﬂowing mobile phase intrinsically triggers the insulin hexamers to dissociate and ultimately to releaseboundzincions(Zn2+) and the various phenolic additives. Based on observations byTeskaetal.[248],Ihy- pothesized two fundamental explanations for the observed SEC column performance issues: i) either the phenolic preservatives were accumulating on the silica, creating amorehydrophobicsurface;and/or,ii)thereleasedzincions were modifying (reacting with) the silica end caps resulting in a modiﬁed surface (i.e. –Si-O-Capped

94 surface to produce –Si-OH groups). Either condition would result in degraded column performance, as they both would introduce unwanted modes of interaction between analytes and the column resin.

Materials and Methods

Materials

Insulin lispro (Humalog Lot: CO74516A; Eli Lilly, Indianapolis, Indiana) was purchased from a local pharmacy. All laboratory chemicals used were analytical grade or higher. Insulin lispro and all chemicals were used before their expiry date. Wa- ter used in mobile phases, formulations and buffers was purified through a Millipore Synergy UV (Millipore, Billerica, MA) filtration unit (MilliQ). Deionized (DI) water was used to rinse the liquid-liquid extraction sample vials.

Silica Resin Incubation

Tosoh G2000SWXL top-oﬀ resin (2.0 g; Tosoh, King of Prussia, PA) was gently shaken to suspend the resin in the manufacturer storage solution and pipetted into a 9mLglasstesttube.Thetubewascentrifuged(2000g,5.0min)andthesupernatant was discarded; the resin was rinsed by adding MilliQ water (3.0 mL) to the settled resin and re-suspended by gentle shaking. For each sample, this process was repeated three times to ensure full removal of the storage solution. After the ﬁnal rinse, the

supernatant was discarded and 3.0 mL of either MilliQ water; 147 uM ZnCl2 in water;

147 uM ZnCl2 and 440 uM EDTA in water; or a 14 mM Na3PO4, 174 mM glycerol, 30 mM meta-cresol in water (referred to as lispro buﬀer) was added. The tubes were capped and shaken to re-suspend the silica resin. Samples were incubated (18-22◦C; 24 hr) on a test tube rotator set to 40 RPM to ensure adequate suspension of the resin during the incubation period. Used resin was removed from degraded columns from the previous stability study [248]; columns were unpacked using a spatula and

95 the dry unpacked resin was slurried in the smallest possible volume of MilliQ water to produce a mixture which could be easily pipetted (approximately 3.0-5.0 mL). No further treatment was implemented on these samples.

Liquid-Liquid Extraction

Used and treated resin samples were extracted by adding suspended resin slurry (approximately 5.0 mL) to a separatory funnel (250 mL), ethyl acetate (EtOAc; 2 vol) was added to the funnel, capped and vigorously shaken. After settling, the aqueous- slurry layer was removed and saved. The organic layer was transferred to a ﬂask and the extraction process was repeated three times. The organiclayerswerecombined, dried over anhydrous MgSO4, ﬁltered and concentrated under reduced pressure (i.e. rotary evaporation). The residue weight was determined and the sample reconstituted in dichloromethane (1.0 mL) and analyzed by GC/MS, or diluted in CDCl3 (1.0 mL) and analyzed by NMR.

Size-Exclusion Chromatography

SEC was performed on an Agilent 1100/1200 HPLC (Agilent Technologies, Santa Clara, CA) detecting UV absorbance (280 nm). Multi-angle light scattering (MALS) measurements were made with a Wyatt DAWN EOS 18-angle MALS detector (Wy- att Technology Corp., Santa Barbra, CA) plumbed in series with the UV detector. Molecular weight calculations were done in Wyatt’s ASTRA software (version 5.3.4). The mobile phase was degassed and filtered (0.2 um filter) before use. Samples were centrifuged at (9650 g, 10 min) before injection (50 uL) to remove large insoluble aggregates. A Tosoh TSK-gel G2000SWXL SEC column (7.8 mm x 300 mm; Tosoh Corp., King of Prussia, PA) was employed at a flow rate of 0.5 mL/min using 0.3 M NaCl, 0.1 M Na3PO4, pH 7.4 with and without ethylene-diamine-tetra-acetic acid (EDTA; 440 uM) as the mobile phase (MP); the HPLC run time was 54 min. A sat-

96 urated uracil-water solution (24 uL) was added to bovine serum albumin (BSA) (1.0 mL, 2.0 mg/mL; Thermo Scientiﬁc, Rockford, IL) and used as my SEC reference standard, and referred to as BSAU. The injection sequence used was as follows: BSAU, 1 injection; insulin lispro, 8 injections; BSAU, 1 injection. This sequence was repeated (24 times) for each mobile phase (MP) condition (with and without 440 um EDTA) resulting in a total of 192 insulin lispro injections and 48 BSAU injections per mobile phase condition. Samples were stored in amber HPLC vials in theauto-samplerat room temperature (18-22◦C) over the nine day experimental analysis duration. These conditions were well within the acceptable temperature conditions and time limits on the prescribing information for insulin lispro [229]. Data were exported from Agilent’s ChemStation software and analyzed with custom scripts written in Matlab and Perl.

Nuclear Magnetic Resonance and Gas Chromatography-Mass Spectrometry

The 1H (400 MHz) and 13C NMR (100 MHz) spectra were recorded using a400

MHz Bruker NMR, Avance III 400 in CDCl3 containing tetra-methyl-silane (TMS; δ =0.0) as an internal standard. Gas chromatography-mass spectrometry (GC/MS) was performed on a Shimadzu GC/MS-QP2010 Plus gas chromatograph mass spectrometer (Shimadzu Scientific Instruments, Inc., Columbia, MD). The GC2010 was equipped with an Rtx-5MS column (0.25 um thickness, 30.0 meter, and diameter of 0.25 mm). The settings were as follows: column temperature was initially 40◦Cwith an injector temperature of 275◦C; the temperature was held at 40◦Cfor4.0minand then ramped at 10◦Cminto280◦C, and then to 300◦Cat2◦C/min and held for 18 min. The ion source temperature was set at 250◦Candtheinterfaceat275◦C. Mass was scanned from 50–750 m/z from 7.0–56.0 min with a scan speed set at 5000. High grade helium was used as the carrier gas and operated in the split-less mode with a pressure of 70.1 kPa, total flow of 14.4 mL/min, column flow was 1.03 mL/min, linear velocity was 36.7 cm/sec, and a purge flow of 3.0 mL/min. Final data were baseline

97 corrected using an asymmetric least squares method implemented in Matlab [251].

Results and Discussion

Size-exclusion chromatography is an important analytical technique but one requires robust methods in order to meet the demands of research and quality/process control. In brief, SEC separations occur fundamentally based upon theanalyte’saccessible volume. Columns are packed with silica particles that have a characteristic size and porosity. Large analytes, for example, are too big (bulky) to enter into the silica particle pores and, as a result, are not readily retained by the column and elute first. Smaller analytes may enter or interact with the pores, which result in longer elution times. Manipulating the overall particle and pore sizes, a researcher may alter the column features to obtain a desired separation. Since silica particles have hydroxyl (Si-OH) groups, many SEC column manufacturers use proprietary techniques to chemically derivatize and “cap” the hydroxy ends of the silica particles (i.e. –Si-O-Capped); a process performed with the intent to avoid ionic or hydrogen bonding interactions with the analytes [131–133]. Generally, this capping eliminates many unfavorable column-analyte interactions, but, in practice, it is very difficult to fully eliminate the effect of exposed surface charge. As a result, researchers will commonly add salt(s) to the mobile phase in order to assist in shielding the analyte from the charged uncapped groups on the surface of the silica [58, 252–255]. This requirement can be problematic, as additional salt can alteraggregationorassembly state of the analyte and confound the SEC analysis [256]. Therefore, when one develops an SEC method, one can spend a significant amount of time optimizing the mobile phase conditions. In the current study, I further investigated insulin formulations via SEC analysis. In a previous study, insulin lispro reference samples were injected onto the SEC column and used to monitor column performance throughout thestudy;thesedata

98 are shown in Figure 37. These reference samples were stored in the product vial at 4 ± 1◦Candinjectedonceforeveryeightexperimentalinsulinsample injections. After approximately 80 total injections, column performanceissueswereapparent (Figure 37; yellow traces) and clearly visible after approximately 100 insulin injections (Figure 37, red traces) [248]. As the injection number increased, the main insulin peak migrated to a longer elution time and developed a post-main peak shoulder ultimately becoming a separate peak (24-27 min). Reversed-phase chromatography (RPC), run in conjunction with SEC during this study, showed no fragments in these samples (data not shown), and was therefore consistent with the notion that the TSK-G2000SWXL column was being degraded during the study. As discussed in the introduction, since commercially available insulin formulations contain phenolic preservatives and zinc ions, I speculated that one or both of these may be triggering the observed SEC column performance issues. To conﬁrm that dilution upon injection of insulin lispro ontotheSECcolumn resulted in the dissociation of hexamers into a smaller-order species—thereby releasing bound zinc ions into solution—I measured multi-angle light scattering (MALS) during column elution (Figure 38). The main insulin peak eluted with aweight-average molecular weight of 14.2±0.3 kDa; results which support the notion that during elution on the SEC column, insulin lispro was minimally assembled and no longer tightly bound the zinc ions found in the hexamer—freeing them to enter into the solution. Iunpackedandextractedresin(seematerialsandmethodssection) from used columns, which displayed the observed SEC column performance issues. The extract from unused (i.e. new) versus used resin were analyzed via classical NMR methods. Representative analysis of the used extracted resin are depicted in Figure 39A (1H-NMR) and Figure 39B (13C-NMR). I did not observe any characteristic aromatic C-H signals (1H-NMR; 7-8 ppm chemical shift range) or aromatic carbon signals(13C-

99 Figure 37: Sixteen insulin lispro drug product peaks eluted from a Tosoh G2000SWXL column during a 128-injection stability study by Teska et al. (2014). Injections are colored from blue to red in order of increasing injection number. Peak position shifted to longer retention times with a post-main peak shoulder becoming very apparent in later injections (24- 27 min). The main insulin peak elutes with a weight average-molecular weight of 14.2±0.3 kDa.

NMR 100–140 ppm range), providing no evidence for the accumulation of phenolic compounds (i.e. meta-cresol) on the column resin. However, the data clearly illus- trated the presence of aliphatic organic compound(s); theseprotonsignalswerenot observed in the extracted unused resin samples (data not shown). Therefore, these NMR data illustrate that degraded proprietary end-cap material could be observed in material extracted from used columns. Furthermore, when the used resin extracts were analyzed by GC/MS, I observed evidence for a variety of small molecular weight compounds (Figure 40, blue trace), but meta-cresol (the preservative in the insulin lispro drug product) was not observed. Hence, these data do notsupportthenotion

100 Figure 38: SEC-MALS data for insulin lispro. The main insulin peak elutes with a weight average-molecular weight of 14.2±0.3 kDa. The expected molecular weight of insulin lispro in the assembled hexamer is 34.8 kDa. that phenolic compounds are accumulating on the column and causing additional non-specific hydrophobic interactions between the insulin analyte and the column surface that could explain the change in chromatographic signal. Inexttreatednew(unusedresin)withmeta-cresol(presentin the lispro buffer) or zinc chloride at concentrations consistent with the insulin lispro drug product. As shown in Figure 40, I present GC/MS chromatograms from extracted samples obtained from new resin treated with i) water only; ii) meta-cresol (in the insulin lispro formulation buffer); iii) ZnCl2 in water; and iv) used resin. Compared to used resin, unused extracted resin (Figure 2, water treated control, black trace) displayed evidence for only a few volatile components, whereas the meta-cresol treated resin clearly presented meta-cresol at tR 11 min. When the new resin was treated with ZnCl2,I observed a variety of volatile organic components also observed from extracted used resin. However, as compared to the used resin samples, the ZnCl2 treated resin samples displayed much lower concentrations. I attribute the differences to two major

101 factors: i) I did not control for zinc exposure in the used resin samples, the used column material had been exposed to far more zinc ions than my new resin zinc ion doping experiments due to the many column injections in the previous stability study; and ii) the used columns were stored for a long period of time (greater than 12 months) after being initially used; hence, it is likely that additional degradation events occurred over time. At this point, the data appeared consistent with the notion that zinc ions were the primary culprit causing the SEC performance issues. If correct, then the addition of a chelator (e.g. EDTA) to the mobile phase should help reducetherateofcolumn

degradation. To test this, I treated new resin with ZnCl2 and EDTA, using a three- fold molar excess of EDTA relative to ZnCl2. I then repeated the GC/MS analysis (Figure 41) and while the resin leached a number of species when treated with water, two species observed at ∼17.3 and ∼18.6 min were significantly decreased when EDTA was included. These results confirm that the column degradation was due, at least in part, to end-cap degradation facilitated by zinc ions. From these observations, I asked the final question: by adding EDTA to the mobile phase, can I maintain better and extended SEC column performance? To address this, I used new columns and compared SEC analysis performed in the presence (Figure 42A) and absence (Figure 42B) of EDTA. When EDTA was present, I observed limited modification to the insulin peak over 240 injections (192 insulin lispro, 48 BSAU); the BSAU standard was completely unaffected (data not shown). However, when I switched to the mobile phase without EDTA, I observed a steady growth in thepost-mainpeak shoulder indicating column degradation. These data are further evidence that the released zinc ions are catalyzing column degradation.

102 Conclusion

My experiments demonstrate that: i) zinc ions are a significant contributor to the observed SEC column degradation when performing insulin sample analysis; and ii) the addition of EDTA, a zinc ion chelating agent, to the mobile phase markedly helped to maintain column performance. At least for zinc-containing insulin samples using these SEC columns, I recommend the addition of EDTA to themobilephase in order to protect the column and help ensure artifacts—due to column degradation events—are not misinterpreted as physiochemical changes in the insulin analyte. These findings also present the possibility that other metals commonly incorporated into biopharmaceutical production, such as tungsten from staked needles in prefilled syringes or copper which may be utilized in cell culture methods, may also possibly contribute to poor analytical SEC column performance. Therefore, I encourage others to investigate metal-catalyzed column surface degradation as a possible issue when developing robust SEC methods.

Acknowledgements

I would like to thank Dr. Nicole Payton for her help and advice regarding GC/MS usage. I would also like to thank my reviewers at the Journal of Pharmaceutical Sciences for their insightful and helpful comments. This research utilized services of the Medicinal Chemistry Core Facility (MCCF) housed within the Department of Pharmaceutical Sciences (DOPS) at the University of Colorado Anschutz Medi- cal Campus. The MCCF receives funding via Colorado Clinical and Translational Sciences Institute grant NIH-NCATS, UL1TR001082.

103 Figure 39: Used extracted resin: (A) 1H-NMR; (B) 13C-NMR. The observed NMR signals were not present in unused resin.

104 Figure 40: GC/MS chromatogram overlays: Extracted used resin (blue); new resin treated with ZnCl2 (green); new resin treated with insulin lispro buﬀer containing meta- cresol (red); and new resin treated with MilliQ water (black). The peak eluting at 11.4 min (red) is meta-cresol.

Figure 41: GC/MS chromatogram overlay: Water treated (green), ZnCl2 treated (blue), and ZnCl2 +EDTA(red)treatednewresinsamples.Specieselutingatapproximately 17.3 and 18.6 min (black arrows) were signiﬁcantly lower in EDTA treatment group.

105 Figure 42: SEC chromatograms of 192 injections of insulin lispro eluted with EDTA (440 uM) in the MP (A) and 192 injections of insulin lispro eluted without EDTA in the MP (B). Injections are colored from blue to red with increasing injection number. Insulin lispro elutes between 20-35 min at a weight-average molecular weight of 14.2±0.3 kDa, and meta-cresol elutes between 35-50 min.

106 CHAPTER V

CONCLUSIONS AND FUTURE DIRECTIONS

Chapter II: New Surfaces in Pre-ﬁlled Syringes

Surfaces in pre-filled syringes can have a substantial directeffect on a protein therapeutic in formulation. In investigating a novel fluoropolymer-based surface, I found that alternative materials can indeed provide significant improvements on the traditional siliconized syringe technology with respect to agitation-induced protein aggregation. In comparison to a traditional siliconized pre-filled syringe, bare glass syringes using the fluoropolymer-surfaced plunger showed much lower levels of sub-visible particle formation during agitation. As discussed in Chapter I, protein aggregates and particles are linked to immunogenicity and this finding highlights the value of new technology in addressing the pressing need to eliminate silicone oil from pre-filled syringes. During the work done in this chapter, I investigated many different techniques and methods to further probe the interactions between the model protein and the fluoropolymer surface. However, the results of these efforts were hampered by limitations in the analytical techniques available to me at the time. Specifically, many of the analytical techniques I attempted to use to measure adsorption failed because of two key incompatibilities with the surface and the technique: the thickness of the fluoropolymer surface, and it’s roughness. The fluoropolymer surface explored in Chapter II was on the order of ∼20µm thick and had variations in surface topology on the order of 2-4µm (estimated from the confocal microscopy experiments). These variations in surface topology may have been inherent to the fluoropolymer surface material or they may have arisen from the process used to affixthesurfacetothemicroscopeslides used for confocal

107 microscopy. Due to the proprietary nature of the fluoropolymer surface, samples of the surface affixed to glass microscope slides were prepared by Gore—I was only able to handle the final prepared surface samples. Many alternative techniques to microscopy I attempted, such as SPR and QCM(-D), failed because of the thickness of the fluoropolymer surface. Others, such as TIRFM and ellipsometry, failed due to the magnitude of the variations in the surface topology. In the coming years, I would expect that more surfaces, similar in nature5 to the fluoropolymer surface I studied will emerge—especially given the promising results during agitation of our model protein with this surface. It may be possible to reduce the “roughness” of the surface by alternate/improved methods of affixing the surface to a glass microscope slide. Reducing the surface roughness would potentially make TIRFM or ellipsometry6 measurements possible, which could provide additional information on both the extent and kinetics of protein adsorption to the surface. The field as a whole would benefit from the development of techniques to measure binding to thick7 surfaces would greatly contribute to the field’s collective ability to investigate protein-surface interactions in delivery devices. SPR would seem to be the most suitable technique for extension to thicker, rough surfaces. Currently, SPR uses the electron resonance in thin gold film to detect adsorption events up to 300nm from the surface of the gold film. Although not my area of expertise, it may be possible that there exists other elements or other metals with a greater magnitude of resonance than gold that would allow for a greater depth of signal penetration in SPR. In theory, this would allow the measurement of adsorption to thicker surfaces or the measurement of interaction events further from a thin surface. 5Or at least bearing similar physical/structural properties such as to experience the same issues I experienced finding a suitable technique with which to measure protein adsorption to the fluoropolymer surface. 6I should note that ellipsometry is extraordinarily sensitive to surface roughness and may not be a suitable technique for the majority of surfaces a protein may experience in the “real world” and may be best for measuring protein interactions with chemically prepared surfaces. 7For the purposes of this discussion, “thick” is defined as surfaces greater than 2 microns in thickness.

108 If technology such as my hypothetical “deep penetration SPR”discussedabove existed, it would be interesting to examine (or re-examine) the following areas:

1. Real-time adsorption of IVIG to the ﬂuoropolymer surface under the conditions:

(a) Without surfactant included in formulation

(b) With surfactant included in formulation

2. Real-time adsorption of IVIG to the ﬂuoropolymer surface with varying levels of formulation ionic strength

3. Real-time adsorption of IVIG to the ﬂuoropolymer surface asafunctionofpH

4. Measure the real-time desorption of IVIG already adsorbed to the ﬂuoropolymer surface by:

(a) Formulations with and without surfactant

(b) Formulations of varying ionic strengths

In addition to more robust techniques to measure protein adsorption to more of the diﬀerent surfaces therapeutic protein formulations are likely to encounter, improvements in analytical techniques to measure the adsorbed protein’s structure would also enhance the ﬁeld’s inderstanding of protein-surface interactions. At the time of writ- ing, the only techniques that can give direct information on the structure of protein adsorbed to a surface are: grazing-angle FTIR, FTIR-ATR, grazing-angle CD and spectroscopic ellipsometry. Both grazing-angle techniques are sensitive to the surface roughness and are less commonly used, presumably because they are challenging to apply both with respect to surface and with respect to real world formulation conditions (i.e. concentration, solution composition, etc.). As discussed above, ellipsometry is even more sensitive to surface roughness, leaving only FTIR-ATR as the most

109 generally applicable technique, robust to both surface type(solongasthesurface material does not adsorb in the amide I/II spectral region) and protein formulation. In this particular mode of use, the application of FTIR-ATR to studying adsorbed protein structure is hampered by FTIR-ATR’s relatively high lower limit of protein concentration (>5mg/mL) needed for adequate signal. Although the protein in the gel layer adsorbed to the surface is relatively concentrated (in comparison to the same thickness “slice” from the bulk solution), the signal from the adsorbed protein may not be strong enough to be detected over the signal from the bulk solution. However, there is a new FTIR instrument, named the Prota-3S, being developed by BioTools Inc specifically for the study of proteins. The Prota-3S has numerous enhancements over a “standard” FTIR including: an MCT detector with upgraded electronics that lower overall measurement noise, and included optical bandpass filters that nearly eliminate the non-linearity inherent to MCT detectors and enhance signal-to-noise in the amide I-IV spectral region. In concert, these enhancements resultinanincreasedsensitivity of approximately an order of magnitude over traditional FTIR instruments, bringing the lower limit of detection down to ∼0.5mg/mL. The added sensitivity of this new instrument makes it posible to better measure the structure of proteins adsorbed to different surfaces. I was able to test a prototype of this instrument to measure thestructureof adsorbed IVIG to the fluoropolymer surface during this work. The data I generated from this early prototype for this work was inconclusive, however, numerous other enhancements have been made since my initial work. My basic measurement technique was to apply a drop of protein solution to the ATR crystal, clamp the fluoropolymer surface to the crystal and make a spectral measurement. In this setup, it was not possible to deconvolute the differential contribution to thespectralsignalfromthe adsorbed protein layer and the small amount of bulk that remained between the crystal and the surface. Given the results of the confocal microscopy experiments (which

110 were completed after the FTIR measurements were made), this could be attained by incubating the fluoropolymer surface with IVIG, rinsing the surface, applying a drop of buffer to the ATR crystal, clamping the rinsed surface to thecrystalandmea- suring the IR spectrum. This experiment would give the structure of the adsorbed layer without interference from any residual protein in the bulk solution. This experiment would also inform future studies such as the effect of formulation conditions on adsorbed structure, the kinetics of adsorbed protein structure change and others. In addition to better instruments to measure protein structure and adsorption, the entire field of therapeutic proteins would benefit from more robust techniques/instruments to measure protein particle formation. During the agitation study, I used MFI and RMM to measure particle formation. RMM is a unique and interesting semi- orthogonal technique to MFI in that it measures particles based on buoyant mass as opposed to optically (as the MFI does). In theory, measuring buoyant mass allows one to discriminate between silicone oil droplets (positively buoyant) and protein particles (negatively buoyant). However, in practice, I found this technique not to be suitableforuseduringthe agitation study in Chapter II. First, as discussed in Chapter II, silicone oil droplets in a protein solution are almost immediately coated with adsorbed protein. Protein adsorption to silicone oil droplets not only changes their overall density, making their size estimates by RMM inaccurate but it also presents the possibility that the overall density of the protein-silicone oil particle may be neutrally buoyant, or at least below the limit of detection of the RMM instrument. This limitation could mean that a sizable portion of the particles detected by the MFI may not bedetectedduring RMM analysis. Second, the microfluidics in the RMM instrumentreadilyretainlarger particles and clog very easily. During my studies I found it necessary to filter samples through a 5µm syringe filter before analysis to help mitigate incessant clogging of the instrument during analysis.

111 These two limitations—to me—leave very serious questions astotheaccuracy, repeatability and overall validity of data generated by the RMM for the type of samples I studied. The unsuitability of RMM for samples generated during the agitation study leaves MFI as the only technique for measuring particles in the 1 to 150 micron size range, which limits the certainty with which one can makeassertionsbasedon particle data. The development of additional techniques to measure particle formation in therapeutic protein formulations, and the reﬁnementofexistingtechniques such as RMM, would reduce the uncertainty of particle measurements and help further enhance ﬁeld’s understanding of particle formation as adegradationpathway for therapeutic proteins. A better understanding of protein particle formation will lead to better strategies and techniques to minimize particle formation in therapeutic protein products and—ultimately—better and safter therapeutic protein products for patients.

Chapter III: Insulin Analog Assembly

Before the publication of the work described in Chapter III on the stability of insulin analogs with depleted levels of phenolic preservatives, differences in assembly state between the three different analogs I studied were not publicly known—although it is very likely that the companies that developed each respective analog had knowledge of their own molecule’s assembly behavior. I showed that not only is each analog its own, distinct molecule—behaving differently from each other and from human insulin— but that each had very different and unique susceptibility to the universal problem of phenolic preservative depletion by the catheter surfaces in continuous insulin infusion pump. This finding, along with the enduring issue of phenolic depletion in insulin pump catheters, highlights the need to examine the complex and multifactorial issue of insulin analog stability more closely. My hope is that my findings will: spur the development of catheter plastics that minimize the amount of absorbed phenolics from

112 the formulation, inform prescribing decisions for patientsusingtheinsulinpump,and encourage a wider discussion of, and research on, the differences in behavior between different insulin analogs. In addition to elucidating the stability and overall self-assembly differences between the three insulin analogs discussed above, my work in Chapter III was—to my knowledge—the first published report of solution insulin assembly using direct boundary modeling of SV-AUC data. SV-AUC is the “gold standard” for determining the solution assembly state for protein molecules. Despite the extreme g-forces, from the protein’s perspective, SV-AUC is actually a very gentle technique requir- ing neither a specific running buffer (that can itself induce conformation or assembly changes) nor a stationary phase (as is the case with SEC). And, with direct boundary modeling techniques, such as those employed in the work described in Chapter III, much more information is able to be gleaned from a given set of experiments. The numerical techniques used to model the boundary movement in SV-AUC data can be very computationally intensive. And, the computational power to widely apply this method to SV-AUC data has only become commonplace in the past decade or so. Previous to the work done in this thesis, less computationally intense and lower resolution numerical methods were employed when analyzing insulin analog assembly by SV-AUC. The SV-AUC experiments done in Chapter III, were used in an attempt to better understand and explain the differences I observed in thermal stability with different levels of phenolic preservative between the three insulin analogs I studied. The interpretation of the SV-AUC data for each of the three differentinsulinanalogs presented in Chapter III was intended to show overall changes in assembly of the different insulin analogs without necessarily determining or subscribing to a particular mechanism of self-association. However, where appropriate, I provided molecular weight estimates for some of the sedimenting species in ordertoprovideareference

113 point for the prevailing “dogma” of insulin self-association following the assembly mechanism originally based on crystal structure data: monomer ↔ dimer ↔ hexamer (M ↔ D ↔ H). This view of insulin self-assembly has been entrenched, inpart, through the prevalent (and popular) publications of Jens Brange [240, 257] and the novelty of crystal structure data at the time the mechanism was proposed [258, 259]. However, a number of studies over many years have questioned the blanket validity of the M ↔ D ↔ H the assembly scheme for insulin. Using concentration gradient static light scattering, Atri et al. [250], expanding on workdonebyMilthropeet al. [260] and have suggested that bovine insulin hexamers indefinitely self-associate (termed isodesmic self-association), an assembly scheme originally proposed for zinc- free, porcine insulin by Pekar and Frank in 1972 [261]. Many others have also proposed isodesmic self-association models, using either the monomer [262] or the dimer [263] as the base unit in the association scheme. In addition, some researchers have also purported to show the existence of insulin tetramers as a intermediate between dimer and hexamer assembly [188, 264]. In light of the many studies challenging the M ↔ D ↔ H assembly scheme, the validity of the M ↔ D ↔ H assembly scheme for insulin may be questionable, or at very least, is still the subject of active debate. Furthermore, my results on the differences between the three insulin analogs I studied, both in their susceptibility to thermal degradation and in overall self-assembly with different levels of phenolic preservatives, suggests that it may be erroneous to assume that insulin analogs self- assemble in an identical scheme to human insulin—even though they perform the same biological/pharmaceutical function. Therefore, outside the “dogma” of insulin assembly, one could approach the interpretation of the SV-AUC data presented in this work (shown in Figure 34) from multiple different molecular perspectives. To inform the molecular interpretation of SV-AUC data, it is helpful to calculate the theoretical limits of sedimentation coefficients for each potential component in

114 the insulin self-assembly scheme to help guide in the interpretation of these data. The observed sedimentation coefficient for a protein is a product of both the molecular weight of the protein and its hydrodynamic shape in solution. Therefore, for a protein of a given molecular weight—assuming ideal sedimentation behavior and no interactions between protein molecules—only changes to the protein’s shape will change its sedimentation coefficient. For theoretical calculations, the shape with the lowest drag and therefore highest sedimentation coefficient, is a perfectly compact sphere; the shape with the highest drag, and therefore the smallest sedimentation coefficient, is a completely unfolded protein. For a given protein of molecular weight M, the theoretical sedimentation coef-

ﬁcient for a completely unfolded protein, s20,w,unfolded, can be calculated as follows [265]:

0.457 s20,w,unfolded =0.296(MkDa) (3)

where MkDa is the molecular weight of the protein units of kilodaltons. Similarly, for a given protein of molecular weight M, the theoretical sedimentation coeﬃcient for a compact sphere can also be calculated [266]. We begin by calculating the volume of a single protein molecule:

M v¯ V = Da (4) NA

where MDa is the molecular weight of the protein in Daltons,v ¯ is the partial speciﬁc volume, and NA is Avagadro’s number. Next, the volume of hydration, Vh,around the protein is calculated:

MDah Vh = (5) NA where h is the hydration ratio—the mass of water per mass of protein (in the following calculations it is assumed to be 0.3). Next, the radius, r, of a sphere of total volume,

115 VT , where VT = V + Vh, is calculated as:

3V 1/3 r = T (6) ! 4π "

Finally, the frictional coeﬃcient, f, for a sphere of radius r is calculated as:

f =6πηr (7)

where η is the viscosity of the solution (in Poise). Equations 4, 5, 6, and 7 can be

combined with the relation VT = V + Vh, and simpliﬁed to yield:

3M(¯v + h) 1/3 f =6πη (8) ! 4NAπ " which can then be used in the Svedberg equation:

MDa(1 − v¯ρ) ssphere = (9) NAf where ρ is the solution density, and f is the frictional coefficient calculated in Equa- tion 8. Using the density (1 g/mL) and viscosity (0.01 Poise) values for water at 20◦C yields the sedimentation coefficient at standard conditions, s20,w,sphere. Using the c(s) distribution for insulin glulisine (Figure 34 panel C) as an example, there are a number of molecular interpretations that could bederivedfromthese sedimentation profiles. For example, using Equations 9 and 3, with the following parameters for insulin glulisine MDa =5823Da,ρ =1g/mL,η =0.01 Poise and v¯ =0.7239 mL/g, the theoretical limits for insulin glulisine sedimentation coefficients can be estimated for multiple possible self-assemblies of insulin glulisine and are shown in Table 3. In consideration of these theoretical limits for different insulin glulisine self-assemblies,

116 Table 3: Theoretical sedimentation coeﬃcient limits for diﬀerent self- assemblies of insulin glulisine.

s20,w,theoretical MW Unfoldeda Sphereb Monomer 5,823 0.66 1.06 Dimer 11,646 0.91c 1.68 Tetramer 23,292 1.25 2.67 Hexamer 34,938 1.50 3.50 12-mer 69,876 2.06 5.56 18-mer 104,814 2.48 7.29 a Calculated using Equation 3. b Calculated using Equation 9. c A “completely unfolded” dimer (or larger assembly) is a theoretical contrivance and not a physically present species. However, one might interpret this value as the completely unfolded structure if all the monomeric units in a dimer were unfolded and cross-linked between their N- and C-termini. in Figure 34 panel C, in the sample with 0% phenolic (red trace), sedimentation of insulin glulisine shows at least two species, one sedimenting at 1.3S and a wider, less resolved peak sedimenting from ∼2–3S.Fortherangeof∼2–3S, it is possible—in theory—that species in this range could vary in molecular weight from tetramer all the way to 18-mer depending on their shape. Therefore, it is possible that this less resolved peak could represent a number of distinct molecular mechanisms. First, this may may be a result of rapid (on the time scale of the SV-AUC run), reversible, self-association between multiple species, perhaps a 12-mer dissociating to two hexamers. Second, this peak may also represent a single assemblystate,forexample a hexamer, that could be in multiple structural conformations, causing it to have different overall shapes and thus sediment at different rates. Third, in consideration of the many isodesmic self-association models that have been proposed, it is possible that insulin glulisine self-assembles into multiple differently assembled hexamers, thus each hexamer may have a different assembled configuration and consequently multiple different hydrodynamic shapes—even though each hexameric assembly has an identical overall molecular weight.

117 From this data set alone, it is impossible to unambiguously determine which possible self-assembly mechanism is most probable for each insulin analog. However, elucidating the differences in assembly scheme—especially with respect to phenolic preservative concentration—for these three insulin analogs would be a very interesting study, and may give further insights into the differences in stability between these three insulin analogs. Two additional experiments would provide a more complete data set that would allow for a better molecular interpretation of the effect of phenol on the assembly of these three insulin analogs. First, one could run SV-AUC at several different insulin analog concentrations. Using insulin glulisine sedimentinginformulationcontaining 0% phenolics as an example, if increasing the protein concentration resulted in a shift of the species sedimenting from ∼2–3S towards higher S-values, this would support the possibility that rapid self-assembly was happening over the time course of the SV- AUC run. This conclusion would be further supported if by decreasing the protein concentration resulted in a shift in the species sedimentingfrom∼2–3S towards lower S-values. This experiment could also be repeated with different phenolic preservative concentrations to assess the effects of phenolic preservative concentration on insulin self-assembly. Additionally, this experiment could be extended to assess the strength of the hexamer coordination around the zinc ions in insulin lispro or insulin aspart (should that be found to be one of the assembled species in solution), by altering the salt concentration in the formulation over a range of concentrations. Additionally, one could use a newer form of direct boundary modeling on the SV-AUC data. As discussed in Chapter III, the direct boundary modeling technique applied to the SV-AUC data presented in this work relies on the assumption that each sedimenting species has the same overall shape (manifest mathematically in the overall shape parameter, f/f0). Recent advances in direct boundary modeling of SV-AUC data allow for the relaxation of the assumption of equivalentshapeforallsedimenting

118 species [267] by allowing the shape parameter, f/f0, to vary over a specified range of values when fitting the data (the resulting fit is termed a c(s, ∗) distribution). This method has been applied to a number of other associating protein systems and has shown promising results [268–270]. Using the sedimentation of insulin glulisine in the formulation with 0% phenolics again as an example (Figure 34 panel C, red trace), the species sedimenting from ∼2–3S could potentially be a single species, for example a hexamer, with multiple different structures and subsequently, multiple

different hydrodynamic shapes (i.e. f/f0 values). In this case, fitting these data using a c(s, ∗)distributionwouldshowmanyspeciesofthesamemolecularweight but with different shapes, allowing the deconvolution of shape and molecular weight information from the fitted sedimentation coefficient. Second, one could run sedimentation equilibrium analytical ultracentrifugation (SE-AUC). In comparison to SV-AUC, SE-AUC allows one to look at thedistribu- tion of species in a samples without the impact of the hydrodynamic shape of the molecules in solution on the interpretation. Running SE-AUC at multiple insulin analog concentrations and phenolic preservative concentrations would allow the global fitting of a specific model of association. The SE-AUC data could also be combined with the SV-AUC data and globally fit using SEDPHAT8, to give an even more complete picture of the self-association scheme of these insulin analogs.

Chapter IV: Eﬀects of Formulations on Surfaces

Juxtapose to my ﬁndings in Chapter III, where the surface altered the therapeutic protein’s formulation, was the work done in Chapter IV where Ifoundthatthefor- mulation itself could also modify surfaces with which it cameincontact.Inthis particular example, I found that certain insulin analog formulations could modify the specialized surface found in size exclusion chromatography (SEC) column resin.

8SEDPHAT is freely available from the National Institutes of Health: http://www. analyticalultracentrifugation.com/sedphat/sedphat.htm

119 In SEC, peak shape and retention time are critical parametersindeterminingthe identity and distribution of the different species within a sample. The SEC analytical technique is used extensively throughout biotechnologyandmaximizingcolumn lifetime is a important topic. During work in Chapter III, I noticed significant degradation in the chromatography column peak shape and retention time over a relatively small number of insulin sample injections—an observation I later found to be a very common and long- standing issue among scientist who have worked on therapeutic insulin extensively. Most experts I talked to on the subject suspected that the phenolic preservatives were accumulating on the column, creating a more hydrophobicsurfacethatresulted in the changes to peak shape and retention time. I investigated this phenomenon and found that it was actually the zinc ions, included in some insulin analog formulations as a stabilizer, that chemically modified the surfaceofthesilicaresininthe chromatography columns, permanently destroying their performance. By including a chelator in the mobile phase, I was able to prevent the deleterious effects of the free zinc ions on the column resin and preserve column performance. This work highlights the importance of the common, but crucial, step of mobile phase optimization during size exclusion chromatography method development. Additionally, my work highlights the need to assess the robustness of a particular column for the particular samples one intends to analyze in addition to selecting columns based on separation characteristics. Finally, my work in this chapter also highlights the need for more robust chemistries used in the passivation of the charged silica resin used in SEC columns columns.

Concluding Remarks

Developing better, more “inert” surfaces for protein therapeutic delivery devices is currently an under-appreciated need in the ﬁeld of protein therapeutics. Due to the

120 complexity of these surfaces, both structurally and chemically, and the complexity of the interactions between protein therapeutics and these surfaces, the field would benefit from more robust techniques and instruments to measure protein adsorption, desorption and structural perturbations to adsorbed molecules. The better we, as a scientific community, understand the interactions between protein formulations and the surfaces they come in contact with, the better equipped wewillbetomitigate any undesirable interactions. This in turn will lead to the development of better therapeutic protein formulations and the development of better materials and surfaces for delivery devices that minimize aggregation of protein therapeutics for patients in the future.

121 REFERENCES

[1] Benjamin Leader, Quentin J Baca, and David E Golan. Protein therapeutics: a summary and pharmacological classiﬁcation. Nature Reviews Drug Discovery, 7(1):21–39, 2008.

[2] Alex K Pavlou and Janice M Reichert. Recombinant protein therapeutics—success rates, market trends and values to 2010. Nature biotechnology, 22(12):1513–1519, 2004.

[3] David B Volkin, Susan Hershenson, Rodney JY Ho, Susumu Uchiyama, Ger- hard Winter, and John F Carpenter. Two decades of publishing excellence in pharmaceutical biotechnology. Journal of pharmaceutical sciences,2014.

[4] Global markets for bioengineered protein drugs. Technical report, bccResearch, 2014.

[5] EC Noble. The eﬀect of pancreatic extract (insulin) on normal rabbits fg bant- ing, ch best, jb collip, jjr macleod and. American Journal of Physiology,62: 162, 1922.

[6] Wolfgang Korte. Catridecacog: a breakthrough in the treatment of congenital factor xiii a-subunit deﬁciency? Journal of blood medicine,5:107,2014.

[7] David J Kuter, James B Bussel, Roger M Lyons, Vinod Pullarkat, Terry B Gernsheimer, Francis M Senecal, Louis M Aledort, James N George, Craig M Kessler, Miguel A Sanz, et al. Eﬃcacy of romiplostim in patients with chronic immune thrombocytopenic purpura: a double-blind randomised controlled trial. The Lancet,371(9610):395–403,2008.

[8] James B Bussel, David J Kuter, Vinod Pullarkat, Roger M Lyons, Matthew Guo, and Janet L Nichol. Safety and eﬃcacy of long-term treatment with romiplostim in thrombocytopenic patients with chronic itp. Blood,113(10): 2161–2171, 2009.

[9] David J Kuter, Mathias Rummel, Ralph Boccia, B Gail Macik, Ingrid Pabinger, Dominik Selleslag, Francesco Rodeghiero, Beng H Chong, XuenaWang,and Dietmar P Berger. Romiplostim or standard of care in patients with immune thrombocytopenia. New England Journal of Medicine,363(20):1889– 1899, 2010.

[10] Wim Jiskoot, Theodore W Randolph, David B Volkin, C Russell Middaugh, Christian Sch¨oneich, Gerhard Winter, Wolfgang Friess, Daan JA Crommelin, and John F Carpenter. Protein instability and immunogenicity: roadblocks to clinical application of injectable protein delivery systems for sustained release. Journal of pharmaceutical sciences,101(3):946–954,2012.

122 [11] Miranda MC van Beers, Wim Jiskoot, and Huub Schellekens. On the role of aggregates in the immunogenicity of recombinant human interferon beta in patients with multiple sclerosis. Journal of Interferon & Cytokine Research,30 (10):767–775, 2010.

[12] Anne S De Groot and David W Scott. Immunogenicity of protein therapeutics. Trends in immunology,28(11):482–490,2007.

[13] Amber Haynes Fradkin, John F Carpenter, and Theodore W Randolph. Im- munogenicity of aggregates of recombinant human growth hormone in mouse models. Journal of pharmaceutical sciences,98(9):3247–3264,2009.

[14] Amy S Rosenberg. Eﬀects of protein aggregates: an immunologic perspective. The AAPS journal, 8(3):E501–E507, 2006.

[15] Andrea Braun, Lia Kwee, Mark A Labow, and Jochem Alsenz. Protein aggregates seem to play a key role among the parameters inﬂuencing the antigenicity of interferon alpha (ifn-α)innormalandtransgenicmice. Pharmaceutical research,14(10):1472–1478,1997.

[16] Miranda MC van Beers, Melody Sauerborn, Francesca Gilli, Vera Brinks, Huub Schellekens, and Wim Jiskoot. Aggregated recombinant human interferon beta induces antibodies but no memory in immune-tolerant transgenic mice. Phar- maceutical research,27(9):1812–1824,2010.

[17] DC Robbins, SM Cooper, SE Fineberg, and PM Mead. Antibodiestocovalent aggregates of insulin in blood of insulin-using diabetic patients. Diabetes,36 (7):838–841, 1987.

[18] G Giovannoni, FE Munschauer, and F Deisenhammer. Neutralising antibodies to interferon beta during the treatment of multiple sclerosis. Journal of Neurology, Neurosurgery & Psychiatry,73(5):465–469,2002.

[19] P Duquette, M Girard, R Dubois, RL Kobler, F Lublin, L Kelley, GS Francis, M Freedman, JI Greenstein, B Mishra, et al. Neutralizing antibodies during treatment of multiple sclerosis with interferon beta-1b: experience during the ﬁrst three years. Neurology,47(4):889–894,1996.

[20] Charles L Bennett, Stefano Luminari, Allen R Nissenson, Martin S Tallman, Stephen A Klinge, Norene McWilliams, June M McKoy, Benjamin Kim, E Al- lison Lyons, Steve M Triﬁlio, et al. Pure red-cell aplasia and epoetin therapy. New England Journal of Medicine,351(14):1403–1408,2004.

[21] Huub Schellekens. Immunologic mechanisms of epo-associated pure red cell aplasia. Best Practice & Research Clinical Haematology,18(3):473–480,2005.

123 [22] Eva Y Chi, Sampathkumar Krishnan, Theodore W Randolph, and John F Carpenter. Physical stability of proteins in aqueous solution: mechanism and driving forces in nonnative protein aggregation. Pharmaceutical research,20 (9):1325–1336, 2003.

[23] Byeong S Chang and Susan Hershenson. Practical approaches to protein formulation development. In Rational Design of Stable Protein Formulations, pages 1–25. Springer, 2002.

[24] Arnoldus WP Vermeer and Willem Norde. The thermal stability of immunoglobulin: unfolding and aggregation of a multi-domain protein. Biophysi- cal Journal,78(1):394–404,2000.

[25] Pinaki Basu, Sampathkumar Krishnan, Renuka Thirumangalathu, Theodore W Randolph, and John F Carpenter. Igg1 aggregation and particleformation induced by silicone-water interfaces on siliconized borosilicate glass beads: A model for siliconized primary containers. Journal of pharmaceutical sciences, 102(3):852–865, 2013.

[26] Renuka Thirumangalathu, Sampathkumar Krishnan, Margaret Speed Ricci, David N Brems, Theodore W Randolph, and John F Carpenter. Silicone oil-and agitation-induced aggregation of a monoclonal antibody in aqueous solution. Journal of pharmaceutical sciences,98(9):3167–3181,2009.

[27] Ken A Dill. Dominant forces in protein folding. Biochemistry,29(31):7133– 7155, 1990.

[28] A Striolo, D Bratko, JZ Wu, N Elvassore, HW Blanch, and JM Prausnitz. Forces between aqueous nonuniformly charged colloids from molecular simulation. The Journal of chemical physics,116(17):7733–7743,2002.

[29] Tsutomu Arakawa and Serge N Timasheﬀ.Mechanismofprotein salting in and salting out by divalent cation salts: balance between hydration and salt binding. Biochemistry,23(25):5912–5923,1984.

[30] Amos M Tsai, John H van Zanten, and Michael J Betenbaugh. Ii. electrostatic eﬀect in the aggregation of heat-denatured rnase a and implications for protein additive design. Biotechnology and bioengineering,59(3):281–285,1998.

[31] Chiaki Nishimura, Vladimir N Uversky, and Anthony L Fink. Eﬀect of salts on the stability and folding of staphylococcal nuclease. Biochemistry,40(7): 2113–2128, 2001.

[32] Donald S Maclean, Quansheng Qian, and C Russell Middaugh. Stabilization of proteins by low molecular weight multi-ions. Journal of pharmaceutical sciences, 91(10):2220–2229, 2002.

124 [33] RA Curtis, J Ulrich, A Montaser, JM Prausnitz, and HW Blanch. Protein– protein interactions in concentrated electrolyte solutions. Biotechnology and bioengineering,79(4):367–380,2002.

[34] Code of federal regulations title 21, chapter i, subchapter f, part 610: Gen- eral biological products standards, 2014. http://www.accessdata.fda.gov/ scripts/cdrh/cfdocs/cfcfr/CFRSearch.cfm?fr=610.15, Accessed: 2015- 06-20.

[35] Yuh-Fun Maa and Chung C Hsu. Aggregation of recombinant human growth hormone induced by phenolic compounds. International journal of pharmaceutics,140(2):155–168,1996.

[36] Jonas Fransson, Dan Hall´en, and Ebba Florin-Robertsson. Solvent eﬀects on the solubility and physical stability of human insulin-like growth factor i. Phar- maceutical research, 14(5):606–612, 1997.

[37] Richard L Remmele Jr, Nancy S Nightlinger, Subhashini Srinivasan, and Wayne R Gombotz. Interleukin-1 receptor (il-1r) liquid formulation development using diﬀerential scanning calorimetry. Pharmaceutical research,15(2): 200–208, 1998.

[38] Xanthe M Lam, Thomas W Patapoﬀ, and Tue H Nguyen. The eﬀect of benzyl alcohol on recombinant human interferon-γ. Pharmaceutical research,14(6): 725–729, 1997.

[39] Renuka Thirumangalathu, Sampathkumar Krishnan, David N Brems, Theodore W Randolph, and John F Carpenter. Eﬀects of ph, temperature, and sucrose on benzyl alcohol-induced aggregation of recombinant human granulocyte colony stimulating factor. Journal of pharmaceutical sciences,95(7): 1480–1497, 2006.

[40] John R Alford, Andrew C Fowler, Deborah S Wuttke, Bruce A Kerwin, Ramil F Latypov, John F Carpenter, and Theodore W Randolph. Eﬀect of benzyl alcohol on recombinant human interleukin-1 receptor antagonist structure and hydrogen–deuterium exchange. Journal of pharmaceutical sciences, 100(10): 4215–4224, 2011.

[41] Regina L Hutchings, Surinder M Singh, Javier Cabello-Villegas, and Kr- ishna MG Mallela. Eﬀect of antimicrobial preservatives on partial protein unfolding and aggregation. Journal of pharmaceutical sciences,102(2):365–376, 2013.

[42] Surinder M Singh, Regina L Hutchings, and Krishna MG Mallela. Mecha- nisms of m-cresol-induced protein aggregation studied using a model protein cytochrome c. Journal of pharmaceutical sciences,100(5):1679–1689,2011.

125 [43] Regina L Bis, Surinder M Singh, Javier Cabello-Villegas, and Krishna MG Mallela. Role of benzyl alcohol in the unfolding and aggregation of interferon α-2a. Journal of pharmaceutical sciences,104(2):407–415,2015.

[44] A Baszkin, MM Boissonnade, A Kamyshny, and S Magdassi. Native and hydrophobically modiﬁed human immunoglobulin g at the air/water interface: Se- quential and competitive adsorption. Journal of colloid and interface science, 239(1):1–9, 2001.

[45] Su-Hwa Chang, Liang-Yu Chen, and Wen-Yih Chen. The eﬀects of denaturants on protein conformation and behavior at air/solution interface. Colloids and Surfaces B: Biointerfaces,41(1):1–6,2005.

[46] Theodore W Randolph and LaToya S Jones. Surfactant-protein interactions. In Rational Design of Stable Protein Formulations, pages 159–175. Springer, 2002.

[47] Hans Elwing, BO Nilsson, Karl-Erik Svensson, Agneta Askendahl, Ulf R Nils- son, and Ingemar Lundstr¨om. Conformational changes of a model protein (com- plement factor 3) adsorbed on hydrophilic and hydrophobic solid surfaces. Jour- nal of colloid and interface science,125(1):139–145,1988.

[48] RJ Green, I Hopkinson, and RAL Jones. Unfolding and intermolecular association in globular proteins adsorbed at interfaces. Langmuir, 15(15):5102–5110, 1999.

[49] L Feng and JD Andrade. Protein adsorption on low-temperature isotropic carbon: I. protein conformational change probed by diﬀerential scanning calorimetry. Journal of biomedical materials research,28(6):735–743,1994.

[50] Vladimir Hlady and Jos Buijs. Protein adsorption on solid surfaces. Current Opinion in Biotechnology, 7(1):72–77, 1996.

[51] Michael Rankl, Thomas Ruckstuhl, Michael Rabe, Georg RJ Artus, Andreas Walser, and Stefan Seeger. Conformational reorientation of immunoglobulin g during nonspeciﬁc interaction with surfaces. ChemPhysChem,7(4):837–846, 2006.

[52] Hanns-Christian Mahler, Robert M¨uller, Wolfgang Frieβ, Aurelie Delille, and Susanne Matheus. Induction and analysis of aggregates in a liquid igg1-antibody formulation. European journal of pharmaceutics and biopharmaceutics,59(3): 407–417, 2005.

[53] Sophie Colombi´e, Alain Gaunand, and Brigitte Lindet. Lysozyme inactivation under mechanical stirring: eﬀect of physical and molecular interfaces. Enzyme and microbial technology,28(9):820–826,2001.

126 [54] Lene Jorgensen, Pernille Bennedsen, Søren Vrønning Hoffmann, Rasmus Lin- nemann Krogh, Charlotte Pinholt, Minna Groenning, Susanne Hostrup, and Jens T Bukrinsky. Adsorption of insulin with varying self-association profiles to a solid teflon surface—influence on protein structure, fibrillation tendency and thermal stability. European Journal of Pharmaceutical Sciences,42(5): 509–516, 2011.

[55] Arpan Nayak, Amit K Dutta, and Georges Belfort. Surface-enhanced nucleation of insulin amyloid ﬁbrillation. Biochemical and biophysical research communications,369(2):303–307,2008.

[56] Hui F Liu, Junfen Ma, Charles Winter, and Robert Bayer. Recovery and pu- riﬁcation process development for monoclonal antibody production. In MAbs, volume 2, pages 480–499. Taylor & Francis, 2010.

[57] Wei Wang, Arun Alphonse Ignatius, and Santosh V Thakkar. Impact of residual impurities and contaminants on protein stability. Journal of pharmaceutical sciences,103(5):1315–1330,2014.

[58] Hanns-Christian Mahler, Wolfgang Friess, Ulla Grauschopf, and Sylvia Kiese. Protein aggregation: pathways, induction factors and analysis. Journal of pharmaceutical sciences,98(9):2909–2934,2009.

[59] Theodore W Randolph, Elise Schiltz, Donn Sederstrom, Daniel Steinmann, Olivier Mozziconacci, Christian Sch¨oneich, Erwin Freund, Margaret S Ricci, John F Carpenter, and Corrine S Lengsfeld. Do not drop: Mechanical shock in vials causes cavitation, protein aggregation, and particleformation.Journal of pharmaceutical sciences,2014.

[60] Ananthakrishnan Sethuraman and Georges Belfort. Protein structural perturbation and aggregation on homogeneous surfaces. Biophysical journal,88(2): 1322–1333, 2005.

[61] Linda O Narhi, Jeremy Schmit, Karoline Bechtold-Peters, and Deepak Sharma. Classiﬁcation of protein aggregates. Journal of pharmaceutical sciences,101(2): 493–498, 2012.

[62] James G Barnard, Satish Singh, Theodore W Randolph, and John F Carpen- ter. Subvisible particle counting provides a sensitive method of detecting and quantifying aggregation of monoclonal antibody caused by freeze-thawing: In- sights into the roles of particles in the protein aggregationpathway.Journal of pharmaceutical sciences,100(2):492–503,2011.

[63] James G. Barnard. Particles in Protein Solutions. PhD thesis, University of Colorado, Anschutz Medical Campus, 2013.

127 [64] James G Barnard, Ken Babcock, and John F Carpenter. Characterization and quantitation of aggregates and particles in interferon-β products: Poten- tial links between product quality attributes and immunogenicity. Journal of pharmaceutical sciences,102(3):915–928,2013.

[65] Vasco Filipe, Wim Jiskoot, Abdul Haﬁd Basmeleh, Andhyk Halim, Huub Schellekens, and Vera Brinks. Immunogenicity of diﬀerent stressed igg monoclonal antibody formulations in immune tolerant transgenic mice. In MAbs, volume 4, pages 740–752. Taylor & Francis, 2012.

[66] Merry Christie, Raul M Torres, Ross M Kedl, Theodore W Randolph, and John F Carpenter. Recombinant murine growth hormone particles are more immunogenic with intravenous than subcutaneous administration. Journal of Pharmaceutical Sciences,103(1):128–139,2014.

[67] Maliheh Shomali, Sultan Tanriverdi, Angelika J Freitag, Julia Engert, Ger- hard Winter, Michael Siedler, Zehra Kaymakcalan, John F Carpenter, and Theodore W Randolph. Dose levels in particulate-containingformulationsim- pact anti-drug antibody responses to murine monoclonal antibody in mice. Journal of pharmaceutical sciences,104(5):1610–1621,2015.

[68] Merry Christie. The Immune Response to Protein Particles. PhD thesis, Uni- versity of Colorado, Anschutz Medical Campus, 2014.

[69] Guidance for industry: Immunogenicity assessment for therapeutic protein products. http://www.fda.gov/downloads/drugs/ guidancecomplianceregulatoryinformation/guidances/ucm338856.pdf, August 2014. Accessed: 2015-01-29.

[70] Miranda van Beers and Muriel Bardor. Minimizing immunogenicity of biopharmaceuticals by controlling critical quality attributes of proteins. Biotechnology journal,7(12):1473–1484,2012.

[71] S Cao, N Jiao, Y Jiang, A Mire-Sluis, and LO Narhi. Sub-visible particle quantitation in protein therapeutics. Pharmeuropa bio & scientiﬁc notes, 2009 (1):73–79, 2009.

[72] John F Carpenter, Theodore W Randolph, WIM Jiskoot, Daan JA Crommelin, C Russell Middaugh, Gerhard Winter, Ying-Xin Fan, Susan Kirshner, Daniela Verthelyi, Steven Kozlowski, et al. Overlooking subvisibleparticlesinthera- peutic protein products: gaps that may compromise product quality. Journal of pharmaceutical sciences,98(4):1201–1205,2009.

[73] Satish K Singh, Nataliya Afonina, Michel Awwad, Karoline Bechtold-Peters, Jeﬀrey T Blue, Danny Chou, Mary Cromwell, Hans-Juergen Krause, Hanns- Christian Mahler, Brian K Meyer, et al. An industry perspective on the monitoring of subvisible particles as a quality attribute for protein therapeutics. Journal of pharmaceutical sciences,99(8):3302–3321,2010.

128 [74] Sagar Makwana, Biswajit Basu, Yogita Makasana, and AbhayDharamsi.Pre- ﬁlled syringes: An innovation in parenteral packaging. International journal of pharmaceutical investigation, 1(4):200, 2011.

[75] Brianne Harrison and Maribel Rios. Big shot: developments in preﬁlled syringes. Advanstar Communications Inc.,2007.

[76] Randy G Lee and Sung Wan Kim. Adsorption of proteins onto hydrophobic polymer surfaces: adsorption isotherms and kinetics. Journal of biomedical materials research, 8(5):251–259, 1974.

[77] Viwat Krisdhasima, Joseph McGuire, and Robert Sproull. Surface hydrophobic inﬂuences on β-lactoglobulin adsorption kinetics. Journal of colloid and interface science,154(2):337–350,1992.

[78] Meredith E Wiseman and Curtis W Frank. Antibody adsorption and orientation on hydrophobic surfaces. Langmuir,28(3):1765–1774,2012.

[79] C Werner, K-J Eichhorn, K Grundke, F Simon, W Gr¨ahlert, and H-J Jacobasch. Insights on structural variations of protein adsorption layers on hydrophobic ﬂuorohydrocarbon polymers gained by spectroscopic ellipsometry (part i). Col- loids and Surfaces A: Physicochemical and Engineering Aspects,156(1):3–17, 1999.

[80] Kazuhiro Nakanishi, Takaharu Sakiyama, and Koreyoshi Imamura. On the adsorption of proteins on solid surfaces, a common but very complicated phenomenon. Journal of Bioscience and Bioengineering,91(3):233–244,2001.

[81] Michael Rabe, Dorinel Verdes, and Stefan Seeger. Understanding protein adsorption phenomena at solid surfaces. Advances in colloid and interface science, 162(1):87–106, 2011.

[82] Jeﬀrey J Gray. The interaction of proteins with solid surfaces. Current opinion in structural biology,14(1):110–115,2004.

[83] Jared S Bee, Michele Davis, Erwin Freund, John F Carpenter, and Theodore W Randolph. Aggregation of a monoclonal antibody induced by adsorption to stainless steel. Biotechnology and bioengineering,105(1):121–129,2010.

[84] Paul Roach, David Farrar, and Carole C Perry. Interpretation of protein adsorption: surface-induced conformational changes. Journal of the American Chemical Society,127(22):8168–8173,2005.

[85] Emanuele Ostuni, Bartosz A Grzybowski, Milan Mrksich, Carmichael S Roberts, and George M Whitesides. Adsorption of proteins to hydrophobic sites on mixed self-assembled monolayers. Langmuir,19(5):1861–1872,2003.

129 [86] Sergii Rudiuk, Laetitia Cohen-Tannoudji, Sylvain Huille, and Christophe Tri- bet. Importance of the dynamics of adsorption and of a transient interfacial stress on the formation of aggregates of igg antibodies. Soft Matter,8(9):2651– 2661, 2012.

[87] Jared S Bee, Daniel K Schwartz, Siwar Trabelsi, Erwin Freund, Jennifer L Stevenson, John F Carpenter, and Theodore W Randolph. Production of particles of therapeutic proteins at the air-water interface during compression/dilation cycles. Soft Matter,8(40):10329–10335,2012.

[88] Johanna Wiesbauer, Ruth Prassl, and Bernd Nidetzky. Renewal of the air– water interface as a critical system parameter of protein stability: aggregation of the human growth hormone and its prevention by surface-active compounds. Langmuir,29(49):15240–15250,2013.

[89] Pinaki Basu, Angela W Blake-Haskins, Kristin B O’Berry, Theodore W Ran- dolph, and John F Carpenter. Albinterferon α2b adsorption to silicone oil–water interfaces: Eﬀects on protein conformation, aggregation, and subvisible particle formation. Journal of pharmaceutical sciences,103(2):427–436,2014.

[90] Alana Gerhardt, Kurt Bonam, Jared S Bee, John F Carpenter, and Theodore W Randolph. Ionic strength aﬀects tertiary structure and aggregation propensity of a monoclonal antibody adsorbed to silicone oil–water interfaces. Journal of pharmaceutical sciences,102(2):429–440,2013.

[91] Latoya S Jones, Allyn Kaufmann, and C Russell Middaugh. Silicone oil induced aggregation of proteins. Journal of pharmaceutical sciences, 94(4):918–927, 2005.

[92] Anthony Lubiniecki, David B Volkin, Marcia Federici, Michael D Bond, Michael L Nedved, Linda Hendricks, Promod Mehndiratta, Mark Bruner, Sud- hir Burman, Paul DalMonte, et al. Comparability assessmentsofprocessand product changes made during development of two diﬀerent monoclonal antibodies. Biologicals,39(1):9–22,2011.

[93] Shyam B Mehta, Rachael Lewus, Jared S Bee, Theodore W Randolph, and John F Carpenter. Gelation of a monoclonal antibody at the silicone oil–water interface and subsequent rupture of the interfacial gel results in aggregation and particle formation. Journal of pharmaceutical sciences,104(4):1282–1290, 2015.

[94] L Danielle. Advanced innovations on a new generation of plastic preﬁlled syringes. ondrugdelivery Magazine,2010.

[95] Alana Gerhardt, Nicole R Mcgraw, Daniel K Schwartz, Jared S Bee, John F Carpenter, and Theodore W Randolph. Protein aggregation and particle formation in preﬁlled glass syringes. Journal of pharmaceutical sciences,103(6): 1601–1612, 2014.

130 [96] Elena Krayukhina, Kouhei Tsumoto, Susumu Uchiyama, and Kiichi Fukui. Ef- fects of syringe material and silicone oil lubrication on thestabilityofpharma- ceutical proteins. Journal of pharmaceutical sciences,104:527–535,2014.

[97] D Brett Ludwig, John F Carpenter, Jean-Bernard Hamel, and Theodore W Randolph. Protein adsorption and excipient eﬀects on kinetic stability of silicone oil emulsions. Journal of pharmaceutical sciences,99(4):1721–1733,2010.

[98] Byeong S Chang, Brent S Kendrick, and John F Carpenter. Surface-induced denaturation of proteins during freezing and its inhibition bysurfactants.Journal of pharmaceutical sciences,85(12):1325–1330,1996.

[99] Lotte Krielgaard, Latoya S Jones, Theodore W Randolph, Sven Frokjaer, James M Flink, Mark C Manning, and John F Carpenter. Eﬀect of tween 20 on freeze-thawing-and agitation-induced aggregation ofrecombinanthuman factor xiii. Journal of pharmaceutical sciences,87(12):1593–1603,1998.

[100] SH Mollmann, U Elofsson, JT Bukrinsky, and S Frokjaer. Displacement of adsorbed insulin by tween 80 monitored using total internal reﬂection ﬂuorescence and ellipsometry. Pharmaceutical research,22(11):1931–1941,2005.

[101] Wayne W Sukow, Howard E Sandberg, Edwin A Lewis, Delbert J Eatough, and Lee D Hansen. Binding of the triton x series of nonionic surfactants to bovine serum albumin. Biochemistry,19(5):912–917,1980.

[102] Narendra B Bam, Jeﬀrey L Cleland, Janet Yang, Mark C Manning, John F Carpenter, Robert F Kelley, and Theodore W Randolph. Tween protects recombinant human growth hormone against agitation-induced damage via hydrophobic interactions. Journal of pharmaceutical sciences,87(12):1554–1559, 1998.

[103] Narendra B Bam, Theodore W Randolph, and Jeﬀrey L Cleland.Stabilityof protein formulations: investigation of surfactant eﬀects by a novel epr spectroscopic technique. Pharmaceutical research, 12(1):2–11, 1995.

[104] Jeﬀrey L Cleland and Theodore W Randolph. Mechanism of polyethylene glycol interaction with the molten globule folding intermediate ofbovinecarbonic anhydrase b. Journal of Biological Chemistry,267(5):3147–3153,1992.

[105] Narendra B Bam, Jeﬀrey L Cleland, and Theodore W Randolph. Molten globule intermediate of recombinant human growth hormone: stabilization with surfactants. Biotechnology progress,12(6):801–809,1996.

[106] Flora Felsovalyi, Sébastien Janvier, Sébastien Jouffray, HervéSoukiassian, and Paolo Mangiagalli. Silicone-oil-based subvisible particles: Their detection, interactions, and regulation in prefilled container closure systems for biopharmaceuticals. Journal of pharmaceutical sciences,101(12):4569–4583,2012.

131 [107] Daniel Schuster. Encyclopedia of emulsion technology, volume 4. CRC Press, 1996.

[108] Gregory Sacha, J Aaron Rogers, and Reagan L Miller. Pre-ﬁlled syringes: a review of the history, manufacturing and challenges. Pharmaceutical development and technology, 20(1):1–11, 2014.

[109] Jan Jezek, Nicholas J Darton, Barry K Derham, Nikki Royle, and Iain Simpson. Biopharmaceutical formulations for pre-ﬁlled delivery devices. Expert opinion on drug delivery,10(6):811–828,2013.

[110] Dennis R Jenke. Extractables and leachables considerations for preﬁlled syringes. Expert opinion on drug delivery,11(10):1591–1600,2014.

[111] SS Qadry, TH Roshdy, DE Knox, and EM Phillips. Model development for o2 and n2 permeation rates through cz-resin vials. International journal of pharmaceutics,188(2):173–179,1999.

[112] Steven L Nail, Wendy Saﬀell-Clemmer, and Michael J Akers. Current issues with parenteral dosage forms. Hospital Pharmacy Europe,34,2007.

[113] Arnoldus WP Vermeer, Maria GEG Bremer, and Willem Norde. Structural changes of igg induced by heat treatment and by adsorption onto a hydrophobic teﬂon surface studied by circular dichroism spectroscopy. Biochimica et Biophysica Acta (BBA)-General Subjects, 1425(1):1–12, 1998.

[114] Arnoldus WP Vermeer, Carla E Giacomelli, and Willem Norde. Adsorption of igg onto hydrophobic teﬂon. diﬀerences between the f ab and f c domains. Biochimica et Biophysica Acta (BBA)-General Subjects,1526(1):61–69,2001.

[115] Susanne H Mollmann, Lene Jorgensen, Jens T Bukrinsky, Ulla Elofsson, Willem Norde, and Sven Frokjaer. Interfacial adsorption of insulin: conformational changes and reversibility of adsorption. european journal of pharmaceutical sciences,27(2):194–204,2006.

[116] Sylvia Kiese, Astrid Papppenberger, Wolfgang Friess, and Hanns-Christian Mahler. Shaken, not stirred: mechanical stress testing of anigg1antibody. Journal of pharmaceutical sciences,97(10):4347–4366,2008.

[117] K. Strauss, H. De Gols, I. Hannet, T.M. Partanen, and A. Frid. A pan-European epidemiologic study of insulin injection technique in patients with diabetes. Practical Diabetes International, 19(3):71–76, 2002.

[118] Guy Dodson and Don Steiner. The role of assembly in insulin’s biosynthesis. Current opinion in structural biology,8(2):189–194,1998.

[119] Irl B Hirsch. Insulin analogues. New England Journal of Medicine,352(2): 174–183, 2005.

132 [120] HE Lebovitz. Diabetic ketoacidosis. The Lancet,345(8952):767–772,1995.

[121] J. Brange, D.R. Owens, S. Kang, and A. Vølund. Monomeric insulins and their experimental and clinical implications. Diabetes Care,13(9):923–954,1990.

[122] Z. Vajo and W.C. Duckworth. Genetically engineered insulin analogs: diabetes in the new millenium. Pharmacological reviews, 52(1):1–10, 2000.

[123] J. Brange. The new era of biotech insulin analogues. Diabetologia,40:48–53, 1997.

[124] A. Wollmer, B. Rannefeld, BR Johansen, KR Hejnaes, P. Balschmidt, FB Hansen, et al. Phenol-promoted structural transformation of insulin in solution. Biological Chemistry Hoppe-Seyler, 368(8):903, 1987.

[125] J.L. Whittingham, D.J. Edwards, A.A. Antson, J.M. Clarkson, and G.G. Dod- son. Interactions of phenol and m-cresol in the insulin hexamer, and their eﬀect on the association properties of b28 pro-asp insulin analogues. Biochemistry, 37(33):11516–11523, 1998.

[126] U. Derewenda, Z. Derewenda, EJ Dodson, GG Dodson, CD Reynolds, GD Smith, C. Sparks, and D. Swenson. Phenol stabilizes more helix in a new symmetrical zinc insulin hexamer. Nature,338(6216):594–596,1989.

[127] SG Melberg, S. Havelund, J. Villumsen, and J. Brange. Insulin compatibility with polymer materials used in external pump infusion systems. Diabetic Medicine, 5(3):243–247, 1988.

[128] E. Chantelau, G. Lange, M. Gasthaus, M. Boxberger, and M. Berger. Inter- action between plastic catheter tubings and regular insulin preparations used for continuous subcutaneous insulin-infusion therapy. Diabetes Care,10(3):348, 1987.

[129] M.R. DeFelippis, M.A. Bell, J.A. Heyob, and S.M. Storms. In vitro stability of insulin lispro in continuous subcutaneous insulin infusion. Diabetes technology & therapeutics,8(3):358–368,2006.

[130] Sadao Mori and Howard G Barth. Size exclusion chromatography. Springer Science & Business Media, 2013.

[131] FE Regnier and R Noel. Glycerolpropylsilane bonded phases in the steric exclusion chromatography of biological macromolecules. Journal of chromatographic science,14(7):316–320,1976.

[132] RJ Blagrove and Maurice J Frenkel. Determination of protein molecular weights in denaturing solvents using glyceryl-cpg. Journal of Chromatography A,132 (3):399–404, 1977.

133 [133] Hugh D Crone and RM Dawson. Residual anionic properties of a covalently substituted controlled-pore glass, glyceryl-cpg. Journal of Chromatography A, 129:91–96, 1976.

[134] Preﬁlled syringes market (glass and plastic): Global industry analysis, size, volume, share, growth, trends and forecast, 2013–2019. Technical report, Trans- parency Market Research, 2013.

[135] Pre-ﬁlled syringes market forecast 2015-2025. Technical report, Visiongain, 2015.

[136] Sumit Sharma, BJ Berne, and Sanat K Kumar. Thermal and structural stability of adsorbed proteins. Biophysical journal,99(4):1157–1165,2010.

[137] A Baszkin, MM Boissonnade, A Kamyshny, and S Magdassi. Adsorption of native and hydrophobically modified human immunoglobulin g on polyethylene solid films: Specific recognition of adsorbed layers. Journal of colloid and interface science,244(1):18–23,2001.

[138] Herbert P Jennissen. Multivalent interaction chromatography as exempliﬁed by the adsorption and desorption of skeletal muscle enzymes on hydrophobic alkyl-agaroses. Journal of Chromatography A,159(1):71–83,1978.

[139] HP Jennissen and G Botzet. Protein binding to two-dimensional hydrophobic binding-site lattices: adsorption hysteresis on immobilized butyl-residues. International Journal of Biological Macromolecules,1(4):171–179,1979.

[140] Herbert P Jennissen. Hydrophobic agaroses: basis for a model of multivalent eﬀector-receptor interactions. Hoppe-Seyler´ s Zeitschrift f¨urphysiologische Chemie,357(2):1727–1734,1976.

[141] JL Brash, S Uniyal, and Q Samak. Exchange of albumin adsorbed on polymer surfaces. Transactions-American Society for Artiﬁcial Internal Organs,20:69– 76, 1973.

[142] John L Brash and Qussay M Samak. Dynamics of interactionsbetweenhuman albumin and polyethylene surface. Journal of Colloid and Interface Science,65 (3):495–504, 1978.

[143] Mark Kastantin, Blake B Langdon, and Daniel K Schwartz. A bottom-up approach to understanding protein layer formation at solid–liquid interfaces. Advances in colloid and interface science, 207:240–252, 2014.

[144] Mark Kastantin, Robert Walder, and Daniel K Schwartz. Identifying mechanisms of interfacial dynamics using single-molecule tracking. Langmuir, 28(34): 12443–12456, 2012.

[145] Blake B Langdon, Mark Kastantin, Robert Walder, and Daniel K Schwartz. Interfacial protein–protein associations. Biomacromolecules,15(1):66–74,2013.

134 [146] Robert Walder and Daniel K Schwartz. Dynamics of protein aggregation at the oil–water interface characterized by single molecule tirf microscopy. Soft Matter,7(17):7616–7622,2011.

[147] Carly F Chisholm, Bao H Nguyen, Kaitlin R Soucie, Raul M Torres, John F Carpenter, and Theodore W Randolph. In vivo analysis of the potency of silicone oil microdroplets as immunological adjuvants in protein formulations. Journal of Pharmaceutical Sciences, 2015. In Press.

[148] P Grueninger. Coated plungers: Peace of mind... Technical report, ONdrugDe- livery, 2011. 34–36 pp.

[149] Victoria Sluzky, Janet A Tamada, Alexander M Klibanov, and Robert Langer. Kinetics of insulin aggregation in aqueous solutions upon agitation in the presence of hydrophobic surfaces. Proceedings of the National Academy of Sciences, 88(21):9377–9381, 1991.

[150] Andreas W Sonesson, Hans Blom, Kai Hassler, Ulla M Elofsson, Thomas H Cal- lisen, Jerker Widengren, and Hjalmar Brismar. Protein–surfactant interactions at hydrophobic interfaces studied with total internal reﬂection ﬂuorescence cor- relation spectroscopy (tir-fcs). Journal of colloid and interface science,317(2): 449–457, 2008.

[151] Avastin [drug package insert]. Genentech, Inc., 2014.

[152] Thermo Scientiﬁc. Instructions: AquasilTM and surfasilTM siliconizing ﬂuids, 2008.

[153] Johannes Schindelin, Ignacio Arganda-Carreras, Erwin Frise, Verena Kaynig, Mark Longair, Tobias Pietzsch, Stephan Preibisch, Curtis Rueden, Stephan Saalfeld, Benjamin Schmid, et al. Fiji: an open-source platform for biological- image analysis. Nature methods,9(7):676–682,2012.

[154] Ronaldo Mohana-Borges, Jerson Lima Silva, and Gonzalo de Prat-Gay. Protein folding in the absence of chemical denaturants reversible pressure denaturation of the noncovalent complex formed by the association of two protein fragments. Journal of Biological Chemistry,274(12):7732–7740,1999.

[155] V. Sluzky, A.M. Klibanov, and R. Langer. Mechanism of insulin aggregation and stabilization in agitated aqueous solutions. Biotechnology and bioengineering, 40(8):895–903, 1992.

[156] Nitin Dixit, Kevin M Maloney, and Devendra S Kalonia. Protein-silicone oil interactions: Comparative eﬀect of nonionic surfactants on the interfacial behavior of a fusion protein. Pharmaceutical research,30(7):1848–1859,2013.

135 [157] Danny K Chou, Rajesh Krishnamurthy, Theodore W Randolph, John F Car- penter, and Mark Cornell Manning. Eﬀects of tween 20⃝R and tween 80⃝R on the stability of albutropin during agitation. Journal of pharmaceutical sciences, 94(6):1368–1381, 2005.

[158] Deepak K Sharma, Peter Oma, and Sampath Krishnan. Silicone microdroplets in protein formulations— detection and enumeration. Pharmaceutical Technology, 33(4):74–79, 2009.

[159] RenéStrehl, Verena Rombach-Riegraf, Manuel Diez, Kamal Egodage, Markus Bluemel, Margit Jeschke, and Atanas V Koulov. Discrimination between silicone oil droplets and protein aggregates in biopharmaceuticals: a novel multi- parametric image filter for sub-visible particles in microflow imaging analysis. Pharmaceutical research,29(2):594–602,2012.

[160] Daniel Weinbuch, Sarah Z¨olls, Michael Wiggenhorn, Wolfgang Friess, Gerhard Winter, Wim Jiskoot, and Andrea Hawe. Micro–ﬂow imaging and resonant mass measurement (archimedes)–complementary methods to quantitatively differentiate protein particles and silicone oil droplets. Journal of pharmaceutical sciences,102(7):2152–2165,2013.

[161] Michelle R Duncan, Joanne M Lee, and Mark P Warchol. Inﬂuence of surfactants upon protein/peptide adsorption to glass and polypropylene. Interna- tional journal of pharmaceutics,120(2):179–188,1995.

[162] Mark´eta Vlˇckov´a, Franka Kalman, and Maria A Schwarz. Pharmaceutical ap- plications of isoelectric focusing on microchip with imaged uv detection. Journal of Chromatography A,1181(1):145–152,2008.

[163] Jai K Kaushik and Rajiv Bhat. Why is trehalose an exceptional protein stabilizer? an analysis of the thermal stability of proteins in the presence of the compatible osmolyte trehalose. Journal of Biological Chemistry, 278(29):26458– 26465, 2003.

[164] EJea Smith, MW Henley, EF Adams, AO DeMarco, JT Cronin, DM Wa- genknecht, G Bates, and M Musolf. Siliconization of parenteral drug packaging components. J. Parent. Sci. Technol, 42:S1–S13, 1988.

[165] Latoya S Jones, Theodore W Randolph, Ulrich Kohnert, Apollon Papadim- itriou, G Winter, Marie-Luise Hagmann, Mark C Manning, and John F Car- penter. The eﬀects of tween 20 and sucrose on the stability of anti-l-selectin during lyophilization and reconstitution. Journal of pharmaceutical sciences, 90(10):1466–1477, 2001.

[166] Jie Chen and Yan Sun. Modeling of the salt eﬀects on hydrophobic adsorption equilibrium of protein. Journal of Chromatography A,992(1):29–40,2003.

136 [167] Renata CF Bonomo, Luis A Minim, Jane SR Coimbra, Rafael CI Fontan, Luis H Mendes da Silva, and Val´eria PR Minim. Hydrophobic interaction adsorption of whey proteins: eﬀect of temperature and salt concentration and thermodynamic analysis. Journal of Chromatography B, 844(1):6–14, 2006.

[168] Lin Zhang, Guofeng Zhao, and Yan Sun. Molecular insightintoproteinconfor- mational transition in hydrophobic charge induction chromatography: A molecular dynamics simulation. The Journal of Physical Chemistry B,113(19):6873– 6880, 2009.

[169] Nathaniel Nelson and Daniel K Schwartz. Specific ion (hofmeister) effects on adsorption, desorption, and diffusion at the solid–aqueous interface. The Journal of Physical Chemistry Letters,4(23):4064–4068,2013.

[170] Paolo Arosio, Baptiste Jaquet, Hua Wu, and Massimo Morbidelli. On the role of salt type and concentration on the stability behavior of a monoclonal antibody solution. Biophysical chemistry, 168:19–27, 2012.

[171] Le Zhang, Helming Tan, R Matthew Fesinmeyer, Cynthia Li, David Catrone, David Le, Richard L Remmele, and Jifeng Zhang. Antibody solubility behavior in monovalent salt solutions reveals speciﬁc anion eﬀects atlowionicstrength. Journal of pharmaceutical sciences,101(3):965–977,2012.

[172] Arpad Szenczi, Jozsef Kardos, Gy¨orgy A Medgyesi, and Peter Zavodszky. The eﬀect of solvent environment on the conformation and stability of human poly- clonal igg in solution. Biologicals, 34(1):5–14, 2006.

[173] Connie M Chung, Jenny D Chiu, Lawreen H Connors, Olga Gursky, Amareth Lim, Andrew B Dykstra, Juris Liepnieks, Merrill D Benson, Catherine E Costello, Martha Skinner, et al. Thermodynamic stability ofaκiimmunoglob- ulin light chain: relevance to multiple myeloma. Biophysical journal,88(6): 4232–4242, 2005.

[174] R Matthew Fesinmeyer, Sabine Hogan, Atul Saluja, Stephen R Brych, Eva Kras, Linda O Narhi, David N Brems, and Yatin R Gokarn. Eﬀect of ions onagitation- and temperature-induced aggregation reactions of antibodies. Pharmaceutical research,26(4):903–913,2009.

[175] Sanjay B Hari, Hollis Lau, Vladimir I Razinkov, Shuang Chen, and Ramil F Latypov. Acid-induced aggregation of human monoclonal igg1 and igg2: Molec- ular mechanism and the eﬀect of solution composition. Biochemistry,49(43): 9328–9338, 2010.

[176] Rebecca K Brummitt, Douglas P Nesta, Liuquan Chang, Susan F Chase, Thomas M Laue, and Christopher J Roberts. Nonnative aggregation of an igg1 antibody in acidic conditions: Part 1. unfolding, colloidal interactions, and formation of high-molecular-weight aggregates. Journal of pharmaceutical sciences,100(6):2087–2103,2011.

137 [177] Sampathkumar Krishnan, Monica M Pallitto, and Margaret S Ricci. Devel- opment of formulations for therapeutic monoclonal antibodies and fc fusion proteins. Formulation and Process Development Strategies for Manufacturing Biopharmaceuticals, pages 383–427, 2010.

[178] Erinc Sahin, Adeola O Grillo, Melissa D Perkins, and Christopher J Roberts. Comparative eﬀects of ph and ionic strength on protein–protein interactions, unfolding, and aggregation for igg1 antibodies. Journal of pharmaceutical sciences,99(12):4830–4848,2010.

[179] Shantanu V Sule, Jason K Cheung, Valentyn Antochshuk, Amardeep S Bhalla, Chakravarthy Narasimhan, Steven Blaisdell, Mohammed Shameem, and Pe- ter M Tessier. Solution ph that minimizes self-association of three monoclonal antibodies is strongly dependent on ionic strength. Molecular pharmaceutics,9 (4):744–751, 2012.

[180] Hirotaka Nishi, Makoto Miyajima, Hiroaki Nakagami, Masanori Noda, Susumu Uchiyama, and Kiichi Fukui. Phase separation of an igg1 antibody solution under a low ionic strength condition. Pharmaceutical research,27(7):1348–1360, 2010.

[181] Branden A Salinas, Hasige A Sathish, Steven M Bishop, Nick Harn, John F Car- penter, and Theodore W Randolph. Understanding and modulating opalescence and viscosity in a monoclonal antibody formulation. Journal of pharmaceutical sciences, 99(1):82–93, 2010.

[182] Thomas M Scherer, Jun Liu, Steven J Shire, and Allen P Minton. Intermolecular interactions of igg1 monoclonal antibodies at high concentrations characterized by light scattering. The Journal of Physical Chemistry B,114(40):12948–12957, 2010.

[183] Abraham George and W William Wilson. Predicting protein crystallization from a dilute solution property. Acta Crystallographica Section D: Biological Crystallography,50(4):361–365,1994.

[184] DrugBank. Intravenous immunoglobulin. http://www.drugbank.ca/drugs/ DB00028. Accessed: 2015-03-30.

[185] Eva Y Chi, Sampathkumar Krishnan, Brent S Kendrick, Byeong S Chang, John F Carpenter, and Theodore W Randolph. Roles of conformational stability and colloidal stability in the aggregation of recombinant human granulocyte colony-stimulating factor. Protein Science, 12(5):903–913, 2003.

[186] Jacob N Israelachvili. Intermolecular and surface forces: revised third edition. Academic press, 2011.

138 [187] Matthew Hoehne, Fauna Samuel, Aichun Dong, Christine Wurth, Hanns- Christian Mahler, John F Carpenter, and Theodore W Randolph. Adsorption of monoclonal antibodies to glass microparticles. Journal of pharmaceutical sciences,100(1):123–132,2011. [188] Y. Pocker and S.B. Biswas. Self-association of insulin and the role of hydrophobic bonding: a thermodynamic model of insulin dimerization. Biochemistry,20 (15):4354–4361, 1981. [189] K. Huus, S. Havelund, H.B. Olsen, M. Van De Weert, and S. Frokjaer. Thermal dissociation and unfolding of insulin. Biochemistry,44(33):11171–11177,2005. [190] J. Brange, L. Andersen, E.D. Laursen, G. Meyn, and E. Rasmussen. Toward understanding insulin fibrillation. Journal of pharmaceutical sciences,86(5): 517–525, 1997. [191] N.C. Kaarsholm, H.C. Ko, and M.F. Dunn. Comparison of solution structural flexibility and zinc binding domains for insulin, proinsulin, and miniproinsulin. Biochemistry,28(10):4427–4435,1989. [192] U Derewenda, Z Derewenda, GG Dodson, RE Hubbard, and F Korber. Molec- ular structure of insulin: the insulin monomer and its assembly. British Medical Bulletin, 45(1):4–18, 1989. [193] S. Rahuel-Clermont, C.A. French, N.C. Kaarsholm, and M.F. Dunn. Mecha- nisms of stabilization of the insulin hexamer through allosteric ligand interactions. Biochemistry,36(19):5837–5845,1997. [194] P. Krüger, G. Gilge, Y. Çabuk, and A. Wollmer. Cooperativity and intermediate states in the t to r-structural transformation of insulin. Biological Chemistry Hoppe-Seyler,371(2):669–674,2009. [195] Mark L Brader, Niels C Kaarsholm, Robert WK Lee, and Michael F Dunn. Characterization of the r-state insulin hexamer and its derivatives. the hexamer is stabilized by heterotropic ligand binding interactions. Biochemistry,30(27): 6636–6645, 1991. [196] Wonjae E Choi, Mark L Brader, Valentin Aguilar, Niels C Kaarsholm, and Michael F Dunn. Allosteric transition of the insulin hexamer is modulated by homotropic and heterotropic interactions. Biochemistry, 32(43):11638–11645, 1993. [197] Curtis R Bloom, Nancy Wu, Alex Dunn, Niels C Kaarsholm, and Michael F Dunn. Comparison of the allosteric properties of the co (ii)-and zn (ii)- substituted insulin hexamers. Biochemistry,37(31):10937–10944,1998. [198] L. Nielsen, S. Frokjaer, J. Brange, V.N. Uversky, and A.L. Fink. Probing the mechanism of insulin fibril formation with insulin mutants. Biochemistry,40 (28):8397–8409, 2001.

139 [199] N.B. Phillips, J. Whittaker, F. Ismail-Beigi, and M.A. Weiss. Insulin ﬁbrillation and protein design: topological resistance of single-chain analogs to thermal degradation with application to a pump reservoir. Journal of Diabetes Science and Technology, 6(2):277–288, 2012.

[200] J. Brange, L. Langkjaer, et al. Chemical stability of insulin. 3. inﬂuence of excipients, formulation, and ph. Acta pharmaceutica nordica, 4(3):149, 1992.

[201] J. Brange and A. Vølund. Insulin analogs with improved pharmacokinetic pro- ﬁles. Advanced drug delivery reviews, 35(2-3):307–335, 1999.

[202] D.N. Brems, L.A. Alter, M.J. Beckage, R.E. Chance, R.D. DiMarchi, L.K. Green, H.B. Long, A.H. Pekar, J.E. Shields, and B.H. Frank. Alteringtheas- sociation properties of insulin by amino acid replacement. Protein engineering, 5(6):527–533, 1992.

[203] E. Ciszak, J.M. Beals, B.H. Frank, J.C. Baker, N.D. Carter, and G.D. Smith. Role of c-terminal b-chain residues in insulin assembly: thestructureofhex- americ lysb28prob29-human insulin. Structure,3(6):615–622,1995.

[204] D.L. Bakaysa, J. Radziuk, H.A. Havel, M.L. Brader, S. Li, S.W. Dodd, J.M. Beals, A.H. Pekar, and D.N. Brems. Physicochemical basis for therapidtime- action of lysb28prob29-insulin: Dissociation of a protein-ligand complex. Pro- tein science,5(12):2521–2531,1996.

[205] B.W. Bode. Comparison of pharmacokinetic properties, physicochemical stability, and pump compatibility of 3 rapid-acting insulin analogues-aspart, lispro, and glulisine. Endocrine Practice,17(2):271–280,2011.

[206] J.H. Anderson, R.L. Brunelle, V.A. Koivisto, M.E. Trautmann, L. Vignati, and R. DiMarchi. Improved mealtime treatment of diabetes mellitus using an insulin analogue. Clinical Therapeutics, 19(1):62–72, 1997.

[207] J.P. Richards, M.P. Stickelmeyer, D.B. Flora, R.E. Chance, B.H. Frank, and M.R. DeFelippis. Self-association properties of monomericinsulinanalogsunder formulation conditions. Pharmaceutical research,15(9):1434–1441,1998.

[208] W.D. Lougheed, B. Zinman, T.R. Strack, L.J. Janis, A.B. Weymouth, E.A. Bernstein, A.M. Korbas, and B.H. Frank. Stability of insulin lispro in insulin infusion systems. Diabetes care,20(7):1061–1065,1997.

[209] M.U. Jars, A. Hvass, and D. Waaben. Insulin aspart (asp b28 human insulin) derivatives formed in pharmaceutical solutions. Pharmaceutical research,19(5): 621–628, 2002.

[210] R.H.A. Becker. Insulin glulisine complementing basal insulins: a review of structure and activity. Diabetes technology & therapeutics, 9(1):109–121, 2007.

140 [211] RJ Woods, J. Alarc´on, E. McVey, and RJ Pettis. Intrinsic ﬁbrillation of fast- acting insulin analogs. Journal of diabetes science and technology,6(2):265, 2012.

[212] B.W. Bode, T.M. Gross, K.R. Thornton, and J.J. Mastrototaro. Continu- ous glucose monitoring used to adjust diabetes therapy improves glycosylated hemoglobin: a pilot study. Diabetes research and clinical practice,46(3):183– 190, 1999.

[213] F.R. Kaufman, L.C. Gibson, M. Halvorson, S. Carpenter, L.K. Fisher, and P. Pitukcheewanont. A pilot study of the continuous glucose monitoring system. Diabetes Care, 24(12):2030, 2001.

[214] D.C. Klonoﬀ. Continuous glucose monitoring. Diabetes care, 28(5):1231, 2005.

[215] DE James, AB Jenkins, EW Kraegen, and DJ Chisholm. Insulin precipitation in artiﬁcial infusion devices. Diabetologia,21(6):554–557,1981.

[216] D.S. Schade, R.P. Eaton, J. DeLongo, L.C. Saland, A.J. Ladman, and G.A. Carlson. Electron microscopy of insulin precipitates. Diabetes Care, 5(1):25– 30, 1982.

[217] JR Brennan, SS Gebhart, and WG Blackard. Pump-induced insulin aggregation. a problem with the biostator. Diabetes,34(4):353–359,1985.

[218] V. Feingold, AB Jenkins, and EW Kraegen. Eﬀect of contact material on vibration-induced insulin aggregation. Diabetologia,27(3):373–378,1984.

[219] J.L. Selam, P. Zirinis, M. Mellet, and J. Mirouze. Stable insulin for implantable delivery systems: in vitro studies with diﬀerent containersandsolvents. Dia- betes care, 10(3):343–347, 1987.

[220] RS Mecklenburg and TS Guinn. Complications of insulin pump therapy: the eﬀect of insulin preparation. Diabetes Care, 8(4):367, 1985.

[221] R. Pryce. Diabetic ketoacidosis caused by exposure of insulin pump to heat and sunlight. BMJ,338,2009.

[222] AW Wright and JA Little. Cannula occlusion with use of insulin lispro and insulin infusion system. Diabetes care,21(5):874–875,1998.

[223] H.A. Wolpert, R.N. Faradji, S. Bonner-Weir, and M.A. Lipes. Metabolic de- compensation in pump users due to lispro insulin precipitation. BMJ,324:1253, 2002.

[224] H. Martindale, J. Marsh, FR Hallett, and AM Albisser. Examination of insulin formulations using quasi-elastic light scattering. Diabetes,31(4):364–366,1982.

141 [225] I. Guilhem, AM Leguerrier, F. Lecordier, JY Poirier, and D. Maugendre. Tech- nical risks with subcutaneous insulin infusion. Diabetes & metabolism,32(3): 279–284, 2006.

[226] R. Renner, A. Pfutzner, M. Trautmann, O. Harzer, K. Sauter, and R. Landgraf. Use of insulin lispro in continuous subcutaneous insulin infusion treatment. results of a multicenter trial. german humalog-csii study group. Diabetes Care, 22(5):784–788, 1999.

[227] WD Lougheed, H. Woulfe-Flanagan, JR Clement, and AM Albisser. Insulin aggregation in artiﬁcial delivery systems. Diabetologia, 19(1):1–9, 1980.

[228] D. Kerr, J. Morton, C. Whately-Smith, J. Everett, and J.P. Begley. Laboratory- Based Non-Clinical Comparison of Occlusion Rates Using Three Rapid-Acting Insulin Analogs in Continuous Subcutaneous Insulin Infusion Catheters Using Low Flow Rates. Journal of Diabetes Science and Technology,2(3):450–455, 2008.

[229] Humalog [drug package insert]. Eli Lilly and Company, 2011.

[230] Novolog [drug package insert]. Novo Nordisk Inc., 2009.

[231] Apidra [drug package insert]. sanoﬁ-aventis U.S. LLC, 2009.

[232] W. Jiskoot. personal email communication, 2011.

[233] B.M. Teska. Unpublished observations. 2011.

[234] R. Tantipolphan, S. Romeijn, J. Engelsman, R. Torosantucci, T. Rasmussen, and W. Jiskoot. Elution behavior of insulin on high-performance size exclusion chromatography at neutral ph. Journal of pharmaceutical and biomedical analysis,52(2):195–202,2010.

[235] W.F. Stafford III. Boundary analysis in sedimentation transport experiments: aprocedureforobtainingsedimentationcoefficient distributions using the time derivative of the concentration profile. Analytical biochemistry,203(2):295–301, 1992.

[236] D.T. Birnbaum, M.A. Kilcomons, M.R. DeFelippis, and J.M. Beals. Assem- bly and dissociation of human insulin and lysb28prob29-insulin hexamers: a comparison study. Pharmaceutical research,14(1):25–36,1997.

[237] Peter Schuck. Size-distribution analysis of macromolecules by sedimentation velocity ultracentrifugation and lamm equation modeling. Biophysical journal, 78(3):1606–1619, 2000.

[238] Beckman Coulter. Chemical resistances for beckman coulter centrifugation products. https://www.beckmancoulter.com/wsrportal/bibliography? docname=chemres.pdf, Oct 2001. Accessed: 2014-05-01.

142 [239] Beckman Coulter. Instructions for use: An-50ti and an-60ti analytical ro- tor, cells and counterbalance. https://www.beckmancoulter.com/wsrportal/ techdocs?docname=LXLA-TB-003, Oct 2011. Accessed: 2014-05-01.

[240] J. Brange. Stability of insulin: studies on the physical and chemical stability of insulin in pharmaceutical formulation. Kluwer Academic Pub, 1994.

[241] M.F. Dunn. Zinc–ligand interactions modulate assemblyandstabilityofthe insulin hexamer–a review. Biometals, 18(4):295–303, 2005.

[242] J Brange. Chemical stability of insulin. 4. mechanismsandkineticsofchemical transformations in pharmaceutical formulation. Acta pharmaceutica nordica,4 (4):209–222, 1992.

[243] Helle B Olsen, Svend Ludvigsen, and Niels C Kaarsholm. Solution structure of an engineered insulin monomer at neutral ph. Biochemistry, 35(27):8836–8845, 1996.

[244] Svend Ludvigsen, Melinda Roy, Henning Thoegersen, and Niels C Kaarsholm. High-resolution structure of an engineered biologically potent insulin monomer, b16 tyr. fwdarw. his, as determined by nuclear magnetic resonance spectroscopy. Biochemistry,33(26):7998–8006,1994.

[245] John F Carpenter, Theodore W Randolph, WIM Jiskoot, Daan JA Crommelin, C Russell Middaugh, Gerhard Winter, Ying-Xin Fan, Susan Kirshner, Daniela Verthelyi, Steven Kozlowski, et al. Overlooking subvisibleparticlesinthera- peutic protein products: Gaps that may compromise product quality. Journal of pharmaceutical sciences,98(4):1201–1205,2008.

[246] B.R. Simler, G. Hui, J.E. Dahl, and B. Perez-Ramirez. Mechanistic complexity of subvisible particle formation: Links to protein aggregation are highly speciﬁc. Journal of Pharmaceutical Sciences,101(11):4140–4154,2012.

[247] GA Heavner, T Arakawa, JS Philo, MA Calmann, and S Labrenz. Protein isolated from biopharmaceutical formulations cannot be used for comparative studies: Follow-up to “a case study using epoetin alfa from epogen and eprex”. Journal of pharmaceutical sciences,96(12):3214–3225,2007.

[248] Brandon M Teska, Javier Alarc´on, Ronald J Pettis, Theodore W Randolph, and John F Carpenter. Eﬀects of phenol and meta-cresol depletion on insulin analog stability at physiological temperature. Journal of pharmaceutical sciences,103 (8):2255–2267, 2014.

[249] R. Payne. personal communication, 2013.

[250] A.K. Attri, C. Fern´andez, and A.P. Minton. Self-association of Zn-insulin at neutral pH: Investigation by concentration gradient-staticanddynamiclight scattering. Biophysical Chemistry,2010.

143 [251] Paul HC Eilers and Hans FM Boelens. Baseline correction with asymmetric least squares smoothing. Leiden University Medical Centre Report, pages 1–24, 2005.

[252] Tsutomu Arakawa, Daisuke Ejima, Tiansheng Li, and John S Philo. The critical role of mobile phase composition in size exclusion chromatography of protein pharmaceuticals. Journal of pharmaceutical sciences,99(4):1674–1692,2010.

[253] W Kopaciewicz and FE Regnier. Nonideal size-exclusion chromatography of proteins: eﬀects of ph at low ionic strength. Analytical biochemistry, 126(1): 8–16, 1982.

[254] Karel Stul´ık,ˇ Vˇera Pacáková, and Marie Tichá. Some potentialities and drawbacks of contemporary size-exclusion chromatography. Journal of biochemical and biophysical methods, 56(1):1–13, 2003.

[255] G Brent Irvine and Christopher Shaw. High-performance gel permeation chromatography of proteins and peptides on columns of tsk-g2000-sw and tsk-g3000- sw: a volatile solvent giving separation based on charge and size of polypeptides. Analytical biochemistry,155(1):141–148,1986.

[256] John F Carpenter, Theodore W Randolph, Wim Jiskoot, Daan JA Crommelin, C Russell Middaugh, and Gerhard Winter. Potential inaccuratequantitation and sizing of protein aggregates by size exclusion chromatography: essential need to use orthogonal methods to assure the quality of therapeutic protein products. Journal of pharmaceutical sciences,99(5):2200–2208,2010.

[257] J. Brange and L. Langkjcer. Insulin structure and stability. Stability and characterization of protein and peptide drugs: case histories,5:315,1993.

[258] TL Blundell, JF Cutﬁeld, SM Cutﬁeld, EJ Dodson, GG Dodson, DC Hodgkin, and DA Mercola. Three-dimensional atomic structure of insulin and its rela- tionship to activity. Diabetes, 21(Supplement 2):492–505, 1972.

[259] Edward N Baker, Thomas L Blundell, John F Cutﬁeld, Susan MCutﬁeld, Eleanor J Dodson, Guy G Dodson, Dorothy M Crowfoot Hodgkin, Roderick E Hubbard, Neil W Isaacs, Colin D Reynolds, et al. The structure of2znpigin- sulin crystals at 1.5a resolution. Philosophical Transactions of the Royal Society of London B: Biological Sciences,319(1195):369–456,1988.

[260] Bruce K Milthorpe, Lawrence W Nichol, and Peter D Jeﬀrey. The polymerization pattern of zinc (ii)-insulin at ph 7.0. Biochimica et Biophysica Acta (BBA)-Protein Structure,495(2):195–202,1977.

[261] Allen H Pekar and Bruce H Frank. Conformation of proinsulin. comparison of insulin and proinsulin self-association at neutral ph. Biochemistry, 11(22): 4013–4016, 1972.

144 [262] A.K. Attri, C. Fernández, and A.P. Minton. pH-dependent self-association of zinc-free insulin characterized by concentration-gradient static light scattering. Biophysical Chemistry,2010. [263] Peter D Jeffrey, Bruce K Milthorpe, and Lawrence W Nichol. Polymerization pattern of insulin at ph 7.0. Biochemistry,15(21):4660–4665,1976. [264] Ewan J Nettleton, Paula Tito, Margaret Sunde, Mario Bouchard, Christo- pher M Dobson, and Carol V Robinson. Characterization of the oligomeric states of insulin in self-assembly and amyloid fibril formation by mass spectrometry. Biophysical journal,79(2):1053–1065,2000. [265] Teng-Chieh Yang and Nasib Karl Maluf. Self-association of the adenoviral l4- 22k protein. Biochemistry,49(45):9830–9838,2010. [266] Rolando DeAngelis. Personal communication, 2015. [267] Patrick H Brown and Peter Schuck. Macromolecular size-and-shape distributions by sedimentation velocity analytical ultracentrifugation. Biophysical journal,90(12):4651–4661,2006. [268] Emre Brookes, Weiming Cao, and Borries Demeler. A two-dimensional spectrum analysis for sedimentation velocity experiments of mixtures with heterogeneity in molecular weight and shape. European Biophysics Journal,39(3): 405–414, 2010. [269] Anastasia V Pivovarova, Natalia A Chebotareva, Ivan S Chernik, Nikolai B Gusev, and Dmitrii I Levitsky. Small heat shock protein hsp27preventsheat- induced aggregation of f-actin by forming soluble complexeswithdenatured actin. Febs Journal,274(22):5937–5948,2007. [270] Erik J Bonten, Yvan Campos, Viateslav Zaitsev, Amanda Nourse, Brett Wad- dell, William Lewis, Garry Taylor, and Alessandra d’Azzo. Heterodimerization of the sialidase neu1 with the chaperone protective protein/cathepsin a pre- vents its premature oligomerization. Journal of Biological Chemistry,284(41): 28430–28441, 2009. [271] Jerry Tsai, Robin Taylor, Cyrus Chothia, and Mark Gerstein. The packing density in proteins: standard radii and volumes. Journal of molecular biology, 290(1):253–266, 1999. [272] Dow Corning. Fluids, emulsions and dispersions chart. http: //www.dowcorning.com/content/publishedlit/Fluids_Emulsions_ Dispersions_Chart.pdf. Accessed: 2014-11-06. [273] Sarita Naik, David Kerr, Joseph Begley, and John Morton. Influence of local skin temperature and choice of insulin analog on catheter occlusion rates during continuous insulin infusion: An exploratory study. Diabetes technology & therapeutics,14(11):1018–1022,2012.

145 [274] Christie R Beatson. personal communication, 2011.

146 APPENDIX A

SUPPLEMENTAL FIGURES FOR CHAPTER II

In Chapter II, resonant mass measurements (RMM) were made on all samples during the agitation study, both during agitaiton in vials and in syringes. However, as discussed in Chapter V, in its current iteration, the RMM instrument has numerous drawbacks that limit the conﬁdence I have in the data generated from it. Neverthe- less, for completeness, I have included the data here with limited discussion. A full discussion of the results is presented in Chapter II.

Methods

Resonant Mass Measurement (RMM)

Negatively and positively buoyant masses for protein and silicone oil particles from 0.5 micron to 5 micron in size were measured on an Archimedes particle metrology system (Malvern Instruments Ltd., Malvern, United Kingdom) using a “micro” sensor chip. Protein density was assumed to be 1.4 g/mL [271], and silicone oil density was assumed to be 0.96 g/mL [272]. Samples were filtered through a 5µm (Millex-SV PVDF, EMD Millipore, Billerica, MA) syringe filter to minimize occlusion of the microfluidics in the instrument or the sensor chip during analysis. Samples were loaded into the system for 20 seconds and analyzed for a total time of 5 minutes. Data were exported from the Archimedes’ ParticleLab software(version1.9.50)and analyzed for total positive and negative buoyant mass using custom computer scripts written in Perl.

147 Results and Discussion

Vial Agitation: IVIG in Glycine Buﬀer

Particles generated from agitating IVIG in glycine buffer, in vials, were analyzed by RMM (Figure 43) in addition to MFI (Figure 4). However, there are some important features, specific to RMM, to note before discussing our results. First, RMM measurements have a slightly smaller size-limit of detection than the MFI, ∼0.5µm versus ∼1.88µm respectively. Second, RMM samples were filtered through a 5µm filter before analysis to prevent occlusion of the instrument microfluidics. This removes a portion of particles for the RMM samples that are present in the MFI samples. This is unavoidable unfortunately, as the microfluidics clog veryeasilywithlargeparticles, and are very difficult to remove subsequently. Third, the RMM instrument measures particles by buoyant mass, which allows us to distinguish protein particles (negatively buoyant) from silicone oil droplets (positively buoyant). However, it is possible that protein molecules could be adsorbed to the silicone oil droplets, and it is therefore possible that some silicone oil-protein particles would be neutrally buoyant and would not be detected by RMM. As measured by RMM, the formation of negatively buoyant particles was observed for both the siliconized stopper and the fluoropolymer surface over the course of the study. The differences in protein particle formation between the RMM and MFI for IVIG agitated in contact with the fluoropolymer surface are likely due differences lower size-limit of detection between the two instruments. A negligible amount of negatively buoyant particle formation was observed in the glass-only system. How- ever, this does indicate that the air-water interface, in combination with the glass interface, was sufficient to produce some protein aggregation during agitation. With all three surfaces, the addition of polysorbate 20 to the formulation greatly decreased protein particle formation.

148 Negatively Buoyant Particles

8,000 Gore Gly Gore Gly+TW 20 6,000 Siliconized Gly Siliconized Gly+TW20 Glass Gly 4,000 Glass Gly+TW 20

2,000

0 Particle Concentration (ng/mL) 0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 24 24 24 24 Agitation Time (hrs) Positively Buoyant Particles 8,000 Gore Gly Gore Gly+TW 20 6,000 Siliconized Gly Siliconized Gly+TW20 Glass Gly 4,000 Glass Gly+TW 20

2,000

0 Particle Concentration (ng/mL) 0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 24 24 24 24 Agitation Time (hrs)

Figure 43: IVIG agitated in vials, in 0.2M Glycine buﬀer (pH 4.25), with the experimental ﬂuoropolymer-based surface (Gore), atraditional siliconized surface and a glass surface as measured by RMM. Data are expressed as total buoyant mass measured per volume sample for negatively buoyant particles (top plot) and positively buoyant particles (bottom plot). Data are plotted as mean ± standard deviation of three independent replicates.

149 Interestingly, very few positively buoyant, silicone oil particles were observed by RMM for the siliconized stopper vials, in contrast to the MFI data. However, it is possible that some of the silicone oil droplets were removed during ﬁltration of the samples before analysis by RMM. No positively buoyant particles were observed for samples agitated in contact with the ﬂuoropolymer surface nor in the glass-only system.

Vial Agitation: IVIG in PBS Buﬀer

By RMM (Figure 44), we observed much greater levels of protein particles for IVIG agitated in the PBS buffer in comparison to agitation in the glycine buffer—especially note the differences in the ordinate axis scale between Figure 44 and Figure 43. The buffer effect was observed for IVIG samples agitated in the presence of all three surfaces. However, with all three surfaces—even in this suboptimal buffer and pH— the addition of polysorbate 20 to the PBS buffer reduced the amount of protein particle formation observed. A negligible amount of positively buoyant particles were observed by measurement with RMM for all three surfaces.

Vial Agitation: Avastin

No particle formation, either positively or negatively buoyant was observed by RMM for Avastin samples agitated in contact with any of the three surfaces tested (data not shown).

Syringe Agitation: IVIG in Glycine Buﬀer

No protein particle formation was observed by RMM for samples agitated in contact with the ﬂuoropolymer interface (Figure 45). However, when samples were agitated in the siliconized syringe, IVIG formed signiﬁcant levels of protein particles during agitation as assessed by MFI and RMM.

150 Negatively Buoyant Particles

80,000 Gore PBS Gore PBS+TW 20 60,000 Siliconized PBS Siliconized PBS+TW20 Glass PBS 40,000 Glass PBS+TW 20

20,000

0 Particle Concentration (ng/mL) 0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 24 24 24 24 Agitation Time (hrs) Positively Buoyant Particles

80,000 Gore PBS Gore PBS+TW 20 60,000 Siliconized PBS Siliconized PBS+TW20 40,000 Glass PBS Glass PBS+TW 20

20,000

0 Particle Concentration (ng/mL) 0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 0 2 4 24 24 24 24 Agitation Time (hrs)

Figure 44: IVIG agitated in vials, in PBS buffer (pH 7.4), with theex- perimental fluoropolymer-based surface (Gore), a traditional siliconized surface and a glass surface as measured by RMM. Data are expressed as total buoyant mass measured per volume sample for negativelybuoyant particles (top plot) and positively buoyant particles (bottom plot). Data are plotted as mean ± standard deviation of three independent replicates. Significant negatively buoyant particle formation was observed with the siliconized stopper in the PBS buffer without surfactant. All other conditions showed no or negligible particle formation by RMM.

151 Negatively Buoyant Particles 30,000 Gore Gly Gore Gly+PS20 20,000 Siliconized Gly Siliconized Gly+PS20

10,000

0 Particle Concentration (ng/mL) 0 1 3 7 0 1 3 7 0 1 3 7 0 1 3 7 Agitation Time (days) Positively Buoyant Particles 30,000 Gore Gly Gore Gly+PS20 20,000 Siliconized Gly Siliconized Gly+PS20

10,000

0 Particle Concentration (ng/mL) 0 1 3 7 0 1 3 7 0 1 3 7 0 1 3 7 Agitation Time (days)

Figure 45: IVIG agitated in 0.2M glycine buﬀer (pH 4.25), in a bare-glass syringe with a ﬂuoropolymer coated plunger and in a traditional siliconized syringe with a siliconized rubber plunger as measured by RMM. Data are expressed as total buoyant mass measured per volume sample for negatively buoyant particles (top plot) and positively buoyant particles (bottom plot). Data are plotted as mean ± standard deviation of three independent replicates except for the siliconized sample without polysorbate 20 on day 7, which is plotted as the mean and range of two independent replicates.

152 Syringe Agitation: IVIG in PBS Buﬀer

Small levels of protein particles were observed by RMM (Figure46)forsamplesagi- tated in contact with the ﬂuoropolymer surface. In siliconized syringes, we observed the formation of both protein particles and silicone oil particles by both MFI (Fig- ures 12 and RMM (Figure 46).

Syringe Agitation: Avastin

I observed no signiﬁcant particle formation for Avastin when agitated with the ﬂuo- ropolymer surface by RMM at 1 mg/mL or at the full formulation concentration (25 mg/mL) (Figure 47). In the siliconized syringes, by RMM, I observed substantial formation of both protein particles and silicone droplets 47).

153 Negatively Buoyant Particles 25,000 Gore PBS 20,000 Gore PBS+PS20 Siliconized PBS 15,000 Siliconized PBS+PS20 10,000

5,000

0 Particle Concentration (ng/mL) 0 1 3 7 0 1 3 7 0 1 3 7 0 1 3 7 Agitation Time (days) Positively Buoyant Particles 25,000 Gore PBS 20,000 Gore PBS+PS20 Siliconized PBS 15,000 Siliconized PBS+PS20 10,000

5,000

0 Particle Concentration (ng/mL) 0 1 3 7 0 1 3 7 0 1 3 7 0 1 3 7 Agitation Time (days)

Figure 46: IVIG agitated in PBS buﬀer (pH 7.4), in a bare-glass syringe with a ﬂuoropolymer coated plunger and in a traditional siliconized syringe with a siliconized rubber plunger as measured by RMM. Data are expressed as total buoyant mass measured per volume sample for negatively buoyant particles (top plot) and positively buoyant particles (bottom plot). Data are plotted as mean ± standard deviation of three independent replicates except for the siliconized sample without polysorbate 20 on day 7, which is plotted as the mean and range of two independent replicates.

154 Negatively Buoyant Particles 5,000 Gore, 25mg/mL 4,000 Gore, 1mg/mL Siliconized, 25mg/mL 3,000 Siliconized, 1mg/mL 2,000

1,000

0 Particle Concentration (ng/mL) 0 3 7 0 3 7 0 3 7 0 3 7 Agitation Time (days) Positively Buoyant Particles 5,000 Gore, 25mg/mL 4,000 Gore, 1mg/mL Siliconized, 25mg/mL 3,000 Siliconized, 1mg/mL 2,000

1,000

0 Particle Concentration (ng/mL) 0 3 7 0 3 7 0 3 7 0 3 7 Agitation Time (days)

Figure 47: Avastin agitated in syringes with the experimental fluoropolymer-based surface plunger (Gore) in a bare glass syringe and a traditional siliconized plunger and siliconized glass surface as measured by RMM. Data are expressed as total buoyant mass measured per volume sample for negatively buoyant particles (top plot) and positively buoyant particles (bottom plot). Data are plotted as mean ± standard deviation of three independent replicates. Silicone oil particles were shed from the siliconized syringes. No significant particle formation was detected during the course of agitation for the fluoropolymer surface.

155 APPENDIX B

EXTENT OF ADSORPTION OF PHENOLIC PRESERVATIVES BY DIFFERENT BRANDS OF COMMONLY USED INSULIN PUMP CATHETERS

Abstract

The thermal stability “fast acting” insulin analogs is aﬀected by the concentration of phenolic preservatives in the insulin formulation. For insulin analogs, phenolic preservatives in the pharmaceutical formulation act both asanantimicrobialagent and as a stabilizing excipient promoting insulin analog self-assembly from the fragile monomer into the more stabile hexameric conformation. Additionally, binding phenolic compounds induces a stabilizing conformational change to the insulin analog monomers within the hexamer that confers additional stability. However, previous studies have documented losses in phenolic preservative concentration by absorption into the insulin infusion pump catheters. However, the most recent results are nearly adecadeold,andgiventhepaceofadvancesinmedicaldevicedevelopment, the subject should be revisited. In this work, I measured the amount of phenolic preservative remaining in six commonly available insulin pump catheter sets. After 46.5 hours of incubation at 37◦C, more than 97% of the formulation level of phenolic preservative were depleted from the formulation in all of the catheter brands except for one.

Introduction

Depletion of phenolic preservatives in the catheters of insulin infusion pumps has been observed for nearly three decades [127–129]. Phenolic preservatives not only confer antimicrobial protection to the formulation, but also, in the case on insulin analogs, promote self-assembly and stabilization of the therapeutic protein [125, 211].

156 Depleted levels of phenolic preservatives may lead to increased aggregation of insulin analogs [248], and ultimately may possibly be an underlying cause of the ongoing problem of catheter occlusion in insulin infusion pumps [205,218–227,273]. To my knowledge, the most recent published work on phenolic depletion was done by DeFelippis et al. in 2006 [129]. Using a laboratory-based simulation of insulin pump use, DeFelippis et al. found that preservative levels for insulin lispro were depleted more than 50% by 48 hours of testing at 37◦Cwithaddedmechanical agitation. DeFelippis et al. noted that the depleted levels of phenolic preservative they observed still were sufficient to ensure antimicrobial protection, but neglected to address the effects of phenolic depletion on insulin stability. Teska et al. [248] and Woods et al. [211] have shown that the concentration of phenolic preservative in the insulin analog formulation can greatly affect the aggregation rate of the analog. For insulin lispro, the insulin analog examined by DeFelippis et al., Teska et al. found that insulin lispro was very susceptible to accelerated degradation during incubation at 37◦Cwhenitsformulationwasdepletedofphenolicpreservatives. In light of the effects of antimicrobial preservative concentration on insulin analog stability, the possible contributions of diminished insulin analog stability to catheter occlusion, any potential changes in catheter plastics in theyearssinceworkdoneby DeFelippis et al. [129], I designed a study aimed at better understanding the extent of phenolic preservative depletion within different brands of currently available insulin infusion pump catheters.

Materials and Methods

Insulin lispro’s formulation buffer consisted of 1.88 mg/mL sodium phosphate buffer with 16 mg/mL glycerol, 3.15 mg/mL meta-cresol (m-cresol) and 0.0197 mg/mL zinc [229]. For this study, zinc was not included in the buffer. All laboratory chemicals used were of analytical grade or higher and were used without further purification.

157 Water used in the insulin lispro buffer was purified through a Millipore Synergy UV (Billerica, MA) filtration unit. Catheter tubing sets were provided by Dr. Sam Ellis from the Sandra Davis Diabetes Center.

Preparation and Incubation of the Catheter Tubing

The incubation time in this study was based on an estimation ofa“worstcase”sce- nario of residence time for a formulation within an insulin pump catheter and was calculated as follows: (1) The interior diameter of each catheter was determined by measuring the outer diameter of the catheter and the height ofthecathetertubing when compressed. (2) The thickness of the catheter wall was estimated by com- pressing the catheter with the calipers, the compressed height measurement was then taken to be twice the thickness of the catheter wall. (3) Two times the thickness of the catheter wall was subtracted from the measurement of the outer diameter yielding the inner diameter. For a 60cm catheter (a common available length of insulin pump catheter) and assuming a low level, 0.5 units of insulin per hour (U/hr), basal flow of insulin with no bolus injections [274], for a 100U/mL insulin analog formulation (3.47mg/mL insulin analog concentration) the residence time of a plug of fluid flowing from the pump to the inject port averaged ∼47 hours (Table 4). Lengths of different brands of catheter tubing were cut to 40cm. This length was selected to provide sufficient volume for UV absorbance measurements of the phenolic preservative content. Lispro buffer containing 3.15mg/mL of the antimicrobial preservative metacresol (m-cresol) was injected through the catheter tubing using a syringe until the tubing was completely filled. The catheter tubing was then clamped on both ends using binder clips and was incubated at 37◦Cfor46.5hours.Follow- ing incubation at 37◦Cfor46.5hours,theformulationwascollectedbydrainingthe catheters into polypropylene microcentrifuge tubes (Thermo Fisher Scientific Inc., Pittsburg, PA) and analyzed by UV-absorbance spectroscopy for remaining m-cresol

158 Table 4: Dimensions and 0.5U/hr residence times of catheter sets studied.

Wall Thickness Residence Time Sample Manufacturer Model Lot Length (cm) ODa (mm) IDb (mm) (mm) at 0.5U/hr (hrs)

A Medtronic Quick-Set Paradigm 9201417 60 1.499 0.394 0.711 47.7

159 B Maersk Medical Deltec Cozmo Essential 536957 110 1.473 0.406 0.660 75.4

C Unomedical Contact Detach 4262389 60 1.499 0.406 0.686 44.3

D Smiths Medical Cleo 90 189X76 61 1.448 0.368 0.711 48.5

E Unomedical Comfort Short 600330 60 1.473 0.381 0.711 47.7

F Medtronic Silhouette 570338 110 1.486 0.432 0.622 66.9 a OD: Outer Diameter b ID: Inner Diameter concentration.

UV Absorbance Spectroscopy

Ultraviolet absorbance was measured on an Agilent 8540 spectrophotometer (Ag- ilent Technologies, Santa Clara, CA) using a 0.1mm pathlength 583.65-Q-0.1/Z15 cell (Starna Cells Inc., Atascadero, CA). Before measurements the spectrometer was zeroed in lispro buﬀer prepared without m-cresol. Samples were measured and absorbance values were compared to the absorbance of the original formulation buﬀer which was stored in a glass container at 4◦C for the duration of the study.

Results and Discussion

Remaining m-cresol in the insulin lispro formulation after incubation at 37◦Cfor46.5 hours in equal lengths of different brands of catheters is shown in Figure 48. For all brands except the Smiths Medical Cleo 90, less than 3% of the formulation m-cresol level remained after the incubation period. The results of this study, building on previous studies [127–129], highlight the enduring issue of phenolic preservative depletion in insulin infusion pump catheters. findings of earlier studies. In light of the recent findings by Teska et al. [248] detailing the effects of phenolic preservative depletion on insulin analog stability, the results of this study underscore the need for the development of new catheter plastics that minimize phenolic depletion.

Limitations and Future Directions

Unfortunately, given material constraints, I was only able tocompleteasinglerepli- cate of each catheter brand which limits these results. In thefutureitwouldbe advantageous to complete the same incubation study with multiple replicates of each catheter brand to establish the variability of the phenolic depletion within on brand.

160 40%

30%

20%

10% Percent Remaining m-Cresol 0%

A B C D E F Catheter

Figure 48: Remaining m-cresol in the lispro formulation buﬀerafter46.5 hours of incubation at 37◦Cindiﬀerent brands of insulin pump catheters. Catheter brands: (A) Medtronic Quick-Set Paradigm, (B) Maersk Medi- cal Deltec Cozmo Essential, (C) Unomedical Contact Detach, (D) Smiths Medical Cleo 90, (E) Unomedical Comfort Short, and (F) Medtronic Sil- houette.

161 Additionally, these catheters are not manufactured by these companies, they are manufactured by another company and re-branded for each company’s product listed above. It may be possible that all of these catheters are sourced from the same manufacturer, although this would need to be investigated. In addition to this study examining the extent of phenolic depletion in insulin pump catheters, it would be interesting to also examine the rate of phenolic depletion. This could be approached in two different ways. First, one could incubate lengths of catheter for different lengths of time and measure the residual phenolics at certain intervals. Second, one could devise a “flow-through” setup todynamicallymeasure the phenolic preservative concentration as it flows through the catheter. This could be accomplished by either using a flow cell on a UV spectrometer, or using the UV detector of a HPLC (or equivalent). This would give a more realistic picture of the phenolic depletion happening in the catheter as seen by patients.

Acknowledgments

I would like to thank Isabel Randolph for her help running the experiments for this study. I would also like to thank Dr. Sam Ellis for gifting me several catheter sets for use in this study.

162