<%BANNER%>

Pharmacokinetic and Pharmacodynamic Modeling of Insulin Following Different Routes of Administration in Healthy and Diab...

Permanent Link: http://ufdc.ufl.edu/UFE0024918/00001

Material Information

Title: Pharmacokinetic and Pharmacodynamic Modeling of Insulin Following Different Routes of Administration in Healthy and Diabetic Subjects
Physical Description: 1 online resource (149 p.)
Language: english
Creator: Potocka, Elzbieta
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: clamp, glucose, inhaled, insulin, pharmacodynamics, pharmacokinetics
Pharmacy -- Dissertations, Academic -- UF
Genre: Pharmaceutical Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Diabetes is one of the largest public health concerns worldwide, with the prevalence of type 1 and type 2 diabetes projected to more than double by the year 2030. Insulin therapy is used routinely in diabetic patients to manage their response to glucose, however, available insulin products do not match the healthy body's insulin concentration-time profile very well. As a result, insulin treatment often results in poor glucose control and associated adverse events. With a number of new insulins in development, an understanding of the insulin pharmacokinetic (PK) profile and its relationship to the resulting effect is an important element to the design of improved insulin therapy. It was the purpose of this research to: 1) characterize insulin PK, 2) determine the effect that demographic variables have on insulin exposure, and 3) to characterize the relationship of insulin pharmacokinetics and pharmacodynamics (PK/PD). To this end, insulin administered via different routes was modeled using a population approach, revealing the two compartment characteristics of insulin pharmacokinetics, with the differences in the shapes of the insulin concentration-time profiles attributable to differences in absorption of the various formulations and routes of administration. For all insulins included in this analysis, age was the only covariate that was found to affect insulin PK regardless of route of administration, with an increase in the central volume of distribution with increasing age. The rate of inhaled insulin absorption from the lung was found to decrease with increasing age, and increasing BMI was associated with a decrease in insulin absorption rate following subcutaneous administration. Analysis of the effect of pharmacokinetically diverse insulins, as determined by the glucose infusion rate (GIR) in glucose clamp studies, found that the relationship between insulin PK and PD was well described by an Emax model, once the hysteresis was collapsed using an effect compartment. The model was expanded to include data from subjects with type 2 diabetes, revealing similar pharmacodynamic parameter estimates for both populations, with an approximately three-fold increase in insulin EC50 in the type 2 diabetic population compared to healthy subjects. This difference was attributed to the decreased insulin sensitivity that is associated with this disease state. The relationships described by the models presented in this dissertation can be readily applied to the development of novel insulins, potentially resulting in improved insulin therapy.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Elzbieta Potocka.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Derendorf, Hartmut C.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024918:00001

Permanent Link: http://ufdc.ufl.edu/UFE0024918/00001

Material Information

Title: Pharmacokinetic and Pharmacodynamic Modeling of Insulin Following Different Routes of Administration in Healthy and Diabetic Subjects
Physical Description: 1 online resource (149 p.)
Language: english
Creator: Potocka, Elzbieta
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: clamp, glucose, inhaled, insulin, pharmacodynamics, pharmacokinetics
Pharmacy -- Dissertations, Academic -- UF
Genre: Pharmaceutical Sciences thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Diabetes is one of the largest public health concerns worldwide, with the prevalence of type 1 and type 2 diabetes projected to more than double by the year 2030. Insulin therapy is used routinely in diabetic patients to manage their response to glucose, however, available insulin products do not match the healthy body's insulin concentration-time profile very well. As a result, insulin treatment often results in poor glucose control and associated adverse events. With a number of new insulins in development, an understanding of the insulin pharmacokinetic (PK) profile and its relationship to the resulting effect is an important element to the design of improved insulin therapy. It was the purpose of this research to: 1) characterize insulin PK, 2) determine the effect that demographic variables have on insulin exposure, and 3) to characterize the relationship of insulin pharmacokinetics and pharmacodynamics (PK/PD). To this end, insulin administered via different routes was modeled using a population approach, revealing the two compartment characteristics of insulin pharmacokinetics, with the differences in the shapes of the insulin concentration-time profiles attributable to differences in absorption of the various formulations and routes of administration. For all insulins included in this analysis, age was the only covariate that was found to affect insulin PK regardless of route of administration, with an increase in the central volume of distribution with increasing age. The rate of inhaled insulin absorption from the lung was found to decrease with increasing age, and increasing BMI was associated with a decrease in insulin absorption rate following subcutaneous administration. Analysis of the effect of pharmacokinetically diverse insulins, as determined by the glucose infusion rate (GIR) in glucose clamp studies, found that the relationship between insulin PK and PD was well described by an Emax model, once the hysteresis was collapsed using an effect compartment. The model was expanded to include data from subjects with type 2 diabetes, revealing similar pharmacodynamic parameter estimates for both populations, with an approximately three-fold increase in insulin EC50 in the type 2 diabetic population compared to healthy subjects. This difference was attributed to the decreased insulin sensitivity that is associated with this disease state. The relationships described by the models presented in this dissertation can be readily applied to the development of novel insulins, potentially resulting in improved insulin therapy.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Elzbieta Potocka.
Thesis: Thesis (Ph.D.)--University of Florida, 2009.
Local: Adviser: Derendorf, Hartmut C.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2011-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024918:00001


This item has the following downloads:


Full Text

PAGE 1

1 PHARMACOKINETIC AND PHARMACODY NAMIC MODELING OF INSULIN FOLLOWING DIFFERENT ROUTES OF ADMINISTRATION IN HEALTHY AND DIABETIC SUBJECTS By ELIZABETH POTOCKA A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2009

PAGE 2

2 2009 Elizabeth Potocka

PAGE 3

3 To my Dad

PAGE 4

4 ACKNOWLEDGMENTS I would like to thank Dr. Hartm ut Derendorf for his encouragement to take the giant leap into the return to graduate school, his guidan ce and advice, as well as his supervision and encouragement during this project. I would also like to thank Dr. Robert Baughman for not only giving me the opportunity to finish my degree, but to experience so many new sides of drug development and to grow to a great extent as a pharmaceutical scientist. His encouragement and support both at work and during my research were invaluable. I would also like to acknowledge and thank the other members of my committee, Dr. Mark Atkinson, Dr. Gunther Hochhaus and Dr. Veronika Butterweck, for their a dvice and support with this project. I would like to thank MannKind Corporation for providing me w ith such exciting data, as well as the means and time to complete my research. My years with MannKind gave me not only the chance to finish my degree, but also an amazing opportunity to broaden my knowledge of the drug development process, including th e exciting and unforgettable experience of contributing as an author to the Technosphere Insulin NDA. I express my deepest gratitude for the support, advice and friendship to all my collea gues, especially Jim Cassidy. Id also like to thank Mella Diaz and Pam Haworth for their efforts in running one of my key research studies. Finally, Id like to extend my thanks to Dr. Klaus Rave and th e rest of the Profil staff, for conducting most of the studies that I have used in my research. Their excellent standards made it that much easier to see the relationships in my modeling work. I would also like to thank my friends in Gainesville, especi ally Viki Keener and Oliver Grundmann. Their help and advice made many challenges easier, a nd their friendship is one of the greatest gifts I leave with. My special thanks go out to Viki, not only for the wonderful friendship, but also opening her home to me on countless occasions. I d also like to thank

PAGE 5

5 Ichiban Sushi for its existence, and for making my years, as well as the subsequent trips to Gainesville, that much brighter. Most importantly, my thanks go to my family. I wanted to thank my parents for instilling in me the drive to see my goals and aspirations through, and for encourag ing me during my years working on this degree, and all the other years of my life. Finally, I wanted to thank my husband, whose patience and support knows no end, and whose encouragement and faith in me were so often the only things th at got me to the finish line.

PAGE 6

6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ............................................................................................................... 4LIST OF TABLES .........................................................................................................................10LIST OF FIGURES .......................................................................................................................11ABSTRACT ...................................................................................................................... .............13 CHAP TER 1 INTRODUCTION .................................................................................................................. 15Diabetes Mellitus ....................................................................................................................15Insulin ....................................................................................................................... ..............16Carbohydrate Metabolism ............................................................................................... 17Fat Metabolism ................................................................................................................18Protein Metabolism and Growth ..................................................................................... 19Normal Physiologic Insulin Secretion as a Component of Carbohydrate Metabolism ... 19Insulin Secretion in the Type 2 Diabetic .........................................................................21Insulin Therapy ................................................................................................................22Clinical Pharmacology St udies Involving Insulin .................................................................. 28Glucose Clamp Studies ....................................................................................................28Data Analysis of the Glucose Infusion Rate ....................................................................29Insulin Pharmacokinetics and Pharmacodynamics ................................................................. 30Insulin Pharmacokinetics ................................................................................................. 30Insulin Clearance .............................................................................................................31Insulin Pharmacodynamics ..............................................................................................32Insulin Pharmacokinetic and Pharmacodynamic Modeling ...................................................33Models of Insulin PK/PD ................................................................................................33Empirical Models and Mechanistic Models .................................................................... 34Effect compartment models ..................................................................................... 34Indirect effect models ............................................................................................... 36Relationship between GIR and blood glucose ......................................................... 38Hypothesis and Objectives ..................................................................................................... 392 PHARMACOKINETIC MODEL FOR INTRAVENOUS, SUBCUT ANEOUS AND INHALED INSULIN IN HEALTHY SUBJECTS ................................................................ 46Background .................................................................................................................... .........46Materials and Methods ...........................................................................................................46Study Population .............................................................................................................46Study Design and Drug Administration ..........................................................................47Drug Administration and Insulin Concentrations ............................................................48Baseline Insulin Correction Methodology ....................................................................... 48

PAGE 7

7 Data Excluded from the Pharmacokinetic Analysis ........................................................ 49Noncompartmental Pharmacokinetic Anal ysis and Absolute Bioavailability ................ 49Pharmacokinetic Model and Data Analysis .................................................................... 50Pharmacokinetic base model ....................................................................................51Error model .............................................................................................................. 51Reparametrization and indi vidual predicted values .................................................52Model fit assessment ................................................................................................ 53Covariate analysis ....................................................................................................53Results .....................................................................................................................................54Study Population .............................................................................................................54Noncompartmental Pharmacokinetic Analysis ............................................................... 55Pharmacokinetic Model ................................................................................................... 55Covariate analysis ....................................................................................................55Final model ............................................................................................................... 56Discussion .................................................................................................................... ...........57Conclusions .............................................................................................................................593 PHARMACODYNAMIC MODEL FOR INTRAVENOUS, SUBCUTANEOUS AND INHALE D INSULIN IN HEALTHY SUBJECTS ................................................................ 66Background .................................................................................................................... .........66Materials and Methods ...........................................................................................................67Study Population .............................................................................................................67Study Design and Drug Administration ..........................................................................67Drug administration and serum insulin concentrations ............................................ 68Glucose infusion rates and blood glucose concentrations ........................................69Baseline Correction and Smoothing Methodology ......................................................... 69Insulin Data for the Pharmacokinetic/Pharmacodynamic Analysis ................................ 69Pharmacodynamic Model and Data Analysis .................................................................. 70Pharmacodynamic model ......................................................................................... 70Error model .............................................................................................................. 71Model Fit Assessment .............................................................................................. 72Results .....................................................................................................................................72Discussion .................................................................................................................... ...........74Conclusions .............................................................................................................................774 POPULATION PHARMACOKINETIC MODEL FOR SUBCUT ANEOUSLY ADMINISTERED REGULAR HUMAN INSULIN, INSULIN LISPRO AND INHALED INSULIN IN HEALTH Y AND DIABETIC SUBJECTS ................................... 85Background .................................................................................................................... .........85Materials and Methods ...........................................................................................................86Study Population and Study Design ................................................................................ 86Healthy volunteers ....................................................................................................87Type 2 diabetic subjects ........................................................................................... 87Type 1 diabetic subjects ........................................................................................... 88Baseline Correction Methodology ...................................................................................89

PAGE 8

8 Data Excluded from the Pharmacokinetic Analysis ........................................................ 89Noncompartmental Pharmacokinetic Anal ysis and Absolute Bioavailability ................ 90Statistical Analysis .......................................................................................................... 91Population Pharmacokinetic Analysis ............................................................................. 92Pharmacokinetic base model ....................................................................................92Error model .............................................................................................................. 93Reparameterization and individual predicted values ............................................... 94Model Fit Assessment .............................................................................................. 95Covariate Analysis ...........................................................................................................95Model Validation .............................................................................................................96Results .....................................................................................................................................96Patient population ............................................................................................................96Noncompartmental Analysis ........................................................................................... 96Statistical Analysis .......................................................................................................... 97ANOVA test for equality of means .......................................................................... 97Differences in means between the four studies ........................................................ 97Population Pharmacokinetic Analysis ............................................................................. 98Base model ...............................................................................................................98Covariate analysis ....................................................................................................98Final model ............................................................................................................. 100Model Validation ...........................................................................................................100Discussion .................................................................................................................... .........101Conclusions ...........................................................................................................................1055 PHARMACODYNAMIC MODEL FOR SU BCUTANEOUSLY ADMINISTERED REGULAR HUMAN INSULIN, INS ULIN LISPRO, AND INHALED INSULIN IN HEALTHY VOLUNTEERS AND TYPE 2 DIABETIC SUBJECTS .................................114Background .................................................................................................................... .......114Materials and Methods .........................................................................................................115Study Population ...........................................................................................................115Study Design and Insulin Concentrations .....................................................................115Healthy volunteers ..................................................................................................115Type 2 diabetic subjects ......................................................................................... 117Glucose Infusion Rates and Bl ood Glucose Concentrations .........................................117Baseline Correction .......................................................................................................118GIR Smoothing Methodology .......................................................................................118Insulin Data for the Pharmacokinetic/Pharmacodynamic Analysis .............................. 118Data Excluded from the Pharmacokinetic Analysis ...................................................... 119Noncompartmental Pharmacokinetic and Pharmacodynamic Analysis ........................ 119Statistical Analysis ........................................................................................................ 120Pharmacokinetic/Pharmacodynamic Model and Data Analysis .................................... 121Pharmacokinetic and pharmacodynamic model ..................................................... 121Error model ............................................................................................................ 122Results ...................................................................................................................................123Patient Population ..........................................................................................................123Pharmacokinetic and Pharmacodynamic Results .......................................................... 123

PAGE 9

9 Noncompartmental analysis ................................................................................... 123Pharmacokinetic and pharmacodynamic analysis ..................................................124Discussion .................................................................................................................... .........125Conclusions ...........................................................................................................................1286 CONCLUSIONS .................................................................................................................. 135LIST OF REFERENCES .............................................................................................................142BIOGRAPHICAL SKETCH .......................................................................................................149

PAGE 10

10 LIST OF TABLES Table page 2-1 Summary of demographics ................................................................................................ 61 2-2 Mean (%CV) noncompartmental pharmacokinetic parameter estimates .......................... 61 2-3 Covariate selection .............................................................................................................61 2-4 Population pharmacokinetic parameters of insulin ............................................................ 62 3-1 Model A insulin population pharmacodynamic parameter estimates ................................ 78 3-2 Model B insulin population pharmacodynamic parameter estimates ................................ 78 4-1 Summary of demographics and baseline characteristics ................................................. 107 4-2 Mean (%CV) noncompartmental insulin pharmacokinetic parameter estimates ............ 107 4-3 Tukey 95% simultaneous confidence intervals................................................................ 107 4-4 Univariate analysis ...........................................................................................................108 4-5 Final model: Backwards elimination .............................................................................. 108 4-6 Population pharmacokinetic parameter estimates of insulin ........................................... 109 5-1 Summary of demographics and baseline characteristics ................................................. 129 5-2 Mean (%CV) noncompartmental pharmacokinetic parameter estimates ........................129 5-3 Mean (%CV) pharm acodynam ic parameters ................................................................... 129 5-4 Population pharmacokinetic parameters of insulin .......................................................... 130 5-5 Pharmacodynamic parameters of insulin in healthy and type 2 diabetic subjects ........... 130

PAGE 11

11 LIST OF FIGURES Figure page 1-1 Relativ e risks for the development of vari ous complications as a function of mean HbA1c during follow up in the DCCT ..............................................................................42 1-2 Insulin and glucose concentrations in h ealthy subjects during a hyperglycem ic clamp study ........................................................................................................................ ......42 1-3 Insulin secretion: Non-diabetic subjects ........................................................................... 43 1-4 Insulin and glucose response in healthy subj ects and subjects w ith type 2 diabetes following meal consum ption ..............................................................................................43 1-5 Insulin and glucose response in subjects with type 2 diabetes following meal consum ption with and without insulin infusion .................................................................44 1-6 GIR vs. time profiles for two different dose strengths of insulin ...................................... 44 1-7 Effect compartment model ................................................................................................. 45 1-8 Indirect pharmacodynamic models .................................................................................... 45 1-9 Pharmacokinetic/pharmacodynamic model diagram ......................................................... 45 2-1 Pharmacokinetic model diagram ........................................................................................62 2-2 Mean insulin concentration-tim e profiles for all dose groups ........................................... 63 2-3 Goodness of fit plots ..........................................................................................................63 2-4 Population predicted concentration-time profiles and observed da ta by dose group ........ 64 2-5 Example observed and predic ted concentration-tim e profiles ...........................................65 2-6 Example observed and predic ted concentration-tim e profiles ...........................................65 3-1 Pharmacokinetic/pharmacodynamic model diagram ......................................................... 79 3-2. Mean insulin and GIRtim e profiles by dose groupp ......................................................... 79 3-3 Hysteresis in the insulin-GIR re lationship ......................................................................... 80 3-4 Goodness of fit plots for Model A ..................................................................................... 80 3-5 Model A: Population predicted and observed GIR values by dose group ........................ 81 3-6 Model A: Model predicted and obs erved GIR values in four subjects ............................. 82

PAGE 12

12 3-7 Goodness of fit plots for Model B ..................................................................................... 82 3-8 Model B: Population predicted and observed GIR values by dose group ........................ 83 3-9 Model B: Predicted and obser ved G IR values in four subjects ........................................ 84 4-1 Pharmacokinetic model diagram ......................................................................................110 4-2 Mean insulin concentration-time profiles by dose and treatment .................................... 110 4-3 Base model goodness of fit plots of pr edicted and individual predicted insulin concentrations ................................................................................................................ ..111 4-4 Base model goodness of fit pl ots of weighted residuals .................................................. 111 4-5 Final model goodness of fit plots of pr edicted and individual predicted insulin concentrations ................................................................................................................ ..112 4-6 Final model goodness of fit pl ots of weighted residuals .................................................. 112 4-7 Simulated covariate effect on insulin pharmacokinetics ................................................. 113 5-1 Pharmacokinetic/pharmacodynamic model diagram ....................................................... 131 5-2 Insulin and GIR-time profiles .......................................................................................... 131 5-3 Dose proportionality assessment ...................................................................................... 132 5-4 Hysteresis in the insulin-GIR relationship for healthy a nd type 2 diabe tic subjects .......132 5-5 Goodness of fit plots for the pharmacokinetic model ...................................................... 133 5-6 Goodness of fit plots for the pharmacodynamic model ................................................... 133 5-7 Individual predicted GIR versus observed GIR by subject ............................................. 134

PAGE 13

13 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy PHARMACOKINETIC AND PHARMACODY NAMIC MODELING OF INSULIN FOLLOWING DIFFERENT ROUTES OF ADMINISTRATION IN HEALTHY AND DIABETIC SUBJECTS By Elizabeth Potocka August 2009 Chair: Hartmut Derendorf Major: Pharmaceutical Sciences Diabetes is one of the largest public health concerns worldwide, with the prevalence of type 1 and type 2 diabetes projected to more than double by the year 2030. Insulin therapy is used routinely in diabetic patients to manage their response to glucose, however, available insulin products do not match the healthy bodys insulin concentration-time profile very well. As a result, insulin treatment often results in poo r glucose control and associated adverse events. With a number of new insulins in development, an understanding of the insulin pharmacokinetic (PK) profile and its relationship to the resulting effect is an im portant element to the design of improved insulin therapy. It was the purpose of this research to: 1) ch aracterize insulin PK, 2) determine the effect that demographic variables have on insulin exposur e, and 3) to characte rize the relationship of insulin pharmacokinetics and pharmacodynamics (PK/PD ). To this end, insulin administered via different routes was modeled using a populati on approach, revealing the two compartment characteristics of insulin pharmacokinetics, with the differences in the shapes of the insulin concentration-time profiles attributable to differe nces in absorption of the various formulations and routes of administration.

PAGE 14

14 For all insulins included in this analysis, age was the only covariate that was found to affect insulin PK regardless of r oute of administration, with an in crease in the central volume of distribution with increasing age. The rate of inhaled insulin absorption from the lung was found to decrease with increasing age, and increasing BM I was associated with a decrease in insulin absorption rate following subcutaneous administration. Analysis of the effect of pharmacokinetica lly diverse insulins, as determined by the glucose infusion rate (GIR) in glucose clamp studi es, found that the relationship between insulin PK and PD was well described by an Emax model, once the hysteresis was collapsed using an effect compartment. The model was expanded to include data from subjects with type 2 diabetes, revealing similar pharmacodynamic parame ter estimates for both populations, with an approximately three-fold increase in insulin EC50 in the type 2 diabetic population compared to healthy subjects. This difference was attributed to the decreased insulin sensitivity that is associated with this disease state. The relationships described by the models pres ented in this dissert ation can be readily applied to the development of novel insulins, po tentially resulting in improved insulin therapy.

PAGE 15

15 CHAPTER 1 INTRODUCTION Diabetes Mellitus Diabetes m ellitus has reached epidemic pr oportions in the Unite d States and, more recently, worldwide. The morbidity and mortality associated with diabetes is anticipated to account for a substantial proportion of health care expenditures, and according to the American Diabetes Association (ADA), the prevalence of diab etes will continue to grow. The number of people in the U.S. with diagnosed diabetes reach ed 17.5 million with the national cost of diabetes exceeding $174 billion in 2007 [1]. This estimate includes $116 billion in medical expenditures attributed to diabetes, as well as $58 billion in reduced national productivity. According to the ADA, approximately $1 out of every 10 health care dollars spent in the U.S. is attributed to diabetes [1]. Looking beyond American borders, th e incidence of diabetes is increasing not only in developed nations, but at an alarmingly high ra te worldwide. The prevalence of type 1 and type 2 diabetes is projected to more than doubl e, from 177 million cases in the year 2000 to 366 million in the year 2030. In fact, by the year 2030, India is projected to have the highest prevalence of diabetes, with an estimated 79 million cases [2]. Diabetes mellitus is a chronic metabolic di sorder characterized by hyperglycemia caused by defective insulin secretion, re sistance to insulin action, or a combination of both. Long-term diabetes-related complications involve microvascular (such as re tinopathy, nephropathy, and neuropathy) and macrovascular (e g, stroke, coronary heart diseas e) complications as well as alterations of lipid and protein metabolism [3]. The primary clinical parameter to assess patient glycemic control is glycosylated hemoglobin (HbA1c), a form of hemoglobin used primarily to identify the average plasma glucose concentrat ion over prolonged periods of time. Landmark studies such as the Diabetes Control and Complications Tr ial (DCCT) and the United Kingdom

PAGE 16

16 Prospective Diabetes Study (UKP DS) have clearly demonstrated the benefit of better glycemic control in subjects with both type 1 and type 2 diabetes [4], as exhibited in the reduction of HbA1c and relative risk of co mplications (Figure 1-1). Most cases of diabetes mellitus fall into the two broad etiologic categories of type 1 or type 2 diabetes mellitus. Type 1 diabetes is characterized by loss of the insulin-producing -cells of the islets of Langerhans in the pa ncreas, leading to a deficiency of insulin. The majority of type 1 diabetes is immune-mediated, where -cell loss results from a T-cell mediated autoimmune attack [5]. Type 2 diabetes mellitus is characterized differently, due to insulin resistance or reduced insulin sensitivity combined with relative ly reduced, and sometimes an absolute, lack of insulin secretion. There are numerous theories as to the exact ca use and mechanism in type 2 diabetes. It is believed that there are both genetic and envir onmental components that affect the risk of developing the disease. There is a high prevalence for developing type 2 diabetes in siblings and offspring of affected indivi duals, and certain ethnic groups, such as Hispanics, African Americans and Polynesian Islander s have a higher incidence of t ype 2 diabetes. Both findings suggest a genetic link, but the genetic mechanis ms are not known. Obesity is also known to predispose individuals to developing type 2 diabet es [6], and other factors include age, family history, lifestyle choices and environmental exposures. Insulin Insulin is a peptide hormone com posed of 51 amino acid resi dues with a molecular weight of 5808 Da. It is produced in the -cells of the islets of Langerhan s in the pancreas, and exerts its actions on global metabolism, including glucos e disposal and utilization by the body, fat metabolism and protein synthesis. Insulin is sy nthesized as a single chain precursor, in which

PAGE 17

17 the A and B chains are connected by C-peptide. C-peptide is secreted in equimolar amounts with insulin, and is often used as a meas ure of acute insulin secretion [6]. Carbohydrate Metabolism The hum an body aims to maintain blood glucose concentrations within the narrow range of 80-115 mg/dL, and insulins primary role is to maintain homeostasis by keeping glucose levels within this range. Glucose is the principal stim ulus to insulin secreti on and, most commonly, the source of this glucose is dietar y. Immediately following a meal, glucose absorbed into the blood is detected by the -cells causing a rapid secret ion of insulin. In turn, this causes a rapid uptake, storage and use of glucose by al most all tissues of the body. Insulin is secreted into the por tal circulation for immediate pr esentation to the liver, which is exposed to the highest insulin concentrations, and which may be a target of the oscillatory nature of insulin secretion [7]. As insulin is secreted, it also directly affects the pancreatic cells and inhibits the production of glucagon, a counterregulatory hormone to insulin, which promotes hepatic conversion of st ored glycogen into glucose. This hormone is also regulated by blood glucose and free fatty acid levels [8]. One of the most important effects of insulin is that which it exerts on the liver. Insulin simultaneously shuts down hepatic glucose produc tion and, at the same time, stimulates glucose uptake by the liver. Insulin enha nces liver glucose absorption by increasing the activity of glucokinase. By stimulating glyc ogen synthase, insulin also causes most of the glucose absorbed after a meal to be stored almost immediately in the liver in the form of glycogen. The liver serves as the largest storage unit of glycogen, whic h is converted into glucose and released back into the circulation to be used as an energy source during times of fasting (glycogenolysis). Elevated insulin concentrations inactivate liver phosphorylase, the principal enzyme responsible for the conversion of glycogen to gl ucose and hence, inhibit glycogenolysis. At

PAGE 18

18 higher concentrations, insulin also inhibits gluc oneogenesis [9], a metabolic pathway that results in the generation of glucose from non-carbohydrate carbon substrates such as lactate, glycerol, and glucogenic amino acids, and is the source of blood glucose after an extended period of fasting, when glycogen stores have been depleted The inhibition of gluconeogenesis is more complex than glycogenolysis. Insulin decreases quantities and activities of enzymes required for gluconeogenesis, but also, its e ffect on fat metabolism limits the required precursors for this metabolic pathway. It has been hypothesized that the disruption of this signaling and the increased presence of free fatty acids (FFAs) in the bloodstream are an im portant factor in the pathophysiology of type 2 diabet es, with a direct impact on insulin sensitivity [10-12]. In the muscle, insulin induces the redist ribution of the GLUT4 transporter from intracellular storage sites to th e plasma membrane. Once at th e cell surface, GLUT4 facilitates the passive diffusion of circulat ing glucose down its concentration gradient into muscle cells, causing rapid transport of glucose into the cells. The abundance of glucos e causes the cells to use glucose preferentially over fatty acids for energy, and glucose not used for energy in the post-prandial period is converted in the muscle cells to glycogen and stored for later use. Fat Metabolism In the presence of insulin, the increase in glucose utilizati on by the bodys tissues autom atically decreases the utilization of fat as an energy source and promotes FFA synthesis by the liver. Insulin directly lowers FFA presen ce in the bloodstream by 1) facilitating the storage of FFAs in adipose tissue and 2) inhibiting the re lease of FFAs from the adipose tissue back into the bloodstream. In the long run, insulins effect on fat metabo lism is equally important to its effect on glucose. When insulin presence is diminishe d, all aspects of fat breakdown and use for energy are enhanced, and FFAs become the main energy substrate used by essentially all body tissues,

PAGE 19

19 with the exception of the brain. This, in turn, l eads to an increase in lipid concentration, in particular cholesterol, in the bloodstream and the development of atherosclerosis in diabetic subjects. Furthermore, the use of fat for ener gy causes the formation of excessive amounts of acetonic acid, which leads to ketosis and acidosis, coma and even death in subjects with severe diabetes. Protein Metabolism and Growth During the hours after a m eal, not only carbohydrates and fats, but also proteins are stored in the tissues, and insulin stimulates transport of many amino acids into cells. It also increases the translation of mRNA, and in creases the rate of transcri ption of selected DNA genetic sequences, thus increasing protei n synthesis and functioning as a growth promoter in certain cases [6]. Finally, insulin inhibits the catabolism of proteins, in particular, in the muscle cells. When insulin is not available, virtually all pr otein storage comes to a halt; protein catabolism increases, protein synthesis stops and large quanti ties of amino acids appear in the plasma, to be used for energy directly or as substrates for gluc oneogenesis. The resul ting protein wasting is a serious effect of severe diab etes, leading to extreme weakne ss as well as many altered organ functions. Normal Physiologic Insulin Secretion as a Component of Carbohydrate Metabolism Nor mal physiologic insulin secretion consis ts of 2 components: a steady low insulin release to maintain basal glucose levels within a narrow range in between meals, and prandialrelated rapid insulin surges, secreted in res ponse to meals in order to control and limit postprandial glucose (PPG) excursions [13]. In sulin response can further be characterized by qualitative and dynamic features, consisting of an initial quick spike in insulin secretion (what has been termed the earlyor first phase) which occurs and reaches a peak quickly, followed

PAGE 20

20 by a more gradual insulin peak (the lat e phase) which corresponds qualitatively and quantitatively to the elevated presence of glucose in the bloodstream [6]. The biphasic nature of insulin secretion has been observed under e xperimental conditions during both an intravenous glucose tolerance test (IVGTT) [14-16] and under hyperglycemic clamp conditions [17]. This is demonstrated in Figure 1-2, where the initial insulin spike subsequent to glucose infusion (termed the first phase) is then followed by a steady response to the elevated presence of glucose in th e bloodstream during the clamp procedure. Physiological models attempting to explai n the above phenomenon includes the storagelimited model developed by Grodsky et al [18] on the basis of experimental data of insulin secretion obtained in vitro from the isolated perfused pancreas. This model assumes that the cell contains two distinct pools of insulin granules: a small (2 %), labile pool accessible for immediate release and a larger (98%), stable pool that feeds slowly into the labile pool. The signal-limited model was developed by Cerasi et al [19]. These investigators abandoned the pool model in favor of the idea that biphasic insulin release is the result of the dynamic interaction between stimulatory and inhibitory events init iated by glucose, each of them having its own kinetics and dose dependence. A clear-cut first-phase insulin response can be elicited in vivo resorting only to stimuli that rapidly elevate blood glucose concentration. Ho wever, such stimuli do not occur naturally following a meal, and when glucose concentrat ion increases graduall y, the insulin response measured in the peripheral blood does not bear a clear sign of a biphasic shape. In healthy individuals, insulin concentrati ons are clearly elevated over the first 60 minutes following a meal (Figure 1-3) but the first-phase becomes blunted in response to the continuous appearance of glucose. In fact, a distinction has been made in describing these early insulin responses, terming

PAGE 21

21 the insulin response to a square wave glucose prof ile as first phase, and the blunted response to a meal-like glucose profile as early phase. Although no clear first phase spike is disti nguishable following a meal, it has been suggested that the same -cell dynamic properties that are able to generate a biphasic secretion in response to a brisk, intravenous gl ucose challenge are still operating in res ponse to the gradual entry of glucose from the gut, and that the diffe rence between first and early phase is driven by glucose appearance rate [17]. Regardless of the mechanism driving the initial surge of insulin following a meal, its presence is believed to be a critical factor in maintaining glucose homeostasis in the healthy individual through the prompt inhibition of endogenous glucose production [17, 20-22]. Work by Cherrington et al [9] has demonstrated that because insulin exerts its effect directly on the liver, it has a significant eff ect on hepatic glucose output within minutes and, hence, the early insulin response is believed to be aimed at quickly shifting glucose metabolism from the fasting to the prandial state. Insulin Secretion in the Type 2 Diabetic A characteristic find ing in the early stages of type 2 diabetes is the loss of first phase insulin secretion in response to intravenous glucose [23-25]. It has been hypothesized that in the early stages of the disease this alteration in the insulin secretion pattern is responsible for the loss of glycemic control, despite an increased insulin production overall. Although the exact mechanism of this relationship has not been comp letely elucidated, a number of theories have been proposed [25-27], all of which suggest the lack of effect on hepatic glucose output, the hypothesized target of the early insulin res ponse. In a hyperglycemic clamp study, Luzi et al demonstrated the almost complete and prol onged suppression of hepatic glucose production (HGP) when the first phase insulin release was simulated, while no difference was observed in peripheral tissue glucose uptake [25] It has been further sugges ted that when the first-phase

PAGE 22

22 insulin response is lacking, the insulin-to-glucagon rati o is rapidly altered in favor of glucagon, thus leading to increased hepatic glucose produc tion and hyperglycemia; restoration of the firstphase insulin profile in plasma, even in the absence of the second pha se, curbs the glucagonmediated rise in plasma glucose concentration [28] Finally, it is clear from clinical studies that subjects with diabetes are unable to match th e magnitude of the early response compared to healthy subjects. This is presumably due to altera tions in insulin secretion patterns, in particular, the loss of the early phase response. The difference in insulin response to a meal was demonstrated by Polonsky, who studied healthy and type 2 diabetic subjec ts matched for age, sex and degree of obesity [13]. In response to the same meal, the overall insulin exposure was not significantly di fferent between groups over the time period studied, but th e magnitude of the insulin res ponse was clearly higher in the healthy subjects immediately following the meal (Figure 1-4). The overa ll glucose levels were approximately 2-3 fold greater in the group with diabetes [13]. In a similar study, Luzio studied subjects with type 2 diabetes [27]. Following a meal, insulin and glucose concentrations were measured. The subjects were then studied ag ain, but an intravenous insulin infusion was administered following the meal, significantl y increasing the insulin concentrations and simulating an early phase insuli n response (Figure 1-5). Blood glucose concentrations were shown to be significantly reduced when the add itional insulin was infuse d, even though the total insulin exposure was only increased over the first hour of the study. This work demonstrated the importance of the insulins early phase in glyc emic control, as well as clearly linked the hyperglycemia observed in type 2 diabetes to its absence. Insulin Therapy Insulin is the m ainstay of diabetes management and is used routinely in both type 1 and type 2 diabetic subjects to mana ge the bodys insulin needs in respons e to glucose. Preparations

PAGE 23

23 of insulin are classified according to their durat ion of action into short, intermediate and long acting. Diabetic subjects with complete or almost complete loss of -cell function, such as type 1 diabetic subjects or subjects w ith advanced type 2 diabetes, re quire insulin replacement for both the basal insulin release and the meal-related insulin surges which are a part of a healthy insulin response. A variety of long acting insulins are co mmercially available, and are combined with prandial insulins, which are administ ered at the time of a meal [6]. Short acting insulins, on the ot her hand, are used for control of prandial glucose in subjects on complete insulin replacement thera py, or subjects requiring only prandial insulin, such as type 2 diabetic subjects with some remaining -cell function. Prandial insulins attempt to reproduce the physiologic se cretion of the pancreatic -cells as closely as possible. Since the shape of the exposure curve in a healthy individual is w holly dependent on glucodynamic changes and blood glucose concentrations, and is a direct response to the needs of the body, it is impossible to truly mimic endogenous insulin respons e. However, certain characteristics, such as a quick peak insulin concentration and an overall exposure that doe s not outlast glucose absorption, would characterize a desirable insu lin pharmacokinetic profile. Unfortunately, commercially available insulin products, administ ered primarily by the subcutaneous route, do not match this profile very well [29]. Subcutaneously administered insulin exhibits a slow and variable absorption profile in contrast to the healthy endogenous postprandial in sulin response, which is characterized by an insulin spike and elevated insulin concentratio ns which match the post-prandially elevated glucose concentrations. Due to this pharmacokinetic difference, exogenous insulin fails to match the early insulin response in thr ee ways: 1) the initial low insulin levels do not match the quick

PAGE 24

24 surge of the early insulin response, 2) slow exogenous insulin absorption does not provide adequate glucose control early following meal i ngestion, and 3) exogenous insulin is associated with a slower decline in blood levels compared to endogenous in sulin, and exposure oftentimes outlasts the timeframe of glucose absorption follow ing a meal. The first two differences result in elevated post prandial glucose c oncentrations (PPG) [30, 31]. Cont rol of PPG is critical, since elevated PPG is an earlier and frequent abnormality in type 2 diabetes and is a stronger predictor of cardiovascular disease than elevation of fa sting blood glucose (FBG) [32]. Furthermore, as HbA1c levels decrease, PPG cont ributes proportionately more a nd more to the glycation of hemoglobin [33]. The difference in exogenous and endogenous insulin duration of insulin exposure is driven by the slow release characteristics of subcut aneous insulin, and results in action profiles which outlast el evated post-prandial blood glucos e. This mismatch in peak delivery and a duration of action that outlasts glucose absorption following a meal can result in postprandial hypoglycemia [34, 35]. Hypoglycemia poses an immediate danger to subjects, as it can lead to coma or even death, but more often is associated with a need to feed the insulin and the resulting insulin therapy-associated weight ga in. Perhaps more importantly, the inability of exogenous insulin to control PPG re sults in the lack of long-term glycemic control, elevated HbA1c levels, and an elevated risk of diabetic complications. Finally, there is evidence that better insulin therapy may be associated with more than providing insulin presence in the bloodstream to cover the glucose absorbed from the meal. Pharmacokinetic properties related le ss to the extent of exposure, a nd rather to its absorption rate and peak most likely play an important role in optimizing the pharmacodynamic response. Although the relationship of different insu lin pharmacokinetics and their respective pharmacodynamics has not been clearly elucidated, it is clear that the relati onship exists, and that

PAGE 25

25 an insulin which more closely resembles hea lthy endogenous insulin secr etion characteristics would result in better pharmacodynamics and improved glycemic control. Types of Prandial Insulins The first insulins to be commercially avai lable were purified proteins from non-human sources, most prominently, forms of porcine insulin. Although still used in other countries, these insulins have been replaced by recombinant human insulin (RHI) in the United States. Administered primarily by the subcutaneous ro ute, RHI is produced by genetic engineering techniques using recombin ant DNA technology [6]. When in solution, both animal source and RHI associates into stable hexamers in the presence of zinc, and commercially available insulins make use of this characteristic to provide stability and to extend the product shelf-life [36]. However, this feature is also a drawback, as the increased molecular weight and steric size of hexamers delay absorption from the subcutaneous injection site, as the molecule ha s to first dissociate into dimers and monomers before it is absorbed [37]. Because of the lengt h of time required for dissociation into the readily absorbed monomer, the absorption profile of subcutaneous RHI is slow, with peak concentrations detected approximately 1.5 to 2.5 hours post-dose, and a peak insulin effect 1.5 to 3.5 hours post-dose [38]. The absorp tion step is so slow in fact, that it drives drug clearance, giving subcutaneous RHI a long presence in th e bloodstream and a 7-8 hour duration of action [38]. The development of rapid acting insulin anal ogs (RAA) was an attempt to address the pharmacokinetic and pharmacodynamic shortcomings of subcutaneously administered RHI. RAAs are insulin molecules containing subtle alterations in amino acid sequence which still bind to the insulin receptor, but have mo re favorable ADME (abs orption, distribution,

PAGE 26

26 metabolism, and excretion) characteristics. Like regular insulin, RAAs exist as a hexamer in commercially available formulations, but unlike RHI, they dissociate in to monomers almost instantaneously following injection, resulting in a much quicker absorption [6]. Eli Lilly & Co. marketed the first insulin analogue, lispro (H umalog ), engineered through recombinant DNA technology so that the penultimate lysine and pro line residues on the C-terminal end of the Bchain were reversed [39]. The side chains of si ngle amino acids at the C-terminal end of the Bchain play a particular role in the self-association of the insulin molecules [40, 41], and when these amino acids are reversed, the binding forces between the two insulin molecules of a dimer are reduced so that the dimerization constant is 3 00 times less than that of human insulin [36]. Lispro can be stabilized as a hexamer in the presence of phenol and zinc; however, the rapid uptake of phenol after injection causes the molecules to quickly dissociate into monomers, which are rapidly absorbed. Although the rapid acting analogues exhibit an earlier peak (tmax of ~50 minutes) and a more rapid clearance compared to subcutaneous RHI, onset of action is still relatively slow compared with prandial glucose absorption (pea k effect at 60-90 minutes post-dose), and the duration of action still exceeds the timing of the return to normal blood glucose levels following a meal (4-6 hours post-dose) [42]. Hence, ev en the RAA insulins do not mimic endogenous insulin release closely enough, and result in substantial post-meal blood glucose excursions and continued incidence of h ypoglycemic events [39]. Inhaled insulins showed promise and were under development by a number of pharmaceutical companies from the late 1990s until the present. Until now, only one inhaled insulin has been approved and marketed. Exubera (Pfizer) was approved in 2006 by the Food and Drug Administration (FDA), and was availa ble commercially until 2007, when Pfizer

PAGE 27

27 withdrew it from the market [43]. The failure of Exubera has been attributed to many factors, ranging from lack of a successful marketing stra tegy, to pricepoint and cost versus benefit considerations, to a cumbersome inhalation devi ce. Whatever the reasons for the lack of Exuberas commercial success, its withdrawal from the market result ed in a rather abrupt end to the development of other similar products with, most prominently, Eli Lilly (AIR Inhaled Insulin) and Novo Nordisk (AERx insulin Diabetes Management System) terminating their prandial inhaled insulin development programs w ithin weeks of the Pfi zer announcement [43]. Exuberas pharmacokinetic (PK) and pharmacodyna mic (PD) profile was similar to that of the subcutaneously administered RAAs, with an insulin peak of about 49 minutes post-dose, an associated peak in activity ranging at approximately 2 hours following dosing, and a duration of action of 6 hours [44]. This similari ty to the RAAs meant that Exubera only offered the convenience of an inhaled formulation that did no t require injection as the only benefit over the existing therapy. The PK and PD properties of the ot her inhaled insulins were similar to that of Exubera [45]. One inhaled insulin still in development is Technosphere Insulin (TI). TI is a novel inhaled insulin whose unique deli very characteristics result in rapid absorption followed by rapid systemic clearance. Followi ng TI administration, insulin Cmax occurs approximately 14 minutes post-dose. Due to its rapid cleara nce, insulin concentrations return to baseline levels much more quickly compared to subcutaneously administered insulin, with little residual effect by three hours post-dose. Data from glucos e clamp studies suggest a much faster onset (within minutes post-dose) and shorter duration of glucodynamic effect (return to baseline by 180 minutes postdose) when compared to subcutaneous insulins [46].

PAGE 28

28 Clinical Pharmacology Studies Involving Insulin The clinical adm inistration of insulin necessitates counteracti ng insulin activity so as to keep the study subject safe. Studies designed to explore the glucodynamic effect of insulin are generally of two types: 1) in sulin action is countered by admini stering the insulin with a meal; or, 2) insulin action is counter ed by an infusion of glucose. This latter study design, commonly referred to as a glucose clamp study, provides the most direct measurement of insulin action. These studies can be subdivided into a few diffe rent types, dependent on the research purpose and/or study conditions. Glucose Clamp Studies Glucose clamp studies are comm only used to st udy diabetes and insulins [9, 17, 35, 39]. The glucose clamp procedure has been a great res earch tool to define several biochemical and feedback mechanisms associated with diabetes, including determination of insulin sensitivity [47], the effects of exercise [48], and counterregulatory and glucagon responses [49]. More recently, the glucose clamp procedure has been used to temporally represent insulin activity. These timeactivity profiles are available for numer ous insulins and insulin analogs in healthy volunteers and subjects with diabetes [50-53]. During a glucose clamp study, subjects are admi nistered insulin, while at the same time they receive a varying infusion of glucose to coun teract insulins action and keep the subject in a constant glycemic state. This methodology allows for the safe dosing of insulin while eliminating the need for a meal to control blood glucose and prevent hypoglycemia. It also allows for a direct way of measuring insulin effects by determining the amount of glucose required to maintain blood glucose levels within a defined range. The subject receives a constant rate insulin infusion to suppr ess the secretion of insulin from the pancreas only during the run-in pe riod, or during the enti re study procedure.

PAGE 29

29 Suppressing endogenous insulin secr etion is important for distingui shing between the effects of the administered insulin and insulin secreted by the pancreas in response to the administered glucose. With endogenous insulin suppressed an d the rate of insulin infusion known, the glucose infusion rate can be used as a measure of th e pharmacodynamic response to the test insulin treatment. The rate of glucos e administration is termed the gl ucose infusion rate or GIR [36]. The clinical setting of these studies makes it easy to obtain numerous blood samples for insulin concentration determination and assessment of insulin pharmacokinetics. It also allows for a direct measurement of blood glucose levels as well as the GIR, both of which have been used as markers of insulin pharmacodynamics. An example of two GIR-time profiles from two different dose strengths of an in sulin preparation is presented in Figure 1-6, and the relative potency can be readily observed. Data Analysis of the Glucose Infusion Rate GIR data has been analyzed in a sim ilar fashion to a noncompartmental analysis of pharmacokinetic data. The total amount of glucose administered is directly related to the overall effect of the insulin administered, and the GIR at any time point is an indi cator of the effect of insulin at that time point. Thus, the maximum GIR (GIRmax) is indicative of the maximum effect, GIR tmax is indicative of its timing, while the area under the glucose infusion time curve (GIR AUC) is indicative of the overall ex tent of the insulin effect. Insulins with greater overall activity have relatively higher GIR AUCs, while insulins with a quicker ons et of activity have a shorter GIR tmax [36]. The relative ease of these calculations and intuitive nature of the GIR-derived parameter interpretation means that extensive use of the da ta has been made in order to compare insulin activity. These pharmacodynamic parameters have been used to asses the onset, duration and

PAGE 30

30 extent of insulin activity, and remain the me thodology of choice when direct pharmacological effect of insulins is compared [36]. Insulin Pharmacokinetics and Pharmacodynamics Insulin Pharmacokinetics Insulin h as been used as a therapeutic ag ent for more than 80 years, and insulin pharmacokinetics have been extensively stud ied since a specific radioimmunoassay became available in 1960 [54]. Alt hough early work suggested a one compartment pharmacokinetic model following intravenous insulin administrati on, advances in analytical technology have made it possible to observe insulin concentrations in the elimination phase. The multiple compartment disposition of intravenous insulin wa s demonstrated in studies utilizing radiolabels over 30 years ago, and more recently, using unlabel ed intravenous insulin and a more sensitive RIA, by Hooper et al [55]. Insulin has been extensively studied following the various routes of administration, with most pharmacokinetic profiles from subcutaneously administered formulations, as this has been the dominant route of administration since insulin therapy was institute d. Unlike intravenously administered insulin, other routes of administrati on have been described, for the most part, using a one compartment model [56-61]. Because the -phase following intravenous dosing lasts about an hour [55], the slow absorption of s ubcutaneously administered insulin obscures the second compartment. Furthermore, because the -half-life following intrav enous dosing is quick (approximately 5-6 minutes) [55], absorption do minates the pharmacokinetics in a phenomenon commonly termed flip-flop kinetics [62], where the terminal phase reflects the slowest step in the sequential/parallel processes of drug absorption, distribution and elimination (ie, the terminal phase reflects absorption not elimination, hence a flip-flop of kinetic processes) [62]. As a result of this, the differences in insulin profiles, especially the terminal phase, observed among

PAGE 31

31 the various non-iv routes of admi nistration reflect the differences in absorption, not elimination. This is well described by asse ssing the clearance of insulin lispro and RHI, two products with differing PK profiles after subc utaneous dosing, but whose clearance, when administered intravenously, is almost identical, indicating that the differences observed in the terminal phase reflect differences in absorption [37]. Insulin Clearance All insulin-sensitive cells rem ove and degrade the hormone. Once an insulin molecule has docked onto the receptor and effected its action, it may be degraded by the cell. Degradation normally involves endocytosis of the insulin-receptor complex followed by the action of insulin degrading enzyme (IDE), which is relatively ub iquitous, and present in all insulin-sensitive tissue. Most evidence supports IDE as the primary degradative mechanism, but other systems (lysosomes and other enzymes) undoubtedly contri bute to insulin metabolism [63]. Since most insulin clearance involves the coupling of insulin and its receptor as a firs t step, insulin clearance is ultimately dependent on the number of insulin receptors, as well as an individuals insulin sensitivity. The liver is the primary site of insulin cleara nce and approximately 50 % of portal insulin is removed during first pass transit [64]. Hepatic uptake is not static, however, and varies with both physiological and pathophysiol ogical factors [63]. The kidne y is another major site of insulin clearance from the systemic circulati on [65], removing approximately 50% of peripheral insulin by two mechanisms: glomerular filt ration and proximal tubul ar reabsorption and degradation [63]. Insulin not cleared by liver and kidney is ultimately removed by other tissues, such as muscle and adipose tissue. Although docking at the insulin recep tor is a first step toward in sulin clearance, removal of insulin from the circulation does not imply im mediate degradation a nd inactivation of the

PAGE 32

32 hormone. A significant amount of receptor-bound insulin is released from the cell and is returned to the circulation intact or partially degraded [ 66]. A model developed by Hovorka et al. [67] estimates that the mean residence time of endogenously secreted insulin is 71 minutes. Of these, 62 minutes is spent bound to the liver re ceptor, 6 minutes is spent bound to peripheral receptors, and 3 minutes is spent in blood or interstitial fluid. With this model, 80% of the total insulin in the body was bound to liv er receptors. Other tissues also transiently bind and can release insulin back into the circulation [63]. Insulin Pharmacodynamics As with m any therapeutic agents, insulin effect is consistent with an Emax model, which is non-proportional in nature [ 68]. Theoretically, the Emax model is based on receptor theory, and is described by a version of the Hill equation: CEC CE Effect 50 max (1-1) where Emax is the maximum effect, C is drug concentration and EC50 is the concentration that causes 50% of the effect. In practice, this equation de scribes a relationship in wh ich, as drug concentrations increase, the effect reaches an asymptote. This is the maximum effect (Emax) the drug can exert, regardless of how far the concen tration is increased. This phenomenon has been observed when multiple doses of insulin are administered, a nd although the relationship remains linear for dose and insulin exposure, a doubling of the dose resu lts in less than doubling of the effect [69]. Hysteresis in Insulin PD Because insulin exerts its activ ity in the periphery (tissue), and measurements for insulin exposure are taken from the central compartment (b lood), there is a disconn ect, or shift, between the rise and fall for insulin concentration in the ce ntral compartment, and the effect that it exerts

PAGE 33

33 peripherally (most often a measure of glucose utilization). The shif t in the PK and PD changes is dependent on the time required to complete the cascade of events composing the cellular mediation of insulin action. Hence, insulins activity cannot be directly related to its presence in the bloodstream, as its pharmacodynamics exhibits a counter-clockwise hysteresis [59]. The hysteresis shape varies following insulins with di fferent pharmacokinetic pr operties, and is most pronounced following intravenous insulin administration. To overcome this complication in determining a relationship between insulin concentration and its effect, a number of techniques have been used to collapse the hysteresis in the PK/PD model development. The effect-compartment link model, first proposed by Sheiner et al in 1979 [70], has successfully been used in a number of models. This model assumes the presence of a hypothetical effect compartment (biophase) a nd proposes that a drug must enter this compartment from the pharmacokinetic (central ) to the peripheral compartment before its pharmacological response is exerted. This model has been criticized by some for the lack of parameters which can readily be translated to describe the physiology and pharmacology of the system [71]. More recently, in direct pharmacodynamic response models have been proposed to describe the insulin pharmacodyna mic response. Both inhibitory [72] and stimulatory [56, 57, 72] models have been used to successfully de scribe insulin action. Such models described insulin effect on the glucose-insulin homeostasis more mechanistically than the biophase models, but tend to require different parameter estimate s for different dose groups [71] which severely restricts any predictive usefulness of the models. Insulin Pharmacokinetic and Pharmacodynamic Modeling Models of Insulin PK/PD Models describing the pharm acokinetic-pha rmacodynamic relationship of insulin have been developed for many purposes. These models ra nge in complexity from a simple description

PAGE 34

34 of the concentration-effect relationship of a single insulin formulation, which tend to be empirically or mechanistically based, to the phys iologically based (PB) models which attempt to describe the physiological situati on as closely as possible for organs, organ systems or even the body as a whole, to the development of an artifi cial pancreas which incorporates a model that completely describes the closed loop system of insulin and glucose. Most of the PB models comprise multiple compartments with complicated blood glucose bioprocesses in order to take into account the counter-regulatory feedback system between insulin and glucose and account for the indir ect relationship between the two [67, 73, 74]. Complex models derived from IVGTT and OGTT [75] using radi olabelled glucose have been used to describe the interrelations hip of insulin and glucose, as we ll as the effect of both glucose and insulin on hepatic glucose pr oduction and tissue disposal. It is the purpose of the work presented here to explore the insulin concentration-effect relationship as determined from glucose clamp da ta, so as to compare the acute insulin effect between different formulations; thus, the more empirical models will be discussed, with examples offered, and the more physiological ly-based pharmacodynamic models will not be presented. Empirical Models and Mechanistic Models Effect compartment models The sim plest models exploring the relations hip between insulin c oncentration and its effect have been based on an Emax relationship between insulin and either the GIR (glucose clamp studies) or blood glucose. As descri bed earlier, under euglycemic glucose clamp conditions, the GIR reflects glucose utilization, and is a measure of total insulin activity. If the contributions of endogenous insulin are suppresse d, the GIR reflects test insulin action, and a number of analyses have been preformed using the GIR as the pharmacodynamic endpoint. The

PAGE 35

35 effect compartment is a hypothetical compartmen t in the model, which makes it possible to account for the delay between insulin dynamics in the central compartment and its action [68]. The model is presented in Figure 1-7. Independent work by Hooper, Tornoe and Woodworth [55, 59, 60] utilized a sequential PK/PD approach, and the hypothetical effect comp artment to take into account the delay in insulin action and effectively collapse the hyste resis observed in their analyses. The work by Hooper et al was based on intravenously administered insulin and, as a resu lt, enables modeling the two compartment pharmacokinetics of insulin [55]. The PD was expressed as a simple gamma-linear model and sigmoidal Emax model with the effect site concentration, described by: e eCEC CE GIR 50 max (1-2) where Emax is the maximum effect, Ce is drug concentration at the effect site, is the shape factor, or Hill coefficient, whos e addition increases the versatility of the model to describe the concentration-effect relationship, and EC50 is the concentration that causes 50% of the effect. Both Tornoe and Woodworth developed pha rmacodynamic models which incorporate the same sigmoidal Emax model to compare subcutaneously admi nistered RHI with other insulins. The model proposed by Tornoe et al estimates the PD parameters individually for Novo Rapid (Novo Nordisk, Bagsvaerd, Denmark) and RHI indi vidually. The model estimates a very similar EC50 and for both insulins, which suggests similar potency for both NovoRapid and RHI [59]. On the other hand, Woodworth et al found that the individually estimated parameters for RHI and NPH (Neutral Protamine Hagedorn) insulin differ substantially in EC50 and [60] using the same model. Although the fits for both insulins are reasonable, the predictive usefulness of their model was lost since the model parameters were estimated for each formulation independently, and cannot be extrapolated to insulins w ith different pharmacokinetic properties.

PAGE 36

36 Indirect effect models Because the effect com partment-based models lack physiological meaning, investigators began using indirect effect models, viewing th em as more mechanistic and less empirical. Studies in animals make it possibl e to explore the relationship betw een the time course of insulin in the blood stream and blood glucose concentrati ons without necessitating either food intake or glucose infusion. This allows for the study of the system without the introduction of exogenous glucose, as well as allows blood glucos e to be the pharmacodynamic endpoint. Interesting work by Lin et al compares the use of the e ffect compartment versus two indirect response models; the stimulatory (incr ease in glucose utiliza tion by the tissues) and inhibitory (decrease in hepatic glucose production) [72]. Unlike the models previously described, the animals (minipigs) were not infused with glucose during the study and blood glucose was used as the pharmacodynamic endpoint. The PK/PD model was sequential, with two compartment characteristics for the insulin-time cu rve. The inhibitory and stimulatory models are presented in Figure 1-8. The two models are described by the following equations: Inhibition model: Rk CpIC Cp k dt dRout in 50 1 (1-3) Stimulation model: R CpSC CpS kk dt dRoutin 50 1max (1-4) where Cp is the plasma insulin concentration, kin is the rate of glucose production, kout is the first order glucose utilization constant IC50 is the insulin concentration that inhibits the maximal production of glucose by 50%, R is the hypoglycemic effect of insulin on blood glucose, Smax is the maximum utilization of blood glucose cont ributed by insulin and SC50 is the insulin concentration that stimulates the maxi mal utilization of blood glucose by 50%.

PAGE 37

37 The authors found that based on the resultant co rrelation coefficient, weighted residuals sum of squares and Akaike information criterion were improved when the indirect models were used as compared to the fit obt ained from the effect compartmen t model. Interestingly, others have noted that the use of the indirect model is better suited for data where counterregulatory mechanisms do not contribute to the system, such as during a glucose clamp procedure [61, 76]. The study conditions most certainly invoked count erregulatory feedback, and this may explain the better fit obtained with th e indirect response model. Gopalakrishnan et al also utilized an indirect model, assuming insulin-induced stimulation of glucose uptake [56]. The data wa s taken from a study in rats, where the animals were administered either subcutaneous or spray-instilled insulin, w ith no additional glucose provided post-dose. The PK and PD components of the analysis were performed in a sequential manner, and the same PD model was successfully applied to both routes of administration. These results are not surprising because ther e appeared to be little difference in the pharmacokinetics of the insulins studied. A lthough the fit was good, the model was never challenged as it was never applied to external data and, in particular, to insulins with different pharmacokinetics properties than those used to develop the model. Subsequently, this model was expanded by Landersdorfer et al and applied to human glucose clamp data from healthy volunteers and s ubjects with T1DM) [77]. In this analysis, the model included an effect compartment to ac count for the delay in insulin action and a stimulatory effect on glucose disposal, making it possible to use blood glucose as the dependent variable. The model is presented in Figure 19. The following equations describe the model: ))()((0 0 0IsIsCeIsIsCpk dt dCee (1-5)

PAGE 38

38 0 )(50 )( 10 0 max ss outinG IsIsSC IsIsS kk dt dG (1-6) where kin is the zero order rate consta nt for glucose input (GIR), kout is the rate of glucose utilization, Is is the insulin concentration, Is0 is the baseline insulin concentration, SC50 is the insulin concentration at half-maximal effect, Smax is the maximal effect, G is glucose concentration and ke0 is the first order rate constant between the central and effect compartments. All data were modeled simultaneously. No si gnificant differences were detected between the inhaled and sc insulins, however, the insulin and GIR profiles were similar, so no differences were expected. As with the work of Gopalakrishnan et al, the model was not challenged with different pharmacokinetic insulin profiles. Howe ver, parameters are estimated for both healthy subjects and subjects with type 1 diabetes, who showed signs of in sulin resistance by higher EC50 and lower Ce estimates. Relationship between GIR and blood glucose Interesting work by Woodworth e t al presents a model which attempts to link a PD model developed using GIR data to gluc ose concentrations [61]. Gluc ose clamp procedures are used extensively to study insulin effect and co mpare the glucodynamic activity of different formulations, however, there is little information on whether GIR accurately represents blood glucose dynamics following insulin administration. Woodworth et al combined data from a glucose clamp study and a study where subjects were dosed with insulin w ithout being clamped, and developed a simple inhibitory Emax model to relate the two: 50 0) ( GIRGIR GIRBGBG BGbase (1-7) where BG is blood glucose at time t from the non-clamp study, GIR is the glucose infusion rate from the clamp procedure at the same time t, GIR50 is the GIR that relates to a 50% maximum

PAGE 39

39 reduction in BG, BGbase is the blood glucose measured at baseline, BG0 is the maximum tolerable reduction in BG, and is the sigmoidicity factor. The Emax model was chosen since both the GIR and blood glucose concentrations were expected to produce a maximum response. Sinc e these responses may correlate to different insulin amounts, the relationship cannot be assumed to be linear, but an Emax model can collapse to a linear relationship if one does not exist. Unfortunately no parameter estimates, or their associated error terms, were provided. However, the model appears predictive when applied to external data, when visually comparing simulate d glucose concentrations overlaid with observed values. The authors conclude that the model is more predictive if c ounterregulatory responses can be avoided. Hypothesis and Objectives The goal of this specific research is the de velopment of a PK/PD m odel for insulins with different pharmacokinetic properties, and the characterization and bridging of the pharmacokinetic and pharmacodynamic properties of a rapidly absorbed inhaled insulin, a subcutaneous insulin and insulin administered intravenously, using a population approach in healthy and type 2 diabetic subjec ts. The main hypothesis to be te sted is that insulin effect is dependent upon the dynamic features of in sulin pharmacokinetics, enabling the insulin pharmacokinetic profile to be used to predict insulin effect. Furthermore, a secondary hypothesis is that insulin effect is altered to some degree in type 2 diabetic subjects, whose sensitivity to insulin has been compromised by the disease state. It is proposed that the development of a PK/PD model will make it possible to: 1. Mathematically express the relati onship between insulin PK and PD 2. Link the pharmacokinetic prope rties of different insulins with their effect 3. Directly compare pharmacokinetically different insulins

PAGE 40

40 4. Account for the effect of different demogr aphic covariates on in sulin pharmacokinetics 5. Elucidate the difference in response to insulin action in healthy vol unteers and subjects with type 2 diabetes 6. Perform model-based simulati on and predictions of insulin activity under different conditions. The development of a successful insulin PK/PD model will aid in understanding the relationship of insulin PK and its effectknow ledge critical to designing improved insulin therapy. The following specific aims are used to further explore the hypothesis: Specific Aim 1: Develop a population pharm acokinetic (PK) model for intravenous, subcutaneous and inhaled insulin in heal thy subjects following a range of doses. A population pharmacokinetic model will be developed for single doses of insulin administered via three different routes of administra tion (inhaled, subcutaneous, and intravenous) to healthy subjects. This population base d model will estimate insulin pharmacokinetic parameters, including relative and absolute bioavailability of inhaled and subcutaneous insulin, as well as identify and incorporate any possi ble demographic covariate effects on insulin pharmacokinetics in hea lthy volunteers. Specific Aim 2: Develop a population pharm acodynamic (PD) model for intravenous, subcutaneous and inhaled insulin in heal thy subjects following a range of doses A population pharmacodynamic model relating seru m insulin concentrations and insulin effect will be elucidated using glucose clamp data from healthy volunteers. Predicted insulin concentrations from Aim #1 will be used as the independent variable in this sequential PK/PD model. The effect variable in the model will be the glucose infusion rate (GIR), a measurement of insulin effect on glucose disposal. A pharm acodynamic model will be developed to describe the temporal dissociation between the pharmacoki netics and pharmacodynamics of insulin. The

PAGE 41

41 model will attempt to describe the shift in hysteresis with respect to the different pharmacokinetic profiles of the three routes of administration, and then to describe the relationship between insulin concentration at the effect site and the response. Specific Aim 3: Develop a population pharmacokinetic model for subcutaneously administered regular human insuli n, insulin lispro, and inhaled insulin in healthy, Type 1 and Type 2 diabetic subjects. A population pharmacokinetic model will be developed for RHI and insulin lispro administered via different routes (inhalation and subcutaneous) to healthy and type 2 diabetic subjects. This popula tion model will estimate insulin pha rmacokinetic parameters, relative bioavailability of inhaled and subcutaneously administered re gular human insulin and a rapid acting analog, as well as determine demographic, route of delivery-specific and disease statespecific covariate effects on insulin pharmacokinetics. Specific Aim 4: Develop a population pharm acodynamic model for inhaled insulin in healthy and Type 2 diabetic subjects. A PK/PD model relating serum insulin concentrations and insulin effect will be developed, using glucose clamp data from healthy and T ype 2 diabetic subjects. Differences in pharmacodynamic parameter estimates will be co mpared across both groups to establish the effect of the disease state on insuli n effect, as described by the GIR.

PAGE 42

42 Figure 1-1 Relative risks for the development of various complications as a function of mean HbA1c during follow up in the DCCT [Source: Skyler, J, Diabetes, 1995. 44(8): p. 968-83] Figure 1-2 Insulin and glucose concentrations in healthy subjects durin g a hyperglycemic clamp study. [Source: Caumo et al, Am J Physiol Endocrinol Metab, 2004. 287(3): p. E37185]

PAGE 43

43 Figure 1-3 Insulin secretion: N on-diabetic subjects [S ource: Aronoff, et al. Diabetes Spectrum. 2004;17: 183-90] 100 200 300 400Glucose (mg/dL)04812162026 Time (h)Meals Diabeticsubjects Controlsubjects 0 20 40 60 80 100 120 48122026 0800 0 0Time (h)Insulin Concentration (uIU/mL) 100 200 300 400Glucose (mg/dL)04812162026 Time (h)Meals Diabeticsubjects Controlsubjects 0 100 200 300 400Glucose (mg/dL)04812162026 Time (h)Meals Diabeticsubjects Controlsubjects 0 20 40 60 80 100 120 48122026 0800 0 0Time (h)Insulin Concentration (uIU/mL) 20 40 60 80 100 120 48122026 0800 0 0Time (h)Insulin Concentration (uIU/mL) Figure 1-4 Insulin and glucose re sponse in healthy subjects and s ubjects with type 2 diabetes following meal consumption [Source: Polonsky, K.S., et al ., N Engl J Med, 1988. 318(19): p. 1231-9]

PAGE 44

44 No insulin Insulin 17.0 11.0 5.0 0 60120180240Glucose (pmol/mL) 23.0 0 Time (min)0.4 0.2 0 60120180240Insulin(pmol/mL) 0 0Time (min)*P<0.05 P<0.01. P<0.01. No insulin Insulin No insulin Insulin 17.0 11.0 5.0 0 60120180240Glucose (pmol/mL) 23.0 0 Time (min)0.4 0.2 0 60120180240Insulin(pmol/mL) 0 0Time (min)*P<0.05 P<0.01. P<0.01. Figure 1-5 Insulin and glucose response in s ubjects with type 2 diabetes following meal consumption with and without in sulin infusion [Source: Luzio et al Diabetes Res, 1991. 16(2): p. 63-7] 050100150200250300350Time (minutes) 0 2 4 6 8 10GIR (mg/kg/min) Figure 1-6 GIR vs. time profiles for two different dose strengths of insulin

PAGE 45

45 ke0EmaxPlasma Insulin (Cp) Effect (Ce) kaGIR k1e ke0EmaxPlasma Insulin (Cp) Effect (Ce) kaGIR k1e Figure 1-7 Effect compartment model A. B. Figure 1-8 Indirect pharmacodynamic models: A) Inhibition and B) Stimulation Figure 1-9 Pharmacokinetic/pharmacodynamic model diagram

PAGE 46

46 CHAPTER 2 PHARMACOKINETIC MODEL FOR INTR AVENOUS, SUBCUT ANEOUS AND INHALED INSULIN IN HEALTHY SUBJECTS Background Insulin disposition following subcutaneous or pulmonary administration has been described by a one compartment pharmacokinetic model [56-61]. Early work also suggested a one compartment model following intravenous admini stration, but with advances in sensitive and specific analytical methodologies, lower concentrat ions of insulin could be quantitated. This increased sensitivity enabled the detection of in sulin concentrations in the elimination phase, revealing multiple compartment disposition followi ng intravenously administered insulin [55]. However, the slow absorption characteristics of both subcutaneously administered insulin and early formulations for delivery via the pulmonary route obscured the distribution phase and the second compartment. Technosphere Insulin (TI), a novel, inhaled, regular human insulin (RHI) whose administration by oral i nhalation results in rapid absorp tion and rapid clearance makes it possible to distinguish the sec ond compartment of the insulin pharmacokinetic profile [78]. The aim of the analysis presented here was to develop a pharmacokinetic model for RHI administered via the intravenous, subcutaneous and inhalation routes, and to demonstrate that two compartment disposition is a characteristic of insulin regardless of route of administration, with the route-dependent concentr ation-time profile differences due primarily to differences in absorption rates. Materials and Methods Study Population Data for this analysis were combined from two separate glucose clamp studies in healthy subjects performed at the same clinical site. Each was a prospective, single-center, open-label, randomized, crossover euglycemic glucose-cl amp study in healthy, non-smoking male and

PAGE 47

47 female volunteers, 18 years of age, with a body mass index of 18 kg/m2 and normal pulmonary function. Each study was local Ethi cs Committee reviewed and approved, and all subjects provided written informed consent prio r to initiation of any st udy-related procedures. Prior to entry into either study, all subjects were administered a physical examination, pulmonary function tests, electrocardiography, and laboratory tests, includi ng urinalysis and screening for drugs of abuse. Study Design and Drug Administration Both studies utilized a euglycemic glucose clamp procedure performe d with the Biostator glucose monitoring and infusion system (Biostator, Life Science Instruments, Elkhart, IN, USA), and a continuous insulin infusion to suppress endogenous insulin production. Following an overnight fast, on each of the treatment days and pr ior to test article administration, the subjects received a 2-hour constant ra te intravenous RHI infusion to establish a serum insulin concentration between 10 U/m L to suppress endogenous insulin secretion. This infusion was continued until the end of each treatment visit. Each subject received a single dose of the test treatment on separate occasions. Study 1: a three-way crossover euglycemic gl ucose-clamp study in five subjects to compare the pharmacokinetics and pharmacodynamics of single doses of 100 U of inhaled TI, 10 U RHI administered subcutaneously and 5 U RHI (Actrapid, Novo Nordisk A/S, Bagsvaerd, Denmark) administered intravenously. Study 2: a four-way crossover euglycemic gluc ose-clamp study in 12 subjects to compare the pharmacokinetics and pharmacodynamics of three different single doses of inhaled TI with a single dose of 10 U RHI (Actrapid, Novo Nordisk A/S, Bagsvaerd, Denmark). Blood glucose was kept consta nt at 90 mg/dL throughout the procedure by a variable infusion of a dextrose solution, controlled by th e Biostator. An additional external pump,

PAGE 48

48 controlled by study personnel, was employed if the Biostator could not meet the glucose infusion requirements to maintain euglycemia. If any glucose was provided by the external pump the GIR from the pump was added to th e GIR of the Biostator. The subjects were maintained in the fasted state until the end of each treatment vis it. The treatment periods were separated by a washout period of 3 to 28 days. Drug Administration and Insulin Concentrations TI was administered using a commercially available inhaler (Model M, Boehringer Ingelheim, Ingelheim, Germany) in Study 1, and via the MedTone Dry Powder Inhaler (Model Alpha, MannKind Corporation, Danbury, CT) in Study 2. RHI insulin was administered by subcutaneous injection in the abdomen. All bloo d samples were drawn from the cubital vein of the arm contralateral to the one used for the continuous insulin infusion and glucose administration via an intravenous catheter. Samples for the determination of insulin concentration were drawn at 120, 90, 60 and 30 minutes before dosing and 0, 1, 3, 7, 12, 20, 30, 45, 60, 90,120, 180, 240, 300 and 360 minutes post-dos e, and analyzed for insulin concentration using radioimmunoassay (RIA) with double determina tions. C-peptide samples were collected at 120, 60 minutes pre-dose and 0, 30, 60, 180, and 300 minutes post-dose. The samples were cooled in an ice-water bath be fore centrifugation. After centrif ugation the plasma samples were immediately frozen and stored at -80C until anal ysis. Glucose infusion rates were registered by the Biostator at every 1 minute from 120 minutes before dosing until 360 minutes post-dose. Baseline Insulin Correction Methodology Insulin concentrations collected at 120, 90, 60 and 30 minutes before dosing were averaged, and subtracted from all subsequently observed insulin concentrations. This method adjusted the serum insulin concentrations to account for the ongoing in sulin infusion. This correction was conducted before the data was an alyzed and used for model development.

PAGE 49

49 Data Excluded from the Pharmacokinetic Analysis Endogenous insulin secretion was assumed to be completely suppressed following insulin administration. However, to ensure that endogenous insulin suppression lasted for the duration of the sampling period, the serum insulin and C-peptide concentration-time profile for each dosing was examined visually. As expected, in subjects with low exposure to the test treatment, endogenous insulin appeared to be c ontributing to the curve in the latter part of the study. Due to the small number of samples collected for C-pep tide assay, a C-peptide correction method [79] could not be applied to the insulin data, and vi sual inspection was used to determine insulin concentrations influenced by the endogenous component. As a resu lt of the lower doses administered in Study 2, a higher incidence of el evated C-peptide concentrations was observed in the latter timepoints following TI treatment, particularly in th e lower dose groups. All insulin concentrations after 180 minutes post-dose in the 25 U dose group, 240 minutes post-dose in the 50 U dose group, and 300 minutes in the 100 U dos e group were excluded, with a total of 51 non-BLQ concentrations excluded from the analysis. Noncompartmental Pharmacokinetic Analysis and Absolute Bioavailability Insulin pharmacokinetics were analyzed using noncompartmental methodology using baseline-corrected insulin concentrations. The following PK parameters were derived using WinNonlin v 5.2 (Pharsight Corporation, Mountain View, CA): observed peak insulin concentration (Cmax), time to peak insulin (tmax), and total insulin exposure as measured by the area under the insulin concentration-time curve from time 0 until the last time point with nonzero insulin concentrations (AUC0-last) calculated by the linear-trap ezoidal method. The terminal elimination half-life (t1/2) was calculated as ln(2)/z in accordance with pharmacokinetic theory [80], where z is the terminal elimination rate cons tant estimated from log-linear regression analysis of the terminal elimination phase of the concentration-time profile, and AUC0was

PAGE 50

50 calculated as AUC0-last + Clast/ z, where Clast is the last observed insulin concentration. Noncompartmental parameter values were used to determine starting estimates for pharmacokinetic modeling, and were compared to parameter values estimated by the final model as part of the model fit assessment. Absolute bioavailability (%F) was determin ed using the mean dose-normalized insulin AUC0-last values, expressed as the group average dose-normalized AUC0-last/average dosenormalized AUC0-last following iv administration. Pharmacokinetic Model and Data Analysis The software package NONMEM (Version VI, Level 1.2, NONMEM Project Group, ICON Development Solutions, USA) was used for the population analysis. The ADVAN6 subroutine and first-order (FO) estimation me thod was used. NONMEM describes the observed concentration-time data in terms of: A number of fixed effect parameters, which may include the mean values of the relevant base pharmacokinetic model parameters or a number of parameters which relate the base model parameters to dem ographic and other covariates; Two types of random effect parameters: (a) 2: the variances of the interindividual variability ( ) within the population, and (b) 2: the variances of the residual intraindividual variability ( ) due to random fluctuations in an individuals parameter values, measurement error, model misspecifica tion, and all sources of error not accounted for by the other parameters. The population or average values of the parameters, the interindividual variances, 2, and the residual variance, 2, were estimated by NONMEM. Subject-specific parameters were calculated by NONMEM using the POSTHOC option. These parameters are empirical Bayesian estimates of the individuals true parameters based on the population parameters and the individuals observed concentrations.

PAGE 51

51 Pharmacokinetic base model Data from all three routes of administrati on were modeled simultaneously. The PK was described by a two compartment model with one (inhaled) or two sequent ial (sc insulin) first order absorption processes and firs t-order elimination [81]. A diag ram of the model is presented in Figure 2-1. The model was described by the following differential equations: 11 1Aka dt dA (2-1) 2211 2AkaAka dt dA (2-2) 31 3Aka dt dATI (2-3) 4 445554223 4AkAkAkAkaAka dt dATI (2-4) 554445 5AkAk dt dA (2-5) The pharmacokinetic structural model was pa rameterized in terms of clearance (CL), volume of distribution in the central compartment (Vc), intercompartmental clearance (Q), the volume of distribution in th e peripheral compartment (Vp), the first order absorption rate constant for TI (kaTI), the two first order absorption rate constants associat ed with the sequential absorption for subcutaneously administered insulin (kasc1and kasc2, depot and a transit compartment) and the absolute bioavailability for subcutaneous insulin and TI (Fsc, FTI). Error model Fixed effects parameters were used to desc ribe the typical population estimates, and an exponential random effect model was used to describe interindividual variability for each model parameter: ii (2-6)

PAGE 52

52 where i is the estimated parameter value for the ith individual, is the fixed effect typical parameter value in the population, and i are individual-specific random effects for the ith individual symmetrica lly distributed with zero mean and variance A combined proportional and additive error model was used to model the residual unexplained variability, as descri bed by the following equation: ij ij ij ijpCCp21)1( ~ (2-7) where Cpij is the observed value of the jth plasma concentration of individual i; ijpC ~ is the predicted jth plasma concentra tion of individual i; and ij is a random variable which represents the discrepancy between the observed and predicted jth concentration. Considerable interoccasion vari ability (IOV) was observed wi thin the TI groups, and was attributed to natural variation in inhalation on different occasions, resulting in differences in relative bioavailability at diffe rent visits, as described by: 3 2 13 2 1OCC OCC OCC FTI TIOCC OCC OCC FTIF (2-8) where OCC1, OCC2 and OCC3 are set to 1 at the corresponding occasion and 0 otherwise. Reparametrization and i ndividual predicted values Individual-specific values of each pharmacokinetic parameter were obtained by Bayesian analysis with the final model. 4VkCL (2-9) 445VkQ (2-10) The following pharmacokinetic parameters coul d subsequently be calculated for each individual according to the following equations based on compartment modeling theory. kkkkkkkk54 5445 54454 ) ( 2 1 (2-11)

PAGE 53

53 kkkkkkkk54 5445 54454 ) ( 2 1 (2-12) 693.0 lifehalf (2-13) 693.0 lifehalf (2-14) ak lifehalf absorption 693.0 (2-15) Model fit assessment Goodness-of-fit was determined by the objec tive function values (OFV) and visual inspection of scatter plots of pr edicted versus observed concentra tions and weighted residuals. For nested models, hypothesis test s were performed based on the likelihood ratio test, in which the change in OFV approximates the 2 distribution. A more comp licated model was preferred when the decrease in OFV was more th an 3.84 (the critical value for the 2 distribution at p< 0.05 with 1 degree of freedom). Covariate analysis Following the determination of the base popul ation model, potential covariates were examined to determine whether they improved the overall fit and reduced variability in the model. These covariates included age, body weight, body height and BMI. Covariates were initially evaluated for possible relationship s with the model estimated pharmacokinetic parameters using the generalized additive model (GAM) procedure in Xpose [82] which incorporates the Akaike Information Criterion (A IC) for covariate identification. Covariates associated with a reduction in the AIC were ev aluated using NONMEM in a stepwise manner. Each covariate added to the base model, and the resulting univariate model was then compared to

PAGE 54

54 the reduced model for significant improvement in fit. Covariates were included in the model if the criteria for nested model cr iteria (a 3.84 point drop in the OFV) was observed (p<0.05). Backward elimination was then performed where each covariate was independently removed from the model to confirm its relevance. An in crease in the OFV of 6.7 (p<0.01) was necessary to confirm that the covariate was significant. Results Study Population All subjects completed the studies and receiv ed all scheduled doses. In Study 1, for one subject in the 5 U iv dose group, the insulin concentration time profile did not match the dosing and sampling times, indicating that possible e rrors may have been made when dosing or sampling times were recorded, and the elapsed time since dose could not be determined with certainty. Since the dosing time could not be determined with certainty, this subject was excluded from the analysis. In Study 2, a subj ect was excluded from an alysis in the 10 U sc treatment group, since three c onsecutive BLQ values were observed in the pharmacokinetic profile following sc RHI treatmen t, suggesting a possible analyti cal error. Furthermore, upon visual inspection of the insulin concentration-tim e profiles, data from one subject appeared to differ markedly from the other subjects, with almost no insulin exposure within the first few hours post-dose, suggesting that the subject may have had difficulty with i nhalation. Statistical analysis determined the subject to be an outlier with respect to at least one pharmacokinetic parameter for each dose group (p< 0.05), and conse quently, this subject was excluded from all the analysis. Dixons test was used on all log-transformed pharmacokinetic parameters, with the exception of tmax, which was tested using non-transformed values and Tukeys test. A total of 650 insulin concentrations from 16 subjects, and 57 profiles were included in the analysis. Demographic data is summarized in Table 2-1.

PAGE 55

55 Noncompartmental Pharmacokinetic Analysis Mean insulin concentration-time profiles are presented in Figure 2-2. Noncompartmental pharmacokinetic parameter estimates are pres ented in Table 2-2. Based on mean dosenormalized AUC0-last, insulin following subcutaneous ad ministration was approximately 53%, and between 10 and 11% following TI administration, and did not appear to depend on TI dose. Terminal insulin half-life and AUC0could not be calculated following subcutaneous dosing due to the prolonged absorption from the tissue, a nd the flip-flop kinetics associated with this route of insulin administration. Pharmacokinetic Model A two-compartment open model with one (inhale d) or two sequential (subcutaneous) first order absorption processes and fi rst-order elimination described the insulin concentration data well. The population typical parameter values in the structural model were: clearance, 43.7 L/hr; volume of distribution in the central compartment, 5.11 L; volume of distribution in the peripheral compartment, 31.6 L. These parameter estimates are in close agreement with values reported in literature for intr avenous insulin [55]. The base model pharmacokinetic parameter estimates are summarized in Table 2-4. Covariate analysis The covariate analysis process is summarized in Table 2-3. A GAM analysis identified BMI as a potential covariate for absorption rate consta nt associated with sc insulin, and body weight as a potential covariate for volume of di stribution in the central compartment; to a lesser extent, age was identified as a potential covariat e for the peripheral volume of distribution. Visual inspection of covariate vs. parameter scat ter plots and graphs of weighted residuals vs. covariates confirmed these findings Because different inhalers we re used in the two studies, a possible effect on TI bioavailability was examined by adding the effect into the univariate

PAGE 56

56 analysis. The lack of trend seen visually was confirmed by a lack of change in OFV when the covariate was tested. The results of a stepwise forward addition showed that following the addition of BMI (p<0.01) as a covariate to the subcutaneous ab sorption rate constant, the further addition of weight or subject age on any of the identified parameters did not significantly contribute to the model. BMI effect was described by the following equation: )65.23/( 1(1, 1 1BMI kakascBMI kasc sc (2-16) where BMI was centered around the population median value. The addition of this covariate decreased the OFV by 9.41 points and also reduced the interindividual CV% on Kasc1 from 57% to 50% (Tables 2-3 and 2-4). Final model The overall model fit was good, with indivi dual predicted versus observed values distributed along the line of unity, and no significant trends in the weighted residuals (Figure 23). Thus, the structural and error models appeared to adequately describe insulin pharmacokinetics and explain the variability in the data. In the final model, the popul ation typical parameter values were: clearance, 43.4 L/hr; volume of distribution in the central compartment, 5.0 L; volum e of distribution in the peripheral compartment, 30.7 L. As with the base model, the parameter estimates are in close agreement with values reported in the litera ture for intravenous insulin [55]. The absolute bioavailability for subcutaneous insulin and TI was 52% and 11 %, respectively, matching both results reported with TI [83] and to results reported with sc RHI [84], as well as the results of the noncompartmental analysis (Table 2-2). Th e pharmacokinetic parameter estimates are summarized in Table 2-4.

PAGE 57

57 The and -half-lives of 5 and 93 minutes, resp ectively, were calculated from the individual predicted parameters and are in close agreement w ith both the insulin half-life estimates following iv administration reported previously [55] and the terminal half-life approximated following noncompartmental analysis (Table 2-2). The population predicted concentration-time profiles are presented in Fi gure 2-4. Following intravenous dosing, the distribution phase was approximately thirty minut es. The apparent distribution phase was longer following TI administration, and undistinguishable from the elimination phase following subcutaneous dosing due to the prolonge d absorption from the injection site. Interoccasion variability following TI admi nistration was approximately 30%. The interindividual %CV for all pharmacokinetic parameter estimates ranged from 10 to 52%. The residual error was described by a combined proportional (25. 1 %CV) and additive (2.27 IU/mL) model. Example observed and predicte d concentration-time profiles are shown in Figure 2-5 and Figure 2-6. Discussion Insulin has been shown to demonstrate a two compartment disposition following intravenous administration, w ith a previously reported -half-life of approximately 5-6 minutes [55]. Because the -phase is short compared to th e duration of absorption following subcutaneous administration, the absorption phas e obscures the initial di stribution, rendering it difficult to distinguish the sec ond compartment. Furthermore, slow absorption dominates the pharmacokinetics of subcutaneously administer ed insulin in a phenomenon commonly termed flip-flop kinetics [59]. As a result of this, the differences in insulin profiles, especially the terminal phase, observed among the various non -iv routes of administration reflect the differences in absorption, not elimination.

PAGE 58

58 Most available assessments of insulin pharmacokinetics use the pharmacokinetic profiles of subcutaneously administered formulations, as this route of administration has been the dominant route since insulins initiation as a therapeutic agen t. Hence, the pharmacokinetic profiles following non-intravenous routes of insu lin administration have been described using a one compartment model [57, 59, 85]. TI is a novel inhaled insulin whose unique delivery char acteristics re sult in rapid absorption and systemic clearance, enabling a cl early distinguishable second compartment in its pharmacokinetic profile. This is unique for an in sulin formulation, and is thought to be due to the combination of: 1) the quick dissolution of the delivery mi croparticles upon c ontact with the lung surface [86]; and, 2) TI is delivered as an insulin monomer to the lung [87] which is the most readily absorbed form of insulin [88]. Because insulin pharmacokinetic properties are expected to be consistent once it is available syst emically [89], the inclusion of data from TI and intravenous dosing, both of which have distinct and phases in their pharmacokinetic profiles, made it possible to demonstrate the two compartm ent disposition of all three routes of insulin administration. Furthermore, it allowed for an estimate of the absorption rate differences between TI and subcutaneously administered insulin. A model incorporating two sequential absorption rates and a transit compartment was used to describe the slower absorption seen with the subcutaneously administered insulin This model was described by Puckett et al [81] and was based on the physiology of subcutaneous insulin administration, where the deport compartment represents the subcutaneous tissu e, and the second compartment represents the interstitial space. Interoccasion inhalation variability can result in differences in the bioavailability of treatments administered through the lung. A lthough this variability may be reduced with

PAGE 59

59 repeated dosing, treatment-nave or inexperienced subjects ma y exhibit significant interoccasion differences in exposure. In this analysis, subject s in one of the studies were administered three different doses of TI on three different occasions The interoccasion variability was successfully modeled to reflect inhalation differences that re sulted in a difference in insulin bioavailability, with no difference in other pharmacokinetic pa rameters. For this group of subjects, the interoccasion variability was estim ated by the model to be approximately 30%. Combined with the variability unaccounted for by the interoccasio n model (11 %CV), the overall interindividual variability was comparable with that of s ubcutaneously administered insulin (52 %CV). Covariate analysis was not expected to result in many findings due to the small number of subjects in this analysis and th e fact that the subjects were yo ung, healthy volunteers. However, even in this population, increas es in BMI were found to decreas e absorption rate when insulin was administered subcutaneously, as reported prev iously by others [90]. This observation may be attributed to the increased thickness of the s ubcutaneous tissue in subjects with higher BMI, which slows absorption from the depot compartment. Since this relationshi p was detected in the healthy population with BMI scores within the normal range, it is expected that a greater impact would be observed on patients taking subcutaneous insulin who have higher BMI, as is often the case in subjects with type 2 diabetes. Conclusions Insulin pharmacokinetics were found to be cons istent with a two-compartment model, with an approximate distribution phase of about ha lf an hour following intravenous dosing. The apparent distribution phase is longer following TI admini stration and undistinguishable following subcutaneous dosing due to the prolon ged duration of absorption. BMI was found to be a significant covariate on in sulin absorption rate following subcutaneous dosing, with an associated decrease in the rate of absorption wi th increasing BMI. In the model presented here,

PAGE 60

60 differences in the shape of the insulin curve fo llowing different routes of administration were successfully attributed to differences in absorption.

PAGE 61

61 Table 2-1 Summary of demographics Demographic Variable Value Age (years) Mean SD 28 4 Weight (kg) Mean SD 76 11 BMI (kg/m2) Mean SD 24 2 Height (cm) Mean SD 179 9 Gender Male N 14 Female N 2 Race Caucasian N 15 Asian N 1 Table 2-2 Mean (%CV) noncompartment al pharmacokinetic parameter estimates Dose Group tmax (h)* Cmax (U/mL) AUC0-last (U/mLmin) Half-life (h) AUC0(U/mLmin) Dose-Normalized AUC0-last (U/mLmin) 5 U iv (n=4) --100.0 (21) 1.7 (23) 103.2 (19) 20.1 10 U sc (n=15) 1.5 30.7 (34) 107.0 (26) --10.7 25 U TI (n=11) 0.2 54.6 (72) 50.0 (61) 1.1 (61) 57.4 (62) 2.0 50 U TI (n=11) 0.2 105.3 (38) 101.5 (39) 1.3 (31) 111.2 (39) 2.0 100 U TI (n=16) 0.33 240.9 (52) 218.6 (43) 1.4 (57) 230.6 (41) 2.2 Median presented for tmax; Table 2-3 Covariate selection Covariate Model Tested OFV Change in OFV Base Model 3695.551 BMI on Kasc1 3686.141 9.410 Body weight on Vc 3694.240 1.311 Age on Vp 3691.658 3.893 BMI on Kasc1 and Age on V p 3682.352 3.789

PAGE 62

62 Table 2-4 Population pharmacokinetic parameters of insulin Pharmacokinetic Parameters Base Model Full Model Parameter Estimate (%RSE) Estimate (%RSE) CL (L/hr) 43.70 (8.6) 43.4 (8.6) Vc (L) 5.11 (10.0) 5.02 (9.6) Q (L/hr) 24.5 (15.4) 23.9 (13.3) V p (L) 31.6 (18.5) 30.7 (15.4) kasc1 (hr-1) 0.52 (17.3) 2.37 (14.4) kasc2 (hr-1) 1.28 (25.4) 1.04 (22.8) kaTI (hr-1) 2.34 (8.7) 2.35 (8.9) Fsc(%) 53 (13.4) 52 (13.0) FTI (%) 11 (12.7) 11 (13.0) BMI covariate effect 0.74 (8.0) Interindividual and Residual Variability (%CV) (%CV) CL 9.2 9.9 Vc 37.8 36.9 Q 31.5 30.7 V p 29.5 27.7 kasc1 57.4 50.1 kaTI 19.1 19.6 Fsc 23.2 24.1 FTI 24.0 23.7 IOV FTI 30.7 30.2 1 26.1 25.1 2 1.3* 1.51 Note: 2 (additive residual error) is expressed in U/mL; Th e magnitude of interindividual and residual variability was expressed as CV%, approximated by th e square root of the variance estimate. Figure 2-1 Pharmacokinetic model diagram

PAGE 63

63 0123456Time (hours) 0 200 400 600Insulin Concentration (uU/mL) 5 U iv 10 U sc 25 U TI 50 U TI 100 U TI Figure 2-2 Mean insulin concentrat ion-time profiles for all dose groups 100 300 500 700 900 Observed Insulin Concentration (uU/mL) 100 300 500 700 900Individual Predicted Insulin Concentration (uU/mL) A 0.1 1.0 10.0 100.0Predicted Insulin Concentration (uU/mL) -5 -3 -1 1 3 5Weighted Residuals B Figure 2-3 Goodness of fit plots. A) Individual predicted versus observed insulin concentrations and B) Predicted insulin concentrat ions versus weighted residuals

PAGE 64

64 0123456Time (hours) 0 40 80 120Insulin Concentration (uU/mL) A 0123456Time (hours) 0 50 100 150Insulin Concentration (uU/mL) B 0123456Time (hours) 0 100 200 300 400Insulin Concentration (uU/mL) C 0123456Time (hours) 0 10 20 30 40 50Insulin Concentration (uU/mL) D 0123456Time (hours) 100101102103Insulin Concentration (uU/mL) E Figure 2-4 Population predicted concentrationtime profiles and observed data by dose group. A) TI 25 U dose, B) TI 50 U dose, C) TI 100 U dose, D) subcutaneous RHI 10 U dose, E) intravenous RHI 5 U dose

PAGE 65

65 0246Time (hours) 0 50 100 150 200Insulin Concentration (uU/mL) sc 10 IU TI 25 U TI 50 U TI 100 UA 0246Time (hours) 100101102Insulin Concentration (uU/mL) sc 10 IU TI 25 U TI 50 U TI 100 UB Figure 2-5 Example observed and predicted concentr ation-time profiles. A) Individual predicted versus observed insulin concentrations for subject 5 (linear scal e) and B) Individual predicted versus observed insulin concentrations for subject 5 (log scale) 0246Time (hours) 10 30 50 70 90Insulin Concentration (uU/mL) sc 10 IU TI 25 U TI 50 U TI 100 U 0246Time (hours) 10-1.0100.0101.0102.0Insulin Concentration (uU/mL) sc 10 IU TI 25 U TI 50 U TI 100 U Figure 2-6 Example observed and predicted concentr ation-time profiles. A) Individual predicted versus observed insulin concentrations for subject 1 (linear scal e) and B) Individual predicted versus observed insulin concentrations for subject 1 (log scale)

PAGE 66

66 CHAPTER 3 PHARMACODYNAMIC MODEL FOR INTRAVENOUS, SUBCUTANEOUS AND INHALE D INSULIN IN HEALTHY SUBJECTS Background The glucose clamp procedure has been us ed to study biochemical and feedback mechanisms affiliated with diabetes, including determination of insulin sensitivity [47], the effects of exercise [48], and counterregulatory and glucagon res ponses [49]. More recently, the glucose clamp procedure has been used to study insulin activity [9, 17, 35, 39]. To assess the response to insulin, plasma glucos e concentration is held constant at basal levels by a variable glucose infusion. Under these steady-state condit ions of euglycemia, the glucose infusion rate equals glucose uptake by all the tissues in the body and is therefore a measure of tissue sensitivity to insulin [14]. By suppressing e ndogenous insulin secreti on one can observe the effects of the administered insulin without confounding additional effect from insulin secreted by the pancreas in response to th e administered glucose. An exogenous insulin infusion employed during the study is assumed to completely supp ress hepatic glucose production, as well as appreciable insulin secretion, allowing the glucose infusion rate (GIR) to be used as a measure of the pharmacodynamic response to the test tr eatment insulin only [36]. Numerous pharmacokinetic/pharmacodynamic (PK/PD) models have been developed to establish a relationship between insulin pharmacokinetics a nd its effect, as assessed by the GIR. Although considerable efforts have been made to model insulin PK/PD, the pharmacokinetic similarity of most insulin formul ations has made it difficult to develop a model which can be applied to pharmacokine tically diverse insulins. Technosphere Insulin (TI) is a novel inhaled insulin whose unique delivery ch aracteristics result in rapid absorption and systemic clearance, even when compared to rapid acting analogs [91]. The aim of this analysis was to develop a PK/PD model for regular human insulin (RHI) following TI, intravenous and

PAGE 67

67 subcutaneous administration, and to better elucid ate the relationship between the vastly different pharmacokinetic profiles of these three insulins with their effect as measured by the GIR. Materials and Methods Study Population Data for this analysis were combined from two glucose clamp studies in healthy subjects performed at the same clinical site. Each st udy was a prospective, sing le-center, open-label, randomized, crossover euglycemic glucose-cl amp study in healthy, non-smoking male and female volunteers, 18 years of age, with a body mass index of 18 kg/m2 and normal pulmonary function. Each study was local ethics committee reviewed and approved, and all subjects provided written informed consent prior to initiating any study-rela ted procedure. Prior to entry into either study, all subjects were administered a physical examination, pulmonary function tests, electrocardiography, and laboratory tests, includi ng urinalysis and screening for drugs of abuse. Study Design and Drug Administration Both studies were euglycemic glucose clamp procedures, and were performed utilizing the Biostator glucose monitoring and infusion system (Biostator, Life Science Instruments, Elkhart, IN, USA), and a continuous insulin infusion to suppress endogenous insulin production. Following an overnight fast and prio r to test article administration, on each of the treatment days the subjects received a 2-hour cons tant rate intravenous RHI infusion to establish a serum insulin concentration between 10 U/mL and suppress e ndogenous insulin secretion. This infusion was continued until the end of each treatment visit. Each subject received a single dose of the test treatment on separate occasions. Study 1: a three-way crossover euglycemic gl ucose-clamp study in five subjects to compare the pharmacokinetics and pharmacodynamics of single doses of 100 U of inhaled TI, 10

PAGE 68

68 U RHI administered subcutaneously and 5 U RHI (Actrapid, Novo Nordisk A/S, Bagsvaerd, Denmark) administered intravenously. Study 2: a four-way crossover euglycemic glucos e-clamp study in 12 subjects to compare single dose of 10 U RHI (Actrapid, Novo Nordisk A/S, Bagsvaerd, Denmark) with single doses of 25, 50 and 100 U of TI. Blood glucose was kept consta nt at 90 mg/dL throughout the procedure by a variable infusion of a dextrose solution, controlled by the Biostator. If the Biostator could not meet the glucose infusion requirements to maintain euglycemia, an additional, external pump controlled by study personnel was employed. If any glucos e was provided by the external pump, that infusion rate was added to the GIR of the Biostato r. The subjects remained fasted until the end of the study. The treatment periods were separa ted by a washout period of 3 to 28 days. Both studies were performed at the same clinical site. Drug administration and seru m insulin concentrations TI was administered using a commercially available inhaler (Model M, Boehringer Ingelheim, Ingelheim, Germany) in Study 1, and via the MedTone Dry Powder Inhaler (Model Alpha, MannKind Corporation, Danbury, CT) in Study 2. RHI insulin was administered by subcutaneous injection in the abdomen. All bloo d samples were drawn from the cubital vein of the arm contralateral to the one used for the continuous insulin infusion and glucose administration via an intravenous catheter. Samples for the determination of insulin concentration were drawn at 120, 90, 60 and 30 minutes before dosing and 0, 1, 3, 7, 12, 20, 30, 45, 60, 90,120, 180, 240, 300 and 360 minutes post-dose, and analyzed for insulin concentration using radioimmunoassay (RIA) with double determina tions. C-peptide samples were collected at 120, 60 minutes pre-dose and 0, 30, 60, 180, and 300 minutes post-dose. The samples were

PAGE 69

69 cooled in an ice-water bath be fore centrifugation. After centrif ugation the plasma samples were immediately frozen and stored at -80C until analysis. Glucose infusion rates and bl ood glucose concentrations Glucose infusion rates were registered by th e Biostator every minute from 120 minutes before dosing until 360 minutes post-dose. The blood glucose (BG) measurements of the Biostator were recalibrated at regular intervals (range, 10 minutes) by the YSI 2300 STAT Plus Glucose Analyzer (YSI Life Sciences, Yellow Springs, OH), which employed the glucose oxidase method. For all subjects, blood glucose concentrations were maintained within the predefined range from the time of dosing until the end of the clamp procedure. Baseline Correction and Smoothing Methodology In order to subtract the contribution of the ongoing insu lin infusion, serum insulin concentrations were corrected for baseline insuli n levels, before being analyzed and used for pharmacokinetic model development. Insulin co ncentrations collected at 120, 90, 60 and 30 minutes before dosing were averaged, and subtr acted from all subseque ntly observed insulin concentrations. The uncorrected GIR values we re smoothed using a 10-minute running average. Insulin Data for the Pharmacokinetic/Pharmacodynamic Analysis Individual predicted insulin pharmacokinetic parameters from the pharmacokinetic model (Chapter 2) were used to simulate the insulin concentrations used in the analysis. The pharmacokinetics was described by a two compartm ent open model with one (inhaled) or two sequential (subcutaneous) first orde r absorption processes and firstorder elimination. Since the pharmacokinetic model was developed using baselin e corrected insulin concentrations and the PK/PD analysis targeted exploring the relationshi p between total insulin concentrations and total GIRs, the simulated insulin values were each adde d to the baseline used for correction at each

PAGE 70

70 visit. Corresponding insulin and GIR values (e very 5 minutes for the first hour and every 10 minutes thereafter) were combined for the analysis. Pharmacodynamic Model and Data Analysis The software package NONMEM (Version VI, Level 1.2, NONMEM Project Group, ICON Development Solutions, USA) was used fo r the population analysis using both the first order conditional estimation method (FOCE) w ith interaction and the subroutine ADVAN 6. NONMEM describes the observed e ffect data in terms of: A number of fixed effect parameters, which may include the mean values of the relevant base model parameters, or a number of parameters which relate the base model parameters to demographic and other covariates; Two types of random effect parameters: (a) 2: the variances of the interindividual variability ( ) within the population, and (b) 2: the variances of the residual intraindividual variability ( ) due to random fluctuations in an individuals parameter values, measurement error, model misspecifica tion, and all sources of error not accounted for by the other parameters. The population or average values of the parameters, the interindividual variances, 2, and the residual variance, 2, were estimated by NONMEM. The values are independent, identically distributed rando m errors with a mean of zero and a variance equal to 2. Subjectspecific parameters were calculated by NONMEM using the FOCE option. These parameters are empirical Bayesian estimates of the indivi duals true parameters based on the population parameters and the individuals observed concentrations. Pharmacodynamic model In a glucose clamp study, blood glucose rema ins constant and the GIR serves as a surrogate for insulin-mediated glucose disposal and is a natural choice for the pharmacodynamic endpoint. The presence of a hysteresis in insuli n effect necessitates th at the time element be removed from the model, so that the hysteresis loop can be effectively collapsed. A simple method of correcting nonsteady-state data to the equivalent of steady-state data (so that a

PAGE 71

71 concentration-response curve can be discerned) is the use of an indi rect model. In this analysis, the effect-compartment model initially proposed by Sheiner et al [70] was used. The effectcompartment model has also previously been suc cessfully applied to similar euglycemic clamp studies [59, 60]. A diagram of the model is presented in Figure 3-1. The pharmacodynamic portion of the model is described by the following equations: CeCpk dt dCee0 (3-1) e eCEC CE EGIR 50 max 0 (3-2) Equation 3-1 describes the relationship between the observed plasma insulin concentration and the concentration at the e ffect site. Effectively, ke0 is a rate constant which describes the delay in effect. The effect comp artment is assumed to receive a negligible mass from the central compartment, thereby not affecting the equati ons for the insulin pharmacokinetic model. Equation 3-2 establishes the relationship between the GIR and insulin concentration at the effect site. The pharmacodynamic structural model was parameterized in terms of E0, the baseline GIR value, Emax, the maximum glucose infusion rate, EC50, the effect site concentration eliciting 50% of the maximal response, the sigmoidicity factor, and Cp and Ce are the plasma and effect site insulin concentrations, respectively. Error model Interindividual variability was describe d by an exponential error model. The intraindividual residual variability of the dependent variable wa s estimated using an additive error model, as described by the following equation: ijij ijVDDV ~ (3-3)

PAGE 72

72 where DVij is the observed value of the jth dependent variable of individual i; ijVD ~ is the predicted jth dependent variab le of individual i; and ij is a random variable which represents the discrepancy between the observed and predicted jth dependent variable value. Model Fit Assessment Goodness-of-fit was determined by the objec tive function values (OFV) and visual inspection of scatter plots of pr edicted versus observed concentra tions as well as the weighted residuals. For nested models, hypothesis tests we re performed based on th e likelihood ratio test, in which the change in OFV approximates the 2 distribution. A more complicated model was preferred when the decrease in OFV was more than 3.84 (the critical value for the 2 distribution at p < 0.05 with 1 degree of freedom). Results Data from both routes of administration were analyzed simultaneously. Mean insulin and GIR-time profiles by dose group are presented in Figure 3-2. When exploring the potential PK/PD model, the relationship between insulin and GIR was first examined by visual inspection. A phase-plot of GIR vs. the plasma insulin conc entration, where data points are connected in chronological order, is shown in Figure 3-3 fo r the pooled data in this analysis. A counterclockwise hysteresis loop is observed indicating a disconnect be tween insulin concentrations observed centrally, and insulin action. The temp oral dissociation between insulin concentration and GIR clearly differs between TI and subcutaneous insulin, as well as insulin administered intravenously. Interestingly, the maximum effect is similar between all th ree treatments, but is observed at very different timepoints (40 and 18 0 minutes post-dose for TI and subcutaneous insulin, respectively, and at 20 minutes following iv administration) and is associated with greatly different insulin concentr ations. Furthermore, the dela y for subcutaneous insulin is

PAGE 73

73 smaller than that for both TI and insulin admini stered iv, since the hys teresis loop is clearly smaller for sc insulin. Pharmacodynamic model: Data from both routes of administration were modeled simultaneously. In Model A, the parameters ke0, EC50 and were estimated individually for each route of insulin administration, whereas in Model B, a single set of parameters was estimated for all three dosing routes. The baseline GIR (E0) was assumed to be comparable because the study conditions were similar and the subject pool was homogenous, and Emax was assumed to be the same for all treatments, as it is an inherent property of insulin. As a result, no treatment-specific parameter estimates were determined for Emax or E0 in Model A. Model A: Model A goodness of fit figures are presen ted in Figure 3-4. The individual predicted versus observed values distributed along the line of un ity, and no significant trends in the weighted residuals. The pharmacokinetic para meter estimates are summarized in Table 3-1. Population predicted GIR-time profiles and observe d GIR values are shown in Figure 3-5, and representative observed and indivi dual predicted GIR-time profiles are shown in Figure 3-6. The interindividual %CV for all parameter estimates ranged from 23 to 54%. The residual error was described by an additive model. Population estimates of ke0 were 0.7, 1.9 and 1.4 h-1 for iv, sc and TI, respectively, and were associated with distribution half-lives of 60, 20 and 30 minutes, respectively. These results indicate that the greatest delay between insuli n concentrations observed centrally and insulin effect occur following iv administration, which is in agreement with the relationship observed in Figure 3-3. Gamma estimates were 6.4, 2.5 and 3. 2 for iv, TI and sc, respectively, and in close agreement to values previously reported in healthy individuals [55]. The EC50 estimated by the

PAGE 74

74 model was associated with the concentration at the effect site (Ce) and was estimated to be 26.6, 41.2 and 32.9 U/mL. Model B: For Model B, the overall model fit was good, with individual predicted versus observed values distributed along the line of unity as well, and no apparent trends in the weighted residuals (Figure 3-7). Thus, the structural and error m odels appeared to adequately describe insulin pharmacodynamics and explain the variability in the data, even without formulations-specific parameter estimates. Ho wever, when compared to Model A, Model B goodness of fit plots show more scatter around the line of unity, and lower individual predicted GIR values overall. In the model, the population typical values were: Emax, 13.1 mg/kg/min; EC50, 32.8 U/mL; gamma, 3.0 and ke0 of 1.6 hr-1. The interindividual %CV for all parameter estimates ranged from 28 to 47%. The residual error was de scribed by an additive m odel, and the residual error had a greater magnitude in Model B (2 mg/kg/min) when compared to Model A (1.68 mg/kg/min), indicating that more of the observed variability was explained by the treatmentspecific parameters and associated interindividual variability in Model A. The pharmacokinetic parameter estimates ar e summarized in Table 3-2. Population predicted GIR-time profiles and observed GI R values are shown in Figure 3-8, and representative observed and indi vidual predicted GIR-time prof iles are shown in Figure 3-9. Discussion The effects of different insulin doses and trea tments are often evaluated by comparing the shape and area under the GIR vs. time curve. As a result, numerous attempts have been made to model the insulin-effect relations hip as derived from clamp study data to more fully understand the differential effects. GIR data have been assessed using a variety of physiologically-based and empirical models, including indirect [56, 57, 72] and effect compartment models [55, 59,

PAGE 75

75 60]. However, previous work focused on either one insulin formulation, insulins with similar pharmacokinetic properties, or explicitly estima ted different pharmacodynamics for insulins with different pharmacokinetics. Hence, the m odels developed provided limited use towards predictions of the effects of insulins with diss imilar insulin concentration profiles than those included in the model. The purpose of the current work was to develop a model that would be simple and non-specific, allowing for the prediction of insulin effect based on insulin pharmacokinetics alone and, thus, applicable to insulins with differing pharmacokinetic properties. To accomplish this, three insulins with differing pharmacokinetic properties were used in model development. As a first step, a model was developed (Mode l A), with individualized estimates for ke0, EC50 and for each formulation. This model was used to obtain reasonable initial estimates for Model B, in which one set of parameters was estimated for all three insulins. Since the predictive performance of Model A was expected to be better, a comparison could be made as to the fit of Model B using magnitude of residual variability and individual fits. Treatment-specific values for insulin Emax and E0 were not estimated in Model A, as they were assumed to be comparable between treatments. Model A parameter estimates varied most notably in the gamma parameter, with the highest value associated with intravenous insulin administration. The other parameter which varied considerably following intravenous administration was ke0, which was lowest in this treatment, and associated with the longest central to effect site equilibrati on half-life. This is most likely due to the relatively quickly changing insulin concentrations following this route of administration, but comparable delay in insulin eff ect for all three treatments, thus resulting in an apparent difference in ke0. However, due to the small numb er of subjects in each treatment

PAGE 76

76 group, in particular only 4 subjects receiving iv insulin, the treatment-specific parameter estimate differences may not be conclusi ve, and should be used with caution for predictive purposes. In Model B, pharmacodynamic parameter variab ility ranged from 27 to 52%, with the greatest variability observed on the sigmoidicity factor. This was an expected result, since Model A estimated different values for the three insulins, and others who estimated formulation-specific differences in insulin fo rmulations saw the greatest differences in pharmacokinetically different insulins in and EC50 estimates [60]. The population ke0 estimate of 1.7 was similar to the estimates for both TI and sc insulin in Model A. This is most likely attributable to the overwhelming majority of the data being obtained following non-iv administration. Nonetheless, the same and ke0 estimates were able to relate fairly well very different insulin concentration-time profiles to the GIR. Model B attempts to simultaneously model the concentration-GIR relationship of insulins with very different pharmacokinetic properties, without individualizing the pharmacodynamic parameters for each formulation. Given the lack of such a model in the literature, this would appear to be the first model constructed in this manner. Alth ough empirical in nature, it is the potential predictive ability of Model B that makes it uniquely valuable, and future work could focus on applying the model to external datasets to determine its accuracy in predicting insulin effect. Using this model, the pharmacokinetic insulin profile can be used to determine the pharmacodynamics, resulting in a straightforward simulation of the activity of insulins with varying pharmacokinetic properties. Although the simple nature of the model is a benefit, it is also a drawback in that the model slightly unde rpredicts the GIR around th e time of peak effect in some patients, especially in subjects who exhi bit a quick and high rise in insulin. In the data assessed here, the model fit is less predictive of maximum effect in certain subjects in the 100 U

PAGE 77

77 TI dose group, and to a greater extent in subj ects receiving insulin intravenously. Under these conditions, the fit is better in Model A, where the formulationspecific parameter estimates are better able to fit the data. However, some unde rprediction still exists, and the reason for this observation may be due to insuli n effect having two components: the stimulation of glucose disposal and the inhibition of glucose production [76]. Th e simple nature of the Emax model cannot account for two processes; it may describe insulin concentr ation-related glucose disposal, which is expected to follow a receptor-driven mo del, but it is unable to account for changes in insulin effects on hepatic glucos e production, which are more immediate and associated with a threshold insulin value [9]. In most cases, the model performs well and is reasonable, with its only weakness being the underprediction at specific conditions. An area of model improvement would be to attempt to account for this additional hump in the GIR cu rve, however, such a modification may detract from the simple nature of the model. Because the population used in this analysis was comprised of healthy volunteers with normal insulin sensitiv ity, the impact on liver glucose output would be expected to be higher than in a diabetic subject, wher e some of the sensitivity is lost. It is therefore possible that the extent of the underprediction may be lessened in the diabetic population. Additional research that would greatly help improve this model would be the comparison of its applicability in the diabetic population, whose insulin sensitivity may be compromised. Conclusions Insulin pharmacodynamics during a glucose clamp procedure in healthy subjects were found to be well described by a simple Emax model when a hypothetical effect compartment was used to collapse the hysteresis in insulin effect. The model was able to successfully describe the GIR response to three pharmacokineti cally diverse insulin formulations.

PAGE 78

78 Table 3-1 Model A insulin population pharmacodynamic parameter estimates Parameter Intravenous Estimate (%RSE) Subcutaneous Estimate (%RSE) Inhaled Estimate (%RSE) E0 (mg/kg/min) 2.7 (10.8) 2.7 (10.8) 2.7 (10.8) Emax (mg/kg/min) 14.6 (14.1) 14.6 (14.1) 14.6 (14.1) EC50 26.6 (18.1) 32.9 (12.1) 41.2 (16.5) Ke0 (h-1) 0.7 (13.9) 1.9 (6.5) 1.4 (12.4) gamma 6.4 (35.4) 3.2 (24.3) 2.5 (19.1) Interindividual and Residual Variability %CV %CV %CV E0 41.6 41.6 41.6 Emax 38.3 38.3 38.3 EC50 31.0 40.2 31.6 Ke0 23.2 17.4 50.3 g amma 54.1 42.5 39.9 IOV 41.6 41.6 41.6 1 1.68 1.68 1.68 Note: 1 (additive residual error) is expressed in mg/kg/min; The magnitude of interindividual and residual variability was expressed as % CV, approximated by the square root of the variance estimate. Table 3-2 Model B insulin population pharmacodynamic parameter estimates Pharmacodynamic Parameters Parameter Estimate (%RSE) E0 (mg/kg/min) 2.9 (12.4) Emax (mg/kg/min) 13.1 (10.0) EC50 32.8 (11.9) Ke0 (h-1) 1.6 (7.7) gamma 3.0 (17.6) Interindividual and Residual Variability %CV E0 41.6 Emax 33.8 EC50 46.7 Ke0 28.1 g amma 35.2 1 2.0 Note: 1 (additive residual error) is expressed in mg/kg/min; The magnitude of interindividual and residual variability was expressed as CV%, approximated by the square root of the variance estimate.

PAGE 79

79 GIR effectInhalation Subcutaneous kcpkpck1ecentralperipheral kaTIkasc2 depot1transit depot2Intravenous kasc1Emax ke0 GIR effectInhalation Subcutaneous kcpkpck1ecentralperipheral kaTIkasc2 depot1transit depot2Intravenous kasc1Emax ke0GIR effectInhalation Subcutaneous kcpkpck1ecentralperipheral kaTIkasc2 depot1transit depot2Intravenous kasc1Emax ke0 Figure 3-1 Pharmacokinetic/pharmacodynamic model diagram 050100150200250300350Time (minutes) 0 200 400 600 800Mean Insulin Concentration (uU/mL) 5 IU iv 10 IU sc 25 IU TI 50 IU TI 100 IU TIA 050100150200250300350Time (minutes) 0 5 10 15 20Mean GIR (mg/kg/min) 5 IU iv 10 IU sc 25 IU TI 50 IU TI 100 IU TIB Figure 3-2 Mean insulin and GI R-time profiles by dose group. A) Mean individual predicted insulin concentrations by dose and B) Mean observed GIR by dose group

PAGE 80

80 100300500700900Insulin Concentration (uU/mL) 0 5 10 15 20GIR (mg/kg/min) 5 U iv 10 U sc 25 U TI 50 U TI 100 U TI Figure 3-3 Hysteresis in the insulin-GIR relationship 0510152025Individual Predict ed GIR (mg/kg/min) 0 10 20 30Obeserved GIR (mg/kg/min) A 2468101214Predicted GIR (mg/kg/min) -6 -4 -2 0 2 4 6Weighted Residuals B Figure 3-4 Goodness of fit plots for Model A. A) Individual predicted GI R versus observed GIR and B) GIR values versus weighted residuals

PAGE 81

81 0246 0246 0246 0246 0246Time (hours) 0 10 20 30GIR (mg/kg/min) DOSE: 5 DOSE: 10 DOSE: 25 DOSE: 50 DOSE: 100 Figure 3-5 Model A: Population predicte d and observed GIR values by dose group

PAGE 82

82 0123456Time (hours) 1 3 5 7GIR (mg/kg/min) 10 IU sc 25 U TI 50 U TI 100 U TIA 0123456Time (hours) 1 6 11 16GIR (mg/kg/min) 10 IU sc 25 U TI 50 IU TI 100 IU TIB 0123456Time (hours) 2 7 12 17 22 27GIR (mg/kg/min) 10 IU sc 25 U TI 50 U TI 100 U TIC 0123456Time (hours) 1 6 11 16GIR (mg/kg/min) 5 IU iv 10 IU sc 100 U TID Figure 3-6 Model A: Model predicted and observed GIR values in four subjects 0 5 101520Individual Predict ed GIR (mg/kg/min) 0 10 20 30Obeserved GIR (mg/kg/min) A 2468101214Predicted GIR (mg/kg/min) -5 -3 -1 1 3 5Weighted Residuals B Figure 3-7 Goodness of fit plots for Model B. A) Individual predicted GI R versus observed GIR and B) GIR values versus weighted residuals

PAGE 83

83 0246 0246 0246 0246 0246Time (hours) 0 10 20 30GIR (mg/kg/min) DOSE: 5 DOSE: 10 DOSE: 25 DOSE: 50 DOSE: 100 Figure 3-8 Model B: Population predicted (red ) and observed (gray) GIR values by dose group

PAGE 84

84 0123456Time (hours) 1 3 5 7GIR (mg/kg/min) 10 IU sc 25 U TI 50 U TI 100 U TIA 0123456Time (hours) 1 6 11 16GIR (mg/kg/min) 10 IU sc 25 U TI 50 IU TI 100 IU TIB. 0123456Time (hours) 1 6 11 16 21 26GIR (mg/kg/min) 10 IU sc 25 U TI 50 U TI 100 U TIC 0123456Time (hours) 1 6 11 16GIR (mg/kg/min) 5 IU iv 10 IU sc 100 U TID Figure 3-9 Model B: Predicted and obs erved GIR values in four subjects

PAGE 85

85 CHAPTER 4 POPULATION PHARMACOKINETIC MODEL FOR SUBCUT ANEOUSLY ADMINISTERED REGULAR HUMAN INSU LIN, INSULIN LISPRO AND INHALED INSULIN IN HEALTHY AND DIABETIC SUBJECTS Background Until recently, the majority of therapeutic insu lins administered via non-intravenous routes have been described by a one compartment pharm acokinetic model, as both the distribution phase and second compartment have been obscu red by slow absorption characteristics. Technosphere Insulin (TI) is a novel inhaled regular human insulin (RHI) whose unique delivery mechanism results in rapid absorpti on and rapid clearance, making it possible to distinguish the second compartment in its pharm acokinetic profile [78]. The inclusion of data following TI dosing has made it possi ble to attribute the differences in the shape of the insulin curve, following various routes of administration, to differences in absorption. As such, insulin disposition following subcutaneous, pulmonary and intravenous administration has been described by a two compartment pharmacokinetic m odel combining data from all three routes of administration (Chapter 2). Having established that insulin pharmacokine tics can be modeled using a two compartment model, and that subcutaneously administered a nd inhaled insulin can be modeled together, the aim of this analysis was to develop a populat ion pharmacokinetic model for RHI and insulin lispro administered via the subcutaneous route, and TI administer ed via inhalation. Population pharmacokinetic modeling has become a staple in all areas of pharmacokinetic research, and can be particularly useful in finding trends and relationships between covariates and pharmacokinetic parameters, as well as identifyi ng and quantifying sources of va riability [92]. Although insulin has been used as therapy for more than 80 year s, no larger scale population analysis has been performed combining healthy subjects and subj ects with diabetes and a large range of

PAGE 86

86 demographic covariates. In this analysis, data from healthy, type 1 and type 2 diabetic subjects was combined, and the model was used to determine the relative bioavailability of the three formulations and to identify any significant covari ates associated with in sulin pharmacokinetics. Materials and Methods Study Population and Study Design A population pharmacokinetic model of insuli n was developed using data from four studies. Three studies were pros pective, open-label, randomized, crossover, euglycemic glucose clamp studies in non-smoking male and female healthy volunteers and subjects with type 2 diabetes. One study was a meal challenge in subject s with type 1 diabetes. Prior to entry into any study, all subjects provided written inform ed consent and were then given a physical examination, pulmonary function tests, electr ocardiography and laborat ory tests, including urinalysis and screening for drugs of abuse. All studies were approved by the local ethics committee. The data from all four studies were combined for modeling purposes, with a total of 103 subjects included in the mode l and 213 pharmacokinetic profiles. All of the studies were conduc ted in a crossover fashion, where each subject received a single dose of each treatment studied. In all stud ies, TI was administered using the MedTone dry powder inhaler, with the exception of Study 1, wh ere TI was administered using a commercially available inhaler (Model M, Bo ehringer Ingelheim, Ingelheim, Germany). Samples for the determination of insulin, lispro a nd c-peptide concentrations were drawn from the cubital vein in the contralateral arm to that used for the in sulin infusion and glucose administration, and analyzed using a radioimmunoassay (RIA). A brie f description of the study design and methods follows.

PAGE 87

87 Healthy volunteers Study 1: This was a three-way crossover euglycem ic glucose-clamp study in five subjects to compare the pharmacokinetics and pharmacodynami cs of single doses of 100 U of inhaled TI, 10 U RHI administered subcutaneously and 5 U RHI (Actrapid, Novo Nordisk A/S, Bagsvaerd, Denmark) administered intravenously. Study 2: This was a four-way crossover euglycem ic glucose-clamp study in 12 subjects to compare the pharmacokinetics and pharmacodynamics of three different si ngle doses of inhaled TI with a single dose of 10 U RHI (Actrapid, Novo Nordisk A/S, Bagsvaerd, Denmark). RHI insulin was administered by subcutaneous injec tion in the abdominal region. Blood glucose was kept constant at 90 mg/dL throughout the glucose clamp by a variable infusion of a dextrose solution. Samples for the determination of insu lin concentration were drawn at 120, 90, 60 and 30 minutes before dosing and 0, 1, 3, 7, 12, 20, 30, 45, 60, 90,120, 180, 240, 300 and 360 minutes post-dose, and C-peptide concentrations were collected at 120 and 60 minutes pre-dose and 0, 30, 60, 180, and 300 minutes post-dose. The subjects were kept fasted until the end of the study. The treatment periods were separated by a washout period of 3 and 28 days. Type 2 diabetic subjects Study 3: This was a prospective, randomized, open label, single site study in 12 nonsmoking, male and female subjects with a di agnosis of type 2 diabetes mellitus for 12 months, a stable antidiabetic regimen with insulin for the previous 3 months and HbA1c 8.5%, to compare the pharmacokinetics and pharmacodynamics of two doses of TI (60 and 90 U) with a single dose of 10 U lispro (Humalog Eli Lilly & Co.). Six subjec ts received 60 U TI and the other six subjects received 90 U TI. All twelve subjects received lispro insulin administered by subcutaneous injection in the abdominal regi on. Following an overnight fast, the subjects received, on each of the treatmen t days, a constant rate intrave nous infusion of either RHI or

PAGE 88

88 insulin lispro (opposite of the t ype of insulin to be used in dosing) during a 4-6-hour run-in period in order to suppress endogenous insulin produc tion. This infusion was continued until the end of the clamp procedure. Blood glucose was kept constant at 110 () mg/dL throughout the last 60 minutes of the run-in period, and for 480 minutes post-dosing via the glucose clamp using the Biostator. Samples for insulin and insulin lispro concentration de termination were drawn at 310, 20 and 5 minutes before dosing and at time 0, 5, 10, 15, 20, 30, 45, 60, 75, 90, 105, 120, 150, 180, 210, 240, 300, 360, 420 and 480 minutes post-do se. The patients were kept fasted until the end of each treatment visit. The treatment periods were separated by a washout period of 7 to 21 days. The studies in healthy a nd type 2 diabetic were performe d at the same clinical site. Type 1 diabetic subjects Study 4: This was an open label, prospectiv e, randomized study in 75 non-smoking male and female subjects with type 1 diabetes (H bA1c<11%) and normal pulmonary function, to compare the pharmacokinetics of a 30 U cartridge delivered using two different models of the MedTone Inhaler. The last dose of long-acting insulins or intermediate-acting insulins was taken in the morning on the day before TI administration. Subsequently, subjects were instructed to manage their BG with intermittent injecti ons of RAA. Subjects utilizing continuous subcutaneous insulin infusion pumps were inst ructed to replace their usual insulin with NovoRapid (Novo Nordisk, Denmark) the day before treatment. All subjects were administered TI directly before the start of a meal (12 oz. Boost Plus). Samples for insulin and insulin lispro concentration determination were drawn at 30, 20, 10 minutes prior to dosing and 0, 3, 6, 9, 12, 15, 20, 25, 30, 45, 60, 75, 90, 105, 120, 150, 180, and 240 minutes after drug administration. The treatment periods were separated by a washout period of 1 day.

PAGE 89

89 Baseline Correction Methodology In order to subtract the cont ribution of the ongoing insulin infusion (studies 1 and 2) and detectable concentrations of insulin still washi ng out from the system (studies 3 and 4), serum insulin concentrations were corrected for baseline insulin levels before being analyzed and used for model development. Any baseline-corrected value below zero was set equal to zero. Due to the differing study conditions, different baselinecorrection strategies had to be used: Studies 1 and 2: Insulin concentrations collected at 120, 90, 60 and 30 minutes before dosing were averaged, and subtracted from all subs equently observed insulin concentrations. Study 3: Baseline correction was applied to account for any endogenous insulin that was not fully suppressed with the baseline infusion. Fo r the TI group, all insuli n concentrations were corrected by subtracting the m ean of the values in the later time points (300 minutes) from each individual time point for each subject. At these time points, most subjects insulin concentration-time profiles were observed to be reasonably flat and uninfluenced by exogenous contribution. Study 4: In accordance with the study analysis plan, the mean concentration of the -30 through 0 time points was used as baseline. Of the 150 pr ofiles, 13 profiles on day 1 had a mean of > 3.6 U/mL as a baseline, and on day 2, only 7 subjects had a baseline > 3.6 U/mL. Data Excluded from the Pharmacokinetic Analysis For studies 1 and 2, each serum insulin a nd C-peptide concentra tion-time profile was examined visually together. Due to the lower do ses administered in Study 2, a higher incidence of elevated C-peptide concen trations was observed in the latter time points following TI treatment, particularly in the lower dose groups All insulin concentr ations after 180 minutes post-dose in the 25 U dose group, 240 minutes post-dose in the 50 U dose group, and 300 minutes in the 100 U dose group were excluded, since endogenous insulin appeared to be

PAGE 90

90 contributing to the curve in the la tter part of the study, with a to tal of 51 non-BLQ concentrations excluded from the analysis. A further 21 con centrations were excluded for pharmacokinetic implausibility, such as concentrations below the quantifiable limit in close proximity to the Cmax, or unexpectedly high concentratio ns at the end of the profile. One subject from study 2 was excluded from the entire analysis, and another a subject was excluded from analysis in the 10 U sc treatment group, since three consecutive BLQ values were observed in the pharmacokinetic profile following sc RHI treatment, suggesting a possible analytical error. A total of 15 profiles were excluded due to lack of test treatmen t exposure or incomplete dosing and/or sampling information. Noncompartmental Pharmacokinetic Analysis and Absolute Bioavailability Insulin pharmacokinetics were analyzed using noncompartmental methodology using baseline-corrected insulin concentrations. The following PK parameters were derived using WinNonlin v 5.2 (Pharsight Corporation, Mountain View, CA): observed peak insulin concentration (Cmax), time to peak insulin (tmax), and total insulin exposure as measured by the area under the insulin concentration-time curve from time 0 until the last timepoint with non-zero insulin concentrations (AUC0-last) as calculated by the linear-tra pezoidal method. The terminal elimination half-life (t1/2) was calculated as ln(2)/ z in accordance with pharmacokinetic theory [80], where z is the terminal elimination rate cons tant estimated from log-linear regression analysis of the terminal elimination phase of the concentration-time profile, and AUC0was calculated as AUC0-last + Clast/ z, where Clast is the last observed insulin concentration. Noncompartmental parameter values were used to determine starting estimates for pharmacokinetic modeling, and were compared to parameter values estimated by the final model as part of the model fit assessment.

PAGE 91

91 Relative bioavailability (%F) was determin ed using the mean dose-normalized insulin AUC0-last values, expressed as the group average dose-normalized AUC0-last/average dosenormalized AUC0-last following lispro administration. Statistical Analysis Dose-normalized insulin AUC0was used in order to inve stigate possible TI exposure differences between studies. Statistical di fferences were concluded if the average log transformed AUC0values were found statistically differe nt by the one-way ANOVA technique, which was performed with the nu ll hypothesis that the study averag e values are the same, at the =0.05 level. The alternative hypothesis was th at some of the group means differ from each other. In the event that statistical differenc es were found by ANOVA, Tukeys pairwise comparison method was used to determine which pair s of studies differ. The family error rate was set to 5%. Results were presented as a set of confidence intervals for the difference between pairs of means, from which the following conclusi ons were drawn: 1) if an interval does not contain zero, there is a statis tically significant difference between the corresponding means and 2) if the interval does contai n zero, the difference between the means is not statistically significant. It was necessary to test the variances of the dose-normalized, log-transformed AUC0values in the four studies for homogeneity in order to accept the results of the ANOVA analysis, since homogeneity of variances is one of the ANOVA assumptions. Levenes test was performed under the null hypothesis that the vari ances are the same and an alternative hypothesis that the at least one of the vari ances was different from the others The data were tested at the =0.05 level. The results of the statistical tests were obtained using MINITAB (v.15.1.30.0).

PAGE 92

92 Population Pharmacokinetic Analysis The software package NONMEM (Version VI, Level 1.2, NONMEM Project Group, ICON Development Solutions, USA) was used fo r the population analysis. NONMEM describes the observed concentration-time data in terms of: A number of fixed effect parameters, which may include the mean values of the relevant base pharmacokinetic model parameters, or a number of parameters which relate the base model parameters to dem ographic and other covariates; Two types of random effect parameters: (a) 2: the variances of the interindividual variability ( ) within the population, and (b) 2: the variances of the residual intraindividual variability ( ) due to random fluctuations in an individuals parameter values, measurement error, model misspecifica tion, and all sources of error not accounted for by the other parameters. The population or average values of the parameters, the interindividual variances, 2, and the residual variance, 2, were estimated by NONMEM. The values are independent, identically distributed rando m errors with mean of zer o and a variance equal to 2. Subjectspecific parameters were calculated by NONMEM using the POSTHOC (FO method) option. These parameters are empirical Bayesian estimates of the individuals true parameters based on the population parameters and the indi viduals observed concentrations. Pharmacokinetic base model Data from all three routes of administrati on were modeled simultaneously. The PK was described by a two compartment model with one (inhaled) or two sequent ial (subcutaneous) first order absorption processes and firs t-order elimination. A diagram of the model is presented in Figure 4-1. The model was described by the fo llowing differential equations: 11 1Aka dt dA (4-1) 2211 2AkaAka dt dA (4-2)

PAGE 93

93 31 3Aka dt dATI (4-3) 4 445554223 4AkAkAkAkaAka dt dATI (4-4) 554445 5AkAk dt dA (4-5) The pharmacokinetic structural model was parameterized in terms of clearance (CL), volume of distribution in the central compartment (Vc), intercompartmental clearance (Q), the volume of distribution in th e peripheral compartment (Vp), the first order absorption rate constant for TI (kaTI), the two first order absorption rate constants associat ed with the sequential absorption for subcutaneously administered insulin (kasc1and kasc2, depot and a transit compartment), sequential absorption for subcutaneously administered lispro (kaLI1and kaLI2), and the absolute bioavailability for su bcutaneous insulin, lispro and TI (Fsc, FLI, FTI). Error model Fixed effects parameters were used to describe the typical population estimates, and an exponential random effect model was used to descri be interindividual vari ability for each model parameter: ii (4-6) where i is the estimated parameter value for the ith individual, is the fixed effect typical parameter value in the population, and i are individual-specific random effects for the ith individual symmetrica lly distributed with zero mean and variance A combined proportional and additive error model was used to model the residual unexplained variability, as descri bed by the following equation: ij ij ij ijpCCp21)1( ~ (4-7)

PAGE 94

94 where Cpij is the observed value of the jth plasma concentration of individual i; ijpC ~ is the predicted jth plasma concentration of individual i; and ij is a random variable which represents the discrepancy between the observed and predicted jth concentration. Considerable interoccasion vari ability (IOV) was observed wi thin the TI groups, and was attributed to natural variability in inhalation on different occasions, resul ting in differences in relative bioavailability at diffe rent visits, as described by: 3 2 13 2 1OCC OCC OCC FTI TIOCC OCC OCC FTIF (4-8) where OCC1, OCC2 and OCC3 are set to 1 at the corresponding occasion and 0 otherwise. Reparameterization and i ndividual predicted values Individual-specific values of each pharmacokinetic parameter were obtained by Bayesian analysis with the final model. 4VkCL (4-9) 445VkQ (4-10) The following pharmacokinetic parameters coul d subsequently be calculated for each individual according to the following equations based on compartment modeling theory. kkkkkkkk54 5445 54454 ) ( 2 1 (4-11) kkkkkkkk54 5445 54454 ) ( 2 1 (4-12) 693.0 lifehalf (4-13) 693.0 lifehalf (4-14)

PAGE 95

95 Model Fit Assessment Goodness-of-fit was determined by the objec tive function values (OFV) and visual inspection of scatter plots of predicted versus observed concen trations as well the weighted residuals. For nested models, hypothesis tests we re performed based on th e likelihood ratio test, in which the change in OFV approximates the 2 distribution. A more complicated model was preferred when the decrease in OFV was more than 3.84 (the critical value for the 2 distribution at p < 0.05 with 1 degree of freedom). Covariate Analysis Following the determination of the base popul ation model, potential covariates were examined to determine whether they improved the overall fit and reduced variability in the model. These covariates included age, body we ight, BMI, disease status (healthy, type of diabetes), gender, study effect a nd insulin type. Furthermore, the pulmonary function values: forced expiratory volume in one second (FEV1) and percent of predicted FEV1 (National Health and Nutrition Examination Survey [NHANES III] ) were assessed for possible effect on TI parameters. Covariates were initially evaluated fo r possible relationships with the model estimated pharmacokinetic parameters using the generalized additive model (GAM) procedure in Xpose [82]. Covariates a ssociated with a reduction in the Akaike Information Criterion (AIC) were evaluated using NONMEM in a stepwise manner. Each covariate added to the base model, and the resulting univariate model was then co mpared to the reduced model for significant improvement in fit. Covariates were included in the model if a 3.84 poi nt drop in the OFV was observed (p<0.05). Backward elimination wa s then performed where each covariate was independently removed from the model to confirm its relevance. An incr ease in the OFV of 6.7 (p<0.01) was necessary to confirm th at the covariate was significant.

PAGE 96

96 Model Validation The precision of the population pharmacokinetic model parameter estimates was assessed by establishing 95% confidence in tervals (CI) using a nonparametr ic bootstrap analysis (Wings for NONMEM; Version 409d) [93]. Subjects were selected repeatedly and at random, with replacement, from the dataset to create a new dataset with the same number but different combination of subjects. The bootstrap resampli ng was repeated 200 fold and only runs that converged successfully were used for the an alysis. The 95% CI was calculated for each pharmacokinetic parameter by using the 2.5t h and 97.5th percentile from 200 bootstrap estimates. The final population pharmacokinetic parameter estimates were compared with the bootstrap estimates. Results Patient population A total of 3,227 insulin concentr ations from 103 subjects and 213 profiles were included in the analysis. The ADVAN6 subroutine and firstorder estimation method were used. Subject demographic data is su mmarized in Table 4-1. Noncompartmental Analysis Mean insulin concentration-time profiles are presented in Figure 4-2. Noncompartmental pharmacokinetic parameter estimates are pres ented in Table 4-2. Based on mean AUC0-last, the relative bioavailability of TI was approximate ly 12% and 20% compared to lispro and RHI, respectively, and approximately 60% when RHI wa s compared to lispro. However, the latter estimate is most likely artificially low, sin ce RHI was sampled only until 6 hours, and lispro was sampled until 8 hours post-dose. Terminal elimination half-life and AUC0could not be calculated following subcutaneous dos ing of either RHI or lispro due to the prolonged absorption from the tissue, and the flip-flop kinetics associat ed with this route of insulin administration.

PAGE 97

97 Exposure in study 4 (30 U TI dose) appeared to be approxima tely 30% to 40% lower when compared to other studies, based on dose-normalized AUC0. The difference was found to be statistically significant when the dose-normalized insulin AUC0was compared using a one-way ANOVA. Statistical Analysis ANOVA test for equality of means The aim of this study was to detect possible statistical differences in average insulin exposure in the four studies in cluded in the analysis. A one -way ANOVA was performed with the null hypothesis that the gr oup average dose-normalized expos ure is the same. The null hypothesis was rejected at the =0.05 level, with a p value of < 0.001 and associated test statistic F= 9.55, in favor of the altern ative hypothesis, that some of the group means are different. Differences in means between the four studies In light of the ANOVA outcome that statistica l differences exist between the studies, the four studies were compared usi ng Tukeys pairwise comparison me thod, and a family error rate of 5%. Results are presented in Table 4-3 as a set of confidence intervals for the difference between pairs of means. From the results pr esented in Table 4-3, a difference is observed between insulin exposure between studies 1 and 4, 2 and 4 as well as 3 and 4, indicating that exposure was different in study 4 (30 U TI dose) when compared to the other studies, as suggested by the results of th e noncompartmental parameter values presented in Table 4-2. The analysis was repeated without data from study 4, to test for differences between the mean exposure in the other studies. A one-w ay ANOVA was performed on the reduced dataset, with the null hypothesis that th e group average dose-normalized exposure is the same. The null hypothesis could not be rejected at the =0.05 level, with a p value of 0.271, and the mean exposure between studies 1, 2 and 3 were conc luded to be no different from each other.

PAGE 98

98 Levenes test was performed to test the varian ces of the four studies for homogeneity. The null hypothesis of equal variances could not be rejected at the =0.05 level, with a p value of 0.175. Therefore, the variances met the homogeneity assumption. Population Pharmacokinetic Analysis Base model A two-compartment open model with one (inhaled) or two sequential (subcutaneous) first order absorption processes and fi rst-order elimination described the insulin concentration data well. The lower exposure in study 4 was incorporat ed in the base model. The population typical values were: apparent clearance, 53.3 L/hr; ap parent volume of dist ribution in the central compartment, 4.6 L; apparent volume of distri bution in the peripheral compartment, 37 L; and relative bioavailability of subcutaneous RHI a nd TI estimated at 75% and 14%, respectively, with an approximate 50% lower exposure in study 4. The base model pharmacokinetic parameter estimates are summarized in Table 4-6. The overall model fit was good, with indivi dual predicted versus observed values distributed along the line of unity, and no significant trends in the weighted residuals (Figure 4-3 and Figure 4-4). Thus, the structural and error m odels appeared to adequa tely describe insulin pharmacokinetics and explain the variability in the data. Covariate analysis A GAM analysis identified the following potential covariates: body weight on insulin clearance, age on the volumes of distribution and TI absorption rate, BMI on subcutaneous absorption, FEV1, body weight and gender on TI absorpti on rate, and body weight and age as covariates on TI bioavailability. Visual inspect ion of covariate vs. parameter scatter plots were also used for covariate identification. All pot ential covariates were included in a univariate analysis, where each covariate was added to th e base model, and asse ssed in terms of OFV

PAGE 99

99 reduction as well as overall reduction in variability. A summary of the univariate analysis is presented in Table 4-4. The results of the univariate analysis showed th at the addition of age as a covariate to the pulmonary absorption rate significantly improve d the model (p<0.01). However, once age had been accounted for, the further addition of gender and weight did not significantly contribute to the model as the addition of gender resulted in an increased %RSE associated with the TI ka estimate as well as an increase in the interindivi dual variability, and the addition of weight did not result in a further decrease in OFV. Hence, neither gender nor weight was not included in further model building. Age effect on TI abso rption rate was described by a power model. ageAge kakaTI TI 36/ (4-15) In agreement with results reported earlier (C hapter 2), BMI was found to be negatively correlated with subcutaneous absorption rate, and was described by a linear model. BMI kasc scBMI ka )24( (4-16) Both age and BMI were centered at the popul ation mean value. The addition of age decreased the OFV by 98.9 points (Table 4-4) and the addition of BMI decreased the OFV by 22.2 points (Table 4-4). Following the inclusion of age on the central volu me of distribution, further addition of age on the peripheral volume did not contribute to the model, resulting in no further decrease in OFV. The addition of age decreased the OFV by 36.5 (Table 4-4). Age was modeled using a linear model, where age was cen tered at the population mean. age VcCAge V )36( (4-17) The remaining significant covariates were each tested for inclusion in the final model using stepwise forward addition. Following the inclusio n of age on TI ka, age on the central volume of

PAGE 100

100 distribution and BMI on subcutaneous absorpti on rate, weight was no longer a significant covariate associated with clea rance (OFV decrease of 0.235). The significance of each of the remaining covariates was confirmed by backwa rds elimination from the full model and is summarized in Table 4-5. Each of the three covariates was found to contribute to the model (p<0.01) and all three were reta ined in the final model. Final model The inclusion of the covariates to the base mode l resulted in a better fit, with an improved distribution of individual pred icted versus observed values along the line of unity, and no significant trends in the weight ed residuals (Figure 4-5 and Fi gure 4-6). The bioavailability relative to lispro for subcutaneous insulin and TI was 72% and 14%. Interoccasion variability following TI administration was approximately 31%. The interindividual %CV for all pharmacokinetic parameter estimates ranged fr om 10.7 to 82.5%. Th e residual error was described by a combined proportional (20.2 %C V) and additive (0.87 U/mL) model. The pharmacokinetic parameter estimates ar e summarized in Table 4-6. The final model was used to simulate insulin concentration-time profiles following administration of 60 U of TI to subjects at age 20, 35 and 65, as well as subjects with BMI values of 20, 23 and 26 kg/m2, receiving 10 U RHI. Due to the small range of BMI in the RHI treated group, and the possible non-linear relationship that may exis t beyond the range studied, no simulations were preformed for BM I values not used in the model development. The magnitude of the covariate effects is depicted in Figure 4-7. Model Validation A nonparametric bootstrap was used to evaluate the stability and pr ecision of the final model parameters. Due to the long run times a ssociated with the final model, the bootstrap analysis was limited to 200 runs. One hundred nine ty-two runs that conver ged successfully were

PAGE 101

101 used for further analysis. The final parameter estim ates are presented in Table 4-6. Overall, the median bootstrap structural and covariate parameter values matched the population typical values well. The confidence interval around the age covariate associated with the volume of distribution was very close to ze ro at the lower limit, suggesti ng that the relationship may be data-set specific, and that it should be used with cauti on for predictive purposes. Discussion In this population pharmacokinetic analysis, da ta from subcutaneous administration of lispro and RHI was combined with data following inhalation (TI). TI ad ministration results in rapid absorption and a rapid systemic clearance, with a cl early distinguishable second compartment in its pharmacokinetic profile. The inclusion of data following TI dosing has made it possible to attribute the differences in the sh ape of the insulin curve following the different routes of administration to differences in absorp tion [94]. In agreement with previously reported results, insulin pharmacokinetics were found to be consistent with a two-compartment model [94]. In this analysis, data from subjects with type 1 and type 2 diabetes treated with TI and lispro were combined with data from healthy individuals dosed with TI and subcutaneously administered insulin. The inclusion of lispro in the model was appropriate, since lispro has been shown to exhibit almost identical pharmacokine tics as RHI when administered intravenously [37], and its more rapid apparent clearance can be attributed to its more rapid absorption rate. The inclusion of lispro data in the model allowed for an estimate of the absorption rate differences between TI, subcutaneously administered insulin as well as lispro. As expected, TI was characterized by the quickest absorption rate, followed by lispro and RHI. Both subcutaneously administered insulins were adeq uately described by two sequential first order absorption rate constants, as described by Puckett et al [81] in a model based on the physiology

PAGE 102

102 of subcutaneous insulin administration, where the depot compartment represents the subcutaneous tissue, and the second compartment repr esents the interstitial space. This model of insulin absorption was applied to data from h ealthy volunteers, as described in Chapter 2. Absorption rate estimates for both TI and subc utaneous RHI were almost identical to the previously developed model, as expected, sin ce the current population model included the data from healthy volunteers used in the earlier analys is (Chapter 2). As also expected, the lispro absorption rate was estimated to be faster than following RHI, indicating that the rapid dissociation of lispro into the re adily absorbable monomeric form increases its absorption rate from the subcutaneous tissue. An ideal design to better explore this difference between the two subcutaneous forms of insulin would be a cro ssover study, which would greatly diminish the interindividual variability that is inherent with subcutaneous dosing and would allow for a better comparison between the two treatments. The inclusion of data from an inhaled produc t in the model made it necessary to account for the considerable interoccasion variability associated with pulmonary drug administration. Inhalation differences from day to day can resu lt in differences in the bioavailability of treatments administered through the lung, and although this variability may be reduced with repeated dosing, treatment-nave or inexperienced subjects ma y exhibit significant interoccasion differences in exposure. In this analysis, subject s in one of the studies were administered three different doses of TI on three different o ccasions and in another study, subjects were administered TI on two different occasions. The interoccasion variability was successfully modeled to reflect inhalation differences that re sulted in a difference in insulin bioavailability, with no difference in other pharmacokinetic pa rameters. For this group of subjects, the

PAGE 103

103 interoccasion variability was estim ated by the model to be approximately 29%, which is similar to values reported previously (Chapter 2). In this analysis, relative bioavailability of TI and subcutaneous RHI was estimated by fixing lispro bioavailability to 1. The absolute bi oavailability of lispro ha s been reported to be between 55-77%, and has been shown to be significantly greater than that of RHI [37]. In the model presented here, subcutaneous RHI and TI ar e estimated to have a relative bioavailability of 72 and 14%, respectively. Assuming the lispro ab solute bioavailability to be 77%, as reported in literature, the parameter estimates presented here match well with those estimated in an analysis reported earlier (Chapter 2), with an ab solute bioavailability adjusted systemic clearance of 40.3 L/hr (previously reported 43.4 L/hr), a central volume of distri bution of 3.7 L (reported 5.0 L) and a peripheral volume of distri bution of 28 L (reported 30 L). The relative bioavailability of TI to RHI was 19.4%, in clos e agreement with previous results where it was found to be 21% [94]. The final population model included age as a significant covariate positively correlated with insulin volume of distributi on in the central compartment, a nd negatively correlated with the TI absorption rate. Pulmonary absorption rate was found to decrease w ith age, and appeared to be independent of any age-associated decreases in pulmonary function, as FEV1 did not contribute to the model significantly when tested in the univariate analysis. Visual inspection of the relationship of age and TI absorption rate revealed asymptotic tendencies with increasing age. The power model appeared to fit the data we ll, indicating that the extent of the effect would be self-limiting. This finding c ould be beneficial to older pati ents, who often times experience a slower gut transit time when comp ared to younger patients. High a nd fast insulin peaks, such as those seen with TI, might cause a mismatch in insulin peak and glucose appearance from the

PAGE 104

104 meal, causing early hypoglycemic events in older patients. The slightly longer absorption and lower overall insulin peak will result in a de creased likelihood of such an event in this population. Unlike differences in age, pulmonary function differences were not found to affect either the pulmonary absorption rate or bioavailability following TI administration. The fact that the analysis did not identify either FEV1 or percent of predicted FEV1 as covariates was not an unexpected result, since the inclusion criteria into the studies in this analysis required pulmonary function in the normal range. An interesting ad dition to the model woul d be the inclusion of subjects with moderately or even severely impaired pulmonary function, to further investigate any potential relationship. Increases in BMI were found to decrease absorption when insulin was administered subcutaneously, as reported previously by others [90]. This observation may be attributed to the increased thickness of the subc utaneous tissue in subjects with higher BMI, which slows absorption from the depot compartment. The small number of subjects, narrow BMI range and homogenous nature of the healt hy population makes it difficult to extrapolate these results, however, it is expected that a greater impact wo uld be observed on patients taking subcutaneous insulin who have higher BMI, as is often th e case in subjects with type 2 diabetes. In this analysis, the inclusion of the covari ates associated with the disease state was difficult and complicated, since data from each population was derived from a different study. Furthermore, different baseline correction methods, which may influence some parameter estimates, were used in each study. Addition of more data to the model may help overcome this issue and would strengthen the validity of the analysis.

PAGE 105

105 Overall, the median bootstrap structural and covariate pa rameter values matched the population typical values well, addi ng a stronger level of confidence in the results. However, the confidence interval around the age covariate associ ated with the volume of distribution was very close to zero at the lower limit, s uggesting that the relatio nship may be data-set specific, and that it should be used with caution for predictive purposes. The bootstrap results of the interindividual and residual error estimates suggest that the accuracy of final model estimates of variability may be limited, and that a larger set of data may be necessary for an accurate assessment of the variability in a model whic h describes the pharmacokinetics of a highly variable substance. Furthermore, the bulk of th e data in this analysis was obtained following TI administration, making it difficu lt to truly gage the variability associated with subcutaneous dosing due to the relatively small number of patients dosed via this route. This is a shortcoming of the model developed, and assessment of variab ilities is very likely representative of only the population used in model development. Finally, it is possible that there are relationships which may account for some of the variability that were impossible to detect with the number of subjects included in the analysis, a problem furt her compounded by the fact that cert ain potential covariates were only derived from one study. Conclusions Data from four studies were included in a population pharmacokinetic analysis of insulin. In agreement with previously reported results, insulin pharmacokinetics were found to be consistent with a two-compartment model. Fo llowing subcutaneous dosing, both sc RHI and lispro were characterized by two sequential first or der absorption rates. No significant differences in insulin pharmacokinetics were obs erved between the health y, type 1 and type 2 diabetic populations. Age was identified as a si gnificant covariate associ ated with the volume of distribution in the central compartment, where volume was found to increase with increasing

PAGE 106

106 age. Age was also found to be related to the rate of pulmonary absorp tion, with increasing age associated with a decrease in absorption rate. Increased BMI was found to be associated with a decrease in insulin absorption rate following subcutaneous administration.

PAGE 107

107 Table 4-1 Summary of demographi cs and baseline characteristics Demographic Variable Diabetes Status Parameter Value Healthy N (# of profiles) 16 (57) Type 1 diabetic N (# of profiles) 75 (136) Type 2 diabetic N (# of profiles) 12 (20) Gender Male N 55 Female N 48 Race Caucasian N 102 Asian N 1 Age (years) Mean SD 36 12 Weight (kg) Mean SD 72 13 BMI (kg/m2) Mean SD 24 3 Height (cm) Mean SD 171 10 FEV1 (L) Mean SD 4 1 Percent of predicted FEV1 (%) Mean SD 98 12 Table 4-2 Mean (%CV) noncompartmental insulin pharmacokinetic parameter estimates Dose Group tmax (h)* Cmax (U/mL) Halflife (h) AUC0-4 (U/mLmin) AUC0-last (U/mLmin) AUC0(U/mLmin) DoseNormalized AUC010 U RHI (n=15) 1.50 30.7 (34) -85.2 (30) 107.0 (26) -10 U lispro (n=9) 1.25 55.5 (21) -147.6 (17) 180.3 (15) --25 U TI (n=11) 0.20 54.6 (72) 1.1 (61) 52.8 (61) 50.0 (61) 57.4 (62) 2.0 30 U TI (n=139) 0.20 52.0 (54) 1.1 (59) 40.2 (47) 40.2 (47) 42.6 (48) 1.4 50 U TI (n=11) 0.20 105.3 (38) 1.3 (31) 102.6 (40) 101.5 (39) 111.2 (39) 2.0 60 U TI (n=6) 0.25 163.1 (43) 1.5 (53) 132.6 (43) 134.1 (43) 139.5 (41) 2.3 90 U TI (n=6) 0.25 217.9 (41) 1.2 (46) 166.9 (35) 168.7 (34) 170.2 (34) 1.9 100 U TI (n=16) 0.33 240.9 (52) 1.4 (57) 213.6 (44) 218.6 (43) 230.6 (41) 2.2 Median presented for tmax; Table 4-3 Tukey 95% simultaneous confidence intervals Compared to Study Lower limit of Conf. interval Center Value Upper limit of Conf. interval Study 1 2 -0.4808 -0.1915 0.09782 3 -0.4861 -0.1652 0.15573 4 -0.6342 -0.3599 -0.08549 Study 2 3 -0.1769 0.0263 0.22956 4 -0.2849 -0.1683 -0.05176 Study 3 4 -0.376 -0.1947 -0.01338

PAGE 108

108 Table 4-4 Univariate analysis Table 4-5 Final model: Backwards elimination Covariate Model Tested OFV Change in OFV Base Model 13760.76 -Age on Vc 13724.22 36.538 Weight on CL 13749.57 11.196 Age on Vp 13760.26 0.502 BMI on kasc 13738.54 22.223 Age on kaTI 13661.83 98.936 Weight on kaTI 13761.16 -0.393 Sex on kaTI 13737.91 22.856 FEV1 on kaTI 13761.31 -0.547 FEV1 on TI F 13761.39 -0.625 Age on TI F 13762.00 -1.238 Body weight on TI F 13760.89 -0.124 Model OFV Change in OFV Final model 13641.22 Final model-age on TI ka 13707.17 -65.945 Final model-age on V 13659.93 -18.709 Final model-BMI on sc RHI ka 13657.96 -16.736

PAGE 109

109 Table 4-6 Population pharmacokinetic parameter estimates of insulin Pharmacokinetic Parameters Ba se Model Full Model Bootstrap Parameter Estimate (%RSE) Estimate (%RSE) Median (95% CI) CL (L/hr) 53.3 (10.0)52.30 (9.6)51.9 (43.0, 69.7) Vc (L) 4.6 (13.5)4.64 (12.8)4.79 (3.83, 6.38) Q (L/hr) 23.9 (12.9)23.00 (12.4)23.3 (18.0, 30.3) V p (L) 37.0 13.9)36.40 (14.0)34.8 (25.7, 47.7) kasc1 (hr-1) 0.6 (47.9)0.68 (24.9)0.76 (0.43, 1.20) kaLI1 (hr-1) 0.7 (28.3)0.73 (36.5)0.72 (0.44, 1.22) kasc2 (hr-1) 1.0 (48.9)0.91 (22.4)0.88 (0.57, 1.60) kaLI2 (hr-1) 1.4 (26.3)1.27 (34.5)1.28 (0.78, 2.15) kaTI (hr-1) 2.6 (5.4)2.39 (5.0)2.44 (2.21, 2.71) Fsc ** 0.75 (13.3)0.72 (12.3)0.71 (0.58, 0.90) FLI ** 1 fixed 1 fixed-FTI ** 0.14 (16.1)0.14(15.9)0.14 (0.10, 0.19) Lower BA of study 4 0.07 (28.6)0.07 (28.2)0.08 (0.05, 0.13) Age on Vc 0.02 (81.8)0.02 (0.001, 0.06) Age on kaTI -0.48 (20.9)-0.47 (-0.69, -0.31) BMI on kasc 0.10 (38.7)0.13 (0.04, 0.42) Interindividual and Residual Variability CL 33.3 30.7 52.8 (33.3, 111) Vc 83.5 82.5 90.8 (80.1, 97.6) Q 70.1 62.6 77.6 (60.6, 108) V p 63.9 63.0 77.7 (61.0, 116) kasc 26.6 21.8 46.2 (28.8, 57.1) kaLI 11.5 10.7 32.8 (6.7, 45.1) kaTI 28.1 19.4 43.2 (5.6, 50.2) Fsc 13.3 13.5 41.2 (8.0, 75.8) FTI -0.0 -IOV FTI 28.9 30.9 58.2 (45.4, 67.3) 1 20.9 20.2 44.8 (42.4, 54.2) 2 0.85 0.870.91 (0.8, 1.05) Note: 2 (additive residual error) is expressed in U/mL; Th e magnitude of interindividual and residual variability was expressed as CV%, approximated by th e square root of the variance estimate. ** F is reported as relative bioavailability to lispro.

PAGE 110

110 Figure 4-1 Pharmacokinetic model diagram 012345Time (hours) 0 50 100 150 200Insulin Concentration (uU/mL) TI 25 U TI 50 U TI 100 U RHI 10 U lispro 10 U TI 60 U TI 90 U TI 30 U Figure 4-2 Mean insulin concentrati on-time profiles by dose and treatment

PAGE 111

111 100 200 300 400Individual Predicted Insulin Concentration (uU/mL) 100 200 300 400Observed Insulin Concentration (uU/mL) A 100 200 300 400Predicted Insulin Concentration (uU/mL) 100 200 300 400Observed Insulin Concentration (uU/mL) B Figure 4-3 Goodness of fit plots of predicted and individual predic ted insulin concen trations. A) Base model individual predic ted versus observed insulin co ncentrations and B) Base model predicted insulin con centrations versus observed insulin concentrations 050100150200250300350 Predicted Insulin Concentration (uU/mL) -10 -5 0 5 10Weighted Residuals A 10-1.0100.0101.0102.023456234562345623456Predicted Insulin Concentration (uU/mL) -10 -5 0 5 10Weighted Residuals B Figure 4-4 Goodness of fit plots of weighted residuals. A) Base model predicted insulin concentrations versus weighted residuals (linear scale) and B) Base model predicted insulin concentrations versus we ighted residuals (log scale)

PAGE 112

112 100 200 300 400Individual Predicted Insulin Concentration (uU/mL) 100 200 300 400Observed Insulin Concentration (uU/mL) A 100 200 300 400Predicted Insulin Concentration (uU/mL) 100 200 300 400Observed Insulin Concentration (uU/mL) B Figure 4-5 Goodness of fit plots of predicted and individual predic ted insulin concen trations. A) Individual predicted versus observed insulin concentrations and B) Predicted insulin concentrations versus weighted residuals 0 100200300400 Predicted Insulin Concentration (uU/mL) -6 -2 2 6 10Weighted Residuals A 10-1.0100.0101.0102.023456234562345623456Predicted Insulin Concentration (uU/mL) -6 -2 2 6 10Weighted Residuals B Figure 4-6 Goodness of fit plots of weighted residuals. A) Final model predicted insulin concentrations versus weighted residuals (linear scale) and B) Predicted insulin concentrations versus wei ghted residuals (log scale)

PAGE 113

113 0123456Time (hours) 0 10 20 30 40Insulin Concentration (uU/mL) BMI 20 BMI 23 BMI 26A. 0.00.51.01.52.02.5Time (hours) 0 50 100 150Insulin Concentration (uU/mL) 20 years 35 years 65 yearsB. 0.00.51.01.52.02.5Time (hours) 0 50 100 150Insulin Concentration (uU/mL) 20 years 35 years 65 yearsC Figure 4-7 Simulated covariate effect on insu lin pharmacokinetics. A) BMI effect on subcutaneous absorption rate B) Age effect on Vc C) Age effect on TI absorption and Vc

PAGE 114

114 CHAPTER 5 PHARMACODYNAMIC MODEL FOR SUBCUT ANEOUSLY ADMINISTERED REGULAR HUMAN INSULIN, INS ULIN LISPRO, AND INHALED INSULIN IN HEALTHY VOLUNTEERS AND TYPE 2 DIABETIC SUBJECTS Background The glucose clamp procedure has been used exte nsively to study insulin activity [9, 17, 35, 39]. During this procedure, patients are administ ered a fixed insulin infusion, while at the same time receiving a varying infusion rate of glucose to counteract insulins action and maintain a constant glycemic state. Suppr essing endogenous insulin secreti on with the insulin infusion is critical to distinguishing between the effects of the administered insulin and insulin secreted by the pancreas in response to the administered gluc ose. As it is assumed that the exogenous insulin infusion completely suppresses hepatic glucos e production as well as appreciable insulin secretion, the glucose infusion rate (GIR) can be used as a measure of the pharmacodynamic response to the administered insulin [36]. This pr ocedure allows for the direct measure of insulin effects by determining the amount of glucose requ ired to maintain blood glucose concentrations within a defined range. A pharmacokinetic/pharmacodynamic (PK/PD ) model was developed for insulin concentrations and the GIR in healthy subjects (C hapter 3). In order to expand the model to the relevant population, glucose clamp data from subj ects with type 2 diabetes was included in the analysis. Subjects with type 2 diabetes exhib it decreased insulin sensitivity [95], and it is expected that the pharmacodynamic response in this population will be different from nondiabetics. The aim of this analysis was the de velopment of a PK/PD model for RHI administered via the inhalation route so as to better refine 1) the model parameter estimates as they differ between a population with type 2 diabetes and healthy subjects, and 2) the impact that the disease state has on insulin eff ect as measured by the GIR.

PAGE 115

115 Materials and Methods Study Population Data from three glucose clamp studies were combined for this analysis. The first two studies were conducted in hea lthy volunteers, and the third study was conducted in type 2 diabetic subjects. Each study was a prosp ective, single-center, open-label, randomized, crossover euglycemic glucose clamp study in non-smoking male and female volunteers with normal pulmonary function. All thr ee studies were performed at the same clinical site. Each was approved by the local ethics committee and was carri ed out in accordance with the principles of the Declaration of Helsinki and of Good Clinical Practice. Subjects gave written informed consent prior to randomization. Prior to entry into the studies all subjects were given a phys ical examination, pulmonary function tests, electrocardiography, and laboratory tests, includi ng urinalysis and screening for drugs of abuse. All blood samples were drawn fr om a forearm vein via an intravenous catheter, placed in the cubital vein of th e arm contralateral to the one used for the continuous insulin infusion and glucose administration. For this analysis, only TI data was used since this was the only treatment used in both the healthy subjects and the subjects with type 2 diabetes, thus only handl ing of TI data is described. However, for completeness, the study conduct for th e studies is described in full, including all treatments. Study Design and Insulin Concentrations Healthy volunteers Both studies were euglycemic glucose clamp procedures performed with the Biostator glucose monitoring and infusion system (Biostator Life Science Instruments, Elkhart, IN, USA), and a continuous insulin infusion to suppress endogenous insulin production. Following an

PAGE 116

116 overnight fast, on each of the treatment days, the subjects received a constant rate intravenous infusion of RHI during a 2-hour run-in period to establish a serum insulin concentration of 10 U/mL. This infusion was continued until the end of the study. Each subject received a single dose of the test treatment on separate occasions. Study 1: This was a three-way cross over euglycemic glucose-clamp study in five subjects to compare the pharmacokinetics and pharmacodynami cs of single doses of 100 U of inhaled TI, 10 U RHI administered subcutaneously and 5 U RHI (Actrapid, Novo Nordisk A/S, Bagsvaerd, Denmark) administered intravenously. Study 2: This was a four-way crossover euglycem ic glucose-clamp study in 12 subjects to compare the pharmacokinetics and pharmacodynamics of three different si ngle doses of inhaled TI with a single dose of 10 U RHI (Actrapid, Novo Nordisk A/S, Bagsvaerd, Denmark). TI was administered using a commercially available inhaler (Model M, Boehringer Ingelheim) in study 1, and using the MedTone Dry Powder Inhaler (MannKind Corporation, Danbury, CT) in study 2. RHI insulin was administered by subcutaneous injection in the abdominal region. Samples for insulin concentr ation determination were drawn at 120, 90, 60 and 30 minutes before dosing and 0, 1, 3, 7, 12, 20, 30, 45, 60, 90,120, 180, 240, 300 and 360 minutes post-dose, and analyzed for insulin co ncentration using radioimmunoassay (RIA) with double determinations. C-peptide concentrations were collected at 120 a nd 60 minutes pre-dose and 0, 30, 60, 180, and 300 minutes post-dose. The samples were cooled in an ice-water bath before centrifugation. After centrifugation the plasma samples were immediately frozen and stored at -80C until analysis. Blood glucose was kept constant at 90 () mg/dL throughout the glucose clamp by a variable infusion of a dextrose solution. The patients were kept in a fasted

PAGE 117

117 state until the end of each trea tment visit. The treatment peri ods were separated by a washout period of 3 and 28 days. Type 2 diabetic subjects Study 3: This was a prospective, randomized, ope n label, single site study in 12 nonsmoking, male and female subjects with a di agnosis of type 2 diabetes mellitus for 12 months, stable antidiabetic regimen with insulin for the previous 3 months and HbA1c 8.5%, to compare the pharmacokinetics and pharmacodynamics of two doses of TI (60 and 90 U) with a single dose of 10 U lispro (Humalog Eli Lilly & Co.). Six subjec ts received 60 U TI and the other six subjects received 90 U TI. All twel ve subjects received lispro administered by subcutaneous injection in the abdominal region. Following an overnight fast, on each of the treatment days, the subjects received a constant ra te intravenous infusion of either RHI or lispro (opposite of the type of insulin to be used in dosing) during a 4-6-hour ru n-in period to suppress endogenous insulin production and to obtain target blood glucose levels of 110 mg/dL. This infusion was continued until the end of the clamp procedure. Blood glucose was kept constant at 110 () mg/dL throughout the glucose clamp by the Biostator. Samples for insulin and lispro concentration determinati on were drawn at 310, 20 and 5 minutes before dosing and at time 0, 5, 10, 15, 20, 30, 45, 60, 75, 90, 105, 120, 150, 180, 210, 240, 300, 360, 420 and 480 minutes post-dose. The patients were kept in a fa sted state until the end of each treatment visit. The treatment periods were separated by a washout period of 7 to 21 days. Glucose Infusion Rates and Blood Glucose Concentrations Glucose infusion rates were registered by th e Biostator every minute from 120 minutes before dosing until 360 minutes post-dose (studies 1 and 2) and from 60 minutes before dosing until 480 minutes post-dose (study 3). If the gluc ose pump could not meet the glucose infusion requirements, an additional external pump cont rolled by study personnel, was employed. The BG

PAGE 118

118 measurements of the Biostator were recalibra ted at regular interval s (range, 10 minutes) by the YSI 2300 STAT Plus Glucose Analyzer (YSI Life Sciences, Yellow Springs, OH), which employed the glucose oxidase method. Baseline Correction Studies 1 and 2: Insulin concentrations collected at 120, 90, 60 and 30 minutes before dosing were averaged, and subtracted from all su bsequently observed insulin concentrations, to account for any incomplete endogenous insulin s uppression and the constant insulin infusion. Any resulting negative values were set to zero. Study 3: Baseline correction was applied to account for any endogenous insulin that was not fully suppressed with the baseline infusion. All insulin concentrations were corrected by subtracting the mean of the values in th e later time points (300 minutes) from each individual time point for each subject. At these time points, most subjects insulin concentrationtime profiles were observed to be reasonably flat and uninfluenced by exog enous contribution. GIR Smoothing Methodology For the pharmacodynamic analysis, GIR values were smoothed using a 10-minute running average. For the noncompartmental GIR analysis, GIR values were first baselinecorrected, where GIR values from 60 minutes prior to dosing were averaged, and subtracted from all subsequently observed GIR values. Resulting ne gative values were set to 0, and the data was smoothed using a 10 minute running average. Insulin Data for the Pharmacokinetic/Pharmacodynamic Analysis Individual predicted insulin pharmacokinetic parameters from the pharmacokinetic model were used to simulate the insulin concentrations used in the analysis. Since the pharmacokinetic model was developed using baseli ne corrected insulin concentra tions and the PK/PD analysis targeted exploring the relationship between tota l insulin concentrations and total GIRs, the

PAGE 119

119 simulated insulin values were each added to the baseline used for correction at each visit. GIR and insulin values ever 5 minutes for the firs t hour and every 10 minut es thereafter were combined for the analysis. Data Excluded from the Pharmacokinetic Analysis For studies 1 and 2, each serum insulin concen tration-time profile was examined visually together with C-peptide concentrations. Due to the lower doses administered in study 2, a higher incidence of elevated C-peptid e concentrations was observed in the latter timepoints following TI treatment, particularly in the lower dose groups All insulin concentra tions after 180 minutes post-dose in the 25 U dose group, 240 minutes post-dose in the 50 U dose group, and 300 minutes in the 100 U dose group were excluded, since endogenous insulin appeared to be contributing to the curve in the la tter part of the study, with a to tal of 51 non-BLQ concentrations were excluded from the analysis. One subject from study 2 was excluded from the analysis. Noncompartmental Pharmacokineti c and Pharmacodynamic Analysis Insulin pharmacokinetics were analyzed using noncompartmental methodology using baseline-corrected insulin concentrations. The following PK parameters were derived using WinNonlin v 5.2 (Pharsight Corporation, Mountain View, CA): observed peak insulin concentration (Cmax), time to peak insulin (tmax), and total insulin exposure as measured by the area under the insulin concentration-time curve from time 0 until the last time point with nonzero insulin concentrations (AUC0-last), calculated by the linear-trap ezoidal method. The terminal elimination half-life (t1/2) was calculated as ln(2)/ z in accordance with pharmacokinetic theory [80], where z is the terminal elimination rate cons tant estimated from log-linear regression analysis of the terminal elimination phase of the concentration-time profile, and AUC0was calculated as calculated as AUC0-last + Clast/ z, where Clast is the last observed insulin concentration. Noncompartmental parameter values were used to determine starting estimates for

PAGE 120

120 pharmacokinetic modeling, and were compared to parameter values estimated by the final model as part of the model fit assessment. Insulin pharmacodynamic (GIR) parameters were derived using WinNonlin v 5.2. The following parameters were determined from th e 10 minute running average smoothed GIR data: observed peak GIR (GIRmax), time to peak GIR (GIR tmax), and the area under the GIR-time curve from time 0 until time t (GIR0-t AUC), calculated by the linear-trapezoidal method. Statistical Analysis Dose-normalized log transformed insulin AUC0was used in order to investigate doseproportionality. Dose-proportionality could not be concluded if the average dose-normalized AUC0values were found statistically different by a one-way ANOVA technique, which was performed with the null hypothesis that the dose av erage log transformed AUC0values are the same, at the =0.05 level. The alterna tive hypothesis was that some of the group means differ from each other. It was necessary to test the variances of the five dose groups for homogeneity in order to accept the results of the ANOVA an alysis, since homogeneity of variances is one of the ANOVA assumptions. Levenes test wa s both performed under the null hypothesis that the variances are the same and an alternative hypothesis that the at least one of the variances was different from the others. The data were tested at the =0.05 level. The results of the statistical tests were obtained using MINITAB (v.15.1.30.0). Linear regression was used to assess whet her or not the 90% CI around the intercept included 0. The CI was calculated as he averag e 1.96 standard erro r of the mean (SEM).

PAGE 121

121 Pharmacokinetic/Pharmacodynami c Model and Data Analysis The software package NONMEM (Version VI, Level 1.2, NONMEM Project Group, ICON Development Solutions, USA) was used fo r the population analysis using both the first order (FO) and the first order cond itional estimation method (FOCE). NONMEM describes the observed concentr ation-time data in terms of: A number of fixed effect parameters, which may include the mean values of the relevant base pharmacokinetic model parameters, or a number of parameters which relate the base model parameters to dem ographic and other covariates; Two types of random effect parameters: (a) 2: the variances of the interindividual variability ( ) within the population, and (b) 2: the variances of the residual intraindividual variability ( ) due to random fluctuations in an individuals parameter values, measurement error, model misspecifica tion, and all sources of error not accounted for by the other parameters. The population or average values of the parameters, the interindividual variances, 2, and the residual variance, 2, were estimated by NONMEM. The values are independent, identically distributed rando m errors with mean of zer o and a variance equal to 2. Subjectspecific parameters were calculated by NONMEM using the POSTHOC (FO method) option. These parameters are empirical Bayesian estimates of the individuals true parameters based on the population parameters and the indi viduals observed concentrations. Pharmacokinetic and pharmacodynamic model Data from both studies were modeled simultaneously, but the pharmacokinetic and pharmacodynamic models were developed sequen tially. The pharmacokinetics was described by a two compartment open model with first order absorption and was parameterized in terms of apparent clearance (CL/F), volume of di stribution in the central compartment (Vc/F), the apparent intercompartmental clearance (Q/F), the volume of distribution in the peripheral compartment (Vp/F) and the first order absorption rate c onstant (ka). NONMEM library

PAGE 122

122 subroutines ADVAN4 and TRANS4 were used. A diagram of the combined PK/PD model is presented in Figure 5-1. In the PD model, the GIR was used as the dependent variable. When exploring the potential PK/PD model, the relationship between insulin and GIR was first examined by visual inspection. The pharmacodynamic model was desc ribed by the following differential equations: CeCpk dt dCee0 (5-1) e eCEC CE EGIR 50 max 0 (5-2) Equation 5-1 describes the relationship between the observed plasma insulin concentration and the concentration at the e ffect site. Effectively, ke0 is a rate constant which describes the delay in effect. The effect comp artment is assumed to receive a negligible mass from the central compartment, thereby not affecting the equati ons for the insulin pharmacokinetic model. Equation 5-2 establishes the relationship between the GIR and insulin concentration at the effect site. The pharmacodynamic structural model was parameterized in terms of E0, the baseline GIR value, Emax, the maximum glucose infusion rate, EC50, the effect site concentration eliciting 50% of the maximal response, the sigmoidicity factor, and Cp and Ce are the plasma and effect site insulin concentrations, respectively. Error model Interindividual variability was describe d by an exponential error model. The intraindividual residual variability of insulin plasma concentration was estimated using a proportional and additive error model (PK) and additive error model (PD).

PAGE 123

123 Results Patient Population A total of 621 insulin concentrations from 28 subjects and 50 profiles were included in the analysis. Demographic data is summarized in Table 5-1. Pharmacokinetic and Pharmacodynamic Results Noncompartmental analysis Mean baseline corrected insulin and uncorrected GIR-time profile s are presented in Figure 5-2 and the PK and PD parameters are presen ted in Table 5-2 and Tabl e 5-3, respectively. Pharmacokinetics: The average time-insulin concentration time profiles increased with increasing dose for insulin following TI administration, and tmax and half-life remained relatively constant with dose (Table 5-2). Insulin exposure following TI, as measured by insulin AUC0-360 and AUC0increased in a proportional manner with increasing dose with the doubling of the dose resulting in an approximate doubling in exposure. An ANOVA and linear regression (Figure 5-3) analyses were performed to a ssess the relationship between dose and insulin exposure. Log-transformed dose-normalized insulin AUC0was analyzed using a one way ANOVA, with no differences found between the do se groups (p=0.985). Levenes test was performed to test the variances of the four studies for homogene ity. The null hypothesis of equal variances could not be rejected at the =0.05 level, with a p value of 0.968. Therefore, the variances met the homogeneity assumption. Due to the high variability observed in this parameter, the r-squared value resulting from th e linear regression was only 0.5061, even with the regression line falling very closely to the mid-points of the parameter values. The intercept estimate was of 1.5216, slope estimate of: 2.2 and associated p < 0.0001. The 90% CI around the intercept (-36.66, 39.7) included 0, and hence dose-proportionality was concluded.

PAGE 124

124 Pharmacodynamics: A phase-plot of GIR vs. the plasma insulin concentration, where data points are connected in chr onological order, is shown in Fi gure 5-4 for the pooled data in this analysis. A counter-clockwise hysteresis loop is observed indica ting a disconnect between insulin concentrations observed centrally a nd insulin action. Unlike the pharmacokinetic profiles, the GIR profiles did not appear dose-proportional from vi sual inspection (Figure 5-2). This was confirmed by an inspec tion of the baseline-corrected GIR parameters, where both the mean GIRmax and mean GIR AUC0-360 were lower in subjects with type 2 diabetes (Table 5-3). Differences in study conditions and patient population resulted in lower GIRs in the subjects with type 2 diabetes. Furthermore, little difference was observed in the mean GIR response between the 60 and 90 U dose groups (Table 5-3), suggesting that the differences in mean concentrations seen in this small sample size did not result in r eadily discernable mean differences in GIR response. Pharmacokinetic and pharmacodynamic analysis Pharmacokinetic model: A two-compartment open model with first order absorption processes and first-order elimination described th e insulin concentration data well. The overall model fit was good, with individual predicted versus observed valu es distributed along the line of unity, and no significant trends in the weighted residuals (Figur e 5-5). Thus, the structural and error models appeared to adequately descri be insulin pharmacokinetics and explain the variability in the data. The population typical values were: apparent clearance, 466 L/hr; apparent volume of distribution in the central compartment, 38.2 L; apparent volume of distribution in the peripheral compartment, 258 L, an intercompartmental clearance of 171 L/hr and an absorption rate of 2.0 h-1. The base model pharmacokinetic parameter estimates are summarized in Table 5-4.

PAGE 125

125 Pharmacodynamic model: For the pharmacodynamic analysis, parameter values estimated by Model A (Chapter 3) developed usin g data from healthy subjects, were used as starting values. To account for the different blood glucose clamp settings, the E0 was estimated separately for the two groups. To explore and ev aluate the differences in the response between the two groups, all other parameters were es timated separately, with the exception of Emax, which was fixed to the value estimated for the healt hy population. The limited range of values and small N associated with the data from the t ype 2 population made the estimation of this parameter difficult, and work previously performed strongly suggests that insulin Emax would not be affected by the disease state [96]. Parameter estimates are presented in Table 5-5. The overall model fit was good, with indivi dual predicted versus observed values distributed along the line of unity, and no significant trends in the weighted residuals (Figure 56) for either the healthy (blue) or type 2 diabetic (orange) subject s. Thus, the structural and error models appeared to adequately describe insuli n pharmacodynamics and explain the variability in the data. Individual predicted GI R values overlaid with observed va lues for all subjects in the analysis are presented in Figure 5-7. Discussion Glucose clamp studies have been used to co mpare the activity of insulins, and GIR data have been assessed using a variety of modes, including indirect [56, 57, 72] and effect compartment models [55, 59, 60]. In previous work, an Emax model was used to describe the PK/PD relationship of insulins with different pharmacokinetics (Chapter 3). Because of its simplicity, and because the different in sulins were modeled simultaneously, the pharmacodynamic parameter estimates from the model developed allow for the prediction of insulin effect completely based on insulin pharm acokinetics alone, and can therefore be used to predict the effect of insulins with differing pha rmacokinetic properties. However, data from

PAGE 126

126 healthy subjects was used for model development, and the pharmacodynamic parameters may not apply to subjects with type 2 diabetes, who exhibit decrease d insulin sensitivity [95]. The work presented here expands the model to the relevant popul ation by including glucose clamp data from subjects with type 2 diabetes. As seen previously, when the pharmacodynamic parameters were estimated independently for each insulin formulation (Chapter 3), interand intra-individual errors decreased when compared to the model where parameter estimates were not individualized fo r each formulation, and the model resulted in improved fit. In this analysis, in order to improv e the accuracy of the fits and better elucidate the difference between healthy individu als and subjects with diabetes, only TI data was included in the analysis. Noncompartmental analysis of the PK da ta showed that dose-proportionality was maintained when the healthy individual da ta was combined with data following TI administration to subjects with type 2 diabetes The two compartment pharmacokinetic model fit the data well, without any need to account for di fferences in disease state between the subjects, and the pharmacokinetic parameter estimates were almost identical to parameter estimates obtained in a model incorporating only the data from the healthy volunteers. However, the response, assessed from the mean GIR curve and the noncompartmental pharmacodynamic parameters, did not appear proportiona l. In fact, the response was lower, on average, in subjects with type 2 diabetes receiving the 90 U TI dos e than healthy volunteers receiving 50 U TI. Although it is impossible to compare the data di rectly due to the di fferent glucose clamp setting (90 mg/dL in healthy subjects and 110 mg/dL in subjects with type 2 diabetes), the difference in response appears markedly different between the two populations in light of the relatively small difference in the target glucose concentration. A literature search was conducted

PAGE 127

127 to try to establish the effect that different cl amp settings would have on the GIR parameters in comparable populations, so that an adjustme nt could be made in the model; however, no published results which would be applicable were found. Baselin e correction for the difference in the glucose needed to maintain the target glucose concentration was therefore assumed to account for the differences in the clamp settings, so that the data could be compared. Although a simplification of the system, the clamp settings were not very different and hence, were not expected to account for the dram atic differences observed. Fo llowing baseline correction, the GIR response was still visibly lower in the type 2 diabetic population. This result is not unexpected due to the insulin resistance observe d in subjects with t ype 2 diabetes, and a decreased response in the diabe tic population has been demons trated previously [96]. When the data was modeled, the pharmacodyna mic parameter estimates obtained in the healthy volunteers (Chapter 3) were used as starting estimates for the model. All of the parameters were estimated separately fo r the two groups, with the exception of Emax, which was fixed to the value estimated for the healthy population. The limite d range of values and small N associated with the data from the type 2 popul ation made the estimati on of this parameter difficult. Although there is a sound basis for this approach in work previously performed by Rizza [96] and the model fit was very good, an valuable improvement to the model would be the addition of more data from diabetic subjects, so that the insulin Emax could be estimated independently in this population. Most of the pharmacodynamic parameters were similar for both groups ( =2.5 and 2.7 for the healthy and type 2 diabetic subjects, respectively, and a ke0 of 1.4 and 1.8 h-1, respectively). The lack of a large difference in ke0 suggests that the distribution ti me is not affected to a large extent by disease state. However, the EC50 and estimate was found to be clearly increased in the

PAGE 128

128 diabetic population. In this set of subjects with diabetes, the magnitude of difference was approximately a three-fold increase in EC50, indicating a significant decrease in insulin sensitivity. Conclusions Insulin pharmacodynamics were found to be well described by a simple Emax model, after a hypothetical effect compartment was used to colla pse the hysteresis in insulin effect observed during clamp procedures in healthy a nd type 2 diabetic subjects. The EC50 parameter estimate was found to be approximately three-fold higher for subjects with type 2 diabetes, most likely due to the insulin resistance that is asso ciated with this disease state.

PAGE 129

129 Table 5-1 Summary of demographics and baseline characteristics Table 5-2 Mean (%CV) noncompartmental pharmacokinetic parameter estimates Dose Group tmax (h)* Cmax (U/mL) AUC0-last (U/mLmin) Half-life (h) AUC0(U/mLmin) Dose-Normalized AUC025 U TI (n=11) 0.2054.6 (72) 50.0 (61) 1.1 (61) 57.4 (62) 2.0 50 U TI (n=11) 0.20105.3 (38) 101.5 (39) 1.3 (31) 111.2 (39) 2.0 60 U TI (n=6) 0.25163.1 (43) 134.1 (43) 1.5 (53) 139.5 (41) 2.3 90 U TI (n=6) 0.25217.9 (41) 168.7 (34) 1.2 (46) 170.2 (34) 1.9 100 U TI (n=16) 0.33240.9 (52) 218.6 (43) 1.4 (57) 230.6 (41) 2.2 Median presented for tmax; Table 5-3 Mean (%CV) pharmacodynamic parameters Demographic Variable Value Diabetes Status N (number of profiles) Healthy N 16 (38) Type 2 diabetic N 12 (12) Gender Male N 24 Female N 4 Race Caucasian 27 Asian N 1 Age (years) Mean SD 40 15 Weight (kg) Mean SD 82 13 BMI (kg/m2) Mean SD 26 4 Height (cm) Mean SD 178 8 FEV1 (L) Mean SD 4 1 Percent of predicted FEV1 (%) Mean SD 99 13 Dose (U) GIR tmax (min) GIRmax (uncorrected) (mg/kg/min) GIR AUC0-360 (uncorrected) (mg/kg) GIRmax (corrected) (mg/kg/min) GIR AUC0-360 (corrected) (mg/kg) 25 (n=11) 50 9.9 (28)1857 (38)8.1 (32) 1221 (44) 50 (n=11) 45 13.2 (36)2230 (32)11.1 (38) 1490 (36) 60 (n=6) 43 8.7 (26)1183 (29)7.6 (29) 825 (28) 90 (n=6) 43 9.1 (57)1228 (42)8.2 (54) 933 (41) 100 (n=16) 38 16.1 (39)3032 (36)13.8 (43) 2220 (40)

PAGE 130

130 Table 5-4 Population pharmacokinetic parameters of insulin Pharmacokinetic Parameters Parameter Values Interindividual and Residual Variability Parameter Estimate (%RSE) Parameter %CV CL (L/hr) 466.0 (9.7) CL 33.6 Vc (L) 38.2 (19.3) Vc 72.9 Q (L/hr) 171.0 (13.9) Q 64.0 V p (L) 258.0 (21.8) V p 67.3 kaTI (hr-1) 2.0 (7.4) ka 22.2 IOV FTI 35.4 1 24.6 2 1.42 Note: 2 (additive residual error) is expressed in U/mL; Th e magnitude of interindividual and residual variability was expressed as CV%, approximated by th e square root of the variance estimate. Table 5-5 Pharmacodynamic parameters of insulin in healthy and t ype 2 diabetic subjects Parameter Healthy Subjects Parameter Estimate (%RSE) Type 2 Diabetic Subjects Parameter Estimate (%RSE) E0 (mg/kg/min) 2.5 (17.1) 1.4 (15.2) Emax (mg/kg/min) 14.4 (14.5) 14.4 fixed EC50 39.9 (15.2) 121.0 (6.5) Ke0 (h-1) 1.4 (14.9) 1.8 (16.4) gamma 2.5 (16.0) 2.7 (16.7) Interindividual and Residual Variability %CV %CV E0 49.5 39.9 Emax 37.0 58.8 EC50 39.0 -Ke0 49.5 53.0 g amma 44.6 49.3 1 1.80 1.16 Note: 1 (additive residual error) is expressed in mg/kg/min; The magnitude of interindividual and residual variability was expressed as CV%, approximated by the square root of the variance estimate.

PAGE 131

131 GIR effectInhalation kcpkpck1ecentralperipheral ka depot Emax ke0GIR effectInhalation kcpkpck1ecentralperipheral ka depot Emax ke0 Figure 5-1 Pharmacokinetic/pharmacodynamic model diagram 050100150200250300Time (minutes) 0 50 100 150 200Insulin Concentration (uU/mL) 25 U healthy (n=11) 50 U healthy (n=11) 100 U healthy (n=16) 60 U T2DM (n=6) 90 U T2DM (n=6)A. 0100200300400500Time (minutes) 0 4 8 12GIR (mg/kg/min) 25 U healthy 50 U healthy 100 U healthy 60 U T2DM 90 U T2DMB. Figure 5-2 Insulin and GIR-time pr ofiles. A) Mean individual insulin concentrations by dose group and B) Mean observed GIR by dose group

PAGE 132

132 20406080100Dose (U) 0 100 200 300AUC0-inf (uU/ml*hr) 1.5216 + 2.2033*x 25506090100Dose (U) 0 1 2 3 4 5Dose-normalized AUC0-inf Figure 5-3 Dose proportionality assessment. A) Dose vs. insulin AUC0with regression line and B) Dose-normalized AUC0vs. dose as box and whiske r plots with median 050100150200Insulin Concentration (uU/mL) 0 4 8 12GIR (mg/kg/min) 25 U healthy 50 U healthy 100 U healthy 60 U T2DM 90 U T2DM Figure 5-4 Hysteresis in the insulin-GIR relations hip for healthy and type 2 diabetic subjects

PAGE 133

133 100200300400Individual Predicted Insulin Concentration (uU/mL) 100 200 300 400Observed Insulin Concentration (uU/mL) A 10-1.0100.0101.0102.0Predicted Insulin Concentration (uU/mL) -5 -3 -1 1 3 5 7Weighted Residuals B Figure 5-5 Goodness of fit plots for the pharmacokine tic model. A) Model individual predicted versus observed insulin concentrations and B) Model predicted insulin concentrations versus weighted residuals 0510152025Individual Predicted GIR (mg/kg/min) 0 10 20 30Obeserved GIR (mg/kg/min) A 02468101214Predicted GIR (mg/kg/min) -6 -4 -2 0 2 4 6Weighted Residuals B Figure 5-6 Goodness of fit plots for the pharmac odynamic model. A) Individual predicted GIR versus observed GIR and B) GIR values vers us weighted residuals (orange=type 2, blue=healthy)

PAGE 134

134 37 37Time (hours) 10 30 10 30 10 30 ID: 1 ID: 2 ID: 3 ID: 4 ID: 5 ID: 6 ID: 7 ID: 8 ID: 9 ID: 10 ID: 11 ID: 43 ID: 44 ID: 45 37 37Time (hours) 10 30 10 30 10 30 ID: 47 ID: 49 ID: 50 ID: 51 ID: 52 ID: 54 ID: 56 ID: 57 ID: 58 ID: 101 ID: 102 ID: 103 ID: 104 ID: 105 Figure 5-7 Individual predicted GIR versus observed GIR by subject (gray symbols=healthy subjects; red symbols=type 2 subjects)

PAGE 135

135 CHAPTER 6 CONCLUSIONS The purpose of this research was to de velop population pharm acokinetic (PK) and pharmacokinetic/pharmacodynamic (PK/PD) models fo r insulins with different PK properties, and to characterize and bridge the PK and PD pr operties of a quickly absorbed inhaled insulin, subcutaneous insulin and insulin administered intravenously, using a population approach in healthy and type 2 diabetic subjec ts. The models examined were empirical in nature, and the PD response was based on GIR data derived from glucose clamp studies. The PK and PK/PD models were developed first in healthy volunteers and then expanded to include subjects with diabetes. The first aim was to characterize insulin pha rmacokinetics. Following intravenous dosing, insulin is characterized by multiple compartment disposition [55], however, the slow absorption characteristics of subcutaneously administered insulin obscure the distribution phase in what is termed flip-flop kinetics, where elimination is driven by a slower absorption process [38], making it impossible to model the second compartment. Technosphere Insulin (TI) is a novel inhaled regular human insulin (RHI) whose unique delivery mechanism results in rapid absorption and rapid clearance, making it possible to distinguish the second compartment of its pharmacokinetic profile [78]. Because insulin ph armacokinetic properties are expected to be consistent once it is available systemically, the in clusion of data from TI and intravenous dosing, both of which have distinct and phases in their pharmacokinetic profiles, made it possible to demonstrate the two compartment disposition of a ll three routes of insulin administration. Data from two studies was pooled for the anal ysis, with a total of 16 healthy subjects treated with insulin administ ered intravenously, subcutane ously and via the lung while undergoing a euglycemic glucose clamp procedur e. Insulin absorption following pulmonary

PAGE 136

136 administration was rapid and well described by a first order absorp tion rate constant; however, two sequential absorption rates and a transit comp artment were used to describe the slower absorption seen with the subcutaneously administ ered insulin. Following intravenous dosing, the distribution phase was approximately thirty minut es. The apparent distribution phase was longer following TI administration, and undistinguishable from the elimination phase following subcutaneous dosing due to the prol onged duration of absorption. In the final model, the popul ation typical values were: clearance, 43.4 L/hr; volume of distribution in the central compartment, 5.0 L; volume of distribution in the peripheral compartment, 30.7 L. The absolute bioavailabil ity for subcutaneous insulin and TI was 52% and 11%, respectively, matching both results reporte d with TI [83] and sc RHI [84]. The and half-lives were calculated from the individual predicted parameters, and were 5 and 93 minutes, respectively, which are in close agreement w ith insulin half-life estimates following iv administration reported previously [55]. Interoc casion variability in bioavailability following TI administration was approximately 30%, indicating moderate differences in insulin exposure on different dosing occasions. Overall, the variability in insulin bioavailability was similar between the two non-iv treatments. Covariate analysis identified BMI as a signi ficant covariate on insulin absorption rate following subcutaneous dosing, with decreased absorption with increasing BMI, as previously reported by others [90]. This finding may be attributed to the increased thickness of the subcutaneous tissue in subjects with higher BMI, which slows absorption from the depot compartment. No other covariates were iden tified, but due to the homogenous nature of the subject population, covariate analysis was limited.

PAGE 137

137 The population model was expanded to include da ta from subjects with both type 1 and type 2 diabetes, as well as data following administ ration of lispro in Aim 3. The same PK model was used to describe the data as in the first aim, and insulin PK wa s found to be consistent with a two-compartment model. The inclus ion of lispro in the model was appropriate, since lispro has been shown to exhibit almost identical pharmacokinetics as RHI when administered intravenously [37], and its more rapid apparent clearance can be attributed to its more rapid absorption rate. As expected, TI was characte rized by the quickest absorption rate, followed by lispro and RHI. Both subcutaneously administer ed insulins were adequately described by two sequential first order absorp tion rate constants. In the model developed, subcutaneous RHI and TI are estimated to have a relative bioavailability of 72% and 14%, respectively, when compared to lispro. Assuming the lispro absolute bioavailability to be 77%, as reported in literature, the parameter estimates match well with those estimated in an analysis reported earlie r (Chapter 2), with an absolute bioavailability adjusted systemic clearance of 40.3 L/hr (previou sly reported 43.4 L/hr), a central volume of distribution of 3.6 L (reported 5. 0 L) and a peripheral volume of distribution of 28 L (reported 30.7 L). Covariate analysis identified age as an im portant covariate positively correlated with insulin volume of distribution in the central compartment. Age was also found to have significant effect on the rate of pulmonary abso rption when insulin was administered into the lung, with a decrease in pulmonary absorption with increasing age, independently of any ageassociated decreases in pulmonary function. This finding could be benefici al to older patients, who have a slower gut transit time and longer absorption period compared to younger patients. In older patients, high and fast insulin peaks might cause a mismatch in insulin peak and glucose

PAGE 138

138 appearance from the meal, causing early hypoglycem ic events, but a slightly longer absorption and lower overall insulin peak will result in less like lihood of early hypoglycemia. Increases in BMI were associated with a decrease in insu lin absorption rate following subcutaneous administration. The small number of subjects, narrow BMI range and homogenous nature of the healthy population makes it difficult to extrapolate these results, however, it is expected that a greater impact would be observe d on patients taking subcutaneous insulin who have higher BMI, as is often the case in subj ects with type 2 diabetes. The second aim was to characterize the pharmac odynamics of the data included in Aim 1. The data originated from two glucose clamp st udies, and the glucose infusion rate (GIR) was used as the pharmacodynamic endpoint. Data following intravenous, subcutaneous and pulmonary administration was modeled simultaneously. An Emax model was found to describe the data well when the hysteresis was first collaps ed using an effect compartment. In the first model (Model A), treatment specific EC50, ke0 and were modeled. The parameter estimates varied most notably in the parameter, with the highest value associated with intravenous insulin administration. The other parameter which varied considerably following intravenous administration was ke0, which was lowest in this treatment, and associated with the longest central to effect site equilibration half-life. This is most likely due to the relatively quickly changing insulin concentrations fo llowing this route of administra tion, but comparable delay in insulin effect for all three treatments. In the second model (Model B), no treatment -specific differences were assumed. Pharmacodynamic parameter variability ranged from 27 to 52%, with the greatest variability observed on the sigmoidicity factor. This was an e xpected result, since this is the parameter that varied most between treatments in the first model. The residual e rror was described by an

PAGE 139

139 additive model in both models A and B, and the residual error ha d a greater magnitude in Model B (2 mg/kg/min) when compared to Model A (1 .68 mg/kg/min), indicatin g that more of the observed variability was explained by the tr eatment-specific parameters and associated interindividual variability in Model A. Although the simple nature of th e second model is a benefit, it is also a drawback in that the model underpredicts the GIR around the time of peak effect in some patients, especially in subjects who exhibit a quick and high rise in insu lin. Hence, the model fit is less predictive of maximum effect in certain subjects in the 100 U TI dose group, and to a greater extent, in subjects receiving insulin intravenously. The r eason for this observation may be due to the fact that insulin effect has two compone nts: the stimulation of glucose disposal, as well as inhibition of hepatic glucose production [76], and the simple nature of the model cannot account for both. It appears that the model may describe insulin concentration-related gluc ose disposal, which is expected to follow a receptor-driven Emax model, it is unable to account for changes in insulin effects on hepatic glucose production, which are more immediate a nd associated with a threshold insulin value [9]. The effects of exogenous insulins are often compared based on the shape of, and area under the GIR vs. time curve, and numerous attemp ts have been made to model the insulin-GIR relationship. However, previous work focused on either one or similar insulin formulations, or explicitly estimated different pharmacodynamics for insulins with different pharmacokinetics, providing limited use towards predicting insulin effect for insulins other than those included in the model. The potential predictive ability of Model B makes it unique since it can easily be applied to simulate the activity of insulins with varying pharmacokinetic properties. In most

PAGE 140

140 cases, the model performs well and is reasonable, with its only weakness being the underprediction at specific conditions. The PD model was expanded to include glucos e clamp data from subjects with type 2 diabetes, who, due to their disease state, exhibit decreased insulin sensitivity [95]. In this analysis, in order to improve th e accuracy of the fits and better elucidate the difference between healthy individuals and subjects wi th diabetes, only TI data was included in the model. The PK model was developed first, with a good fit of a two compartment model, and with no difference in the PK between the two popul ations. This finding was confirmed by a noncompartmental analysis, which showed that dose-pro portionality for both insulin AUC and Cmax in the pooled dataset. However, the response, assessed from the mean GIR curve and the noncompartmental pharmacodynamic parameters, did not appear propor tional. Although it is impossible to compare the data directly due to the different target glucose concentrations duri ng the clamp for the two populations, the difference in response appeared markedly different between the two groups, even though the blood glucose target differed only slightly. This result is not unexpected due to the insulin resistance observed in subjects with type 2 diabetes. When the data was modeled, the insulin EC50 estimate was found to be approximately thre e-fold higher in the diabetic population, and was indicative of a significant decrease in insulin sensitivity due to the disease state. Little difference was observed in ke0, which suggests that the distributi on time is largely independent of the disease state. In conclusion, insulin was found to be well described by a two compartment pharmacokinetic model, with the differences in the insulin profiles attributable to absorption differences for the different routes of admi nistration. Insulin pharmacodynamics were well

PAGE 141

141 described by an Emax model, related to effect site concen tration. Some underprediction of the GIR resulted with high and quick in sulin peaks. Model fit was better when the model estimated different EC50, ke0 and values for each of the insulin fo rmulations, however, when one common set of parameters were used, the model still de scribed the data well, with a modestly greater underprediction of the GIR associated at some peaks, in particular, those associated with intravenous dosing. Although th e increase in underp rediction is a weakness of the simpler model, its ability to predict insulin effect based on insulin pharmacokinetics alone makes the model much more useful in practical applications. The extension of this model to subjects with type 2 diabetes found this population to ha ve an approximately three-fold higher EC50 when compared to healthy subjects, most likely due to the decreased insulin se nsitivity associated with the disease state.

PAGE 142

142 LIST OF REFERENCES [1] Economic costs of diabetes in th e U.S. In 2007. Diabetes Care. 2008; 31: 596-615 [2] Wild S, Roglic G, Green A, Sicree R, Ki ng H. Global prevalence of diabetes: estimates for the year 2000 and projections for 2030. Diabetes Care. 2004; 27: 1047-1053 [3] Weyer C, Bogardus C, Mott DM, Pratley RE The natural history of insulin secretory dysfunction and insulin resistance in the pat hogenesis of type 2 diabetes mellitus. The Journal of Clinical Investigation. 1999; 104: 787-794 [4] The relationship of glycemic exposure (HbA1c) to the risk of development and progression of retinopathy in the diabetes control and complica tions trial. Diabetes. 1995; 44: 968-983 [5] Rother KI. Diabetes treatment--bridgi ng the divide. The New England Journal of Medicine. 2007; 356: 1499-1501 [6] Goodman LS, Gilman A, Brunton LL, Lazo JS, Parker KL. Goodman & Gilman's the Pharmacological Basis of Therapeutics. 11th edn. New York: McGraw-Hill, 2006. [7] Matveyenko AV. Liver Cells Listening to Beta Cells. American Diabetes Association 69th Scientific Sessi ons. New Orleans, 2009 [8] Guyton AC, Hall JE. Textbook of medical physiology. 11th edn. Philadelphia: Elsevier Saunders, 2006. [9] Cherrington AD, Sindelar D, Edgerton D, Steiner K, McGuinness OP. Physiological consequences of phasic insulin release in the normal animal. Diabetes. 2002; 51 Suppl 1: S103-108 [10] Wyne KL. Free fatty acids and type 2 di abetes mellitus. The American Journal of Medicine. 2003; 115 Suppl 8A: 29S-36S [11] Boden G, Lebed B, Schatz M, Homko C, Le mieux S. Effects of acute changes of plasma free fatty acids on intramyocellular fat content and insulin resistance in healthy subjects. Diabetes. 2001; 50: 1612-1617 [12] McGarry JD. Banting lecture 2001: dysregul ation of fatty acid meta bolism in the etiology of type 2 diabetes. Diabetes. 2002; 51: 7-18 [13] Polonsky KS, Given BD, Hirsch LJ, et al. Abnormal patterns of insulin secretion in noninsulin-dependent diabetes mellitus. The New England Journal of Medicine. 1988; 318: 1231-1239 [14] DeFronzo RA, Tobin JD, Andres R. Gluc ose clamp technique: a method for quantifying insulin secretion and resistance. The Am erican Journal of Physiology. 1979; 237: E214223

PAGE 143

143 [15] Lerner RL, Porte D, Jr. Relationship between intrave nous glucose loads, insulin responses and glucose disappearance rate. The Journal of Clin ical Endocrinology and Metabolism. 1971; 33: 409-417 [16] Porte D, Jr., Pupo AA. Insulin responses to glucose: evidence for a two pool system in man. The Journal of Clin ical Investigation. 1969; 48: 2309-2319 [17] Caumo A, Luzi L. First-phase insulin secre tion: does it exist in real life? Considerations on shape and function. American Journal of Physiology. 2004; 287: E371-385 [18] Grodsky GM. A threshold distribution hypothe sis for packet storage of insulin and its mathematical modeling. The Journal of clinical investigation. 1972; 51: 2047-2059 [19] Cerasi E. Potentiation of insulin release by glucose in man. I. Quantitative analysis of the enhancement of glucose-induced insulin secretion by pretreat ment with glucose in normal subjects. Acta Endocrinologica. 1975; 79: 483-501 [20] Mitrakou A, Kelley D, Mokan M, et al. Role of reduced suppression of glucose production and diminished early insulin releas e in impaired glucose tolerance. The New England Journal of Medicine. 1992; 326: 22-29 [21] Mitrakou A, Kelley D, Veneman T, et al. Contribution of abnormal muscle and liver glucose metabolism to postprandial hyperglycemia in NIDDM. Diabetes. 1990; 39: 13811390 [22] Shah P, Basu A, Basu R, Rizza R. Impact of lack of suppression of glucagon on glucose tolerance in humans. The Ameri canJournal of Physiology. 1999; 277: E283-290 [23] Ganda OP, Srikanta S, Brink SJ, et al. Differential sensitivity to beta-cell secretagogues in "early," type I diabetes mellitus. Diabetes. 1984; 33: 516-521 [24] Garvey WT, Revers RR, Kolterman OG, Rubenstein AH, Olefsky JM. Modulation of insulin secretion by insulin and glucose in type II diabetes mellitus. The Journal of Clinical Endocrinology and Metabolism. 1985; 60: 559-568 [25] Luzi L, DeFronzo RA. Effect of loss of first-phase insulin secretion on hepatic glucose production and tissue glucose disposal in human s. The American Journal of Physiology. 1989; 257: E241-246 [26] Del Prato S, Marchetti P, Bonadonna RC. Phasic insulin release and metabolic regulation in type 2 diabetes. Diabetes. 2002; 51 Suppl 1: S109-116 [27] Luzio SD, Owens DR, Vora J, Dolben J, Smith H. Intravenous insulin simulates early insulin peak and reduces post-prandial hyperg lycaemia/hyperinsulinaemia in type 2 (noninsulin-dependent) diabetes mellitus. Diabet es Research (Edinburgh, Scotland). 1991; 16: 63-67

PAGE 144

144 [28] Steiner KE, Mouton SM, Williams PE, Lacy WW, Cherrington AD. Relative importance of firstand second-phase insu lin secretion in glucose homeostasis in conscious dog. II. Effects on gluconeogenesis. Diabetes. 1986; 35: 776-784 [29] Guerci B, Sauvanet JP. Subcutaneous insu lin: pharmacokinetic va riability and glycemic variability. Diabetes & Metabolism. 2005; 31: 4S7-4S24 [30] Gualandi-Signorini AM, Giorgi G. Insulin formulations--a review. European Review for Medical and Pharmacological Sciences. 2001; 5: 73-83 [31] Hoffman A, Ziv E. Pharmacokinetic cons iderations of new insulin formulations and routes of administration. C linical Pharmacokinetics. 1997; 33: 285-301 [32] Bonora E, Muggeo M. Postprandial blood gl ucose as a risk factor for cardiovascular disease in Type II diabetes: the epidem iological evidence. Diabetologia. 2001; 44: 21072114 [33] Monnier L, Lapinski H, Colette C. Contributions of fa sting and postprandial plasma glucose increments to the overall diurnal hype rglycemia of type 2 diabetic patients: variations with increasing levels of HbA(1c). Diabetes Care. 2003; 26: 881-885 [34] Dimitriadis GD, Gerich JE. Importance of timing of preprandial subcutaneous insulin administration in the management of diabetes mellitus. Diabetes Care. 1983; 6: 374-377 [35] Lean ME, Ng LL, Tennison BR. Interval be tween insulin injection and eating in relation to blood glucose control in adul t diabetics. British Medical Jo urnal (Clinical research ed. 1985; 290: 105-108 [36] Heinemann L. Time-Action Profiles of Insu lin Preparations. Mainz, Germany: Kirchheim & Co GmbH, 2004. [37] Heinemann L, Woodworth, J. Pharmacoki netics and Glucodynamics of Insulin Lispro. Drugs of Today: Prous Science, 1998:23-36 [38] Actrapid 100 IU/mL, Solutino for Injection in a Vial, Summary of Product Characteristics. Novo Nordisk Limited, 2007 [39] Hirsch IB. Insulin analogues. The New England journal of medicine. 2005; 352: 174-183 [40] Brange J, Owens DR, Kang S, Volund A. M onomeric insulins and their experimental and clinical implications Diabetes Care. 1990; 13: 923-954 [41] Brange J, Ribel U, Hansen JF, et al. Monomeric insulins obtaine d by protein engineering and their medical implications. Nature. 1988; 333: 679-682 [42] Joslin EP, Kahn CR. Joslin's diabetes mellitus. 14th edn. Philadelphia: Lippincott Williams & Willkins, 2005.

PAGE 145

145 [43] Mitri J, Pittas AG. Inhaled insulin--what went wrong. Na ture Clinical Practice. 2009; 5: 24-25 [44] Pfizer Inc. Exubera Prescribing Information. 2006 [45] Heinemann L, Heise,Tim Review: Current status of the development of inhaled insulin The British Journal of Diabet es & Vascular Disease. 2004; 4: 295-301 [46] Boss AH, Yu, Wen, Ellarman, Karen. Prandial Insulin: Is Inhaled Enough? Drug Development Research. 2008; 69: 138-142 [47] DeFronzo RA, Binder C, Wahren J, Felig P, Ferrannini E, Faber OK. Sensitivity of insulin secretion to feedb ack inhibition by hyperinsulin aemia. Acta Endocrinologica. 1981; 98: 81-86 [48] Koivisto VA, DeFronzo R, Hendler R, Felig P. The difference in insulin sensitivity and metabolic response to acute exercise in trai ned and sedentary subj ects. Current Problems in Clinical Biochemistry. 1982; 11: 122-132 [49] DeFronzo RA, Hendler R, Christensen N. Stimulation of counterregulatory hormonal responses in diabetic man by a fall in glucose concentration. Diabetes. 1980; 29: 125-131 [50] Heinemann L, Linkeschova R, Rave K, Ho mpesch B, Sedlak M, Heise T. Time-action profile of the long-acting insu lin analog insulin glargine (HOE901) in comparison with those of NPH insulin and placebo. Diabetes Care. 2000; 23: 644-649 [51] Heise T, Heinemann L. Rapid and longacting analogues as an approach to improve insulin therapy: an evidence-based medicine assessment. Current Pharmaceutical Design. 2001; 7: 1303-1325 [52] Heise T, Weyer C, Serwas A, et al. Time-action profiles of novel premixed preparations of insulin lispro and NPL insulin. Diabetes Care. 1998; 21: 800-803 [53] Howey DC, Bowsher RR, Brunelle RL, Woodworth JR. [Lys(B28), Pro(B29)]-human insulin. A rapidly absorbed analogue of human insulin. Diabetes. 1994; 43: 396-402 [54] Yalow RS, Berson SA. Immunoassay of endoge nous plasma insulin in man. The Journal of Clinical Investigation. 1960; 39: 1157-1175 [55] Hooper SA. Peptides, Peptoids, and Prot eins. Cinncinnati, OH: Harvery Whitney Books, 1991. [56] Gopalakrishnan M, Suarez S, Hickey AJ, Gobburu JV. Popul ation pharmacokineticpharmacodynamic modeling of subcutaneous and pulmonary insulin in rats. Journal of Pharmacokinetics and Pharmacodynamics. 2005; 32: 485-500 [57] Landsdorfer C. PK/PD Mode ling of Glucose Clamp Effects of Inhaled Insulin. Tucson, AZ: American Conference on Phar macometrics, March 9-12, 2008

PAGE 146

146 [58] Osterberg O, Erichsen L, Ingwersen SH, Pl um A, Poulsen HE, Vicini P. Pharmacokinetic and pharmacodynamic properties of insulin aspart and human insulin. Journal of Pharmacokinetics and Pharmacodynamics. 2003; 30: 221-235 [59] Tornoe CW, Jacobsen JL, Madsen H. Grey-box pharmacokinetic/pharmacodynamic modelling of a euglycaemic clamp study. J ournal of Mathematical Biology. 2004; 48: 591-604 [60] Woodworth JR, Howey DC, Bowsher RR. Establishment of time-action profiles for regular and NPH insulin using pharmac odynamic modeling. Diabetes Care. 1994; 17: 6469 [61] Woodworth JR, Howey DC, Bowsher RR, Brunelle RL. Relating glucose clamp profiles to reduction of blood glucose after insulin admi nistration. Diabetes Technology & Therapeutics. 2004; 6: 147-153 [62] Sakagami M. Insulin disposition in the l ung following oral inhalati on in humans : a metaanalysis of its pharmacokinetics. Clinical Pharmacokinetics. 2004; 43: 539-552 [63] Duckworth WC, Bennett RG, Hamel FG. In sulin degradation: progress and potential. Endocrine Reviews. 1998; 19: 608-624 [64] Duckworth WC, Hamel FG, Peavy DE. Hepatic metabolism of insulin. The American Journal of Medicine. 1988; 85: 71-76 [65] Rabkin R, Ryan MP, Duckworth WC. The renal metabolism of insulin. Diabetologia. 1984; 27: 351-357 [66] Levy J, Olefsky JM. The effect of insulin concentration on retroendocytosis in isolated rat adipocytes. Endocrinology. 1987; 120: 450-456 [67] Hovorka R, Powrie JK, Smith GD, Sonksen PH, Carson ER, Jones RH. Fivecompartment model of insulin kinetics and its use to investigate action of chloroquine in NIDDM. The American Journa l of Physiology. 1993; 265: E162-175 [68] Meibohm B, Derendorf H. Basic c oncepts of pharmacokinetic/pharmacodynamic (PK/PD) modelling. Internati onal Journal of Clinical Ph armacology and Therapeutics. 1997; 35: 401-413 [69] Rave K, Potocka, E., Heinemann, L., Heise, T., Boss, A., Marino, M., Costello, D., Chen, R. Pharmacokinetics and linear e xposure of AFRESA compared with the subcutaneous injection of regular human insulin. Diabetes, Obesity and Metabolism. 2009: [70] Sheiner LB, Stanski DR, Vozeh S, M iller RD, Ham J. Simultaneous modeling of pharmacokinetics and pharmacodynamics: appl ication to d-tubocur arine. Clinical Pharmacology and Therapeutics. 1979; 25: 358-371

PAGE 147

147 [71] Landersdorfer C. Modeling Diverse Anti-Diab etic Drug Effects Using Indirect Response Models: Review. Population Approach Group in Europe. Dunedin, New Zealand: Abstracts of the Annual Meeting 2008 [72] Lin S, Chien YW. Pharmacokinetic-pha rmacodynamic modelling of insulin: comparison of indirect pharmacodynamic response with e ffect-compartment link models. The Journal of Pharmacy and Pharmacology. 2002; 54: 791-800 [73] Tolic IM, Mosekilde E, Sturis J. Modeling the insulin-glucose feedback system: the significance of pulsatile insu lin secretion. Journal of Theoretical Biology. 2000; 207: 361-375 [74] Berger M, Rodbard D. Computer simula tion of plasma insulin and glucose dynamics after subcutaneous insulin in jection. Diabetes Care. 1989; 12: 725-736 [75] Thomaseth K, Pavan A, Berria R, Glass L, DeFronzo R, Gastaldelli A. Model-based assessment of insulin sensitivity of glucos e disposal and endogenous glucose production from double-tracer oral glucose tolerance test. Computer Methods and Programs in Biomedicine. 2008; 89: 132-140 [76] Landersdorfer CB, Jusko WJ. Pharmacoki netic/pharmacodynamic modelling in diabetes mellitus. Clinical Pharmacokinetics. 2008; 47: 417-448 [77] Landersdorfer C. PK/PD Modeling of Gl ucose Clamp Effects of Inhaled Insulin. American Conference on Pharmacometrics Tucson, AZ, 2008 [78] Rave KM, Nosek L, de la Pena A, et al. Dose response of inhaled dry-powder insulin and dose equivalence to subcutaneous in sulin lispro. Diabetes Care. 2005; 28: 2400-2405 [79] Owens DR. Human insulin. Intermediate-ac ting insulin preparations initial ratio method. In: Owens DR, ed. Clinical Pharm acological Study in Normal Man: MTD Press Limited, 1986:138 [80] Muir KT GR. Non-compartmental analysis In: Bonate P HD, ed. Pharmacokinetics in Drug Development: Clinical Study Design a nd Analysis. Arlington, VA: American Association of Pharmaceutical Scientists, 2004:235-266 [81] Puckett WR, Lightfoot EN. A model for multiple subcutaneous insulin injections developed from individual diab etic patient data. The Ameri can Journal of Physiology. 1995; 269: E1115-1124 [82] Jonsson EN, Karlsson MO. Xpose--an S-PLUS based population pharmacokinetic/pharmacodynamic model building aid for NONMEM. Computer Methods and Programs in Biomedicine. 1999; 58: 51-64 [83] Steiner S, Pfutzner A, Wilson BR, Harzer O, Heinemann L, Rave K. Technosphere/Insulin--proof of concept st udy with a new insulin formulation for pulmonary delivery. Exp Clin Endocrinol Diabetes. 2002; 110: 17-21

PAGE 148

148 [84] Humalog Insulin Lispro Injection, U SP (rDNA origin), 100 units per mL (U-100). Prescribing Information: Eli Lilly and Company, 2009 [85] Nucci G, Cobelli C. Models of subcutaneous insulin kinetics. A crit ical review. Computer Methods and Programs in Biomedicine. 2000; 62: 249-257 [86] Richardson PC, Boss AH. Technosphere insulin technology. Diabetes Technology & Therapeutics. 2007; 9 Suppl 1: S65-72 [87] Leone-Bay A. Technosphere/Insulin: mimicking endogenous insulin release. In: Rathbone M HJ, Roberts M, Lane M, Eds, ed. Modified Release Drug Delivery Technology. 2 edn. New York, NY: Informa Healthcare USA, Inc., 2008 [88] Brain JD. Inhalation, deposition, and fate of insulin a nd other therapeutic proteins. Diabetes Technology & Therapeutics. 2007; 9 Suppl 1: S4-S15 [89] Gad SC. Drug safety evaluation. New York: J. Wiley, 2002:xi, 1007 [90] Sindelka G, Heinemann L, Berger M, Fr enck W, Chantelau E. Effect of insulin concentration, subcutaneous fat thickness a nd skin temperature on subcutaneous insulin absorption in healthy subj ects. Diabetologia. 1994; 37: 377-380 [91] Cassidy J, BAughman, R, Richardson, P. AFRESA (Technosphere Insulin) Dosage Strengths Are Interchangeable. American Di abetes Association Annual Meeting, New Orleans, LA: American Diabetes Association, 2009 [92] Bonate PL. Pharmacokinetic-Pharmacodynami c Modeling and Simulation. S.l.: Springer Science+Business Media, 2006. [93] Parke J, Holford NH, Charles BG. A procedure for generating bootstrap samples for the validation of nonlinear mixed-effects popul ation models. Computer Methods and Programs in Biomedicine. 1999; 59: 19-29 [94] Potocka E, Baughman, R., Derendorf, H. Population Pharmacokine tic Model of Regular Human Insulin Following Different Rout es of Administration. 2009 AAPS National Biotechnology Conference, Seattle, WA: Am erican Association of Pharmaceutical Scientists, June, 2009 [95] Bavenholm PN, Pigon J, Ostenson CG, Efendi c S. Insulin sensitivity of suppression of endogenous glucose production is the single most important determinant of glucose tolerance. Diabetes. 2001; 50: 1449-1454 [96] Rizza RA, Mandarino LJ, Gerich JE. Mechan ism and significance of insulin resistance in non-insulin-dependent diabetes mellitus. Diabetes. 1981; 30: 990-995

PAGE 149

149 BIOGRAPHICAL SKETCH Elizabeth Po tocka received her Bachelor of Sc ience degree in mathematics at Fairfield University in 1994. After graduation, Elizabeth joined Boehringer Ingelheim Pharmaceuticals, Inc, in Ridgefield, CT, in 1997, where she worked in the Department of Drug Metabolism and Pharmacokinetics. Her responsib ilities primarily incl uded interpretation of pharmacokinetic and pharmacodynamic data from the clinical program, as well as population modeling and simulation work in support of drug development. She join ed the University of Florida doctoral program under the supervision of Dr. Hartmut Derendorf in the Department of Pharmaceutics, College of Pharmacy, in August 2004. In 2007, Elizabeth joined MannKind Corporation as a Pharmacokineticist in the Experimental Pharma cology Department, where her responsibilities encompassed the design and interpretation of clinical studies for pharmacokinetics and pharmacodynamics as well as pharmacokinetic a nd pharmacodynamic modeling of insulin. She is a contributing author to a number of peer-review ed presentations and publications, as well as a contributing author to the Technosphere Insulin New Drug Application. She received her Ph.D. in pharmaceutical sciences from the University of Florida in August 2009, where her primary research focus included population pharmac okinetics and pharmacodynamics of insulin.