Population pharmacokinetic/pharmacodynamic (PK/PD) modeling of depot testosterone cypionate in healthy male subjects

University of Iowa Iowa Research Online Theses and Dissertations Summer 2016 Population pharmacokinetic/pharmacodynamic (PK/PD) modeling of depot testosterone cypionate in healthy male subjects Youwei Bi University of Iowa Copyright 2016 Youwei Bi This dissertation is available at Iowa Research Online: https://ir.uiowa.edu/etd/5716 Recommended Citation Bi, Youwei. "Population pharmacokinetic/pharmacodynamic (PK/PD) modeling of depot testosterone cypionate in healthy male subjects." PhD (Doctor of Philosophy) thesis, University of Iowa, 2016. https://ir.uiowa.edu/etd/5716. Follow this and additional works at: https://ir.uiowa.edu/etd Part of the Pharmacy and Pharmaceutical Sciences Commons

POPULATION PHARMACOKINETIC/PHARMACODYNAMIC (PK/PD) MODELING OF DEPOT TESTOSTERONE CYPIONATE IN HEALTHY MALE SUBJECTS by Youwei Bi A thesis submitted in partial fulfillment of the requirements for the Doctor of Philosophy degree in Pharmacy in the Graduate College of The University of Iowa August 2016 Thesis Supervisor: Associate Professor Daryl J. Murry

Graduate College The University of Iowa Iowa City, Iowa CERTIFICATE OF APPROVAL PH.D. THESIS This is to certify that the Ph.D. thesis of Youwei Bi has been approved by the Examining Committee for the thesis requirement for the Doctor of Philosophy degree in Pharmacy at the August 2016 graduation. Thesis Committee: Daryl J. Murry, Thesis Supervisor Lawrence L. Fleckenstein Gary Milavetz Nicole K. Brogden Brian J. Smith

ACKNOWLEDGEMENTS Foremost, I would like to express my sincere gratitude to my advisor Dr. Daryl J. Murry for his continuous support for my Ph.D. study and research, for his patient guidance, inspiration, motivation and extensive knowledge. I want to thank him for providing me with various research projects, fellowship training at FDA and valuable advice on my longterm career development. Besides my advisor, I want to thank my dissertation committee members, Dr. Lawrence L. Fleckenstein, Dr. Gary Milavetz, Dr. Nicole K. Brogden and Dr. Brian J. Smith for their helpful advice, insightful comments and encouragement. My sincere gratitude also goes to Dr. Guohua An for offering me research and publication opportunity in PBPK model of gefitinib in mice and scale-up to humans. I am thankful to all the current and previous labmates and fellow colleagues: Dr. Xiaofeng Wang, Dr. Carrie Ann Morris, Dr. Amal Ayyoub, Yu Jiang, Nattawut Leelakanok, Megan Kelchen and Ronlida Raymond D Cunha. I really appreciate their friendship, helping hands, advice and all the fun we had in the past 5 years. I want to express my deepest thanks to my parents Lvhai Bi, Yongfeng Wang for their unconditional love, care, support and understanding throughout my life. I would like to give my heartfelt thanks to my beloved wife, Wenjing Han for trusting me, encouraging me, believing in me and loving me. I also want to thank my parents-in-law, Yaocai Han and Jinping Xu for their trust and love. Last but not least, I want to thank College of Pharmacy, College of Public Health and University of Iowa for the excellent coursework, research training and financial support. ii

ABSTRACT Depot intramuscularly administered testosterone cypionate (TC) is indicated for treatment of hypogonadism in males. However, illegal use of TC and other anabolic steroids in athletic competition has been occurring for over 50 years. A randomized threearm clinical trial was conducted to investigate long-term abuse effects of TC. The objective of the thesis is to apply modeling approaches to characterize the pharmacokinetics of long-term TC injections and identify side effects in healthy male volunteers. Given the known inhibitory effect of testosterone on HPG axis, an indirect effect model was applied to describe the suppression of luteinizing hormone and spermatogenesis. A polynomial change point mixed effects model was developed to describe the change in weight and lipid profiles after weekly injection of testosterone cypionate. A linear one-compartment model with first-order absorption best described the concentration-time profile of testosterone obtained from 31 healthy males. The population clearance estimates for total and free testosterone were 2.42*10 3 and 6.03*10 5 L/day, respectively. Weight, albumin and their changes from baseline were identified as significant covariates for total testosterone. The estimated potency of total testosterone with respect to LH suppression was 9.38ng/ml. Model simulation showed that suppression of luteinizing hormone and spermatogenesis after TC injection was more severe and of greater duration at the highest testosterone dose level. Model simulation showed that both 250mg and 500mg would result in an average increase of body weight of 3.5kg at 8 weeks after dosing. Total lipids decreased after TC administration however there was no statistically significant difference in the lipid change between three dose groups, which precludes any definite conclusion on the effect of long-term TC administration on lipid profiles. iii

PUBLIC ABSTRACT Depot intramuscularly administered testosterone cypionate (TC) is indicated for treatment of hypogonadism in males. However, illegal use of TC and other anabolic steroids in athletic competition has been occurring for over 50 years. A randomized threearm clinical trial was conducted to investigate long-term abuse effects of TC. The objective of the thesis is to apply modeling approaches to characterize the pharmacokinetics of long-term TC injections and identify side effects in healthy male volunteers. Given the known inhibitory effect of testosterone on HPG axis, an indirect effect model was applied to describe the suppression of luteinizing hormone and spermatogenesis. A polynomial change point mixed effects model was developed to describe the change in weight and lipid profiles after weekly injection of testosterone cypionate. A linear one-compartment model with first-order absorption best described the concentration-time profile of testosterone obtained from 31 healthy males. The population clearance estimates for total and free testosterone were 2.42*10 3 and 6.03*10 5 L/day, respectively. Weight, albumin and their changes from baseline were identified as significant covariates for total testosterone. The estimated potency of total testosterone with respect to LH suppression was 9.38ng/ml. Model simulation showed that suppression of luteinizing hormone and spermatogenesis after TC injection was more severe and of greater duration at the highest testosterone dose level. Model simulation showed that both 250mg and 500mg would result in an average increase of body weight of 3.5kg at 8 weeks after dosing. Total lipids decreased after TC administration however there was no statistically significant difference in the lipid change between three dose groups, which precludes any definite conclusion on the effect of long-term TC administration on lipid profiles. iv

TABLE OF CONTENTS LIST OF TABLES ix LIST OF FIGURES...x LIST OF ABBREVIATONS......xiii CHAPTER I INDICTION AND ABUSE OF TESTOSTERONE THERAPY 1 1.1 Testosterone Therapy... 1 1.1.1 Male Hypogonadism... 1 1.1.2 Testosterone replacement therapy: General Information... 2 1.2 Pharmacology of testosterone preparations... 2 1.3 Current Testosterone Therapies... 3 1.3.1 Implants... 3 1.3.2 Oral administration... 4 1.3.3 Intramuscular injections... 5 1.3.4 Transdermal applications... 7 1.3.4.1 Transdermal patches... 7 1.3.4.2 Transdermal gels... 8 1.4 Illegal use of steroids outside its indications... 10 1.4.1 Abuse of anabolic steroids... 10 1.4.2 Effect of anabolic steroids on healthy body... 11 1.4.2.1 Limitations of research on the effect of AAS on human body... 11 1.4.2.2 Body composition... 11 1.4.2.3 Adverse effects... 11 CHAPTER II REVIEW OF POPULATION PHARMACOKINETIC ANALYSES 15 2.1 Background and Concept... 15 2.2 Population Pharmacokinetics... 16 2.2.1 Nonlinear Mixed Effects Modeling (NLME)... 17 v

2.2.1.1 Introduction... 17 2.2.1.2 Key components... 18 2.2.1.3 Covariate Analysis... 20 2.3 Thesis overview and objective... 22 CHAPTER III POPULATION PHARMACOKINETICS OF TESTOSTERONE CYPIONATE IN HEALTHY MALE VOLUNTEERS........24 3.1 Pharmacokinetic properties of Testosterone cypionate... 24 3.1.1 Absorption... 26 3.1.2 Distribution... 25 3.1.3 Metabolism... 25 3.1.4 Excretion... 25 3.2 Objective... 26 3.3 Study Design... 26 3.3.1 Subjects Enrollment and Randomization... 26 3.3.2 Inclusion and Exclusion Criteria... 27 3.3.3 Chronological Design... 27 3.3.4 Study medication... 28 3.3.5 Sampling Schedule... 28 3.3.6 Assay of quantification... 28 3.3.7 Monitoring of subjects... 29 3.3.8 Study Summary... 29 3.4 Methods... 30 3.4.1 Population Pharmacokinetic Analysis... 30 3.4.2 Base Pharmacokinetic model building... 30 3.4.3 Covariate model development... 31 3.4.4 Model fit evaluation... 33 3.4.5 Monte Carlo simulation... 33 3.5 Results... 34 vi

3.5.1 Subjects... 34 3.5.2 Model development... 34 3.6 Discussion... 38 3.7 Conclusions... 43 CHAPTER IV EFFECT OF TESTOSTERONE CYPIONATE ON LH HORMONE AND SPERMATOGENESIS..62 4.1 Objective... 62 4.2 Methods... 62 4.2.1 Subjects Enrollment and Randomization... 62 4.2.2 Inclusion and Exclusion Criteria... 63 4.2.3 Chronological Design... 63 4.2.4 LHRH stimulation test... 63 4.2.5 Assessment of spermatogenesis... 64 4.2.6 Population Pharmacodynamic model building... 64 4.3 Results... 65 4.3.1 PD model for suppression of LH... 65 4.3.2 PD model for suppression of spermatogenesis... 67 4.3 Discussion... 69 4.4 Conclusions... 72 CHAPTER V EFFECT OF TESTOSTERONE CYPIONATE ON BODY WEIGHT AND LIPID PROFILES..95 5.1 Introduction... 95 5.2 Objective... 96 5.3 Methods... 96 5.3.1 Safety monitoring and lab measurements... 96 5.3.2 Model building... 97 vii

5.4 Results... 97 5.4.1 Change in weight... 97 5.4.2 Change in lipid profiles... 99 5.5 Discussion... 100 5.6 Conclusions... 101 APPENDIX NONMEM CONTROL FILE FOR THE FINAL MODEL....115 1 Final PK model for total testosterone (Chapter III)... 115 2 Final PD model for LH suppression (Chapter IV)... 118 3 Final PD model for suppression of spermatogenesis (Chapter IV)... 121 4 Polynomial change point model for change in weight (Chapter V)... 123 5 Polynomial change point model for change in triglycerides (Chapter V)... 125 REFERENCES...127 viii

LIST OF TABLES Table 1. Half-lives of different preparation of testosterone formulations... 13 Table 2. Baseline demographic characteristics, protein levels and laboratory test values of 31 subjects... 44 Table 3. Population pharmacokinetic estimates and bootstrap results for total testosterone in 31 volunteers.... 45 Table 4. Population pharmacokinetic estimates and bootstrap results for free testosterone in 31 volunteers.... 46 Table 5. Population pharmacodynamic estimates and bootstrap results of LH suppression following long-term dosing of testosterone cypionate.... 73 Table 6. Point estimates of level of sperm production compared to baseline (0-100%) at each time point for three dose groups by the indirect response PD model. 100% means same to baseline level, 0% means total inhibition.... 74 Table 7. Model estimates of polynomial change point mixed model for change in weight from baseline.... 103 Table 8. Model estimates of polynomial change point mixed model for change in lipid profiles from baseline.... 104 ix

LIST OF FIGURES Figure 1. Molecular structure of testosterone and its esters... 14 Figure 2. Components of a nonlinear mixed effects model... 23 Figure 3. Overview of testosterone metabolism... 47 Figure 4. Goodness-of-fit plots of total testosterone concentrations..... 48 Figure 5. Predicted and observed individual concentration-time profiles of total testosterone in 100mg, 250mg and 500mg... 49 Figure 6. Prediction-corrected visual predictive check for total testosterone obtained from 1000 simulations stratified on sampling period... 52 Figure 7. Scatterplot of NPDE versus time and population predictions of total testosterone....53 Figure 8. Goodness-of-fit plots of free testosterone concentrations.... 54 Figure 9. Predicted and observed individual concentration-time profiles of free testosterone in 100mg, 250mg and 500mg... 55 Figure 10. Prediction-corrected visual predictive check for free testosterone obtained from 1000 simulations stratified on sampling period... 58 Figure 11. Scatterplot of NPDE versus time and population predictions of free testosterone... 59 Figure 12. Simulated median total testosterone concentration versus study time based on 1000 simulations... 60 Figure 13. Effects of significant covariates on clearance and volume of distribution in total testosterone... 61 Figure 14. Goodness-of-fit plots of LH hormone..... 75 x

Figure 15. Predicted and observed individual concentration-time profiles of LH hormone in 100mg, 250mg and 500mg.... 76 Figure 16. Visual predictive check for LH obtained from 1000 simulations... 79 Figure 17. Median suppression level of LH in 100mg, 250mg and 500mg TC from 1000 simulations... 80 Figure 18. Variability of LH suppression in 100mg,250mg and 500mg TC from 1000 simulations... 81 Figure 19. Effects of significant covariates on pharmacodynamic parameters of suppression of luteinizing hormone... 82 Figure 20. Median and variability of suppression level of LH stratified on 4 quartiles of baseline weight obtained from 1000 simulations... 83 Figure 21. Goodness-of-fit plots of sperm counts... 84 Figure 22. Goodness-of-fit plots of sperm motility... 85 Figure 23. Predicted and observed individual concentration-time profiles of sperm counts in 100mg, 250mg and 500mg... 86 Figure 24. Predicted and observed individual concentration-time profiles of sperm motility in 100mg, 250mg and 500mg... 89 Figure 25. Visual predictive check for sperm counts obtained from 1000 simulations... 92 Figure 26. Visual predictive check for sperm motility obtained from 1000 simulations... 93 Figure 27. Predicted level of suppression of spermatogenesis over time during and after TC dosing based on 1000 simulations... 94 Figure 28. Goodness-of-fit plots of change in weight... 105 Figure 29. Predicted and observed individual concentration-time profiles of change in weight in 100mg, 250mg and 500mg... 106 xi

Figure 30. Visual predictive check for change in weight obtained from 1000 simulations... 109 Figure 31. Predicted change in weight over time during and after TC dosing based on 1000 simulations... 110 Figure 32. Plots of Observations versus individual predictions for triglycerides, HDL, LDL, cholesterol and HCT... 111 Figure 33. Visual predictive check for change in lipid profiles (Triglycerides, HDL, LDL, Cholesterol, HCT) obtained from 1000 simulations... 112 Figure 34. Average predicted change in lipid profiles (triglycerides, HDL, LDL, cholesterol and HCT) over time during and after TC dosing based on 1000 simulations.. 113 Figure 35. Average predicted change of triglycerides stratified on 4 quartiles of baseline level during and after TC dosing... 114 xii

LIST OF ABBREVIATIONS AAS ADME ALT AST AUC BCOV BLA BMI CCOV CL C max CL/F COV CV CWRES FDA FOCE FSH GGT HDL HPG HPLC IM IMP IMPMAP INH IOV IPRED Anabolic androgenic steroids Absorption, Distribution, Metabolism, Excretion Alanine Aminotransferase Aspartate Aminotransferase Area under the curve Baseline Covariate Biological License Application Body mass index Change in Covariate from baseline Absolute clear ance Maximum concentration Apparent clearance Covariate Coefficient of variation Conditionally weighted residual Food and Drug Administration First order conditional estimation Follicle-stimulation hormone Gamma-Glutamyl Transpeptidase High-density lipoprotein Hypothalamic-pituitary-gonadal High performance liquid chromatography Intramascular injection Importance sampling Importance sampling assisted by mode a posteriori Inhibition Inter-occasion variability Individual prediction xiii

IRB IWRES KA LDL LH LHRH LOQ LRT MLE NDA NLME NPDE OFV PCVPC PD PK PPK PRED PSN SCM SAEM SHBG TBG TC TE T max V d V d /F VPC Institutional review board Individual weighted residual Absorption rate Low-density lipoprotein Luteinizing hormone Luteinizing hormone releasing hormone Limit of quantification Likelihood ratio test Maximum likelihood estimation New Drug Application Nonlinear Mixed Effects Modeling Normalized prediction distribution errors Objective function value Prediction-corrected visual predictive check Pharmacodynamics Pharmacokinetics Population Pharmacokinetics Population prediction Perl-speaks-NONMEM Stepwise covariate model Stochastic approximation expectation-maximization Sex hormone-binding globulin Thyroxine-binding globulin Testosterone cypionate Testosterone enanthate Time of maximum concentration Absolute volume of distribution Apparent volume of distribution Visual predictive check xiv

CHAPTER I INDICTION AND ABUSE OF TESTOSTERONE THERAPY 1.1 Testosterone Therapy 1.1.1 Male Hypogonadism Hypogonadism is a clinical syndrome complex defined by low testosterone or low sperm production. (1) It is caused by hypothalamic, pituitary, or testicular organ disorders either congenital or acquired after birth. It is categorized as primary and secondary hypogonadism depending on whether pathology occurs at the testicular (primary) or pituitary-hypothalamic (secondary) level. Aging is the most relevant factor associated with hypogonadism. Vermeulen and Kaufman estimate that 7% of men between the ages of 40 to 60 years old have total testosterone concentrations lower than normal and that this percentage increases to 35% for men older than 80 years of age. (2) A number of co-morbid conditions besides aging may also be associated with a decreased testosterone concentration. These include diabetes mellitus, myocardial infarction, chronic heart failure, chronic renal failure, AIDS, and liver cirrhosis among others. (3) Male hypogonadism manifests as many symptoms reflecting the physiological functions of decreased testosterone in men. The most reported symptoms include decreased libido, decreased strength and endurance, erectile dysfunction and mood fluctuations. (4) Total serum testosterone status is the most widely used lab measurement for the diagnosis of hypogonadism in males. Although it is ideal to directly measure free testosterone concentration, it is difficult to measure free testosterone in the clinical setting. Therefore it is advisable to measure total testosterone concentration and the sex hormone-binding globulin (SHBG) to calculate the free testosterone. Since serum testosterone has a diurnal variation and peak levels occur between 8:00 a.m. and 10:00 1

a.m. (5), the serum sample should be obtained during this period. The general clinical guideline is that a patient can benefit from testosterone treatment if they have a total testosterone level below 3 ng/ml or a free testosterone level below 65 pg/ml. As the percentage of older age population in developed countries is increasing, hypogonadism or testosterone deficiency will become a more commonly diagnosed disorder in middleaged and older men. At present testosterone deficiency is underdiagnosed in most countries and many patients are unaware of their disease and go untreated. (5) 1.1.2 Testosterone replacement therapy: General Information Testosterone replacement therapy has been used to treat male hypogonadism for more than 6 decades; the goal of therapy is to restore the patients testosterone concentration to the normal range. Testosterone replacement therapy has been shown to alleviate the symptoms associated with testosterone deficiency without serious side effects or safety concerns, thereby allowing patients to regain their quality of life. (6, 7) 1.2 Pharmacology of testosterone preparations Testosterone, like all other androgens, has the basic structure of androstane, which consists of three cyclohexane, one cyclopentane ring and a methyl group at positions 10 and 13. The chemical structure of testosterone is characterized by an oxy group in position 3, a hydroxyl group in position 17 and a double bond in position 4 (8) (Figure 1) Oral administration of testosterone itself is not therapeutically effective as 90% of the hormone is metabolized and inactivated in the liver before it reaches the target organs. After the first-pass metabolism the half-life of testosterone is less than 30 minutes. (9) Therefore modifications to the testosterone therapy have been used alone or in combinations to increase the amount of testosterone that can reach the systemic circulation and target site, three most common approaches are listed below: 1) Alternative routes of administration 2

2) Esterification at position 17 3) Chemical modification of testosterone Different routes of administration of testosterone preparation will be discussed in the following sections in this chapter. 1.3 Current Testosterone Therapies According to the WHO Guidelines for Use of Androgens in Men (1992) (8), the ideal properties of testosterone treatment should include: safety, efficacy, value of money, convenient administration, a good release profile, flexibility of dosing and normalization of testosterone levels. Since testosterone was first isolated and synthesized in the 1930s, numerous routes of delivery have been used in testosterone therapy over the last 6 decades, including intramuscular injections, transdermal systems and buccal preparations of testosterone. The selection of the formulation should be decided by both the patient and physician based on patient preference, age, type of hypogonadism and co-morbid conditions. The therapeutic goal of testosterone supplementation is to maintain a serum testosterone level in the middle to lower normal testosterone range, but the target level is usually higher for older males to have a therapeutic effect. The pharmacokinetic properties for these testosterone therapies vary significantly, it is essential to understand the pharmacokinetic behavior of each drug product and their differences for all the formulations in order to achieve optimal therapeutic effects. This chapter focuses on the review of the pharmacokinetic behaviors of different routes of delivery and its general tolerability among the patients with hypogonadism. 1.3.1 Implants T-pellets are the oldest form of testosterone replacement therapy and were developed in the 1940s, however the early pellets were cast, broke easily and had a 3

short duration of action. Later, new pellets were developed around 1980s. These new pellets were fused at elevated temperatures resulting in improved stability and prolonged activity.(10) The pharmacokinetics of 6 new subcutaneous testosterone pellets, each containing 200mg of fused crystalline testosterone, have been evaluated in multiple randomized clinical trials at durations of at least half a year (10, 11). An initial short burst release of testosterone was observed around 12 hours after implantation with peak total concentration of 11.5 ±1.1 ng/ml. Physiological levels were maintained until day 63, serum testosterone levels gradually declined thereafter and was close to baseline values on day 300. The testosterone elimination profile was characterized by an apparent terminal half-life of 70.8 ± 10.7 days. (10) The absorption of testosterone from pellets closely approximated zero-order release kinetics and exhibited an absorption half-life of 74.7 days. The bioavailability of testosterone pellets was near complete by day 189 (95% ± 0.84%). (10) However, the insertion of implants under the abdominal skin requires minor surgery, and 11% patients experience side effects including extrusions (8.5%), bleeding (2.3%) and infection episodes (0.6%) (12). The overall rate of adverse events was related to the number of implants (P=0.031) and levels of physical activities at work (P=0.030). But there were no significant effects of age, weight, body mass index (BMI) or any relationship to the site of implantation, wound dressing and diagnosis (12). 1.3.2 Oral administration Oral administration of unmodified testosterone would appear to be the first choice for testosterone replacement therapy as it is very convenient. However, although unmodified testosterone is absorbed well from the gut it goes through extensive firstpass metabolism and is inactivated before it reaches systemic circulation. In hypogonadal men with normal liver function, 400-600mg doses of unmodified testosterone must be taken through oral administration, a dose almost 100 fold higher 4

than the normal production of testosterone in a healthy male (8). The oral administration of testosterone is simply not safe, practical or economical and therefore has not become a generally accepted method for therapeutic dosing. Several attempts have been made to modify testosterone in order to improve its absorption and delay metabolism by liver. When testosterone is esterified with an undecanoic acid at the 17β-position, its route of absorption from gut is preferentially shifted to the lymphatic system and absorption is improved (13, 14). Pharmacokinetics of testosterone undecanoate were tested in 12 normal men and 8 hypogonadal men before and at hourly intervals after the single-dose oral administration. (15) On average maximum testosterone levels were observed five hours after administration. High interindividual variability was observed in the time when maximum concentrations were reached, as well as in the maximum concentrations, which ranged from 4.9 to 11.5 ng/ml. Testosterone concentrations declined to baseline levels 4h after T max in the normal men and 2h in the patients. This study showed that this modification still suffers from high inter-individual variability of oral bioavailability, serum levels and a short halflife. Similar pharmacokinetic behavior was observed in a two-month multiple-dose study of testosterone undecanoate. (16) Based on these observations, even with 3 capsules per day, only short-lived testosterone peaks with substantial variability can be obtained. A ten-year safety study of oral testosterone undecanoate showed that upon annual measurements of 33 patients there is no increased hepatic metabolism over time. Digital examination of the prostate did not reveal any sign of prostate tumor, and no major adverse events were noted due to testosterone treatment. (17, 18) However, other modifications like 17α methyltesterone and fluoxymesterone were found to be associated with hepatic toxicity and therefore were withdrawn from the market (19, 20). 1.3.3 Intramuscular injections Intramuscular injections of testosterone esters are the most widely used form of testosterone replacement therapy, in part due to the convenience and that it is 5

relatively inexpensive. While unmodified testosterone requires multiple injections a day due to its very short half-life, the esterification of testosterone at position 17 with acid increases the lipid solubility, prolongs its half-life in the body. This depot effect has been shown to be proportional to the length of the ester side chain (21), therefore testosterone esters with longer acid chains have longer biological half-lives. After absorption from the IM depot, the testosterone ester is hydrolysed in plasma, which is much faster than the release form the injection site. The metabolism of the testosterone ester to unmodified testosterone also occurs rapidly so that the Intravenous pharmacokinetic profiles of testosterone enanthate and testosterone are nearly idential (22). There are a number of testosterone esters available in the market. The half-life and dosing intervals of different preparations of testosterone formulations are summarized in table 1. Testosterone concentrations following IM injection of some short-acting preparations will surge up to several-fold above the normal testosterone upper limit due to the initial burst, but will decrease to sub-physiological level rapidly after cessation of dosing. In clinical practice, short and long-acting esters have been combined in order to overcome this shortcoming. (23) Schurmeyer and Nieschlag noted that a single IM testosterone enanthate (TE) 200 mg dose equivalent to 140 mg of free testosterone administered to normal males produced supra-physiologic testosterone levels in serum and saliva as early as 2 hours following injection, reaching peak levels 4 to 5 times above basal values between 8 and 24 hours post dose.(24) Nine days after injection serum and salivary levels of testosterone returned to baseline. In a male contraception study, 33 healthy adult men aged 21-41 years were administered 200 mg IM injection of TE weekly for up to 18 months. (25) Compared to baseline, pre-injection levels of total testosterone increased 2.5-fold, reaching steady state values after approximately 12 weeks of treatment. Peak 6

concentrations of total testosterone increased by 5-fold, and free and non-shbg-bound testosterone levels increased 10-fold after 16 weeks. Overall the deep IM injection of testosterone esters is generally safe and well tolerated with infrequent dosing schedules and satisfactory therapeutic effects, although it may cause patient discomfort and local pain during the injection (26). 1.3.4 Transdermal applications 1.3.4.1 Transdermal patches Since skin easily absorbs androgen and other drugs, the transdermal application of testosterone therapy is commonly utilized. It has been shown that the scrotum has the highest testosterone absorption, which is about 40-fold higher than the forearm (27). People have utilized this advantage to develop transdermal patches for administration on scrotal skin. These patches loaded with 10 or 15mg testosterone were attached to between 40 to 60 cm 2 scrotal skin area to provide hypogonadal men with sufficient amounts of testosterone to maintain normal physiological hormone level (28, 29). Long-term therapy with daily administration of scrotal patches in 11 men diagnosed with hypogonadism produced steady-state serum levels of testosterone within the normal range without significant adverse side effects up to 10 years (30). However, it is not very convenient for the daily use of these transdermal patches on the scrotal skin as it requires hair clipping or shaving to optimize the effect and adherence. The use of transdermal patches on the non-scrotal skin requires enhancers to facilitate testosterone passage through skin. The pharmacokinetics, efficacy and safety of a testosterone transdermal patch were compared with IM injections of testosterone for the treatment of male hypogonadism in a 24-week multicenter, randomized study. (31) Mean morning testosterone concentrations were within the normal range in greater proportion of patients in the testosterone patch group (range, 77-100%) than the IM injection group (range, 19-84%). With the permeation enhancer testosterone patches can deliver 2.5mg testosterone per day and achieve the normal physiologic 7

serum testosterone concentration, and no excessive stimulation of endogenous testosterone production was observed. Skin irritation from the patches was reported by 60% of patients. The most common adverse events were pruritus (27.3%), local skin reactions (18.2%), and allergic contact dermatitis at the application site (6.0%). The irritation typically resolved within 1 or 2 days after patch removal, and only 9% of patients discontinued treatment due to skin reactions. However, in a clinical audit of the acceptability and efficacy of the transdermal therapy for male hypogonadism in the UK, about 84% patients experienced adverse effects in the transdermal therapy, only 22% of the patients chose to continue with the testosterone patches. These patches were judged to be too noisy, visually indiscreet, embarrassing, unpleasant to apply and socially unacceptable (32). Overall, the inconvenience of scrotal-skin patches, the poor local tolerability and obtrusive nature of the non-genital patches limits its applicability and acceptability for many hypogonadal patients. 1.3.4.2 Transdermal gels Testosterone gel is another application of transdermal delivery; it was approved for clinical use in hypogonadism in 2000. The gel is applied to the torso and upper arms with the dose titriated for each individual. The gel usually dries in less than 5 minutes without a visible residue on the skin. The daily application of these testosterone gels increased serum testosterone concentrations into the normal range within one hour and resulted in steady-state serum concentrations 4-5 fold above baseline for the duration of gel application (33). Long-term pharmacokinetic profiles of 2 doses of testosterone gel (50 and 100mg) were evaluated in 227 hypogonadal men. (34) The dose was titrated up (50mg to 75mg) or down (100mg to 75mg) if the serum testosterone concentrations were outside of the adult normal range after 90 days of treatment. Mean serum steady state testosterone concentrations remained stable over 180 days of daily gel application and were higher in 8

the 100mg group. The dosing regimen of gel application was further investigated in order to most effectively maintain the serum testosterone concentrations within the normal reference range of 3-11.4 ng/ml in treated patients (35). The 3g/2% dose of daily testosterone gel application was found to result in serum testosterone in the normal range for most hypogonadal men. The two most common prescription topical gel formulations are AndroGel (testoseterone gel 1.62%) and Testim (testosterone gel 1%). The pharmacokinetic profiles of these two gel formulations were compared in a randomized crossover study (36). C max and AUC 0 24 estimates for total testosterone were greater following the application of Testim compared to AndroGel. The confidence intervals for C max and AUC 0 24 were not wholly contained within the bioequivalence limits and they are not bioequivalent with Testitim offering higher serum testosterone levels and greater bioavailability than AndroGel. Testosterone gel has been shown to be safe and well tolerated in patients. Longterm treatment with testosterone gel has a good safety profile with respect to the prostate and hematocrit levels. (37, 38) A similar percentage of serious treatmentrelated adverse events occurred in the testosterone gel (2.1%) and placebo groups (2.5%) after daily application of gel to 274 men for 182 days. (39) The advantages of transdermal gels over patches are superior local tolerability, convenience of usage and the ability to deliver testosterone dose-dependently. In studies comparing testosterone gel to the patch, the incidence of dermatological reactions was much higher in the testosterone patch group. Skin irritation was reported in only 5.5% of subjects treated with Androgel compared to 66% of those treated with testosterone patch (Androderm) in a 6-month study of 227 hypogonal patients (33). Lack of skin irritation was further confirmed in a longer follow-up study up to 42 months. Mild local skin irritation occurred in 12 subjects among 163 hypogonal men, resulting in discontinuation in only 9

one subject (38). Similar safety profiles confirmed these observations in both shortterm and long-term studies of Testim. (37, 40, 41) A concern for testosterone gel is the interpersonal transfer of drug during skinto-skin contact. Both Androgel and Testim have the potential to raise testosterone concentrations in individuals after skin contact with the treated patient even several hours after application. (42) The potential risks of secondary exposure are more harmful in children between 9 months to 5 years. Reported adverse events include enlargement of the genitalia, advanced bone age and aggressive behavior. (42) After observing and evaluating these risks, the US Food and Drug Administration (FDA) has demanded a black box to be added in the instructions of the testosterone gel application regarding the potential interpersonal skin-to-skin transfer. (43) However, even with the risk of interpersonal transfer, superior local tolerability and dose flexibility still make testosterone gel highly desirable over other testosterone formulations for treatment of low testosterone. As a result, testosterone gel has gained a significant market share in both Europe and United States although its price is much higher than the IM injection. 1.4 Illegal use of steroids outside its indications 1.4.1 Abuse of anabolic steroids Anabolic androgenic steroids (AAS) are synthetic derivatives of testosterone. Aside from the therapeutic indications discussed in the previous chapter, the illegal use of AAS by athletes to gain an unfair advantage in athletic competitions has been reported for over 50 years. (44-46) The abuse of anabolic steroids by healthy people is of great medical concern, and numerous studies have shown that short-term use (several weeks at a time) or excessive dosing of anabolic steroids can lead to serious, sometimes irreversible, health risks. (47-50). 10

1.4.2 Effect of anabolic steroids on healthy body 1.4.2.1 Limitations of research on the effect of AAS on human body The dose of testosterone required for treatment of low testosterone is substantially lower than the dose used illegally by athletes to improve performance. Since it is not ethical to expose healthy volunteers to supra-therapeutic dosages of testosterone for the sole purpose of studying its effects, there is a paucity of literature regarding the pharmacokinetics and toxicities in this setting. Moreover, most studies evaluating abuse effects of androgen steroids are observational studies and don t met the golden standard quality of randomized, double-blind placebo-controlled study design. The untoward effects of AAS on healthy body in the literature therefore may underestimate the effects in real life. 1.4.2.2 Body composition The most common reason that AAS are used illegally is due to their effect on body composition. Several studies have shown that bodyweight may increase by 2 to 5 kg after a short-term use (<3 months) of AAS.(51-53) This increase in bodyweight is mainly attributed to the increase in lean body weight, with no change in fat mass observed. (52, 54, 55) 1.4.2.3 Adverse effects The use of AAS also will have negative effects on health status. Many studies have been conducted to investigate the side effects of AAS on the human body. The most common self-reported side effects include increased sexual drive, occurrence of acne, increased body hair and more aggressive behaviors. (56, 57) Testosterone is a very important hormone in hypothalamic-pituitary-gonadal (HPG) axis. Testosterone secretion stimulates the production of gonadotropins, luteinizing hormone (LH) and follicle-stimulation hormone (FSH). Ingestion of exogenous testosterone disturbs the secretion of endogenous testosterone, LH and FSH. 11

Suppression of gonadotropin production can then cause testicular atrophy and reduce the semen counts and quality. Numerous studies have shown that AAS may dramatically reduce the sperm motility and counts per unit within 24 hours of dosing. (25, 58-60) Long-term ingestion of supra-therapeutic doses of AAS could potentially lead to infertility. (60-62) After cessation of AAS dosing, spermatogenesis usually will recover to the normal level, but the exact time of recovery may vary depending on the dosing regimen. In a study of 6-month long-term AAS administration the recovery of spermatogenesis took between 4 to 5 months. (58, 63) The association between abuse of AAS and cardiovascular risks has been equivocal. In several studies an elevation of blood pressure has been observed after administration of high doses of AAS, (52, 64-66) however in other studies no alteration in BP has been observed. (67-69) Low high-density lipoprotein (HDL) level has been recognized as an independent risk factor for cardiovascular disease. (70, 71) Risk for coronary artery disease increases sharply as HDL levels fall below 40mg/dL.(72) Several studies have shown serum levels of HDL were remarkably reduced after AAS administration. (73-75) Conversely, the administration of AAS has been shown to increase the serum level of low-density lipoprotein (LDL). (64, 65, 76) 12

Table 1. Half-lives of different preparations of testosterone formulations Testosterone derivatives Route of Terminal elimination Dosing Regimen Administration half-life Testosterone undecanoate (oral) Oral 150 minutes 2-4 capsules of 40mg per day Testosterone propionate IM injection 0.8 days 3 days Testosterone enanthate IM injection 4.5 days 2 weeks Testosterone cypionate IM injection 7.3 days 2 weeks Testosterone buciclate IM injection 29.5 days 8-12 weeks Testosterone IM injection 23.7 days 8-12 weeks undecanoate Testosterone patch Transdermal 10-100 minutes 1 or 2 patches per day Testosterone gel Transdermal 10-100 minutes 5 to 10 g gel per day 13

Figure 1. Molecular structure of testosterone and its esters Testosterone Testosterone undecanoate Testosterone propionate Testosterone enanthate Testosterone cypionate 14

CHAPTER II REVIEW OF POPULATION PHARMACOKINETIC ANALYSES 2.1 Background and Concept Pharmacometrics is the emerging science that quantifies drug, disease and trial information to aid efficient drug development and regulatory decisions. (77) It uses a mathematical model to quantify the relationship between xenobiotics and humans based on principles of pharmacology, physiology and disease pathophysiology. These mathematical models include drug models (pharmacokinetic and pharmacodynamics models), disease models and trial models. Drug models are the most typical pharmacometric analyses, they are also referred to terms as: concentration-effect, dose-response and PK-PD relationships. (78) Pharmacometric analyses have become an increasingly important component of New Drug Application (NDA) and Biological License Application (BLA) submissions to the US FDA to support labeling, study design and regulatory approval decisions. (79) In the recent survey of NDAs/BLAs submitted to FDA from 2000 to 2008, pharmacometric analyses influenced more than 60% of the submissions in various therapeutic areas including cardiovascular, neurology, anti-viral, rheumatology diseases and etc. The number of submissions where pharmacometric analyses were included has also grown over time: it increased by 6-fold over 9 years, from 45 submissions in 5 years (2000 to 2004) to 87 submissions in 2 years (2007-2008). The recent survey of NDAs/BLAs submitted to FDA showed that pharmacometric analyses mainly contributed to four areas; optimization for dosing in the labeling; Evaluation of doses for future drug development; Improvement of clinical trial design; and provision of evidence of efficacy.(79) The most significant strength of pharmacometric analyses lies in its ability to integrate knowledge across different disciplines and programs to aid drug development and/or regulatory decisions. 2.2 Population Pharmacokinetics 15

Population pharmacokinetic analyses are the most frequent analyses (57%) among pharmacometric analyses submitted by the sponsor to US FDA between January 2000 and December 2008. (79) It is defined as the study of the sources and correlates of variability in drug concentrations among individuals who are the target patient population receiving clinically relevant doses of a drug of interest.(80) The overall purpose of population pharmacokinetics is to identify variability in patient demographical, pathophysiological and response characteristics which will alter doseconcentration relationships and adjust dosage accordingly if such change may lead to clinically meaningful shifts in therapeutic index. (80) Population pharmacokinetic analyses have several advantages compared to traditional approaches. It allows for analysis and interpretation of pharmacokinetic parameters from sparse data, dense data, or a combination of both. In late stage (Phase 2/3) drug development, samples obtained per subject are usually limited because of ethical and/or medical reasons. The PPK analyses allow one to integrate heterogeneous data types obtained from varying sources, such as data from studies with unbalanced design, data from different phases or centers, or data from various patient populations (e.g. elderly, pediatric population). (81) The population approach can also provide more precise estimates of variability, such as inter-individual, inter-occasion and residual variability than those estimated in the traditional methods. (81) There are at least two levels of hierarchy in the population model framework. At the individual level, a specified structure model is given to estimate the mean and variance of the individual-specific parameters. All the residual errors of observations from an individual are assumed to follow a normal distribution. At the population level, a more complicated statistical model is required to estimate the measurement error and inter-subject variability respectively. Simply put, a population model is a collection of individual models; while it assumes the structural pharmacokinetic model is qualitatively same for all individuals, the specific pharmacokinetic parameters such as clearance and 16

volume of distribution vary among individuals. In the population model, individual pharmacokinetic parameters are often described by a function of typical population parameter and individual specific covariates. Between-subject variability is usually added to account for the random variation of individual parameter vectors around the population prediction. (81) Two common approaches have been employed to perform population pharmacokinetic analysis: the two-stage approach and nonlinear mixed effects modeling. The two-stage approach involves estimation of pharmacokinetic parameters through nonlinear regression using individual concentration-time data at first stage. These individual parameters will be the input data for calculation of mean, variance and covariance for summary statistics of relevant pharmacokinetic parameters at second stage. Random effects, variance and covariance, are likely to be biased in the two-stage approach. (82-84) In this chapter the main focus will be on nonlinear mixed effects modeling approach. 2.2.1 Nonlinear Mixed Effects Modeling (NLME) 2.2.1.1 Introduction Nonlienar mixed-effects modeling approaches (NLME) consisted of 92% of all population pharmacokinetic methodologies published from 2002 to 2004. (85) It was first introduced by Lewis B. Sheiner, a clinician, and Stuart Beal, a statistician. In the 1970 s they successfully utilized NLME approach in dose optimization for long-term anticoagulation therapy with warfarin and digoxin. (86) In their several papers they showed convincing evidence that NLME approach could adequately describe the pharmacokinetics of digoxin and address the effect of renal impairment on drug s elimination from sparse data where only 2 to 3 samples are available per patient. (82-84, 87, 88) These pivotal publications laid the foundation for population pharmacokinetics. The software used to analyze digoxin pharmacokinetics was later extended to modeling software NONMEM. (89) Currently NONMEM is the most widely used software to 17

conduct pharmacometric analyses in both industry and regulatory agencies and for majority of the pharmacometricians is the functional gold standard (86, 90-93), despite NLME approach is also available in other software such as MONOLIX, ADAPT 5, Phoenix NLME and R. 2.2.1.2 Key components In general, NLME includes two components: fixed effects and random effects. (Figure 2) Fixed effects consist of population estimates of pharmacokinetic parameters describing the underlying structural model or the effects of covariates on model parameters. It does not vary across subjects. Random effects describe the variation between typical values, which usually consist of inter-subject variability, inter-occasion variability and residual variability. The variability in pharmacokinetics is known to vary over time. Estimation of inter-occasion variability is one approach to account for this variability due to unknown or unobserved covariate. Failure to account IOV may lead to biased parameter estimates and inflated residual variability. (94) The remaining variability, which cannot be explained by the fixed effect or random effect model, is lumped into the residual error model. Residual variability takes into account the model misspecification and measurement errors. The most common residual error models are additive error models, proportional error models and combination of both. The residual error model describes the difference between individual predictions and individual observations, which can be shown in the following equation: Y obs = Y IPRED ε prop + ε add both ε prop and ε add are assumed to follow normal distributions with mean of 0 and variance of σ 2. The mathematical expression for structural model in NLME can be expressed in the following equation. 18

Y ij = f ij (θ, x ij, η i ) where y ij is the jth observation of the ith patient; f is a function describing the pharmacokinetic profile of the compound of interest; θ is the vector of fixed effect parameters which usually include population estimate of pharmacokinetic parameters; x ij is a vector of known values such as dose and time; η i is the vector of random effect parameters describing the variability of individual parameters around typical population estimate. The η i estimates are assumed to be normally distributed with the mean of 0 and variance of ω 2. Individual parameters θ i can be described from different random effect models: 1 Additive model: θ i = θ + η i 2 Proportional model: θ i = θ (1 + η i ) 3 Exponential model: θ i = θ exp (η i ) Since most pharmacokinetic parameters are based on physiology, the exponential model is the most common model to constrain individual parameters θ i to be always positive. In NLME, maximum likelihood estimation (MLE) is used to estimate the pharmacokinetic parameters. The pharmacokinetic parameters are derived to maximize likelihood of joint probability of observations under the proposed population model. Because of the nonlinear dependence of observations on parameters η and ε in the random effects, even the simplest mixed-effect problems are analytically intractable. Various algorithms were developed to solve this problem numerically. First-order conditional estimation with interaction is the most widely used estimation method in NLME, it has the advantage of providing highly reproducible values and is computationally rapid for simple PK model. (95) Other estimation methods, such as Monte Carlo importance sampling (IMP), importance sampling assisted by mode a posteriori (IMPMAP), stochastic approximation expectation-maximization (SAEM), are 19

more complicated and computationally intensive, but tend to be more stable and faster for complex PK/PD model. (95) 2.2.1.3 Covariate Analysis One of the biggest strengths in population pharmacokinetic lies in its ability to identify covariates that can explain random variability in pharmacokinetic parameters and response between individuals. The relationship between pharmacokinetic parameters and covariates such as sex, bodyweight or genetic polymorphism are quantified in the covariate model. A covariate model can help understand the causes of variability and allow for better therapeutic use and better study design of future trials. Common parameterization to explore the relationship between PK parameters and covariates include linear model, power model and exponential model. For continuous covariates, covariates are often centered to improve the stability of estimation. These models can be described by: 1 Linear model : θ i = θ 1 + θ 2 (COV COV median ) 2 Power model : θ i = θ 1 (COV/COV median ) θ 2 3 Exponential model: θ 3 = θ 1 exp (θ 2 (COV COV median )) For categorical covariates, these models can be described by: 1 Linear model : θ i = θ 1 + θ 2 COV COV 2 Power model : θ i = θ 1 θ 2 3 Exponential model: θ 3 = θ 1 exp (θ 2 COV) All included covariates should have physiological plausibility. There are in general two common approaches in building a covariate model: stepwise selection and full covariate approach. Stepwise selection consists of forward addition and backward elimination. In the forward addition step, all covariate-parameter relationships with different parameterization were screened univariately using likelihood ratio test. The covariate relationship which resulted in largest decline in OFV was added to the base model. 20

These steps are continued until no significant covariates are left. Backward elimination step starts with the final model from forward addition step, covariates are removed one at a time in turn. The covariate that is not significant and has the smallest impact on the model will be dropped. These steps are repeated until all non-significant covariates are removed from the final model. Stepwise selection approach requires a large number of model runs. Luckily automated stepwise selection model building procedure is available in the SCM procedure in Perl-speaks-NONMEM. (96) There are some disadvantages present in the stepwise selection procedure: P-values are difficult to adjust and interpret for multiple-comparisons; The presence of collinear or correlated predictors may cause the confidence intervals to be falsely narrow; Statistically significant effects are not necessarily equal to clinically meaningful effect, likewise lack of statistical significance does not necessarily indicate lack of effect. The full covariate approach is another popular alternative to test the significant covariates on pharmacokinetic parameters. After defining a stable base model, it adds all pre-specified covariates into the base model and estimates their effects. The prespecified covariates are selected based on scientific or clinical interest, mechanistic plausibility or prior knowledge about the relationship between covariates and parameters. The inferences are based on the point estimates and confidence intervals or the posterior distribution of estimated covariate effects. (SEs, bootstrap, Bayes) (97) For example, a covariate is claimed to be clinically important if its interival or posterior distribution of the estimated covariate effect result in change in relevant parameter greater than 20%. After testing all pre-specified covariates those covariate relationships that are not statistically and clinically significant can be dropped from the model. The full covariate approach had several advantages compared to stepwise selection: the covariate selection is not subject to the choice of p-values and problems of shrinkage; It helps to understand the reasons why some covariates have no impact; The point and interval estimates can be used to assess the clinical importance of 21

specified covariates. The caveats for this approach are that it requires rational selection of potential covariates to test in the full model, and re-parameterization is often necessary to stabilize the model. 2.3 Thesis overview and objective The overall objective of this thesis is to apply sophisticated modeling approaches to characterize the pharmacokinetic profile of long-term dosing of testosterone cypionate in healthy males and evaluate its pharmacodynamic effects on various physiological functions. In chapter 3, we develop a population pharmacokinetic model to describe and predict the concentration-time profile of total and free testosterone after long-term dosing of TC. Through developing a PPK model for testosterone cypionate we will further explore and identify clinically relevant covariates associated with its pharmacokinetic variability. The specific aim of chapter 4 is to develop a PK-PD model to describe and predict the suppression and subsequent recovery of luteinizing hormone and spermatogenesis following TC dosing. The PK-PD model will help with the mechanistic understanding of testosterone action and the underlying regulatory action of HPG axis. In the last chapter, we use a population modeling approach to fit the longitudinal change of body weight and lipid profiles and evaluate how they are affected by the longterm abuse of testosterone injection. 22

Figure 2. Components of a nonlinear mixed effects model. 23

CHAPTER III POPULATION PHARMACOKINETICS OF TESTOSTERONE CYPIONATE IN HEALTHY MALE VOLUNTEERS 3.1 Pharmacokinetic properties of Testosterone cypionate 3.1.1 Absorption IM injection generally results in lower but more sustained plasma concentrations than IV injection, partly because a depot" will form in the muscle tissue and acts as a repository for the drug when a solution is injected. The rate of absorption of IM injection depends partly on physiological factors such as depth of injection and local blood flow supply. (98, 99) It is also affected by many formulation factors. Aqueous solutions usually provide the fastest absorption rate into the systematic circulation. Drug can be formulated as suspensions or emulsions to slow the absorption rate. It can also be modified to the salt form to take advantage of the slow dissolution rate and low solubility. Absorption form IM injection is a complex process and we need to take into account the various physiological and formulation factors when evaluating its absorption profile. (98, 99) Esterification of testosterone at position 17 increases the lipid solubility of molecule and prolongs its activity. Following IM injection in an oily solution, testosterone cypionate is slowly absorbed into the systematic circulation and then rapidly hydrolysed to testosterone in plasma. Pharmacokinetics of a single injection of 200mg testosterone cypionate was studied in a randomized cross-over study of 6 healthy young males. Mean serum testosterone concentrations increased dramatically to 3 times the baseline levels at 1 day and then decreased gradually afterwards. The testosterone concentration returned to baseline level by day 10. (8) Similar results were observed in a clinical study treating low testosterone in 11 hypogonadal males aged 28 to 74 years of age. A single 24

IM injection of 200mg TC caused a threefold increase in mean serum testosterones at days 2 to 5 (1108 ± 440 ng/dl) and a slow decline to basal by days 13 to 14. (100) 3.1.2 Distribution Testosterone is highly protein-bound endogenous hormone. About 69% of the circulating testosterone is permanently bound to sex hormone binding globulin (SHBG) and is not biologically active. Another 30 percent of testosterone is weakly bound to albumin; the last 1 percent of testosterone is free and biologically active. (101) Weakly albumin-bound testosterone could easily dissociate and be absorbed by the tissues along with the free testosterone. The sum of these two is refereed as bioavailable testosterone. (23) 3.1.3 Metabolism Testosterone is converted to two important hormones, dihydrotestosterone and estradiol, through two different pathways. It is reduced to dihydrotestosterone in specific tissues like skin and prostate. The oxidation to estradiol mainly takes place in adipose tissue and about 20% in testes in men. (Figure 3) (102) The metabolism of testosterone and dihydrotestosterone takes place in 90% in the liver. Its extraction ratio is relatively high. Human CYP3A4 is one of the most important and abundant oxidizing enzyme in the liver. It accounts for about 40% of the total CYP in liver microsomes. CYP3A4 is the major metabolizing enzyme for various xenobiotics and drugs such as simvastatin, alprazolam, calcium channel blockers. (103, 104) In addition, it is also responsible for the metabolism of endogenous compounds such as testosterone. (105, 106) 3.1.4 Excretion After reduction and oxidization in the liver, about 90% of testosterone given intramuscularly is excreted in the urine as glucuronic and sulfuric acid conjugates and its metabolites; another 6% is excreted in the feces, mostly in the unconjugated form. (107) 25

MacIndoe and coworkers conducted a randomized double-blind study examining the long-term effects of weekly injections of testosterone cypionate (TC) upon the suppression of the hypothalamic-pituitary-gonadal axis and its subsequent recovery in healthy male subjects. (108) Testosterone half-lives were calculated in 29 male patients who had at least three free and total testosterone samples each collected between day 2 through day 14 following the last TC injection. There was no difference in the mean value of testosterone elimination half-life between 100 mg (6.9 ± 1.3 days), 250 mg (7.3 ± 2.9 days) and 500 mg (6.7 ± 2.7 days) doses. The overall mean exogenous testosterone elimination half-life was 6.9 ± 2.3 days. Thus approximately 46 days or 6 to 7 weeks were required for all subjects to reach steady state after the start of the injection protocol. 3.2 Objective To date there is no population pharmacokinetic model available for the exogenous administration of testosterone ester. We evaluated the pharmacokinetics of testosterone as part of a study conducted by MacIndoe and coworkers. The main objective of the present analysis in this chapter was to (i) develop a population pharmacokinetic model to adequately describe and predict the concentration-time profile of total and free testosterone after long-term dosing of TC (ii) to explore and identify clinically relevant covariates associated with its pharmacokinetic variability. The PPK model analysis will help with the mechanistic understanding of PK characteristics of testosterone and the underlying regulatory action of HPG axis. 3.3 Study Design 3.3.1 Subjects Enrollment and Randomization A cohort of healthy male volunteers between the ages of 18 and 40 years old was recruited. The study was approved by the institutional review board (IRB) of the 26

University of Iowa College of Medicine and review committee for the Clinical Research Center at the University of Iowa Hospitals and Clinics. All study subjects reviewed the study summary and signed written informed consent. Subjects were aware they would have a one-third chance of receiving one of the study doses. They were also aware they would receive weekly injections of testosterone cypionate (TC) and placebo each for 14 weeks but were not informed of the sequence. The diet in this study was not controlled considering the long-term nature of the study. All subjects were on their own diets. 3.3.2 Inclusion and Exclusion Criteria Subjects received a physical health examination including electrocardiogram, blood tests and a psychiatric evaluation prior to the beginning of the study. Subjects were excluded if they had any of the following: baseline blood pressure greater than 140 systolic or greater than 90 diastolic; any chronic medical condition requiring ongoing assessment and treatment like diabetes, hypothyroidism, peptic ulcer disease; significant biochemical abnormality in liver function tests, serum cholesterol, and serum HDL cholesterol at entry; the presence of an abnormal entry semenogram, anemia; a renal function or electrolyte abnormality; evidence of recent mood, anxiety, substance abuse or psychotic disorder. In addition, potential subjects completed the Buss-Durkee Hostility Inventory and subjects with total scores greater than 30 for the total and greater than 15 on the aggression subscale on this measure were excluded from study participation (44). 3.3.3 Chronological Design The chronological design of the study included two consecutive weekly injections of an identically appearing TC placebo, followed by 14 consecutive weekly injections of testosterone cypionate. 14 weeks were chosen to mimic the common cycling duration observed by illicit steroid users in the community. Following the last weekly injection of active agent the subjects received 12 consecutive weeks of TC placebo injections. (46) 27

Pituitary gonadotropin secretion was evaluated by conducting luteinizing hormone releasing hormone (LHRH) stimulation tests biweekly 7 days after the last injection of either placebo or testosterone. Venous blood samples for LH and FSH were obtained before, 30, 60, and 120 min after the intravenous infusion of 100 mcg LHRH.(109) 3.3.4 Study medication Depo -testosterone (testosterone cypionate) (200 mg/ml) was the proprietary product utilized for the study. It is a white or creamy crystalline power, odorless and stable in the air. It is insoluble in water but freely soluble in alcohol, ether and vegetable oils. The diluent (0.2 ml benzyl benzoate, 9.45 mg benzyl alcohol in 560 mg of cottonseed oil/ml) was prepared by the Pharmaceutical Services division of University of Iowa College of Pharmacy. The final dosage form was a 2.5 ml IM injection containing the correct amount of TC in cottonseed oil (TC, weeks 2-15) or 2.5 ml IM injection of cottonseed oil (Placebo, weeks 1, 2, and 16-27). 3.3.5 Sampling Schedule All blood samples were collected between 7 AM and 9 AM to minimize the potential chronotropic variations in endogenous testosterone values. Serum total (tt) and free testosterone (ft) were sampled at biweekly intervals 7 days following the last injection. The elimination of tt and ft at each TC dose were assessed from hormone values measured in samples drawn at immediately post dose, 8h and then daily for 26 consecutive days after the last active TC injection. Not all samples at planned visits were available because of conflicts in the subjects schedule. However, at least 3 samples are available per person between day 2 and day 14 after last dose to allow for the calculation of pharmacokinetic parameters. 3.3.6 Assay of quantification All samples were stored at -20 C until assay. Serum LH, FSH, tt and ft were determined with Coat-A-Count radioimmunoassay kits. (110) The method was tested to 28

have reliable results in screening blood and urine for drugs of abuse with superior costeffectiveness. (111) The lower limits of detection were 4 ng/dl, 0.015 ng/dl, 0.15 miu/ml, and 0.10 miu/ml for total testosterone, free testosterone, LH, and FSH, respectively. For the free testosterone assay the inter-assay and intra-assay coefficients of variation were 11.2% and 5.5%. The mean and standard deviations of free testosterone quality control samples were 0.19 ± 0.02 ng/dl, 1.37 ± 0.12 ng/dl, and 3.27 ± 0.49 ng/dl. For total testosterone the mean and standard deviations for the quality control samples were 86 ± 9.9 ng/dl, 571 ± 40.8 ng/dl, and 1152 ± 50.4 ng/dl. Active metabolites dihydrotestosterone and estradiol were not measured and quantified in this study. 3.3.7 Monitoring of subjects Subjects were monitored for potential medical and psychiatric problems that would trigger study discontinuation for safety reasons. Vital signs were obtained at each visit. General chemistry profiles including liver function tests, hematocrits, and fasting lipid profiles were accessed biweekly throughout the study. Urine drug screens were obtained at baseline, week 1, 8, 24, and 28. All subjects were examined by a physician and study endocrinologist at week 16 and week 28. Biweekly psychiatric monitoring and ratings were performed by the study psychiatrist for prevention of the development of significant psychopathology. 3.3.8 Study Summary As mentioned in the section 1.4.2.1, there are several common limitations in research of effects of AAS on human body; the dosing regimen used in most studies is far below daily practices in gym or in athletic competitions. Only few studies met the golden standard quality of randomized, double-blind, placebo-controlled study design. In almost every study subjects were exposed to AAS for a short period. None of these common limitations are present in the current study; this is a randomized double-blind 29

clinical trial in which 31 patients were randomized into 3 dose groups. The high dosing regimens 250 mg/week and 500 mg/week resemble the dose illegally used by bodybuilders and athletes. Subjects were exposed to weekly injections of TC for 14 weeks consecutively, which is the common cycle duration observed by illicit steroid users in the community. (44) These features enable the current study to accurately estimate the long-term effects of AAS on human body. 3.4 Methods 3.4.1 Population Pharmacokinetic Analysis Nonlinear mixed effects modeling was performed using NONMEM version 7.2 (ICON Development Solutions, Ellicott City, MD). The first-order conditional estimation (FOCE) method was used throughout the Population PK model building with subroutine ADVAN6. NONMEM outputs were processed using Pirana (112) and Xpose version 4.5.3 (Uppsala University, Uppsala, Sweden) (113). R version 3.0.1 (Free Software Foundation, Vienna, Austria) (114) was used for statistical analysis and final report. Model selection was based on the following criteria: physiological understanding of the HPG axis, plausibility and precision of model estimates, goodness of fit plots, likelihood ratio test (LRT), successful convergence and model stability. A model was a better fit to the observation if the objective function value decreased 3.84, which corresponds to p-value of 0.05 in a chi-square distribution with degree of freedom equal to 1. 3.4.2 Base Pharmacokinetic model building One-compartment or two-compartment models were evaluated as the pharmacokinetic model for weekly IM injection of testosterone cypionate based on the concentration-time profile and pharmaceutical property of the compound. Different absorption models including mixed zero and first-order, dual depot model, two part rate 30

model (parallel/sequential K0KA) (115) and Michaelis-Menten models have been fit to the model since TC as an oil suspension is believed to have a more complex absorption profile than a first-order process. A lag time parameter was also tested to describe the possible absorption delays. It was believed after ingestion of exogenous testosterone administration, endogenous testosterone secretion would be suppressed as a result of regulation by the feedback loop from HPG axis. Several attempts including positive feedback of LH and negative feedback of testosterone itself were tried to describe the inhibition of secretion of endogenous testosterone. Inter-individual variability was modeled by assuming individual pharmacokinetic parameters followed log-normal distribution with mean equal to population estimate. Different residual error models were also tried to fit the data. 3.4.3 Covariate model development After the optimal base model was determined, covariate models were conducted to screen potential clinically relevant covariates on pharmacokinetic parameters. In the current analysis, stepwise selection was chosen to conduct direct covariate testing. Potential covariates were selected based on the physiological plausibility and prior knowledge about the relationship between covariate and parameter. Continuous covariates available for screening and testing included subject demographic characteristics (weight and age), renal function indicators (blood urea nitrogen, serum creatinine, creatinine clearance and bun-to-creatinine ratio), protein concentration levels (albumin, globulin, total protein, thyroxine-binding globulin (TBG) and sex hormone-binding globulin) and laboratory test values (bilirubin, alkaline phosphatase, lactate dehydrogenase, AST, ALT, GGT, triglicerides, cholesterol, HDL and LDL). The height of subjects was not available at the time of analysis, therefore LBW and BMI cannot be calculated for the covariate analysis. 31

They were first screened using visual inspection of scatterplot of parameter or IIV (ETA values) versus covariates. Then direct covariate testing was conducted in the model using a stepwise selection procedure. Linear, exponential and power models were tested for all continuous and categorical covariates. A decrease in OFV of 3.84 (P<0.05) was used as the cutoff value to include a covariate in stepwise forward addition. During backward elimination, for a covariate to remain in the model, an increase of OFV at least 10.83 (P<0.001) was needed. Covariates and its parameterizations with PK parameters were automatically selected using SCM procedure available in the PSN. (96) Several covariates such as weight, serum creatinine and albumin concentration changed over time during the study period. Thus we separated the original covariates into baseline values and its change over time and tested its effects on the pharmacokinetic parameters separately. The general equation with a power model for baseline covariate and a linear model for change in covariate from baseline was (116) shown in the following equation: P = θ p (BCOV/BCOV median ) θ BCOV (1 + CCOV θ CCOV ). where θ p represents the typical value of parameter P in the population,bcov is the baseline value of the covariate and CCOV is the difference from baseline at each sampling time. θ BCOV describes the fractional change in typical value P with difference between BCOV and the median baseline value BCOV median in the sample. θ DCOV describes the effect of change in covariate from baseline on the typical value. The stepwise selection procedure identified covariates that had statistically significant effects on the pharmacokinetic parameters. To assess clinical relevance of the covariate effects, we evaluated the 95% CI for covariate effects obtained by 1000 simulation from multivariate normal distribution with mean and variance-covariance matrix equal to estimates in the final model. The interval of 0.8 to 1.25 was selected to indicate a clinically relevant effect of a covariate 32

on a parameter, if the interval or posterior distribution of covariate effect was outside of this range then the covariate was considered to be clinically relevant. The number and percentage of missing covariates were also recorded and previous measurement of the covariate was used to fill in the missing value assuming the change was small during testosterone administration. 3.4.4 Model fit evaluation The model was evaluated by the following goodness-of-fit plots, including plots of observations versus population predictions, observations versus individual plots, conditionally weighted residuals versus time and absolute of individual weighted residuals versus individual predictions. The model stability and precision of final model parameters were evaluated in non-parametric bootstrap analysis of 1000 samples with replacement. The predictive performance of the final model was evaluated using prediction-corrected visual predictive check (PCVPC) with 1000 simulations. 5% percentile, median and 95% percentile of observations was then compared with 90% prediction intervals of the corresponding percentile obtained from the simulation. The percentage of observations outside of 90% prediction interval for each percentile was also calculated to assess the model predictive performance. In addition, normalized prediction distribution errors (NPDE) developed by Mentré F (117) were computed to evaluate the PK model. 1000 simulations were conducted beforehand in NONMEM as the validation dataset. Under the null hypothesis, if the validation dataset is large enough, the NPDE should follow the N(0,1) distribution. NPDE was also plotted against time and predicted testosterone concentrations, no trend should be seen if model adequately describes the validation dataset. 3.4.5 Monte Carlo simulation Monte Carlo simulation is the generation of random objects from certain probability distributions to obtain many quantities of interest using methods of 33

statistical inference. (118) Monte Carlo simulations have been used widely in the evaluation of drug effect, assessment of dose-response relationship and design and optimization of dosing strategy. (119-121) The simulations can reflect the expected range of variability in response under the assumptions of the proposed model. (118) In the current analysis, Monte Carlo simulations were conducted in NONMEM. Population estimates of fixed effects and standard errors (SE) from random effects in the final model were fixed in the simulation mode in NONMEM. Pseudo-random numbers from uniform distribution are generated in NONMEM as the seed for Monte Carlo simulations. The median and its 95% CI obtained from 1000 simulation replicates were used to summarize the results and interpret the question of interest. 3.5 Results 3.5.1 Subjects A total of 31 healthy male subjects were randomized into the 100 mg (n=10), 250 mg (n=10), and 500 mg (n=11) dose groups. A total of 726 free testosterone and 729 total testosterone serum samples were available from these healthy male subjects. Among these observations 298 free and 299 total testosterone samples were obtained within 26 days from the last dose, the main pharmacokinetic parameters were determined from these observations. Baseline demographic characteristics and laboratory tests for all subjects are summarized in Table 2. The ratio of total to free testosterone was found to be constant during the study periods. 3.5.2 Model development The data was best fit using one-compartment model with linear elimination and first-order absorption. After cessation of TC administration, total and free testosterone concentrations were lower than pretreatment levels temporarily and then slowly returned back to pre-dose level. The low testosterone concentrations were believed to 34

result from the decrease of endogenous testosterone secretion regulated by the feedback loop from HPG axis. The inhibition of endogenous testosterone was modeled with constant reduction of its secretion rate in pre-specified interval from day 28 to day 133. The objective function value for free and total testosterone PK model decreased 118 and 90 after accounting for the inhibition of endogenous testosterone secretion accordingly. A statistically significant improvement of fit was also observed when the inhibition of testosterone secretion was modeled to be different across three dose levels. The differential equation describing the testosterone pharmacokinetics is derived from the principles of mass balance. The rate of change of testosterone is equal to the rate of secretion and absorption minus the rate of elimination, as shown below: d(abs) = ka ABS d(t) d(te) d(t) = β i + ka ABS CL V Te Where d(x)/d(y) represents the derivative of x at y, Te is amount of testosterone in the body, β i symbolizes the testosterone secretion rate, basic pharmacokinetic parameters are CL the systemic clearance, V the central volume of distribution and Ka the first-order absorption rate. The total testosterone secretion rate β i is equal to β base in normal situation. When LH concentration is inhibited, the secretion of testosterone is estimated to be β decrease for three dosing regimens. It is well known that in men testosterone concentrations exhibit a 24-hour cycle, with maximal values at 4-6 a.m., dropping later in the day and evening by 15-70% and then rising again in the night. (122) However, since there was only one sample drawn per day at most, and all samples were collected from 7 a.m. to 9a.m. to minimize the potential variations of testosterone. The 24-hour variation of testosterone has not been incorporated into the model. 35

The mean and precision of final pharmacokinetic parameters for total testosterone were shown in Table 3. The CL/F and V/F for total testosterone were estimated to be 2.42 (kl/day) and 16.8 kl in typical healthy male with median covariate values. Half-life was calculated to be 4.81 days for total testosterone. The estimated secretion rate for endogenous testosterone in baseline were 11.9 mg/day in the PK model, and was estimated to decrease to 6.72, 3.01 and 2.64 mg/day after administration of 100mg, 250mg and 500mg doses of TC, respectively. For the remaining variability that cannot be explained by the covariate analysis, a combined additive and proportional model was applied to model the residual error. Baseline weight, baseline albumin and change in albumin from baseline are significant covariates for total testosterone. Power models for baseline weight and albumin, linear model for change in albumin and exponential model for change in weight result in lowest objective function values. The equations for covariate models for total testosterone are described below: Baseline albumin on Vd : Valbumin = ( Albumin θ 15 ) 4.6 Change in albumin from baseline on Vd : Vcalbumin = (1 + θ 14 (Calbumin + 0.2)) Weight on Vd : Weight on CL : V wt = ( WT 85 ) θ12 CL wt = ( WT 85 ) θ11 Change in weight from baseline on CL : 36

CL cwt = Exp(θ 13 (CWT 2.9)) The inter-individual variability of V d /F and CL/F decreased from 32.2% and 14.1% to 28.0% and 9.5% after including all significant covariates into the model. For significant covariates weight, albumin and globulin, 30(6.05%), 9(1.81%) and 10(2.02%) out of 496 observations were missing and replaced by the previous measurements. The covariance between different pharmacokinetic parameters was found not significant and therefore not included in the final model. The goodness-of-fit plots of final PK models for total testosterone are shown in Figure 4. Individual model predictions in whole period and intensive sampling period are shown for three dosing regimens separately in Figure 5. The smoothing lines of observations versus predictions are in line with the identity line, no obvious trend is shown in the plot of CWRES versus Time and IWRES versus IPRED. Bootstrap estimates and its 95% confidence intervals are provided in Tables 2. All the model estimates fell within the 95% CI of bootstrap estimates, indicating good precision in model parameter estimation. Due to different dosing regimens utilized in the study, the prediction-corrected visual predictive check (PC-VPC) instead of standard VPC was used. PC-VPC was stratified on sampling period to better evaluate the predictive performance of model (Figure 6). The 5 th, 50 th and 95 th percentiles of observations are all well within the 95% CI of same percentiles obtained from 1000 simulations, indicating good predictive performance for the final model. Numerical predictive check also shows that false positive rate and false negative rate for intensive sampling periods are 1.67% and 4.01% respectively. For the test of normalized prediction distribution of errors (NPDE), the t-test failed to reject that the mean of NPDE is significantly different from 0 in total testosterone, and no obvious trend was seen in the scatter plot of NPDE versus time and predicted testosterone concentrations (Figure 7). 37

The PK model developed for free testosterone is very similar to the total testosterone model. The mean and precision of final pharmacokinetic parameters for free testosterone are shown in Table 4. Baseline weight and change in globulin from baseline are significant covariates for free testosterone. The goodness-of-fit plots, individual model predictions, PC-VPC for free testosterone and NPDE diagnostic plots are shown in Figures 8 to 11. 3.6 Discussion The pharmacokinetics of total and free testosterone was satisfactorily described by one-compartmental model with first-order absorption. Initially different absorption models such as mixed zero-order and first-order were tried since we expect a sustained absorption profile from an IM injection, however all modeling attempts failed expect first-order absorption model, mainly because we have only 1-2 samples available per subject during the absorption phase. First-order absorption model was used widely to approximate the absorption profile of IM injection in population PK analysis (123-125), although a mixed zero-order and first-order absorption model with lag time (126) would be more plausible considering the high lipophilicity of testosterone and oil IM injection. The precision of first-order absorption rate Ka is 36.2%, which is relatively lower compared to the precision of CL and V. This further indicates the instability of the absorption model. The flip-flop pharmacokinetics occurs with extravascularly administered drugs when the rate of absorption is much slower than the rate of elimination. It can cause difficulties in estimation and interpretation of pharmacokinetic parameters if not correctly recognized. (127) Although IV administration of testosterone is not a feasible route of administration in the market, the absorption of oral testosterone is fairly quickly. Therefore the elimination profile of testosterone tablet should be 38

approximately parallel to the IV injection. The half-life of testosterone undecanoate is around 150 minutes, (42) which is very different from the half-life estimates of 4.81 days of IM injection of TC by the population PK analysis. The large difference in half-lives suggests that flip-flop pharmacokinetics are occurring after IM administration of testosterone cypionate, which is also reported in the literature (128) and IM injection of nandrolone. (129) For the flip-flop kinetics of IM injection of TC, diffusion and release from oily depot site is the rate-limiting step. The difference in elimination profiles between gel and IM injection reflects the difference in the absorption rates between two formulations rather than the difference in elimination rates. Median simulated testosterone concentrations for a typical subject with median covariate value were plotted versus time in three dose groups in Figure 12. Panel A shows the simulated median PK profile in the whole study period; panel B shows the simulated PK profile in the first 4 dosing periods; panel C shows the simulated testosterone concentrations versus time after last active injection. From panel B we can see that testosterone concentrations reach the steady state after the second dosing. From panel C the median T max from a single IM injection of TC is about 2 days with 95% CI (0.5,6.5) days, which agrees well with the reported T max in the literature. (8, 100) Based on model estimates of secretion rate of endogenous testosterone in the proposed PK model, suppression of basal endogenous testosterone was most severe in the highest dose group. This was consistent with the observation that it took shorter time to observe the loss of basal and LHRH-stimulated LH in 250mg and 500mg group. Thus the suppression of endogenous testosterone induced by TC administration was separately estimated in different dose groups in pre-specified intervals. The interval from day 28 to day 133 was selected because most LH and FSH concentrations were undetectable during this period, so there would be no or minimal positive feedback of LH on testosterone secretion. 39

Alternative models were tried to fit the suppression of endogenous testosterone, including positive feedback of LH or negative feedback of testosterone itself using sigmoid E max model, but they all failed to converge. One possible reason of failure of convergence in the positive feedback model was that most LH concentrations (629 out of 710) were below LOQ and undetectable in the intensive sampling period. The positive feedback of LH on testosterone secretion was lost during the intensive sampling part of the study and it was hard to fit an E max model when there is actually no effect. The negative testosterone feedback model also failed to converge. One possible explanation is that testosterone concentrations keeps changing after the start of the injection protocol, but the suppression of endogenous testosterone is stable after administration of high doses of exogenous testosterone. Thus it was hard to develop a sigmoid E max model to link the testosterone concentrations to the suppression of endogenous secretion. We evaluated the relation between the AUC in the intensive sampling period with magnitude of LH suppression, however no significant improvement was found in terms of objective function value and goodness-of-fit plots. The estimated basal endogenous testosterone secretion rate in pharmacokinetic models for total and free testosterone models was 11.9 and 13 mg/day, respectively. Other studies suggest that men produce between 4 to 11.8 mg of testosterone per day (130), which is slightly lower than our model estimates. But since there is a wide inherent variability in normal testosterone range between 300 and 1000 ng/dl (131), our model estimates of endogenous testosterone secretion rate are still consistent and are close to the reported physiological value. The V/F for total testosterone was estimated to be 16.8 kl in typical healthy male with median covariate values. This value is much larger than the physiological value of human, mainly because of the high lipophilicity and high protein binding of testosterone. After the IM injection of TC, most drugs will stay in the tissue and body fat, 40

the rest will enter systematic circulation and 99% of them will bind to sex hormonebinding globulin and albumin, only a small amount of testosterone will be free drug in the blood stream. It is also worth noticing that in flip-flop kinetics of TC the absorption rate is much slower than the elimination rate, the PK estimates from current compartmental analysis does not truly reflect the elimination characteristics of testosterone in the body. The estimated half-lives were 4.81 and 4.79 days for total and free testosterone respectively, which were shorter than the mean elimination half-life of 6.9 days previously reported using a non-compartmental analysis. (108) Such inconsistency was believed to result from failure to consider endogenous testosterone production when calculating the half-life using non-compartmental methods. Since testosterone is the endogenous substance and its secretion rate is regulated by the LH-testosterone feedback loop system in HPG axis, it is difficult to calculate the pharmacokinetic parameters of TC using traditional non-compartmental methods, especially when endogenous testosterone secretion rate is suppressed during the course of TC administration. The population pharmacokinetic approach allowed us to estimate the endogenous testosterone production rate during and after the time course of TC administration. It also allows for the estimation of PK parameters even in subjects who have sparse data in the intensive sampling period after the last TC injection. Therefore the model-based estimate of clearance, volume of distribution and its secondary parameter half-life are believed to be more accurate than the estimates in the noncompartmental method. Inclusion of significant covariates resulted in improved goodness-of-fit plots and lower inter-individual variability on the main pharmacokinetic parameters for both total and free testosterone. In the PPK model for total testosterone, baseline weight, baseline 41

albumin and change in albumin from baseline are significant covariates for volume of distribution; weight and change in weight from baseline are significant covariates for clearance. Their effects on clearance and volume of distribution are illustrated in a forest plot showing the effect of extreme values (5 th and 95 th ) on the parameter relative to a typical patient (Figure 13). As shown in the forest plot, the 95% CI of covariate effects for albumin, change in albumin from baseline and change in weight from baseline was almost fully inside the interval of 0.8 to 1.25, indicating these covariates are not clinically important for the tested parameters. On the other hand, the 95% CI for the effect of weight on both V and CL are broad and spans across the specified interval. The point estimates and majority of the CI for effect of weight fall within the clinically relevant interval, which suggests weight is likely to be a clinically important covariate for the pharmacokinetics of testosterone cypionate. The current pop-pk model was built based on the data in healthy subjects. The majority of the patients seeking testosterone therapy, however, are elderly patients. The pharmacokinetic behavior in elderly patients is different than that in the heathy young men. A series of physiological changes will occur as people get older. These changes include cardiac structure and function, blood flow, renal system and body composition, which will affect ADME process in the body. (132) Testosterone is a lipophilic drug with high protein binding, the distribution of which is greatly influenced by protein binding affinity and total body fat. In elderly patients there is a progressive reduction in total body water and lean body mass, resulting in increase in total body fat. The volume of distribution of testosterone in elderly patients is therefore expected to increase. Change in body composition might also affect the diffusion and release rate of TC from the oily depot. In addition, a substantial increase in SHBG in elderly patients was reported in the literature. (133) A higher dose in elderly patients might be necessary in order to achieve optimal therapeutic effects because of the ensuing 42

decrease in free testosterone concentrations. Besides the change in testosterone distribution, a reduction in metabolic clearance of testosterone is also expected in elderly patients. This is probably due to the decrease in hepatic blood flow and phases 1 metabolism activity. 3.7 Conclusions A one-compartment model of linear elimination with first-order absorption best fit the pharmacokinetic profile of long-term weekly injection of testosterone cypionate. The inhibition of endogenous testosterone was modeled with constant reduction of its secretion rate in pre-specified interval from day 28 to day 133. Baseline weight, baseline albumin and change in albumin from baseline are significant covariates for total testosterone. The CL/F and V/F for total testosterone were estimated to be 2.42 (kl/day) and 16.8 kl in typical healthy male with median covariate values. Half-life was calculated to be 4.81 days for total testosterone. The estimation of PK parameters was believed to be more accurate than non-compartmental method as the inhibition of endogenous testosterone secretion resulting from exogenous TC dosing was incorporated into the model. This model identifies weight, albumin and their change from baseline as sources of pharmacokinetic variability and therefore allows for a better prediction of the PK profile of testosterone cypionate in the individual level. 43

Table 2. Baseline demographic characteristics, protein levels and laboratory test values of 31 subjects. Characteristics Median (Range) Age (yr) 28 (21-39) Weight (kg) 85.3 (60.7-114.8) Sex (M:F) Male (31:0) Blood urea nitrogen (mg/dl) 17.4 (7-81) Serum Creatinine (mg/dl) 1.07 (0.8-1.5) Creatinine Clearance (ml/min) 125.0 (84.3-178.6) Albumin (g/dl) 4.6 (1.6-6.9) Globulin (g/dl) 2.9 (2-3.5) Thyroxine-binding globulin (ug/ml) 19.6 (10-31) Sex hormone-binding globulin (nmol/l) 25.8 (11-48) Alkaline phosphatase (IU/L) 65.1 (41-98) Lactate dehydrogenase (IU/L) 151.9 (111-196) AST (IU/L) 22.3 (12-53) ALT (IU/L) 23.7 (10-52) GGT (IU/L) 26.0 (10-58) Total Bilirubin (mg/dl) 0.93 (0.4-2.5) Direct Bilirubin (mg/dl) 0.27 (0.1-0.5) Triglyceride (mg/dl) 159.8 (40-538) Cholesterol (mg/dl) 189.3 (139-241) HDL (mg/dl) 38.4 (21-70) LDL (mg/dl) 121.4 (58-186) 44

Table 3. Population pharmacokinetic estimates and bootstrap results for total testosterone in 31 volunteers. Parameter Final model Bootstrap Estimate RSE(%) Mean RSE(%) 95% CI CL/F (kl/day) 2.42 3.7 2.42 3.7 (2.25,2.60) V/F (kl) 16.8 10.7 16.1 8.6 (14.0,19.5) Ka (day 1 ) 1.67 36.2 1.73 28.1 (0.71,2.62) β i (mg/day) 11.9 6.6 11.9 7.5 (10.1,13.6) β decrease 100mg 6.72 12.8 6.82 12.5 (5.06,8.39) β decrease 250mg 3.01 52.2 3.59 47.0 (-0.29,6.32) β decrease 500mg 2.64 72 2.88 46.9 (-0.01,5.30) θ Balbumin V -1.55 10.1-1.40 61.1 (-3.23,0.13) θ Calbumin V -0.27 6.3-0.25 24.2 (-0.39,-0.15) θ BWT V 1.36 39.5 1.49 34.1 (0.36,2.35) θ BWT CL 0.789 15.6 0.815 15.9 (0.54,1.05) θ CWT CL 0.017 34 0.0174 46.5 (0.001,0.033) σ 2 additive 0.532 89.5 0.636 59.1 (-0.204,1.268) σ 2 proportional 0.057 25.8 0.051 22.6 (0.034,0.080) IIV_CL 9.5% 57.6 9.0% 49.6 (2.9%, 12.7%) IIV_V 28.0% 44.9 28.1% 49.0 (4.3%, 40.6%) IIV_Ka 74.0% 60.9 73.2% 59.2 (19.1%, 122.3%) IIV_β i 26.7% 44.5 23.1% 40.0 (13.2%, 31.5%) 45

Table 4. Population pharmacokinetic estimates and bootstrap results for free testosterone in 31 volunteers. Parameter Final model Bootstrap Estimate RSE(%) Mean RSE(%) 95% CI CL/F (10 5 L/day) 6.03 4.2 6.11 3.7 (5.69,6.65) V/F (10 5 L) 41.7 8.2 41.8 8.4 (35.2,50.1) Ka (day 1 ) 1.43 26.2 1.58 48.1 (0.88,3.91) β i (mg/day) 13.0 5.1 13.0 4.4 (11.7,14.3) β decrease 100mg 7.87 13.3 8.03 12.7 (5.95,0.88) β decrease 250mg 2.98 63.1 3.44 49.5 (1.18,8.78) β decrease 500mg 0 Fix θ Cglobulin V -0.62 19.2-0.55 22.9 (-0.73,-0.27) θ BWT CL 0.49 30.2 0.52 27.5 (0.2,0.8) σ 2 additive 0.22 89.5 0.23 29.4 (0.10,0.37) σ 2 proportional 0.037 25.8 0.035 16.9 (0.025,0.052) IIV_CL 10.4% 33.6 9.4% 41.7 (3.8%, 13.7%) IIV_V 36.9% 28.8 36.1% 34.7 (19.1%, 47.6%) IIV_Ka 70.0% 69.1 83.0% 75.6 (10.9%, 153.3%) IIV_β i 18.2% 40.1 17.9% 35.6 (10.0%, 23.2%) 46

Figure 3. Overview of testosterone metabolism 47

Figure 4. Goodness-of-fit plots of total testosterone concentrations. Red lines are smoothing lines. Black lines are lines of identity. Blue dots are observed values. 48

Figure 5. Predicted and observed individual concentration-time profiles of total testosterone in 100mg, 250mg and 500mg. 49

50

51

Figure 6. Prediction-corrected visual predictive check for total testosterone obtained from 1000 simulations stratified on sampling period. 5 th, 50 th and 95 th percentiles of observations (Black solid and dashed line). 95% confidence intervals of 5 th, 50 th and 95 th percentiles of simulated data (orange, green and blue areas). 52

Figure 7. Scatterplot of NPDE versus time and population predictions of total testosterone. Red lines are smoothing lines of NPDE versus time or population predictions. 53

Figure 8. Goodness-of-fit plots of free testosterone concentrations. Red lines are smoothing lines. Black lines are lines of identity. Blue dots are observed values. 54

Figure 9. Predicted and observed individual concentration-time profiles of free testosterone in 100mg, 250mg and 500mg. 55

56

57

Figure 10. Prediction-corrected visual predictive check for free testosterone obtained from 1000 simulations stratified on sampling period. 5 th, 50 th and 95 th percentiles of observations (Black solid and dashed line). 95% confidence intervals of 5 th, 50 th and 95 th percentiles of simulated data (orange, green and blue areas). 58

Figure 11. Scatterplot of NPDE versus time and population predictions of free testosterone. Red lines are smoothing lines of NPDE versus time or population predictions. 59

Figure 12. Simulated median total testosterone concentration versus study time based on 1000 simulations. Solid line: Median total testosterone in 3 dose groups 60

Figure 13. Effects of significant covariates on clearance and volume of distribution in total testosterone. 61

CHAPTER IV EFFECT OF TESTOSTERONE CYPIONATE ON LH HORMONE AND SPERMATOGENESIS 4.1 Objective Long-term illegal use of AAS might have negative effects on health status. Numerous studies have shown that long-term use or excessive dosing of anabolic steroids can lead to serious, sometimes irreversible, health risks. One of the most significant physiological changes induced by the use of TC is a dose-dependent impairment of normal testicular androgen secretion and spermatogenesis (134-136). This effect is believed to result from the suppression of circulating luteinizing hormone (LH) and follicle stimulating hormone (FSH) through feedback loop system of hypothalamic-pituitary-gonadal (HPG) axis (136, 137). Although several studies have demonstrated the suppression of gonadal function in the experimental setting, the extent and variability of such suppression remains poorly defined and its relationship to testosterone exposure has not been quantified. (108, 138-140) The objective of this chapter is to develop a PK-PD model to describe and predict the suppression and subsequent recovery of luteinizing hormone and gonadal function following TC dosing. The PK-PD model will help with the mechanistic understanding of testosterone action and the underlying regulatory action of HPG axis. 4.2 Methods 4.2.1 Subjects Enrollment and Randomization A cohort of healthy male volunteers between the ages of 18 and 40 years old was recruited. All study subjects reviewed the study summary and signed written informed consent. Subjects were aware they would have a one-third chance of receiving 62

one of the study doses. They were also aware they would receive weekly injections of testosterone cypionate (TC) and placebo each for 14 weeks but were not informed of the sequence. 4.2.2 Inclusion and Exclusion Criteria Subjects received a physical health examination including electrocardiogram, blood tests and a psychiatric evaluation prior to the beginning of the study. Subjects were excluded it they had any of the following: baseline blood pressure greater than 140 systolic or greater than 90 diastolic; any chronic medical condition requiring ongoing assessment and treatment like diabetes, hypothyroidism, peptic ulcer disease; significant biochemical abnormality in liver function tests, serum cholesterol, and serum HDL cholesterol at entry; the presence of an abnormal entry semenogram, anemia; a renal function or electrolyte abnormality; evidence of recent mood, anxiety, substance abuse or psychotic disorder. In addition, potential subjects completed the Buss-Durkee Hostility Inventory and subjects with scores greater than 30 on the total scale and greater than 15 on the aggression subscale on this measure were excluded from study participation (44). 4.2.3 Chronological Design The chronological design of the study included two consecutive weekly injections of an identically appearing TC placebo, followed by 14 consecutive weekly injections of testosterone cypionate. 14 weeks were chosen to mimic the common cycling duration observed by illicit steroid users in the community. Following the last weekly injection of active agent the subjects received 12 consecutive weeks of TC placebo injections. (46) 4.2.4 LHRH stimulation test Pituitary gonadotropin secretion was assessed by performing biweekly leutinizing hormone releasing hormone (LHRH) stimulation tests 7 days after the last experimental injection of either placebo or testosterone. The LHRH test involves intravenous administration of 2.5 ug/kg LHRH and determination of peak LH and FSH 63

responses in a 2-hour period after dosing. (141) LHRH stimulation test is an important tool in the diagnosis of hypogonadism and children with puberty disorder. (142) Venous blood samples for LH and FSH were obtained before and 30, 60, and 120 min after the intravenous infusion of 100 mcg LHRH. 4.2.5 Assessment of spermatogenesis Semen samples were obtained by masturbation and assessed for total and motile sperm concentrations at baseline (week 0), weeks 2, 16, 21, 28 and in some subjects at week 40. Subjects were asked to be sexually abstinent for 24 hours before the sample. 4.2.6 Population Pharmacodynamic model building The PK/PD data analysis was performed sequentially. The PK parameters for total and free testosterone was first estimated with the previously specified model (Chapter 3), then PD model for suppression of LH and spermatogenesis was sequentially fitted conditionally on fixed individual PK parameters. Because of the variability in the PD endpoints, natural log-transformation was applied to the data before the PD analysis. The population pharmacodynamic model was fitted using NONMEM version 7.2 (ICON Development Solutions, Ellicott City, MD). NONMEM outputs were processed using Pirana (112) and Xpose version 4.5.3 (Uppsala University, Uppsala, Sweden) (113). R version 3.0.1 (Free Software Foundation, Vienna, Austria) (114) was used for statistical analysis and final report generation. LH concentrations obtained at various time points (before, 30 minutes, 60 minutes, 120 minutes) were evaluated as PD endpoints of gonadal function. Given the inhibitory effects of testosterone on LH production, the following indirect response models were fitted to the observations: empirical linear model, exponential model, direct response model and indirect response model. A Linear model and sigmoid Emax model were evaluated to quantify the effect of testosterone concentrations on 64

suppression of LH. To test whether there was a delayed testosterone effect on LH suppression, an effect compartment or transit model were added to the structural PD model. The PD model for suppression of spermatogenesis was tested in the same procedure as LH. As active metabolites dihydrotestosterone and estradiol were not measured and quantified in the study, it is assumed that elevation of testosterone concentration was the major driving force that led to LH suppression. Model selection was based on the same criteria of PK model building described in section 3.4.1. Potential covariates were screened and tested using stepwise selection on the PD parameters after the best base model was chosen. The details of stepwise selection can be found in section 3.4.3. 4.3 Results 4.3.1 PD model for suppression of LH The LH concentration at 30 minutes after stimulation test (LH30) was chosen as the PD endpoint. There are a total of 379 LH serum samples (9-13 samples per subject) in the PD dataset. The time interval between two sampling points varies from one week to 3 weeks. Given the known inhibitory effect of testosterone on LH production, an indirect effect model best described the observations: d(lh 30 ) d(t) = K in (1 INH) K out LH Where d(x)/d(y) represents the derivative of x at y, LH 30 is the LH concentration 30 minutes after the stimulation test, K in is the constant zero-order production rate, K out is the first-order degradation rate. INH describes the inhibitory effect of testosterone on λ synthesis of LH using sigmoid E max model: INH=C e / (C λ e + IC λ 50 ). Inclusion of an effect compartment results in better model fitting and lower objective function value. C e is the 65

hypothetical effect compartment concentration accounting for the lag between surge of testosterone and loss of LH. IC 50 is the testosterone concentration when LH synthesis is inhibited by half. E max is fixed to 1 as LH synthesis is fully inhibited during the study. Baseline weight and thyroxin were found to be significant covariates for IC 50 and Kin. Based on the model estimates, a subject with higher weight at baseline would have a lower IC 50, and a subject with higher thyroxin would have a higher Kin. The final PD model estimates and bootstrap results for LH suppression are shown in the Table 5; IC 50 and λ were estimated to be 9.42 ng/ml and 9.95, respectively. Goodness-of-fit plots are shown in Figure 14, no obvious bias can be found in the residual diagnostic plots. The individual predictions in all subjects agreed well with the observations, as shown in Figure 15. Visual predictive check generated from 1000 simulations is shown in Figure 16. The 95% CI of 5 th, 50 th and 95 th percentiles from 1000 simulations well captured the corresponding percentiles of observations, which suggests model simulation adequately describes the magnitude of LH suppression in the whole study period. Simulation in NLME is a very useful tool to evaluate and summarize the central tendency and variability in dose-response relationship. The median and variability of LH suppression over the study period in three dose groups are shown in Figure 17 and Figure 18. Based on the simulation results, LH is greatly suppressed in both the 250mg and 500mg dose group; LH is completely suppressed for 13 weeks (from 5 th to 18 th week) in 250mg group and 15 weeks (from 5 th to 20 th week) in 500mg group, the level of median suppression surpasses 95% during these periods. There is higher consistency in suppression of LH in 500mg group compared to 250mg group. e.g. The 95% CI for suppression of LH at the 15 th week is (99%,100%) in the 500mg group compared to (78%,100%) in 250mg group. In contrast, the level of suppression of LH is lower and more variable in 100mg. The highest level of suppression occurs at week 15 and its median suppression level is 77% with 95% CI (0.4%, 100%). 66

The duration of LH suppression is longer in the high dose group. LH suppression can be observed as early as the 3 rd week in the 500 mg dose group, and the median suppression is 52.8% compared to 0.3% in the 100mg and 25.1% in the 250mg group. This suppression of LH continues until the 22 nd week when values return to baseline. The level of suppression drops below 50% at the 24 th week in 500mg group, which is later than the 21 st week in 250mg and the 18 th week in 100mg group. 4.3.2 PD model for suppression of spermatogenesis Because of vasectomies in two patients, semenogram data although not always complete was available in 29 subjects. A total of 159 sperm samples were obtained at baseline, week 2, 16, 21, 28 and in some subjects at week 40. Both sperm counts and sperm motility were measured and used as the endpoint in the PD analysis. An indirect response model was established to characterize the change of spermatogenesis during and after TC dosing: d(sperm) d(t) = K in (1 INH) K out LH where sperm is sperm counts or motility, K in is the constant zero-order production rate for spermatogenesis, K out is the first-order degradation rate. INH, the inhibition of testosterone dosing on spermatogenesis, was estimated separately at each time point in the PD model. The goodness-of-fit plots of sperm counts and motility are shown in Figure 21 to 22. Individual predictions are shown in Figure 23 to 24, there is no obvious bias between predictions and observations. The visual predictive check in Figure 25 to 26 shows the model simulation is able to describe the suppression of spermatogenesis after long-term abuse of testosterone injection. The predicted suppression levels of sperm counts and motility during and after TC dosing based on 1000 simulations are comparable (Figure 27). The simulation results show that spermatogenesis production rate in the 250 and 500mg dose group was 67

estimated to be fully suppressed (suppression level larger than 95%) by the end of dosing (15 th week). In contrast, the suppression level in 100mg group was variable; some subjects showed near to normal sperm level while others had no detectable motile sperms. Interestingly, subjects with near to normal sperm levels were the only subjects who maintained their basal FSH throughout the study. Thus we separately estimated the spermatogenesis of subjects in 100mg based upon their basal FSH level; production rate K in of subjects with undetectable basal FSH levels in 100mg group was estimated to be fully inhibited, while in subjects who have maintained their FSH production K in was estimated to be 50.5% (95% C.I [8.4%,92.6%]) of the original level. The model estimates of spermatogenesis suppression in terms of sperm motility at different time points in each dose group are listed in the table 6. Originally the model parameters for spermatogenesis suppression at the end of dosing (15 th week) were estimated to be very close to 0 for subjects with undetectable basal FSH concentration in 100mg group and all subjects in 250mg, which made the estimation of variancecovariance matrix difficult and unstable. These estimates were therefore fixed to 0 in the final model. At the 21 st week of the study, almost all exogenous testosterone was eliminated from the body, but spermatogenesis was still found to be severely inhibited in the two high dose groups, with the mean suppression level estimated to be 93.5% and 95.7%. By the 28 th week and 40 th week the suppression level is getting lower and returning to baseline. At the end of the study (40 th week), the level of spermatogenesis in the 100mg and 250mg group has almost fully recovered, with K in estimated comparable to baseline (117% and 93.2% respectively). However, the production rate K in only recovered to about half [54.9%, 95% CI (19.8%, 90%)] by the 40 th week in 500mg group. The parameter estimates in the PD model for sperm counts were comparable and listed in table 6. 68

4.3 Discussion An indirect response model was used to describe the change of PD marker LH30, LH concentrations obtained 30 minutes after stimulation test, during and after TC dosing. LH30 was chosen as the PD endpoint as it is more representative of the level of LH synthesis after TC dosing when compared to baseline LH values. A sigmoid E max model was applied to link testosterone concentrations with its inhibitory effect on LH synthesis. The addition of an effect compartment accounted for the lag between loss of LH and surge of testosterone. LH concentration after stimulation test rather than baseline LH was chosen as the PD marker because response of luteinizing hormone to the stimulation test was different across all three dose groups while most basal LH concentrations were undetectable in all subjects. Subjects in 100mg group began to have measureable LH concentrations 30 minutes after the stimulation test but LH concentrations remained close to 0 in most subjects in two high dose groups. The lag between loss of basal and stimulated LH levels has been previously observed and one of the proposed explanation was that there are reservoirs of previously formed LH in the body which become depleted over time (143). Exogenous testosterone administration may also affect the ability of these reservoirs to release LH and such effects should be incorporated and tested in the PD model. Initially LH30 was chosen as the PD endpoint to evaluate the inhibitory effect of TC dosing on LH synthesis. After establishment and validation of the final PD model, LH60 and LH120 were also fitted as the PD endpoints with same model structure linked to total or free testosterone and compared with LH30. All models were found to have comparable performance and parameter estimation. The Monte Carlo simulations further illustrate the magnitude and variability of LH suppression in the three different dosing groups. Overall the level of LH suppression lasts longer in 250mg and 500mg group (Figures 17 and 18); the median suppression level for 2 high doses was 99% during TC dosing period, while it was only 68% in the 100mg group. Also the duration of severe LH suppression based on simulations, defined 69

as when the magnitude of LH suppression is larger than 95%, was longer in 500mg group (15 weeks) compared to 250mg group (13 weeks) and 100mg group (0 weeks). There is substantial variability in LH suppression at the end of dosing (15 th week) in 100mg group; the 1 st quartile of suppression level is equal to 25.1%. On the other hand, the 1 st quartiles of suppression level in 2 high dose groups are 98.9% and 100.0% respectively, which indicates LH synthesis in almost all subjects in these 2 groups is completely suppressed during TC dosing. Baseline weight and thyroxin are significant covariates for IC 50 and Kin. Their effects are illustrated in a forest plot showing the effect of extreme values (5 th and 95 th ) on the parameter relative to a typical patient (Figure 19). As shown in the forest plot, a subject with higher weight at baseline would have a lower IC 50. The 95% CI of covariate effects for weight and thyroxin were almost fully outside of the specified interval 0.8 to 1.25, indicating potentially clinical significance. 1000 simulations were further conducted to illustrate these covariate effects. LH suppression tends to start earlier and end later in patients with baseline weight greater than the 3 rd quartile (95.1kg) (Figure 20). Although the 95% CIs of LH suppression in the first three quartiles of baseline weight are wide and almost identical, the variability for LH suppression in the patients with baseline weight larger than 3 rd quartile (95.1kg) is narrow, suggesting consistent high suppression of LH during dosing period. Based on the estimated covariate effects and simulation results, LH suppression is more severe in the subjects with a large body weight at baseline. In addition, we were able to describe the inhibition and subsequent recovery of spermatogenesis during and after TC dosing in our model. Spermatogenesis was not linked to testosterone concentration by PK-PD model because sperm data was collected in only 1 time point during testosterone surge, other sperm samples were collected after the last testosterone injection. During model building, effect and transit 70

compartments were tried to account for the delayed effect of testosterone administration on spermatogenesis, but both models failed to converge or produce reasonable estimates. Therefore the spermatogenesis was estimated individually at each time point in each dose group in the final PD model. With this approach, we were able to describe the suppression and subsequent recovery of spermatogenesis in each dose group separately. Monte Carlo simulations also suggest that suppression of sperm motility in subjects in 250mg and 500mg group after the end of dosing (15 th and 21 st week) was severe (>95%) and consistent. But the suppression in 100mg group was quite variable; the spermatogenesis level is totally suppressed in some subjects but near to normal in others. FSH level was found to be related to the variable spermatogenesis suppression in 100mg group. At the end of study, spermatogenesis recovered to baseline level in 100mg and 250mg group, but remained suppressed in the 500mg group. In the simple ANOVA analysis, because of small sample size and large variability in the sperm data, no significant statistical difference was found in sperm motility or count when comparing three dose groups (Kruskal-Wallis Nonparametric ANOVA (KW) = 4.2368, p =0.12, df = 2). But in the model-based approach, we were able to build a mechanistic model based on regulatory feedback actions in HPG axis to test and characterize the differences in suppression of spermatogenesis between the three doses. One limitation is that although the PD model is accurate in describing the level of spermatogenesis in specific time points for 100, 250 and 500mg dose, it does not have the ability to extrapolate the prediction beyond study time or to other dose groups. Another limitation is that the study doesn t control other factors which might affect spermatogenesis. Several lifestyle choices such as obesity and smoking and 71

environmental factors like exposure to chemicals, pollutants and combustion products are reported to negatively affect adult testes and spermatogenesis. (144, 145) In addition, research has shown that genetic disorders account for about 15-30% cases of male infertility and are responsible for the majority of the idiopathic cases. (145) Stress, physical exercise and weight loss have also been found to correlate with semen quality. (146-149) Among these factors physical exercise and environmental factors were the most possible confounders we fail to control in the current analysis. 4.4 Conclusions In summary, we developed an indirect pharmacodynamic response model to characterize the suppression of LH synthesis and spermatogenesis during TC dosing and its subsequent recovery afterwards in healthy males. The estimated potency (IC50) of total testosterone with respect to LH suppression was 9.38ng/ml. The Monte Carlo simulations showed that the suppression of LH synthesis and spermatogenesis was more severe and of greater duration in 250mg and 500mg dose groups. 72

Table 5. Population pharmacodynamic estimates and bootstrap results of LH suppression following long-term dosing of testosterone cypionate. Parameter Final model Bootstrap Estimate RSE (%) Mean RSE (%) 95% CI Kin (IU/day) 1.98 11.3 2.01 11.4 (1.55,2.48) Kout (L/day) 0.13 7 0.13 6.4 (0.11,0.14) IC 50 (ng/ml) 9.38 5.9 9.28 7.0 (8.35,10.78) λ 9.85 14.9 10.45 18.1 (7.62,15.78) k e0 (ml/day) 0.06 7.4 0.06 8.4 (0.05,0.07) σ 2 additive 0.38 11.1 0.37 11.5 (0.28,0.45) θ wt IC50-1.39 30.4-1.44 36.4 (-2.76,-0.41) θ thyroxin Kin 1.2 11.6 1.3 24.7 (0.62,2.12) IIV_Kin 45.2% 43.7 44.4% 46.8 (20.3,64.4) IIV_IC 50 33.8% 26.8 31.0% 34.8 (18.8,41.1) IIV_λ 32.9% 82.1 40.0% 70.9 (15.2,75.3) IIV_k e0 41.7% 40.9 42.1% 38.3 (26.3,58.3) 73

Table 6. Point estimates of level of sperm production compared to baseline (0-100%) at each time point for three dose groups by the indirect response PD model. 100% means same to baseline level, 0% means total inhibition. Time (week) Sperm [0,2] (2,14] (2,14] (14,21] (14,21] (21,28] (28,40] motility FSH conc. =0 FSH conc. >0 FSH conc. =0 FSH conc. >0 100mg 100 0 (fixed) 50.5 56.9 11 66.6 117 250mg 100 0 (fixed) 6.5 51.8 93.2 500mg 100 1.92 4.3 32.1 54.9 Sperm counts 100mg 100 0 (fixed) 70.7 40.1 11.6 75 138 250mg 100 0 (fixed) 5.3 46.9 85.2 500mg 100 3.3 3.8 31.9 63 74

Figure 14. Goodness-of-fit plots of LH hormone. Red lines are smoothing lines. Black lines are lines of identity. Blue dots are observed values. 75

Figure 15. Predicted and observed individual concentration-time profiles of LH hormone in 100mg, 250mg and 500mg. 76

77

78

Figure 16. Visual predictive check for LH obtained from 1000 simulations. 5 th, 50 th and 95 th percentiles of observations (Black solid and dashed line). 95% confidence intervals of 5 th, 50 th and 95 th percentiles of simulated data (orange, green and blue areas). 79

Figure 17. Median suppression level of LH in 100mg, 250mg and 500mg TC from 1000 simulations. 80

Figure 18. Variability of LH suppression in 100mg,250mg and 500mg TC from 1000 simulations. Green, blue and red solid lines: median simulated level of LH inhibition in 100, 250 and 500mg groups. Green, blue and red shaded areas: 95% confidence intervals of simulated level of LH synthesis in 100, 250 and 500mg groups. 81

Figure 19. Effects of significant covariates on pharmacodynamic parameters of suppression of luteinizing hormone. 82

Figure 20. Median and variability of suppression level of LH stratified on 4 quartiles of baseline weight obtained from 1000 simulations. Green, blue, red and black solid lines: median simulated level of LH inhibition stratified on 4 quartiles of baseline weight. Green, blue, red and black shaded areas: 95% confidence intervals of simulated level of LH synthesis stratified on 4 quartiles of baseline weight. 83

Figure 21. Goodness-of-fit plots of sperm counts. Red lines are smoothing lines. Black lines are lines of identity. Blue dots are observed values. 84

Figure 22. Goodness-of-fit plots of sperm motility. Red lines are smoothing lines. Black lines are lines of identity. Blue dots are observed values. 85

Figure 23. Predicted and observed individual concentration-time profiles of sperm counts in 100mg, 250mg and 500mg. 86

87

88

Figure 24. Predicted and observed individual concentration-time profiles of sperm motility in 100mg, 250mg and 500mg. 89

90

91

Figure 25. Visual predictive check for sperm counts obtained from 1000 simulations. 5 th, 50 th and 95 th percentiles of observations (Black solid and dashed line). 95% confidence intervals of 5 th, 50 th and 95 th percentiles of simulated data (orange, green and blue areas). 92

Figure 26. Visual predictive check for sperm motility obtained from 1000 simulations. 5 th, 50 th and 95 th percentiles of observations (Black solid and dashed line). 95% confidence intervals of 5 th, 50 th and 95 th percentiles of simulated data (orange, green and blue areas). 93

Figure 27. Predicted level of suppression of spermatogenesis over time during and after TC dosing based on 1000 simulations. Median (solid line), 95% CI (shaded area). (Left panel: Sperm counts, Right panel: Sperm motility). 94

CHAPTER V EFFECT OF TESTOSTERONE CYPIONATE ON BODY WEIGHT AND LIPID PROFILES 5.1 Introduction We have characterized the effect of TC on suppression of LH and spermatogenesis using population pharmacodynamic approach in the previous chapter. Besides its effect on HPG axis, anabolic steroids also affect the human body in many other ways including an increase in body weight and strength, the most visible and significant effect on human body (51-53). The effect of testosterone on lipid profiles is ambiguous, whether abuse of anabolic steroids will lead to increased risks for cardiovascular diseases remains a great medical concern. In the January 2014, U.S. FDA issued a warning about safety of testosterone products (150). In this communication, the FDA states they have started the investigation of the risk of stroke, heart attack and death in men taking FDA-approved testosterone products. The alert is based on the publication of two separate studies (151, 152) that suggested an increased risk of cardiovascular events among patients taking testosterone therapy, which further suggested there is still uncertainty about the effect of testosterone therapy on cardiovascular diseases even though the therapy has been approved for more than 50 years. It is ideal to use cardiovascular events as the primary endpoint to study the correlation between potential factors and risks of cardiovascular disease. However, it requires a large number of subjects and long study time to achieve statistical significance because of rare events and is therefore often difficult to design in the clinical practice. High density lipoprotein cholesterol (HDL) has been identified as a strong, independent predictor of risks for coronary heart disease by many cohort 95

studies and meta-analyses. (153-157) HDL screening has been recommended for risk assessment and management of cardiovascular diseases. (158-160) In this study HDL along with other lipids (LDL, triglyceride, cholesterol) serve as surrogate markers for risk of cardiovascular disease. 5.2 Objective Since it is not ethical to expose healthy volunteers to supra-therapeutic dosages for the sole purpose of studying its effects, most studies are limited in their ability to truly estimate the effect of abuse of AAS in real life; only a few studies have met the current scientific quality standards of randomized, double-blind placebo-controlled study design. The dosing regimen in most studies is usually far below the daily practice in illicit steroid user community. Therefore this three-arm randomized clinical trial offers very precious data for us to evaluate the real-life impact of long-term abuse of steroids on human body. In this chapter we aim to use population modeling approach to fit the longitudinal change of weight and lipid profiles and interpret how individuals respond to long-term TC dosing. 5.3 Methods 5.3.1 Safety monitoring and lab measurements Subjects were carefully monitored for potential medical and psychiatric problems that would trigger discontinuation from the study for safety reasons. Vital signs were obtained at each weekly visit. General chemistry profiles like liver function studies, hematocrits, and fasting lipid profiles were accessed biweekly throughout the study. 96

5.3.2 Model building As the lab measurements over time is highly correlated with each other in the same individual, and these biological values such as weight and lipid profiles are approximately normally distributed, mixed effect modeling was conducted in NONMEM to allow different change in response between individuals. Different empirical models such as linear model, exponential model, weibull model, indirect response model, inverse Bateman model and their combinations are tested and compared with each other. Linear or nonlinear mixed effect model was fitted using NONMEM version 7.2 (ICON Development Solutions, Ellicott City, MD). NONMEM outputs were processed using Pirana (112) and Xpose version 4.5.3 (Uppsala University, Uppsala, Sweden) (113). R version 3.0.1 (Free Software Foundation, Vienna, Austria) (114) was used for statistical analysis and final report. Likelihood ratio test (LRT) was used to discriminate between different proposed models. A model was a better fit to the observation if the objective function value decreased 3.84, which corresponds to p-value of 0.05 in a chi-square distribution with degree of freedom equal to 1. After the best base model was established, stepwise selection was conducted to screen and test the potential significant covariates on the linear and quadratic effect. 5.4 Results 5.4.1 Change in weight A linear mixed effect model with change point in slope was found to best describe the change in weight. The chronological design of the clinical trial involves 14 weekly injections of TC dosing and subsequent 12 weekly injections of placebo. We would expect the change of weight to be different after cessation of TC dosing. Therefore a change point (t 0 = 16) was introduced into the linear mixed model. It is very likely that weight could increase rapidly during the first few weeks of use and then 97

reach a plateau. A quadratic polynomial term to describe this nonlinear relationship was tested and added to the linear model. The general equation for the polynomial change point mixed model can be described in the following regression. μ = β 0 + β 1 t + β 2 t 2 + β 3 [(t t 0 ) I] + β 4 [(t t 0 ) 2 I] I = { 0 t < t 0 1 t t 0 where μ is the change in weight, β 0 is the intercept, β 1 and β 2 represent the coefficients of linear and quadratic effect, β 3 and β 4 represent the coefficients of change in linear and quadratic effect after cessation of TC dosing. IIV (inter-individual variability) was added to these coefficients to allow the linear and quadratic effects to vary among different subjects. As we are fitting the model to the baseline-adjusted response, the coefficient for intercept was tested to be non-significant as all change from baseline is equal to 0 in the beginning. Coefficients β 1, β 2 and β 4 were found to be different among three dose groups; e.g. β 1 was fixed to 0 for 100mg and estimated to 0.60 kg/week for 250mg and 500mg dose groups. Model estimates are listed in table 7. Basic goodness-of-fit plots, plots of individual predictions versus observations and VPC stratified on dose can be found in Figures 28-30. The diagnostic plots indicate the model is able to describe the weight change in all three dose groups. No significant covariate was identified. 1000 Monte Carlo simulations were conducted to summarize the effect of long-term TC dosing on weight. (Figure 29) Simulation results show that subjects taking weekly injection of 250mg and 500mg TC have a significant increase in weight during the dosing period, the median increase in weight from baseline is 3.24 kg by week 12, compared to 0.4kg in 100mg group at the same time. Although the median increase in weight is indistinguishable between 250mg and 500mg, the confidence intervals obtained from 1000 simulations showed that the increase in weight in 500mg is more consistent and 98

less variable; the 95% CI for increase in weight at 12 th week is (1.0kg, 5.7kg) in 500mg compared to (-2.8kg, 9.0kg) in 250mg group. After cessation of TC dosing, weight was found to decrease below baseline values. The median decrease in weight from baseline at 28 th week is 1.85kg, 1.76kg and 1.35kg in 100mg, 250mg and 500mg group respectively. 5.4.2 Change in lipid profiles A linear mixed effect model with change point in slope was also the best model to describe the change in lipid profiles, namely triglycerides, LDL, HDL, cholesterol and hematocrit (HCT). Model estimates were listed in Table 8. The individual predictions were found to agree well with the observations (Figure 32), visual predictive check also showed model is able to adequately describe the change in all lipid profiles (Figure 33). No significant covariates are found in the stepwise selection procedure except for triglycerides. In the linear model for triglycerides, baseline triglycerides are found to strongly correlate with the slope of change and its quadratic effect. To better summarize the response of lipid profiles to long-term TC dosing, median change from baseline for all lipid profiles from 1000 simulations are presented in Figure 34. In general, the lipid profiles decreased after TC administration. The median change in triglycerides, HDL, LDL and cholesterol based on model simulation is -16.0, -2.7, -5.4, -11.2 mg/dl after last active TC injection (16 th week). The profile of simulated median change in triglycerides over time was also stratified on 4 quartiles of the baseline triglycerides values to show how the change varies among patients with different baseline values. (Figure 35) There is a different profile between subjects with high baseline triglycersides (>3 rd quartile) and the rest subjects with median to low baseline. For subjects whose baseline is lower than 174 mg/dl there is a very small decrease at the end of dosing (16 th week), while for patients whose baseline is greater than 174 mg/dl, the median decrease in triglycerides at the end of dosing is 85.6 mg/dl. 99

5.5 Discussion Increase in body weight and strength is the most typical reason why people take risks to use anabolic steroids illegally (50). In order to understand how the body weight reacts to different TC dosing regimens and how it recovers after cessation of doing, we built a polynomial change point mixed effect model to successfully describe the change in weight from baseline in three different dosing regimens. The addition of a change point at 16 th week enables the model to fit the decrease in weight after cessation of doing. Likelihood ratio test shows that a linear model with a quadratic effect is able to better fit the observations, which agrees with the hypothesis that weight increase fast at first few weeks then slowly reach a plateau. The increase in weight following the 250mg and 500mg dose was significantly greater than that for the 100mg dose, but there was no difference in the mean increase rate of weight gain between two high dose groups. The model simulation predicts that in 250mg and 500mg group the maximum increase of weight from baseline occurs at 10 th week with median value equal to 3.26kg. Simulation also shows that weight will decrease to lower than baseline value at all three dose groups after cessation of dosing; At 12 weeks after last TC injection, weight was predicted to be lower than baseline with median decrease 1.76kg. The possible reasons for such decreases include the inhibition of endogenous testosterone secretion or the reaction of human body to the cessation of long-term TC dosing. Failure to discriminate the weight increase between two high dose groups suggests that 250mg dosing regimen is high enough to result in the same effect on body weight as 500mg, although the estimate of weight increase is more variable in 250mg. A polynomial change point mixed effects model also adequately describes the change in lipid profiles during study periods. All lipids (Triglycerides, HDL, LDL and cholesterol) were found to decrease first and return to baseline later. However 100

statistical significance is not always equal to clinical importance. There is a high biological variation in the lipid test results. The normal day-day variation of HDL, LDL and total cholesterol is smaller than 5% but it can as high as 20% for triglycerides. (161) Many studies have documented a seasonal variation although these results are discordant. (162) Posture, diet and medications are also found to affect the test results of lipid profiles. Taking into account the high biological variation of lipid test results in the long-term study and the fact model doesn t detect any difference in lipid change between three dose groups, we must pay extra caution when interpreting the effect of testosterone injection on lipid profiles. Therefore although the model suggests that there is a tendency for a reduction in total lipid concentrations after TC administration, failure to identify any difference in change between dose groups suggests more studies are needed to conclude any causal relationship or clinically meaningful effect. Baseline triglyceride was found to be the only significant covariate identified for lipid profile during model building. The change in triglycerides in subjects with very high baseline values was found to be different from the rest of subjects. Triglycerides in individuals with very high baseline values experienced a decrease after weekly injection of TC (Figure 35). 5.6 Conclusions In summary, a polynomial mixed effect model with change point was developed to successfully describe the change in weight and lipid profiles after weekly injection of testosterone cypionate. Model evaluation and simulation show that both 250mg and 500mg TC would incur an average increase of body weight of 3.5kg at 8 weeks after dosing. A polynomial change point model also identifies that there is a tendency for lipid decrease after TC administration. However, no difference was found in the lipid change 101

between three dose groups, which precludes any definite conclusion on the effect of long-term TC administration on lipid profiles. 102

Table 7. Model estimates of polynomial change point mixed model for change in weight from baseline. Parameter Final model Bootstrap Estimate Mean RSE (%) 95% CI 0 Fix β 1 100mg β 1 250mg/500mg β 2 100mg β 2 250mg/500mg 0.60 0.62 11.6 (0.50,0.79) 0 Fix -0.027-0.027 16.4 (-0.037,-0.020) β 3-0.32-0.31 30.0 (-0.50,-0.13) 0.011 0.011 31.2 (0.004,0.019) β 4 100mg β 4 250mg/500mg 0.046 0.046 12.0 (0.037,0.058) σ 2 additive 0.883 0.890 6.3 (0.759,0.991) IIV_β 1 100mg/250mg 45.2% 48.5% 34.6 (33.9%, 65.2%) IIV_β 1 500mg 20.4% 20.3% 53.2 (7.0%, 31.3%) IIV_β 2 100mg/250mg 2.2% 2.2% 25.2 (1.6%, 2.8%) IIV_β 2 500mg 1.8% 1.8% 33.7 (1.1%, 2.5%) IIV_β 3 50.5% 49.8% 28.2 (33.3%, 63.9%) IIV_β 4 2.4% 2.4% 26.0 (1.8%, 3.0%) 103

Table 8. Model estimates of polynomial change point mixed model for change in lipid profiles from baseline. Parameter Triglycerides HDL LDL Cholesterol HCT β 1-2.37-0.97-0.46-2.66-0.03 β 2 0.13 0.05 0.007 0.121 0.008 β 3-3.37 0.233-0.08-1.07-0.53 β 4 1.69-0.076 0.017-0.098 / σ 2 additive 44.5 5.75 14.4 14.7 1.98 η β1 170 2.37 16.6 18.1 0.019 η β2 0.50 0.006 0.047 0.055 / η β3 24.8 1.63 11.1 15.1 0.035 η β4 1.6 0.007 0.045 0.046 / θ BTRI β1 1.90 / / / / θ BTRI β2 1.73 / / / / θ BTRI β3 1.59 / / / / θ BTRI β4-5.95 / / / / 104

Figure 28. Goodness-of-fit plots of change in weight. Red lines are smoothing lines. Black lines are lines of identity. Blue dots are observed values. 105

Figure 29. Predicted and observed individual concentration-time profiles of change in weight in 100mg, 250mg and 500mg. 106

107

108

Figure 30. Visual predictive check for change in weight obtained from 1000 simulations. 5 th, 50 th and 95 th percentiles of observations (Black solid and dashed line). 95% confidence intervals of 5 th, 50 th and 95 th percentiles of simulated data (orange, green and blue areas). 109

Figure 31. Predicted change in weight over time during and after TC dosing based on 1000 simulations. Median (solid line), 95% CI (shaded area). (Left panel: Sperm counts, Right panel: Sperm motility). 110

Figure 32. Plots of Observations versus individual predictions for triglycerides, HDL, LDL, cholesterol and HCT. Red lines are smoothing lines. Black lines are lines of identity. Blue dots are observed values. 111

Figure 33. Visual predictive check for change in lipid profiles (Triglycerides, HDL, LDL, Cholesterol, HCT) obtained from 1000 simulations. 5 th, 50 th and 95 th percentiles of observations (Black solid and dashed line). 95% confidence intervals of 5 th, 50 th and 95 th percentiles of simulated data (orange, green and blue areas). 112