Milne.ruc.dk
- I, OM OG MED MATEMATIK OG FYSIK I, OM OG MED MATEMA Semi-Mechanistic Pharmacokineticand Pharmacodynamic Modelling
of a Novel Human Recombinant
Follicle Stimulating Hormone
Trine Høyer Rose Roskilde University Department of Science and Environment nr. 502 - 2016
DK - 4000 Roskilde Roskilde University, Department of Science, Systems and Models, IMFUFA P.O. Box 260, DK - 4000 Roskilde Tel: 4674 2263 Fax: 4674 3020 Semi-Mechanistic Pharmacokinetic and Pharmacodynamic Modelling of a NovelHuman Recombinant Folllicle Stimulating Hormone By: Trine Høyer Rose IMFUFA tekst nr. 502/ 2015 – 182 sider – A novel recombinant follicle stimulating hormone (rFSH), FE 999049, expressed in a cell- line of human fetal retinal origin (PER.C6 R ) is in development at Ferring Pharmaceuticals for controlled ovarian stimulation in infertility treatment to achieve functional oocytes for assistedreproductive technologies. In this PhD, population pharmacokinetic (PK) and pharmacody-namic (PD) modelling with nonlinear mixed effects models was used to analyse the PK and PDproperties of FE 999049.
The first PK model based on first-in-human single dose data with FE 999049 was a one- compartment model with a delayed absorption described by a transit compartment. A similarmodel was then used to describe the repeated dose pharmacokinetics of FE999049 observed aftermultiple dosing in a second phase I study. The model differed by having an endogenous FSHcontribution to the total measured FSH concentration. Furthermore, progesterone baseline levelshad an inhibitory feedback effect on the estimated endogenous FSH baseline value. Over time theinhibin B levels further suppressed the endogenous FSH production. In a semi-mechanistic PKPDmodel, the rFSH, endogenous FSH, and inhibin B concentrations were modelled simultaneouslywith total FSH stimulating the inhibin B production rate, and inhibin B levels inhibiting theendogenous FSH production.
Body weight was a covariate explaining some of the variation in apparent clearance and apparent volume of distribution in all models. Through simulations it was found that havingbody weight as a covariate at the PK parameters resulted in an overall decrease in drug exposureand inhibin B response with increasing body weight.
In this work it was furthermore identified that endogenous FSH levels have to be considered in the model else parameters can become biased and total FSH exposure is under-estimated inthe model, especially at baseline. Additionally, this thesis demonstrates it is of importance toaccount for the hormone dynamics and negative feedback from the ovarian hormones in orderto accurately describe the change in endogenous FSH levels over time. The standard method ofbaseline correcting data might therefore not be adequate as it will result in an overcorrection atlater time points since the endogenous level decrease during treatment.
Semi-Mechanistic Pharmacokinetic and Pharmacodynamic Modelling of a Novel Human Recombinant Follicle Stimulating Hormone Dissertation for the degree of Doctor of Philosophy in applied mathematics Semi-Mechanistic Pharmacokinetic and Pharmacodynamic Modelling of a Novel Human Recombinant Follicle Stimulating Hormone Trine Høyer Rose Department of Science, Systems, and Models Roskilde University Experimental Medicine Ferring Pharmaceuticals A/S International PharmaScience Center Semi-Mechanistic Pharmacokinetic and Pharmacodynamic Modellingof a Novel Human Recombinant Follicle Stimulating Hormone PhD dissertation c 2015 Trine Høyer Rose All graphics included were made by Trine Høyer Rose Front cover: Silhouettes from Ferring Pharmaceuticals Viggo Andreasen, PhD, Associate Professor (Chairman of the Committee)Department of Science, Systems, and ModelsRoskilde University, Denmark Professor Piet Hein van der GraafAcademic Center for Drug ResearchLeiden University, The Netherlands Professor Dr. Franz KappelInstitute for Mathematics and Scientific ComputingUniversity of Graz, Austria Professor Johnny T. OttesenDepartment of Science, Systems, and ModelsRoskilde University, Denmark Experimental MedicineFerring Pharmaceuticals A/S, Denmark Lars Erichsen, PhDExperimental MedicineFerring Pharmaceuticals A/S, Denmark With an interest in applied mathematics, in particular modelling within humanphysiology, pharmacokinetic (PK) and pharmacodynamic (PD) modelling in drugdevelopment was the ideal topic for my PhD work. The project was an industrialPhD as a collaboration between the department of Science, Systems, and Modelsat Roskilde University and Experimental Medicine in Ferring Pharmaceuticals. Itwas a privilege having both an academic and an industrial affiliation and veryinteresting working in these two different research environments.
Data from three clinical trials with a recombinant follicle stimulating hormone (rFSH), FE 999049, for use in infertility treatment were provided by Ferring phar-maceuticals for analyses and development of population PK and PKPD models. Asmodelling is an interdisciplinary field the optimal results are achieved with knowl-edge from all the fields as well as expertise in software for model implementationand data analysis. This PhD thesis demonstrates how the multiple disciplinesare needed for understanding the system of interest and developing mathematicalmodels. Furthermore, the focus is on how these models can be used in acquir-ing knowledge about an investigational medicinal product in development and themodels potential use in simulations and for predicting drug response.
The thesis starts with a general introduction to the fields and the overall back- ground of infertility, the drug development process, and need for modelling andsimulation (Chapters 1-2). In the next chapters the more extensive theory be-hind this work is covered. The underlying physiology for the female reproductiveendocrine system is described in Chapter with an outline of the major hor-mones involved, their interaction, and role in follicular development. Chapter gives a historic perspective of the development in gonadotropin therapy and FSHcompounds. The chapter also provides an overview of the important factors in-fluencing the PK and PD properties of FSH compounds as well as markers forovarian response that have potential in predicting treatment outcome and indi-vidualising treatment. This is important knowledge in determining what relationsand covariates are relevant to include in the modelling and for dose selection.
General methods for population data analysis and development of nonlinear mixed effects models are presented in Chapter A review of current mathematicalmodels with FSH products is given in Chapter which leads to what is relevantto investigate in this PhD and the specific research objectives in Chapter Chapter lists the three clinical trials with generated data and analysis meth- ods used in this PhD for development of the PK and PKPD models. The finalmodels and parameters are reported in Chapter along with illustrations of theresults and use of the models in simulations. Lastly the thesis is rounded of witha discussion (Chapter and conclusions (Chapter of the findings from this work and put into perspective of the current available knowledge.
The research resulted in three manuscripts, one for each of the clinical trials, included at the back of the thesis: Population Pharmacokinetic Modelling of FE 999049, a Recom-binant Human Follicle Stimulating Hormone, in Healthy Womenafter Single Ascending DosesRose, T.H., R¨ oshammar, D., Erichsen, L., Grundemar, L., and Characterisation of Population Pharmacokinetics and Endoge-nous Follicle Stimulating Hormone (FSH) Levels after MultipleDosing of a Recombinant Human FSH, FE 999049, in HealthyWomenRose, T.H., R¨ oshammar, D., Erichsen, L., Grundemar, L., and Semi-Mechanistic Pharmacokinetic-Pharmacodynamic Modellingof Inhibin B Levels after Multiple Doses of the Recombinant Hu-man Follicle Stimulating Hormone FE 999049 in Infertile WomenRose, T.H., R¨ oshammar, D., Karlsson, M.O., Erichsen, L., Grun- demar, L., and Ottesen, J.T.
This industrial PhD was financed by Ferring Pharmaceuticals with funding fromthe Innovation Fund Denmark under their 3-year industrial PhD programme.
Infertility is an increasing worldwide problem and it can be devastating for a couplenot being able to conceive a child on their own. Controlled administration of re-combinant follicle stimulating hormone (rFSH) can be used in infertility treatmentto achieve functional oocytes for assisted reproductive technologies. A novel rFSH,FE 999049, expressed in a cell-line of human fetal retinal origin (PER.C6 R development at Ferring Pharmaceuticals. Drug development is a long and costlyprocess, it is therefore important to extract and utilise all possible informationfrom clinical trials.
In this PhD, population pharmacokinetic (PK) and pharmacodynamic (PD) modelling of data from three clinical trials with FE 999049 was used to analyse thePK and PD properties of FE 999049. A population approach with nonlinear mixedeffects models was used in order to facilitate identification of variation betweensubjects and causes hereof.
The first PK model based on the first-in-human single dose data with FE 999049 was a one-compartment model with a delayed absorption described by a transitcompartment. Body weight was a covariate explaining some of the variation inapparent clearance (CL/F) and apparent volume of distribution (V/F). A similarmodel was then used to describe the repeated dose pharmacokinetics of FE999049observed after multiple dosing in a second phase I study. A few adjustmentswere made from the first model according to differences in study designs. Themodel differed by having an endogenous FSH contribution to the total measuredFSH concentration. Furthermore, feedback from ovarian hormones at endogenousFSH levels were identified and accounted for in the model. Progesterone baselinelevels had an inhibitory effect on the estimated endogenous FSH baseline value.
Over time the inhibin B levels further suppressed the endogenous FSH production.
Body weight was confirmed as a covariate at CL/F and V/F. The interactionbetween rFSH, endogenous FSH, and inhibin B was further investigated in a semi-mechanistic PKPD model. Modelling the hormone concentrations simultaneouslyfacilitated incorporation of continuous stimulation and feedback over time. Themodel adequately described the hormone dynamics with total FSH stimulatingthe inhibin B production rate, and inhibin B levels inhibiting the endogenous FSHproduction.
Through simulations of different patient and dosing scenarios it was found that having body weight as a covariate at the PK parameters resulted in an overalldecrease in drug exposure with increasing body weight. A doubling of weightrequired almost a doubling of dose to get same exposure. Consequently, with FSHstimulating inhibin B production, the inhibin B response was lower for higher body In this work, it was furthermore identified that endogenous FSH levels have to be considered in the model else parameters can become biased and total FSHexposure is under-estimated in the model, especially at baseline. Additionally,this thesis demonstrates it is of importance to account for the hormone dynamicsand negative feedback from the ovarian hormones in order to accurately describethe change in endogenous FSH levels over time. The standard method of baselinecorrecting data might therefore not be adequate as it will result in an overcorrectionat later time points since the endogenous level decrease during treatment.
A simultaneous dose-concentration-response model for inhibin B levels can be of value in quantifying and predicting ovarian response in clinical studies andclinical settings since inhibin B is an indicator of follicular growth and the earliestmeasured ovarian hormone response to FSH treatment.
e (Danish summary) Fertiliteten falder med alderen, og med en generel tendens til at kvinder bliverældre inden de f˚ ar børn, er der et øget behov for fertilitetsbehandling. Det kan være ardt psykisk pres ikke selv at være i stand til at blive gravid naturligt og skulle a hormonbehandling. Derfor er der behov for nye hormonpræparater med innovative behandlingsforløb tilrettet den enkelte patient.
Ferring Pharmaceuticals udvikler et rekombinant follikelstimulerende hormon (rFSH), FE 999049, fra en human cellelinie (PER.C6 R ), der skal injiceres sub- kutant til kvinden for at modne flere æg til brug i reagensglasbefrugtning (IVF)og mikroinsemination (ICSI). I denne PhD analyseres FE 999049's egenskaberog variation i resulterende koncentration og effekt ved hjælp af populations far-makokinetisk (PK) og farmakodynamisk (PD) modellering af data fra tre kliniskeforsøg.
Den første model baseret p˚ a det første humane datasæt fra enkelt dosering af FE 999049 i raske frivillige var en en-kompartments PK model med en transit-kompartment i absorptionsprocessen for en forsinket optagelse. Det var nødvendigtmed nogle f˚ a tilpasninger af modellen før den kunne bruges til at beskrive FE 999049 farmakokinetikken efter multiple doser i det andet fase I studie p˚ af forskellene i forsøgsdesignet. Dette inkluderede et endogent FSH bidrag til densamlede m˚ alte FSH koncentration. Den endogene FSH startværdi inden dosering af FE 999049 var negativt korreleret med progesteron startværdien. Inhibin Bhæmmede yderligere FSH produktionen over tid via en negativ feedback loop. Tilsidst var en semi-mekanistisk PKPD model udviklet fra data fra et fase II multipledosis studie for nærmere at undersøge forholdet mellem FE 999049, endogent FSHog inhibin B respons. Hormon koncentrationerne blev modelleret simultant for atmuliggøre beskrivelsen af hormondynamikken over tid med FSH, der stimulererinhibin B produktionen i ovarierne, og inhibin B der hæmmer den endogene FSHproduktion i hypofysen.
aende i de tre modeller var kvindens kropsvægt en statistisk sig- nifikant kovariat p˚ a parametrene clearance og fordelingsvolumenet. Modellerne blev brugt til at simulere FSH koncentrationen og inhibin B respons i kvinder medforskellige kropsvægte. B˚ ade FSH og inhibin B koncentrationen faldt med stigende Resultaterne i denne PhD indikerer, at det er vigtigt at inkludere endogent FSH i modellering og tage højde for den endogene hormondynamik, der for˚ variation i koncentrationen over tid, ellers kan der forekomme bias i parameter-estimaterne, og FSH koncentrationen bliver underestimeret. Dette s˚ om hvorvidt standardmetoden med baseline korrektion for et endogent substans e (Danish summary) er valid, da FSH afviger betydeligt fra baseline værdien over tid.
Inhibin B er en indikator for follikeludviklingen og er den tidligste respons variabel, der er blevet m˚ alt efter FSH behandling. Derfor kan PKPD modellen være brugbar i forudsigelsen af ovariernes respons, idet inhibin B bliver prædikteretsom en kontinuer variabel over tid simultant med FSH koncentrationen ud fra dosis.
I am deeply thankful to my supervisors who have made this PhD possible: Johnny T. Ottesen for supporting me from the beginning and giving me expert academic advice. We have had many mathematical discussions that often couldextend into interesting areas even outside the scope of this thesis. Thank you foryour commitment and constructive input and ideas.
Lars Erichsen for introducing me to the great field of population PKPD mod- elling and giving me a variety of exciting research possibilities. Thank you forguiding and supporting me and sharing your expert modelling knowledge.
oshammar for taking over after Lars Erichsen and bringing new great ideas to the project. Thank you for believing in me and giving me excellent feed-back. I am most grateful for that you in a busy work day still prioritised tomotivate and support me, and especially for continuing doing so even after youleft Ferring.
I owe a special thanks to Mats Karlsson for welcoming me to Uppsala and supervising me for three months at first as well as extra visits throughout myPhD. Also for inviting me to most interesting courses and for further answeringmy questions continuously over email. I want to express my sincere gratitude forall your help, your input to the PKPD model has been invaluable.
I extend my thanks to the rest of the pharmacometric group in Uppsala for all your kindness and for including me in everything from day one, both in all socialand academic events. You are truly an amazing group of people, kind and warm,and extremely talented research scientists and modellers.
To my fellow Ph.D. students and colleagues at Roskilde University, thank you for sharing knowledge and common PhD issues, and most importantly for greatsocial times.
I would also like to thank my co-workers in Experimental Medicine, Ferring, as well as people in the rFSH project group, who have been kind to answer anyquestions I would have, and review my articles and chapters in the thesis. Youhave been most helpful, thank you.
A warm thanks to my family and friends for never loosing faith in me but always encouraging me and for understanding my absence in the busy periods of my PhD.
To my boyfriend Tobias, thank you so much for your patience and support, for always making me smile and being able to cheer me up even after really hard andlong working days. I love you.
Antral follicle count Assisted reproductive technologies Below quantification limit Chinese hamster ovary Combined oral contraceptives Empirical Bayes estimates European Medicines Agency U.S. Food and Drug Administration First order estimation First order conditional estimation Follicle stimulating hormone Gonadotropin releasing hormone Human chorionic gonadotropin Human menopausal gonadotropin Human pituitary gonadotropin In vitro fertilisation Intracytoplasmic sperm injection International units Luteinizing hormone Lower limit of quantification Model-based drug development Modelling and simulation Nonlinear mixed effect modelling Objective function value Pearl speaks NONMEM Recombinant follicle stimulating hormone Relative standard error Visual predictive check World health organisation Challenges in drug development are to find a potent compound, establish its ef-ficient and safe dose with a minimum of side effects, while keeping costs down.
Regulatory authorities outline guidance documents with recommended approachesfor drug development to assist the industry, as well as setting high demands andrequirements for every step in the development process. This includes how to findfirst-in-human safe starting dose guidance on study design, data analysis,and reporting of results To get a new drug approved, substantial evidencefor its safety and beneficial effect is thus required. To obtain this, drug develop-ment becomes a long, expensive, and cumbersome process. It is therefore essentialto optimise the design and interpretation of clinical trials as well as extracting asmuch information as possible from the resulting data about the safety, tolerability,and effect of a new investigational medicinal product. An important tool for thisis mathematical modelling and simulation.
When a drug enters the body it is absorbed, distributed throughout the body, and in the end cleared from the system by metabolism or excretion. These phar-macokinetic (PK) processes determine the drug concentration in the blood, whichalso is called drug exposure. The drug concentration depends on the dose andvaries over time. In addition, variation between patients occur due to person spe-cific factors. This can be anything from body weight, age, and gender to morecomplicated influences like disease pathology, hormone levels and other intrinsicfactors, as well as extrinsic factors such as concomitant use of other drugs, smok-ing, or food intake. It is crucial that the drug concentration is high enough to givethe intended drug effect over time but still without causing severe side effects. Thepharmacodynamics of a drug is the effect caused by the drug, both beneficial andadverse. The drug effect or response to the drug is registered using pharmacody-namic (PD) endpoints which can be biological markers in the body, clinical efficacymeasures, or clinical outcomes including unwanted side effects. There might be adelay before an effect can be measured or observed if it is the result of receptorbinding and intracellular signalling cascades, or if the clinical outcome evolve overtime, e.g. reducing tumour size.
Mathematical models can describe and estimate the drug concentration over time (PK profile) for a given dose and identify factors (covariates) influencing theresults. In a complete PKPD model the PK model for the concentration-timeprofile is related to observed effects in form of PD endpoints for a description of dose-concentration-response and their combined time course. A population ap-proach with nonlinear mixed effects modelling provides specific information aboutindividuals relative to the population mean, types of variability, and opportunityof examine and perhaps quantify potential covariates' influence at the variability.
Appropriate models satisfactorily describing the observed data can then be usedfor simulation to predict patient outcome, estimate the optimal dose for individualsbased on the most significant factors, or to evaluate different study designs.
This PhD thesis concerns population PK and PKPD modelling of FE 999049, which is a recombinant follicle stimulating hormone (rFSH) expressed in a cell-line of human fetal retinal origin (PER.C6 R ). It is under development at Fer- ring Pharmaceuticals to be used for controlled ovarian stimulation in order toinduce multiple follicular development in women undergoing in vitro fertilisation(IVF)/intracytoplasmic sperm injection (ICSI) treatment. This quantitative workaims at further characterising the PK and PD properties of FE 999049 and vari-ation in drug exposure using nonlinear mixed effects modelling. Furthermore, theinfluence of endogenous hormone levels and covariates will be investigated in orderto facilitate simulation and prediction of infertility treatment outcomes.
Infertility is an increasing health issue and there is a need for better treatments tocatch the great diversity in causes and types of infertility. There are some marketedproducts for infertility treatment but there is still a medical need for new drugswith innovative treatment protocols. To market a new drug it is often desiredit should be superior to existing ones or present new opportunities for treatmentstrategies. Model-based drug development is a useful approach for assessment ofa new drug's potential, dose selection, and study design considerations both ininternal decision making and in regulatory interactions.
The world health organisation (WHO) has defined primary infertility as the in-ability to conceive within two years of trying to obtain a pregnancy, and secondaryinfertility as inability to become pregnant again after having an earlier birth Clinicians often define infertility as inability to become pregnant after one year oftrying The infertility prevalence varies between countries and regions. In addition, the different definitions and whether all women, married women, or only child-seekingwomen are used as basis for the calculation, adds to the difference in reportedprevalences. Some estimate that 1 out of 10 couples suffers from infertility, whichworldwide is equivalent to more than 80 million people Another article hasstated that it is 80 million couples rather that individuals A large systemicanalysis of health surveys from 1990 to 2010 accessed infertility, measured asnot having a live birth over a 5-year exposure period. They found that the overallrates in those years were the same but due to population growth the absolutenumber of infertile couples was 48.5 million in 2010, an increase of 6.5 millionsince 1990. The secondary infertility increased with age from 2.6% in child-seekingwomen of age 20-24 years to 27.1% for age 40-44 years, whereas primary infertilityhad a mean value around 2.5% and decreased slightly with age.
With the clinical definition, 15% of couples are unable to conceive within the first year, constituting in the US alone at least 6 million infertile couples (ReviewBased on data from 2198 infertile couples the cause of infertility was due tofemale ovulation disorders in 17.6% of the cases while 25.6% had unexplained infer- tility. Treatment options for these infertility types include hormone therapy, sincethe cause is not ovarian failure and the ovaries are still responsive to stimulation.
The other registered infertility factors were tubal disease (23.1%), endometriosis(6.6%), luteal phase, cerivical, and uterine defect (3.2%), and male factors (23.9%).
A Dutch study with 726 couples found that 25.9% and 30% had ovulation disordersand unexplained infertility, respectively In this study 12.9% of the coupleshad infertility caused by tubal disease, 3.6% by endometriosis, 27.7% by cervicalfactor, and 30% had male infertility factors. They experienced that the mean ageof females visiting an infertility clinic increased from 29.1 years in 1985-1993 to31.2 years in 2002-2003 Similar trends were observed at a Dutch fertility clinicIn 1985, 256 women visited the clinic with a mean age of 27.7 years, by 2008these numbers had increased to 594 women and 31.4 years, respectively. A furtherindication of an age shift was that the proportion of women ≥ 35 years increasedfrom 7.9% to 31.2%. Even though these results are from The Netherlands, thestudies refer to that the trend of delaying childbirth is observed in most developedcountries, so this should be representative for other western countries.
In developing countries the prevalence is general higher but also with great variations between regions and countries. In a report by WHO it is estimatedthat overall more than 1 out of 4 married women in developing countries areinfertile adding up to 186 million women whereof 18 million had primary infertility.
A similar trend of an increase in secondary infertility with age was observed buthere reaching 62% of women trying to become pregnant at age 45-49 years. Thereproductive age range in this report is 15 to 49 years, thus wider than the others,and is the reason for that the reported number is larger than the worldwide number.
The number also include 17% self-reported infecundity from women who havenever menstruated or not menstruated for at least 5 years and postmenopausalwomen. WHO compared the estimate with previous data and opposed to westerncountries infertility has overall decreased over time. Sexual transmitted diseasescould potentially affect infertility but WHO found no trend with extent of HIV.
A possible explanation for the decline could be due to the progress within assistedreproductive technologies (ART) and easier accessibility for women to infertilitytreatment leading to an increase of women seeking help.
Definitions of infertility and thereby prevalence varies but one thing is certain, infertility is a major worldwide problem and for those affected it can become a lifecrisis with psychological pressure and a great loss. The age of women starting tohave children have increased and as fertility decreases with age, the demand forinfertility treatment also increases. The cause of infertility varies greatly and up to30% have unexplained infertility, so one kind of treatment do not fit all. Hormonetreatment can be used for ovulation disorders, which accounts for around 20%of the cases, and if personal factors are taken into account for individualising 2.2 Drug Development treatment, success rates may increase. In unexplained infertility where no damageto ovaries are found, individualised hormone treatment could possibly be useful fortiming of mature oocytes and maybe diminish the psychological stress factor whengetting personal treatment. Hence, hormone therapy can potentially be used in50% of all infertile cases. A new hormone compound with an individualised specifictreatment protocol have thus potential for being marketed.
Development of a new drug starts with a drug discovery phase where targetsand possible therapies are identified through research based on knowledge of thetherapeutic area and disease pathology Thousands of compounds arebeing tested in order to find candidates of therapeutic value. The most promising,could be several hundred, compounds are selected for further preclinical in vitrolaboratory tests and in vivo animal testing to assess the drug's safety profile andeffect. It also has to be considered if the candidate drug can be developed at alarge scale for marketing. At the end only a few compounds enter clinical trials.
In Phase I clinical trials doses are administered to healthy volunteers to deter- mine whether the drug is safe in humans, as well as investigating its pharmacoki-netic properties. The potential dose range is examined in single ascending dosestudies with sequential groups given gradually increased doses starting from a lowdose in the first group. A safe dose can be tested in a multiple dose study forfurther assurance of safety and effect after repeated administration. Several trialscan be performed to test the drug in different population groups. Phase I studiestypically include in the order of 20 to 100 subjects Phase II clinical trials continue to evaluate safety but overall focus is on effec- tiveness in patients, PKPD relationship, and short-term side effects. This phase isoften divided into two parts: a smaller phase IIa proof of concept study in about100 patients to assure the potential of the drug, and a larger phase IIb study(100-500 patients) where an optimal dose is established. If successful in provingan effective dose with acceptable side effects the drug can proceed into phase III.
Based on the results and evaluations, a large scale phase III trial in a diverse tar-get population group (thousands of patients) is undertaken to establish significantconfirmatory evidence of drug safety, efficacy, and verify optimal dose of the drugin all subpopulations. More than one phase III trial can be required The overall timeline to get regulatory approval for a new drug is 10-15 years.
All along the way communication with regulatory authorities, review boards, ethiccommittees, and local authorities occur for monitoring and approval of trials toensure all requirements are fulfilled, and most importantly that trial participantsare not exposed to unnecessary risks. After final approval and marketing of a drug, additional studies are carried out with the purpose of continuing documentingbeneficial and adverse effects, since when marketed a much larger population isexposed to the drug. A postmarketing phase IV trial can also be conducted toinvestigate long-term safety or effect in a specific subgroup of patients All these precautions and stepwise evaluations through the trials also aim to kill non-effective and non-safe drugs as early as possible, but some drugs are stillterminated late in the development process. To get one drug approved, money andtime have thus been spend on thousands of other compounds, and even clinicaltrials for other candidates that ended up being discarded. There is not one numberfor what the average amount of money spend in total on research and developmentis before one new drug is approved. Different cost estimates and trends over theyears were listed in a large systemic review of published articles from 1979 to2010 that gave estimates of drug development costs. They operated with two typesof costs: cash and capitalised costs. The cash number is the actual costs spend onresearch and development for one new drug to reach the market. Capitalised costsinclude in addition opportunity costs, which is the amount of money that couldhave been earned from investing the actual cost elsewhere instead of spending themat research and development. Wide variation was observed and an overall increasein costs over the years from 92 to 883.6 million US$ cash with a correspondingcapitalised cost estimate of 161 and 1799 million US$, respectively.
The U.S. Food and Drug Administration (FDA) reported in 2004 the number to be between 0.8 and 1.7 billion US$ In 2009 estimates of drug developmentand marketing costs for one drug was estimated to be in the range 1.3 - 1.7 billionUS$. That is approximately a doubling of the costs since 2003 It is primarilythe costs of clinical trials that are increasing. These ranges might be for capitalizedcosts if following the above defined numbers, but it is not mentioned what theseestimates include.
In the beginning of this century the pharmaceutical industry increased research and development and was successful in marketing many of the new drugs discoveredWith several drugs at the market it became harder to develop and prove thebenefits of a new drug to those already marketed. Hence larger trials with morepatients had to be conducted in order to establish if there is a small improvementin the new drug compared to old drugs.
It also seemed like a maximum for the success rate of drugs that entered clinical investigation was reached. The success rate had increased from 12% for drugs beingdeveloped in the 1960s and 70s to 24% for drugs entering clinical trials in the 1990s.
It then decreased to 11.7% again in 2010 (Review The regulatory demandshave also been increased causing a longer development process and sometimesrequires extra or bigger trials making it harder and more expensive to market adrug. These criteria cannot be circumvented, but a way to lower costs is catching 2.2 Drug Development and stopping non promising drugs as early as possible and before entering morecostly late stage development phases In addition, costs may be reduced byexpediting the clinical development program by more efficient use of the generateddata and knowledge about the molecule.
The problem with these challenges, inefficiencies, and rising costs in drug de- velopment has also been addressed by FDA in 2004 In the report it is stressedthat there is an urgent need for improving the drug development path. Scientificresearch and collaboration across fields and institutions are essential for buildingknowledge and finding better tools and technologies for all levels in drug develop-ment. As a toolkit for better effectiveness in data management and analysis, FDAproposes model-based drug development (MBDD) as an important approach thatpotentially can improve the process significantly.
Model-Based Drug Development It was not without challenges to apply modelling and simulation (M&S) to drugdevelopment. For proper implementation was needed software, knowledge, andsufficient quality data with the right measurements of interest. Another challengewas to convince people it was worth using resources and time at this complicatedinterdisciplinary task, and incorporate it in clinical trial protocols and reports Population modelling was, in spite of implementation difficulties, reviewed to beof great value in every phase of drug development to obtain important informa-tion about PK and PD properties, and provide insight in individual response andsubgroups at risk with need for different dosage regimen.
In 1997, Lewis B. Sheiner introduced the concept of learn-confirm cycles to the drug development phases for a more efficient and informative process Phase I trials are a first informative step for investigating quantitative questionsto gather knowledge and learn about the drug's properties. The phase I studydesigns also fit nicely in the learning scheme with either a wide range of dosesin a diverse population group for PK analysis or as a multiple dose study withseveral endpoints measured to get an initial indication of the pharmacodynamicsand toxic effect. What is learnt about safe and tolerable doses should then beconfirmed in a phase IIa study. After the first learn-confirm cycle the results areevaluated to see whether it was possible to properly confirm satisfactory efficacy,and on this basis judged if the drug is qualified to proceed in further development.
The second learn-confirm cycle consists of phase IIb (dose finding) and phaseIII/IV (confirm dosage regimens). If the benefit/risk profile is acceptable the drugis eligible for approval. Data from learning trials is analysed by a populationmodelling approach and used to design the confirmatory trials, where statisticaltests are performed with a null hypothesis of no efficient treatment effect of thedrug. Software and capable scientists to perform this model-based data analyses and simulations were in short supply at the time The paradigm did notpresent any single new design ideas, but a development framework including amore science oriented focus instead of mostly having a confirmatory motivationfor proving efficacy with the goal of approval. The need for model-based analysisto achieve a greater understanding through learning was emphasised.
This paradigm is comprised in MBDD except with a less sharp distinction of learning and confirming phases in two cycles, but rather a continuous view oflearning throughout all phases while confirming previous results The firststep in implementing models in drug development was moving from an empiricaldecision making to a model-aided development process, where models are usedonly occasionally and primarily in later phases for decision confirmation and sup-porting labelling The further transition to MBDD is difficult, but optimallymodelling should be applied throughout drug development. Starting in preclinicaldevelopment for evaluation and identification, and proceeding into translationalPKPD modelling for extrapolation and scaling to humans. In the clinical phasesmodelling should be used for analytical and predictive purposes as well as forpost-marketing surveillance. Pharmaceutical companies doing so may be able tosafely enter patient studies faster with increased success rate and at a lower cost.
Sheiner and Steimer (2000) gave an extensive list of examples of M&S in drugdevelopment proving that modelling was gaining impact.
The usefulness of modelling in phase I was not acknowledged initially, therefore an expert meeting was held to evaluate M&S in phase I Modelling is invalu-able in some phase I tasks like handling missing or censored data and describingcomplex exposure-biomarker relationship. In particular for sparse data, popula-tion modelling is the only way to achieve proper information and interpretationof data. Additionally, a population approach with nonlinear mixed effects mod-elling enable analysis and quantification of variability and covariate effects. Otheradvantages of mathematical models are they facilitate data integration from dif-ferent clinical trials, enable clinical trial simulations to test different study designs,predict results, and extrapolate results to future patients, dose selection, and aidin better informed development decisions of go, pause, or stop based on quanti-tative decision criteria of treatment effect Physiological-based modellingand mechanistic models that incorporate knowledge of the biological system andphysiological parameters are increasingly used. Mechanistic models quantitativelydescribe (some of) the processes from drug administration to effect, e.g. drugdistribution, receptor mechanisms, interactions with endogenous substances, andfeedback mechanisms. Hereby increasing prediction accuracy and improve extrap-olation properties of the model For successful implementation of M&S, PKPD modelling specialists, biostatis- ticians, and clinical research experimentalists have to work closely together with 2.2 Drug Development mutual trust and acceptance of challenges and contributions from each group. Ittakes time to overcome scientific differences between each group's custom methodsand language. Through tight collaboration with clear communication and inter-actions, an understanding of the benefits of MBDD can be achieved as well asconfidence in model-informed decisions and trials. When successful, the modelscan be used to support registration of the drug, and convince research and develop-ment management of M&S is a powerful tool for more efficient drug developmentPlanning trials properly is very important as with a wrong dose selection,subjects, duration, or time of sample measurements, the results can be insufficientfor analysis or conclusion. The trial will then have to be repeated with a differentdesign which is a costly affair. MBDD can thus help avoiding unnecessary extradevelopment costs. MBDD has even been used to skip phase IIa completely byutilising prior knowledge Using a MBDD approach can also reduce the num-ber of required patients and shorten the study duration To obtain this, itrequired a willingness from the company to change, a training program for the in-volved employees, and support from senior management. The effort was rewardedwith increased probability of success in the studies and a yearly cost reduction of$100 million.
Another important factor for extending the use of M&S is to communicate the results in an understandable way not only to the development team and man-agement, but also very importantly to the regulatory authorities. In 1999 FDAissued a guidance document on how and when to use population PK approach andhow protocol, data, and reports should be handled It was followed in 2003by a guidance on using PKPD modelling in exposure-response analysis Euro-pean Medicines Agency (EMA) released a guideline in 2007 for reporting resultsfrom population PK analyses and have since had numerous presentations andworkshops on M&S. The great usefulness of M&S was acknowledged by regulatoryauthorities, and hereby establishing M&S as an important tool in drug develop-ment. Today this is widely recognised, though still not implemented as a standardtool for every new drug.
Two recent papers evaluate whether the impact of M&S at drug de- velopment has been as great as expected. In their perspective to some extendit has, but it can become even greater. The papers point out that a problem isthat in some instances it may be difficult to internally convince some companiesof the value of a model based approach. A continuing problem is the lack of un-derstanding of the modelling results for the non-modelling-specialist members ofthe development team. This is due to the modellers lack of presentation skills. Itis emphasised that optimal usage of M&S can only be achieved through propertiming and team work from the beginning to ensure that all are on board with thegoals and what the purpose of the M&S is. Else the risk is that at the end the generated results from M&S are hard to believe for the rest of the team. It is alsoof uttermost importance that the modellers can present the results in line withthe goals and in an understandable way. In addition Bonate stress that thetrue MBDD revolution will not happen until the M&S results are being presentedoutside the clinical pharmacology field to make it more broadly understood andhereby get greater influence.
The Female Reproductive In order to satisfactorily describe the impact of the pharmacological drug therapyin a mechanistically correct hence useful model, it is necessary to understand theunderlying physiology and anatomy of the system of interest. A comprehension ofthe components and drug response factors in the system will also enable identifi-cation of relevant PD endpoints. Thus the reproductive endocrinology involvingFSH and follicular development will be presented in this chapter. At first the over-all structure and components will be introduced, thereafter more details about thehormonal interplay and development steps are described.
Anatomical and Physiological Overview The reproductive endocrine system consists of the hypothalamic-pituitary-gonadalaxis and involves several hormones from the brain and ovaries. Interactions be-tween the hormones occur at all levels through feedback loops, resulting in com-plicated dynamics which eventually leads to maturation of an oocyte and timedovulation. The major hormones involved in the dynamics and control of repro-ductive function in females are gonadotropin releasing hormone (GnRH) from thehypothalamus, follicle stimulating hormone (FSH) and luteinizing hormone (LH)from the anterior pituitary gland, and inhibin B, progesterone, and estradiol fromthe ovarian follicles. Many other hormones, neurotransmitters, and additionalcomponents are involved in the underlying mechanisms behind the effects butthese details are beyond the scope of this chapter. The overall connections in thereproductive system is illustrated in Figure The effects of the hormones willbe described in the following sections.
The hypothalamus is part of the diencephalon at the base of the brain and plays an essential role in neurohormonal control of the endocrine system by influ-encing the pituitary function (Review The pituitary gland, also calledhypophysis, is an extension of the hypothalamus consisting of an anterior and aposterior part with distinct functions. Magnocellular neurons in the hypothala-mus have long axons that extend to the posterior pituitary where neurohormonesare transferred and subsequently released for regulation of homeostasis and stim- 3 The Female Reproductive Endocrinology Ovaries Inhibin B Figure 3.1: A simplified diagram of the hypothalamic-pituitary-gonadal axis in the female withthe major hormones and feedback loops. GnRH: gonadotropin releasing hormone, FSH: folliclestimulating hormone, LH: luteinizing hormone. A + indicates a stimulatory effect and a - aninhibitory/supressive effect.
ulation of lactation and uterine contraction. A neural connection does not existbetween the anterior pituitary gland and the hypothalamus. Instead, hormones aretransported through blood vessels making up the the hypothalamic-hypophysealportal system. It originates in a capillary bed at the median eminence in thehypothalamus and is connected via long portal veins to a capillary bed in theanterior pituitary. Hypothalamic parvocellular neurons terminates at the medianeminence releasing neuropeptide hormones to be taken up by the portal system.
These hypophysiotropic hormones affect the anterior pituitary release of hormonesinvolved in growth, metabolism, lactation, stress, sexual differentiation, and repro-duction. The pituitary hormones exert their effect at endocrine glands and organsthroughout the body.
The principal target organs for pituitary hormones affecting reproduction in females are the ovaries, which contain follicles of different sizes. A growing follicleconsists of an oocyte surrounded by layers of granulosa cells and theca cells. Thefollicles and oocyte develop under tight hormonal control and usually only onefollicle per menstrual cycle reach full maturation. At ovulation the mature oocyteis released and transported to the uterus by the fallopian tube. The remaining celllayers of the ovulatory follicle become the corpus luteum, that secretes hormonespreparing the uterus for potential implantation of a fertilised egg and exert feed-back control to the brain. If the oocyte is not fertilised, menstruation occur andthe corpus luteum degenerate.
3.1 Anatomical and Physiological Overview Reproductive Hormones of the Brain In reproduction the hypophysiotropic hormone secreted by hypothalamic neuronsis the decapeptide GnRH (Review Upon reaching the anterior pituitaryvia the portal system GnRH binds to specific GnRH receptors and stimulate thesynthesis and release of FSH and LH. GnRH can enhance it's own effect by up-regulation of its receptors at the pituitary. GnRH is secreted in a pulsatile mannerwith changing frequency during the menstrual cycle in response to integrated neu-ronal signals and hormonal feedback as well as influence from other substrates.
The different frequencies is a way for one hormone to control different release oftwo hormones from the same gland and even the same cells. In general low GnRHpulse frequencies increase FSH secretion and higher frequencies favours LH secre-tion, but other hormones also affect the net production and secretion. Anotherreason for the pulsatile secretion is that the sensitivity of pituitary GnRH recep-tors decrease if exposed to constant levels of GnRH, and ultimately the pituitary isrendered unresponsive to GnRH. Therefore, controlled GnRH pulses are essentialfor normal FSH and LH secretion and thus reproductive function, since the majorfunctions of FSH and LH are to induce follicular development and ovulation bystimulating the ovaries.
FSH and LH are gonadotropins belonging to the glycoprotein hormone family that in addition include human chorionic gonadotropin (HCG) and thyroid stimu-lating hormone (Review Glycoproteins are heterodimers consisting of twoamino acid chains, an α- and β-chain. The α-chain is common to all glycoproteinsbut the β-chains differ which provide unique specificities, properties, and effectsof each hormone. The polypeptide subunits are glycosylated to different extentand there exist various isoforms of each glycoprotein giving rise to altered struc-ture and activity. The higher glycosylation the longer half-life but lower receptoraffinity than the more basic isoforms (Review FSH and LH are synthesised by gonadotrope cells (gonadotropes) of the an- terior pituitary gland. These cells are the only ones that express the genes forthe FSH and LH β subunits. An increased synthesis does not necessarily meanincreased hormone secretion as they can be stored in secretory vesicles and laterreleased by exocytosis. They are endocrine tropic hormones since they exert theireffect after transported in the circulation to the target organ, the ovaries. In theovary FSH and LH binds to cell surface receptors at follicular cells activating in-tracellular signalling leading to follicular growth, production of ovarian hormones,oocyte maturation, and ovulation.
The effect of FSH and the necessity of it in maturation of follicles was made clear in 1931 by Fevold et al. but studies investigating reproductive functionstarted even earlier. The first study indicating a role of the pituitary in gonadalfunction date back to over 100 years ago In the 1920's Zondek and colleagues 3 The Female Reproductive Endocrinology had extensively studied the anterior pituitary and hormones involved in ovarianfunction At the same time a different group demonstrated that the anteriorpituitary affect sexual maturity (Review It was generally believed that theanterior pituitary secretes two distinct hormones affecting the ovaries. Fevold etal set out an experiment to prove it and separated the two hormones fromthe anterior pituitary: a gonad stimulating factor and a luteinizing factor. Theyestablished their individual effects in rats. Giving the gonad stimulator on its ownresulted in growth of the follicles and was found to be similar to Prolan A - whichwas the name Zondek had given FSH. The luteinizing hormone could not stimulatethe ovaries on its own, but if given after the gonad stimulator hormone such thatgrowing follicles were present, the hormone caused luteinization and creation ofcorpus luteum. The hormones separated were FSH and LH and their overall effectswere thus established in 1931.
FSH is the key hormone in follicular development and growth. It stimulates the granulosa cells to increase FSH receptor expression, it induces LH receptorsat the granulosa cells, and increases progesterone, estradiol, and inhibin B pro-duction by increasing enzymatic activity (Review In addition to causingluteinization by a high LH surge and subsequently corpus luteum formation andmaintenance, LH stimulates production of estrogens and progesterone by the cor-pus luteum. Before ovulation, LH stimulates production of progesterone in thetheca cells as well as androgens, which are transferred to the granulosa cells forconversion to estrogens under stimulation by FSH. The theca cells cannot performthe final aromatisation of androgens to estrogens, since they lack the enzyme aro-matase responsible for the conversion. The granulosa cells, on the other hand,cannot synthesise androgens so these have to be received from the adjacent thecacells. Hence, both LH stimulation of the theca cells and FSH stimulation of thegranulosa cells are required for estrogen production. This is the two-cells-two-gonadotropin concept. Later in the follicular development when LH receptors areexpressed at granulosa cells due to FSH stimulation, LH acts in synergy with FSHat the granulosa cells too. Just before ovulation LH induce progesterone receptorsat the granulosa cells.
The hormones produced by the ovaries in response to gonadotropin stimulationinclude inhibin B, estradiol, and progesterone (Review Inhibin B is apeptide hormone consisting of two peptide subunits. In the same peptide familyare also inhibin A, activin, and follistatin, which are involved in regulation of FSHlevels. The major effect of inhibin B is its important role in inhibiting FSH syn-thesis and secretion via negative feedback to the pituitary. Furthermore inhibinB can both reduce the number of pituitary GnRH receptors and block the GnRH 3.2 Follicular Development induced up-regulation of GnRH receptors. Other effects of inhibin B include en-hancing the stimulating effect of LH at the theca cells for increased production ofandrogens. Inhibin B is produced by the small follicles and is thus an early markerfor follicular development The main source of inhibin B production is thegranulosa cells, but it is also produced in other tissues.
Estradiol and progesterone are steroid hormones derived from cholesterol by steroidogenesis. The process consists of different paths of alterations of cholesterolby enzymatic activity. After a few steps, progesterone is produced and can act asa precursor for androgens, which in turn can be transformed to estrogens. Pro-gesterone is secreted by both granulosa and theca cells, whereas estradiol is themain estrogen secreted only by the granulosa cells. They are smaller than peptidehormones and can thus enter the target cells for evoking their effect, which is bothparacrine, autocrine, and endocrine through feedback loops.
Progesterone is only produced in small amounts before ovulation but when high levels of estradiol are present, low levels of progesterone are involved in stimula-tion of LH and FSH secretion for the timing of high surges and ovulation. Afterovulation progesterone plays an important role in preparation of the uterus forimplantation and in a negative feedback signal at LH and FSH secretion. At thehypothalamic level progesterone suppress the GnRH secretion, and it reduce thepituitary response to GnRH.
Estradiol plays a significant role in escalating ovarian growth and cell prolif- eration. It acts on granulosa cells in a self-promoting manner by increasing thenumber of its own as well as LH and FSH receptors and by stimulating aromataseactivity. Besides local effect in the ovaries, estradiol is also involved in longer feed-back loops for regulation of gonadotropin production and secretion both direct andindirect. Low levels of estradiol have a negative feedback effect at FSH and LHsecretion, while high levels stimulate LH secretion. To exerts the positive feedbackat LH, highly elevated estradiol levels have to be sustained over a longer period.
The high estradiol level is essential in causing the ovulatory LH surge in multipleways. It amplify the GnRH effect at the pituitary, increase the number of GnRHreceptors, and up-regulate the intracellular signalling system in the gonadotropes.
Estradiol also induces progesterone receptor expression and hereby enhance theeffect of progesterone.
Follicular Development Follicular development is a continuing process happening from fetal life untilmenopause with both gonadotropin independent and dependent stages (ReviewOnly aproximately 400 out of the two million primordial follicles presentat birth will become fully mature and ovulate. A primordial follicle contains an 3 The Female Reproductive Endocrinology oocyte at a resting state with a single layer of spindle shaped pre-granulosa cellssurrounding it. At all times resting primordial follicles enter the pool of growingfollicles independent of gonadotropin stimulation. Likewise, oocytes are lost con-tinuously at any point in the development through atresia by apoptosis which isprogrammed cell death of the follicular cells. It takes several months for a follicleto reach a fully mature state.
In the first development step from a primordial follicle to a primary follicle the oocyte is slightly enlarged and the surrounding cells proliferate to cuboidal shapedgranulosa cells (Figure When several layers of granulosa cells are formedthe follicle develop into a secondary follicle. Surrounding stroma cells that haveacquired its own blood supply form an outer layer of the follicle which differentiateinto two theca layers. The theca externa contains smooth muscle and collagenand is involved in contraction causing the rupture of the follicle at ovulation.
The theca interna develop LH receptors and are the cells later producing ovarianhormones upon LH stimulation. At this stage the granulosa cells start to expressFSH and estrogen receptors and progress into a preantral follicle. The receptorsenables the follicle to respond to FSH and LH stimulation and to produce ovarianhormones. The preantral follicles represent the pool of available follicles that canbe recruited for further maturation by FSH stimulation. The following stages aregonadotropin dependent and do therefore only happen after the hypothalamus-pituitary connection is fully develop in puberty.
In between the granulosa cells, cavities start to form with follicular fluid con- taining hormones secreted by the granulosa cells as well as hormones like FSH fromthe circulation. In the early antral follicle these cavities eventually merges into anantrum creating a confined environment around the oocyte. With the follicularhormone production the growth is accelerated due to positive auto- and paracrinefeedback loops by estradiol. This include up-regulation of FSH and estradiol re-ceptors and consequently increased responsiveness of the follicular cells and pro-motion of further proliferation, growth, and hormone secretion. The antrum growsthroughout different stages of antral follicles and create a hormone rich environ-ment required for oocyte maturation. In the presence of both FSH and estrogen,LH receptors are induced at granulosa cells of large follicles. These follicles cantherefore also be stimulated by LH causing same effect as FSH in the granulosacell. Due to this extra source of stimulation and to the escalating self-stimulatinggrowth events caused by estradiol the most advanced follicle will gain further dom-inance. It hereby become the dominant follicle selected to continue maturation toa preovulatory follicle while less developed follicles undergo atresia. LH is crucialfor the final maturation of the dominant follicle as it ensures a supply of androgensfrom the theca cells required for estradiol production, and it supports stimulationof the granulosa cells to sustain growth and hormone production while FSH levels 3.2 Follicular Development Figure 3.2: Illustration of the follicular development from a primordial follicle to a maturepreovulatory follicle. The oocyte is initially surrounded by a single layer of granulosa cells. Inthe development the layers of granulosa cells increase, theca cells layers are added, and lastly theantrum is formed - a cavity with hormone-rich follicular fluid.
decline. Induction of LH receptors at granulosa cells are therefore essential in thepreovulatory follicle. Hence, selection of the dominant follicle to continue matura-tion of the final stages to ovulation depends on the follicle's estrogen content in thefollicular fluid, the induction of LH receptors, and high number of FSH receptors.
The granulosa and theca cells change in formation in the final step along withnumerous intracellular changes preparing the follicle for rupture. The LH surgeprime the oocyte of the preovulatory follicle to initiate completion of maturationand cause luteinization of the granulosa cells. With the release of the oocyte theremnants of the follicle form the corpus luteum.
A certain level of FSH is needed to recruit the cohort of preantral follicles forfurther development. This threshold theory was first proposed in 1978 by Brown(Review The FSH level needed to sustain growth is lower than thatrequired for initiating growth hence the controlled period with elevated FSH levelensures that only a small cohort starts to grow. There is a different threshold levelfor each follicle and it can change as the follicle grow and determine the faith ofthe follicle. Evidence suggest that the dominant follicle is more sensitive to FSH 3 The Female Reproductive Endocrinology and has a lower threshold than the others and thus survive a decline in FSH level,whereas the smaller follicles in gonadotropin dependent development will undergoatresia as the FSH level falls below their thresholds.
The duration of elevated FSH level above the threshold is also important.
A short elevation by exogenous administration of FSH does not interfere withthe selection of one dominant follicle, but a moderate elevation for a longer timecause multiple follicles to continue growing to the mature state The resultssupported that more mature follicles become more sensitive to FSH stimulationand thus gain dominance. The duration also affect the number of follicles initiallystarting to grow.
Therefore a FSH window rather than a threshold has been proposed (Review FSH levels and individual thresholds play an important role in not only re- cruitment of preantral follicles, but also selection of the dominant follicle, whichfollicles continue to grow, and which undergo atresia.
Menstrual Cycle and Hormone Dynamics The menstrual cycle lasts on average 28 days and consists of two phases of ap-proximately 14 days: the follicular phase and the luteal phase. Ovulation marksthe transition between the phases, and menstruation causes the cycle to start over(Review . Complicated hormone dynamics, as illustrated in Figure control the events throughout the cycle.
The follicular phase is initiated by an increase in FSH concentration that allow the group of preantral follicles to grow. As the growing follicles start to produceinhibin B, the FSH level decrease due to the negative feedback. Inhibin B ispredominantly secreted in the early to mid follicular phase by the small folliclesand thus decrease in the late follicular phase when more and more follicles becomeatresic. The onset of estradiol production starts later since secondary follicles withtheca cells have to be formed. Unlike inhibin B, estradiol production increaseswith the growing follicles. The dominant follicle produces estradiol in continuingincreasing amounts, which causes estradiol levels to rise rapidly. The negativefeedback of inhibin B at first and later by estradiol at FSH secretion adds to theadvancement of the dominant follicle and atresia of less FSH sensitive follicles. Onthe contrary, estradiol's effect at LH changes from inhibitory to stimulatory andconsequently LH levels begin to rise in the late follicular phase. The maturingpreovulatory follicle secrete low levels of progesterone that augment the positivefeedback of estradiol at LH production and cause the midcycle increase in FSH aswell. This burst of FSH ensures completion of implementation of LH receptors atthe granulosa cells and final maturation.
3.3 Menstrual Cycle and Hormone Dynamics Hormone concentration Figure 3.3: Illustration of the reproductive hormone concentration dynamics through the men-strual cycle.
The scales are arbitrary and the magnitudes are not comparable between the hormones only the dynamics relative to each other. FSH: follicle stimulating hormone, LH:luteinizing hormone, Prog: progesterone, Est: estradiol, Inh B: inhibin B.
The timing of ovulation is controlled by the maturing follicle and the levels of secreted steroid hormones by the follicle indicate when it is mature and ready forovulation. The low level of progesterone contributes to the timing of the LH surgethat cause ovulation and the beginning of the luteal phase. With the expression ofLH receptors at granulosa cells, LH takes over stimulating progesterone productionand changing the inhibin production from inhibin B to inhibin A. In addition theLH receptors enable LH to cause luteinization of the granulosa cells. Without FSHstimulation and rupture of the of the dominant follicle the estradiol levels decrease.
An inhibin B peak is observed just after ovulation and is most likely caused by theruptured dominant preovulatory follicle and release of its hormone-rich follicularfluid. After ovulation the remnants of the follicle is stimulated to progress intothe corpus luteum which is an endocrine gland producing high levels of estradiol,progesterone, and inhibin A under LH stimulation. Thus inhibin B levels are lowafter ovulation and hereby the suppressive effect at FSH is diminished. However,FSH concentrations are still low due to suppression from the increasing estradioland progesterone levels. Progesterone blocks the GnRH surges in the hypothala-mus and suppresses the LH release in the pituitary which causes the rapid declinein LH concentration after the surge. If no fertilisation of the ovulated egg occur 3 The Female Reproductive Endocrinology the corpus luteum degenerates and the progesterone and estradiol levels declineallowing FSH levels to increase in the late luteal phase. This increase in FSH isimportant for the initiation of the next cycle.
Gonadotropin Therapy Since the discovery of gonadotropins' role in reproduction, several sources andmethods for measuring and extracting the hormones have been tested and devel-oped in different directions. Purity, batch-to-batch consistency, tolerability, safety,high potency and activity were desired properties of the preparations for effectiveinfertility treatment. However, the method and source also needed to be eligible forlarge scale production, without being too complex or expensive. Several productsare marketed today but the search continues for better products and innovativetreatment strategies.
Cause of infertility varies greatly, it is therefore important to diagnose the patient to determine if the patient at all is eligible for gonadotropin treatment. Inaddition, it is preferable to find an optimal individual dosing scheme according topatient-specific factors to increase success rate in pregnancy. For this is neededreliable determinants of ovarian reserve and response to treatment. If a goodpredictor can be identified, there could be therapeutic value in adjusting individualdoses according to its level.
Thus knowledge of what hormones, factors, and personal demographics that have potential in diagnosing or predicting infertilityextend and response is necessary.
The Road to Recombinant FSHPreparations For nearly a century gonadotropins have been extracted from various sourcesincluding animal pituitaries, pregnant mares' serum, human urine, and humanpostmortem pituitary glands. When it became possible to identify, isolate, andassess the amount of the hormones, more could be learned about variations inhormone levels within and between individuals and causes hereof. In addition, thegonadotropins' effects could be tested when administered in animals, as well asin humans with intention of developing gonadotropin preparations for infertilitytreatment.
4 Gonadotropin Therapy Human Gonadotropins Animal preparations could be used in humans and successful treatments werereported, but formation of antibodies was also detected (Review It wassuggested that if treatment was carried out while carefully monitoring individualfollicular development and with an improved two phase method, animal productscould still be used With time several groups concluded that to avoid formationof antibodies, gonadotropins had to be of human origin Despite this,animal preparations were used for several decades before they were withdrawn(Review is mostly a reprint of .
Much was learned from the animal extraction methods that could be adapted to preparations of human gonadotropins. In 1949 Li et al. extracted FSH fromsheep pituitaries. When administered to rats it stimulated follicular developmentand showed no effect of any other hormones than FSH, suggesting purity of thepituitary FSH extract. It took another decade before a similar method was used toextract human pituitary gonadotropin (HPG) from human autopsy pituitaries asstarting material for purification and separation of FSH and LH A partiallypurified human pituitary FSH preparation with high activity showed promisingclinical effect in 7 female patients At that time it was the preparation withthe highest FSH activity It was possible to purify the preparation even further to get an activity over 2000 times higher than urinary derived standardpreparations In the following years several successful studies in women wereconducted with HPG but as postmortem human pituitary glandswere only available in limited amounts, HPG could not be produced on a largescale. It was used for ovarian stimulation for 30 years, but after several reportedcases of Creutzfeldt-Jakob disease HPG was withdrawn from the market (Review Another hormone used in gonadotropin therapy is human chorionic gonadotropin (HCG) extracted from human placenta, pregnant women's urine or produced byrecombinant technologies. As early as 1931, only a few years after its discovery,the first HCG product was marketed (Review It is used as a luteinizingfactor and does not contain any FSH activity, hence HCG has no effect on theovaries unless FSH is present or prior treatment with an FSH preparation hasoccurred HCG is still used today in combination therapy to induce ovulationand sustain corpus luteum.
Human Menopausal Gonadotropins Methods to extract gonadotropins from human urine date back to the 1930s butthey were tedious and expensive with varying yield and content. The first uri-nary preparations had very little activity and were toxic to laboratory animals 4.1 The Road to Recombinant FSH Preparations (Review Therefore, before gonadotropin preparations could be adminis-tered to humans it was necessary to develop methods for hormone extraction andpurification from human sources yielding safe and efficient preparations.
The three major urinary extraction methods for gonadotropins were alcohol- precipitation method kaolin-adsorption method and ultrafiltration When the kaolin-adsorption method was combined with chromatography greatimprovements were achieved And when further treating the urine extractswith tri-calcium phosphate the toxicity could be reduced Urine from bothmales and females of different ages were used.
With both ultrafiltration and alcohol-precipitation methods it was possible to quantify the hormone amountsin urine to differentiate between normal levels, hyper- and hypogonadotropic syn-dromes Menopause is one of the hypogonadal syndromes with increased gonadotropin excretion into the urine. After administration to both intact andhypophysectomised immature rodents it was also observed that extracts frompostmenopausal women's urine have higher gonadotropin activity than urine fromnormal females In addition it became clear that human menopausal go-nadotropin (HMG) contain both FSH and LH activity Postmenopausalwomen's pituitary glands and urine were also reported to have higher FSH contentcompared to other ages and males Based on these findings it was concludedthat extracting HMG from post-menopausal women's urine give higher yield ofgonadotropins and is thus the best urinary source for gonadotropin preparations(Review Several of the urinary extraction methods were discussed to be used for a stan- dard HMG reference preparation, which was needed for comparison in developmentof new and hopefully better methods. Not only one, but several standard prepa-rations were established and research groups used different standard preparationsas reference A clear comparison between studies was therefore still notpossible but it did add some to a more systemic review of preparations. Newpurification methods were developed to get overall substantially higher activitythan the references and obtain a purified HMG safe for human injections Comparative studies of different HMG preparations suggested that both a certainlevel of FSH and FSH/LH ratio was needed for an appropriate ovarian response In the 1960s the clinical use of HMG, often in combination with HCG, was extensive. Administration of HMG in women with amenorrhoea or anovulationinduced ovarian response and ovulation with numerous pregnancies following treat-ment These studies confirmed FSH as being the important therapeuticfactor for clinical effectiveness. In the first HMG preparations FSH and LH onlyaccounted for 5 % of the total protein content. In addition, FSH exists in the bodyin different isoforms and the proportion of isoforms changes over time causing vary- 4 Gonadotropin Therapy ing isoform profiles in the urinary extract (Review Hence these preparationshad low bioactivity and high batch-to-batch variability with large amounts of uri-nary proteins and unidentified substances potentially causing side effects Asa consequence, HMGs were only available for intramuscular (i.m.) administration.
Further development in the fabrication techniques and by applying methods used at pituitary tissue enabled separation and achievement of a partially puri-fied FSH from postmenopausal urine Starting with existing preparations andapplying different methods Donini et al. managed to increase potency andFSH/LH ratio, and by binding and removing LH with chromatography they at-tained an apparently pure FSH preparation At last was achieved a highlypurified FSH product with higher content of FSH (about 95 %) almost withoutany polluting proteins As a result of the lower contaminating content,risk of injecting site reactions were reduced, the bioacivity became higher, thuslower amounts needed to be injected and subcutaneous (s.c.) injection was madepossible.
In 1989 advances in DNA technologies enabled production of recombinant FSH(rFSH) from Chinese Hamster Ovarian (CHO) cell lines This controlled fab-rication process eradicate all non-specific proteins normally found in the urine,rendering a pure FSH preparation completely deprived of other gonadotropins aswell. Adverse effects and allergenic reactions were thus reduced making it suitablefor s.c. administration. In addition, specificity and batch-to-batch consistencyincreased and the source is sustainable hence can meet any market demands (Re-view , is mostly reprinted in . When devoid of LH activityrFSH preparations allow for controlled pure FSH monotherapy. This is an ad-vantage as many patients have sufficient LH concentration and do only need FSHfor proper ovarian stimulation. Mixed gonadotropin preparations can in thesepatients even cause adverse effects and reduce success rate in fertilization andpregnancy because elevated LH concentrations can inhibit the stimulatory effectof FSH on the ovaries. High levels of LH can cause off-timed ovulation resulting inpoor quality embryos and early pregnancy termination. Only hypogonadotrophichypogonadism patients need exogenous administration of LH for proper estradiolproduction (Review Hereby the disadvantages with urinary productswere overcome.
Biological characteristics of rFSH are similar to pituitary and urinary FSH preparations but the isoform profile is more consistent and resemblant of the nat-ural circulating FSH with greater number of basic isoforms. Consequently, thepotency and specific FSH activity of rFSH are higher In the followingyears numerous studies were conducted with different rFSH products, doses and 4.2 Individualised Treatment routes of administration. Clinical use of rFSH proved to be safe with promisingoutcome The pharmacokinetics of rFSH were comparable with urinaryFSH but efficacy and pregnancy rates were higher However, in a meta-analysis of six studies higher clinical pregnancy rates were reported with HMGthan with rFSH in a long treatment protocol, but they could not conclude if itwas the case for ongoing pregnancy rates and live births It was generallybelieved that rFSH would be the future in infertility treatment.
The discovery of the impact of different isoforms and glycosylation on FSH activity and clearance led to development of a long acting rFSH (Corifollitropinalfa, Elonva) that is safe and effective with increased bioactivity and sustainedhalf-life Other research groups have also used different techniques todevelop a long acting rFSH Today existing marketed rFSH products include Gonal-F Puregon and Elonva (Corifollitropin alfa) Following the marketing of the productsother companies have made biosimilars, that have the same active substance asan existing approved biological medicine which is used as reference. Bemfola and Oveleap are biosimilars to Gonal-F, and Fertavid is the same as Puregon.
All these drugs are approved by EMA. Gonal-F and Puregon (with the nameFollistim) are the only ones approved by FDA as well. A novel rFSH, FE 999049,is under development at Ferring Pharmaceuticals. FE 999049 differs from theother rFSH product as it is expressed in a cell line of human fetal retinal origin(PER.C6 R ). It has been demonstrated that FE 999049 has different PK and PD properties compared to Gonal-F To achieve optimised infertility treatmentand to increase pregnancy rates it is intended to investigate the possibility ofindividualised FE 999049 dosing.
Individualised Treatment Gonadotropin therapy cannot be used for every type of infertility, thus a diagnose isneeded prior to treatment. If the cause is primary ovarian failure, administration ofgonadotropins is of no use. Patients with intact ovaries and follicles but who sufferfrom e.g. oligo-anovulation, inadequate ovarian response, insufficient amounts orimbalanced ratio of gonadotropins may benefit from gonadotropin treatment. Thatis, the cause need to be functional and not ovarian failure such that the ovariesare responsive with primordial follicles for FSH to exert its stimulatory effect.
Remarkably early in the gonadotropin therapy history it was known that not all infertile patients could become pregnant using gonadotropin stimulation In addition there was awareness of possible benefits of individualised treat-ment and dose adjustment according to patients' characteristics Already in the first study with HPG it also became clear that several injections 4 Gonadotropin Therapy were needed for prolonged exposure to achieve proper therapeutic stimulation A single high dose of 375 IU urinary FSH elevated endogenous FSH levels by 1.9times compared to normal levels but it was still just a single dominant follicle thatreached full maturation. If instead a low dose, 75 IU of recombinant or urinaryFSH, was given continuously in the follicular phase of the menstrual cycle, growthof multiple follicles was induced Thus, suggesting that duration of ex-posure is just as important as the dose itself for multiple follicular development.
Studies comparing treatment with current rFSH products indicate that differ- ent doses may be needed for individuals. Increasing daily dose of Puregon from 150IU to 250 IU gave only a small increase in oocytes retrieved and the number de-creased with age Even though the higher daily dose shortened the treatmentperiod the total overall amount used was higher. In both dose groups some womenhad insufficient ovarian response and some women were at risk of hyperstimulation.
Another study compared multiple doses of 150 IU with 225 IU Gonal-F in youngand older women. They found similar results for oocytes retrieved between dosingand age groups, and concluded that a higher rFSH dose did not compensate for areduced number of follicles in older women These results reveal that womenwithin same age group and with similar personal characteristic do not respond inthe same way to treatment.
Awareness that one protocol does not fit all, but there is a need for indi- vidualisation and knowledge of the influential factors on ovarian stimulation isincreasing. It is necessary to identify poor and excessive responders to optimisepregnancy rates and reduce cancellation rates. It has even suggested to be un-acceptable to start treatment without knowing the individual patient's potentialsand risks. (Review The necessary dose and treatment length should vary according to individual response in order to get the FSH concentration within the threshold window forprolonged time to induce proper follicular development and not risking ovarianhyperstimulation. What dose to give, how to identify it, based on which criteria,and if it should be changed during the treatment period are major challenges forindividualising infertility treatment.
Predictors of Ovarian Response Diagnosing the cause of infertility is not always possible. As a minimum it canbe checked if the patients have appropriate ovarian reserve with primordial fol-licles eligible for stimulation. Furthermore, indicators for ovarian sensitivity andmagnitude of response would be useful for selecting the individual dose. Fertilitydecreases with age and comes to a hold at menopause. This is due to changesin hormone productions and decline of the primordial follicle pool since birth.
Hormone levels or follicle number could therefore be a better indicator for fertility 4.2 Individualised Treatment than age itself. It is greatly debated which hormone levels and indicators of ovarianreserve are reliable, and which can be used as predicting treatment outcome.
There exists over 20 tests using hormone levels, patient history, and more complicated tests for measuring ovarian function and response to certain stimuli(Review . In a large systemic review from 2006 where the predictivecapability of available ovarian reserve and response tests were evaluated, it wasconcluded that all tests only performed modest to poor. Basal FSH, inhibin B, andantral follicle count (AFC) was though judged to be potentially useful as initialscreening test. There were only included two studies using anti-m¨ (AMH) as predictor. The conclusion for AMH was therefore not final and as itperformed moderately, further studies were suggested for obtaining evidence of itspotential to perform better than the other tests.
Previously it has been shown that elevated FSH baseline is an indicator of reduced response to gonadotropin treatment. A reason for this is probably thatthe threshold has increased and hereby is needed a higher dose to achieve a suf-ficient circulating FSH level for ovarian stimulation (Review Howles et al.
proposed after investigating 15 potential predictive factors that dose shouldbe based on not one but four factors: basal FSH concentration, body mass in-dex (BMI), age, and AFC. This algorithm was subsequently used in a pilot studyyielding good results for oocytes retrieved and pregnancy rates. They suggestedthat further adjustments of dose could reduce cancellation due to hyper- or sub-optimal response Other studies support that a combination of AFC anda selection of hormone levels such as FSH, estradiol, and/or inhibin B optimisethe predictive information instead of using only one marker A problemwith these can be the complexity of assessing all variables and calculating a dosebased on the findings. It would set high demands to the clinic in form of ability toperform all the necessary tests, analyse the samples, evaluate results, and executethe calculations by entering the appropriate values into an algorithm.
As more is learned about ovarian function and hormonal influence the supe- riority of AMH is gaining evidence as an indicator for ovarian reserve (Reviewand as a consistent predictor for poor response as it decrease with ageand ovarian function and correlates with AFC, which also has high value as adeterminant Results from several studies have shown that poor respon-ders have lower AMH level, and some also found that they additionally have lowerAFC, higher FSH levels and age than normal responders. Confirmatory to thisit is also observed that high responders have higher AMH level, higher AFC andlower FSH levels than normal responders Using AMH level as anovarian reserve test is believed to have advantages as it is easy to accurately accessand for low levels AMH has low variation during the menstrual cycle. In a cohortstudy with 20 women AMH levels above 1 ng/mL indicated a young well- 4 Gonadotropin Therapy functioning ovary and the AMH level varied during the menstrual cycle. Whereaswomen with low almost steady AMH levels also had lower inhibin B levels andshorter menstrual cycles, and was identified as having an aging ovary with im-paired function. It was therefore suggested that if AMH is measured at any dayto be below 1 ng/mL it imply reduced ovarian reserve.
Even though several studies point at AMH as the best single predictor of re- sponse and treatment outcome the other factors are still in question. Often base-line measurements have been used in assessment of the predictive value of hormonelevels, but having markers reflecting the degree of response during treatment canbe useful. Inhibin B has the potential of being a predictor for ovarian responsebecause inhibin B levels correlates with follicular development and has been iden-tified as the first PD marker to change after gonadotropin treatment Anincrease in inhibin B levels is the earliest hormone increase observed, the increase issteeper and a maximum level is obtained faster than the other hormones. InhibinB's role as a potential predictor is further supported by that the measured inhibinB baseline and more significantly the rise in inhibin B 24 hours after gonadotropinadministration are both higher in good responders Furthermore, inhibinB levels 24 hours after rFSH administration correlates with AFC and oocytes re-trieved Several studies have investigated how the change in inhibin B duringthe treatment correlates with treatment endpoints to establish its role as a markerfor ovarian response. Number of oocytes retrieved correlates with the increase ininhibin B the first day of treatment in down-regulated subjects with inhibinB levels at days 4-6, 7-8, and 9-10 during treatment as well as with inhibinB levels at day 6 and 8 and the difference between the two days A thresh-old of 300 pg/mL inhibin B increase from day 6 to day 8 was also proposed todiffer between poor (below the threshold) and normal responders Othershave suggested a connection between inhibin B, AMH and oocyte quality basedon findings that the inhibin B rise from day 3 to 4 and AMH correlate with bothoocytes retrieved and number of eggs fertilised Another way to personalise treatment is to calculate dose according to personal demographics that do not give a measure for infertility but instead affect the PKor PD properties of the drug. Giving dose per body size, either measured asbody weight, BMI, or body surface area, is a common dosing protocol for manydrugs. In gonadotropin therapy it seems like body weight has been chosen asthe measure to describe body size influence at the necessary gonadotropin dose.
Studies with urinary and recombinant FSH have shown a correlation betweenserum FSH or the PK parameters and body weight In recentyears dosing Corifollitropin alpha based on body weight has also been proposed When an understanding of what predictors may be important is established, 4.2 Individualised Treatment they can be incorporated in the clinical study protocol to ensure the wanted mea-surements of the factors are available for further testing its influence on treat-ment. With sufficient data generated for potential predictors the significance canbe tested or validated in the analysis and modelling of data. Using the developedmodel for simulations different scenarios can be tested for different values of theidentified factors. Subsequently, when an influential predictor has been identifiedwith certainty, criteria can be set for initial dose stratification, goals or adjustmentsduring the trial.
From a modelling and simulation perspective influence of all available factors can be tested, but for creating the recommended dosing strategy the complexityneed to be considered as dosing should be guided by an easy to use table. Itis not durable if the clinicians have to perform numerous complicated tests andmeasurements, analyse the results and run an advanced computer program in orderto find the dose for a patient. It is therefore important to have both knowledge ofthe biological system and what has clinical relevance when developing models.
Pharmacometric Modelling The pharmacometric discipline in drug development utilises mathematical mod-els and simulations based on knowledge of the physiological system, the disease,and pharmacology. The models are used to quantitatively analyse, describe, andclarify results from clinical trials in order to better understand the drug's PKand PD properties. Important knowledge, maybe otherwise inaccessible, can beachieved through analysis of the models and by performing model simulations totest theories and scenarios.
Drug exposure and response often differ between subjects after dose adminis- tration in a clinical trial. If a variation is observed between individuals in a studypopulation it is important to investigate the causes of the variation and how it canbe quantified and incorporated in the model parameters.
Modelling is an interdisciplinary field, therefore, knowledge from multiple scien- tific areas is needed to develop a valid and useful model with justified assumptionsto obtain a simplified representation of reality. In the previous chapters the back-ground knowledge needed for this PhD thesis has been presented and include thedynamics and function of the reproductive hormone system, the infertility therapyarea, and what biomarkers that are relevant to observe to register treatment effect.
In addition, is needed an understanding of methods and software for data anal- ysis, model development, estimation, and evaluation. These modelling techniquesare introduced in this chapter with focus on specific methods for model implemen-tation in the software NONMEM Population Approach It is important to study the drug in the intended target population, but this isespecially difficult for vulnerable sub-populations like some elderly, paediatric, orintensive care patients. In addition, samples are costly and complicates the logisticsof clinical trials since patients have to stay longer at the clinical. Therefore, datafrom clinical trials do not always include a high number of samples per individualand thus poses challenges in obtaining useful information about the drug from theclinical data. The analysis of such data requires special techniques and software 5 Pharmacometric Modelling When analysing clinical population data the objective is to describe the typical behaviour of the study population as well as variability between subjects. Previ-ously, standard analysis approaches were na¨ıve pooling of data and the standardtwo stage approach Na¨ıve pooling can be used when data contain very fewsamples per individual. All data is analysed together disregarding which individu-als the samples come from. Hereby is only obtained information about the typicalparameter values for the population as one unity and nothing about variationsin the population. In the two stage approach subjects are analysed one by oneto obtain individual parameters and require an extensive number of observationsper individual. Average population parameters are then calculated as mean ofindividual parameters with respective variances to describe the population vari-ability. The advantages with these method are that they are simple, familiar, andstraightforward. However, they have been shown to result in bias of the parameterestimates when data is not suitable and because they do not differentiate betweentypes of variability, if at all considered.
Observational data from clinical settings, besides potentially being sparse, can be nonstringent, nonhomogenous, and imbalanced, hence not suited for restrictedapproaches New analysis methods were therefore needed. A more efficientway for analysing population data is using nonlinear mixed effects modelling toquantify the PK and PD properties of the drug. With this method data is pooledfor estimating population PK parameters but, as opposed to na¨ıve pooling, itis explicitly tracked which individuals observations belong to, in order to obtainthe variability and distribution of individual parameters The populationapproach with nonlinear mixed effects models facilitate identification of variabil-ity as well as differentiation between interindividual variability (IIV) and unex-plained variability which includes measurement errors, intraindividual variability,and model misspecification. Furthermore, it gives the possibility of quantitativelyanalyse and identify factors causing the variability, and opens for opportunitiesto include more physiological correct descriptions for mechanism- or physiological-based modelling. The population PKPD approach is an analysis technique forsolving the issue: "How to learn what we need to know to administer drugs opti-mally in clinical settings" (quote An important progress in the population modelling field was the development of NONMEM by Beal and Sheiner - a software for implementing nonlinearmixed effects models and estimating model parameters. Notably with their threearticles in the early 1980's on evaluation of methods for estimating populationpharmacokinetic parameters Sheiner and Beal sat ground for thepopulation PK approach and showed the great usefulness of NONMEM, even forsparse clinical data, and its superiority to the standard methods. The developmentof NONMEM started in 1972 and up till his death in 2004, Lewis B. Sheiner played 5.2 Nonlinear Mixed Effects Modelling a pivotal role in this development (for an excellent review see the years others have made valuable contribution to the further development ofNONMEM Other software for analysing population data exist (reviewsbut NONMEM is the most widely used. Within the first decade of thepopulation modelling approach's existence, more than 40 studies reported resultsobtained from population modelling Hereafter published articles concerningpopulation PK and NONMEM per year have increased exponentially Nonlinear Mixed Effects Modelling A nonlinear mixed effects model consists of fixed effects, random effects, and in-dependent variables. In a population PKPD framework the model is made up ofthree sub-models; the structural model, the statistical model, and the covariatemodel. The fixed effects are the PK, PD, and covariate effect parameters The PK parameters describe the processes controlling the time-course of drug con-centration in the blood or tissue of interest, and the PD parameters are relatedto the effect of the drug. The average parameter values in a population charac-terise a typical individual in the population and are therefore also called typicalpopulation parameters. The random effects are divided into interindividual andintraindividual variability, which is the variability between subjects and residualerrors, respectively. Covariates, time, dose, and other variables in the study designare independent variables in the model.
A structural PK model consists of a number of compartments involved in thedrug's path through the body. The change in drug amount in each compartmentis described by a differential equation with structural model parameters, which arethe fixed effects in the model and denoted by θ's.
A one-compartment model after an i.v dose is the simplest PK model. It is illustrated in Figure and the associated differential equation is dA1(t) = −kA1(t) , A1(0) = dose.
The elimination rate, k, is clearance (CL) divided by the volume of distribution(V), and the drug concentration at time t is the amount in the central compartment(A1(t)) divided by V. The initial condition for the differential equation is dose.
For a two-compartment i.v. model an extra equation is added for the peripheral compartment (A2) and the drug amount flow from and to the central compartmentfollows the distribution rate constants k12 and k21, respectively. In Figure 5 Pharmacometric Modelling Figure 5.1: Drug distribution after an intravenous dose following a) a one-compartment or b)a two-compartment model.
is illustrated a two-compartment i.v. model with elimination from the centralcompartment. The dynamical system for the model then becomes dA1(t) = −kA1(t) − k12A1(t) + k21A2(t) , A1(0) = dose.
dA2(t) = k12A1(t) − k21A2(t) , A2(0) = 0.
The terms one- and two-compartment PK models refer to the distribution of the drug, but after other administration routes than i.v. the complete modelcontain more compartments. When the dose is given orally, s.c., or i.m. a com-partment representing the dosing site is included, from where the drug is absorbedto the central compartment with the rate constant ka. A transit model with extracompartments can be added to describe a delay in the absorption process. An ex-ample of this type of model is represented by the compartment diagram in Figurewith the respective differential equations: dA1(t) = −ktrA1(t) dTR1(t) = ktr(A1(t) − TR1(t)) dTRn(t) = ktrTRn−1(t) − kaTRn(t) dA2(t) = kaTRn(t) − kA2(t).
5.2 Nonlinear Mixed Effects Modelling Transit 1 ktr Figure 5.2: A one-compartment model after e.g. a subcutaneous dose. The absorption processfrom the dosing site is delayed by n transit compartments.
The initial condition at the dosing site is dose, and the others are zero. The rateconstant for the flow into the transit compartments is (ktr). In an absorption modelthe complete dose might not be absorbed, and without i.v. data the bioavailability(F) is not known, hence the PK parameters that can be estimated are the apparentvolume of distribution (V/F) and apparent clearance (CL/F). The tradition in thefield is to denote this a one-compartment model despite the additional absorptionand transit compartments.
The compartments do not necessarily represent an anatomically defined part of the body but rather a collection or a symbolic compartment. With moving froman empirical to a semi-mechanistic and mechanistic approach, parameters andcompartments will to a larger degree be representative for physiological processesand anatomical parts. In a physiological-based model all compartments correspondto an organ or tissue in the system of interest and the parameters will have aphysiological interpretation.
5 Pharmacometric Modelling Figure 5.3: Illustration of variation between subjects in a population. The PK profile ofindividual subjects (blue lines) and the population mean profile (purple line).
Statistical Model Individual PK profiles are likely to vary from the population mean for a typicalindividual in the observed population (Figure In order to describe this vari-ability between subjects in the population, random effects are added to the typicalmodel parameters where needed. One way to add the random effect for IIV at atypical population parameter (θ) to obtain the ith subject's individual parameter(θi) is with an exponential model θi = θ exp(ηi).
The individual random effect (ηi) comes from an approximately normal distribu-tion with mean zero and variance ω2 for describing the IIV of the parameter. Thecovariance matrix for all IIVs is denoted Ω. The variability between two parame-ters can be correlated and is incorporated in the model by adding a covariance forthe interindividual random effects in Ω. The advantage of this model over e.g. anadditive is that the individual parameter cannot become negative.
In addition, individual model prediction at time tij may differ from the cor- responding jth observation yij by a residual error εij (Figure It is the in-traindividual variability and covers the unexplained model misspecification, suchas unknown influencing factors that are not possible to describe in the model. Theresidual errors are assumed normally distributed with mean zero, variance σ2, andcovariance matrix Σ. Together, the random effects make up the statistical model,which is also called the error model or stochastic model.
A third type of variability may be considered, the interoccasion variability, which can be important to model if subjects are studied at different occasions overtime such that the individual parameters can change during the study period 5.2 Nonlinear Mixed Effects Modelling Figure 5.4: The difference between observations (points) and model prediction (line) for indi-vidual i at time tij is the residual error εij - a random effect in the model.
The last part is the covariate model where potential factors influencing the vari-ation in parameters and drug effect are identified. Covariates can for example bepersonal demographics, disease parameters, or other factors in the target system.
Covariate analysis can furthermore be used to identify sub-groups of patients whodo not get optimal effect due to distinctive characteristics and therefore wouldneed different doses. Thus, the covariate model will further increase the predictivecapability of the model.
In the model potential influential factors can be tested for significance as a covariate to explain some of the IIV in a parameter. Continous covariates can forexample be added in a linear, exponential, or power relation, and often the effect isnormalised or centred to a standard value of the covariate like population mean ormedian. A typical covariate at CL and V is body weight, since body compositioncan affect drug metabolism and distribution. It might be underlying mechanismsresponsible for the impact of body weight, but such detailed measurement aremostly not available thus body weight is expressive for the differences. The effectcan be included in the model with allometric scaling at the parameters where WTi is the ith subject's body weight, WTst is the normalisation value, andALθ is the allometric values: 0.75 for CL and 1 for V. The same expression can beused for other covariates and other parameters, but then ALθ is estimated as thepower exponent for the effect. Other functions like linear and exponential can also 5 Pharmacometric Modelling be tested for describing a covariate effect. For categorical covariates (e.g. gender)different θ's are used for the categories and potentially also different η's.
The model prediction for individual i can be presented as a function of time, theother independent variables (xij), and the vector of individual i's parameters (Θi).
The individual observations are described by adding the residual errors to themodel prediction yij = f (tij, xij, Θi) + εij.
In general an individual parameter is given as a function of the typical populationparameter, individual random effect, and individual associated values of covariates(ci) θi = h(θ, ηi, ci), where the random and covariate effects potentially are absent.
PKPD models are not as straight forward and simple as PK models. There doexist a number of predefined effect models but the variation by combinations andalterations of the models gives many options. Three major types of PKPD modelsare direct link, indirect link, and indirect response models. A few examples willbe given here, but for extensive reviews see Concentration-effect or equivalently named exposure-response models are PKPD models describing the effect caused by the drug concentration and the combinedtime-course. In direct link models the drug concentration is directly included inan expression for the effect. The site of action where the effect occur is not neces-sarily the blood (or another body fluid) where the drug concentration is typicallymeasured. Therefore, steady state conditions and equilibrium between the con-centration at the effect site and the blood is often assumed in order to directlymodel the effect of the drug concentration.
It can simply be an all-or-none model where the effect occur above a certain concentration threshold and not below. Or the effect can for example be describedby a linear model, a power model, an Emax model, or different sigmoid Emax modelswith Hill coefficients. The equations can include a baseline effect if an effect ispresent before any drug is administered. A linear or power model might not seemfeasible as the effect has no limit which is not physiological plausible. Anyhow itcan in a certain concentration area be the best description of the effect and can beseen as an approximation to an Emax model. Such approximations may be suitablefor concentrations (C) much lower than EC50, the concentration where half of the 5.3 Pharmacokinetic-Pharmacodynamic Modelling maximum effect (Emax) is obtained. Consequently the sigmoid Emax model withHill coefficient λ is approximately reduced by one parameter to a power function: with k = Emax/ECλ .
If substituting drug concentration with dose in the direct link model, the model becomes a dose-response model, which is a simple direct PD model describing thechange in effect with dose and bypassing any variability in the pharmacokinetics.
Indirect link models introduce a time delay for the effect to occur. The delayed response can be incorporated in the model by linking the PK model to the PDeffect by an effect compartment. It is a hypothetical effect site that does not affectthe drug amount balance in the PK model but represents drug distribution fromthe central compartment. The PK model is therefore unchanged and the followingequation is added for the effect compartment dCe = k1eCc − ke0Ce, where Cc and Ce is the concentration in the central and effect compartment, re-spectively. It might not be possible to differentiate between the first order distri-bution rate constants k1e and ke0, hence they are often set to be equal. The effectcompartment concentration is then used in the PD model.
The pharmacodynamic response can be time-dependent when the effect is not directly related to drug concentration. In such situations an indirect responsemodel is needed where the drug concentration indirectly affect the response vari-able. It consists of a turnover model as the basic model for describing the response(R) over time as a separate differential equation with a zero order production rateconstant (kin) and a first order rate constant for elimination (kout) dR = kin − koutR.
In the absence of drug R(0) = kin/kout initially. An inhibitory or stimulatoryfunction of drug concentration is then added to either of the rate constants to get the change in response over time. Stimulating the input or inhibiting theoutput cause an increase in the response and inhibiting the input of stimulatingthe output cause a decrease in the response.
Dose-concentration-response models characterise the full PKPD picture of how the effect changes in relations to drug concentration and in turn to dose and overtime.
Irrespective of the type of PKPD model, a sequential or a simultaneous mod- elling approach can be chosen. In a sequential model the PK model is developed 5 Pharmacometric Modelling first and then either the estimated PK parameters or predicted concentrations areused in the PD model. In simultaneous modelling all PKPD model parameters areestimated at once. As this can include many parameters, priori knowledge of thevalues might have to be used depending on the size of and information in the data.
As opposed to sequential modelling, simultaneous modelling gives the opportunityof (semi-)mechanistic modelling with feedback to the PK model.
Prediction is of great importance in clinical drug development and models for predictive purposes should be mechanistic rather than empirical Inaddition, population modelling with random effects should be chosen over mod-elling of only population averages, as it also favours incorporation of physiology.
With a mechanistic modelling approach a more accurate model can potentially beachieved for increased precision in predictions and further understanding of thesystem.
Estimation Processes Fitting the model to the data is done by estimation processes searching for pa-rameter estimates that maximises the probability of data (y) occurring under themodel: P(y Θ, Ω, Σ). Maximum likelihood estimation methods are commonly usedfor obtaining model parameter estimates for the best model fit to the data. Giventhe data and model the likelihood function provides a measure for how likely aspecific set of model parameters are. For a nonlinear mixed effects model the in-dividual likelihood of the model parameters for the individual data is defined asthe marginal likelihood Li(Θ, Ω, Σ yi) = p(yi Θ, η, Σ) p(η Ω) dη, consisting of the density functions of the individual data (yi) and the individualrandom effects (η). The log-likelihood function can be used in the estimationprocess to obtain a more numerically stable function. The log-likelihood for themodel parameters for the whole dataset is given as a sum of all individual log-likelihoods l(Θ, Ω, Σ y) = log Li(Θ, Ω, Σ yi).
This integral can rarely be solved analytical so in the software NONMEM the log-likelihood function is numerically approximated by different types of lineari-sation and iterative procedures The NONMEM objective function value(OFV) is approximately proportional to -2log-likelihood of data given the modeland parameters, and is thus sought to be minimised. Some of the specific estima-tion methods for NONMEM are first order (FO), first order conditional estimation 5.4 Estimation Processes (FOCE), and Laplacian method. The FO estimation method was the first avail-able method, in which a first order Taylor series expansion of the log-likelihoodaround the expected values of the parameters, i.e. without the random effects,is used. The fixed effect population parameters are then estimated by extendedleast squares. The random IIV effects can be obtained posthoc as empirical Bayesestimates (EBE) of the η's. In the FOCE method the linearisation is made aroundthe conditional estimates of the η's, i.e. η estimates obtained conditionally on thepopulation parameters, and therefore simultaneously obtain population parameterestimates and EBE's. FOCE with interaction takes into account the interactionbetween IIV and residual error. The Laplacian method is similar to FOCE but usessecond order Taylor series linearisation instead of first order and hereby increasesaccuracy but also computational complexity.
Data Requirements The dose is not explicitly contained in the differential equations of the model, butgiven as initial conditions. Therefore, the time of dose administration, amount,and number of doses for each subject has to be included in the data, as wellas specifying dose as the input to a compartment in the model depending on theroute of administration. Data is allowed to be sparse, imbalanced, and incomplete,because when using the population approach information from all individuals areused, and full individual profiles can be obtained even though there are time pointswith missing measurements for some individuals. Missing covariate values will haveto be dealt with in order to test the effect. Ways of doing this can be to use themean or median of the population values.
The laboratory equipment and assay for measuring the concentration in the blood samples has a lower limit of quantification (LLOQ) for the lowest measurableconcentration that is validated. When this limit is reached the observation isreported as below quantification limit (BQL) in the dataset meaning that theconcentration is somewhere between zero and LLOQ. Thus BQL observations donot have a concentration value and therefore need to be changed or removed beforethe data is used for estimation of model parameters. Depending on the proportionthe BQL measurements constitute of data, they might have to be accounted forin the model development to avoid biased parameter estimates. There are sevendifferent ways to handle BQL data points as listed by Stuart Beal in 2001 There are four simple methods where all BQL observations are either discarded(M1), replaced with LLOQ/2 (M5), or replaced with zero (M7). In the M6 methodthe BQL observation is set to LLOQ/2 if it follows an observation with a valueabove LLOQ and it is discarded if it comes after another BQL observation, thus inconsecutive BQL observations the first is set to LLOQ/2 and the rest discarded.
The M2-M4 methods are likelihood based approaches where the BQL observations 5 Pharmacometric Modelling are discarded (M2) or treated as censored (M3,M4). In most cases the preferredmethod for handling BQL observations is M3 as it gives the least bias, in particularwith a high BQL proportion in the data With the M3 method BQLdata are censored observations and treated as categorical data. These are includedin the likelihood function for the estimation as the likelihood that the observationis truly below LLOQ. Therefore, from the estimation process for these observationsa probability is returned and not a value for a predicted concentration .
Model Development and Evaluation As a guideline for model development the structural model, then the statisticalmodel, and lastly the covariate model is developed. The sub-models are not sep-arate models, therefore significance of previously included of excluded parameterscan change. It can take several loops of refinement, challenging, and evaluatingprevious steps before a satisfactory model is achieved. A proposed modelling struc-ture is developing the structural model with relevant residual errors (first-stagemodel). The second-stage model is addition of IIV and covariates for obtainingindividual parameters In reality, the residual errors are included in thestructural model development as a minimum, but often IIV is also added to someparameters from the beginning when there are clear variations in the population.
The interaction between the sub-models have been studied with different datasets,and the sub-models were found to be highly intertwined and the choice of onesub-model affected the other significantly A revised model building strat-egy was suggested after this observation. From the structural models evaluated, asimple supportable but not necessarily the best model should be chosen for covari-ate analysis. If a more complicated structural model was observed to have someadvantages, the found covariate model should then be tested with this model. Asthe structural model has been changed left out covariates should be reconsideredand the significance of included covariates should be retested.
Besides deciding on a modelling strategy, evaluation criteria are needed in order to choose the best model. For a quantitative assessment for choosing between twomodels and evaluate the model fit both statistical and graphical tools are used.
There are shortcomings to the different evaluation methods and it is thereforeimportant to use several for a proper diagnosis and challenge the model by usingsimulations Statistical Methods The difference in OFV between two nested models is approximately χ2-distributed,with degrees of freedom equal to the number of differing parameters. To determine 5.5 Model Development and Evaluation which of two nested models are statistically significant best, a χ2-test can thereforebe performed on their OFVs and used for selection of inclusion (or deletion) of amodel parameter. For 1 degree of freedom and 5% significance level the decreasein OFV from the reduced model has to be greater than 3.84 for choosing the largermodel.
The model selection should not purely be based on the OFV value. A logical evaluation of parameters should be performed to get an overview of if the param-eters are reasonable and if they make biological sense when a more mechanisticapproach has been used. A measure for a parameter's reliability is the relativestandard error (RSE = SE(θ)/θ).
It should also be checked if the assumption of the individual random effects having a mean of zero holds. NONMEM returns a probability for if the true meanof the η distribution for the IIV on a population parameter is zero. This p-value hasto be greater than 5%. If the individual information is scarce the phenomenon ofη-shrinkage can occur, where the individual parameters shrink towards the typicalparameter values It is defined as shrinkage = 1 − where ηEBE are the EBE estimates of the η's and ω is the standard deviation ofthe η's. In case of high η-shrinkage (above 20 or 30 %), the interpretation of theEBE-based diagnostic plots (presented below) should be made carefully, as falserelationships might have been induced. For even higher values, the individualrandom effect model should be reconsidered as it can also mean that the model iswrong.
Graphical Methods Several diagnostic and goodness-of-fit plots are useful for evaluating how well themodel describe data. Plotting mean of predictions and mean of observations cangive an overview of the trend in model fit over time, e.g if the baseline value isover- or under-predicted, or if the model do not catch the change over time. Onefault of this plot is it does not show if individuals have different curves that all arepoorly predicted but by chance the means match. Observed and predicted PK orPD profiles can be plotted for each individual instead to further inspect model fit.
The visual predictive check (VPC) is a diagnostic plot using simulations to check the model's predictability. First, 1000 datasets for model predictions aresimulated using the distribution of the estimated parameters, interindividual vari-ability and model residual errors. Then, observed data is compared to the sim-ulated model predictions by plotting the 2.5th, 50th, and 97.5th percentiles ofobservations and the 95% confidence intervals (CI) for the corresponding model 5 Pharmacometric Modelling prediction percentiles. When observations include BQL data points a two panelVPC for illustrating both continuous and categorical (BQL) data has to be usedThe top panel displays the observations above LLOQ like a normal VPC.
Since a value is not estimated for BQL data the evaluation of these is insteadshown in a second panel as the median observed proportion of BQL observationswith the 95% CI for the median of simulations.
Goodness-of-fit plots are the dependent variable versus predictions and different residual plots. In the first type the observed values are plotted against individualpredictions or population predictions for a typical individual to get informationof the variability in the data. Different types of residuals can be plotted againstprediction or an independent variable. Individual weighted residuals are definedas IWRES = ij − ˆ yij is the individual prediction of the individual observation yij and σ is the error magnitude (standard deviation) of the residual error model. These residualsmay not reveal model misspecification and are only useful when individual data isinformative Standard residuals are the difference between observationsand population predictions and lack information at the individual level and caninaccurately show residual trends. The weighted residuals (WRES) are normalisedto residual variability and thus more explanatory of the model but the calculationis based on the FO estimation method. With the development of improved estima-tion methods the FO method is rarely used, hence conditional weighted residuals(CWRES) based on the FOCE method are better for diagnostic purposes Correlations between parameter's random effects can be identified by plotting EBE's against EBE's, and covariate relationships can be investigated by plottingboth parameters and random effects against potential covariates. In these EBEdiagnostic plots it is important to remember to consider the size of η-shrinkage.
Modelling of FSH Compounds The aim of PK and PKPD modelling of FSH compounds in drug developmentvaries from comparing different FSH products, acquire knowledge of the drug'sproperties, quantify covariate influence, and to dose finding purposes. Treatmentstrategy and optimal dose can vary greatly between infertile patients due to thegreat difference in cause and extent of infertility. It is therefore valuable to describethe time course and variations in drug concentration and effects properly, andultimately be able to predict the outcome of a given dose in a specific woman andthereby provide individual dosing schemes. In this chapter is reviewed which PKand PKPD models of FSH compounds exist, and at the end the foundation forthe research questions in this PhD thesis is given.
Existing Mathematical Models The pharmacokinetic properties of gonadotropin preparations have been studiedextensively, but existing PK and PKPD models for FSH compounds are limitedand even fewer population models with nonlinear mixed effects modelling exist.
Pharmacokinetic Models Several studies by le Cotonnec et al. use modelling to describe the pharmacokineticbehaviour of urinary and recombinant FSH in pituitary down-regulated female vol-unteers. Two studies investigated and compared the pharmacokinetics of urinaryhuman FSH in a standard form (Metrodin) and in a highly purified form (MetrodinHP) A series of three connected studies investigated thePK and PD properties of Gonal-F. The first study compared Gonal-F withMetrodin. The second study investigated the pharmacokinetics of Gonal-Fafter different routes of administration and after multiple doses. The third studyused the multiple dose data from the second study to develop a PKPD model(described below).
It was found that recombinant and urinary FSH have similar PK properties.
The studies also included non-compartmental analysis where the endogenous FSHwas handled by baseline correcting data. The two pharmacokinetic analysis ap-proaches gave similar results.
6 Modelling of FSH Compounds The four PK studies and type of models are listed in Table The models were exponential models where an exponential equation describes the drug con-centration in a one-compartment model and a two-compartment model consistsof a biexponential equation. If an endogenous FSH contribution was modelled itwas simply included as a linear function of time, and an initial baseline value wasadded to the equation for an intercept. The resulting PK models after s.c. ori.m. administration of both urinary and recombinant FSH were one-compartmentmodels with first order absorption, whereas after i.v. doses a two-compartmentmodel was found to describe data best. These studies describe data well and iden-tify a large variation in the PK parameters, but lack consideration of covariatesand alternative models. When using a population approach with nonlinar mixedeffects modelling instead of subject-by-subject analysis, parameters with interindi-vidual variation can be specified and it is possible to further investigate causes ofthe observed variation by covariate analysis.
Four published articles with population PK models developed using the popu- lation approach with nonlinear mixed effects modelling and NONMEM software toanalyse data after dosing with urinary or recombinant FSH were retrieved (TableKarlsson et al. developed population PK models using data from threedifferent studies with infertile patients or pituitary down-regulated healthy femalevolunteers receiving urinary human FSH (u-hFSH) or rFSH. It was not reportedwhich specific FSH products were used, only the type of FSH. Data from i.m.
and s.c. dosing was combined with i.v. data when available for the modelling.
After i.m. administration of u-hFSH and s.c. administration of rFSH (study 1)the resulting model was a one-compartment disposition model, whereas after i.v.
doses of rFSH (study 2) and u-hFSH (study 3) a two-compartment model wasused. These results are in accordance with the previous results with exponentialmodels. It was though found that when the i.m. or s.c. data was combined withthe i.v. data for the same subjects, and hereby increasing the number of samplesfor each individual, a two-compartment model could adequately describe the data.
The first study where one-compartment models were identified had few samplesper individual. It was concluded that the exogenous FSH pharmacokinetics werebest described by a two-compartment model if the data was "rich", meaning thatthere were an extensive number of samples. The model structure does not dependon the route of administration as it is related to the properties of the drug but itis the data that dictates how many compartments that can be characterised. Theconsistency in number of compartments in the model with route of administrationwas therefore due to the richness of data.
Covariate analysis was only performed in the first study dataset with patients receiving u-hFSH or rFSH. An influence of body weight at CL was significant in themodel with the u-hFSH data. BMI was a covariate at ka and creatinine clearance 6.1 Existing Mathematical Models (CLCR) at FSH baseline in the rFSH data. In both models the FSH baseline wasfound to be best described by an exponential decline over time. The second datasetwith healthy female volunteers given rFSH was divided into four subsets (see Tableand four models were developed with different extent of linear decline in FSHbaseline over time. In the last dataset with single dose u-hFSH the baseline wasalmost constant over time.
Three studies investigating the dose-response relationship and pharmacody- namic properties of Corifollitropin alfa (a long acting rFSH) after a single s.c.
dose in infertile patients used population PK models and NON-MEM to estimate PK parameters. Development of a pharmacokinetic model wasnot the primary objective in these studies, thus no details about the developmentwas provided. A one-compartment model with first order absorption was used,and in two of the studies body weight was included as a covariate atCL and V. Another article with Corifollitropin alfa used a population PKapproach to analyse data, but it was purely result oriented purpose and only thePK parameters obtained from the model were presented. Non-modelling studieswith rFSH products have found a negative correlation between serum rFSH levelsand body weight further supporting the role of body weight as a factorin determining the pharmacokinetics of FSH compounds.
A few PKPD models for urinary and recombinant FSH are listed in Table Karlsson et al. used a population PD model to predict the follicular growthin patients after dosing with u-hFSH and rFSH. It was tested whether FSH con-centration could be included in the model as a separate term with either the totalendogenous and exogenous FSH concentration, baseline corrected levels, or as nor-malised to pretreatment levels. Numerous models and covariate relationships wereevaluated. The change in total follicular volume (TFV) was found to be indepen-dent of FSH concentration, and the best model was an Emax model for TFV plusa constant term. It was possible to explain some of the interindividual variabilityin the Emax model parameters with pretreatment FSH level and baseline estradiol.
Inhibin B and estradiol levels have been described after multiple s.c. doses of Gonal-F in pituitary down-regulated healthy female volunteers with a sequentialPKPD model The PK model used had been developed in a previous pa-per (described above) from where the estimated PK parameters were fixedin the PKPD models.The PK model was linked to the PD models with an effectcompartment and the PD responses were calculated from the FSH concentrationsin the effect compartment. Both PD models describing inhibin B and estradiolconcentrations were power functions. They did not use a population modellingapproach but fitted the model to individual data in order to obtain the PD pa- 6 Modelling of FSH Compounds rameters. A large interindividual variation in the estimated inhibin B responseand parameters were observed but since no covariate analysis were performed, nofactors causing the variation were identified. A high correlation between TFVwith both maximal inhibin B and with estradiol levels was found, but there wasno correlation between FSH concentrations and any of the maximal effects for thePD markers. Thus, the high variation between subjects in PD parameters wasnot due to pharmacokinetic variations but different pharmacodynamic sensitivity.
It is therefore not enough to adjust dose after variations in FSH concentrationsbut the response should also be taken into account. These findings of relationbetween PD responses and large PD variation were confirmed with a new datasetwhere almost the same PKPD models were used. The only adjustment wasdue to a measurable FSH baseline concentration and therefore a constant term forthe effect at baseline was added to the power function in the PKPD models. Inaddition was found that large fluctuations in FSH baseline production over timeinfluenced the pharmacokinetics indicated by high variation in the individual pa-rameter estimates and in predicted FSH concentration at late time points, wherethe FSH concentration is mostly due to endogenous FSH.
The fourth and last PKPD article found is a more extensive but still empirical model for Corifollitropin alfa by de Greef et al. A population PK modeland four PD submodels were generated sequentially and then combined to predictovarian response and used for optimal dose selection. NONMEM was only usedto develop the PKPD model for Corifollitropin alfa (PK) and inhibin B (PD)concentration-time profiles. For the last three PD models for follicular volume,cancellation rate, and number of oocytes were used SAS. The PK model was basedon 1263 Corifollitropin alfa concentrations from three clinical trials with single s.cdosing (in fact the three mentioned above with PK models Bothone- and two-compartment models were tested, but since the data had sparsesampling a two-compartment model could not be fitted to data. Of the potentialcovariates: body weight, BMI, height, and age, was body weight at CL and Vthe only significant relationships. The difference from the above two models isthat the inhibin B concentration is linked to the predicted Corifollitropin alfaconcentration by an indirect response model. In which the production of inhibin Bis stimulated by individual Corifollitropin alfa concentrations following a sigmoidalEmax function and the elimination over time is also modelled explicitly in thedifferential equation for inhibin B. The model development was guided purely bydata and to correct for an undershoot in inhibin B below the baseline values, ahypothetical modulator that stimulated the elimination of inhibin B was includedto further lower the level. All parameters had interindividual variability, and agewas a covariate at the Emax parameter.
In the development of Corifollitropin alfa extensive modelling has been used, 6.2 This Research and the development program has been reported as an example of successful imple-mentation of MBDD A combination of PKPD population models, statisticalregression models, and simulation was used to analyse Corifollitropin alfa trialdata and make better informed decisions. It resulted in a dosage regimen of 150µg and 100 µg for subjects weighing >60 kg and ≤60 kg, respectively A particular interest in this PhD project on modelling of FE 999049 in drug de-velopment was to investigate the influence of endogenously produced FSH. A highendogenous FSH level at the beginning of treatment could influence the phar-macokinetic characterisation of the drug, add to the variations in exposure, andeven more so if the level changes over time. Endogenous FSH should therefore beconsidered in the modelling.
The standard data analysis method is to baseline correct data when the drug is a naturally occurring substance in the body. It seems unlikely that the endoge-nous FSH production stay unchanged from the initial baseline value throughouttreatment since the ovarian hormones are known to alter the endogenous FSH pro-duction and release from the anterior pituitary. Subtracting baseline values fromconcentrations at all other time points may give lower drug concentrations thanthey actual are. Hence, using baseline corrected data can potentially induce a biasin the model estimates. If the gonadotropin down-regulation of subjects in clinicaltrials has been completely successful any interference can be avoided and it mightbe possible to keep the baseline value steady. The influence of endogenous FSHtherefore depends on the type of trial and should be considered according to thestudy design as well.
Some of the existing models with FSH compounds considered endogenous FSH either as a constant level or that changed over time, but influence from otherhormones was not included. All the existing PKPD models listed in Table wereempirical sequential models, where no feedback to the pharmacokinetics could beincorporated.
A more mechanistic modelling approach may be warranted to account for this hormone dynamics and variations in endogenous FSH both between subjects andover time. In addition, variations in response to treatment can be caused bypersonal demographics, hormone levels, the great difference in infertility type,and a variety of other factors. As seen in the modelling articles discussed abovelarge variations are observed in both PK concentrations, PD markers, and thusmodel parameters. Another focus in the model development is to identify possiblecovariates for explaining some of the variations in drug concentration and effectsbetween subjects.
6 Modelling of FSH Compounds A full systems pharmacology model including all hormones would be rather complex and require an extensive dataset for proper estimation of all parameters.
To have a more focused goal within this PhD framework inhibin B was chosen asthe ovarian hormone to incorporate in the model as it has a purely inhibitory effectat FSH. In addition, inhibin B is an important marker in early drug developmentand is the first hormone to increase upon exogenous FSH stimulation with highcorrelation to follicular development as discussed in section The overall aim of the thesis was to develop population PK and PKPD modelsfrom clinically observed FE 999049 data using nonlinear mixed effects modellingto acquire a better understanding of the drug's PK and PKPD properties. Inaddition, the objective was to investigate the influence of endogenous FSH bytaking the reproductive endocrine system dynamics into account.
Three specific aims of the analyses were to: 1. Describe the population pharmacokinetics of FE 999049 after single dose administration and examine if part of any interindividual variability in thePK model can be explained by potential person specific covariates (Paper I).
2. Describe the population pharmacokinetics of FE 999049 after multiple dos- ing and evaluate the influence of endogenous FSH levels on the FE 999049pharmacokinetics (Paper II).
3. Develop a semi-mechanistic PKPD model describing the relationship be- tween FSH and inhibin B when accounting for variations in endogenous FSH(Paper III).
Data from three clinical trials with FE 999049, two phase I studies (CS01 andCS02) and one phase II study (study 000009), were available for the analyses.
In this chapter a summary of each clinical study, the generated data, and mod-elling strategy is given. For further details about the clinical studies the reader isreferred to Olsson et al. Olsson et al. and Arce et al. respec-tively. CS02 and study 000009 included a Chinese hamster ovary (CHO)-derivedrFSH marketed product GONAL-F (follitropin alfa, EMD Serono) as an activecomparator. In this work, focusing on the PK and PD properties of FE 999049,subjects receiving GONAL-F were excluded. The intention was to develop a PKmodel based on single dose data from CS01 (paper I). This model structure shouldoverall be confirmed after repeated dosing with the CS02 data (paper II), whereadditional information could be gained from the different study design and extrahormone measurements. Finally, using the FE 999049 population PK knowledgeacquired from phase I data, a simultaneous PKPD model with inhibin B as PDendpoint was developed based on phase II data (paper III).
The trials were performed by Ferring Pharmaceuticals. Each study protocol wasapproved by independent investigational review boards, regulatory authorities andlocal ethics committees. The studies were performed according to the Helsinkideclaration and good clinical practice. Prior to the studies all participants signedinformed consent forms.
The phase I studies included healthy female volunteers aged 21-40 years with anormal menstrual cycle, a BMI 18-29 kg/m2, and using combined monophasic oralcontraceptives (COC). Table gives an overview of the subjects who receivedFE 999049 in each trial. In order to avoid any interference with endogenous FSHlevels, they were gonadotropin suppressed throughout the study. To verify a properlow endogenous hormone level, serum FSH had to be below 5 IU/L on day -3 andday -1 before dose administration. If not the subjects were excluded.
CS01 was the first-in-human study with FE 999049. It was a randomised, double-blind, placebo controlled, sequential single ascending dose study investigating thesafety, tolerability, and pharmacokinetics of FE 999049. 40 subjects received asingle subcutaneous abdominal injection of either 37.5, 75, 150, 225, or 450 IUFE 999049 (equivalent to 2.19, 4.38, 8.76, 13.14, and 26.27 µg, respectively) orplacebo. On day -14 before the start of the trial all subjects were switched tothe same high-dose COC (OGESTREL 0.5/50), which were taken continuouslythroughout the study to suppress endogenous FSH.
Blood samples for measurement of serum FSH concentration were collected 60 and 30 minutes prior to administration, immediately before administration,at every 4 hours the first 48 hours and subsequent every day up to 9 days afteradministration.
The second clinical study, CS02, was a randomised, double-blind, active control,multiple dose study for investigating the safety, tolerability, immunogenicity, phar-macokinetics and pharmacodynamics of FE 999049. A GnRH agonist (LUPRONDEPOT, 1-month depot) was used to suppress the endogenous FSH levels. Eventhough it is an agonist i.e. it exerts the same effect as GnRH and stimulates thepituitary gland to secrete gonadotropins, it can be used to suppress the endoge-nous production of FSH. This is due to the fact that the sensitivity of the GnRHreceptors in the pituitary gland decrease under constant exposure to GnRH (ago-nist) and ultimately the receptors become unresponsive. As a consequence GnRHstimulated gonadotropin secretion ceases. 49 subjects were given daily subcuta-neous doses of 225 IU for 7 days of either FE 999049 (dose equivalent to 14.69 µg)or GONAL-F.
Blood samples were collected 60 and 30 minutes prior to administration, im- mediately before administration, and once a day for 15 days after the first admin-istration. Serum FSH, Inhibin B, estradiol, progesterone, and LH concentrationswere measured in the blood samples. In addition, at day 6 after administration ofthe last dose the FSH concentration was measured every 4th hour for the following2 days.
The 000009 phase IIb dose finding study was a randomised, controlled, assessor-blind, parallel group, multicentre, multiple dose study assessing the dose-response 8.1 Clinical Trials Table 8.1: Number of subjects in each clinical study who received FE 999049 with the meanvalue of age, body weight, and body mass index (BMI) followed by range in brackets. Fur-ther details about subjects in each trial can be found in the manuscripts paper I, II, and III,respectively, at the end of this thesis.
relationship of FE 999049 in women undergoing an ART programme. 265 womenwho had been diagnosed with tubal infertility, infertility related to endometriosisstage I/II, unexplained infertility, or have a partner diagnosed with male factorinfertility were included. Table provides an overview of the personal demo-graphics for the subjects who received FE 999049. Randomisation of patientswere stratified according to AMH levels at time of screening: a concentration of5.0-14.9 pmol/L was defined as low and 15.0-44.9 pmol/L as high AMH level.
They received daily subcutaneous doses of either 5.18, 6.90, 8.63, 10.35, or 12.08µg (equivalent to 90, 120, 150, 180 and 210 IU, respectively) FE 999049 or 11 µg(150 IU) GONAL-F. To prevent a premature LH surge, a GnRH antagonist (0.25mg ganirelix acetate, ORGALUTRAN, MSD / Schering-Plough) was initiated onstimulation day 5 after the first dose and given daily throughout the stimulationperiod. Blood samples for FSH, LH, inhibin A, inhibin B, estradiol, progesterone,and testosterone measurements were collected immediately before the first admin-istration, at day 3 and day 5 after the first dose, and hereafter at least every secondday. When three follicles ≥15 mm were observed, visits and monitoring were per-formed daily. Doses were given until three or more follicles with a diameter ≥17mm were observed or for a maximum of 16 days. The cycle would be cancelled ifthere were either too many (more than 35 follicles ≥12 mm) or too few (less thanthree follicles ≥10 mm at day 10) growing follicles.
Analysis of serum FSH concentrations from the phase I studies was performed atFerring Pharmaceuticals A/S, Copenhagen, Denmark, with a validated immunoas-say based on electrochemiluminescence (MSD sectorTM Imager 2400) with a lowerlimit of quantification (LLOQ) of 0.075 µg/L. FSH samples from the Phase II000009 study were analysed at ICON Central Laboratories, Dublin, Ireland, with achemiluminescent immunometric assay (IMMULITE 2500 FSH (ROCHE), LLOQ: Time after dose (Days) Figure 8.1: The observed serum FSH concentration over time for all subjects in each treatmentgroup in CS01. Lines are mean values with standard error (SE) bars. Observed BQL measure-ments were set to LLOQ/2 in the plot and the day of administration to day 0. The grey linerepresents the LLOQ of 0.075 µg/L.
Time after dose (Days) Figure 8.2: The three subjects that are assumed to not have fully suppressed endogenous FSHdue to the second increase in observed FSH concentration after day 3. Subject 12 and 13 arefrom the 4.38 µg dose group and subject 24 is from the 8.76 µg dose group.
0.0052 µg/L). Inhibin B was measured by an enzyme linked immunosorbent assay(Gen II ELISA (Beckman Coulter)) with an LLOQ of 4.8 pg/mL.
To inspect the trend in the datasets and look for any abnormalities, plots for individual concentration-time profiles and mean serum FSH concentrations againsttime were initially created. Mean of observations for each of the 5 treatment groupsin CS01 is shown in Figure The 4.38 µg dose group have an odd second increasein FSH concentration starting at day 3 after administration. This pattern wasexamined further in the individual concentration-time profiles, where two subjects 8.2 Data Analysis & Modelling Table 8.2: Specifications about each of the three modelling datasets. First is given how manyFE 999049 dose groups, the dose range, and number of subjects in each dataset. Duration ishow many days subjects were monitored in the study. For the 000009 study the number is themaximum number of stimulation days. Number of total FSH and inhibin B measurements ineach dataset is listed. For FSH is also given how many of the measurements were below thequantification limit (BQL) with the percentage in brackets.
were identified to cause this increase. In addition, one subject in the 8.76 µg dosegroup showed a similar tendency (Figure A likely explanation for this increaseis that their endogenous FSH was not fully suppressed and hereby camouflagingthe true rFSH concentration after treatment. In particular the elimination rateof the drug cannot be properly estimated from these subjects. These subjects'concentration-time profiles were considered to not reflect the PK profile for FE999049 and were thus excluded from the analysis. Table gives details abouteach dataset used for modelling.
Data Analysis & Modelling Each of the datasets were analysed separately and described in paper I, paper II,and paper III, respectively, using the population PK(PD) modelling approach withnonlinear mixed effects models. The models were implemented in NONMEM 7.2.0(Icon Development Solutions, USA) using first order conditional estimationwith interaction for model parameter estimation. The statistical program R ver-sion 2.11.1 was used for data management, as well as making all graphicalrepresentations. VPCs were performed using PsN and plotted usingXpose The datasets were not balanced since measurements were randomly missing forsome subjects. In addition, FSH was measured more frequently than inhibin B inCS02. No action were taken for the single missing values in the modelling esti-mation process since they were missing at random. When the observed hormoneconcentrations were needed for testing covariate relationships the missing valueswere filled out in the following way. A subject without a baseline value were giventhe median of the population's baselines. For later time points the last observationwas carried forward such that a missing value was given the last measured value.
Baseline measurements were taken before dosing, and therefore represent the initial endogenous FSH level. All subjects in the CS01 study had a BQL measure-ment as FSH baseline value, meaning that their endogenous FSH level was notmeasurable. Whereas CS02 subjects had higher and measurable baseline values.
Due to the different extent of FSH BQL measurements in CS01 and CS02 datathey were handled differently in the model development. In paper I the M3 methodwas used since over 40% of the CS01 data were BQL. With only three BQLpoints in the CS02 data, it was chosen to exclude them (M1 method for themodel development in paper II.
Modelling Strategy For the population PK structural model, one- and two-compartment distributionmodels were tested. A potential delay in the absorption process was examinedby transit compartments and lag time. It was checked if an endogenous FSHcontribution to the total FSH concentration could be identified, if the endogenouslevel changed during the course of the trial, and if it influenced the results.
A combined additive and proportional error model was used for the residual If one of the parameters was too small or poorly estimated a reduced error model was tested. Variations in exposure between subjects were establishedby adding random effects (IIV) to the parameters, subsequently any correlationsbetween the identified random effects were examined.
The parameters and their random effects were plotted against potential covari- ates to investigate whether some of the IIV could be explained. If a trend wasseen in the plots, the parameter-covariate relationship was tested in the model.
After adding covariates it was checked if any of the IIV random effect parameterscould be removed. Covariates available in the CS01 data were limited to bodyweight, age, and height. In CS02 there were additionally measured LH, proges-terone, estradiol, and inhibin B hormone levels. The baseline values for the latterthree hormones, that are produced by the ovaries and affect the FSH productionby feedback mechanisms, were tested as covariates at the baseline FSH. Inhibin B 8.2 Data Analysis & Modelling was also tested as a time-varying covariate at the endogenous FSH level using thelongitudinal data. To get comparable population parameters between studies 65kgwas chosen as the normalisation factor when testing body weight as a covariatesince it was an average standard weight.
First the structural model was developed, then the random effects matrix iden- tified, and lastly covariates were added. However, the sub-models are intertwinedi.e. the choice of one sub-model affects the others, therefore, continuously dur-ing the model development several iterations of checking the significance of earlieradded or discarded model components were performed.
In the 000009 study PK data was sparse with a limited number of samples col-lected. Therefore, ka and V were fixed to the values from paper II in the PKPDmodel (paper III). The focus in paper III was to describe the exposure-responserelationship for FSH and inhibin B taking the hormone dynamics into account. Asimultaneous modelling approach was therefore used. Generally, the same mod-elling principles used for the PK models were applied in the PKPD model. Withthe extension of linking the PK model to the inhibin B concentration, the mod-elling becomes more complicated with additional types of interactions between thehormones and relations in the model to explore. The FSH stimulation of inhibinB was tried to be modelled as both an indirect and an indirect delayed effect withdifferent functions describing the effect. Likewise for the inhibin B inhibition ofendogenous FSH.
A turnover model, where no change in concentrations at time zero is assumed when the equations fulfil the initial condition, was used for the FSH and inhibin Bconcentrations. The intention of administering the GnRH antagonist was to avoida premature LH surge, but this also inhibits the production of FSH. To follow theprotocol this inhibition of endogenous FSH after day 5 should be included in themodel.
In addition to the covariates tested in the other models, any influence of AMH level was examined. Data was log-transformed and separate error models for FSHand inhibin B was used.
Evaluation Criteria Model development was guided by changes in the OFV, and graphical model as-sessments by goodness-of-fit plots and VPCs. A significance level of 0.05 was usedfor the χ2-test to discriminate between nested models. For two non-nested mod-els with equal number of parameters the model with lowest OFV was chosen. In addition, the precision of parameter estimates expressed as relative standard er-ror (RSE) were considered as well as keeping in mind if the parameter value wasrealistic. For testing covariates the significance was also evaluated by the changein %CV, i.e. how much the unexplained IIV decreased from adding the covariate.
The results of the data analyses and modelling are described in the three manuscripts,one for each of the three aims of this PhD thesis. Paper I and paper II presents thePK model developed from the phase I CS01 and CS02 data, respectively. PaperIII describes the PKPD model that relates the FSH concentration to the inhibinB PD response in the phase II 000009 study.
Pharmacokinetic Properties of FE 999049 There were some differences between the two phase I PK studies in the observedFSH levels at baseline, hence the datasets were not integrated but two separatePK models were developed in paper I and paper II with two alternative methodsfor handling BQL data. In paper I the FE 999049 PK was described followinga single dose (CS01 data). In paper 2 the PK was described following multipleadministration (CS02 data).
Pharmacokinetic Model (Paper I) From the CS01 data the best structural model was found to be a one-compartmentdistribution model with first order absorption. In order to adequately describe theabsorption phase a one compartment transit model was introduced to add a timedelay in the absorption from the dosing site to the central compartment. The PKmodel is described by the differential equations drFSHDS(t) = −ktrrFSHDS(t) drFSHT R(t) = ktrrFSHDS(t) − karFSHTR(t) = karFSHT R(t) − k rFSH(t).
rFSHDS(t), rFSHT R(t), and rFSH(t) is the FE 999049 amount left at the dosingsite, in the transit compartment, and in the central compartment at time t, respec-tively. The pre-dose amount in the central and transit compartment is zero, butthe dosing site is initiated by the amount of dose given. Since data was obtained after subcutaneous dosing, the parameters estimated were CL/F and V/F. To ob-tain the predicted serum concentration in the central compartment the amountrFSH(t) is divided by V/F. The absorption rate from the dosing site into the tran-sit compartment is ktr. FE 999049 is absorbed to the central compartment withrate ka, from where it is eliminated with rate k, which is given by CL/V . Anendogenous FSH supply to the central compartment could not be identified by themodel, and the measured FSH serum concentrations could therefore be regarded asFE 999049 with an insignificant amount of endogenous FSH. An IIV was detectedat CL/F, V/F, and ka, and the variances for the IIV random effect parameters onCL/F and V/F are positively correlated. A combined additive and proportionalerror model was necessary to describe the residual error.
No influence of dose at the parameters were found, indicating that the phar- macokinetics is linear. The only covariate identified with statistical significancewas body weight allometrically scaled at CL/F and V/F. A standard weight of 65kg was chosen as normalisation factor for the covariate effect. The importance ofbody weight being a factor influencing the parameters was further supported bya reduction in the unexplained IIV from 31.5 to 28.2% CV for CL/F and from46.4 to 44.3% CV for V/F upon inclusion of body weight in the model. Figureillustrates how CL/F and its random effect varied with weight. Adding bodyweight as a covariate do not change the individual parameters but the trend inthe random effects with weight is gone, implying that the relationship is describedproperly.
Final model parameters are listed in Table The graph with mean observed data and typical model predictions for each dose (Figure together with thediagnostic plots (Figure - indicate the model describes data well. Con-ditional weighted residuals cannot be calculated when the M3 method is used,because BQL data points do not provide a prediction but a probability. Instead,individual weighted residuals was used for the diagnostic plots. A two panel VPCillustrating both continuous and categorical (BQL) data was used For thismodel the 2.5th percentile of observations is not shown since it solely consists ofBQL points and therefore falls outside the plotting area for the top panel.
Using the PK model based on CS01 single dose data, the expected time-course of FE 999049 following multiple dose administration was simulated for three womenwith different body weights to illustrate the impact of body weight on FSH ex-posure. The resulting PK profiles in Figure vary substantially for the threeindividuals. The inclusion of body weight at CL/F and V/F clearly makes a differ-ence in the model predicted drug concentration. Through simulations of differentdoses it was found that on average a 100 kg woman would need doses of 18 µg toget a similar exposure as a 50 kg woman receiving 10 µg doses (Figure 9.1 Pharmacokinetic Properties of FE 999049 Figure 9.1: Illustration of covariate effect of body weight in the first model from the CS01data. Individual clearance (CL) values and respective individual random effects (ηCL) againstbody weight (points) in top and bottom panels, respectively, before (left) and after (right) bodyweight was added as a covariate. The relationship is illustrated with a smooth lowess line (brokenpurple) and a linear regression line (blue).
Table 9.1: Typical population pharmacokinetic parameter estimates obtained from modellingof CS01 data (paper I) with the relative standard error (RSE) in brackets. For CL/F and V/Fthe value is the typical value for a woman weighing 65 kg. The interindividual variability (IIV)is listed as the percentage coefficient of variation (CV) with RSE in brackets. F: bioavailability,CL/F: apparent clearance, V/F: apparent volume of distribution, ktr: absorption rate from thedosing site, ka: absorption rate to the central compartment.
Time after dose (Days) Figure 9.2: Illustration of observed FSH concentrations and model predictions for each treat-ment group from the model based on CS01 data. Points are mean of observations with standarderror (SE) bars. Observed BQL measurements were set to LLOQ/2 in the plot. Lines are typicalmodel prediction. The grey line represents the LLOQ of 0.075 µg/L.
Individual Predictions Figure 9.3: Goodness of fit plots for the model based on CS01 data. Left: Observationsagainst population predictions (purple points *) and individual predictions (blue points +) withthe unity line. The grey line represents the LLOQ of 0.075 µg/L. Right: Individual residualsagainst individual predictions (points) with a smooth lowess line.
9.1 Pharmacokinetic Properties of FE 999049 Time after dose (Days) Figure 9.4: Two panel visual predictive check for the CS01 model. The top panel shows theobservations above LLOQ (points) and the 50th and 97.5th percentiles of observations (purplelines). The 2.5th percentile of observations is not shown since it solely consists of BQL points.
The shaded areas are the simulated 95% confidence intervals (CI) for the 2.5th, 50th, and 97.5thpercentiles. The grey line represents the LLOQ of 0.075 µg/L. In the bottom panel the blue lineis the fraction of BQL observations with the 95% CI for the median from simulations.
µ 0.8ation ( 0.6 Time after first dose (Days) Figure 9.5: Simulations of body weight effect based on the model from CS01 single dose data.
The expected FSH concentration simulated after multiple dosing of 10 µg FE 999049 for threesubjects with different body weights.
µ 0.8ation ( 0.6 Time after first dose (Days) Figure 9.6: Simulations based on the model from CS01 single dose data of required dose to getsame exposure in two different women. Simulation results of the expected FSH concentrationafter multiple dosing of 10 µg FE 999049 to a woman weighing 50 kg (blue line) and 18 µg to awoman weighing 100 kg (purple line).
Modelling Endogenous FSH Levels (Paper II) The PK model developed from the CS01 data could not entirely be used as astarting point for paper II, since there were measurable endogenous FSH concen-trations at baseline (FSHbl) before dosing in the CS02 data that would have to beaccounted for. Additionally, there was no need for using the M3 method for theCS02 data. Therefore, initially a reduced model in form of a one-compartmentdistribution PK model with first order absorption and a combined additive andproportional error model was used as starting model. An endogenous FSH supplyhad to be added at first to correct for a clear under-prediction, especially at base-line. The total FSH amount at time t in the central compartment was therefore asum of the endogenous FSH and the exogenously administered rFSH: FSH(t) = FSHen(t) + rFSH(t).
The endogenous FSH was included in the model as a contribution to the differentialequation for the central compartment with zero order production rate constantkendo and was assumed to have the same elimination rate constant as FE 999049.
A random effect for IIV was found to be significant at CL/F, V/F, ktr, and FSHbl.
CL/F and V/F were positively correlated, and body weight normalised to 65 kg wasan allometrically scaled covariate at CL/F and V/F with statistical significance.
Adding a transit compartment at this stage was not significant.
The individual estimated FSHbl values were evaluated against observed base- line values of estradiol, inhibin B, and progesterone (Figure . Of these proges- 9.1 Pharmacokinetic Properties of FE 999049 Lowess lineLinear regression line Estradiol (pg/mL) Inhibin B (pg/mL) Progesterone (ng/mL) Baseline concentrations Figure 9.7: Relationship in the CS02 model and data between individual estimated endoge-nous baseline values FSHbl and observed estradiol, inhibin B, and progesterone baseline values(points), respectively. Illustrated with a smooth lowess line (broken purple) and a linear regres-sion line (blue).
terone baseline was the only significant baseline relationship in the model whenincluded as a negative correlation with FSHbl. A linear, exponential, Imax, andpower function was tested for describing the covariate relationship. Of these apower function was most significant (Figure Furthermore, observed inhibinB levels (InhB(t)) was a time-varying inhibitory covariate at the endogenous FSHproduction and was best described by an Imax function (see equation Figureillustrates how the inhibin B concentration suppress the model predicted en-dogenous FSH concentration over time. It was re-tested if a transit compartmentcould be added in the absorption process and it was significant better after the co-variate relations had been added. The overall model structure from first in humandata was thus confirmed. Accordingly, the PK model based on CS02 data consistsof equation and from the first PK model for the absorption process, andin order to incorporate the study design differences the change in FSH amount inthe central compartment is described by + karFSHT R(t) − k FSH(t). (9.5) Values of FSHbl was estimated for each subject to be the initial concentration inthe central compartment before dosing, thus the initial condition for equation in amount is FSHbl*V/F. The parameter IC50 is the concentration yielding half ofmaximum inhibin B suppression. It was assumed that at time zero the change inFSH amount is zero as a turnover model and consequently kendo is given by theother parameters at pre-dose values.
Typical model prediction Progesterone ( µg/L) Figure 9.8: Individual estimated endogenous baseline values FSHbl from the CS02 modelagainst observed progesterone baseline values (points) with a smooth lowess line (broken purpleline). The blue solid line is the typical population relationship described by a power function.
Model FSH predictionsObserved FSH baselineObserved Inhibin B Inhibin B Concentr Endogenous FSH concentr Time after first dose (Days) Figure 9.9: Individual hormone concentration profiles over time for subjects in CS02. Thebroken blue line is the observed endogenous FSH baseline level when assuming it is constantthroughout the trial. The solid blue line is the model predicted endogenous FSH level obtainedwith inhibition by the observed inhibin B levels (purple line) over time. The number at eachsubplot is the subject ID number.
9.1 Pharmacokinetic Properties of FE 999049 Figure 9.10: Compartment diagram of the PK model from the CS02 data. Contributions tothe total FSH amount in the central compartment (FSH(t)) are FE 999049 from the transitcompartment (rFSHTR(t)) and endogenous FSH (FSHen(t)). The endogenous FSH productionrate (kendo) is inhibited by inhibin B concentrations (InhB(t)). ktr: absorption rate from thedosing site, ka: absorption rate to the central compartment, k: elimination rate.
The model is illustrated by the compartment diagram in Figure and the parameters are listed in Table The VPC in Figure indicates that themodel and estimated parameters adequately describe data.
The final model was re-evaluated to see if any covariate relationships or corre- lations had become excessive after others had been added. A significant increasein OFV as a result of removing any of the effects revealed that the model could notbe reduced (Table Moreover, removing body weight as a covariate increasedthe unexplained IIV from 15.6 to 18.1% CV for CL/F and from 18.4 to 22.0% CVfor V/F. Not including inhibin B as a covariate in the final model resulted in anincrease from 83.4 to 164.9% CV at ktr. IIV at FSHbl increased from 27.8 to 32.6%CV when the progesterone effect was removed.
Using the model based on CS02 multiple dose data the effect of having body weight as a covariate was evaluated by simulating the expected FSH exposure forthree subjects with different weights. In the simulation the same observed inhibinB and progesterone values were used for all three subjects and simulations wereperformed with frequent time points to get the full dosing profile after 7 dailydoses of 10 µg FE 999049. The simulated PK profiles follow the relationship oflower exposure with higher body weight (Figure Table 9.2: Typical population parameter estimates obtained from modelling of the CS02 data(paper II) with the relative standard error (RSE) in brackets. For CL/F and V/F the valueis the typical value for a woman weighing 65 kg. The interindividual variability (IIV) is listedas the percentage coefficient of variation (CV) with RSE in brackets. F: bioavailability, CL/F:apparent clearance, V/F: apparent volume of distribution, ktr: absorption rate from the dosingsite, ka: absorption rate to the central compartment, FSHbl: endogenous FSH baseline, Progblef:power exponent for progesterone baseline covariate effect, IC50: inhibin B concentration yieldinghalf suppression.
Time after first dose (Days) Figure 9.11: Visual predictive check for the CS02 model showing the FSH observations (points)and the 2.5th, 50th, and 97.5th percentiles of observations (purple lines). The shaded areas arethe simulated 95% confidence intervals (CI) for the 2.5th, 50th, and 97.5th percentiles.
9.1 Pharmacokinetic Properties of FE 999049 WT at CL/F and V/F Progesterone effect Table 9.3: The resulting increase in objective function value (dOFV) when removing covariates,random effect η's for IIV, or the correlation between CL/F and V/F (cov(CL/F,V/F)). Whenremoving η at either CL/F or V/F it is also necessary to remove the correlation, therefore is df =2. WT: body weight, df: degrees of freedom, F: bioavailability, CL/F: apparent clearance, V/F:apparent volume of distribution, ktr: absorption rate from the dosing site, FSHbl: endogenousFSH baseline.
Time after first dose (Days) Figure 9.12: Illustration of body weight effect on FSH exposure using the model from CS02multiple dose data. The FSH concentration-time profiles are obtained from simulations of threesubjects with different body weights receiving multiple dosing of 10 µg FE 999049.
A simultaneous dose-concentration-response PKPD model with inhibin B as PDendpoint was developed from the phase II 000009 data. An indirect responseturnover model was used to describe the inhibin B response upon FSH stimulation.
The PKPD relationship was modelled simultaneously in order to include a negativefeedback of inhibin B concentration at the endogenous FSH production rate.
The PK model from paper II without hormone covariates was used as a starting point. In the 000009 data the PK information was very sparse and not enoughto estimate all PK parameters, so ka and V/F were fixed to values from paper II.
After re-testing the PK structure in the simultaneous PKPD model the PK partwas the one from paper II with a few alterations.
An Imax function was most significant in describing the inhibition of endogenous FSH by the predicted inhibin B concentrations as opposed to using the observedas in paper II. The GnRH antagonist given to suppress the LH secretion from day5 also affects the endogenous FSH. Therefore, an inhibitory effect at kendo from theantagonist had to be added after day 5 for the model to be in accordance with thetrial design. This meant two types of suppression at the endogenous FSH whichcaused estimation problems. It was tested if the suppression was total after day 5or if one of the effects took over, but both were significant. So IC50 for the inhibinB inhibitory effect on endogenous FSH was fixed to the value obtained in paper IIfor a more stable model.
A power function with exponent λ for the FSH stimulation of the inhibin B production rate kin was found to describe data best. The power function was givenby where the total FSH amount (FSH(t)) is divided by V/F to achieve concentrationsand it is normalised with endogenous FSH baseline concentrations.
For both FSH stimulation and inhibin B inhibition linear and exponential rela- tionships were tested but they either gave significant worse OFV, a worse fit withparameters poorly estimated, or the model became unstable. It was neither pos-sible having an Emax model describing the FSH stimulation because the EC50 andEmax parameters varied greatly and was estimated poorly with high RSE. Therewere indications of a delayed response but data did not support estimation of aneffect compartment.
9.2 Semi-Mechanistic Dose-Concentration-Response Model (Paper III) The model compartment diagram is illustrated in Figure and the corre- sponding differential equations for the model are drFSHDS(t) = − ktrrFSHDS(t) drFSHT R(t) = ktrrFSHDS(t) − karFSHTR(t) = kendo(1 − ANTAef) 1 − where ANTAef is the suppressive effect of the GnRH antagonist administered afterday 5 and is given by Individual inhibin B baseline values (InhBbl) were estimated as initial conditionsfor equation since pre-dose inhibin B concentrations were present. Initialconditions for equation were estimated individual endogenous FSH baselineconcentrations (FSHbl) multiplied by V/F to get FSH amount. The eliminationrate constant kout for inhibin B varied greatly with poor estimation and was there-fore fixed. Since it was a turnover model the inhibin B production rate kin wasgiven by the other parameters at initial conditions.
The parameters CL/F, ktr, FSHbl, InhBbl, and λ varied between subjects and an IIV random effect was added at these parameters. An IIV could not be identifiedfor the fixed V/F parameter, but body weight was still tested and found to be astatistically significant covariate at V/F. Furthermore, body weight was a covariateat CL/F, ktr, and λ with a reduction in the variation from 27.2, 51.1, and 48.9%CV to 23.1, 48.3, and 42.7% CV, respectively. A full Ω covariance matrix wassignificant better and all correlations were well estimated with no extreme values.
Separate combined additive and proportional error models were added for FSHand inhibin B.
The PKPD model parameters are given in Table The mean predictions fit observations nicely (Figure but in the VPC (Figure an over-predictionfor the lowest dose and an under-prediction for the highest dose is observed for Figure 9.13: Compartment diagram of the PKPD model in paper III. Contributions to the totalFSH amount in the central compartment (FSH(t)) are FE 999049 from the transit compartment(rFSHTR(t)) and endogenous FSH (FSHen(t)). The endogenous FSH production rate (kendo) isinhibited by predicted inhibin B concentrations (InhB(t)) and after day 5 also by a gonadotropinreleasing hormone (GnRH) antagonist. The inhibin B production rate (kin) is stimulated byFSH. ktr: rFSH absorption rate from the dosing site, ka: rFSH absorption rate to the centralcompartment, k: FSH elimination rate from the central compartment, kout inhibin B eliminationrate.
9.2 Semi-Mechanistic Dose-Concentration-Response Model (Paper III) Power exponent for body weight at Table 9.4: Typical population parameter estimates obtained from modelling of the 000009 data(paper III) with the relative standard error (RSE) in brackets. For CL/F, V/F, ktr, and λ thevalue is the typical value for a woman weighing 65 kg. The interindividual variability (IIV) islisted as the percentage coefficient of variation (CV) with RSE in brackets. F: bioavailability,CL/F: apparent clearance, V/F: apparent volume of distribution, ktr: absorption rate from thedosing site, ka: absorption rate to the central compartment, FSHbl: endogenous FSH baseline,InhBbl: inhibin B baseline, IC50: inhibin B concentration yielding half of maximum suppression,GnRHanta: suppressive effect of the gonadotropin releasing hormone (GnRH) antagonist, kout:inhibin B elimination rate constant, λ: the power exponent for the FSH stimulation at inhibinB. * fixed to values from paper II. † parameter fixed value.
FSH concentrations and inhibin B is over-predicted for the highest dose. Since thefrequency subjects come in to the clinic for measurements depends on fulfilmentof pre-set criteria for follicle number and size, different number of subjects aremeasured per day. There are even days with only one subject and therefore theobserved percentiles in the VPC collapse to the same value. A log-linear relation-ship was tested for the FSH stimulation instead of a power function. It improvedthe VPCs for inhibin B by reducing the over-prediction, but the OFV increasedby 100.
Simulations of FSH and inhibin B concentrations after multiple dosing in pa- tients of different body weights were performed using the simultaneous PKPD Time after first dose (Days) Inhibin B concent Time after first dose (Days) Figure 9.14: Illustration of observed FSH (top) and inhibin B (bottom) concentrations andmodel predictions for each treatment group in the 000009 study. Points are mean of observationswith standard error (SE) bars. Lines are typical model predictions.
9.2 Semi-Mechanistic Dose-Concentration-Response Model (Paper III) Time after first dose (Days) Inhibin B concentr 4000 Time after first dose (Days) Figure 9.15: Visual predictive check for FSH at the top and inhibin B at the bottom fromthe PKPD model. The points are observations with purple lines for the 2.5th, 50th, and 97.5thpercentiles of observations. The shaded areas are the simulated 95% confidence intervals (CI)for the 2.5th, 50th, and 97.5th percentiles.
50 kg 75 kg100 kg Time after first dose (Days) Inhibin B concentr Time after first dose (Days) Figure 9.16: Illustration of simulated FSH (top) and inhibin B (bottom) concentrations forthree patients of different weights. The patients received 7 doses of 10 µg FE 999049.
model. The resulting concentration-time profiles decrease with increasing bodyweight for both hormones (Figure A sigmoidal Emax dose-response model based on the 000009 study data was used for simulations to obtain 95% CI for the inhibin B response to the differentdoses. Compared to 95% CI of observations the interval is narrower with increasedprecision in the prediction (Figure Due to the over- and under-predictionsadditional refinements may be needed before the simultaneous PKPD model isused for simulation of dose-response relationship.
9.2 Semi-Mechanistic Dose-Concentration-Response Model (Paper III) Mean with 95% interval
ObservationsEmax model Figure 9.17: Illustration of how modelling of dose-response can increase precision in inhibinB predictions. Blue bars are 95 % confidence intervals (CI) of observations around the mean.
The purple line is the mean model prediction with 95 % CI for the doses from simulations witha sigmoidal Emax dose-response model.
Discussion & Perspectives Drug development is an ongoing search for better, more efficient, cheaper, safer, ordifferent drugs from existing ones, as well as developing new treatment strategies,or finding compounds for new therapy areas. For new compounds the minimaleffective dose should be identified to avoid side effects. A drug with great efficacymight not be utilised to its full capacity if not administered optimally. What theoptimal treatment strategy is, is very likely to differ between patients. Major chal-lenges for individualising treatment are to identify how the dose should be adjustedbased on patient-specific factors, what the dosing intervals should be, and whetherchanges should be made in long term treatment. An important tool for a moreefficient and informative drug development path is using MBDD with a populationmodelling approach for analysing data, simulating different scenarios, and productcomparison in order to assess the potential of a new drug and investigate likelytreatment strategies. In addition, significance of potential factors influencing drugexposure and effect can be quantified. Thus MBDD can support decision making,dose selection, and study designs.
In this PhD thesis the PK and PD properties of FE 999049 were characterised through development of population PK and PKPD models based on clinical data.
FE 999049 differ from marketed rFSH products by being expressed in a human cellline instead of a CHO cell line. The cause of infertility varies greatly, it is there-fore important not only to find a safe dose but also an optimal individual dosingscheme according to patient-specific factors to increase success rate in pregnancy.
Hence, there is a demand for infertility therapy drugs with innovative personalisedtreatment strategies to catch the great diversity in causes and extent of infertility.
Based on the first-in-human data with FE 999049 (CS01) an initial PK model was developed in paper I to investigate the pharmacokinetics of FE 999049 aftersingle dose administration. The resulting model was a one-compartment distribu-tion model with first order elimination and a delayed absorption through a transitcompartment. The pharmacokinetics of a drug is characterised by its PK param-eters, i.e. it should usually be described by one PK model. However, what can beobserved and tested in the model and how many parameters that can be estimatedis data driven, thus it depends greatly on the study design. Due to differences inbaseline values, number of BQL measurements, and method for handling the BQLmeasurements, the first model from paper I could not directly be confirmed inpaper II with the CS02 dataset. After adjusted according to differences in study 10 Discussion & Perspectives design, the overall model structure was confirmed in paper II. The PK datasetswere relatively sparse and no i.v. data was used. That these data only support aone-compartment PK model for describing the pharmacokinetics of an FSH prod-uct is in agreement with previous results from the existing FSH models discussedin Chapter section Using the knowledge about the reproductive hormone dynamics presented in Chapter it was investigated if any of the ovarian hormones could be identifiedas a covariate in the model. It was possible to include the inhibitory effect ofprogesterone and inhibin B at endogenous FSH in the model developed from theCS02 data in paper II. As discussed in Chapter section inhibin B has beenlisted in the literature as one of the valuable predictors of ovarian response togonadotropin therapy, and was therefore chosen as an interesting biomarker forassessment in this PhD. The observed influence of inhibin B at endogenous FSHin paper II was further investigated in paper III, where the dynamics betweenFSH and inhibin B were described by a simultaneous dose-concentration-responsePKPD model based on the phase II 000009 data.
In all three models body weight was found to be a statistically significant covariate at CL/F and V/F, resulting in lower drug exposure with higher bodyweight. These findings are in accordance with the existing FSH modelling resultspresented in Chapter and support the evidence for body weight being a factoraffecting the pharmacokinetics of FSH compounds. In addition in paper III bodyweight was found to affect the absorption rate and the FSH stimulation of inhibinB.
The two lowest doses in the CS01 data did not seem to fulfil dose linearity as their mean profiles fell together (Figure and 70% of the 220 measurementswere BQL. It was considered if these doses at all would add any information tothe analysis, or even worse cause bias of parameters. This could be true if theBQL measurements were ignored. The possibility of including BQL measurementsis one of the advantages of population modelling. By using the M3 method theBQL measurements were accounted for, such that the information - even thoughlimited - from the two lowest doses were utilised. No information is lost and lessbias is induced as no inaccurate assumptions of the values are used, e.g. assumingthey are zero or LLOQ/2 (see Chapter section All pre-dose FSH measurements in the CS01 data were BQL indicating that endogenous FSH was successfully suppressed by the COCs. Nonetheless, threesubjects showed an additional peak of FSH levels several days after FE 999049administration, which was very unlikely to be caused by FE 999049.
There is no confirmatory conclusion for the reason of this second peak, a pos- sible explanation is lack of adherence in these three subjects to oral contraceptivesfor suppression of endogenous FSH. As a first rule, all data should be included but these three individuals were excluded from the analysis as their FSH levelswere judged not to reflect the pharmacokinetics of the exogenously administeredFE 999049.
The fact that choices of error and covariate model affect the structural model (see Chapter section was experienced in the model development in paper II.
At first the model from paper I could not be used with the CS02 data. Adding atransit compartment to an initial basic model improved the parameter estimatesbut the OFV was worse. When the best covariate relationships were found and theendogenous FSH was properly described, the structural model was re-evaluated,and it was then significantly better with a transit compartment in the absorptionprocess.
Differences in the absorption rates and their IIV were observed. In paper I an IIV was placed on ka and in paper II the IIV was higher and at ktr. Moving therandom effect in any of the models to the other absorption rate constant resultedin a slightly higher OFV and worse precision. It was neither possible to have anIIV at both rate constants. There were indications of variation in both absorptionrate constants but in the current studies it was not possible to identify both. Wheninhibin B was not included as a time-varying covariate in the model in paper II,the variation at ktr increased. This could imply that if the endogenous FSH inputis not described properly, the variation in FE 999049 absorption increases in orderto explain the variation otherwise caused by differences in endogenous FSH levelsbetween subjects.
Pooling of the two phase I data-sets and including i.v. data could maybe facilitate enough information for a two-compartment model to be supported bythe data and a clearer identification of variations in the absorption processes.
Few samples per subject in the phase II 000009 data also affected the model development. Approximations, assumptions, and fixation of parameters were nec-essary in order to get a stable PKPD model in paper III. The PK information inthe 000009 data was very sparse and not enough to estimate all PK parameters,so ka and V/F were fixed to values from paper II. The PK parameters from paperII were chosen over the values from paper I, since the CS02 data include moreFSH measurements and few BQL observations. In addition IC50 and kout had tobe fixed to stabilise the model.
The inclusion of a full covariance matrix indicates correlation between PK and PD parameters and significant variation between subjects. It was tested ifIIV should be placed on more or other parameters, but no better combinationwas found. In the reproductive system FSH and inhibin B are correlated, thusit makes sense the model parameters describing the relationship are as well. Inaddition, other ovarian hormones are involved in the dynamics as described inChapter The over- and under-predictions observed can thus be due to that the 10 Discussion & Perspectives full endocrine dynamics is not considered. Furthermore, the data was not sufficientto support more advanced functions for describing the relations. In earlier modelswhere an effect compartment was included and the FSH stimulation of inhibinB was modelled with an Emax model, the VPCs were better with less over- andunder-prediction. In spite of the better fit such models were discarded due topoorly estimated parameters and model instability. More data would probably beneeded to improve the model and for describing the FSH-inhibin B relationshipmore accurately, which might require more complicated functions, e.g. an Emaxmodel for the FSH stimulation of inhibin B production rate.
Subjects did not stay in the trial the same number of days, which is not ac- counted for in the model and can add to the reason for imprecise predictions athigh values. An option could be to model this as drop-out subjects. A few ex-tremely high inhibin B concentrations were observed, so it was looked into if theseaffected the results. The subjects with extreme values were identified and excludedbut it did not improve the model or change the VPCs.
Through simulations the models were used to predict outcomes of different doses to women of different body weights. In all three models the simulationswere performed for three subjects weighing 50 kg, 75 kg, and 100 kg who received7 daily doses of FE 999049. Consistently the simulated FSH exposure decreasedwith increasing body weight (Figure The second PK model basedon CS02 data included an endogenous FSH supply and initial exposure levelstherefore started above zero and was likewise slightly higher over time comparedto the first PK model based on CS01 data where no endogenous FSH levels weremeasurable. The PKPD model was based on patient data from a phase II trial.
The patients would likely have different endocrine profiles and they have initialhigher endogenous levels than the gonadotropin suppressed female volunteers inthe phase I trials. This is also reflected in the simulations where the PK profileslook different from the ones obtained with the PK models. After dosing the FSHexposure decrease to an endogenous level. In the PKPD model it can becomelower than pre-dose levels due to suppression by the GnRH antagonist and inhibinB levels. Since FSH stimulates the inhibin B production the simulated inhibinB response in the PKPD model also decrease with increasing body weight. Inaddition, body weight affects the FSH stimulation of inhibin B thus the extentof difference between the concentration-time profiles are different for inhibin Bresponse compared to FSH exposure. Body weight play a role in the variation inexposure and thus response but other factors also affect the exposure. EndogenousFSH and variation in its level over time is likely an influential factor too.
The PKPD model would need more work before further simulations can be performed. Then, the potential of inhibin B as a marker for ovarian responsecould be investigated and be related to later clinical PD endpoints like oocytes retrieved. It could be interesting to test the hypothesis from the literature dis-cussed in Chapter about how poor and good responders are related to inhibinB levels and changes during treatment. There are evidence of inhibin B being thefirst PD marker measurable during infertility treatment with FSH products andan indicator for initial follicle development in response to FSH. Hence, it couldbe useful to be able to predict inhibin B levels in a patient to determine if thedose is effective. Individual FSH thresholds cannot be measured easily but whensimulating both FSH and inhibin B, the inhibin B increase implies that follicleshave started their gonadotropin dependent growth and thus the FSH thresholdhas been surpassed. In addition, the model could be used to simulate the typicaldose-response relationship for inhibin B to see what dose range is required to geta proper response in inhibin B. If the inhibin B response is not sufficient it in-dicates that no antral follicles have been recruited and entered the gonadotropindependent growth phase.
In decision making and dose selection is needed to know the response of dif- ferent doses. A pre-set criterion for a minimum response or a desired responseinterval can be set. If only looking at observed data it can be difficult to de-termine whether the criterion has been met due to an often large variation inthe population. Modelling and simulation can potentially improve precision bypredicting typical response. Using a sigmoidal Emax dose-response model basedon the 000009 study data, better precision in the predicted response to differentdoses was achieved compared to confidence intervals for observed data obtainedwith traditional statistical methods. The precision would be expected to improvefurther when using the semi-mechanistic PKPD model to simulate the FE 999049dose-inhibin B response compared to when using the empirical sigmoidal Emaxdose-response model, and can thus improve decision making and provide a bettertool for dose selection.
Data from three clinical trials were used to characterise the PK and PD propertiesof FE 999049, a novel human rFSH for controlled ovarian stimulation in ART.
Using a population approach with nonlinear mixed effects models the PK profileof FE 999049 was successfully described by a one-compartment model, both aftersingle and repeated administration. A semi-mechanistic PKPD model was devel-oped to describe the dynamics between rFSH, inhibin B, and endogenous FSH.
Inhibin B was chosen as the PD endpoint because it has been suggested to be theearliest response marker to FSH treatment and an indicator of follicular growth.
From these model-based exploratory analyses it can be concluded that • is a statistically significant covariate explaining some of the variation in the PK parameters CL/F and V/F in the models.
– The resulting effect is that FE 999049 exposure decrease with in- creasing body weight: A patient weighing twice as much as anotherpatient would need a 1.8 times higher dose to get the same FSHconcentration.
• affects the FSH stimulation of inhibin B in the PKPD model.
– With higher body weight the same FSH concentration has a higher stimulatory impact on inhibin B production rate.
2. Endogenous FSH levels • can significantly contribute to the measured serum FSH concentrations, hence if not accounted for in the model – the FSH exposure is under-predicted.
– a bias is induced in the PK parameter estimates.
• depends on inhibin B and progesterone levels.
• change over time and has to be described properly in the model, other- – unexplained parameter variation increase, for the absorption rate constant it is doubled.
– a transit compartment cannot be included, thus the structural model change.
– a bias is induced in the PK parameter estimates.
• are thus not constant throughout treatment and baseline correcting data might be wrong.
• significantly inhibit endogenous FSH production rate.
• production rate constant is stimulated by total FSH levels.
• response decrease with increasing body weight.
• levels vary greatly between subjects.
These results support the current findings presented in the literature of body weight being an important factor in dosing of FSH products and resulting ex-posure. Furthermore, it was identified that inhibin B response, stimulated byFSH exposure, also decrease with increasing body weight. The advantage of thepopulation PK models developed in this PhD work is that they can be used forsimulating different scenarios of body weight and dose, they take into account theendogenous FSH for more accurate estimation of parameters and prediction, andhave specifically identified variation in the population parameters.
The PKPD model has the potential of predicting a specific patients PK and PD profile, simulate overall dose-response relationship, and hereby aid in decisionmaking. For FSH products in the development phase, such a model can giveindications of required dosing range or support a go/no-go decision based on if asufficient inhibin B response can be achieved.
Thus, population PKPD modelling is a useful tool for analysing clinical data and the possible applications seems endless. Even though MBDD is a growingfield it is not utilized fully in the pharmaceutical industry. Overall in this thesisit was demonstrated how population modelling can be used to gain informationfrom clinical data and it has been emphasised that many disciplines need to beintegrated in order to produce reliable and useful models.
European Medicines Agency 2007., Guideline on Strategies to Identifyand Mitigate Risks for First-in-Human Clinical Trials with InvestigationalMedicinal Products. U.S. Food and Drug Administration 2005., Guidance for industry: Estimat-ing the maximum safe starting dose in initial clinical trials for therapeuticsin adult healthy volunteers. European Medicines Agency 2007., Guideline on Reporting the Resultsof U.S. Food and Drug Administration 1999., Guidance for Industry: Rutstein, S. O. and Shah, I. H. 2004., Infecundity, infertility, and childless-ness in developing countries. Demographic and Health Surveys (DHS) Com-parative reports No. 9. Calverton, Maryland, USA: ORC Macro and theWorld Health Organization. Zegers-Hochschild, F., Schwarze, J., and Alam, V. 2008. Infertility. Heggen-hougen, K. (ed.), International Encyclopedia of Public Health, pp. 576–587,Academic Press.
Smith, S., Pfeifer, S., and Collins, J. 2003. Diagnosis and management offemale infertility. JAMA, 290:13: 1767–1770.
van der Poel, S. 2012. Historical walk: the HRP Special Programme andinfertility. Gynecol. Obstet. Invest , 74:3: 218–227.
Mascarenhas, M. N., Flaxman, S. R., Boerma, T., Vanderpoel, S., andStevens, G. A. 2012. National, Regional, and Global Trends in InfertilityPrevalence Since 1990: A Systematic Analysis of 277 Health Surveys. PLoSMed., 9:12: 1–12.
Snick, H., Snick, T., Evers, J., and Collins, J. 1997. The spontaneous preg-nancy prognosis in untreated subfertile couples: the Walcheren primary carestudy. Hum. Reprod., 12:7: 1582–1588.
Snick, H., Evers, J., and Collins, J. 2005. An update on the age of subfertilecouples in Walcheren: age at registration mirrors increasing age at first birth.
Hum. Reprod., 20:2: 572–573.
de Graaff, A., Land, J., Kessels, A., and Evers, J. 2011. Demographic ageshift toward later conception results in an increased age in the subfertilepopulation and an increased demand for medical care. Fertil. Steril., 95:1:61–63.
U.S. Food and Drug Administration 2004., Innovation or Stagnation:Challenge and Opportunity on the Critical Path to New Medical Products.
Pharmaceutical Research and Manufacturers of America 2007., Drug Dis-covery and Development: Understanding the R&D Process. Lipsky, M. S. and Sharp, L. K. 2001. From idea to market: the drug approvalprocess. J. Am. Board Fam. Pract., 14:5: 362–367.
U.S. Food and Drug Administration 2014., The FDA's Drug Review Pro-cess: Ensuring Drugs Are Safe and Effective. Morgan, S., Grootendorst, P., Lexchin, J., Cunningham, C., and Greyson,D. 2011. The cost of drug development: a systematic review. Health Policy,100:1: 4–17.
Collier, R. 2009. Rapidly rising clinical trial costs worry researchers. CMAJ ,180:3: 277–278.
Vozeh, S., Steimer, J., Rowland, M., Morselli, P., Mentre, F., Balant, L.,and Aarons, L. 1996. The use of population pharmacokinetics in drug devel-opment. Clin. Pharmacokinet., 30:2: 81–93.
Sheiner, L. 1997. Learning versus confirming in clinical drug development.
Clin. Pharmacol. Ther., 61:3: 275–291.
Lalonde, R., et al. 2007. Model-based drug development. Clin. Pharmacol.
Ther., 82:1: 21–32.
Zhang, L., Pfister, M., and Meibohm, B. 2008. Concepts and Challenges inQuantitative Pharmacology and Model-Based Drug Development. AAPS J.,10:4: 552–559.
Sheiner, L. and Steimer, J. 2000. Pharmacokinetic/pharmacodynamic mod-eling in drug development. Annu. Rev. Pharmacol.Toxicol., 40: 67–95.
Aarons, L., Karlsson, M., Mentre, F., Rombout, F., Steimer, J., and vanPeer, A. 2001. Role of modelling and simulation in Phase I drug development.
Eur. J. Pharm. Sci., 13:2: 115–122.
Gieschke, R. and Steimer, J. 2000. Pharmacometrics: modelling and simu-lation tools to improve decision making in clinical drug development. Eur.
J. Drug Metab. Pharmacokinet., 25:1: 49–58.
Breimer, D. 2008. PK/PD modelling and beyond: impact on drug develop-ment. Pharm. Res., 25:12: 2720–2722.
Ploeger, B., Van der Graaf, P., and Danhof, M. 2009. Incorporating re-ceptor theory in mechanism-based pharmacokinetic-pharmacodynamic (PK-PD) modeling. Drug Metab. Pharmacokinet., 24:1: 3–15.
Zandvliet, A. S., de Greef, R., de Haan, A. F., Ijzerman-Boon, P. C.,Marintcheva-Petrova, M. Z., Mannaerts, B. M., and Kerbusch, T. 2011. Keysof collaboration to enhance efficiency and impact of modeling and simulation.
AAPS Adv. Pharm. Sci., pp. 131–148, no. 7, Springer Verlag.
Milligan, P., et al. 2013. Model-based drug development: a rational approachto efficiently accelerate drug development. Clin. Pharmacol. Ther., 93:6:502–514.
Allerheiligen, S. R. B. 2014. Impact of Modeling and Simulation: Myth orFact? Clin. Pharmacol. Ther., 96:4: 413–415.
Bonate, P. L. 2014. What Happened to the Modeling and Simulation Revo-lution? Clin. Pharmacol. Ther., 96:4: 416–417.
Strauss, J. F. and Barbieri, R. L. (eds.) 2009. Yen and Jaffe's Reproduc-tive Endocrinology: Physiology, Pathophysiology, and Clinical Management ,chap. 1, 2, 4, 5, 7, 8. Saunders Elsevier, 6th edition.
Fritz, M. A. and Speroff, L. 2011. Clinical Gynecologic Endocrinology andInfertility, chap. 2, 3, 5, 6. Lippincott Williams and Wilkins, eighth edition.
Levy, T. and Ben-Rafael, Z. 1996. Pharmacokinetics of gonadotropin ther-apy. Reprod. Med. Rev., 5:1: 13–35.
Simoni, M. and Nieschlag, E. 1995. FSH in therapy: Physiological basis, newpreparations and clinical use. Reprod. Med. Rev., 4:3: 163–177.
Fevold, H. 1931. The gonad stimulating and the luteinizing hormones of theanterior lobe of the hypophesis. Am. J. Physiol. – Legacy Content , 97:2:291–301.
Crowe, S., Cushing, H., and Homans, J. 1910. Experimental Hypophysec-tomy. Bull. Johns Hopkins Hospital , 21:230: 127–169.
Zondek, B. and Aschheim, S. 1925. Experimentelle Untersuchungen ¨ Funktion und das Hormon des Ovariums, gepr¨ uft am biologischen Testob- akologie, 125:3: 581–585.
Uber die Funktion des Ovariums. Zeitschrift f¨ akologie, pp. 372–380.
Ludwig, M., Felberbaum, R., Diedrich, K., and Lunenfeld, B. 2002. Ovar-ian stimulation: from basic science to clinical application. Reprod. Biomed.
Online., 5 Suppl 1: 73–86.
Eldar-Geva, T., Robertson, D., Cahir, N., Groome, N., Gabbe, M., Maclach-lan, V., and Healy, D. 2000. Relationship between serum inhibin A and Band ovarian follicle development after a daily fixed dose administration ofrecombinant follicle-stimulating hormone. J. Clin. Endocrinol. Metab., 85:2:607–613.
Ben-Rafael, M., Levy, M., and Schoemaker, M. 1998. Pharmacokineticsof Follicle-Stimulating Hormone: Clinical Significance. Fertil. Steril., 69:3,Supplement 2: 40S–49S.
Schipper, I., Hop, W., and Fauser, B. 1998. The follicle-stimulating hormone(FSH) threshold/window concept examined by different interventions withexogenous FSH during the follicular phase of the normal menstrual cycle: du-ration, rather than magnitude, of FSH increase affects follicle development.
J. Clin. Endocrinol. Metab., 83:4: 1292–1298.
Macklon, N. and Fauser, B. 1998. Follicle development during the normalmenstrual cycle. Maturitas, 30:2: 181–188.
Lunenfeld, B. 1963. Treatment of anovulation by human gonadotropins. J.
Int. Fed. Gynaecol. Obstet., 1:1: 153–167.
Igarashi, M. and Matsumoto, S. 1957. Induction of human ovulation by in-dividualized gonadotrophin therapy in two phases. Am. J. Obstet. Gynecol.,73:6: 1294–1298.
Jungck, E. and Brown, W. 1952. Human pituitary gonadotropin for clinicaluse preparation and lack of antihormone formation. Fertil. Steril., 3:3: 224–229.
Maddock, W., Leach, R., Tokuyama, I., Paulsen, C., Nelson, W., Jungck,E., and Heller, C. 1956. Antihormone formation in patients receiving go-nadotrophin therapy. Acta Endocrinol. Suppl. (Copenh), 23:Suppl 28: 55–62.
Gemzell, C. 1965. Induction of ovulation with human gonadotropins. RecentProg. Horm. Res., 21: 179–204.
Lunenfeld, B. 2004. Historical perspectives in gonadotrophin therapy. Hum.
Reprod. Update, 10:6: 453–467.
Lunenfeld, B. 2002. Development of gonadotrophins for clinical use. Reprod.
Biomed. Online, 4 Suppl 1: 11–17.
Li, C., Simpson, M., and Evans, H. 1949. Isolation of pituitary follicle-stimulating hormone (FSH). Science (New York, N.Y.), 109:2835: 445–446.
Li, C. H., Squire, P. G., and Gr¨ oschel, U. 1960. Purification and proper- ties of human pituitary follicle-stimulating and interstitial cell-stimulatinghormones. Arch. Biochem. Biophys., 86:1: 110–116.
Gemzell, C., Diczfalusy, E., and Tillinger, G. 1958. Clinical effect of humanpituitary follicle-stimulating hormone (FSH). J. Clin. Endocrinol. Metab.,18:12: 1333–1348.
Schmidt-Elmendorff, H., Loraine, J., and Bell, E. 1962. Studies on the bi-ological activity of various gonadotrophin preparations. J. Endocrinol., 24:349–357.
Gemzell, C. 1962. Induction of ovulation with human pituitary go-nadotrophins. Fertil. Steril., 13: 153–168.
Rosemberg, E., Coleman, J., Gibree, N., and Macgillivray, W. 1962. Clinicaleffect of gonadotropins of human origin. Fertil. Steril., 13: 220–236.
Grant, A., Meares, S., Ferguson, K., and Wallace, A. 1964. The treatmentof anovulation by human pituitary gonadotropin. Aust. NZ J. Obstet. Gy-naecol., 4:1: 39–42.
Dumble, L. and Klein, R. 1992. Creutzfeldt-Jakob legacy for Australianwomen treated with human pituitary gonadotropins. Lancet , 340:8823: 847–848.
The Practice Committee of the American Society for Reproductive Medicine2008. Gonadotropin preparations: past, present, and future perspectives.
Fertil. Steril., 90:Suppl 3: S13–S20.
Loraine, J. A. 1952. Recent developments in the clinical application of hor-mone assay. J. Obstet. Gynaecol., 59:4: 535–557.
Loraine, J. A. 1956. Bioassay of pituitary and placental gonadotropins in re-lation to clinical problems in man. S.Harris, R. (ed.), Vitamins & Hormones,vol. 14, pp. 305–357, Academic Press.
Heller, C. and Chandler, R. 1942. Gonadotropic hormone: Modification ofthe alcohol-precipitation assay method. J. Clin. Endocrinol. Metab., 2:4:252–253.
Dekanski, J. 1949. The kaolin-absorption method for the quantitative assayof urinary gonadotrophins. Br. J. Exp. Pathol., 30:4: 272–282.
Gorbman, A. 1945. Ultrafiltration of urine for collection and biological assayof excreted hypophyseal hormones. Endocrinology, 37:3: 177–190.
Albert, A. and Rogers, D. 1951. An improved ultrafilter for the concentrationof urinary gonadotropin. J. Lab. Clin. Med., 37:3: 485–487.
Butt, W. and Crooke, A. 1953. Chromatography of Urinary Gonadotrophins.
Ciba Foundation Symposium - Bioassay of Anterior Pituitary and Adrenocor-tical Hormones (Colloquia on Endocrinology), pp. 44–51, no. 6, John Wiley& Sons, Ltd.
Loraine, J. and Brown, J. 1954. Some observations on the estimation ofgonadotrophins in human urine. Acta Endocrinol. (Copenh), 17:1-4: 250–263.
Jungck, E., Maddock, W., and Heller, C. 1947. Gonadotropic hormone; com-parison of ultrafiltration and alcohol-precipitation methods of recovery fromurine. J. Clin. Endocrinol. Metab., 7:1: 1–10.
Dekanski, J. 1950. The effects of human pituitary gonadotrophins on hy-pophysectomized rats. Br. J. Exp. Pathol., 31:6: 813–818.
Borth, R., Lunenfeld, V., and de, W. H. 1954. [Gonadotropic activity ofurinary extracts of climacteric females]. Experientia, 10:6: 266–268.
Borth, R., Lunenfeld, B., Riotton, G., and de, W. H. 1957. [Gonadotropicactivity of female urine extracts in menopause]. Experientia, 13:3: 115–117.
Bahn, R., Lorenz, N., Bennett, W., and Albert, A. 1953. Gonadotropins ofthe pituitary of postmenopausal women. Endocrinology, 53:4: 455–457.
Loraine, J. and Brown, J. 1956. Further observations on the estimation ofurinary gonadotropins in non-pregnant human subjects. J. Clin. Endocrinol.
Metab, 16:9: 1180–1195.
Albert, A., Borth, R., Diczfalusy, E., Loraine, J., Lunenfeld, B., McArthur,J., and Rosemberg, E. 1958. Collaborative assays of two urinary preparationsof human pituitary gonadotropin. J. Clin. Endocrinol. Metab., 18:10: 1117–1123.
Benz, F., Borth, R., Brown, P., Crooke, A., Dekanski, J., Diczfalusy, E.,Loraine, J., Lunenfeld, B., and Schuler, W. 1959. Collaborative assay oftwo gonadotrophin preparations from human postmenopausal urine. J. En-docrinol., 19: 158–163.
Department of Biological Standards 1960. International Reference Prepara-tion for human menopausal gonadotrophin. Bull. World Health Organ., 22:563–564.
Rosemberg, E. and Engel, I. 1961. Comparative activities of urinary go-nadotropin preparations from postmenopausal women. J. Clin. Endocrinol.
Metab, 21: 603–610.
Albert, A., Derner, I., Stellmacher, V., Leiferman, J., and Barnum, J. 1961.
Purification of pituitary gonadotropin from urine of postmenopausal women.
J. Clin. Endocrinol. Metab., 21: 1260–1266.
Donini, P., Puzzuoli, D., and Montezemolo, R. 1964. Purification of go-nadotrophin from human menopausal urine. Acta Endocrinol.(Copenh), 45:321–328.
Rosemberg, E., Lewis, W., and Solod, E. 1964. Comparative activities ofhuman urinary postmenopausal gonadotropins. J. Clin. Endocrinol. Metab,24: 673–675.
Rosemberg, E., Coleman, J., Demany, M., and Garcia, C. 1963. Clinicaleffect of human urinary postmenopausal gonado-tropin. J. Clin. Endocrinol.
Metab, 23: 181–190.
Diczfalusy, E., Johannisson, E., Tillinger, K., and Bettendorf, G. 1964. Com-parison of the clinical and steroid metabolic effect of human pituitary- andurinary gonadotrophins in amenorrhoeic women. Acta Endocrinol.(Copenh),45:Suppl 90: 35–56.
Furuhjelm, M., Lunell, N., and Odeblad, E. 1966. Ovarian stimulation byhuman menopausal gonadotrophin in patients with amenorrhoea and anovu-lation. Acta Obstet. Gynecol. Scand., 45:1: 63–87.
Rabau, E., David, A., Serr, D., Mashiach, S., and Lunenfeld, B. 1967. Humanmenopausal gonadotropins for anovulation and sterility. Results of 7 yearsof treatment. Am. J. Obstet. Gynecol., 98:1: 92–98.
Rosemberg, E. and Nwe, T. 1968. Induction of ovulation with human post-menopausal gonadotropin. Experience with 23 patients. Fertil. Steril., 19:2:197–205.
Giudice, E., Crisci, C., Eshkol, A., and Papoian, R. 1994. Compositionof commercial gonadotrophin preparations extracted from human post-menopausal urine: characterization of non-gonadotrophin proteins. Hum.
Reprod., 9:12: 2291–2299.
Reichert, J., L.E. and Parlow, A. 1964. Partial purification and separation ofurinary gonadotrophins of nonpregnant humans. J. Clin. Endocrinol. Metab,24: 1040–1048.
Donini, P., Puzzuoli, D., D'Alessio, I., Lunenfeld, B., Eshkol, A., and Parlow,A. 1966. Purification and separation of follicle stimulating hormone (FSH)and luteinizing hormone (LH) from human postmenopausal gonadotrophin(HMG). I. Separation of FSH and LH by electrophoresis, chromatographyand gel filtration procedures. Acta Endocrinol.(Copenh), 52:2: 169–185.
Donini, P., Puzzuoli, D., D'Alessio, I., Lunenfeld, B., Eshkol, A., and Parlow,A. 1966. Purification and separation of follicle stimulating hormone (FSH)and luteinizing hormone (LH) from human postmenopausal gonadotrophin(HMG). II. Preparation of biological apparently pure FSH by selective bind-ing of the LH with an anti-HGG serum and subsequent chromatography.
Acta Endocrinol.(Copenh), 52:2: 186–198.
le Cotonnec, J.-Y., Porchet, H., Beltrami, V., and Howles, C. 1993. Compar-ative pharmacokinetics of two urinary human follicle stimulating hormonepreparations in healthy female and male volunteers. Hum. Reprod., 8:10:1604–1611.
Keene, J., Matzuk, M., Otani, T., Fauser, B., Galway, A., Hsueh, A., andBoime, I. 1989. Expression of biologically active human follitropin in Chinesehamster ovary cells. J. Biol. Chem., 264:9: 4769–4775.
Fauser, B. 1997. Recombinant human FSH: A new era in the developmentof drugs for stimulation of ovarian function. Middle East Fertil. Soc. J., 2:2:95–102.
Loumaye, E., Campbell, R., and SalatBaroux, J. 1995. Human follicle-stimulating hormone produced by recombinant DNA technology: A reviewfor clinicians. Hum. Reprod. Update, 1:2: 188–199.
Loumaye, E., Martineau, I., Piazzi, A., O'Dea, L., Ince, S., Howles, C.,Decosterd, G., Van Loon, K., and Galazka, A. 1996. Clinical assessmentof human gonadotrophins produced by recombinant DNA technology. Hum.
Reprod., 11 Suppl 1: 95–107.
Fauser, B. 1998. Developments in human recombinant follicle stimulatinghormone technology: are we going in the right direction? Hum. Reprod., 13:36–46.
Eshkol, A. 1996. Recombinant gonadotrophins: an introduction. Hum. Re-prod., 11 Suppl 1: 89–94.
Kol, S. 2005. To add or not to add LH: consideration of LH concentrationchanges in individual patients. Reprod. Biomed. Online, 11:6: 664–666.
[100] Howles, C. 2000. Role of LH and FSH in ovarian function. Mol. Cell. En- docrinol., 161:1-2: 25–30.
[101] Mannaerts, B., Deleeuw, R., Geelen, J., Vanravestein, A., Vanwezenbeek, P., Schuurs, A., and Kloosterboer, H. 1991. Comparative in vitro and in vivo studies on the biological characteristics of recombinant human follicle-stimulating-hormone. Endocrinology, 129:5: 2623–2630.
[102] Flack, M., Bennet, A., and Nisula, B. 1991. Production of recombinant FSH in a human cell line biologic and immunologic properties. Maggi, M. andGeenen, V. (eds.), Serona Symposia Publications from Raven Press; v. 76 ,pp. 35–40, Raven Pr.
[103] Flack, M., Bennet, A., Froehlich, J., Anasti, J., and Nisula, B. 1994.
Increased biological-activity due to basic isoforms in recombinant humanfollicle-stimulating-hormone produced in a human cell-line. J. Clin. En-docrinol. Metab., 79:3: 756–760.
[104] Prevost, R. 1998. Recombinant follicle-stimulating hormone: New biotech- nology for infertility. Pharmacotherapy, 18:5: 1001–1010.
[105] Devroey, P., Mannaerts, B., Smitz, J., Coelingh, B. H., and Van, S. A. 1994.
Clinical outcome of a pilot efficacy study on recombinant human follicle-stimulating hormone (Org 32489) combined with various gonadotrophin-releasing hormone agonist regimens. Hum. Reprod., 9:6: 1064–1069.
[106] Mannaerts, B., Shoham, Z., Schoot, D., Bouchard, P., Harlin, J., Fauser, B., Jacobs, H., Rombout, F., and Coelingh, B. H. 1993. Single-dose pharma-cokinetics and pharmacodynamics of recombinant human follicle-stimulatinghormone (Org 32489*) in gonadotropin-deficient volunteers. Fertil. Steril., 59:1: 108–114.
[107] Out, M., Driessen, P., Mannaerts, M., and Coelingh, M., Bennink 1998.
Recombinant follicle-stimulating hormone (Follitropin beta, Puregon) yieldshigher pregnancy rates in in vitro fertilization than urinary gonadotropins.
Fertil. Steril., 69:2, Supplement 1: 40S–44S.
[108] Goldfarb, J. M. and Desai, N. 2003. Follitropin-alpha versus human menopausal gonadotropin in an in vitro fertilization program. Fertil. Steril.,80:5: 1094–1099.
[109] Daya, S. 2002. Updated meta-analysis of recombinant follicle-stimulating hormone (FSH) versus urinary FSH for ovarian stimulation in assisted re-production. Fertil. Steril., 77:4: 711–714.
[110] van Wely, M., Westergaard, L. G., Bossuyt, P. M., and van der Veen, F.
2003. Effectiveness of human menopausal gonadotropin versus recombinantfollicle-stimulating hormone for controlled ovarian hyperstimulation in as-sisted reproductive cycles: a meta-analysis. Fertil. Steril., 80:5: 1086–1093.
[111] Duijkers, I. J. M., Klipping, C., Boerrigter, P. J., Machielsen, C. S., de Bie, J. J., and Voortman, G. 2002. Single dose pharmacokinetics and effects onfollicular growth and serum hormones of a long-acting recombinant FSHpreparation (FSH-CTP) in healthy pituitary-suppressed females. Hum. Re-prod., 17:8: 1987–1993.
[112] Loutradis, D., Drakakis, P., Vlismas, A., and Antsaklis, A. 2009. Corifol- litropin alfa, a long-acting follicle-stimulating hormone agonist for the treat-ment of infertility. Curr. Opin. Investig. Drugs, 10:4: 372–380.
[113] van Schanke, A., van de Wetering-Krebbers S.F.M., Bos, E., and Sloot, W.
2010. Absorption, distribution, metabolism and excretion of corifollitropinalfa, a recombinant hormone with a sustained follicle-stimulating activity.
Pharmacology., 85:2: 77–87.
[114] Verbost, P., Sloot, W., Rose, U., de, L. R., Hanssen, R., and Verheijden, G. 2011. Pharmacologic profiling of corifollitropin alfa, the first developedsustained follicle stimulant. Eur. J. Pharmacol., 651:1-3: 227–233.
[115] Fauser, B., Mannaerts, B., Devroey, P., Leader, A., Boime, I., and Baird, D. 2009. Advances in recombinant DNA technology: corifollitropin alfa, ahybrid molecule with sustained follicle-stimulating activity and reduced in-jection frequency. Hum. Reprod. Update, 15:3: 309–321.
[116] Klein, J., Lobel, L., Pollak, S., Lustbader, B., Ogden, R., Sauer, M., and Lustbader, J. 2003. Development and characterization of a long-acting re-combinant hFSH agonist. Hum. Reprod., 18:1: 50–56.
[117] Howles, C. 1996. Genetic engineering of human FSH (Gonal-F). Hum. Re- prod. Update, 2:2: 172–191.
[118] Olijve, W., de Boer, W., Mulders, J., and van Wezenbeek, P. 1996. Molecular biology and biochemistry of human recombinant follicle stimulating hormone(Puregon). Mol. Hum. Reprod., 2:5: 371–382.
[119] Wolzt, M., Gouya, G., Sator, M., Hemetsberger, T., Irps, C., Rettenbacher, M., and Vcelar, B. 2015. Comparison of pharmacokinetic and safety profilesbetween Bemfola and Gonal-f after subcutaneous application. Eur. J. DrugMetab. Pharmacokinet., epub ahead of print.
[120] Olsson, H., Sandstr¨ om, R., and Grundemar, L. 2014. Different pharmacoki- netic and pharmacodynamic properties of recombinant follicle-stimulatinghormone (rFSH) derived from a human cell line compared with rFSH froma non-human cell line. J. Clin. Pharmacol., 54:11: 1299 – 1307.
[121] Borth, R. and Kim, M. 1970. Human gonadotrophins: a review of recent developments. Can. Med. Assoc. J., 102:11: 1173–1178.
[122] Hohmann, F., Laven, J., de Jong, F., and Eijkemans, M. 2001. Low-dose exogenous FSH initiated during the early, mid or late follicular phase caninduce multiple dominant follicle development. Hum. Reprod., 16:5: 846–854.
[123] The Latin-American Puregon IVF Study Group 2001. A double-blind clin- ical trial comparing a fixed daily dose of 150 and 250 IU of recombinantfollicle-stimulating hormone in women undergoing in vitro fertilization. Fer-til. Steril., 76:5: 950–956.
[124] Yong, P. Y., Brett, S., Baird, D. T., and Thong, K. 2003. A prospective ran- domized clinical trial comparing 150 IU and 225 IU of recombinant follicle-stimulating hormone (Gonal-F*) in a fixed-dose regimen for controlled ovar-ian stimulation in in vitro fertilization treatment. Fertil. Steril., 79:2: 308–315.
[125] Nardo, L. G., Fleming, R., Howles, C. M., Bosch, E., Hamamah, S., Ubaldi, F. M., Hugues, J. N., Balen, A. H., and Nelson, S. M. 2011. Conventionalovarian stimulation no longer exists: welcome to the age of individualizedovarian stimulation. Reprod. Biomed. Online, 23:2: 141–148.
[126] Bosch, E. and Ezcurra, D. 2011. Individualised controlled ovarian stimula- tion (iCOS): maximising success rates for assisted reproductive technologypatients. Reprod. Biol. Endocrinol., 9: 82–.
[127] Johnson, N., Bagrie, E., Coomarasamy, A., Bhattacharya, S., Shelling, A., Jessop, S., Farquhar, C., and Khan, K. 2006. Ovarian reserve tests for pre-dicting fertility outcomes for assisted reproductive technology: the Interna-tional Systematic Collaboration of Ovarian Reserve Evaluation protocol fora systematic review of ovarian reserve test accuracy. BJOG , 113:12: 1472–1480.
[128] Broekmans, F., Kwee, J., Hendriks, D., Mol, B., and Lambalk, C. 2006. A systematic review of tests predicting ovarian reserve and IVF outcome. Hum.
Reprod. Update., 12:6: 685–718.
[129] Howles, C., Saunders, H., Alam, V., and Engrand, P. 2006. Predictive factors and a corresponding treatment algorithm for controlled ovarian stimulationin patients treated with recombinant human follicle stimulating hormone(follitropin alfa) during assisted reproduction technology (ART) procedures.
An analysis of 1378 patients. Curr. Med. Res. Opin., 22:5: 907–918.
[130] Olivennes, F., Howies, C., Borini, A., Germond, M., Trew, G., Wikland, M., Zegers-Hochschild, F., Saunders, H., and Alam, V. 2009. IndividualizingFSH dose for assisted reproduction using a novel algorithm: the CONSORTstudy. Reprod. Biomed. Online, 18:2: 195–204.
[131] Bancsi, L. F., Broekmans, F. J., Eijkemans, M. J., de Jong, F. H., Habbema, J., and te Velde, E. R. 2002. Predictors of poor ovarian response in in vitrofertilization: a prospective study comparing basal markers of ovarian reserve.
Fertil. Steril., 77:2: 328–336.
[132] Muttukrishna, S., McGarrigle, H., Wakim, R., Khadum, I., Ranieri, D., and Serhal, P. 2005. Antral follicle count, anti-m¨ ullerian hormone and inhibin B: predictors of ovarian response in assisted reproductive technology? BJOG ,112:10: 1384–1390.
[133] Eldar-Geva, T., Ben-Chetrit, A., Spitz, I., Rabinowitz, R., Markowitz, E., Mimoni, T., Gal, M., Zylber-Haran, E., and Margalioth, E. 2005. Dynamicassays of inhibin B, anti-M¨ ullerian hormone and estradiol following FSH stimulation and ovarian ultrasonography as predictors of IVF outcome. Hum.
Reprod., 20:11: 3178–3183.
[134] Themmen, A. 2005. Anti-M¨ ullerian hormone: its role in follicular growth initiation and survival and as an ovarian reserve marker. J. Natl. CancerInst. Monogr., 34: 18–21.
[135] van Rooij, I. A., Broekmans, F. J., Scheffer, G. J., Looman, C. W., Habbema, J., de Jong, F. H., Fauser, B. J., Themmen, A. P., and te Velde, E. R. 2005.
Serum antim¨ ullerian hormone levels best reflect the reproductive decline with age in normal women with proven fertility: A longitudinal study. Fertil.
Steril., 83:4: 979–987.
[136] Hazout, A., Bouchard, P., Seifer, D. B., Aussage, P., Junca, A. M., and Cohen-Bacrie, P. 2004. Serum antim¨ ullerian hormone/m¨ substance appears to be a more discriminatory marker of assisted repro-ductive technology outcome than follicle-stimulating hormone, inhibin B, orestradiol. Fertil. Steril., 82:5: 1323–1329.
[137] Majumder, K., Gelbaya, T. A., Laing, I., and Nardo, L. G. 2010. The use of ullerian hormone and antral follicle count to predict the potential of oocytes and embryos. Eur. J. Obstet. Gynecol. Reprod. Biol., 150:2: 166–170.
[138] Jayaprakasan, K., Campbell, B., Hopkisson, J., Johnson, I., and Raine- Fenning, N. 2010. A prospective, comparative analysis of anti-M¨ mone, inhibin-B, and three-dimensional ultrasound determinants of ovarian reserve in the prediction of poor response to controlled ovarian stimulation.
Fertil. Steril., 93:3: 855–864.
[139] Choi, M., Yoo, J., Kim, H., Cha, S., Park, C., Yang, K., Song, I., Koong, M., and Kang, I. 2011. Serum anti-M¨ ullerian hormone levels as a predictor of the ovarian response and IVF outcomes. Clin. Exp. Reprod. Med., 38:3:153–158.
[140] Nardo, L., Gelbaya, T., Wilkinson, H., Roberts, S., Yates, A., Pemberton, P., and Laing, I. 2009. Circulating basal anti-M¨ ullerian hormone levels as predictor of ovarian response in women undergoing ovarian stimulation forin vitro fertilization. Fertil. Steril., 92:5: 1586–1593.
[141] Andersen, A., Witjes, H., Gordon, K., and Mannaerts, B. 2011. Predictive factors of ovarian response and clinical outcome after IVF/ICSI followinga rFSH/GnRH antagonist protocol with or without oral contraceptive pre-treatment. Hum. Reprod., 26:12: 3413–3423.
[142] Sowers, M., McConnell, D., Gast, K., Zheng, H., Nan, B., McCarthy, J. D., and Randolph, J. F. 2010. Anti-M¨ ullerian hormone and inhibin B variability during normal menstrual cycles. Fertil. Steril., 94:4: 1482–1486.
[143] Porchet, H., le Cotonnec, J.-Y., and Loumaye, E. 1994. Clinical Phar- macology of Recombinant Human Follicle-Stimulating Hormone. III.
Pharmacokinetic-Pharmacodynamic Modeling After Repeated Subcuta-neous Administration. Fertil. Steril., 61:4: 687 – 95.
[144] Dzik, A., Lambert-Messerlian, G., Izzo, V. M., Soares, J. B., Pinotti, J. A., and Seifer, D. B. 2000. Inhibin B response to EFORT is associated withthe outcome of oocyte retrieval in the subsequent in vitro fertilization cycle.
Fertil. Steril., 74:6: 1114–1117.
[145] Yong, P., Baird, D., Thong, K., McNeilly, A., and Anderson, R. 2003.
Prospective analysis of the relationships between the ovarian follicle cohortand basal FSH concentration, the inhibin response to exogenous FSH andovarian follicle number at different stages of the normal menstrual cycle andafter pituitary down-regulation. Hum. Reprod., 18:1: 35–44.
[146] Decanter, C., Pigny, P., Lefebvre, C., Thomas, P., Leroy, M., and Dewailly, D. 2009. Serum inhibin B during controlled ovarian hyperstimulation: anadditional criterion for deciding whether to proceed with egg retrieval. Fertil.
Steril., 91:6: 2419–2425.
[147] Mannaerts, B., Rombout, F., Out, H., and Coelingh, B. H. 1996. Clini- cal profiling of recombinant follicle stimulating hormone (rFSH; Puregon):relationship between serum FSH and efficacy. Hum. Reprod. Update., 2:2:153–161.
[148] Karlsson, M., Wade, J., Loumaye, E., and Munafo, A. 1998. The popula- tion pharmacokinetics of recombinant- and urinary-human follicle stimulat-ing hormone in women. Br. J. Clin. Pharmacol., 45:1: 13–20.
[149] Corifollitropin Alfa Dose-finding Study Group 2008. A randomized dose- response trial of a single injection of corifollitropin alfa to sustain multifol-licular growth during controlled ovarian stimulation. Hum. Reprod., 23:11:2484–2492.
[150] The corifollitropin alfa Ensure study group 2010. Corifollitropin alfa for ovar- ian stimulation in IVF: A randomized trial in lower-body-weight women.
Reprod. Biomed. Online, 21:1: 66–76.
[151] de Greef, R., Zandvliet, A., de Haan, A., Ijzerman-Boon, P., Marintcheva- Petrova, M., and Mannaerts, B. 2010. Dose selection of corifollitropin alfa bymodeling and simulation in controlled ovarian stimulation. Clin. Pharmacol.
Ther., 88:1: 79–87.
[152] Ledger, W. L., Fauser, B. C., Devroey, P., Zandvliet, A. S., and Mannaerts, B. M. 2011. Corifollitropin alfa doses based on body weight: clinical overviewof drug exposure and ovarian response. Reprod. Biomed. Online, 23:2: 150–159.
[153] Beal, S., Sheiner, L., Boeckmann, A., and Bauer, R. 2011. NONMEM User's Guides. (1989-2011). Icon Development Solutions, Ellicott City, MD, USA.
[154] Aarons, L. 1992. Population pharmacokinetics. Int. J. Clin. Pharmacol.
Ther. Toxicol., 30:11: 520–522.
[155] Aarons, L. 1996. Population approaches/sparse data analysis for human vari- ability in kinetics and dynamics. Environ. Toxicol. Pharmacol., 2:2-3: 197–199.
[156] Sheiner, L. and Ludden, T. 1992. Population pharmacokinetics/dynamics.
Annu. Rev. Pharmacol. Toxicol., 32: 185–209.
[157] Sheiner, L. B. and Beal, S. L. 1980. Evaluation of methods for estimating population pharmacokinetics parameters. I. Michaelis-Menten model: rou-tine clinical pharmacokinetic data. J. Pharmacokinet. Biopharm., 8:6: 553–571.
[158] Beal, S. and Sheiner, L. 1980. The NONMEM system. Am. Stat., 34: 118– [159] Sheiner, L. and Beal, S. 1981. Evaluation of methods for estimating popula- tion pharmacokinetic parameters. II. Biexponential model and experimentalpharmacokinetic data. J. Pharmacokinet. Biopharm., 9:5: 635–651.
[160] Sheiner, L. B. and Beal, S. L. 1983. Evaluation of methods for estimat- ing population pharmacokinetic parameters. III. Monoexponential model:routine clinical pharmacokinetic data. J. Pharmacokinet. Biopharm., 11:3:303–319.
[161] Pillai, G., Mentre, F., and Steimer, J. 2005. Non-linear mixed effects model- ing - from methodology and software development to driving implementationin drug development science. J. Pharmacokinet. Pharmacodyn., 32:2: 161–183.
[162] Aarons, L., Balant, L., Mentre, F., Morselli, P., Rowland, M., Steimer, J., and Vozeh, S. 1994. Population approaches in drug development. Report onan expert meeting to discuss population pharmacokinetic/pharmacodynamicsoftware. Eur. J. Clin. Pharmacol., 46:5: 389–391.
[163] Aarons, L. 1999. Software for population pharmacokinetics and pharmaco- dynamics. Clin. Pharmacokinet., 36:4: 255–264.
[164] Bonate, P. 2005. Recommended reading in population pharmacokinetic phar- macodynamics. AAPS J., 7:2: E363–E373.
[165] Sheiner, L. and Grasela, T. 1991. An introduction to mixed effect modeling: Concepts, definitions, and justification. J. Pharmacokinet. Biopharm., 19:3Suppl.: 11S–24S.
[166] Karlsson, M. and Sheiner, L. 1993. The importance of modeling interocca- sion variability in population pharmacokinetic analyses. J. Pharmacokinet.
Biopharm., 21:6: 735–750.
[167] Meibohm, B. and Derendorf, H. 1997. Basic concepts of pharmacoki- netic/pharmacodynamic (PK/PD) modelling. Int. J. Clin. Pharmacol.
Ther., 35:10: 401–413.
[168] Holford, N. H. and Sheiner, L. B. 1981. Pharmacokinetic and pharmacody- namic modeling in vivo. CRC Crit. Rev. Bioeng., 5:4: 273–322.
[169] Jusko, W. and Ko, H. 1994. Physiologic indirect response models characterize diverse types of pharmacodynamic effects. Clin. Pharmacol. Ther., 56:4:406–419.
[170] Sheiner, L. and Wakefield, J. 1999. Population modelling in drug develop- ment. Stat. Methods Med. Res., 8:3: 183–193.
[171] Beal, S. 2001. Ways to Fit a PK Model with Some Data Below the Quan- tification Limit. J. Pharmacokinet. Pharmacodyn., 28:5: 481–504.
[172] Ahn, J., Karlsson, M., Dunne, A., and Ludden, T. 2008. Likelihood based approaches to handling data below the quantification limit using NONMEMVI. J. Pharmacokinet. Pharmacodyn., 35:4: 401–421.
[173] Bergstrand, M. and Karlsson, M. 2009. Handling data below the limit of quantification in mixed effect models. AAPS J., 11:2: 371–380.
[174] Wade, J., Beal, S., and Sambol, N. 1994. Interaction between structural, statistical, and covariate models in population pharmacokinetic analysis. J.
Pharmacokinet. Biopharm., 22:2: 165–177.
[175] Karlsson, M. and Savic, R. 2007. Diagnosing model diagnostics. Clin. Phar- macol. Ther., 82:1: 17–20.
[176] Savic, R. and Karlsson, M. 2009. Importance of shrinkage in empirical bayes estimates for diagnostics: problems and solutions. AAPS J., 11:3: 558–569.
[177] Hooker, A., Staatz, C., and Karlsson, M. 2007. Conditional weighted resid- uals (CWRES): a model diagnostic for the FOCE method. Pharm. Res., 24:12: 2187–2197.
[178] le Cotonnec, J.-Y., Porchet, H., Beltrami, V., Khan, A., Toon, S., and Row- land, M. 1995. Comprehensive pharmacokinetics of urinary human folliclestimulating hormone in healthy female volunteers. Pharm. Res., 12:6: 844–850.
[179] le Cotonnec, J.-Y., Porchet, H., Beltrami, V., Khan, A., Toon, S., and Rowland, M. 1994. Clinical pharmacology of recombinant human follicle-stimulating hormone (FSH). I. Comparative pharmacokinetics with urinaryhuman FSH. Fertil. Steril., 61:4: 669–678.
[180] le Cotonnec, J.-Y., Porchet, H., Beltrami, V., Khan, A., Toon, S., and Rowland, M. 1994. Clinical pharmacology of recombinant human follicle-stimulating hormone. II. Single doses and steady state pharmacokinetics.
Fertil. Steril., 61:4: 679–686.
[181] Devroey, P., Fauser, B. C., Platteau, P., Beckers, N. G., Dhont, M., and Mannaerts, B. M. 2004. Induction of multiple follicular development by asingle dose of long-acting recombinant follicle-Stimulating hormone (FSH-CTP, corifollitropin alfa) for controlled ovarian stimulation before in vitrofertilization. J. Clin. Endocrinol. Metab., 89:5: 2062–2070.
[182] Balen, A., Mulders, A., Fauser, B., Schoot, B., Renier, M., Devroey, P., Struijs, M., and Mannaerts, B. 2004. Pharmacodynamics of a single lowdose of long-acting recombinant follicle-stimulating hormone (FSH-carboxyterminal peptide, corifollitropin alfa) in women with World Health Organi-zation group II anovulatory infertility. J. Clin. Endocrinol. Metab., 89:12:6297–6304.
[183] Fauser, B. C., Alper, M. M., Ledger, W., Schoolcraft, W. B., Zandvliet, A., and Mannaerts, B. M. 2010. Pharmacokinetics and follicular dynamics ofcorifollitropin alfa versus recombinant FSH during ovarian stimulation forIVF. Reprod. Biomed. Online, 21:5: 593–601.
[184] Karlsson, M., Wade, J., Loumaye, E., and Munafo, A. 1997. A population model for the follicular growth in women treated with follicle stimulatinghormone. Clin. Pharmacol. Ther., 62:6: 665–674.
[185] le Cotonnec, J.-Y., Loumaye, E., Porchet, H. C., Beltrami, V., and Mu- nafo, A. 1998. Pharmacokinetic and pharmacodynamic interactions betweenrecombinant human luteinizing hormone and recombinant human follicle-stimulating hormone. Fertil. Steril., 69:2: 201–209.
[186] Olsson, H., Sandstr¨ om, R., and Bagger, Y. 2015. Dose-Exposure Propor- tionality of a Novel Recombinant Follicle-Stimulating Hormone (rFSH), FE999049, Derived from a Human Cell Line, with Comparison Between Cau-casian and Japanese Women After Subcutaneous Administration. Clin. DrugInvest., 35:4: 247–253.
[187] Arce, J.-C., Nyboe Andersen, A., Fern´ anchez, M., Visnova, H., Bosch, E., Garc´ıa-Velasco, J. A., Barri, P., de Sutter, P., Klein, B. M.,and Fauser, B. C. 2014. Ovarian response to recombinant human follicle-stimulating hormone: a randomized, antim¨ ullerian hormone-stratified, dose- response trial in women undergoing in vitro fertilization/intracytoplasmicsperm injection. Fertil. Steril., 102:6: 1633–1640.e5.
[188] R Development Core Team 2010. R: A Language and Environment for Statis- tical Computing. R Foundation for Statistical Computing, Vienna, Austria,pp. –.
[189] Lindbom, L., Ribbing, J., and Jonsson, E. 2004. Perl-speaks-NONMEM (PsN) - a Perl module for NONMEM related programming. Comput. Meth-ods Programs Biomed., 75:2: 85–94.
[190] Lindbom, L., Pihlgren, P., and Jonsson, E. 2005. PsN-Toolkit–a collection of computer intensive statistical methods for non-linear mixed effect modelingusing NONMEM. Comput. Methods Programs Biomed., 79:3: 241–257.
[191] Jonsson, E. and Karlsson, M. 1999. Xpose-an s-plus based population phar- macokinetic/pharmacodynamic model building aid for nonmem. Comput.
Methods Programs Biomed., 58:1: 51–64.
Population Pharmacokinetic Modelling of FE 999049, a
Recombinant Human Follicle Stimulating Hormone, in
Healthy Women after Single Ascending Doses
Trine Høyer Rose1,2, Daniel Röshammar1, Lars Erichsen1, Lars Grundemar1, Johnny T. Ottesen2
1Experimental Medicine, Ferring Pharmaceuticals A/S, Denmark 2Department of Science, Systems, and Models, Roskilde University, Denmark Corresponding author: Trine Høyer Rose, Ferring Pharmaceuticals A/S, Kay Fiskers Plads 11,
DK-2300 Copenhagen S, Denmark. E-mail: [email protected]
Abstract
Background and Objective
The purpose of this analysis was to develop a population pharmacokinetic model for a novel recombinant
human follicle stimulating hormone (FE 999049) expressed from a human cell line of fetal retinal origin
(PER.C6®) developed for controlled ovarian stimulation prior to assisted reproductive technologies.
Methods
Serum FSH levels were measured following a single subcutaneous FE 999049 injection of 37.5, 75, 150, 225,
or 450 IU in 27 pituitary suppressed healthy female subjects participating in this first-in-human single
ascending dose trial. Data was analysed by nonlinear mixed effects population pharmacokinetic modelling in
NONMEM 7.2.0.
Results
A one-compartment model with first order absorption and elimination rates was found to best describe the
data. A transit model was introduced to describe a delay in the absorption process. The apparent clearance and
volume of distribution estimates were found to increase with body weight. Body weight was included as an
allometrically scaled covariate with a power exponent of 0.75 for clearance and 1 for the volume of
distribution.
Conclusions
The single dose pharmacokinetics of FE 999049 were adequately described by a population pharmacokinetic
model. The average drug concentration at steady-state is expected to be reduced with increasing body weight.
Key Points
The population pharmacokinetics of the novel recombinant human follicle stimulating hormone FE 999049 have been characterised in healthy females after single ascending dosing Follicle stimulating hormone measurements below the quantification limit were accounted for Body weight influences exposure to FE 999049 and may be an important factor for dosage 1. Introduction
The female reproductive function is controlled by periodically regulated production, secretion, and interaction of hormones in the hypothalamic-pituitary-gonadal axis. Of particular importance are the two gonadotropins follicle stimulating hormone (FSH) and luteinizing hormone (LH), both of which are released from the anterior pituitary gland upon stimulation by gonadotropin-releasing hormone (GnRH) from the hypothalamus. During the normal menstrual cycle, the combined actions of FSH and LH induce development of a single dominant follicle, ovarian hormone production, oocyte maturation, and ovulation. In the management of infertility, exogenous FSH administration is used to induce monofollicular development and ovulation in anovulatory women [1], or multiple follicular development to allow selection of embryos for transfer in women undergoing in vitro fertilisation (IVF)/intracytoplasmic sperm injection (ICSI) treatment [2]. Urinary menopausal gonadotropin preparations that contain both FSH and LH activity are being extensively used for controlled ovarian stimulation (COS) in IVF/ICSI treatment. In 1989 advances in DNA technologies enabled development of a recombinant FSH (rFSH) generated in Chinese Hamster Ovarian (CHO) cell lines which expressed the genes encoding human FSH [3]. Since then several rFSH products have been marketed including two original rFSH compounds originating from CHO cell lines [4;5] and a long acting rFSH [6]. A novel recombinant human FSH (rhFSH, FE 999049) expressed from a human fetal retinal cell line (PER.C6®, Crucell, Leiden, The Netherlands) is under development by Ferring Pharmaceuticals A/S for patients undergoing COS for IVF/ICSI. Dose proportionality has been shown for maximum concentration (Cmax) and area under the concentration-time curve (AUC) by non-compartmental analysis (NCA) of data from both a single ascending dose phase 1 trial with Caucasian women and one with Japanese women [7]. The purpose of this analysis was to describe the population pharmacokinetics of FE 999049 based on the Caucasian trial in Olsson et al.[7]. When using a population modelling approach with nonlinear mixed effects models, as opposed to NCA, it is possible to investigate variation in the population and to identify potential covariates explaining some of the variability in drug exposure. 2. Materials and Methods
2.1 Clinical trial design
The first-in-human trial with FE 999049 was a randomised, double-blind, placebo controlled, sequential single dose escalation trial investigating the safety, tolerability, and pharmacokinetics. The trial was carried out in accordance with the Declaration of Helsinki and Good Clinical Practice. It was approved by regulatory authorities and local ethics committees. All subjects gave written informed consent to participate. The trial is the ‘Caucasian study' described in Olsson et al. [7]. In summary, the trial population consisted of 40 healthy female volunteers aged 21-35 years with a normal menstrual cycle and a body mass index (BMI) ranging 18-29 kg/m2 (Table 1). They received a single subcutaneous abdominal injection of 37.5, 75, 150, 225, or 450 IU FE 999049 or placebo. In each dose group there were 8 women whereof 2 were given placebo. To avoid any interference with endogenous FSH levels during the trial, all volunteers were pituitary down-regulated by means of a combined oral contraceptives (COC). To ensure similar down-regulation in all subjects, they were all switched from their COC to the same high-dose COC (OGESTREL 0.5/50, 50 µg of ethinyl estradiol, 0.5 mg norgestrel, Watson Pharma Inc.) 14 days before drug administration. Blood samples for measurement of serum FSH concentration were collected 60 and 30 minutes prior to administration, immediately before administration, at every 4 hours the first 48 hours and subsequent every day up to 9 days after administration. Determination of serum FSH concentrations was performed at Ferring Pharmaceuticals A/S with a validated immunoassay based on electrochemiluminescence (MSD sectorTM Imager 2400) with a lower limit of quantification (LLOQ) of 0.075 µg/L. FE 999049 was dosed in IU and for this analysis converted to μg units (2.2, 4.4, 8.8, 13.1 and 26.3 μg) using the specific activity in order to estimate the serum concentrations of FE 999049 in µg/L units. Prior to the modelling, mean serum FSH concentrations versus time were plotted for each of the 5 treatment groups including all subjects (Figure 1). A second increase in FSH concentration starting at day 3 after administration was observed for the 4.4 μg dose group. From the individual concentration-time profiles (not shown) two subjects were identified to cause this increase. One subject in the 8.8 μg dose group had a second increase in FSH concentration at day two. Their concentration-time-profiles were considered not to reflect the pharmacokinetic (PK) profile for the exogenously administered rhFSH (FE 999049) and were excluded from the analysis. A total of 594 samples from 27 individuals were included in the analysis. 258 measurements, constituting over 40% of the total data points, were below the quantification limit (BQL). In the 2.2, 4.4, 8.8, 13.1 and 26.3 μg dose group 67, 74, 39, 23, and 23% of the measurements were BQL, respectively. 2.3 Pharmacokinetic Modelling
A population PK model was developed using nonlinear mixed effects modelling. This included finding a structural model together with appropriate error models describing interindividual and residual variability. To increase the predictive capability of the model, it was checked if part of the interindividual variability (IIV) in parameter estimates could be explained by covariates (body weight, age, and dose). Model development was guided by changes in the NONMEM objective function value (OFV), precision of parameter estimates, and graphical model goodness-of-fit assessments including visual predictive checks (VPC). The OFV is approximately proportional to -2log likelihood. The difference in OFV between two nested models is approximately χ2-distributed, with degrees of freedom equal to the difference in the number of parameters. Based on this, the statistical significance for inclusion/exclusion of a model parameter can be judged. For this descriptive analysis a significance level of 0.05 was used for discrimination among nested models and covariate testing. BQL measurements were accounted for in the analysis using the M3 method since for a high proportion of BQL it is the preferred method out of the 7 existing methods [8-10]. With the M3 method BQL data are censored observations and treated as categorical data. These are included in the likelihood function for the model parameter estimation as the likelihood that the observation is truly BQL. The sensitivity of the model parameter estimates to the BQL method used was evaluated by comparing the estimates from the final model to those estimated when the BQL measurements were ignored (M1), set to LLOQ/2 (M5), or set to zero (M7). The final model was used for simulations to illustrate the FE 999049 concentration-time profile and steady state exposure following repated administartion of 10 µg. The average steady state exposure was calculated as 2.4 Software
The models were implemented and parameters estimated in NONMEM 7.2.0 (Icon Development Solutions, USA) [11] using first order conditional estimation with interaction. Data handling and graphical representations were performed in R version 2.11.1 [12]. VPCs were performed using PsN [13;14] and plotted using Xpose [15]. 3. Results
A one-compartment distribution model with first order absorption and a transit model for adding a delay in the absorption was found to adequately describe data. The time-course of serum FE 999049 concentration after dosing was described by the differential equations (1) - (3), one for each of the three compartments representing the dosing site, the transit model, and the central compartment, respectively. At time t, Ai(t) is the FE 999049 amount in the ith compartment. The absorption rate from the dosing site and the transit compartment is ktr and ka, respectively. The elimination rate of FE 999049, k, is clearance (CL) divided by the volume of distribution (V). Since it is the amount tracked in the equations, predicted serum FE 999049 concentrations are calculated as A3(t)/V. Data is obtained after subcutaneous dosing so the bioavailability (F) is not known. The CL and V estimated here are therefore the apparent clearance (CL/F) and apparent volume of distribution (V/F). A combined additive and proportional error model was used to describe the residual error. The parameters CL/F, V/F and ka varied between subjects. A variation in ktr was also observed, but given the current data it was not possible to include separate IIV on both parameters describing the absorption process. It was chosen to keep the variability on ka because this model had the best OFV compared to the model with variability on only ktr. For parameters with IIV the ith subject's individual parameter, Pi, is where Ptv is the typical population parameter and ηi is the individual random effect from an approximately normal distribution with mean zero and variance ω 2 p for the IIV. A positive correlation was identified between CL/F and V/F by a statistically significant improvement in OFV when adding a covariance between the two parameters. Body weight was found to be a statistically significant covariate on CL/F and V/F and was included in the model parameters as where WTi is the ith subject's body weight and ALp is the allometric values: 0.75 for CL/F and 1 for V/F. Adding body weight as a covariate caused a drop of -2.94 in OFV which was considered significantly better since no extra parameters were added. In addition, the coefficient of variation (CV) for the unexplained IIV was reduced from 31.4 to 28.2% CV for CL/F and from 46.4 to 44.3% CV for V/F. Not including body weight in the final model at either CL/F or V/F increased the OFV by 10.8 and 4.4, respectively. Subject age did not further explain any of the IIV. There was neither any influence of dose at the parameters indicating that the pharmacokinetics is linear. The final model parameters are listed in Table 2. The mean observed FSH data and typical model predictions are shown for each dose level in Figure 2a. In the diagnostic VPC plot (Figure 2b) the observed data is compared to model predictions based on 1000 simulated trial datasets using the final model. Since the data consisted of more than 40% BQL measurements that were included in the model by the M3 method, a two panel VPC illustrating both continuous and categorical (BQL) data was used[10]. The top panel displays the observations above LLOQ. The 2.5th, 50th, and 97.5th percentiles of observations and the 95% confidence intervals (CI) for the corresponding model predictions are plotted. In the bottom panel the observed and predicted proportion of BQL observations are visualised. Both the points for observations against individual and population predictions fall around the unity line and there is no trend observed in the model residuals (Figure 3). When re-estimating the final model parameters using a simpler method than M3 for handling BQL measurements slight changes in the parameter estimates were observed. With the M1 or M5 method the PK parameters changed by less than 10% from the parameters obtained with the M3 method. On the contrary, applying the M7 method instead of M3 caused greater parameter changes from the values in the final model and the RSEs increased. The largest change was an increase in the CL/F estimate by 22% to 0.524 L and with a fourfold increase in its RSE. Thus making M7 the least precise method for this model. The impact of body weight on the expected FE 999049 concentration following multiple dose administration was investigated using the final model for simulations. In Figure 4a, illustrating the time-course of FE 999049 concentration in three subjects of different weights, it is shown how the concentration decreases with increasing body weight. Taking IIV into considerations there are large overlaps in the average steady state concentrations across the three weight groups (Figure 4b). However, the FSH exposure appears to be lower in the majority of subjects with body weight of 100 kg compared with FSH levels in subjects with body weight of 50 kg. 4. Discussion
In the present study, population PK modelling was carried out to characterise the pharmacokinetics of FE 999049 after single ascending doses in healthy women. The objective was to get an initial understanding of the time-course of drug exposure and the magnitude of inter-individual variability through a modelling approach. In addition, to optimise ovarian response to treatment with FE 999049, it was examined whether patient-specific variables can aid in the design of individualised dosing schemes. The pharmacokinetics of FE 999049 were described by a one-compartment distribution model with first-order absorption and elimination. These findings are in accordance with previous results that have shown that FSH, either as urinary or recombinant preparation, follows a one-compartment model after s.c. or i.m. administration [16-20]. Some of these studies have found exogenous FSH pharmacokinetics to be best described by a two-compartment model if the data is "rich", i.e. with extensive number of samples, or if doses are given intravenously [17-19]. A two-compartment distribution model was also tested here. As this was a first single dose trial with few subjects, the data generated was not sufficient to give successful estimation of the extra parameters for the peripheral distribution compartment. A transit model with one compartment was introduced to describe a somewhat prolonged absorption of FE 999049 causing an apparent delay for the measurable change in serum FE 999049 concentration. It is conceivable that this extra transit time in the absorption process could be attributed to the lymphatic system, since proteins given subcutaneously are usually absorbed through the lymphatic system [21]. It was tested whether the FE 999049 absorption could be described by alternative models. Adding an extra transit compartment in the final model increased the OFV by 0.002 and is thus worse. Using a lag-time instead of a transit compartment decreased the OFV by 0.085, but these two models are not nested and the OFV cannot be compared with statistically significance. Since the models have the same number of parameters and basically no difference in neither OFV nor the model fit graphs, it was chosen to keep the transit model because it is more mechanistic correct than a lag-time. As part of the model development it was checked whether any covariates could be identified. Body weight was found to be the cause of some of the variation in FE 999049 concentration after treatment. It is consistent with previous analyses that have shown a relation between serum FSH and the PK parameters with body weight [16;17;22;23]. In the current study, with the power exponent fixed to allometric values, body weight was a significant covariate and could explain some of the IIV in CL/F and V/F indicated by a reduction from 31.4% CV and 46.4% CV to 28.2% CV and 44.3% CV, respectively. The marginal effect of adding body weight as a covariate is likely due to the limited number of subjects in this first-in-human trial with a relatively narrow body weight range. Among the potential covariates (age, body weight, and dose) body weight was the only covariate identified. In order to avoid interference with endogenous FSH in the analysis of the FE 999049 pharmacokinetics, all subjects in this trial were pituitary suppressed by means of COC. Since all pre-dose FSH measurements were BQL the measured serum FSH concentrations were exclusively reflecting exogenous FSH from FE 999049. Nonetheless, three subjects showed an additional peak of FSH levels several days after administration of FE 999049, which could likely be due to endogenous FSH levels not being fully suppressed. These individuals were excluded from the analysis as their FSH levels were judged not to reflect the pharmacokinetics of the exogenously administered FE 999049. The initiation of the additional FSH peak occurred 3 days after FE 999049 administration which was the day after the trial subjects were discharged from the residential stay in the clinic. A possibly explanation for the later secondary increase in FSH levels could be poor compliance to taking OGESTREL after discharge from the clinic, however, such protocol deviation was not reported. In future studies where subjects are not pituitary suppressed or down-regulated, endogenous FSH levels have to be considered in the modelling. Especially in phase 2 studies where patients could have varying and measurable endogenous FSH levels influencing the total FSH concentration. When available, the significance of other reproductive hormones such as inhibin B, estradiol, and progesterone should also be studied. Since these hormones influence FSH levels over time they could potentially further explain the observed variation in the FSH concentration profile between subjects. The identified lower drug exposure with higher body weight should be further quantified in future models from other clinical studies with FE 999049. It must also be related to any subsequent effects
that possibly could add variability in clinical efficacy endpoints (e.g. number of oocytes retrieved,
successful implantation rate and pregnancy rate) in order to judge if there is a therapeutic value in
individualising dosing based on patient's body weight. Modelling has previously been used to set two
different rFSH dosage regimens for subjects weighing more or less than 60 kg [16;24].
5. Conclusion
A population PK model was successfully developed for FE 999049 using data from a single ascending
dosing trial in healthy female subjects. There were indications of that FE 999049 exposure decreases
with increasing body weight. When also considering findings from the literature body weight can be
an important factor to consider in efforts to develop individualised dosing regimens for optimised
treatment outcomes. However, in order to confirm the influence of body weight at FE 999049
exposure the model should be updated using data from subsequent clinical trials including multiple
dose trials and trials involving patients, where the body weight range is likely to be wider. In addition
the relationship between drug exposure and clinical efficacy/safety parameters must be established.
Compliance with Ethical Standards
Funding: This work was supported by Innovation Fund Denmark (Industrial PhD case number
11-117436).
Conflicts of Interest: Trine Høyer Rose, Daniel Röshammar, Lars Erichsen, and Lars Grundemar are
all current or former employees at Ferring Pharmaceuticals A/S.
Ethical approval: All procedures performed in studies involving human participants were in
accordance with the ethical standards of the institutional and/or national research committee and with
the 1964 Helsinki declaration and its later amendments or comparable ethical standards.
Informed consent: Informed consent was obtained from all individual participants included in the
study.
1. Homburg R. Clomiphene citrate--end of an era? A mini-review. Hum Reprod 2005; 20:2043-2051. 2. Macklon NS, Stouffer RL, Giudice LC, Fauser BC. The science behind 25 years of ovarian stimulation for in vitro fertilization. Endocr Rev 2006; 27:170-207. 3. Keene JL, Matzuk MM, Otani T, Fauser BC, Galway AB, Hsueh AJ et al. Expression of biologically active human follitropin in Chinese hamster ovary cells. J Biol Chem 1989; 264:4769-4775. 4. Howles CM. Genetic engineering of human FSH (Gonal-F). Hum Reprod Update 1996; 2:172-191. 5. Olijve W, de Boer W, Mulders JW, van Wezenbeek PM. Molecular biology and biochemistry of human recombinant follicle stimulating hormone (Puregon). Mol Hum Reprod 1996; 2:371-382. 6. Fauser BCJM, Mannaerts BMJL, Devroey P, Leader A, Boime I, Baird DT. Advances in recombinant DNA technology: corifollitropin alfa, a hybrid molecule with sustained follicle-stimulating activity and reduced injection frequency. Hum Reprod Update 2009; 15:309-321. 7. Olsson H, Sandström R, Bagger Y. Dose-Exposure Proportionality of a Novel Recombinant Follicle-Stimulating Hormone (rFSH), FE 999049, Derived from a Human Cell Line, with Comparison Between Caucasian and Japanese Women After Subcutaneous Administration. Clin Drug Investig 2015; 8. Ahn JE, Karlsson MO, Dunne A, Ludden TM. Likelihood based approaches to handling data below the quantification limit using NONMEM VI. J Pharmacokinet Pharmacodyn 2008; 35:401-421. 9. Beal S. Ways to Fit a PK Model with Some Data Below the Quantification Limit. J Pharmacokinet Pharmacodyn 2001; 28:481-504. 10. Bergstrand M, Karlsson MO. Handling data below the limit of quantification in mixed effect models. AAPS J 2009; 11:371-380. 11. Beal S, Sheiner LB, Boeckmann A, Bauer RJ. NONMEM User's Guides. (1989-2011). 2011. Icon Development Solutions, Ellicott City, MD, USA. 12. R Development Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria; 2010. 13. Lindbom L, Pihlgren P, Jonsson EN. PsN-Toolkit -- a collection of computer intensive statistical methods for non-linear mixed effect modeling using NONMEM. Comput Methods Programs Biomed 2005; 79:241-257. 14. Lindbom L, Ribbing J, Jonsson EN. Perl-speaks-NONMEM (PsN) - a Perl module for NONMEM related programming. Comput Methods Programs Biomed 2004; 75:85-94. 15. Jonsson EN, Karlsson MO. Xpose–an s-plus based population pharmacokinetic/pharmacodynamic model building aid for nonmem. Comput Methods Programs Biomed 1999; 58:51-64. 16. de Greef R, Zandvliet AS, de Haan AF, Ijzerman-Boon PC, Marintcheva-Petrova M, Mannaerts BM. Dose selection of corifollitropin alfa by modeling and simulation in controlled ovarian stimulation. Clin Pharmacol Ther 2010; 88:79-87. 17. Karlsson MO, Wade JR, Loumaye E, Munafo A. The population pharmacokinetics of recombinant- and urinary-human follicle stimulating hormone in women. Br J Clin Pharmacol 1998; 45:13-20. 18. le Cotonnec JY, Porchet HC, Beltrami V, Howles C. Comparative pharmacokinetics of two urinary human follicle stimulating hormone preparations in healthy female and male volunteers. Hum Reprod 1993; 8:1604-1611. 19. le Cotonnec JY, Porchet HC, Beltrami V, Khan A, Toon S, Rowland M. Clinical pharmacology of recombinant human follicle-stimulating hormone. II. Single doses and steady state pharmacokinetics. Fertil Steril 1994; 61:679-686. 20. le Cotonnec J-Y, Loumaye E, Porchet HC, Beltrami V, Munafo A. Pharmacokinetic and pharmacodynamic interactions between recombinant human luteinizing hormone and recombinant human follicle-stimulating hormone. Fertil Steril 1998; 69:201-209. 21. Porter CJH, Charman SA. Lymphatic transport of proteins after subcutaneous administration. J Pharm Sci 2000; 89:297-310. 22. Mannaerts B, Shoham Z, Schoot D, Bouchard P, Harlin J, Fauser B et al. Single-dose pharmacokinetics and pharmacodynamics of recombinant human follicle-stimulating hormone (Org 32489*) in gonadotropin-deficient volunteers. Fertil Steril 1993; 59:108-114. 23. Mannaerts BM, Rombout F, Out HJ, Coelingh BH. Clinical profiling of recombinant follicle stimulating hormone (rFSH; Puregon): relationship between serum FSH and efficacy. Hum Reprod Update 1996; 24. Ledger WL, Fauser BCJM, Devroey P, Zandvliet AS, Mannaerts BMJL. Corifollitropin alfa doses based on body weight: clinical overview of drug exposure and ovarian response. Reprod BioMed Online 2011;
Table 1. Summary of subject characteristics
Dose (µg)
Age (years)
Height (cm)
(154.9-166.0) (157.5-175.3) (154.9-182.9) (152.4-162.6) (152.4-170.2) Weight (kg)
BMI (kg/m2)
Personal demographics for all the subjects in the five dosing groups. The values are mean with range in brackets. n: number of subjects
in each group, BMI: body mass index.
Table 2. Pharmacokinetic parameter estimates
Parameter
Estimate (RSE%) IIV CV% (RSE%)
Shrinkage (%)
CL/F (L/h)
ktr (h-1)
ka (h-1)
Typical population pharmacokinetic parameter estimates obtained from modelling with the relative standard error (RSE) in brackets. F: bioavailability, CL/F: apparent clearance, V/F: apparent volume of distribution, ktr: absorption rate from the dosing site to the transit compartment, and ka: absorption rate to the central compartment. For CL/F and V/F the value is the typical value for a woman weighing 65 kg. The interindividual variability (IIV) is listed as the coefficient of variation (CV) with RSE in brackets and corresponding ETA shrinkage in percentage.
Figure 1. The observed FSH concentration shown as mean of all subjects with standard error (SE)
bars for each treatment group. The grey line represents the LLOQ of 0.075 µg/L. Observed BQL measurements were plotted as LLOQ/2. Figure 2. Model predictions compared to observations. (a) Points are mean of
observations with standard error (SE) bars. Lines are typical model predictions for each treatment group. The grey line represents the LLOQ of 0.075 µg/L. Observed BQL measurements were plotted as LLOQ/2. (b) Two panel visual predictive check for all
dose groups together. The top panel shows the observations above LLOQ (points) and the 50th and 97.5th percentiles of observations (purple lines). The 2.5th percentile of observations is not shown since it solely consists of BQL points. The shaded areas are the simulated 95% confidence intervals (CI) for the 2.5th, 50th, and 97.5th percentiles. The grey line represents the LLOQ of 0.075 µg/L. In the bottom panel the blue line is the fraction of BQL observations with the 95% CI for the median from simulations. Figure 3. Goodness of fit plots. (a) Observations against population predictions (purple
points *) and individual predictions (blue points +) with the unity line. The grey line represents the LLOQ of 0.075 µg/L. (b) Individual residuals against individual
predictions (points) with a smooth lowess line. Figure 4. Effect of body weight at the FSH concentration. (a) Simulation of the typical
expected FSH concentration after multiple dosing of 10 µg FE 999049 for three subjects with different body weights. (b) Boxplot of the average steady state concentration
obtained from 1000 simulations for each weight group. Characterisation of Population Pharmacokinetics and Endogenous
Follicle Stimulating Hormone (FSH) Levels after Multiple Dosing of a
Recombinant Human FSH, FE 999049, in Healthy Women
Trine Høyer Rose1,2, Daniel Röshammar1, Lars Erichsen1, Lars Grundemar1, Johnny T. Ottesen2 1. Experimental Medicine, Ferring Pharmaceuticals A/S, Denmark 2. Department of Science, Systems and Models, Roskilde University, Denmark
Corresponding author: Trine Høyer Rose, Ferring Pharmaceuticals A/S, Kay Fiskers Plads 11,
DK-2300 Copenhagen S, Denmark. E-mail: [email protected]
Abstract
Background and Objective
To characterise the population pharmacokinetics of FE 999049, a novel recombinant human follicle
stimulating hormone (rhFSH), after multiple dosing in healthy women considering endogenous FSH levels.
Methods
Longitudinal measurements of FSH, luteinizing hormone, progesterone, estradiol, and inhibin B levels were
collected after repeated subcutaneous dosing with 225 IU FE 999049 in 24 gonadotropin down-regulated
healthy women participating in a phase 1 trial. The FSH population pharmacokinetics were evaluated using
nonlinear mixed effects modelling and NONMEM 7.2.0.
Results
The measured FSH levels were modelled as a sum of endogenous FSH and the administered FE 999049. In
this analysis the FE 999049 population pharmacokinetics were best described by a one-compartment model
with first order absorption and elimination. A delay in the absorption was described by a transit model. The
apparent clearance and volume of distribution was found to increase with body weight in accordance to an
allometrically scaled power exponent of 0.75 and 1, respectively. The endogenous FSH levels were
described by a turnover model. Endogenous FSH baseline levels were observed to be lower in individuals
with higher baseline progesterone levels. The endogenous FSH levels were further suppressed over time with
increasing inhibin B levels.
Conclusions
It can be of importance to account for endogenous FSH levels for accurate estimation of exogenously
administered FSH pharmacokinetic parameters. Moreover, the endogenous FSH levels can be affected by
reproductive hormones with time. Thus, correcting measured total FSH concentrations by the observed
endogenous FSH baseline value at all time points may be incorrect.
Key Points
The multiple dose pharmacokinetics of FE 999049 have been described accounting for endogenous follicle stimulating hormone (FSH) levels. The exposure to FE 999049 was influenced by body weight. Endogenous FSH levels were influenced by progesterone and inhibin B levels. When characterising the pharmacokinetics of recombinant FSH products the time varying contribution of endogenous FSH may be important to consider. 1. Introduction
Follicle stimulating hormone (FSH) is a gonadotropin synthesised and secreted by the anterior pituitary
gland. The major function of FSH is to regulate the reproductive processes by stimulating the gonads. In
females, FSH stimulates follicular development in the ovaries and production of inhibin B, progesterone, and
oestrogens by the ovarian follicular granulosa cells. Luteinizing hormone (LH), another gonadotropin from
the anterior pituitary, stimulates the theca cells of the follicles to deliver androgens to the granulosa cells for
conversion to oestrogens. LH is also responsible for ovulation of the dominant follicle that has reached a full
mature preovulatory stage. The ovarian hormones promote further follicular development as well as exerting
negative and positive feedback loops to the hypothalamus and pituitary affecting the gonadotropin
production and secretion. In addition, gonadotropin secretion is stimulated by gonadotropin-releasing
hormone (GnRH) produced in the hypothalamus.
Female infertility can be caused by numerous factors at any level from the hypothalamus to the ovaries and uterus. Gonadotropin therapy with either menotropins or recombinant FSH (rFSH) preparations can be used for infertility treatment when the cause is not primary ovarian failure, such that the ovaries are still responsive with primordial follicles. The purpose of controlled ovarian stimulation with daily administration of gonadotropins prior to assisted reproductive technologies such as in vitro fertilisation (IVF) or intracytoplasmic sperm injection (ICSI) is to obtain an adequate number of oocytes per retrieval with the minimum risks for the woman [1]. An appropriate ovarian response leading to availability of several embryos makes it possible to select the best one(s) for transfer. Recently, a novel recombinant human FSH (rhFSH, FE 999049, Ferring Pharmaceuticals A/S) has been expressed for the first time in a human cell line (PER.C6®, Crucell, Leiden, The Netherlands), while existing
rFSH preparations in clinical use (e.g. follitropin alfa and follitropin beta) are derived from Chinese hamster
ovary cell lines (CHO). Previously, in a population pharmacokinetic analysis of first-in-human data after
single ascending doses [2], body weight was identified as a factor that negatively correlates with serum FE
999049 concentration. In the present work the FE 999049 population pharmacokinetics after multiple dosing
are characterised. In addition, the endogenous FSH contribution to the total FSH levels and the covariate
influence of other reproductive hormones are evaluated.
2. Methods
2.1 Clinical Trial Design and Data
Data was generated in a randomised, double-blind, active control, multiple dose trial with the aim to
investigate the safety, tolerability, immunogenicity, pharmacokinetics, and pharmacodynamics of FE 999049
in healthy women. The trial was performed according to the Helsinki declaration and good clinical practice.
It was approved by regulatory authorities and local ethics committees. All subjects gave written informed
consent to participate. The trial has been described in more detail in a recent publication comparing the
pharmacokinetic and pharmacodynamic properties of FE 999049 and GONAL-F (follitropin alfa, EMD
Serono) using non-compartmental analysis (NCA) [3]. Briefly, 49 healthy women were given daily
subcutaneous doses of 225 IU rFSH for 7 days. 24 out of the 49 women were treated with FE 999049 and 25
women received GONAL-F as an active comparator. Prior to the trial (day -28 and -14) subjects were given
two doses of a GnRH agonist (LUPRON DEPOT, 1-month depot) to down-regulate endogenous FSH.
Blood samples for FSH, inhibin B, estradiol, progesterone, and LH measurements were collected 60 and
30 minutes prior to administration of FE 999049, immediately before administration, and once a day for 15
days. In addition, after administration of the last dose on day 6 and until day 8 the FSH concentration was
measured every 4th hour. Analysis of serum FSH concentrations was performed at Ferring Pharmaceuticals
A/S with a validated immunoassay based on electrochemiluminescence (MSD sectorTM Imager 2400)
with a lower limit of quantification (LLOQ) of 0.075 µg/L.
The present analysis included data from the 24 women receiving FE 999049. Three out of the 672 FSH
measurements (0.4 %) were below the quantification limit and excluded from further analysis. The personal
demographics and baseline characteristics for the included subjects are listed in Table 1. Subjects with
missing hormone baseline values were given the median population baseline value. Between day 6 and 8,
FSH was measured every 4th hour and the other hormones were measured only once a day, leaving missing
hormone values in between. To fill out the extra time points the last measured hormone values were carried
forward.
2.2 Pharmacokinetic Modelling
The PK model was developed using nonlinear mixed effects modelling, where both the population
parameters, interindividual variability (IIV), and residual errors are estimated. For parameters with IIV
the ith subject's individual parameter, θi, is log-normal distributed:
where θ is the typical population parameter and ηi is the individual random effect from an approximately
normal distribution with mean zero and variance ω2 for describing the IIV of the parameter. In the model
potential influential factors can be tested for significance as a covariate to explain some of the IIV in a
parameter. Thus the set of individual parameters, Θi, is given as a function of the typical population
parameters, Θ, individual values of the covariates, ci, and random effects, ηi:
The residual errors, ε, are assumed normally distributed with mean zero and variance σ2, and are the unexplained deviations of model predictions from the observations:
yij is the observation at time tij and the subscript ij denotes the jth number for subject i. The individual model
prediction f (·) at time tij is calculated from the individual parameters, Θi.
Model development was guided by changes in the NONMEM objective function value (OFV), precision
of parameter estimates, and graphical model goodness-of-fit assessments including visual predictive checks
(VPC). In the VPC the observed data is compared to model predictions based on 1000 simulated trial
datasets using the final model. It displays the observations and the 2.5th, 50th, and 97.5th percentiles of
observations and the 95% confidence intervals (CI) for the corresponding model predictions are plotted. The
OFV is approximately proportional to -2log likelihood. The difference in OFV between two nested models is
approximately χ2-distributed, with degrees of freedom (df) equal to the difference in the number of
parameters. Based on this, the statistical significance for inclusion/exclusion of a model parameter can be
judged. A significance level of 0.05 was used for discrimination among nested models and covariate testing.
It corresponds to a 3.84 change in OFV for 1 df.
Body weight, age, and hormone (estradiol, inhibin B, progesterone) values at baseline were tested as
potential covariates explaining some of the IIV in parameter estimates. Inhibin B was also tested as a
covariate over time, since it has a purely inhibitory effect at FSH. Besides a decrease in OFV the significance
of a covariate was also evaluated by looking at the reduction in the IIV measured as coefficient of variation
(CV) for the parameter's random effect.
2.3 Software
The models were implemented and parameters estimated in NONMEM 7.2.0 (Icon Development Solutions,
USA) [4]. Data handling and graphical representations were performed in R version 2.11.1 [5]. VPCs were
performed using PsN [6;7] and plotted using Xpose [8].
3. Results
Measurable FSH levels before drug administration indicated that endogenous FSH was not fully suppressed
in this trial. To obtain accurate PK parameter estimates it was therefore necessary to model the total FSH at
time t as the sum of the endogenous FSH and the exogenously administered rhFSH:
The FE 999049 absorption was found to be delayed and a transit compartment was introduced with absorption rate ktr. The change over time in the FE 999049 amount in the central compartment was given by where rhFSHT(t) is the amount entering from the transit compartment with rate constant ka. The elimination rate constant k is given by clearance (CL) divided by volume of distribution (V). Since data is obtained after subcutaneous dosing the bioavailability (F) is not known. The CL and V estimated here are therefore the apparent clearance (CL/F) and apparent volume of distribution (V/F). The endogenous FSH was assumed to have the same elimination rate constant from the central compartment as FE 999049 and a zero order production rate constant kin. The change in endogenous FSH amount over time in the central compartment was described as The endogenous baseline FSH concentration (FSHbl) was estimated for each subject to be the initial concentration in the central compartment before dosing. IIV was introduced for CL/F, V/F, ktr, and FSHbl. The variances of the IIV on CL/F and V/F were positively correlated. Body weight was an allometrically scaled covariate at CL/F and V/F with the power exponents fixed to 0.75 and 1, respectively. Adding body weight as a covariate reduced CV for the unexplained IIV from 18.1 to 15.6% CV for CL/F and from 22.0 to 18.4% CV for V/F. The measured hormone levels were assumed to not affect the rFSH concentration but only tested as covariates on the endogenous baseline FSH parameter (FSHbl). A trend towards lower estimated FSH baseline levels were seen in individuals with higher progesterone baseline levels (Figure 1). The progesterone baseline effect was confirmed to be a statistically significant covariate when modelled as an inhibitory power function at the FSHbl parameter. This decreased IIV in FSHbl from 32.6 to 27.8% CV. Moreover, observed inhibin B levels (InhB(t)) suppressed endogenous FSH production rate (kin) over time when introduced in the model as a time-varying covariate. An Imax function with a parameter InhBef for the inhibin B concentration yielding half of maximum suppression was found to best describe the relationship. The equation for the total FSH amount then becomes
Individual profiles for the resulting model predicted endogenous FSH level and observed inhibin B are
shown in Figure 2. Accounting for the inhibin B suppression of the endogenous FSH contribution in this way
increased the V/F estimate from 18.9 to 24.3 L. Not including inhibin B as a covariate in the final model
resulted in an increase from 83.4 to 164.9% CV at ktr.
When the best model was found in a forward development process it was checked if all the elements still
were significant or if the model could be reduced. In Table 2 is listed the resulting increase in OFV from
removing the elements. No reduction was possible. The best structural model is illustrated in Figure 3. A
combined additive and proportional error model was used to describe the residual error. The final model
parameter estimates are displayed in Table 3. The VPC plot (Figure 4) based on 1000 simulations, shows that
the model predicts the FSH concentration PK profile well with appropriate variation.
4. Discussion
The FE 999049 population pharmacokinetics have previously been characterised after single dose
administration [2]. In accordance with those results, the current analysis showed that after multiple FE
999049 dosing the population pharmacokinetics were also best described by a one-compartment model with
first order elimination and absorption through a transit compartment. There was a correlation between CL/F
and V/F, and body weight was an allometrically scaled covariate for both CL/F and V/F. Hereby, it was
confirmed that lower average steady-state drug exposure in females correlates with higher body weight after
repeated doses, which is in agreement with single dose FE 999049 results.
In contrast to the single ascending dose trial the FSH down-regulation in this trial was not complete,
since a measurable endogenous FSH level was detected before any drug administration. The observed
endogenous FSH contributes to total FSH measurements and thereby affects the evaluation of rhFSH
concentrations in the trial, thus it had to be accounted for in the model. To adjust the model accordingly it
was necessary to add an endogenous supply of FSH to the central compartment for proper prediction of total
FSH concentration.
When evaluating the pharmacokinetics of drugs that are naturally occurring substances, data is often
baseline corrected in order to get values only representing the exogenously administered drug. However,
when dealing with hormones there may be fluctuations from the endogenous baseline value over time due to
various feedback control systems. Even with gonadotropin down-regulation, the endogenous FSH level will
most likely change over time as progesterone, inhibin B, and estradiol exert inhibitory and stimulatory
feedback loops at the gonadotropin production and release. The potential impact of varying endogenous FSH
levels at the pharmacokinetic assessment of FE 999049 was tested by incorporating different covariate
relationships in the model. Before any drug was administered there was a tendency towards a decreased
endogenous FSH baseline with increasing progesterone baseline. A power function for the relation between
progesterone baseline and FSHbl was shown to be significantly better than using a linear, an exponential, or
an Imax function. When introduced as a time-varying covariate inhibin B was found to suppress the
endogenous FSH production rate over time with an Imax function being the most significant relationship.
When accounting for the inhibitory effect of inhibin B over time, the resulting model-predicted endogenous
FSH concentration profiles indicated that the observed FSH baseline is an overestimate of the endogenous
FSH level at all other time points than zero (Figure 2).
When multiple covariates and correlations have been added to the model, it is possible that initial
significant relations have become redundant and hence can be removed from the model. It was therefore
checked if the model could be reduced. Removing any relations resulted in an increase in OFV, thus all
relations were still significant. Removing WT from both CL/F and V/F increased OFV less than removing
only one of them. This could be due to that there is a correlation between the two parameters.
The population PK parameters estimated with nonlinear mixed effect modelling for single dose and
multiple doses of FE 999049 are similar. In the modelling of the first in human data CL/F and V/F for a 65
kg woman was found to be 0.430 L/h and 28.0 L, respectively [2]. In this trial CL/F was 0.423 L/h and V/F
was 24.3 L. By incorporating the inhibin B dynamics in the model and suppressing the endogenous FSH
contribution instead of letting the FSH baseline be constant throughout the trial, V/F increased from 18.9 to
24.3 L. This suggests that without proper estimation of endogenous FSH levels over time a bias in V/F may
be obtained.
5. Conclusion
The multiple dose FE 999049 population pharmacokinetics were in agreement with results obtained after
single dose administration [2]. It was confirmed that after repeated drug administration, drug exposure also
appears lower in females with higher body weight.
When subjects do not have fully suppressed endogenous FSH levels, it affects total FSH concentration,
thus inclusion of endogenous FSH levels in the PK modelling of FSH preparations could be important. In
addition, the endogenous FSH level will possibly not be constant over time and quantifying the influence of
the endocrine hormone dynamics during the trial may well be essential for proper estimation of
pharmacokinetic parameters. The standard method of baseline correcting data does not account for a
variation over time and could potentially cause an underestimation of serum drug concentration and hence
inaccurate parameter estimation.
To better explore and link the time-varying impact of inhibin B on the endogenous FSH levels a more
mechanistic modelling approach may be warranted for including indirect delayed response and hormone
feedback mechanisms in a PKPD model and simultaneously quantify the inhibin B response to rFSH
stimulation.
Compliance with Ethical Standards
Funding: This work was supported by Innovation Fund Denmark (Industrial PhD case number 11-
117436).
Conflicts of Interest: Trine Høyer Rose, Daniel Röshammar, Lars Erichsen, and Lars Grundemar
are all current or former employees at Ferring Pharmaceuticals A/S.
Ethical approval: All procedures performed in studies involving human participants were in
accordance with the ethical standards of the institutional and/or national research committee and
with the 1964 Helsinki declaration and its later amendments or comparable ethical standards.
Informed consent: Informed consent was obtained from all individual participants included in the
study.
1. Macklon NS, Stouffer RL, Giudice LC, Fauser BC. The science behind 25 years of ovarian stimulation for in vitro fertilization. Endocr Rev 2006; 27:170-207. 2. Rose TH, Röshammar D, Erichsen L, Grundemar L, Ottesen JT. Population Pharmacokinetic Modelling of FE 999049, a Human Recombinant Follicle Stimulating Hormone, in Healthy Women after Single Ascending Doses. submitted 2015. 3. Olsson H, Sandström R, Grundemar L. Different pharmacokinetic and pharmacodynamic properties of recombinant follicle-stimulating hormone (rFSH) derived from a human cell line compared with rFSH from a non-human cell line. The Journal of Clinical Pharmacology 2014; 54:1299-1307. 4. Beal S, Sheiner LB, Boeckmann A, Bauer RJ. NONMEM User's Guides. (1989-2011). 2011. Icon Development Solutions, Ellicott City, MD, USA. 5. R Development Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria; 2010. 6. Lindbom L, Ribbing J, Jonsson EN. Perl-speaks-NONMEM (PsN) - a Perl module for NONMEM related programming. Comput Methods Programs Biomed 2004; 75:85-94. 7. Lindbom L, Pihlgren P, Jonsson EN. PsN-Toolkit -- a collection of computer intensive statistical methods for non-linear mixed effect modeling using NONMEM. Comput Methods Programs Biomed 2005; 79:241-257. 8. Jonsson EN, Karlsson MO. Xpose–an s-plus based population pharmacokinetic/pharmacodynamic model building aid for nonmem. Comput Methods Programs Biomed 1999; 58:51-64.
Table 1. Summary of subject characteristics
Age (years)
Height (cm)
163.5 (149.0 - 175.3) Weight (kg)
BMI (kg/m2)
FSH (µg/L)
0.211 (0.089 - 0.376) E2 (pg/mL)
LH (IU/L)
Prog (µg/L)
InhB (pg/mL)
Demographics and baseline hormone levels for the 24 subjects receiving FE999049. The values are mean with range in brackets. BMI is body mass index. FSH, E2, LH, Prog, and InhB is the measured baseline concentration of follicle stimulating hormone, estradiol, luteinizing hormone, progesterone, and inhibin B, respectively.
Table 2. Change in objective function value for reduced models
Removing
WT at CL/F
WT at CL/F and V/F
Progesterone effect
Inhibin B effect
The resulting change in objective function value (dOFV) when removing covariates or the correlation between CL/F and V/F (cov(CL/F,V/F)). F: bioavailability, CL/F: apparent clearance, V/F: apparent volume of distribution, WT: body weight, df: degrees of freedom. Table 3. Pharmacokinetic parameter estimates
Parameter
Estimate (RSE%) IIV CV% (RSE%) Shrinkage (%)
CL/F (L/h)
ktr (h-1)
ka (h-1)
FSHbl (µg/L)
Progblef
Typical population parameter estimates obtained from modelling with the relative standard error (RSE) in brackets. F is the bioavailability and the parameters are the apparent clearance (CL/F), apparent volume of distribution (V/F), absorption rate from the dosing site (ktr), absorption rate from the transit compartment (ka), endogenous FSH baseline (FSHbl), power exponent for progesterone baseline covariate effect (Progblef), and inhibin B time-varying covariate effect (InhBef). For CL/F and V/F the value is the typical value for a woman weighing 65 kg. The interindividual variability (IIV) is listed as the percentage coefficient of variation (CV) with RSE in brackets and corresponding ETA shrinkage in percentage. Figure 1 The relationship between endogenous FSH and progesterone at baseline. Points are individual predicted endogenous FSH
baseline values (FSHbl) and observed progesterone baseline values with a smooth lowess trend line (broken line). The solid line is the
power function used in the model describing the typical population relationship for the effect of progesterone baseline at the
parameter FSHbl.
Figure 2 Individual hormone concentration profiles over time. The broken blue line is the level for the observed endogenous FSH
baseline. The solid blue line is the model predicted endogenous FSH level when not assuming it to be constant throughout the trial
but predicting the level based on what the observed inhibin B levels (purple line) over time are based on the model. The number at
each subplot is the subject ID number.
Figure 3 A compartment diagram showing the pharmacokinetic model for FE 999049. It illustrates the contribution of endogenous
FSH (FSHen(t)) to the central compartment with a production rate kin being suppressed by inhibin B levels (InhB(t)) over time. FE
999049 after administration is absorbed from the dosing site with a rate ktr to a transit compartment in which the amount is
rhFSHT(t). The absorption rate from the transit compartment to the central compartment is ka, where total FSH(t) is measured without
differing between the sources. From here the elimination rate for the total FSH is k.
Figure 4 Visual predictive check for the final model. It shows the individual observed FSH concentrations (points) and the 2.5th,
50th, and 97.5th percentiles of observations (lines). The shaded areas are the 95% confidence intervals for the 2.5th, 50th, and 97.5th
percentiles of the simulations.
Semi-Mechanistic Pharmacokinetic-Pharmacodynamic Modelling of
Inhibin B Levels after Multiple Doses of the Recombinant Human
Follicle Stimulating Hormone FE 999049 in Infertile Women
Trine Høyer Rose1,2, Daniel Röshammar1, Mats O. Karlsson3, Lars Erichsen1, Lars Grundemar1, Johnny T.
Ottesen2
1. Experimental Medicine, Ferring Pharmaceuticals A/S, Denmark
2. Department of Science, Systems and Models, Roskilde University, Denmark
3. Department of Pharmaceutical Biosciences, Uppsala University, Sweden
Abstract
Objectives: To develop a semi-mechanistic population pharmacokinetic-pharmacodynamic (PKPD) model
describing the relationship between follicle stimulating hormone (FSH) and inhibin B in infertile women
receiving multiple dosing of FE 999049 - a recombinant human FSH (rhFSH).
Methods: Data from a FE 999049 phase 2 multiple dose trial including 222 infertile women was analysed
using non-linear mixed effect modelling in NONMEM 7.2.0. The patients received daily subcutaneous doses
of either 5.2, 6.9, 8.6, 10.3, or 12.1 µg FE 999049 for a maximum of 16 days. Data contained 1160 FSH
measurements and 1155 inhibin B measurements. The FE 999049 concentration-time profile and inhibin B
levels were modelled simultaneously in a PKPD model taken the dynamics of the reproductive hormone
system into consideration.
Results: The FE 999049 population pharmacokinetics were described by a one compartment distribution
model with a transit compartment for a delayed absorption. The total measured FSH concentration consisted
of both an endogenous FSH contribution and the exogenously administered rhFSH (FE 999049). The total
FSH level stimulated the inhibin B production rate in an indirect turnover response model, and inhibin B
level simultaneously exerted a negative feedback loop at the endogenous FSH production. Body weight was
a significant covariate in the model which resulted in lower FSH exposure as well as lower inhibin B
response with higher body weight.
Conclusion: The semi-mechanistic PKPD model can be used to evaluate longitudinal FSH dose-exposure-
inhibin B relationship over time as an early marker of response in clinical infertility studies. A decrease in
inhibin B response is also seen with increasing body weight in the same way as FSH exposure is affected by
body weight.
Introduction
FE 999049 is a novel recombinant human FSH (rhFSH, Ferring Pharmaceuticals A/S) expressed in a cell-
line of human fetal retinal origin (PER.C6®, Crucell, Leiden, The Netherlands). FE 999049 is intended for
controlled ovarian stimulation by subcutaneous administration to induce maturation of multiple follicles for
in vitro fertilization (IVF)/intracytoplasmic sperm injection (ICSI) treatment. Hormone therapy with
exogenous administration of follicle stimulating hormone (FSH) can be used for infertility treatment when
the cause is not ovarian failure and the ovaries are still responsive to stimulation and contain functional
primordial follicles. This is the case for female ovulation disorders and unexplained infertility which
accounts for 40-50% of infertility cases1, 2.
FSH is a gonadotropin produced by the anterior pituitary gland upon stimulation by gonadotropin releasing hormone (GnRH) from the hypothalamus. FSH is a key hormone in female reproductive function as it stimulates the ovaries to induce follicular development and hormone production. One of the ovarian hormones produced is inhibin B which exerts a negative feedback loop at the FSH production. Inhibin B has been identified as the earliest measureable hormone marker of response to gonadotropin treatment3, 4 and there is substantial evidence that suggests inhibin B is a promising predictor for ovarian response. Inhibin B baseline and the rise in inhibin B after gonadotropin administration are higher in good responders5, 6. Number of oocytes retrieved correlates with inhibin B levels and change in inhibin B during treatment3, 7, 8. Furthermore, inhibin B correlates with both oocyte quality and number of eggs fertilised9, total follicular volume4, and antral follicle count8. It could therefore be valuable to predict and follow the inhibin B response throughout treatment. The population pharmacokinetics of FE 999049 have recently been described after single and repeated administration (paper I and II). The objective of the current analysis was to develop a semi-mechanistic population pharmacokinetic-pharmacodynamic (PKPD) model simultaneously describing the dynamics between endogenous FSH, exogenous rhFSH, and inhibin B in patients receiving repeated administration of FE 999049 in a phase 2b dose finding trial. Body weight has been identified to be an important patient specific factor influencing the population PK of FE 999049 (paper I and II). Subjects with high body weight has in general been reported to experience lower drug exposure compared to subjects with low body weight following administration of the same dose. An additional objective of this paper was to explore how the body weight related differences in drug exposure translates into between-patient differences in the pharmacodynamic response to FSH treatment measured as the biomarker inhibin B. Methods
Clinical Trial Design and Data
The data used for the analysis was collected in a phase 2b dose finding trial. It was a randomised, controlled,
assessor-blinded, parallel-group, multicentre, multiple dose trial assessing the dose-response relationship of
FE 999049 in women undergoing IVF/ICSI. The trial was performed according to the ethical principles in
the Helsinki declaration and in compliance with good clinical practice and regulatory requirements. The trial
was approved by local regulatory authorities and independent ethics committees. All subjects gave written
informed consent to participate.
The trial has been described more in detail in an earlier publication10. In summary, 265 women with tubal infertility, unexplained infertility, infertility related to endometriosis stage I/II, or with partners diagnosed with male factor infertility were included in the trial. They received daily subcutaneous doses of either 5.2 (n=42), 6.9 (n=45), 8.6 (n=44), 10.3 (n=45), or 12.1 µg (n=46) (90, 120, 150, 180 or 210 IU) FE 999049 or 11 µg (n=43) (150 IU) GONAL- F (follitropin alfa, EMD Serono) for a maximum of 16 days. Randomisation was stratified according to antimüllerian hormone (AMH) level at screening (low: 5.0-14.9 pmol/L and high: 15.0-44.9 pmol/L). From day 5 after the first dose and throughout the stimulation period a GnRH antagonist (0.25 mg ganirelix acetate, ORGALUTRAN, MSD / Schering-Plough) was given to prevent a premature luteinising hormone (LH) surge. Blood samples for FSH, LH, inhibin A, inhibin B, estradiol, progesterone, and testosterone measurements were collected immediately before the first administration, at day 3 and day 5 after the first dose, and hereafter at least every second day. When 3 follicles of ≥15 mm were observed, visits had to be performed on a daily basis. Each patient's treatment length depended on the individual response. Doses were given until three or more follicles with a diameter ≥17 mm were observed. The cycle would be cancelled if there were either too many (more than 35 follicles ≥12 mm) or too few (less than three follicles ≥10 mm at day 10) growing follicles. Analysis of serum FSH concentrations was performed at ICON Central Laboratories, Dublin, Ireland, with a chemiluminescent immunometric assay (IMMULITE 2500 FSH (ROCHE), with a lower limit of quantification (LLOQ) of 0.0052 µg/L. Inhibin B was measured by an enzyme linked immunosorbent assay (Gen II ELISA (Beckman Coulter)) with an LLOQ of 4.8 pg/mL. The current analysis included data from the 222 women receiving FE 999049 giving a total of FSH 1160 measurements and 1155 inhibin B measurements. The personal demographics for the included subjects are listed in Table 1. Population PKPD Modelling
The population PKPD model was developed using a nonlinear mixed effects modelling. The population PK
of FSH and the exposure-response relationship between FSH and inhibin B were described in a simultaneous
PKPD model developed by a semi-mechanistic modelling approach in order to incorporate the hormone
dynamics with feedback mechanisms in the longitudinal data.
Endogenous FSH baseline levels were detectable and had to be included in the model as a contribution to the central compartment such that total FSH concentration was modelled as a sum of endogenous FSH and exogenous administered rhFSH. Following the dynamics in the reproductive endocrine system total FSH should stimulation the production of inhibin B, and endogenous FSH production should be inhibited by inhibin B. Different functions were tested for describing the stimulating and inhibitory processes both with and without an effect compartment for a delayed indirect response. A GnRH antagonist was administered at day 5 to avoid a premature LH surge, but this also inhibits the production of FSH. To follow the protocol this inhibition of endogenous FSH after day 5 should be added in the model. Turnover models were used to describe the FSH and inhibin B concentrations. When the dose is given subcutaneously, the bioavailability (F) is not known. The parameters estimated here are therefore the apparent clearance (CL/F) and apparent volume of distribution (V/F). In the current dataset there were few observations per individual, which was not sufficient to estimate all PK parameters, therefore V/F and the absorption rate constant (ka) were fixed to the values found in the PK model from phase 1 data (paper II). Individual parameters (θi) were obtained from typical population parameters (θ) with addition of an exponential function of an individual random effect (ηi) from an approximately normal distribution with mean zero and variance ω2. Body weight was tested as a covariate by including individual body weight (WTi) normalised to 65kg in a power function. Parameters with both IIV and body weight as a covariate were described by where Pθ is the power exponent for the weight effect at the parameter θ. The intraindividual variability was modelled with a combined additive and proportional residual error model. Data was log-transformed and separate error models for FSH and inhibin B were used. Model Evaluation
The models were implemented and parameters estimated in NONMEM 7.2.0 (Icon Development Solutions,
USA)11. Model development was guided by changes in the NONMEM objective function value (OFV),
precision of parameter estimates, and graphical model goodness-of-fit assessments. The OFV is
approximately proportional to -2log likelihood. The difference in OFV between two nested models is
approximately χ2-distributed, with degrees of freedom (df) equal to the difference in the number of
parameters. Based on this, the statistical significance for inclusion/exclusion of a model parameter can be
judged. A significance level of 0.05 was used for discrimination among nested models and covariate testing.
It corresponds to a 3.84 change in OFV for 1 df. Besides a decrease in OFV the significance of a covariate
was also evaluated by looking at the reduction in the IIV measured as coefficient of variation (CV) for the
parameter's random effect.
Graphical evaluations include population mean concentration-time profiles, individual profiles, residual plots, and visual predictive checks (VPC). In the VPC the observed data is compared to model predictions based on 1000 simulated trial datasets using the final model. It displays the observations and the 2.5th, 50th, and 97.5th percentiles of observations as well as the 95% confidence intervals (CI) for the corresponding model predictions. Data handling and graphical representations were performed in R version 2.11.112. VPCs were performed using PsN13, 14 and plotted using Xpose15.
Results
The PK part of the model was found to be a one-compartment distribution model where the total amount of
FSH was modelled as a sum of the endogenous FSH and the administered rhFSH in form of FE 999049.
The absorption of FE 999049 was found to be first order with rate ktr to a transit compartment from where it
entered the central compartment with rate constant ka. The elimination rate constant k is given by CL/F
divided by V/F and was assumed to be the same for endogenous FSH and FE 999049.
An inhibitory effect from the GnRH antagonist at the endogenous FSH production rate constant (kendo) were added after day 5 to be in accordance with the trial design. An exponential function, a linear function, log-linear, and several types of Emax models were tested for describing the FSH stimulation of inhibin B, but either the fit was poor or data did not support parameter estimation of the more advanced functions. A power function with the FSH concentration input normalised with baseline values and power exponent λ was found to describe data best. It was tested if an effect compartment could be included to induce an indirect delayed response. There were indications of an improved fit hereby but the OFV was not significant better hence the effect compartment was left out. Likewise was tested several functions for the inhibition of endogenous FSH by the predicted inhibin B concentrations where an Imax function was found most suitable with the best OFV. When assuming full suppression is possible the Imax function is given by where IC50 is the inhibin B concentration giving half suppression. The structural model was described by the differential equations (2)-(5), one for each of the four compartments in the model as illustrated in Figure 1. ANTAef is the effect of the GnRH antagonist and is given by rhFSHDS(t) and rhFSHTR(t) is the FE 999049 amount left at the dosing site and in the transit compartment at time t in days, respectively. FSH(t) is the total amount of FSH in the central compartment, thus the FSH concentration is obtained by dividing the amount with V/F. InhB(t) is the inhibin B concentration at time t. Inhibin B had a production rate constant kin and was eliminated by the rate constant kout. The endogenous FSH baseline concentration (FSHbl) and inhibin B baseline concentration (InhBbl) was estimated for each subject to be the initial concentration in the central compartment (equation (4)) and in the inhibin B compartment (equation (5)) before dosing, respectively. A random effect for IIV was introduced for CL/F, ktr, FSHbl, InhBbl, and λ. Choices of stimulation and inhibiting feedback functions greatly affected the other parameter estimates and a full covariance matrix was therefore tested and found to be significant better. A combined additive and proportional error model was used to describe the residual error for FSH and inhibin B separately. Body weight was an allometrically scaled covariate at CL/F and V/F with the power exponent Pθ in equation (1) fixed to 0.75 and 1, respectively. Adding body weight as a covariate reduced CV for the unexplained IIV from 27.2 to 23.1% CV for CL/F. Body weight was also found to be a significant covariate at ktr and λ with a reduction from 51.1, and 48.9% CV to 48.3, and 42.7% CV, respectively. The estimated model parameters are listed in Table 2. Mean concentration profiles are shown in Figure 2 and VPCs in Figure 3 with separate graphs for FSH and inhibin B for each dose. The mean predictions fit observations nicely but in the VPC an over-prediction for the lowest dose and an under-prediction for the highest dose is observed for FSH concentrations, and inhibin B is over-predicted for the highest dose. Since the frequency subjects come in to the clinic for measurements depends on fulfilment of pre-set criteria for follicle number and size, different number of subjects are measured per day. There are even days with only one subject from a dosing group and therefore the observed percentiles in the VPC collapse to the same value. FSH and inhibin B concentration-time profiles were simulated for three subjects weighing 50 kg, 75 kg, and 100 kg, respectively, after 7 doses of 10 µg FE 999049 using the model with typical parameters
(Figure 4). The FSH exposure as well as inhibin B response decrease with increasing body weight.
Discussion
The FE 999049 dose-concentration-inhibin B response relationship was described by a semi-mechanistic
PKPD model with incorporation of endogenous FSH. The total FSH concentration stimulates the inhibin B
production rate. In return the inhibin B levels suppress endogenous FSH production rate over time when
modelled as a simultaneous negative feedback. The PK information in the data was very sparse and not
enough to estimate all PK parameters, so ka and V/F were fixed to values from paper II. In addition IC50 and
kout for inhibin B had to be fixed to stabilise the model.
The inclusion of a full covariance matrix indicates correlation between PK and PD parameters and significant variation between subjects. In the reproductive system FSH and inhibin B are correlated, thus it makes sense the model parameters describing the relationship are as well. Different functions for the stimulatory and inhibitory dynamics between FSH and inhibin B were tested both with and without an effect compartment. It could be necessary with an effect compartment but the current data could not support the estimation of an extra parameter for the effect compartment nor a more complex stimulation function. De Greef et al.16 described the FSH stimulation of inhibin B production by a sigmoidal Emax function in an indirect response model. However, even with a large dataset they had to fix the Hill coefficient due to a large standard error. In addition, it was a sequential model where a population PK model was developed first and then the FSH concentrations were used in the PD model. It was mostly an empirical model and they did not account for endogenous FSH. Their model did not predict inhibin B levels well at first and to catch an undershoot in inhibin B below the baseline values at later time points, a hypothetical modulator that stimulated the elimination of inhibin B was included to further lower the predicted level. Together with three PD models for follicular volume, cancellation rate, and number of oocytes, they used the model to simulate ovarian response and dose selection. Inhibin B levels have also been described by others after multiple s.c. doses of Gonal-F in pituitary down-regulated healthy female volunteers with a sequential PKPD model4. They did not use a population modelling approach but fitted the model to individual data. The PK model was previously described by an exponential equation with the same data17 and the estimated PK parameters for each subject were fixed in the PKPD model. Subsequently, the PK model was linked with an effect compartment to the PD model in which the inhibin B response was calculated by a power function from the FSH concentrations in the effect compartment. A high interindividual variation in inhibin B response was observed but no covariate analysis were performed hence no factors causing the variation were identified. They found no correlation between FSH concentrations and inhibin B, and thus concluded that the high variation between subjects in PD parameters was not due to pharmacokinetic variations but different pharmacodynamic sensitivity. It is therefore not enough to adjust dose after variations in FSH concentrations but the response should also be taken into account. Another study used the same PKPD model with a new dataset18 and confirmed the results. However, in that study there was a measurable endogenous FSH at baseline and a constant term was thus added to the power function in the PD model for the effect at baseline. In addition they suggested that variation in pharmacokinetics was caused by large fluctuations in endogenous FSH production over time. In these previous models, inhibin B has been described with different methods and functions, but they were all sequential PKPD models. They too observed great variation in response and one group concluded that endogenous FSH matter. To our knowledge this work is the first PKPD model for FSH and inhibin B modelling the concentrations simultaneously with incorporation of change in endogenous FSH over time after exogenous FSH administration. Unfortunately, the data was not sufficient to properly describe the full dynamics. Furthermore, the data was unbalanced with few samples per subject, who in addition did not stay in the trial the same number of days. When this is not accounted for in the model, it can cause imprecise prediction for later time points, in particular for higher doses where subjects might sooner reach the follicle criteria. In order to investigate the impact of body weight on FSH exposure and inhibin B response simulations were performed for three patients weighing 50 kg, 75 kg, and 100 kg who received 7 daily doses of FE 999049. The simulated FSH exposure decreased with increasing body weight in agreement with previous findings (paper I). After the last dose, the FSH exposure decrease to an endogenous level lower than the pre-dose level due to suppression by the GnRH antagonist and inhibin B levels. Since FSH stimulates the inhibin B production the simulated inhibin B response also decrease with increasing body weight. In addition, body weight affects the FSH stimulation of inhibin B thus the extent of difference between the concentration-time profiles are different for inhibin B response compared to FSH exposure. Body weight play a role in the variation in exposure and thus response but other factors are also involved. Endogenous FSH and variation in its level over time is likely an influential factor too. There are evidence of inhibin B being the first biomarker possible to measure in infertility treatment with FSH products and thus a predictor for follicle development in response to FSH3, 4, 8. Hence, it
is useful to be able to predict inhibin B levels in a patient to determine if the dose is effective and antral
follicles have been recruited and entered the gonadotropin dependent growth phase. The model from this
work could be used in simulations to predict outcomes of different doses to women of different weights.
Then, the potential of inhibin B as a marker for ovarian response could be investigated and be related to later
clinical PD endpoints like oocytes retrieved. In addition, the model could be used to simulate the typical
dose-response relationship for inhibin B to see what dose range is required to get a proper response in inhibin
B. Using the model should increase precision compared to only looking at data with standard statistical
methods or when using a simple empirical dose-response model, and can thus improve decision making and
dose selection.
Declaration of Conflicting Interests
Trine Høyer Rose, Daniel Röshammar, Lars Erichsen, and Lars Grundemar are all current or former
employees at Ferring Pharmaceuticals A/S.
(1) Snick HK, Snick TS, Evers JL, Collins JA. The spontaneous pregnancy prognosis in untreated subfertile couples: the Walcheren primary care study. Hum Reprod 1997;12(7):1582-1588. (2) Smith S, Pfeifer SM, Collins JA. Diagnosis and management of female infertility. JAMA 2003;290(13):1767-1770. (3) Eldar-Geva T, Robertson DM, Cahir N et al. Relationship between serum inhibin A and B and ovarian follicle development after a daily fixed dose administration of recombinant follicle-stimulating hormone. J Clin Endocrinol Metab 2000;85(2):607-613. (4) Porchet HC, le Cotonnec J-Y, Loumaye E. Clinical Pharmacology of Recombinant Human Follicle-Stimulating Hormone. III. Pharmacokinetic-Pharmacodynamic Modeling After Repeated Subcutaneous Administration. Fertil Steril 1994;61(4):687-695. (5) Dzik A, Lambert-Messerlian G, Izzo VM, Soares JB, Pinotti JA, Seifer DB. Inhibin B response to EFORT is associated with the outcome of oocyte retrieval in the subsequent in vitro fertilization cycle. Fertil Steril 2000;74(6):1114-1117. (6) Eldar-Geva T, Ben-Chetrit A, Spitz IM et al. Dynamic assays of inhibin B, anti-Müllerian hormone and estradiol following FSH stimulation and ovarian ultrasonography as predictors of IVF outcome. Hum Reprod 2005;20(11):3178-3183. (7) Decanter C, Pigny P, Lefebvre C, Thomas P, Leroy M, Dewailly D. Serum inhibin B during controlled ovarian hyperstimulation: an additional criterion for deciding whether to proceed with egg retrieval. Fertil Steril 2009;91(6):2419- (8) Yong PY, Baird DT, Thong KJ, McNeilly AS, Anderson RA. Prospective analysis of the relationships between the ovarian follicle cohort and basal FSH concentration, the inhibin response to exogenous FSH and ovarian follicle number at different stages of the normal menstrual cycle and after pituitary down-regulation. Hum Reprod 2003;18(1):35-44. (9) Muttukrishna S, McGarrigle H, Wakim R, Khadum I, Ranieri DM, Serhal P. Antral follicle count, anti-müllerian hormone and inhibin B: predictors of ovarian response in assisted reproductive technology? BJOG 2005;112(10):1384-1390. (10) Arce JC, Nyboe Andersen A, Fernández-Sánchez M et al. Ovarian response to recombinant human follicle-stimulating hormone: a randomized, antimüllerian hormonestratified, doseresponse trial in women undergoing in vitro fertilization/intracytoplasmic sperm injection. Fertil Steril 2014;102(6):1633-1640. (11) Beal S, Sheiner LB, Boeckmann A, Bauer RJ. NONMEM User's Guides. (1989-2011). Icon Development Solutions, Ellicott City, MD, USA; 2011. (12) R Development Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria; 2010. (13) Lindbom L, Ribbing J, Jonsson EN. Perl-speaks-NONMEM (PsN) - a Perl module for NONMEM related programming. Comput Methods Programs Biomed 2004;75(2):85-94. (14) Lindbom L, Pihlgren P, Jonsson EN. PsN-Toolkit -- a collection of computer intensive statistical methods for non-linear mixed effect modeling using NONMEM. Comput Methods Programs Biomed 2005;79(3):241-257. (15) Jonsson EN, Karlsson MO. Xpose–an s-plus based population pharmacokinetic/pharmacodynamic model building aid for nonmem. Comput Methods Programs Biomed 1999;58(1):51-64. (16) de Greef R, Zandvliet AS, de Haan AF, Ijzerman-Boon PC, Marintcheva-Petrova M, Mannaerts BM. Dose selection of corifollitropin alfa by modeling and simulation in controlled ovarian stimulation. Clin Pharmacol Ther 2010;88(1):79-87. (17) le Cotonnec JY, Porchet HC, Beltrami V, Khan A, Toon S, Rowland M. Clinical pharmacology of recombinant human follicle-stimulating hormone. II. Single doses and steady state pharmacokinetics. Fertil Steril 1994;61(4):679-686. (18) le Cotonnec J-Y, Loumaye E, Porchet HC, Beltrami V, Munafo A. Pharmacokinetic and pharmacodynamic interactions between recombinant human luteinizing hormone and recombinant human follicle-stimulating hormone. Fertil Steril 1998;69(2):201-209. Dose (µg)
Age (years)
Weight (kg)
BMI (kg/m2)
FSH (µg/L)
InhB (pg/mL)
Table 1 Personal demographics for the 222 patients receiving FE 999049 used in the modelling. The values are mean with range in
brackets for each of the 5 dose groups. n indicate the number of patients in each group. BMI is body mass index. FSH and InhB is the
measured baseline concentration of follicle stimulating hormone and inhibin B, respectively.
Parameter
Estimate (RSE%) IIV CV% (RSE%)
CL/F (L/h)
ktr (h-1)
ka (h-1)
FSHbl (µg/L)
InhBbl (pg/mL)
GnRHanta
Power exponent Pθ for body weight at
CL/F
ktr
Table 2 Typical population parameter estimates obtained from modelling with the relative standard error (RSE) in brackets. The
interindividual variability (IIV) is listed as the percentage coefficient of variation (CV) with RSE in brackets. F is the bioavailability
and the parameters are the apparent clearance (CL/F), apparent volume of distribution (V/F), absorption rate from the dosing site
(ktr), absorption rate from the transit compartment (ka), endogenous FSH baseline (FSHbl), inhibin B baseline (InhBbl), inhibin B
concentration yielding half of maximum suppression (IC50), suppressive effect of the gonadotropin releasing hormone antagonist
(GnRHanta), inhibin B elimination rate (kout), and the power exponent for the FSH stimulation at inhibin B (λ). For CL/F, V/F, ktr, and
λ the value is the typical value for a woman weighing 65 kg and Pθ is the power exponent from equation (1) for the body weight
effect at the parameters.
a fixed to values from paper II. b fixed value. c allometric values.
Figure 1 Compartment diagram of the PKPD model . Contributions to the total FSH amount in the central compartment (FSH(t)) are
FE 999049 from the transit compartment (rhFSHTR(t)) and endogenous FSH (FSHen(t)). The endogenous FSH production rate (kendo)
is inhibited by predicted inhibin B concentrations (InhB(t)) and after day 5 also by a gonadotropin releasing hormone (GnRH)
antagonist. The inhibin B production rate (kin) is stimulated by FSH.
ktr: rFSH absorption rate from the dosing site, ka: rFSH absorption rate to the central compartment, k: FSH elimination rate from the
central compartment, kout inhibin B elimination rate.
Figure 2 Illustration of observed FSH (top) and inhibin B (bottom) concentrations and model predictions for each treatment group.
Points are mean of observations with standard error (SE) bars. Lines are typical model predictions.
Figure 3 Visual predictive checks for FSH (top) and inhibin B (bottom). The points are observations with purple lines for the 2.5th,
50th, and 97.5th percentiles of observations. The shaded areas are the simulated 95% confidence intervals (CI) obtained with the
model for the 2.5th, 50th, and 97.5th percentiles.
Figure 4 Illustration of simulated FSH (top) and inhibin B (bottom) concentrations for three patients of different weight. The patients
received 7 daily doses of 10 µg FE 999049.
Source: http://milne.ruc.dk/imfufatekster/pdf/502.pdf
Nowelia couverture 22
Sphagnum magellanicum cladus aducus group ( Bryopsida : Amblystegiaceae ) , 12 ill., 248 VOTRE VOYAGE NOUS INTERESSE…… Si vous envisagez de voyager prochainement, pensez à récolter Revue spécialisée de bryologie quelques échantillons de bryophytes et à nous les faire parvenir dès votre Numéro 22 — juin 2002 ISSN: (1377 - 8412)
Jasn2011060543 1.10
Upregulation of miR-142-3p in Peripheral BloodMononuclear Cells of Operationally Tolerant Patientswith a Renal Transplant Richard Danger,*† Annaïck Pallier,* Magali Giral,*†‡ Marc Martínez-Llordella,§Juan José Lozano, Nicolas Degauque,* Alberto Sanchez-Fueyo,§ Jean-Paul Soulillou,*†‡and Sophie Brouard*‡ *Institut National de la Santé Et de la Recherche Médicale UMR643 and Institut de Transplantation Urologie,Néphrologie, Nantes, France; †Université de Nantes, Nantes, France; ‡Centre Hospitalier Universitaire Hôtel-Dieu,Nantes, France; §Liver Unit, Hospital Clinic Barcelona, CIBEREHD, Barcelona, Spain; and Bioinformatics Platform,CIBEREHD, Barcelona, Spain