PK/PD modeling of 5‐hydroxytryptophan (5‐HTP) challenge test with cortisol measurement in serum and saliva

Abstract This research was planned to build a Pharmacokinetic/Pharmacodynamic (PK/PD) model of 5‐hydroxytryptophan (5‐HTP) challenge study including a circadian rhythm component of cortisol and to predict serum cortisol based on saliva cortisol. Data from three 5‐HTP challenge studies in healthy volunteers were collected. Serum 5‐HTP, saliva, and serum cortisol were sampled as PK and PD marker. The population PK/PD modeling approach was applied. A baseline model of serum cortisol was built to assess the circadian rhythm before a pharmacodynamic model was used to evaluate the drug effect of the 5‐HTP on cortisol. Finally, linear and power function relationships were tested to predict serum cortisol based on saliva cortisol. The PK of 5‐HTP could be described using a one‐compartment model with a transit compartment. The typical value for clearance was 20.40 L h−1 and showed inter‐study variability. A cosine function was chosen and properly described the circadian rhythm of serum cortisol. A linear approximation model was applied to fit the 5‐HTP PD effect on cortisol data with a slope of 4.16 ng mL−1 h. A power function provided a better description than a linear function to relate the saliva and serum cortisol. In conclusion, a circadian rhythm component was built in the PK/PD model of the 5‐HTP challenge test which could better improve the understanding of the stimulating effect on HPA with cortisol change. After the 5‐HTP challenge, saliva cortisol correlated well with serum cortisol and was predictable by a population PK‐PD model.


| INTRODUC TI ON
The hypothalamus-pituitary-adrenal axis (HPA) has been associated with the neurobiology of mood and anxiety disorder, etc. 1 A reliable and well-characterized pharmacological challenge test that quantitatively evaluates the function of central components of the HPA axis, would, therefore, be a useful tool to evaluate its role health and psychiatric disease. In addition, such challenge study could be helpful delineating endophenotypal characteristics of clinical psychiatric phenomena and guiding the development of innovative central nervous system (CNS) drugs along a rational path. 5-hydroxytryptophan (5-HTP) is converted from tryptophan, an essential amino acid, by tryptophan hydroxylase and it is further converted to serotonin. 5-HTP has been used in alternative medicine as a possibly effective aid in treating depression or fibromyalgia and its potential applications under research include insomnia, alcohol withdrawal, migraine, premenstrual syndrome, binge-eating related to obesity, attention deficit disorder, cerebellar ataxia, and muscle spasms in the mouth. 2-7 5-HTP is also reported as a challenge test to examine central serotonergic function, with cortisol and prolactin release used as a measure of response, as well as the excretion of the metabolite 5-hydroxy-indoleacetic acid. [8][9][10][11][12] In previous studies, we have demonstrated reproducible, concentration-dependent pharmacodynamic effects with acceptable variability associated with a serotonergic function test in healthy volunteers using 5-hydroxytryptophan (5-HTP). 8,9 Carbidopa and granisetron were co-administrated with 5-HTP. Carbidopa prevented the peripheral conversion of 5-HTP to 5-hydroxytryptamine (5-HT) which would preclude brain penetration, while granisetron limited serotonergic side-effects such as gastro-intestinal stimulation and vomiting without influencing the neuroendocrine response or 5-HTP pharmacokinetics. 9 The 5-HTP challenge test is used to quantify central serotonergic (5-HT) neurotransmission by elevating central 5-HT. The increase of 5-HT activates the HPA axis which then releases corticotrophin (CRH), adrenocorticotrophic hormone (ACTH), and cortisol step by step. Cortisol was tested as the neuroendocrine endpoint as a key and a downstream steroid hormone of the HPA axis which is involved in stress and different diseases. [13][14][15] In human plasma, the major fraction (about 70%) of cortisol is bound to corticosteroid-binding globulin (CBG), approx. 20% is bound to albumin and 10% is unbound. 16 Several observations have led to the conclusion that only the unbound cortisol is able to penetrate the intracellular compartment and that the CBG-cortisol complex has no direct hormonal activity. 17,18 Cortisol measurement in blood samples is easily interfered with stress. 19 The invasive blood sampling can cause an up-swing of cortisol serum concentration.
The up-swing artificially generates a high concentration and cannot reflect the true concentration. One way to avoid this artificially high concentration cause by blood draw may be using non-invasive measurement like salivary sampling. There are already cases and attempts to use saliva to the predict serum/plasma concentration of drugs in many therapeutic and research areas, such as antituberculosis drugs, anticonvulsants, antiepileptic drugs, psychobiological agents, etc. 20-23 Salivary sampling is a non-invasive patient-friendly method, which offers new possibilities for cortisol measurement since it can also be sampled when volunteers or patients are at home. Salivary cortisol concentration reflects the biologically active serum unbound cortisol level and is thus unaffected by elevations in CBG, which confuse the interpretation of serum cortisol levels. 24 As a result, another advantage of testing salivary sampling is that the distribution of cortisol from blood to saliva generally occurs by passive diffusion and different researches have shown that the salivary cortisol concentration correlates well with the serum-free cortisol concentration throughout the physiological concentration range. 9,24-27 Under normal physiological condition without a pharmacological challenge, the relationship between serum and saliva cortisol has already been studied with regression analysis. 28,29 However, after a challenge of 5-HTP, the higher range of cortisol in both serum and saliva should further be studied.
Another complicating factor of cortisol after the 5-HTP challenge is the fact that the concentration of cortisol follows circadian rhythm. Plasma concentration of cortisol reaches peak concentrations in the morning (6 AM to 10 AM) and trough concentration at night between 8 PM and 2 AM. [30][31][32][33] As a result, the change from baseline of cortisol after the 5-HTP challenge is the mixed additive effect of drug response and circadian rhythm. To better understand the effect of 5-HTP, a circadian rhythm factor should be peeling off from the total change after baseline.

What is known about this subject?
We have previously demonstrated reproducible, concentration-dependent pharmacodynamic effects with acceptable variability associated with a serotonergic function test in healthy volunteers using 5-hydroxytryptophan to guide the development of the novel compound that target central components of the HPA axis. The evaluation of the circadian rhythm effect of cortisol which is the biomarker of the challenge test and the possibility of using saliva cortisol as an alternative monitor metric will assist our understanding of this challenging test.

What this study adds?
This study retrospectively collected the data of three trials of the 5-HTP challenge test in healthy volunteers.
Population PK/PD modeling which chose both serum and saliva cortisol as observations was constructed incorporating the circadian rhythm of cortisol. This improved the understanding of the 5-HTP stimulating effect on the HPA axis and provided the possibility of applying the salivary sampling of cortisol as a monitor metric due to its less burdensome and better feasibility.
In the current study, we retrospectively collected data from three studies conducted in our research center within 5 years. The combination of data was based on the fact that all three trials were designed similarly so that the heterogeneity of these trials was small.
Data were pooled to enable us to utilize a population pharmacokinetic (PK)/ pharmacodynamic (PD) modeling approach for addressing the aforementioned issues. PK/PD modeling is an approach to characterize the concentration-time profile and the relationship between concentrations and effects using a mathematical model.

Model estimation can be based on both individuals and populations.
The assumption that all individual concentration-effect relationships can be described with the same structural model is based on the notion that the drug activates the same pharmacological system in all subjects (or systems for different responses). PK/PD modeling is performed by a non-linear mixed effect modeling approach which provides the estimates of the population average parameters (assuming that each individual can be described using the same struc-

| Clinical trial design
We retrospectively included three studies (CHDR0204, CHDR0612, and CHDR0716). The three trials were randomized, double-blind, double-dummy placebo-controlled, crossover trial, and were performed at CHDR. The combination of 5-HTP (200 mg), CBD (100 mg + 50 mg), and granisetron (2 mg) was orally administrated. Carbidopa was administrated to prevent peripheral carboxylation which can stabilize the PK of 5-HTP and granisetron was administrated as an antiemetic to reduce the systemic side-effects of 5-htp. 8,9 The sampling time started before the administration of 5-HTP and finished 5 to 9 hours after 5-HTP administration. The 5-HTP challenge trial designs scheme were separately described in previous publications and summarized in In total, 35 healthy male volunteers participated in these studies. Their blood and saliva samples were collected. The plasma of 5-HTP, total serum cortisol, and saliva cortisol concentrations were measured. Study medications and biochemical methodologies of PK and PD measurements can be found in previous publications as well. 8,12,36,37 The demographic data of subjects were also collected. Additionally, in two studies, subjects' corticosteroid-binding protein (CBG) concentrations were measured. Then free serum cortisol could be calculated later using the method of Coolens et al 26 The Coolens equation is based on the total serum cortisol and CBG concentrations, considering the affinity of cortisol for CBG and albumin as below: where U is the free serum cortisol concentration, G is the CBG, and T is the total serum cortisol. CBG was analyzed using a radioimmunoassay kit from the BioSource (Nivelles, Belgium) at the Xendo Drug Development BV, Groningen, The Netherlands.

| Population approach
The population approach using nonlinear mixed-effects models was performed using NONMEM 7.1.0. The method used was the Firstorder conditional estimation (FOCE). Parameters were estimated with possible inter-individual variability (IIV) in the followed statistical model. IIV was exponentially expressed using Equation 3: In the above equation, P ij represented the j th basic pharmacokinetic parameter of the i th individual. All the values of P ij were assumed to be log-normally distributed. P TV j was the typical population value of the j th parameter and η ij is the deviation of P ij from P TV j with a mean of 0, and an estimated variance of 2 j . Both proportional and additive error model were tested to describe the residual unexplained variance between the observed concentrations and predictions from the model. The combined residual error model which combines proportional and additive error structures was also tested. The residual error statistical model followed Observations including both PK and PD model, respectively. ε 1 and ε 2 represented random deviation between the predicted and observed concentration, with a zero mean and variances of 2 1 and 2 2 .

| PK/PD modeling
The population approach was applied to analyze both PK and PD data. The compartmental model was used to fit the pharmacokinetic profile of 5-HTP during the challenge test. One-, two-, and three- A two-step modeling approach was used. First, the PK model for 5-HTP was built to obtain the estimated PK parameters based on OFV and goodness of fit. The PK model was only built to optimally describe the PK profile. Second, the PK/PD model was built. Individual empirical Bayes' estimates were determined to describe the concentration profile and used in the subsequent PK/PD analyses.
To build the PD model of stimulated serum cortisol concentration, a baseline model of serum cortisol was first built to assess the circadian rhythm based on the data of the placebo group using a cosine function as Equation 5. 38 BSL 0 was an initial baseline value. AMP was the amplitude of the cosine term. n could be different values (such as 4, 8, 12, 16, 24, etc) and the final chosen value should be suggested by the model fitting procedure. BSL was the total apparent baseline.
Then, a sigmoid model was selected to model the drug effect of the 5-HTP challenge as Equation 6. There was no reported evidence that the circadian rhythm of cortisol is affected by stress or drugs. As a result, total plasma cortisol (as E in Equation 6) was calculated as the sum of the effect of 5-HTP challenge part and baseline level of cortisol. [38][39][40][41] where E max is the maximum stimulation effect of 5-HTP and EC50 is the 5-HTP concentration producing 50% of maximum stimulation. In the modeling process, a shift in the circadian rhythm was discovered visibly between-day variability. Accordingly, inter-occasion variability was included to describe the day to day differences of the individual baselines.
Linear and power functional relationships were used to predict the saliva cortisol based on serum cortisol which was presented by total serum cortisol or free serum cortisol separately.
β serves as a simple scaling factor. γ is called either the exponent or the power, which determines the function's rates of growth or decay and the function's overall shape and behavior. If γ equals 1, the relationship becomes linear. C sal represents saliva cortisol con- (3)

| Subject information
We included a total of 35 healthy male volunteers. The numbers of each type of observation in three trials are shown in Table 1   The PK parameters are presented with parameter estimate, relative standard error, and inter-individual variability in Table 2 (also refer to Supplement material S1). Inter-individual variability (IIV) was identified on the apparent clearance (CL/F) and absorption rate (Ka).

Only proportional residual error was included in the model. VPC
showed that most of the data fell within the 95% prediction interval and were symmetrically distributed around the median (Figure 3), which suggested that the final model adequately described the majority of the data. Diagnostic plots also supported that the model fitted 5-HTP PK data properly ( Figure S1). In diagnostic plots, different colors were marked for three studies, from which it was shown that the final model fitted observations equally acceptable in all studies.  (Figure 4 and Figure S2). In diagnostic plots, different colors were marked for three studies, from which it is shown that the final model fit observations equally acceptable in all studies.

| Relationship between serum and saliva cortisol
A power function (Equation 7) provided a better description than a linear function to relate saliva cortisol with serum cortisol.
Additionally, free serum cortisol was a better predictor for saliva cortisol than total serum cortisol. The parameters are presented with parameter estimate, relative standard error, and inter-individual variability in Table 2 (also refer to Supplement material S3). A VPC was performed to verify the model performance which showed that most of the data fell within the 95% prediction interval and were symmetrically distributed around the median. The VPC and diagnostic plots are shown in Figure 5 and Figure S3. In diagnostic plots, different colors were marked for different studies, from which it is shown that the final model fit observations equally acceptable in all studies. Only CHDR0607 and CHDR0712 included saliva cortisol data so that only data of these two trials were used to build the correlation model of cortisol in saliva and serum.

| D ISCUSS I ON
Our aims were to develop a population PK/PD model for the ef- The current model provided a PK model that can predict PK exposure in population and individual levels, so that it may serve as a tool   and no sex differences in cortisol responses were observed under physical stress. 51,52 Similarly, no sex differences in adrenocortical activity could be observed in studies exposing healthy subjects to mild psychosocial stress. [55][56][57][58] There was also a study reporting difference in the saliva cortisol level between male and female subjects. 59 These potential gender differences mentioned above may lead to difficulty in directly applying the current PK/PD model in the female subject. Further study with female should be recruited and studied.
In conclusion, the PK/PD model, including a cosine function with a trend served as a simplified model to describe part of the circadian rhythm, could describe and predict the total serum cortisol concentration for the proposed dose level in the 5-HTP challenge test, but limitations existed when extrapolating to higher dose levels. The relationship between saliva cortisol and serum cortisol was well characterized by a power function. The results provide a rationale to sample cortisol from saliva as an alternative of serum.

ACK N OWLED G EM ENT
The authors like to thank Dr Jan Freijer for his valuable advice during the research design, data analysis, and manuscript preparation.
Authors Contribution: Prof. K Burggraaf and Prof. W Zhao coordinated and supervised data collection and critically revised the manuscript. Dr Z Guan performed analysis and drafted the initial manuscript, Drs. G E Jacobs, J Van Pelt, and J M A Van Gerven collected the clinical data and reviewed the manuscript. All authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.

E TH I C S TATEM ENT
The study was approved by the Medical Ethics Committee of Leiden University Medical Centre and performed according to the Good Clinical Practice and International Conference on Harmonization guidelines. Further utilization of the data in later scientific research from these studies was noticed to all subjects in informed consents that were available per request.

CO N FLI C T O F I NTE R E S T S
All authors have completed the Unified Competing Interest form at http://www.icmje.org/coi_discl osure.pdf (available on request from the corresponding author) and declare no support from any organization for the submitted work, no financial relationships with any organizations that might have an interest in the submitted work in the previous 3 years and no other relationships or activities that could appear to have influenced the submitted work.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.