Bull Math Biol (2014) 76:136156

DOI 10.1007/s11538-013-9913-7
O R I G I N A L A RT I C L E

Modeling Endocrine Control of the PituitaryOvarian

Axis: Androgenic Influence and Chaotic Dynamics
Angelean O. Hendrix Claude L. Hughes
James F. Selgrade

Received: 4 January 2013 / Accepted: 8 October 2013 / Published online: 22 November 2013
Society for Mathematical Biology 2013

Abstract Mathematical models of the hypothalamus-pituitary-ovarian axis in women

were first developed by Schlosser and Selgrade in 1999, with subsequent models of
Harris-Clark et al. (Bull. Math. Biol. 65(1):157173, 2003) and Pasteur and Selgrade
(Understanding the dynamics of biological systems: lessons learned from integrative
systems biology, Springer, London, pp. 3858, 2011). These models produce periodic
in-silico representation of luteinizing hormone (LH), follicle stimulating hormone
(FSH), estradiol (E2), progesterone (P4), inhibin A (InhA), and inhibin B (InhB).
Polycystic ovarian syndrome (PCOS), a leading cause of cycle irregularities, is seen
as primarily a hyper-androgenic disorder. Therefore, including androgens into the
model is necessary to produce simulations relevant to women with PCOS. Because
testosterone (T) is the dominant female androgen, we focus our efforts on modeling pituitary feedback and inter-ovarian follicular growth properties as functions of
circulating total T levels. Optimized parameters simultaneously simulate LH, FSH,
E2, P4, InhA, and InhB levels of Welt et al. (J. Clin. Endocrinol. Metab. 84(1):105
111, 1999) and total T levels of Sinha-Hikim et al. (J. Clin. Endocrinol. Metab.
Support by NSF grants DMS-0920927 and DMS-1225607.
This material is based upon work supported by the National Science Foundation Graduate Research
Fellowship under Grant No. DGE-0750733.

A.O. Hendrix ( ) C.L. Hughes

Department of Mathematics, North Carolina State University, Raleigh, NC, USA
e-mail: aohendri@ncsu.edu
C.L. Hughes
Cardiovascular and Metabolic Unit, Quintiles, Inc., RTP, NC, USA
C.L. Hughes
Department of Obstetrics and Gynecology, Duke University Medical Center, Durham, NC, USA
J.F. Selgrade
Department of Mathematics and Biomathematics Program, North Carolina State University, Raleigh,
NC, USA

Modeling Endocrine Control of the PituitaryOvarian Axis

137

83(4):13121318, 1998). The resulting model is a system of 16 ordinary differential equations, with at least one stable periodic solution. Maciel et al. (J. Clin. Endocrinol. Metab. 89(11):53215327, 2004) hypothesized that retarded early follicle
growth resulting in stockpiling of preantral follicles contributes to PCOS etiology.
We present our investigations of this hypothesis and show that varying a follicular growth parameter produces preantral stockpiling and a period-doubling cascade
resulting in apparent chaotic menstrual cycle behavior. The new model may allow investigators to study possible interventions returning acyclic patients to regular cycles
and guide developments of individualized treatments for PCOS patients.
Keywords Modeling Androgens

1 Introduction
Periodic fluctuations in pituitary and ovarian hormones regulate the human female reproductive cycle. Gonadotropin-releasing hormone (GnRH) from the hypothalamus
prompts the pituitary to produce follicle stimulating hormone (FSH) and luteinizing
hormone (LH) which control ovarian follicular development and sex hormone secretion (Yen 1999; Hotchkiss and Knobil 1994). The ovaries secrete estradiol (E2),
progesterone (P4), inhibin A (InhA), inhibin B (InhB) and androgens which in turn
affect the synthesis and release of FSH and LH (Karch et al. 1973; Liu and Yen
1983) via positive and negative feedback relationships. Models for hormonal regulation of the menstrual cycle have been constructed using systems of ordinary differential equations where state variables represent serum hormone levels or different
stages of monthly ovarian development, e.g., Harris-Clark et al. (2003); Reinecke
and Deuflhard (2007), and Pasteur and Selgrade (2011). The model presented here
expands on the models of Harris-Clark et al. (2003) and Pasteur and Selgrade (2011)
by including the effects of the androgen testosterone (T) on the brain and on the
ovaries.
Pulses of gonadotropin-releasing hormone (GnRH) produced by the hypothalamus
on a time scale of minutes and hours cause pulses of FSH and LH to be produced by
the pituitary. Because the ovaries respond to average daily blood levels (Odell 1979),
our models track average daily concentrations of FSH and LH. Hence, we lump the
effects of the hypothalamus and the pituitary together and just consider the synthesis
and release of FSH and LH on the time scale of days. This simplification results in
models, which give good fit to the daily serum data of McLachlan et al. (1990), and
to the data of Welt et al. (1999), and which avoid the complication of multiple time
scales. These models exhibit novel features, e.g., two stable periodic solutionsone
ovulatory and the other non-ovulatory (Harris-Clark et al. 2003). The nonovulatory
cycle lacks a surge in LH and has a contraceptive level of E2. Model simulations
illustrate how exogenous P4 may be used to perturb the non-ovulatory cycle to the
ovulatory cycle and how exogenous E2 may be used to perturb the ovulatory cycle
to the nonovulatory cycle (see Harris-Clark et al. 2003). The latter is an example of
how environmental estrogens may disrupt a womans menstrual cycle. In addition,

Modeling Endocrine Control of the PituitaryOvarian Axis

137

83(4):13121318, 1998). The resulting model is a system of 16 ordinary differential equations, with at least one stable periodic solution. Maciel et al. (J. Clin. Endocrinol. Metab. 89(11):53215327, 2004) hypothesized that retarded early follicle
growth resulting in stockpiling of preantral follicles contributes to PCOS etiology.
We present our investigations of this hypothesis and show that varying a follicular growth parameter produces preantral stockpiling and a period-doubling cascade
resulting in apparent chaotic menstrual cycle behavior. The new model may allow investigators to study possible interventions returning acyclic patients to regular cycles
and guide developments of individualized treatments for PCOS patients.
Keywords Modeling Androgens

1 Introduction
Periodic fluctuations in pituitary and ovarian hormones regulate the human female reproductive cycle. Gonadotropin-releasing hormone (GnRH) from the hypothalamus
prompts the pituitary to produce follicle stimulating hormone (FSH) and luteinizing
hormone (LH) which control ovarian follicular development and sex hormone secretion (Yen 1999; Hotchkiss and Knobil 1994). The ovaries secrete estradiol (E2),
progesterone (P4), inhibin A (InhA), inhibin B (InhB) and androgens which in turn
affect the synthesis and release of FSH and LH (Karch et al. 1973; Liu and Yen
1983) via positive and negative feedback relationships. Models for hormonal regulation of the menstrual cycle have been constructed using systems of ordinary differential equations where state variables represent serum hormone levels or different
stages of monthly ovarian development, e.g., Harris-Clark et al. (2003); Reinecke
and Deuflhard (2007), and Pasteur and Selgrade (2011). The model presented here
expands on the models of Harris-Clark et al. (2003) and Pasteur and Selgrade (2011)
by including the effects of the androgen testosterone (T) on the brain and on the
ovaries.
Pulses of gonadotropin-releasing hormone (GnRH) produced by the hypothalamus
on a time scale of minutes and hours cause pulses of FSH and LH to be produced by
the pituitary. Because the ovaries respond to average daily blood levels (Odell 1979),
our models track average daily concentrations of FSH and LH. Hence, we lump the
effects of the hypothalamus and the pituitary together and just consider the synthesis
and release of FSH and LH on the time scale of days. This simplification results in
models, which give good fit to the daily serum data of McLachlan et al. (1990), and
to the data of Welt et al. (1999), and which avoid the complication of multiple time
scales. These models exhibit novel features, e.g., two stable periodic solutionsone
ovulatory and the other non-ovulatory (Harris-Clark et al. 2003). The nonovulatory
cycle lacks a surge in LH and has a contraceptive level of E2. Model simulations
illustrate how exogenous P4 may be used to perturb the non-ovulatory cycle to the
ovulatory cycle and how exogenous E2 may be used to perturb the ovulatory cycle
to the nonovulatory cycle (see Harris-Clark et al. 2003). The latter is an example of
how environmental estrogens may disrupt a womans menstrual cycle. In addition,

138

A.O. Hendrix et al.

Selgrade (2010) and Margolskee and Selgrade (2011) explain the bistable behavior
by analyzing bifurcation diagrams with respect to a parameter, which measures the
LH response of the pituitary to E2 priming. The review article by Vetharaniam et al.
(2010), describes many models of hormonal control of the female reproductive cycle
and discusses their strengths and limitations.
The development of the follicle which releases its ovum in a specific menstrual cycle begins at least 60 days before that cycle (Nussey and Whitehead 2001). To model
hormonal regulation of these early growing follicles, preantral, and early antral stages
must be included. Abnormal development during this period of early growth may result in cycle irregularities such as polycystic ovarian syndrome (PCOS), a leading
cause of female infertility (Alvarez-Blasco et al. 2006; Azziz et al. 2004; Yen 1999;
Franks et al. 2008). In fact, Maciel et al. (2004) reported a stockpiling of preantral
follicles in women with PCOS as compared to normally cycling women. The development of preantral follicles is gonadotropin independent but intraovarian factors
(Skinner 2005; Reddy et al. 2010; Maciel et al. 2004) influence the early growth and
the transition to the antral stage. Also, androgen receptors appear before FSH receptors (Rice et al. 2007), so in this study we consider the effects of testosterone (T) on
preantral and early antral follicles. Understanding how variations in hormone levels
and key ovarian growth parameters alter early follicular development may predict
disordered ovulation 3 months later.
To this end, we describe a mathematical model for menstrual cycle regulation,
which includes 12 distinct stages of ovarian development and 7 pituitary and ovarian
hormones. Our model builds on previous models (e.g., see Harris-Clark et al. 2003;
Pasteur 2008; Schlosser and Selgrade 2000; Selgrade and Schlosser 1999; Selgrade
et al. 2009) for endocrine control of the cycle but adds three new stages of follicular
development and includes the effects of T on the ovaries and on the pituitary. First, we
discuss the biological background and the model equations. Second, parameters are
identified using the data of Welt et al. (1999) and Sinha-Hikim et al. (1998). Third,
model simulations are presented and compared to data. Finally, we illustrate how
decreasing the transfer parameter between the preantral and early antral follicular
stages results in the stockpiling of preantral follicles and a significant alteration in
hormone levels. In fact, by varying this parameter we demonstrate a period-doubling
cascade of bifurcations that results in chaotic hormone fluctuations.

2 Biological Background and Model Development

2.1 Follicular Growth and Ovarian Stages
The normal adult menstrual cycle is approximately 28 days in duration and can be
divided into two phases: follicular (menstruation to ovulation) and luteal (ovulation
to menstruation). A cohort of primordial follicles is activated at least 60 days before
the LH surge promotes ovulation of the dominant follicle from that particular cohort
(Gougeon 1986). Upon activation, this cohort migrates to the medullary region of the
ovary, returning to the surface of the ovary in a process known as emergence as
maturation proceeds (Fihri et al. 2004). During this time, follicles are classified into

Modeling Endocrine Control of the PituitaryOvarian Axis

139

at least 4 categories: preantral, early antral, antral, and recruited. The preantral stage
accounts for the majority of the gonadotropin independent follicular growth. Thereafter, follicles begin to develop an antrum and sensitivity to FSH increases rapidly
(Gougeon 1986). A significant portion of these antral follicles degenerate through
atresia while larger follicles with additional FSH receptors enter a recruitment phase.
This development establishes the beginning the follicular phase of the monthly cycle. During this 14-day period, evidence suggests FSH must rise above a threshold
level for 5 days during which a select subgroup of follicles experiences rapid growth.
This span of time in the menstrual cycle is referred to as the FSH window (Fauser and
Van Heusden 1997). The dominant follicle (the one destined for ovulation) is selected
from this cohort by its acquisition of LH receptors. As serum FSH begins to decline,
nondominant follicles begin to degenerate. The dominant follicle reaches a diameter
of about 20 mm just prior to mid cycle and secretes large amounts of E2, which promotes pituitary LH production. Once LH reaches surge levels, the dominant follicle
releases its ovum and is transformed into the corpus luteum (CL). CL tissue develops
a yellow appearance, increases in mass until day nineteen of the cycle and produces
high levels of E2, P4, and InhA. Ovarian steroid production peaks at day twenty-one
and, in the absence of pregnancy, declines with the regression of the CL during the
remainder of the luteal phase of the menstrual cycle.
2.2 Intraovarian Growth Regulation
Histological samples of ovarian tissue can reveal diverse numbers of follicles from
1 mm to 20 mm in diameter during most of the menstrual cycle (Gougeon 1986).
As current ultrasound technology is best at identifying follicles greater than 2 mm
(Jonard and Dewailly 2004), it is common to consider this as the point of primary
follicle activation. A slight elevation in ovarian mass can be observed with the appearance of the dominant follicle (Lass 1999) with the peak in ovarian mass occurring at approximately day 19 of the cycle (Lass 1999) due to CL development. These
findings suggest the existence of a signaling mechanism that maintains the total mass
near steady state. Local factors identified in follicular fluid analysis are commonly
accepted as major contributors in this regulation. It is hypothesized that follicles in a
more advanced growth stage regulate activation of a new wave of follicles through the
use of insulin like growth factors (Lucy 2011), transforming growth factor (TGF1)
(Kuo et al. 2011), granulosa-theca cell factors (Orisaka et al. 2009), antimllerian
hormone (Franks et al. 2008), and androgens (Rice et al. 2007). Follicular fluid concentrations of growth factors are significant and change dynamically, but it seems implausible that corresponding changes in serum levels will be detectable. We therefore
focus our efforts on serum levels of T with the knowledge that androgen receptors
have been found in granulosa cells of developing follicles as small as 0.2 mm, before FSH receptors appear (Rice et al. 2007). This allows us to capture fluctuations in
follicular growth rates by integrating T levels into a subsystem of ovarian follicular
development that self-regulates its total mass.
2.3 Ovarian Modeling Technique
In our attempt to model the ovarian mass of preantral and small antral follicles, we
utilize mathematical theories of mass action kinetics. Often used to describe chemical

140

A.O. Hendrix et al.

Fig. 1 State variables

representing 12 stages of
follicular development are
shown. Follicle growth begins
with the PrA1 stage and
continues in a clockwise
direction for a complete cycle.
The pie chart indicates the
timing of the regulatory effects
of luteinizing hormone (LH) and
follicle stimulating hormone
(FSH) on follicular development

reactions where the total mass or volume of a system remains constant while the individual components change dynamically, the mass action approach allows us to reflect
total mass steady state regulation. The model equations are built to reflect an interdependent shift of mass through three stages of follicular maturity, represented by
the state variables PrA1 (preantral follicle 1), PrA2 (preantral follicle 2), and SmAn
(small antral follicle). This approach captures the intraovarian effects on early follicular development reviewed in Sect. 2.2. Transitions through these stages depend on
the masses of adjacent stages and available hormone levels (see Fig. 1). We assume
that proximity between follicles determines the magnitude of interfollicular signaling with the most significant effects coming from the subsequent maturity levels as
the follicles migrate toward the outer cortex of the ovary. This allows us to emulate
intraovarian signaling when follicular fluid levels of growth factors cannot be quantified. Differential equations are then mechanistically constructed for each of our state
variables.
The growth term in our first stage, PrA1, has a constant rate, m1 , of primordial
follicle recruitment as suggested by Gougeon (1986). We introduce a T dependent
transfer term that scales the product of the current mass of preantral follicles (PrA1)
with the mass of the next stage of maturity (PrA2). This term reflects the appearance
of androgen receptors before gonadotropin growth begins and contains an exponential, , introduced to increase the rate of stimulatory response to T (Rice et al. 2007).
The equation for PrA1 becomes:
d
PrA1 = m1 m2 T PrA1 PrA2
(1)
dt
The product PrA1 PrA2 quantifies the intraovarian growth factors and becomes
the growth term for PrA2. We then shift our mass from PrA2 utilizing an FSH dependent threshold term (e.g., see Zeleznik 2004)
FSH
m3
.

KmFSH + FSH

Modeling Endocrine Control of the PituitaryOvarian Axis

141

This gradual acquisition of receptors and saturation behavior is translated mathematically through a Hill function whose product with the current stage, PrA2, and
following stage, SmAn, defines our next transfer term.
FSH
d
PrA2 SmAn
PrA2 = m2 T PrA1 PrA2 m3
dt
KmFSH + FSH

(2)

As the mass of PrA2 is dependent on that of SmAn, there exists an indirect effect
of SmAn on our first stage, PrA1. This assumes that migration increases the distance
from SmAn follicles to the PrA1 follicles and, therefore, the interfollicular signaling
between the two decreases. The last stage of mass action dependence, SmAn, provides
the small antral follicle mass available for recruitment whose growth is partially regulated by follicles in PrA2, and indirectly by follicles in PrA1 through its inclusion
in the equation for PrA2.
FSH
d
SmAn = m3
PrA2 SmAn b FSH  SmAn RcF
dt
KmFSH + FSH

(3)

For the decay term, we assume that the rate of FSH receptor acquisition rapidly increases to a point directly proportional to the natural rise in follicular phase FSH at the
beginning of the follicular phase (Gougeon 1986), rather than a threshold response
used in the previous stage. Similarly, the transfer of mass from SmAn is affected by
the existing mass in the subsequent stage RcF whose growth is affected by SmAn
directly (see Eq. (4)) and by PrA1 and PrA2 indirectly (see Eqs. (2), (3)).
To reflect the increasing ovarian mass during the follicular and luteal phases of the
cycle, linear growth and decay terms are employed in 9 different stages which represent ovarian development during the monthly cycle (Fig. 1). RcF represents recruited
follicles with additional growth stimulated by FSH and transfer dependent on the
appearance of LH receptors. Our equation for RcF marks transition from the mass
action terms to linear compartmental terms as in Harris-Clark et al. (2003). These
compartmental terms permit total ovarian mass to increase as the ovaries approach
the time at which a dominant follicle is selected. DmF and OvF represent the dominant and ovulatory follicle. CL1 and CL2 portray the transition to the corpus luteum.
The luteal phase consists of the four stages Luti, i = 1, . . . , 4 representing the regression of the CL and conclusion of the current monthly cycle (Fig. 1). These 9 stages
correspond to the ovarian model of Harris-Clark et al. (2003). The growth and decay of these stages are influenced by the gonadotropins as indicated in the following
differential equations:


d
RcF = b FSH  SmAn RcF + c1 FSH c2 LH RcF
dt


d
DmF = c2 LH RcF + c3 LH c4 LH DmF
dt
d
OvF = c4 LH DmF c5 LH OvF
dt
d
CL1 = c5 LH OvF d1 CL1
dt
d
CL2 = d1 CL1 d2 CL2
dt

(4)
(5)
(6)
(7)
(8)

142

A.O. Hendrix et al.

d
(9)
Lut1 = d2 CL2 k1 Lut1
dt
d
Lut2 = k1 Lut1 k2 Lut2
(10)
dt
d
Lut3 = k2 Lut2 k3 Lut3
(11)
dt
d
Lut4 = k3 Lut3 k4 Lut4
(12)
dt
Clearance from the blood for the ovarian hormones is on a fast time scale (Baird
et al. 1969) as compared to ovarian development and clearance for the pituitary hormones. Hence, we assume that circulating levels of the ovarian hormones are maintained at a quasisteady state (Keener and Sneyd 2009) as did Bogumil et al. (1972).
Implementation of this approach results in using linear combinations of the 12 ovarian stages to obtain serum levels of E2, P 4, InhA, InhB, and T . The following five
auxiliary equations result:
Auxiliary Equations (A)
E2 = e0 + e1 DmF + e2 Lut4

(A1)

P 4 = p1 Lut3 + p2 Lut4

(A2)

InhA = h0 + h1 OvF + h2 Lut2 + h3 Lut3

InhB = j1 + j2 PrA2 + j3 SmAn + j4 RcF + j5 CL1

(A3)
j6

(A4)

T = t1 + t2 PrA2 + t3 SmAn + t4 RcF + t5 DmF

+ t6 OvF + t7 Lut1 + t8 Lut3

(A5)

The equation for circulating T was originally constructed to be a linear combination of ovarian stages (112). However, we assume that hormones are not produced
by newly activated follicles in PrA1 and parameter fitting to data revealed little to no
synthesis of T from the ovarian stages CL1, CL2, Lut2, and Lut4.
2.4 Androgenic Feedback on the Pituitary
Recent studies of T feedback suggest that it is an important component in the modulation of LH secretion by the pituitary in women. In human females, elevated T
is significantly correlated with elevated basal LH and dampened LH surge (Fauser
and Van Heusden 1997; Taylor et al. 1997). Animal models have shown that T is a
necessary component in priming the pituitary for GnRH induced LH synthesis, e.g.,
see Pielecka et al. (2006) and Yasin et al. (1996). Ovariectomized rats subjected to
exogenous GnRH pulses showed significantly greater LH mRNA response when
pretreated with T implants versus controls (Yasin et al. 1996). Treatment levels, similar to those documented at proestrous, demonstrate a 3-fold increase in the LH
mRNA response to GnRH when compared to untreated controls. Yasin et al. (1996)
noted that their results were independent of aromatization of T to estradiol. Hence,
we suggest that T may directly stimulate LH synthesis and will include such an effect
in our model.

Modeling Endocrine Control of the PituitaryOvarian Axis

143

Fig. 2 Ovarian control of the GnRH modulated pituitary synthesis and release of LH and FSH. Stimulatory and inhibitory effects are denoted by + and signs, respectively

2.5 Hypothalumus-Pituitary Modeling Technique

Schlosser and Selgrade (2000) proposed a pair of differential equations for each gonadotropin to model the hypothalamically controlled pituitary response to ovarian
hormones. These equations described the synthesis in the pituitary, the release into
the blood and clearance from the blood of LH and FSH; see Fig. 2. The coupling of
hypothalamic and pituitary actions allowed the system to predict serum levels of gonadotropins on a daily time scale consistent with published clinical data (McLachlan
et al. 1990 and Welt et al. 1999). In these differential equations, the state variables
RPLH and RPFSH represent the amounts of gonadotropins synthesized in the pituitary
via GnRH signaling and the state variables LH and FSH represent blood concentrations. The original model of Schlosser and Selgrade (2000) assumed a baseline LH
synthesis rate v0 independent of E2. Motivated by Yasin et al. (1996), for our model
we assume that this baseline rate depends on T; see Eq. (13).
a

E)
v0 T (t dT ) + v1 (KmaE2(td
a
d
LH +E2(tdE ) )
RPLH =
dt
(1 + P 4(tdP ) )

KiLH

(1 + cLHp P 4 )
RPLH
kLH
(1 + cLH e E2)
(1 + cLHp P 4 )
1
d
LH = kLH
RPLH rLH LH
dt
v
(1 + cLH e E2)
d
VFSH
RPFSH =
InhA(td
InhA
InhB )
dt
1 + ( KiFSHa ) ) + ( InhB(td
)
KiFSHb
(1 + cFSHp P 4)
RPFSH
(1 + cFSHe E2 )
(1 + cFSHp P 4)
d
1
FSH = kFSH
RPFSH rFSH FSH
dt
v
(1 + cFSHe E2 )
kF SH

(13)
(14)

(15)
(16)

144

A.O. Hendrix et al.

Through changes in GnRH pulse frequency and amplitude, LH exhibits a biphasic

response to E2 (Liu and Yen 1983), so to account for this the model assumes that the
effect of E2 on LH synthesis is different than the effect on LH release. E2 inhibits
release (see the denominator of the second term of Eq. (13)) but at high levels E2
promotes synthesis (see the numerator of the first term of (13)). On the other hand,
P4 inhibits GnRH modulated LH synthesis but promotes its release from the pituitary.
The release term in (13) appears as a growth term in (14), where it is divided by blood
volume v. Equations (15) and (16) for FSH are similar except the synthesis term has
constant growth and inhibition due to InhA and InhB. Since hormone synthesis is a
biochemical process more complicated than hormone release, we allow for discrete
time-delays, dE , dP , dT , dInhA , and dInhB , in the synthesis terms.
2.6 Parameter Identification and Periodic Solution
Parameter estimation began with the best-fit parameters reported by Pasteur (2008).
Nelder-Meade algorithm with least squares approximation optimized final parameters
against data for normal mean serum levels for E2, P4, LH, FSH, InhA, and InhB as
published by Welt et al. (1999) and data for normal serum Total T levels as reported
in Sinha-Hikim et al. (1998). Supplementation of the Welt data set was necessary as
a single data set for all seven serum hormones was not available at the time of this
publication. The original data reported by Sinha-Hikim et al. (1998) included nine
data points over 26 days. Therefore, linear extrapolation was used to generate 28
data points (see Fig. 5) for optimal parameter estimation. For comparison purposes
we converted the Sinha-Hikim data from nmol/L to ng/dL. This conversion takes the
range of values from (0.7, 1.6) to (20, 47). The best-fit parameter set is detailed in
the Appendix, Tables 1, 2, 3 and 4.
The discrete time-delays present in the model (dT , dE , dP , dInhA , and dInhB ) necessitate the use of a delay differential equation solver DDE23 by Shampine and
Thompson available through MatLab 2010a (2012) to numerically approximate a
solution. Day 1 values from Welt et al. (1999) served as initial conditions for LH
and FSH equations. Numerical simulations with the best-fit parameter set exhibit an
asymptotically stable periodic solution of period 29 days (Fig. 6) consistent with reports of average cycle lengths between 26 and 32 days (Gougeon 1986). Once this
stable periodic solution was identified, the remainder of the initial conditions were
determined by the values from the last day of simulation. Rounded to four significant
digits, our initial condition vector is detailed in the Appendix, Table 5.

3 Results
3.1 Numerical Simulations
Given the parameters listed in Tables 1 through 4, Figs. 34 represent serum concentrations of LH, FSH, E2, P4, InhA, and InhB for 58 days as predicted by the model,
with daily data for mean serum levels from Welt et al. (1999) for comparison (the 28
data values reported in Welt et al. 1999 are repeated here to obtain 58 data values).

Modeling Endocrine Control of the PituitaryOvarian Axis

145

Fig. 3 Two 29 day cycles for (a) LH data from Welt et al. () and (b) FSH data from Welt et al. () are
presented with current model simulations (solid curves)

Fig. 4 Two 29 day cycles for (a) E2 data from Welt et al. () , (b) P4 data from Welt et al. (), (c) InhA
data from Welt et al. (), (d) InhB data from Welt et al. () are presented with model simulations (solid
curves)

Two complete 29 day cycles are presented for each hormone to support the stability
conclusion.
One can observe the qualitative similarities to clinical observations. Serum T concentrations for a 29 day simulation follow in Fig. 5. Approximated T levels increase
from a day 3 level of 30 ng/dL to a maximum level of 44 ng/dL on day 11 of the

146

A.O. Hendrix et al.

Fig. 5 The nine T data points () from Sinha-Hikim et al. (1998) are graphed with linear interpolation
( ) and current model simulation (solid curve)

cycle. After ovulation, T slowly declines, plateauing from day 17 to 22 at approximately 34 ng/dL. At the end of the luteal phase, levels continue to decline before a
slight rebound in circulating levels is observed during the luteal to follicular transition. Figure 5 presents data extracted from Sinha-Hikim et al. (1998) against model
predictions for comparison.
3.2 Modeling Predictions Regarding Stockpiling of Follicles
Irregularities in cycle length are commonly seen in patients with PCOS. As a common
cause of infertility, PCOS affects up to 10 % of reproductive age women and significantly correlates with increased risk of Type II diabetes and its associated morbidity
(Alvarez-Blasco et al. 2006; Panidis et al. 2006; Azziz et al. 2004). Common phenotypes of PCOS also include elevated androgens and the appearance of polycystic
ovaries on ultrasound (Azziz et al. 2009), both significant components in the etiology of PCOS (Ropelato et al. 2009; Medina and Nestler 1998; Hillier and Ross 1979).
Histological studies of tissue samples taken from clinically diagnosed PCOS patients,
recently reported by Maciel et al. (2004), suggests a stockpiling of preantral follicles when compared to controls. Their findings show a significantly (P = 0.001)
increased number of follicles comprised of an oocyte and a single layer of cuboidal
granulosa cells, i.e., primary follicles. It is hypothesized that the increase in primary
follicle numbers is due to a longer growth pattern during this stage, represented by
PrA1 in our model. We investigate m2 as a possible model parameter to test this hypothesis. Because m2 is measurement of mass transferred out of the primary stage,
decreasing m2 should delay primary follicle maturation, permitting additional growth
of primary follicles.
For comparison, we begin by presenting LH concentrations from the stable periodic solution that best fits data from Welt and Sinha-Hikim in Fig. 6, i.e., the solution

Modeling Endocrine Control of the PituitaryOvarian Axis

147

Fig. 6 (a) Simulated serum LH IU/L over 7 months with reference lines for 75 % and 50 % of mean Welt
surge levels (b) First 3 stages of follicular development, PrA1, PrA2, and SmAn

with parameters listed in Tables 1 through 4. This simulation is graphed for 7 months
showing a stable solution having an average cycle length of approximately 29 days,
consistent with reports from Baerwald et al. (2003) and Gougeon (1986). We assume
ovulation coincides with the LH surge given a follicle of sufficient size is available for
rupture. Hence, we present Fig. 6 to depict 7 ovulations and the first 3 ovarian stages
for comparison with Figs. 7, 8, and 9, where the parameter m2 has decreased. As the
exact surge level of LH necessary to induce ovulation is unique to each woman, we
identify (by thick horizontal lines in Fig. 6) LH levels at 75 % and 50 % of the mean
maximum LH level reported by Welt et al. (1999) for possible threshold references.
While the ovarian mass values for PrA1 through SmAn (Fig. 6) are unitless, we note
that the first preantral follicle mass peaks at approximately 20 units before transferring to the second androgen dependent preantral follicle mass. Follicular mass during
this time is presented to demonstrate the interaction between the mass of developing
follicles and ovulation assuming that surge levels of LH are indicative of the existence
and timing of ovulation.
Figure 7 demonstrates the effect of reducing m2 by approximately 50 % of the
best fit value in Table 3. Analysis of the resulting behavior, over 7 months, reveals
an LH surge exceeding 150 IU/L followed by a surge approximately 20 IU/L lower.
While each peak surge occurs monthly, the pattern of alternating surge levels takes
over 2 months to repeat. Similarly, the first preantral follicle mass begins to oscillate
with a maximum mass of 30 units that results in lower mass transfer to the second
preantral mass. In the normal case, our cycle length and the time for LH levels to
return to the peak level of the previous cycle were the same. Reducing the transfer
rate from PrA1 approximately doubles the time between peaks of the same magnitude, a phenomenon know as a period-doubling event (the period of the solution to
our model equations is now 64 days). Identification of this behavior is mathematically significant for systems of this size and complexity. Although period-doubling
often occurs in systems of nonlinear equations, it has rarely been demonstrated in a
physiological model, which predicts data in the literature. As demonstrated in Fig. 8,

148

A.O. Hendrix et al.

Fig. 7 (a) Serum LH level results after a 50 % reduction in parameter m2 from best-fit value in Table 3
(b) Note the increase in PrA1 from Fig. 6

Fig. 8 (a) Serum LH level results after a 70 % reduction parameter m2 best-fit value in Table 3 (b) First
three stages of follicular development

reducing m2 to 30 % of the best-fit value in Table 3, results in LH surges of four

distinct values. In this simulation, the pattern of peak variation now repeats every
5.5 months with an average time between surges of approximately 40 days (another
period-doubling has occurred). Examination of the resulting follicular mass reveals
a distinct pattern of elevated PrA1, stockpiling of preantral follicles that completes
its transfer to subsequent stages over a period of approximately 80 days. This suggests correlation between preantral growth and irregular menstrual cycles consistent
with the stockpiling hypothesis of Maciel et al. (2004). Mathematically, this behavior indicates the existence of a period-doubling cascade of bifurcations as m2 is
decreased.

Modeling Endocrine Control of the PituitaryOvarian Axis

149

Fig. 9 Chaotic serum LH levels observed over 3 years as m2 is reduced to 25 % of the best-fit value
in Table 3. If ovulation occurs for an LH surge over 90 mg/L (75 % of the normal LH surge) then 5
ovulations occur per year. If ovulation occurs for an LH surge over 60 mg/L, then 8 occur per year

Complex dynamical systems with period-doubling cascades are often associated

with chaotic behavior (Sander and Yorke 2012). Investigating this phenomenon motivated numerical experiments with additional decreases in m2 to identify behavior
consistent with chaotic attractors. Reducing m2 by an additional 5 % causes the disappearance of the stable 4-cycle shown in Fig. 8 and the appearance of an apparent
chaotic attractor as demonstrated in Fig. 9. Also, a common characteristic of chaotic
behavior, i.e., the existence of a stable 3-cycle, appears in a very narrow range of
m2 values near 26 % of the value reported in Table 3. (Due to the lack of biological
relevance of the 3-cycle solution, the results are not presented, merely noted.) If one
assumes an LH threshold at 75 % of the reported mean, as demonstrated by the top
horizontal line in Fig. 9, then the solution presented would ovulate approximately 5
times per year given the availability of a dominant follicle at the time of LH surge.
Reducing the threshold assumption to 50 % increases the frequency of ovulation to
an average 8 cycles annually over the 3-year window presented. These frequencies
are consistent with a clinical diagnosis of oligomenorrhea, infrequent menstruation
with 4 to 8 menstrual cycles per year, a primary phenotype of PCOS. The appearance
of low amplitude peaks that do not exceed threshold values observed in Fig. 9 may
be consistent with nonovulatory LH surges discussed in Baerwald et al. (2012).

4 Summary and Discussion

This study presents a nonlinear mechanistic model for the endocrine regulation of the
human menstrual cycle, which consists of 16 delay differential equations and 5 auxiliary equations, with 71 parameters identified using data from the literature. Model
simulations predict serum levels of LH, FSH, E2, P4, InhA, InhB, and T consistent
with biological data (Welt et al. 1999; Sinha-Hikim et al. 1998). The asymptotically

150

A.O. Hendrix et al.

stable periodic solution resulting from the best-fit parameter set (Tables 1 through 4)
has a period of 29 days, consistent with current literature. Moreover, increases in
qualitative accuracy are obtained through the use of mass action kinetic theory to describe preantral follicular through early antral follicular stage growth. We believe this
is the first model of its type to be able to predict serum T levels in a mathematical
context that allows analysis of bifurcations and stable solutions of various periods.
A benefit of this approach is the identification of stable solutions that may resemble hormonal profiles consistent with some of the 16 PCOS phenotypes as defined
in the Rotterdam consensus document (Consensus workshop group 2004). While it
is generally agreed that T plays an important role in PCOS, only 8 of the 16 PCOS
phenotypes present with elevated androgens (Consensus workshop group 2004). It is
hypothesized that extended preantral follicular development could play an important
role in explaining the appearance of polycystic ovaries (Maciel et al. 2004), a significant criterion in normal patients. Our approach of reflecting intraovarian follicular
growth regulation systemically allows researchers to investigate effects of preantral
growth abnormalities on menstrual cycle behavior. In our investigation, observations
from increasing preantral follicle growth duration support the hypothesis presented
by Maciel et al. (2004) that a stockpiling of immature follicles may be significant in
the etiology of PCOS. Figures 6 through 9 reinforce this conclusion by demonstrating
that the reduction of m2 from its best-fit value in Table 3, which delays the transfer
from preantral growth to androgen dependent growth, results in a visible stockpiling of preantral follicular mass (Figs. 7 and 8) and in irregular cyclicity (Fig. 9).
Furthermore, the identification of a period-doubling cascade of bifurcations leading to apparent chaotic behavior (Fig. 9) may actually increase accuracy in representing clinical findings of women with PCOS and/or oligomenorrhea over previous
models. A recent paper by Derry and Derry (2010) presents a time series analysis of
longitudinal menstrual cycle length data for 40 women over 20 to 40 years. They concluded that the human menstrual cycle is the result of chaos in a nonlinear dynamical
system (Derry and Derry 2010) with only 5 degrees of freedom. They further referenced specifically the model of Harris-Clark et al. (2003) with the assertion that any
model producing only perfectly periodic menstrual cycles is, at best, incomplete
(Derry and Derry 2010). It is our belief that our current approach meets their criteria
for a model of the human menstrual cycle that displays biologically relevant random
behavior independent of external interference. It also shows that follicular dynamics cannot be abandoned in the quest for accuracy, rather it is the interplay between
intraovarian mass and endocrine regulation that explains this behavior.
We anticipate the emerging collaborations with clinical endocrinologists and experimentalists will soon provide essential data necessary to refine additional parameter sets that can simulate additional PCOS phenotypes. Preliminary investigations
include manipulating serum T levels, through altering t1 , as a reflection of excess
adrenal androgen production, and t3 , as a reflection of increased insulin dependent
theca cell T synthesis. These studies support the consideration of androgenic feedback on LH synthesis and encourage further examination of its effect on FSH synthesis. We believe these investigations will lend further understanding to PCOS phenotypes and adrenal disorders that present with elevated androgens.

Modeling Endocrine Control of the PituitaryOvarian Axis

151

The basic assumption that the hypothalamuspituitaryovarian axis is in itself a

closed autonomous dynamical system presenting stable yet possibly chaotic behavior
has led us to the development and presentation of our current model for endocrine
regulation of female reproduction. Numerical approximations of stable solutions can
be quickly computed as compactness was a major consideration in model development. These attributes make it a prime candidate for in silico investigations of cycle
regularity. We have shown that modeling biological mechanisms, while considering
clinical challenges and computational costs can lead to mathematically rich dynamics that present new model-derived insights that may guide future development of
innovative individualized therapeutic interventions for women with PCOS.

Appendix: Parameters and Inital Conditions

Table 1 LH subsystem
parameters

Table 2 FSH subsystem

parameters

Parameter

Value

Unit

v0LH

33.3

/day

v1LH

160.75

/day

KiLH

13.6

L/nmol

KmLH

47.33

ng/L

kLH

19.99

1/day

cLH P

0.98

L/nmol

cLH E

0.9

L/ng

dP

days

dT

days

dE

days

0.92

dimensionless

dimensionless

6.07

dimensionless

rLH

14

1/day

2.5

liters

Parameter

Value

Unit

vFSH

283.99

/day

KiFSH a

6.38

/day

KiFSH b

3000

L/nmol

kFSH

3.41

1/day

cFSH p

1.26

L/nmol

cFSH e

0.16

L/ng

0.84

dimensionless

dInhA

days

dInhB

days

rFSH

8.21

1/day

152

A.O. Hendrix et al.

Table 3 Ovarian subsystem parameters

Parameter

Value

Unit

0.7100

dimensionless

0.6928

dimensionless

0.0002

dimensionless

0.952

dimensionless

0.0189

L/day

c1

0.0871

L/g

c2

0.1251

1/day

c3

0.0533

1/day

c4

0.0371

1/day

c5

0.4813

1/day

d1

0.7053

1/day

Parameter

Value

Unit

d2

0.6488

1/day

k1

0.7089

1/day

k2

0.8712

1/day

k3

1.038

1/day

k4

1.052

1/day

m1

0.9274

1/day

1.338 103

1/day

m2

1.162

1/day

m3

0.15

1/day

KmFSH

7.533

1/day

1/day

0.451

1/day

Table 4 Auxiliary equations parameter set

Parameter

Value

Unit

j3

3.738

1/L

j4

3.411

1/L

j5

1/L

1/kL

j6

0.0012

1/L

U/L

t1

21.92

ng/dL

0.0251

(nmol/L)/g

t2

0.3721

1/dL

0.06315

U/L

t3

0.2961

1/dL

h3

0.1584

U/L

t4

0.4538

1/dL

p1

0.2548

(nmol/L)/g

t5

0.0411

1/dL

p2

0.1276

(nmol/L)/g

t6

0.7029

1/dL

j1

27.21

pg/L

t7

0.0459

1/dL

j2

1.885

1/L

t8

0.1998

1/dL

RPLH

1071

Parameter

Value

Unit

e0

38.25

ng/L

e1

2.3

1/kL

e2

2.724

h0

0.0525

h1
h2

Table 5 Initial conditions

State variable

Initial condition

RPLH

1071

LH

12.0

RPFSH

155.6

FSH

11.40

PrA1

0.6112

PrA2

15.11

Modeling Endocrine Control of the PituitaryOvarian Axis

Table 5 (Continued)

State variable

153
Initial condition

SmAn

12.36

RcF

0.1511

DmF

1.337

OvF

1.168

CL1

1.003

CL2

1.865

Lut1

3.265

Lut2

4.071

Lut3

6.055

Lut4

9.029

References
Alvarez-Blasco, F., Botella-Carretero, J. I., San Milln, J. L., & Escobar-Morreale, H. F. (2006). Prevalence and characteristics of the polycystic ovary syndrome in overweight and obese women. Arch.
Intern. Med., 166(19), 20812086. http://www.ncbi.nlm.nih.gov/pubmed/17060537. doi:10.1001/
archinte.166.19.2081.
Azziz, R., Woods, K. S., Reyna, R., Key, T. J., Knochenhauer, E. S., & Yildiz, B. O. (2004).
The prevalence and features of the polycystic ovary syndrome in an unselected population. J. Clin. Endocrinol. Metab., 89(6), 27452749. http://jcem.endojournals.org/cgi/content/
abstract/89/6/2745. doi:10.1210/jc.2003-032046.
Azziz, R., Carmina, E., Dewailly, D., Diamanti-Kandarakis, E., Escobar-Morreale, H. F., Futterweit, W.,
Janssen, O. E., Legro, R. S., Norman, R. J., Taylor, A. E., & Witchel, S. F. (2009). The androgen excess and PCOS society criteria for the polycystic ovary syndrome: the complete task force report.
Fertil. Steril., 91(2), 456488. http://www.sciencedirect.com/science/article/B6T6K-4TRR8Y51/2/1838ea8df3b5f094e2d10e1b791a881b.
Baerwald, A. R., Adams, G. P., & Pierson, R. a. (2003). A new model for ovarian
follicular development during the human menstrual cycle. Fertil. Steril., 80(1), 116
122. http://linkinghub.elsevier.com/retrieve/pii/S0015028203005442. http://www.ncbi.nlm.nih.gov/
pubmed/12849812. doi:10.1016/S0015-0282(03)00544-2.
Baerwald, A. R., Adams, G. P., & Pierson, R. A. (2012). Ovarian antral folliculogenesis during
the human menstrual cycle: a review. Hum. Reprod. Updat., 18(1), 7391. http://www.ncbi.nlm.
nih.gov/pubmed/22068695. doi:10.1093/humupd/dmr039.
Baird, R. T., Horton, R., Longcope, C., & Tait, J. F. (1969). Steroid dynamics under steady-state conditions.
Recent Prog. Horm. Res., 25, 611664.
Bogumil, R. J., Ferin, M., Rootenberg, J., Speroff, L., & Vande Wiele, R. L. (1972). Mathematical studies
of the human menstrual cycle. I. Formulation of a mathematical model. J. Clin. Endocrinol. Metab.,
35(1), 126143. http://www.ncbi.nlm.nih.gov/pubmed/5064156.
Derry, G., & Derry, P. (2010). Characterization of chaotic dynamics in the human menstrual cycle. Nonlinear Biomed. Phys., 4(1), 5. http://www.nonlinearbiomedphys.com/content/4/1/5. doi:10.
1186/1753-4631-4-5.
Fauser, B. C., & Van Heusden, A. M. (1997). Manipulation of human ovarian function: physiological concepts and clinical consequences. Endocr. Rev., 18(1), 71106. http://publishing.
eur.nl/ir/darenet/asset/8648/9034787.pdf http://www.ncbi.nlm.nih.gov/pubmed/9034787.
Fihri, A. F., Hajji, K. H., & Lakhdissi, H. (2004). Histological exploration of follicular population of the
Moroccan bovine (Oulmes-Zaers) breed. Afr. J. Biotechnol., 3, 294298.
Franks, S., Stark, J., & Hardy, K. (2008). Follicle dynamics and anovulation in polycystic ovary syndrome. Hum. Reprod. Updat., 14(4), 112. http://humupd.oxfordjournals.org/content/early/2008/
05/22/humupd.dmn015.full.pdf+html.
Gougeon, A. (1986). Dynamics of follicular growth in the human: a model from preliminary results. Hum.
Reprod. (Oxford, England), 1(2), 8187. http://www.ncbi.nlm.nih.gov/pubmed/3558758.

154

A.O. Hendrix et al.

Consensus workshop group, T.R.E.S.P. (2004). 2004 Rotterdam 1: revised 2003 consensus on diagnostic
criteria and long-term health risks related to polycystic ovary syndrome (PCOS). Hum. Reprod.,
19(1), 4147.
Harris-Clark, L., Schlosser, P. M., & Selgrade, J. F. (2003). Multiple stable periodic solutions in a
model for hormonal control of the menstrual cycle. Bull. Math. Biol., 65(1), 157173. http://
www.ncbi.nlm.nih.gov/pubmed/12597121. doi:10.1006/bulm.2002.0326.
Hillier, S. G., & Ross, G. T. (1979). Effects of exogenous testosterone on ovarian weight, follicular morphology and intraovarian progesterone concentration in estrogen-primed hypophysectomized immature female rats. Biol. Reprod., 20, 261268.
Hotchkiss, J., & Knobil, E. (1994). The menstrual cycle and its neuroendocrine control. In E. Knobil & J.
Neill (Eds.), The physiology of reproduction (2nd ed., pp. 711750). New York: Raven Press.
Jonard, S., & Dewailly, D. (2004). The follicular excess in polycystic ovaries, due to intra-ovarian hyperandrogenism, may be the main culprit for the follicular arrest. Hum. Reprod. Updat., 10(2), 107117.
http://humupd.oxfordjournals.org/cgi/content/abstract/10/2/107. doi:10.1093/humupd/dmh010.
Karch, F., Dierschke, D., Weick, R., Yamaji, T., Hotchkiss, J., & Knobil, E. (1973). Positive and negative
feedback control by estrogen of luteinizing hormone secretion in the rhesus monkey. Endocrinology,
92(3), 799804. http://endo.endojournals.org/cgi/content/abstract/92/3/799. doi:10.1210/endo-923-799.
Keener, J., & Sneyd, J. (2009). Mathematical physiology I: cellular physiology (2nd ed.). New York:
Springer.
Kuo, S. W., Ke, F. C., Chang, G. D., Lee, M. T., & Hwang, J. J. (2011). Potential role of
follicle-stimulating hormone (FSH) and transforming growth factor (TGF1) in the regulation of
ovarian angiogenesis. J. Cell. Physiol., 226(6), 16081619. http://www.ncbi.nlm.nih.gov/pubmed/
20945379. doi:10.1002/jcp.22491.
Lass, A. (1999). The role of ovarian volume in reproductive medicine. Hum. Reprod. Updat.,
5(3), 256266. http://humupd.oxfordjournals.org/cgi/content/abstract/5/3/256. doi:10.1093/humupd/
5.3.256.
Liu, J. H., & Yen, S. S. (1983). Induction of midcycle surge by ovarian steroids in women: a critical
evaluation. J. Clin. Endocrinol., 57(4), 797802.
Lucy, M. C. (2011). Growth hormone regulation of follicular growth. Reprod. Fertil. Dev., 24(1), 1928.
http://www.ncbi.nlm.nih.gov/pubmed/22394714. doi:10.1071/RD11903.
Maciel, G. A. R., Baracat, E. C., Benda, J. A., Markham, S. M., Hensinger, K., Chang, R. J., &
Erickson, G. F. (2004). Stockpiling of transitional and classic primary follicles in ovaries of
women with polycystic ovary syndrome. J. Clin. Endocrinol. Metab., 89(11), 53215327. http://
www.ncbi.nlm.nih.gov/pubmed/15531477. doi:10.1210/jc.2004-0643.
Margolskee, A., & Selgrade, J. F. (2011). Dynamics and bifurcation of a model for hormonal control
of the menstrual cycle with inhibin delay. Math. Biosci., 234(2), 95107. http://www.ncbi.nlm.
nih.gov/pubmed/21939671. doi:10.1016/j.mbs.2011.09.001.
Matlab (2012): version 7.14 (R2012a). Natick, Massachusetts: The MathWorks Inc.
McLachlan, R., Cohen, N., Dahl, K., Bremner, W., & Soules, M. (1990). Serum inhibin levels during
the periovulatory interval in normal women: relationships with sex steroid and gonadotrophin levels.
Clin. Endocrinol., 1(32), 3948.
Medina, F., & Nestler, J. (1998). Insulin stimulates testosterone biosynthesis by human theca cells from
women with polycystic ovary syndrome by activating its own receptor and using inositoglycan mediators as the signal transduction system. J. Clin. Endocrinol. Metab., 83(6), 20012005.
Nussey, S. S., & Whitehead, S. A. (2001). Endocrinology: an integrated approach. London: Taylor &
Francis.
Odell, W. (1979). The reproductive system in women. In L. DeGroot (Ed.), Endocrinology (pp. 1393
1400). New York: Grune & Stratton.
Orisaka, M., Tajima, K., Tsang, B. K., & Kotsuji, F. (2009). Oocyte-granulosa-theca cell interactions during preantral follicular development. J. Ovarian Res., 2(1), 9. http://www.
pubmedcentral.nih.gov/articlerender.fcgi?artid=2715405&tool=pmcentrez&rendertype=abstract.
doi:10.1186/1757-2215-2-9.
Panidis, D., Macut, D., Farmakiotis, D., Rousso, D., Kourtis, A., Katsikis, I., Spanos, N., Petakov,
M., Bjekic, J., & Damjanovic, S. (2006). Indices of insulin sensitivity, beta cell function and serum proinsulin levels in the polycystic ovary syndrome. Eur. J. Obstet. Gynecol. Reprod. Biol., 127(1), 99105. http://www.ncbi.nlm.nih.gov/pubmed/16460870. doi:10.
1016/j.ejogrb.2005.12.016.

Modeling Endocrine Control of the PituitaryOvarian Axis

155

Pasteur, R. (2008). A multiple-inhibin model for the human menstrual cycle. Ph.D. thesis, North Carolina
State University. http://www.lib.ncsu.edu/resolver/1840.16/5587.
Pasteur, R., & Selgrade, J. (2011). A deterministic, mathematical model for hormonal control of the menstrual cycle. In W. Dubitzky, J. Soutate, & H. Fuss (Eds.), Understanding the dynamics of biological
systems: lessons learned from integrative systems biology (pp. 3858). London: Springer.
Pielecka, J., Quaynor, S. D., & Moenter, S. M. (2006). Androgens increase gonadotropin-releasing hormone neuron firing activity in females and interfere with progesterone negative feedback. Endocrinology, 147(3), 14741479. http://www.ncbi.nlm.nih.gov/pubmed/16339200. doi:10.1210/en.20051029.
Reddy, P., Zheng, W., & Liu, K. (2010). Mechanisms maintaining the dormancy and survival
of mammalian primordial follicles. Trends Endocrinol. Metab., 21(2), 96103. http://dx.doi.
org/10.1016/j.tem.2009.10.001. doi:10.1016/j.tem.2009.10.001.
Reinecke, I., & Deuflhard, P. (2007). A complex mathematical model of the human menstrual cycle. J. Theoret. Biol., 247(2), 303330. http://www.ncbi.nlm.nih.gov/pubmed/17448501.
doi:10.1016/j.jtbi.2007.03.011.
Rice, S., Ojha, K., Whitehead, S., & Mason, H. (2007). Stage-specific expression of androgen
receptor, follicle-stimulating hormone receptor, and anti-Mllerian hormone type II receptor
in single, isolated, human preantral follicles: relevance to polycystic ovaries. J. Clin. Endocrinol. Metab., 92(3), 10341040. http://jcem.endojournals.org/content/92/3/1034.abstract. doi:10.
1210/jc.2006-1697.
Ropelato, M. G., Garca Rudaz, M. C., Escobar, M. E., Bengolea, S. V., Calcagno, M. L., Veldhuis, J. D., & Barontini, M. (2009). Acute effects of testosterone infusion on the serum luteinizing hormone profile in eumenorrheic and polycystic ovary syndrome adolescents. J. Clin. Endocrinol. Metab., 94(9), 36023610. http://jcem.endojournals.org/content/94/9/3602.abstract. doi:10.
1210/jc.2009-0402.
Sander, E., & Yorke, J. (2012). Connecting period-doubling cascades to chaos.
Int.
J.
Bifurc.
Chaos,
22,
129.
http://www.worldscientific.com/doi/abs/10.1142/
S0218127412500228.
Schlosser, P., & Selgrade, J. (2000). A model of gonadotropin regulation during the menstrual cycle in
women: qualitative features. Environ. Health Perspect., 108(5), 873881.
Selgrade, J. F. (2010). Bifurcation analysis of a model for hormonal regulation of the menstrual cycle. Math. Biosci., 225(2), 108114. http://www.ncbi.nlm.nih.gov/pubmed/20219482.
doi:10.1016/j.mbs.2010.02.004.
Selgrade, J. F., & Schlosser, P. M. (1999). A model for the production of ovarian hormones during the menstrual cycle. Fields Inst. Commun., 21, 429446. http://www4.ncsu.edu/~selgrade/FIC_
Selgrade.pdf.
Selgrade, J. F., Harris, L. A., & Pasteur, R. D. (2009). A model for hormonal control of the menstrual
cycle: structural consistency but sensitivity with regard to data. J. Theoret. Biol., 260(4), 572580.
http://dx.doi.org/10.1016/j.jtbi.2009.06.017. doi:10.1016/j.jtbi.2009.06.017.
Sinha-Hikim, I., Arver, S., Beall, G., Shen, R., Guerrero, M., Sattler, F., Shikuma, C., Nelson, J. C.,
Landgren, B. M., Mazer, N.a., & Bhasin, S. (1998). The use of a sensitive equilibrium dialysis method for the measurement of free testosterone levels in healthy, cycling women and in
human immunodeficiency virus-infected women. J. Clin. Endocrinol. Metab., 83(4), 13121318.
http://www.ncbi.nlm.nih.gov/pubmed/9543161.
Skinner, M. K. (2005). Regulation of primordial follicle assembly and development. Hum. Reprod. Update,
11(5), 461471. http://humupd.oxfordjournals.org/content/11/5/461.short.
Taylor, A., Mccourt, B., Martin, K. A., Anderson, E. J., Adams, J. M., Schoenfeld, D., & Hall, J. E. (1997).
Determinants of abnormal gonadotropin secretion in clinically defined women with polycystic ovary
syndrome. J. Clin. Endocrinol. Metab., 82(7), 22482256.
Vetharaniam, I., Peterson, a. J., McNatty, K. P., & Soboleva, T. K. (2010). Modelling female reproductive function in farmed animals. Anim. Reprod. Sci., 122(34), 164173. http://www.ncbi.nlm.
nih.gov/pubmed/20832205. doi:10.1016/j.anireprosci.2010.08.015.
Welt, C. K., McNicholl, D. J., Taylor, A. E., & Hall, J. E. (1999). Female reproductive aging is
marked by decreased secretion of dimeric inhibin. J. Clin. Endocrinol. Metab., 84(1), 105111.
http://www.ncbi.nlm.nih.gov/pubmed/9920069.
Yasin, M., Dalkin, A., Haisenleder, D., & Marshall, J. (1996). Testosterone is required for gonadotropinreleasing hormone stimulation of luteinizing hormone-beta messenger ribonucleic acid expression in
female rats. Endocrinology, 137(4). http://endo.endojournals.org/content/137/4/1265.short.

156

A.O. Hendrix et al.

Yen, S. S. (1999). The human menstrual cycle: neuroendocrine regulation. In S. S. C. Yen, R. B. Jaffe,
& R. L. Barbieri (Eds.), Reproductive endocrinology. physiology, pathophysiology and clinical management (4th ed., pp. 191217). Philadelphia: Sauders.
Zeleznik, A. J. (2004). The physiology of follicle selection. Reprod. Biol. Endocrinol., 2, 31.
http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=442133&tool=pmcentrez&rendertype=
abstract. doi:10.1186/1477-7827-2-31.