Encoding and Decoding Mechanisms of Pulsatile Hormone Secretion


  • J. J. Walker,

    1. Henry Wellcome Laboratories for Integrative Neuroscience and Endocrinology, University of Bristol, Bristol, UK.
    2. Bristol Centre for Applied Nonlinear Mathematics, Department of Engineering Mathematics, University of Bristol, Bristol, UK.
    Search for more papers by this author
  • J. R. Terry,

    1. Henry Wellcome Laboratories for Integrative Neuroscience and Endocrinology, University of Bristol, Bristol, UK.
    2. Bristol Centre for Applied Nonlinear Mathematics, Department of Engineering Mathematics, University of Bristol, Bristol, UK.
    Search for more papers by this author
  • K. Tsaneva-Atanasova,

    1. Bristol Centre for Applied Nonlinear Mathematics, Department of Engineering Mathematics, University of Bristol, Bristol, UK.
    Search for more papers by this author
  • S. P. Armstrong,

    1. Henry Wellcome Laboratories for Integrative Neuroscience and Endocrinology, University of Bristol, Bristol, UK.
    Search for more papers by this author
  • C. A. McArdle,

    1. Henry Wellcome Laboratories for Integrative Neuroscience and Endocrinology, University of Bristol, Bristol, UK.
    Search for more papers by this author
  • S. L. Lightman

    1. Henry Wellcome Laboratories for Integrative Neuroscience and Endocrinology, University of Bristol, Bristol, UK.
    Search for more papers by this author

Jamie J. Walker, Henry Wellcome Laboratories for Integrative Neuroscience and Endocrinology, University of Bristol, Dorothy Hodgkin Building, Whitson Street, Bristol BS1 3NY, UK (e-mail: jamie.walker@bristol.ac.uk).


Ultradian pulsatile hormone secretion underlies the activity of most neuroendocrine systems, including the hypothalamic-pituitary adrenal (HPA) and gonadal (HPG) axes, and this pulsatile mode of signalling permits the encoding of information through both amplitude and frequency modulation. In the HPA axis, glucocorticoid pulse amplitude increases in anticipation of waking, and, in the HPG axis, changing gonadotrophin-releasing hormone pulse frequency is the primary means by which the body alters its reproductive status during development (i.e. puberty). The prevalence of hormone pulsatility raises two crucial questions: how are ultradian pulses encoded (or generated) by these systems, and how are these pulses decoded (or interpreted) at their target sites? We have looked at mechanisms within the HPA axis responsible for encoding the pulsatile mode of glucocorticoid signalling that we observe in vivo. We review evidence regarding the ‘hypothalamic pulse generator’ hypothesis, and describe an alternative model for pulse generation, which involves steroid feedback-dependent endogenous rhythmic activity throughout the HPA axis. We consider the decoding of hormone pulsatility by taking the HPG axis as a model system and focussing on molecular mechanisms of frequency decoding by pituitary gonadotrophs.


Rhythms are fundamental in nature and endocrine rhythms span time frames ranging from milliseconds to years. Many of these are organised by endogenous mechanisms, including the circadian (approximately 24-h period) cycles of many hormones that free-run under constant conditions (1), and even slower circannual (approximately yearly) endocrine rhythms that are maintained under photoperiodic clamping (2). Hormone secretion is also rhythmic over much shorter time frames, as illustrated by neuroendocrine systems where oscillators controlling neuronal membrane potential, with time frames of milliseconds to seconds, control the coordinated activity of hypothalamic neuronal networks. These characteristically secrete hormone releasing factors into hypophysial portal blood in a pulsatile fashion with ultradian (period < 24 h) time frames of minutes to hours, which in turn drive the secretion of hormones from the anterior pituitary with similar time frames. The pituitary hormones act on target tissues, including the adrenals and gonads, which synthesise and secrete glucocorticoid hormones and gonadal steroids in a pulsatile fashion. The magnitude of this pulsatile hypothalamic-pituitary target gland activity is modulated by slower rhythms. Inputs from the suprachiasmatic nucleus (SCN), for example, result in a circadian variation in glucocorticoid pulse amplitude. In addition to daily changes, reproductive hormones also show oestrus/menstrual temporal profiles as well as circannual cycles in seasonal breeders.

Oscillatory signals are well described for both intracellular and cell-to-cell communication in many biological systems (3). When compared with steady-state signalling, pulsatile signalling allows greater control, is more robust to degradation, and is generally more energy efficient (4). Moreover, pulsatile signalling provides target sites with a quiescent interpulse interval allowing target receptor recovery, and is therefore essential for maintaining tissue responsiveness (5). Pulsatile signals may also be more ‘information rich’ in the sense that pulse amplitude, pulse duration, pulse shape and interpulse interval can all theoretically provide information to the target cell, in comparison to amplitude alone provided by a steady-state input. Most importantly, pulsatile signalling permits not only amplitude, but also frequency modulation, and the potential advantage of this is amply illustrated by comparing the information provided by a black and white television image (where each pixel has a given signal amplitude) and a colour television image (where each pixel has a given frequency as well as amplitude).

Ultradian pulsatility underlies the secretion of most hormones (6). Despite this, in many neuroendocrine systems, the physiological implication of these rhythms and the mechanisms underlying their generation (encoding) and interpretation (decoding) at target cells are still unclear. Here, we begin to address both of these questions. Specifically, we have studied mechanisms within the hypothalamic-pituitary-adrenal (HPA) axis responsible for encoding glucocorticoid pulsatility. We review the ‘hypothalamic pulse generator’ hypothesis, and also discuss an alternative model for pulse generation that involves steroid feedback-dependent endogenous rhythmic activity throughout the HPA axis. We also consider the second question concerning the decoding of hormone pulsatility, taking the hypothalamic-pituitary-gonadal (HPG) axis as a model system and focussing on molecular mechanisms of frequency decoding by pituitary gonadotrophs.

Encoding glucocorticoid pulsatility

A pulsatile mode of signalling underlies the activity of the HPA axis, a system crucial for maintaining basal and stress-related homeostasis by regulating the circulating levels of vital glucocorticoid hormones (cortisol in man, corticosterone in rodents). Glucocorticoids govern a broad range of physiological functions, including the regulation of cardiovascular, metabolic, cognitive and immunological activity (7–10). The regulation of HPA activity depends on multiple inputs (Fig. 1). In the hypothalamus, the paraventricular nucleus (PVN) receives an indirect input from the SCN which regulates circadian variation in HPA activity, as well as afferent information from brainstem nuclei responding to physical stressors such as hypotension and inflammation, and from limbic areas of the central nervous system that respond to cognitive and emotional stressors (11). The PVN regulates corticotroph activity in the anterior pituitary via two neuropeptides [corticotrophin-releasing hormone (CRH) and arginine vasopressin (AVP)], which are released into the pituitary portal circulation and hence the anterior pituitary. Upon stimulation, corticotrophs secrete adrenocorticotrophic hormone (ACTH) into the general circulation, through which it accesses cells of the adrenal cortex initiating the synthesis and secretion of glucocorticoids.

Figure 1.

 Glucocorticoid release is regulated by the hypothalamic-pituitary adrenal (HPA) axis. Neurones of the paraventricular nucleus (PVN) receive circadian signals from the suprachiasmatic nucleus (SCN), as well as information from brainstem nuclei responding to physical stressors, and from limbic areas of the central nervous system (CNS) that respond to cognitive and emotional stressors. These neurones project to the median eminence where they release corticotrophin-releasing hormone (CRH) and arginine vasopressin (AVP) into the portal blood. These peptides act on corticotroph cells in the anterior pituitary to secrete adrenocorticotrophic hormone (ACTH) into the general circulation, which in turn stimulates the synthesis and release of glucocorticoid hormones (CORT) from cells of the adrenal cortex. Glucocorticoids feed back at the pituitary and hypothalamus to inhibit ACTH and CRH/AVP secretion, respectively.

Glucocorticoid hormones act at target sites via their two cognate receptors [the glucocorticoid receptor (GR) and mineralocorticoid receptor (MR)], the expression of which is widespread in areas regulating HPA activity (12–14). The classical effect of glucocorticoids is via activation of their receptors that translocate to the nucleus acting as ligand dependent transcription factors binding to glucocorticoid response elements at promoter regions of glucocorticoid sensitive genes (15–18). In addition to these classic genomic effects, there has been considerable recent interest in rapid nongenomic mechanisms through which glucocorticoids can also act (19,20).

In the basal state, glucocorticoids are released in an ultradian pulsatile fashion from the adrenal cortex, which results in rapidly changing levels of hormone concentration observable in blood plasma (21–24) (Fig. 2), as well as within target tissues such as the brain (25,26). The classic circadian glucocorticoid rhythm, which peaks in the morning in man (27) and evening in the rodent (28), is the result of amplitude (and to a lesser extent frequency) modulation of the underlying ultradian rhythm (22,23,29–33). A particularly striking feature of glucocorticoid pulsatility is the appearance of a more distinct ultradian rhythm during the peak of the circadian cycle, which operates at a frequency of approximately one pulse per hour in rats (see shaded region 2 of Fig. 3). Variations in amplitude and frequency of the ultradian rhythm not only compose the circadian rhythm, but also characterise changes in HPA activity that occur during early life programming (34), chronic stress (35), lactation and ageing (36), and a number of other physiological and pathological conditions (37).

Figure 2.

 Human data for adrenocorticotrophic hormone (ACTH) and cortisol from a healthy male sampled at 10-min intervals, adapted from (24). Concentration time series demonstrates strong concordance between rhythmic secretion of ACTH and cortisol, as well as a short time lag in ACTH-induced cortisol secretion.

Figure 3.

 Different ways in which the hypothalamic-pituitary adrenal axis encodes glucocorticoid pulsatility. In region 1, where mean corticotrophin-releasing hormone (CRH) levels are low, glucocorticoid pulsatility may well reflect the irregular high-frequency fluctuations in CRH and other corticotrophin-releasing factors acting on the anterior pituitary. In region 2, higher mean levels of CRH drive on the anterior pituitary are sufficient to excite the intrinsic rhythmicity of the pituitary-adrenal loop, which gives rise to the more distinct approximately hourly rhythm in glucocorticoid secretion. CRH data are adapted from a previous study (51); glucocorticoid data adapted from a previous study (21). All data were obtained from male Sprague–Dawley rats.

The importance of glucocorticoid pulsatility resides in its ability to provide a digital signalling system that can respond rapidly to stress or changes in environmental conditions. A recent study by Stavreva et al. (38) clearly shows that the pattern of glucocorticoid presented to a tissue is critical for its transcriptional response, and this effect is seen both in vivo and in vitro. The details of the mechanisms underlying both genomic and nongenomic signalling of glucocorticoid hormones are beyond the scope of this review, but they not only act directly on multiple genes throughout the body, but also have major effects on the ‘clock genes’ and through them on circadian physiology (39,40).

Despite the significance of glucocorticoid pulsatility (36), surprisingly little is known about the mechanisms that encode this rhythm. Although the role of the SCN is well founded in regulating circadian HPA activity (41,42), evidence suggests it is not required for the generation of ultradian rhythmicity. Indeed, lesioning of the SCN completely abolishes the circadian glucocorticoid rhythm in both the adrenals and blood plasma (43–45), although it has little effect on the ultradian rhythm, which persists at amplitudes and frequencies comparable to those observed at the peak of the native circadian rhythm (E. Waite, personal communication). Furthermore, pulsatility in the HPA axis appears to be regulated independently of the ultradian rhythmicity found in other neuroendocrine systems. Glucocorticoids and luteinising hormone (LH) are both secreted in pulses with similar frequencies, for example, although their release is not concurrent (37).

Given the wealth of evidence showing that the circadian pacemaker resides within the hypothalamus (46), it is perhaps not surprising that it has also been assumed that the ultradian rhythm was the result of a hypothalamic neural pulse generator. We review this hypothesis, covering what we feel to be the most important experimental studies and go on to discuss recent developments from theoretical studies, which have led us to propose a new model for the encoding mechanisms of glucocorticoid pulsatility.

The ‘hypothalamic pulse generator’ hypothesis

Neural signalling to the anterior pituitary is encoded in the dynamic patterns of hypothalamic neuropeptides released into the portal circulation (47). Both in man and the rat, the major ACTH secretagogues are CRH and to a lesser extent AVP, although oxytocin, norepinephrine and epinephrine may modulate corticotroph activity (48). Moreover, CRH is the only corticotrophin-releasing factor known to regulate pro-opiomelanocortin gene expression in the rat (48). Whilst, CRH is also an important ACTH secretagogue in ovine species there is evidence to suggest that the predominant corticotroph secretagogue is AVP (49).

Push–pull perfusion studies of the median eminence in freely moving rats show that CRH follows an irregular pulsatile mode of secretion, with a frequency of approximately three pulses per hour (50). Interestingly, no significant difference is seen in CRH pulse frequency between the morning and evening, indicating a relatively steady frequency over the diurnal period (51). On the other hand, mean CRH concentration levels do vary significantly over the 24-h period, with evening concentration levels being almost two-fold greater than those of the morning (51). These increased levels result from the fact that mean pulse nadir and peak levels, as well as pulse amplitude, and to a lesser extent pulse duration, are all significantly higher in the evening (51).

An episodic mode of secretion for CRH and AVP has also been demonstrated in vivo in a number of other species under basal conditions (48). In the portal blood of unrestrained conscious rams, a pulsatile pattern of CRH and AVP secretion is evident (52), and measurement of AVP in the pituitary venous effluent of the unanesthetised horse also reveals an episodic secretory pattern (53). In conscious sheep also, both CRH and AVP display a pulsatile secretory pattern in the portal blood (54).

The regulatory oscillator(s) underlying the pulsatile mode of CRH secretion remains to be defined. It is certainly interesting to compare the pulses of CRH with the much more organised pulsatile secretion of gonadotrophin-releasing hormone (GnRH) into portal blood. GnRH and LH release has a relatively consistent frequency, which correlates well with organised increases in electrical activity in networks of GnRH-producing neurones (55). By contrast, the irregularity of episodic CRH secretion not only suggests a different underlying mechanism, but may also reflect an inability of CRH-producing neurones to effectively synchronise with one another in a coherent way (51).

ACTH pulsatility and its relationship to CRH

Given this episodic release of CRH, it was not surprising that pulsatile patterns of ACTH measured in blood plasma were also found in many species (4). What has been more surprising however, is that in most cases, the relationship between pulsatile neuropeptides and ACTH rhythmicity is far from straightforward.

In the rat, two ultradian ACTH rhythms are found simultaneously (56). Concentrations of ACTH in blood plasma display fast episodic bursts of variable amplitude that occur at a frequency of approximately three pulses per hour (so-called ‘micropulses’), as well as larger and more prolonged episodes of secretion that appear approximately every 1–2 h (we shall refer to this as ‘ultradian’). The two rhythms in ACTH are related in the sense that it is in fact the rhythmic variation in micropulse amplitude that makes up the slower ultradian rhythm (56).

Blockade of endogenous CRH by passive immunoneutralisation in the rat results in a significant reduction in micropulse amplitude without any effect on micropulse frequency (57), which could be attributed to the pulsatile release of other corticotrophin-releasing factors, an intrinsic rhythmicity of the corticotroph, or a combination of these factors (58). Moreover, blockade of endogenous CRH practically completely destroys the slower ultradian rhythm (57), which has prompted speculation that the slower ultradian rhythm in ACTH is driven by rhythmic secretion of CRH (57). However, there are currently no solid data that support this hypothesis. In fact, this hypothesis is actually contradictory to data on the frequency of pulsatile CRH measured in rats of the same strain (male Sprague-Dawley), which was three pulses per hour across the circadian cycle (50,51,59), and which clearly correlates well with the ACTH micropulse frequency.

In larger species, which allow for the simultaneous measurements of portal blood and blood plasma, a straightforward connection between pulsatile CRH/AVP and pulses of ACTH is also absent. In rams, for example, a clear relationship between portal levels of AVP/CRH and temporal patterns of ACTH is not seen (52). In sheep, a large proportion of CRH/AVP pulses are not followed by a significant rise in ACTH, and a significant number of ACTH pulses are not preceded by a pulse in CRH or AVP (54). Moreover, in sheep that have undergone surgical disconnection of the hypothalamus from the pituitary, pulsatility in ACTH and cortisol is maintained (60).

Taken together, of these diverse experimental studies highlight the lack of any clear causal relationship between changes in portal CRH concentrations and the ultradian rhythmicity of ACTH and glucocorticoid hormones. This suggests that the generation of HPA ultradian activity must involve other factors, presumably at a sub-hypothalamic level.

Network encoding of pulsatility

Some biological systems are endogenously rhythmic and oscillate under the influence of constant stimulation, or even in the absence of stimulation; such systems are often referred to as ‘pacemakers’ (61). In hepatocytes (liver cells), for example, continuous stimulation with physiological levels of AVP induces intracellular Ca2+ oscillations, the frequency of which increases with increasing concentrations of agonist (62). The endogenous rhythmicity of these systems finds its roots in the underlying regulatory mechanisms that govern the dynamics of that system (3), and these often involve some form of feedback, which may be positive and/or negative, nonlinear in nature, and is often time-delayed (3,61,63). Indeed, circadian time-keeping in neurones of the SCN is achieved through an intricate molecular circuitry involving ‘clock genes’ that take part in a complex regulatory network consisting of transcriptional and translational feedback loops (64). Feedback also underlies the generation of ultradian rhythmicity at the cellular level. For example, serum treatment of cultured cells induces oscillatory expression of the transcription factor Hes1 at ultradian frequencies (period approximately 2 h), which is the result of a negative feedback loop whereby Hes1 binds directly to regulatory sequences in the Hes1 promoter, thereby repressing transcription of its own gene (65). The network comprising the HPA axis also features some of the properties that are common to pacemaker circuitry. In particular, the anterior pituitary, PVN and higher centres are all targets for negative feedback by circulating glucocorticoids (66–69). In addition to feedback, delays are inherent in the HPA network, which arise from transmission times through the blood, as well as delayed response times.

The existence of pacemaker traits in the HPA network has prompted speculation that the generation of pulsatile glucocorticoid secretion is not necessarily a result of the neural pulse generator in the hypothalamus, but may actually arise from the complex network of excitatory and inhibitory interactions between CRH, ACTH and the glucocorticoids. This view is supported by a number of studies that have developed mathematical models characterising the dynamic interactions between HPA hormones (70–72). Most of these studies assume that feedback at the level of both the pituitary and the hypothalamus gives rise to the ultradian rhythm. If this were the case, then CRH, ACTH and glucocorticoids would all oscillate over the same time frame. However, there is no good evidence for this. Specifically, in the rat, the approximately hourly rhythm observed in ACTH (56) and corticosterone (22) is not observable in CRH during either the nadir or peak of the circadian rhythm (51) and, if glucocorticoid feedback at the hypothalamic level was important in generating ultradian rhythmicity, then removal of this feedback should ablate CRH pulsatility. However, in cultured explants of the macaque hypothalamus, the pulsatile release of CRH persists both in the presence and absence of glucocorticoids (73) and, in conscious rats, CRH pulsatility in the median eminence is maintained following adrenalectomy (59).

We have recently taken a mathematical approach to explore the encoding mechanisms underlying glucocorticoid pulsatility. Our mathematical model builds upon one describing dynamic interactions within the HPA axis (74), and focuses on the effects of nonlinear feedback of glucocorticoids mediated by GR at the level of the anterior pituitary. Furthermore, we incorporate the delayed response of glucocorticoid secretion following ACTH stimulation (75), which results from the lack of releasable pools of glucocorticoids and the need to synthesise the hormone before secretion.

Our theoretical results do not discount the possibility of a hypothalamic pulse generator but suggest that ultradian pulsatility in ACTH and glucocorticoids can also occur even in the absence of pulsatile input from the hypothalamus, providing that the mean levels of hypothalamic drive are within a certain range. Thus, for very low or high levels of constant hypothalamic stimulation, the mathematical model predicts a steady-state response in ACTH and glucocorticoid levels, whereas, for intermediate levels of hypothalamic drive, sustained oscillations in ACTH and glucocorticoid levels occur (76). These oscillations are born out of the excitatory–inhibitory loop formed by the interactions between the anterior pituitary and the adrenal gland. Implicit in this idea is that there is a close coupling between ultradian rhythms in ACTH and glucocorticoids, and there is indeed good experimental evidence supporting such a relationship (Fig. 2).

Although yet to be tested in vivo, it follows from our work that the HPA axis may have multiple ways of encoding glucocorticoid pulsatility (Fig. 3). During the nadir of the circadian rhythm, when mean CRH levels are low, our mathematical results suggest that fluctuations in ACTH and glucocorticoids most likely reflect the activity of episodic CRH and other corticotrophin-releasing factors. During the peak of the circadian cycle, however, when mean CRH levels are significantly higher, the theoretical model predicts that these higher mean CRH levels are sufficient to generate the endogenous hourly rhythm in ACTH and glucocorticoids, an intrinsic property of the pituitary-adrenal system (76).

Under both basal and nonbasal conditions, the HPA axis functions as a closed-loop control system, heavily influenced by negative feedback from circulating glucocorticoids. The regulation of pulsatility in the HPA axis likely involves a number of factors. The complex network of excitatory and inhibitory connections coupled with the pulsatile activity of hypothalamic-releasing factors renders the task of unravelling mechanisms regulating glucocorticoid pulsatility extremely difficult. Experimental studies have so far been unsuccessful in identifying either the mechanistic or anatomical origin of the ultradian rhythm. We have been able to demonstrate, using a mathematical approach, that glucocorticoid negative feedback at the level of the anterior pituitary may not just be important for the homeostatic regulation of optimal levels of ACTH and glucocorticoids, but also could actually be involved in generating ultradian HPA activity. The implications of this finding are that the pulsatile patterns of glucocorticoids that we observe in blood plasma may well reflect the integrated activity of pulsatile hypothalamic forcing on an endogenously rhythmic pituitary-adrenal system.

Decoding GnRH pulsatility

In the preceding sections, the HPA axis was used as a model system to consider physiological mechanisms generating pulsatile glucocorticoid signals. Recent work (38,77) has demonstrated that target cells are sensitive to the pattern of glucocorticoid to which they are exposed, although the molecular mechanisms used by cells to decode pulsatile glucocorticoid signals have not yet been extensively explored. Accordingly, we now focus on the HPG axis as a model system for exploring pulsatile signal decoding.

The neuroendocrine network regulating HPG activity has many similarities to that which mediates HPA activity. In particular, it too is characterised by an ultradian pattern of hormone secretion. The main role of the HPG axis is in regulating reproductive function. Following its release from the hypothalamus, GnRH acts via seven transmembrane region (7TM) receptors to stimulate the synthesis and secretion of LH and follicle-stimulating hormone (FSH) from gonadotrophs in the anterior pituitary. It acts via type I GnRH receptors (GnRHRs) to stimulate phospholipase C, activating protein kinases C (PKC) and mobilising Ca2+. This leads to activation of mitogen-activated protein kinase (MAPK) pathways and Ca2+ effectors such as calmodulin, mediating effects of GnRH on exocytotic gonadotrophin secretion, as well as on the expression of many genes including those for the gonadotrophin subunits (78–80). Following their secretion into the general circulation, LH and FSH act to mediate the control of gametogenesis and hormone secretion from the gonads. In a similar manner to regulation of the HPA axis, feedback mechanisms (involving gonadal steroids and proteins such as activins, inhibins and follistatins) influence GnRH secretion and/or GnRH action at the pituitary (81,82).

GnRH is synthesised in, and secreted from, a relatively small number (100 s) of hypothalamic neurones and has long been known to be secreted in brief pulses. GnRH pulse frequency varies under different physiological conditions. For example, it varies over the menstrual cycle with pulses on average every 6 h in mid- to late-luteal phases and every 90 min during follicular and early luteal phases (83). GnRH pulse frequencies are higher in rats and mice with physiological pulse intervals of 8–240 min (84). Importantly, GnRH effects on its target cells depend upon pulse frequency as illustrated by early studies showing that constant GnRH suppresses LH and FSH secretion, whereas the restoration of GnRH pulses restores gonadotrophin secretion (85). Changing GnRH pulse frequency is also the primary means by which the body alters its reproductive status during development, with an increase in GnRH frequency driving the increased gametogenesis and gonadal steroid production at puberty (86). Moreover, stimulation paradigm is crucial for therapeutic manipulation of this system because pulsatile stimulation with GnRH agonists is used to stimulate gonadotrophin secretion in assisted reproduction, whereas sustained treatment ultimately reduces gonadotrophin secretion and this underlies agonist efficacy against steroid hormone-dependent cancers (87).

Although frequency decoding is fundamental to the physiology and pharmacology of the HPG axis, the mechanisms are poorly understood. Most recent work has focused on effects of GnRH on expression of gonadotrophin subunit genes, which [in both gonadotrophs and LβT2 cells; a gonadotroph lineage cell line expressing GnRHR, beta subunit of luteinising hormone (LHβ) and beta subunit of follicle-stimulating hormone (FSHβ)] are sensitive to GnRH pulse frequency. Increasing GnRH pulse frequency to physiological levels increases its effects on LHβ, FSHβ and GnRHR expression but, as frequency is further increased to super-physiological levels, transcription is reduced (84,88–94). Computational models (95–101) lacking negative feedback show frequency-dependence but not bell-shaped frequency-response relationships (Fig. 4). Indeed, it is generally considered that such bell-shaped frequency-response relationships require feedback mechanisms that could include GnRHR down-regulation, induction of RGS (regulator of G-protein signalling)-2, inhibition of Ca2+ channels by the calmodulin-dependent G-protein Kir/Gem, or induction of MAPK phosphatases (MKPs). Rapid homologous receptor desensitisation can be excluded as a mechanism because type I mammalian GnRHR do not show this behaviour. They lack the C-terminal tails that mediate phosphorylation, arrestin binding and desensitisation of numerous other 7TM receptors (79,102–104). Alternative decoding mechanisms involve interplay between Egr-1 [an extracellular signal-regulated kinase (ERK)-activated transcription factor providing a down-stream readout for ERK activation] and a co-regulator (Nab-2) at the LHβ promoter. In this model, low GnRH pulse frequency causes transient Egr-1 expression, causing expression of Nab-2 that inhibits LHβ expression, whereas, at high pulse frequencies, more sustained increases in Egr-1 quench Nab-2 and increase LHβ transcription (105). For the FSHβ promoter, similar interplay between c-Fos and the co-regulator TG-interacting factor has been proposed to underlie preferential activation at low GnRH pulse frequency (82). Another possibility, suggested by Ciccone et al. (106), is that cAMP response element-binding protein and inducible cAMP early repressor (ICER) are important. In this model, high pulse frequencies preferentially induce ICER causing transcriptional repression by competing for a cAMP response element site in the FSHβ promoter (106).

Figure 4.

 Frequency decoding mechanisms. (a) Possible responses to a square wave pulsatile stimulation with 5-min pulse duration and 120-, 60- or 30-min pulse intervals, eliciting sequential responses (1–3) in a simple vectorial signalling pathway (in which the pulsatile stimulus causes response 1, which causes response 2, which in turn causes response 3). With such pulsatile inputs, upstream responses may have rapid onset and offset, following the input in a process known as digital tracking (rows 1 and 2). However, downstream responses with slower kinetics may not have not returned to baseline before repeat stimulation and this can result in cumulative saw-tooth responses in a process known as integrative tracking (row 3). Integrative tracking can increase the efficiency of signalling and the differential kinetics at different response levels can underlie differences in frequency-dependence (110) as illustrated in (b). This shows the relationship between pulse frequency and responses calculated as the area under the curve for the input signal (A) or three sequential responses [B, C and D, which correspond to responses 1, 2 and 3 in (a)] in a vectorial signalling cascade, again using a 5-min stimulus (so that the maximal frequency of 12 pulses/hour is identical to continuous stimulation). Note that maximal activation of response b is only seen with continuous stimulation, whereas near maximal activation of response D occurs at one pulse/hour. Importantly, these processes can not alone explain the genuine frequency decoding (i.e. dependence on interpulse interval that is independent of cumulative pulse duration) that is evident in the bell-shaped frequency-response relationships for gonadotrophin-releasing hormone (GnRH) effects on gonadotrophin subunit and GnRH receptor (GnRHR) expression [i.e. all the curves in (b) reach the same maximum as pulse frequency increases]. Such frequency decoding is assumed to require positive or negative feedback or feedforward loops (96).

The results described above highlight three distinct possibilities: that GnRH frequency decoding reflects feedback effects on signal generation in the cytoplasm; that frequency decoding occurs at the level of the transcriptome; or that frequency decoding reflects both of the above. Using conventional techniques, it has been difficult to test the first of these possibilities, although we have developed live cell readouts for signalling pathways implicated in frequency decoding, and have used these to explore GnRH signalling, as outlined below.

We note here that it is important to distinguish between ‘frequency-dependence’ and ‘frequency-decoding’. There are many systems in which increasing pulse frequency increases responses but this could reflect either the increase in cumulative pulse duration or the reduction in interpulse interval (96). True frequency decoders sense interpulse interval independently of cumulative pulse duration (Fig. 4), as illustrated by GnRH effects on expression of genes encoding rodent LHβ and FSHβ, both of which are increased more effectively at low or intermediate GnRH pulse frequency than at high frequency or with sustained stimulation (84,88–94).

Nuclear factor of activated T-cells (NFAT) signalling

GnRHR-mediated activation of the Ca2+/calmodulin pathway can influence gonadotrophin subunit gene expression (107–109), and mechanisms by which calmodulins and their effectors interpret frequency-encoded Ca2+ signals are well established (110–113). More recently, NFATs, transcription factors activated by Ca2+/calmodulin-dependent activation of the protein phosphatase calcineurin (which dephosphorylates NFAT), have been implicated in transcriptional regulation by GnRH (114–116). This is of particular interest in light of the role of NFATs as frequency decoders in other systems (117–120). When GnRHR expressing HeLa cells were transduced with NFAT2-EFP (a reporter that translocates to the nucleus after Ca2+/calmodulin/calcineurin activation), GnRH caused dose-dependent NFAT2-EFP translocation, although the effect was slower than the underlying Ca2+ response (121). With pulsatile stimulation (5-min pulses, 1-h intervals), responses were reproducible with no measurable desensitisation or reduction in GnRHR expression. Varying the GnRHR number influenced response amplitude but not kinetics and, again, no desensitisation was seen. With 5-min GnRH pulses at 1- or 2-h intervals, the NFAT-EFP translocation responses returned to basal or near basal values between stimuli, so that the responses simply followed the stimuli (albeit with slower kinetics) in a process known as digital tracking (110). However, at high pulse frequency (30 min), responses had not returned to basal values before repeat stimulation and this led to cumulative of ‘saw-tooth’ responses, in a process known as integrative tracking. GnRH also caused dose- and frequency-dependent activation of αGSU-, LHβ- and FSHβ-luc reporters and these responses were inhibited by cyclosporin A (an inhibitor of calcineurin, the Ca2+/calmodulin sensitive phosphatase that causes NFAT translocation to the nucleus), indicating calcineurin-dependence. Pulsatile GnRH also activated an NFAT-responsive luc reporter, although its effect was proportional to cumulative pulse duration (121).

ERK signalling

Similar to many other 7TM receptors, GnRHR activate the Raf/MEK/ERK cassette (i.e. the basic components of the best characterised mitogen-activated protein kinase pathway, in which the protein kinase Raf phosphorylates and activates the protein kinase MEK causing it, in turn, to phosphorylate and activate ERK) (78–80). Activated ERKs can translocate to the nucleus where they phosphorylate transcription factors to control gene expression. GnRH activates ERK1/2, and ERKs can mediate GnRHR-stimulated transcription of the common αGSU subunit, as well as the LHβ and FSHβ (122–126). ERKs can mediate responses to pulsatile GnRH stimulation (107,127,128), pituitary-targeted ERK knockout can cause infertility (129), and the ERK cascade functions as a frequency decoder in other systems (130–132). To explore ERK signal dynamics, we developed a model in which small interfering RNAs (siRNAs) are used to prevent the expression of endogenous ERK1/2 and recombinant adenovirus are used to add back an ERK2-GFP reporter at a physiological expression level (133–135). With this system, we found that pulsatile GnRH causes dose- and frequency-dependent ERK2-GFP translocation to the nucleus. The responses were rapid and transient and therefore showed only digital tracking (Fig. 4). They also failed to desensitise under any condition tested (dose, frequency and receptor number varied). GnRH also caused dose- and frequency-dependent activation of an Egr1-responsive luc reporter (used as a downstream readout for ERK activation), although the responses were again proportional to pulse frequency (with no evidence of a bell-shaped frequency-response relationship). This response, similar to effects of pulsatile GnRH on FSHβ-luc, was inhibited by siRNA-mediated knockdown of endogenous ERKs (133). ERK responses are modulated by many ERK interacting proteins, including dual-specificity phosphatases (DUSPs), some of which can both scaffold and inactivate ERKs. Using siRNA to target DUSPs, we found that 12 of the 16 DUSPs expressed in HeLa cells influenced ERK responses to sustained stimulation with GnRH or a PKC activator (134,135). Moreover, GnRH can increase expression of nuclear-inducible MKP family DUSPs (125,136–138) and a recent computational model illustrated the potential for pulse frequency decoding by MAPK pathways and inducible phosphatases (98). Accordingly, we tested for effects of cycloheximide (to prevent nuclear-inducible MKPs induction), and used GFP fusions containing ERK mutations (D319N, which prevents D-domain-dependent binding of MKPs, and K52R, which prevents catalytic activity). These had little or no effect on the ERK translocation responses arguing against a role for MKPs or ERK-mediated feedback in shaping ERK activation (133).

Our data show that GnRH effects on the Ca2+/calmodulin/calcineurin/NFAT signalling and Ras/Raf/ERK signalling are frequency-dependent, and are consistent with roles for these pathways in mediating pulsatile GnRH effects on gonadotrophin expression. Nevertheless, these pathways do not appear to act as genuine frequency-decoders of GnRH signalling in the models used (HeLa and/or LβT2 gonadotrophs). We were unable to find evidence for the negative feedback assumed to underlie such frequency decoding (i.e. no desensitisation was seen and responses to pulsatile GnRH were not MKP-dependent) despite the fact that genuine GnRH frequency-decoding occurs in both models (121,133). These data differ from those obtained by using a mathematical model that predicts cellular responses to GnRH by simultaneous solution of differential equations describing various aspects of the GnRH signalling to LH secretion. Solving these nonlinear equations by machine computation predicts that GnRH effects will desensitise with pulsatile stimulation and that the extent of this desensitisation will increase with GnRH dose, frequency and receptor number (101,121), yet no such desensitisation was seen with either of our imaging assays. However, a reduction in cell surface GnRHR number is the primary cause of desensitisation in this computational model and we have found that a 5-min exposure to GnRH does not reduce cell surface GnRHR number in the HeLa cell model (121). To further explore this mathematically, we have adapted a published computational model for GnRHR signalling (101) by removal of receptor down-regulation as negative feedback mechanism and have extended it to incorporate two cellular compartments representing the cytoplasm and the nucleus, respectively. We couple both compartments by incorporating into the model the fluxes of activated/inactivated NFAT and ERK1/2 across the nuclear envelope. Thus, the model includes a number of time-dependent evolution equations for all the key players involved in GnRH signalling as outlined above. Accordingly, we have already calibrated preliminary model parameters using experimental measurements of the nuclear to cytosolic ratios of total NFAT (121) and ERK1/2 concentrations (133) as shown in Fig. 5. This model provides an excellent fit for the experimental data but, as noted above, digital or integrative tracking is seen with no evidence of adaptation that might underlie genuine frequency decoding, so we are currently extending the model to incorporate the action of NFAT and an ERK1/2-dependent transcription factor (e.g. Egr-1) as convergent inputs on GSU promoters.

Figure 5.

 Modelling pulsatile gonadotrophin-releasing hormone receptors (GnRHR) signalling. Pulsatile stimulation for 5 min with varied GnRH pulse frequency at 30-min intervals, hourly intervals, or every 2 h, as indicated. (a) The data shown in the upper panel are the normalised nuclear to cytosolic (N : C) ratio of NFAT2-EFP fluorescence intensity (121). The bottom panel illustrated preliminary model simulations of the whole-cell model NFAT response. (b). The data shown are the N : C ratio of ERK2-GFP fluorescence intensity (Tsaneva-Atanasova K, Mina P, Caunt C, Armstrong S, McArdle C, unpublished data). The bottom panel illustrated preliminary model simulations of the whole-cell model ERK1/2 response (133). The data with GnRH at 1-h or 30-min intervals are offset by 1 or 2 units on the vertical axis for clarity.

Most of the previously developed mathematical models concentrate on a particular aspect of GnRH-signalling rather than on the system as a whole. Earlier theoretical work on modulation of pre-synthesised gonadotrophins was based on GnRHR desensitisation kinetics (97). Furthermore, LH release per se has been the focus of several theoretical studies where frequency-dependent modulation of secretion is a consequence of the presumed decrease in GnRHR plasma membrane expression (95,101). Most importantly, it is now clear that type I mammalian GnRHR do not undergo rapid desensitisation (103) and, although agonists do stimulate GnRHR trafficking (139), agonist-induced down-regulation of cell surface GnRHR was not seen with brief GnRH activation (121). Most work on GnRHR signalling has been with sustained stimulation paradigms so that signalling with physiological (pulsatile) activation remains relatively poorly understood, as does the mathematical basis for frequency decoding in this system. In this context, only one study (96) recapitulates experimentally-determined transcriptional characteristics of GSU synthesis; nonetheless, it reflects a ‘top-down’ approach by considering the complex GnRH-signalling as a ‘black-box’, subsuming this process within empirical assumptions regarding translational delays and putative GnRH network modules. There are very few computational models of transcriptional GSU gene regulation that have been developed from the ‘bottom-up’ approach, building on the wealth of knowledge for the intracellular pathways activated by GnRH. Ruf et al. (99) used single cell imaging to monitor effects of GnRH on ERK and down-stream transcriptional responses. However, their modelling concentrated only on the ERK1/2 pathway in populations of cells and did not take into account pulsatile GnRH stimulation. The effects of GnRH on the Ca2+/calmodulin/calcineurin/NFAT signalling and Ras/Raf/ERK signalling are frequency-dependant and can be regarded as sub-modules acting within the overall GnRH signalling pathway.

Concluding remarks

HPA and HPG axes are characterised by pulsatile hormone secretion and digital hormonal signalling systems. We have used these two model systems to demonstrate two major facets of neuroendocrine signalling: mechanisms by which they can encode signals, well exemplified in the HPA axis, and mechanisms by which cells in target tissues decode them, as described for the regulation of gonadotrophin secretion. Mathematical modelling has been key to elucidation of these signalling mechanisms, which highlights the power of modelling to investigate the organisation of complex neuroendocrine systems.


This work was supported by the Engineering and Physical Sciences Research Council (EPSRC) grant EP/E032249/1, and Wellcome Trust grants 084588 and 074112/Z/04/Z.