Brain mechanisms of HPA axis regulation: neurocircuitry and feedback in context Richard Kvetnansky lecture

Abstract

Regulation of stress reactivity is a fundamental priority of all organisms. Stress responses are critical for survival, yet can also cause physical and psychological damage. This review provides a synopsis of brain mechanisms designed to control physiological responses to stress, focusing primarily on glucocorticoid secretion via the hypothalamo-pituitary-adrenocortical (HPA) axis. The literature provides strong support for multi-faceted control of HPA axis responses, involving both direct and indirect actions at paraventricular nucleus (PVN) corticotropin releasing hormone neurons driving the secretory cascade. The PVN is directly excited by afferents from brainstem and hypothalamic circuits, likely relaying information on homeostatic challenge. Amygdala subnuclei drive HPA axis responses indirectly via disinhibition, mediated by GABAergic relays onto PVN-projecting neurons in the hypothalamus and bed nucleus of the stria terminalis (BST). Inhibition of stressor-evoked HPA axis responses is mediated by an elaborate network of glucocorticoid receptor (GR)-containing circuits, providing a distributed negative feedback signal that inhibits PVN neurons. Prefrontal and hippocampal neurons play a major role in HPA axis inhibition, again mediated by hypothalamic and BST GABAergic relays to the PVN. The complexity of the regulatory process suggests that information on stressors is integrated across functional disparate brain circuits prior to accessing the PVN, with regions such as the BST in prime position to relay contextual information provided by these sources into appropriate HPA activation. Dysregulation of the HPA in disease is likely a product of inappropriate checks and balances between excitatory and inhibitory inputs ultimately impacting PVN output.

Keywords:

• HPA Axis
• glucocorticoids
• amygdala
• hippocampus
• prefrontal cortex
• bed nucleus of the stria terminalis

It was a great honor for the senior author to present the inaugural Richard Kvetnansky lecture at the 12th International Symposium on Catecholamines and Other Neurotransmitters in Stress, held at Smolenice Castle in the summer of 2019. Richard was a consummate scientist and a strong advocate for the field of stress biology, and his absence is keenly felt.

The goal of this review is to provide a framework for understanding neurobiological mechanisms governing glucocorticoid responses to adverse events. For the purposes of framing the review with respect to the “problem” of stress, we begin with a brief conceptual introduction. The remainder of this contribution will be directed toward discussion of the neural basis of neuroendocrine responses to stressors, which has been a prevailing emphasis of our laboratory’ for the last 30+ years.

The “Stress” Concept: Origin and Evolution

Interest in stress harkens back to the early days of experimental medicine. The concept that physiological responses drive adaptation originated with Claude Bernard’s early consideration of the “milieu interieur” in the late 1800s (Bernard, 1865), and was a driving factor in Walter Cannon’s original exposition of homeostasis (Cannon, 1939). Selye first popularized the term “stress”-borrowed from engineering- to describe the “non-specific responses of the body to any demand upon it.” He based this definition on clinical observations, where he noted a common spectrum of symptoms present in a variety of disease conditions. These clinical observations were translated into basic research leading to a landmark paper published in Nature in 1936, showing that a variety of noxious conditions presented common physiological reactions, including adrenal hypertrophy, atrophy of lymphoid organs and gastric ulceration in rats (Selye, 1936). The former two observations are the result of elevated glucocorticoid secretion, which will be a guiding topic of this review.

Definition of “stress” emerged some time later (Selye, 1956, 1950). The definition above specifies that “stress” is defined as a “response,” rather than a cause. Over the years this definition has become clouded, as “stress” had moved from consequence to antecedent in popular parlance (indeed as fuzzily defined, stress can be both cause and effect). In deference to Selye’s definition, we will refer to causal factors as “stressors” (rather than “stress”).

Selye also noted that pleasurable or appetitive events are able to generate physiological indices of stress. He subsequently distinguished “distress” and “eustress,” with the former referring to responses to noxious, adverse or aversive stimuli, typical of what is usually thought of as “stress.” “Eustress” essentially refers to responses generated in positive as opposed to negative contexts, and reinforces the notion that physiological responses are required to help the organism perform optimally in both conditions. The distress-eustress continuum has proven hard to reconcile and difficult for the literature to navigate (Selye, 1975). Indeed, a consensus paper suggests limiting discussion of stress to negative conditions, so as to minimize ambiguous definitions (Koolhaas et al., 2011).

The distress-eustress distinction underscores the role of both neuroendocrine and autonomic regulation in energy balance. One can argue that the major so-called “stress systems,” the hypothalamo-pituitary-adrenocortical (HPA) and sympathoadrenomedullary axes, are primarily concerned with metabolism. Indeed, both systems mobilize energy at the level of the liver (gluconeogenesis, proteolysis and lipolysis in the case of glucocorticoids, glycogenolysis and glycolysis in the case of epinephrine), in an attempt to provide resources allowing the organism to adapt to challenges, be they in a negative or positive context. Indeed, this appears to be a principle role for both systems on an hour-to-hour and day-by-day schedule: their relationship with stressors is but one aspect of their primary function.

Activation of stress effectors can occur in response or anticipation of discrete events (acute stressors) or in prolonged fashion when confronted with persistent or intermittent stress exposures (chronic stress). Responses to acute stressors are generally considered adaptive, having the goal of mobilizing resources to meet bodily needs. In contrast, chronic activation of stress effectors can trigger or exacerbate pathologies, due to prolonged systemic drive, alterations in metabolic processes and/or immune dysregulation. As discussed below, pathways controlling acute and chronic responses to stressors can differ substantially, and one should not assume that chronic stress is simply the summation of individual acute responses.

Another important concept in stress is “allostasis,” a term defined by (Sterling & Eyer, 1988). “Allostasis” was initially proposed to explain the process of “stability through change,” as opposed to homeostasis, defined as the constant return to balance of the internal milieu in response to environmental stimuli. They postulated that the parameters proposed to be constant by the concept of homeostasis, could, in fact, have more than one optimal point of balance, depending on the external and internal circumstances the individual faces at any given moment.

This concept was later adopted by Bruce McEwen, who extended it by defining allostatic load as the "cost of chronic exposure to fluctuating or heightened neural or neuroendocrine response resulting from repeated or chronic environmental challenge” (McEwen & Stellar, 1993). These concepts contributed to include the cognitive functions of the individual (i.e. the memory and emotional experience with a particular stressor) as an important component of the adaptation machinery that is required for allostatic processes (Korte et al., 2005).

It is important here to consider the concept of emotion, since emotions drive both generation and interpretation of stressors. Early work in the field proposes that the physical responses are the root cause of emotions (the James-Lange theory of emotion; James, 1994). Although subsequently challenged (e.g. by Cannon, 1927), the physiology and emotion connection remains strong (James, 1994), and indeed has found new life in the somatic marker hypothesis championed by Damasio, which posits that physiological reactions can affect decision making (Damasio, 1996). The key feature to consider here is that emotions almost certainly occur in the context of stressors, and importantly that the physiological reactions occurring during stressors can be encoded as part of the interpreted emotion.

Dynamics of HPA axis regulation

Activation of the HPA axis occurs in a number of contexts. Importantly, the HPA axis exhibits a marked circadian rhythm, with peak secretion generally corresponding with the onset of the active part of the day-night cycle (e.g. onset of the light cycle in diurnal animals, onset of the dark cycle in nocturnal animals; Cascio et al., 1987; Jacobson et al., 1988). The circadian rise is thought to be important for mobilizing energy to meet the energetic needs of an awake, behaving organism. Glucocorticoids also rise in anticipation of meals (Feillet et al., 2006). The circadian peak is guided by input from the suprachiasmatic nucleus, based on endogenous rhythmicity and/or light cues (Cascio et al., 1987). Note that neither circadian nor metabolic release of glucocorticoids necessarily reflect a response to “stressors” per se, but rather fulfill the metabolic function of the of HPA axis (putting the “gluco(se)” in “glucocorticoid” through glyconeogenesis, promoting release of free fatty acids and amino acids by lipolysis and proteolysis, respectively).

Activation by stressors co-opts this important metabolic system to introduce glucocorticoids in order to meet external or internal challenges, real or perceived (Myers et al., 2014b). The important consideration here is that the HPA axis is NOT a canonical “stress system,” but rather a bodily system recruited by stressors. Consequently, all measures of HPA activation need to be weighed with respect to both stressor-elicited secretion and endogenous release patterns. The most striking need is consideration of circadian timing, which is critical to appreciating the meaning of baseline glucocorticoid changes in the context of stress and disease. Indeed, the current gold standard for measurement of human HPA axis reactivity is the cortisol awakening response (CAR), timed across a narrow window corresponding to awakening rather than clock time (Federenko et al., 2004; Stalder et al., 2016). Unfortunately, a sizable literature on human stress biology emerged prior to appreciation of the significance of the CAR, and we have not even begun to tackle the significance of anything resembling the CAR in animal models. Moreover, these concerns call the value of single-point assessment of HPA axis hormones into question, particularly during the active phase of the circadian cycle.

Activation of the HPA axis is mediated by a discrete population of neuroendocrine corticotropin-releasing hormone (CRH) neurons in the paraventricular nucleus of the hypothalamus (PVN; Antoni, 1986; Herman et al., 2016). These neurons produce other neuropeptides (e.g. arginine vasopressin (AVP) that enhance the efficacy of down-stream CRH action (Gillies et al., 1982). Hypophysiotrophic PVN CRH neurons project to the external lamina of the median eminence, whereby CRH and co-stored peptides/transmitters are released into portal veins and subsequently access anterior pituitary corticotropes to drive adrenocorticotropic hormone (ACTH) release. ACTH then travels to the adrenal cortex via the systemic circulation, driving synthesis and secretion of glucocorticoids and completing what is effectively a three-step amplification process (fg/ml CRH to pg/ml ACTH to ng/ml glucocorticoids; Herman et al., 2016). It is important to note that the amplification process is subject to adjustment at both the pituitary and adrenal. For example, glucocorticoid feedback (below) can block pituitary release of ACTH (Keller-Wood & Dallman, 1984), and sympathetic activation can enhance adrenal glucocorticoid production (Jasper & Engeland, 1997; Ulrich-Lai et al., 2006), respectively. Local effects at both the pituitary and adrenal are powerful modifiers of HPA output, and are unfortunately incompletely understood.

Glucocorticoids signal via interaction with two primary receptors, the mineralocorticoid receptor and glucocorticoid receptor (MR and GR; de Kloet et al., 1998). These receptors act via any of several mechanisms: as ligand activated transcription factors to drive or inhibit gene transcription; as intranuclear inhibitors of other transcription factors; or as membrane receptors providing for rapid glucocorticoid signaling (de Kloet et al., 2018; Oitzl et al., 1997). The array of signaling mechanisms provides for HPA axis influence on target systems in the time-frame of minutes (membrane) to days or even weeks (transcription; de Kloet et al., 2008). The MR is extensively bound by relative low levels of glucocorticoids, with nuclear-signaling receptors saturated at low circadian levels of glucocorticoids and leading some to suggest that it is important in rhythmic actions of hormone (Bradbury et al., 1991; De Kloet & Reul, 1987). Recent evidence suggests that the affinity of the membrane-bound receptor may be substantially lower, allowing it to operate at higher, stress levels of glucocorticoid secretion (Joëls, 2006). The GR is traditionally thought to govern the metabolic and stress actions of glucocorticoids, as this receptor is extensively occupied only at the circadian peak or following stress activation of hormone release (Reul & de Kloet, 1985). There is some evidence to suggest that the two receptors heterodimerize in the cell nucleus, which may affect which genes are expressed or inhibited (Trapp et al., 1994), although this has been debated of late (Pooley et al., 2020). Of note, RNAseq studies indicate that despite the high homology present in the DNA binding domain of MR and GR, occupation of the respective receptors produces vastly different patterns of gene transcription (Van Weert et al., 2017).

Glucocorticoid receptors are widely (but not ubiquitously) expressed in the central nervous system (Ahima & Harlan, 1990; Herman, 1993), and have physiological actions in nearly every region tested, ranging from the brainstem to prefrontal cortex. Similarly, MR is also expressed in a variety of regions (Ahima et al., 1991), including several where colocalization with GR is documented (e.g. hippocampus; de Kloet et al., 1998) or highly likely.

It is important to note that corticosteroids are released from the adrenal in a pulsatile manner, with a frequency of roughly 3 hours, entrained in accordance with pulsatile pituitary ACTH release (Sarabdjitsingh et al., 2012; Windle et al., 2013). The net impact of a stressor on corticosteroid release is dependent on pulsatility, as imposition on the rising phase of a pulse will enhance net secretion of steroid, whereas the opposite is true on the falling phase (Lightman et al., 2020). Pulsatile release is critical for appropriate glucocorticoid signaling, as blocking pulsatile release by constant delivery of glucocorticoids impairs GR nuclear translocation and thereby modifies glucocorticoid actions on transcription (Russell et al., 2015; Sarabdjitsingh et al., 2010).

Feedback Control of the HPA Axis

Activation of the HPA axis is limited by glucocorticoid negative feedback inhibition at several levels, ranging from peripheral tissues (e.g. fat) to brain (de Kloet & Herman, 2018; Myers et al., 2012). Negative feedback is best thought of as a multifaceted process, with glucocorticoid signals communicated largely by GR binding in target cells. Similarly, GR in pituitary corticotropes reduce expression of ACTH, limiting the capacity of the brain to drive the HPA axis, and as we shall see below, GR acts in the brain (e.g. hypothalamus, hippocampus, prefrontal cortex, etc.) to limit activation of HPA axis responses to stressors (Ding et al., 2019; Herman et al., 2016; Keller-Wood & Dallman, 1984; Levin et al., 1988). Finally, glucocorticoids can also regulate the HPA axis via peripheral mechanisms: for example, our group has shown that deletion of GR in adipocytes inhibits the HPA axis, likely through actions of end-products generated by GR action (e.g. free fatty acids generated by lipolysis; de Kloet & Herman, 2018).

Glucocorticoid feedback acts across multiple time domains via multiple mechanisms (Keller-Wood & Dallman, 1984). Traditional genomic effects occur at relatively long latencies (hours) and can initiate transcriptional and functional changes that persist beyond the period of active glucocorticoid secretion. Transcriptional actions can involve direct DNA binding to either increase (glucocorticoid response elements, GREs) or decrease (negative or nGREs) gene transcription, and perhaps are the most commonly studied. Ultimate transcriptional actions of GR (or MR) will depend on interactions with nuclear coactivators (e.g. Src) that control histone acetylation/deacetylation and ultimately associations with the transcriptional machinery (Lachize et al., 2009; Meijer et al., 2005). Additional evidence supports the ability of GR to bind other nuclear transcription factors (e.g. AP1, NFkB complexes) and modulate their transcriptional actions, providing for non-GRE dependent effects on gene expression (McKay & Cidlowski, 1998; Yang-Yen et al., 1990).

A wealth of data dating back to early studies by Dallman indicates the ability of glucocorticoids to inhibit HPA axis drive within minutes, so-called “fast feedback” (Dallman & Yates, 1969). In vitro electrophysiological studies from the Tasker group found that PVN neurons are inhibited by glucocorticoids at the level of the membrane, accomplished via G-protein dependent release of endocannabinoids and nitric oxide, which inhibit glutamatergic and enhance GABAergic synaptic currents, respectively, in the PVN (Di et al., 2003). Our group subsequently verified endocannabinoid-mediated membrane fast feedback inhibition in vivo as well (Evanson et al., 2010). Glucocorticoid fast feedback is blocked by PVN deletion of GR in vitro and in vivo, indicating that actions are likely mediated by membrane-associated GRs (Nahar et al., 2015; Solomon et al., 2015). While immunohistochemical studies support the existence of GR (or GR-like) molecules at or near the cell membrane (Jafari et al., 2012; Johnson et al., 2005), the exact nature of the signaling process is unknown. Rapid GR action is also observed at the level of the corticotropes, where fast feedback appears to be mediated by annexin 1 A signaling (Buckingham et al., 2003).

Recent data suggest that GR may also signal via ligand-independent mechanisms. Use of non-membrane-permeant GR ligands (dexamethasone-bovine serum albumin conjugate, DEX-BSA) have revealed the ability of membrane-bound GR to elicit Akt mediated nuclear translocation of unliganded GR. Nuclear GR translocated as a result of DEX-BSA treatment does not appear to directly drive transcription via binding to consensus glucocorticoid response elements (GREs) and promotes transcription of genes distinct those regulated by DEX alone, implying interaction with alternative transcription mechanisms (Rainville et al., 2019). The data indicate further complexities associated with cellular GR signaling as a result of membrane GR binding.

Hypothalamic Mechanisms of HPA Axis Regulation

In general, glucocorticoid feedback directly or indirectly inhibits PVN neurons driving the neuroendocrine cascade. Electrophysiological studies show that neurons in the PVN are inhibited, as indicated by a reduction in mEPSC frequency and enhanced mIPSC frequency during the rapid negative feedback of the HPA axis (Di et al., 2003; Joëls, 2006). Lesion and implant studies suggest strong hypothalamic involvement in feedback, e.g. local implants of glucocorticoid pellets in the region of the PVN can attenuate corticosterone and ACTH responses to stressors (corticosterone; Feldman et al., 1992). In addition, local dexamethasone implants inhibit enhanced CRH and AVP immunoreactivity (Kovacs & Mezey, 1987) and mRNA expression driven by adrenalectomy (Sawchenko, 1987). Conversely, conditional knockdown of GR in PVN (and SON) neurons (using Sim1-driven expression of Cre recombinase to delete exon 1 C-2 or exon3) enhanced both basal and stress-induced ACTH and corticosterone release in male mice (Laryea et al., 2013). Results in females are more variable, depending on the exon targeted for deletion: females with deletion of exon 1 C-2 exhibit minimal HPA axis dysfunction, whereas the impact of exon 3 deletion is equivalent to males, suggesting a potential sex difference in the importance of GR (Laryea et al., 2013; Solomon et al., 2015). As noted above, Sim1-mediated deletion of GR blocks rapid inhibitory effects of glucocorticoids on PVN neurons (Nahar et al., 2015), suggesting that fast feedback actions may be linked to GR in or near the cell membrane. Overall, actions at the PVN are clearly glucocorticoid negative feedback effects.

Drive of the HPA axis may also be inhibited by input from PVN-projecting hypothalamic neurons. A large proportion of hypothalamic neurons are GABAergic, and more than 50% of synaptic inputs on PVN neurons are inhibitory, suggesting a prominent role for local inhibition on HPA axis regulation (Decavel & Van Den Pol, 1990). Indeed, the PVN is surrounded by a shell of GABAergic neurons. These local circuit neurons are in the projection fields of afferents from regions such as the hippocampus, lateral septum and raphe nuclei (Herman et al., 2002). Blockade of ionotropic glutamate receptors in this peri-PVN region increases corticosterone responses to stress (Ziegler & Herman, 2000), suggesting that these local neurons may be involved in mediating trans-synaptic inhibition by relaying extrahypothalamic input (Figure 1).

Figure 1. Neural mechanisms of acute stress inhibition. As noted, the CRH containing region of the medial parvocellular paraventricular nucleus (PVN) receives substantial inhibitory input from hypothalamic (medial preoptic nucleus (mPOA), dorsomedial nucleus (DMH), periPVN zone and medial forebrain structures (bed nucleus of the stria terminalis, BST). The regions receive excitatory inputs from forebrain structures such as the IL infralimbic (IL) and prelimbic (PL) cortices and the ventral subiculum (vSUB), which are thought to mediate trans-synaptic inhibition of HPA axis stress responses. Upstream limbic pathways may also limit drive of the PVN by way of local, intranuclear inhibition of HPA axis excitatory circuits, e.g. the nucleus of the solitary track (NTS) and/or posterior hypothalamus (PH). Open red circles and red lines: inhibitory (e.g. GABAergic) neurons/connections; closed green circles and green lines: excitatory (e.g. glutamatergic) neurons and connections. Figure modified from Herman et al. (2005), with permission, and were created with Biorender.com.

Several nuclei within the hypothalamus act to modulate HPA axis drive, including the medial preoptic area (mPOA), dorsomedial hypothalamic (DMH) and posterior hypothalamic (PH) nuclei (Figure 1). Lesion studies suggest that the medial preoptic area plays a role in inhibition of the HPA axis following stress exposure. This hypothalamic region is rich in gonadal steroid receptors, and is thought to mediate the inhibitory effects of testosterone on HPA axis reactivity (Viau & Meaney, 1996). The DMH sends both glutamatergic and GABAergic inputs to the PVN (Bailey & Dimicco, 2001; Roland & Sawchenko, 1993), and indeed, its role in HPA axis integration appears to vary by subregion: inhibition of the ventromedial component reduces corticosterone responses to stressors, whereas dorsolateral DMH activation drives ACTH release (Bailey & Dimicco, 2001; Herman et al., 2003). GABAergic neurons in the ventromedial DMH are Fos-activated by acute stress and project to the PVN (Cullinan et al., 2008), suggesting a local circuit capable of stress inhibition.

Recent work from our group and that of Campeau note a strong stress-regulatory PVN input from the posterior hypothalamic area (PH; Myers et al., 2016; Nyhuis et al., 2016). This region is particularly interesting given its close link to drive of autonomic responses and defensive behaviors. Targeted inactivation and activation studies (using muscimol and bicuculline, respectively) indicate that the output of PH (in particular its rostral component) activates ACTH secretion, corticosterone release and PVN Fos expression elicited by acute noise or restraint, and drives aggressive and threat avoidance behaviors (Myers et al., 2016; Nyhuis et al., 2016). These effects appear to be mediated by direct excitatory projections to PVN CRH neurons, likely to be glutamatergic and/or peptidergic in phenotype. The effects of PH inactivation appear to inhibit habituated HPA axis responses to noise and restraint, suggesting temporally distinct actions on HPA axis regulation. Anatomical data indicate that stress-activated neurons in the infralimbic cortex (IL) and lateral septum send projections to the PH, with the IL preferentially targeting resident GABA neurons that likely convey intra-nuclear inhibition to PH output neurons (Myers et al., 2016; Nyhuis et al., 2016; Figure 2).

Figure 2. Neural mechanisms of acute stress excitation. Data suggest PVN neurons can be driven by neurons communicating homeostatic challenge, including the nucleus of the solitary tract (NTS), among others. The PVN also has numerous connections with hypothalamic nuclei and subcortical telencephalic structures, including excitatory (PH, anterior BST) and inhibitory (POA, DMH, periPVN, anteroventral BST, posterior BST) inputs. Inhibitory input to the PVN provides a substantial inhibitory tone, which can be disrupted by inhibition from upstream sites such as the medial and central amygdaloid nuclei (MeA, CeA), providing a mechanism for trans-synaptic disinhibition from the limbic forebrain. There is also some evidence suggesting that some cortical regions, such as the infralimbic region (IL) of the medial prefrontal cortex (PL), may also provide trans-synaptic excitation, perhaps via relays in the brainstem. There is less evidence for excitatory input from other forebrain stress circuits, such as the ventral subiculum (vSUB), prelimbic division of the mPFC or paraventricular thalamus (PVT). Input from limbic regions may also access the PVN by interaction with local interneurons in the PVN surround (periPVN). See Figure 1 legend for abbreviations and symbol definitions). Figure modified from Herman et al. (2005), with permission, and were created with using Biorender.com.

The hypothalamus also plays a major role in homeostatic regulation, being largely responsible for coordination of metabolism, thermoregulation, fluid/electrolyte balance and sleep. Disruption of any of the above processes can activate the HPA axis, consistent with the need to provide energy to meet current or potential emergencies. For example, negative energy balance promotes activation of melanocortin neurons in the arcuate nucleus that may in turn activate CRH neurons and result in increased ACTH and corticosterone release (Bell et al., 2000; Liu et al., 2007).

Top-down regulation by cortical and limbic region

Hippocampus

Glucocorticoid signaling in the hippocampus is perhaps a prime example of distant regulatory processes controlling the PVN. Lesion, stimulation and steroid infusion studies were used in the 1970s to 1990s to support a preeminent role of the hippocampus in feedback inhibition of the HPA axis (Herman et al., 2003; Jacobson & Sapolsky, 1991). This hypothesis was further supported by rich expression of GR as well as MR in hippocampal neurons (Herman et al., 1989; Van Eekelen et al., 1988). Lesion studies (including our own) indicate that extensive hippocampal damage, fornix sections or ibotenic acid lesions of the dorsal and ventral hippocampus increase PVN CRH mRNA expression and enhance corticosterone responses to stress (see summary in Herman et al., 2003; Radley & Sawchenko, 2011). These results were not replicated in all studies (Bradbury et al., 1993; Coover et al., 1971; Tuvnes et al., 2003), perhaps related to stimulus-dependent hippocampal control of HPA axis responses. For example, hippocampal damage causes hypersecretion of corticosterone when exposed to stimuli connected with novel environments/situations (e.g. restraint, open field exposure, novelty), but not to interoceptive cues (e.g. ether inhalation, hypoxia; Herman et al., 1998; Mueller et al., 2004; Radley & Sawchenko, 2011). Together, these data argue against a generalized or obligate role for the hippocampus in HPA axis negative feedback.

The ventral hippocampus is implicated in inhibition of the HPA axis, via relays to cortical output neurons in the ventral subiculum (SUBv). Neurons in this region are stress-sensitive and project to a number of sub-cortical regions that in turn project to the PVN, but not the PVN proper, including the bed nucleus of the stria terminalis (BST), the medial preoptic area, and dorsomedial and posterior hypothalamic nuclei (Cullinan et al., 1995, 1993; Myers et al., 2016; Figure 1). Importantly, PVN-projecting neurons are apposed by terminals of SUBv projection neurons, suggestive of synaptic relay. Moreover, the vast majority of PVN inputs from the BST and preoptic area are GABAergic in nature, implying a mechanism for trans-synaptic inhibition of the HPA axis (Cullinan et al., 1993). The nature of these connections indicates that inhibitory effects of the hippocampus on the HPA axis require an intermediary neuron.

Finally, the temporal dynamics of ventral subiculum regulation of HPA axis function are worthy of comment. Specific damage to this region (as opposed to whole hippocampal lesion) results in an enhanced or prolonged peak in corticosterone release (Herman et al., 1998, 1995), rather than delayed shut-off reported following large hippocampal lesions (Sapolsky et al., 1984). The impact of SUBv lesions appears to enhance responsiveness to stressors rather than return to baseline. Thus, actions of the SUBv likely trigger trans-synaptic PVN inhibition, functioning to shut down the corticosterone response more quickly.

Prefrontal cortex

Lesion studies indicate that the infralimbic (IL) and prelimbic (PL) divisions of the medial prefrontal cortex function as modulators of HPA axis output (Note that some of the earlier studies do not parse medial prefrontal subregions, in which case the region will be referred to as IL/PL or the ventromedial prefrontal cortex (vmPFC). Damage to or inactivation of the IL/PL and PL increase corticosterone and ACTH responses to restraint or novelty but not hypoxia or ether, suggesting that like the hippocampus, these regions integrate stimuli relevant to anticipated threat rather than physical challenge (Diorio et al., 1993; Figueiredo et al., 2003). However, closer examination of the literature reveals a more nuanced role in HPA integration: in response to repeated challenge, damage to the right (but not left) vmPFC reduces corticosterone secretion (Sullivan & Gratton, 1999), suggesting an excitatory (and lateralized!) role for this region with repeated homotypic stress.

The IL and PL express both GR and MR in abundance (Ahima et al., 1991; Ahima & Harlan, 1990; Herman, 1993) and GR-containing subpopulations of neurons are activated under conditions of acute stress (Ostrander et al., 2003). vmPFC GR is down-regulated under conditions of chronic stress (Mizoguchi et al., 2003), and chronic stress drives increased IL and PL deltaFosB expression (Flak et al., 2012), consistent with a putative role in temporal processing of repeated stressors. (i.e. chronic stress; Figure 3). Corticosterone implants into the vmPFC inhibit HPA axis responses to restraint but not ether (Diorio et al., 1993), suggesting that the vmPFC uses glucocorticoid signals to modify output relevant to anticipatory stimuli. We have gone on to demonstrate regional specificity of GR signaling via viral vector-based knockdown of GR in the IL and PL. Our data indicate that GR knockdown in both regions increases HPA responsiveness to acute restraint (McKlveen et al., 2013). More recently we have used a recently-developed GR flox rat in conjunction with a CaMKII-driven Cre recombinase to confirm that specific deletion of GR in PL projection neurons enhances acute stress reactivity (Scheimann et al., 2019). Importantly, knockdown of GR in the IL (but not PL) was sufficient to enhance HPA axis reactivity (as well as passive coping behavior) in the context of chronic stress, indicative of a role in processing stress chronicity (McKlveen et al., 2013). These data are recapitulated by knockdown of glutamate vesicular packaging in IL neurons (Myers et al., 2017), attesting to a role for the IL in control of HPA axis responses to repeated challenge (Figure 2).

Figure 3. Neural mechanisms controlling chronic stress regulation of the HPA axis. Pathways responsible for drive of the HPA axis under chronic stress are not as well understood as those mediating acute response. There is strong evidence that the PVT, which is not involved in acute stress excitation or inhibition, is required for both stress habituation and stress facilitation, suggesting a role in communicating stress chronicity. Importantly, the PVT has extensive reciprocal projections to the IL, PL and vSUB, as well as projections to the area of the BST. Neuronal activation studies indicate the existence of a small network of structures that are differentially activated by chronic unpredictable stress (relative to restraint), including the IL, PL, PH and NTS. The NTS itself appears to contribute to chronic stress-related HPA drive via peptidergic neurons. Importantly, the PH and NTS are both connected with the IL, and both mediate acute stress excitation, suggesting a possible integrated circuit mediating chronic stress drive. Finally, chronic stress increases tone of CRH-expressing stress circuitry in the CeA, suggesting that CRH systems may be recruited by chronic stress and participate in HPA axis hyperdrive. See Figure 1 legend for abbreviations and symbol definitions). Figure modified from Herman et al. (2005), with permission, and were created with Biorender.com.

In all cases, IL and PL manipulations fail to alter baseline HPA axis activity and, unlike the hippocampus, do not appear to affect long-term shut-off. Primary actions are evident as potentiation or prolongation of peak corticosterone activation. As was the case for the hippocampus, the IL and PL have few if any direct projections to the PVN, showing strong projections to many of the same structures innervated by the SUBv, most notably the BST (Radley et al., 2009).

Finally, it appears that IL/PL GR enhances activation of cortical projection neuron output, which would be roughly consistent with drive of trans-synaptic inhibition. GR activation can enhance glutamatergic transmission in layer V PL pyramidal neurons after acute stress. This effect is thought to be mediated by increased expression of post synaptic AMPA and NMDA receptors (Joëls et al., 2012; Yuen et al., 2011, 2009). Similar effects on glutamatergic signaling following stress and subsequent GR activation are also seen in the hippocampus (Karst & Joëls, 2005). Further work in PL suggests that GR binding inhibits GABAergic interneurons, thereby affording activation of cortical outflow (Hill et al., 2011). Recent data from our group demonstrates that chronic stress enhances inhibitory input onto IL projection neurons, an observation that is correlated with reduced GR expression in cortical interneurons (McKlveen et al., 2016).

Amygdala

In contrast to the hippocampus and vmPFC, the HPA axis appears to be activated by input from the amygdala, mediated primarily by the medial (MeA) and central (CeA) nuclei (Ulrich-Lai & Herman, 2009). The MeA and CeA lack substantial connectivity with the PVN, and appear to modulate HPA axis function by interactions with intermediary neurons in the same nuclei receiving input from the hippocampus and vmPFC, e.g. the BST, DMH, PH and mPOA (Myers et al., 2016, 2014a; Figure 2). As the majority of MeA and CeA projections are GABAergic (Swanson & Petrovich, 1998), it is thought that activation of the HPA axis by these structures is mediated by disinhibition, i.e. blockade of tonic PVN inhibition via direct GABAergic projections in these regions. Whereas there is overlap of limbic innervation to all subcortical regions targeted by descending HPA axis-regulatory regions, the relative weighting of projections differs across the various subcortical targets (Herman et al., 2003; Ulrich-Lai & Herman, 2009).

The MeA and CeA appear to have complementary roles in mediating different stressor modalities. Lesions of the MeA diminish stressor responses to psychogenic (restraint) but not systemic (interleukin-1 beta) stimuli, whereas CeA damage impairs responses to systemic (IL-1beta) but not psychogenic (restraint) stimuli (Dayas et al., 2001, 1999). Stressor selectivity is also evident in patterns of cellular activation in the two nuclei, with the MeA evincing strong Fos activation by psychogenic stress such as restraint, the CeA by immune stimuli, hypovolemia and pain (Dayas et al., 2001).

Complementary roles of MeA and CeA afferents are commensurate with the very different biological processes controlled by the two regions. The MeA is in receipt of considerable olfactory input and is thought to be critical in control of social behavior and aggression (Chen et al., 2019; Haller, 2018; Sah et al., 2003). The CeA receives information from the thoracic and abdominal viscera and is an important regulator of autonomic nervous system responses (Viltart et al., 2006; Browning and Travagli, 2014). Both regions receive excitatory input from the basolateral amygdala (BLA; Sah et al., 2003), which is in receipt of a variety of multimodal sensory input and suggests it may play a “gate-keeper” function in driving MeA and CeA neurons controlling HPA axis stressor responses. Indeed, the BLA acts as a critical node in many of the behaviors associated with stress, including fear and anxiety (Janak and Tye, 2015). The exact role of the BLA in the control of the HPA axis has been difficult to pin down, possibly due to intrinsic heterogeneity of neurons responsive to positive or negative valence.

In contrast to the hippocampus and PFC, chronic glucocorticoids and stressor exposure stimulate dendritic plasticity in the BLA (McEwen et al., 2016; Vyas et al., 2002). Increased plasticity (in the form of dendritic hypertrophy and greater synaptic connectivity) here is thought to promote, rather than oppose, stress responsivity, especially to chronic stressors (Ashokan et al., 2016; Mitra et al., 2005). These effects have been linked to increased GR translocation and activation of downstream pathways in the BLA (Novaes et al., 2017). Moreover, in contrast with actions in the PVN, glucocorticoids promote CRH synthesis in the CeA, which is linked to chronic stress-related pathology (Dallman et al., 2003; Figure 3). The molecular mechanism underlying positive regulation of CRH is linked to tissue-specific alternative splicing of the GR binding partner Src, which dictates the directionality of GR transcription in this region (Lachize et al., 2009; Zalachoras et al., 2013).

Thalamus

The posterior paraventricular nucleus of the thalamus (PVT) appears to be selectively involved in control of HPA axis responses to chronic rather than acute stress. Lesions of this region block sensitization of the HPA axis by chronic cold exposure, and prohibit habituation of HPA axis responses to homotypic stress (Bhatnagar et al., 2002; Bhatnagar & Dallman, 1998). Moreover, habituation of HPA axis responses can be blocked by chronic inhibition of both GR and MR in this region, suggesting that glucocorticoid signals are essential to this process (Jaferi et al., 2003). The PVT has extensive projections to amygdala, prefrontal and ventral subiculum regions involved in HPA axis regulation (above), and in turn receives input from each region (Li & Kirouac, 2012; Vertes et al., 2015). Given the strong data supporting effects of the PVT on HPA drive by chronic stress, it is possible that these reciprocal projections may play a role in tuning the overall responsiveness of the HPA axis to repeated or prolonged stress (Hsu et al., 2014), perhaps via coordinated action across excitatory and inhibitory stress nodes.

Hindbrain mechanisms: nucleus of the solitary tract (NTS)

Physiological adversity reliably drives activation of the HPA axis, likely as a mechanism to generate energy resources to meet the challenge. Homeostatic perturbations are relayed in part by neurons resident in the hindbrain, which are responsible for relaying interceptive sensory information to the PVN. The NTS plays a particularly important role in this process, as it receives ascending vagal information from the thoracic and abdominal viscera (and perhaps immune system) and heavily innervates the PVN (Myers et al., 2017). The NTS sends heavy catecholaminergic (noradrenergic (NE) and adrenergic (E; Cunningham et al., 1990; Cunningham & Sawchenko, 1988) and non-catecholaminergic (for example, glucagon-like peptide-1 (GLP-1; Ghosal et al., 2013) projections to the PVN, preferentially targeting the CRH-containing medial parvocellular subdivision. Importantly, catecholaminergic NTS neurons do no co-express GLP-1 (Maniscalco & Rinaman, 2017), indicating that these cell populations are distinct and raising the possibility of differential actions on the HPA axis (see below).

Damage to ascending NTS pathways thought to contain PVN-projecting axons (6-hydroxydopamine lesions of ventral noradrenergic bundle/NTS or saporin-conjugated anti-DBH infusion directly into the NTS) cause reductions in ACTH and corticosterone release and/or PVN Fos induction to homeostatic perturbations (e.g. ether, cytokine injection, glucose deprivation) but not stressors of a more psychogenic nature (restraint, footshock, swim; Bundzikova-Osacka et al., 2015; Flak et al., 2014; Gaillet et al., 1993, 1991; Ritter et al., 2003), suggesting stressor specificity. Notably this specificity does not extend to NTS GLP-1 neurons, which modulate responses to psychogenic as well as physical stressors (Ghosal et al., 2017). Overall, the literature indicates that direct NTS projections to the PVN stimulate HPA axis responses to acute stressors (Figure 2) in a neuronal cell type-specific manner.

The NTS also mediates HPA axis responses to chronic stress exposure. Chronic variable stress (CVS) promotes drive of TH mRNA and protein expression in NTS catecholaminergic neurons, but markedly reduces expression of the GLP-1 precursor preproglucagon (GCG; Zhang et al., 2010). Despite presumptive increases in biosynthesis, lesions of ascending NTS NE and E neurons do not attenuate enhanced HPA axis drive following CVS (Flak et al., 2014). In contrast, intracerebroventricular supplementation of GLP-1 in rat enhances HPA axis drive following CVS. In contrast, infusion of a GLP-1 receptor antagonists of selective deletion of the GLP-1 receptor in the PVN blocks CVS potentiation of HPA axis stress responses (Tauchi et al., 2008). Thus, GLP-1 but not catecholamine neurons appear to be required for chronic stress-induced HPA axis sensitization (Figure 3).

Catecholaminergic and GLP-1 neurons in the NTS express GR (Harfstrand et al., 1986; Rinaman, 2011), suggesting the potential for glucocorticoid regulation of output. Local implants of pellets containing the GR antagonist mifepristone increase drive of the HPA axis by an acute stressor, whereas implants of corticosterone reduce stress-induced corticosterone secretion. Moreover, local mifepristone implants increase resting corticosterone secretion following chronic stress and potentiate HPA axis sensitization following CVS (Bechtold et al., 2009; Ghosal et al., 2014), consistent with blockade of negative feedback effects on the NTS. These data are consistent with GR-mediated feedback inhibition of HPA axis responding via the NTS. Given that 1) GLP-1 neurons are differentially implicated in potentiation of responses to chronic stress, and 2) GCG mRNA expression and GLP-1 immunoreactivity are decreased by CVS exposure (Ghosal et al., 2017; Zhang et al., 2010), it is plausible that effects of GR are mediated by attenuation of GLP-1 drive to PVN neurons.

The role of the NTS in both physical vs. psychogenic stress is somewhat at odds with the notion of specific circuitry mediating two distinct classes of stressors, a hypothesis put forward by our group and Paul Sawchenko’s in the late 1990s (Herman et al., 1998; Li et al., 1996). The newer data suggest that the NTS may instead be a common conduit of stress integration of multiple modalities, with psychogenic stressor responses and chronic stress reactivity likely controlled by non-catecholaminergic neurons, perhaps via descending inputs from limbic projection regions (such as the ventromedial prefrontal cortex and central amygdaloid nuclei; Schwaber et al., 1982; van der Kooy et al., 1984).

Putting It All Together: Role of the Bed Nucleus of the Stria Terminalis

One common feature shared by all of the above regions is a connection with the BST (Figures 1–3). The BST is known to play a role in anxiety processing (as opposed to fear), becoming important in generation of responses to stimuli that do not necessarily predict a defined outcome (Avery et al., 2016; Walker et al., 2003; a function well connected with general conceptualization of stress). Recent work highlights a role for the BST in the development and expression of contextual conditioning, implicating the BST as an interpreter of contextual information (Goode & Maren, 2017; Luyck et al., 2020), a process consistent with generating responses in anticipation of potential threat.

The BST is intimately involved in control of HPA axis responses to stressful stimuli. Early work from our group demonstrated a complex role for the BST in stress: lesions of the anteromedial divisions of the BST inhibit HPA axis responses to stress, whereas lesion of posterior nuclei enhanced HPA axis stressor reactivity (Choi et al., 2007). Recent work further identifies a specified role for the anterolateral region in HPA axis activation, conferring inhibition of stressor responses via PL projections (Radley et al., 2009; Radley & Johnson, 2018). Notably, the role of the anteromedial region appears to change under conditions of chronic drive, becoming stress-inhibitory rather than stress-excitatory (Choi et al., 2008). The mechanism underlying this switch is unclear but could be related to the complex neurotransmitter content of BST-PVN connections, which express both GABA and CRH, the latter a predominantly excitatory neuropeptide. As neuropeptides are generally secreted at more intense levels of stimulation than classical transmitters (Mains & Eipper, 1999), it is possible that alterations in BST drive in acute vs. chronic stress may differential favor downstream activation or inhibition of PVN neurons. It is also important to consider that chronic stress and/or corticosterone administration can reverse the chloride gradient in PVN neurons, causing GABA to be excitatory rather than inhibitory (Bains et al., 2015; Inoue & Bains, 2014).

The BST receives input from all the purported stress regulatory regions listed above. The distribution of inputs can vary somewhat by BST subregion. For example, the CeA heavily innervates the anterolateral and anteroventral regions of the BST, whereas the MeA projects heavily to posterior subnuclei (Dong et al., 2001). In some cases, inputs from limbic regions can innervate the same BST neurons: in fact, dual tracing studies suggest direct innervation of individual BST neurons by both PL and SUBv projectors, consistent with the ability of BST neurons to summate influences from different HPA regulatory pathways (Radley & Sawchenko, 2011). Connections between the IL/PL and BST are thought to mediate inhibition of HPA axis responses to acute stressors (Radley et al., 2009; Spencer et al., 2005). Moreover, the anterior BST has extensive interactions with the NTS, being a major target of both catecholaminergic and GLP-1ergic neurons (Bundzikova-Osacka et al., 2015; Ghosal et al., 2013; Maniscalco & Rinaman, 2017). Thus, the BST is at a crossroads for handling of information from a variety of stress effector systems.

The intrinsic organization of the BST is itself complex. Anterior subnuclei are generally associated with anxiety and emotional responses, including HPA axis activation, and appear to mediate BST potentiation of fear and drug relapse and reinstatement (Miles & Maren, 2019; Ressler et al., 2011), whereas the posterior BST subnuclei have rich connections with structures linked with agonistic and social behavior (e.g. MeA; Dong et al., 2001). In addition, the BST also has intrinsic interconnections across subnuclei that may differentially interface with input from prefrontal and amygdala projections (Gungor & Paré, 2016).

The BST is known to play a major role in processing of emotional responses to context. Prior studies indicate that the BST plays a major role in non-associative sensitization of startle (Davis et al., 2010; Walker & Davis, 1997) and in expression of contextual fear (Ali et al., 2012; Sullivan et al., 2004; Zimmerman & Maren, 2011).The BST is also critical for reinstatement of contextual fear responses following extinction (Goode et al., 2015; Waddell et al., 2006). Finally, activation of the BST is required for stress-related relapse of drug and alcohol self-administration (see reviews of Centanni et al., 2019; Goode & Maren, 2019). Together the data suggest that the BST is involved in processing information on the overall significance of stressors, integrating inputs from corticolimbic as well as hindbrain structures involved in determining the stimulus significance and salience. The net output of the BST then broadly affects how the organism responds to stress, adjusting behavioral, autonomic as well as HPA axis reactions in accordance with its “interpretation” of the summated neuronal input

Summary

Regulation of the HPA axis is of critical importance to the organism, given the behavioral and energetic signal carried by glucocorticoid hormones. Glucocorticoids act in a variety of body compartments and by numerous neuronal systems to achieve a net signaling level that optimizes the response of the organism to stressors. Secretion is regulated by diverse somatic and psychological signals and held in check by a diverse feedback system designed to “dial in” responses to best meet the real or perceived challenge. The checks and balances system is by no means perfect, and disruption by chronic drive or disease can move response magnitude and/or signaling capacity into either inadequate or over-reactive ranges, with potentially important consequences. In some cases, responses to stressors may be “adaptive” in nature: for example, habituation of stressor responses limits potential damage associated with glucocorticoid over-reactivity, and one can argue that sensitization of HPA axis responses to chronic stress may promote readiness of the organism to cope with a “now hostile” world. In contrast, inappropriate secretory activity resulting from developmental adversity, aging or disease may result in out-of-context hypo- or hypersecretion, generating “stress” reactions under conditions where stress responses are not beneficial (or even harmful, i.e. allostatic overload). Overall, the weight of evidence indicates that management of stress is above all a network problem, requiring integration of a substantial array of signals that link psychological responses to the internal state of the organism. Consequently, convergent inputs from regions such as the BST, hypothalamic nuclei and perhaps the NTS send sensory-processed and salience-adjusted information to the PVN, where it is integrated into a go- no go response. This response is then adjusted by neuronal and hormonal feedback signals designed to provide temporal control of the responses. It is this exquisite balance of appropriate drive and hormonal/neuronal inhibition that provides life-long efficiency of stressor responses and limits disruptions in physiology and behavior.

Acknowledgements

Submitted for the Special Issue: 12th International Symposium on Catecholamines and Other Neurotransmitters in Stress.

Disclosure statement

The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Additional information

Funding

This work was supported by a Veterans Administration Merit Award to JPH [I01BX003858] and National Institutes of Health [grants MH049698, MH101729 and MH119814].