A CLD is a graphical model used to represent relationships between different variables (e.g., factors, processes, subsystem states, aggregate quantities) of a given system. In a CLD, variables and the relationships between them, are represented by nodes and directed edges, respectively, i.e., directed arrows that indicate an influence of the variable at the tail of the arrow (source variable), on the variable at the head of the arrow (target variable). The directed edges of a CLD are usually marked with a polarity indicating either a positive influence (an increase of the source variable induces an increase of the target variable), or a negative influence (an increase of the source variable induces a decrease of the target variable). By depicting how different elements influence other elements of the modelled system, CLDs provide a quasi-dynamic description of the possible outcomes of the evolution of a given variable on the system and therefore allow for a qualitative model of the progression from a given input to a given output.

CLDs reveal feedback loops that correspond to the influence of a variable’s output on the same variable (e.g., variable A influences the evolution of variable B, which in turn influences the evolution of variable A). Reinforcing loops describe amplifying mechanisms and are characterized by an even number of negative influences. Balancing feedback loops describe mechanisms that oppose further change in a certain direction with an action in the opposite direction and are characterized by odd numbers of negative influences. When feedback loops are interconnected, each feedback loop’s individual influence on the whole system can progressively increase or decrease, for instance an amplifying feedback loop can become dominant over a balancing feedback loop and over time, thus drive an entire system towards imbalance.

Feedback loops are consequently an important part of CLDs and are critical for understanding the properties of a system and in describing the conditions under which homeostatic biological mechanisms are impaired and a biological system can change from a healthy into a pathophysiological state.

2.1. CLD development

The current CLD was built by a modelling team with expertise in public health, health inequalities, complexity science, (neuro)endocrinology, chronobiology, metabolism, obesity and T2D pathophysiology. The development of the CLD was based on a literature review translated into a conceptual model realized by NM. The CLD and its description were then presented to seven additional researchers with expertise in (neuro)endocrinology, neurobiology, stress and plasticity of the brain, chronobiology, metabolism, gastroenterology, obesity and T2D pathophysiology. These domain experts were consulted through semi-structured interviews of 60–90 min conducted by NM and MN and/or written feedback. Their feedback was discussed in the modelling team and integrated in the paper by adapting and finalizing the CLD.

2.2. Literature review

Our literature review was organized following different steps that we identified from recommendations given by Miller et al. (2009) and we included clinical, laboratory and epidemiological studies and reviews.

Miller et al. (2009) recommended to use an “‘approach that reverse engineers’ adverse health outcomes into their specific biological determinants, and then identifies psychologically-modulated, neuroendocrine and immunological dynamics that modulate those pathological processes at the cellular and molecular levels”.

Building on the work of these authors, our literature review was organized according to the following items:

1.

identify “the most proximal biological pathways linked to clinical disease outcomes (i.e., mechanism of pathogenesis)” (Miller et al., 2009);

2.

identify “psychologically modulated neuroendocrine dynamics” (Miller et al., 2009);

3.

identify how these neuroendocrine dynamics modulate biological pathways leading to T2D.

2.3. Scope

Stress is a multidimensional construct which can be described through three different components: 1) stressors, i.e., stimuli that are hypothesized to induce distress or elicit a stress response in the body, 2) the processing of stressors which includes the subjective experience of stress and underlies interactions between cognitive evaluations and emotional/affective states, and 3) the biological stress responses, i.e. bodily or hormonal physiological responses in an individual who is exposed to a stressor (Kelly and Ismail, 2015, Ursin, 1991).

These different dimensions of stress interact, at various levels and spatiotemporal scales, via feedforward and feedback loops, ultimately aiming at restoring homeostasis. They do so through behavioral and physiological adaptations (de Kloet et al., 2019, Levine, 2005), and by interacting with an individual’s context, behavior and physical environment (Epel et al., 2018). Fig. 1 maps the relations between these different aspects and defines the scope of the CLD built in the current paper.

psychological stress. These mediators are then thought to activate secondary effectors, including the immune, cardiovascular, glucose and lipid regulatory systems. Due to its gradual accumulation, T2D can then be considered a tertiary and final outcome (McEwen, 1998, McEwen, 2000, McEwen, 1993). Regarding the non-linear interplay between acute and chronic stress, we focused more extensively on markers of HPA axis activity and the regulation and evolution of HPA axis activity in the context of chronic stress, the HPA axis reactivity appearing to be the most widely investigated aspect in relation to stress. In relation with behaviors, we included systems regulating food intake and circadian rhythms as they interact with the HPA axis and SNS and have been associated with T2D pathogenesis. We included temporal scales ranging from minutes/hours to months/years and spatial scales ranging from molecular to tissue scales. Finally, we detailed processes occurring in the periphery during acute stress and remained relatively generic regarding processes occurring in the central nervous system (CNS), since current evidence is not developed or consistent enough to infer possible mechanisms linking these aspects to HPA axis regulation and further on to T2D pathogenesis.

For panel C (Fig. 1), we included differences mainly related to sex/gender. Sex differences refer to biology-related differences caused by differences in e.g., sex chromosomes, sex-specific gene expression, sex hormones and their actions on biological processes, while gender differences emerge from sociocultural processes (Kautzky-Willer et al., 2016).

For panel D (Fig. 1), we focused on behaviors relevant to T2D and to HPA axis responsivity, including circadian and feeding-related behavior (Stenvers et al., 2019).

2.4. CLD format

The CLD (shown in Fig. 2) was rendered in Wondershare EdrawMax (version 10.5.4). In the current section we describe the graphical structure of the CLD.

2.

feedback loops related to how the stress response could influence stress regulating systems and thus subsequent stress responses.

3.

feedback loops related to how alterations in the periphery, which can be due to past psychogenic stress responses, can activate the stress systems.

Feedback loops revealed by the CLD can be within specific biological scales, e.g., contain only processes at molecular scales, or across multiple scales, e.g., link processes at molecular scales to processes at tissue and network scales. They vary in complexity as they can involve multiple steps before a variable is fed back to its own derivative. We distinguished short-loops (e.g., containing two or less variables) by using the loop symbols “R” for reinforcing and “B” for balancing. These loops are within the same scale (molecular/intracellular). Most larger loops involve processes occurring at different spatial and/or temporal scales. Loops across different scales are composed of arrows going from left to right indicating an evolution towards more long-lasting processes, and of arrows going from right to left, linking long-term modifications and larger spatial scale variables to short-term processes and smaller spatial scales processes. Larger loops also link variables in distinct clusters (e.g., relations between acute stress responses, in clusters B1 and B4, and resulting modifications of stress-related systems that influence subsequent stress responses, in clusters B3 and B6).

3. CLD narrative

In line with the different steps followed during the reviewing process (section 2.2) and the different types of feedback loops that we identified (section 2.5), the descriptive narrative of the CLD in Fig. 2 was structured into four subsections. Section 3.1. focuses on T2D pathogenesis. Section 3.2. describes how alterations induced by the repetition of the stress response could influence T2D pathogenesis. Section 3.3 focuses on the regulation of stress responses, and more specifically HPA axis stress responses at different spatiotemporal scales. In section 3.4 we describe possible drivers modulating the regulation of HPA stress responses and their relations to T2D pathogenesis. Relevant feedback loops revealed by the CLD are described in the narrative.

3.1. Mechanisms underlying T2D pathogenesis

Most often, the pathogenesis of T2D is characterized by insulin resistance (link 178), which refers to reduced insulin effects on glucose uptake by insulin target tissues (adipose tissue, liver and skeletal muscle), which in turn leads to an increased metabolic demand on pancreatic β-cells and hyperinsulinemia. The over-activity of pancreatic β-cells and the consequences of reduced glucose uptake by insulin target tissues result in the progressive deterioration of β-cell function (link 177) and the establishment of sustained hyperglycemia (Galicia-Garcia et al., 2020, Schwartz et al., 2017).

One major risk for the occurrence of insulin resistance is (abdominal) obesity or excessive adiposity (Zheng et al., 2018) with more than 90 % of patients with T2D being obese or overweight (Bramante et al., 2017). Globally T2D is more prevalent in men than women and in Europe, men are also diagnosed at lower body mass index (BMI) and younger ages than women. However, obesity, which is the most important risk factor for T2D, is more prevalent in women with sex differences in obesity rates varying between countries (Kautzky-Willer et al., 2016) (link 179). Age is also an important risk factor for T2D onset (link 178), although, the prevalence of T2D in adolescents and young adults (below 40 years old) is increasing significantly (Kautzky-Willer et al., 2016, Lascar et al., 2018).

Chronic low-grade inflammation can result from excessive adiposity (especially from excess visceral adipose tissue (VAT), i.e., around the abdominal organs) and has been implicated in obesity-driven insulin resistance and T2D. Although the underlying mechanisms are not completely clear, adipose tissue expansion, characterized by adipocyte hyperplasia and/or hypertrophy (links 108–110) can result in multiple outcomes, e.g., hypoxia, mechanical stress, adipocyte death, that can initiate an inflammatory response (links 157–162) (Longo et al., 2019, Zatterale et al., 2020).

Obesity and/or over-nutrition can also result in increased blood cholesterol and triglycerides (hyperlipidemia), transported by lipoproteins such as the very low-density lipoprotein (VLDL) produced by the liver (links 101, 102). Hydrolysis of triglycerides increases the concentrations of free fatty acids (FFAs) (link 92), which promote the expression of pro-inflammatory cytokines (link 105) (Hotamisligil, 2017, Tripathy et al., 2003). Excessive adiposity can also increase the amount of circulating FFAs via lipolysis (link 89).

Multiple studies have shown an altered expression of pro- or anti-inflammatory adipokines and cytokines, immune receptors and intracellular mediators of inflammation in obese humans and animal models of obesity. In particular, adipose tissue macrophages play an important role in obesity-driven inflammation. Increased infiltration of macrophages (which can constitute up to 40% of cells in the adipose tissue of obese subjects) and their shift towards “pro-inflammatory” phenotypes lead to the secretion of cytokines inducing insulin resistance in adipocytes (e.g., tumor necrosis factor-α (TNF-α) and interleukins (IL)-6 and −1β) (link 133) (Hotamisligil, 2017, Zatterale et al., 2020). Although a wide spectrum of different macrophage profiles has been observed in obese humans and animals, a binary model is often used, which distinguishes “pro-inflammatory” M1 macrophages from “anti-inflammatory” M2 macrophages, which are linked to tissue remodeling and the resolution of inflammation. Obesity-driven inflammation is therefore thought to result from an imbalance between M1 and M2 macrophages, induced by conditions arising from excess adipose tissue accumulation (Castoldi et al., 2016, Catrysse and van Loo, 2018, Hotamisligil, 2017).

Other immune cell types, including eosinophils and lymphoid cells are also hypothesized to regulate the differentiation of adipose tissue macrophages into an M1 or an M2 phenotype (Hotamisligil, 2017, Sun et al., 2012). Although the actions of several inflammatory mediators have been associated with insulin resistance, their influence on insulin sensitivity may depend on duration, dose of exposure and target sites. For instance, increased blood levels of IL-6 have been associated with obesity and correlate with T2D risk. However, in-vivo animal studies have also shown that IL-6 acutely promotes insulin signaling in muscle. Duration of exposure and dose likely depend on complex interactions between different immune mediators and feedback mechanisms regulating their production rates (Hotamisligil, 2017).

Five possible feedback loops are included in the CLD and depicted in Fig. 3. The first feedback loop corresponds to the recruitment and differentiation of additional pro-inflammatory immune cells by pro-inflammatory signaling, resulting in an amplifying feedback loop (loop 131 in Fig. 3). In contrast in the second feedback loop, pro-inflammatory cytokines can later also promote the expression of anti-inflammatory cytokines, which will in turn inhibit the production of pro-inflammatory cytokines resulting in a balancing feedback loop (loop 127, 128 in Fig. 3). The third feedback loop corresponds to the induction of oxidative stress by pro-inflammatory processes via the production of reactive oxygen species by immune cells (Mittal et al., 2014). Uncontrolled levels of reactive oxygen species (ROS) (or oxidative stress which is defined as an imbalance between the production of ROS (link 136) and the capacity of the antioxidant system to detoxify them (link 143)) can result in the activation of pro-inflammatory processes which in turn can generate more ROS, constituting an amplifying feedback loop (loop 129, 130 in Fig. 3). ROS levels can be controlled by a fourth feedback loop: ROS activate antioxidant defenses which eliminate them, constituting again a balancing feedback loop (loop 142, 143 in Fig. 3).

Fig. 3. Reinforcing and balancing short feedback loops. From left to right: pro-inflammatory cytokines can promote the expression of anti-inflammatory cytokines which will in turn inhibit the production of pro-inflammatory cytokines (balancing feedback loop 127, 128); production of reactive oxygen species (ROS) in the context of pro-inflammatory signaling promotes pro-inflammatory signaling (reinforcing feedback loop 129, 130); activation of antioxidant defenses by ROS which eliminate ROS (balancing feedback loop 142, 143) ; endoplasmic reticulum (ER) stress can lead to the production of ROS and ROS contribute to ER stress (reinforcing feedback loop 154, 155); ER stress activates the unfolding protein response (UPR) which restores the ER folding capacity (balancing feedback loop 139, 140).

Inflammatory mediators produced at the level of the adipose tissue can have systemic effects and induce insulin resistance in the liver and skeletal muscle. Moreover, accumulation of fat in ectopic tissues (e.g., the liver and skeletal muscle) and in VAT (links 111, 112) also leads to a local expression of pro-inflammatory cytokines and inflammation (link 113) which can affect hepatic and muscle insulin sensitivity (link 133) (Longo et al., 2019).

Men show a higher accumulation of VAT and liver fat compared to women of similar age and BMI. Women have more subcutaneous adipose tissue (SAT) which is more likely to accumulate in the gluteal-femoral region. In both sexes, central/abdominal (i.e., subcutaneous upper body and visceral) adipose tissue is associated with an increased risk for T2D while lower body (gluteal-femoral) fat deposition is linked with decreased metabolic risk. In addition, VAT has greater rates of lipolysis and lipogenesis than SAT and the more pronounced VAT accumulation in men correlates with higher FFAs, and triglyceride (TG) levels. These differences could contribute to explaining the higher susceptibility for men to develop T2D at lower BMI and younger age, although with age and post-menopause, women are more likely to accumulate VAT (Kautzky-Willer et al., 2016, Tramunt et al., 2020). Another factor could be a higher sensitivity to insulin in women, as observed in a large cohort study in normoglycemic individuals that found higher insulin sensitivity in women than in men, even after controlling for age and BMI (Kautzky-Willer et al., 2012). Studies in rodents reported a protective effect of estrogens against diet-induced insulin resistance in the liver and skeletal muscle (Tramunt et al., 2020). In humans, this sex difference in insulin sensitivity disappears with the development of T2D (Tura et al., 2018).

In addition to stimulating glucose uptake, insulin has multiple other functions such as the regulation of gene expression and enzymatic activity and the modulation of appetite and energy homeostasis (Petersen and Shulman, 2018). Insulin actions are exerted on multiple target tissues and mediated by complex intracellular signaling cascades. In skeletal muscle insulin triggers the translocation of the glucose transporter GLUT4 to promote glucose uptake and stimulate glucose oxidation and glycogenesis (link 74) (Huang and Czech, 2007). Impairment of insulin signaling in skeletal muscle results in decreased glucose uptake. In the liver, insulin inhibits gluconeogenesis (link 76) (Fazakerley et al., 2019). As a result, when hepatic insulin resistance occurs, glucose production is not properly inhibited, resulting in increased blood glucose (link 84). In adipose tissue, insulin exerts anti-lipolytic effects (link 75) inhibiting the release of FFAs by adipocytes and stimulates the uptake of glucose via GLUT4 and lipogenesis (link 77) (Fazakerley et al., 2019). Therefore, insulin resistance in the adipose tissue can lead to increased circulating FFAs (link 90).

Glucose is the main trigger of insulin release by β-cells and also regulates transcription and translation processes involved in insulin synthesis (link 99) (Cerf, 2013). FFAs can also stimulate insulin secretion (link 100) (Cen et al., 2016, Nolan et al., 2006). Increased glycemia and circulating FFAs due to insulin resistance therefore increase the demand on β-cells. β-cells’ mass and insulin production and secretion are consequently increased to compensate for insulin resistance leading to hyperinsulinemia (link 71 in Fig. 4). Hyperinsulinemia in turn can induce insulin resistance (link 80 in Fig. 4). The observation of fasting hyperinsulinemia in normoglycemic obese subjects has led to the hypothesis that hyperinsulinemia induced by FFAs could be an initial trigger leading to insulin resistance (Fryk et al., 2021). In both these hypotheses, insulin resistance and hyperinsulinemia influence one another due to amplifying feedback loops, for instance loop 71, 80, 73, 76, 84, 99, shown in Fig. 4, depicts how hepatic insulin resistance can result in higher glucose levels as a consequence of uninhibited hepatic glucose production (loop 71, 80, 73, 74, 87, 99 and loop 71, 80, 73, 77, 88, 99 in Fig. 4 describe the same phenomenon, respectively, in relation to glucose uptake by muscle and adipose tissue).

Fig. 4. Reinforcing feedback loops describing how insulin resistance in peripheral systems (corresponding to lowered insulin sensitivity) can induce hyperinsulinemia (elevated blood insulin levels) and how hyperinsulinemia can in turn induce insulin resistance.

In the long-term, an increased demand on β-cells and the deleterious consequences of high glucose levels (glucotoxicity) and high levels of blood FFAs (lipotoxicity) can lead to β-cell dysfunction.

Insulin production by β-cells (link 71) involves the folding of the precursor proinsulin to insulin in the endoplasmic reticulum (ER). As the ER folding capacity is limited, a high physiological demand (e.g., hyperglycemia), or disturbances in protein folding, can lead to the accumulation of misfolded proteins in the ER, a process defined as “ER stress” (link 105). The unfolded protein response (UPR) is a cellular defense mechanism, which aims to restore ER folding capacity and protein homeostasis, constituting a balancing feedback loop (loop 139–140 in Fig. 3). However, upon persistent activation and chronic ER stress, the UPR signaling system can switch to induce (apoptotic) cell death (link 163) (Adams et al., 2019). Another result of the higher production of insulin by β-cells, could be amyloid stress (amylin is a peptide co-secreted with insulin (link 107) (Mather et al., 2002)), which would be induced by high levels of β-cell amylin, and has been proposed to also contribute to ER stress (link 146) and ROS production (link 145) (Christensen and Gannon, 2019, Stumvoll et al., 2005, Swisa et al., 2017).

High glucose and FFAs concentrations in the circulation can further lead to oxidative stress due to an increased generation of ROS, for instance through increased oxidative metabolism (link 135) or due to the formation of advanced glycation products (AGEs) (links 134) (Bloemer et al., 2014). In addition to being part of amplifying feedback loops with inflammatory processes, ROS can cause oxidative damage to cell components and alter cellular functions (link 150). ROS also trigger signaling pathways by inducing transcription factor activation, gene transcription (link 152) and epigenetic modifications (link 151) (Schwartz et al., 2017). ROS induced gene transcription can ubiquitously promote cell proliferation, hypertrophy, loss of function and even apoptosis. These effects of ROS can be diminished by scavenging mechanisms (i.e. antioxidant systems) which eliminate ROS (link 143) or repair mechanisms (e.g., DNA repair processes) which counteract ROS-induced damage (link 166) (Chapple, 1997, Lee et al., 2004). β-cells, both of rodents and humans, have been reported to be highly vulnerable to ROS because of their low expression level of classical antioxidant enzymes (e.g., superoxide dismutases) in comparison to other cell types (Benáková et al., 2021, Swisa et al., 2017).

β-cell damage and dysfunction lead to an insufficient production of insulin to regulate blood glucose and lipid levels, resulting in multiple amplifying feedback loops which can progressively deteriorate pancreatic function. In the CLD (Fig. 2), these loops correspond to links going from cluster B4 (short-term processes) to cluster B6 (short and long-term modifications in the periphery) and subsequently from cluster B6 to cluster B5 (e.g., links 136, 152, 171), which corresponds to state variables regulating processes in cluster B4 (e.g., link 174).

β-cell failure has been traditionally associated with massive β-cell death (link 169) and decreases of more than 60% in β-cell mass have been reported to occur in T2D (Butler et al., 2003). More recent studies suggest that β-cells in T2D might in fact de-differentiate (link 165) and even gain characteristics of other pancreatic islet cell types, a process that might in principle be reversible, depending on the state of the β-cells (Swisa et al., 2017). In normoglycemic individuals, women show higher insulin secretion capacity than men. Specifically, endogenous estrogens stimulate the synthesis and secretion of insulin and preserve the function of the β-cells against metabolic or oxidative stress. However, similar impairments in β-cell function in T2D have been reported in both sexes (Tramunt et al., 2020).

Other pathways and feedback mechanisms contributing to T2D involve hypothalamic dysfunction, gastro-intestinal disturbances, defects in glucagon metabolism and circadian misalignment (Schwartz et al., 2017).

Hypothalamic dysfunction plays an important role in the development of T2D and obesity. Multiple studies have specifically investigated the hypothalamic infundibular nucleus (IFN), which is equivalent to the arcuate nucleus (ARC) in rodents. The IFN or ARC contain two neuronal populations involved in the regulation of energy homeostasis and food intake: anorectic pro-opiomelanocortin (POMC) expressing neurons, and orectic neuropeptide Y (NPY)/agouti-related protein–expressing (AgRP-expressing) neurons. These populations of neurons contain receptors that bind leptin or insulin. Diet-induced obesity and T2D have been associated with an imbalance between POMC and NPY/AgRP neurons (Alkemade et al., 2012, Kalsbeek et al., 2020). Animal studies have shown that inflammation in the hypothalamus plays an important role in the relationships between hypothalamic dysfunction and in the development of obesity and T2D, however the relevance of findings on hypothalamic inflammatory mechanisms in rodents remains so far unknown for humans (Kalsbeek et al., 2020).

Inflammation could impair the function of ARC neurons and might also induce insulin resistance. In rodents, insulin induces transcription of anorectic peptides in the ARC (e.g., α-melanocyte stimulating hormone (αMSH) produced by POMC neurons) (link 122) and promotes the synthesis of leptin by adipose tissue (link 79), which in turn can also increase transcription of anorectic peptides (link 117, 121) and inhibit transcription of NPY and AgRP (link 115, 120) (Diepenbroek et al., 2013). Leptin also inhibits the synthesis and secretion of insulin (link 97) while increasing insulin sensitivity (link 89) (Amitani et al., 2013). In the rodent brain, injection of NPY and AgRP lead to reduced insulin sensitivity (link 180 in Fig. 5) and increased glucose production, while injection of αMSH increases insulin action (link 181 in Fig. 5) (Diepenbroek et al., 2013).

and hyperlipidemia, and insulin resistance, which could evolve towards T2D onset in the long-term if positive feedback loops described in section 3.2.1, e.g., feedback loops contributing to the progression of insulin resistance and β-cell dysfunction, become dominant over balancing loops.

3.2.2. Effects on immune and inflammatory processes

GCs have pleiotropic effects on various immune processes that have traditionally been attributed to GR-mediated alterations in gene expression. The actions of GCs on inflammation are known to favor anti-inflammatory (link 123), and inhibit pro-inflammatory processes (link 125) (Cain and Cidlowski, 2017)). For instance, GCs inhibit the pro-inflammatory cytokines IL-1β and TNF-α (increased levels of these cytokines have been associated with increased T2D risk) (link 54) and promote the differentiation of macrophages into, anti-inflammatory, M2 phenotypes (which are assumed to be downregulated in obesity) (Cain and Cidlowski, 2017).

Although GCs actions on immunity are mostly described in terms of suppressing adaptive immunity and promoting innate immunity, GCs can also enhance certain adaptive immunity processes (e.g., favoring the differentiation of Th cells into a Th2 phenotype, which also promotes the differentiation of macrophages into an M2 phenotype) and enhance the reactivity of innate immunity to danger signals, exerting a permissive effect on inflammatory processes (link 124). Specifically, GCs promote the expression of toll-like receptors (TLRs) 2 and 4 (Busillo and Cidlowski, 2013, Chinenov and Rogatsky, 2007, Newton et al., 2017), which have also been characterized as receptors for FFAs and are involved in obesity-driven inflammation (Hotamisligil, 2017).

GCs upregulate the expression of another receptor involved in the recognition of danger signals, the inflammasome (Busillo and Cidlowski, 2013) which has been shown to play an important role in the development of diet-induced insulin resistance and pancreatic β-cell deterioration in mouse models and in human diabetes (Hotamisligil, 2017). This potentiation of the immune system can again indirectly enhance pro-inflammatory processes. In addition, GCs can inhibit many wound-healing processes (Cain and Cidlowski, 2017) which could delay the resolution of certain inflammatory responses.

GCs could also exert pro-inflammatory effects through the presence of MRs in specific immune cell types. Indeed, MR promotes the activation of macrophages to a pro-inflammatory M1 phenotype and regulates the differentiation of Th cells into a Th1 and Th17 phenotypes, which are also pro-inflammatory. MR also downregulates anti-inflammatory T regulatory lymphocytes. As macrophages do not express 11β-HSD2, the pro-inflammatory influence of MR on macrophages is likely to be induced by GCs rather than via aldosterone, whereas the possibility that T lymphocytes express 11β-HSD2 remains an open question (Bene et al., 2014).

Catecholamines are also involved in the regulation of multiple immune-related processes including immune cell activation, proliferation and apoptosis. Notably, effects of catecholamines on immune and inflammatory processes are bidirectional: while β-adrenergic receptor signaling has been mostly associated with anti-inflammatory effects, α-adrenergic receptor stimulation has been linked to pro-inflammatory effects (Barnes et al., 2015, Elenkov, 2007).

Acute stress, through the release of norepinephrine, can also transiently increase IL-6 levels and other inflammatory mediators in the circulation (Barnes et al., 2015, Elenkov, 2007).

Depending on the dose, duration and the general context (e.g., state of metabolic and immune systems) under which GCs and catecholamines are elevated, acute stress responses could favor a suppression or enhancement of inflammatory processes. The permissive actions of GCs on immunity (link 124 in Fig. 6) could lead to a condition of chronic inflammation, provided sufficient triggers enable its initiation, and provided the inhibitory actions of pro-resolving factors are insufficient. For example, amplifying feedback loops 143, 129, 135, 149, 175 in Fig. 6 favor pro-inflammatory processes because of endotoxemia (see section 3.2.1), or because of a chronic triggering of pro-inflammatory processes to which GCs elevations can also contribute (links 124, 126 in Fig. 6). Repeated elevations of plasma GCs could e.g., enhance (links 66–68 in Fig. 6) excess visceral fat depots (links 161, 162 in Fig. 6) or hyperlipidemia-induced inflammation (link 105 in Fig. 6).

dependent on the time of exposure to stressors.

Plasma GCs levels also follow a marked ultradian rhythm with an approximate frequency of 60–90 min that is superimposed on the circadian GC rhythm (Fitzsimons et al., 2016, Russell and Lightman, 2019). In rodents, blocking the activity of the SCN removes circadian rhythms of corticosterone but does not affect the ultradian rhythmicity (Waite et al., 2012). In contrast, GC ultradian rhythmicity depends on the ultradian rhythmicity of ACTH (Kalafatakis et al., 2019). GC ultradian rhythms have been shown to play an important role for metabolic function in rodent models (Kalafatakis et al., 2019, Oster et al., 2017) and also for regulating GC mediated-genomic actions (Russell and Lightman, 2019). Repetition of increased levels of GCs during exposure to stress could also interfere with GC ultradian rhythm-mediated regulation of, e.g., metabolism, and through this pathway might increase the risk for T2D.

3.3. Regulation of the HPA axis activity at different spatiotemporal scales

HPA axis responses to acute psychosocial stress show large intra- and inter-individual variability. A meta-analysis of 208 laboratory stress studies showed that in humans, motivated performance tasks reliably induced HPA axis responses (ACTH and cortisol) if they were perceived as uncontrollable or characterized by a social-evaluative threat. Tasks having both components were associated with the largest increase in hormone levels and the highest recovery times (Dickerson and Kemeny, 2004, Kudielka et al., 2009). The acute reactivity approach assumes that affective experiences modulate acute responses to stress. When such affective experiences occur repeatedly, they are thought to increase the intensity or duration of the stress response. As such, they contribute to biological changes that accumulate over time and result in an ‘allostatic load’ on several biological systems. In particular, anticipatory reactions can lead to a heightened response before exposure to a stressor, whereas rumination would lead to a delayed recovery following stress (Epel et al., 2018, McEwen, 1998, McEwen, 2000). It has also been proposed that women show a greater cortisol response to interpersonal stressor whereas men would be more sensitive to achievement stressors, although this finding remains inconsistent (Zänkert et al., 2019).

Higher elevations of GCs levels and a delayed return to baseline enhances the area under the curve of total GC exposure, and could negatively impact peripheral systems e.g., immunity and glucose and lipid metabolism-related systems.

In humans, chronic stress has been associated with both lower and higher basal levels of cortisol. Also, the cortisol responses have been reported to be either prolonged or blunted (Epel et al., 2018, Hackett and Steptoe, 2017, Miller et al., 2007). While chronic stress can sensitize the responses to new stressors, several (pre-)clinical studies show a decline of the HPA response to a psychological stressor with repeated exposure to a homotypic stressor. This decline has been defined as “habituation”, which refers to a form of non-associative learning (Grissom and Bhatnagar, 2009). However, habituation of the HPA axis probably only partially explains this decline.

The observed decline of HPA responses could depend on the perceived stressor controllability (e.g., if the perceived control over a stressor increases, current stress responses (during exposure) or subsequent stress responses to the same or similar stressor might be attenuated), and could also depend on other aspects of cognitive and emotional processing of the stressor (e.g., changes in vigilance, or fear of being evaluated or rejected). In the acute reactivity approach, a lack of habituation results in more intense and longer stress responses, which, when repeated, tend to increase the allostatic load (Epel et al., 2018, McEwen, 1998, McEwen, 2000). Also, the timing of stress exposure and the negative feedback inhibition could determine response magnitude (Grissom and Bhatnagar, 2009). A study in rodents, e.g., shows that re-exposure to a stressor before GCs return to baseline, decreases the magnitude of the subsequent HPA response to a stressor (de Souza and van Loon, 1982). In addition, this phenomenon might be regulated by alterations of stress-related neuronal circuits (Grissom and Bhatnagar, 2009).

In relation to stress, many MRI studies in humans have focused on the consequences of post-traumatic stress disorder (PTSD) and depression on the brain, and less so on those of chronic stress per se (Chattarji et al., 2015, Czéh and Lucassen, 2007, Lucassen et al., 2014). Studies focusing more specifically on chronic stress exposure alone, suggested similar outcomes to those associated with PTSD. For instance, chronic occupational stress has been associated with decreases in the volume of hippocampal and mPFC regions, alterations of functional connectivity between the AG and PFC regions and an increase in the volume of AG regions. Alterations of the PFC and HPC volumes seem to be reversible after a recovery period, while alterations of amygdala volume may persist longer (links 39, 47) (Blix et al., 2013, Golkar et al., 2014, Savic, 2015, Savic et al., 2018).

Similar alterations have been reported in older adults after exposure to chronic stress and in adults who have been exposed to stress in early life (Ansell et al., 2012, Gianaros et al., 2007, Hanson et al., 2012). Alterations of volumes could be preceded by detectable, yet still reversible alterations in function: alterations in mPFC function were e.g., associated with impaired attentional control in healthy adults exposed to one month of psychosocial stress. Both the functional and behavioral changes were reversible after another month (Liston et al., 2009).

These alterations in volume and function of limbic structure regulating the HPA axis response to stress might impact HPA axis responses and be associated with the higher, lower or more blunted cortisol responses to stress after exposure to chronic stress (links 40, 41, 42, 43). In humans, the roles of the different limbic structures in the initiation and termination of the HPA axis response to stress (links 9–11 in Fig. 7) are not fully understood and likely depend on the characteristics of stressors and multiple other factors, such as psychological aspects or chronic stress history.

Fig. 7. Feedback loops regulating plasma levels of GCs: inhibition of the stress response via feedback in the HPA axis and in the limbic system and regulation of GCs circadian rhythm.

Higher levels of cortisol during psychological stress exposure have been associated with a decreased activity in the ventromedial prefrontal cortex (vmPFC) and perigenual anterior cingulate cortex (Harrewijn et al., 2020), while increased activity in the orbitofrontal mPFC (part of the vmPFC) has been associated with higher cortisol levels during psychological stress exposure (Dedovic et al., 2009). Different and bidirectional roles of the different subareas of the vmPFC on HPA axis responses have been proposed in both humans and rodents (Dedovic et al., 2009).

Altered activities in the HPC, amygdala and inferior frontal gyrus, found in relation to higher levels of cortisol during psychological stress, are inconsistent across studies (increased or decreased activity) (Dedovic et al., 2009, Harrewijn et al., 2020). Increased cortisol responses have been linked to increased amygdala activity in studies using fear-evoking images, and decreased amygdala activity occurs after stress that combined an arithmetic task and social evaluation. The amygdala might be specifically involved in fear reaction and not necessarily in responses to psychosocial stress (Dedovic et al., 2009, Muscatell and Eisenberger, 2012). Both positive and negative correlations have been reported between HPC activation and the magnitude of the cortisol response (Dedovic et al., 2009, Harrewijn et al., 2020, Kern et al., 2008); higher levels were associated with a decreased activity of the HPC during tasks combining cognitive and social evaluation, and with an increased HPC activity after fear evoking paradigms (Harrewijn et al., 2020). Stressor controllability has been further shown to modulate fear extinction in humans (Hartley et al., 2014). Studies more specifically investigating the impact of controllability have used physical mild electric stressors and stressor control decreased responses in brain regions related to the processing of threatening signals (Limbachia et al., 2021).

Uncontrollable stress might therefore result in higher and longer HPA axis responses to acute stress and if repeated, amplify the responsivity of the HPA axis to subsequent stress exposure. However, physiological processes and gain-of-control over a stressor might also lead to a lower responsivity of the HPA axis when re-exposed to a homotypic stressor.

Biology-related changes in HPA axis reactivity and stress response recovery may be due to the specific impact GCs and neurotransmitters (link 39), released during exposure to uncontrollable stress, could have on the CNS including the HPA axis.

Feedback loops corresponding to these mechanisms are represented in a generic manner in the CLD. For instance, loops 37/38, 39, 40, 41/42 in Fig. 9 describe how repeated exposure to high cortisol and other stress mediators, as occurs during chronic stress, may alter the physiology of the brain. GR occupancy can further repress certain target genes and contribute to GR downregulation. This could modify subsequent stress responses if those occur before the restoration of GRs density. Because the termination of the stress response is an important consequence of GR occupancy in the PVN and pituitary, such a local downregulation could lead to a longer GCs stress response and subsequently also alter peripheral GC actions. The extent, nature and persistence of these modifications likely depend on the frequency of exposure to psychological stressors, in addition to the intensity of the induced responses (links 44–47).

Signaling from the periphery, e.g., via immune responses, via the microbiome, or via metabolic signals, can also activate the HPA axis or modulate its activity. In particular, cytokines, such as IL-1β, IL-6, and TNF-α, were shown to increase the release of CRH in the PVN in rodents, indicative of an activation of the HPA axis (Dunn, 2000, Fan et al., 2021). The subsequent release of GCs can in turn influence immune and inflammatory processes generating feedback loops, that can be balancing or unbalancing (see section 3.2.1). For instance, loops 33, 23, 14, 16, 17, 48, 125 and 33, 23, 14, 16, 17, 48, 124, 127 in Fig. 8 are balancing. Another loop including cortisol plasma levels and pro-inflammatory signaling, in contrast, can be reinforcing: GCs may also exert a permissive effect on the expression of pro-inflammatory cytokines via immune potentiation (loop 33, 23, 14, 16, 17, 48, 125, 126 in Fig. 8).