Hormone circuit analysis explains why most HPA drugs fail for mood disorders and predicts the few that work

  1. Tomer Milo
  2. Shiraz Nir Halber
  3. Moriya Raz
  4. Avi Mayo
  5. Uri Alon

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

Elevated cortisol causes morbidity in chronic stress and mood disorders, including metabolic and cardiovascular diseases. There is therefore interest in developing drugs that lower cortisol by targeting its endocrine pathway, the hypothalamic-pituitary-adrenal (HPA) axis. Several promising HPA-modulating drugs have, however, failed to lower long-term cortisol in mood disorders such as major depressive disorder despite their effectiveness in situations where high cortisol is caused by a tumor (Cushing’s syndrome). Why these drugs failed is not well understood. Here we use a mathematical model of the HPA axis to show that the pituitary and the adrenal glands compensate for the effect of drugs by adjusting their functional mass, a feedback compensation that is absent in Cushing tumors. To find potential drug targets, we carried out a systematic in silico analysis of points of intervention in the HPA axis. We find that only two interventions that target corticotropin-releasing hormone (CRH) can lower long-term cortisol. Other drug targets either fail to lower cortisol due to gland-mass compensation or lower cortisol but harm other aspects of the HPA axis. Thus, we identify potential drug targets, including CRH-neutralizing antibodies and CRH-synthesis inhibitors, for lowering long-term cortisol in mood disorders and in those suffering from chronic stress. More generally, this study indicates that understanding the slow compensatory mechanisms in endocrine axes can be crucial in order to prioritize drug targets.

Article activity feed

  1. Review Commons

    Note: This response was posted by the corresponding author to Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    General Statements
    The reviewer comments helped us improve the paper by including new computations, figures, and analyses related to vasopressin, drug dosages, and treatment cessation. We have also removed confusing terminology from the text. We believe that the paper is now more comprehensive, clear, and rigorous.

    Reply to the reviewers

    Reviewer #1 (Evidence, reproducibility and clarity):

    The authors address the question of lowering long-term elevated cortisol levels by affecting the parameters in a previously published mathematical model of the hypothalamic-pituitary-adrenal (HPA) axis. The parameters are related to various pathways. The elevation in cortisol levels is related to diseases e.g. mood disorders and Cushing's syndrome.
    The authors conducted a systematic in silico analysis of various points of intervention in the HPA axis. They found that only two interventions targeting corticotropin-releasing hormone (CRH) can lower long-term cortisol. Other drug targets either fail to lower cortisol due to gland-mass compensation or lower cortisol but harm other aspects of the HPA axis. Thus, they identify potential drug targets, including CRH-neutralizing antibodies and CRH synthesis inhibitors, for lowering long-term cortisol in mood disorders and in those suffering from chronic stress.
    The method used is in silico investigations of the mathematical model.
    The draft is well written with a single typo in line 270. I have no further comments!

    Response: The typo is fixed.

    Reviewer #1 (Significance):

    In silico predictions without direct use of data is a weakness but the conducted analysis is convincing. An improved understanding of why some drugs work and others do not is important and is postulated to agree with clinical evidence.

    Response: We thank the reviewer for this endorsement.

    Reviewer #2 (Evidence, reproducibility and clarity):

    Summary
    The authors utilise a mathematical model of the hypothalamic-pituitary-adrenal axis to address the utility of interventions altering its various outputs (CRH, ACTH and cortisol) to ameliorate axis disruption in response to chronic stress. They show that a lowering of circulating CRH by either blocking its synthesis or increasing its clearance is effective at returning the HPA axis to basal activity at all levels. In contrast, interventions altering ACTH or cortisol production, their circulating levels or actions are ineffective in the model. This is consistent with data on the long-term efficacy of drugs reducing excess corticosteroids in patients and animal models. The use of mathematical models to describe complex interactions in endocrine systems is a valuable advance in our understanding of potential mechanisms and therapies and this is an excellent example.

    Response: We thank the reviewer for this endorsement.

    Major comments

    1. The model of the HPA axis that the authors have described previously is a little simplistic when considering the known physiology. Specifically, this model ignores the contribution of vasopressin to the axis, which has been described as being the primary hypothalamic factor driving HPA axis activity in chronic stress (see doi.org/10.1016/S0079-6123(08)00403-2). Including this may be beyond the scope of the current model, however it should be considered and at least commented on. It is notable that the model fits the clinical and animal model data, which may suggest that the contribution of vasopressin in the long term may be overestimated, possibly as a result of differential effects of the two hypothalamic factors, with CRH driving ACTH release and POMC gene expression, whilst vasopressin only increases ACTH release without augmenting POMC expression. This is worthy of discussion.

    Response: We thank the reviewer for this comment which helped us discuss vasopressin. We agree that adding it as a variable in the model is beyond the scope of the current study. We describe its effects in the introduction and discussion sections. Interestingly, when one considers the best characterized effect of vasopressin, namely enhancing CRH-dependent ACTH release, one can use this model to investigate the effects of inhibiting vasopressin. We predict that vasopressin inhibition is unlikely to be an effective strategy for lowering long-term cortisol and alleviating stress-related mental disorders, as evidenced by the failure of clinical trials.

    In the introduction we add:

    1. “CRH stimulates the secretion of adrenocorticotropic hormone (ACTH) by corticotroph cells in the anterior pituitary, an effect enhanced by vasopressin (Aguilera et al, 2008; Antoni, 2017).” (lines 35-37)
    2. Clinical trials for two vasopressin 1b receptor antagonist candidates, SSR149415 and TS-121, in the table of HPA-related clinical trials (Table 1)

    In the discussion we add (lines 398-409): ”One important factor not explicitly considered in the model is the contribution of vasopressin to the axis. Vasopressin potentiates the CRH-dependent release of ACTH from pituitary corticotrophs by acting on the V1b receptor (V1bR) (Aguilera et al, 2008; Antoni, 2017). Including this hormone explicitly is beyond the current scope. However, a simple analysis indicates that the effect of elevated vasopressin can be modeled by increasing the ACTH secretion parameter b2. This suggests that vasopressin V1b receptor antagonists should have effects similar to inhibitors of ACTH production. As such, vasopressin receptor antagonists should be compensated by the HPA axis without long-term effects on cortisol. Accordingly, V1bR antagonists did not show statistically significant efficacy in clinical trials for major depressive disorder and generalized anxiety disorder (Griebel et al, 2012; Chaki, 2021; Kamiya et al, 2020). However, vasopressin may have additional relevant effects on the HPA axis and the central nervous system which warrant a more detailed modeling analysis.”

    1. The model that this study relies on is dependent on slow changes in the various levels of the endocrine axis and the authors have focused on alterations in cell number as the process leading to a prolongation of their dysfunction. For the stress axis, the evidence for changes in corticotroph cell number is weak and the recent paper of Lopez et al (DOI: 10.1126/sciadv.abe44) suggests that chronic stress, at least over a period of 3 weeks does not lead to an alteration in the number of corticotrophs, despite cell population changes in the adrenal gland. There are other processes which could lead to prolonged alteration of corticotroph output and it would be better to focus (as the authors have in places) on functional mass, rather than cell number which may suggest it is not the trophic effect of CRH that is important for increased functional mass.

    Response: We thank the reviewer for this. We now refer only to functional mass changes. We corrected all places in which hyperplasia of corticotrophs is mentioned. We also state in lines 125-126 that the model is agnostic as to whether growth in functional mass is due to hyperplasia or hypertrophy.
    We also added a citation for Lopez et al. 2021 (line 86) to support the growth of cortisol-secreting cells in the zona fasciculata of the adrenal gland under stress conditions.

    1. The parameters in the model for interventions are described as simply being less than or greater than one- to what extent are the effects of these interventions dependent on their specific value? For example, presumably if the I1 value is close to zero, then the CRH-synthesis inhibitor would be ineffective. Likewise, if it were close to 1 then there would be negligible release of CRH in response to stress, and the preservation of a response to acute stress would be lost. Can the authors show the range of values for I1, C1 and A1 where the interventions are effective at normalising HPA axis function whilst (for I1 and A1) still preserving the acute stress response?

    Response: We thank the reviewer for this comment that helped us to add a new section in the results on dose response, and three new figures (Figure 4, Figure S2 and Figure S3):

    “CRH interventions have a dose-dependent response in the model
    We computed the effects of drug doses by varying the relevant model parameter, where zero dose means no change in the parameter and high doses mean large changes in the parameter. We find that both candidate interventions for lowering cortisol - CRH-synthesis inhibitors and CRH-blocking antibodies - cause a dose-dependent reduction of steady-state cortisol (Figure 4A). This indicates that putative treatment may require finding the appropriate dose to return the patients to their normal cortisol baseline range. Other drug candidates have no effect on long-term cortisol steady state (Figure S2).

    At all doses, the steady states of CRH and ACTH remain normal (Figure 4B-C). The acute stress response, defined as peak cortisol upon acute stress input relative to steady-state cortisol, is dose dependent (Figure 4D and Figure S3). At a dose that returns cortisol to the normal range, the acute response is also normalized.

    We also tested the effects of abrupt treatment cessation. For both CRH interventions, stopping treatment led to a rapid return to hypercortisolemia (Figure 4E-F and Figure S4).

    Figure 4. Predicted effective interventions have a dose-dependent effect on cortisol, and cortisol abruptly rises when treatment is ceased. (A) Cortisol steady state in the model upon changes in doses of CRH-synthesis inhibitors and CRH-blocking antibodies. (B-C) The same changes in drug doses have no effect on ACTH (B) and CRH (C) steady state levels. (D) Cortisol peak response to an acute stress relative to steady state for different drug doses. (E-F) HPA dynamics upon cessation of CRH-synthesis inhibitors (E) and anti-CRH antibodies (F) after 50 days.”

    In the supplemental information:

    “Cortisol dose response to HPA-targeting drugs

    Figure S2. Cortisol steady state dose response to HPA-targeting drugs, related to Figure 4.

    Figure S3. Cortisol peak response to acute stressor under varying concentrations of HPA-targeting drugs, related to Figure 4.”

    1. In the models that the authors describe with CRH interventions, what is the impact of stopping the intervention on axis output in the short and long-term? Presumably ceasing the use of CRH antagonists would lead to much more severe axis dysregulation than CRH neutralising antibodies or CRH synthesis inhibitors.

    Response: We have now added new analysis on drug cessation (new figure 4E-F, Figure S4). After a 50 day treatment, sudden cessation caused a rapid return to hypercortisolemia:
    We added in lines 277-278: “We also tested the effects of abrupt treatment cessation. For both CRH interventions, stopping treatment led to a rapid return to hypercortisolemia (Figure 4E-F).”

    Reviewer #2 (Significance):

    Whilst the study builds on the use of a previously described mathematical model, its utility in identifying potential targets for therapy within the important area of chronic stress makes it an important example of the value of the modelling approach to decisions on appropriate targets for therapy. The model does not include important known factors which have been described as being important in the HPA axis response to chronic stress and would be considerably improved if these could be incorporated.
    The study builds on conceptual insights into the role a delayed or slow functional mass change might play in dysregulation of endocrine axes and this could be applied to other physiological systems and will be of interest to modellers and physiologists alike. The authors are leaders in this field and there are few other modellers considering systems level interactions over this timescale.

    Response: We thank the reviewer for this endorsement.

    As a pituitary physiologist, my review has focused on the interactions between the various players in HPA axis function, I do not have the expertise to comment on mathematical modelling aspects.

    Reviewer #3 (Evidence, reproducibility and clarity):

    This extremely interesting paper asks why various attempts to treat depression and bipolar disorder with glucocorticoid antagonists or cortisol synthesis inhibitors have failed. The starting point for their analysis is a simple computational model that, importantly, includes the facts that CRH stimulates not only ACTH release but also corticotroph growth and ACTH stimulates not only cortisol production but also the growth of cells in the adrenal cortex. They call this the "gland mass model". According to the model, if the hypothalamus receives a continuous stress input, all of the HPA hormones will be elevated-CRH transiently and the others in a sustained fashion. Adding a sufficient dose of a CRH inhibitor (decreasing the rate constant b1 in the model) or a CRH antibody (increasing the rate constant a1) normalizes the hormone levels, whereas blocking cortisol function or production does not. This is demonstrated by numerical simulations and backed up by deriving analytical expressions for the hormone concentrations at steady state. The paper provides a plausible explanation for why past therapeutic efforts have failed and points to a couple of approaches that might succeed. These conclusions are hypotheses-they haven't been tested experimentally and we really don't know how accurately the system is described by this nice, simple model-but they are really intriguing hypotheses that could lead to therapeutic breakthroughs. I strongly recommend publication.

    Response: We thank the reviewer for this endorsement.

    My only criticisms are minor:

    1. The authors should specify what exact change in the model's parameters they are making to implement their therapeutic interventions. E.g. in Fig 1B top left and 2A, what is the change in the value of b1 that corresponds to the addition of a CRH-synthesis inhibitor? (I'd guess it's being dropped to zero, but if this is stated, I missed it)

    Response: We thank the reviewer for that comment which helped us to clarify what is the required parameter change to normalize cortisol. We have now added in lines 173-175: “According to equation (1), as a general guideline, treating cortisol levels that are x-fold higher than baseline requires a drug dose that alters the relevant parameter (e.g., CRH production or removal rate) by a similar x-fold.”

    1. I think it would also be useful to show a dose-response relationship for the various interventions.

    Response: We thank the reviewer for this comment that helped us to add a new section in the results on dose response, and three new figures (Figure 4, Figure S2 and Figure S3):

    “CRH interventions have a dose-dependent response in the model
    We computed the effects of drug doses by varying the relevant model parameter, where zero dose means no change in the parameter and high doses mean large changes in the parameter. We find that both candidate interventions for lowering cortisol - CRH-synthesis inhibitors and CRH-blocking antibodies - cause a dose-dependent reduction of steady-state cortisol (Figure 4A). This indicates that putative treatment may require finding the appropriate dose to return the patients to their normal cortisol baseline range. Other drug candidates have no effect on long-term cortisol steady state (Figure S2).

    At all doses, the steady states of CRH and ACTH remain normal (Figure 4B-C). The acute stress response, defined as peak cortisol upon acute stress input relative to steady-state cortisol, is dose dependent (Figure 4D and Figure S3). At a dose that returns cortisol to the normal range, the acute response is also normalized.

    We also tested the effects of abrupt treatment cessation. For both CRH interventions, stopping treatment led to a rapid return to hypercortisolemia (Figure 4E-F and Figure S4).

    Figure 4. Predicted effective interventions have a dose-dependent effect on cortisol, and cortisol abruptly rises when treatment is ceased. (A) Cortisol steady state in the model upon changes in doses of CRH-synthesis inhibitors and CRH-blocking antibodies. (B-C) The same changes in drug doses have no effect on ACTH (B) and CRH (C) steady state levels. (D) Cortisol peak response to an acute stress relative to steady state for different drug doses. (E-F) HPA dynamics upon cessation of CRH-synthesis inhibitors (E) and anti-CRH antibodies (F) after 50 days.”

    In the supplemental information:

    “Cortisol dose response to HPA-targeting drugs

    Figure S2. Cortisol steady state dose response to HPA-targeting drugs, related to Figure 4.

    Figure S3. Cortisol peak response to acute stressor under varying concentrations of HPA-targeting drugs, related to Figure 4.”

    ****Referees cross-commenting****

    It looks like we are all enthusiastic about this work.

    Response: Thank you.

    Reviewer #3 (Significance):

    Strengths: It's a beautiful new insight on a really important topic, extracted from a simple, understandable mathematical model of the HPA axis.

    Weaknesses: It is based on a model and the model could be wrong. This does not however diminish my enthusiasm for this provocative work.

    Advance: It is highly original.

    Audience: I hope attracts a wide audience--modelers, endocrinologists, psychiatrists, drug developers.

    My expertise: I am a systems biologist, have taught psychopharmacology to medical students, and have an interest in endocrine signaling.

  2. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #3

    Evidence, reproducibility and clarity

    This extremely interesting paper asks why various attempts to treat depression and bipolar disorder with glucocorticoid antagonists or cortisol synthesis inhibitors have failed. The starting point for their analysis is a simple computational model that, importantly, includes the facts that CRH stimulates not only ACTH release but also corticotroph growth and ACTH stimulates not only cortisol production but also the growth of cells in the adrenal cortex. They call this the "gland mass model". According to the model, if the hypothalamus receives a continuous stress input, all of the HPA hormones will be elevated-CRH transiently and the others in a sustained fashion. Adding a sufficient dose of a CRH inhibitor (decreasing the rate constant b1 in the model) or a CRH antibody (increasing the rate constant a1) normalizes the hormone levels, whereas blocking cortisol function or production does not. This is demonstrated by numerical simulations and backed up by deriving analytical expressions for the hormone concentrations at steady state. The paper provides a plausible explanation for why past therapeutic efforts have failed and points to a couple of approaches that might succeed. These conclusions are hypotheses-they haven't been tested experimentally and we really don't know how accurately the system is described by this nice, simple model-but they are really intriguing hypotheses that could lead to therapeutic breakthroughs. I strongly recommend publication.

    My only criticisms are minor:

    1. The authors should specify what exact change in the model's parameters they are making to implement their therapeutic interventions. E.g. in Fig 1B top left and 2A, what is the change in the value of b1 that corresponds to the addition of a CRH-synthesis inhibitor? (I'd guess it's being dropped to zero, but if this is stated, I missed it)
    2. I think it would also be useful to show a dose-response relationship for the various interventions.

    Referees cross-commenting

    It looks like we are all enthusiastic about this work.

    Significance

    Strengths: It's a beautiful new insight on a really important topic, extracted from a simple, understandable mathematical model of the HPA axis.

    Weaknesses: It is based on a model and the model could be wrong. This does not however diminish my enthusiasm for this provocative work.

    Advance: It is highly original.

    Audience: I hope attracts a wide audience--modelers, endocrinologists, psychiatrists, drug developers.

    My expertise: I am a systems biologist, have taught psychopharmacology to medical students, and have an interest in endocrine signaling.

  3. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #2

    Evidence, reproducibility and clarity

    Summary

    The authors utilise a mathematical model of the hypothalamic-pituitary-adrenal axis to address the utility of interventions altering its various outputs (CRH, ACTH and cortisol) to ameliorate axis disruption in response to chronic stress. They show that a lowering of circulating CRH by either blocking its synthesis or increasing its clearance is effective at returning the HPA axis to basal activity at all levels. In contrast, interventions altering ACTH or cortisol production, their circulating levels or actions are ineffective in the model. This is consistent with data on the long-term efficacy of drugs reducing excess corticosteroids in patients and animal models. The use of mathematical models to describe complex interactions in endocrine systems is a valuable advance in our understanding of potential mechanisms and therapies and this is an excellent example.

    Major comments

    1. The model of the HPA axis that the authors have described previously is a little simplistic when considering the known physiology. Specifically, this model ignores the contribution of vasopressin to the axis, which has been described as being the primary hypothalamic factor driving HPA axis activity in chronic stress (see doi.org/10.1016/S0079-6123(08)00403-2). Including this may be beyond the scope of the current model, however it should be considered and at least commented on. It is notable that the model fits the clinical and animal model data, which may suggest that the contribution of vasopressin in the long term may be overestimated, possibly as a result of differential effects of the two hypothalamic factors, with CRH driving ACTH release and POMC gene expression, whilst vasopressin only increases ACTH release without augmenting POMC expression. This is worthy of discussion.
    2. The model that this study relies on is dependent on slow changes in the various levels of the endocrine axis and the authors have focused on alterations in cell number as the process leading to a prolongation of their dysfunction. For the stress axis, the evidence for changes in corticotroph cell number is weak and the recent paper of Lopez et al (DOI: 10.1126/sciadv.abe44) suggests that chronic stress, at least over a period of 3 weeks does not lead to an alteration in the number of corticotrophs, despite cell population changes in the adrenal gland. There are other processes which could lead to prolonged alteration of corticotroph output and it would be better to focus (as the authors have in places) on functional mass, rather than cell number which may suggest it is not the trophic effect of CRH that is important for increased functional mass.
    3. The parameters in the model for interventions are described as simply being less than or greater than one- to what extent are the effects of these interventions dependent on their specific value? For example, presumably if the I1 value is close to zero, then the CRH-synthesis inhibitor would be ineffective. Likewise, if it were close to 1 then there would be negligible release of CRH in response to stress, and the preservation of a response to acute stress would be lost. Can the authors show the range of values for I1, C1 and A1 where the interventions are effective at normalising HPA axis function whilst (for I1 and A1) still preserving the acute stress response?
    4. In the models that the authors describe with CRH interventions, what is the impact of stopping the intervention on axis output in the short and long-term? Presumably ceasing the use of CRH antagonists would lead to much more severe axis dysregulation than CRH neutralising antibodies or CRH synthesis inhibitors.

    Significance

    Whilst the study builds on the use of a previously described mathematical model, its utility in identifying potential targets for therapy within the important area of chronic stress makes it an important example of the value of the modelling approach to decisions on appropriate targets for therapy. The model does not include important known factors which have been described as being important in the HPA axis response to chronic stress and would be considerably improved if these could be incorporated.
    The study builds on conceptual insights into the role a delayed or slow functional mass change might play in dysregulation of endocrine axes and this could be applied to other physiological systems and will be of interest to modellers and physiologists alike. The authors are leaders in this field and there are few other modellers considering systems level interactions over this timescale.

    As a pituitary physiologist, my review has focused on the interactions between the various players in HPA axis function, I do not have the expertise to comment on mathematical modelling aspects.

  4. Review Commons

    Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

    Learn more at Review Commons

    Referee #1

    Evidence, reproducibility and clarity

    The authors address the question of lowering long-term elevated cortisol levels by affecting the parameters in a previously published mathematical model of the hypothalamic-pituitary-adrenal (HPA) axis. The parameters are related to various pathways. The elevation in cortisol levels is related to diseases e.g. mood disorders and Cushing's syndrome.
    The authors conducted a systematic in silico analysis of various points of intervention in the HPA axis. They found that only two interventions targeting corticotropin-releasing hormone (CRH) can lower long-term cortisol. Other drug targets either fail to lower cortisol due to gland-mass compensation or lower cortisol but harm other aspects of the HPA axis. Thus, they identify potential drug targets, including CRH-neutralizing antibodies and CRH synthesis inhibitors, for lowering long-term cortisol in mood disorders and in those suffering from chronic stress.
    The method used is in silico investigations of the mathematical model.
    The draft is well written with a single typo in line 270. I have no further comments!

    Significance

    In silico predictions without direct use of data is a weakness but the conducted analysis is convincing. An improved understanding of why some drugs work and others do not is important and is postulated to agree with clinical evidence.

  5. Version published to 10.1101/2024.06.10.598205v1 on bioRxiv