5D Library › HPA-Axis Coupled Oscillations, Cortisol Awakening Response Flattening and Burnout Recovery Kinetics · 中文

Chapter 1. A Loop with a Delay

Three hormones, one brake, and a transit time

The axis is a chain of three steps with a brake wrapped around it. The hypothalamus releases a hormone, the pituitary answers with corticotropin, the adrenal answers with cortisol, and cortisol travels back to suppress the first step. Nothing in that description is controversial. What is worth computing is the consequence of the brake arriving late, because a negative feedback with a delay does not merely settle more slowly; past a certain delay it stops settling at all and begins to oscillate.

That is not a hypothetical in this system. The axis does oscillate, in pulses roughly an hour apart, and Walker, Terry and Lightman asked where those pulses come from, showing that a delayed pituitary-adrenal loop can generate them without any separate pulse generator upstream (Walker et al. 2010). The question this chapter puts to that account is quantitative rather than qualitative: with the clearance rates fixed at their measured values rather than fitted, does the loop produce pulses at the observed rate?

Evidence

Published. Walker, Terry and Lightman showed that a delayed feedback loop between the pituitary and the adrenal is sufficient to generate ultradian pulses of corticotropin and cortisol, and analysed the conditions under which the oscillation exists (Walker et al. 2010).

The model below takes each clearance rate as the natural logarithm of two divided by a published plasma half-life, so there is nothing to tune there. The feedback is a Hill term. The only free quantity is the delay.

Calculation

# Chapter 1. p01: The three-stage loop with a delay. Releasing hormone
# drives corticotropin, corticotropin drives cortisol, and cortisol
# feeds back to suppress the releasing hormone after a transit delay.
# Each clearance rate below is the natural logarithm of two divided by a
# published plasma half-life, so the time constants are not free
# parameters. The feedback is a Hill term of exponent N. The loop is
# integrated forward by small steps with a history buffer for the delay.
# What is asked is: at what delay does the loop stop settling and start
# oscillating, and what period does it then have?
HL_CORT, HL_ACTH, HL_CRH = 70.0, 15.0, 10.0      # minutes
N, GAIN, K = 8.0, 20.0, 1.0
T, DT = 300.0, 0.002                              # hours
LN2 = 0.6931471805599453

BC = LN2 / (HL_CORT / 60.0)
QA = LN2 / (HL_ACTH / 60.0)
MH = LN2 / (HL_CRH / 60.0)
KC, PA, KH = BC * 1.6, QA, MH * GAIN

def simulate(tau, scale=1.0):
    bc, qa, mh = BC * scale, QA * scale, MH * scale
    kc, pa, kh = bc * 1.6, qa, mh * GAIN
    lag = int(round(tau / DT))
    c = [1.0] * (lag + 1)
    a = h = 0.6
    tail = []
    steps = int(T / DT)
    for i in range(steps):
        cl = c[-1 - lag]
        f = 1.0 / (1.0 + (cl / K) ** N)
        h += DT * (kh * f - mh * h)
        a += DT * (pa * h - qa * a)
        c.append(c[-1] + DT * (kc * a - bc * c[-1]))
        if i * DT > T * 0.6:
            tail.append(c[-1])
    amp = 0.5 * (max(tail) - min(tail))
    mean = sum(tail) / len(tail)
    cross = []
    for i in range(1, len(tail)):
        if (tail[i - 1] - mean) < 0.0 <= (tail[i] - mean):
            cross.append(i * DT)
    per = ((cross[-1] - cross[0]) / (len(cross) - 1)
           if len(cross) > 2 else 0.0)
    return amp, per, mean

if __name__ == "__main__":
    print("   clearance rates per hour from the published half-lives:")
    print("   cortisol {:.4f}, corticotropin {:.4f}, releasing hormone"
          " {:.4f}".format(BC, QA, MH))
    print("   delay h   amplitude   period h   mean")
    for tau in (0.00, 0.05, 0.10, 0.15, 0.20, 0.25, 0.33, 0.50):
        amp, per, mean = simulate(tau)
        print("   {:<10.2f}{:<12.4f}{:<11.3f}{:.4f}"
              .format(tau, amp, per, mean))
    lo, hi = 0.0, 0.50
    for _ in range(34):
        mid = 0.5 * (lo + hi)
        if simulate(mid)[0] > 0.01:
            hi = mid
        else:
            lo = mid
    crit = 0.5 * (lo + hi)
    print("   the loop starts to oscillate at a delay of {:.4f} hours"
          .format(crit))
    print("   which is {:.1f} minutes".format(crit * 60.0))
    amp, per, mean = simulate(0.25)
    print("   at a delay of 0.25 hours the period is {:.3f} hours"
          .format(per))
    print("   which is {:.1f} minutes".format(per * 60.0))
    print("   the observed pulse interval in people is 60 to 90 minutes,")
    print("   so this loop, built only from plasma half-lives, is about")
    print("   {:.2f} times too slow".format(per * 60.0 / 75.0))
    print("   amplitude there is {:.4f} about a mean of {:.4f},"
          .format(amp, mean))
    print("   a relative swing of {:.4f}".format(amp / mean))
#    clearance rates per hour from the published half-lives:
#    cortisol 0.5941, corticotropin 2.7726, releasing hormone 4.1589
#    delay h   amplitude   period h   mean
#    0.00      0.0000      0.000      1.4621
#    0.05      0.0000      0.000      1.4621
#    0.10      0.0000      0.000      1.4621
#    0.15      0.0001      2.312      1.4621
#    0.20      0.2420      2.539      1.5289
#    0.25      0.4348      2.859      1.6396
#    0.33      0.7305      3.374      1.8467
#    0.50      1.4957      4.441      2.3653
#    the loop starts to oscillate at a delay of 0.1630 hours
#    which is 9.8 minutes
#    at a delay of 0.25 hours the period is 2.859 hours
#    which is 171.5 minutes
#    the observed pulse interval in people is 60 to 90 minutes,
#    so this loop, built only from plasma half-lives, is about
#    2.29 times too slow
#    amplitude there is 0.4348 about a mean of 1.6396,
#    a relative swing of 0.2652

The loop settles for small delays and oscillates for large ones, and the boundary is at a delay of 9.8 minutes. At a fifteen-minute delay, which is the right order for the transit from pituitary to adrenal and back, the period is 171.5 minutes with a relative swing of 0.2652 about a mean of 1.6396. The observed interval between pulses in people is sixty to ninety minutes, so this loop is 2.29 times too slow.

That is a real discrepancy and not a rounding problem. It is also the useful kind, because the size of the gap constrains the repair.

The amplitude and period of the loop against the delay, with the boundary marked (p01).

Which repair is cheap

Two things could close the gap. The signal the pituitary actually reads might turn over faster than total plasma cortisol does, which would mean the clearances in the model are too slow. Or the whole loop might run faster, delay included, which would mean the oscillator is not the systemic loop at all but something tighter and more central.

Calculation

# Chapter 1. p02: What would have to be true for the period to match.
# p01 found that a loop built from plasma half-lives oscillates with a
# period about twice the observed pulse interval. Two repairs are
# scanned. The first speeds up the clearances only, which is the same as
# saying the signal the pituitary reads turns over faster than total
# plasma cortisol; the delay is left where it is. The second speeds up
# the whole loop, clearances and delay together. Only one of them can
# reach the observed period, and which one it is says something about
# where the pulse generator has to live.
HL_CORT, HL_ACTH, HL_CRH = 70.0, 15.0, 10.0
N, GAIN, K = 8.0, 20.0, 1.0
T, DT = 300.0, 0.002
TARGET = 75.0           # minutes, the middle of the observed band
LN2 = 0.6931471805599453
BC, QA, MH = (LN2 / (HL_CORT / 60.0), LN2 / (HL_ACTH / 60.0),
              LN2 / (HL_CRH / 60.0))

def simulate(tau, scale=1.0):
    bc, qa, mh = BC * scale, QA * scale, MH * scale
    kc, pa, kh = bc * 1.6, qa, mh * GAIN
    lag = int(round(tau / DT))
    c = [1.0] * (lag + 1)
    a = h = 0.6
    tail = []
    for i in range(int(T / DT)):
        cl = c[-1 - lag]
        f = 1.0 / (1.0 + (cl / K) ** N)
        h += DT * (kh * f - mh * h)
        a += DT * (pa * h - qa * a)
        c.append(c[-1] + DT * (kc * a - bc * c[-1]))
        if i * DT > T * 0.6:
            tail.append(c[-1])
    amp = 0.5 * (max(tail) - min(tail))
    mean = sum(tail) / len(tail)
    cross = []
    for i in range(1, len(tail)):
        if (tail[i - 1] - mean) < 0.0 <= (tail[i] - mean):
            cross.append(i * DT)
    per = ((cross[-1] - cross[0]) / (len(cross) - 1)
           if len(cross) > 2 else 0.0)
    return amp, per, mean

if __name__ == "__main__":
    print("   faster turnover at a delay of 0.25 hours:")
    print("   scale   period min   amplitude   effective cortisol half-life")
    for s in (1.0, 1.2, 1.4, 1.6, 1.8, 2.0, 2.4):
        amp, per, mean = simulate(0.25, s)
        print("   {:<8.1f}{:<13.1f}{:<12.4f}{:.1f}"
              .format(s, per * 60.0, amp, HL_CORT / s))
    print("   pushed much further, the period stops falling:")
    for s in (4.0, 8.0, 16.0, 32.0):
        amp, per, mean = simulate(0.25, s)
        print("   scale {:<7.1f}period {:.1f} minutes".format(s,
                                                              per * 60.0))
    print("   it is settling near {:.1f} minutes, which is {:.2f} times"
          .format(simulate(0.25, 32.0)[1] * 60.0,
                  simulate(0.25, 32.0)[1] / 0.25))
    print("   the delay. The target of {:.0f} minutes is reached near a"
          .format(TARGET))
    lo, hi = 1.0, 32.0
    for _ in range(34):
        mid = 0.5 * (lo + hi)
        if simulate(0.25, mid)[1] * 60.0 > TARGET:
            lo = mid
        else:
            hi = mid
    sc = 0.5 * (lo + hi)
    print("   scale of {:.3f}, an effective cortisol half-life of {:.1f}"
          .format(sc, HL_CORT / sc))
    print("   minutes, which no measurement of the hormone supports")
    print("   the whole loop faster, delay included:")
    print("   scale   delay min   period min   amplitude")
    for s in (1.0, 1.5, 2.0, 2.29, 3.0):
        amp, per, mean = simulate(0.25 / s, s)
        print("   {:<8.2f}{:<12.1f}{:<13.1f}{:.4f}"
              .format(s, 0.25 / s * 60.0, per * 60.0, amp))
    lo, hi = 1.0, 6.0
    for _ in range(34):
        mid = 0.5 * (lo + hi)
        if simulate(0.25 / mid, mid)[1] * 60.0 > TARGET:
            lo = mid
        else:
            hi = mid
    s = 0.5 * (lo + hi)
    print("   the factor that reproduces a {:.0f} minute period: {:.4f}"
          .format(TARGET, s))
    print("   which is far more modest than the {:.3f} the first repair"
          .format(sc))
    print("   needed, so the pulse generator is better placed above the")
    print("   pituitary than in the clearance of the hormone")
    print("   it needs an effective cortisol half-life of {:.1f} minutes"
          .format(HL_CORT / s))
    print("   and a delay of {:.1f} minutes".format(0.25 / s * 60.0))
    print("   scaling the whole loop leaves the shape untouched, so the")
    print("   relative swing stays at {:.4f}"
          .format(simulate(0.25 / s, s)[0] / simulate(0.25 / s, s)[2]))
    print("   against {:.4f} at the published rates"
          .format(simulate(0.25, 1.0)[0] / simulate(0.25, 1.0)[2]))
#    faster turnover at a delay of 0.25 hours:
#    scale   period min   amplitude   effective cortisol half-life
#    1.0     171.5        0.4348      70.0
#    1.2     158.9        0.6160      58.3
#    1.4     149.9        0.8073      50.0
#    1.6     143.0        1.0153      43.8
#    1.8     137.5        1.2425      38.9
#    2.0     132.9        1.4899      35.0
#    2.4     125.3        2.0418      29.2
#    pushed much further, the period stops falling:
#    scale 4.0    period 103.7 minutes
#    scale 8.0    period 74.4 minutes
#    scale 16.0   period 53.8 minutes
#    scale 32.0   period 42.0 minutes
#    it is settling near 42.0 minutes, which is 2.80 times
#    the delay. The target of 75 minutes is reached near a
#    scale of 7.866, an effective cortisol half-life of 8.9
#    minutes, which no measurement of the hormone supports
#    the whole loop faster, delay included:
#    scale   delay min   period min   amplitude
#    1.00    15.0        171.5        0.4348
#    1.50    10.0        113.8        0.4291
#    2.00    7.5         85.0         0.4233
#    2.29    6.6         74.8         0.4362
#    3.00    5.0         56.9         0.4335
#    the factor that reproduces a 75 minute period: 2.2767
#    which is far more modest than the 7.866 the first repair
#    needed, so the pulse generator is better placed above the
#    pituitary than in the clearance of the hormone
#    it needs an effective cortisol half-life of 30.7 minutes
#    and a delay of 6.6 minutes
#    scaling the whole loop leaves the shape untouched, so the
#    relative swing stays at 0.2632
#    against 0.2652 at the published rates

Speeding the clearances alone works eventually but at an absurd price. The period falls from 171.5 minutes at the published rates to 103.7 at four times those rates and 42.0 at thirty-two times, and it reaches the target of seventy-five minutes near a scale of 7.866, which implies an effective cortisol half-life of 8.9 minutes. No measurement of the hormone supports that. The period also stops falling as the scale rises, settling near 2.80 times the delay, because with the delay fixed the delay eventually dominates.

Speeding the whole loop is far cheaper. Scaling clearances and delay together scales the period exactly, so the factor needed is simply the discrepancy itself: 2.2767, which asks for an effective cortisol half-life of 30.7 minutes and a delay of 6.6 minutes. Both are inside the range that a tighter, more central loop could plausibly have, and scaling the whole loop leaves the shape untouched, so the relative swing stays at 0.2632 against 0.2652 at the published rates.

Hypothesis · not established

The author’s interpretation (not yet tested): the ultradian pulse generator sits above the pituitary, in a loop whose delay and clearances are both about 2.2767 times faster than the systemic ones, rather than in the systemic pituitary-adrenal loop; the systemic loop is driven by it rather than generating the rhythm. What would show it wrong: if pulses of the same period survived in a preparation in which the hypothalamic contribution was removed and only the pituitary-adrenal loop remained, with plasma half-lives unchanged, the systemic loop would be sufficient after all and this reading would be wrong.

Summary

A loop built only from published half-lives begins to oscillate above a delay of 9.8 minutes and, at a fifteen-minute delay, has a period of 171.5 minutes, which is 2.29 times slower than the observed pulse interval. Closing that gap by speeding the clearances alone would need a factor of 7.866 and an effective half-life of 8.9 minutes; closing it by speeding the whole loop needs only 2.2767. Chapter 2 asks what the pulses are for.

References

Walker, Jamie J., John R. Terry, and Stafford L. Lightman. 2010. “Origin of Ultradian Pulsatility in the Hypothalamic-Pituitary-Adrenal Axis.” Proceedings of the Royal Society B 277 (1688): 1627 to 1633.

Fan, Y. (2026). HPA-Axis Coupled Oscillations, Cortisol Awakening Response Flattening and Burnout Recovery Kinetics: Glucocorticoid Receptor Downregulation, Ultradian Pulsatility and Allostatic Load Reset [下丘脑垂体肾上腺轴耦合振荡、皮质醇觉醒反应扁平化与倦怠恢复动力学] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090252 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).