5D Library › Neurobiology of Ritual, Spatial Boundary Transitions and Computational Biomedicine of Self-Transcendence · 中文

Chapter 2 Stress Axis, Breathing and Inflammation: Which Parts Are Equations and Which Are Evidence

From “allostatic load” to a model one can solve

The second chapter of the framework sets out to do three things: solve equations for HPA-axis allostatic load and the flattening of the diurnal cortisol rhythm with fourth-order Runge–Kutta (RK4); connect “0.1-Hz Mayer-wave resonance breathing” at about six breaths a minute, via the baroreflex and RMSSD, to the cholinergic anti-inflammatory pathway mediated by α7 nicotinic receptors (TNF-α, IL-6); and write down the kinetics of endogenous β-endorphin and oxytocin. The evidence behind these three tasks differs greatly, and this chapter takes them one at a time.

First the concept. McEwen (1998) defined allostatic load as the cumulative cost the body pays for coping with repeated or prolonged stress: stress mediators (cortisol, catecholamines, inflammatory cytokines) are protective in the short term, and their long-term dysregulation is damaging (McEwen 1998). In research it is usually an index, the number of a dozen or so biomarkers (blood pressure, waist–hip ratio, glycated haemoglobin, cortisol and others) beyond risk quantiles. It is not a state variable that can be written straight into a differential equation. Book 14 discussed the bistability of the CRH–ACTH–cortisol negative-feedback loop and how allostatic-load indices are computed; that is only recalled here, not repeated.

An illustrative HPA model

Gupta and colleagues (2007) added the glucocorticoid receptor to an HPA-axis model and obtained bistability (Gupta et al. 2007); Sriram, Rodriguez-Fernandez and Doyle (2012) built on it to fit cortisol and ACTH data from healthy people and from people with depression and post-traumatic stress disorder, showing that different parameter regions give differently shaped diurnal rhythms (Sriram et al. 2012). This book does not reproduce their model; it writes a simpler three-variable illustrative model (the author’s model):

dCdt=D(t)(1+S)1+(F/Ki)2−Cτc,dAdt=C1+(F/Ki)2−Aτa,dFdt=A−Fτf,\frac{dC}{dt} = \frac{D(t)(1+S)}{1 + (F/K_i)^2} – \frac{C}{\tau_c},\quad \frac{dA}{dt} = \frac{C}{1 + (F/K_i)^2} – \frac{A}{\tau_a},\quad \frac{dF}{dt} = A – \frac{F}{\tau_f},

where CC, AA and FF are CRH, ACTH and cortisol, D(t)=1+0.9cos⁡(2π(t−6.5)/24)D(t) = 1 + 0.9\cos\bigl(2\pi(t-6.5)/24\bigr) is the circadian drive, SS a sustained extra drive, KiK_i the half-saturation constant of negative feedback (a larger KiK_i means feedback resistance), and the three time constants correspond to half-lives of about 4, 10 and 65 minutes. As in many field studies, the diurnal slope is the least-squares slope of ln⁡F\ln F on clock time between 07:00 and 22:00. The program integrates 8 days with RK4 at a step of 0.005 hours and keeps the last day.

Calculation

# A three-variable HPA-axis model solved with fourth-order Runge-Kutta (Chapter 2).
# The author's simplified model in the spirit of Gupta et al. (2007) and Sriram et al.
# (2012); it is NOT their model and its parameters are illustrative, not fitted.
#   dC/dt = D(t) (1 + S) / (1 + (F/Ki)^2) - C/tc        CRH, circadian drive D(t)
#   dA/dt = C / (1 + (F/Ki)^2) - A/ta                   ACTH
#   dF/dt = A - F/tf                                     cortisol
# D(t) = 1 + 0.9 cos(2 pi (t - 6.5)/24). Time in hours; half-lives about 4 min (CRH),
# 10 min (ACTH) and 65 min (cortisol). Diurnal slope: least-squares slope of ln F
# against clock time from 07:00 to 22:00, as in many field studies (ln units per hour).
import math

TC, TA, TF = 4 / 60 / math.log(2), 10 / 60 / math.log(2), 65 / 60 / math.log(2)

def rhs(t, y, S, Ki, evening):
    c, a, f = y
    d = 1 + 0.9 * math.cos(2 * math.pi * (t - 6.5) / 24)
    if evening and 17 <= t % 24 < 23:
        d += evening                                  # stress load late in the day
    fb = 1 / (1 + (f / Ki) ** 2)
    return (d * (1 + S) * fb - c / TC, c * fb - a / TA, a - f / TF)

def rk4(S=0.0, Ki=0.04, evening=0.0, days=8, dt=0.005):
    y, t, out = (0.1, 0.1, 0.1), 0.0, []
    for _ in range(round(days * 24 / dt)):
        f = lambda tt, yy: rhs(tt, yy, S, Ki, evening)
        k1 = f(t, y)
        k2 = f(t + dt / 2, tuple(v + dt / 2 * k for v, k in zip(y, k1)))
        k3 = f(t + dt / 2, tuple(v + dt / 2 * k for v, k in zip(y, k2)))
        k4 = f(t + dt, tuple(v + dt * k for v, k in zip(y, k3)))
        y = tuple(v + dt / 6 * (p + 2 * q + 2 * r + s)
                  for v, p, q, r, s in zip(y, k1, k2, k3, k4))
        t += dt
        if t > (days - 1) * 24 - 1e-9:
            out.append((t - (days - 1) * 24, y[2]))
    return out

def slope(day):
    pts = [(t, math.log(f)) for t, f in day if 7 <= t <= 22]
    mt = math.fsum(t for t, _ in pts) / len(pts)
    my = math.fsum(v for _, v in pts) / len(pts)
    return (math.fsum((t - mt) * (v - my) for t, v in pts)
            / math.fsum((t - mt) ** 2 for t, _ in pts))

def at(day, h):
    return min(day, key=lambda p: abs(p[0] - h))[1]

SCEN = (("baseline", {}), ("constant extra drive S = 0.5", {"S": 0.5}),
        ("evening load 17-23 h (+1.0)", {"evening": 1.0}),
        ("feedback resistance Ki x 2", {"Ki": 0.08}),
        ("S = 0.5 and Ki x 2", {"S": 0.5, "Ki": 0.08}))
print("cortisol in units of baseline F(08:00) = 100")
print("scenario                       F(08h)  F(22h)  ratio  slope (ln/h)  mean F")
res = {}
for name, kw in SCEN:
    day = rk4(**kw); res[name] = day
    u = 100 / at(res["baseline"], 8)
    print(f"{name:30s} {u * at(day, 8):6.1f} {u * at(day, 22):7.1f}"
          f" {at(day, 8) / at(day, 22):6.2f} {slope(day):11.3f}"
          f" {u * math.fsum(f for _, f in day) / len(day):8.1f}")
fine = rk4(dt=0.0025)
err = max(abs(a[1] - b[1]) for a, b in zip(res["baseline"], fine[0::2]))
print(f"RK4 check: largest change in F when the step is halved:"
      f" {err * 100 / at(res['baseline'], 8):.1e} units")
# cortisol in units of baseline F(08:00) = 100
# scenario                       F(08h)  F(22h)  ratio  slope (ln/h)  mean F
# baseline                        100.0    28.6   3.50      -0.137     66.7
# constant extra drive S = 0.5    117.3    40.6   2.89      -0.122     81.3
# evening load 17-23 h (+1.0)     100.0    85.0   1.18      -0.030     80.9
# feedback resistance Ki x 2      143.9    30.0   4.80      -0.160     89.2
# S = 0.5 and Ki x 2              176.2    44.0   4.01      -0.147    114.0
# RK4 check: largest change in F when the step is halved: 5.7e-10 units

Halving the step changes cortisol by at most 5.7×10−105.7\times10^{-10} units, so numerical error is negligible. The results hold a surprise for the framework. At baseline the morning-to-evening ratio is 3.50 and the slope −0.137. A sustained extra drive (S=0.5S = 0.5) flattens the slope slightly (−0.122). An evening load (drive raised by 1.0 between 17:00 and 23:00) flattens it most (−0.030), raising 22:00 cortisol from 28.6 to 85.0 while 08:00 is unchanged. Feedback resistance (doubling KiK_i) actually steepens the slope (−0.160) and raises morning cortisol to 143.9. So, in this model, “chronic stress flattens the diurnal cortisol rhythm” is not automatic: it depends on when in the day the load falls and on whether drive or feedback is impaired. The left panel of Figure 2 shows three cases.

Evidence

Cortisol slope and health (meta-analysis of human observational studies). The meta-analysis of Adam and colleagues (2017) pooled 179 associations from 80 studies and found flatter diurnal cortisol slopes associated with poorer mental and physical health, with an average effect size r=0.147r = 0.147 (Adam et al. 2017). Most of the studies were cross-sectional or longitudinal observational studies, with outcomes including depression, fatigue, cardiovascular markers and mortality. An rr of 0.147 means the slope accounts for about 2 % of the variance in outcome; and a flat slope may be a consequence of illness, of sleep timing or of inaccurately timed samples, so it cannot be read as “steepen the slope and health will improve”.

Resonance breathing: the delay of the baroreflex loop

The framework calls breathing at about six breaths a minute “0.1-Hz Mayer-wave resonance”. Two concepts need separating. Mayer waves are spontaneous oscillations of arterial pressure at about 0.1 Hz, generally linked to sympathetic control of the vessels and the baroreflex loop (Julien 2006). Resonance breathing means that, at a certain breathing rate, the amplitude of heart-rate oscillation is greatest. The explanation of Vaschillo, Vaschillo and Lehrer is that breathing causes blood-pressure fluctuations, the baroreflex changes heart rate after a delay, and heart rate changes blood pressure in turn; when the breathing period matches the loop delay, the effects add in phase and the amplitude peaks (Vaschillo et al. 2006; Lehrer and Gevirtz 2014). So resonance comes from the delay of the baroreflex loop; it is close to the frequency of Mayer waves but is not the same thing, and each person’s resonance frequency differs and usually has to be measured individually. Book 14 mentioned resonance at about six breaths a minute; here a linear loop model shows where it comes from and what it does to measurement (the author’s model).

Let breathing at angular frequency ω\omega drive blood pressure (gain BB) and heart rate (a direct vagal path, low-pass, gain RR, time constant TvT_v); let heart rate act on blood pressure through a vascular low-pass (gain KK, time constant TwT_w), and blood pressure act on heart rate through the baroreflex with gain −G-G and delay τ\tau. The closed-loop heart-rate amplitude is

H(ω)=R/(1+iωTv)−Ge−iωτB1+GKe−iωτ/(1+iωTw).H(\omega) = \frac{R/(1+i\omega T_v) – G e^{-i\omega\tau} B}{1 + G K e^{-i\omega\tau}/(1+i\omega T_w)}.

When the loop phase approaches −π-\pi, the denominator shrinks and the amplitude grows. The program then generates 5 minutes of beat-to-beat RR intervals at each rate (mean heart rate 65 per minute, beat noise 10 ms) and computes RMSSD and SDNN.

Calculation

# Resonance breathing in a linear baroreflex loop, and what RMSSD does (Chapter 2).
# The author's illustrative model (after the resonance idea of Vaschillo, Lehrer et al.):
# breathing at angular frequency w drives blood pressure (gain B) and heart rate directly
# (vagal low-pass, gain R, time constant Tv); heart rate feeds back on pressure through a
# Windkessel low-pass (gain K, time constant Tw); pressure acts on heart rate through the
# baroreflex with gain -G and delay tau. Closed-loop heart-rate amplitude per unit breath:
#   H(w) = (R/(1+iwTv) - G e^{-iw tau} B) / (1 + G K e^{-iw tau} / (1 + iwTw))
# Then 5-minute RR series (65 bpm mean, beat noise 10 ms) give RMSSD and SDNN.
import cmath, math, random

B, R, TV, K, TW, G = 2.0, 2.0, 1.5, 1.0, 2.0, 1.0

def hr_amp(f, tau=3.5):
    w = 2 * math.pi * f; d = cmath.exp(-1j * w * tau)
    return abs((R / (1 + 1j * w * TV) - G * d * B) / (1 + G * K * d / (1 + 1j * w * TW)))

def rr_series(rng, f, amp, minutes=5, hr0=65.0, noise=10.0):
    t, rr = 0.0, []
    while t < minutes * 60:
        hr = hr0 + amp * math.sin(2 * math.pi * f * t)
        x = 60000 / hr + rng.gauss(0, noise); rr.append(x); t += x / 1000
    return rr

def rmssd(rr):
    return math.sqrt(math.fsum((b - a) ** 2 for a, b in zip(rr, rr[1:])) / (len(rr) - 1))

def sdnn(rr):
    m = math.fsum(rr) / len(rr)
    return math.sqrt(math.fsum((x - m) ** 2 for x in rr) / (len(rr) - 1))

rng = random.Random(6)
print("breaths/min  HR amplitude (bpm)  RMSSD (ms)  ln RMSSD  SDNN (ms)")
rows = []
for bpm in (3, 4, 5, 5.5, 6, 6.5, 7, 8, 10, 12, 15):
    f = bpm / 60; a = hr_amp(f)
    rr = rr_series(rng, f, a); rows.append((bpm, a, rmssd(rr), sdnn(rr)))
    print(f"{bpm:9.1f} {a:15.2f} {rmssd(rr):15.1f} {math.log(rmssd(rr)):9.2f}"
          f" {sdnn(rr):9.1f}")
best = max(rows, key=lambda r: r[1]); top_r = max(rows, key=lambda r: r[2])
print(f"largest HR oscillation at {best[0]} breaths/min; largest RMSSD at"
      f" {top_r[0]} breaths/min")
print("resonance frequency against baroreflex delay tau (fine grid 2-15 breaths/min):")
for tau in (2.5, 3.0, 3.5, 4.0, 4.5):
    grid = [2 + 0.01 * i for i in range(1301)]
    fb = max(grid, key=lambda b: hr_amp(b / 60, tau))
    print(f"   tau = {tau:.1f} s: peak at {fb:.2f} breaths/min ({fb / 60:.3f} Hz)")
rr0 = rr_series(random.Random(1), 0.25, hr_amp(0.25))
rr6 = rr_series(random.Random(1), 0.1, hr_amp(0.1))
print(f"same model person recorded at 15/min vs paced 6/min: RMSSD {rmssd(rr0):.1f}"
      f" vs {rmssd(rr6):.1f} ms")
print("   (the whole difference comes from the breathing pattern during the recording)")
# breaths/min  HR amplitude (bpm)  RMSSD (ms)  ln RMSSD  SDNN (ms)
#       3.0            1.00            14.1      2.65      13.9
#       4.0            1.92            16.1      2.78      21.9
#       5.0            3.88            23.6      3.16      40.3
#       5.5            5.31            31.9      3.46      54.8
#       6.0            6.12            38.3      3.65      62.4
#       6.5            5.71            38.4      3.65      58.4
#       7.0            4.90            36.9      3.61      51.0
#       8.0            3.71            31.1      3.44      38.0
#      10.0            2.67            28.8      3.36      28.9
#      12.0            2.24            27.8      3.33      24.4
#      15.0            1.88            29.9      3.40      22.0
# largest HR oscillation at 6 breaths/min; largest RMSSD at 6.5 breaths/min
# resonance frequency against baroreflex delay tau (fine grid 2-15 breaths/min):
#    tau = 2.5 s: peak at 8.12 breaths/min (0.135 Hz)
#    tau = 3.0 s: peak at 6.90 breaths/min (0.115 Hz)
#    tau = 3.5 s: peak at 6.05 breaths/min (0.101 Hz)
#    tau = 4.0 s: peak at 5.42 breaths/min (0.090 Hz)
#    tau = 4.5 s: peak at 4.92 breaths/min (0.082 Hz)
# same model person recorded at 15/min vs paced 6/min: RMSSD 27.8 vs 38.3 ms
#    (the whole difference comes from the breathing pattern during the recording)

With τ=3.5\tau = 3.5 s, heart-rate amplitude is largest at six breaths a minute (6.12 beats per minute), more than three times that at fifteen (1.88); SDNN also peaks at six (62.4 ms). But the RMSSD peak is flat: 38.3 ms at six, 38.4 at six and a half, 29.9 at fifteen. Going from fifteen to six breaths a minute, heart-rate amplitude grows 3.3-fold and SDNN 2.8-fold, but RMSSD only about 1.3-fold. The reason is that RMSSD uses differences between neighbouring beats, and a slow oscillation with a 10-second period changes little from one beat to the next, so RMSSD is relatively insensitive to it; relatively insensitive is not unchanged, and, as shown below, it still changes clearly with breathing during the recording. The resonance moves with the delay: from 2.5 to 4.5 s, the peak moves from 8.12 to 4.92 breaths a minute. The right panel of Figure 2 shows the curves for three delays.

The last two lines matter for the design of Chapter 8: the same model person has an RMSSD of 27.8 ms when recorded while breathing fifteen times a minute and 38.3 ms when breathing six times a minute. The whole difference comes from breathing during the recording. Guidelines on heart-rate variability have long warned that breathing rate and tidal volume affect short-term indices and that researchers should control, or at least report, breathing (Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology 1996; Laborde et al. 2017). If one arm of a trial is still breathing slowly during the recording while another breathes spontaneously, a difference in RMSSD cannot be read as “better recovery”.

Left: one day of cortisol in the illustrative HPA model (baseline, evening load, extra drive with feedback resistance; baseline 08:00 = 100). Right: heart-rate amplitude against breathing rate in the baroreflex-loop model for three delays (the author’s model, illustrative parameters).

Evidence

Slow breathing and heart-rate variability (human studies). The systematic review and meta-analysis of Laborde and colleagues (2022) included 223 studies (172 measuring effects during breathing, 16 immediately after a single session and 49 after multi-session interventions) and found that slow breathing increased vagally mediated heart-rate variability during practice, after a single session and after multiple sessions (Laborde et al. 2022). The systematic review of Zaccaro and colleagues (2018) also linked slow breathing to higher heart-rate variability and to improvements in autonomic and emotional measures, though the quality of the included studies varied (Zaccaro et al. 2018). These results support “slow breathing raises vagal indices during and shortly after practice”; on their own they say neither how long the rise lasts nor whether it changes inflammation or disease outcomes.

The cholinergic anti-inflammatory pathway: animals, devices and breathing

The framework links breathing, RMSSD and the “α7nAChR cholinergic anti-inflammatory pathway” into a chain. The evidence for each link differs in strength.

Evidence

Animal and mechanistic evidence. Borovikova and colleagues (2000) found in rats that electrical stimulation of the vagus nerve attenuated the systemic inflammatory response and the rise in TNF caused by endotoxin (Borovikova et al. 2000); Wang and colleagues (2003) found in mice that the α7 nicotinic acetylcholine receptor on macrophages is required for this inhibition (Wang et al. 2003); Tracey (2002) summarised these results as the “inflammatory reflex” (Tracey 2002). These are animal and mechanistic studies: they show that vagal efferent activity can inhibit cytokine production, but the stimulation strengths, time courses and endotoxin doses are far from everyday breathing practice in people.

Evidence

Human evidence. Koopman and colleagues (2016) implanted vagus-nerve stimulators in patients with rheumatoid arthritis and reported inhibited TNF production and lower disease activity; this was a small open-label study of an implanted device, not of breathing (Koopman et al. 2016). Kox and colleagues (2014) trained 12 men in a programme of breathing techniques, cold exposure and meditation and then injected them, together with 12 untrained men, with endotoxin: the trained group had higher adrenaline, higher levels of the anti-inflammatory cytokine IL-10, lower TNF-α, IL-6 and IL-8, and fewer flu-like symptoms (Kox et al. 2014). The authors attributed the effect to activation of the sympathetic nervous system and adrenaline rather than to the vagus. The study is often cited as “breathing fights inflammation”, but its breathing was hyperventilation with breath-holding, not slow breathing, and its mechanism was not the vagal–α7 pathway of the framework. In checking for this book we found no consistent evidence that slow-breathing practice reliably lowers blood TNF-α or IL-6 in healthy people or patients.

The next program uses two illustrative models to show why the size of an effect in people cannot be read off animal results. Part A is a toy: an endotoxin-like stimulus decays over time, TNF production is inhibited by vagal activity vv by a factor 1/(1+v/Kv)1/(1+v/K_v), and IL-6 follows TNF. Part B is a one-compartment hormone model: secretion at a constant rate during a 12-minute practice, then first-order elimination.

Calculation

# Two schematic kinetic models (Chapter 2). Parameters are illustrative, NOT estimates.
# A. Cholinergic anti-inflammatory pathway as a toy: an endotoxin-like stimulus E(t)
#    decays with a 1-hour time constant; TNF is produced at rate E/(1 + v/Kv) and cleared
#    with a 20-minute half-life; IL-6 follows TNF with a 1-hour half-life. v is efferent
#    vagal activity (arbitrary units). How much does TNF fall when v rises from 0 to 1?
#    The answer is set by Kv, which is unknown in humans, and by timing.
# B. A one-compartment hormone pulse: secretion at a constant rate for 12 minutes, then
#    first-order elimination with half-life h. What share of the peak is still there when
#    blood is drawn at various times after the practice ends?
import math

def toxin(v, kv, hours=8.0, dt=0.001, window=None):
    e, tnf, il6, t, peak, auc = 1.0, 0.0, 0.0, 0.0, 0.0, []
    kt, ki = math.log(2) / (20 / 60), math.log(2) / 1.0
    while t < hours:
        vv = v if window is None or window[0] <= t < window[1] else 0.0
        de = -e; dtnf = e / (1 + vv / kv) - kt * tnf; dil = tnf - ki * il6
        e += dt * de; tnf += dt * dtnf; il6 += dt * dil; t += dt
        peak = max(peak, tnf); auc.append(il6 * dt)
    return peak, math.fsum(auc)

print("A  TNF peak and IL-6 area, relative to v = 0 (same Kv)")
for kv in (0.5, 2.0, 10.0):
    p0, a0 = toxin(0.0, kv)
    row = []
    for v in (0.5, 1.0, 2.0):
        p, a = toxin(v, kv); row.append(f"v={v:.1f}: {p / p0:5.2f}/{a / a0:5.2f}")
    print(f"   Kv = {kv:4.1f}  " + "   ".join(row))

p0, a0 = toxin(0.0, 2.0)
print("   a 12-minute rise to v = 1 (Kv = 2), starting at various times after"
      " the stimulus")
for st in (0.0, 0.5, 1.0, 2.0):
    p, a = toxin(1.0, 2.0, window=(st, st + 0.2))
    print(f"   start +{st:3.1f} h: TNF peak {p / p0:5.2f}, IL-6 area {a / a0:5.2f}")

def pulse(h, t_after, dur=12.0, rate=1.0):
    k = math.log(2) / h
    peak = rate / k * (1 - math.exp(-k * dur))            # at the end of secretion
    return peak, peak * math.exp(-k * t_after)

print("B  share of the end-of-practice level left at the blood draw")
print("   half-life      +5 min   +20 min   +40 min   +60 min")
for h in (4.0, 10.0, 25.0, 60.0):
    cells = [pulse(h, t)[1] / pulse(h, 0)[1] for t in (5, 20, 40, 60)]
    print(f"   {h:5.0f} min " + "".join(f"{c:10.3f}" for c in cells))
# A  TNF peak and IL-6 area, relative to v = 0 (same Kv)
#    Kv =  0.5  v=0.5:  0.50/ 0.50   v=1.0:  0.33/ 0.33   v=2.0:  0.20/ 0.20
#    Kv =  2.0  v=0.5:  0.80/ 0.80   v=1.0:  0.67/ 0.67   v=2.0:  0.50/ 0.50
#    Kv = 10.0  v=0.5:  0.95/ 0.95   v=1.0:  0.91/ 0.91   v=2.0:  0.83/ 0.83
#    a 12-minute rise to v = 1 (Kv = 2), starting at various times after the stimulus
#    start +0.0 h: TNF peak  0.93, IL-6 area  0.94
#    start +0.5 h: TNF peak  0.96, IL-6 area  0.96
#    start +1.0 h: TNF peak  1.00, IL-6 area  0.98
#    start +2.0 h: TNF peak  1.00, IL-6 area  0.99
# B  share of the end-of-practice level left at the blood draw
#    half-life      +5 min   +20 min   +40 min   +60 min
#        4 min      0.420     0.031     0.001     0.000
#       10 min      0.707     0.250     0.062     0.016
#       25 min      0.871     0.574     0.330     0.189
#       60 min      0.944     0.794     0.630     0.500

The conclusion of part A is set almost entirely by a parameter, KvK_v, that is unknown in people: when vv rises from 0 to 1, peak TNF falls to 0.33 with Kv=0.5K_v = 0.5 but only to 0.91 with Kv=10K_v = 10. Timing matters too: with Kv=2K_v = 2, a 12-minute rise in vagal activity that starts with the stimulus lowers the TNF peak only to 0.93; one that starts an hour after the stimulus leaves the peak unchanged (1.00). So any claim that “breathing practice lowers TNF by X %” without human measurements is just a choice of parameters.

Part B shows a measurement problem. If a hormone has a plasma half-life of only 4 minutes, blood drawn 20 minutes after the practice ends holds 3.1 % of the end-of-practice level; with a half-life of 25 minutes, 57.4 %. The half-lives of β-endorphin and oxytocin, and whether central and peripheral release move together, are empirical questions to be checked separately; this book writes them only as illustrative parameters, not as measured values. No change in plasma does not prove no change in the brain; a change in plasma does not prove that it came from the brain or that it relates to a change in mood.

Hypothesis · not established · Slow breathing and inflammatory markers

The author’s hypothesis (not established): in healthy adults, eight weeks of slow breathing at each person’s resonance frequency, 15 minutes twice a day, lowers the ex vivo TNF-α production of endotoxin-stimulated whole blood compared with time-matched relaxation with spontaneous breathing. What would show it wrong: in a pre-registered randomised controlled trial, a confidence interval for the between-group difference in TNF-α production that lies within the pre-specified smallest effect of interest; or a difference that appears only on practice days and disappears after a one-week washout.

Chapter summary

Allostatic load is a multi-marker index, not a variable in an equation; the book’s HPA equations are an illustrative model, and the RK4 numerical error is negligible. In this model the diurnal cortisol slope flattens mainly with an evening load (from −0.137 to −0.030), while feedback resistance steepens it; in human meta-analysis a flatter slope is associated with poorer health with r=0.147r = 0.147, an association. Resonance breathing comes from the delay of the baroreflex loop; it is close to, but not the same as, Mayer waves, and resonance frequencies differ between people; RMSSD is relatively insensitive to slow oscillations (it changes far less than SDNN), yet breathing during the recording still changes it clearly. Evidence for the cholinergic anti-inflammatory pathway comes mainly from animals and implanted stimulators; evidence that breathing practice lowers inflammatory cytokines in people is thin; the kinetics of endogenous peptides in this book are only illustrative.

References

Adam, Emma K., Meghan E. Quinn, Royette Tavernier, Mollie T. McQuillan, Katie A. Dahlke, and Kirsten E. Gilbert. 2017. “Diurnal Cortisol Slopes and Mental and Physical Health Outcomes: A Systematic Review and Meta-Analysis.” Psychoneuroendocrinology 83: 25–41. https://doi.org/10.1016/j.psyneuen.2017.05.018.
Borovikova, Lyudmila V., Svetlana Ivanova, Minghuang Zhang, et al. 2000. “Vagus Nerve Stimulation Attenuates the Systemic Inflammatory Response to Endotoxin.” Nature 405 (6785): 458–62. https://doi.org/10.1038/35013070.
Gupta, Suresh, Eric Aslakson, Brian M. Gurbaxani, and Suzanne D. Vernon. 2007. “Inclusion of the Glucocorticoid Receptor in a Hypothalamic Pituitary Adrenal Axis Model Reveals Bistability.” Theoretical Biology and Medical Modelling 4: 8. https://doi.org/10.1186/1742-4682-4-8.
Julien, Claude. 2006. “The Enigma of Mayer Waves: Facts and Models.” Cardiovascular Research 70 (1): 12–21.
Koopman, Frieda A., Sangeeta S. Chavan, Sanda Miljko, et al. 2016. “Vagus Nerve Stimulation Inhibits Cytokine Production and Attenuates Disease Severity in Rheumatoid Arthritis.” Proceedings of the National Academy of Sciences 113 (29): 8284–89. https://doi.org/10.1073/pnas.1605635113.
Kox, Matthijs, Lucas T. van Eijk, Jelle Zwaag, et al. 2014. “Voluntary Activation of the Sympathetic Nervous System and Attenuation of the Innate Immune Response in Humans.” Proceedings of the National Academy of Sciences 111 (20): 7379–84. https://doi.org/10.1073/pnas.1322174111.
Laborde, Sylvain, Mark S. Allen, Uirassu Borges, et al. 2022. “Effects of Voluntary Slow Breathing on Heart Rate and Heart Rate Variability: A Systematic Review and a Meta-Analysis.” Neuroscience and Biobehavioral Reviews 138: 104711. https://doi.org/10.1016/j.neubiorev.2022.104711.
Laborde, Sylvain, Emma Mosley, and Julian F. Thayer. 2017. “Heart Rate Variability and Cardiac Vagal Tone in Psychophysiological Research: Recommendations for Experiment Planning, Data Analysis, and Data Reporting.” Frontiers in Psychology 8: 213. https://doi.org/10.3389/fpsyg.2017.00213.
Lehrer, Paul M., and Richard Gevirtz. 2014. “Heart Rate Variability Biofeedback: How and Why Does It Work?” Frontiers in Psychology 5: 756. https://doi.org/10.3389/fpsyg.2014.00756.
McEwen, Bruce S. 1998. “Protective and Damaging Effects of Stress Mediators.” New England Journal of Medicine 338 (3): 171–79. https://doi.org/10.1056/NEJM199801153380307.
Sriram, Krishna, Maria Rodriguez-Fernandez, and III Doyle Francis J. 2012. “Modeling Cortisol Dynamics in the Neuro-Endocrine Axis Distinguishes Normal, Depression, and Post-Traumatic Stress Disorder (PTSD) in Humans.” PLoS Computational Biology 8 (2): e1002379. https://doi.org/10.1371/journal.pcbi.1002379.
Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology. 1996. “Heart Rate Variability: Standards of Measurement, Physiological Interpretation and Clinical Use.” Circulation 93 (5): 1043–65. https://doi.org/10.1161/01.CIR.93.5.1043.
Tracey, Kevin J. 2002. “The Inflammatory Reflex.” Nature 420 (6917): 853–59. https://doi.org/10.1038/nature01321.
Vaschillo, Evgeny G., Bronya Vaschillo, and Paul M. Lehrer. 2006. “Characteristics of Resonance in Heart Rate Variability Stimulated by Biofeedback.” Applied Psychophysiology and Biofeedback 31 (2): 129–42.
Wang, Hong, Man Yu, Mahendar Ochani, et al. 2003. “Nicotinic Acetylcholine Receptor α\alpha7 Subunit Is an Essential Regulator of Inflammation.” Nature 421 (6921): 384–88. https://doi.org/10.1038/nature01339.
Zaccaro, Andrea, Andrea Piarulli, Marco Laurino, et al. 2018. “How Breath-Control Can Change Your Life: A Systematic Review on Psycho-Physiological Correlates of Slow Breathing.” Frontiers in Human Neuroscience 12: 353. https://doi.org/10.3389/fnhum.2018.00353.

Fan, Y. (2026). Neurobiology of Ritual, Spatial Boundary Transitions and Computational Biomedicine of Self-Transcendence: From Predictive-Coding Free-Energy Minimisation, Default-Mode Network Decoupling and Vagal–Endocrine Restoration to Classical Chinese Medicine on Shifting the Spirit (Yijing Bianqi) [仪式神经学、空间边界切换与自我转移的计算生物医学] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23839910 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).