5D Library › 复合补剂相互作用、时间疗法给药窗口与代谢酶竞争 · English

第四章 受体在听的时候

时辰为什么会有关系

时钟可以通过两条路改变结果。它可以改变有多少物质进入血液,这是前几章处理的;它也可以改变身体对给定量的反应有多大,而这是另一个问题,有另一套文献。Smolensky 与 Peppas 综述了这个领域,而机制处处相同:受体的密度、某个转录程序的可用性,或某个器官自身的节律,在一天之中变化,于是恒定的浓度并不产生恒定的效应 (Smolensky and Peppas 2007)。Czeisler 等人确立了人体起搏器有多精确、周期有多接近二十四小时,而这正是使时序可重复到足以据此排程的原因 (Czeisler et al. 1999)。Lewy 等人绘出了褪黑素的相位反应曲线,这就是同一种激素在傍晚给与在清晨给会把生物钟推向相反方向的原因 (Lewy et al. 1992)。

临床上的利害不是假想的。Hermida 等人把服用降压药的时辰随机化,并报告了心血管结局上的差别,这是时间疗法所产生过的最强一类论断 (Hermida et al. 2010)。

证据

已发表。 Smolensky 与 Peppas 综述时间生物学与时间治疗学,以及给药时辰改变效应的那些机制 (Smolensky and Peppas 2007)。Czeisler 等人测量了人体昼夜起搏器的稳定性与周期 (Czeisler et al. 1999)。Lewy 等人确立了褪黑素给药的相位反应曲线 (Lewy et al. 1992)。Hermida 等人把抗高血压治疗的给药时辰随机化,并报告了心血管风险上的差别 (Hermida et al. 2010)。

注

本章不是什么。 下面那些曲线是皮质醇与褪黑素的示意形状,加上一个示意的敏感度模型,不是测量。目的是弄清时辰在原理上能有多大关系,以及有用的窗口有多宽,这些是结构性问题。任何读者都不应从本章推断某种药物的服用时辰;一种真实药物的时间疗法文献是针对那种药物的。

推导

# Chapter 4. p07: When the receptor is listening. Two things vary across
# the day: the concentration a dose produces, which depends only on when
# it was swallowed, and the sensitivity of what it acts on, which does
# not. The effect of a dose is the integral of the product. Cortisol
# rises before waking and peaks in the morning, and the transcriptional
# window that mitochondrial cofactors act in follows it; melatonin rises
# in the late evening, and the inhibitory tone that a calming agent adds
# to follows that. The plasma curve is a one-compartment absorption and
# elimination model with the times below. The day wraps around, so a
# dose late at night is credited with its effect after midnight.
TMAX, HALFLIFE = 1.5, 4.0          # hours
STEP = 1.0 / 60.0                  # one minute
E = 2.718281828459045

def plasma(dt):
    # dt is hours since the dose; a difference of two exponentials
    if dt < 0.0:
        return 0.0
    ka = 2.0 / TMAX
    ke = 0.6931471805599453 / HALFLIFE
    return (E ** (-ke * dt) - E ** (-ka * dt)) * ka / (ka - ke)

def cortisol(t):
    # peak near the eighth hour, trough in the late evening
    x = (t - 8.0) % 24.0
    if x > 12.0:
        x -= 24.0
    return 0.12 + 0.88 * E ** (-(x / 3.6) ** 2)

def melatonin(t):
    x = (t - 3.0) % 24.0
    if x > 12.0:
        x -= 24.0
    return 0.04 + 0.96 * E ** (-(x / 3.2) ** 2)

SENS = {"morning cofactor": cortisol, "evening calmer": melatonin}

def effect(hour, which):
    f = SENS[which]
    total = 0.0
    n = int(round(48.0 / STEP))
    for i in range(n):
        t = i * STEP
        total += plasma(t) * f((hour + t) % 24.0)
    return total * STEP

if __name__ == "__main__":
    print("   hour   morning cofactor   evening calmer")
    best = {k: (-1.0, None) for k in SENS}
    worst = {k: (1e9, None) for k in SENS}
    for h in range(24):
        row = []
        for k in ("morning cofactor", "evening calmer"):
            v = effect(float(h), k)
            row.append(v)
            if v > best[k][0]:
                best[k] = (v, h)
            if v < worst[k][0]:
                worst[k] = (v, h)
        print("   {:<7d}{:<19.4f}{:.4f}".format(h, row[0], row[1]))
    for k in ("morning cofactor", "evening calmer"):
        b, hb = best[k]
        w, hw = worst[k]
        print("   {}: best hour {}, worst hour {}".format(k, hb, hw))
        print("      effect {:.4f} against {:.4f}, a ratio of {:.2f}"
              .format(b, w, b / w))
        print("      dosing at the worst hour loses {:.4f}"
              .format(1.0 - w / b))
    gap = (best["evening calmer"][1] - best["morning cofactor"][1]) % 24
    print("   hours forward from the first best time to the second: {}"
          .format(gap))
    print("   as a phase angle: {:.1f} degrees".format(360.0 * gap / 24.0))
    print("   the shorter way round is {} hours".format(min(gap, 24 - gap)))
    print("   best hour inside waking hours 6 to 22, and what it costs:")
    for k in ("morning cofactor", "evening calmer"):
        hw = max(range(6, 23), key=lambda h: effect(float(h), k))
        v = effect(float(hw), k)
        print("   {:<20}hour {:<4d}effect {:.4f}, loss {:.4f}"
              .format(k, hw, v, 1.0 - v / best[k][0]))
    both = max(range(24), key=lambda h: effect(float(h), "morning cofactor")
               + effect(float(h), "evening calmer"))
    s_both = (effect(float(both), "morning cofactor")
              + effect(float(both), "evening calmer"))
    s_split = best["morning cofactor"][0] + best["evening calmer"][0]
    print("   best single hour for both at once: {}".format(both))
    print("   its total effect {:.4f} against {:.4f} when split"
          .format(s_both, s_split))
    print("   so taking both together costs {:.4f}"
          .format(1.0 - s_both / s_split))
#    hour   morning cofactor   evening calmer
#    0      2.4732             3.1151
#    1      2.7710             2.9183
#    2      3.0740             2.4915
#    3      3.3398             1.9291
#    4      3.5073             1.3725
#    5      3.5124             0.9384
#    6      3.3170             0.6710
#    7      2.9364             0.5481
#    8      2.4438             0.5202
#    9      1.9437             0.5437
#    10     1.5277             0.5938
#    11     1.2426             0.6602
#    12     1.0868             0.7410
#    13     1.0298             0.8375
#    14     1.0362             0.9522
#    15     1.0796             1.0887
#    16     1.1459             1.2509
#    17     1.2297             1.4432
#    18     1.3309             1.6705
#    19     1.4516             1.9357
#    20     1.5951             2.2367
#    21     1.7655             2.5580
#    22     1.9668             2.8603
#    23     2.2026             3.0746
#    morning cofactor: best hour 5, worst hour 13
#       effect 3.5124 against 1.0298, a ratio of 3.41
#       dosing at the worst hour loses 0.7068
#    evening calmer: best hour 0, worst hour 8
#       effect 3.1151 against 0.5202, a ratio of 5.99
#       dosing at the worst hour loses 0.8330
#    hours forward from the first best time to the second: 19
#    as a phase angle: 285.0 degrees
#    the shorter way round is 5 hours
#    best hour inside waking hours 6 to 22, and what it costs:
#    morning cofactor    hour 6   effect 3.3170, loss 0.0556
#    evening calmer      hour 22  effect 2.8603, loss 0.0818
#    best single hour for both at once: 1
#    its total effect 5.6893 against 6.6275 when split
#    so taking both together costs 0.1416

一剂的效应,是它所产生的浓度与它所作用之物的敏感度,两者之积在一天上的积分。逐小时算出来,晨间辅因子在第 5 小时给药时见顶,达到 3.5124,而在最差的时辰是 1.0298,比值 3.41,所以在最差时辰给药损失 0.7068。夜间镇静剂在第 0 小时给药时见顶,为 3.1151,对最差的 0.5202,比值 5.99,最差时辰损失 0.8330。对一项分文不花的干预而言,这些是很大的数字。

然而,最佳时辰是清晨五点与午夜,而诚实的观察是:前者没人醒着,后者少有人清醒。限制在醒着的时辰内,最佳时间变成第 6 小时与第 22 小时,相对无约束最优只花 0.0556 与 0.0818。身为一个有工作的人所带来的实际损失不到十分之一,这才是有用的发现。

一剂的效应对服用的时辰,两类都在图中(p07)。

窗口有多宽

一个定位到小时的最优点,如果它周围的曲线很尖,就不可执行;如果曲线很平,又几乎不值得算。p08 把同一个模型放到一分钟的网格上,量出宽度。

推导

# Chapter 4. p08: The optimal phase, and how wide the window is. The
# same two sensitivity curves and the same plasma curve as p07, now on a
# one-minute grid, so the best moment is located to the minute rather
# than to the hour. What matters in practice is not the peak but the
# width of the window around it inside which little is lost, because a
# schedule nobody can keep is worth nothing. Two widths are reported:
# the window that costs less than one part in twenty, and the one that
# costs less than one part in ten.
TMAX, HALFLIFE = 1.5, 4.0
STEP = 1.0 / 60.0
E = 2.718281828459045
TOL = (0.05, 0.10)

def plasma(dt):
    if dt < 0.0:
        return 0.0
    ka = 2.0 / TMAX
    ke = 0.6931471805599453 / HALFLIFE
    return (E ** (-ke * dt) - E ** (-ka * dt)) * ka / (ka - ke)

def cortisol(t):
    x = (t - 8.0) % 24.0
    if x > 12.0:
        x -= 24.0
    return 0.12 + 0.88 * E ** (-(x / 3.6) ** 2)

def melatonin(t):
    x = (t - 3.0) % 24.0
    if x > 12.0:
        x -= 24.0
    return 0.04 + 0.96 * E ** (-(x / 3.2) ** 2)

SENS = (("morning cofactor", cortisol), ("evening calmer", melatonin))
GRID = 24 * 60

def curve(f):
    n = int(round(48.0 / STEP))
    weights = [plasma(i * STEP) for i in range(n)]
    out = []
    for m in range(GRID):
        hour = m / 60.0
        total = 0.0
        for i in range(n):
            total += weights[i] * f((hour + i * STEP) % 24.0)
        out.append(total * STEP)
    return out

def clock(minutes):
    return "{:02d}:{:02d}".format(int(minutes) // 60, int(minutes) % 60)

if __name__ == "__main__":
    peaks = {}
    for name, f in SENS:
        c = curve(f)
        best = max(range(GRID), key=lambda m: c[m])
        peaks[name] = (best, c[best], c)
        print("   {}: best moment {}, effect {:.4f}"
              .format(name, clock(best), c[best]))
        for tol in TOL:
            thr = c[best] * (1.0 - tol)
            lo = best
            while c[(lo - 1) % GRID] >= thr:
                lo = (lo - 1) % GRID
            hi = best
            while c[(hi + 1) % GRID] >= thr:
                hi = (hi + 1) % GRID
            width = ((hi - lo) % GRID) / 60.0
            print("      within {:.2f} of the peak: {} to {}, {:.2f} hours"
                  .format(tol, clock(lo), clock(hi), width))
    a, b = peaks["morning cofactor"], peaks["evening calmer"]
    gap = ((b[0] - a[0]) % GRID) / 60.0
    print("   phase difference: {:.2f} hours, {:.1f} degrees"
          .format(gap, 360.0 * gap / 24.0))
    print("   cost of an error, as a share of the peak effect:")
    print("   error h   morning cofactor   evening calmer")
    for err in (0.5, 1.0, 2.0, 3.0, 6.0, 12.0):
        row = []
        for name, _ in SENS:
            m0, peak, c = peaks[name]
            k = int(round(err * 60.0))
            late = c[(m0 + k) % GRID]
            early = c[(m0 - k) % GRID]
            row.append(1.0 - min(late, early) / peak)
        print("   {:<10.1f}{:<19.4f}{:.4f}".format(err, row[0], row[1]))
    print("   a six-hour error costs {:.4f} on the cofactor and {:.4f}"
          .format(1.0 - min(peaks["morning cofactor"][2][
              (peaks["morning cofactor"][0] + 360) % GRID],
              peaks["morning cofactor"][2][
                  (peaks["morning cofactor"][0] - 360) % GRID])
              / peaks["morning cofactor"][1],
              1.0 - min(peaks["evening calmer"][2][
                  (peaks["evening calmer"][0] + 360) % GRID],
                  peaks["evening calmer"][2][
                      (peaks["evening calmer"][0] - 360) % GRID])
              / peaks["evening calmer"][1]))
    print("   on the calmer, which is the price of one schedule slip")
#    morning cofactor: best moment 04:32, effect 3.5337
#       within 0.05 of the peak: 03:05 to 05:51, 2.77 hours
#       within 0.10 of the peak: 02:23 to 06:24, 4.02 hours
#    evening calmer: best moment 23:42, effect 3.1259
#       within 0.05 of the peak: 22:27 to 00:49, 2.37 hours
#       within 0.10 of the peak: 21:50 to 01:17, 3.45 hours
#    phase difference: 19.17 hours, 287.5 degrees
#    cost of an error, as a share of the peak effect:
#    error h   morning cofactor   evening calmer
#    0.5       0.0069             0.0097
#    1.0       0.0283             0.0391
#    2.0       0.1132             0.1556
#    3.0       0.2412             0.3270
#    6.0       0.6156             0.7652
#    12.0      0.6637             0.7712
#    a six-hour error costs 0.6156 on the cofactor and 0.7652
#    on the calmer, which is the price of one schedule slip

最佳时刻是晨间辅因子的 04:32 与夜间镇静剂的 23:42,两者的相位差是 19.17 小时,也就是 287.5 度。损失不到二十分之一的窗口,辅因子宽 2.77 小时,从 03:05 到 05:51;镇静剂宽 2.37 小时,从 22:27 到 00:49。把容差放宽到十分之一,则分别为 4.02 与 3.45 小时。

那张误差表才是要记住的部分。半小时花 0.0069 与 0.0097,一小时花 0.0283 与 0.0391,两小时花 0.1132 与 0.1556,而六小时花 0.6156 与 0.7652。迟一小时几乎免费;迟六小时则把大部分剂量扔掉。而六小时的偏差,恰恰就是前一晚忘了、第二天中午补上时所发生的事,于是随之而来的规则是:跳过,不要挪。

假说 · 尚未验证

作者的诠释(尚未验证): 对敏感度在一天之中变化的药剂,一剂漏服后在偏离三小时以上时补上,交付量低于跳过它、按时服下一剂,所以服药说明应当写「跳过」而不是「想起就补」。什么会证明它错: 如果对某种药剂而言,在刻意挪动一剂之后测得的药效学终点并不比跳过一剂更差,那么对那种药剂就不需要这条跳过规则。

小结

在最差而非最佳时辰给药,晨间那一类花 0.7068,夜间那一类花 0.8330;但把选择限制在醒着的时辰内,只花 0.0556 与 0.0818。最优点是 04:32 与 23:42,相位差 19.17 小时即 287.5 度,而损失不到二十分之一的窗口宽 2.77 与 2.37 小时。一小时的误差花费不到 0.04,六小时的误差花 0.6156 与 0.7652。第五章转向肝脏,时钟与相互作用在那里相遇。

参考文献

Czeisler, Charles A., Jeanne F. Duffy, Theresa L. Shanahan, et al. 1999. “Stability, Precision, and near-24-Hour Period of the Human Circadian Pacemaker.” Science 284 (5423): 2177 to 2181.
Hermida, Ramon C., Diana E. Ayala, Artemio Mojon, and Jose R. Fernandez. 2010. “Influence of Circadian Time of Hypertension Treatment on Cardiovascular Risk: Results of the MAPEC Study.” Chronobiology International 27 (8): 1629 to 1651.
Lewy, Alfred J., Saeeduddin Ahmed, Jeanne M. Jackson, and Robert L. Sack. 1992. “Melatonin Shifts Human Circadian Rhythms According to a Phase-Response Curve.” Chronobiology International 9 (5): 380 to 392.
Smolensky, Michael H., and Nicholas A. Peppas. 2007. “Chronobiology, Drug Delivery, and Chronotherapeutics.” Advanced Drug Delivery Reviews 59 (9 to 10): 828 to 851.

Fan, Y. (2026). Polypharmacy Supplement Interactions, Chronotherapy Dosing Windows and Cytochrome Enzyme Competition: Michaelis-Menten Inhibition Matrices, Chelation Absorption Losses and Circadian Receptor Sensitivity [复合补剂相互作用、时间疗法给药窗口与代谢酶竞争] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090250 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).