5D Library › 亚知觉色胺药代动力学、5-HT2A 受体脱敏与神经可塑性窗口的生物数学模型 · English

第六章 可塑性窗口:闸门,不是驱动

已知的部分

证据

结构可塑性(细胞与动物实验)。 Ly 等人 2018 年报告这一类化合物在培养神经元与大鼠中促进树突棘密度与树突复杂度的增加 (Ly et al. 2018)。Shao 等人用在体双光子成像报告赛洛西宾在小鼠额叶皮层引起树突棘的快速且持久的生长 (Shao et al. 2021)。Olson 把这一类化合物归为「psychoplastogen」,即促可塑性药物 (Olson 2018)。

证据

机制上的两个意外。 第一,Hesselgrave 等人发现小鼠中赛洛西宾的抗抑郁样行为与突触效应不被 5-HT2 受体拮抗剂阻断 (Hesselgrave et al. 2021),这与「效应必经 5-HT2A 占据」不一致。第二,Moliner 等人报告这类化合物可以直接结合 BDNF 的受体 TrkB,并以此促进可塑性 (Moliner et al. 2023)。两项结果都提示存在不经 5-HT2A 的通路。

注

这对剂量推理的影响。 如果存在不经 5-HT2A 的通路,那么第一章以受体占据为基础的全套剂量推理就不适用于那条通路,而两条通路的剂量—反应关系没有理由相同。这使情况更不确定,不是更确定。本书据此不对任何暴露水平下「可塑性是否被激发」作判断。

窗口需要输入

近年最有意思的结果不是「可塑性增加了」,而是关于窗口的。

证据

关键期的重开(动物实验)。 Nardou 等人报告,致幻剂重开小鼠社会奖赏学习的关键期,而重开窗口的持续时间在不同化合物之间不同 (Nardou et al. 2023)。要紧的是这个框架的含义:窗口的作用是允许学习发生,而学习还需要有东西可学——需要在窗口期内有相应的经验输入。

把这写成模型。一个可塑性许可状态 WW 在暴露后打开并以速率 kck_c 关闭;突触权重 SS 只在 WW 打开且输入 II 存在时变化:

dWdt=astim(t)(1−W)−kcW,dSdt=ηWI(t)(Starget−S)−γS.\frac{dW}{dt} = a\,\text{stim}(t)(1 – W) – k_c W, \qquad \frac{dS}{dt} = \eta\,W\,I(t)\,(S_{\text{target}} – S) – \gamma S .

推导

# Chapter 6. A "window of plasticity", written as a two-state model, and what
# a window can and cannot do on its own. A plasticity-permissive state W opens
# after an exposure and closes with rate kc; synaptic weight S changes only
# while W is open AND an input I is present:
#   dW/dt = a * stim(t) * (1 - W) - kc * W
#   dS/dt = eta * W * I(t) * (target - S) - decay * S
# The point is in the second equation: the window is a GATE, not a drive. With
# I = 0 the weight does not move however wide the window. So "a window was
# opened" is not by itself a mechanism of benefit, and the thing that has to
# be scheduled is the input, not only the exposure. Author's model;
# dimensionless; time in days.
import math

DT = 0.001
A_STIM, ETA, DECAY, TARGET = 8.0, 0.9, 0.012, 1.0

def run(open_half_life, inputs, exposures=(0.0,), t_end=30.0, input_amp=1.0):
    """inputs: list of (start, duration) in days when the practice/input is
    present."""
    kc = math.log(2.0) / open_half_life
    W, S = 0.0, 0.0
    peakW, auc_W, auc_WI = 0.0, 0.0, 0.0
    for i in range(int(t_end / DT)):
        t = i * DT
        stim = 1.0 if any(abs(t - e) < 0.05 for e in exposures) else 0.0
        I = input_amp if any(s <= t < s + d for s, d in inputs) else 0.0
        W += DT * (A_STIM * stim * (1.0 - W) - kc * W)
        S += DT * (ETA * W * I * (TARGET - S) - DECAY * S)
        peakW = max(peakW, W)
        auc_W += W * DT
        auc_WI += W * I * DT
    return peakW, auc_W, auc_WI, S

if __name__ == "__main__":
    print(
        "One exposure at day 0. Window half-life varied. The input"
        " ('practice') is"
    )
    print(
        "present on days 0-2 in the first block and absent in the second.\n"
    )
    print(
        "   window half-life (d)   peak W   AUC of W   AUC of W x I   final S"
    )
    for hl in (0.25, 0.5, 1.0, 2.0, 4.0):
        pw, aw, awi, S = run(hl, inputs=[(0.0, 2.0)])
        print(
            "   {:20.2f} {:9.3f} {:10.3f} {:14.3f} {:9.4f}".format(
                hl, pw, aw, awi, S
            )
        )

    print("\n   Same exposures, NO input at all:")
    print(
        "   window half-life (d)   peak W   AUC of W   AUC of W x I   final S"
    )
    for hl in (0.25, 0.5, 1.0, 2.0, 4.0):
        pw, aw, awi, S = run(hl, inputs=[])
        print(
            "   {:20.2f} {:9.3f} {:10.3f} {:14.3f} {:9.4f}".format(
                hl, pw, aw, awi, S
            )
        )

    print(
        "\n   The window is identical in the two blocks - same peak, same"
        " area - and"
    )
    print(
        "   the weight ends at 0 in the second. A measurement of the"
        " window (a"
    )
    print(
        "   marker, spine density, a plasticity assay) therefore cannot"
        " predict"
    )
    print(
        "   whether anything was learned; it reports that the gate was open."
    )

    print("\nTiming of the input relative to a window of half-life 1 day:")
    print(
        "   input window (days)   AUC of W x I   final S   S relative to best"
    )
    rows = []
    for start, dur in (
        (0.0, 2.0),
        (1.0, 2.0),
        (2.0, 2.0),
        (4.0, 2.0),
        (7.0, 2.0),
        (0.0, 7.0),
    ):
        pw, aw, awi, S = run(1.0, inputs=[(start, dur)])
        rows.append((start, dur, awi, S))
    best = max(r[3] for r in rows)
    for start, dur, awi, S in rows:
        print("   day {:.0f} to {:.0f} {:20.3f} {:9.4f} {:17.1%}"
              .format(start, start + dur, awi, S, S / best))

    d02 = [r for r in rows if r[0] == 0.0 and r[1] == 2.0][0][3]
    d07 = [r for r in rows if r[1] == 7.0][0][3]
    d46 = [r for r in rows if r[0] == 4.0][0][3]
    d79 = [r for r in rows if r[0] == 7.0][0][3]
    print(
        "\n   What matters is LATENCY, not duration. Two days of input"
        " starting at"
    )
    print(
        "   the exposure reaches {:.0%} of the best case; the same two days"
        " delayed".format(d02 / best)
    )
    print(
        "   to days 4-6 reaches {:.0%} and to days 7-9 {:.0%}. Extending the"
        " input".format(d46 / best, d79 / best)
    )
    print(
        "   from two days to seven adds only {:+.0%}, because the window has"
        " closed".format(d07 / d02 - 1.0)
    )
    print(
        "   and the extra days fall outside it. So within this model there"
        " is a"
    )
    print(
        "   schedule that matters, and it is a schedule of PRACTICE"
        " relative to"
    )
    print(
        "   exposure - not the dosing schedule that popular accounts discuss."
    )

    print(
        "\nRepeated exposures with input always present, window half-life 1"
        " day:"
    )
    print("   exposures (days)   final S")
    for label, ex in (("day 0 only", (0.0,)), ("days 0, 3", (0.0, 3.0)),
                      ("days 0, 3, 6", (0.0, 3.0, 6.0)),
                      ("days 0, 1, 2, 3", (0.0, 1.0, 2.0, 3.0))):
        pw, aw, awi, S = run(1.0, inputs=[(0.0, 30.0)], exposures=ex)
        print("   {:18s} {:9.4f}".format(label, S))
    print(
        "\n   NOTE what this model does NOT include: any dependence of the"
        " window on"
    )
    print(
        "   receptor density. Chapter 4 showed receptors ratchet down under"
        " tight"
    )
    print(
        "   schedules, so a model combining the two would predict diminishing"
    )
    print(
        "   windows. Writing that model requires the recovery half-life"
        " tau_R, which"
    )
    print(
        "   is unmeasured, so this book does not write it and does not"
        " report a"
    )
    print("   number for the best schedule.")
# One exposure at day 0. Window half-life varied. The input ('practice') is
# present on days 0-2 in the first block and absent in the second.
#
#    window half-life (d)   peak W   AUC of W   AUC of W x I   final S
#                    0.25     0.311      0.120          0.120    0.0717
#                    0.50     0.320      0.240          0.224    0.1285
#                    1.00     0.326      0.478          0.357    0.1936
#                    2.00     0.328      0.955          0.474    0.2451
#                    4.00     0.329      1.899          0.554    0.2774
#
#    Same exposures, NO input at all:
#    window half-life (d)   peak W   AUC of W   AUC of W x I   final S
#                    0.25     0.311      0.120          0.000    0.0000
#                    0.50     0.320      0.240          0.000    0.0000
#                    1.00     0.326      0.478          0.000    0.0000
#                    2.00     0.328      0.955          0.000    0.0000
#                    4.00     0.329      1.899          0.000    0.0000
#
#    The window is identical in the two blocks - same peak, same area - and
#    the weight ends at 0 in the second. A measurement of the window (a
#    marker, spine density, a plasticity assay) therefore cannot predict
#    whether anything was learned; it reports that the gate was open.
#
# Timing of the input relative to a window of half-life 1 day:
#    input window (days)   AUC of W x I   final S   S relative to best
#    day 0 to 2                0.357    0.1936             78.4%
#    day 1 to 3                0.182    0.1078             43.6%
#    day 2 to 4                0.091    0.0568             23.0%
#    day 4 to 6                0.023    0.0150              6.1%
#    day 7 to 9                0.003    0.0020              0.8%
#    day 0 to 7                0.474    0.2471            100.0%
#
#    What matters is LATENCY, not duration. Two days of input starting at
#    the exposure reaches 78% of the best case; the same two days delayed
#    to days 4-6 reaches 6% and to days 7-9 1%. Extending the input
#    from two days to seven adds only +28%, because the window has closed
#    and the extra days fall outside it. So within this model there is a
#    schedule that matters, and it is a schedule of PRACTICE relative to
#    exposure - not the dosing schedule that popular accounts discuss.
#
# Repeated exposures with input always present, window half-life 1 day:
#    exposures (days)   final S
#    day 0 only            0.2489
#    days 0, 3             0.4937
#    days 0, 3, 6          0.6274
#    days 0, 1, 2, 3       0.6326
#
#    NOTE what this model does NOT include: any dependence of the window on
#    receptor density. Chapter 4 showed receptors ratchet down under tight
#    schedules, so a model combining the two would predict diminishing
#    windows. Writing that model requires the recovery half-life tau_R, which
#    is unmeasured, so this book does not write it and does not report a
#    number for the best schedule.

第一、二张表并排给出本章的要点。两张表的窗口完全相同——同样的峰、同样的面积——而第二张表里最终的突触权重是 0。所以

窗口是闸门,不是驱动。对窗口的测量(某个标志物、树突棘密度、某项可塑性测定)不能预测是否学到了东西;它只报告闸门开过。

这对「打开了可塑性窗口」这句话是一个直接的限定:它本身不是获益的机制陈述,而只是获益的一个必要条件。

左:可塑性许可状态 W 在一次暴露后的时间进程,关闭半衰期 0.25、1、4 天。右:最终突触权重随输入起始时间的变化(窗口半衰期 1 天,输入时长固定为两天)。输入的延迟远比其时长重要。

延迟比时长重要

第三张表给出一个可操作的结果。窗口半衰期为 1 天时,紧接暴露的两天输入达到最佳情形的 78%;同样的两天延到第 4–6 天只有 6%,延到第 7–9 天只有 0.8%。而把输入从两天延长到七天只增加约 28%,因为后面的日子已经在窗口之外。

所以这里确实有一个「日程」问题——但它是练习相对于暴露的日程,不是给药日程。这与第四章合起来说明一件事:流行说法把注意力放在给药间隔上,而按这两章的模型,更要紧的变量是暴露之后做了什么、什么时候做。

假说 · 尚未验证

作者的假说(尚未确立): 在任何能产生可检出获益的安排中,获益的大小更强地取决于暴露后数日内结构化行为输入(练习、治疗、学习任务)的有无与时点,而不是取决于给药间隔。什么样的结果能证明它是错的: 在事先登记的 2×2 因子设计中(两种给药间隔 × 有无结构化输入),若输入主效应的 95% 置信区间包含零而间隔主效应的置信区间排除零,则本假说错。注意这个设计本身不需要知道任何剂量:两个水平都可以取自已获批准的研究方案。

注

本章模型不包含的东西。 这个模型里窗口的宽度与受体密度无关。第四章说明受体在密集日程下会棘轮下降,所以一个把两者合在一起的模型会预测窗口随日程递减。写那个模型需要恢复半衰期 τR\tau_R,而它未被测量;因此本书不写那个模型,也不给出任何「最佳日程」的数字。这是一处有意的留白,而不是遗漏。

本章小结

结构可塑性的证据是实在的:培养神经元与动物中的树突棘与树突复杂度增加,在体成像显示快速且持久的棘生长。机制上有两个意外:小鼠的突触与行为效应不被 5-HT2 拮抗剂阻断,而这类化合物可直接结合 TrkB;两者都提示存在不经 5-HT2A 的通路,这使以受体占据为基础的剂量推理更不适用。关键期重开的框架给出本章的主结果:窗口是闸门而不是驱动——在模型中,没有输入时无论窗口多宽突触权重都停在 0,所以对窗口的测量不能预测是否学到了东西。输入的延迟远比其时长重要:紧接暴露的两天达到最佳的 78%,延到第 4–6 天只有 6%,而把两天延长到七天只增加约 28%。因此真正要紧的日程是练习相对于暴露的日程,不是给药日程。把窗口与受体棘轮合在一起的模型需要未被测量的恢复半衰期,本书不写。

参考文献

Hesselgrave, Natalie, Timothy A. Troppoli, Andreas B. Wulff, Anthony B. Cole, and Scott M. Thompson. 2021. “Harnessing Psilocybin: Antidepressant-Like Behavioral and Synaptic Actions of Psilocybin Are Independent of 5-HT2R Activation in Mice.” Proceedings of the National Academy of Sciences 118 (17): e2022489118. https://doi.org/10.1073/pnas.2022489118.
Ly, Calvin, Alexandra C. Greb, Lindsay P. Cameron, Jonathan M. Wong, Eden V. Barragan, Paige C. Wilson, Kyle F. Burbach, et al. 2018. “Psychedelics Promote Structural and Functional Neural Plasticity.” Cell Reports 23 (11): 3170–82. https://doi.org/10.1016/j.celrep.2018.05.022.
Moliner, Rafael, Mykhailo Girych, Cecilia A. Brunello, Vera Kovaleva, Caroline Biojone, Giray Enkavi, Lina Antenucci, et al. 2023. “Psychedelics Promote Plasticity by Directly Binding to BDNF Receptor TrkB.” Nature Neuroscience 26: 1032–41. https://doi.org/10.1038/s41593-023-01316-5.
Nardou, Romain, Eastman Sawyer, Young Jun Song, Makenzie Wilkinson, Yasmin Padovan-Hernandez, Junia Lara de Deus, Noelle Wright, et al. 2023. “Psychedelics Reopen the Social Reward Learning Critical Period.” Nature 618: 790–98. https://doi.org/10.1038/s41586-023-06204-3.
Olson, David E. 2018. “Psychoplastogens: A Promising Class of Plasticity-Promoting Neurotherapeutics.” Journal of Experimental Neuroscience 12: 1179069518800508. https://doi.org/10.1177/1179069518800508.
Shao, Ling-Xiao, Clara Liao, Ian Gregg, Pasha A. Davoudian, Neil K. Savalia, Kristina Delagarza, and Alex C. Kwan. 2021. “Psilocybin Induces Rapid and Persistent Growth of Dendritic Spines in Frontal Cortex in Vivo.” Neuron 109 (16): 2535–44. https://doi.org/10.1016/j.neuron.2021.06.008.

Fan, Y. (2026). Biomathematical Models of Sub-Perceptual Tryptamine Pharmacokinetics, 5-HT2A Receptor Desensitisation and Neuroplastic Window Dynamics: From Michaelis–Menten Kinetics of Prodrug Deacetylation and Whole-Brain Connectivity Entropy to the Microdose Tolerance Ratchet, with a Textual Examination of the Dose Boundary in the Classical Materia Medica [亚知觉色胺药代动力学、5-HT2A 受体脱敏与神经可塑性窗口的生物数学模型] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24019347 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).