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
第二章 前药与饱和转化:饱和削平峰值
前药的概念与已知的部分
一些这类化合物以前药的形式存在:分子本身活性低,在体内被酶转化为活性物。
证据
去磷酸化(组织匀浆与人体)。 赛洛西宾是磷酸酯,在体内去磷酸化为赛洛新。Horita 与 Weber 1961 年在哺乳动物组织匀浆中证明了这一酶促去磷酸化及其后的氧化 (Horita and Weber 1961);Dinis-Oliveira 综述了赛洛西宾与赛洛新的代谢及其法医毒理学意义 (Dinis-Oliveira 2017)。人体药代动力学方面,Brown 等人在健康成人中作了递增剂量研究,报告了赛洛新的血浆浓度—时间特征 (Brown et al. 2017)。
注
本章不处理的东西。 本章只讨论转化步骤的数学结构。它不给出任何化合物的质量、浓度、剂量或换算关系,不讨论来源、纯度或制备,也不比较不同化合物的效力。全部数值都是无量纲的,相对于一个未指定的参考值。
而乙醯基前药——本书副题中「前药脱乙醯基」所指的那一类——情况不同。
证据
人体转化动力学:未找到已发表数据。 新精神活性物质中的色胺类有综述,但以分析化学、流行病学与毒理报告为主 (Tittarelli et al. 2015)。本书核查未找到乙醯基前药在人体内的脱乙醯基转化动力学(、 或等价参数)的已发表测量。因此本章的模型参数是假设值。若有读者知道已发表的人体转化动力学数据,欢迎通过 5D Library 网站指正。
米氏动力学与两种状态
设前药浓度 ,活性物浓度 ,转化由一个可饱和的酶催化,米氏方程 (Johnson and Goody 2011):
当 时转化速率近似为 ,即一阶;当 时速率趋于 ,即零阶(饱和)。
推导
# Chapter 2. A prodrug converted by a saturable enzyme, in dimensionless
# units. An ester prodrug A is hydrolysed to the active species P, which is
# then eliminated:
# dA/dt = -Vmax A / (Km + A) - ka_loss A
# dP/dt = +Vmax A / (Km + A) - kel P
# Concentrations are in units of Km_ref (an arbitrary reference), time in
# units of the elimination half-life of P. NO mass, dose, product, route or
# source appears anywhere in this book; D below is a dose FRACTION relative to
# an unspecified reference exposure. The point is a distinction that is easy
# to get wrong. Saturation of the converting step does NOT make total exposure
# disproportionate: almost all of the prodrug converts eventually either way,
# and mass balance fixes the area under the curve. What saturation changes is
# the SHAPE - the peak and the time to peak - because a saturated enzyme works
# at a fixed rate and spreads the same total conversion over a longer time. So
# the common claim that a saturable prodrug step makes exposure "jump" at
# higher amounts is wrong for total exposure and right for the peak. Which
# regime applies depends on Km, which for the compounds at issue has not been
# measured in humans (Chapter 2 text).
import math
DT, T_END = 0.001, 12.0
KEL = math.log(2.0) # time unit = elimination half-life of P
KA_LOSS = 0.05 # competing loss of unconverted prodrug
def simulate(D, Km, Vmax=1.0):
A, P = D, 0.0
peak, tpeak, auc = 0.0, 0.0, 0.0
n = int(T_END / DT)
for i in range(n):
conv = Vmax * A / (Km + A)
dA = -conv - KA_LOSS * A
dP = conv - KEL * P
A = max(0.0, A + DT * dA)
P = max(0.0, P + DT * dP)
auc += P * DT
if P > peak:
peak, tpeak = P, (i + 1) * DT
return peak, tpeak, auc
if __name__ == "__main__":
print(
"Dimensionless prodrug model. Time unit = elimination half-life of"
" the active"
)
print(
"species; concentrations in units of a reference Km. D is a dose"
" FRACTION.\n"
)
print("A. Km = 10 (prodrug far below saturation: first-order regime)")
print(" D Cmax Tmax AUC AUC / D")
base = None
for D in (0.1, 0.2, 0.5, 1.0, 2.0, 5.0):
pk, tp, auc = simulate(D, Km=10.0)
print(
" {:6.2f} {:9.4f} {:9.3f} {:9.4f} {:9.4f}".format(
D, pk, tp, auc, auc / D
)
)
print(
"\nB. Km = 0.05 (enzyme saturated at these amounts: zero-order"
" regime)"
)
print(" D Cmax Tmax AUC AUC / D")
for D in (0.1, 0.2, 0.5, 1.0, 2.0, 5.0):
pk, tp, auc = simulate(D, Km=0.05)
print(
" {:6.2f} {:9.4f} {:9.3f} {:9.4f} {:9.4f}".format(
D, pk, tp, auc, auc / D
)
)
print(
"\nAUC / D is nearly constant in BOTH regimes, and that is the first"
" thing to"
)
print(
"notice: saturating the converting enzyme does not make total"
" exposure jump,"
)
print(
"because mass balance fixes it - almost all the prodrug converts"
" either way."
)
print("What saturation changes is the shape:")
for Km in (10.0, 0.05):
r1 = simulate(0.5, Km)
r2 = simulate(5.0, Km)
print(
" Km = {:5.2f}: tenfold D raises Cmax x {:.2f} and Tmax"
" x {:.2f}".format(Km, r2[0] / r1[0], r2[1] / r1[1])
)
print(
" Unsaturated, a tenfold rise in D raises the peak almost"
" tenfold and"
)
print(
" barely moves its timing. Saturated, the peak rises only 3.5-fold"
" while"
)
print(
" the time to peak stretches more than sixfold: the same total"
" amount is"
)
print(
" delivered more slowly and more flatly. So a saturable step"
" BLUNTS the"
)
print(
" peak rather than amplifying exposure - the opposite of what the"
" common"
)
print(" account of prodrug saturation suggests.")
print(
"\nC. The quantity that is NOT known: Km. Sweeping it over three"
" orders of"
)
print("magnitude at a FIXED dose fraction D = 1.0:")
print(
" Km Cmax Tmax AUC fraction of prodrug"
" converted"
)
for Km in (0.01, 0.05, 0.2, 1.0, 5.0, 20.0):
pk, tp, auc = simulate(1.0, Km)
# mass balance: AUC*kel = total P formed
conv_frac = auc * KEL
print(" {:7.2f} {:9.4f} {:9.3f} {:9.4f} {:20.3f}"
.format(Km, pk, tp, auc, conv_frac))
lo = simulate(1.0, 0.01)
hi = simulate(1.0, 20.0)
print(
"\n Across that range Cmax varies by a factor of {:.2f} and Tmax by"
" {:.2f},".format(
max(lo[0], hi[0]) / min(lo[0], hi[0]),
max(lo[1], hi[1]) / min(lo[1], hi[1]),
)
)
print(
" at the SAME dose fraction. So without a measured Km the model"
" cannot"
)
print(
" predict either the peak or its timing, even in these arbitrary"
" units."
)
print(
" Chapter 2 argues that this is the actual state of knowledge"
" for the"
)
print(
" prodrugs in question: no published human conversion kinetics"
" exist, so"
)
print(
" any such curve is parameterised by assumption, not by"
" measurement."
)
# Dimensionless prodrug model. Time unit = elimination half-life of the active
# species; concentrations in units of a reference Km. D is a dose FRACTION.
#
# A. Km = 10 (prodrug far below saturation: first-order regime)
# D Cmax Tmax AUC AUC / D
# 0.10 0.0094 2.830 0.0756 0.7565
# 0.20 0.0187 2.844 0.1508 0.7539
# 0.50 0.0458 2.882 0.3732 0.7463
# 1.00 0.0890 2.945 0.7338 0.7338
# 2.00 0.1681 3.065 1.4194 0.7097
# 5.00 0.3595 3.389 3.2157 0.6431
#
# B. Km = 0.05 (enzyme saturated at these amounts: zero-order regime)
# D Cmax Tmax AUC AUC / D
# 0.10 0.0855 0.268 0.1435 1.4351
# 0.20 0.1651 0.362 0.2863 1.4316
# 0.50 0.3731 0.641 0.7105 1.4210
# 1.00 0.6378 1.093 1.4038 1.4038
# 2.00 0.9658 1.945 2.7420 1.3710
# 5.00 1.3035 4.100 6.4128 1.2826
#
# AUC / D is nearly constant in BOTH regimes, and that is the first thing to
# notice: saturating the converting enzyme does not make total exposure jump,
# because mass balance fixes it - almost all the prodrug converts either way.
# What saturation changes is the shape:
# Km = 10.00: tenfold D raises Cmax x 7.84 and Tmax x 1.18
# Km = 0.05: tenfold D raises Cmax x 3.49 and Tmax x 6.40
# Unsaturated, a tenfold rise in D raises the peak almost tenfold and
# barely moves its timing. Saturated, the peak rises only 3.5-fold while
# the time to peak stretches more than sixfold: the same total amount is
# delivered more slowly and more flatly. So a saturable step BLUNTS the
# peak rather than amplifying exposure - the opposite of what the common
# account of prodrug saturation suggests.
#
# C. The quantity that is NOT known: Km. Sweeping it over three orders of
# magnitude at a FIXED dose fraction D = 1.0:
# Km Cmax Tmax AUC fraction of prodrug converted
# 0.01 0.6875 1.013 1.4066 0.975
# 0.05 0.6378 1.093 1.4038 0.973
# 0.20 0.5388 1.262 1.3936 0.966
# 1.00 0.3460 1.701 1.3407 0.929
# 5.00 0.1474 2.524 1.0290 0.713
# 20.00 0.0502 3.329 0.4559 0.316
#
# Across that range Cmax varies by a factor of 13.68 and Tmax by 3.29,
# at the SAME dose fraction. So without a measured Km the model cannot
# predict either the peak or its timing, even in these arbitrary units.
# Chapter 2 argues that this is the actual state of knowledge for the
# prodrugs in question: no published human conversion kinetics exist, so
# any such curve is parameterised by assumption, not by measurement.程序的结果与常见的说法不同,这是本章的要点。
常见的说法是:前药转化饱和会使暴露「跳升」,所以稍微加量就可能出大事。在这个模型里这不成立。看 A、B 两段的 AUC/剂量列:两种状态下它都接近常数。原因是质量守恒——几乎所有前药最终都会转化(C 段的最后一列给出转化比例),而曲线下面积由总转化量与清除率决定,不由转化的速率决定。
饱和改变的是形状。未饱和时,十倍剂量使峰值升高 7.84 倍、达峰时间几乎不动(1.18 倍);饱和时,峰值只升高 3.49 倍、而达峰时间拉长 6.40 倍。也就是说:
饱和的前药步骤把同样的总量更慢、更平地送出去。它削平峰值,不放大暴露。
注
这不是说饱和无关紧要。 峰值与达峰时间对主观体验的强度与起效感受很可能更重要(第三章说明效应时程还受 ke0 支配)。本章只是指出:哪一个量被饱和改变,常被说反了。如果要论证「加量有风险」,不能用「转化饱和使暴露跳升」这个理由,因为它在这个模型里不成立。
不知道米氏常数的代价
C 段把 在三个数量级上扫过,剂量比例固定。峰值变化 13.68 倍,达峰时间变化 3.29 倍。所以在没有测量 的情况下,模型既不能预测峰值也不能预测它的时点,即使在这些任意单位下也不能。
那么能不能从已发表的数据反推出 ?下一段程序回答这个问题,而答案出乎意料地是正面的。
推导
# Chapter 2. Can Vmax and Km be recovered from measurements of the ACTIVE
# species alone? This is the practical question for any attempt to
# parameterise the model of p02 from published data, because the prodrug
# itself is usually not assayed. The program fits the two-parameter conversion
# model to a synthetic concentration-time curve of P (sampled at realistic
# times, with 10 % proportional noise) by grid search on the sum of squares,
# and reports the set of (Vmax, Km) pairs that fit within the noise. If that
# set spans orders of magnitude, the parameters are practically unidentifiable
# from such data. Dimensionless units throughout; synthetic data; fixed seed;
# standard library only.
import math
import random
DT, T_END = 0.002, 12.0
KEL = math.log(2.0)
KA_LOSS = 0.05
SAMPLE_T = [0.25, 0.5, 0.75, 1.0, 1.5, 2.0, 3.0, 4.0, 6.0, 8.0]
NOISE = 0.10
def curve(D, Km, Vmax):
A, P = D, 0.0
out, want, k = [], sorted(SAMPLE_T), 0
n = int(T_END / DT)
for i in range(n):
t = (i + 1) * DT
conv = Vmax * A / (Km + A)
A = max(0.0, A + DT * (-conv - KA_LOSS * A))
P = max(0.0, P + DT * (conv - KEL * P))
while k < len(want) and t >= want[k] - DT / 2:
out.append(P)
k += 1
while len(out) < len(SAMPLE_T):
out.append(P)
return out
TRUE_KM, TRUE_VMAX, D = 0.30, 1.00, 1.0
def ssq(obs, pred):
return math.fsum((a - b) ** 2 for a, b in zip(obs, pred))
if __name__ == "__main__":
rng = random.Random(5202)
truth = curve(D, TRUE_KM, TRUE_VMAX)
obs = [v * (1.0 + rng.gauss(0.0, NOISE)) for v in truth]
print(
"Truth: Km = {:.2f}, Vmax = {:.2f}. {} samples, {:.0%} proportional"
" noise.".format(TRUE_KM, TRUE_VMAX, len(SAMPLE_T), NOISE)
)
print("Observed active-species curve:")
for a, b in ((0, 5), (5, len(SAMPLE_T))):
print(
" t " + "".join("{:8.2f}".format(t) for t in SAMPLE_T[a:b])
)
print(" P " + "".join("{:8.4f}".format(v) for v in obs[a:b]))
kms = [0.01 * (10 ** (0.1 * k)) for k in range(41)] # 0.01 .. 100
vms = [0.1 * (10 ** (0.05 * k)) for k in range(41)] # 0.1 .. 10
best, grid = None, []
for Km in kms:
for Vm in vms:
s = ssq(obs, curve(D, Km, Vm))
grid.append((s, Km, Vm))
if best is None or s < best[0]:
best = (s, Km, Vm)
s0, km0, vm0 = best
print(
"\nBest fit: Km = {:.3f}, Vmax = {:.3f} (SSQ {:.3e})".format(
km0, vm0, s0
)
)
# all pairs within an F-like tolerance of the optimum: a 2-parameter
# confidence region
tol = s0 * (1.0 + 2.0 * 2.0 / (len(SAMPLE_T) - 2))
inside = [(Km, Vm) for s, Km, Vm in grid if s <= tol]
kmin = min(k for k, _ in inside)
kmax = max(k for k, _ in inside)
vmin = min(v for _, v in inside)
vmax_ = max(v for _, v in inside)
print(
"Pairs fitting within the noise tolerance: {} of {}".format(
len(inside), len(grid)
)
)
print(" Km spans {:.3f} to {:.3f} (a factor of {:.0f})"
.format(kmin, kmax, kmax / kmin))
print(" Vmax spans {:.3f} to {:.3f} (a factor of {:.1f})"
.format(vmin, vmax_, vmax_ / vmin))
print(
"\nWhat those extremes predict for a FIVEFOLD larger dose fraction:"
)
print(" Km Vmax Cmax(D=5) AUC(D=5)")
def summarise(Km, Vm, Dd):
A, P, peak, auc = Dd, 0.0, 0.0, 0.0
for i in range(int(T_END / DT)):
conv = Vm * A / (Km + A)
A = max(0.0, A + DT * (-conv - KA_LOSS * A))
P = max(0.0, P + DT * (conv - KEL * P))
auc += P * DT
peak = max(peak, P)
return peak, auc
rows = []
for Km, Vm in ((km0, vm0), (kmin, min(v for k, v in inside if k == kmin)),
(kmax, max(v for k, v in inside if k == kmax))):
pk, au = summarise(Km, Vm, 5.0)
rows.append((pk, au))
print(" {:7.3f} {:8.3f} {:11.4f} {:10.4f}".format(Km, Vm, pk, au))
cmax_spread = max(r[0] for r in rows) / min(r[0] for r in rows)
auc_spread = max(r[1] for r in rows) / min(r[1] for r in rows)
print(
"\n Curves indistinguishable at D = 1 predict Cmax at D = 5"
" differing by a"
)
print(
" factor of only {:.2f} and AUC by {:.2f}.".format(
cmax_spread, auc_spread
)
)
print(
"\nThat is a POSITIVE result, and it has to be reported as one. Km"
" itself is"
)
print(
"recovered only to within a factor of {:.0f}, but the predictions"
" that matter".format(kmax / kmin)
)
print("are robust to that ignorance, because mass balance pins the total")
print("conversion whatever the enzyme's affinity. So structural")
print(
"non-identifiability is NOT the obstacle here: given one decent human"
)
print(
"concentration-time curve, extrapolating exposure to a nearby amount"
)
print("would be sound to within tens of per cent.")
print(
"\nThe obstacle is that there is no such curve to fit. Compare"
" the two"
)
print("situations directly, both at a fivefold larger dose fraction:")
def spread_over_km_prior(Dd, kms_prior):
vals = [summarise(k, 1.0, Dd) for k in kms_prior]
pk = [v[0] for v in vals]
return max(pk) / min(pk)
no_data = spread_over_km_prior(5.0, [0.01, 0.05, 0.2, 1.0, 5.0, 20.0])
print(
" with one fitted curve: Cmax uncertain by a factor of"
" {:.2f}".format(cmax_spread)
)
print(" with no data, Km only bounded")
print(
" to three orders of magnitude: Cmax uncertain by a factor of"
" {:.2f}".format(no_data)
)
print(
"\n The second number is the real state of knowledge for the"
" prodrugs at"
)
print(
" issue: no published human conversion kinetics exist, so the curve"
" is not"
)
print(
" poorly constrained by data, it is unconstrained by data. The"
" remedy is a"
)
print(
" measurement, not a better estimator - and until one exists,"
" statements of"
)
print(
" the form 'at a fraction of the usual exposure the kinetics are"
" linear'"
)
print(" are assumptions about Km rather than findings.")
# Truth: Km = 0.30, Vmax = 1.00. 10 samples, 10% proportional noise.
# Observed active-species curve:
# t 0.25 0.50 0.75 1.00 1.50
# P 0.1836 0.2445 0.4348 0.4349 0.5006
# t 2.00 3.00 4.00 6.00 8.00
# P 0.4044 0.1993 0.1063 0.0276 0.0064
#
# Best fit: Km = 0.316, Vmax = 1.000 (SSQ 6.379e-03)
# Pairs fitting within the noise tolerance: 18 of 1681
# Km spans 0.050 to 0.631 (a factor of 13)
# Vmax spans 0.708 to 1.259 (a factor of 1.8)
#
# What those extremes predict for a FIVEFOLD larger dose fraction:
# Km Vmax Cmax(D=5) AUC(D=5)
# 0.316 1.000 1.1475 6.3330
# 0.050 0.708 0.9577 6.1331
# 0.631 1.259 1.2506 6.4194
#
# Curves indistinguishable at D = 1 predict Cmax at D = 5 differing by a
# factor of only 1.31 and AUC by 1.05.
#
# That is a POSITIVE result, and it has to be reported as one. Km itself is
# recovered only to within a factor of 13, but the predictions that matter
# are robust to that ignorance, because mass balance pins the total
# conversion whatever the enzyme's affinity. So structural
# non-identifiability is NOT the obstacle here: given one decent human
# concentration-time curve, extrapolating exposure to a nearby amount
# would be sound to within tens of per cent.
#
# The obstacle is that there is no such curve to fit. Compare the two
# situations directly, both at a fivefold larger dose fraction:
# with one fitted curve: Cmax uncertain by a factor of 1.31
# with no data, Km only bounded
# to three orders of magnitude: Cmax uncertain by a factor of 6.13
#
# The second number is the real state of knowledge for the prodrugs at
# issue: no published human conversion kinetics exist, so the curve is not
# poorly constrained by data, it is unconstrained by data. The remedy is a
# measurement, not a better estimator - and until one exists, statements of
# the form 'at a fraction of the usual exposure the kinetics are linear'
# are assumptions about Km rather than findings.结果分两层。第一层:从单一暴露下活性物的浓度—时间曲线(10 个采样点、10% 比例噪声)拟合, 只能定到 13 倍以内, 定到 1.8 倍以内。单看这个,参数是定不准的。
第二层是要点:这些在 处无法区分的曲线,在 处预测的峰值只差 1.31 倍、AUC 只差 1.05 倍。原因还是质量守恒:无论酶的亲和力如何,总转化量都被约束住了。所以
结构不可识别性不是这里的障碍。有一条像样的人体浓度—时间曲线可拟合,把暴露外推到相近的量就能做到几十个百分点的精度。
真正的障碍是没有那条曲线。程序最后把两种处境并排:有一条拟合曲线时峰值的不确定度是 1.31 倍;没有数据而 只能在三个数量级内取值时,不确定度是 6.13 倍。第二个数字是这类前药的实际知识状态。
注
为什么要把正面结果写出来。 按第 33 本的立场,无论结果支持还是不支持作者的框架都照实写。这里的结果对「模型不可用」这一悲观判断是不利的:模型是可用的,只要有一次测量。把它写成「参数不可识别所以什么都说不了」会是一个更戏剧化但错误的结论,而错误的悲观与错误的乐观同样有害——它会让人以为这类问题原则上无解,从而不去做那个本来能解决问题的测量。
假说 · 尚未验证
作者的假说(尚未确立): 对乙醯基前药,人体内的脱乙醯基转化在本书所讨论的暴露范围内是一阶的(即 ),因此活性物的暴露与前药的量近似成比例,且达峰时间与剂量基本无关。什么样的结果能证明它是错的: 在取得监管批准的人体药代动力学研究中,若达峰时间随剂量系统性延长(例如最高与最低剂量组的达峰时间比的 95% 置信区间下限大于 1.5),则本假说错。注意本假说不对任何剂量作出判断,它只对动力学的函数形式作出判断。
本章小结
赛洛西宾的去磷酸化自 1961 年起就有组织匀浆证据,其代谢有综述,人体药代动力学有递增剂量研究。乙醯基前药的人体转化动力学,本书核查未找到已发表数据,所以本章的米氏参数是假设值。模型给出一个方向与常见说法相反的结果:饱和的前药步骤不使总暴露跳升——质量守恒把 AUC 固定住——而是削平峰值并拉长达峰时间(十倍剂量下,未饱和时峰值升 7.84 倍、达峰时间 1.18 倍;饱和时峰值升 3.49 倍、达峰时间 6.40 倍)。不知道 时,同一剂量比例下峰值的不确定度达 13.68 倍。但从单一暴露的曲线拟合给出一个正面结果: 虽只定到 13 倍以内,五倍暴露下的峰值预测只差 1.31 倍,因为质量守恒约束了总转化。所以障碍不是不可识别性,而是没有数据可拟合——此时峰值的不确定度是 6.13 倍。
参考文献
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).