5D Library › 胰岛素胰高血糖素双稳态、内脏脂解动力学与线性规划配餐 · English

第一章 开关:十四到二十小时从哪里来

肝糖原到底能支撑多久

两餐之间,肝脏是唯一把葡萄糖送进血液的器官。它有两条来路:分解自己的糖原,以及用乳酸、甘油与氨基酸重新造出葡萄糖。只有第一条动用储量,而储量很小。在断食成年人身上做的磁共振测量,把餐后的肝糖原定在一百克上下,并显示它稳定下降,随着时间推移,糖异生承担了总输出中越来越大的一份 (Rothman et al. 1991)。Cahill 关于饥饿时燃料使用的长篇综述从另一面给出同样的图景:大脑的需求几乎恒定,所以唯一的问题是哪一条来路在为它付账 (Cahill 2006)。

证据

已发表。 Rothman 等人用碳十三磁共振在一次断食中追踪肝糖原,并直接把糖原分解与糖异生分开,而不是推算这个比例 (Rothman et al. 1991)。Cahill 的综述列出饥饿数日里燃料的更替次序,以及酮体开始承担大脑大部分供给的那个节点 (Cahill 2006)。

这就让那个流行的说法可以被检验了。如果储量是若干克,而糖原分解以每小时若干克的速度把它抽干,耗尽时间就是这个商。两个数字在人与人之间、在同一个人身上都会变,所以诚实的答案是一张格子,不是一个点。

推导

# Chapter 1. p01: How long the liver's glycogen lasts. The liver is the
# only organ that releases glucose into the blood between meals. The
# time to run the store down is the store divided by the part of
# hepatic glucose output that glycogenolysis supplies; gluconeogenesis
# supplies the rest and does not draw on the store. The grid below
# spans the stores and drain rates reported for healthy adults.
STORE = (80, 90, 100, 110, 120)        # g of liver glycogen after a meal
DRAIN = (4.5, 5.0, 5.5, 6.0, 6.5)      # g per hour from glycogenolysis
LOW, HIGH = 14.0, 20.0                 # the window the popular claim uses

def quantile(values, a):
    s = sorted(values)
    return s[int(a * (len(s) - 1))]

if __name__ == "__main__":
    print("   time to empty the store, hours: store (g) by drain (g/h)")
    print("   store" + "".join("{:>8.1f}".format(d) for d in DRAIN))
    cells = []
    for s in STORE:
        row = [s / d for d in DRAIN]
        cells.extend(row)
        print("   {:<5d}".format(s)
              + "".join("{:>8.1f}".format(h) for h in row))
    inside = [h for h in cells if LOW <= h <= HIGH]
    print("   cells inside the {:.0f} to {:.0f} hour window: {} of {}"
          .format(LOW, HIGH, len(inside), len(cells)))
    print("   share of the grid inside that window: {:.3f}"
          .format(len(inside) / len(cells)))
    print("   median of the grid: {:.1f} hours".format(quantile(cells, 0.5)))
    print("   full spread: {:.1f} to {:.1f} hours"
          .format(min(cells), max(cells)))
    print("   a 100 g store at 5.5 g/h: {:.1f} hours"
          .format(100 / 5.5))
#    time to empty the store, hours: store (g) by drain (g/h)
#    store     4.5     5.0     5.5     6.0     6.5
#    80       17.8    16.0    14.5    13.3    12.3
#    90       20.0    18.0    16.4    15.0    13.8
#    100      22.2    20.0    18.2    16.7    15.4
#    110      24.4    22.0    20.0    18.3    16.9
#    120      26.7    24.0    21.8    20.0    18.5
#    cells inside the 14 to 20 hour window: 16 of 25
#    share of the grid inside that window: 0.640
#    median of the grid: 18.2 hours
#    full spread: 12.3 to 26.7 hours
#    a 100 g store at 5.5 g/h: 18.2 hours

那张格子的二十五格里,有十六格落在十四到二十小时的窗口内,占 0.640。格子的中位数是 18.2 小时,整个跨度从最小储量配最快消耗的 12.3 小时,到最大储量配最慢消耗的 26.7 小时。一百克的储量按每小时 5.5 克抽取,18.2 小时见底。所以网上引用的那个窗口是一个合理区间的中段,而这个区间宽到接近两倍。任何把方案建立在精确小时数上的人,依赖的是一个在任何单个人身上都没有被知道到那个精度的量。

互相关掉对方的两个激素

第二个说法讲的是变化的形状,不是它的时刻。胰岛素下降让脂肪离开储存;释放出来的燃料,加上同时升高的胰高血糖素,反过来进一步压低胰岛素分泌。Unger 以现代形式陈述了这种互为倒数的安排 (Unger 1971)。这一类互相抑制给不出阈值。它给出一个有两个稳定设定的开关,而两个设定之间隔着一道壁垒,于是系统记得自己是从哪一侧来的。

把燃料信号写成胰岛素张力的递减函数,把胰岛素分泌写成葡萄糖的递增函数与燃料信号的递减函数,其中有一部分分泌是任何燃料信号都关不掉的。在稳态下,这一对收缩成一个关于胰岛素张力的单变量方程,它的根就是稳定设定:

x=a(E+1−E1+(B1+xN)N),a=Amaxgg+KGx = a\left(E + \frac{1-E}{1 + \left(\dfrac{B}{1+x^{N}}\right)^{N}}\right), \qquad a = \frac{A_{\max} g}{g + K_{G}}

推导

# Chapter 1. p02: A two-hormone switch with two stable settings. Write
# x for insulin tone and y for the fatty-acid and glucagon tone that
# rises when fat is mobilised. Insulin suppresses lipolysis, and the
# fuels lipolysis releases suppress insulin secretion, so the pair is a
# mutual-inhibition toggle: y = B / (1 + x**N) at a steady state, and
# x = a * (E + (1 - E) / (1 + y**N)), where E is the part of insulin
# secretion that no fuel signal can switch off and a = AMAX * g /
# (g + KG) is the drive from blood glucose g. The steady settings of x
# are the roots of f below. Three roots mean two stable settings with an
# unstable one between them, and the switch is then hysteretic: glucose
# must fall well below the level at which the fed state took hold.
N, B, E = 3.0, 4.0, 0.15
AMAX, KG = 8.0, 1.0
GRID, XMAX = 120001, 9.0

def drive(g):
    return AMAX * g / (g + KG)

def fuel(x):
    return B / (1.0 + x ** N)

def f(x, a):
    return a * (E + (1.0 - E) / (1.0 + fuel(x) ** N)) - x

def roots(a):
    out, px, pf = [], 0.0, f(0.0, a)
    for i in range(1, GRID):
        x = XMAX * i / (GRID - 1)
        fx = f(x, a)
        if (pf < 0.0) != (fx < 0.0):
            lo, hi = px, x
            for _ in range(60):
                mid = 0.5 * (lo + hi)
                if (f(lo, a) < 0.0) != (f(mid, a) < 0.0):
                    hi = mid
                else:
                    lo = mid
            out.append(0.5 * (lo + hi))
        px, pf = x, fx
    return out

def threshold(lo, hi):
    # lo and hi must have different numbers of steady settings
    base = len(roots(drive(lo))) == 3
    for _ in range(44):
        mid = 0.5 * (lo + hi)
        if (len(roots(drive(mid))) == 3) == base:
            lo = mid
        else:
            hi = mid
    return 0.5 * (lo + hi)

if __name__ == "__main__":
    print("   Hill exponent N = {:.1f}, unswitchable secretion E = {:.2f}"
          .format(N, E))
    print("   glucose g   steady settings of insulin tone x")
    for i in range(11):
        g = 0.20 + 0.10 * i
        r = roots(drive(g))
        print("   {:<12.2f}".format(g)
              + "  ".join("{:.3f}".format(v) for v in r))
    g_on = threshold(0.20, 0.80)
    g_off = threshold(0.80, 1.60)
    print("   fed state first becomes possible at g = {:.4f}".format(g_on))
    print("   fasted state is lost above g = {:.4f}".format(g_off))
    print("   width of the hysteresis in g: {:.4f}".format(g_off - g_on))
    print("   ratio of the two glucose thresholds: {:.2f}"
          .format(g_off / g_on))
    r = roots(drive(0.80))
    print("   at g = 0.80 the stable insulin tones are {:.3f} and {:.3f}"
          .format(r[0], r[2]))
    print("   fatty-acid tone there: {:.3f} fasted, {:.3f} fed"
          .format(fuel(r[0]), fuel(r[2])))
    print("   ratio of the two fatty-acid tones: {:.1f}"
          .format(fuel(r[0]) / fuel(r[2])))
#    Hill exponent N = 3.0, unswitchable secretion E = 0.15
#    glucose g   steady settings of insulin tone x
#    0.20        0.218
#    0.30        0.303
#    0.40        0.378  1.548  2.214
#    0.50        0.445  1.387  2.647
#    0.60        0.506  1.293  2.992
#    0.70        0.564  1.221  3.290
#    0.80        0.621  1.159  3.554
#    0.90        0.678  1.099  3.788
#    1.00        0.742  1.036  3.999
#    1.10        0.840  0.940  4.190
#    1.20        4.363
#    fed state first becomes possible at g = 0.3554
#    fasted state is lost above g = 1.1142
#    width of the hysteresis in g: 0.7589
#    ratio of the two glucose thresholds: 3.14
#    at g = 0.80 the stable insulin tones are 0.621 and 3.554
#    fatty-acid tone there: 3.229 fasted, 0.087 fed
#    ratio of the two fatty-acid tones: 37.0

希尔指数取 3.0、关不掉的那一份取 0.15 时,进食态最早在葡萄糖驱动 0.3554 处成为可能,而空腹态一直存活到 1.1142。两个门槛相差 3.14 倍,驱动上的回差宽度是 0.7589。在重叠区内,驱动为 0.80 时,两个稳定的胰岛素张力是 0.621 与 3.554,与之相伴的脂肪酸张力是 3.229 对 0.087,比值 37.0。换句话说,同一个葡萄糖水平既与一具正在用力动员脂肪的身体相容,也与一具完全没有在动员脂肪的身体相容。究竟是哪一个,取决于近期的进食史,而不取决于当下这一次测量。

胰岛素张力的稳定设定对葡萄糖驱动,以及两个设定共存的那段重叠区(p02)。

假说 · 尚未验证

作者的诠释(尚未验证): 因为这个安排带回差,临床上有用的量不是断食保持了多少小时,而是复食之前葡萄糖低谷的深度,因为只有低于下门槛的低谷才把系统推上空腹分支,此后它就停在那里,哪怕当下的驱动本来不足以把它送上去。什么会证明它错: 在断食总小时数相匹配的人群里,如果复食后测到的动员能由总断食小时数解释而不能由葡萄糖的最低值解释,或者两者预测得一样好,那么回差这一读法就没有添加任何东西。

小结

十四到二十小时这个窗口,是一个在合理参数下从 12.3 到 26.7 小时的区间的中段,而那张格子只有 0.640 落在窗口内。窗口末端的那个变化不是阈值,而是一个有两个稳定设定的开关,两个门槛在葡萄糖驱动上相差 3.14 倍,所以同一个葡萄糖读数与两种相当不同的脂肪动员速率都相容。第二章问的是,开关已经动了之后,脂肪细胞内部必须发生什么。

参考文献

Cahill, George F. 2006. “Fuel Metabolism in Starvation.” Annual Review of Nutrition 26: 1 to 22.
Rothman, Douglas L., Inger Magnusson, Laurie D. Katz, Robert G. Shulman, and Gerald I. Shulman. 1991. “Quantitation of Hepatic Glycogenolysis and Gluconeogenesis in Fasting Humans with 13C NMR.” Science 254 (5031): 573 to 576.
Unger, Roger H. 1971. “Glucagon Physiology and Pathophysiology.” The New England Journal of Medicine 285 (8): 443 to 449.

Fan, Y. (2026). Insulin-Glucagon Bistability, Visceral Lipolysis Kinetics and Linear-Programming Diet Optimization: Hepatic Glycogen Depletion Times, Triglyceride-Glucose Indices and Micronutrient-Constrained Menus [胰岛素胰高血糖素双稳态、内脏脂解动力学与线性规划配餐] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090248 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).