Part II: Systems Neuroscience | Chapter 5

Sleep & Falling-Asleep Mechanisms

Sleep-onset neurobiology — how the brain switches from wake to sleep, the chemistry of drowsiness, EEG stages, and the two-process model of sleep regulation.

Overview

Falling asleep is not a passive fading of consciousness — it is an actively orchestrated switch in the central nervous system. A small cluster of GABAergic neurons in the ventrolateral preoptic nucleus(VLPO) silences the ascending arousal system, while adenosine accumulated across the waking day provides the homeostatic pressure and the suprachiasmatic-nucleus circadian clock modulates the timing. Within minutes, EEG synchronises, thalamocortical oscillations shift from gamma into alpha then sleep spindles, heart rate drops, and muscle tone collapses.

This chapter covers the two-process model of sleep regulation (Borbély 1982), the flip-flop wake-sleep switch (Saper 2005), adenosine accumulation, the chemistry of melatonin and histamine, EEG stage architecture, and the neural substrates of NREM and REM sleep.

1. The Two-Process Model of Sleep Regulation

Borbély’s (1982) two-process model posits that the timing and duration of sleep arise from the interaction of two independent regulators: a homeostatic process (S) that accumulates monotonically during wake and dissipates during sleep, and a circadian process (C) driven by the suprachiasmatic nucleus (SCN), providing a daily oscillating threshold. Sleep onset occurs when S rises to meet the circadian upper threshold; sleep offset occurs when S decays to the lower threshold.

Derivation 1: Homeostatic Process Dynamics

During wake, sleep pressure builds toward an upper asymptote S_u, following an exponential-saturation law:

\[ S(t) = S_u - (S_u - S(t_0))\,e^{-(t-t_0)/\tau_w} \]

During sleep, S dissipates toward lower asymptote S_l on a faster timescale:

\[ S(t) = S_l + (S(t_0) - S_l)\,e^{-(t-t_0)/\tau_s} \]

Human fits: τ_w ≈ 18 h,τ_s ≈ 4.2 h. The circadian threshold pair is modeled as \( C_{u,l}(t) = m_{u,l} + a\sin(2\pi(t-\phi)/24) \) with phase φ ≈ 18 h (near dusk).

Sleep pressure = adenosine: S has a molecular identity — extracellular adenosine concentration in the basal forebrain increases across the waking day by roughly 40% (Porkka-Heiskanen 1997). Caffeine blocks A₁/A_2A adenosine receptors, masking the accumulated pressure; once caffeine clears, the underlying S is unchanged, producing the “rebound” after an all-nighter.

Python

script.py56 lines

import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt

# Borbely 1982 two-process model of sleep regulation
# S = homeostatic sleep pressure; C = circadian process

hours = np.linspace(0, 72, 2000)            # 3 days
# Homeostatic Process S (exponential saturation/decay)
S_u, S_l = 1.0, 0.17                         # upper/lower asymptotes
tau_wake, tau_sleep = 18.0, 4.2              # time constants (hours)

S = np.zeros_like(hours)
state = 'wake'                               # 'wake' or 'sleep'
S[0] = 0.3
for i in range(1, len(hours)):
    dt = hours[i] - hours[i-1]
    if state == 'wake':
        S[i] = S_u - (S_u - S[i-1]) * np.exp(-dt/tau_wake)
    else:
        S[i] = S_l + (S[i-1] - S_l) * np.exp(-dt/tau_sleep)
    # Circadian threshold crossings
    C_upper = 0.95 + 0.07*np.sin(2*np.pi*(hours[i]-18)/24)
    C_lower = 0.17 + 0.07*np.sin(2*np.pi*(hours[i]-18)/24)
    if state == 'wake' and S[i] >= C_upper:
        state = 'sleep'
    elif state == 'sleep' and S[i] <= C_lower:
        state = 'wake'

# Circadian envelopes
C_up = 0.95 + 0.07*np.sin(2*np.pi*(hours-18)/24)
C_lo = 0.17 + 0.07*np.sin(2*np.pi*(hours-18)/24)

fig, ax = plt.subplots(figsize=(12, 6), facecolor='#0a0a1a')
ax.set_facecolor('#111827')
ax.tick_params(colors='#cbd5e1')
for s in ax.spines.values(): s.set_color('#334155')

ax.plot(hours, S, color='#a78bfa', lw=2.2, label='Homeostatic pressure S(t)')
ax.plot(hours, C_up, color='#fbbf24', lw=1.5, linestyle='--', label='Upper circadian threshold')
ax.plot(hours, C_lo, color='#34d399', lw=1.5, linestyle='--', label='Lower circadian threshold')
ax.fill_between(hours, C_lo, C_up, color='#64748b', alpha=0.1)
ax.set_xlabel('Time (hours)', color='#94a3b8')
ax.set_ylabel('Sleep drive (a.u.)', color='#94a3b8')
ax.set_title('Borbely Two-Process Model of Sleep Regulation', color='#f5d0fe', fontweight='bold')
ax.legend(facecolor='#1e293b', edgecolor='#334155', labelcolor='#cbd5e1')
ax.grid(True, color='#334155', alpha=0.35)

# Shaded sleep episodes
plt.tight_layout()
plt.savefig('output.png', dpi=120, bbox_inches='tight', facecolor='#0a0a1a')
print(f'Peak S (night onset): {S.max():.3f}')
print(f'Trough S (after sleep): {S.min():.3f}')
print(f'Mean S across 72 h: {S.mean():.3f}')
print(f'Energy integral over 72 h: {np.trapezoid(S, hours):.2f} a.u.-hours')

Click Run to execute the Python code

Code will be executed with Python 3 on the server

2. The Flip-Flop Wake-Sleep Switch

Saper et al. (2005) proposed that the transition from wake to sleep is implemented by mutual inhibition between two populations: the VLPO (ventrolateral preoptic nucleus, GABAergic + galaninergic, sleep-promoting) and the ARAS (ascending reticular activating system — cholinergic LDT/PPT, noradrenergic LC, serotonergic DRN, dopaminergic VTA, histaminergic TMN, wake-promoting). Like an electronic flip-flop, the bistable switch eliminates intermediate states: you are either clearly awake or clearly asleep, with rapid transitions between them.

Derivation 2: Bistability from Mutual Inhibition

Let V = VLPO firing rate, A = ARAS firing rate. Each population has a saturating response to total input minus the other’s inhibitory drive:

\[ \tau_V \frac{dV}{dt} = -V + \tanh\!\bigl(I_{sleep} - k_{VA}\,A\bigr) \]

\[ \tau_A \frac{dA}{dt} = -A + \tanh\!\bigl(I_{wake} - k_{AV}\,V\bigr) \]

With k_VA, k_AV sufficiently large (typically > 1), the system has two stable fixed points — (V high, A low) = sleep and (V low, A high) = wake — separated by an unstable saddle. Smooth drift of I_sleep (rising with adenosine) andI_wake (falling with fatigue) eventually destabilises the wake fixed point, producing a fast transition.

Orexin/hypocretin from lateral hypothalamus stabilises the wake state against premature transitions;narcolepsy with cataplexy is caused by autoimmune destruction of orexin neurons, making the flip-flop hypersensitive (Lin 1999 dog narcolepsy hypocretin-receptor-2 mutation; Peyron 2000 human orexin loss).

Python

script.py74 lines

import numpy as np
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
from scipy.integrate import odeint

# Flip-flop wake-sleep switch (Saper 2005)
# VLPO (ventrolateral preoptic nucleus) firing = sleep-promoting
# ARAS (ascending reticular activating system) firing = wake-promoting
# Mutual inhibition produces bistability

def flipflop(state, t, I_wake, I_sleep):
    V, A = state                              # VLPO, ARAS firing rates
    kva, kav = 1.5, 1.5                       # mutual inhibition gains
    tauV, tauA = 2.0, 1.0                     # time constants (min)
    dV = (-V + np.tanh(I_sleep - kva*A)) / tauV
    dA = (-A + np.tanh(I_wake - kav*V)) / tauA
    return [dV, dA]

# Simulate a wake → sleep transition triggered by rising sleep pressure
t = np.linspace(0, 90, 3000)
# Homeostatic sleep pressure ramp (wake drive decreasing)
I_wake  = 1.2 - 0.012 * t                     # wake drive drops with fatigue
I_sleep = 0.5 + 0.016 * t                     # VLPO input rises with adenosine

sol = odeint(flipflop, [0.05, 0.95], t, args=(I_wake[0], I_sleep[0]))
# Re-integrate with time-varying inputs
V = np.zeros_like(t); A = np.zeros_like(t)
V[0], A[0] = 0.05, 0.95
dt = t[1] - t[0]
for i in range(1, len(t)):
    kva, kav = 1.5, 1.5
    tauV, tauA = 2.0, 1.0
    V[i] = V[i-1] + dt * (-V[i-1] + np.tanh(I_sleep[i-1] - kva*A[i-1])) / tauV
    A[i] = A[i-1] + dt * (-A[i-1] + np.tanh(I_wake[i-1] - kav*V[i-1])) / tauA

# Detect switch point
switch_idx = np.argmax(V > 0.5)
t_switch = t[switch_idx]

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(13, 5), facecolor='#0a0a1a')
for ax in (ax1, ax2):
    ax.set_facecolor('#111827'); ax.tick_params(colors='#cbd5e1')
    for s in ax.spines.values(): s.set_color('#334155')

ax1.plot(t, V, color='#a78bfa', lw=2.2, label='VLPO (sleep)')
ax1.plot(t, A, color='#fbbf24', lw=2.2, label='ARAS (wake)')
ax1.axvline(t_switch, color='#ef4444', linestyle='--', label=f'Transition @ t={t_switch:.1f} min')
ax1.set_xlabel('Time (min)', color='#94a3b8')
ax1.set_ylabel('Firing rate (normalised)', color='#94a3b8')
ax1.set_title('Flip-Flop Wake-Sleep Switch', color='#f5d0fe', fontweight='bold')
ax1.legend(facecolor='#1e293b', edgecolor='#334155', labelcolor='#cbd5e1')
ax1.grid(True, color='#334155', alpha=0.35)

# Phase portrait
V_grid, A_grid = np.meshgrid(np.linspace(-0.2, 1.2, 25), np.linspace(-0.2, 1.2, 25))
dV = (-V_grid + np.tanh(0.5 - 1.5*A_grid)) / 2.0
dA = (-A_grid + np.tanh(1.2 - 1.5*V_grid)) / 1.0
ax2.streamplot(V_grid, A_grid, dV, dA, color='#64748b', density=1.2, linewidth=0.8)
ax2.plot(V, A, color='#ef4444', lw=1.5, alpha=0.8)
ax2.scatter([V[0]], [A[0]], color='#34d399', s=80, zorder=5, label='Start (wake)')
ax2.scatter([V[-1]], [A[-1]], color='#a78bfa', s=80, zorder=5, label='End (sleep)')
ax2.set_xlabel('VLPO activity', color='#94a3b8')
ax2.set_ylabel('ARAS activity', color='#94a3b8')
ax2.set_title('Phase portrait', color='#f5d0fe', fontweight='bold')
ax2.legend(facecolor='#1e293b', edgecolor='#334155', labelcolor='#cbd5e1')
ax2.grid(True, color='#334155', alpha=0.35)

plt.tight_layout()
plt.savefig('output.png', dpi=120, bbox_inches='tight', facecolor='#0a0a1a')
print(f'Transition at t = {t_switch:.1f} min')
print(f'ARAS drop ΔA = {A[0] - A[-1]:.3f}')
print(f'VLPO rise ΔV = {V[-1] - V[0]:.3f}')
print(f'Integrated wake drive: {np.trapezoid(A, t):.1f} a.u.-min')

Click Run to execute the Python code

Code will be executed with Python 3 on the server

3. The Chemistry of Falling Asleep

A small vocabulary of neuromodulators carries the sleep-wake signal:

Transmitter	Source nucleus	Role	Pharmacology
Adenosine	Basal forebrain (metabolic)	Homeostatic pressure	A1/A2A antagonist = caffeine
GABA / galanin	VLPO	Sleep promotion	GABA-A PAM = benzodiazepines
Histamine	Tuberomammillary nucleus (TMN)	Wake promotion	H1 blockade = doxylamine drowsiness
Acetylcholine	LDT / PPT	Wake + REM	Muscarinic
Norepinephrine	Locus coeruleus	Wake (mostly)	NE-reuptake in modafinil
Serotonin	Dorsal raphe	Wake + SWS suppression	5-HT2A ligands alter architecture
Orexin/hypocretin	Lateral hypothalamus	Wake stabiliser	OX2R-A = suvorexant sleep aid
Melatonin	Pineal gland	Circadian phase cue	Ramelteon (MT1/MT2 agonist)
Prostaglandin D2	Leptomeninges	Sleep promotion via VLPO	DP1 agonism induces sleep

Melatonin is synthesised from serotonin via N-acetyltransferase and hydroxyindole-O-methyltransferase in the pineal gland, peaking at ~03:00 with intensity inversely coupled to retinal light exposure via the retinohypothalamic tract → SCN → pineal. Exogenous 0.3 mg timed ~6 h before sleep phase-advances the circadian clock; higher doses are hypnotic but over-ride endogenous rhythm (Lewy 1998, Zhdanova 2001).

4. EEG Stages of Sleep Onset

From eyes-closed wake to established N1 lasts ~5–15 minutes in young adults. The EEG passes through:

Wake (eyes open): desynchronised gamma/beta (15–40 Hz), low amplitude (10–30 µV).
Wake (eyes closed): alpha rhythm dominant (8–12 Hz, 30–50 µV) over occipital cortex.
N1 (stage 1): alpha drops out, replaced by low-amplitude mixed frequency (4–7 Hz theta). Hypnagogic hallucinations and myoclonic jerks possible here.
N2: sleep spindles (11–15 Hz, 0.5–3 s bursts, thalamic reticular origin) + K-complexes (sharp sentinel waves).
N3 (slow-wave sleep): delta activity < 2 Hz dominating > 20% of epoch, up/down states, glymphatic clearance active.
REM: returns to wake-like EEG but with atonia, rapid eye movements, dreaming.

Sleep spindles arise from thalamic reticular nucleus GABAergic oscillators in reciprocal connection with thalamocortical relay cells (Steriade 1993). Their frequency (~13 Hz) correlates with memory consolidation during subsequent N3 (Fogel 2011).

5. Slow Oscillations & Glymphatic Clearance

During N3, cortical neurons alternate between depolarised “up states” (active) and hyperpolarised “down states” (silent), producing the < 1 Hz slow oscillation (Steriade 1993). These travel as waves across cortex. Xie et al. 2013 showed that the glymphatic system (astrocyte AQP4-mediated ISF exchange) increases flow ~60% during N3, driven by expanded interstitial space (~60% → 24% cerebral blood volume fraction). Amyloid-β clearance is doubled during sleep; chronic sleep disruption impairs Aβ clearance, implicated in Alzheimer’s risk (Ju 2017, Shokri-Kojori 2018).

6. REM Sleep Mechanisms

REM is generated by a second flip-flop between REM-off neurons (ventrolateral periaqueductal grey + lateral pontine tegmentum) and REM-on neurons (sublaterodorsal nucleus + PPT/LDT cholinergic). REM-on activity produces:

Cortical activation indistinguishable from wake on EEG
Rapid eye movements via oculomotor nuclei
Muscle atonia via glycinergic inhibition at motor neurons (REM behaviour disorder = loss of this atonia, α-synucleinopathy early sign)
Pontine waves (PGO) triggering dream imagery

7. Why You Can’t Fall Asleep

Common disruptions to the sleep-onset machinery:

Caffeine: A1/A2A antagonism; half-life 5–6 h. Coffee at 15:00 still 25% active at 02:00.
Blue-spectrum light: 460–480 nm stimulates ipRGCs → SCN → suppresses melatonin 50% at 100 lux (Lockley 2003).
Cortisol elevation: stress-driven HPA activation maintains ARAS drive; CBT-I is first-line chronic insomnia therapy.
Orexin over-activity: DORAs (dual orexin receptor antagonists — suvorexant, lemborexant, daridorexant) dampen wake signal without GABAergic sedation.
Circadian misalignment: shift work and jet lag decouple C from the environmental light cycle.

References

Borbély, A. A. (1982). A two process model of sleep regulation. Human Neurobiology, 1, 195–204.
Saper, C. B., Scammell, T. E. & Lu, J. (2005). Hypothalamic regulation of sleep and circadian rhythms. Nature, 437, 1257–1263.
Porkka-Heiskanen, T. et al. (1997). Adenosine: a mediator of the sleep-inducing effects of prolonged wakefulness. Science, 276, 1265–1268.
Peyron, C. et al. (2000). A mutation in a case of early-onset narcolepsy and a generalized absence of hypocretin peptides. Nature Medicine, 6, 991–997.
Steriade, M., McCormick, D. A. & Sejnowski, T. J. (1993). Thalamocortical oscillations in the sleeping and aroused brain. Science, 262, 679–685.
Xie, L. et al. (2013). Sleep drives metabolite clearance from the adult brain. Science, 342, 373–377.
Lewy, A. J. et al. (1998). The human phase response curve to melatonin is about 12 hours out of phase with the phase response curve to light. Chronobiology International, 15, 71–83.
Scammell, T. E., Arrigoni, E. & Lipton, J. O. (2017). Neural circuitry of wakefulness and sleep. Neuron, 93, 747–765.

← Attention & Consciousness Part III: Computational Neuroscience →