Neuroscience · Hypothesis
The shape of need
In addiction the difference that matters may not be how strongly the system responds, but how slowly it returns to baseline — and how much of it never returns at all.
The question
In people with addiction, is the dynamics of need and its recovery measurable in a repeatable way, and does it distinguish those who relapse from those who hold?
This is not the question of who becomes addicted: answering that requires observing people before they lose control. This first phase does not, and says so.
Preliminary content of the proposal history
What is known and what is not
Level is not mechanism. CRF1 receptor antagonists were tested in humans and failed instructively: pexacerfont left craving untouched despite reaching 90% receptor occupancy, and verucerfont fully blocked the stress axis on the dexamethasone-CRF test while need stayed unchanged. The trial was stopped for futility. The stress system can be switched off, measured as hormone level, and the need remains.
Astrocytes move. During withdrawal astrocytic processes physically retract from synapses and re-approach on a timeline that tracks drug seeking. Not a level: a dynamic, with a time constant.
Alcohol touches water transport. Chronic medium-dose intake raises AQP4 expression while reducing its polarisation at the endfeet; low doses improve glymphatic function, high doses suppress it.
Sleep predicts relapse, but treating it does not prevent relapse. Sixty per cent of patients with pre-treatment insomnia relapsed within five months against thirty per cent without; some sleep abnormalities persist after two years of abstinence. Yet insomnia interventions improve sleep without improving abstinence rates. This finding stays in view, not hidden.
The glymphatic bridge is not ours. Human studies already link glymphatic function and addiction using the DTI-ALPS index, in heroin patients on substitution treatment, in alcohol use disorder and in methamphetamine. They are cross-sectional and correlational, and their own authors write that the link between index and glymphatic system needs strengthening.
What does not appear to have been done: linking glymphatic function to recovery dynamics after a perturbation, manipulating sleep to see whether the residue changes, and building the mouse-to-human bridge with an AQP4 model in hand.
- association Wang L, Qin Y, Li X et al. (2023). Glymphatic-system function is associated with addiction and relapse in heroin dependents undergoing methadone maintenance treatment Brain Sciences Il ponte glinfatico-dipendenza esiste già in letteratura umana. Trasversale.
- causal evidence Lundgaard I, Wang W, Eberhardt A et al. (2018). Beneficial effects of low alcohol exposure, but adverse effects of high alcohol intake on glymphatic function Scientific Reports Dose bassa migliora, dose alta sopprime. AQP4 depolarizzata con consumo cronico.
- causal evidence Schwandt ML, Cortes CR, Kwako LE et al. (2016). The CRF1 antagonist verucerfont in anxious alcohol-dependent women: translation of neuroendocrine, but not of anti-craving effects Neuropsychopharmacology Asse HPA bloccato, craving invariato. Studio interrotto per futilità.
- causal evidence Kwako LE, Schwandt ML, Ramchandani VA et al. (2015). The corticotropin releasing hormone-1 receptor antagonist pexacerfont in alcohol dependence: a randomized controlled experimental medicine study Neuropsychopharmacology Nessun effetto sul craving nonostante occupazione recettoriale del 90%.
- association Brower KJ, Aldrich MS, Robinson EA et al. (2001). Insomnia, self-medication, and relapse to alcoholism American Journal of Psychiatry Ricaduta 60% con insonnia contro 30% senza.
Preliminary content of the proposal history
The hypothesis
The need that follows last use differs not only in intensity but in shape over time: speed of rise, time to peak, and above all the share that never returns to baseline.
Those who relapse do not necessarily react more strongly: they return more slowly, or not at all. The difference that matters is in the tail, not the peak.
The bridging hypothesis, stated as hypothesis and not as premise: that incomplete recovery depends in part on a nightly perivascular clearance that does not complete, a mechanism resting on AQP4 polarisation at astrocytic endfeet.
Preliminary content of the proposal history
How it can be disproved
The project stops, and says so, if any of the following happens.
The parameters are not stable. Measured twice weeks apart, the same person yields different curves. Below the pre-specified preliminary threshold of 0.70 intraclass correlation — ICC(2,1), two-way random effects, single measure — the parameter is not sufficiently stable for the intended use.
They add nothing. The parameters do not improve outcome prediction beyond what is already in the clinical record.
They measure the substance, not the person. Differences are fully explained by current medication, days since admission, or comorbidity.
The biological bridge fails. In the mouse: fragment sleep and measure whether the residue changes, comparing animals with normal AQP4 against knockouts. If the residue is identical, the hypothesis falls — in an experiment a glial neuroscience laboratory knows how to run.
Each of these outcomes is publishable. None is a failure of the method: they are answers.
- contested Miller MB, Donahue ML, Carey KB, Scott-Sheldon LAJ (2017). Insomnia treatment in the context of alcohol use disorder: a systematic review and meta-analysis Drug and Alcohol Dependence Il sonno migliora, i tassi di astinenza no. Anello debole della catena causale.
Preliminary content of the proposal history
Proposed mechanism
Two glial axes, both dynamic. No study appears to have measured them together in the same animal, but this is not a systematic search: it must be verified with a proper review before being presented as novel.
The first is the geometry of perisynaptic processes: they retract during withdrawal and re-approach on a timeline that tracks behaviour.
The second is perivascular water transport, where AQP4 polarisation at the endfeet matters more than total amount.
The question that follows: do the two axes move together? If so, they are two faces of one remodelling of the glial membrane, and incomplete recovery has a single mechanism. If their timings differ, they are distinct processes and the hypothesis must be rewritten.
- causal evidence Mestre H, Hablitz LM, Xavier AL et al. (2018). Aquaporin-4-dependent glymphatic solute transport in the rodent brain eLife Replica multi-laboratorio su quattro linee knockout indipendenti.
- causal evidence Scofield MD, Kalivas PW et al. (2016). Cocaine self-administration and extinction leads to reduced glial glutamate transporter expression and morphological changes within the nucleus accumbens Biological Psychiatry Retrazione dei prolungamenti astrocitari dalle sinapsi durante l'astinenza.
- causal evidence Xie L, Kang H, Xu Q et al. (2013). Sleep drives metabolite clearance from the adult brain Science La clearance aumenta durante il sonno.
- association Kalivas PW (2009). The glutamate homeostasis hypothesis of addiction Nature Reviews Neuroscience Riduzione di GLT-1 e xCT sugli astrociti.
Preliminary content of the proposal history
How it is measured
Short, repeatable tasks, not a reconstruction of personal history. The person performs tasks; instruments record.
Need is not measured by asking how bad it feels, but as the price a person will pay to make the discomfort stop: how much work, how much waiting, how much forgoing. It is a behaviour, it has a unit, it repeats days apart.
In structured treatment the time since last use is known and verifiable: there is at last a certain time zero, and life has already supplied the perturbation. Each person becomes their own control, and for the first gate — within-person repeatability — that is enough. It stops being enough the moment one wants to say what is abnormal: defining a reference range requires comparison groups, exposed people who have not lost control and low-exposure controls.
Two gates, in order. First within-person repeatability; only then predictive value. No hypothesis about mechanism is tested before the first gate.
Preliminary content of the proposal history
Instruments and data
This section sets out what is acquired, with which instrument, at what rate and with what error. It is what makes the rest verifiable: without it, the response curve is a drawing.
The observed state
For each person i and each instant t a vector of measurements is recorded:
x(i, t) = [ HRV, skin conductance, EEG, pupil, sleep, cortisol, activity, … ]
It is not a score and must never be collapsed into one. Each component carries its own unit, the instrument that produced it, the sampling rate and a declared uncertainty. Two components sharing a unit but coming from different instruments stay two distinct components.
| Quantity | Instrument | Rate | Error to declare |
|---|---|---|---|
| Inter-beat intervals | contact ECG, not wrist photoplethysmography | continuous | R-peak detection, lost samples |
| Skin conductance | palmar or plantar electrodes | 4–32 Hz | drift, motion artefacts |
| Electroencephalogram | frontal headband or cap | 128–256 Hz | per-electrode impedance, staging against polysomnography |
| Pupil diameter | pupillometry | 60–250 Hz | ambient luminance, interpolated blinks |
| Sleep | wearable EEG plus actigraphy | 30 s epochs | per-stage agreement, fragmentation |
| Cortisol | saliva at fixed times | discrete points | verified waking adherence, within-day variability |
| Substance exposure | blood marker or transdermal sensor | weekly or continuous | time window, sensor latency |
Two rules hold for all of them. The latency of each channel must be measured, not assumed: a signal arriving four hundred undeclared milliseconds late produces dynamics that do not exist. And raw data are never overwritten: reconstructions live in derived layers, marked as such, with the version of the algorithm that produced them.
The perturbation
u(t) denotes the controlled stimulus. Only repeatable and describable stimuli are admitted:
- exposure to a substance-associated cue, against neutral cues;
- a reward task, and omission of an expected reward;
- an inhibitory control task;
- a measured night of sleep, possibly fragmented under control;
- a defined treatment, with known dose and timing.
Time zero must be verifiable. In structured treatment, time since last use is; in the rest of life, almost never.
The response
R(i, t | u) = x(i, t) − x(i, 0)
How far the measurements move from their own starting value after the perturbation. From this curve, and not from absolute values, the quantities of interest are extracted:
- peak amplitude;
- time to peak;
- rise rate;
- half-time of recovery;
- area under the curve;
- residue after a pre-set time;
- oscillation;
- within-person repeatability.
The last is not a property of the person but of the measurement, and it comes before all the others.
Only at the end, Z
Once real data exist and the parameters have passed the stability gate, a composite index can be built:
Z(i) = β₀ + β₁·A(i) + β₂·T(i) + β₃·E(i) + …
where A is amplitude, T recovery time, E residue. The β coefficients are estimated from the data, not chosen.
Three things must be said now, before Z exists. It is not a biological substance: it is a combination of measurements. It is not a diagnosis, and it does not describe who you are. And it is discarded if it does not improve prediction in people never used to build it — validation is on unseen data, or Z measures itself.
The first job
It is not to show that the equation works. It is to establish which measurements are stable and repeatable enough to deserve a place in it.
Concretely: for each component of the vector, measure the same person twice weeks apart and compute the intraclass correlation. Anything below the preliminary threshold of 0.70 does not enter, however promising it looks.
Preliminary content of the proposal history
Skills and feasibility
The project has two arms with two different leads, and the link between them is a hypothesis to be demonstrated, not a given.
The human arm needs a clinical principal investigator with access to people in treatment, a psychophysiologist for task design, and a time-series statistician. Without the first there is neither ethics approval nor recruitment.
The biological arm needs a glial neuroscience laboratory with in vivo two-photon imaging and AQP4 models. This is where no other instrument substitutes: not MRI, not blood.
Infrastructure for organising the research and storing its data is available. The analysis engine exists as a prototype on synthetic data: it has never been validated on real measurements, and designing it is part of the work to be done.
- causal evidence Bordoni L, Thorén AE, Gutiérrez-Jiménez E et al. (2023). Deletion of aquaporin-4 improves capillary blood flow distribution in brain edema Glia Metodo e modello del braccio biologico proposto.
Preliminary content of the proposal history
Open questions
The glymphatic model itself is contested: work exists disputing the extent and nature of that flow, and it must be cited.
The DTI-ALPS index is a weak surrogate, as its own users say.
The step from mouse to human, in this scheme, is the hypothesis — not an already welded link.
Still to decide: which substances for the second and third tracks, which functional form for the curve, how to operationalise the price of relief tolerably for someone in early withdrawal.
- association Dai X, Gao L, Zhang J et al. (2024). Investigating DTI-ALPS index and its association with cognitive impairments in patients with alcohol use disorder Journal of Psychiatric Research Gli autori stessi chiedono un legame più solido fra indice e sistema glinfatico.
Preliminary content of the proposal history