The reward system runs hotter — and only for food#
After a night without sleep, a tired brain does not simply want more of everything. It wants food more, specifically, and the cleanest demonstration of that specificity comes from a study that gave people something else to want as a control.
Thirty-two lean, healthy men completed a repeated-measures crossover — one night of habitual sleep, one night of total sleep deprivation — and performed a value-based decision task inside an fMRI scanner, bidding real money on snack foods and on trinkets. The sleep state × reward category interaction was significant (F(1,31) = 5.48, P = 0.03): willingness to pay rose after deprivation for food, and not for the trinkets1. The imaging matched the bidding. Valuation signals in the hypothalamus increased food-specifically (right hypothalamus z = 3.27, P = 0.018; left z = 2.92, P = 0.048, both small-volume corrected), and the right amygdala showed increased coupling with the left hypothalamus (z = 3.50, P = 0.011).
That is the answer in one experiment: sleep loss is not a general appetite for reward. It is a targeted upgrade to how much food is worth. This article is the imaging layer under why you crave junk food when you're tired, which establishes the behaviour and the endocannabinoid mechanism; here we ask which circuits, which foods, and how reliably any of it replicates.
The brain valued food more while hunger stayed flat#
The most striking thing in Rihm's data is a null result sitting next to the positive ones.
Subjective hunger ratings did not differ significantly between the sleep-deprived and rested sessions. Des-acyl ghrelin did rise after deprivation (P = 0.04) — but changes in ghrelin were not correlated with changes in fMRI activity, and neither the behavioural nor the imaging results correlated with acyl, des-acyl or total ghrelin.
The same men, on the same food, bid more money after a sleepless night — while reporting no more hunger, and with the hunger hormone explaining none of it.
So the valuation moved without the felt state moving with it. That is a genuine dissociation, and it reframes what the experience of tired eating actually is: not a stomach signal you are failing to resist, but a pricing error you cannot feel. You are not hungrier. The food is worth more.
The caveats are real. Thirty-two men, all lean, all young, after total sleep deprivation — a harsher exposure than a short night, and a sample that cannot speak for women or for people with obesity. And the ghrelin non-correlation should be read as this study failing to find a link rather than as proof there isn't one; the broader audit of how badly the hormone story carries the behaviour is in the cluster pillar.
Which circuits, exactly#
The wider imaging literature maps the regions, using a different and more common design: show pictures of food, measure the response.
Twenty-six normal-weight adults completed a crossover of six nights at 4 hours in bed against six at 9 hours. Neuronal activity in response to food stimuli was greater after restricted sleep across a broad reward and salience network — putamen and nucleus accumbens, thalamus, insula, orbitofrontal cortex, precuneus and cingulate gyrus, with peak z-scores from 2.57 to 3.79 at a cluster-level threshold of P < 0.012.
| Study | Design | n | What moved |
|---|---|---|---|
| St-Onge, 2012 | 4 h vs 9 h × 6 nights, food images | 26 | Putamen/NAcc, thalamus, insula, OFC, precuneus, cingulate |
| St-Onge, 2014 | 4 h vs 9 h × 5 nights, healthy vs unhealthy food | 25 | Unhealthy-food-specific activation incl. OFC, right insula |
| Rihm, 2019 | Total deprivation, bidding on food vs trinkets | 32 | Hypothalamic valuation, amygdala–hypothalamic coupling |
| DiFrancesco, 2024 | ~6.5 h vs ~9 h × 3 weeks, adolescents | 39 | One cluster: VTA/substantia nigra |
Read down the right-hand column and something becomes obvious that individual papers obscure: these studies do not agree on a circuit. They agree on a direction. Different tasks, different exposures and different age groups light up substantially different maps, which is the normal condition of a small-sample imaging literature and not a scandal — but it does mean "the brain's reward center" is doing a lot of rhetorical work that the data does not support. There is no single region here that every study finds.
"Junk food specifically" was a claim before it was a finding#
Here is a detail worth knowing about the most-cited study in this area, because it is routinely used for a conclusion it cannot support.
St-Onge's 2012 paper is the standard citation for tired brains preferring calorie-dense food. Its authors were explicit that it could not show this: the study investigated neuronal activity in response to visual food stimuli but did not distinguish between high- and low-calorie foods. Food versus non-food was the contrast. The selectivity everyone quotes it for was not tested in it.
What makes this a good story rather than a gotcha is that the same group went and tested it. Twenty-five normal-weight habitual 7-to-9-hour sleepers were scanned in the fasted state after five nights at 4 hours or 9 hours in bed, viewing healthy foods, unhealthy foods and objects in a block design. After restricted sleep, unhealthy foods produced greater activation than healthy foods in the superior and middle temporal gyri, middle and superior frontal gyri, left inferior parietal lobule, orbitofrontal cortex and right insula. After habitual sleep, the same images produced no marked unhealthy-food-specific pattern3.
That is the finding the 2012 paper is usually credited with, arriving two years later in a study designed to produce it. The claim is now supported — and it is supported by 25 people in one laboratory, which is worth holding in mind before treating it as bedrock.
How reliably does it replicate?#
Not as reliably as the confident write-ups suggest, and the most informative test is the one that changed the population.
Thirty-nine healthy adolescents aged 14 to 17 completed a three-week within-subjects crossover of roughly 6.5 hours nightly against roughly 9 hours, with fMRI during a visual food paradigm. The result was a single voxel cluster responding more to food under sleep restriction, covering the left ventral tegmental area and substantia nigra and extending into hippocampus, parahippocampus and fusiform structures (cluster-level P = 0.0072 FWE-corrected; small-volume correction within substantia nigra P = 0.0030, 35 voxels). No change in food appeal ratings was detected, with p values across conditions running from 0.27 to 0.78. The authors stated plainly that their study did not discern differences in several aspects of core reward circuitry that adult imaging studies had observed4.
That is a partial replication at best: the direction survived, the map largely did not, and the subjective ratings again refused to move. Three explanations are live and the study cannot choose between them — adolescent reward circuitry differs from adult, a milder exposure (6.5 hours, not 4 or none) produces a milder signal, or the adult findings were somewhat inflated by small samples and flexible analysis, as much of the early fMRI literature was. All three are plausible. None of them reverses the direction of the effect.
What should raise your confidence is convergence across methods rather than within one. A supermarket experiment, a scanner and a blood assay independently describe the same pull, which is the argument the cravings article makes with the endocannabinoid and purchasing data. Imaging is the weakest of those legs on its own and the most explanatory in combination.
What a valuation problem implies that a hunger problem doesn't#
If the deficit were hunger, eating more would fix it. The imaging says it isn't, and that changes the countermeasure.
Twice now — in Rihm's adults and DiFrancesco's adolescents — the subjective ratings stayed flat while the neural or behavioural valuation moved. A signal you cannot feel is a signal you cannot correct for in the moment, which is precisely why introspection fails here: nothing feels different at 9pm on a tired Tuesday except that the biscuits seem like a better idea than usual, and "seeming like a better idea" is exactly what a raised valuation signal is. The parallel failure on the control side, where the braking system weakens at the same moment, is how sleep loss weakens food willpower.
The practical consequence is that the useful decisions are the ones made at a distance — earlier in the day, in a shop rather than a kitchen, about what is in the building rather than what to do about it now. That is not a motivational point; it is a structural one. A valuation system running 30% hot on doughnuts will win any argument conducted in front of a doughnut, and it will lose easily to a decision made yesterday.
It also argues for treating a badly slept day as a known condition rather than a personal failure — a distinction that matters for whether people keep going, which is the practical thread in managing hunger during weight loss. The neural story is not an excuse. It is a schedule: you now know roughly when your judgement about food will be worse, and that is actionable in a way that guilt never is.
FAQ#
Does sleep loss make everything more tempting, or just food?#
Just food, in the one experiment designed to tell them apart. Men bidding real money on snacks and on trinkets after a night of total sleep deprivation showed a significant sleep state × reward category interaction (F(1,31) = 5.48, P = 0.03), with willingness to pay rising for food alone, and hypothalamic valuation signals increasing food-specifically1. That was 32 lean young men, so treat it as one good demonstration rather than a settled generalisation.
If my hunger hasn't changed, why does the food look better?#
Because the two are measurably separable. In the bidding study, subjective hunger ratings did not differ between the sleep-deprived and rested sessions even as valuation and neural signals rose, and ghrelin changes correlated with neither1. Adolescents under three weeks of sleep restriction likewise showed altered neural food responses with no change in food appeal ratings4. What shifts is what the food is worth, not what your stomach reports.
Do brain scans prove tired people eat more junk food?#
They support it without proving it, and the most-quoted study is weaker evidence than its reputation. St-Onge's 2012 scan found broad reward-network activation to food images but explicitly did not distinguish high- from low-calorie foods2. The follow-up that did make that contrast found unhealthy-food-specific activation after restricted sleep in 25 people3. These are small single-laboratory samples, and an adolescent replication recovered only one cluster4.
Sources#
- Rihm JS, Menz MM, Schultz H, et al. Sleep deprivation selectively upregulates an amygdala-hypothalamic circuit involved in food reward. J Neurosci. 2019;39(5):888-899.
- St-Onge MP, McReynolds A, Trivedi ZB, Roberts AL, Sy M, Hirsch J. Sleep restriction leads to increased activation of brain regions sensitive to food stimuli. Am J Clin Nutr. 2012;95(4):818-824.
- St-Onge MP, Wolfe S, Sy M, Shechter A, Hirsch J. Sleep restriction increases the neuronal response to unhealthy food in normal-weight individuals. Int J Obes (Lond). 2014;38(3):411-416.
- DiFrancesco MW, Alsameen M, St-Onge MP, Duraccio KM, Beebe DW. Altered neuronal response to visual food stimuli in adolescents undergoing chronic sleep restriction. Sleep. 2024;47(4):zsad036.
