The 3-gram bedtime glycine dose the sleep trials actually used

Every glycine-powder sleep trial gave 3 g near lights-out. Andrew Huberman's 2 g comes from his own sleep stack, and no trial has tested it.² The trials behind the 3 g dose are three short Japanese studies, mostly run by Ajinomoto scientists, that enrolled 36 to 40 people between them, depending on which count you accept for the first trial.³¹⁴⁵¹⁶ Their best result is faster entry into slow-wave sleep, with no change in how much of the night it fills. The one non-Ajinomoto trial, funded by the collagen maker Rousselot, gave 15 g of collagen, a similar glycine load. It found fewer awakenings, no faster sleep onset, and no difference in core temperature between collagen and placebo.⁶

01Three grams is the only sleep dose with a trial

If you copied Huberman's 2026 supplement list, you copied 2 g. Inagawa, Yamadera, and Bannai all gave 3 g of flavored glycine against a flavored placebo within an hour of lights-out. Bannai dosed 30 minutes before bed.³¹⁴ Huberman takes 2 g with 100 mg GABA every third or fourth night, alongside magnesium threonate, apigenin, and theanine, because daily use blunts the stack for him.² Treat that as one neuroscientist's diary.

Ajinomoto scientists are authors on all three powder trials. Ajinomoto's Kentaro Inagawa is first author of the 2006 trial and a coauthor of the 2007 one. Makoto Bannai of Ajinomoto is a coauthor of the 2007 trial and corresponding author of the 2012 trial, whose volunteers were Ajinomoto employees.³¹⁴ The results still count. The only test of a similar glycine dose from outside Ajinomoto is the Rousselot-funded collagen trial below.⁶

02What the Japanese glycine trials and one collagen trial found

Inagawa's 2006 crossover used the St Mary's Hospital Sleep Questionnaire and the Space-Aeromedicine Fatigue Checklist the next morning. Fatigue, liveliness, and clear-headedness improved. There was no polysomnography.³ Bannai and Kawai's later mini-review describes 19 women aged 24 to 53, mean 31.1 years.⁵ Norway's VKM risk assessment describes the same trial as 15 women taking glycine or placebo for four nights.¹⁶ I could not reconcile the two counts, so I give both.

Yamadera's 2007 study is the only powder trial with a sleep lab. Eleven volunteers, eight women and three men, aged 30 to 57, mean 40.5, mean PSQI 8.07, took 3 g within an hour of bedtime for two consecutive lab nights. Polysomnography shortened latency to stage 2 (p=0.01) and to stage 3 slow-wave sleep (p=0.019). The share of the night in each stage did not change. REM latency did not change.¹ People fell into stage 2 and slow-wave sleep faster. They did not spend a larger fraction of the night there. Subjective sleep satisfaction (p=0.046), difficulty getting to sleep (p=0.008), estimated time to fall asleep (p=0.002), and sleep efficacy (p=0.018) all improved. Daytime sleepiness fell. Memory-recognition performance rose.¹

Bannai's 2012 crossover enrolled 10 healthy Ajinomoto men and analyzed seven. Mean age 40.6, PSQI 4.7. Time in bed was cut 25% for three nights, 7.3 hours down to 5.5, with 3 g taken 30 minutes before bed. Day 1 visual-analog fatigue improved (p=0.022). Sleepiness only trended (p=0.098). By day 3 the subjective effects were gone. Psychomotor vigilance was faster on days 1 and 3. There was no polysomnography. The rat arm of the same paper, 2 g/kg, left plasma melatonin unchanged.⁴ If your nights are short, the lost sleep is the bigger problem, and the sleep-restriction playbook for muscle shows what it costs. Glycine is a next-morning patch at best.

Soh's 2023 systematic review put the sleep signal in those three short Japanese studies and flagged small samples with high bias risk.⁷ That is as far as the powder evidence goes.

Thomas 2024 is the one check from outside Ajinomoto, and it is mixed. Thirteen athletic men aged 24 ± 4, Athens Insomnia Scale 9 ± 2, took 15 g of bovine-hide collagen peptides, about 3.5 g glycine, or placebo with no food about an hour before bed for seven nights. Night-7 polysomnography in 12 men found fewer awakenings, 21.3 ± 9.7 versus 29.3 ± 13.8 (p=0.028). Subjective awakenings over the week were 1.3 ± 1.5 versus 1.9 ± 0.6 (p=0.023). Sleep onset latency, sleep efficiency, total sleep time, architecture, wake after sleep onset, actigraphy, Karolinska Sleepiness Scale, fatigue, subjective sleep quality, cortisol, IL-6, and hs-CRP did not differ. Core temperature fell overnight in both conditions and did not differ between collagen and placebo.⁶ Rousselot, which makes collagen, funded the trial and supplied the supplements. Fewer awakenings is a real finding. It is not a replication of Yamadera's faster entry into stage 2 and slow-wave sleep, and it shows no glycine-specific cooling.

No trial tested 2 g. The only non-Ajinomoto trial did not reproduce the faster sleep onset. Thomas found no difference in core temperature between collagen and placebo.⁶ For the human cooling report, Kawai 2015 cites a Nagao meeting abstract.¹⁷

03The cooling pathway is mapped in rats

Kawai's 2015 rat paper is the mechanism everyone quotes. Oral glycine at 2 g/kg raised NREM in the first 90 minutes from 15.8 to 35.0 minutes and shortened NREM latency from 54.7 to 36.8 minutes, with a simultaneous drop in core temperature. Plantar cutaneous blood flow rose at 1 and 2 g/kg. NMDA antagonists blocked that blood-flow rise. Strychnine, a glycine-receptor antagonist, did not. Lesioning the suprachiasmatic nucleus abolished the sleep and temperature effects.⁸ The story is NMDA receptors in the SCN shell, peripheral vasodilation, heat loss, then sleep. That is rodent pharmacology at 2 g/kg, not a 3 g human equivalent. Do not convert it.

The human temperature drop quoted online traces to Kawai 2015, which cites a 2007 Nagao meeting abstract for the claim that glycine lowers core temperature in people.¹⁷ Do not quote a human °C figure from that citation. Thomas 2024 measured core temperature after a glycine-matched collagen dose. It fell overnight after both collagen and placebo, and the two conditions did not differ.⁶ Overnight cooling happens anyway. The trial found nothing extra from the collagen. The internet version of glycine, a 3 g scoop that cools you into sleep, is a rat paper plus an abstract. Evening heat can still help sleep through a different cooling sequence, which is the sauna timing problem. Glycine has not earned that claim in people.

The SCN finding does tie glycine to circadian rhythm circuitry in rats. Bannai's rat arm still found no melatonin change at 2 g/kg.⁴ Do not take glycine as a melatonin replacement.

04Food and collagen math for a 3-gram night

Mixed diets already supply about 3 to 5 g of glycine a day.⁴ A 3 g bolus puts close to a full day's food glycine into one dose at bedtime. The trials tested that timing. None of them corrected a deficiency.

Collagen is the convenient scoop. Glycine is about one-third of collagen's amino acid residues, and it is the lightest amino acid, so by weight it makes up closer to a fifth. Lee 2024 listed hydrolyzed collagen at 21.0% glycine by weight, so 15 g is about 3.15 g.⁹ Thomas used 15 g of bovine-hide peptides as about 3.5 g glycine.⁶ Marketing that calls collagen "one-third glycine" overstates the scoop. Dry unsweetened gelatin is 19.049 g glycine per 100 g in USDA SR Legacy, so 15 g of powder is about 2.86 g.¹⁰ Close. Not identical.

Whole milk is about 0.06 g per 100 g. Firm tofu is about 0.73 g per 100 g, so you would need about 400 g to approach 3 g.¹⁰ Bone broth varies too much to assign a number.

Magnesium glycinate at 200 mg elemental magnesium is a magnesium form, dose, and timing choice, not a glycine dose. Pure magnesium bisglycinate is about 14% magnesium by weight. Each magnesium atom, 24.3 g per mole, carries two glycines at 75.07 g per mole each. So 200 mg of magnesium comes with about 1.2 g of glycine.¹⁸ Buffered products that blend in magnesium oxide carry less. Either way it does not replace the 3 g sleep trial.

Gannon's 2002 crossover in nine people is a metabolic side note, not a sleep finding. About 4.6 g of glycine taken with 25 g of glucose cut the glucose area response by more than half versus glucose alone.¹⁴ That is a meal-time observation. It does not tell you what 3 g does at lights-out.

05A bedtime collagen scoop is not the tendon protocol

People buy one 15 g collagen scoop and run it as both a sleep dose and a tendon dose.

Shaw's tendon work times gelatin or collagen peptides about an hour before loading, with vitamin C in the original biomarker study. The collagen protocol for tendons and ligaments is a loading-window substrate. You take it so amino acids are in the blood when the tissue is being stressed. A scoop in a dark kitchen at 10 p.m. on a rest day does not run that protocol. Pre-sleep protein is a different question, and collagen is a poor muscle protein besides.

Thomas gave 15 g of collagen at bedtime for sleep, not for tendon remodeling.⁶ You can use one scoop for glycine grams. You cannot pretend the tendon study and the sleep study are the same experiment because the gram number matches.

06Safety at 3 grams, 9 grams, and 60 grams

Norway's VKM found no adverse events reported in Inagawa's and Yamadera's trials of 3 g for up to four nights. VKM also noted that these studies were short and tracked adverse effects only to a limited extent, so they say little about long-term use.¹⁶ Inagawa's open 2006 follow-up gave 12 people seven days of glycine, with 9 g within an hour of bed on the last three nights. Some reported mild soft stool and abdominal pain. There was no daytime sleepiness carry-over.¹¹ Triple the sleep dose and the gut can notice. That observation comes from 12 people in one open-label study, so it does not set a safe upper limit.

Heresco-Levy's 1999 schizophrenia trial used 0.8 g per kg per day, about 50 to 60 g, as an NMDA co-agonist on top of antipsychotics.¹⁵ That is psychiatry. It is useful only as proof that gram-level glycine has been swallowed at doses an order of magnitude above a bedtime scoop. It is not a sleep protocol.

FDA status is two different rules. 21 CFR 170.50 rescinded GRAS for added glycine in food after animal data and rising use.¹² 21 CFR 172.320 still allows aminoacetic acid as a nutrient additive in conventional foods, under listed limits. Added glycine plus the glycine already in the food, free or bound in protein, cannot exceed 3.5% by weight of the finished food's total protein.¹³ That rule covers glycine added to improve the protein quality of a food that is already a significant protein source. It is not a general authorization for a glycine powder or a dietary supplement. Do not call a 3 g bedtime powder "GRAS" and stop there.

07When glycine is the wrong fix for bad sleep

A late espresso in a slow CYP1A2 metabolizer can outlast any bedtime scoop, and sleep hygiene decides more than glycine does. The Japanese volunteers had sleep complaints or a deliberately shortened week. None had a clinical insomnia diagnosis, and the trials lasted two to four nights.⁷ Thomas's men had sleep complaints too, and his trial ran seven nights.⁶ Nobody in this literature took glycine for months. If the real problem is caffeine timing or a chaotic bedtime, fix that first. Three grams of glycine will not cover for either.

08Stacks that have never been tested

I found no RCT of Huberman's bundle, 2 g glycine plus 100 mg GABA with magnesium threonate, apigenin, and theanine, every third or fourth night.² I found no trial of glycine plus magnesium glycinate either. The powder literature is 3 g of glycine, alone, near bed. Thomas is 15 g of collagen, alone, near bed. Adding a second compound is a new experiment you are running on yourself.

If you want the trial, take 3 g of glycine powder within an hour of lights-out for a few nights and judge the morning. Do not treat Huberman's 2 g, a magnesium glycinate capsule, or a collagen scoop on a rest day as the same night.