[An LLM was used for help with research. All words are mine.]
Benquo's "Is fever a symptom of glycine deficiency?" argues that most people have a substantial dietary glycine deficit.[1] I think most research cited in favor of that proposition is bad (or fails to support the claim). See Raemon's comment and Benquo's response for some limitations of the original post which I won't be addressing. On to Benquo's post:
Meléndez-Hevia et al. 2009
This is what Benquo cites for the 10g/day glycine deficit in his post. A few epistemic spot checks left me thinking this study is worthless.[2] I initially planned to check more of its claims, but I no longer consider it worth my while. An LLM I discussed this with found many more issues with the paper than the ones mentioned below. I've included only those I checked myself against the cited papers. If you want, you can discuss this paper at more length with your own shoggoth.
Authors calculated the supposed deficit thusly:[3]
Now, to examine the supposed glycine need for collagen synthesis—table 3:
Footnotes d–g explain how the crucial rightmost column is calculated. Allow me to walk you through it: (Muscle is used as an example) 523 grams of collagen in muscle (column 1) times 3.52% (column 2). We get 18.4g/day (column 3).
Now what's that "Procollagen cycle% column"? It's synonymous with intracellular collagen degradation. Quoting Rennard 1982:
The intracellular degradation of newly synthesized collagen is a cellular pathway that accounts for the destruction of 10-60% of collagen synthesized by a variety of cell types prior to secretion. This pathway can serve in a regulatory role to limit the secretion of defective molecules, and, [...], regulates the amount and type of collagens secreted.
The calculations implicitly assume that 95.3% of muscle collagen is degraded before secretion from the cells. Carrying on:
18.4 g/day times gives ~391.5 g/day (column 6)
The authors assume a 95% recycling rate of glycine: 391.5 g/day times (1 - 0.95) times 0.261[4] gives ~5.11 g/day (rightmost column)
See the problem?
Procollagen cycle%
If, instead of 95.3%, we used 90%, the estimated glycine need would drop to 47% of the original[5] For the two highest supposed consumers of glycine (muscle and skeleton [6]) the final result is VERY sensitive to the assumptions about intracellular collagen degradation.
In light of this, we'd want the cited sources to demonstrate that the given numbers are precise.
Mays et al. 1991 Rats! Conveniently, this one is cited for four tissue types:
Meléndez-Hevia used data from 15-month-old rats' tissues[8]. These choices maximize the estimated glycine need in the original table 3. Look at those values! THEY ARE ALL OVER THE PLACE! Taking data from the youngest group—If we plug 26.2% into our muscle calculation, the requirement shrinks to 6.4% of the original.[9] The relative drops in estimated requirements of other tissues are also huge, except for skin where it can at most shrink to ~half of the original. Moving on.
For skeleton Bienkowski et al. 1978a was cited. The measurements were on cultured human fetal lung fibroblasts. Under Figure 2[10] authors find that30% of newly synthesized collagen was degraded. In a separate experiment degradation rates of 22 ± 3% and 33 ± 5% were observed in the control conditions and in the intervention arm, respectively. (Figure 3[11]) You may have noticed that those numbers are not equal to 90% Meléndez-Hevia et al. used. As far as I can tell, there is only one sentence in the paper that has this percentage and it's:
Evaluation of the size distribution of the peptides derived from degraded collagen demonstrated that more than 90% of the degraded collagen was reduced to less than 500 daltons within 6 h (data not shown).
The passage only pertains to already degraded collagen. The 90% used by Meléndez-Hevia et al. is contradicted by the study they cite. Plugging in 33% into the "Procollagen cycle%" column would reduce the glycine requirements for bone to 15% of the original.[12]
I did not verify the "procollagen cycle%" numbers for other tissues—they contribute minute amounts to the total of 12 grams. For bone&muscle, it seems overwhelmingly likely that the collagen degradation rate used by Meléndez-Hevia et al. massively overstate the glycine need in humans.
Cartilage FSR:
The source for the fractional synthesis rate (FSR) column is Repo and Mitchell 1971 which used rabbits. We have human data[13]—if you'd allow me to quote from the abstract:
[V]irtually, no replacement of the collagen matrix happened after skeletal maturity and that neither [osteoarthritis] nor tissue damage, per se, influenced collagen turnover. [...] The collagen matrix of human articular cartilage is an essentially permanent structure that has no significant turnover in adults, even with the occurrence of disease.
Sounds conclusive. I admit it doesn't affect the obtained total glycine need much.
Meléndez-Hevia et al. were writing in 2009, so not citing research from 2016 seems quite forgivable. All the same, the cited study doesn't say what Meléndez-Hevia et al. are suggesting—even for rabbit cartilage. What's in the paper?
They injected rabbits' knee joints with radioactive[14] proline and checked total radioactivity and radioactivity from hydroxyproline in cartilage samples[15] as a function of time. On page 4 of the (Repo and Mitchell) pdf we read:
However, by twenty-four days 45 per cent [sic] of the retained activity was recovered as labelled hydroxyproline.
Look at table 3 above; I'm guessing that Meléndez reasoned: "45% is almost 50%, therefore the average lifetime is 24 days"[16] NO! This is not close to what the fragment is saying. The passage says: for cartilage sampled at 24 days [17]. Nothing more.
I'm not sure if the collagen average lifetime of (rabbit) cartilage could be estimated at all from the data provided. But even if it could, it's definitely not 24 days; just look at Figure 5 from Repo and Mitchell:
Muscle FSR:
Three things were cited. Laurent 1982—Rabbits! Table 3 of this study gives 2.7-3.3% FSR /day. Palmer et al. 1980 Rabbits again—Laurent 1982 cited this and wrote:
They obtained rates which ranged from 2.1 to 9.2%/day, for muscles of young rabbits. These values were in general higher than those obtained here, but this is not surprising, since the present study used older rabbits and collagen synthesis rates are known to decrease with age
Converting the per hour rate to daily: 0.018 * 24 = 0.432% FSR /day.
Babraj et al. 2005 isn't cited but the abstract says they found a similar 0.016%/h[19] FSR for healthy young men. Back to Meléndez-Hevia, based on those three studies they put muscle FSR at an oddly specific 3.52%/day. Whence this number? I dunno. I'll just note that the human data suggest an FSR roughly 8-9 times smaller than the one used in the table.
Skeleton FSR:
Babraj et al. 2005 is the source for the FSR. Based on Table 2 and the abstract, the paper gives two numbers: one for a fasted state the second during feeding[20]—0.066%/h and 0.11%/h respectively. They picked the higher number and implicitly assumed that this rate of synthesis continues throughout 24 hours for the whole skeleton collagen content. I'll quote from the discussion section:
Our results provide new insights into human bone collagen metabolism: they suggest that, in the iliac crest at least, its synthesis (and hence its turnover) is faster (∼1.6%/day)[21] than expected, given generally accepted values of adult cortical and trabecular bone remodeling, respectively, of 3–25% per year(22) or of bone calcium turnover of ∼8–15% per year.(23,24) The discrepancy in values may be caused because of errors in our measurements, because the values for iliac crest collagen synthesis are unrepresentative of bone generally, or because the prevailing paradigm is in error and values obtained from bone histomorphometry and bone markers are unrepresentative of bone collagen turnover for reasons not yet understood.
This fragment and the rest of the discussion section suggest that the authors are aware that the "prevailing paradigm" is at odds with their own results and they don't have a good idea why. I don't know either, of course.
Recycling rate
Another thing you might have noticed is that the final number is also highly reliant on the recycling rate. How does Meléndez-Hevia justify this number?
The gross total of 924 g collagen corresponds to a consumption of 240 g glycine (if there is no recycling of glycine), a huge amount, [more than we eat/synthesize]. Clearly, therefore, much of the glycine produced from collagen breakdown and from the procollagen cycle must be recycled, more than 98%, if there is to be no deficit at all. It is very unlikely that the proportion can be as high as this [...]. Moreover, the measured value for cysteine reutilization is only 85% (section 4.1.4), and although the rate for glycine may be higher, there is no reason to suppose it is so much higher. The right-hand column of table 3 gives the amount of glycine that is needed if 95% of the glycine released in the procollagen cycle is recycled[.]
They start by taking their stratospheric estimate of daily collagen synthesis at face-value. And then... they pull 95% out of a hat. As the authors suggest, the actual recycling rate might be lower—the significance of that is unclear since the 924 g/day figure is unequivocally preposterous.
(I didn't investigate whether the 85% recycling rate for cysteine checks out.)
Conclusion on Meléndez-Hevia et al.
Table 3 used throughout was labeled as "maximal", but what of it? It uses maxima drawn from different species, ages, nutritional states etc.—the result isn't tethered to any coherent physiological state— let alone a plausible human one. Meléndez-Hevia et al. then carry this "maximum" into Table 4 and the discussion section as evidence of an actual glycine deficit.
In my eyes, Meléndez-Hevia et al. not only failed to demonstrate a 10-gram daily glycine deficit; they failed to convincingly argue that there's a deficit at all.
Other studies
Benquo also cites other things in support of his claim.
15g gelatin before [jumping rope] doubled a collagen synthesis marker in blood (Shaw et al. 2017) [n = 8]
Looks like the result for PINP[22] here was p<0.05 compared with placebo and p<0.01 compared with 5 grams of gelatin.
I found three quasi-replications: Lis & Baar 2019 n = 10. Jumping rope too. They had three intervention groups[23] No significant effect was found
Lee et al. 2024: n = 10. Rather than jumping rope, they were doing barbell back squats. Results: 30 vs 15 grams p = 0.013; 30 vs 0: p = 0.002; 15 vs 0 gave a null.
Nulty et al. 2024: n = 8; leg press; for PINP: 30 vs 15 p = 0.033; 30 vs 0 p<0.001; 15 vs 0 p = 0.024;
The replications don't look too bad. Does the observed rise of a marker in blood cause any outcome of interest?
In a systematic review and meta-analysis by Bischof et al. 2024 the authors found a few statistically significant effects[24], while also noting:
The GRADE approach revealed a moderate certainty of evidence for body composition, a very low certainty for tendon morphology and mechanical properties, and a low certainty for the remaining.
Bischof et al. seems more of interest than the study cited by Benquo, since they analyzed weeks-long studies and they evaluated outcomes—not markers. With more studies and subjects.
Overall, all this seems quite inconclusive to me. But even supposing that gelatin&exercise results hold up, are they strong evidence of a glycine deficiency? Usually, claims of the form: "Supplement X helps with outcome Y" are not treated as synonymous with "The median person has an X deficiency". Even when the former claim is well established. As I understand it, "deficiency" means that a lack of a nutrient impairs the body's ‘normal’ functioning.
Whey protein supplementation helps with muscle gain, is that indicative of a protein deficiency? How about creatine? Is that a deficiency too?
Whole-body glutathione utilization in healthy adults is about 25 µmol/(kg·h) (Wu et al. 2004), which for a 70 kg person works out to roughly 3.2g of glycine per day consumed by glutathione turnover alone. Under metabolic stress this rate increases substantially: Darmaun et al. 2005 found a 1.7× increase in glutathione fractional synthesis rate in poorly controlled type 1 diabetics, which would bring the glycine demand from glutathione to roughly 5.4g/day.
The 3.2 g/day figure assumes a 0% recycling rate after glutathione utilization. Seems dubious.
Funnily enough Meléndez-Hevia et al. 2009 considered glutathione synthesis as well (section 4.1.4). They used an 85% recycling rate and got a daily need of 0.567 grams[25] of glycine for that purpose.
I don't really trust the scientific rigor of the Meléndez-Hevia study at this point, but still, it stands to reason that 3.2g/day is a large overestimate.
As to the second claim, the cited study did find an increase in glutathione FSR but the subjects with poorly controlled diabetes had a smaller glutathione pool to turn over—this netted out in no difference in the absolute glutathione synthesis rates:
Admittedly, the standard error bars for ASR are pretty wide.
5-oxoproline
Benquo writes:
Others have noted that markers of glycine deficiency appear in the urine of vegetarians, people on low-protein diets, children recovering from malnourishment, and pregnant women.
This statement seems to come from Meléndez-Hevia et al. 2009 discussion section. See the footnote for the fragment[26]. Before proceeding, I'll just note that urinary 5-oxoproline isn't a confirmed measure of glycine deficiency, but it stands to reason they should be related.
Children recovering from malnourishment
What does Persaud et al. 1996 say? It's an n = 12 study of toddlers[27] recovering from severe malnutrition. They found that:
Urinary 5-L-oxoproline was similar at admission and after recovery, but was increased significantly during rapid catch up growth. There was a significant relationship between the rate of weight gain and 5-L-oxoproline excretion in urine.
In a separate protocol with different children, the authors found:
In nine children (aged 15 5 mo), the effect of oral supplementation with glycine, [1.7 [28] for 48 h] during rapid catch-up growth on 5-Loxoprolinuria and blood glutathione concentration was determined. In seven of the nine children weight gain was less than 17 and following oral glycine supplements 5-L-oxoproline excretion was reduced up to 64% and blood glutathione concentration increased up to 100%. In the two children who were gaining weight at a rate >17 , glycine supplementation was associated with a further increase in 5-L-oxo proline excretion and a decrease in blood glutathione.
The limitation is that the second protocol had no control group, and the authors reported only 2 p-values:
The urinary excretion of 5-L-oxoproline was 178 ±68 µmol/mmol creatinine, with a statistically significant decrease to 137 ±70 µmol/mmol creatinine (P < 0.018) after the 48 h of supplementation.
And a p<0.01 for a subgroup analysis.
I'm unimpressed and I'm pretty sure this study was underpowered to detect the effects it wanted—assuming they're real.
They measured 5-oxoproline excretion—with 3 measurements of interest: (Bolding mine. Heavily edited for brevity)
Male and female omnivores. Complete collections of urine over 24 h were taken from 25 males and 25 females, aged 20 to 45 y[.] [...] Male and female vegetarians. Complete collections of urine over 24 h were taken from 12 males and 12 females, aged 20 to 45 y[.] [...] [6] normal males, aged 20 to 24 y, were studied during each of four periods of 5 d each (Meakins and Jackson 1996). For each period the subjects were provided with a diet consisting of [ a) 70g/day of protein. b) <30g/day of protein.[29] c) <30g/day of protein with the addition of 6.9 g of urea. d) <30g/day of protein with the addition of 13.8 g of urea. Total daily nitrogen consumption in diets a, b, c, d was 11.2 g, 4.8 g, 8.0 g, 11.2 g respectively] [...] A further group of [9] normal adults (three males and six females, aged 21 to 26 y) made complete collections of urine for 24 h after 5 d of consuming a diet designed to provide [<40g/day of protein. Nitrogen—6.4g]
First, the omnivore vs vegetarian comparison. The difference between the omnivore group and vegetarian males was p<0.0001. Compared with vegetarian females—p = 0.005. Very impressive p-values. My guess is that the difference is caused by lower total protein consumption and/or digestibility[30].
What about those controlled diets?
There was a significant inverse linear relationship between dietary nitrogen and urinary 5-L-oxoproline (r = -0.315, P = 0.005, n = 81)[.]
There are some confusing aspects about the results, but I see no reason to dig further, because, well, consider this statement: "After a period of inadequate protein intake, a marker of [amino acid x] deficiency increases". If you don't get enough protein from your diet, your body will make an implicit tradeoff between cutting back on beneficial processes requiring amino acids and cannibalizing its own tissues to get them. This should hold even for amino acids whose synthesis is not bottlenecked in any way. The Estimated Average Requirement (EAR)[31] for protein (sedentary individuals) is 0.66g/day per kg of body weight—46.2 grams for a 70kg male. In light of this, the result seems unremarkable and unlikely to shed light on dietary requirements of glycine in particular.
Glycine, a Dispensable Amino Acid, Is Conditionally Indispensable in Late Stages of Human Pregnancy. The women were given a standardized diet with 0.88 of protein and varying amounts of glycine.
The results for late stage pregnancy are in C & D in the figure below:
No p-value provided but the variance explained isn't too bad. The supposed breakpoints of and for glycine sufficiency translate to 1.95 and 2.69 grams of glycine respectively, in this sample.[33] Those values are below median glycine intakes of both vegans and omnivores in Dietrich et al. 2022 (see footnote 30).
In the discussion section the authors also cite observational research showing higher urinary 5-oxoproline levels in pregnant women vs non-pregnant women. This result seems quite plausible but not decisive by itself.
To wrap up this section, glycine is typically considered conditionally essential—i.e. normally the body can produce adequate amounts for functioning but under some, well, conditions it needs to be supplied in the diet. I see no reason to doubt this. But at the same time, none of those studies constitute much evidence that the median person has a glycine deficiency.
The amino acids sarcosine and glycine [...] have also been assessed as adjunctive treatments for schizophrenia, [...] Neither sarcosine (at 2 g/day) or glycine (at 2.8‐60 g/day)[34] had any effect on positive symptoms, although both did significantly reduce total psychopathology as an adjunctive to antipsychotic treatment ([...] glycine: N=159, n=6, SMD=0.66, 95% CI: 0.04‐1.28, p=0.04, I2not reported)57.
The effects on negative symptoms fell short of statistical significance ([...] glycine: N=268, n=7, SMD=0.39, 95% CI: –0.11 to 0.9, p=0.13)57. However, significant benefits for negative symptoms were observed in individuals treated with non‐clozapine antipsychotics ([...] glycine: N=219, n=5, SMD=0.60, p=0.05; CIs and I2not provided)57.[35]
As an adjunctive to clozapine treatment (N=58, n=3)58, glycine was ineffective for positive (SMD=0.63, 95% CI: –0.21 to 1.48, I2not reported), negative (SMD=0.03, 95% CI: –0.51 to 0.57, I2not reported) and total symptoms scores (SMD=0.32, 95% CI: –0.2 to 0.84, I2not reported).
Doesn't look very strong to my eyes—one p = 0.04 and a p = 0.05 + 2 decent p-values for a subgroup of studies.
The rest of the justifications in Benquo's post consist of animal research and theoretical mechanistic models. I won't be engaging with those parts—I get back to it a bit in the "Closing thoughts" section.
Other than Benquo's, there was a post last month on LessWrong where the author claims glycine supplementation greatly improved his sleep. To reiterate: claims of the form: "Supplement X helps with outcome Y" are not synonymous with "The median person has an X deficiency".
Still, how robust are those sleep results?
Glycine&sleep
I couldn't find the fulltext of this scoping review. However, the shoggoth helped me find the 3 studies that tested the effects of 3g of glycine on sleep-related measures.
Inagawa et al. 2006 4 out of 5 authors were affiliated with Ajinomoto Co.—a company selling amino acids, among them, glycine. 15 out of 19 subjects completed the study—all of them Ajinomoto Co. employees. The effects were assessed using two questionnaires. They reported an effect on the total scores of the SAM Fatigue Checklist (P = 0.031). Also, they list statistically significant differences on 3 out of 24 questionnaire items, but you'd expect to find some results there even knowing that the null hypothesis is true.
No effect on sleep time or sleep latency was detected.
There's more. The paper doesn't say whether the test was one- or two-sided, but GPT 6.1 Sol thinks it was one-sided.[36] A two-sided analysis would have left the authors with a single p < 0.05[37] result—on the item "very lively" from the SAM Fatigue Checklist.[38]
3 out of 6 authors affiliated with Ajinomoto Co. n = 11. Results in order: (no correction for multiple testing was used by the authors)
They used the same sleep questionnaire as in the 2006 study. Significant results for 3 of the 14 questions in the questionnaire and for a sleep-efficiency measure were reported.
On the polysomnography results they found p < 0.05 for stage 2 and stage 3 sleep onset, but not for REM. They also reported no significant effect on sleep architecture.
Two different measurements of daytime sleepiness were used at 8:00, 10:00, 12:00, 21:00, 23:00. Only one of the two measures showed a significant result, at 10:00 (p < 0.01).[39]
Two different measures of cognitive function were used at 8:00, 10:00, 12:00. The only significant result was on a memory recognition task at 12:00. (also at p < 0.01)
3 out of 5 authors affiliated with Ajinomoto Co. Data from 7 out of 10 initial subjects was analyzed. Again, all of them Ajinomoto Co. employees.
The participants were deliberately sleep-restricted during the experimental period. (average of 5.5 hours in bed) Measurements were taken on days 1 and 3 of the 3-day supplement protocol. Quoting from the results section:
Data from the VAS test showed significant effects on fatigue on day 1 (p = 0.022); however, no significant effects were observed for sleepiness (p = 0.098, Table 1). Post hoc analysis revealed a significant improvement in fatigue at 14:00 (p = 0.022). [...] Table 1 also shows the results of the questionnaire for fatigue complaints. No significant effects were observed between the placebo and glycine treatment groups for any measurement; [...] On day 3, no significant difference was observed between the placebo and glycine treatment groups.
They also used 4 cognitive function tests (measuring correct response rates and reaction time separately for each). For reaction time in one task, they reported p = 0.003 on day 1 and p = 0.05 on day 3. The other seven measures showed no significant overall treatment effect. See footnote for all the results. [40]
Interpretation
The evidence base is abysmal—we have 3 studies. All of them affiliated with one company. All of them with low n. The results of the 2006 study are basically what you'd expect under the hypothesis that glycine doesn't help sleep. The 2007 and 2012 studies used the classic trick of measuring a gorrilion outcomes and using those that came back significant as evidence.
Closing thoughts
My opinions on the following statements.
"In some conditions the body can be glycine deficient"
Certainly. As I understand it, there's no doubt about the conditionally essential status of glycine in the research literature.
"Glycine helps with outcome X" I very tentatively conclude that some of the Gelatin&exercise results might hold up. I see very little reason to think glycine improves sleep or reduces the need for it.
"The median person has a >=10 gram daily glycine deficit" Preposterous until proven otherwise.
Furthermore, I concede that you can find plenty of limitations in most RCTs. Some of them sensible, some of them not—but I don't think I'm being unreasonable here or making isolated demands for rigor.
Most supplements don't work. If you disagree you should probably be popping tens, if not hundreds, of different ones daily[41].
I think the evidence base for glycine is bad, but not uniquely bad. Most of the studies examined in this piece on the supposed benefits of glycine supplementation are shaky by themselves. The whole literature is also afflicted with an unknown amount of publication bias. Animal studies offer little evidence. Translation is uncertain, and they share many of the methodological problems of human research—p-hacking, publication bias, small sample sizes. Mechanistic just-so stories should not by themselves be treated as evidence either. I don't think they distinguish truth from falsehood. The claim: "Nutrient X is involved in beneficial process Y" is basically always true—but it's not possible to establish a priori whether taking more X will result in more process Y happening in the body[42]. Or that more of process Y would causally influence general health.
Of course, supplements can (semi-stochastically) help particular people without being beneficial for the population as a whole. But if you want to claim a general and predictable benefit of a supplement—the burden of proof is on you. And I don't think that any sensible evidentiary bar was cleared for glycine's effect on sleep or any other specific outcome.
Pardon the nitpick, but in Benquo's footnote 5 we read:
Meléndez-Hevia’s flux model estimates that baseline collagen turnover is the single largest consumer of glycine at roughly 14.5g/day of the ~15g total need.
As can be seen from the table below—it's 12g/day of a 14.5g total.
Two clear signs of glycine deficiency come from studies of the metabolism of glutathione and benzoic acid. Under normal conditions, only a small fraction of 5-oxoproline, a product of glutathione metabolism (figure 10), is excreted[.] It is also excreted in the urine in large quantities in conditions of glycine deficiency (Persaud et al. 1996), and this symptom can be corrected by supplementing the diet with glycine; the authors proposed that assay of 5-oxoproline in urine supplies a way of measuring glycine deficiency. High amounts of this metabolite have been found in the urine of vegetarians (186% of normal values for males, and 136% for females), similar to the amounts in subjects with a low-protein diet (25 g/day), who had 141% of the 5-oxoproline found in subjects with a high-protein diet (70 g/day) (Jackson et al. 1996).
In the "SUBJECTS AND ANALYTICAL METHODS" section, the authors describe those lower protein groups as consuming "no more than" 30g/day of protein, but later in the study they switch to writing that they consumed 25 grams daily.
It seems dubious that the difference is caused by a glycine deficiency specifically vs inadequate protein in general (or something else entirely). In Schmidt et al. 2015 (each subgroup was n = 98), both vegetarians and vegans had higher plasma glycine concentrations than omnivores—105% and 116% respectively. They found substantially lower glycine consumption (73%, 84%), but as the authors themselves note, their method of estimating amino acid consumption may have been inadequate for veg*ns (a retrospective food questionnaire). Do we have anything better?
Dietrich et al. 2022 (n = 72, split evenly) compared the amino acid intake of omnivores and vegans (not vegetarians) using a weighed food protocol across 3 days. Glycine intake was 41.1 vs 45.6 (~10% lower) in vegans and omnivores which was not significant after correction for multiple comparisons. The total protein intake was 1 vs 1.2 (~17% lower). In this study vegans had higher glycine concentrations as well. 134% of that of omnivores.
Not mentioned in the review but the cited meta-analysis also found benefits for positive symptoms at p = 0.01 and total symptoms at p rounded to 0.00 (for non-clozapine antipsychotics).
Based on the Figure 1 from the study. I did not independently verify this. I personally believe this is most likely true and the rest of this section is written accordingly.
I think ivermectin doesn’t work. I think that it looks like it works, because it has lots of positive studies and a few big-name endorsements. But our current scientific method is so weak and error-prone that any chemical which gets raised to researchers’ attentions and studied in depth will get approximately this amount of positive results and buzz. Look through the thirty different chemicals featured on the sidebar of the ivmmeta site if you don’t believe me.
[An LLM was used for help with research. All words are mine.]
Benquo's "Is fever a symptom of glycine deficiency?" argues that most people have a substantial dietary glycine deficit.[1] I think most research cited in favor of that proposition is bad (or fails to support the claim).
See Raemon's comment and Benquo's response for some limitations of the original post which I won't be addressing.
On to Benquo's post:
Meléndez-Hevia et al. 2009
This is what Benquo cites for the 10g/day glycine deficit in his post.
A few epistemic spot checks left me thinking this study is worthless.[2] I initially planned to check more of its claims, but I no longer consider it worth my while.
An LLM I discussed this with found many more issues with the paper than the ones mentioned below. I've included only those I checked myself against the cited papers.
If you want, you can discuss this paper at more length with your own shoggoth.
Authors calculated the supposed deficit thusly:[3]
Now, to examine the supposed glycine need for collagen synthesis—table 3:
Footnotes d–g explain how the crucial rightmost column is calculated.
Allow me to walk you through it: (Muscle is used as an example)
523 grams of collagen in muscle (column 1) times 3.52% (column 2). We get 18.4g/day (column 3).
Now what's that "Procollagen cycle% column"? It's synonymous with intracellular collagen degradation. Quoting Rennard 1982:
The calculations implicitly assume that 95.3% of muscle collagen is degraded before secretion from the cells. Carrying on:
18.4 g/day times
The authors assume a 95% recycling rate of glycine:
391.5 g/day times (1 - 0.95) times 0.261[4] gives ~5.11 g/day (rightmost column)
See the problem?
Procollagen cycle%
If, instead of 95.3%, we used 90%, the estimated glycine need would drop to 47% of the original[5]
For the two highest supposed consumers of glycine (muscle and skeleton [6]) the final result is VERY sensitive to the assumptions about intracellular collagen degradation.
In light of this, we'd want the cited sources to demonstrate that the given numbers are precise.
Mays et al. 1991
Rats!
Conveniently, this one is cited for four tissue types:
How were those values picked?
Table 5:
Meléndez-Hevia used data from 15-month-old rats' tissues[8]. These choices maximize the estimated glycine need in the original table 3.
Look at those values! THEY ARE ALL OVER THE PLACE!
Taking data from the youngest group—If we plug 26.2% into our muscle calculation, the requirement shrinks to 6.4% of the original.[9] The relative drops in estimated requirements of other tissues are also huge, except for skin where it can at most shrink to ~half of the original.
Moving on.
For skeleton Bienkowski et al. 1978a was cited.
The measurements were on cultured human fetal lung fibroblasts.
Under Figure 2[10] authors find that 30% of newly synthesized collagen was degraded.
In a separate experiment degradation rates of 22 ± 3% and 33 ± 5% were observed in the control conditions and in the intervention arm, respectively. (Figure 3[11])
You may have noticed that those numbers are not equal to 90% Meléndez-Hevia et al. used. As far as I can tell, there is only one sentence in the paper that has this percentage and it's:
The passage only pertains to already degraded collagen.
The 90% used by Meléndez-Hevia et al. is contradicted by the study they cite.
Plugging in 33% into the "Procollagen cycle%" column would reduce the glycine requirements for bone to 15% of the original.[12]
I did not verify the "procollagen cycle%" numbers for other tissues—they contribute minute amounts to the total of 12 grams.
For bone&muscle, it seems overwhelmingly likely that the collagen degradation rate used by Meléndez-Hevia et al. massively overstate the glycine need in humans.
Cartilage FSR:
The source for the fractional synthesis rate (FSR) column is Repo and Mitchell 1971 which used rabbits. We have human data[13]—if you'd allow me to quote from the abstract:
Sounds conclusive. I admit it doesn't affect the obtained total glycine need much.
Meléndez-Hevia et al. were writing in 2009, so not citing research from 2016 seems quite forgivable. All the same, the cited study doesn't say what Meléndez-Hevia et al. are suggesting—even for rabbit cartilage. What's in the paper?
They injected rabbits' knee joints with radioactive[14] proline and checked total radioactivity and radioactivity from hydroxyproline in cartilage samples[15] as a function of time. On page 4 of the (Repo and Mitchell) pdf we read:
Look at table 3 above; I'm guessing that Meléndez reasoned: "45% is almost 50%, therefore the average lifetime is 24 days"[16]
NO!
This is not close to what the fragment is saying.
The passage says:
for cartilage sampled at 24 days
Nothing more.
I'm not sure if the collagen average lifetime of (rabbit) cartilage could be estimated at all from the data provided. But even if it could, it's definitely not 24 days; just look at Figure 5 from Repo and Mitchell:
Muscle FSR:
Three things were cited.
Laurent 1982—Rabbits! Table 3 of this study gives 2.7-3.3% FSR /day.
Palmer et al. 1980 Rabbits again—Laurent 1982 cited this and wrote:
(I only took a brief glance at the studies above)
Waterlow 2006 is a book. From page 272:[18]
Converting the per hour rate to daily: 0.018 * 24 = 0.432% FSR /day.
Babraj et al. 2005 isn't cited but the abstract says they found a similar 0.016%/h[19] FSR for healthy young men.
Back to Meléndez-Hevia, based on those three studies they put muscle FSR at an oddly specific 3.52%/day. Whence this number? I dunno.
I'll just note that the human data suggest an FSR roughly 8-9 times smaller than the one used in the table.
Skeleton FSR:
Babraj et al. 2005 is the source for the FSR.
Based on Table 2 and the abstract, the paper gives two numbers: one for a fasted state the second during feeding[20]—0.066%/h and 0.11%/h respectively.
They picked the higher number and implicitly assumed that this rate of synthesis continues throughout 24 hours for the whole skeleton collagen content.
I'll quote from the discussion section:
This fragment and the rest of the discussion section suggest that the authors are aware that the "prevailing paradigm" is at odds with their own results and they don't have a good idea why.
I don't know either, of course.
Recycling rate
Another thing you might have noticed is that the final number is also highly reliant on the recycling rate. How does Meléndez-Hevia justify this number?
They start by taking their stratospheric estimate of daily collagen synthesis at face-value. And then... they pull 95% out of a hat.
As the authors suggest, the actual recycling rate might be lower—the significance of that is unclear since the 924 g/day figure is unequivocally preposterous.
(I didn't investigate whether the 85% recycling rate for cysteine checks out.)
Conclusion on Meléndez-Hevia et al.
Table 3 used throughout was labeled as "maximal", but what of it? It uses maxima drawn from different species, ages, nutritional states etc.—the result isn't tethered to any coherent physiological state— let alone a plausible human one. Meléndez-Hevia et al. then carry this "maximum" into Table 4 and the discussion section as evidence of an actual glycine deficit.
In my eyes, Meléndez-Hevia et al. not only failed to demonstrate a 10-gram daily glycine deficit; they failed to convincingly argue that there's a deficit at all.
Other studies
Benquo also cites other things in support of his claim.
Gelatin&exercise
From footnote 1:
Looks like the result for PINP[22] here was p<0.05 compared with placebo and p<0.01 compared with 5 grams of gelatin.
I found three quasi-replications:
Lis & Baar 2019 n = 10. Jumping rope too.
They had three intervention groups[23]
No significant effect was found
Lee et al. 2024: n = 10. Rather than jumping rope, they were doing barbell back squats. Results:
30 vs 15 grams p = 0.013;
30 vs 0: p = 0.002;
15 vs 0 gave a null.
Nulty et al. 2024: n = 8; leg press; for PINP:
30 vs 15 p = 0.033;
30 vs 0 p<0.001;
15 vs 0 p = 0.024;
The replications don't look too bad.
Does the observed rise of a marker in blood cause any outcome of interest?
In a systematic review and meta-analysis by Bischof et al. 2024 the authors found a few statistically significant effects[24], while also noting:
Bischof et al. seems more of interest than the study cited by Benquo, since they analyzed weeks-long studies and they evaluated outcomes—not markers. With more studies and subjects.
Overall, all this seems quite inconclusive to me. But even supposing that gelatin&exercise results hold up, are they strong evidence of a glycine deficiency?
Usually, claims of the form: "Supplement X helps with outcome Y" are not treated as synonymous with "The median person has an X deficiency". Even when the former claim is well established.
As I understand it, "deficiency" means that a lack of a nutrient impairs the body's ‘normal’ functioning.
Whey protein supplementation helps with muscle gain, is that indicative of a protein deficiency?
How about creatine? Is that a deficiency too?
Glutathione
From footnote 4:
The 3.2 g/day figure assumes a 0% recycling rate after glutathione utilization. Seems dubious.
Funnily enough Meléndez-Hevia et al. 2009 considered glutathione synthesis as well (section 4.1.4). They used an 85% recycling rate and got a daily need of 0.567 grams[25] of glycine for that purpose.
I don't really trust the scientific rigor of the Meléndez-Hevia study at this point, but still, it stands to reason that 3.2g/day is a large overestimate.
As to the second claim, the cited study did find an increase in glutathione FSR but the subjects with poorly controlled diabetes had a smaller glutathione pool to turn over—this netted out in no difference in the absolute glutathione synthesis rates:
Admittedly, the standard error bars for ASR are pretty wide.
5-oxoproline
Benquo writes:
This statement seems to come from Meléndez-Hevia et al. 2009 discussion section. See the footnote for the fragment[26].
Before proceeding, I'll just note that urinary 5-oxoproline isn't a confirmed measure of glycine deficiency, but it stands to reason they should be related.
Children recovering from malnourishment
What does Persaud et al. 1996 say?
It's an n = 12 study of toddlers[27] recovering from severe malnutrition. They found that:
In a separate protocol with different children, the authors found:
The limitation is that the second protocol had no control group, and the authors reported only 2 p-values:
And a p<0.01 for a subgroup analysis.
I'm unimpressed and I'm pretty sure this study was underpowered to detect the effects it wanted—assuming they're real.
Jackson et al. 1996
They measured 5-oxoproline excretion—with 3 measurements of interest:
(Bolding mine. Heavily edited for brevity)
First, the omnivore vs vegetarian comparison.
The difference between the omnivore group and vegetarian males was p<0.0001. Compared with vegetarian females—p = 0.005.
Very impressive p-values. My guess is that the difference is caused by lower total protein consumption and/or digestibility[30].
What about those controlled diets?
There are some confusing aspects about the results, but I see no reason to dig further, because, well, consider this statement:
"After a period of inadequate protein intake, a marker of [amino acid x] deficiency increases".
If you don't get enough protein from your diet, your body will make an implicit tradeoff between cutting back on beneficial processes requiring amino acids and cannibalizing its own tissues to get them. This should hold even for amino acids whose synthesis is not bottlenecked in any way.
The Estimated Average Requirement (EAR)[31] for protein (sedentary individuals) is 0.66g/day per kg of body weight—46.2 grams for a 70kg male.
In light of this, the result seems unremarkable and unlikely to shed light on dietary requirements of glycine in particular.
Pregnancy
Rasmussen et al. 2021[32], titled:
Glycine, a Dispensable Amino Acid, Is Conditionally Indispensable in Late Stages of Human Pregnancy.
The women were given a standardized diet with 0.88
The results for late stage pregnancy are in C & D in the figure below:
No p-value provided but the variance explained isn't too bad.
The supposed breakpoints of
In the discussion section the authors also cite observational research showing higher urinary 5-oxoproline levels in pregnant women vs non-pregnant women.
This result seems quite plausible but not decisive by itself.
To wrap up this section, glycine is typically considered conditionally essential—i.e. normally the body can produce adequate amounts for functioning but under some, well, conditions it needs to be supplied in the diet. I see no reason to doubt this. But at the same time, none of those studies constitute much evidence that the median person has a glycine deficiency.
Anything else?
There's also a non-randomized trial on glycine's effect on respiratory infections and schizophrenia trials. Benquo acknowledges the limitations of the former.
Regarding the latter, quoting this meta-review of meta-analyses:
(bolding mine)
Doesn't look very strong to my eyes—one p = 0.04 and a p = 0.05 + 2 decent p-values for a subgroup of studies.
The rest of the justifications in Benquo's post consist of animal research and theoretical mechanistic models. I won't be engaging with those parts—I get back to it a bit in the "Closing thoughts" section.
Other than Benquo's, there was a post last month on LessWrong where the author claims glycine supplementation greatly improved his sleep. To reiterate: claims of the form: "Supplement X helps with outcome Y" are not synonymous with "The median person has an X deficiency".
Still, how robust are those sleep results?
Glycine&sleep
I couldn't find the fulltext of this scoping review. However, the shoggoth helped me find the 3 studies that tested the effects of 3g of glycine on sleep-related measures.
Inagawa et al. 2006
4 out of 5 authors were affiliated with Ajinomoto Co.—a company selling amino acids, among them, glycine.
15 out of 19 subjects completed the study—all of them Ajinomoto Co. employees.
The effects were assessed using two questionnaires.
They reported an effect on the total scores of the SAM Fatigue Checklist (P = 0.031).
Also, they list statistically significant differences on 3 out of 24 questionnaire items, but you'd expect to find some results there even knowing that the null hypothesis is true.
No effect on sleep time or sleep latency was detected.
There's more.
The paper doesn't say whether the test was one- or two-sided, but GPT 6.1 Sol thinks it was one-sided.[36]
A two-sided analysis would have left the authors with a single p < 0.05[37] result—on the item "very lively" from the SAM Fatigue Checklist.[38]
Yamadera et al. 2007
3 out of 6 authors affiliated with Ajinomoto Co.
n = 11.
Results in order: (no correction for multiple testing was used by the authors)
They used the same sleep questionnaire as in the 2006 study. Significant results for 3 of the 14 questions in the questionnaire and for a sleep-efficiency measure were reported.
On the polysomnography results they found p < 0.05 for stage 2 and stage 3 sleep onset, but not for REM. They also reported no significant effect on sleep architecture.
Two different measurements of daytime sleepiness were used at 8:00, 10:00, 12:00, 21:00, 23:00. Only one of the two measures showed a significant result, at 10:00 (p < 0.01).[39]
Two different measures of cognitive function were used at 8:00, 10:00, 12:00. The only significant result was on a memory recognition task at 12:00. (also at p < 0.01)
Bannai et al. 2012
3 out of 5 authors affiliated with Ajinomoto Co.
Data from 7 out of 10 initial subjects was analyzed. Again, all of them Ajinomoto Co. employees.
The participants were deliberately sleep-restricted during the experimental period. (average of 5.5 hours in bed)
Measurements were taken on days 1 and 3 of the 3-day supplement protocol. Quoting from the results section:
They also used 4 cognitive function tests (measuring correct response rates and reaction time separately for each). For reaction time in one task, they reported p = 0.003 on day 1 and p = 0.05 on day 3. The other seven measures showed no significant overall treatment effect. See footnote for all the results. [40]
Interpretation
The evidence base is abysmal—we have 3 studies.
All of them affiliated with one company.
All of them with low n.
The results of the 2006 study are basically what you'd expect under the hypothesis that glycine doesn't help sleep.
The 2007 and 2012 studies used the classic trick of measuring a gorrilion outcomes and using those that came back significant as evidence.
Closing thoughts
My opinions on the following statements.
"In some conditions the body can be glycine deficient"
Certainly. As I understand it, there's no doubt about the conditionally essential status of glycine in the research literature.
"Glycine helps with outcome X"
I very tentatively conclude that some of the Gelatin&exercise results might hold up. I see very little reason to think glycine improves sleep or reduces the need for it.
"The median person has a >=10 gram daily glycine deficit"
Preposterous until proven otherwise.
Furthermore, I concede that you can find plenty of limitations in most RCTs. Some of them sensible, some of them not—but I don't think I'm being unreasonable here or making isolated demands for rigor.
Most supplements don't work. If you disagree you should probably be popping tens, if not hundreds, of different ones daily[41].
I think the evidence base for glycine is bad, but not uniquely bad. Most of the studies examined in this piece on the supposed benefits of glycine supplementation are shaky by themselves. The whole literature is also afflicted with an unknown amount of publication bias.
Animal studies offer little evidence. Translation is uncertain, and they share many of the methodological problems of human research—p-hacking, publication bias, small sample sizes.
Mechanistic just-so stories should not by themselves be treated as evidence either. I don't think they distinguish truth from falsehood. The claim: "Nutrient X is involved in beneficial process Y" is basically always true—but it's not possible to establish a priori whether taking more X will result in more process Y happening in the body[42]. Or that more of process Y would causally influence general health.
Of course, supplements can (semi-stochastically) help particular people without being beneficial for the population as a whole. But if you want to claim a general and predictable benefit of a supplement—the burden of proof is on you. And I don't think that any sensible evidentiary bar was cleared for glycine's effect on sleep or any other specific outcome.
In the "Practical Considerations" sections—taking 10-15 grams of glycine daily is recommended.
Maybe even negative-value—disinformation.
Pardon the nitpick, but in Benquo's footnote 5 we read:
As can be seen from the table below—it's 12g/day of a 14.5g total.
Free-glycine equivalent per gram of collagen
5.1 grams/day vs 2.4
Accounting for 10.75 of the 12 gram total
An error by Melendez-Hevia et al. The cited study gives 56.3%
n = 6
5.1 grams/day vs 0.32
5.65 g/day vs 0.84
Heinemeier et al. 2016
Tritiated—containing tritium.
Hydroxyproline is used (approximately) only in collagen.
For fractional synthesis %/day: 100/24 = 4.1(6), which the authors rounded down to 4.16
All other sources include unhydroxylated proline incorporated into collagen, proline in other proteins and metabolites of proline.
I'm guessing this is the part they wanted to cite
0.384%/day
Intravenous feeding, 4 hours in length. Labeled proline was infused over 3 minutes at hour 2.
Using the 0.066%/h value.
The aforementioned synthesis marker
15 grams of:
a) gelatin
b) hydrolyzed collagen
c) a 50/50 mix of both
Applying the 0% recycling "correction" to that figure, we get a comparable 3.78 g/day requirement.
7-26 months of age.
0.127
In the "SUBJECTS AND ANALYTICAL METHODS" section, the authors describe those lower protein groups as consuming "no more than" 30g/day of protein, but later in the study they switch to writing that they consumed 25 grams daily.
It seems dubious that the difference is caused by a glycine deficiency specifically vs inadequate protein in general (or something else entirely).
In Schmidt et al. 2015 (each subgroup was n = 98), both vegetarians and vegans had higher plasma glycine concentrations than omnivores—105% and 116% respectively. They found substantially lower glycine consumption (73%, 84%), but as the authors themselves note, their method of estimating amino acid consumption may have been inadequate for veg*ns (a retrospective food questionnaire).
Do we have anything better?
Dietrich et al. 2022 (n = 72, split evenly) compared the amino acid intake of omnivores and vegans (not vegetarians) using a weighed food protocol across 3 days.
Glycine intake was 41.1 vs 45.6
The total protein intake was 1 vs 1.2
In this study vegans had higher glycine concentrations as well. 134% of that of omnivores.
The daily intake level estimated to meet the basic nutritional requirement of 50% of healthy individuals
Not cited by either Benquo or Melendez-Hevia et al.
These are total test-diet glycine intakes, not additional supplement doses.
Certainly a transcription error. The lowest-dose trial in the cited meta-analysis was at 28 g/day
Not mentioned in the review but the cited meta-analysis also found benefits for positive symptoms at p = 0.01 and total symptoms at p rounded to 0.00 (for non-clozapine antipsychotics).
Based on the Figure 1 from the study.
I did not independently verify this. I personally believe this is most likely true and the rest of this section is written accordingly.
P = 0.016
Assuming the authors listed every statistically significant effect they could find, which I find likely.
Scott Alexander makes a similar argument in the context of Covid meds:
(bolding mine)
For a person on a non-atrocious diet.