The MSG Evidence Tracker
Ask the internet whether MSG is safe and you will get certainty in both directions. We wanted the receipts instead. So we pulled every record PubMed holds on MSG and human symptoms, read the abstracts, and sorted the studies by how they were run. What falls out is not a tie. The result tracks the method: the better the study design, the quieter the symptoms.
By Travis Leleu. Published October 24, 2026; revised October 28, 2026. This page will be updated as the underlying review continues.
What we pulled
Nine PubMed searches (MSG crossed with headache, "Chinese restaurant syndrome," symptom complex, adverse effects, asthma, urticaria, and metabolic outcomes, plus sweeps for challenge trials and reviews) returned 538 hits. Deduplicated, that is 341 unique records. Screening for dietary MSG in humans left 236 records that passed the first screening, of which 69 carry our dataset's own "peripheral" tag, leaving 167 in the screened core: 55 records classified as challenge or trial studies in the first pass (see the verified core below), 62 reviews including 12 systematic reviews or meta-analyses, 16 case reports, 10 dietary-intake epidemiology studies, and the rest letters, mechanistic studies, and commentaries, spanning 1969 to 2026. Full-text reading showed that at least 6 of the 55 first-pass records are not dietary MSG challenges at all, and 1 more is unusable as challenge evidence, leaving a verified core of 48 dietary MSG challenge records. The seven set aside from the challenge count are two reviews (Tarlo 1993; Beausoleil 2007), the Sinitkul 2018 transplant cohort, two Kumar 2015 glutamate-injection studies, the Pacor 2004 additive panel, and Kim 2015, which has no MSG-free arm. The 105 records screened out, 80 with no dietary-MSG mention (mostly endogenous-glutamate migraine neurochemistry), 22 animal studies, and 3 in vitro studies, are flagged in our data rather than deleted.
Funding disclosure, first
Read this before the pattern below. The International Glutamate Technical Committee (IGTC) is the glutamate manufacturers' body. The IGTC funded Geha 2000 (footnote: "Supported by a grant from the International Glutamate Technical Committee"), Woessner 1999 (IGTC plus the NIH-funded General Clinical Research Center at Scripps Clinic), Schwartzstein 1987 (per the Cochrane review's risk-of-bias assessment), and Stevenson 2000 (IGTC grants). Obayashi and Nagamura 2016 was authored from the IGTC itself: the PubMed record lists the IGTC as the lead author's affiliation with an Ajinomoto email address, and the IGTC financed the publication fee. Yang 1997 was IGTC-supported, with IGTC-supplied MSG. The Brosnan 2014 review sat in an IGTC-sponsored symposium. Two papers need exact wording: Simon 2000 was published in the IGTC-financed symposium supplement, and a paper-specific support line is unconfirmed, so we do not assert one; and we make no funding claim about Woods 1998, for which no verified funding statement was found in the sources checked.
Stevenson, who wrote the methodological review used to dismiss the positive asthma literature, also co-authored the Woessner null study and Rajan 2014. He is a participant in this literature, not a neutral referee standing outside it.
A funding map over all 341 papers in the corpus found that only 90 of 341 papers have any funding information at all, and only 22 have a documented funder or conflict tie (9 of them industry-linked). The other 251 papers (74%) are dark, holding 60% of all citations and 75.9% of PageRank mass. That last figure carries an age confound, stated plainly: PageRank tilts old, the top-50 papers by PageRank have a median year of 2000, and 36 of the top 50 are funding-unknown, so age drives both influence and non-disclosure. The record's finding: in this corpus the funder does not predict the findings. Yang 1997, the strongest positive challenge study, and Geha 2000, the flagship null, are both IGTC-tied, and Fisher exact tests find no association, with the honest caveat that only 9 industry-linked papers are documented anywhere, so the tests are nearly powerless. At abstract level funding is nearly invisible: no on-topic abstract states any funding; the single apparent value was a mis-extracted case-history sentence, now cleared.
The pattern, in four steps
- Unblinded reports, anecdotes, and single-blind challenges periodically report symptoms from MSG.
- Double-blind, placebo-controlled challenges mostly fail to reproduce those symptoms, including in people who believe they are MSG-sensitive.
- The partial exception is a narrow, artificial protocol: large doses, roughly 2.5 to 5 grams and up, at or above about 50 mg per kilogram of body weight, given without food, usually in solution or capsules. Under those conditions some well-blinded studies record more short-lived symptoms (headache, flushing, sensitivity around the skull) than placebo. Even there, the largest study's substantial initial response collapsed on retesting, and the responses were not reproducible with food, in two subjects.
- For calibration: a typical serving of food with added MSG contains under 0.5 grams (FDA figures), and average added-MSG intake is about 0.55 grams a day, against roughly 13 grams a day of glutamate arriving from ordinary protein foods.
In our dataset's design field for the 48-record verified core, 28 records state a double-blind design and 31 mention a placebo; 15 carry no blinding label in that field. A cautious read of abstract wording (not an effect-size analysis, see Limits below) applied to the verified core counts 13 finding no MSG effect, 7 reporting an effect, 4 mixed, and 24 that cannot be classified from the abstract alone, with Yang 1997 counted as effect-finding after a correction to our dataset. Treat that as a wording count only: the same heuristic inverted Yang before the correction, so it is not a vote count and not evidence of effect size.
Key studies
| Study | Design | What happened |
|---|---|---|
| Geha et al., 2000, J. Allergy and Clinical Immunology (companion review: J. Nutrition, 2000) | Multicenter double-blind, placebo-controlled challenge, 130 people who identified themselves as MSG-sensitive | In Protocol A, 50 of 130 (38.5%) responded to MSG only, against 17 (13.1%) to placebo only, a substantial initial response rate that collapsed on retesting. Responses were inconsistent between sessions and not reproducible on retesting. In Protocol D, with food, both subjects (n = 2) responded to 1 of 3 MSG challenges, each time with different symptoms: not reproducible with food, in two subjects. |
| Yang et al., 1997, J. Allergy and Clinical Immunology | Double-blind, placebo-controlled challenge in a citrus drink, 61 self-identified sensitive subjects entered, 36 underwent dose rechallenge | 5 g MSG in a 200 mL strongly citrus-tasting beverage, not capsules. Both halves in one paragraph: the initial symptom-incidence comparison was null (MSG-only 36.1% vs placebo-only 24.6%, p = 0.324), while the positive reading rests on severity scores and the rechallenge phase, with an apparent 2.5 g threshold. The strongest single result on that side of the ledger, and part of why the honest summary is "not reproduced consistently" rather than "never observed." |
| Simon, 2000 | 65 chronic urticaria patients, 2.5 g challenge | 2 patients reacted in single-blind challenge. Zero of them reacted in double-blind challenge. Conclusion: MSG is at most an unusual (under 3%) trigger of chronic hives. |
| Woessner et al., 1999 | Single-blind screening challenge, 100 asthma patients | Zero confirmed cases of MSG-provoked asthma. Double-blind rechallenge was reserved for positive screens; there were none, so no double-blind MSG challenge occurred in this study. A null from a single-blind study is conservative, since any bias would favor reactions. Fits the wider asthma record: early positive reports were single-blind, and rigorous challenges (Schwartzstein 1987, n = 12; Woods 1998, n = 12) found nothing. |
| Aarhus group, 2010 (Baad-Hansen et al.) and 2013 (Shimada et al.) | Double-blind crossover experiments, 75 to 150 mg/kg in soda, 14 healthy volunteers in each study | High-dose MSG in drink produced more headache than placebo, under conditions far from a meal. Keep the two studies distinct: the 2010 study found subjectively reported tenderness around the skull without robust measured change, while the 2013 study found reduced pressure-pain thresholds, with headache in 8 of 14 subjects on MSG against 2 of 14 on placebo. "Fasted" is not stated in these abstracts and is not claimed here. |
| Obayashi and Nagamura, 2016, J. Headache and Pain | Systematic review of the human headache literature | A causal relationship between MSG and headache is unproven, citing absent or improper blinding and inconsistent findings across the challenge literature. The review's own count is 4 of 7 without-food studies significant, and that without-food subset is internally inconsistent (see Headache below). The review was authored from the IGTC itself; see Funding disclosure above. |
By condition
The symptom complex once called "Chinese restaurant syndrome." Reviews from Kenney (1986) through Freeman (2006) conclude that trials have failed to identify a consistent relationship between MSG and the symptom constellation. The Geha multicenter study above is the defining test: a substantial initial response that collapsed on retesting, and responses not reproducible with food, in two subjects.
Asthma. A repeating pattern of early alarm and later calm. Early positive reports were single-blind, and Stevenson's methodological critique of them is attributed to him as a participant in the literature (see Funding disclosure). Woessner 1999 was single-blind screening, not double-blind. The Cochrane review (2012) found only 24 subjects across all eligible randomized trials and no significant difference, concluding there is no evidence to support MSG avoidance in chronic asthma, while noting the evidence base is thin.
Hives and swelling. The same two-step repeats: a few positive single-blind challenges that do not survive double-blind rechallenge (Simon 2000 above; Rajan 2014, 100 patients, found additive sensitivity in under 1%).
Headache. The genuinely mixed area, and this page will not pretend otherwise. The Aarhus experiments show a real signal at high doses without food. The 2016 systematic review finds meal-level causality unproven, on its own count of 4 of 7 without-food studies significant, while its without-food subset is internally inconsistent: significant differences at 2.5 and 5.0 g per 200 mL in one study, none at 5.0 g per 200 mL in another, and none at 5, 8, or 12 g per 100 mL in Rosenblum's 99 men. Both sentences are true at once.
Weight and metabolism. No challenge trials exist here; the record is observational and split. Two analyses of Chinese cohorts found positive associations between MSG intake and weight gain; studies in Jiangsu and Vietnam found none, and one found higher MSG intake associated with lower rates of high blood sugar. MSG intake tracks total diet, sodium, and eating out, so confounding is heavy. Verdict: unresolved, and weaker evidence than either side's headlines suggest. The Brosnan 2014 review arguing against the obesity hypothesis sat in an IGTC-sponsored symposium (see Funding disclosure).
What the regulators concluded
| Body | Position |
|---|---|
| FDA (United States) | MSG is "generally recognized as safe" (GRAS). FDA states that scientists have not been able to consistently trigger reactions in studies of self-identified sensitive people, and that it has never confirmed MSG as the cause of reported adverse events. |
| FASEB review for the FDA (1995) | MSG is safe. Possible short-term, mild symptoms (headache, numbness, flushing, tingling, palpitations, drowsiness) in some sensitive individuals at 3 grams or more taken without food, an exposure FDA notes is unlikely at one sitting when a typical serving holds under 0.5 grams. |
| JECFA, the FAO/WHO expert committee (1988) | Assigned glutamates an ADI of "not specified," its lowest-concern category, finding that total dietary intake at normal use levels does not represent a hazard. The EU's Scientific Committee for Food reached a similar position in 1991. |
| EFSA (European Union, 2017) | The most conservative assessment. Set a group acceptable daily intake of 30 mg per kilogram of body weight per day, derived from the highest animal dose showing no adverse effects, and noted that high consumers, especially young children, may exceed it. Recommended the EU review permitted levels. |
The two positions, stated plainly and without a verdict on motive: FDA and JECFA impose no numerical limit. EFSA derived a 30 mg per kilogram per day ADI from the highest animal no-effect dose with uncertainty factors and found high consumers may exceed it. Both positions are presented here as stated by their authors.
Limits of this tracker
- Abstract-only, in part. Most fields in our dataset come from the PubMed abstract and metadata. At abstract level, within the 48-record verified core, only 16 abstracts state the dose, 12 state the vehicle (food, capsule, or drink), and 19 state the sample size. A full-text pass covered 67 target records, and 30 of those carry full-text extractions. The rest of the verified core still rests on abstracts, and everything else is marked "needs full text" rather than guessed.
- The direction labels are a wording heuristic, not a meta-analysis. No effect sizes, confidence intervals, or risk-of-bias scores were extracted. The heuristic inverted Yang 1997 before correction, which is why no tally computed on the 55-record first pass is published here. The verified-core wording count above carries the same caveat.
- Funding at abstract level is unusable. No on-topic abstract states any funding; the single apparent value was a mis-extracted case-history sentence, now cleared. Funding facts on this page come from full texts, the Cochrane risk-of-bias assessment, and the funding map, as cited in the Funding disclosure section above.
- Diet trials need narrower claims. Brandley 2022: the blinded MSG-vs-placebo difference appeared only in the severe-anxiety subgroup, not in the minimal/mild or moderate groups. Fiebel 2023: the blood-pressure improvement belongs to the open diet phase; its blinded challenge found no acute MSG effect on blood pressure, so citing it the other way would invert its result.
- Genuinely mixed areas stay mixed: headache (a real high-dose experimental signal versus unproven meal-level causality) and weight (a split observational record with heavy confounding).
- Coverage. PubMed only, with an English-abstract bias, and 40 on-topic records (mostly pre-1990) have no abstract at all, so their rows rest on title and metadata. Regulatory dossiers, including the full FASEB report, sit outside this pull.
The short version lives on our facts about MSG page. This tracker is the long version, kept honest on purpose: a safety case that hides the other side's best studies is not worth much, and this record does not need the help.
