Sulforaphane
Sulforaphane, an isothiocyanate released from broccoli glucoraphanin
Written by Aaron CuhaReviewed Sep 2026
Also known as: SFN, Broccoli sprout extract, Glucoraphanin (precursor), Sulphoraphane
Three randomised trials with prespecified primary endpoints, three misses: PSA slope in 78 men after prostatectomy, fasting glucose in 74 adults with prediabetes, and the autism replication in 45 children. The cleanest positive human result is a urinary detoxification biomarker, a 63.2 percent rise in benzene mercapturic acid excretion in Qidong, China.
Overview
Sulforaphane has more genuinely randomised human data behind it than almost any other supplement-aisle compound, which makes it a useful test case, because the pattern in that data is consistent and it is not the pattern the marketing describes.
The trials that measured a biomarker got results. A 170-person randomised trial of broccoli sprout beverage in Qidong, China found that the highest dose raised urinary excretion of benzene mercapturic acids by 63.2 percent over ten days, with the half and one fifth doses not significantly different from placebo. That is a real, dose-dependent, placebo-controlled human finding about the detoxification of an airborne carcinogen.
The trials that measured something closer to an outcome did not. In 78 men with rising PSA after radical prostatectomy, 60 mg of stabilised free sulforaphane daily for six months did not reach the primary endpoint, though PSA doubling time was 28.9 months on sulforaphane against 15.5 on placebo as a secondary result. In 74 drug-naive adults with prediabetes, twelve weeks of broccoli sprout extract did not meet its prespecified 0.3 mmol/L fasting glucose target, landing at 0.2 mmol/L. And the 2014 autism trial that produced the compound's best headline, a 34 percent improvement on the Aberrant Behavior Checklist in 29 young men, was followed by a 2021 trial in 45 children in which the primary outcome measure showed no significant difference at either seven or fifteen weeks.
The other thing worth knowing before buying is that the label and the dose are not the same thing. Sulforaphane itself is unstable, so most products sell glucoraphanin, its inert precursor, which needs the enzyme myrosinase to convert. When sprouts or seeds with their own active myrosinase intact were given directly, sulforaphane was three to four fold more bioavailable than the same molar dose of glucoraphanin delivered without active plant enzyme.
Mechanism of action
Sulforaphane modifies cysteine residues on KEAP1, which releases the transcription factor NRF2 to enter the nucleus and drive a large battery of phase 2 detoxification and antioxidant genes, including glutathione S-transferases and NAD(P)H quinone dehydrogenase 1. This is why its most reproducible human effect is on the conjugation and excretion of electrophilic toxicants rather than on any disease process: the pathway it activates is literally a disposal system. Separate work identified sulforaphane by matching a type 2 diabetes liver gene expression signature against a library of 3,800 drug signatures, and found it suppressed hepatic glucose production through NRF2 nuclear translocation with reduced expression of gluconeogenic enzymes. In the plant it does not exist. Broccoli stores the stable glucosinolate glucoraphanin in one compartment and the enzyme myrosinase in another, and sulforaphane is generated when the tissue is damaged, which is the relevance of chewing, chopping and of whether a capsule contains active enzyme.
Human evidence
Unusually large for a supplement: multiple randomised placebo-controlled trials across detoxification, glycaemia, prostate cancer recurrence and autism, several publicly funded. The consistent pattern is that biomarker endpoints move and prespecified primary endpoints do not.
- Qidong, China, 170 adults, 10 days: urinary benzene mercapturic acids up 63.2 percent on the full dose, not significantly different from placebo at half and one fifth doses.
- Prostate biochemical recurrence, 78 men, 6 months at 60 mg daily: primary endpoint not reached. PSA doubling time 28.9 against 15.5 months as a secondary outcome.
- Prediabetes, 74 adults, 12 weeks: prespecified 0.3 mmol/L fasting glucose target not met, overall effect 0.2 mmol/L.
- Autism, 44 young men, 18 weeks: 34 percent improvement on the Aberrant Behavior Checklist against under 3.3 percent on placebo, with scores returning toward baseline after discontinuation.
- Autism replication, 45 children analysed, 15 weeks: primary outcome not significant; a secondary caregiver-rated scale did improve.
- Bioavailability: sprouts and seeds with active myrosinase delivered 3 to 4 fold more sulforaphane than an equimolar glucoraphanin dose without it.
- No human trial has measured lifespan, mortality, cancer incidence or any ageing endpoint.
What this does not tell you: Every positive result above is a surrogate. Raised urinary excretion of a benzene conjugate is evidence that a disposal pathway was switched on, not evidence that anybody got less cancer, and the authors describe it as a strategy that portends reduced risk rather than one shown to reduce risk. The two autism trials point in different directions on their primary measures, and the earlier one carries a patent and licensing conflict of interest that the paper discloses. The prostate result is a doubling time, which is a rate of change in a marker, in men whose disease had already recurred.
Reading the research record
A surrogate endpoint is a measurement that stands in for the thing you care about, on the assumption that moving it moves the outcome. That assumption has failed often enough to be treated as a claim rather than a given. Homocysteine is the canonical case: it predicts cardiovascular disease strongly in observational data, it can be lowered reliably by several cheap interventions, and when trials lowered it and then counted heart attacks, the events did not fall. The lesson is not that markers are worthless. It is that a trial reporting a marker has answered a smaller question than the one the buyer is asking, and the page should say which question was answered.
Sulforaphane is the best-funded compound in this batch and the one where the primary endpoint record is easiest to read, because it keeps being written down. Three prespecified primary endpoints, three misses, and in all three cases the secondary and exploratory analyses gave something quotable. That is not fraud and it is not even unusual; it is what happens when a compound has a real but modest effect and trials are sized on optimistic assumptions. It does mean that anybody citing the PSA doubling time, the responder subgroup in prediabetes, or the Aberrant Behavior Checklist improvement in children is citing a secondary result from a trial that failed its primary one, and almost nobody says so.
The bioavailability problem is separate and more practical. What you buy is nearly always glucoraphanin, and how much sulforaphane it delivers depends on whether the product supplies active myrosinase and on your own gut flora. Two products with identical glucoraphanin labels can deliver different doses, which means a self-experiment on this compound has a dose you cannot read off the bottle.
The evidence, charted
Fig. 1 · evidence composition
7of 7 citations (100%) are in people
Every citation cited here is Human work; no other study type is cited on this page. This count is of our own citation list and understates any literature larger than the sources we cite.
Fig. 2 · evidence over time
Evidence spans 6 distinct years, 2014 to 2025, counted from the citation list on this page.
Fig. 3 · legal status at a glance
US
Not approved
UK
Not approved
AU
Not approved
CA
Not approved
Not approved in any of the four jurisdictions shown. A jurisdiction's classification is a regulatory fact, not a verdict on the science; see Legal status below for the full text and any notes.
Key studies & citations
- Human2019
Dose-dependent detoxication of the airborne pollutant benzene in a randomized trial of broccoli sprout beverage in Qidong, China.
170 adults randomised to placebo (n = 55) or one of three concentrations of broccoli sprout beverage for 10 consecutive days. Urinary benzene mercapturic acids rose significantly only in the high dose group, by 63.2 percent. The one half dose (+11.3 percent) and one fifth dose (-6.4 percent) were not significantly different from placebo. Median 24 hour urinary sulforaphane metabolite output was 24.6, 10.3 and 4.3 micromol across the three doses. The endpoint is a urinary detoxification biomarker, not a cancer rate. NIH funded, registered as NCT02656420.
American Journal of Clinical Nutrition - Human2015
Effect of Sulforaphane in Men with Biochemical Recurrence after Radical Prostatectomy.
78 men, mean age 69, with rising PSA after radical prostatectomy, randomised double blind to 60 mg free sulforaphane daily for 6 months. The paper states plainly that the primary endpoint was not reached. Secondary results favoured sulforaphane: PSA doubling time 28.9 months against 15.5 on placebo, and PSA rises above 20 percent at month 6 in 44.4 percent against 71.8 percent. PSA slopes were the same in both arms once treatment stopped.
Cancer Prevention Research - Human2025
Effect of broccoli sprout extract and baseline gut microbiota on fasting blood glucose in prediabetes: a randomized, placebo-controlled trial.
74 drug-naive adults with prediabetes randomised to broccoli sprout extract (n = 35) or placebo (n = 39) for 12 weeks. The authors report that the extract did not meet the prespecified primary outcome of a 0.3 mmol/L fall in fasting blood glucose; the overall effect was 0.2 mmol/L (95 percent CI -0.44 to -0.01, p = 0.04). Gastrointestinal side effects, no severe adverse events. A responder subgroup was identified in exploratory analysis, which is hypothesis generating rather than confirmatory.
Nature Microbiology - Human2014
Sulforaphane treatment of autism spectrum disorder (ASD).
44 young men aged 13 to 27 with moderate to severe autism, 29 on sulforaphane at 50 to 150 micromol daily and 15 on placebo for 18 weeks. Aberrant Behavior Checklist scores improved 34 percent against under 3.3 percent on placebo (p < 0.001) and Social Responsiveness Scale 17 percent (p = 0.017); scores rose back toward baseline after stopping. The declared conflict of interest states that Johns Hopkins holds patent applications and has licensed glucosinolate-rich sprouts and seeds to a company whose chief executive is the son of one of the authors.
Proceedings of the National Academy of Sciences - Human2021
Randomized controlled trial of sulforaphane and metabolite discovery in children with Autism Spectrum Disorder.
The attempted replication in children, 57 randomised and 45 analysed, aged 3 to 12. Treatment effects on the primary outcome, the Ohio Autism Clinical Impressions Scale, were not significant at 7 or 15 weeks (Cohen's d 0.21 and 0.10, both confidence intervals crossing zero). A secondary caregiver-rated measure, the Aberrant Behavior Checklist, did improve significantly at 15 weeks. The authors state that clinical effects were less notable in children than in their earlier trial in young men. US Department of Defense funded.
Molecular Autism - Human2015
Sulforaphane Bioavailability from Glucoraphanin-Rich Broccoli: Control by Active Endogenous Myrosinase.
Human volunteers given broccoli sprouts or seeds with endogenous myrosinase still active produced 3 to 4 fold more bioavailable sulforaphane than the same molar dose of glucoraphanin delivered without active plant enzyme. A commercial glucoraphanin supplement gave urinary sulforaphane metabolite output equivalent to a boiled and lyophilised sprout extract. This is the reason a label dose of glucoraphanin does not tell you the delivered sulforaphane dose.
PLoS One - Human2017
Sulforaphane reduces hepatic glucose production and improves glucose control in patients with type 2 diabetes.
A multi-part paper. Sulforaphane was identified computationally by matching a type 2 diabetes liver disease signature against 3,800 drug signatures; it suppressed glucose production in hepatic cells via NRF2 nuclear translocation, attenuated glucose intolerance in diabetic animals by a magnitude similar to metformin, and in the final human component, given as concentrated broccoli sprout extract, reduced fasting blood glucose and HbA1c in obese patients with dysregulated type 2 diabetes. The human arm is the smallest part of the paper and is the part that the later prediabetes trial did not confirm on its primary endpoint.
Science Translational Medicine
Frequently asked questions
Does sulforaphane prevent cancer?
No trial has measured cancer incidence. What has been measured, in a randomised trial of 170 people in Qidong, China, is that a broccoli sprout beverage increased urinary excretion of benzene conjugates by 63.2 percent at its highest dose. That is evidence that a detoxification pathway was activated. The step from there to fewer cancers has not been tested in humans.
Should I take broccoli sprout extract or eat broccoli sprouts?
On the bioavailability data, sprouts and seeds with their own myrosinase enzyme intact delivered three to four times more sulforaphane than the same molar dose of glucoraphanin without active enzyme. If you buy an extract, whether it supplies active myrosinase is the variable that decides your actual dose, and most labels state glucoraphanin rather than delivered sulforaphane.
Did it work for autism?
The 2014 trial in 29 young men showed a 34 percent improvement on the Aberrant Behavior Checklist against under 3.3 percent on placebo. The 2021 trial in children, by overlapping authors, did not show a significant effect on its primary outcome measure at 7 or 15 weeks, though a caregiver-rated secondary scale improved. The honest summary is one positive trial in young men and a failed primary endpoint on replication in children.
Does it lower blood sugar?
In people with established type 2 diabetes, a concentrated extract reduced fasting glucose and HbA1c in the human arm of a 2017 paper. In 74 people with prediabetes, a 2025 randomised trial did not meet its prespecified 0.3 mmol/L fasting glucose endpoint, coming in at 0.2 mmol/L. It is not a substitute for a diabetes medication and has never been compared head to head with one in people.
Is it safe?
Tolerability in the trials was good, with gastrointestinal side effects the most common complaint and no severe adverse events reported in the prediabetes trial. The longest randomised exposures here are six to eighteen months, so nothing is known about years of daily use.