Sulforaphane and Autism: Human Trials, Mixed Outcomes, and the 2025 Meta-Analysis

Bottom line: Sulforaphane has a real randomized-trial program in autism spectrum disorder, and a 2025 meta-analysis found small pooled benefits across several symptom domains. Individual trials are notably mixed, including null primary outcomes, so the evidence is more accurately described as emerging and inconsistent than established treatment efficacy.

The 2014 trial created the excitement

A placebo-controlled double-blind trial in adolescent and young adult males with moderate-to-severe autism reported substantial improvements on several behavioral scales after 18 weeks of sulforaphane-rich treatment.[1]

Scores moved back toward baseline after treatment stopped.

That pattern was biologically and clinically interesting and triggered substantial follow-up research.

The later pediatric trial was less impressive

A 2021 randomized trial in children aged 3 to 12 used a primary clinician-rated autism measure.[2]

The primary outcome did not significantly differ between sulforaphane and placebo.

Some caregiver-rated secondary outcomes improved, while other measures did not.

That trial is important precisely because it did not simply reproduce the original headline result.

The larger multicenter study was mixed

A later multicenter randomized study enrolled 108 children.[3]

Caregiver-rated scales did not significantly differ between groups, while clinician-rated measures showed improvement.

When rater type changes the apparent result, interpretation becomes more complicated.

It may reflect measurement sensitivity, expectancy, different observation settings, or chance.

The 2025 meta-analysis

A 2025 systematic review and meta-analysis included six randomized trials.[4]

The pooled estimates favored sulforaphane for total symptoms and several domains, including aberrant behavior and aspects of social interaction or communication.

The average effects were generally small to moderate rather than transformative.

The authors also emphasized uncertainty related to assessment tools, treatment duration, and study design.

Preparation and exposure vary

Sulforaphane research may use broccoli-sprout preparations, glucoraphanin plus myrosinase systems, or other formulations.

Actual sulforaphane exposure can differ substantially depending on formulation and conversion efficiency.

A "broccoli supplement" is therefore not automatically equivalent to the intervention used in a clinical trial.

What the evidence does not show

The current literature does not establish that sulforaphane replaces established behavioral, educational, communication, or medical support.

It also does not identify a universal responder profile.

Autism is heterogeneous, and trial averages can conceal large individual differences.

Safety

Across the randomized literature, sulforaphane has generally been tolerated reasonably well under study conditions.[2][3][4]

That does not automatically establish long-term safety at every dose or in every medical context.

What would make the mixed trial record easier to interpret?

Future trials would benefit from harmonized, prespecified primary outcomes and careful reporting of who rated each measure. When clinician-rated and caregiver-rated scales point in different directions, the disagreement is part of the result rather than noise to be hidden.

Preparation identity also matters because different glucoraphanin, myrosinase, and sulforaphane delivery systems can produce different exposure. Replication with well-characterized formulations would make cross-trial comparisons more meaningful.

Most importantly, a pooled average should not be translated into a claim about any individual person. The existing trials answer a population-level research question under defined study conditions; they do not establish a universal response or replace individualized support and care.

Verdict

Sulforaphane has emerging human evidence in autism, stronger than mechanism-only supplement claims but not consistent enough for a simple treatment verdict.

The most accurate summary is: multiple RCTs, mixed individual trial results, small pooled benefit in the 2025 synthesis, and a clear need for replication and formulation-specific evidence.