Effect of broccoli sprout extract and baseline gut microbiota on fasting blood glucose in prediabetes: a randomized, placebo-controlled trial

Abstract

More effective treatments are needed for impaired fasting glucose or glucose intolerance, known as prediabetes. Sulforaphane is an isothiocyanate that reduces hepatic gluconeogenesis in individuals with type 2 diabetes and is well tolerated when provided as a broccoli sprout extract (BSE). Here we report a randomized, double-blind, placebo-controlled trial in which drug-naive individuals with prediabetes were treated with BSE (n = 35) or placebo (n = 39) once daily for 12 weeks. The primary outcome was a 0.3 mmol l^-1 reduction in fasting blood glucose compared with placebo from baseline to week 12. Gastro-intestinal side effects but no severe adverse events were observed in response to treatment. BSE did not meet the prespecified primary outcome, and the overall effect in individuals with prediabetes was a 0.2 mmol l^-1 reduction in fasting blood glucose (95% confidence interval -0.44 to -0.01; P = 0.04). Exploratory analyses to identify subgroups revealed that individuals with mild obesity, low insulin resistance and reduced insulin secretion had a pronounced response (0.4 mmol l^-1 reduction) and were consequently referred to as responders. Gut microbiota analysis further revealed an association between baseline gut microbiota and pathophysiology and that responders had a different gut microbiota composition. Genomic analyses confirmed that responders had a higher abundance of a Bacteroides-encoded transcriptional regulator required for the conversion of the inactive precursor to bioactive sulforaphane. The abundance of this gene operon correlated with sulforaphane serum concentration. These findings suggest a combined influence of host pathophysiology and gut microbiota on metabolic treatment response, and exploratory analyses need to be confirmed in future trials. ClinicalTrials.gov registration: NCT03763240 .

Fig. 1. Study profile as a CONSORT diagram.

Number of individuals randomized and assigned to BSE and placebo, respectively. In addition to the reasons provided for study discontinuation, a full list of reported adverse events for all participants is presented in Table 2. Source data

Fig. 2. Participant distribution and gut microbiota composition in clusters.

a, Distribution of study participants in pathophysiological clusters according to the clustering methodology in ref. (n = 89). b, Cluster distribution of participants in the replication cohort (n = 164). c, PCo analysis of Bray–Curtis dissimilarities at the species level in the three clusters of study participants (P = 0.02; n = 67 samples from 15 participants with SIRD-like characteristics, 11 with MOD-like characteristics and 41 with MARD-like characteristics). Standard error bars for the principal coordinates are denoted. d,e, Significantly altered taxa (at adjusted P < 0.05) in the microbiota between participants with MARD-like and MOD-like characteristics (d) and between participants with MARD-like and SIRD-like characteristics (e). The x axis denotes log(fold change) of the taxa in the cluster with MARD-like characteristics compared with the clusters with MOD-like and SIRD-like characteristics, respectively, as indicated. Clostridium spiroforme is in the process of getting renamed and therefore within brackets. The legend in c also applies to d and e. Source data

Fig. 3. Gut microbiota profiles of participants with different responses to BSE.

a, PCo analysis of Bray–Curtis dissimilarities of gut microbiota at the species level at baseline between participants who showed a pronounced response (defined as a reduction of fasting blood glucose of at least 0.3 mmol l⁻¹; n = 13) and a less pronounced response (n = 22) to BSE treatment (P = 0.001). Standard error bars for the principal coordinates are denoted. b, PCo analysis of species‐level Bray–Curtis dissimilarities at baseline and post-treatment in participants with a pronounced (n = 13) and less pronounced (n = 22) response to BSE (no significant differences between baseline and post-treatment; standard error bars are denoted). c, Significantly altered taxa at baseline (at adjusted P < 0.05) between participants with a pronounced (blue) and less pronounced (yellow) response to BSE. The x axis denotes log(fold change) of the taxa. Source data

Extended Data Fig. 1. Distribution of primary and secondary variables in all participants.

Box plots show individual data points, medians (straight lines inside box) and means (cross marking) with hinges representing lower and upper quartile. Data are from baseline and post-treatment in the placebo (n = 39) and BSE (n = 35) groups. Individuals with similar values are represented by the same circle. Source data

Extended Data Fig. 2. Distribution of cholesterol, triglycerides, insulin and C-peptide in all participants.

Box plots show individual data points, medians (straight lines inside box) and means (cross marking) with hinges representing lower and upper quartile. Data are from baseline and post-treatment in the placebo (n = 39) and BSE (n = 35) groups. Individuals with similar values are represented by the same circle. Source data

Extended Data Fig. 3. Distribution of primary and secondary variables in MARD-like participants.

Box plots show individual data points, medians (straight lines inside box) and means (cross marking) with hinges representing lower and upper quartile. Data are from baseline and post-treatment in the placebo (n = 20) and BSE (n = 24) groups of MARD participants. Individuals with similar values are represented by the same circle. Source data

Extended Data Fig. 4. Distribution of cholesterol, triglycerides, insulin and C-peptide in MARD-like participants.

Box plots show individual data points, medians (straight lines inside box) and means (cross marking) with hinges representing lower and upper quartile. Data are from baseline and post-treatment in the placebo (n = 20) and BSE (n = 24) groups of MARD participants. Individuals with similar values are represented by the same circle. Source data

Extended Data Fig. 5. Dissimilarities between BSE and placebo.

Principal coordinates analysis of species-level Bray-Curtis dissimilarities at baseline and post-treatment between study participants randomized to BSE and placebo, respectively. No significant differences were observed. Source data

Extended Data Fig. 6. Relative importance of clinical and microbiota features in predicting the response to BSE.

a, The relative importance of baseline variables (features) in predicting the change in fasting glucose after treatment with BSE. The measures of relative feature importance sum up to 1 and were generated by XGBoost. All participants receiving BSE were included in these analyses to investigate the importance of continuous baseline variables across all individuals. b, SHAP summary plot for the impact of baseline variables in predicting the change in fasting glucose in response to BSE treatment. Each point in the figure corresponds to a participant, and the SHAP value reflects the impact of the baseline variable in predicting the change in fasting glucose for that individual. For example, in the row corresponding to alkaline phosphatase, all individuals receiving BSE are plotted in accordance with how much baseline alkaline phosphatase predicts the change in fasting glucose. The panel is arranged based on the mean of the absolute SHAP values for each baseline variable. c, The impact of baseline clinical and microbiota variables on the glycemic response to BSE treatment using a distance‐based redundancy analysis model, calculated by the capscale function in R (the vegan package with the significance level based on the anova.cca function and PERMANOVA test with 10,000 permutations; P = 0.1). BMI is body mass index, blood glucose denotes fasting blood glucose, ALT is alkaline phosphatase, GGT is gamma‐glutamyl transferase, and HOMA‐IR is the homeostasis model assessment‐2 estimate of insulin resistance. Source data