Sulforaphane and Sulforaphane-Nitrile Metabolism in Humans Following Broccoli Sprout Consumption: Inter-individual Variation, Association with Gut Microbiome Composition, and Differential Bioactivity

Abstract

Scope

The glucosinolate glucoraphanin from broccoli is converted to sulforaphane (SFN) or sulforaphane-nitrile (SFN-NIT) by plant enzymes or the gut microbiome. Human feeding studies typically observe high inter-individual variation in absorption and excretion of SFN, however, the source of this variation is not fully known. To address this, we conducted a human feeding trial to comprehensively evaluate inter-individual variation in the absorption and excretion of all known SFN metabolites in urine, plasma, and stool, and tested the hypothesis that gut microbiome composition influences inter-individual variation in total SFN excretion.

Methods and Results

Participants (n = 55) consumed a single serving of broccoli or alfalfa sprouts and plasma, stool, and total urine were collected over 72 hours for quantification of SFN metabolites and gut microbiome profiling using 16S gene sequencing. SFN-NIT excretion was markedly slower than SFN excretion (72h vs 24h). Members of genus Bifidobacterium, Dorea, and Ruminococcus torques were positively associated with SFN metabolite excretion while members of genus Alistipes and Blautia had a negative association.

Conclusion

This is the first report of SFN-NIT metabolite levels in human plasma, urine and stool following consumption of broccoli sprouts. Our results help explain factors driving inter-individual variation in SFN metabolism and are relevant for precision nutrition.

Introduction

Cruciferous vegetables are associated with numerous health benefits, most notably with cancer prevention. Cruciferous vegetables, such as broccoli, Brussels sprouts, kale, and cauliflower ubiquitously contain a class of phytochemicals known as glucosinolates ^[1,2]. Upon damage to the cell well, the enzyme myrosinase is released, cleaving the glucose from the glucosinolate, and yielding the primary hydrolysis products isothiocyanates (ITCs) and nitriles (NITs) ^[1,2]. Inter-individual variation in the amount of ITC’s absorbed by participants who consumed a controlled amount of glucosinolates has long been noted in clinical studies of cruciferous vegetables ^[3-6]. Differences in the amount and form of the glucosinolate hydrolysis products (ITCs vs NITs) a person receives may affect the efficacy of these dietary compounds in the prevention or suppression of chronic diseases ^[3-6]. Isothiocyanates and their metabolites are bioactive, activating a multitude of anticancer mechanisms. Isothiocyanates are known to upregulate cytoprotective and phase 2 drug metabolizing enzymes, like Heme Oxygenase 1 (HO-1), NADPH quinone dehydrogenase 1 (NQO1), and glutathione S transferase (GST), by activating Nuclear factor erythroid 2–related factor 2 (Nrf2)^[7-16]. Isothiocyanates also halt cell proliferation and induce cell cycle arrest in cancer cells and alter epigenetic regulation via histone deacetylase enzymes (HDACs) making them a powerful dietary cancer chemopreventive agent ^[17,18]. Like other electrophiles, isothiocyanates are metabolized via the mercapturic acid pathways to dithiocarbamate metabolites (DTCs) and excreted through the kidneys into the urine primarily conjugated to N-acetyl cysteine (NAC) or into the bile conjugated to glutathione (GSH) and NAC ^[19-21]. Glucosinolate-derived nitriles, on the other hand, are widely accepted to be biologically inactive but, the pharmacokinetics of SFN-NIT in vivo has not been fully explored despite the fact that it is a hydrolysis product of glucoraphanin and may be contributing to inter-individual variation in benefits received from broccoli sprout consumption ^[4,5,22-24]. Furthermore, while urinary excretion of SFN is widely accepted to be the primary route of excretion, studies examining biliary excretion of glucosinolate metabolites are more limited in humans (ileostomy patients and in vitro fecal culture models) but to our knowledge the abundance of SFN metabolites in human stool are still unknown^[4,19,25,26]. Thus, empirical evidence is needed to validate that biliary, and thus excretion through stool, is only a minor path of excretion in humans.

Glucoraphanin, the primary glucosinolate in broccoli sprouts, is one of the most heavily studied glucosinolates and yields the isothiocyanate sulforaphane (SFN) and the nitrile sulforaphane-nitrile (SFN-NIT). There is strong evidence supporting the efficacy of SFN for cancer prevention in cell culture and animal models, however, human studies present equivocal evidence possibly stemming from the high levels of inter-individual variation in absorption, metabolism and excretion of SFN that is observed in clinical trials ^[6,25,27-30]. Work involving glutathione-S-transferase polymorphisms has not fully explained differences in SFN metabolite production. Other studies have reported factors, such as BMI, explaining differences in SFN metabolite production ^[30-33]. Accumulating evidence points towards the gut microbiome as an important player in the hydrolysis of glucosinolates to ITCs and NITs, and as a potential driver of inter-individual variation in the absorption and excretion of SFN-metabolites ^{[25,31,32,34-36]}. Work in vitro with monocultures has shown that individual microbes preferentially yield isothiocyanates or nitriles from glucosinolates, however, the relevance of these microbes to the human gastrointestinal tract is unclear ^[35,37,38]. Work examining the impacts of cruciferous vegetables on gut microbiome composition has shown that consumption of cruciferous vegetables can alter the composition of the microbiome, but these studies do not measure isothiocyanate metabolites ^[34,39-42]. Our group has previously shown that microbiome composition not only influences the production of nitriles from glucosinolates, but also influences the metabolism of a broad range of phytochemicals from cruciferous vegetables in vitro^[25,43].

To investigate the relationship among diet, microbiome composition, and SFN and SFN-NIT metabolism and excretion in humans, we conducted a human feeding trial followed by measurement of SFN metabolites and 16S gene sequencing. While 16S sequencing only captures the composition of bacteria in the gut, we use the term microbiome to describe bacterial community composition in this study due to its historical use in the field. Our goals were to: 1) Understand the metabolism of SFN and SFN-NIT in human plasma, urine, and stool, 2) Describe alterations to the gut microbiome by broccoli consumption, and 3) identify relationships between dietary factors, microbiome composition, and SFN and SFN-NIT excretion. We hypothesize that 1) SFN-NIT metabolism will be markedly different than SFN metabolism and urine will be the primary route of excretion of SFN, 2) A single serving of broccoli will minimally alter microbiome composition, and 3) That specific microbes and dietary factors will be associated with SFN and SFN-NIT excretion. Additionally, we utilized cell culture methods to further evaluate the bioactivity of SFN-NIT for cancer prevention, specifically focusing on HDAC activity and gene induction through Nrf2 in prostate cancer cells. Through this work, we have begun to understand the link between microbiome composition and glucosinolate hydrolysis products to explain factors which drive inter-individual variation observed in human feeding trials, and have provided the foundations in pushing the field of nutrition further towards precision nutrition and medicine.

Experimental Section

Subjects

Fifty-five healthy women and men, 19-60 years old, were recruited in Corvallis, Oregon and signed an informed consent document. The study was conducted in the Linus Pauling Institute and the Moore Family Center metabolic kitchen at Oregon State University. Exclusion criteria included 1) tobacco use; 2) BMI <18.5 or >30.0 kg/m²; 3) pregnancy or breastfeeding; 4) use of oral antibiotic medication (within past 6 months); 5) extensive vigorous exercise (7+ hours per week); 6) use of medications to control cholesterol levels or fat absorption; 7) a history of significant acute or chronic illness and, bariatric surgery and, gastrointestinal procedures or disorders. Eligibility of subjects was confirmed. All study activities were approved by the Oregon State University Institutional Review Board (IRB #8343, NCT04641026).

Study Design

Subjects were randomized to two treatment groups, receiving either fresh broccoli sprouts or fresh alfalfa sprouts. The study was organized as 6 cohorts of subjects and took place between May and November 2021. Subjects consumed fresh broccoli sprouts (40.5 g on average) containing 100 μmol SFN equivalents. The alfalfa sprouts subjects consumed an equivalent amount to match the broccoli sprout treatment group. Long-term consumption of sprouts has been shown to improve the gut microbiome’s capability of convert glucosinolates to ITCs. To limit alterations to the gut microbiome and identify members of the gut microbiome which may contribute to inter-individual differences in SFN-metabolism, we opted to use a single serving of broccoli sprouts. Additionally, previous work has demonstrated that a single serving of broccoli is capable of altering the abundance of specific genera in the gut ^[25,43]. Lastly, as we are interested in the pharmacokinetics of SFN-NIT in humans, using a single dose allowed for the most straightforward and simple analysis. As previously published, subjects consumed sprouts with a standardized breakfast and fasted for at least 8 h prior to the meal ^[16]. One week before and throughout the sample collection period, subjects self-reported dietary intake and were instructed to avoid consuming foods, beverages, and supplements containing cruciferous vegetables, and live/active cultures, or probiotics. Diet records were analyzed using Food Processor^® SQL (EHSA, Salem, OR).

Preparation of Sprouts

The sprouts were hydroponically grown in the Moore Family Center metabolic kitchen in Corvallis, OR. Broccoli sprouting seed (Brassica oleracea var. italica) was from the Sprout House (Kingston, NY) and True Leaf Market (Salt Lake City, UT). Alfalfa sprouting seed (Medicago sativa) was from the Sprout House. For cultivation seed was sanitized with calcium hypochlorite (20,000 ppm, 15 minutes), rinsed, soaked overnight, and then rinsed twice daily for 5 additional days. Sprouts were harvested on day 6 and refrigerated until use. SFN contents in alfalfa or broccoli sprouts were analyzed on the day of harvest for every subject cohort. Sprouts samples were heated to 60°C in water for 10 mins to inhibit epithiospecifier protein activity ^[44], followed by homogenization and 1 h incubation in the presence of 2 mg/ml Sinapis alba thioglucosidase (Sigma-Aldrich) at 60°C. Following incubation, samples were centrifuged (16,000 x g, 5 min, 25°C), and supernatants were filtered using 0.22-μm nylon Spin-X filters (VWR, Radnor, PA). Amount of SFN present in sprouts was determined by mass spectrometry analyses (see section below).

Biospecimen Sample Collection

Baseline 0 h spot urine collections were obtained prior to sprout consumption. Complete urine collections were obtained following sprout consumption and collections were 0-3, 3-6, 6-24, 24-48, and 48,-72 h post consumption. To stabilize SFN metabolites urine jugs contained granulated boric acid and were kept cold ^[16,45]. Upon receipt, urine was acidified with TFA (10% v/v) and stored at −80°C. Whole blood (10 mL) was collected by venipuncture into lithium heparin vacutainers (Becton, Dickinson and Company, Franklin Lakes, NJ) at 0, 3, 6, 24, 48, and 72 h post consumption. Plasma fractions were collected and stored at −80°C. Stool samples were, collected by subjects into OMNIgene-gut kits (DNA Genotek, Ottawa, ON) and cryoELITE tissue vials (Wheaton, Millville NJ). Fecal samples were collected in the evening, or morning before sprouts were consumed (0 h) and then again before the 24, 48 and 72 h study visit. On receipt, cryoELITE tissue vials and OMNIgene-gut tubes were stored at −80°C.

Human Sulforaphane and Sulforaphane Metabolite Quantification

Plasma and urine samples were processed as previously described ^[16]. They were thawed and centrifuged (16,000 x g, 5 min, 4°C) to precipitate proteins and filtered using 0.22-μm Spin-X filters. Frozen fecal samples were partially thawed, small amounts of fecal materials (~150-200 mg wet weight) were transferred to 2 ml homogenization tubes (Precellys CKMix Tissue Homogenizing Kit, Cayman Chemicals). Fecal samples were acidified in 250 μl water containing 70% methanol/10% formic acid to stabilize SFN in the samples. Acidified samples were dried in a desiccator and weighed. Dried fecal materials (~100-150mg) were resuspended in 0.1% formic acid (w/v) at a ratio of 1:5 (w/v), and the mixture homogenized using a Precellys tissue homogenizer. Samples were centrifuged (16,000 x g, 10 min, 4°C), and supernatants were filtered using 0.22-μm Spin-X filters. All processed samples were diluted in 0.1% formic acid in water (v/v) for mass spectrometric analyses.

Detection of SFN and SFN metabolites (SFN-Cys, SFN-NAC, SFN-GSH, SFN-CG, SFN-NIT) has previously been published ^[25,46] with slight modifications. Briefly, LC-MS/MS was performed using a Shimadzu system (Shimadzu, Columbia, MD) coupled to a QTRAP 4000 mass spectrometer (Sciex, Framingham, MA) employing multiple reaction monitoring (MRM) transitions for metabolites detection at the OSU Mass Spectrometry Center. MRM analysis was conducted in the positive ionization mode for SFN and its metabolites. The following precursor and product ions were used to detect SFN and SFN metabolites: SFN (178 > 114), SFN-glutathione (SFN-GSH, 485 > 179), SFN-cysteine (SFN-Cys, 299 > 114), SFN N-acetyl-L-cysteine (SFN-NAC, 341 > 114), SFN-cysteinylglycine (SFN-CG, 356 > 114), and SFN-nitrile (SFN-NIT,146 > 98). Quantification was determined against known standards.

Pharmacokinetic Analysis

Noncompartmental pharmacokinetic analysis was conducted in R version 4.2.1 using the package PKNCA (v0.10.1), as previously described ^[16,47]. Briefly, analysis was conducted separately for SFN and its mercapturic acid pathway metabolites, and for SFN-NIT. AUC was calculated between 0 and 72 hours using the last measurable observation (AUC_last). T_1/2 was calculated using a minimum of 2 timepoints with T_max being included within those points.

In vitro Testing of SFN and SFN-Nitrile Effects on Gene Expression and HDAC Activity

Androgen-independent (PC-3) prostate cancer cells were obtained from American Type Tissue Collection (Manassas, VA) and cultured as previously described ^[48]. Cells were treated with SFN (15 μM) or SFN-nitrile (15, 45, 135 μM) (LKT Laboratories, St. Paul, MN) for 24h or 48h. Control cells were treated with DMSO matching the highest DMSO concentration present in treatment groups (0.27% v/v). 15 μM SFN has previously been shown to inhibit HDAC activities and induce phase II enzyme in PC3 cells, and this concentration is thought to be achievable from consuming 1 to 2 cups of broccoli sprouts ^[49]. Changes in NQO-1 and HO-1 gene expression were determined using quantitative real time PCR using primers specific for NQO-1 and HO-1, with GAPDH as housekeeping normalization control using reagents, software and materials previously described^[50]. HDAC activities were measured 48h post treatment using RIPA buffer, Pierce BCA protein assay kit, and HDAC fluorometric activity assay kit (Cayman Chemical).

Microbial Sequencing

DNA from fecal samples collected in OMNIgene-gut tubes were isolated using QIAamp PowerFecal Pro DNA kit (Qiagen). Fecal DNA concentrations were measured using Qubit dsDNA HS assay kit (Invitrogen, Waltham, MA, USA). PCR was used to amplify the 16S rRNA gene at the V4V5 region using the Earth Microbiome Project 16S Illumina Amplicon Protocol ^[51]. Barcoded amplicons were quantitated, pooled, and sequenced using an Illumina MiSeq instrument. 16S amplicon sequencing was performed at the Center for Quantitative Life Sciences core facilities (Oregon State University) using established methods ^[52]. This approach yielded 300 bp, paired end amplicon sequences at a target sequencing depth of 50,000 reads per sample (n = 317). Data preprocessing and identification of amplicon sequence variations (ASVs) was conducted using the DADA2 pipeline, as implemented in R (v4.1.0) ^[53]. Briefly, reads were filtered for expected errors (maxEE = 2) followed by a merging of paired reads and a removal of chimeric ASVs. Taxonomy was assigned using the Silva database v138.1 with the naïve Bayesian classifier built into DADA2 ^[53,54].

Microbiome Data Management and Quantification of ASVs

All statistical analysis was conducted in R version 4.2.1, unless otherwise noted. The Benjamini-Hochberg procedure was used for multiple testing correction and an adjusted p-value of 0.05 was used as the significance cutoff ^[55]. Previous work in Bacteroides thetaiotaomicron has shown strain-level differences in GLS hydrolyzing capabilities ^[56]. While detecting strain-level differences with 16S sequencing is unfeasible, to capture microbiome composition at the most specific level possible, we conducted our analysis at the ASV level as opposed to agglomerating our data to the genus or species level. To minimize noise in the dataset, sparse ASVs, which were those observed fewer than 3 times in less than 20% of the samples and with a mean relative abundance across all samples less than 0.001%, were filtered out and yielded a final dataset of 317 ASVs. Rarefaction curves using the vegan package (v2.6-2) in R were built on filtered data to ensure all samples were sufficiently sequenced (Figure S1) ^[57].

Diversity Analysis and Visualization

The R packages phyloseq (v1.4) and ggplot2 (v3.3.6) were used to visualize and calculate alpha-diversity metrics using unfiltered data rarefied to an evening sequencing depth ^[58,59]. Alpha diversity was assessed using observed ASVs, Shannon diversity index, and Simpson diversity index. Differences in alpha diversity were assessed using a generalized linear mixed model as implemented by lmerTest (v3.1-3) ^[60]. Grams of sprouts consumed was used to control for cohort effect. Beta diversity of filtered data was analyzed using Principal Coordinate Analysis (PCoA) and based on Jaccard distance ^[58]. Permutation analysis of variance (PERMANOVA) was conducted using the adonis2 function from the vegan package (v2.6-2) ^[57].

Differential Abundance Analysis

To identify genera which were differentially abundant between treatment groups, we utilized a generalized linear mixed-effect model, as implemented by the package lmerTest ^[60,61]. All samples were rarefied to an even depth and data was transformed using the center log ratio (clr) prior to fitting the model. The clr-transformed abundance of each ASV was used as the response variable with the treatment group, interaction, and their interaction as the predictor variable. Grams of sprouts consumed was used to control for a cohort effect. Participant effect was accounted for as a random effect. One model was built for each ASV. The Benjamini-Hochberg procedure was used to account for multiple tests ^[55].

Multiblock Sparse Partial Least Squares Analysis

To integrate microbiome, SFN metabolite, and dietary data, a sparse partial least squares (sPLS) model was used. SFN metabolite levels in urine were normalized to μmol excreted using the volume of urine produced for each individual. To capture total variation, the sum of detected metabolites across all 72 hours was calculated. To capture different components of GLS metabolism, three categories of metabolites were made: total dithiocarbamates containing SFN and its downstream metabolites (SFN-GSH, SFN-Cys, SFN-CG, SFN-NAC), SFN-NIT, and all metabolites . Dietary data for each participant was exported from Food Processor^® SQL and all nutrients were normalized to kCals consumed. For each nutrient, the mean consumption over the 7 days leading up to the study was calculated. Baseline (0 h) microbiome data was clr-transformed. Diet and urine blocks were log transformed to normalize the data and all three blocks were autoscaled to center and scale the data.

A sparse partial least squares (sPLS) model was built in canonical mode, as implemented in mixOmics (v6.20.0) ^[62]. Pairwise canonical sPLS models were built between each block of data and a correlation cutoff of 0.8 between components of each block were used to guide model design. The resulting design matrix connected the diet block to the microbiome block and the microbiome block to the SFN-metabolites, but not diet to SFN-metabolites. Component correlation between blocks was maximized to tune the model. This tuning resulted in 55 and 5 features kept on the first and second components of the microbiome block, respectively; 25 and 5 features kept on the first and second components of the diet block; and all features kept on both components of SFN-metabolite block. Variables were extracted from the first component and a loading cutoff of 0.1 was used for further investigation yielding 38% of ASVs from the model and 60% of dietary factors.

Correlation and Regression Analysis

Dietary components and microbes selected by the sPLS model were further investigated by Spearman’s correlation coefficient in R using untransformed diet and SFN metabolites and clr-transformed microbiome data. To visualize the results, pheatmap (v1.0.12) and unsupervised clustering based on Euclidean distance were used. To evaluate how much variation in GLS metabolism could be explained by the gut microbiome, a regression approach was used. Specifically, backwards stepwise selection based on AIC, implemented via the stepAIC function in the MASS R package (v7.5-38.0), was used to identify the most parsimonious model with the fewest needed predictor variables. The null model contained only grams of sprouts consumed while the full model contained grams of sprouts consumed and all 21 ASVs selected by the sPLS model.

Data and Code Availability

R scripts used for data analysis and relevant data are available in the github repository “bouranij/Bouranis_SFN_Microbiome”. Raw 16S reads are available in the NCBI SRA under BioProject PRJNA953404.

Results

Participant Characteristics

The mean age and BMI of both the alfalfa and the broccoli treatment groups was 31 years and 24 kg/m². The alfalfa group contained 16 females and 9 males and the broccoli group contained 21 females and 9 males. Overall, there were no significant differences between alfalfa and broccoli treatment groups in age, BMI, nor racial composition (Table S1). Total intake of calories and macronutrients, and other dietary components were similar between alfalfa and broccoli treatment groups (Table S2). Participant intake records indicated compliance with avoiding confounding food items. In general, the broccoli and alfalfa sprouts were well tolerated by subjects although mild abdominal discomfort (ex. gas, feeling full, upset stomach) was experienced by 20% participants in each treatment group. Discomfort was resolved in subjects 3-24 h post sprout consumption.

Distribution of SFN-Metabolites

No SFN-metabolites were detected in the alfalfa group or at baseline suggesting adherence to dietary protocols. We next evaluated the distribution of SFN-metabolites in plasma, urine, and stool samples over time (Figure 1). In plasma, the dominant metabolite at all timepoints was SFN-NIT which was also the only detectable metabolite at the 24, 48, and 72-hour timepoints. In urine, the distribution of metabolites was markedly different from plasma. SFN-NAC was the primary metabolite detected between 0-3, 3-6, and 6-24 hours while SFN-NIT was the dominant metabolite between 24-48 and 48-72 hours. In stool, at the 24-hour timepoint, free SFN was the primary metabolite detected followed by SFN-NAC and SFN-NIT. At 48 hours, over 95% of the detected metabolites in stool were free SFN and at 72 hours only free SFN was detected. This is in stark contrast to plasma and urine where free SFN made up only a small fraction of detectable metabolites. SFN metabolites in stool were three orders of magnitudes lower than those detected in plasma and urine (nmols/g vs μmols/dL and μmols, respectively) indicating the primary route of excretion for both dithiocarbamates and nitriles is through the urine as opposed to through biliary excretion. As expected, high levels of inter-individual variation were observed for all metabolites in all compartments (Figure 2).

Figure 1:

Distribution of SFN-metabolites in A) plasma, B) urine, C) Stool of broccoli consumers (n = 30). SFN: Free-sulforaphane, SFN-CG: Sulforaphane-CysGly, SFN-Cys: Sulforaphne-Cys, SFN-GSH: Sulforaphane-Glutathione, SFN-NAC: Sulforaphane-N-AcetylCysteine, SFN-NIT: Sulforaphane-Nitrile.

Figure 2:

Inter-individual variation in SFN is visualized with subject-level time-course curves of A) SFN-NIT in plasma, B) SFN-Cys in plasma, C) SFN-NAC in plasma, D) SFN-NIT in urine, E) SFN-NAC in urine, F) Total SFN metabolites in stool of broccoli consumers. Each point represents and individual (n=30) and black lines represent the average curve of all individuals.

Pharmacokinetic analysis was conducted separately on each metabolite (Table 1). Average C_max of SFN and DTC metabolites was 0.077 μM with a range of 0.0107 – 0.11 μM while SFN-NIT had a C_max of 1.05 ± 0.602 μM. Average t_max of SFN and DTC metabolites was 3.1 hours with a range of 3 to 3.2 hours and SFN-NIT had an average t_max of 3.2 ± 0.76 hours. AUC of DTC metabolites ranged from 0.0533 to 0.418 μmol x h/L with an average of 0.264 μmol x h/L while that of SFN-NIT was 15.4 ± 11.7 μmol x h/L. Lastly, the half-life of DTC metabolites ranged from 1.82 to 5.23 hours with an average of 2.66 hours while SFN-NIT was 9.89 ± 3.4 hours. Average excretion of DTC metabolites over 72 hours was 17.4 μmol with a range of 0.00174 to 70 μmol while SFN-NIT was 19.34 ± 9.26 μmol. Overall, these results indicate that while SFN-NIT is the dominant metabolite in circulation, this is due to markedly slower excretion and elimination from the body. Additionally, the primary metabolite to be generated from consumption of glucoraphanin from broccoli sprouts, as indicated from total excretion, is SFN which is rapidly metabolized to its downstream mercapturic acid pathway metabolite SFN-NAC.

Table 1:

Pharmacokinetics of Glucosinolate Metabolites

Pharmacokinetics of Glucosinolate Metabolites
	Pharmacokinetic Parameter	Free-SFN	SFN-GSH	SFN-CG	SFN-Cys	SFN-NAC	SFN-NIT
Plasma Metabolites	Cmax (μM)	0.0107 ± 0.00551	0.0737 ± 0.0423	0.103 ± 0.0666	0.0885 ± 0.0681	0.11 ± 0.0495	1.05 ± 0.602
	Tmax (h)	3	3.10 ± 0.548	3	3	3.2 ± 0.761	3.2 ± 0.76
	AUC(0-72h) (μmol x h/L)	0.0533 ± 0.0251	0.238 ± 0.137	0.331 ± 0.222	0.281 ± 0.229	0.418 ± 0.253	15.4 ± 11.7
	t1/2 (h)	2.28 ± 0.874	1.99 ± 0.888	2.01 ± 0.801	1.82 ± 0.700	5.23 ± 5.40	9.89 ± 3.4
Urinary Metabolites	Total Excretion (μmol)	4.54 ± 2.85	0.00174 ± 0.00474	0.0241 ± 0.0873	12.5 ± 5.85	70.0 ± 25.2	19.34 ± 9.26
	Faction Absorbed	0.0454 ± 0.0285	0.0000174 ± 0.0000474	0.000241 ± 0.000873	0.125 ± 0.0585	0.700 ± 0.252	0.193 ± 0.09

SFN-NIT Is Not Biologically Active in PC-3 Cells

SFN-NIT concentration was higher in plasma than we previously anticipated, thus we sought to confirm lack of biological activity. Previous early studies have shown SFN-NIT induced quinone reductase activity only at higher doses in Hepa 1c1c7 cells ^[5,22,23]. We choose aggressive prostate cancer cells (PC-3 cells) to do this work because they are known to be sensitive to SFN treatment, SFN-NIT bioactivity has not been tested in these cells before, and it allowed us to evaluate if SFN-NIT treatment could affect both known genetic and epigenetic targets (HDAC activity) ^[5,16,22]. SFN-NIT had no impact on HDAC activity (15 – 135 μM), while 15 μM SFN (positive control) significantly reduced HDAC activity compared to a vehicle-control (Figure 3A). Similarly, at all concentrations SFN-NIT did not significantly increase transcript abundance of Nrf2-related genes (NQO-1 and HO-1) at neither the 24 nor 48-hour timepoints compared to a vehicle control. Conversely, SFN significantly increased expression of both NQO-1 and HO-1 at both 24 and 48 hours after treatment (Figure 3B-C) and decreased HDAC activity (Figure 3A). In summary, these results indicate that SFN-NIT is not bioactive in PC-3 cells and thus does not confer the chemoprotective properties of cruciferous vegetables that SFN does.

Figure 3:

Comparison of SFN and SFN-NIT bioactivity in PC-3 prostate cancer cells A) HDAC activity following incubation with a vehicle control (DMSO), SFN, or SFN-NIT for 48h. B) NQO-1 and C) HO-1 gene expression in PC-3 cells (by qPCR) following incubation with vehicle control (DMSO), SFN, or SFN-NIT for 24 (light grey) or 48 (dark grey) hours. A-C) Bars represent mean ± SEM (n=4)

Consumption of a Single Serving of Broccoli or Alfalfa sprouts did not Alter Gut Microbiome Composition

We next set out to evaluate if the gut microbiome composition influences the inter-individual variation in SFN excretion we observed. In order to do this work we first needed to establish if the microbial composition of participants changed over time. Gut microbiome alpha- and beta diversity measures for each treatment group were evaluated to determine if a single serving of sprouts significantly impacts microbiome composition. No differences in observed ASVs, Shannon index, nor Simpson index of alpha diversity were detected both between groups at each timepoint, nor within groups between timepoints. Similarly, no differences in beta diversity, as calculated by Jaccard distance, were detected between groups at any timepoint, nor within groups between any timepoint (Figure S2). Lastly, differential abundance testing was conducted and no ASVs were found to be differentially abundant between treatment groups at any timepoint or between timepoints with treatment groups. Taken together, these findings suggest that consumption of a single serving of broccoli or alfalfa sprouts did not significantly alter gut microbiome composition.

Dietary Factors and Microbiome Composition Influences SFN Hydrolysis

To understand the relationship between gut microbiome composition, other dietary factors, and glucoraphanin hydrolysis products, a sparse partial least squares (sPLS) canonical analysis was performed. To capture different aspects of glucoraphanin metabolism, metabolites were broken into three categories: DTCs, SFN-NIT, and all metabolites (DTCs + SFN-NIT). 21 ASVs and 15 dietary factors with an absolute loading score greater than 0.1 were selected by the model suggesting they are the most important variables influencing microbiome composition and SFN-metabolite excretion. The selected ASVs included Dorea longicatena, multiple species of Alistipes, multiple unannotated species of Bifidobacterium, and multiple members of the Ruminococcus torques group. Dietary components included grams of mono- and di-saccharides, “other” carbohydrates (i.e. carbohydrates which are not a mono-, or di-saccharide nor fiber), total fiber, total soluble fiber, alcohol, saturated fat, and cholesterol consumed. The features selected by the sPLS analysis were further analyzed using spearman’s correlation. Dorea longicatena, members of genus Roseburia and Bifidobacterium, were positively correlated with dietary components such as fiber, “other” carbs, and micronutrients like folate, manganese, and copper, while negatively correlated with cholesterol, alcohol, and saturated fat (Figure S3). Members of genus Blautia and Alistipes had an inverse trend displaying positive correlations with cholesterol, saturated fat, and alcohol and a negative correlation with fiber, carbohydrates, and micronutrients. Negative correlations with SFN metabolites were found for members of genus Alistipes and Blautia while members of Bifidobacterium, Dorea longicatena, Rominococcus torques, and Roseburia had positive correlations (Figure 4, Figure S5). Taken together, these findings suggest that dietary factors such as carbohydrates, fiber, and micronutrients may promote the growth of GLS-metabolizing bacteria, such as those belonging to the genera Bifidobacterium and Roseburia. Additionally, bacteria which metabolize glucoraphanin to SFN may serve a similar ecological niche in the gut and utilize similar substrates for energy production.

Figure 4:

Spearman’s rank correlation between individual ASVs selected by the sPLS model and total SFN-metabolites in urine (n = 30). ASV names are presented as genus_species. Unannotated genera are shown as ‘f_family_ASV#’ and unannotated species as ‘g_genera_AVS#’.

To understand the extent to which the selected microbes explain inter-individual differences in SFN excretion in the study population, a regression model was built using backwards stepwise AIC. Of the 21 ASVs which entered the full model, 4 were removed coming from genera Lachnospiraceae UG-004 (ASV111), Lachnospiraceae CAG-56 (ASV62), Ruminococcus torques (ASV527), and Alistipes putredinis (ASV16). One reason for this removal may be due to high collinearity between these ASVs and others in the model. The adjusted R² of the model was 0.81 indicating 81% of the inter-individual variation in SFN excretion in our study population can be explained by the selected microbes. This observation indicates that the representation of these taxa in the gut may predict SFN excretion by an individual, possibly because these taxa impact SFN bioavailability.

Discussion

In this study, we sought to understand human pharmacokinetics of SFN-NIT generated from consumption of broccoli sprouts, describe the impact of a single serving of broccoli or alfalfa sprouts on gut microbiome composition, and evaluate the influence of gut microbiome composition on inter-individual variation of SFN metabolite excretion. To our knowledge, this study is the first to examine levels of SFN-NIT in human plasma and the first to measure any SFN metabolites in human stool samples. High levels of inter-individual variation were observed for metabolite production in plasma, urine, and stool samples, especially for SFN-NIT metabolism and excretion. We also confirm the marked difference in bioactivity of SFN-NIT compared to SFN and to our knowledge, this is the first study to report on the impact of SFN-NIT on HDAC activity and on Nrf2-targets in prostate cancer cells.

Unlike in plasma and urine where mercapturic acid pathways metabolites are primarily detected, the dominant metabolite detected in stool was free SFN. Indeed, members of the gut microbiome could hydrolyze glucoraphanin to SFN. However, the SFN in stool could instead be from enterohepatic circulation. One excretion mechanism of SFN and its metabolites is through the bile, following transformation through the mercapturic acid pathways with the primary metabolites excreted in bile being SFN-GSH and SFN-NAC ^[19]. Zhang et al. showed that SFN-GSH readily dissociates from glutathione to form free SFN, thus, the SFN detected in stool may be through this mechanism ^[63]. This excretion mechanism through the bile may also explain the relatively high levels of SFN-NAC observed in stool at 24 hours post-consumption. In this study, we also found SFN-NIT was significantly slower to clear circulation from plasma than SFN and its DTC metabolites. Little is known about SFN-NIT metabolism in vivo, however, its long duration in circulation suggests binding to plasma proteins or more prominent enterohepatic circulation. On the other hand, isothiocyanates, and other electrophilic xenobiotics, are known to upregulate phase 2 drug metabolizing enzymes, effectively inducing their own excretion ^[64-68].

We did not observe any changes to gut microbiome composition with broccoli or alfalfa sprout consumption. While it is reasonable to expect no change after only one exposure to a food it was important to establish, as others have shown that longer term dietary exposure to broccoli, broccoli sprouts, brussels sprouts, daikon radish, and other cruciferous vegetables, can shift the microbiome ^{[25,39-41,43,69,70]}. We did observe relationships between microbes and sulforaphane metabolites. We observed a positive correlation between a member of genus Roseburia, members of genus Bifidobacterium, Bacteroides vulgatus, members of genus Ruminococcus torques group, and Dorea longicatena and SFN metabolites, and these same microbes were also positively correlated with consumption of carbohydrates, fiber, and micronutrients. Conversely, members of genus Alistipes and a member of genus Blautia were negatively correlated with SFN metabolites. In congruence to our findings, in vitro work has shown members of Bifidobacterium and Bacteroides are capable of converting glucosinolates into isothiocyanates and nitriles. Likewise, members of Bacteroides have been shown to increase with broccoli consumption ^[39,71-73]. Additionally, members of Bifidobacterium are known to increase with fiber consumption ^[74]. In fact, in vitro work has indicated that Bifidobacterium exclusively converts glucosinolates into nitriles ^[71]. We observed moderate positive correlations between members of genus Bifidobacterium (rho 0.46-0.5) while weak trends were observed with DTCs (rho 0.15-0.17). Work conducted on Bacteroides thetaiotaomicron has shown that enzymes capable of hydrolyzing glucosinolates to isothiocyanates are also capable of hydrolyzing carbohydrates including cellobiose and maltose ^[56]. Furthermore, prolonged consumption of broccoli has been shown to increase the efficiency of the gut microbiome to hydrolyze glucosinolates to isothiocyanates. It has been shown that this change is induced by broccoli with or without glucosinolates (i.e. fiber) while not induced by isolated glucosinolates ^[69,70]. Taken together even though individual ASVs do not change in abundance with short-term broccoli consumption, microbial metabolic pathways may be impacted altering the bioavailability of SFN and influencing host health. Longer-term feeding studies with cruciferous vegetables have shown that members of Blautia decrease in abundance with cruciferous vegetables intake while members of Alistipes have shown mixed responses to cruciferous vegetable consumption ^{[34,40-42,69]}. Interestingly, multiple studies have reported a decrease in members of both Dorea and Roseburia with cruciferous vegetable consumption, however, we observed some of the highest positive correlations between glucosinolate metabolites and these bacteria in this study ^[43,69]. All in all, our findings, and those in the literature, point to evidence of shared pathways between glucosinolate hydrolysis and carbohydrate hydrolysis in microbes. Additionally, diet can affect not only the abundance of individual microbes, but also the expression of metabolic pathways impacting the bioavailability of dietary compounds.

We previously hypothesized that metabolic conversion of glucoraphanin into SFN and SFN-NIT by the gut microbiome was inversely related ^[25]. Additionally, we previously observed a positive relationship between members of family Clostridiaceae and SFN-NIT abundance, possibly due to desulfation of glucosinolates to desulfoglucosinolates ^[25]. In contrast, in this study we did not see similar trends and instead the three ASVs of family Clostridiaceae that we did observe had correlations of differing strengths and directions with SFN-NIT and DTC levels (Figure S4). In fact, we observed positive correlations between total SFN-NIT and DTC excretion in urine (r = 0.4). These findings challenge our previous hypotheses and may stem from differences between the in vivo nature of this work and the in vitro approach previously used. A major limitation of this study is that the contribution of the gut microbiome in the conversion of glucoraphanin to SFN and SFN-NIT cannot be fully teased apart from that of plant enzymes. Myrosinase and epithiospecific protein from the plants is still present contributing to SFN and SFN-NIT conversion and pre-existing desulfoglucosinolates within the sprouts could be converted to SFN-NIT further complicating detection of the microbiome-effect ^[44,75,76]. Human feeding experiments conducted with GLS-extracts in the absence of myrosinase will help to further clarify the relationship between SFN-NIT and SFN generated by the gut microbiome. Additionally, while our linear regression model yielded a relatively high R², this measure simply describes the fit of our model and cannot be extrapolated to other study populations. Future work is needed to further explore the relationship between DTCs and SFN-NIT in other populations to evaluate the generalizability of our findings.

This work pushes the field of nutrition further towards individualized-nutrition and precision medicine by beginning to describe factors in vivo which drive inter-individual variation in the generation of isothiocyanates from broccoli. The driver of this variation appears to be the gut microbiome with multiple members of genus Bifidobacterium, Roseburia, Alistipes, and Blautia putatively playing roles. The high R² of our linear regression model suggests that variation in SFN excretion is highly linked to gut microbiome composition and thus personalized differences in microbiome composition greatly influence glucosinolate metabolism. However, this high R² cannot be extrapolated to other populations and thus more work in larger and more diverse populations is needed. The long duration of SFN-NIT in plasma and its low abundance in stool supports in vitro work suggesting it’s lack of bioactivity, and thus high SFN-NIT producers may gain less cancer-preventative effects from consuming broccoli. Future mechanistic studies are needed to further elucidate the role of the gut microbiome in glucosinolate metabolism and tease apart the influence of the microbial enzymes and enzymes and compounds in the plants.

Supplementary Material

SI

Figure S1: Rarefaction curves of filtered 16S amplicons at the ASV levels. Each curve represents one stool sample (n=317).

Figure S2: A single serving of broccoli nor alfalfa sprouts impacted microbiome composition. A) Principal Coordinate Analysis (PCoA) of alfalfa and broccoli consumers microbiomes 24 hours after consumption (n = 55). B) PCoA of broccoli consumer (n = 30) microbiomes 0, 24, 48, and 72 hours after consumption. PCoA plots were generated using JACCARD distance on the presence/absence ASV data.

Figure S3: Spearman’s rank correlation between individual ASVs and dietary factors selected by the sPLS model. ASV names are presented as genus_species. Unannotated genera are shown as ‘f_family_ASV#’ and unannotated species as ‘g_genera_AVS#’. Dietary factors are as follows: diet_Chol_mg: milligrams dietary cholesterol; diet_SatFat_g: grams of dietary saturated fat; diet_Alc_g: grams of dietary alchol; diet_TotSolFib_g: grams of total soluble fiber; diet_MonSac_g: grams of monosaccharides; diet_Disacc_g: grams of disaccharides; diet_BetaCaro_mcg: micrograms of beta carotene; diet_vit_B1_mg: milligrams of vitamin B1 (Thiamin); diet_Vit_B6_mg: milligrams of vitamin B6; diet_Fol.DFE_mcg_DFE: micrograms of dietary folate equivalents; diet_Mang_mg: milligrams of dietary manganese; diet_OCarb_g: grams of “other” carbohydrates; diet_TotFib_g: grams of total fiber; diet_Iron_mg: milligrams of dietary iron; diet_Copp_mg: milligrams of copper.

Figure S4: Spearman’s rank correlation between all members of family Clostridiaceae detected in the samples of broccoli consumer (n = 303) and SFN-metabolites. ASV names are presented as genus_species. Unannotated genera are shown as ‘f_family_ASV#’ and unannotated species as ‘g_genera_AVS#’.

Figure S5: Relationships between center log ratio (clr)-transformed microbes and total SFN-metabolites (A-C), SFN-NIT (D-E), and DTCs (F).

Abbreviations Used:

AIC

Akaike Information Criterion

ASV

Amplicon Sequence Variant

CG

CysGly

clr

center log ratio

Cys

Cysteine

DTC

Dithiocarbamate

GSH

Glutathione

GST

Glutathione S transfrase

HDAC

Histone deacetylase enzymes

HO-1

Heme Oxygenase 1

ITC

Isothiocyanate

NAC

N-Acetyl Cysteine

NIT

Nitrile

NQO1

NADPH Quinone Dehydrogenase 1

Nrf2

Nuclear factor erythroid 2–related factor 2

PCoA

Principal Coordinate Analysis

PERMANOVA

Permutation analysis of variance

SFN

Sulfroaphane

SFN-NIT

Sulforaphane-Nitrile

sPLS

Sparse Partial Least Squares