Isothiocyanates are detected in human synovial fluid following broccoli consumption and can affect the tissues of the knee joint

Osteoarthritis is a major cause of disability and there is no current pharmaceutical treatment which can prevent the disease or slow its progression. Dietary advice or supplementation is clearly an attractive option since it has low toxicity and ease of implementation on a population level. We have previously demonstrated that sulforaphane, a dietary isothiocyanate derived from its glucosinolate precursor which is found in broccoli, can prevent cartilage destruction in cells, in in vitro and in vivo models of osteoarthritis. As the next phase of this research, we enrolled 40 patients with knee osteoarthritis undergoing total knee replacement into a proof-of-principle trial. Patients were randomised to either a low or high glucosinolate diet for 14 days prior to surgery. We detected ITCs in the synovial fluid of the high glucosinolate group, but not the low glucosinolate group. This was mirrored by an increase in ITCs and specifically sulforaphane in the plasma. Proteomic analysis of synovial fluid showed significantly distinct profiles between groups with 125 differentially expressed proteins. The functional consequence of this diet will now be tested in a clinical trial.


Study design.
This was a proof-of-principle; randomised controlled trial designed to test whether bioactive molecules derived from broccoli, eaten prior to joint replacement, can be detected in the articular joint and mediate changes.
Patients followed a 7-day washout period, (no consumption of broccoli, cabbage, cauliflower, arugula (rocket), canola/rapeseed, Brussels sprouts, radish, horseradish, kale, turnip root, greens, watercress, pak choi, collard greens, rutabaga (swede) vegetables). The washout diet was either continued (low glucosinolate group) or patients consumed 100 g of a high glucosinolate broccoli 15 per day for 14 days prior to TKR (high glucosinolate group). Patients were allocated to a high or low glucosinolate group by block randomisation (blocks of five). The trial nurses screened and recruited patients and were blinded to the pre-allocated study numbers. Throughout the study, patients had telephone access to a trial nurse for advice and received a telephone call to monitor compliance.
Blood was sampled at the end of the washout period (baseline) and collected into K 2 EDTA. Patients allocated to the high glucosinolate group were given written instructions, a vegetable steamer and a demonstration for preparation of the broccoli. All patients were given a validated 7-day food diary with guidance for completion. A second blood sample was taken on the day of TKR surgery. Synovial fluid, cartilage and fat tissues were obtained at surgery. Synovial fluid was aspirated into a sterile, pre-chilled container immediately prior to surgery, taking care to avoid blood contamination. Cartilage and fat tissues were removed as routine for TKR and immediately placed into separate containers with pre-chilled, sterile phosphate buffered saline (PBS). All tissue samples were transported at 4 °C in sterile conditions and processed or stored within 1 hour. Synovial fluid was centrifuged at 200 g, 5 minutes at 4 °C and the supernatants stored at −80 °C. Undamaged cartilage was dissected at full thickness from the bone, and both cartilage and fat were snap frozen and stored at −80 °C for RNA extraction, or placed directly into culture for in vitro degradation analysis.
Blood plasma and synovial fluid were analysed for ITC levels. Analyses of synovial fluid proteomics, in vitro cartilage degradation, joint tissue gene expression profiling and GST genotyping were performed. Sulforaphane detection: Plasma. Plasma samples were analysed for SFN and its conjugates using a method that has been described previously [13] with slight modifications. Briefly plasma samples (100 µl) were prepared by adding 20 µl precooled (4 °C) trichloroacetic acid followed by centrifugation at 11 600 g at 4 °C for 10 min. The injection volume was 5 µl, the HPLC column was a Luna (3 µm particle size, 100 × 2 mm from Phenomenex) using mobile phase flow 0.25 ml/min, the mobile phase were consisted of ammonium acetate buffer (13 mmol/L, pH4; solvent A) and acetonitrile plus 0.1% acetic acid (solvent B) in a linear gradient from 5% B to 35% B over 5 min with 6 min re equilibration time. Agilent 6490 Mass spectrometry analysis was performed with the use of electrospray ionization in positive ion mode with nitrogen gas and nitrogen sheath gas temperatures 160 °C and 400 °C, respectively, gas flow 16 l/min, Nebulizer pressure 30 psi, sheath gas flow 12 l/min and capillary voltage 4000 V.
Isothiocyanate detection: Synovial fluid. Synovial fluid was diluted in water if necessary and processed as for plasma with the inclusion of at least 10 mins centrifugation following the deproteinisation step. Sensitivity of the assay in synovial fluid was 3.8 pmols.
Proteomic analysis of synovial fluid. A subset of 18 synovial fluid samples were prepared as described previously 18 . Briefly, samples were hyaluronidase treated, filtered and total protein quantified using Bradford reagent. Synovial fluid samples containing 6 mg total protein were enriched for low abundance proteins using ProteoMiner ™ Small Capacity kit (BioRad) according to the manufacturer's instructions and analysed by SDS-PAGE. For LC-MS/MS, samples were trypsin digested and loaded on a 2 hour gradient LC-MS/MS analysis using a NanoAcquity ™ ultra-performance LC (Waters) and LTQ-Orbitrap-Velos (Thermo-Fisher Scientific) 19 . One sample was excluded for technical reasons (blebbing).
Progenesis ™ LC-MS software (Waters, Manchester, UK) was used for label-free quantification 18 . Data were searched using Mascot against the Unihuman Reviewed database. Peptide matches above an identity threshold were adjusted to give a false discovery rate (FDR) of 1% before the protein identifications being re-imported into Progenesis-QI ™ . Adjusted ANOVA values of p < 0.05 and regulation of >2-fold were regarded as significant.
Neopeptides were identified as previously described adjusted to 1% FDR 20 .
In vitro cartilage degradation. Cartilage was dissected into ~5 mm 3 chips and cultured overnight in quadruplicate in Dulbecco's modified Eagle's medium (DMEM; GlutaMAX) supplemented with 1,000 IU/ml penicillin, and 100 µg/ml streptomycin at 37 °C, 5% v/v CO 2 . Culture media were refreshed at day 1 and day 7, with an overall culture of 14 days. Remaining cartilage was papain digested. Culture media and digested cartilage were analysed for glycosaminoglycan and hydroxyproline content as described 21,22 .
Gene expression analysis. Cartilage (200 mg) was crushed under liquid nitrogen to a fine powder and total RNA was extracted using TRIzol ® reagent (Invitrogen) and RNeasy mini kit (Qiagen), as described 23,24 .
Approximately 100 mg of fat tissue was homogenised in 600 µl RLT buffer (Qiagen) using the TissueLyser LT (Qiagen) and spherical ball bearings. Total RNA was extracted using the RNeasy mini kit (Qiagen) according to the manufacturer's instructions.
RNA was quantified using the NanoDrop 2000 spectrophotometer (Thermo Scientific) and the quality assessed using the Experion RNA StdSens analysis kit (Bio-Rad). cDNA was synthesised using 500 ng of RNA and Superscript II reverse transcriptase (Invitrogen) with random hexamers according to the manufacturer's instructions.
Relative quantification of gene expression was measured using the 7900HT Fast Real Time PCR System (Applied Biosystems) and a custom designed Taqman Low Density Array (TLDA). Genes were selected by analysis of microarray data from SFN treated SW1353 cells, primary human articular chondrocytes and human articular cartilage explants (data not shown). Assay details are included in Supplementary Table S1. Cycling conditions were 50 °C, 20 s, 95 °C, 10 min, 95 °C, 15 s, and 60 °C, 1 min × 40 cycles. Data were analysed using DataAssist Software v3.01 (Life Technologies), normalised to 18 S rRNA and expressed as 2 −ΔCt .
Biomarkers. Plasma samples were analysed for Coll 2-1, (a peptide from the triple helix of type II collagen) using ELISA 25 , and hsCRP (high sensitivity C-reactive protein) using a particle-enhanced immunoturbidimetric assay. Analyses were performed by the Bioanalytical Facility (BAF) at the University of East Anglia. Synovial fluid samples were analysed for cytokine protein expression using the Cytometric Bead Array Flex Set (BD Biosciences) and was performed by Flow Cytometry Services, University of East Anglia.
Diet diary. All patients completed a validated 7-day food diary 26 . Dietary data was analysed using WISP V4.0 (Tinuviel Software).
Genotyping. DNA was extracted from plasma using QIAamp DNA Blood Mini Kit (Qiagen) according to manufacturer's instructions. Copy number variation for members of the glutathione S-transferase superfamily, (GSTP1, GSTM1 and GSTT1) was detected using TaqMan ® Copy Number Assays (Life Technologies cat: Hs01541162_cn, Hs02575461_cn, Hs00010004) and analysed using CopyCaller ® software v2.0 (Life Technologies).
Data availability. The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Statistical testing.
High and low glucosinolate groups were compared using unpaired t-test (two tailed) and when applied, false discovery rate (FDR) was calculated using the Benjamini-Hochberg FDR method. It was noted that within the plasma samples of the low glucosinolate group, all except five samples of the post intervention samples had undetectable levels of SFN. Undetected points were treated as zero for the comparisons of plasma ITCs or SFN, and synovial fluid ITC. Since ITCs were not detected in the low glucosinolate group of the synovial fluid samples, no statistical testing was performed. Differences observed between groups remains the same and conclusions are also identifical using 'limit of detection' values instead of zero. All data was analysed using GraphPad Prism version 5.00 for Windows, GraphPad Software, San Diego California USA, unless otherwise stated in the methods. Graphpad Prism calculated descriptive statistics to include confidence intervals and normality testing (D' Agostino & Pearson omnibus normality test). Data with non-Gaussian distribution were transformed prior to further statistical testing. A p value < 0.05 was considered significant.
Ethics approval. This study protocol was approved by the NRES Committee East of England, Cambridge South, UK.

Results
Patients and baseline characteristics. A total of 268 patients were screened for eligibility where 143 (53%) were eligible, of which 40 patients (15%) were recruited. A flow diagram to show the process of recruitment and parallel randomisation of the trial is shown in Fig. 1, and more details are given in Supplementary Table S2. Drug use was not recorded. Patient characteristics were males n = 17, (low glucosinolate 7, high glucosinolate 10,), females n = 20 (low glucosinolate 13, high glucosinolate 7). Low glucosinolate group median age 71 (range 50-89), high glucosinolate group median age 73 (range 58-84).
Isothiocyanate levels in plasma and synovial fluid. Total ITCs measured in the high glucosinolate broccoli were 1.8 µmols g −1 , so patients consumed 180 µmols per day. Patients in the high glucosinolate group attained greater increases in plasma total ITC concentrations between baseline and post-intervention: high glucosinolate group (mean fold change: 2.20, 95% CI: 1.45 to 2.92) compared to the low glucosinolate group (mean fold change: 0.90, 95% CI: 0.68 to 1.12), p < 0.0001 (Table 2). This confirmed compliance with the intervention. Sulforaphane was also significantly increased in the plasma of the high glucosinolate group (mean change: 0.32, 95% CI: 0.20 to 0.44) compared to low glucosinolate (mean change: 0.02, 95% CI: −0.01 to 0.04), p < 0.0001 (Table 2); since SFN in many post-intervention samples in the low glucosinolate group was undetected and therefore treated as zero, fold change cannot be calculated. Importantly, ITCs were detected in the synovial fluids of the high glucosinolate group (mean 496.6 nM, 95% CI: 419.1 to 574.2) but not in the low glucosinolate group (

Proteomics.
Peptides identified by LC-MS/MS in the synovial fluid samples (low glucosinolate n = 8, high glucosinolate n = 9) were analysed using Progenesis ™ QI software (Waters, Manchester, UK). 382 quantifiable proteins were identified, 348 with >1 peptide. One hundred and twenty-five proteins were identified with >1 peptide and differentially expressed between high and low glucosinolate diets (p < 0.05, 1% FDR). An FDR of 5% revealed 29 differentially expressed proteins identified by >1 peptide with a minimum fold change of >2. In the high glucosinolate group 18 proteins had decreased expression whilst 11 proteins were increased compared to the low glucosinolate group. Twelve proteins are commonly found in plasma, 10 of which were decreased in the high glucosinolate group (Table 3). Aside from common plasma proteins, the top four differentially expressed proteins, were neuroblastoma suppressor of tumorigenicity 1 (NBL1), histidine protein methyltransferase 1 homolog (METTL18), calreticulin (CALR) and SWI/SNF complex subunit SMARCC1 (SMARCC1) each p < 0.0001.
Neopeptides. Neopeptides identified in synovial fluid unique to high and low glucosinolate groups were identified for aggrecan, biglycan, decorin, fibromodulin, cartilage oligomeric matrix protein and a number of collagens.
There was no correlation between diet and neopeptide abundance, (an indicator of proteolysis), in synovial fluid (see Supplementary Table S3). This holds true whether considering neopeptides unique or common to either intervention.    In vitro cartilage degradation. No significant differences were observed for in vitro rates of cartilage degradation (glycosaminoglycan and hydroxyproline loss) between the high and low glucosinolate interventions (Table 4).
Diet Diaries. Nutrient intakes were largely similar between groups however, a significant increased intake of vitamin C (p = 9.96 × 10 −06 ) was observed following adjustment for multiple testing using a 1% FDR (Supplementary Table S5).
Adverse events. There was one serious adverse event of a fractured humerus considered unrelated to the intervention. In terms of minor events, one participant reported bloating (high glucosinolate group), although the participant continued with the intervention.

Discussion
Our results provide the first evidence that following broccoli intake, the bioactive constituent ITCs reach the synovial fluid at concentrations with biological impact on the articular joint tissues, and alter the synovial fluid protein profile. This study clearly demonstrates that a dietary bioactive with chondroprotective properties can penetrate the knee in osteoarthritis. These data support our previous mouse model work, where we showed that a SFN-rich diet can provide chondroprotection in a mouse model of OA, though SFN and its metabolites were not measured directly in the joint. Mean total ITC levels were 4.47-fold lower in the synovial fluid than the plasma. The concentration of small molecules in synovial fluid often reflects that of plasma since permeability into synovial fluid occurs through free diffusion from the endothelium of the synovial microvasculature and surrounding interstitium. The synovial lining has no basement membrane and therefore small molecule diffusion is limited only by the volume of intercellular space 27 , however, many non-steroidal anti-inflammatory drugs also show lower concentrations in synovial fluid than plasma 28,29 . This reflects the kinetics of elimination of the drug, repeat administration and also protein binding. In this study we did not standardize the timing of broccoli intake compared to sampling of blood or synovial fluid. It is also possible that the diet modifies synovial fluid volume which was not monitored in this study, though this is unlikely to be significant. Plasma levels for ITCs and SFN were higher than expected given the limited available published data. In the current study the timing of the broccoli meal during the day was not controlled. This and possible differences in the rate of metabolism by each patient's unique microbiome 30 may explain the greater variation observed for plasma ITC levels in the high glucosinolate group. Twelve of the 29 synovial fluid proteins with significant difference between groups are commonly found in plasma, and 10 of  these were decreased in the high glucosinolate group. In the inflamed joint, changes in the permeability of the synovium have been described, whereby increases in protein and solute concentration within the joint cavity have been observed 31,32 . Resolution of aberrant synovial permeability by the high glucosinolate diet may offer one possible explanation for a reduction in the number of plasma proteins observed in this group, though this remains to be confirmed. Furthermore, a number of the proteins highlighted in our study (table 3) have been reported in human OA synovial fluid previously [33][34][35][36] . Common plasma proteins aside, there were a number of interesting proteins decreased in abundance with the high glucosinolate diet including NBL1, CALR, SMARCC1 and type VI collagen. NBL1 is an antagonist of bone morphogenetic proteins (BMPs), key in cartilage homeostasis. Increased gene expression of NBL1 in cartilage is reported in mouse and rat models of OA and in human OA cartilage [37][38][39][40] . Calreticulin has functions outside of calcium regulation 41 . Calreticulin is increased in the plasma and synovial fluid of RA patients and this correlated with swollen joint count and disease activity score 42 . SMARCC1 is a regulator of gene transcription through chromatin remodelling, and is found in exosomes in cancer with no reported function in the joint [43][44][45] . Type VI collagen is a structural component of the perichondral matrix within articular cartilage. It has been reported in synovial fluid, with no significant difference between OA and control 33 . Methyltransferase-like 18 (METTL18) was significantly increased in abundance in the high glucosinolate group. It is a putative methyltransferase, and interacts with molecular chaperones 46 . Individual neopeptides were not reproducibly identified across all samples and this technology requires further development to be conclusive.
Gene expression in either cartilage or fat was not significantly different between groups when data were adjusted for multiple testing. Genes shown to be significant in unadjusted data are therefore of uncertain importance.
The collagen degradation biomarker Coll 2-1 has previously been shown to decrease after supplementation with a form of curcumin 47 , however only at the 85 day time point and not at the earlier 14 day time point.
Validated diet diaries showed that patients consumed similar levels of energy, fat, protein and fibre during the intervention. The high glucosinolate group consumed significantly higher levels of vitamin C, likely due to the high vitamin C content of broccoli. We therefore cannot rule out the contribution of vitamin C to the effects observed in the study. Future studies involving broccoli intake should include a control for vitamin C levels.
The interaction of GST genotype and ITC metabolism measured as rate of urinary secretion has been reported previously with differing associations 13,[48][49][50] . It has been suggested that specific GST null genotypes may confer a protective effect through a lowered rate of metabolism that extends tissue exposure time. Our cohort showed no associations of genotype with study outcomes though numbers in each subgroup were small (data not shown).

Limitations.
Our study was a proof-of-principle study and limitations of the trial design were the short intervention time and unspecified timing of broccoli intake. These limitations and/or the use of end-stage OA patients whose tissue gene expression maybe beyond modification by dietary intervention, may have contributed to the small changes in gene expression observed. Similarly, cartilage degradation rates in vitro were likely affected by rapid metabolism of dietary SFN in these cultures. The link between ITCs in the joint and changes in the synovial fluid may be indirect and this requires further investigation.

Conclusion
This is the first human study to show that increased broccoli intake results in ITC uptake into the joint, with concomitant changes in the joint. Coupled with our earlier data showing efficacy of SFN in both in vitro and in vivo laboratory models of OA, this supports the need for an appropriately designed clinical trial to determine the impact of dietary SFN in OA.