Clerodane furanoditerpenoids as the probable cause of toxic hepatitis induced by Tinospora crispa

Tinospora crispa is a popular traditional herbal plant commonly used throughout the world for treatment of various diseases, in particular type 2 diabetes mellitus. We report here a new case of toxic hepatitis in a 57-year old male patient in the French West Indies following the consumption of two aqueous extracts of fresh Tinospora crispa stems. It thus differs from two previously reported cases that concerned the chronic intake of powdered dry stems delivered in solid oral dosage forms (i.e. pellets and tablets). Liquid Chromatography-Diode Array Detection-Mass Spectrometry (LC/DAD/MS) analyses were performed on an aqueous extract of the offending sample that mimics the swallowed preparation. They revealed the presence of species-specific molecular marker borapetoside C (1) and thus enabled an unambiguous phytochemical identification. The exploration of tandem MS/MS data obtained by ultra-high performance liquid chromatography/electrospray ionization quadrupole time-of-flight mass spectrometry (UHPLC-ESI-QTOF-HRMS) allowed the identification of 17 additional cis-clerodane-type furanoditerpenoid lactones, analogues of 1. These results support the hypothesis that the mechanisms underlying hepatotoxicity of Tinospora crispa are the same as those encountered with furanoditerpenoids-containing plants such as Teucrium chamaedrys or Dioscorea bulbifera. In the context of type 2 diabetes treatment, we recommend that Tinospora crispa intake should be more closely monitored for signs of hepatotoxicity.


Laboratory Investigations
Identification of plant sample. The sample provided by the patient consisted of fresh stems of T. crispa as ascertained by comparison of both macroscopic botanical features and phytochemical profile, with those of an authentic herbal reference standard 4 7 .
Chromatographic profiles of both the analysed and the reference samples were next established by High-Performance Liquid Chromatography with Diode-Array Detection-Mass Spectrometry HPLC-DAD-MS for their corresponding CH 2 Cl 2 extracts (Fig. 2).  Table 1. Evolution of hepatic laboratory parameters. The first laboratory test was performed one week after the last intake of T. crispa decoction. Borapetoside C (1) 8,9 (synonym tinocrisposide), a species-specific molecular marker of Tinospora crispa was unambiguously identified in the herbal reference standard as well as in the suspect sample by comparison of the t R , UV spectrum and MS data with a reference compound isolated from the plant (t R 55.2 min; λ max : 214 nm; m/z 559: Na adduct [M + Na] + , 537: pseudo-molecular ion [M + H] + ; fragment ions at m/z 375, 357, 339, 325, 311, 307, 297, 279, 251, 205, 187 and 159). Interestingly, this compound appeared as the main peak on the LC-UV (214 nm) chromatogram of the analysed sample (Fig. 2, lower chromatogram). The amount of this compound was clearly found higher in the fresh stems than in the older reference sample, moreover consisting of dried stems (Fig. 2, upper chromatogram).
Phytochemical analysis of an aqueous extract that mimics the traditional preparation. We next investigated whether borapetoside C (1) and other potentially toxic related clerodane furanoditerpenoids could be identified in an aqueous extract of fresh stems that mimics the traditional preparation.
First of all, the peak assigned to 1 was also found predominant in the chromatogram obtained by HPLC/ DAD/MS and depicted in Fig. 2 (middle chromatogram). However, this finding was not really surprising since glycosides are essentially water-soluble.
Subsequently, a targeted qualitative analysis of the furanoditerpenoids structurally related to 1 was carried out in two ways: by using ultra-high performance liquid chromatography/electrospray ionization quadrupole time-of-flight mass spectrometry (UHPLC-ESI-QTOF-MS/MS) and by analysing the generated data on the basis of the exact mass of the compounds and their fragmentation pathways, in combination with a molecular networking approach for the whole dataset processing and especially their filtering. For this purpose, the data analysis portal of the Global Natural Products Social Molecular Networking (GNPS) web-based platform (http://gnps. ucsd.edu) was used. This approach constitutes a powerful and innovative tool for the dereplication of natural products and consequently for the identification of new ones. Building of molecular networks is based on relatedness analysis within tandem MS/MS data. It enables to visualize them as a molecular network clustering together data for structurally related molecules. MS/MS spectra are represented as nodes connected together by edges symbolizing close similarities between them 5,6 . Consequently, introducing this step in the workflow allowed an efficient data filtering and facilitated the targeting of compounds structurally analogous to borapetoside C (1).
The molecular network generated with the MS/MS data acquired from both the CH 2 Cl 2 and aqueous extracts is depicted in Fig. 5. Within the molecular network we focused our attention on the cluster exhibiting a central node associated with the precursor mass and the t R of borapetoside C (1). It should be noted that under the adopted conditions of "one-shot" analysis (i.e. without use of enriched fractions and optimization of the collision energy) several nodes were found to correspond exclusively to fragment ion spectra and are thus labelled with the corresponding Compounds corresponding to peaks 13 and 14 exhibited the same pseudomolecular ion at m/z 537.2332/537.2338, respectively (calculated for C 27 H 37 O 11 : 537.2330, Δppm = 0.3/1.4) as that of 1 and both could be tentatively assigned to borapetoside E 10,11 or to unknown stereoisomers of borapetosides C and E. Peaks 9, 11 and 12, could be tentatively assigned to borapetoside D (=6′ → 1″)-O-β-D-glucopyranosyl borapetoside E) 10,11 15 : 681.2753, Δppm = −0.6) and has not been previously reported. The mass difference observed between the protonated ions of this compound and that of 1 was 145 uma, which could be reasonably attributed to one additional ester group of a hydroxylated dicarboxylic acid (C 6 H 10 O 5 ).
A second group of borapetoside C analogues is represented by compounds bearing a hydroxyl group at the C-2 position such as borapetoside B. They all shared common fragmentation pathways with probably similar reactions to those described for 1, as illustrated in Fig. 7 for borapetoside B. However, protonated parent ions and fragment ions displayed a characteristic difference of mass of 16 uma attributed to oxygen in comparison to analogous ions observed in the tandem mass spectra of 1. Compounds belonging to this group are significantly more hydrophilic and were eluted between 7.0 and 12.0 min (peaks 1-6 and 10).
Compounds corresponding to peaks 5 (t R of 9.1 min) and 6 (t R of 9.3 min) exhibited the same pseudomolecular ion at m/z 553.2277/553.2279, respectively (calculated for C 27 H 37 O 12 : 553.2280, Δppm = −0.5/−0.1) and both could be tentatively assigned to borapetoside B 8 or its reported isomers (i.e. rumphioside I 11 and the diastereosiomer 3 13   . This new compound is most likely the corresponding analogue of either borapetoside B or its isomers, with the lactone ring opened and converted to 5-hydroxypentanoic acid moiety. A last group corresponds to compounds eluted at 9.7 (peak No 7) and 11.7 min (peak No 8). These compounds shared some fragment ions with borapetoside C (1) and its close structural analogues particularly those at m/z 251, 159 and 131, but also exhibited in their MS/MS tandem spectra several specific fragment ions. Peak No 8 was tentatively assigned to the non-glycosidic furanoditerpenoid crispene B 12  We concluded from these data that this compound was an unknown O-glycosyl derivative of crispene B or of its isomers (such as, for example, the corresponding carboxylic acid of 1).
Interestingly, apart from mass and chromatographic information provided by molecular network exploration, node coloring indicates that all these compounds were detected in the aqueous extract. This observation is consistent with the presence of highly polar/hydrophilic molecules such as, for instance, borapetosides D and H which bear two sugar units, as well as borapetoside B and rumphioside I with a free hydroxyl group at the C2 position.    . Molecular network generated with UHPLC-ESI-QTOF-HRMS 2 data from the CH 2 Cl 2 and aqueous extracts of T. crispa stems and visualized using Cytoscape software. Cosine similarity score cutoff was of 0.6. Nodes are labelled with parent or precursor mass value and their size are linked to the number of spectra (molecular formula of corresponding ions are indicated beside nodes). Nodes coloring: nodes are represented as pies and each color represents a group of spectrum files associated with an extract: red: G1/aqueous extract; blue: G2/CH 2 Cl 2 extract. Edges are annotated with mass difference and their thickness depends on the cosine score ranging between 0.6 (minimum accepted similarity between spectra) and 1 (maximum similarity between spectra). See Material and Methods for supplementary details.

Discussion
Among the large spectrum of traditional use of T. crispa reported in the literature, its anti-diabetic activity has raised a special interest for researchers all around the world considering the growing number of type 2 diabetes cases, all the more so as its use in South-Asia has been widespread for a long time. We have discovered that the plant is also consumed in the French West Indies (and probably throughout the Caribbean Arc due to the historic intermingling of Caribbean populations). Indeed, a survey has shown that 60% of diabetic patients in Martinique (French West Indies) use plants in addition to their glucose lowering drugs, among them T. crispa was the second plant the most used 14 . However, like several alternative medicine therapies for diabetes, finding  reliable information about safety and benefits remains difficult. Notably, adverse effects associated with traditional medicine are often not well documented.
In this report, we present a case of toxic hepatitis following occasional consumption of a stem aqueous extract of the T. crispa. Hepatitis induced by T. crispa is established, based on absence of medical history, clinical exam and biological results. We noted that the patient had been exposed to several pesticides the next day after the last ingestion of the aqueous extract. Yet, to the best of our knowledge acute cutaneous or respiratory exposure to those pesticides (containing myclobutanil, glyphosate, acetamipride, abamectin and linuron) could not explain the onset of hepatitis. Hepatitis was reversible and biological hepatic enzymes returned to normal after a few weeks without any specific treatment. This conclusion was corroborated by calculation of Roussel Uclaf Causality Assessment Method (RUCAM) score. With a calculated score of +6, the causal relationship between T.c. aqueous extract intake and hepatotoxicity was "probable" 15 .
Two previous cases of acute hepatitis have been reported with T. crispa but with a different method of administration consisting of chronic use of tablets or pellets of the plant. The first case of hepatitis concerned a 37-year-old woman who had consumed T. crispa tablets (bought in Indonesia) for 10 weeks 3 . The second case concerned a 49-year-old man who had orally taken pellets of T.crispa (bought on a Vietnamese market 4 ) for 4 weeks. In both cases, hepatitis was reversible after a few weeks. These case reports are in accordance with changes in biochemical parameters observed during some clinical trials assessing the effect of T. crispa in diabetic patients and in patients with metabolic syndrome. Marked elevation of liver enzymes (that returned to normal after discontinuing T. crispa) has been observed in 2 of the 20 patients treated for 6 months with a capsule form at a dosage of 1 gram thrice daily 16 . Similarly, a double-blind, placebo-controlled trial using a crossover design found an elevation of more than 3 times baseline levels of ALT and AST in 6 of the 36 patients who received 250 mg T. crispa dry powder capsule twice a day for 2 months 17 .
We hypothesize that two mechanisms may have contributed to hepatotoxicity induced by T. crispa, considering the close structural similarity observed between borapetosides and i. furanoditerpenoids like 8-Epidiosbulbin E acetate (EEA) and ii. teucrin A or teuchmaedryn A. The first mechanism is direct liver toxicity induced by metabolic activation of T. crispa furanoditerpenoids. Such a mechanism of toxicity has been described with EEA, a norclerodane furanoditerpenoid structurally very similar to borapetosides 18 . Cytochrome P450 metabolic activation of the furan moiety of EEA may be responsible for the formation of electrophilic species, leading to a dose-dependent hepatotoxicity. Other similar examples of bioactivation of furanoditerpenoids correlated with hepatotoxicity are reported in literature 18,19 . The second toxicological mechanism may be idiosyncratic, as it has been reported that only a few people exposed may develop hepatitis 16,17 . Such a mechanism may be suspected in our case, as hepatitis occurred after reexposure to T. crispa and was associated with fever, a typical symptom of idiosyncratic immunoallergic toxic hepatitis. Moreover, there is a close structural analogy between borapetosides and teucrin A or teuchmaedryn A, two clerodanes found in Teucrium chamaedrys L. (Lamiaceae). In-vivo and in-vitro studies have demonstrated that those teucrin A or teuchmaedryn A induce hepatotoxicity, consecutive to both direct toxicity and a secondary immune reaction with antoantibody formation 20 . Future studies using isolated compounds will be necessary to ascertain the exact mechanism of toxicity of the furanoditerpenoids from Tinospora crispa stems.
The present UHPLC-ESI-QTOF-MS/MS metabolomic investigation performed on the aqueous stem extract of T. crispa led to the detection of 18 furanoditerpenoids structurally related to the major compound, namely borapetoside C (1). A targeted analysis of MS/MS data was facilitated by introducing a molecular networking approach in the data analysis workflow. MN is becoming one of the most efficient tools for the analysis of untargeted MS/MS data, allowing dereplication of complex extracts and exploration of molecular diversity 21 . We have showed here that MN is also valuable for targeted metabolite fingerprinting. To the best of our knowledge this is the first example of application of MN for investigating plant suspected of being toxic and toxins thereof.
Hepatic trouble should be put in perspective with the potential hypoglycaemic effect of the plant. Promising results have been obtained by in vitro and in vivo tests with T. crispa. Noor et al. have observed insulin secretory rates in alloxan-diabetic rats and insulin release from rat islets and HIT-T15 beta cells in vitro with the extract of stem plant 22 . Since this publication, its hypoglycaemic mechanism of action has been attributed to diterpenoids (borapetol B, borapetoside A and C) isolated from the plant that could stimulate insulin release from pancreatic β-cell. Using an oral glucose tolerance test, an increase of 2-fold of plasma insulin release has been observed after oral administration of borapetol B to Wistar rats and spontaneously type 2 diabetic Goto-Kakizaki rats compared to placebo 23 . Similarly, administration of borapetoside A to diet-induced type 2 diabetes mellitus mice lowered the plasma glucose level in a dose-depended manner with results similar to the administration of metformin for the dose range between 0.3 and 10 mg/kg 24 . A potential mechanism of improvement of peripheral glucose uptake, notably via an increase in glucose utilization of skeletal muscle and liver, has also been advanced following in vitro tests [24][25][26] . To date, these promising non-clinical studies have unfortunately not been confirmed by clinical trials applied especially to type 2 diabetic patients. Indeed, Klangjareonchai et al. showed that glucose and insulin areas under the curve were not different with or without ingestion of 125 or 250 g of T. crispa dry power capsule 27 , while Sangsuwan et al. in a randomized, double blind, placebo-controlled trial found no difference on HbA1c or fasting plasma glucose between patients treated with 1 gram thrice daily Tinospora crispa powder in capsule for 6 months and patients treated with placebo 16 .
In total, the Benefit-Risk of the use of Tinospora crispa for the treatment of type 2 diabetes appears negative at this step, despite the hope raised by non-clinical studies. There remain uncertainties on the efficacy on blood glucose control in real-life setting while the risk of hepatic trouble is established, both for an acute and for a chronic use and independently of the method of administration (stem in powder capsule or used as an aqueous extract). Furthermore, toxic mechanisms may associate both dose-response relationship and idiosyncratic effects, therefore making therapeutic use of T. crispa hazardous, in so far as diabetic patients are often treated simultaneously with statins, drugs known to potentially increase liver enzymes.

Conclusion
Chronic or occasional use of T. crispa stem could induce toxic hepatitis, reversible after a few weeks without any specific treatment. Despite its promising results suggesting an increase in insulin release in non-clinical studies, its traditional use should be avoided for diabetic patients for lack of demonstrated benefit obtained by dedicated well-controlled clinical trials. Several furanoditerpenoids have been detected and putatively identified by UHPLC-ESI-QTOF-MS/MS in an aqueous extract of fresh stems that mimics the traditional preparation by combining classical data exploration and molecular networking approach. Even in non-optimized conditions of data acquisition, molecular networking constitutes a powerful and useful tool facilitating the data filtering. Further studies are still needed to confirm the putative toxicity of furanoditerpenoids and to elucidate subjacent mechanisms.

Materials and Methods
Questioning and biochemical analyses. Informed consent was obtained from the patient. The Faculté de Pharmacie de Paris, Université Paris Descartes approved the experimental protocol and it was carried out in accordance with all relevant guidelines and regulations. Biochemical analyses were performed by standard methods using automated techniques.

UHPLC-HRMS and MS 2 experiments.
Separations were performed using an Ultimate 3000 RSLC system equipped with a binary pump, an autosampler and a thermostated column compartment, equipped with a diode array detector (195-800 nm) (Dionex, Germering, Germany). Components were separated on a C18 Luna Omega column of 150 mm × 2.1 mm with a particle size of 1.6 μm (Phenomenex, Le Pecq, France). The mobile phase was made up of 0.1% formic acid in water (phase A), and 0.08% formic acid in acetonitrile (phase B). A solvent gradient was applied as follows: 0-0.1 min: 3% B, 0.1-26 min: 3-80% B, 26-26.5 min: 80-95% B, 26.5-29.5 min: 95% B, 29.5-30 min: 95-3% B, and finally 30-33 min: 3% B. The column was introduced in a thermostated compartment heated at 40 °C. All solutions were prepared in MeOH (2 mg/mL) and filtered through 0.2 μm nylon filter disk prior the injection. The injection volume was set at 1 µL and the flow rate was set at 500 μL/min. MS experiments were carried out on a maXis UHR-Q-TOF mass spectrometer (Bruker, Bremen, Germany) in positive electrospray ionization (ESI) mode. Capillary voltage was set at +4.5 kV. The flows of nebulizing and drying gas (nitrogen) were respectively set at 2.0 bar and 9.0 L/min and drying gas was heated at 200 °C. Mass spectra were recorded in the range 50-1650 m/z. MS/MS experiments were conducted using data dependent acquisition (DDA) mode (auto-MS/MS) in a mass window from m/z 150-1200. Three precursor ions with intensities higher than 400 au were selected per fragmentation cycle among the most intense ions to be fragmented. These three precursor ions were allowed to be selected for two consecutive cycles and were then placed on an exclusion list for 0.05 min. The collision energy was set at 35 eV and was applied as follows: 88% of the collision energy was applied during half of the fragmentation cycle, and 117% of the collision energy was applied during the half remaining cycle time. Raw data were converted to mzXML format using CompassXport software (Bruker, Bremen, Germany) and then processed using MZmine version 2 28 .
Molecular Network Analysis. A molecular network was created using the online workflow at GNPS (https:// gnps.ucsd.edu/ProteoSAFe/static/gnps-splash.jsp) 6 . The data was filtered by removing all MS/MS peaks within +/−17 Da of the precursor m/z. MS/MS spectra were window filtered by choosing only the top 6 peaks in the +/−50 Da window throughout the spectrum. The data was then clustered with MS-Cluster with a parent mass tolerance of 0.05 Da and a MS/MS fragment ion tolerance of 0.05 Da to create consensus spectra 29 . Further, consensus spectra that contained less than 1 spectrum were discarded. A network was then created where edges were filtered to have a cosine score above 0.6 and more than 2 matched peaks. Further edges between two nodes were kept in the network if and only if each of the nodes appeared in each other's respective top 20 most similar nodes. The spectra in the network were then searched against GNPS' spectral libraries. The library spectra were Scientific RepoRTS | (2018) 8:13520 | DOI:10.1038/s41598-018-31815-6 filtered in the same manner as the input data. All matches kept between network spectra and library spectra were required to have a score above 0.7 and at least 6 matched peaks. Molecular network parameters are available at https://gnps.ucsd.edu/ProteoSAFe/status.jsp?task=2817c0f504f741f781f645eb3ee27d95.