Toxicological assessment of SGLT2 inhibitors metabolites using in silico approach

: Sodium-glucose cotransporter 2 inhibitors (SGLT2i) are the latest class of drugs approved to treat type 2 DM (T2DM). Although adverse effects are often caused by a metabolite rather than the drug itself, only the safety assessment of disproportionate drug metabolites is usually performed, which is of particular concern for drugs of chronic use, such as SGLT2i. Bearing this in mind, in silico tools are efficient strategies to reveal the risk assessment of metabolites, being endorsed by many regulatory agencies. Thereby, the goal of this study was to apply in silico methods to provide the metabolites toxicity assessment of the SGLT2i. Toxicological assessment from SGLT2i metabolites retrieved from the literature was estimated using the structure and/or statistical-based alert implemented in DataWarrior and ADMET predictor TM softwares. The drugs and their metabolites displayed no mutagenic, tumorigenic or cardiotoxic risks. Still, M1-2 and M3-1 were recognized as potential hepatotoxic compounds and M1-2, M1-3, M3-1, M3-2, M3-3 and M4-3, were estimated to have very toxic LD 50 values in rats. All SGLT2i and the metabolites M3-4, M4-1 and M4-2, were predicted to have reproductive toxicity. These results support the awareness that metabolites may be potential mediators of drug-induced of the therapeutic agents.


INTRODUCTION
Diabetes Mellitus (DM) is a chronic progressive metabolic disorder with an increasing prevalence worldwide. Type 2 DM (T2DM) is the most common, accounting for around 90% of diabetes cases. T2DM is a non-insulin dependent DM, caused by insulin decreased sensitivity of target tissues (WHO 2020). As a result of this, blood glucose concentration increases, promoting increased excretion of both sodium and glucose in the urine (Guyton & Hall 2006).
The adverse events of SGLT2i include symptomatic hypotension, hypoglycemia, urinary tract infections, and mycotic infections (Halimi & Vergès 2014). Furthermore, DM is associated with chronic liver disease increase associated with cirrhosis, non-alcoholic fatty liver disease, alcoholic cirrhosis, chronic hepatitis C (CHC), and hemochromatosis (Li et al. 2019, Kita et al. 2007. Previous reports also pointed out that T2DM is associated with an increased incidence of overall cancer (Tsilidis et al. 2015, Yuan et al. 2020. Phase III clinical trials with dapagliflozin (2) reported an imbalance of bladder cancer in men and breast cancer in women, which delayed its FDA approval (Burki 2012, Scheen 2014). An increased risk of bladder cancer was also observed in the individuals taking either empagliflozin (3) or dapagliflozin (2) (Tang et al. 2017, Scheen 2014. Corroborating these findings, a recent analysis reported a high number of cases of bladder cancer among users of SGLT2i (García et al. 2021). These adverse effects are a significant public health problem and a challenge for the pharmaceutical industry. Thus, further studies are required to support these pieces of evidence related to the SGLT2i therapy and long-term outcomes (Shao et al. 2020).
It is well-established that adverse effects often occur due to a metabolite rather than the drug itself (Park et al. 2001, Thompson et al. 2016, Mumtaz & Durkin 1992, Luffer-Atlas & Atrakchi 2017. Regulatory agencies recommend performing the safety assessment only for disproportionate drug metabolites, i.e., those present at > 10% of total drug-related human exposure at steady-state, while no tests are performed for the remaining metabolites. This lack of toxicological data is of particular concern since their contribution to the parent drug's overall toxicity remains unknown, particularly for metabolites of chronic use drugs (FDA 2020, Luffer-Atlas & Atrakchi 2017). Nowadays, computational methods play a vital role in the safety assessment of molecules with challenging isolation, quantification, or synthesis. In silico toxicology is one of the alternatives to animal testing to toxicity assessment that uses computational resources to organize, analyze, model, simulate, visualize, or predict the toxicity of chemicals (Raies & Bajic 2016). The use of computer-based models using machine learning and structural alert to predict toxicity has increased significantly due to improvements in the performance of the models and their ease of use (De Mello et al. 2018, Myatt et al. 2018, Graham et al. 2021. Also, in silico studies are being endorsed by regulatory agencies, as they are typically based on human data, with an enhancement of interspecies transferability. Mutagenicity, carcinogenicity, acute oral toxicity, liver adverse effects, allergenic skin sensitization, and reproductive toxicity are some of the toxicity endpoints usually estimated for chemicals (Archibald et al. 2018, Vedani & Smiesko 2009. In this context, in silico methods is an efficient strategy to assess the safety profile of SGLT2i metabolites in humans.
T h e m u t age n i c i t y, h e pa to to x i c i t y, cardiotoxicity, reproductive toxicity and acute toxicity endpoints were evaluated using statistical-based models implemented in ADMET predictor TM version 9.5 (Simulations Plus, Inc., Lancaster, CA, USA, 2019).
The mutagenicity endpoint was predicted based on the Ames Test. This test applied models of Artificial Neural Network Ensembles as qualitative models for five strains of Salmonella (TA97 or TA1537, TA98, TA100, TA102, and TA1535 strains) with and without microsomal activation (Bakhtyari et al. 2013, Honma et al. 2019. Hepatotoxicity parameters were studied using five relevant biomarkers: alkaline phosphatase (ALP), serum glutamic oxaloacetic transaminase (SGOT), serum glutamic pyruvic transaminase (SGPT), gamma-glutamyl transferase (GGT), and lactate dehydrogenase (LDH) (Abreu et al. 2020, Garcia et al. 2021). The Cardiotoxicity model predicted the likelihood that a compound will block the hERG channel, related to ventricular arrhythmias and sudden death (Garcia et al. 2021). The reproductive toxicity endpoint includes anything that disturbs the reproductive process of organisms, including adverse effects to sexual organs, behavior, ease of conception, teratogenicity, and developmental toxicity to offspring before or after birth. Lastly, the acute toxicity model predicted the capability of the amount of orally administered chemical (mg/ kg body weight) required to kill 50% of the rats population within 24 h of exposure (LD 50 ).
Besides statistical-based models, an expert rule-based method was also performed to evaluate mutagenicity and tumorigenicity using DataWarrior software (Sander et al. 2015, Guerra et al. 2017. The mutagenicity prediction includes 20 specific test systems, both in vitro and in vivo, with tested organisms including bacteria, molds, yeast, protozoa, insects, and mammalian cell lines. The tumorigenic effect data is predicted considering three criterias: carcinogenic by RTECS (Registry of Toxic Effects of Chemical Substances), neoplastic by RTECS, and equivocal tumorigenic results (Von Korff & Sander 2006, CDC 2011.

Compilation of SGLT2i metabolites
In vitro metabolic profiles from liver microsomes and hepatocyte incubations can be poor predictors of in vivo circulating major human metabolites (Luffer-Atlas & Atrakchi 2017). Thus, we retrieved from literature data of metabolic profile from SGLT2i after single oral dose administration to healthy humans (Mamidi et al. 2014, Obermeier et al. 2010, Kasichayanula et al. 2014, Chen et al. 2015, Miao et al. 2013. The in vivo biotransformation of empagliflozin (3) resulted in six metabolites described by Chen et al. (2015), but one of them is related as an oxidation/dehydrogenation metabolite and did not have its structure completely elucidated. Thereby, two metabolites were products from Phase I metabolism (M3-1 and M3-2) and three of them from glucuronidation by UGT1A9 (M3-3, M3-4, and M3-5) ( Figure 4).

In silico toxicity assessment of metabolites
The in silico toxicity assessment of SGLT2i and its metabolites were performed using two methodologies, that complement each other, a statistical-based and an expert rule-based, from ADMET Predictor TM (Simulations Plus 2019) and DataWarrior, respectively (Sander et al. 2015, Guerra et al. 2017 (Table I). Statistical-based models apply machine learning algorithms to analyze the correlations between molecular structures and biological activity. Expert rulebased models find the compounds most similar to the parent compound based on similarity while leaving the selected structure untouched using fragments from known drugs (Goel & Valerio Jr 2020, Sander et al. 2015. The mutagenicity predictions indicated that neither the drugs nor their metabolites might be mutagenic, considering results from the statistical-based model (Table I). In the expert rule-based analysis, ertugliflozin (4) and its metabolites presented mutagenic potential (Table I). Also, DataWarrior results indicated that all SGLT2i and metabolites are not tumorigenic based on RTECS criteria (Table I).

DISCUSSION
Nowadays, the increasing interest and acceptance of in silico methods are providing their inclusion for regulatory purposes. Based on the FDA recommended studies for assessing the safety of the disproportionate metabolites (FDA 2020), we conducted mutagenicity, tumorigenicity, and general toxicity studies for the SGLT2i metabolites using established in silico approaches (National Research Council 2007). Mutagenicity evaluations were carried out using two different approaches: statisticalbased and expert-system rule-based (ICH 2015). The statistical-based model presented nonmutagenicity results for all drugs and their metabolites corroborating the data from drug registration (Table I). In the expert rule-based analysis, ertugliflozin (4) and its metabolites (M4-1 to M4-4) presented structural alert, which suggest the potential mutagenicity of the dioxabicyclo[3.2.1]octane moiety. Indeed, non-mutagenic data was attributed to this group in the literature and the statistical-based method does not indicate this toxicity. In order to rationalize the final conclusion, the expert knowledge was applied to support this evidence, deciding that non-mutagenic results were linked to these metabolites (ICH 2015).
Carcinogenicity studies should be conducted on metabolites of drugs that are used regularly for at least 6 months or for the treatment of chronic diseases. Hence, we also conducted carcinogenicity evaluation of SGLT2i  , Tang et al. 2017. For this endpoint, SGLT2i and its metabolites were outside the applicability domain (AD) of the carcinogenicity model implemented in ADMET Predictor TM . In other words, these molecules were out-of-scope and the predictions were not considered due to its low reliability (Simulations Plus 2019, Ruiz et al. 2017, El-Saadi et al. 2015. None of the SGLT2i metabolites were predicted to interact with the hERG potassium channel. These data may be supported by the clinical evidence for cardioprotective effects of SGLT2i (Table II)  Evidence for the cardiovascular benefits of SGLT2i continues to accumulate. Treatment with SGLT2 inhibitors has been linked to a reduced risk of heart failure and cardiovascular death in clinical trials (Sayour et al. 2021).
While some parent drugs can directly cause hepatotoxicity, it is generally the metabolites of these compounds that lead to liver injury (Tarantino et al. 2009). Thus, it is crucial to evaluate the hepatotoxicity of SGTL2i metabolites. Due to SGOT and SGPT increased levels, canagliflozin and empagliflozin metabolites (M1-2 and M3-1, respectively) are potential hepatotoxic compounds (Table II). Indeed, in multiple large randomized controlled trials, the hepatotoxicity of SGLT2i was unproven but suspected to be a rare cause of clinically apparent liver injury due to serum enzyme elevations (Livertox 2012). Furthermore, it has been suggested that for every 10 SGPT cases reported in a clinical trial, there will be one case of more severe liver injury. This case develops once the drug is of chronic use (Bell & Chalasani 2009). Since the canagliflozin (1) and empagliflozin (3) metabolites (M1-2 and M3-1, respectively) cause hepatotoxicity, it may affect the metabolism of other drugs (Bell & Chalasani 2009, Tarantino et al. 2009). In addition to M1-2 hepatotoxicity, the starting dose of canagliflozin (1) is 100 mg/day orally (Janssen Pharmaceutical Companies 2013). Therefore, these outcomes lead to a possible safety concern, as daily doses of ≥ 50 mg are significantly more likely to cause liver injury (Lammert et al. 2008 Ingelheim Pharmaceuticals 2014), which leads to a less relevant safety concern compared with canagliflozin metabolite. The acute toxicity endpoint predicts the amount of orally administered substance (in mg/kg body weight) required to kill 50% of the rats tested (LD 50 ) (Simulations Plus 2019). The rat oral LD 50 ADMET Predictor TM model is supported by data from two sources, CDC's Registry of Toxic Effects of Chemical Substances (RTECS), and the ChemIDplus database (Ruiz et al. 2012). One of the most common scales used for the final interpretation of acute rat toxicity results is the Hodge and Sterner scale (Hodge & Sterner 2005, Erhirhie et al. 2018, Ruiz et al. 2012. According to toxicity classes of Hodge & Sterner (2005), the O-glucuronide metabolites of canagliflozin M1-2 and M1-3, empagliflozin (3) and its metabolites M3-1, M3-2 and M3-3, and ertugliflozin metabolite M4-3 were estimated to have worrying toxicity, with LD 50 values <500mg/kg. The remaining SGLT2i and its metabolites were estimated to have moderately toxic level, with LD 50 values (500-5000 mg/kg) ( Table II). The Globally Harmonized System (GHS) of Classification and Labelling of Chemicals criteria are used to determine the nature and the relative severity of the hazard of a chemical substance or mixture. According to GHS criteria, chemicals are assigned to one of the five toxicity categories on the basis of LD 50 (oral, dermal) or LC 50 (inhalation) from acute toxicity studies (United Nations 2021). The LD 50 of the O-glucuronide metabolites of canagliflozin M1-2 and M1-3 places them within the category 3 of the Organization for Economic Cooperation and Development (OECD) GHS classification system (LD 50 <300 mg/kg), while the SGLT2i and the remaining metabolites were classified in the less severe hazard category 4 (300 < LD 50 < 2000 mg/kg). Furthermore, the European Union regulation is currently furthering alternative toxicity studies that decrease the use of animals.
All SGLT2i, the empagliflozin metabolite M3-4, and the ertugliflozin metabolites M4-1 and M4-2 presented reproductive toxicity risk (Table  II). Clinical trial evidence (Monami et al. 2014, Rizzi & Trevisan 2016, Zaccardi et al. 2016 and post-marketing safety analysis corroborates these findings (Raschi et al. 2017). According to this analysis, SGLT2i are associated with a high report of reproductive adverse effects in the international pharmacovigilance databases. Signals of reproductive events in the postmarketing analysis were largely in agreement with data obtained from pre-approval RCTs (Randomized control trials) (Monami et al. 2014, Rizzi & Trevisan 2016, Zaccardi et al. 2016. In addition to this clinical evidence, these results reinforced the guidance that prescribers should be aware of these common safety issues and should monitor patients to avoid them. These results support the awareness that metabolites may be potential mediators of drug-induced toxicities of the therapeutic agents and should be structurally and toxicologically characterized.

CONCLUSIONS
It is well-established that metabolite rather than the drug itself causes adverse reactions. Herein, we showed that all SGLT2i and its metabolites were non-tumorigenic, non-mutagenic and non-cardiotoxic. However, particular attention must be given to canagliflozin and empagliflozin metabolites M1-2, M1-3, M3-1, M3-2 and M3-3, which present hepatotoxicity (M1-2 and M3-1) and high LD 50 values according to Hodge and Sterner scale (M1-2, M1-3, M3-1, M3-2 and M3-3). In agreement with clinical trials evidence and post-marketing analysis, all SGLT2i, the empagliflozin M3-4, and ertugliflozin M4-1 and M4-2 were predicted to have reproductive toxicity. Concerning these endpoints, our in silico results support the awareness that even minority metabolites may be potential mediators of drug-induced toxicities, especially for drugs of chronic use, and thus should be evaluated by sponsors and regulators.