The structures of the test compounds were searched for in the PubChem database (https://pubchem.ncbi.nlm.nih.gov/). The compounds’ Simplified Molecular-Input Line-Entry System strings and Chemical Abstracts Service (CAS) numbers were obtained for the in silico evaluation. We considered the prediction of potentially mutagenic impurities using two in silico methods, an expert rule-based method, and a statistics-based method, as recommended by ICH M7 guidelines (ICH 2017).

We used more than two models to consolidate expert decisions and increase confidence in the predictions. Using additional models can aid in evaluating prediction accuracy and increase confidence in results. Additionally, software, such as QSAR Toolbox, was used to evaluate structural alerts in the molecules. The presence of detailed comments discussing the understanding of the hazard presented by the chemical class can also contribute to the accuracy of the assessment. These comments provide information relevant to the review, such as mitigating features. Therefore, multiple genotoxicity endpoints for vildagliptin and its impurities were evaluated using the following software considered on statistically-based (SB), expert rule-based (RB), or structural alert detection (SA) (Tugcu et al. 2021): US EPA TEST (v.5.1.1) (Toxicity Estimation Software Tool. Washington, DC: United States Environmental Protection Agency) (SB, RB), VEGA-QSAR v.1.2.3 (www.vega-qsar.eu) (SB, RB), ToxRead (v.0.23) (RB, SA), OCHEM Open predictor (Sushko et al. 2011) (SB), ProTox3.0 (Banerjee et al. 2018) (SB), Danish QSARDB (http://qsardb.food.dtu.dk/db/index.html) (SB, RB), and QSAR Toolbox v.4.5 (http://www.qsartoolbox.org/)(SA). Some of these software for toxicity prediction use machine learning algorithms in their QSAR models (e.g., ProTox 3.0), while others are based on read-across methodology (e.g., ToxRead).

Vildagliptin cyclic amidine (impurity E, CAS No. 1789703–37-2, purity 98%, Batch No. VL/I-927), vildagliptin diketopiperazine (impurity F, CAS No. 1789703–36-1, purity 97%, Batch No. VL/I-931), and vildagliptin amide (CAS No. 565453–39-6, purity 94%, Batch No. VL/I-700) were kindly provided by Helba Pharma company (Istanbul, Türkiye).

Salmonella typhimurium bacterial strains and the post-mitochondrial fraction (S9) prepared from rat liver were supplied by the Moltox Molecular Toxicology, Inc. (NC, USA). 2-aminofluorene was obtained from Merck (Hohenbrunn, Germany), and nutrient broth was supplied by Hi Media Laboratories Ltd (Mumbai, India). Histidine was provided by Fluka (USA).

The standard plate incorporation Ames test was performed according to the protocol outlined in Organisation for Economic Co-operation and Development (OECD) No. 471 (2020) and Maron and Ames (1983), and in an accredited laboratory by TÜRKAK TS EN ISO/IEC 17025 (No: AB-1764-T). Five histidine-auxotrophic S. typhimurium strains (TA98, TA97a, TA100, TA102, and TA1535) were cultured following the respective provider’s instructions. The assay was performed in triplicate, as two independent experiments, in the presence and absence of an external metabolic activation system from Aroclor™ 1254-treated rats (S9 fraction), and at five concentrations (1–5000 µg/plate). Appropriate vehicle controls (methanol: 50 µL/plate) and positive controls (without S9 fraction: 20 µg/plate 4-nitro-o-phenylenediamine for TA98 and TA97a, 1 µg/plate sodium azide for TA100 and TA1535, 0.5 µg/plate mitomycin-C for TA102, and with S9 fraction: 5 µg/plate 2-aminofluorene for TA98, TA97a and TA100, and 10 µg/plate 2-aminoanthracene for TA102 and TA1535) were included in both assays. A positive response was defined as a number of revertant colonies that was at least twice (1.5 for TA102) the number of colonies observed in the negative control.

Notably, for the sake of transparency, it is important to highlight that in the initial Ames test, Vildagliptin amide showed inconclusive results in the TA97a strain with S9 at concentrations of 1 and 10 µg/plate. To ensure reproducibility, follow-up experiments were conducted with all three impurities under the same conditions. These tests confirmed that none of the impurities exhibited mutagenic potential, in line with OECD 471 guidelines. While the follow-up experiments utilized historical positive control data from the initial assay rather than new concurrent controls, the historical values demonstrated sufficient effect size to validate the findings. Additionally, variability in the initial results may have stemmed from conducting assays on separate days, which required averaging control values. To address this, the follow-up experiments were performed on the same day to enhance consistency. The updated data, now included in Table 2, reaffirm the non-mutagenic nature of the investigated impurities.

The MNT was performed following the OECD guideline (No. 487) for testing chemicals (OECD 2023) in an accredited laboratory (TÜRKAK TS EN ISO/IEC 17025 (No: AB-1764-T)).

Chinese hamster ovary (CHO-K1) cell line (ATCC; CCL-61) was cultured in Ham’s medium F12 supplemented with 10% (v/v) fetal bovine serum (FBS) and 1% (v/v) antibiotics (10,000 U/ml penicillin and 50 mg/ml streptomycin). Cells were cultured at 37°C in a humidified incubator with 5% CO2 and were passaged twice a week using 0.25% trypsin solution. Cell culture reagents were obtained from (Gibco, Carlsbad, CA).

Briefly, cells were seeded in 6-well plate at a density of 2 × 105 cells/well for 24 h. After the cells were attached to the surface, they were incubated with different concentrations (100–500 µg/ml) of the compound. After 24 h of exposure, cytochalasin B (3 µg/mL) was added to each well to inhibit cytokinesis for 24 h. Cells were then harvested and fixed twice with methanol acetic acid (3:1) solution. The fixed cell suspension was spread onto pre-cleaned microscope slides and left to air-dry. Cells were stained for 5 min with 5% Giemsa in Sorensen Buffer (v/v). Subsequently, the microscope slides were rinsed in tap water and left to air-dry at room temperature overnight. The potential of the metabolites of vildagliptin impurities to induce micronuclei was also evaluated in the presence of an external metabolic activation system from AroclorTM1254-treated rats (S9-fraction, final concentration of 0.34%) after a 4-h exposure in CHO cells. Appropriate vehicle controls (DMSO: 1%) and positive controls (24 h without S9-fraction: ethyl methanesulfonate, 4 h with S9-fraction: benzo[a]pyrene) were included in all experiments. The number of micronuclei was established by analyzing 1000 binucleated cells per culture per treatment, following the methods previously described by Fenech (2000). Additionally, cells were assessed for the Nuclear Division Index by scoring cells with 1–4 nuclei, using the formula: Nuclear Division Index = M1 + 2(M2) + 3(M3) + 4(M4)/N, where M1–M4 represent the number of cells with 1–4 nuclei, and N is the total number of viable cells analyzed.

The cytotoxicity of the samples was analyzed to establish the concentration levels for MNT. The results showed that all tested samples were not cytotoxic up to 1000 µg/ml. Therefore, the highest concentration of 500 µg/ml was selected for MNT according to the ICH S2 guidelines (2012).

Group comparisons were conducted using SPSS 20 software, with experimental results expressed as mean ± standard deviation (SD). In vitro experiment outcomes were represented as the mean of triplicate measurements ± SD. For comparisons between positive controls and test groups with negative controls in in vitro assays, Dunnett’s multiple comparison test was applied (Dunnett and Crisafio 1995). p < 0.05 was considered statistically significant.