Introduction

Methylglyoxal (MGO), a highly reactive α-oxoaldehyde compound, is extensively found in various foods and beverages, particularly those rich in sugar and fat, as well as in fermented products1,2. MGO is generated in food through several processes such as the Maillard reaction, autoxidation of sugars, caramelization, lipid oxidation, and/or microbial fermentation during thermal processing and food storage3,4. In addition to these exogenous sources, MGO is also endogenously formed as a by-product, primarily from glycolysis through the spontaneous degradation of triosephosphate intermediates5. In a healthy adult human, it has been estimated that 3 mmol MGO is formed endogenously each day6. However, the body maintains MGO at relatively low levels, with concentrations reported to be in the range of 0.05–0.6 μM in plasma and 1–4 μM in cells, primarily through the GSH-dependent glyoxalase system consisting of the enzymes glyoxalase 1 (Glo1) and glyoxalase 2 (Glo2), which detoxifies MGO into d-lactate (Fig. 1)7,8,9,10. Elevated in vivo levels of MGO can occur under certain pathological conditions, such as hyperglycemia, that disrupt the balance between the formation and detoxification of MGO10. It was reported that the plasma levels of MGO in type 2 diabetic patients can be up to 30% higher than in healthy individuals7.

Fig. 1: Schematic presentation of the intracellular detoxification of MGO by the GSH-dependent glyoxalase system.
figure 1

MGO can react with the thiol group of GSH both non-enzymatically and through a reaction catalyzed by GSTs to form a hemithioacetal, which can subsequently be converted by Glo1 and Glo2 into D-lactate. GSH glutathione, GSTs glutathione‐S‐transferases, Glo1 glyoxalase 1, Glo2 glyoxalase 2.

Full size image

MGO is a highly reactive precursor for advanced glycation end products (AGEs), whose formation has been linked to various chronic diseases3,11. Elevated in vivo levels of MGO can increase AGE formation by interacting with the nucleophilic sites on DNA and proteins, leading to structural and functional changes in these macromolecules, which contributes to cellular and tissue dysfunction and ultimately could play a role in various diseases, including diabetes and multiple neurodegenerative disorders (e.g., Alzheimer’s disease and Parkinson’s disease)3,10,12. This concern is exacerbated by the dietary intake of exogenous MGO and the resulting AGE formation, which may further increase human exposure and the potential related health risks13. However, the extent to which dietary MGO contributes to the total body burden and impacts human health remains unclear.

Clinical studies have shown that higher serum MGO levels are associated with accelerated cognitive decline in older adults14. In Alzheimer’s disease patients, elevated levels of MGO and the resulting AGEs have been detected in the cerebrospinal fluid15,16. Similarly, elevated levels of MGO-derived AGEs have been identified in the nigra neurons of Parkinson’s disease patients17. These clinical findings suggest a pathological role for MGO in the context of neurodegeneration. Moreover, various in vitro studies have shown that exposure to exogenous MGO can induce cellular injury and toxicity in different animal and human neuronal cell models18,19,20,21,22,23. Neuronal cells, in particular, are susceptible to the toxic effects of MGO, a vulnerability that is potentially related to their high-energy glucose metabolism, combined with relatively lower levels of glutathione (GSH) and lower activities of glyoxalase enzymes compared to what is found in astrocytes24,25,26,27. Exposure of neuronal cells to exogenous MGO has been observed to decrease cell viability and mitochondrial redox activity, as well as to increase reactive oxygen species (ROS) production and apoptosis in in vitro studies20,28,29. Furthermore, the activation of the receptor for AGEs (RAGE) by AGEs formed in a reaction of DNA and proteins with MGO, is potentially linked to these adverse effects29,30,31. These in vitro findings reveal the neurotoxicity of MGO and suggest mechanisms underlying its toxicity that may play a role in neurodegeneration. Apart from its well-documented neurotoxicity, the genotoxic potential of MGO represents another critical concern. Studies have shown that DNA adducts induced by MGO could increase the mutation rate in E. coli cells, as well as G:C to C:G and T:A transversions in the supF gene of simian kidney cells32,33. Besides, it was found that MGO-derived DNA adducts N2-(1-carboxyethyl)-2’-deoxyguanosine (N2-CEdG) exhibited a lesion rate of one per 107 nucleosides at an endogenous background level of MGO in WM-266-4 human melanoma cells, and exposure to exogenous MGO resulted in a concentration-dependent increase in these N2-CEdG levels34.

However, these in vitro neurotoxicity and genotoxicity data cannot be applied directly for risk assessment in humans. The concentration-response data obtained from these in vitro assays require further translation into dose-response data to determine safe exposure levels in humans. This translation can be achieved using new approach methodologies (NAMs) such as quantitative in vitro–in vivo extrapolation (QIVIVE) using physiologically based kinetic (PBK) modeling and reverse dosimetry, methods proven effective in predicting in vivo toxicity of various compounds for a variety of endpoints35,36,37. The PBK modeling facilitated reverse dosimetry approach contributes to the development of non-animal-based methods and offers a predictive tool to assess whether systemic exposure of the host to MGO might result in internal concentrations that could cause toxicity. This study aimed to use available in vivo kinetic mouse data, in vitro neurotoxicity and genotoxicity data, and in vitro–in silico-based reverse dosimetry for quantitative translation of concentration-response curves for the toxicity of MGO into in vivo dose-response predictions in humans. This can help in understanding the effects of exogenous versus endogenous MGO in humans.

In this study, the first step was to develop a mouse PBK model and evaluate the model predictions with the literature available experimental data for blood levels of MGO38. Based on the evaluated mouse model, a human PBK model was developed and applied to translate the concentration-response data from the literature for the toxicity of MGO in human neuronal cells and human melanoma cells to predict in vivo dose-response curves for the neurotoxicity and genotoxicity of MGO in humans. Based on these predicted in vivo neurotoxicity and genotoxicity data, a risk assessment was performed using the margin of exposure (MOE) approach to evaluate the risks from dietary and endogenous MGO.

Methods

PBK modeling-facilitated reverse dosimetry approach

The PBK modeling-facilitated reverse dosimetry approach was applied to predict in vivo dose-dependent neurotoxicity and genotoxicity induced by MGO. This approach involved the following steps: (1) development of a mouse PBK model for MGO to allow the prediction of the in vivo kinetics of MGO in mice, (2) evaluation of the mouse PBK model by comparing predicted kinetic data for MGO with available in vivo literature data, (3) defining a human PBK model for MGO based on the mouse PBK model developed, (4) extrapolation of the in vitro concentration-response data from the literature for MGO neurotoxicity and genotoxicity to in vivo dose-response curves using human PBK model-facilitated reverse dosimetry, and (5) risk assessment of MGO from daily dietary intake and endogenous formation via an MOE approach based on the predicted in vivo neurotoxicity and genotoxicity data for MGO.

Development of a PBK model for MGO in mice

We utilized our previously developed PBK models for 17β-estradiol and bisphenol A as the basis to create a new PBK model describing the kinetics of MGO in mice37. The conceptual mouse PBK model for MGO is presented in Fig. 2. In the PBK model, processes such as gastrointestinal transport, absorption, and clearance were all assumed to show first-order kinetics and were modeled as a function of an intrinsic kinetic constant times the local concentration (k × [C]). This model consisted of separate compartments for the stomach, intestine, liver, fat, blood, rapidly perfused tissue (e.g., brain, lungs, heart, and kidneys), and slowly perfused tissue (muscle, skin, and bone). The PBK model enables the simulation of systemic blood concentrations of MGO following its oral administration. The physiological and anatomical parameters, such as the fraction of tissue volumes and blood flows to tissues were obtained from the literature39,40 and are summarized in Supplementary Table S1.

Fig. 2: Schematic presentation of the PBK model for MGO.
figure 2

The conceptual mouse PBK model for MGO consisted of separate compartments representing the stomach, intestine, liver, fat, blood, rapidly perfused tissue (e.g., brain, lungs, heart, and kidneys), and slowly perfused tissue (e.g., muscle, skin, and bone).

Full size image

This model accounted for stomach emptying following oral administration of MGO and divided the intestinal compartment into seven sub-compartments to simulate intestinal transit of MGO. The apparent permeability coefficient (Papp) value for MGO, determined to be 1.09 × 10−6 cm/s from Caco-2 transport studies conducted in our lab41, was utilized to define the intestinal absorption of MGO from these sub-compartments. To extrapolate the Caco-2 Papp value to an in vivo Papp value for humans and mice, the following equations were used (Eqs. 1–2):

$${\rm{Log}}\left(P_{{\rm{app}},{{\rm{human}}\; {\rm{in}}\; {\rm{vivo}}}}\right)\,=\,0.6836\times {\rm{Log}}\left({P}_{{\rm{app}},{\rm{Caco}}-2}\right)-0.5579$$
(1)
$${P}_{{\rm{app}},{\rm{mouse\; in\; vivo}}}={P}_{{\rm{app}},{\rm{human\; in\; vivo}}}/3.6$$
(2)

in which Eq. 1 scales the in vitro Caco-2 apparent permeability (Papp,Caco-2) to an in vivo human effective permeability (Papp,human in vivo) for passively absorbed compounds42, and Eq. 2 scales the human effective permeability (Papp,human in vivo) to mouse effective permeability (Papp, mouse in vivo) using the equation originally defined for rats43. It was assumed that rat and mouse permeabilities would be the same. The intestinal absorption rate of MGO was subsequently determined using the equation: absorption rate (μmol/h) = Papp,mouse in vivo (cm/h) × surface area of the mouse small intestine (cm2) × luminal concentration of MGO (μmol/cm3)37. The luminal concentration of MGO in each sub-compartment was calculated by dividing the amount of MGO by the volume of the sub-compartment. Additionally, it was reported that blood MGO levels in mice began to rise above basal levels 50–55 min following a single oral dose of MGO at 50 mg/kg bw, 100 mg/kg bw, or 200 mg/kg bw38. Accordingly, a lag time of 55 min was included in the model to account for this delayed response.

To describe the distribution of MGO across different tissues, the tissue/blood partition coefficients were estimated using the Rodgers and Rowland method44 via the online QIVIVE tool (https://www.qivivetools.wur.nl/, accessed January 2024)45. The input parameters for the estimation included the logarithmic octanol–water partition coefficient (LogP) and molecular weight (72.06 g/mol) of MGO. The LogP value used for MGO, 0.196, was sourced from the online platform Chemicalize developed by ChemAxon (https://chemicalize.com/, accessed January 2024). The blood-to-plasma ratio (BPr) is often assumed to be 1 for neutral compounds46. MGO is a neutral compound, and a BPr of 1 for MGO is also supported by the findings from a previous study, which reported that MGO concentrations in plasma and whole blood were nearly identical in mice following oral administration of exogenous MGO38. Consequently, correction was not necessary when comparing MGO concentrations in blood and plasma. The assumption was also made that the distribution of MGO in rats mirrors that in mice, and therefore the rat tissue/plasma partition coefficients estimated by the QIVIVE tool were directly applied to the mouse PBK model. The tissue/blood coefficients used for the mouse PBK model are presented in Supplementary Table S1.

The in vivo clearance or elimination of MGO is primarily mediated through the intracellular glyoxalase system, which is ubiquitously present in the cytoplasm of mammalian cells47. Consequently, all tissues possess the capability to metabolize MGO. Given that apparent total body clearance data for MGO in mice are available in the literature, it was assumed that the clearance for each organ could be derived based on its respective tissue volume fraction. Hence, the clearance for each tissue was estimated by multiplying the literature-reported apparent total body clearance (CLapp) of MGO with the respective tissue volume fraction38, reflecting the individual metabolic contributions of each tissue to the overall clearance of MGO.

The reported CLapp values for MGO in mice were reported to increase from 7.69 L/h/kg to 8.28 L/h/kg to 15.54 L/h/kg when the oral dose was increased from 50 mg/kg bw to 100 mg/kg bw to 200 mg/kg bw38. Notably, no significant differences in blood concentrations were observed between the two higher doses38. Ghosh et al. derived these CLapp values from the half-life of MGO in 12 mice, but the fact that the value varied with the dose actually suggests that factors other than clearance could influence total circulating concentrations and consequently the half-life38. Given the similar maximum blood concentrations (Cmax) observed after administering 100 mg/kg bw and 200 mg/kg bw, it is likely that there is an influence of saturated uptake or limited gastrointestinal solubility, rather than enhanced clearance with increasing dose. Therefore, the CLapp value from the lowest dose (50 mg/kg bw) was used in the PBK model. To better fit the model to the observed in vivo data at higher doses (100 mg/kg bw and 200 mg/kg bw), a fraction absorbed (fa) correction was incorporated. However, the literature reported CLapp values without fa correction at higher dose levels were also applied to allow comparison of model predictions by the two approaches.

The model equations were coded and numerically integrated into Berkeley Madonna 10.6.1 (UC Berkeley, CA, USA), utilizing Rosenbrock’s algorithm for stiff systems. The model code can be found in the Supporting Information and on Figshare (https://doi.org/10.6084/m9.figshare.27160080.v1).

PBK model evaluation

The performance of the developed mouse PBK model was assessed by comparing model-predicted time-dependent blood concentrations of MGO with reported in vivo time-dependent blood concentrations in mice upon a single oral dose of 50 mg/kg bw, 100 mg/kg bw, or 200 mg/kg bw of MGO38. The software TechDig 2.0 was utilized to digitize and transform graphical data from the cited studies into the numerical format for analysis.

Sensitivity analysis of the PBK model

A sensitivity analysis was conducted to determine which model parameters exert the greatest influence on the predicted maximum blood concentration (Cmax) of MGO. Normalized sensitivity coefficients (SCs) were determined using the formula: SC = (Cʹ − C)/(Pʹ − P) × (P/C), in which C is the initial model output, Cʹ represents the output after parameter adjustment, P represents the initial parameter value, and Pʹ represents the adjusted parameter value48. A 5% increase in each parameter value was implemented to evaluate the impact of parameter adjustments on the Cmax predictions of MGO. The sensitivity analysis was conducted for a single oral administration of 50 mg/kg bw, 100 mg/kg bw, or 200 mg/kg bw MGO, aligning with the doses used in the in vivo kinetic study for model evaluation. Each parameter’s influence was examined separately by modifying one parameter at a time while maintaining the others constant.

Human PBK model

Based on the evaluated mouse PBK model, a human PBK model was developed. The physiological and physicochemical parameters for the human model were obtained from the literature40,45, and are listed in Supplementary Table S1. The in vivo Papp value for the human model was determined as described in the section development of a PBK model for MGO in mice above. The CLapp values of MGO for humans were assumed to be the same as those for mice, based on the fact that detoxification of dicarbonyls by the glyoxalase system represents universal basal mammalian physiology49. Comparable blood Glo1 activity has been reported in humans and rats, further supporting this assumption50,51. Consequently, a CLapp of 7.69 L/h/kg, derived from data at the lowest dose (50 mg/kg bw) in the mouse in vivo study was applied in the human model. Given that this model will be mainly used to translate in vitro MGO neurotoxicity and genotoxicity data into in vivo dose-response data, and considering the potential differences in the lag phase between mice and humans, as well as the fact that the lag phase does not impact the predicted Cmax and is irrelevant for QIVIVE, the lag phase was omitted from the human model. For further details, the human model script can be found in the Supporting Information and on Figshare (https://doi.org/10.6084/m9.figshare.27160080.v1).

Translation of the in vitro concentration-response curves for MGO into in vivo dose-response curves using the PBK model developed for humans

In vitro concentration-response data on MGO-induced neurotoxicity, utilizing human neuronal cell models, were collected (see “Results” section). Human primary neuronal-like cells (hNLCs) transdifferentiated from human mesenchymal stem cells exhibited the greatest sensitivity to MGO (see “Results” section), more accurately reflecting the physiological and biochemical properties of neuronal cells compared to other neuronal cell models23. Hence, in vitro concentration-response data from hNLCs including effects of MGO on mitochondrial function (measured using the MTT viability assay), cytotoxicity (assessed via the Trypan Blue exclusion method), and apoptosis (evaluated using nuclear staining with Hoechst 33258 dye), were selected for establishing human in vivo dose-response relationships for MGO-induced neurotoxicity23. Additionally, in vitro concentration-response data on MGO-induced DNA adduct formation in WM-266-4 human melanoma cells reflecting the potential genotoxicity of MGO were also chosen for in vivo translation34.

In this study, in vivo dose-dependent toxicity by MGO was assumed to be dependent on the Cmax of MGO reached in blood, and thus the PBK modeling-based reverse dosimetry approach was applied to predict the dose levels that were required to reach the respective effective Cmax concentrations of MGO in blood. The translation from the in vitro assay concentrations to corresponding exogenous dose levels was performed without correcting for endogenous MGO formation based on the assumption that the endogenous formation of MGO would be comparable in cells in vitro and in vivo. Indeed intracellular endogenous MGO formation is known to occur in vivo but also in in vitro cell models8. Additionally, the toxicity of MGO was attributed to the unbound fraction (fub) of MGO, requiring a correction for protein binding to address differences in fub between human blood and the in vitro assay medium. Consequently, the in vitro unbound concentration was set equal to the in vivo unbound Cmax, and the in vivo unbound Cmax derived from oral MGO administration was calculated using the equation:

$${C}_{{\rm{in\; vivo}}}\times {f}_{{\rm{un}},{\rm{in\; vivo}}}={C}_{{\rm{in\; vitro}}}\times {f}_{{\rm{un}},{\rm{in\; vitro}}}$$

where Cin vivo is the maximal total MGO concentration resulting from oral MGO administration, and Cin vitro is the total MGO concentration used in the in vitro assay. The fraction unbound in the assay medium and in blood, fun,in vitro, and fun,in vivo, respectively, take into account the difference in protein binding in the assay medium and human blood. The fun,in vivo value, was calculated using the online QIVIVE tool45 and amounted to 0.71. The fub,in vitro value was calculated based on the assumption that the unbound fraction depends linearly on the protein content, and can be derived from the fub, in vivo52. This was determined using the equation: fub,in vitro = 1 − (1 − fub,in vivo) × (protein content in vitro/protein content in vivo)53. In this equation, the protein content in human blood is approximately 8%54, while the protein content in the assay medium used to generate the in vitro data intended for in vivo translation was considered 0, due to the use of the serum-free medium in these assays.

Based on the above, reverse dosimetry on each effective concentration Cin vivo was conducted utilizing the developed human PBK model to obtain the corresponding oral MGO dose levels, thereby facilitating the conversion of in vitro concentration-response data into in vivo dose-response curves.

Determination of points of departure (PODs) based on the predicted in vivo dose-response curves

To determine the PODs for MGO, benchmark dose modeling (BMD) analysis was applied to the dose-response curves predicted by the human PBK model. The analysis was conducted using the US EPA’s Benchmark Dose Software 3.2. Exponential and hill models were employed due to their suitability for modeling continuous data. The lower and upper 90% confidence limits of the BMD (BMDL10 and BMDU10, respectively), which correspond to an extra 10% response above the background compared to the control, were selected from the model with the best fit, as reflected by the lowest Akaike Information Criterion. The BMDL10 values were used as PODs for MGO risk assessment as further described in the next section.

Estimation of margins of exposure (MOEs) for MGO resulting from exogenous and endogenous MGO exposure

An MOE approach was employed for risk assessment of both endogenous and exogenous exposure to MGO. MOEs were calculated by dividing the derived PODs (BMDL10) by the estimated daily intake or the estimated endogenous formation of MGO in healthy individuals and diabetic patients. The estimated daily MGO intake levels, as reported by Hellwig et al.3 and Degen et al.2, are 0.03 mg/kg bw and 0.07–0.29 mg/kg bw for a 70 kg person, respectively. Levels of daily endogenous MGO formation are estimated at 3.09 mg/kg bw for healthy adult individuals and range from 6.01 mg/kg bw to 12.35 mg/kg bw for diabetic patients, as reported by Rabbani and Thornalley6. In the MOE calculations, the highest reported daily intake level of 0.29 mg/kg bw from Degen et al.2 and the highest level of estimated daily endogenous formation of 12.35 mg/kg bw for diabetic patients from Rabbani and Thornalley6 were used to represent a conservative, worst-case scenario. For healthy individuals, the reported daily endogenous formation of 3.09 mg/kg bw6 was used.

Results

PBK model evaluation

For model evaluation, in a first approach, CLapp values corresponding to each administered dose (50 mg/kg bw, 100 mg/kg bw, or 200 mg/kg bw of MGO) were employed without additional fraction absorbed (fa) corrections (Fig. 3a–c). The predicted Cmax and Tmax of MGO in mouse blood by the PBK model and the reported in vivo values are detailed in Table 1. The PBK model predictions for blood concentrations of MGO matched the reported concentrations well, with predictions falling within a two-fold difference of the in vivo data38, indicating that the model was able to adequately predict the time-dependent blood concentrations of MGO at the given dose levels (Fig. 3a–c and Table 1).

Fig. 3: Comparison of reported38 and PBK model-predicted time-dependent blood concentrations of MGO in mice following a single oral dose of 50 mg/kg bw, 100 mg/kg bw, or 200 mg/kg bw MGO.
figure 3

a–c CLapp values reported for the respective dose38 were used for all three doses without additional fraction absorbed (fa) correction. d–e The CLapp value reported for the lowest dose (50 mg/kg bw) was used for doses at 100 mg/kg bw and 200 mg/kg bw, with also a fraction absorbed (fa) correction applied at 200 mg/kg bw to achieve a better fit (blue line in panel e). The reported blood concentrations of MGO were corrected for endogenously formed MGO by subtracting the basal in vivo blood level of MGO amounting to 13.7 μM38.

Full size image
Table 1 PBK model-predicted Cmax and Tmax of MGO compared to in vivo mouse data as reported by Ghosh et al.38
Full size table

However, it is important to recognize that organ clearance should ideally be modeled as a function of the local concentration, multiplied by an intrinsic kinetic constant that does not vary with the dose. As described in the section development of a PBK model for MGO in mice, the similar in vivo Cmax observed in mice after administering 100 mg/kg bw and 200 mg/kg bw MGO indicates that there is likely an influence of saturated uptake or limited gastrointestinal solubility. Consequently, the CLapp value from the lowest dose (50 mg/kg bw) was applied for higher doses (100 mg/kg bw and 200 mg/kg bw), resulting in the predicted Cmax for each respective dose being 1.05 and 2.02 times higher than the reported Cmax (Fig. 3d, e and Table 1). For the 200 mg/kg bw dose, an fa correction (fa = 0.45) was introduced to include a limited oral bioavailability and achieve a better fit, resulting in a predicted Cmax that is 1.02 times higher than the reported value.

In the human model, a CLapp of 7.69 L/h/kg from the lowest dose (50 mg/kg bw) was utilized. For human PBK model-based reverse dosimetry, a fa correction (fa = 0.45) was applied specifically to doses above 200 mg/kg bw. Applying this correction to model these higher dose levels was necessary because the in vitro dataset used for QIVIVE in this study contains concentrations much higher than typical in vivo exposures, corresponding to oral doses exceeding 200 mg/kg bw.

Sensitivity analysis of the PBK model

The performance of the mouse PBK model was further assessed by a sensitivity analysis to identify the parameters that are most influential on the Cmax predictions of MGO in blood following a single oral administration of 50 mg/kg bw, 100 mg/kg bw, or 200 mg/kg bw. The parameters with an absolute normalized SC value of 0.1 or greater are presented in Fig. 4. The results show that, for all three doses, the Cmax predictions for MGO appeared to be most sensitive to the parameters related to gastrointestinal transport and absorption, including the surface area of the intestinal compartment (SAin), the volume for each compartment of intestines (Vin), the stomach emptying rate (ksto), the Papp value, and the transfer rate to the next compartment within the intestines (kin). Additionally, the CLapp value was also identified as one of the influential parameters affecting the Cmax predictions for MGO.

Fig. 4: Sensitivity analysis for the predicted Cmax of MGO after oral administration of 50 mg/kg bw, 100 mg/kg bw, or 200 mg/kg bw MGO.
figure 4

Model parameters with an absolute normalized SC value ≥ 0.1 are displayed. VLc fraction of liver tissue, VBc fraction of blood, VRc fraction of rapidly perfused tissue, VSc fraction of slowly perfused tissue, QC cardiac output, QLc fraction of blood flow to liver, QRc fraction of blood flow to rapidly perfused tissue, QSc fraction of blood flow to slowly perfused tissue, ksto stomach emptying rate, Papp apparent permeability coefficient obtained from the Caco-2 model, Vin volume for each compartment of intestines, SAin surface area of the intestinal compartment, kin transfer rate to next compartment within the intestines, CLapp apparent total body clearance in mice.

Full size image

Translation of the in vitro concentration-response curves for neurotoxicity and genotoxicity of MGO to in vivo dose-response curves and BMD analysis of predicted dose-response data

Upon evaluating the mouse PBK model for MGO, the model code was adapted to define a human PBK model for predicting MGO-induced neurotoxicity and genotoxicity in humans from in vitro data. Figure 5a summarizes the in vitro concentration-response data on MGO-induced neurotoxicity obtained with various human neuronal cell models found in the literature. hNLCs, were found to be the most sensitive to MGO, displaying the lowest EC50 value for the cytotoxicity after 48 h of exposure, which was 220.8 μM, as detailed in Table 2. These hNLCs are considered to mimic the physiological and biochemical properties of neuronal cells more accurately than other tested cell models23. Consequently, the in vitro dataset for hNLCs, which includes the effects of MGO on mitochondrial function, cytotoxicity, and apoptosis, was chosen to establish in vivo dose-response data for humans. Furthermore, the dataset from Yuan et al., describing DNA adduct (R-N2-CEdG and S-N2-CEdG) formation induced by MGO in WM-266-4 cells55, was selected for in vivo translation for the genotoxicity of MGO (Fig. 5b). The in vitro results in Fig. 5 reveal that the toxicity of MGO quantified by cytotoxicity and apoptosis in hNLCs started to occur at concentrations of 10 μM onwards. Mitochondrial dysfunction and cytotoxicity in hNLCs, and DNA adduct formation in WM-266-4 cells appeared to occur at somewhat higher concentrations.

Fig. 5: Overview of in vitro concentration-response data for MGO neurotoxicity and DNA adduct formation found in literature.
figure 5

a Neurotoxicity of MGO in various human neuronal cell models including SH-SY5Y cells20, SK-N-MC cells22, SK-N-SH cells21, M17 cells and iPSC-derived neurons73, and hNLCs23; b DNA adduct (R-N2-CEdG and S-N2-CEdG) formation induced by MGO in WM-266-4 cells34. Data points are shown as mean ± (SD or SEM), where available.

Full size image
Table 2 EC50 values derived from the in vitro data on the neurotoxicity of MGO in human neuronal cell models reported in the literature
Full size table

Figure 6 shows the dose-response curves derived from translating the in vitro concentration-response data using PBK modeling-facilitated reverse-dosimetry with the developed human PBK model. Subsequent BMD analysis of these predicted in vivo dose-response curves revealed BMDL10 values of 251 mg/kg bw and 254 mg/kg bw for R-N2-CEdG and S-N2-CEdG formation, respectively, and a somewhat higher BMDL10 of 304 mg/kg bw associated with apoptosis in neuronal cells (48 h exposure). The estimated BMDL10 values for mitochondrial function (48 h exposure) and cytotoxicity (48 h exposure) amounted to 1366 mg/kg bw and 590 mg/kg bw, respectively (Table 3).

Fig. 6: Predicted in vivo dose-response curves for MGO-induced neurotoxicity and DNA adduct formation upon oral administration of MGO to humans.
figure 6

a Predicted dose-response curves for MGO-induced neurotoxicity, acquired through PBK modeling-facilitated reverse dosimetry of in vitro toxicity data on MGO-induced neurotoxicity in hNLCs reported by Coccini et al. 23; b Predicted dose-response curves for MGO-induced DNA adduct formation, acquired through PBK modeling-facilitated reverse dosimetry of in vitro toxicity data on DNA adduct formation induced by MGO in WM-266-4 cells reported by Yuan et al. 34 Data points are shown as mean ± (SD or SEM), where available.

Full size image
Table 3 Summary of estimated PODs taken as the BMDL10 derived from the predicted dose-response curves and MOEs for neurotoxicity (mitochondrial function, cytotoxicity, and apoptosis), and genotoxicity (DNA adduct formation) induced by MGO
Full size table

Risk assessment of exogenous and endogenous exposure to MGO using the MOE approach

Figure 7 displays the predicted BMDL10-BMDU10 ranges for MGO-induced neurotoxicity and DNA adduct formation, and also the dose levels at which the MOE relative to the respective BMDL10 is 100 (for mitochondrial function, cytotoxicity, and apoptosis) or 10,000 (for DNA adduct formation) in comparison to the estimated daily dietary intake and endogenous formation of MGO in humans. The results indicate that both the estimated daily intake and endogenous formation of MGO in healthy and diabetic individuals are below the BMDL10 for all the endpoints. Especially, the estimated dietary intake of MGO is approximately three orders of magnitude lower than the predicted BMDL10 values. Drawing from the European Food Safety Authority (EFSA)’s risk assessment for acrylamide, which is neurotoxic and genotoxic, this study adopted a similar MOE approach for further risk characterization for MGO56. Accordingly, an MOE of 100 or above for neurotoxicity endpoints including mitochondrial function, cytotoxicity, and apoptosis, was considered of no health concern. This margin accounts for uncertainties and variability within the human population (default uncertainty value of 10) and the incorporation of NAMs (assumed uncertainty value of also 10). Considering that DNA adduct formation implies genotoxicity effects, an MOE of 10,000 or higher for DNA adduct formation was employed, accounting for interindividual variability (value of 10), the use of NAMs (value of 10), uncertainties related to the carcinogenic processes (value of 10), and use of the BMDL10 value (value of 10), as indicative of no health concern. The green vertical lines in Fig. 7 represent the dose levels where the MOE relative to the respective BMDL10 is 100 (for mitochondrial function, cytotoxicity, and apoptosis) or 10,000 (for DNA adduct formation). Comparison of these dose levels to the estimated daily dietary intakes reveals that the estimated daily intake of MGO falls far below the green vertical lines representing the dose levels where the MOE relative to the respective BMDL10 is 100 for endpoints including mitochondrial function, cytotoxicity, and apoptosis, indicating that the systematic Cmax values resulting from daily intake of MGO would not be high enough to cause mitochondrial dysfunction, cytotoxicity, and apoptosis in neuronal cells. However, endogenous MGO levels in healthy individuals, exceed this safety threshold (green vertical line) for apoptosis (48 h exposure), indicating that a potential risk of inducing apoptosis by endogenously formed MGO cannot be excluded. In diabetic patients, endogenous MGO levels not only surpass the dose level where the MOE relative to the BMDL10 is 100 for apoptosis but also are above the respective dose levels that result in an MOE of 100 for cytotoxicity, indicating that a concern can no longer be excluded. It is also important to note that for R-N2-CEdG and S-N2-CEdG formation both the estimated daily intake and the endogenous formation of MGO in healthy and diabetic individuals are above the green vertical lines representing the dose levels where the MOE relative to the respective BMDL10 for these endpoints is 10,000, indicating a concern for MGO-induced DNA modification from both dietary intake and endogenous formation cannot be excluded, with concerns for diabetic patients being higher than those for the healthy adult population. The numbers of the calculated MOE values for the dietary intake or endogenous formation of MGO in healthy individuals and diabetic patients for all the endpoints used in this study can be found in Table 3.

Fig. 7: Comparison of estimated daily MGO intake and endogenous formation in healthy individuals and diabetic patients with predicted BMDL10 values.
figure 7

The estimated daily MGO intake dose levels were taken from Hellwig et al.3 and Degen et al.2 and amount to 0.03 mg/kg bw and 0.07–0.29 mg/kg bw for a 70 kg person, respectively. The estimated daily endogenous MGO formation levels in healthy adult individuals and diabetic patients were taken from Rabbani and Thornalley and amounted to 3.09 mg/kg bw, and 6.01–12.35 mg/kg bw for a 70 kg person, respectively6. The green vertical lines represent the dose levels where the MOE relative to the respective BMDL10 is 100 (for mitochondrial function, cytotoxicity, and apoptosis) or 10,000 (for DNA adduct formation). For further details see text.

Full size image

Discussion

PBK modeling-facilitated reverse dosimetry has proven to be a promising NAM for extrapolating in vitro data to predict in vivo toxicity of chemicals in humans, providing significant potential in chemical risk assessment35,36,37. This study aimed to apply the PBK modeling-facilitated reverse dosimetry approach as a NAM to predict the neurotoxicity and genotoxicity of MGO in humans, based on the in vitro toxicity data obtained in a human neuronal cell model (hNLCs)23, and in WM-266-4 human melanoma cells34. The in vitro toxicity endpoints evaluated included mitochondrial function, cytotoxicity, apoptosis, and DNA adduct formation.

The human PBK model for MGO presented in this study was developed based on a mouse PBK model. Evaluation of the mouse PBK model performance for MGO demonstrated its adequacy for predicting kinetic data at the given dose levels (50 mg/kg bw, 100 mg/kg bw, and 200 mg/kg bw). The kinetic parameters for MGO used in this model, including parameters such as SAin, Vin, the Papp value, and CLapp value, were identified as the most influential parameters for the prediction of the Cmax of MGO in the sensitivity analysis. The seven-compartment GI-tract model, featuring parameters such as SAin and Vin, was developed in our lab and integrated within the PBK model code. It has been successfully applied to a broad spectrum of compounds, demonstrating robustness and reliability37,53,57,58,59. The absorption rate constant derived from the Papp value, determined in our laboratory using the Caco-2 model, amounted to 0.25 h−1 41. This value is in line with the average apparent absorption rate constant (ka) from the reported mouse in vivo data amounting to 0.24 h−1 38. Since these values are comparable, employing either ka or the Papp value to define MGO’s absorption rate in the PBK model does not significantly affect the predicted Cmax. With respect to the CLapp, the reported CLapp values for the three doses varied and increased with doses38. The increase in CLapp with increasing dose appeared to be more substantial at dose levels above 100 mg/kg bw and may be less pronounced at lower dose levels. Moreover, for the reasons detailed in the sections development of a PBK model for MGO in mice and PBK model evaluation, the lowest available CLapp value of 7.69 L/h/kg appears most suitable for PBK modeling at dose levels in the range of the estimated daily dietary MGO intake. Even when the CLapp value would be tenfold lower, it can be calculated (Supplementary Fig. S1 and Supplementary Table S2) that the predicted BMDL10 values for endpoints including mitochondrial function, cytotoxicity, and apoptosis will still be over two orders of magnitude above the estimated daily dietary MGO intake, and the MOEs are still high enough to be of no concern.

The current study assumed that the in vivo dose-dependent neurotoxicity induced by MGO is based on the maximal unbound concentration (Cmax) of MGO in the blood, rather than incorporating a separate compartment for the central nervous system (CNS) in the PBK model to predict the maximal concentration achievable in the CNS. Given the variable rates of MGO formation and detoxification across different cell types, and their potentially differing susceptibilities to MGO27,60, ideally, risk assessment would include tissue-specific MGO concentrations resulting from endogenous and exogenous sources. However, comprehensive data on the rate of MGO formation in different tissues are currently lacking. Besides, at the present state-of-the-art, the extent to which MGO can enter the CNS is not yet clearly defined. If the transport to the CNS is limited, the MOEs resulting from daily dietary intake would be even larger than what has now been estimated. In the present study, we utilized estimates of systematic endogenous MGO formation and dietary intake for the subsequent risk assessment. While this approach has its limitations, it allows the use of available data to estimate the potential risks of MGO-related neurotoxicity under realistic exogenous and endogenous exposure conditions.

In the risk assessment of MGO in the present study, it is worth noting that the value of 10 was chosen to account for uncertainties related to use of NAMs, aligning with the default uncertainty factors of 10 for both interspecies and interindividual differences. Also these default uncertainty factors were originally not based on a robust scientific basis although at present they are universally accepted by the scientific community. Use of a factor 10 to account for uncertainties accompanying use of NAMs was suggested before to account for the variability and uncertainties inherent in translating results from novel or non-traditional models to human health outcomes61,62. In our study, we have adopted this value to err on the side of caution, ensuring that our risk assessment remains protective of human health despite the potential uncertainties involved with the use of NAMs.

The results reported in the literature and depicted in Fig. 7 indicate that the reported endogenous MGO formation in healthy adults (ca. 216 mg equal to 3.09 mg/kg bw for a 70 kg person) exceeds the estimated daily dietary intake of MGO (5–20 mg equal to 0.07–0.29 mg/kg bw for a 70 kg person) by more than tenfold2,6, indicating that the dietary contribution to the total MGO exposure is relatively low. This observation aligns with findings from a previous study where the consumption of honey containing 37 mg of MGO, a quantity exceeding the estimated daily intake levels and equal to 0.53 mg/kg bw for a 70 kg person, did not result in increased urinary excretion of MGO and its metabolite d-lactate63. This suggests that the absorption of dietary MGO might be limited. Furthermore, the results obtained in the current study show that the MOEs resulting from estimated daily dietary intake of MGO for the neurotoxicity endpoints ranged from 304 to 1366, which was above the safety limit of 100, suggesting that the systemic concentrations resulting from daily MGO intake are unlikely to reach levels capable of inducing mitochondrial dysfunction, cytotoxicity, and apoptosis in neuronal cells. In contrast, the MOE value for the apoptosis (48 h exposure) resulting from daily endogenous MGO formation in healthy individuals, amounted to 98, and for diabetic patients, whose endogenous MGO formation is reported to be two to four times higher than that in healthy individuals6, the MOEs for the apoptosis and also for cytotoxicity were below the safety limit of 100 (Table 3), suggesting neurotoxicity reflected by these two endpoints induced by endogenously formed MGO cannot be excluded. These results are also supported by previous research which linked the progression of diabetic neuropathies, a major complication of diabetes, to increased in vivo levels of dicarbonyl compounds64, further emphasizing the critical role of endogenous MGO levels in these adverse health outcomes. The MOEs for DNA adduct formation from both dietary intake and endogenous formation in healthy individuals and diabetic patients were all below 10,000, indicating that concerns about MGO’s potential to increase the risk of developing cancer via a genotoxic mode of action cannot be excluded, especially not for diabetic patients for which as a result of endogenous MGO formation an MOE of only 20 was obtained. Previous meta-analyses revealed that diabetic patients were at a higher risk of developing several types of cancer65, and endogenous accumulation of AGEs formed from protein and DNA modifications by MGO may potentially be linked to cancer development66.

It is relevant to note that the average plasma levels from endogenously formed MGO amounted to 0.29 μM for healthy individuals as derived from aggregating data across multiple studies, reporting values of 0.21 μM, 0.13 μM, and 0.52 μM7,8,9. In diabetic patients, the plasma MGO levels were found to be more than 30% higher than those in healthy individuals7,9. However, it appeared that in the PBK modeling the internal concentrations resulting from an exogenous dose as high as 50 mg/kg bw, 100 mg/kg bw, and 200 mg/kg bw were far above the endogenous plasma concentrations (0.29 μM). This can be attributed to the fact that the dose levels used for the modeling (and used in the mouse kinetic in vivo study) are much higher than the dietary intake levels. The PBK model predicted internal Cmax levels resulting from realistic estimated daily dietary intake were found to be more than threefold lower than the endogenous plasma MGO levels. This observation also suggests that in vitro studies employing MGO concentrations in the high mM range may not accurately reflect realistic dietary exposure to MGO. Besides, considering the more critical role of endogenous MGO levels in potential adverse health effects, future studies could focus on in vitro models that mimic increased intracellular MGO formation, aiming to provide deeper insights into the mechanisms underlying how increased endogenous MGO formation contributes to disease processes and identify mitigation strategies, particularly for vulnerable populations like diabetics.

It is also crucial to recognize that while our study suggests the neurotoxicity risk from dietary MGO might be minimal, exposure to other dicarbonyl compounds, such as glyoxal (GO) and 3-deoxyglucosone (3-DG), is also prevalent and could potentially add to the risk of toxicity caused by these dicarbonyl AGE precursors in humans. Among these, MGO was identified as the most reactive in terms of cytotoxicity and ROS formation in a previous study67. The estimated worst-case daily intake for glyoxal is around 10 mg68, whereas, for 3-DG, dietary intake levels are notably higher, ranging from 20 to 160 mg per day2. A study found that upon consuming 82 mg 3-DG in a honey matrix, a significant proportion (10–15%) of the ingested 3-DG and its metabolite (3-deoxyfructose) was excreted in the urine69. This suggests that the contribution of dietary 3-DG to the total in vivo 3-DG exposome may be more substantial than what was observed for MGO. Given these findings, it would be of interest for future research to establish relative potency values for GO and 3-DG relative to MGO to more accurately assess the health risks from combined exposure to dicarbonyl compounds.

The current PBK model for MGO also has several limitations. Firstly, due to the limited in vivo kinetic data for MGO, the performance of the developed mouse PBK model was evaluated by comparing the model-predicted time-dependent blood concentrations of MGO with the in vivo time-dependent blood concentrations reported by Ghosh et al. for mice38. Although the model-predicted blood concentrations matched well with the in vivo data reported by Ghosh et al.38, this may partly originate from the fact that the rate constants for clearance of MGO used in the mouse PBK model were defined in the study by Ghosh et al. by fitting to the experimental data. To increase the reliability of the PBK model, future studies could independently determine the rate constant for clearance, which can be determined through in vitro assays with relevant tissue samples. Any deviations predicted by this newly parameterized PBK model from the data from the in vivo study can then be critically analyzed to further refine the model’s predictive accuracy. Particularly, the parameter for clearance, identified by the sensitivity analysis as the most influential on Cmax predictions, should be quantified with a high level of accuracy to enhance the predictive reliability of the model. Moreover, future studies could also consider running a Monte Carlo simulation taking into account the variability within the population for these influential parameters. Doing so would not only help estimate error margins, thereby enhancing the reliability and robustness of the model’s predictions, but it would also aid in understanding how interindividual variability in these parameters impacts the model’s outcomes70. Additionally, in the human PBK model, the same CLapp values were used for MGO as those defined for the mouse model, based on the fact that detoxification of dicarbonyls by the glyoxalase system represents universal basal mammalian physiology49. However, it would still be beneficial for future studies to compare the detoxification capacities of dicarbonyls across different species using in vitro assays, to further validate and support the validity of this assumption made when defining the human PBK model. Lastly, for the evaluation of genotoxicity risks in the present study, data from the literature quantifying MGO-induced DNA adducts in WM-266-4 human melanoma cells after a 3-h exposure to exogenous MGO at concentrations ranging from 0 μM to 1250 μM were used for QIVIVE34. It should be noted that, while this data set is the only one currently available for in vitro genotoxicity of MGO, melanoma cells, as a tumor cell line, appear to be not ideally suited for defining concentration- response curves for QIVIVE when assessing genotoxicity. This is due to different basal levels of MGO formation and detoxification capabilities of tumor cells71, which may affect the in vitro effective concentrations. Furthermore, a 3-h exposure period may not adequately capture the full spectrum of genotoxic effects, which often require longer durations to manifest72. Consequently, future studies should consider using more physiologically relevant in vitro models, such as human peripheral blood lymphocytes (HPBLs) from healthy donors, for more accurate assessments of genotoxic potential of MGO. Nonetheless, our newly defined human PBK model for MGO remains a valuable tool for QIVIVE and subsequent risk assessment for MGO, providing essential insights despite the limitations of the current in vitro toxicity data sets available for QIVIVE.

Taking all together, this study demonstrates a proof of principle on how to integrate in vitro data and in silico PBK modeling to predict in vivo kinetics of dicarbonyl compounds and characterize their dose levels causing toxicity in vivo in humans. The results show that the MOEs resulting from daily dietary intake of MGO did not raise a concern for endpoints for neurotoxicity including mitochondrial function, cytotoxicity, and apoptosis, while those for DNA adduct formation could not exclude a concern over genotoxicity. In comparison, endogenous MGO formation, especially under diabetic conditions, resulted in MOEs that raised a concern not only for genotoxicity but also for some of the neurotoxicity endpoints evaluated. Thus, the results also point at the importance of taking the endogenous levels into account in risk assessment of MGO.