The role of trust in the use of artificial intelligence for chemical risk assessment.

Risk assessment of chemicals is a time-consuming process and needs to be optimized to ensure all chemicals are timely evaluated and regulated. This transition could be stimulated by valuable applications of in silico Artificial Intelligence (AI)/Machine Learning (ML) models. However, implementation of AI/ML models in risk assessment is lagging behind. Most AI/ML models are considered 'black boxes' that lack mechanistical explainability, causing risk assessors to have insufficient trust in their predictions. Here, we explore 'trust' as an essential factor towards regulatory acceptance of AI/ML models. We provide an overview of the elements of trust, including technical and beyond-technical aspects, and highlight elements that are considered most important to build trust by risk assessors. The results provide recommendations for risk assessors and computational modelers for future development of AI/ML models, including: 1) Keep models simple and interpretable; 2) Offer transparency in the data and data curation; 3) Clearly define and communicate the scope/intended purpose; 4) Define adoption criteria; 5) Make models accessible and user-friendly; 6) Demonstrate the added value in practical settings; and 7) Engage in interdisciplinary settings. These recommendations should ideally be acknowledged in future developments to stimulate trust and acceptance of AI/ML models for regulatory purposes.

10% Body weight (gain) change as criterion for the maximum tolerated dose: A critical analysis.

The concept of the Maximum Tolerated Dose (MTD) was introduced in the seventies for carcinogenicity testing and was defined as the highest dose inducing clear toxicity, but not mortality by causes other than cancer. As estimation of the MTD in a carcinogenicity study, the highest dose that causes a 10% decrease in body weight compared to control animals over the course of a 90-day study, was formulated as a suitable criterion. This criterion was not seen as indicator of excessive toxicity but as a means to avoid false negative outcomes in a carcinogenicity study, as tumor formation may be reduced when body weight is significantly decreased. The body weight-based MTD criterion, however, turned up in carcinogenicity test guidelines and guidance (e.g., from OECD) as the highest dose that causes a 10% decrease in body weight gain relative to controls. Moreover, the 10% decrease in body weight gain criterion for MTD also ended up in test guidelines and guidances for toxicity endpoints other than carcinogenicity, so outside the context it was intended for. A 10% decrease in body weight gain relative to controls is however not a biologically relevant effect as it corresponds to less than 3% body weight reduction relative to controls in a 90-day study, which is within the normal variation in body weight. It therefore should certainly not be considered as a condition of excessive toxicity. Using the 10% lower weight gain criterion and incorrectly associating it with excessive toxicity has major implications for top dose selection in regulatory safety studies, resulting in tests performed at doses too low to elicit toxicity. This negatively impacts the reliability of studies and their regulatory usability; moreover, it results in a waste of experimental animals, which is ethically highly undesirable. Hence, our plea is to remove this MTD criterion for top dose selection in test guidelines and guidances for toxicity endpoints other than carcinogenicity and to reinstall the original 10% decrease in body weight criterion in test guidelines and guidances for carcinogenicity.

Implementation of a Human Renal Proximal Tubule on a Chip for Nephrotoxicity and Drug Interaction Studies.

Proximal tubule epithelial cells (PTEC) are susceptible to drug-induced kidney injury (DIKI). Cell-based, two-dimensional (2D) in vitro PTEC models are often poor predictors of DIKI, probably due to the lack of physiological architecture and flow. Here, we assessed a high throughput, 3D microfluidic platform (Nephroscreen) for the detection of DIKI in pharmaceutical development. This system was established with four model nephrotoxic drugs (cisplatin, tenofovir, tobramycin and cyclosporin A) and tested with eight pharmaceutical compounds. Measured parameters included cell viability, release of lactate dehydrogenase (LDH) and N-acetyl-ß-d-glucosaminidase (NAG), barrier integrity, release of specific miRNAs, and gene expression of toxicity markers. Drug-transporter interactions for P-gp and MRP2/4 were also determined. The most predictive read outs for DIKI were a combination of cell viability, LDH and miRNA release. In conclusion, Nephroscreen detected DIKI in a robust manner, is compatible with automated pipetting, proved to be amenable to long-term experiments, and was easily transferred between laboratories. This proof-of-concept-study demonstrated the usability and reproducibility of Nephroscreen for the detection of DIKI and drug-transporter interactions. Nephroscreen it represents a valuable tool towards replacing animal testing and supporting the 3Rs (Reduce, Refine and Replace animal experimentation).

Flow stimulates drug transport in a human kidney proximal tubule-on-a-chip independent of primary cilia.

BACKGROUND: Kidney disease modeling and assessment of drug-induced kidney injury can be advanced using three-dimensional (3D) microfluidic models that recapitulate in vivo characteristics. Fluid shear stress (FSS) has been depicted as main modulator improving in vitro physiology in proximal tubule epithelial cells (PTECs). We aimed to elucidate the role of FSS and primary cilia on transport activity and morphology in PTECs. METHODS: Human conditionally immortalized PTEC (ciPTEC-parent) was cultured in a microfluidic 3D device, the OrganoPlate, under a physiological peak FSS of 2.0 dyne/cm2 or low peak FSS of 0.5 dyne/cm2. Upon a 9-day exposure to FSS, albumin-FITC uptake, activity of P-glycoprotein (P-gp) and multidrug resistance-associated proteins 2/4 (MRP2/4), cytotoxicity and cell morphology were determined. RESULTS: A primary cilium knock-out cell model, ciPTEC-KIF3α-/-, was successfully established via CRISPR-Cas9 genome editing. Under physiological peak FSS, albumin-FITC uptake (p = .04) and P-gp efflux (p = .002) were increased as compared to low FSS. Remarkably, a higher albumin-FITC uptake (p = .03) and similar trends in activity of P-gp and MRP2/4 were observed in ciPTEC-KIF3α-/-. FSS induced cell elongation corresponding with the direction of flow in both cell models, but had no effect on cyclosporine A-induced cytotoxicity. CONCLUSIONS: FSS increased albumin uptake, P-gp efflux and cell elongation, but this was not attributed to a mechanosensitive mechanism related to primary cilia in PTECs, but likely to microvilli present at the apical membrane. GENERAL SIGNIFICANCE: FSS-induced improvements in biological characteristics and activity in PTECs was not mediated through a primary cilium-related mechanism.

Organic anion transporters 1 and 3 influence cellular energy metabolism in renal proximal tubule cells.

Organic anion transporters (OATs) 1 and 3 are, besides being uptake transporters, key in several cellular metabolic pathways. The underlying mechanisms are largely unknown. Hence, we used human conditionally immortalized proximal tubule epithelial cells (ciPTEC) overexpressing OAT1 or OAT3 to gain insight into these mechanisms. In ciPTEC-OAT1 and -OAT3, extracellular lactate levels were decreased (by 77% and 71%, respectively), while intracellular ATP levels remained unchanged, suggesting a shift towards an oxidative phenotype upon OAT1 or OAT3 overexpression. This was confirmed by increased respiration of ciPTEC-OAT1 and -OAT3 (1.4-fold), a decreased sensitivity to respiratory inhibition, and characterized by a higher demand on mitochondrial oxidative capacity. In-depth profiling of tricarboxylic acid (TCA) cycle metabolites revealed reduced levels of intermediates converging into α-ketoglutarate in ciPTEC-OAT1 and -OAT3, which via 2-hydroxyglutarate metabolism explains the increased respiration. These interactions with TCA cycle metabolites were in agreement with metabolomic network modeling studies published earlier. Further studies using OAT or oxidative phosphorylation (OXPHOS) inhibitors confirmed our idea that OATs are responsible for increased use and synthesis of α-ketoglutarate. In conclusion, our results indicate an increased α-ketoglutarate efflux by OAT1 and OAT3, resulting in a metabolic shift towards an oxidative phenotype.

Nephrotoxicity and Kidney Transport Assessment on 3D Perfused Proximal Tubules.

Proximal tubules in the kidney play a crucial role in reabsorbing and eliminating substrates from the body into the urine, leading to high local concentrations of xenobiotics. This makes the proximal tubule a major target for drug toxicity that needs to be evaluated during the drug development process. Here, we describe an advanced in vitro model consisting of fully polarized renal proximal tubular epithelial cells cultured in a microfluidic system. Up to 40 leak-tight tubules were cultured on this platform that provides access to the basolateral as well as the apical side of the epithelial cells. Exposure to the nephrotoxicant cisplatin caused a dose-dependent disruption of the epithelial barrier, a decrease in viability, an increase in effluent LDH activity, and changes in expression of tight-junction marker zona-occludence 1, actin, and DNA-damage marker H2A.X, as detected by immunostaining. Activity and inhibition of the efflux pumps P-glycoprotein (P-gp) and multidrug resistance protein (MRP) were demonstrated using fluorescence-based transporter assays. In addition, the transepithelial transport function from the basolateral to the apical side of the proximal tubule was studied. The apparent permeability of the fluorescent P-gp substrate rhodamine 123 was decreased by 35% by co-incubation with cyclosporin A. Furthermore, the activity of the glucose transporter SGLT2 was demonstrated using the fluorescent glucose analog 6-NBDG which was sensitive to inhibition by phlorizin. Our results demonstrate that we developed a functional 3D perfused proximal tubule model with advanced renal epithelial characteristics that can be used for drug screening studies.

Screening of Drug-Transporter Interactions in a 3D Microfluidic Renal Proximal Tubule on a Chip.

Drug-transporter interactions could impact renal drug clearance and should ideally be detected in early stages of drug development to avoid toxicity-related withdrawals in later stages. This requires reliable and robust assays for which current high-throughput screenings have, however, poor predictability. Kidney-on-a-chip platforms have the potential to improve predictability, but often lack compatibility with high-content detection platforms. Here, we combined conditionally immortalized proximal tubule epithelial cells overexpressing organic anion transporter 1 (ciPTEC-OAT1) with the microfluidic titer plate OrganoPlate to develop a screenings assay for renal drug-transporter interactions. In this platform, apical localization of F-actin and intracellular tight-junction protein zonula occludens-1 (ZO-1) indicated appropriate cell polarization. Gene expression levels of the drug transporters organic anion transporter 1 (OAT1; SLC22A6), organic cation transporter 2 (OCT2; SLC22A2), P-glycoprotein (P-gp; ABCB1), and multidrug resistance-associated protein 2 and 4 (MRP2/4; ABCC2/4) were similar levels to 2D static cultures. Functionality of the efflux transporters P-gp and MRP2/4 was studied as proof-of-concept for 3D assays using calcein-AM and 5-chloromethylfluorescein-diacetate (CMFDA), respectively. Confocal imaging demonstrated a 4.4 ± 0.2-fold increase in calcein accumulation upon P-gp inhibition using PSC833. For MRP2/4, a 3.0 ± 0.2-fold increased accumulation of glutathione-methylfluorescein (GS-MF) was observed upon inhibition with a combination of PSC833, MK571, and KO143. Semi-quantitative image processing methods for P-gp and MRP2/4 was demonstrated with corresponding Z'-factors of 0.1 ± 0.3 and 0.4 ± 0.1, respectively. In conclusion, we demonstrate a 3D microfluidic PTEC model valuable for screening of drug-transporter interactions that further allows multiplexing of endpoint read-outs for drug-transporter interactions and toxicity.

CYP-epoxygenase metabolites of docosahexaenoic acid protect HL-1 cardiac cells against LPS-induced cytotoxicity Through SIRT1.

Bacterial LPS is an environmental toxin capable of promoting various cardiac complications. Current evidence suggests that LPS-induced myocardial dysfunction emerges as a consequence of compromised quality of cardiac mitochondria. Docosahexaenoic acid (DHA, 22:6n3) is an n-3 polyunsaturated fatty acid (PUFA), which produces a broad spectrum of intrinsic physiological effects including regulation of cell survival and death mechanisms. Although, numerous studies revealed fundamentally beneficial effects of DHA on cardiovascular system, it remains unknown whether these effects were produced by DHA or one of its possibly more potent metabolites. Emerging evidence indicates that cytochrome P450 (CYP) epoxygenase metabolites of DHA, epoxydocosapentaenoic acids (EDPs), produce more potent biological activity compared to its precursor DHA. In this study we investigated whether DHA and its metabolite 19,20-EDP could protect HL-1 cardiac cells against LPS-induced cytotoxicity. We provide evidence that exogenously added or DHA-derived EDPs promote mitochondrial biogenesis and function in HL-1 cardiac cells. Our results illustrate the CYP epoxygenase metabolite of DHA, 19,20-EDP, confers extensive protection to HL-1 cardiac cells against LPS-induced cytotoxicity via activation of SIRT1.

PPARγ signaling is required for mediating EETs protective effects in neonatal cardiomyocytes exposed to LPS.

Lipopolysaccharide (LPS) is a bacterial wall endotoxin producing many pathophysiological conditions including myocardial inflammation leading to cardiotoxicity. Epoxyeicosatrienoic acids (EETs) are biologically active metabolites of arachidonic acids capable of activating protective cellular pathways in response to stress stimuli. EETs evoke a plethora of pathways limiting impairments of cellular structures, reducing cell death, and promoting anti-inflammatory reactions in various cell types. Considering EETs are capable of producing various biological protective effects, we hypothesized that EETs would protect rat neonatal cardiomyocytes (NCM) against LPS-induced cytotoxicity. In this study, we used a dual-acting, synthetic analog of EETs, UA-8 [13-(3-propylureido)tridec-8-enoic acid], possessing both EET-mimetic and soluble epoxide hydrolase selective inhibitory properties and 14,15-EET as a model of canonical EET molecules. We found that both UA-8 and 14,15-EET significantly improved cell viability and mitochondrial function of cardiomyocytes exposed to LPS. Furthermore, treatment with UA-8 or 14,15-EET resulted in significant attenuation of LPS-triggered pro-inflammatory response, caspase-3 activation and reduction in the total antioxidant capacity in cardiomyocytes. Importantly, EET-mediated effects were significantly reduced by pharmacological inhibition of peroxisome proliferator-activated receptors γ (PPARγ) suggesting that PPARγ signaling was required for EETs exerted protective effects. Data presented in the current study demonstrate that activation of PPARγ signaling plays a crucial role in EET-mediated protection against LPS-cytotoxicity in cardiomyocytes.