COSMOperm: Mechanistic Prediction of Passive Membrane Permeability for Neutral Compounds and Ions and Its pH Dependence.

We present a new and entirely mechanistic COSMOperm method to predict passive membrane permeabilities for neutral compounds, as well as anions and cations. The COSMOperm approach is based on compound-specific free energy profiles within a membrane of interest from COSMO-RS (conductor-like screening model for realistic solvation) calculations. These are combined with membrane layer-specific diffusion coefficients, for example, in the water phase, the polar head groups, and the alkyl tails of biochemical phospholipid bilayers. COSMO-RS utilizes first-principle quantum chemical structures and physically sound intermolecular interactions (electrostatic, hydrogen bond, and van der Waals). For this reason, it is unbiased toward different application scenarios, such as in cosmetics and industrial chemical or pharmaceutical industries. A fully predictive calculation of passive permeation through phospholipid bilayer membranes results in a performance of r2 = 0.92; rmsd = 0.90 log10 units for neutral compounds and anions, as compared to gold standard black lipid membrane experiments. It will be demonstrated that new membrane types can be generated by the related COSMOplex method and directly used for permeability studies by COSMOperm.

QSAR for baseline toxicity and classification of specific modes of action of ionizable organic chemicals in the zebrafish embryo toxicity test.

The fish embryo toxicity (FET) test with the zebrafish Danio rerio is widely used to assess the acute toxicity of chemicals thereby serving as animal alternative to the acute fish toxicity test. The minimal toxicity of neutral chemicals in the FET can be predicted with a previously published Quantitative Structure-Activity Relationship (QSAR) based on the liposome-water partition coefficient Klipw. Such a QSAR may serve to plan toxicity testing and to evaluate whether an observed effect is caused by a specific mode of action (MoA). The applicability domain of this QSAR was extended to ionizable organic chemicals (IOC) without any modification of slope and intercept simply by replacing the Klipw with the speciation-corrected liposome-water distribution ratio (Dlipw(pH)) as descriptor for the uptake into the embryo. FET LC50 values of IOCs were extracted from an existing FET database and published literature. IOCs were selected that are present concomitantly as neutral and charged, species, i.e., acids with an acidity constant pKa <10 and bases with pKa>5. IOCs were grouped according to their putative MoA of acute aquatic toxicity. The toxic ratios (TR) in the FET were derived by of the experimental FET-LC50 in comparison with the baseline toxicity QSAR. Baseline toxicants were confirmed to align well with the FET baseline toxicity QSAR (TR < 10). Chemicals identified to act as specific or reactive chemicals with the toxic ratio analysis in the FET test (TR > 10) were generally consistent with MoA classification for acute fish toxicity with a few exceptions that were suspected to have had issues with the stability of the pH during testing. One critical aspect for the effect analysis of ionizable chemicals is the pH, since the difference between pH and pKa determines the speciation and thereby the Dlipw(pH).

Screening tools for the bioconcentration potential of monovalent organic ions in fish.

Currently the bioaccumulation potential of organic chemicals is assessed in a first tier approach via their octanol-water partition coefficient. This approach has been developed for neutral chemicals and cannot work for ionizable and ionic chemicals because the latter have different sorption-mechanisms and -preferences. Thus, suitable screening tools for the bioconcentration potential of ionic and ionizable chemicals need to be developed because it cannot be expected that these chemicals are non-bioaccumulative per se. Here, we present such screening tools for monovalent ions and ionizable chemicals based on calibrated sorption models for membrane lipids, structural proteins and albumin. The molecular descriptors used for these models arise from quantum chemical calculations and are based on COSMO-RS theory. When we applied our screening tools to 1839 preselected chemicals from the REACH registration data base, we identified 187 chemicals as potentially bioconcentrating (still ignoring any kind of metabolism). Among these were carbon and sulphur based aromatic and aliphatic acids mostly with a rather high molecular surface area. We hope that this outcome will trigger further research on ion specific sorption mechanisms and lead to a re-evaluation of the bioconcentration potential of ionic chemicals.

Equilibrium biopartitioning of organic anions - A case study for humans and fish.

In this work we combine partition coefficients between water and membrane lipid, storage lipid, the plasma protein albumin as well as structural protein with the tissue dependent fraction of the respective phases in order to obtain a clearer picture on the relevance of various biological tissues for the bioaccumulation of 31 organic anions. Most of the partition coefficients are based on experimental data, supplemented by some predicted ones. The data suggest that the plasma protein, albumin, will be the major sorption matrix in mammals. Only small fractions of the studied chemicals will occur freely dissolved in an organism. For the investigated acids with pKa <5, partitioning is dominated by the ionic species rather than the corresponding neutral species. Bioconcentration in fish is not expected to occur for many of these acids unless pH in the aqueous environment is low or specific sorption mechanisms are relevant. In contrast, biomagnification in terrestrial mammals would be expected for most organic anions if they are not sufficiently metabolized. We conclude that sorption is important for the toxicokinetics of ionizable organic chemicals and the dominating sorbing matrices are quite different from those for neutral species.

Predicting apparent passive permeability of Caco-2 and MDCK cell-monolayers: A mechanistic model.

Experimentally derived apparent permeabilities, Papp, through cell monolayers such as Caco-2 and MDCK are considered to be an in-vitro gold standard for assessing the uptake efficiency of drugs. Here, we present a mechanistic model that describes 'passive' Papp values (i.e., neglecting active transport) by accounting for the different resistances solutes encounter when permeating a cell monolayer. We described three parallel permeation pathways, namely a cytosolic-, paracellular-, and lateral route, each of which consists of a number of serial resistances. These resistances were accounted for via a mechanistic depiction of the underlying processes that are largely based on literature work. For the present Papp dataset, about as much chemicals are dominated by the cytosolic route as were dominated by the paracellular route, while the lateral route was negligible. For the cytosolic route by far the most chemicals found their main resistance in the various water layers and not in the membrane. Although correlations within the subclasses of chemicals dominated by a specific permeation route were rather poor, we could overall satisfyingly predict Papp for 151 chemicals at a pH of 7.4 (R2 = 0.77, RMSE = 0.48). For a specific evaluation of the intrinsic membrane permeability, Pm, a second experimental dataset based on experiments with black lipid membranes, BLM, was evaluated. Pm could be predicted for 37 chemicals with R2 = 0.91 and RMSE = 0.64 log units.

Baseline toxicity and ion-trapping models to describe the pH-dependence of bacterial toxicity of pharmaceuticals.

In numerous studies on the toxicity of ionisable organic chemicals, it has been shown that the toxicity was typically higher, when larger fractions of the neutral species were present. This observation was explained in some cases by slower uptake of charged species. In other cases it was suggested that the neutral species has intrinsically higher toxicity than the charged species or is alone responsible for the toxicity. However, even permanently charged and organic chemicals with multiple acid and base functional groups and zwitterions are toxic. We set out to reconcile the divergent views and to compare the various existing models for describing the pH-dependence of toxicity with the goal to derive one model that is valid independent of the type and number of charges on the molecule. To achieve this goal we measured the cytotoxicity of 18 acidic, 15 basic and 9 multiprotic/zwitterionic pharmaceuticals at pH 5.5 to pH 9 with the bioluminescence inhibition test using Aliivibrio fischeri (Microtox assay). This assay is useful for an evaluation of various models to describe pH-dependent toxicity because the majority of chemicals act as baseline toxicants in this 30 min cytotoxicity assay. Therefore baseline toxicity with constant membrane concentrations of the sum of all chemical species of approximately 200 mmol kglip-1 served for the validation of the suitability of the various tested models. We confirmed that most tested pharmaceuticals acted as baseline toxicants in this assay at all examined pH values, when toxicity was modeled with a mixture model of concentration addition between the neutral species and all charged species. An ion trapping model, that assumes that the membrane permeability of charged species is kinetically limited, improved model predictions for some pharmaceuticals and pH values. However, neither unhindered uptake nor no uptake of the charged species were ideal models; the reality lies presumably between the two limiting cases with a slower uptake of the charged species than the neutral species. For practical applications a previously developed QSAR model with the ionisation-corrected liposome-water distribution ratio as the sole physicochemical descriptor proved to be generally applicable for all ionisable organic chemicals including those with multiple charges and zwitterions.

Assessing the toxicity of ionic liquids - Application of the critical membrane concentration approach.

Charged organic chemicals are a prevailing challenge for toxicity modelling. In this contribution we strive to recapitulate the lessons learned from the well-known modelling of narcosis (or baseline toxicity) of neutral chemicals and apply the concept to charged chemicals. First we reevaluate the organism- and chemical independent critical membrane concentration causing 50% mortality,.cmemtox, based on a critical revision of a previously published toxicity dataset for neutral chemicals. In accordance to values reported in the literature we find a mean value for cmemtox of roughly 100 mmol/kg (membrane lipid) for a broad variety of 42 aquatic organisms (333 different chemicals), albeit with a considerable scatter. Then we apply this concept to permanently charged ionic liquids (ILs). Using COSMOmic, a quantum mechanically based mechanistic model that makes use of the COSMO-RS theory, we predict membrane-water partition coefficients (Kmem/w) of the anionic and cationic IL components. Doing so, cmemtox(total) for permanently charged ILs can be estimated assuming independent, concentration additive contributions of the cationic and its respective anionic species. The resulting values for some of the toxicity data for ionic liquids are consistent with the expected range for baseline toxicity for neutral chemicals while other values are consistently greater or smaller. Based on the calculation of toxic ratios we identify ILs that exert a specific mode of toxic action. Limitations of the modelling approach especially but not exclusively due to the use of nominal concentrations instead of freely-dissolved concentrations in the published literature are critically discussed.

Modeling Exposure in the Tox21 in Vitro Bioassays.

High-throughput in vitro bioassays are becoming increasingly important in the risk characterization of anthropogenic chemicals. Large databases gather nominal effect concentrations (Cnom) for diverse modes of action. However, the biologically effective concentration can substantially deviate due to differences in chemical partitioning. In this study, we modeled freely dissolved (Cfree), cellular (Ccell), and membrane concentrations (Cmem) in the Tox21 GeneBLAzer bioassays for a set of neutral and ionogenic organic chemicals covering a large physicochemical space. Cells and medium constituents were experimentally characterized for their lipid and protein content, and partition constants were either collected from the literature or predicted by mechanistic models. The chemicals exhibited multifaceted partitioning to proteins and lipids with distribution ratios spanning over 8 orders of magnitude. Modeled Cfree deviated over 5 orders of magnitude from Cnom and can be compared to in vivo effect data, environmental concentrations, and the unbound fraction in plasma, which is needed for the in vitro to in vivo extrapolation. Ccell was relatively constant for chemicals with membrane lipid-water distribution ratios of 1000 or higher and proportional to Cnom. Representing a sum parameter for exposure that integrates the entire dose from intracellular partitioning, Ccell is particularly suitable for the effect characterization of chemicals with multiple target sites and the calculation of their relative effect potencies. Effective membrane concentrations indicated that the specific effects of very hydrophobic chemicals in multiple bioassays are occurring at concentrations close to baseline toxicity. The equilibrium partitioning model including all relevant system parameters and a generic bioassay setup is attached as an excel workbook to this paper and can readily be applied to diverse in vitro bioassays.

General baseline toxicity QSAR for nonpolar, polar and ionisable chemicals and their mixtures in the bioluminescence inhibition assay with Aliivibrio fischeri.

The Microtox assay, a bioluminescence inhibition assay with the marine bacterium Aliivibrio fischeri, is one of the most popular bioassays for assessing the cytotoxicity of organic chemicals, mixtures and environmental samples. Most environmental chemicals act as baseline toxicants in this short-term screening assay, which is typically run with only 30 min of exposure duration. Numerous Quantitative Structure-Activity Relationships (QSARs) exist for the Microtox assay for nonpolar and polar narcosis. However, typical water pollutants, which have highly diverse structures covering a wide range of hydrophobicity and speciation from neutral to anionic and cationic, are often outside the applicability domain of these QSARs. To include all types of environmentally relevant organic pollutants we developed a general baseline toxicity QSAR using liposome-water distribution ratios as descriptors. Previous limitations in availability of experimental liposome-water partition constants were overcome by reliable prediction models based on polyparameter linear free energy relationships for neutral chemicals and the COSMOmic model for charged chemicals. With this QSAR and targeted mixture experiments we could demonstrate that ionisable chemicals fall in the applicability domain. Most investigated water pollutants acted as baseline toxicants in this bioassay, with the few outliers identified as uncouplers or reactive toxicants. The main limitation of the Microtox assay is that chemicals with a high melting point and/or high hydrophobicity were outside of the applicability domain because of their low water solubility. We quantitatively derived a solubility cut-off but also demonstrated with mixture experiments that chemicals inactive on their own can contribute to mixture toxicity, which is highly relevant for complex environmental mixtures, where these chemicals may be present at concentrations below the solubility cut-off.

Comparison of different models predicting the phospholipid-membrane water partition coefficients of charged compounds.

A large fraction of commercially used chemicals is ionizable. This results in the need for mechanistic models to describe the physicochemical properties of ions, like the membrane-water partition coefficient (K(mw)), which is related to toxicity and bioaccumulation. In this work we compare 3 different and already existing modelling approaches to describe the liposome-water partition coefficient (K(lipw)) of organic ions, including 36 cations, 56 anions, 2 divalent cations and 2 zwitterions (plus 207 neutral compounds for ensuring model consistency). 1) The empirical correlation with the octanol-water partition coefficient of the corresponding neutral species yielded better results for the prediction of anions (RMSE = 0.79) than for cations (RMSE = 1.14). Though describing most anions reasonably well, the lack of mechanistic basis and the poor performance for cations constrain the usage of this model. 2) The polyparameter linear free energy relationship (pp-LFER) model performs worse (RMSE = 1.26/1.12 for anions/cations). The different physicochemical environments, due to different sorption depths into the membrane of the different species, cannot be described with a single pp-LFER model. 3) COSMOmic is based on quantum chemistry and fluid phase thermodynamics and has the widest applicability domain. It was the only model applicable for multiply charged ions and gave the best results for anions (RMSE = 0.66) and cations (RMSE = 0.71). We expect COSMOmic to contribute to a better estimation of the environmental risk of ionizable emerging pollutants.

Prediction of Phospholipid-Water Partition Coefficients of Ionic Organic Chemicals Using the Mechanistic Model COSMOmic.

The partition coefficient of chemicals from water to phospholipid membrane, K(lipw), is of central importance for various fields. For neutral organic molecules, log K(lipw) correlates with the log of bulk solvent-water partition coefficients such as the octanol-water partition coefficient. However, this is not the case for charged compounds, for which a mechanistic modeling approach is highly necessary. In this work, we extend the model COSMOmic, which adapts the COSMO-RS theory for anisotropic phases and has been shown to reliably predict K(lipw) for neutral compounds, to the use of ionic compounds. To make the COSMOmic model applicable for ionic solutes, we implemented the internal membrane dipole potential in COSMOmic. We empirically optimized the potential with experimental K(lipw) data of 161 neutral and 75 ionic compounds, yielding potential shapes that agree well with experimentally determined potentials from the literature. This model refinement has no negative effect on the prediction accuracy of neutral compounds (root-mean-square error, RMSE = 0.62 log units), while it highly improves the prediction of ions (RMSE = 0.70 log units). The refined COSMOmic is, to our knowledge, the first mechanistic model that predicts K(lipw) of both ionic and neutral species with accuracies better than 1 log unit.