High-Throughput IVIVE Daily Dose Equivalent

A goal within the Tox21 collaboration is to research, develop, validate, and translate innovative test methods that will better predict how chemicals may affect humans and the environment. Tox21 and ToxCast efforts have screened ~10,000 chemicals with limited biological information in hundreds of high-throughput screening (HTS) in vitro assays. Translating these HTS data into a common language among toxicologists, specifically risk assessors, is an important next step toward making the data useful. This website allows users to estimate doses at which chemical-biological interactions are "likely", "possible", and "possible w/10-fold safety factor" for humans in vivo.

This web-application is a companion to Sipes NS et. al., 2017

Approach

A goal within the Tox21 collaboration is to research, develop, validate, and translate innovative test methods that will better predict how chemicals may affect humans and the environment. Toward this effort, ~10,000 chemicals in hundreds of HTS assays have been screened to elucidate mechanisms of chemical­-biological interactions as a majority of the chemicals previously had limited biological activity information. These data were processed into activity concentrations at half maximal efficacy (AC50­), for hazard characterization. Translating these HTS AC50s into a common language among toxicologists, specifically risk assessors, is an important next step toward making the data useful.

In understanding chemical­-biological interactions, risk assessors evaluate the dose-­response relationship, often expressed in terms of external dose (i.e., mg/kg/day), between a chemical and the incidence of an adverse toxicological response, such as cancer or developmental defects. With a mechanistic or pathway approach, we use chemical concentrations and efficacies, often expressed in µM and percent­ or fold-­change with respect to the positive control, respectively, to describe chemical­-biological interactions above a noise threshold. In vitro to in vivo extrapolation (IVIVE) applications are available to take the chemical concentration from in vitro studies that are defined by an AC50 and estimate the dose given to the organism to achieve a blood concentration equal to the AC50.

Using the IVIVE method, we can predict daily dosing scenarios assuming the blood concentration to elicit a biological effect is equivalent to the Tox21/ToxCast HTS activity metric (i.e., AC50) for any chemical. Is this assumption correct? How much of it in the blood is needed to have an effect? To address this, we borrow an approach used by the FDA in translating in vitro assay results to clinical relevance for pharmaceutical research applications. For example, FDA guidance documents have been drafted to predict whether a chemical can interfere with in vitro metabolizing enzymes and transporters adequately to disrupt normal metabolic processes by calculating the ratio of peak plasma an efficacy cutoff, to minimize false positives (e.g., 40% of positive control response) and has proven to be a useful estimate of the potential for chemicals to elicit clinically ­relevant drug-­drug concentrations (Cmax) of the chemical and dividing by the observed in vitro potency (Ki). (Draft Guidance for Industry 2006, link). The interaction is "likely" if the calculated ratio is >1 (blood concentrations equivalent or higher than in vitro half-­maximal applied concentrations), "possible" if 0.1 < X < 1, or "possible w/10-fold safety factor" if the ratio is < 0.1 (blood concentrations more than 10­-fold lower than requisite in vitro half­-maximal applied concentrations).

Model and Parameters

The High-Throughput ­Toxicokinetics (HTTK) R-­package (link) was used to perform in vitro to in vivo extrapolation of HTS data to daily human equivalent doses [Wambaugh J et. al., 2015; Pearce R et. al., 2017].

  • A 3-compartment model was used to predict in vivo doses needed to achieve blood concentrations analogous to in vitro HTS concentrations.
  • This model assumes rapid oral absorption (1/hr), 100% bioavailability, Schmitt's method for chemical partitioning, and metabolism is a "well-stirred" approximation. 
  • Minimal chemical­-specific input parameters are required, including in silico estimated physicochemical parameters (e.g., hydrophobicity, acid/base dissociation constants) as well as traditionally two in vitro measured parameters, fraction of the chemical unbound in plasma (fup) and the intrinsic hepatic clearance (CLint) to determine the equivalent doses. CLint is needed in the model to estimate how much the liver is metabolizing the parent and fup is needed in the model to estimate tissue to unbound fraction in the plasma partition coefficients. The in vitro parameters traditionally used are located within the HTTK package (chem.physical_and_invitro.data) with references.

The analyses with the Tox21/ToxCast data in the IVIVE approach has been and is currently limited to ~500 chemicals with in vitro measured CLint & fup values and cannot be applied to the entire Tox21 chemical library, unless they are predicted.

  • In silico estimates of CLint and fup parameters were estimated using a QSAR approach from ADMET Predictor (Simulations Plus, Inc.)
  • Sipes NS et. al. 2017 found these parameters to be comparable to in vitro values (74% and 62% within 10-fold of each other for fup and CLint, respectively). Total clearance values calculated using the HTTK package with in silico or in vitro values were compared to in vivo human measured values and found to be comparable (88% and 75% within 10-fold, respectively).

HTS data

This approach requires robust interactions, as we are evaluating HTS data and nuanced data should be avoided in a clear step. We reduced false-positive data by keeping only good quality Tox21/ToxCast data (i.e., ones with limited curve-fitting issues and efficacies > 40%). Specifically,

  • Tox21 data was obtained using an established method to exclude chemical-biological interactions that were confounded by cytotoxicity (measured at the same time), autofluorescence, or a weak/noisy signal (Hsieh JH et. al., 2015).
  • ToxCast specific data were obtained from the US EPA’s invitrodb_v2 database (October 2015, INVITRODB_V2_LEVEL5 folder, link)

Data downloads correspond to:

  • Download HTS data: 65,039 active chemical-assay pairs (which can be further filtered to 56,135 active calls by removing efficacies < 40% or < 2 fold, as was performed in this web application). This file contains AC50s, efficacies and flags. These data correspond to Table S4 from Sipes NS et. al. 2017.
  • Download HTS data w/ExpoCast + extras: 57,721 active chemical assay pairs (which can be further filtered to 49,789 active calls by removing efficacies < 40% or < 2 fold, as was performed in this web application) that also had ExpoCast dose estimates. This file contains the same information above as well as ExpoCast dose estimates, chemical property estimations, dose likelihoods, and uncertainty. These data correspond to Table S5 from Sipes NS et. al., 2017.