15 In Silico Prediction of the Point of Departure (POD) …
309
modeling with ToxCast or Tox21 data, see [34] for an example. The POD derived
this way is concerning the dose in an in vitro assay environment. Shah et al. [35]
studied the use of ToxCast data to reconstruct dynamic cell-state trajectories and
estimate the in vitro POD. In this study, the authors evaluated the effects of 967
chemicals in multiple doses on HepG2 cells over a 72-h exposure period using high
content imaging (HCI). For each chemical, HCI endpoints including various protein
readings, mitochondrial properties, cell cycle indicators, and other cell properties
were used to define a cell-state trajectory. It is posited that if the effect of the chemical
is not intolerable, the cells tend to recover to their original states after a period of
perturbation. Tipping points were identified as concentration-dependent transitions
in system recovery, beyond which the potential for recovery will be lost. The authors
argue that the tipping point can serve as a point of departure to provide information
about the effects of new chemicals and about critical concentrations at which cellular
responses fail to recover to the pre-perturbation levels. This is potentially useful for
screening a large number of chemicals for prioritization.
15.4 Predict In Vivo PODs with In Vitro Assays
Though defining in vitro PODs as in [35] has significant potential for prioritization
in screening for a large number of chemicals, it is often of interest to predict in vivo
PODs. If this can be done directly with in vitro assay data, it will greatly advance
the vision of Tox21. To do this, however, one needs to relate the concentration
used in vitro assays to the oral dose in animals or humans. One approach is to use
in vitro-in vivo extrapolation (IVIVE) techniques based on toxicokinetics. Sipes
et al. [36] discussed a strategy in relating the peak plasma concentration (C max )
to the half-maximal effective concentration (AC 50 ) from in vitro assays to assess
chemical–biological interactions. The C max value can be related to oral dose with
models for toxicokinetics, which is implemented in the R package HTTK [37]. In
principle, a similar approach can be applied with in vitro PODs instead of AC 50 ,
though it has not been attempted.
Wang [34] took a predictive learning approach to this problem, i.e., using a large
number of in vitro assay endpoints to infer the in vivo point of departure. In this paper,
a robust learning approach was developed to infer the in vivo point of departure (POD)
with in vitro assay endpoints from ToxCast and Tox21 projects. First, the in vitro
dose response data were utilized to derive the in vitro PODs for several hundred
chemicals following the BMD approach. These were combined with in vivo PODs
from ToxRefDB regarding the rat and mouse liver to build a high-dimensional robust
regression model. The advantage of this approach is the separation of chemicals into
a majority, well-predicted set; and a minority, outlier set. Prominent relationships
will then become apparent in the majority set (Fig. 15.2). For both mouse and ratliver PODs, over 93% of chemicals have inferred values from in vitro PODs that
are within ±1 of the in vivo PODs on the log 10 scale (the target learning region,
or TLR) and with R
2 values of 0.80 (rats) and 0.78 (mice) for these chemicals.
309
modeling with ToxCast or Tox21 data, see [34] for an example. The POD derived
this way is concerning the dose in an in vitro assay environment. Shah et al. [35]
studied the use of ToxCast data to reconstruct dynamic cell-state trajectories and
estimate the in vitro POD. In this study, the authors evaluated the effects of 967
chemicals in multiple doses on HepG2 cells over a 72-h exposure period using high
content imaging (HCI). For each chemical, HCI endpoints including various protein
readings, mitochondrial properties, cell cycle indicators, and other cell properties
were used to define a cell-state trajectory. It is posited that if the effect of the chemical
is not intolerable, the cells tend to recover to their original states after a period of
perturbation. Tipping points were identified as concentration-dependent transitions
in system recovery, beyond which the potential for recovery will be lost. The authors
argue that the tipping point can serve as a point of departure to provide information
about the effects of new chemicals and about critical concentrations at which cellular
responses fail to recover to the pre-perturbation levels. This is potentially useful for
screening a large number of chemicals for prioritization.
15.4 Predict In Vivo PODs with In Vitro Assays
Though defining in vitro PODs as in [35] has significant potential for prioritization
in screening for a large number of chemicals, it is often of interest to predict in vivo
PODs. If this can be done directly with in vitro assay data, it will greatly advance
the vision of Tox21. To do this, however, one needs to relate the concentration
used in vitro assays to the oral dose in animals or humans. One approach is to use
in vitro-in vivo extrapolation (IVIVE) techniques based on toxicokinetics. Sipes
et al. [36] discussed a strategy in relating the peak plasma concentration (C max )
to the half-maximal effective concentration (AC 50 ) from in vitro assays to assess
chemical–biological interactions. The C max value can be related to oral dose with
models for toxicokinetics, which is implemented in the R package HTTK [37]. In
principle, a similar approach can be applied with in vitro PODs instead of AC 50 ,
though it has not been attempted.
Wang [34] took a predictive learning approach to this problem, i.e., using a large
number of in vitro assay endpoints to infer the in vivo point of departure. In this paper,
a robust learning approach was developed to infer the in vivo point of departure (POD)
with in vitro assay endpoints from ToxCast and Tox21 projects. First, the in vitro
dose response data were utilized to derive the in vitro PODs for several hundred
chemicals following the BMD approach. These were combined with in vivo PODs
from ToxRefDB regarding the rat and mouse liver to build a high-dimensional robust
regression model. The advantage of this approach is the separation of chemicals into
a majority, well-predicted set; and a minority, outlier set. Prominent relationships
will then become apparent in the majority set (Fig. 15.2). For both mouse and ratliver PODs, over 93% of chemicals have inferred values from in vitro PODs that
are within ±1 of the in vivo PODs on the log 10 scale (the target learning region,
or TLR) and with R
2 values of 0.80 (rats) and 0.78 (mice) for these chemicals.