Exposure Modeling Leading to Discovery

Models can often present us with remarkable learning experiences.   Indeed, combining model development with the necessary aspects of experimentation to feed a model can lead to some important discoveries.  This week I am going to recount a discovery that my colleagues and I had while developing a model of the indoor air exposure potential from the off-gassing of pesticide from treated wood.

The pesticide was an important product for the company I was working for at the time and as the Manager of Human Health Risk Assessment, it was my responsibility to conduct a risk assessment on the use of the product as a wood preservative used on wood in the indoor environment.   

Our first experiment was to put the treated wood into a glass chamber and measure its concentration in the chamber’s air.  From this and other experiments, the generation rate of the off-gassing was estimated. 

An early lesson in all this is that models and some simulations are not reality but simply a portrayal of reality.   If they are reasonably portrayals, we can learn something.  Indeed, glass chambers are not real rooms but these results provided a lot of information about the ultimate use of treated wood used indoors in residences.

The system was complicated enough that we decided to do dynamic modeling of the chemical in compartments in a manner that is completely analogous to physiologically based pharmacokinetic (PBPK) models (https://en.wikipedia.org/wiki/Physiologically_based_pharmacokinetic_modelling). That is, each compartment is conceptually constructed to describe the instantaneous and integrated rate of pesticide input to and output from it.   While in any compartment, there was an option to describe changes or reactions of the pesticide during its time in that compartment.

The compartments we chose in our glass chamber experiment were:

·         The treated wood

·         The air space in the chamber

·         The chamber walls

A remarkable bit of technology that allowed us to do this was Advance Continuous Simulation Language (ACSL) software  (https://en.wikipedia.org/wiki/AEgis_Technologies).   At the time the best way to run this software was on mini-computers (known colloquially as “pizza boxes” because of their shape and size) running UNIX with a program subscription from  the Dow Chemical Company called SimuSolv.    I mention this because there was a lot of expense (10s of thousands of dollars) involved in getting a “pizza box”, having the technical IT support to run UNIX on this minicomputer and the cost of the license for SimuSolv.    Today you can get a PC license for ACSL for as little as $500  (http://www.acslx.com/sales/).   I am not trying to sell or advertise ACSL.   My intent is only to point out that ACSL is a good modeling tool that we went to a lot of trouble and expense to use it but it has gotten a lot more affordable over the last 20 years.

We ran a number of experiments to feed this model and during our initial analysis could not make the model work!

In our initial model, we assumed that there was NO reaction of the pesticide while it was in any compartment.   As a result the model predicted exposures that were about 5 times higher than what was actually measured in the glass chamber.   Clearly, the model got it wrong and needed to be refined to account for the "lost" material.

We asked the synthesis chemists about the stability of the pesticide in air or on glass surfaces and they said that it should be very stable for the time frame we were measuring (a few hundred hours).  

 

Because our model did not work, and notwithstanding the Chemists' comments, we hypothesized that perhaps that the combination of long residence time on the internal chamber glass surface combined with the large surface area-to-volume residue of the pesticide film on the glass could indeed lead to degradation.  This degradation would come from reaction with oxygen or trace amounts of tropospheric ozone or other reactive species present in the untreated suburban air used to ventilate the chamber. 

We changed the model to allow for degradation while on the chamber walls and SimuSolv allowed us to optimize the model for the degradation rate that provided the best fit to the data.    This eventually led to 0.005/hr as the estimated rate of degradation.  In 100 hours this predicts that 50% of the deposited pesticide would have degraded to other, typically less toxic species.

The Chemists congratulated us on our model fit but said that they did not believe that degradation was occurring.    That led to another series of experiments where we demonstrated after putting essentially pure pesticide on glass it was significantly transformed to almost a dozen chromatographically distinct species of compounds after prolonged exposure to ambient air.   Clearly, the significant rate of degradation that the model predicted was occurring.

The initial failure of the model allowed us to discover this important mechanism that was driving the concentration in the glass chamber.

Clearly, most of us do not live in glass houses and subsequent experiments with real rooms showed much stronger effects presumably from absorption (probably with degradation) were in play; however, the lesson here should not be lost.   Modeling can lead to some important discoveries.

We published most of the above work in the AIHA J and I would be happy to send a copy to anyone who requests it from me at:  mjayjock@gmail.com