Molecular shape similarity and “Bath Salts”: Anatomy of a collaboration

Being so intrigued by the mechanics of scientific collaboration it is a fun excercise to rewind and find out how one started work on a project in the first place. Lets take the result of a recent multigroup collaboration led by my long time friend and collaborator Matthew Krasowski (Dept of Pathology, University of Iowa). This collaboration in question led to a recent publication “Cross-reactivity studies and predictive modeling of “Bath Salts” and other amphetamine-type stimulants with amphetamine screening immunoassays” published in the past few weeks. More on that later. Since starting collaborating in the mid 2000’s we have also published extensively on using computational approaches for understanding nuclear receptor evoloution, then in 2007 we started looking at a different topic. Yes we both have day jobs! but over the years have been working on trying to understand one of the issues that plagues using immunoassays, that is cross reactivity or interference when used for detection of drugs in assays for therapeutic drug monitoring or drugs of abuse/toxicology. Perhaps this sounds a bit arcane, not exactly the hunt for a cure for an important disease but at its heart is the study of molecular shape and it is important in the clinical setting in particular for emergency medicine.

An immunoassay used in these scenarios, uses an antibody that is commonly raised against a single molecule (see figure below), e.g. a drug you want to detect (also called the antigen), say digoxin. The immunoassay is very sensitive to the molecule in question and such assays are widely used in an automated fashion in clinical testing laboratories in hospitals. Over the years however false positives have been deteremined for different immunoassays. For example, herbal products interfering with the digoxin immunoassay. Our rather simplistic hypothesis was that similar molecules to the compound (antigen) being detected by the immunoassay, would be more likely to cross react with the immunoassay. Interestingly, as far as we can ascertain nobody had tested this idea before. So we set about to compare molecules with 2D and 3D molecule similarity methods. One approach created a string of descriptors for each molecule (essentially a molecule specific fingerprint) and compared it to the immunoassay  compound of interest. Another approach created a pharmacophore shape for each compound and used these for comparison. We found that similarity using 2D MDL fingerprints worked well after looking at many different therapeutic drug monitoring and drugs of abuse assays. After publishing several papers and a book chapter on this concept nothing much happened, or at least so I thought.

Unbeknownst to me, “Bath Salts” or designer drugs (or other so called ‘legal highs’ or ‘plant food’) containing substituted cathinones (basically amphetamine like molecules) were sweeping the USA and elsewhere in 2010-2011. Bath salts are so called in an attempt to skirt laws so that  they could be legally sold with the disclaimer “not for human consumption”. These were creating a high level of interest at poison centers. Matthew was contacted in May 2011 by Robert Goetz at one such poison center in Cincinatti, he mentioned that products like Spice, Mr Nice Guy etc. containing mephedrone, and 3,4- methylenedioxypyrovalerone and other synthetic cannibinoids were creating a spike in poison center reports. He wondered if we had tried our approach on these sorts of molecules. This set in process a collaboration over the next year and a half with Matthew Petri (UCSF), Kara Lynch (Kaiser Permanente Northern California), and co workers in their labs. They tested the cross reactivity of 42 amphetamines and amphetamine like molecules with 3 different commercially available immunoassays called CEDIA, EMIT II and AxSYM (each is a registered copyright of their manufacturers, Thermo Fisher Scientific,  Siemens and Abbott Diagnostics, respectively). It should be noted that the CEDIA assay detects amphetamine and ecstacy (MDMA) While AxSYM detects amphetamine and methamphetamine.bath salts slide In parallel we compared 2D and 3D methods at predicting cross reactivity at concentrations of 5,000, 20,000 and 100,000 ng/ml and looked at consensus predictions (combining different methods to make a prediction). The experimental cross reactivity work suggested the different immunoassays showed differential detection of different compound classes. As for the computational methods, the 2D similarity using a consensus of amphetamine and MDMA and  amphetamine and MDA had receiver operator curve values of 0.97 (the ideal value would be 1). We also used the computational methods to predict a further 261 compounds for future testing.

The collaboration showed  conclusively that the most common “Bath Salts” were undetectable (i.e. unlikely to cross react) by common immunoassays for amphetamines used in Poison centers, and they would require mass spectrometry methods for detection instead. However, grouping compounds by structral class showed some immunoassays were better than others, a quite surprising finding. We also identified that compounds with a 2D similarity  score greater than 0.55 were likely to cross react with the different immunoassays. This would enable us to reliably suggest which compounds in future will need other techniques for detection.

So what next? We have an admittedly small dataset here that can be used for machine learning models to predict cross reactivity and it would be interesting to see if these are any improvement on the 2D similarity method. Perhaps obtaining more compounds and testing crossreactivity would be an ideal next study, however availability of pure compounds is probably not in the legal commercial domain. In some ways the many studies we have done to date both experimentally and computationally suggest some fuzziness or flexibility in the antigen binding sites of antibodies. Perhaps as good as any example to show the flexibility of the lock and key hypothesis. One tangental thought. Is there a compound that we can use as an antigen that would detect a greater breadth of bath salts and related compounds? Perhaps that would be something Thermo Fisher Scientific,  Siemens and Abbott Diagnostics, may want to look at.

From my perspective, this collaboration was a great example of a project that involved a great deal of communication via email consisting of excel sheets of molecular similarity data for hundreds of compounds, Receiver operator characteristic plots and drafts of the results and final manuscripts. Perhaps not the most efficient way to do it, but because of the long time scale involved it seemed to work. It would not be too difficult from the computational side to blog such a collaboration while in progress. Something to try next. This was also another example of how someone can see a publication and that sparks another collaboration.

We plan on making the molecule structures for the >200 bath salt compounds publically available in the near furture with all the data in the supplemental files. So watch this space. It will be interesting to see what springs from this work too.





Leave a Reply

Your email address will not be published.

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>