Multivariate analysis applied to complex biological medicines
A biological medicine is a term for a medicinal compound that is derived from a living organism. By nature they are complex and are often heterogenous in composition and their biological activity. Some of the oldest pharmaceutical products are biologicals, for example insulin and heparin. The former is now produced recombinantly, with technology being at a point that this can be considered a defined chemical entity. This is not the case for the latter, however. Heparin is heterogenous polysaccharide, that is extracted from the intestinal mucosa of animals, primarily porcine, although there is also a significant market though for non-porcine heparin. In 2008 heparin was adulterated with another sulfated polysaccharide. Unfortunately, this event was disastrous, resulting in a global public health emergency. This was the impetuous to apply modern analytical techniques, principally NMR spectroscopy, and multivariate analyses to monitor heparin. Initially, traditional unsupervised multivariate analysis (principal component analysis) was applied to the problem. This was able to distinguish animal heparins from each other, and could also separate adulterated heparin from what was considered bona fide heparin. Taught multivariate analysis function by training the analysis to look for specific patterns within the dataset of interest. If this approach was to be applied to heparin, or any other biological medicine, it would have to be taught to find every possible alien signal. The opposite approach would be more efficient; defining the complex heterogenous material by a library of bona fide spectra and then filtering test samples with these spectra to revea alien features that are not consistent with the reference library. This is the basis of an approach termed spectral filtering, which has been applied to 1D- and 2D-NMR spectra, and that has been very successful in extracting the spectral features of adulterants in heparin, as well as being able to differentiate supposedly similar biosimilar products. In essence, the filtered spectrum is determined by subtracting the covariance matrix of the library spectra from the covariance matrix of the library spectra plus the test spectrum. These approaches are universal and could be applied to biological medicines such as vaccine polysaccharides and monoclonal antibodies.
- This article is part of the themed collection: Challenges in analysis of complex natural mixtures