13C NMR as a foundation for machine learning models of polysaccharides.

Abstract

Polysaccharide structural assignment via nuclear magnetic resonance (NMR) spectroscopy remains an analytical challenge due to spectral overlap because of limited chemical shift dispersion. This challenge is exacerbated by the wide-spread use of proton (¹H) detection. Progress has also been hindered by the dispersed nature of carbohydrate databases and the restricted applicability of most prediction tools, which provide limited atom-specific residue discrimination in larger polymers. These limitations hamper the efficient characterization of glycosyltransferase (GT) activity, which depends on defining donor-acceptor substrate pairs and accurately identifying corresponding products. Here, we demonstrate the utility of high field ¹³C-detected NMR for the assignment of two homohexamers. Laminarihexaose and xylohexaose were analyzed using 2D heteronuclear correlation and correlation via long-range coupling (COLOC) experiments to evaluate whether complete residue level assignments could be achieved. The COLOC experiment revealed long range correlations that were not resolved using heteronuclear multiple bond correlation, and the 1D ¹³C spectra provided exceptional resolution, including distinct shoulders corresponding to residue specific chemical shift differences previously assumed to be indistinguishable. These findings suggest that high field ¹³C NMR can provide the nuanced atom level information required to train machine learning models for predicting chemical shift assignments of large, complex, and highly degenerate polysaccharides. Such models offer a promising framework for rapid structural identification of GT reaction products, enabling progress of high throughput characterization of plant derived polysaccharides central to renewable biomaterial development.