Computational Study of Glycosaminoglycan Specificity for Growth Factor and Chemokine Family Members

Abstract

Glycosaminoglycans (GAGs), linear polysaccharides with repeating disaccharide units, bind to proteins through primarily electrostatic interactions, which tend to induce considerable non‐specificity of recognition. Yet, GAG–protein interactions are known to regulate various physiological and pathological functions. To date, only one sequence the heparin pentasaccharide sequence that binds to antithrombin has been identified as a specific sequence with therapeutic promise. Are specific GAG sequences so rare despite the millions of natural occurring GAG sequences? Our hypothesis has been that GAG sequences that bind to proteins are not rare but it has been difficult to identify them due to the phenomenal heterogeneity of GAGs and the near impossibility of synthesizing all possible natural (and unnatural) sequences. For example, a natural hexasaccharide sequence belonging to the heparan sulfate family can be of 46,656 types. To overcome this, our lab has developed a computational strategy called as Combinatorial Virtual Library Screening (CVLS), which is exhibit impressive application in identifying sequences displaying a high level of specificity against chosen targets. Here, we extend the applicability of CVLS by adding molecular dynamics (MD) simulations as a powerful approach to identify specific and non‐specific GAG sequences for a particular target. Application of CVLS‐MD approach to well‐known GAG target, fibroblast growth factor 2 (FGF2) and chemokine CXCL13 led to unique understanding on the nature of GAG recognition and origin of specificity for therapeutic targeting. Libraries of 2592 tetrasaccharides (HS04) and 46,656 hexasaccharides (HS06) were screened against both protein targets. CVLS identified 14 HS04 sequences as highly promising sequences for CXCL13, whereas FGF2 was found to be better recognized by 9 HS06 sequences. The co‐complexes of these sequences were studied using MD in the presence of explicit water and analyzed to understand specificity of interaction through analysis of direct and indirect hydrogen bonding, single residue energy decomposition and binding free energy. There was a strong correlation of CVLS‐MD predictions with experimental data. Interesting, FGF2–HS06 co‐complex exhibited differences in hydrogen bonding pattern and binding free energy due to IdoA2S ring puckering (both 1C4 to 2SO), whereas CXCL13 showed preferential recognition of one HS04 binding site of the two possible on protein surface. Most importantly, the CVLS–MD approach identified HS sequences showing specific as well as plastic (non‐specific) hydrogen bonding interactions in solution. The in silico predictions corresponded very well with solution NMR chemical shift perturbation (CSP) data resulting in identification of a specific HS04 sequence that stabilizes the chemokine dimer. Overall, this combined CVLS–MD approach is likely to serve as a promising strategy for the identification of specific and non‐specific GAG sequences against various target proteins.Support or Funding InformationThis work was supported in part by grants from the NIH including HL107152, HL090586 and HL128639. We thank Lortat‐Jacob H, University of Grenoble Alpes, CNRS, CEA, IBS, 38000 Grenoble, France hugues.lortat-jacob@ibs.fr. for experimental dataThis abstract is from the Experimental Biology 2018 Meeting. There is no full text article associated with this abstract published in The FASEB Journal.