A continuous fractionation-chemometric framework for resolving collinearity in polysaccharide structure-activity relationships: application to Syzygium aromaticum glycoprotein 9224P.

The glycoprotein 9224P from Syzygium aromaticum (clove) exhibits anti-SARS-CoV-2 activity and binds to RNA-dependent RNA polymerase (RdRp), yet its broad molecular-weight (Mw) distribution has obscured which structural features drive this activity. Conventional fractionation pools consecutive gel-filtration eluates, thereby averaging out the continuous variation in molecular weight and monosaccharide composition. Moreover, because monosaccharide abundances are expressed as molar percentages, the resulting closure constraint can induce spurious correlations that complicate structure-activity analysis. To address these challenges, 9224P was subjected to three parallel Sephacryl S-300 HR runs, yielding 25 consecutive fractions (apparent Mw 5-57 kDa). Correlation analyses were performed on centered log-ratio (CLR)-transformed compositions, whereas principal component analysis (PCA) and partial least squares regression (PLS-R) were based on biologically interpretable isometric log-ratio (ILR) coordinates representing GalA versus neutral sugars (ILR1), Rha versus Ara and Gal (ILR2), and Ara versus Gal (ILR3). Fractions with apparent Mw ≤ 10 kDa showed significantly greater antiviral activity than those with Mw > 30 kDa (mean inhibition 62% versus -47%; p < 0.001). The CLR coordinate of GalA was positively correlated with inhibition (r = 0.76). A two-latent-variable PLS-R model identified apparent Mw (VIP = 1.22) and ILR1 (VIP = 1.14) as the most influential predictors, with ILR2 (VIP = 1.06) as a possible secondary contributor (R2 = 0.68, Q2 = 0.56). ILR-PCA separated high- and low-activity fractions along PC1 (67.3% of the variance). These findings suggest that lower apparent Mw, a higher GalA-to-neutral-sugar balance, and relative Rha enrichment are candidate quality attributes for 9224P.