Correcting for sparsity and non‐independence in glycomic data through a systems biology framework

Abstract

Glycans are fundamental cellular building blocks, involved in many organismal functions. Advances in glycomics are elucidating the roles of glycans, but it remains challenging to properly analyze large glycomics datasets, due to the fundamental nature of these data. The sparsity and dependency problem mainly lies in the detected glycans are treated as independent units without considering their many shared features of intermediate biosynthetic steps or glycan substructures. When glycoforms change significantly due to the genetic perturbation, the glycoprofiles share fewer glycans, which makes it not possible to quantitatively compare the profiles. We address these challenges with GlyCompare, a glycomic data analysis approach that leverages shared biosynthetic pathway intermediates to correct for sparsity and non‐independence in glycomics. Specifically, quantities of measured glycans are propagated to intermediate glycan substructures, which enables direct comparison of different glycoprofiles and increases statistical power. Using GlyCompare, we studied diverse N‐glycan profiles from glycoengineered erythropoietin. We obtained biologically meaningful clustering of mutant cell glycoprofiles and identified knockout‐specific effects of fucosyltransferase/sialyltransferase mutants on tetra‐antennary structures. We further analyzed human milk oligosaccharide profiles and identified novel impacts, the mother’s secretor‐status, on fucosylation and sialylation. Our substructure‐oriented approach will enable researchers to take full advantage of the growing power and size of glycomics data.Support or Funding InformationThis work was conducted with support from the Novo Nordisk Foundation provided to the Center for Biosustainability at the Technical University of Denmark (NNF10CC1016517: A.W.T.C.), NIGMS (R35 GM119850: B.B., B.K.P., N.E.L.), NICHD (R21 HD080682: L.B.) and USDA (USDA/ARS 6250‐6001; M.W.H)The problem we had in comparing glycoprofiles and it can be solved by GlyCompare.a, the glycan differences between WT and KO. Mgat4a/4b are the present/absent of the third branch targeted by Mgat4 family. The glycans are different which makes the glycoprofiles quantitatively incomparable. b, WT and KO. Mgat4a/4b glycoprofiles share part of biosynthesis steps and part of substructures, which makes the comparison possible in enzymatic perspective and substructure perspective. c, Sixteen glycoprofiles from glycoengineered recombinant EPO cluster poorly when based solely on raw glycan abundance. d, GlyCompare was used to cluster EPO glycoprofiles with selected‐substructure vectors, resulting in three dominant clusters of glycoprofiles and a few individuals that have severe changes in their glycan structural pattern (distance threshold=0.5) and thirty‐five clusters of glycan substructures.Figure 1GlyCompare summarizes the structural and enzymatical differences across KO glycoprofiles.a, a representative substructure table derives from the clustering table in Fig 1d. The substructures are reordered basing on the glycan structure complexity, followed by the number of branches, the degree of galactosylation, sialylation, and fucosylation. The significantly differential‐synthesized glycan substructures are illustrated by zscore‐scaled abundance of thirty‐five glycan substructures, compared with WT. b, KO.Fut8 and KO.St3gal4/6 are used to illustrate the synthetic changes compared to the WT. The rescaled substructure clusters’ abundance in c are visualized with a simplified network. In KO. Fut8, the abundance of non‐fucosylated LacNac‐elongated structure are all increased. In KO. St3gal4/6, the non‐sialic acid LacNAc‐elongated structure are all increased. c, the product/substrate ratio changes imply the enzyme activity changes and help explain the substructure changes. KO. St3gal4/6 loses the total reaction in A3SIAT and has increased enzyme activity in iGNT. KO. Fut8 loses the reaction in A6FUCT completely and has increased enzyme activity in iGNT.Figure 2