Growing Glycans in Rosetta: Accuratede novoglycan modeling, density fitting, and rational sequon design

Abstract

Carbohydrates and glycoproteins modulate key biological functions. Computational approaches inform function to aid in carbohydrate structure prediction, structure determination, and design. However, experimental structure determination of sugar polymers is notoriously difficult as glycans can sample a wide range of low energy conformations, thus limiting the study of glycan-mediated molecular interactions. In this work, we expanded the RosettaCarbohydrate framework, developed and benchmarked effective tools for glycan modeling and design, and extended the Rosetta software suite to better aid in structural analysis and benchmarking tasks through the SimpleMetrics framework. We developed a glycan-modeling algorithm, GlycanTreeModeler, that computationally builds glycans layer-by-layer, using adaptive kernel density estimates (KDE) of common glycan conformations derived from data in the Protein Data Bank (PDB) and from quantum mechanics (QM) calculations. After a rigorous optimization of kinematic and energetic considerations to improve near-native sampling enrichment and decoy discrimination, GlycanTreeModeler was benchmarked on a test set of diverse glycan structures, or "trees". Structures predicted by GlycanTreeModeler agreed with native structures at high accuracy for both de novo modeling and experimental density-guided building. GlycanTreeModeler algorithms and associated tools were employed to design de novo glycan trees into a protein nanoparticle vaccine that are able to direct the immune response by shielding regions of the scaffold from antibody recognition. This work will inform glycoprotein model prediction, aid in both X-ray and electron microscopy density solutions and refinement, and help lead the way towards a new era of computational glycobiology.

Antigen- and scaffold-specific antibody responses to protein nanoparticle immunogens

“…To introduce NxT/S potential N-linked glycosylation sites (PNGS) into the exposed surfaces of the I53_dn5A pentamer and the I53_dn5B trimer, we used a custom “sugarcoat” protocol that we recently developed 1 as part of the Rosetta macromolecular modeling and design software. 68 , 69 Sequences corresponding to design models containing a single inserted NxT/S sequon, modeled with and without a Man9 glycan tree present, that had a Rosetta “total_energy” <500, as well as <0.25 Å and <0.40 Å backbone (Cα) root-mean-square deviation (RMSD) compared with the parent I53_dn5A and I53_dn5B design models, respectively, were tested for protein expression and glycosylation ( Figures S1 A and S1B).…”

“… 99 https://www.rbvi.ucsf.edu/chimerax/ Prism GraphPad https://www.graphpad.com/scientific-software/prism/ FlowJo v10 FlowJo https://www.flowjo.com RosettaScripts “sugarcoat” code Adolf-Bryfogle et al. 1 N/A Other EM supplies 300 mesh grids Ted Pella Cat# 01843-F Filter paper Cytiva Cat# 1004047 Uranyl formate SPI Chem Cat# 02545-AA Superdex 200 Increase SEC column Cytiva Cat# 28-9909-44 Superose 6 Increase SEC column Cytiva Cat# 29091596 Talon resin TaKaRa Cat# 635652 Excel resin Cytiva Cat# 17371203 Isoflurane USP Patterson Cat# 07-893-1389 EndoSafe LAL Test Cartridges Charles River Labs Cat# PTS20005F Lemo21(DE3) New England BioLabs Cat# C2528J Isopropyl-B-D-thiogalactoside (IPTG) Sigma-Aldrich Cat# I6758 Kanamycin Sulfate Sigma-Aldrich Cat# K1876 HisTrap FF Cytiva Cat#17525501 …”

Accurate de novo design of membrane-traversing macrocycles

“…For some runs, we also used the AddCompositionConstraintMover to limit the minimum and maximum number of allowed prolines, D-prolines, and some bulky hydrophobic amino acids in the designed peptides (Hosseinzadeh et al, 2017). Given the difficulty in synthesizing peptides with multiple N-methylated amino acids, we filtered the design models based on total number N-methylated amino acids in the design models and lack of any exposed NH groups in the final designed state using Rosetta SimpleMetrics (Adolf-Bryfogle et al, 2021).…”

A versatile Halo- and SNAP-tagged BMP/TGFβ receptor library for quantification of cell surface ligand binding

“…The ten best variants were discriminated by plotting the RMSD of all heavy backbone atoms vs. the Rosetta-Score (using scoring function ref2015) of the final 500 variants. The RMSD was calculated against the input structures (either BMPR2 or ALK2) using the RMSDMetric mover (Adolf-Bryfogle et al, 2021) in RosettaScripts. The ten variants with low REU-values (score) and low RMSD were visually inspected, and the most reliable structure based on convincing intramolecular interactions and comparison to known BMP-receptor-ECDs was used for receptor-ligand-docking.…”