Fidelity of glycan-code synthesis in a realistic Golgi: extending optimal information coding to the branched N-glycosylation network, molecular noise, and self-assembled cisternae

Yadav, Vagne, Sens, Iyengar and Rao (eLife 2022) modeled Golgi N-glycan processing as an information-coding channel and showed, in a simplified mean-field model of line-like reaction sequences with the cisternal number as a free parameter, that synthesizing a complex target glycan distribution with high fidelity requires multiple cisternae, precise enzyme partitioning, and an optimal (intermediate) enzyme specificity. The paper itself poses the open follow-up: does this link between glycan-code fidelity and the Golgi machinery survive, and what new constraints appear, when the model is made realistic — the full branched N-glycosylation network with multiple glycoprotein substrates, stochastic kinetics at low copy numbers, mechanistic parameter constraints, discrete enzyme allocation across cisternae, a cisternal number that self-assembles from membrane dynamics rather than being imposed, and non-equilibrium transport that could act as error correction. Prior work leaves these load-bearing aspects open: the closest follow-up (Satyajith and Thattai, 2026) resolves multi-glycan enzyme-partitioning trade-offs in a yield-based Pareto framing (jointly optimizing the production yields of several glycans rather than the information fidelity of a target distribution) on idealized trees, not the fidelity-to-target-distribution objective; related Golgi models treat cisternal shape or membrane self-organization without coupling them to the fidelity of the glycan code.

Background

The Golgi apparatus is the eukaryotic organelle where many secreted and membrane proteins receive their final post-translational modifications. It is organized as a stack of flattened membrane-bound sacs called cisternae; cargo enters at the cis face, is modified while traversing successive cisternae, and exits at the trans face. In N-glycosylation — the modification studied here — branched sugar chains (glycans) are built and trimmed on proteins by glycosyltransferases and glycosidases (enzymes that respectively attach sugars to, and cleave sugars from, the growing glycan) that reside in specific cisternae, together with inter-cisternal transport of the cargo. Because there is no nucleic-acid-like template for glycans, the set of glycan structures on a cell's surface (its glycome) is determined by which enzymes act, where, and for how long; glycans nevertheless form a reproducible, cell-type-specific 'molecular code' recognized by glycan-binding proteins. Two modeling traditions address this system. Elaborated kinetic network models (Umaña and Bailey 1997; Krambeck and Betenbaugh 2005; Krambeck et al. 2009; Fisher et al. 2019) represent thousands of reactions and oligosaccharide structures with Michaelis–Menten kinetics and predict glycoform distributions from enzyme activities, but they take the enzymes' cisternal arrangement as input and do not ask what the information carried by glycans requires of the organelle. Algorithmic compartment models (Jaiman and Thattai 2020) showed that successive compartments let promiscuous, stochastic enzymes produce a controlled rather than smeared product spectrum. Yadav, Vagne, Sens, Iyengar and Rao (eLife 2022; arXiv:2005.08740) reframed the question information-theoretically: they define the fidelity of a synthesis mechanism as the minimum normalized Kullback–Leibler divergence (KL divergence is the standard information-theoretic distance between two probability distributions; normalizing by the target's entropy makes it a relative error) between the synthesized surface glycan distribution and a prescribed target distribution, and define the complexity of a target as the number of Gaussian-mixture-model components (a standard density model built by superposing Gaussian peaks) needed to approximate it to the noise floor. Optimizing a simplified model — line-like sequential reactions, mean-field (deterministic) kinetics, enzyme specificity summarized by one parameter $\sigma$, and the number of cisternae $N_C$ treated as a tunable parameter rather than a self-assembled structure — they found that (i) complex targets cannot be reached with few cisternae no matter how the enzymatic parameters are tuned, so the glycan code's fidelity functionally constrains $N_C$; (ii) for fixed enzyme number $N_E$ and $N_C$ there is an optimal specificity: overly promiscuous enzymes create wrong glycans, overly specific ones cannot build diversity; (iii) the fidelity landscape has stiff directions — parameter combinations along which fidelity varies sharply (transport rate, specificity) — and sloppy directions, along which fidelity is nearly flat (many reaction rates), implying robustness and degeneracy. The authors explicitly flag what the framework has not yet established: the model is mean-field and ignores noise from low copy numbers of enzymes and substrates, from multiple substrates competing for the same enzymes, and from stochastic transport; the reaction sequences are linear caricatures of the branched N-glycosylation network, which in reality involves many glycoprotein substrates and parallel pathways; enzyme partitioning across cisternae is inferred indirectly from effective rates rather than determined by an explicit discrete-allocation optimization; optimized effective rates are constrained only by box bounds derived from Michaelis–Menten parameters ($V_{max}$, the maximal reaction rate, and $K_M$, the substrate concentration at which the rate is half of $V_{max}$) rather than by mechanistic physical constraints; the cisternal number is imposed, whereas a real Golgi assembles and maintains a definite number of cisternae through membrane fission, fusion and transport (a dynamics studied in non-equilibrium Golgi models such as Sachdeva, Barma and Rao 2016 and Vagne, Vrel and Sens 2020, but never coupled to glycan fidelity); and non-equilibrium transport might additionally serve as a proofreading or error-correction mechanism (proofreading here means spending energy to suppress synthesis errors, as in kinetic proofreading) with optimal adaptive strategies when the target distribution varies in time. The open problem is therefore whether the fidelity–machinery link, established only in the simplified setting, extends to the realistic system, and what constraints on the Golgi's structure and enzyme repertoire high-fidelity glycan-code synthesis imposes there.

Problem Statement

Extend the information-coding framework of Yadav et al. (2022) from its simplified setting to a mechanistically realistic Golgi and determine what the requirement of high-fidelity glycan-code synthesis then constrains. The single scientific objective is a quantitative account, at the level of the fidelity metric (minimum normalized KL divergence between the synthesized and prescribed target glycan distributions), of how glycan-code fidelity trades off against — and constrains — the number of Golgi cisternae, the number and specificity of processing enzymes, their partitioning across cisternae, and inter-cisternal transport, as a function of the prescribed target's complexity (the number of Gaussian-mixture components needed to approximate it) and across the realistic parameter space of the branched network — enzyme repertoire and specificities, their discrete partitioning across cisternae, enzyme and substrate copy numbers, and the transport and membrane-dynamics rates — as the general relation over that target-complexity class and parameter space, not a result at a single self-chosen target and parameter point. The simplifications the source analysis relied on are removed jointly and within one framework, with each component serving that common objective: (i) replace the line-like reaction sequences with the branched N-glycosylation network, including multiple glycoprotein substrates competing for shared enzymes and parallel pathways, and determine optimal enzyme localization by explicit discrete (Boolean or integer) allocation across cisternae rather than by inference from effective rates, under mechanistically meaningful kinetic constraints rather than box bounds on aggregated rates; (ii) replace mean-field kinetics with stochastic dynamics, so that fidelity is evaluated in the presence of molecular noise from low copy numbers of enzymes and substrates and from stochastic transport; (iii) couple the chemical processing to non-equilibrium membrane dynamics (cisternal fission, fusion and transport) so that the number of cisternae is an emergent, self-assembled outcome — and ask whether the fidelity requirement then selects a definite cisternal number, rather than treating it as a free parameter; and (iv) within the same realistic model, characterize whether non-equilibrium inter-cisternal transport provides proofreading or error correction that improves fidelity, including the trade-off between robustness to a time-varying target environment and fidelity to a fixed complex target. The components of this account — the fidelity landscape and its stiff/sloppy geometry in the realistic parameter space, the existence and location of optimal enzyme specificities and partitions, the derived functional constraints on cisternal number and enzyme repertoire, and the error-correction capacity of transport — jointly bear on one question: does the glycan-information channel of the real Golgi have the capacity and fidelity properties the simplified model predicted, and what machinery does achieving a complex target glycome actually require? The four extensions are not separate questions: the account is sought in the regime where the branched stochastic network with discrete enzyme allocation and mechanistic kinetic constraints, the stochastic low-copy-number kinetics, the self-assembled cisternal number, and the transport error-correction capacity are all present together in one framework.

References

1. Yadav A, Vagne Q, Sens P, Iyengar G, Rao M (2022) 'Glycan processing in the Golgi as optimal information coding that constrains cisternal number and enzyme specificity', eLife 11:e76757, doi:10.7554/eLife.76757, https://doi.org/10.7554/eLife.76757 (preprint: arXiv:2005.08740, doi:10.48550/arXiv.2005.08740, posted under the title 'Glycan processing in the Golgi -- optimal information coding and constraints on cisternal number and enzyme specificity')
2. Umaña P, Bailey JE (1997) 'A mathematical model of N-linked glycoform biosynthesis', Biotechnology and Bioengineering 55:890–908, doi:10.1002/(SICI)1097-0290(19970920)55:6<890::AID-BIT7>3.0.CO;2-B, https://doi.org/10.1002/(SICI)1097-0290(19970920)55:6%3C890::AID-BIT7%3E3.0.CO;2-B
3. Krambeck FJ, Betenbaugh MJ (2005) 'A mathematical model of N-linked glycosylation', Biotechnology and Bioengineering 92:711–728, doi:10.1002/bit.20645, https://doi.org/10.1002/bit.20645
4. Krambeck FJ, Bennun SV, Narang S, Choi S, Yarema KJ, Betenbaugh MJ (2009) 'A mathematical model to derive N-glycan structures and cellular enzyme activities from mass spectrometric data', Glycobiology 19:1163–1175, doi:10.1093/glycob/cwp081, https://doi.org/10.1093/glycob/cwp081
5. Fisher P, Spencer H, Thomas-Oates J, Wood AJ, Ungar D (2019) 'Modeling Glycan Processing Reveals Golgi-Enzyme Homeostasis upon Trafficking Defects and Cellular Differentiation', Cell Reports 27:1231–1243, doi:10.1016/j.celrep.2019.03.107, https://doi.org/10.1016/j.celrep.2019.03.107
6. Jaiman A, Thattai M (2020) 'Golgi compartments enable controlled biomolecular assembly using promiscuous enzymes', eLife 9:e49573, doi:10.7554/eLife.49573, https://doi.org/10.7554/eLife.49573
7. Sachdeva H, Barma M, Rao M (2016) 'Nonequilibrium description of de novo biogenesis and transport through Golgi-like cisternae', Scientific Reports 6:38840, doi:10.1038/srep38840, https://doi.org/10.1038/srep38840
8. Vagne Q, Vrel J-P, Sens P (2020) 'A minimal self-organisation model of the Golgi apparatus', eLife 9:e47318, doi:10.7554/eLife.47318, https://doi.org/10.7554/eLife.47318
9. Satyajith A, Thattai M (2026) 'Pareto-optimal synthesis of multiple glycans in Golgi compartments', Biophysical Journal 125(14):3755–3765, doi:10.1016/j.bpj.2026.06.019 (preprint doi:10.1101/2025.03.14.643235), https://doi.org/10.1016/j.bpj.2026.06.019
10. Revell CK, Lowe M, Stevenson NL, Jensen OE (2026) 'Morphological determinants of glycosylation efficiency in Golgi cisternae', PLOS Computational Biology 22(3):e1013993, doi:10.1371/journal.pcbi.1013993, https://doi.org/10.1371/journal.pcbi.1013993
11. Floyd C, Dinner AR, Murugan A, Vaikuntanathan S (2025) 'Limits on the computational expressivity of non-equilibrium biophysical processes', Nature Communications 16:7184, doi:10.1038/s41467-025-61873-0, https://doi.org/10.1038/s41467-025-61873-0
12. Akune-Taylor Y, Kon A, Aoki-Kinoshita KF (2024) 'In silico simulation of glycosylation and related pathways', Analytical and Bioanalytical Chemistry 416(16), doi:10.1007/s00216-024-05331-8, https://doi.org/10.1007/s00216-024-05331-8
13. Iyer KS, Prabhakara C, Mayor S, Rao M (2023) 'Cellular compartmentalisation and receptor promiscuity as a strategy for accurate and robust inference of position during morphogenesis', eLife 12:e79257, doi:10.7554/elife.79257, https://doi.org/10.7554/elife.79257
14. Fuchsberger FF, Kim D, Baranova N, Vrban H, Kagelmacher M, Wawrzinek R, Rademacher C (2023) 'Information transfer in mammalian glycan-based communication', eLife 12:e69415, doi:10.7554/elife.69415, https://doi.org/10.7554/elife.69415
15. Joshi P, Satyajith A, Panda D, R RB, Thattai M, Balasubramanian N (2024) 'Cell-surface glycans are quantitative reporters of Golgi dysfunction in single cells', bioRxiv 2024.06.17.599374, doi:10.1101/2024.06.17.599374, https://doi.org/10.1101/2024.06.17.599374