The Epigenetic Dimension of Protein Structure Is an Intrinsic Weakness of the AlphaFold Program

Abstract

One of the most important lessons we have learned from sequencing the human genome is that not all proteins have a 3D structure. In fact, a large part of the human proteome is made up of intrinsically disordered proteins (IDPs) which can adopt multiple structures, and therefore, multiple functions, depending on the ligands with which they interact. Under these conditions, one can wonder about the value of algorithms developed for predicting the structure of proteins, in particular AlphaFold, an AI which claims to have solved the problem of protein structure. In a recent study, we highlighted a particular weakness of AlphaFold for membrane proteins. Based on this observation, we have proposed a paradigm, referred to as "Epigenetic Dimension of Protein Structure" (EDPS), which takes into account all environmental parameters that control the structure of a protein beyond the amino acid sequence (hence "epigenetic"). In this new study, we compare the reliability of the AlphaFold and Robetta algorithms' predictions for a new set of membrane proteins involved in human pathologies. We found that Robetta was generally more accurate than AlphaFold for ascribing a membrane-compatible topology. Raft lipids (e.g., gangliosides), which control the structural dynamics of membrane protein structure through chaperone effects, were identified as major actors of the EDPS paradigm. We conclude that the epigenetic dimension of a protein structure is an intrinsic weakness of AI-based protein structure prediction, especially AlphaFold, which warrants further development.

Keywords: AI; alphafold; ganglioside; lipid rafts; membrane; molecular modeling; pathology; protein structure; therapy.

Figure 1

Comparison of the structure of the epidermal growth factor receptor retrieved from Alpha Fold (depicted as cartoon colored in blue) or modelized via Ab-initio calculation on Robetta (red) (A). Molecular model of the insertion of the receptor obtained by Robetta in a lipid membrane environment (B). Comparison of the structure of the epidermal growth factor receptor resolved by Cryo-EM (PDB: 7SYD, resolved from the residue 25 to 638) ((C), green) with the structure obtained by Robetta ((D), red) and AlphaFold2 ((E), blue). In (E), the arrow points to an unstructured region that was predicted and inserted by the AlphaFold2 algorithm between the two extra-cellular domains.

Figure 2

Comparison of the structure of human synaptic vesicle glycoprotein C (h-SV2C) predicted by AlphaFold2 (blue) and Robetta (red). Insertion of the model predicted by Robetta into a lipid bilayer (A). Crystal structure of the luminal domain of h-SV2C in complex with BoNT/A1 (B). Structural alignment of h-SV2C predicted by AlphaFold2 and Robetta with the crystal structure of h-SV2C (C). Comparison of the energy of interactions of each h-SV2C-BoNT/A1 complex (D).

Figure 3

Structural conformational changes of BoNT/A1-h-SV2C complex involving the Robetta model (red) or the AlphaFold2 model (blue) after energy minimization.

Figure 4

Comparison of the structure of human synaptotagmin 1 (h-SYT1) (top panel) and APP (bottom panel) retrieved from AlphaFold2 (blue) or generated by ab-initio modeling with Robetta (red).

Figure 5

Structural comparison of the spatial organization of the domains of botulinum neurotoxin A obtained from Xray diffraction (PDB: 3BTA) (left), AlphaFold2 (middle) and Robetta (right) and molecular modeling of each structure with its membrane receptor human synaptic vesicle glycoprotein C (h-SV2C) in a neural membrane context. The toxin receptor h-SV2C is depicted as a cartoon colored in black. The phosphate atom of each POPC lipid is shown as brown spheres. GT1b molecules are represented as orange sticks and the lipid tail of POPC molecules are shown as thin blue lines.

Figure 6

Structural comparison of the spatial organization of the domains of botulinum neurotoxin B obtained from Xray diffraction (PDB: 2NP0) (left), AlphaFold2 (middle) and Robetta (right) and molecular modeling of each structure with its membrane receptor human synaptotagmin 1 in a neural membrane context. The toxin receptor h-SYT1 is depicted as cartoon colored in black. The phosphate atom of each POPC lipids is shown as brown spheres. The GT1b molecules are represented as orange sticks, the lipid tails of POPC molecules are shown as thin blue lines and the N-glycan and O-glycan of h-SYT1 are depicted as purple and red spheres respectively.

Figure 7

Comparison of Xray diffraction (PDB: 1HAQ), AlphaFold2 and Robetta models for complement factor H. In the case of this highly flexible protein, AlphaFold2 predicts a globular shape whereas Robetta’s model is rather elongated.

Figure 8

Lipid chaperone effect as a major parameter of the EDPS paradigm. The extracellular domain of synaptotagmin-2 (h-SYT2) is totally disordered when bound to ceramide (Cer), whereas it acquires a α-helix structure when bound to ganglioside GT1b. Both models were obtained with Hyperchem and submitted to energy minimization with the Polak–Ribière algorithm according to the protocol used for BoNT/A1-h-SV2C as described in Materials and Methods. Lipid molecular structures have been schematized for clarity. The global shape and volume of these lipids are directly responsible for these typical chaperone effects. This mechanism accounts for the critical role played by raft lipids on protein structure, illustrating the EDPS paradigm. TM, transmembrane domain.