Does AlphaFold2 model proteins’ intracellular conformations? An experimental test using cross-linking mass spectrometry of endogenous ciliary proteins

Abstract

A major goal in structural biology is to understand protein assemblies in their biologically relevant states. Here, we investigate whether AlphaFold2 structure predictions match native protein conformations. We chemically cross-linked proteins in situ within intact Tetrahymena thermophila cilia and native ciliary extracts and identified 1,225 intramolecular cross-links within the 100 best-sampled proteins to provide a benchmark of distance restraints obeyed by proteins in their native assemblies. The corresponding AlphaFold2 structure predictions were highly concordant, positioning 86.2% of cross-linked residues within Cɑ-to-Cɑ distances of 30 Å, consistent with the known cross-linker length. 43% of the proteins showed no violations. Most inconsistencies occurred in low-confidence regions or between domains of the structure prediction. For basal body protein BBC118, cross-links combined with the predicted structure revealed domain packing satisfying both data. Overall, AlphaFold2 predicted biological structures with low predicted aligned error corresponded to more correct native structures. However, we observe cases where rigid body domains are oriented incorrectly, suggesting that combining structure prediction with experimental information will better reveal biologically relevant conformations.

Native doublet microtubules from Tetrahymena thermophila reveal the importance of outer junction proteins

“…We also examined side-chain fitting of all protein models and cross-links identified from cross-linking mass spectrometry of the cilia in situ (Supplementary Fig. 2A, B and Supplementary Table 5) 31 . With such an approach, we could identify and validate almost all Tetrahymena MIPs and show that they are forming a weaving network inside the DMT (Table 2 and Supplementary Tables 1-4).…”

Integrative modeling reveals the molecular architecture of the Intraflagellar Transport A (IFT-A) complex

“…Using AlphaFold’s predicted aligned error (PAE) confidence scores as a guide (as detailed in (56)), we constructed high-confidence models of interacting domains between IFT121-IFT122, IFT121-IFT139, IFT121-IFT43, IFT122-IFT139, IFT122-IFT140, IFT122-IFT144, and IFT140-IFT144 ( Figure 2—figure supplement 2A ). The combination of cross-linked residue pairs, AlphaFold2 monomer structures, and AlphaFold-Multimer domain-domain interaction models provided sufficient spatial restraints to build an initial model of IFT-A, which we improved by using a four-step integrative modeling approach (57–59) ( Figures 2 , Figure 2—figure supplement 3, Figure 2—figure supplement 4, and Methods ).…”

“…Pairwise interacting protein structures were predicted using the 2.1.2 version/release of AlphaFold-Multimer (55) as implemented on Texas Advanced Computing Center (TACC) Maverick2 and Frontera (110) GPU computer clusters. Predicted aligned error (PAE) plots were used as in (56) to determine interaction interfaces to be represented as rigid bodies in integrative modeling. We use the calibration curve from our previous studies to select pairwise interactions with PAE of less than 3.5 Å.…”

“…AlphaFold-Multimer (55) was used to model pairs of IFT-A proteins and compute the predicted aligned error (PAE, reported in Å, where darker colors indicate higher confidence in the relative spatial positions of two amino acid residues) between all pairs of amino acid residues in both structures. Using these PAE scores and thresholds determined by comparing PAE scores to independent structural evidence (56), we defined boundaries of well-predicted interacting domains to be used as rigid bodies for subsequent integrative modeling (see Methods ), as for the case shown for IFT121-IFT122 ( top panels ). The white box indicates the extracted interacting region, which in this case lies within larger well-predicted monomeric structures; the complete IFT121-122 interaction model is drawn at right.…”

Cross-linking mass spectrometry for mapping protein complex topologies in situ

“…In addition to integrative modelling approaches, the increased power of protein structure prediction in addressing multi-protein complexes mean that novel PPIs discovered by cross-linking MS can be structurally predicted, with predictions independently validated by fitting of models to the experimental distance restraints [ 20 , 106 , 107 ]. Cross-linking MS can be combined with other techniques to bring structural biology approaches directly in situ , in order to study the structure and function of molecular machines in their native environment.…”