CoChat
Sign Up
Cited by 574Protein Structure and DynamicsEnzyme Structure and FunctionReceptor Mechanisms and Signaling

Predicting multiple conformations via sequence clustering and AlphaFold2

Open access
Wayment-Steele, Hannah K., Ojoawo, Adedolapo, Otten, Renee, Apitz, Julia M., Pitsawong, Warintra, Hömberger, Marc, Ovchinnikov, Sergey, Colwell, Lucy +1 more·2024·Nature
DOI 10.1038/s41586-023-06832-9PMC PMC10808063
Open this paper in CoChat — free account

Abstract

AlphaFold2 (ref. 1 ) has revolutionized structural biology by accurately predicting single structures of proteins. However, a protein’s biological function often depends on multiple conformational substates 2 , and disease-causing point mutations often cause population changes within these substates 3,4 . We demonstrate that clustering a multiple-sequence alignment by sequence similarity enables AlphaFold2 to sample alternative states of known metamorphic proteins with high confidence. Using this method, named AF-Cluster, we investigated the evolutionary distribution of predicted structures for the metamorphic protein KaiB 5 and found that predictions of both conformations were distributed in clusters across the KaiB family. We used nuclear magnetic resonance spectroscopy to confirm an AF-Cluster prediction: a cyanobacteria KaiB variant is stabilized in the opposite state compared with the more widely studied variant. To test AF-Cluster’s sensitivity to point mutations, we designed and experimentally verified a set of three mutations predicted to flip KaiB from Rhodobacter sphaeroides from the ground to the fold-switched state. Finally, screening for alternative states in protein families without known fold switching identified a putative alternative state for the oxidoreductase Mpt53 in Mycobacterium tuberculosis . Further development of such bioinformatic methods in tandem with experiments will probably have a considerable impact on predicting protein energy landscapes, essential for illuminating biological function.

AI snapshot

Written from the paper’s full text.

TL;DR

AF-Cluster clusters a protein multiple-sequence alignment by sequence similarity and feeds each cluster to AlphaFold2, enabling sampling of alternative conformational states for metamorphic proteins. The authors experimentally validate predicted KaiB states and designed point mutations, and identify a putative alternative state for Mpt53.

Key findings

  • AF-Cluster predicted both ground and fold-switched states for KaiB, RfaH and MAD2; for RfaH, top-five-model plDDT was 84.2 for the active state and 73.9 for the autoinhibited state, versus 68.6 for the autoinhibited state from the complete MSA.
  • NMR confirmed the AF-Cluster prediction for KaiBTV-4: CS-Rosetta models were within 1.8 ± 0.3 Å r.m.s.d. to the AF-Cluster model, and the NMR structure matched with 1.89 ± 0.13 Å r.m.s.d. and average pairwise r.m.s.d. of 0.7 Å over backbone atoms.
  • Three mutations (I68R, V83D, N84R) switched KaiBRS from the ground to the fold-switched state; NMR showed the mutant occupied the FS state at 86% versus 11% in wild type, and the ground state at 14% versus 89% in wild type.
  • Screening 628 protein families identified a putative alternative state for M. tuberculosis Mpt53; in the GA/GB system, AF-Cluster correctly predicted the most stable fold for 10 out of 12 point-mutation sets, whereas default AF2 predicted 8 out of 12.

Methods

  • MSAs were generated with ColabFold/MMseqs2, sequences with >25% gaps removed, then clustered by edit distance using DBSCAN; each cluster was used as input to AlphaFold2, with MSA Transformer used to probe evolutionary couplings.
  • Experimental validation used NMR spectroscopy, CS-Rosetta, NOESY-based structure calculation, SEC–MALS, and a curated KaiB phylogenetic tree of 487 variants.
  • The fold-switch screen used 628 proteins of 48–150 residues with >1,000 MSA sequences from a database of MSAs associated with crystal structures; Mpt53 homologues were analyzed with DALI and phylogenetic analysis.

Limitations

  • The paper states that pLDDT cannot be used as a measure of free energy; the thermodynamically disfavoured KaiB FS state still had higher pLDDT than the ground state, and the number of models per state reflects construct abundance rather than Boltzmann weight.
  • The paper states that AF-Cluster could not predict the oligomeric states of selecase, lymphotactin or CLIC1, and that conformational substates absent from the evolutionary signal may require alternative methods.
  • The paper notes that novel computationally identified folds may be difficult to validate experimentally if the triggering stimulus is unknown; the Mpt53 alternative state is a prediction without experimental validation in this work.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41586-023-06832-9

Cites

90 of 90 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 49 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    Predictions of both conformations of KaiB were distributed in clusters across the KaiB family.

    “Using this method, named AF-Cluster, we investigated the evolutionary distribution of predicted structures for the metamorphic protein KaiB5 and found that predictions of both conformations were distributed in clusters across the KaiB family.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [5] Chang, Y. G. et al. Circadian rhythms. A protein fold switch joins the circadian oscillator to clock output in cyanobacteria. Science 349, 324–328 (2015). source

      Not checked in the free run.

  • Findingown data

    Clustering a multiple-sequence alignment by sequence similarity enables AlphaFold2 to sample alternative states of known metamorphic proteins with high confidence.

    “We demonstrate that clustering a multiple-sequence alignment by sequence similarity enables AlphaFold2 to sample alternative states of known metamorphic proteins with high confidence.”
    Quote in the paperSubstantial claim
  • Findingown data

    A set of three designed mutations flips KaiB from Rhodobacter sphaeroides from the ground to the fold-switched state, experimentally verified.

    “To test AF-Cluster’s sensitivity to point mutations, we designed and experimentally verified a set of three mutations predicted to flip KaiB from Rhodobacter sphaeroides from the ground to the fold-switched state.”
    Quote in the paperSubstantial claim
  • Findingown data

    Screening for alternative states in protein families without known fold switching identified a putative alternative state for the oxidoreductase Mpt53 in Mycobacterium tuberculosis.

    “Finally, screening for alternative states in protein families without known fold switching identified a putative alternative state for the oxidoreductase Mpt53 in Mycobacterium tuberculosis.”
    Quote in the paperSubstantial claim
  • Findingown data

    A cyanobacteria KaiB variant is stabilized in the opposite state compared with the more widely studied variant, as confirmed by NMR.

    “We used nuclear magnetic resonance spectroscopy to confirm an AF-Cluster prediction: a cyanobacteria KaiB variant is stabilized in the opposite state compared with the more widely studied variant.”
    Quote in the paperSubstantial claim
  • May combine two claims. 33
  • Classified as probably not a substantial claim. 21
  • The quote only partly states this claim. 1

  • May combine two claims.

    Findingown data

    The average plDDT for the top five models for each RfaH state (84.2 active, 73.9 autoinhibited) was higher than the plDDT of the autoinhibited state by the complete MSA (68.6), suggesting that clustering resulted in deconvolving conflicting sets of couplings.

    “Notably, the average plDDT for the top five models for each state (84.2 for the active state, 73.9 for the autoinhibited) was higher than the plDDT of the autoinhibited state by the complete MSA (plDDT of 68.6), suggesting that clustering resulted in deconvolving conflicting sets of couplings.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The MSA clusters contained evolutionary couplings for both states, and the score based on contact maps correlated with the r.m.s.d. in AF2.

    “We used the same set of clusters to make predictions using the unsupervised deep learning model MSA Transformer36 and found that these clusters contained evolutionary couplings for both states, and the score based on contact maps correlated with the root mean squared deviation (r.m.s.d.) in AF2 (Methods and Extended Data Fig. 3).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [36] Rao, R. M. et al. MSA Transformer. In Proc. International Conference on Machine Learning 8844–8856 (PMLR, 2021). source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    AF2 predictions from MSA clusters comprised a distribution of structures, with the highest-scored regions corresponding to the ground and FS state.

    “Notably, we found that the AF2 predictions from our MSA clusters comprised a distribution of structures, with the highest-scored regions of the distribution corresponding to the ground and FS state.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Two prominent states are observed in AF-Cluster models for Mpt53 that correspond to the largest-sized MSA clusters, and both have plDDT values statistically significantly higher than the rest of the set.

    “Two prominent states are observed that correspond to the largest-sized MSA clusters (Extended Data Fig. 7a), and both of which have plDDT values that are statistically significantly higher than the rest of the set (Extended Data Fig. 7b).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Clusters predicting the FS state were enriched for arginine at position 68, whereas clusters predicting the ground state at position 68 were enriched for leucine.

    “For example, clusters predicting the FS state were enriched for arginine at position 68, whereas clusters predicting the ground state at position 68 were enriched for leucine, a switch between a charged and a hydrophobic residue.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For uniformly subsampled MSAs of size 10, 1 out of 500 samples was within 3 Å of the ground state, with lower confidence than the MSA cluster samples; uniformly subsampled MSAs of size 100 did not sample the ground state at all.

    “We found that, for uniformly subsampled MSAs of size 10, 1 out of 500 samples was within 3 Å of the ground state, with lower confidence than the MSA cluster samples (Extended Data Fig. 2e). Uniformly subsampled MSAs of size 100 did not sample the ground state at all.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    KaiB variants from R. sphaeroides, T. elongatus and S. elongatus were predicted in the ground state, confirming their characterized circadian-rhythm function.

    “For example, variants from R. sphaeroides32, T. elongatus5 and Synechococcus elongatus31 all were predicted in the ground state, confirming their characterized circadian-rhythm function.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [32] Pitsawong, W. et al. From primordial clocks to circadian oscillators. Nature 616, 183–189 (2023). source

      Not checked in the free run.

    • cites [5] Chang, Y. G. et al. Circadian rhythms. A protein fold switch joins the circadian oscillator to clock output in cyanobacteria. Science 349, 324–328 (2015). source

      Not checked in the free run.

    • cites [31] Ishiura, M. et al. Expression of a gene cluster kaiABC as a circadian feedback process in cyanobacteria. Science 281, 1519–1523 (1998). source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    The mutant KaiBRS-3m occupies the FS state with a population of 86% (versus 11% in WT) and the ground state with a population of 14% (versus 89% in WT).

    “Comparison of the average peak intensity ratios of the assignable minor (ground state) peaks to those of the major state (FS) peaks show that the mutant occupies the FS state with a population of 86% (versus 11% in the WT), and the ground state with a population of 14% (versus 89% in the WT) (Fig. 3f).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    One Mpt53 state corresponds to the known thioredoxin-like conformation, whereas the other corresponds to a conformation with a different secondary structure layout.

    “One state corresponds to the known thioredoxin-like conformation of Mpt53 (ref. 42), whereas the other state corresponds to a conformation with a different secondary structure layout (Fig. 5d,e).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [42] Goulding, C. W. et al. Gram-positive DsbE proteins function differently from Gram-negative DsbE homologs. A structure to function analysis of DsbE from Mycobacterium tuberculosis. J. Biol. Chem. 279, 3516–3524 (2004). source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    The closest PDB structure found (PDB: 3EMX) adopted a similar secondary structure to the Mpt53 alternative structure; it is an unspecified thioredoxin from Aeropyrum pernix with no associated publication.

    “The closest structure that we found (PDB: 3EMX) adopted a similar secondary structure to the Mpt53 alternative structure. This structure is of an unspecified thioredoxin from the archaea Aeropyrum pernix with no associated publication.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites 3EMX

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    Predicting the 3D structure of KaiB using just the 50 MSA sequences closest by edit distance to KaiBTE resulted in a prediction of the ground state; using the closest 100 sequences returned to predicting the FS state.

    “Notably, predicting the 3D structure of KaiB using just the 50 MSA sequences that are closest by number of mutations (hereafter, edit distance) to KaiBTE resulted in a prediction of the ground state (Fig. 1c (right)); however, predicting the 3D structure of KaiBTE using the closest 100 sequences returned to predicting the FS state.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    KaiB-4 variants were evolutionarily close to KaiB-1 variants yet were predicted by AF-Cluster to primarily occupy the FS state.

    “Notably, we noticed that KaiB-4 variants were evolutionarily close to the better-studied KaiB-1 variants involved in the circadian clock, yet the KaiB-4 variants were predicted by AF-Cluster to primarily occupy the FS state (Fig. 2c).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the second Mpt53 state, strand β1 replaces β5 within the β-sheet, α-helix α4 is displaced to the opposite side of the β-sheet, and α5 is rotated.

    “In the second state, strand β1 replaces β5 within the β-sheet. The α-helix α4 is displaced to the opposite side of the β-sheet, and α5 is rotated.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Using AF-Cluster, both states are sampled and scored with high confidence by AF2's learned plDDT measure.

    “Importantly, we show that, using our method, AF-Cluster, both states are sampled and scored with high confidence by AF2’s learned predicted local distance difference test (plDDT) measure.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    The 1H-15N HSQC spectra in both WT and KaiBRS-3m indicate the presence of major and minor state peaks, with the populations appearing to be flipped.

    “The 1H-15N heteronuclear single quantum coherence (HSQC) spectra in both the wild-type (WT) and KaiBRS-3m indicate the presence of major and minor state peaks, with the populations appearing to be flipped (Fig. 3d).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    KaiBTV-4 was found to be stably folded at 35 °C and showed peak duplication for many peaks corresponding to a major stable and minor unfolded state.

    “KaiBTV-4 was found to be stably folded at 35 °C and, after backbone assignments, we found peak duplication for many peaks corresponding to a major stable and minor unfolded state (Extended Data Fig. 4).”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingreasoning

    For epsilon > epsmax, more sequences are clustered, so the number of clusters is decreasing because clusters are merged.

    “For epsilon > epsmax, more sequences are clustered, so the number of clusters is decreasing because clusters are merged.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    A total of 479 sequences were hits for both the known and alternative state, with 368 exclusively for the known and 208 exclusively for the alternative state.

    “A total of 479 of these were hits for both the known and alternative state, with 368 exclusively for the known and 208 exclusively for the alternative state.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The closest structure homologues to the known Mpt53 state are clustered, while the closest homologues to the alternative state are dispersed across the phylogenetic tree.

    “The resulting tree is shown in Extended Data Fig. 9, and demonstrates that, while the closest structure homologues to the known state are clustered, the closest homologues to the alternative state are dispersed across the tree.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Findingown data

    Clusters from AF-Cluster scored higher to the ground state than randomly sampled clusters.

    “We found that clusters from AF-Cluster scored higher to the ground state (Extended Data Fig. 3b)”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Seven out of the depicted nine proteins had a helix in an analogous spot to the α-4 helix location in the Mpt53 alternative structure.

    “Seven out of the depicted nine proteins had a helix in an analogous spot to the α-4 helix location in the Mpt53 alternative structure.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The highest-pLDDT model from AF-Cluster correctly predicted the most stable folds for 10 out of 12 sets of point mutations in the GA/GB protein system, whereas default AF2 correctly predicted 8 out of 12.

    “We found that the highest-pLDDT model from AF-Cluster correctly predicted the most stable folds for 10 out of 12, whereas default AF2 correctly predicted 8 out of 12.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For both states, AUC scores correlated with the r.m.s.d. to the FS state from AF2 (ground state: Spearman R = −0.32, P = 2 × 10−9; FS state: Spearman R = −0.34, P = 4 × 10−10).

    “For both states, we found that the AUC scores correlated with the r.m.s.d. to the FS state from AF2 (ground state: Spearman R = −0.32, P = 2 × 10−9; FS state: Spearman R = −0.34, P = 4 × 10−10)”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Findingown data

    The union of matches after CD-HIT filtering resulted in 1,055 sequences remaining.

    “We took the union of these two sets and applied CD-HIT65 with default parameters to filter for highly similar sequences. This resulted in 1,055 sequences remaining.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [65] Fu, L., Niu, B., Zhu, Z., Wu, S. & Li, W. CD-HIT: accelerated for clustering the next-generation sequencing data. Bioinformatics 28, 3150–3152 (2012). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    A few structures had higher weighted r.m.s.d. for the alternative Mpt53 state than for the known Mpt53 state.

    “A few structures had higher weighted r.m.s.d. for the alternative Mpt53 state than for the known Mpt53 state (Extended Data Fig. 7e (orange points) and 7f (structures)).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    DALI screening resulted in 1,822 matches for the Mpt53 known state and 1,245 matches for the Mpt53 alternative state.

    “This resulted in 1,822 matches for the Mpt53 known state and 1,245 matches for the Mpt53 alternative state (Extended Data Fig. 7d).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    A KaiB variant in T. elongatus vestitus is phylogenetically close to the known fold-switching KaiB for which the ground state is thermodynamically favoured, yet was predicted and experimentally corroborated to be stabilized in the FS state.

    “We identified a KaiB variant in T. elongatus vestitus that is phylogenetically close to the known fold-switching KaiB for which the ground state is thermodynamically favoured, yet was predicted and experimentally corroborated to be stabilized in the FS state.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The premise of methods to infer structure based on evolutionary couplings is that amino acids co-evolve in patterns reflective of the underlying structure because they exist and evolve in the context of 3D structure.

    “The premise of methods to infer structure based on evolutionary couplings is that, because amino acids exist and evolve in the context of 3D structure, they are not free to evolve independently, but instead co-evolve in patterns reflective of the underlying structure.”
    Quote in the paperNot a substantial claim0 of 4 sources checked
    • cites [7] Altschuh, D., Lesk, A. M., Bloomer, A. C. & Klug, A. Correlation of co-ordinated amino acid substitutions with function in viruses related to tobacco mosaic virus. J. Mol. Biol. 193, 693–707 (1987). source

      Not checked in the free run.

    • cites [8] Lockless, S. W. & Ranganathan, R. Evolutionarily conserved pathways of energetic connectivity in protein families. Science 286, 295–299 (1999). source

      Not checked in the free run.

    • cites [9] Marks, D. S. et al. Protein 3D structure computed from evolutionary sequence variation. PLoS ONE 6, e28766 (2011). source

      Not checked in the free run.

    • cites [10] Morcos, F. et al. Direct-coupling analysis of residue coevolution captures native contacts across many protein families. Proc. Natl Acad. Sci. USA 108, E1293–E1301 (2011). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Understanding the mechanistic basis of any protein's functions requires understanding the complete set of conformational substates it can adopt.

    “Understanding the mechanistic basis of any protein’s functions requires understanding the complete set of conformational substates that it can adopt2.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [2] Henzler-Wildman, K. & Kern, D. Dynamic personalities of proteins. Nature 450, 964–972 (2007). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Contacts corresponding to multiple states can be observed at the level of entire MSAs for membrane proteins, ligand-induced conformational changes and multimerization-induced conformational changes.

    “Indeed, methods for inferring evolutionary couplings have already demonstrated that contacts corresponding to multiple states can be observed at the level of entire MSAs for membrane proteins15, ligand-induced conformational changes16 and multimerization-induced conformational changes17.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [15] Hopf, T. A. et al. Three-dimensional structures of membrane proteins from genomic sequencing. Cell 149, 1607–1621 (2012). source

      Not checked in the free run.

    • cites [16] Morcos, F., Jana, B., Hwa, T. & Onuchic, J. N. Coevolutionary signals across protein lineages help capture multiple protein conformations. Proc. Natl Acad. Sci. USA 110, 20533–20538 (2013). source

      Not checked in the free run.

    • cites [17] Uguzzoni, G. et al. Large-scale identification of coevolution signals across homo-oligomeric protein interfaces by direct coupling analysis. Proc. Natl Acad. Sci. USA 114, E2662–E2671 (2017). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    KaiB contains only 108 residues and undergoes a conformational change affecting the secondary structure of around 40 residues in its C-terminal part, switching between a canonical thioredoxin-like structure and a unique alternative conformation.

    “For example, although the metamorphic protein KaiB contains only 108 residues, it undergoes a conformational change that affects the secondary structure of around 40 residues in its C-terminal part, switching between a canonical thioredoxin-like structure and a unique alternative conformation5”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [5] Chang, Y. G. et al. Circadian rhythms. A protein fold switch joins the circadian oscillator to clock output in cyanobacteria. Science 349, 324–328 (2015). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    A computational analysis of the PDB suggested that between 0.5% and 4% of all proteins are fold switching.

    “A computational analysis of the Protein Data Bank (PDB) that identified changes in secondary structure between protein models sharing the same sequence suggested that between 0.5% and 4% of all proteins are fold switching29.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [29] Porter, L. L. & Looger, L. L. Extant fold-switching proteins are widespread. Proc. Natl Acad. Sci. USA 115, 5968–5973 (2018). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Mpt53 is an extracellular single-domain enzyme suggested to ensure correct folding of several cell-wall and extracellular protein substrates in M. tuberculosis by catalysing disulfide oxidation.

    “Mpt53 is an extracellular single-domain enzyme that is suggested to ensure correct folding of several cell-wall and extracellular protein substrates in M. tuberculosis by catalysing disulfide oxidation42.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [42] Goulding, C. W. et al. Gram-positive DsbE proteins function differently from Gram-negative DsbE homologs. A structure to function analysis of DsbE from Mycobacterium tuberculosis. J. Biol. Chem. 279, 3516–3524 (2004). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Fewer than ten metamorphic protein families have been thoroughly experimentally characterized, spanning a diverse range of functions.

    “Fewer than ten metamorphic protein families have been thoroughly experimentally characterized21, spanning a diverse range of functions.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [21] Dishman, A. F. & Volkman, B. F. Design and discovery of metamorphic proteins. Curr. Opin. Struct. Biol. 74, 102380 (2022). source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    In RfaH, the C-terminal domain interconverts between an α-helix bundle and a β-barrel through binding to functional partners.

    “In RfaH, the C-terminal domain (CTD) interconverts between an α-helix bundle and a β-barrel through binding to functional partners23.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [23] Zuber, P. K., Schweimer, K., Rosch, P., Artsimovitch, I. & Knauer, S. H. Reversible fold-switching controls the functional cycle of the antitermination factor RfaH. Nat. Commun. 10, 702 (2019). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationown data

    Related sequences for any given protein target might contain a signal for more than one biologically relevant structure, and deep-learning methods can be used to detect and analyse these multiple conformational states.

    “Our results demonstrate that, in the oncoming age of AF2-enabled structural biology, related sequences for any given protein target might contain a signal for more than one biologically relevant structure, and that deep-learning methods can be used to detect and analyse these multiple conformational states.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    KaiB is a circadian-rhythm protein found in cyanobacteria and proteobacteria that adopts two conformations with distinct secondary structures as part of its function.

    “KaiB is a circadian-rhythm protein found in cyanobacteria5,31 and proteobacteria32 that adopts two conformations with distinct secondary structures as part of its function”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [5] Chang, Y. G. et al. Circadian rhythms. A protein fold switch joins the circadian oscillator to clock output in cyanobacteria. Science 349, 324–328 (2015). source

      Not checked in the free run.

    • cites [31] Ishiura, M. et al. Expression of a gene cluster kaiABC as a circadian feedback process in cyanobacteria. Science 281, 1519–1523 (1998). source

      Not checked in the free run.

    • cites [32] Pitsawong, W. et al. From primordial clocks to circadian oscillators. Nature 616, 183–189 (2023). source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    The thermodynamically favoured state for KaiB from T. elongatus is the ground state; the FS structure was first solved in a complex with KaiC and could be solved for isolated KaiB only by introducing stabilizing mutations.

    “The thermodynamically favoured state for KaiB from T. elongatus (KaiBTE) is the ground state; the FS structure was first solved in a complex with KaiC33, and could be solved for the isolated KaiB only by introducing stabilizing mutations to this variant33.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [33] Tseng, R. et al. Structural basis of the day-night transition in a bacterial circadian clock. Science 355, 1174–1180 (2017). source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    Previous studies have identified evolutionary couplings corresponding to multiple states of domain-based conformational changes.

    “As previous studies have identified evolutionary couplings corresponding to multiple states of domain-based conformational changes15,16,20”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [15] Hopf, T. A. et al. Three-dimensional structures of membrane proteins from genomic sequencing. Cell 149, 1607–1621 (2012). source

      Not checked in the free run.

    • cites [16] Morcos, F., Jana, B., Hwa, T. & Onuchic, J. N. Coevolutionary signals across protein lineages help capture multiple protein conformations. Proc. Natl Acad. Sci. USA 110, 20533–20538 (2013). source

      Not checked in the free run.

    • cites [20] Malinverni, D. & Barducci, A. Coevolutionary analysis of protein subfamilies by sequence reweighting. Entropy 21, 1127 (2020). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductnone given

    The assignments from I-PINE were verified and some were adjusted manually in POKY.

    “The assignments from I-PINE were verified and some were adjusted manually in POKY.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    MSA Transformer is an unsupervised learning method; its contact predictions purely reflect evolutionary couplings learned in sequences, rather than being supervised on structure as is the case AF2.

    “MSA transformer is an unsupervised learning method, which signifies that its contact predictions purely reflect evolutionary couplings learned in sequences, rather than being supervised on structure as is the case AF2.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [36] Rao, R. M. et al. MSA Transformer. In Proc. International Conference on Machine Learning 8844–8856 (PMLR, 2021). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Mpt53 oxidizes the human kinase TAK1, which was shown to trigger an immune response.

    “Mpt53 oxidizes the human kinase TAK1, which was shown to trigger an immune response52.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [52] Wang, L. et al. Oxidization of TGFbeta-activated kinase by MPT53 is required for immunity to Mycobacterium tuberculosis. Nat. Microbiol. 4, 1378–1388 (2019). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    The final structures were validated using the wwPDB validation tool and the Protein Structure Validation Suite (PSVS).

    “The final structures were validated using the wwPDB validation tool79,80 (https://validate-rcsb-east.wwpdb.org/validservice/) and the Protein Structure Validation Suite (PSVS)81.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [79] Xu, W. et al. Announcing the launch of Protein Data Bank China as an associate member of the Worldwide Protein Data Bank Partnership. Acta Crystallogr. D 79, 792–795 (2023). source

      Not checked in the free run.

    • cites [80] ww, P. D. B. c. Protein Data Bank: the single global archive for 3D macromolecular structure data. Nucleic Acids Res. 47, D520–D528 (2019). source

      Not checked in the free run.

    • cites [81] Bhattacharya, A., Tejero, R. & Montelione, G. T. Evaluating protein structures determined by structural genomics consortia. Proteins 66, 778–795 (2007). source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    Given that known metamorphic proteins often switch folds through cellular stimuli, it may in general be difficult to experimentally validate novel folds identified through computational methods if the stimulus is unknown.

    “Given that known metamorphic proteins often switch folds through cellular stimuli, it may in general be difficult to experimentally validate novel folds identified through computational methods if the stimulus—whether pH, redox reaction or a binding partner—is unknown.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Conformational substates not present in the evolutionary signal may require alternative methods.

    “However, conformational substates not present in the evolutionary signal may require alternative methods.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    The triple mutation was confirmed by DNA sequencing using GeneWiz primers.

    “The triple mutation was confirmed by DNA sequencing using GeneWiz primers.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    If the previous estimate that 0.5-4% of all proteins contain fold-switching domains is accurate, this would correspond to approximately 1-8 million fold-switching proteins with possible alternative states that would not be predicted by the default AF2 method.

    “If the previous estimate29 that 0.5–4% of all proteins contain fold-switching domains is accurate, this would correspond to approximately 1–8 million fold-switching proteins with possible alternative states that would not be predicted by the default AF2 method.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [29] Porter, L. L. & Looger, L. L. Extant fold-switching proteins are widespread. Proc. Natl Acad. Sci. USA 115, 5968–5973 (2018). source

      Not checked in the free run.

  • May combine two claims.

    Interpretationreasoning

    The number of models returned for each state from AF-Cluster reflects the abundance of constructs reflecting different states across the protein family, which cannot be interpreted as that state's Boltzmann weight.

    “Second, the number of models returned for each state from AF-Cluster will reflect the abundance of constructs reflecting different states across the protein family, which cannot be interpreted as that state’s Boltzmann weight.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductown data

    All of the samples were purified to homogeneity with a single band at ~10 kDa on a Bis-Tris 4-12% gradient SDS-PAGE gel.

    “All of the samples were purified to homogeneity with a single band at ~10 kDa on a Bis-Tris 4–12% gradient SDS–PAGE gel (GenScript).”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Disease-causing point mutations are often due to population changes of protein substates.

    “Disease-causing point mutations are often due to population changes of protein substates3,4 and there is therefore great interest for methods to predict the effects of point mutations on structural ensembles and free energy.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [3] Wang, Z. & Moult, J. SNPs, protein structure, and disease. Hum. Mutat. 17, 263–270 (2001). source

      Not checked in the free run.

    • cites [4] Stein, A., Fowler, D. M., Hartmann-Petersen, R. & Lindorff-Larsen, K. Biophysical and mechanistic models for disease-causing protein variants. Trends Biochem. Sci. 44, 575–588 (2019). source

      Not checked in the free run.

50 of 50

  • The paper relies on this without support.

    Proteins must evolve in the context of the multiple conformational states that they adopt.

    “However, proteins must evolve in the context of the multiple conformational states that they adopt.”

  • The paper relies on this without support.

    There is a need for methods that deconvolve signal from multiple states if they are not already both present at the level of the entire MSA.

    “However, there is a need for methods that deconvolve signal from multiple states if they are not already both present at the level of the entire MSA.”
2 of 2

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Wayment-Steele et al found that clustering MSAs by sequence similarity lets AF2 sample known alternative states of KaiB, RfaH, and Mad2.

    “Wayment-Steele et al found that clustering MSAs by sequence similarity enables AF2 to sample known alternative states of KaiB, RfaH, and Mad247”

    From The power and pitfalls of AlphaFold2 for structure prediction beyond rigid globular proteins (2024) · cites this paper as [47]

  • Initial efforts to enhance conformational sampling have involved altering the number of sequences used to generate shallower MSA representations, masking coevolutionary information from MSAs, and splitting conflicting coevolutionary signals by clustering MSAs.

    “Initial efforts to enhance sampling of different conformations have involved altering the number of sequences used to generate shallower MSA representations, masking coevolutionary information provided by MSAs, and splitting conflicting coevolutionary signals by clustering MSAs46,47,105.”

    From The power and pitfalls of AlphaFold2 for structure prediction beyond rigid globular proteins (2024) · cites this paper as [47]

  • The curated dataset of 74 protein-antibody complexes released after 2024 with residues less than 800 ensures no overlap with Boltz-2's training data.

    “we curated a dataset of 74 complexes released after 2024 and with residues less than 800 for efficiency, ensuring no overlap with Boltz-2's training data [22].”

    From Reshaping biomolecular structure prediction through strategic conformational exploration with HelixFold-S1 (2026) · cites this paper as [22]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • The tested KaiBRS triple mutant has two different third mutations

    Major

    The results identify the fold-switching triple mutant as I68R/V83D/N84R, but the protein-expression methods say the triple mutant used in the study was I68R/V83D/N84A. These specify different proteins for the central experimental validation.

  • The fold-switch screening uses inconsistent minimum protein lengths

    Minor

    The main text and figure describe screening proteins starting at 48 amino acids, whereas the methods specify a minimum of 52 residues for the same set of 628 proteins.

2 of 2

Cites

Flagged first, then by citations

Research works printed but not linked: 3

  1. Ester, M., Kriegel, H.-P., Sander, J. & Xu, X. A density-based algorithm for discovering clusters in large spatial databases with noise. In Proc. 2nd International Conference on Knowledge Discovery and Data Mining (KDD-96) 226–231 (ACM, 1996).
  2. DeLano, W. L. Pymol: an open-source molecular graphics tool. CCP4 Newsl. Protein Crystallogr. 40, 82–92 (2002).
  3. Pedregosa, F. et al. Scikit-learn: machine learning in Python. J. Mach. Learn. Res. 12, 2825–2830 (2011).

Cited by

Sorted by citations

Related

Authors
Wayment-Steele, H. K., Ojoawo, A., Otten, R., Apitz, J. M., Pitsawong, W., Hömberger, M., Ovchinnikov, S., Colwell, L., & Kern, D.
Affiliations
  • Howard Hughes Medical Institute (US)
  • Harvard University (US)
  • Google (United States) (US)
Corresponding author
Kern, Dorothee
ORCID
Venue
Nature
Publisher
Springer Science and Business Media LLC
Published
2023-11-13
Volume
625
Issue
7996
Pages
832-839
ISSN
  • 0028-0836 (ISSN-L)
  • 1476-4687
DOI
10.1038/s41586-023-06832-9
PMC
PMC10808063
OpenAlex
W4388627063
Semantic Scholar
6d464c44cad5eb182ae580adfff8624562e457b8
Cited by
545
References
90
Citations by year
2026: 175 · 2025: 202 · 2024: 160 · 2023: 8
FWCI
48.74
Type
journal article
Language
en
Open access
Licence
Locations
  • Nature (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • Apollo (University of Cambridge) (submittedVersion, cc-by)
Funders
National Institute of General Medical Sciences (P41GM111135, T32GM135126, R24GM141526)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Protein Structure and Dynamics
  • Physical Sciences › Materials Science › Materials Chemistry › Enzyme Structure and Function
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Receptor Mechanisms and Signaling
Keywords
protein conformational switching, sequence clustering, AlphaFold2, KaiB, fold switching, point mutations, protein energy landscape, oxidative enzymes, protein family evolution
MeSH terms
Machine Learning, Antigens, Bacterial, Bacterial Proteins, Mutation, Protein Conformation, Proteins, Rhodobacter sphaeroides, Cluster Analysis, Sequence Alignment, Protein Folding

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-07.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-07.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review