AlphaFold predictions of fold-switched conformations are driven by structure memorization.

Recent work suggests that AlphaFold (AF)-a deep learning-based model that can accurately infer protein structure from sequence-may discern important features of folded protein energy landscapes, defined by the diversity and frequency of different conformations in the folded state. Here, we test the limits of its predictive power on fold-switching proteins, which assume two structures with regions of distinct secondary and/or tertiary structure. We find that (1) AF is a weak predictor of fold switching and (2) some of its successes result from memorization of training-set structures rather than learned protein energetics. Combining >280,000 models from several implementations of AF2 and AF3, a 35% success rate was achieved for fold switchers likely in AF's training sets. AF2's confidence metrics selected against models consistent with experimentally determined fold-switching structures and failed to discriminate between low and high energy conformations. Further, AF captured only one out of seven experimentally confirmed fold switchers outside of its training sets despite extensive sampling of an additional ~280,000 models. Several observations indicate that AF2 has memorized structural information during training, and AF3 misassigns coevolutionary restraints. These limitations constrain the scope of successful predictions, highlighting the need for physically based methods that readily predict multiple protein conformations.

Metamorphic proteins and how to find them.

In the last two decades, our existing notion that most foldable proteins have a unique native state has been challenged by the discovery of metamorphic proteins, which reversibly interconvert between multiple, sometimes highly dissimilar, native states. As the number of known metamorphic proteins increases, several computational and experimental strategies have emerged for gaining insights about their refolding processes and identifying unknown metamorphic proteins amongst the known proteome. In this review, we describe the current advances in biophysically and functionally ascertaining the structural interconversions of metamorphic proteins and how coevolution can be harnessed to identify novel metamorphic proteins from sequence information. We also discuss the challenges and ongoing efforts in using artificial intelligence-based protein structure prediction methods to discover metamorphic proteins and predict their corresponding three-dimensional structures.

Sequence clustering confounds AlphaFold2.

Though typically associated with a single folded state, some globular proteins remodel their secondary and/or tertiary structures in response to cellular stimuli. AlphaFold21 (AF2) readily generates one dominant protein structure for these fold-switching (a.k.a. metamorphic) proteins2, but it often fails to predict their alternative experimentally observed structures3,4. Wayment-Steele, et al. steered AF2 to predict alternative structures of a few metamorphic proteins using a method they call AF-cluster5. However, their Paper lacks some essential controls needed to assess AF-cluster's reliability. We find that these controls show AF-cluster to be a poor predictor of metamorphic proteins. First, closer examination of the Paper's results reveals that random sequence sampling outperforms sequence clustering, challenging the claim that AF-cluster works by "deconvolving conflicting sets of couplings." Further, we observe that AF-cluster mistakes some single-folding KaiB homologs for fold switchers, a critical flaw bound to mislead users. Finally, proper error analysis reveals that AF-cluster predicts many correct structures with low confidence and some experimentally unobserved conformations with confidences similar to experimentally observed ones. For these reasons, we suggest using ColabFold6-based random sequence sampling7-augmented by other predictive approaches-as a more accurate and less computationally intense alternative to AF-cluster.

ColabFold predicts alternative protein structures from single sequences, coevolution unnecessary for AF-cluster.

Though typically associated with a single folded state, globular proteins are dynamic and often assume alternative or transient structures important for their functions1,2. Wayment-Steele, et al. steered ColabFold3 to predict alternative structures of several proteins using a method they call AF-cluster4. They propose that AF-cluster "enables ColabFold to sample alternate states of known metamorphic proteins with high confidence" by first clustering multiple sequence alignments (MSAs) in a way that "deconvolves" coevolutionary information specific to different conformations and then using these clusters as input for ColabFold. Contrary to this Coevolution Assumption, clustered MSAs are not needed to make these predictions. Rather, these alternative structures can be predicted from single sequences and/or sequence similarity, indicating that coevolutionary information is unnecessary for predictive success and may not be used at all. These results suggest that AF-cluster's predictive scope is likely limited to sequences with distinct-yet-homologous structures within ColabFold's training set.

SSDraw: Software for generating comparative protein secondary structure diagrams.

The program SSDraw generates publication-quality protein secondary structure diagrams from three-dimensional protein structures. To depict relationships between secondary structure and other protein features, diagrams can be colored by conservation score, B-factor, or custom scoring. Diagrams of homologous proteins can be registered according to an input multiple sequence alignment. Linear visualization allows the user to stack registered diagrams, facilitating comparison of secondary structure and other properties among homologous proteins. SSDraw can be used to compare secondary structures of homologous proteins with both conserved and divergent folds. It can also generate one secondary structure diagram from an input protein structure of interest. The source code can be downloaded (https://github.com/ncbi/SSDraw) and run locally for rapid structure generation, while a Google Colab notebook allows easy use.

SSDraw: software for generating comparative protein secondary structure diagrams.

The program SSDraw generates publication-quality protein secondary structure diagrams from three-dimensional protein structures. To depict relationships between secondary structure and other protein features, diagrams can be colored by conservation score, B-factor, or custom scoring. Diagrams of homologous proteins can be registered according to an input multiple sequence alignment. Linear visualization allows the user to stack registered diagrams, facilitating comparison of secondary structure and other properties among homologous proteins. SSDraw can be used to compare secondary structures of homologous proteins with both conserved and divergent folds. It can also generate one secondary structure diagram from an input protein structure of interest. The source code can be downloaded (https://github.com/ethanchen1301/SSDraw) and run locally for rapid structure generation, while a Google Colab notebook allows easy use.

Evolutionary selection of proteins with two folds.

Although most globular proteins fold into a single stable structure, an increasing number have been shown to remodel their secondary and tertiary structures in response to cellular stimuli. State-of-the-art algorithms predict that these fold-switching proteins adopt only one stable structure, missing their functionally critical alternative folds. Why these algorithms predict a single fold is unclear, but all of them infer protein structure from coevolved amino acid pairs. Here, we hypothesize that coevolutionary signatures are being missed. Suspecting that single-fold variants could be masking these signatures, we developed an approach, called Alternative Contact Enhancement (ACE), to search both highly diverse protein superfamilies-composed of single-fold and fold-switching variants-and protein subfamilies with more fold-switching variants. ACE successfully revealed coevolution of amino acid pairs uniquely corresponding to both conformations of 56/56 fold-switching proteins from distinct families. Then, we used ACE-derived contacts to (1) predict two experimentally consistent conformations of a candidate protein with unsolved structure and (2) develop a blind prediction pipeline for fold-switching proteins. The discovery of widespread dual-fold coevolution indicates that fold-switching sequences have been preserved by natural selection, implying that their functionalities provide evolutionary advantage and paving the way for predictions of diverse protein structures from single sequences.

Fluid protein fold space and its implications.

Fold-switching proteins, which remodel their secondary and tertiary structures in response to cellular stimuli, suggest a new view of protein fold space. For decades, experimental evidence has indicated that protein fold space is discrete: dissimilar folds are encoded by dissimilar amino acid sequences. Challenging this assumption, fold-switching proteins interconnect discrete groups of dissimilar protein folds, making protein fold space fluid. Three recent observations support the concept of fluid fold space: (1) some amino acid sequences interconvert between folds with distinct secondary structures, (2) some naturally occurring sequences have switched folds by stepwise mutation, and (3) fold switching is evolutionarily selected and likely confers advantage. These observations indicate that minor amino acid sequence modifications can transform protein structure and function. Consequently, proteomic structural and functional diversity may be expanded by alternative splicing, small nucleotide polymorphisms, post-translational modifications, and modified translation rates.

Identification of a covert evolutionary pathway between two protein folds.

Although homologous protein sequences are expected to adopt similar structures, some amino acid substitutions can interconvert α-helices and ß-sheets. Such fold switching may have occurred over evolutionary history, but supporting evidence has been limited by the: (1) abundance and diversity of sequenced genes, (2) quantity of experimentally determined protein structures, and (3) assumptions underlying the statistical methods used to infer homology. Here, we overcome these barriers by applying multiple statistical methods to a family of ~600,000 bacterial response regulator proteins. We find that their homologous DNA-binding subunits assume divergent structures: helix-turn-helix versus α-helix + ß-sheet (winged helix). Phylogenetic analyses, ancestral sequence reconstruction, and AlphaFold2 models indicate that amino acid substitutions facilitated a switch from helix-turn-helix into winged helix. This structural transformation likely expanded DNA-binding specificity. Our approach uncovers an evolutionary pathway between two protein folds and provides a methodology to identify secondary structure switching in other protein families.

Distinguishing features of fold-switching proteins.

Though many folded proteins assume one stable structure that performs one function, a small-but-increasing number remodel their secondary and tertiary structures and change their functions in response to cellular stimuli. These fold-switching proteins regulate biological processes and are associated with autoimmune dysfunction, severe acute respiratory syndrome coronavirus-2 infection, and more. Despite their biological importance, it is difficult to computationally predict fold switching. With the aim of advancing computational prediction and experimental characterization of fold switchers, this review discusses several features that distinguish fold-switching proteins from their single-fold and intrinsically disordered counterparts. First, the isolated structures of fold switchers are less stable and more heterogeneous than single folders but more stable and less heterogeneous than intrinsically disordered proteins (IDPs). Second, the sequences of single fold, fold switching, and intrinsically disordered proteins can evolve at distinct rates. Third, proteins from these three classes are best predicted using different computational techniques. Finally, late-breaking results suggest that single folders, fold switchers, and IDPs have distinct patterns of residue-residue coevolution. The review closes by discussing high-throughput and medium-throughput experimental approaches that might be used to identify new fold-switching proteins.

Evolutionary selection of proteins with two folds.

Although most globular proteins fold into a single stable structure 1 , an increasing number have been shown to remodel their secondary and tertiary structures in response to cellular stimuli 2 . State-of-the-art algorithms 3-5 predict that these fold-switching proteins assume only one stable structure 6,7 , missing their functionally critical alternative folds. Why these algorithms predict a single fold is unclear, but all of them infer protein structure from coevolved amino acid pairs. Here, we hypothesize that coevolutionary signatures are being missed. Suspecting that over-represented single-fold sequences may be masking these signatures, we developed an approach to search both highly diverse protein superfamilies-composed of single-fold and fold-switching variants-and protein subfamilies with more fold-switching variants. This approach successfully revealed coevolution of amino acid pairs uniquely corresponding to both conformations of 56/58 fold-switching proteins from distinct families. Then, using a set of coevolved amino acid pairs predicted by our approach, we successfully biased AlphaFold2 5 to predict two experimentally consistent conformations of a candidate protein with unsolved structure. The discovery of widespread dual-fold coevolution indicates that fold-switching sequences have been preserved by natural selection, implying that their functionalities provide evolutionary advantage and paving the way for predictions of diverse protein structures from single sequences.

AlphaFold2 has more to learn about protein energy landscapes.

Recent work suggests that AlphaFold2 (AF2)-a deep learning-based model that can accurately infer protein structure from sequence-may discern important features of folded protein energy landscapes, defined by the diversity and frequency of different conformations in the folded state. Here, we test the limits of its predictive power on fold-switching proteins, which assume two structures with regions of distinct secondary and/or tertiary structure. Using several implementations of AF2, including two published enhanced sampling approaches, we generated >280,000 models of 93 fold-switching proteins whose experimentally determined conformations were likely in AF2's training set. Combining all models, AF2 predicted fold switching with a modest success rate of ~25%, indicating that it does not readily sample both experimentally characterized conformations of most fold switchers. Further, AF2's confidence metrics selected against models consistent with experimentally determined fold-switching conformations in favor of inconsistent models. Accordingly, these confidence metrics-though suggested to evaluate protein energetics reliably-did not discriminate between low and high energy states of fold-switching proteins. We then evaluated AF2's performance on seven fold-switching proteins outside of its training set, generating >159,000 models in total. Fold switching was accurately predicted in one of seven targets with moderate confidence. Further, AF2 demonstrated no ability to predict alternative conformations of two newly discovered targets without homologs in the set of 93 fold switchers. These results indicate that AF2 has more to learn about the underlying energetics of protein ensembles and highlight the need for further developments of methods that readily predict multiple protein conformations.

Many dissimilar NusG protein domains switch between α-helix and ß-sheet folds.

Folded proteins are assumed to be built upon fixed scaffolds of secondary structure, α-helices and ß-sheets. Experimentally determined structures of >58,000 non-redundant proteins support this assumption, though it has recently been challenged by ~100 fold-switching proteins. Though ostensibly rare, these proteins raise the question of how many uncharacterized proteins have shapeshifting-rather than fixed-secondary structures. Here, we use a comparative sequence-based approach to predict fold switching in the universally conserved NusG transcription factor family, one member of which has a 50-residue regulatory subunit experimentally shown to switch between α-helical and ß-sheet folds. Our approach predicts that 24% of sequences in this family undergo similar α-helix ⇌ ß-sheet transitions. While these predictions cannot be reproduced by other state-of-the-art computational methods, they are confirmed by circular dichroism and nuclear magnetic resonance spectroscopy for 10 out of 10 sequence-diverse variants. This work suggests that fold switching may be a pervasive mechanism of transcriptional regulation in all kingdoms of life.

Measuring Ostracism-Induced Changes in Consumption of Palatable Food: Feasibility of a Novel Behavioral Task.

Purpose: Ostracism is a highly aversive interpersonal experience. Previous research suggests that it can increase consumption of highly palatable food in some individuals, but decrease it in others. Thus, we developed the Cyberball-Milkshake Task (CMT), to facilitate research investigating individual differences in ostracism's effects on consumption of highly palatable food. We present data on feasibility for the CMT in a sample of young adult women. Materials and Methods: Participants were 22 women, 18-30 years old, reporting very low or very high levels of emotional eating at screening. Participants performed the CMT, which consisted of 12 trials. Each trial included: playing a round of Cyberball (a computerized game of catch with fictitious "other participants" programmed to either include or exclude the participant); viewing a chocolate image; and then consuming a participant-determined amount of milkshake. Participants subsequently played an additional inclusion and exclusion round of Cyberball, each immediately followed by questionnaires assessing current mood and recent Cyberball experience. Results: Cyberball exclusion (vs. inclusion) was associated with large, significant increases in reported ostracism and threats to self-esteem; exclusion's effects on affect were in the expected direction (e.g., increased negative affect), but generally small and non-significant. Milkshake intake was measurable for 95% of participants, on 96% of trials. Intake decreased quadratically across trials, with a steep negative slope for low trial numbers that decreased to the point of being flat for the highest trial numbers. Discussion: The CMT is a generally feasible approach to investigating ostracism's effects on consumption of highly palatable food. The feasibility (and validity) of the CMT may benefit from modification (e.g., fewer trials and longer rounds of Cyberball). Future research should examine whether performance on a modified version of the CMT predicts real-world behavior in a larger sample.

AlphaFold2 fails to predict protein fold switching.

AlphaFold2 has revolutionized protein structure prediction by leveraging sequence information to rapidly model protein folds with atomic-level accuracy. Nevertheless, previous work has shown that these predictions tend to be inaccurate for structurally heterogeneous proteins. To systematically assess factors that contribute to this inaccuracy, we tested AlphaFold2's performance on 98-fold-switching proteins, which assume at least two distinct-yet-stable secondary and tertiary structures. Topological similarities were quantified between five predicted and two experimentally determined structures of each fold-switching protein. Overall, 94% of AlphaFold2 predictions captured one experimentally determined conformation but not the other. Despite these biased results, AlphaFold2's estimated confidences were moderate-to-high for 74% of fold-switching residues, a result that contrasts with overall low confidences for intrinsically disordered proteins, which are also structurally heterogeneous. To investigate factors contributing to this disparity, we quantified sequence variation within the multiple sequence alignments used to generate AlphaFold2's predictions of fold-switching and intrinsically disordered proteins. Unlike intrinsically disordered regions, whose sequence alignments show low conservation, fold-switching regions had conservation rates statistically similar to canonical single-fold proteins. Furthermore, intrinsically disordered regions had systematically lower prediction confidences than either fold-switching or single-fold proteins, regardless of sequence conservation. AlphaFold2's high prediction confidences for fold switchers indicate that it uses sophisticated pattern recognition to search for one most probable conformer rather than protein biophysics to model a protein's structural ensemble. Thus, it is not surprising that its predictions often fail for proteins whose properties are not fully apparent from solved protein structures. Our results emphasize the need to look at protein structure as an ensemble and suggest that systematic examination of fold-switching sequences may reveal propensities for multiple stable secondary and tertiary structures.

A sequence-based method for predicting extant fold switchers that undergo α-helix ↔ ß-strand transitions.

Extant fold-switching proteins remodel their secondary structures and change their functions in response to cellular stimuli, regulating biological processes and affecting human health. Despite their biological importance, these proteins remain understudied. Predictive methods are needed to expedite the process of discovering and characterizing more of these shapeshifting proteins. Most previous approaches require a solved structure or all-atom simulations, greatly constraining their use. Here, we propose a high-throughput sequence-based method for predicting extant fold switchers that transition from α-helix in one conformation to ß-strand in the other. This method leverages two previous observations: (a) α-helix â†” ß-strand prediction discrepancies from JPred4 are a robust predictor of fold switching, and (b) the fold-switching regions (FSRs) of some extant fold switchers have different secondary structure propensities when expressed by themselves (isolated FSRs) than when expressed within the context of their parent protein (contextualized FSRs). Combining these two observations, we ran JPred4 on 99-fold-switching proteins and found strong correspondence between predicted and experimentally observed α-helix â†” ß-strand discrepancies. To test the overall robustness of this finding, we randomly selected regions of proteins not expected to switch folds (single-fold proteins) and found significantly fewer predicted α-helix â†” ß-strand discrepancies. Combining these discrepancies with the overall percentage of predicted secondary structure, we developed a classifier to identify extant fold switchers (Matthews correlation coefficient of .71). Although this classifier had a high false-negative rate (7/17), its false-positive rate was very low (2/136), suggesting that it can be used to predict a subset of extant fold switchers from a multitude of available genomic sequences.

Predictable fold switching by the SARS-CoV-2 protein ORF9b.

Extant fold-switching proteins remodel their secondary structures and change their functions in response to environmental stimuli. These shapeshifting proteins regulate biological processes and are associated with a number of diseases, including tuberculosis, cancer, Alzheimer's, and autoimmune disorders. Thus, predictive methods are needed to identify more fold-switching proteins, especially since all naturally occurring instances have been discovered by chance. In response to this need, two high-throughput predictive methods have recently been developed. Here we test them on ORF9b, a newly discovered fold switcher and potential therapeutic target from the Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). Promisingly, both methods correctly indicate that ORF9b switches folds. We then tested the same two methods on ORF9b1, the ORF9b homolog from SARS-CoV-1. Again, both methods predict that ORF9b1 switches folds, a finding consistent with experimental binding studies. Together, these results (a) demonstrate that protein fold switching can be predicted using high-throughput computational approaches and (b) suggest that fold switching might be a general characteristic of ORF9b homologs.

A high-throughput predictive method for sequence-similar fold switchers.

Although most experimentally characterized proteins with similar sequences assume the same folds and perform similar functions, an increasing number of exceptions is emerging. One class of exceptions comprises sequence-similar fold switchers, whose secondary structures shift from α-helix <-> ß-sheet through a small number of mutations, a sequence insertion, or a deletion. Predictive methods for identifying sequence-similar fold switchers are desirable because some are associated with disease and/or can perform different functions in cells. Here, we use homology-based secondary structure predictions to identify sequence-similar fold switchers from their amino acid sequences alone. To do this, we predicted the secondary structures of sequence-similar fold switchers using three different homology-based secondary structure predictors: PSIPRED, JPred4, and SPIDER3. We found that α-helix <-> ß-strand prediction discrepancies from JPred4 discriminated between the different conformations of sequence-similar fold switchers with high statistical significance (P < 1.8*10-19 ). Thus, we used these discrepancies as a classifier and found that they can often robustly discriminate between sequence-similar fold switchers and sequence-similar proteins that maintain the same folds (Matthews Correlation Coefficient of 0.82). We found that JPred4 is a more robust predictor of sequence-similar fold switchers because of (a) the curated sequence database it uses to produce multiple sequence alignments and (b) its use of sequence profiles based on Hidden Markov Models. Our results indicate that inconsistencies between JPred4 secondary structure predictions can be used to identify some sequence-similar fold switchers from their sequences alone. Thus, the negative information from inconsistent secondary structure predictions can potentially be leveraged to identify sequence-similar fold switchers from the broad base of genomic sequences.

Functional and Regulatory Roles of Fold-Switching Proteins.

Fold-switching proteins respond to cellular stimuli by remodeling their secondary structures and changing their functions. Whereas several previous reviews have focused on various structural, physical-chemical, and evolutionary aspects of this newly emerging class of proteins, this minireview focuses on how fold switching modulates protein function and regulates biological processes. It first compares and contrasts fold switchers with other known types of proteins. Second, it presents examples of how various proteins can change their functions through fold switching. Third, it demonstrates that fold switchers can regulate biological processes by discussing two proteins, RfaH and KaiB, whose dramatic secondary structure remodeling events directly affect gene expression and a circadian clock, respectively. Finally, this minireview discusses how the field of protein fold switching might advance.