Artificial Intelligence: Exploring the conformational diversity of proteins

An artificial intelligence-based method can predict distinct conformational states of membrane transporters and receptors.
  1. Avner Schlessinger  Is a corresponding author
  2. Massimiliano Bonomi  Is a corresponding author
  1. Department of Pharmacological Sciences, Icahn School of Medicine at Mount Sinai, United States
  2. Department of Structural Biology and Chemistry, Institut Pasteur, Université Paris Cité, France

The human body contains a vast number of different proteins that carry out distinct roles. Proteins are made up of combinations of 20 amino acids, each with different physicochemical properties. The number and sequence of amino acids in a protein determine how it will fold into the specific three-dimensional structure or shape that the protein needs to perform its role.

Several proteins, including membrane proteins, do not simply fold into a single conformation. Instead, they need to be able to ‘flip’ between different conformations to do their job. Innovations in the experimental techniques used to determine protein structures – such as cryo-electron microscopy, nuclear magnetic resonance spectroscopy or X-ray crystallography –have provided valuable insights into the different conformations of many membrane proteins. However, these methods are costly and time consuming.

Using computational methods to predict the structures of proteins could allow scientists to fill the gap between protein sequence and structural knowledge, without having to rely on expensive experimental methods (Baker and Sali, 2001). Recently, an artificial intelligence-based method to predict protein structures, called AlphaFold2 (AF2), has taken structural biology by storm (Jumper et al., 2021).

AF2 emerged as a valuable tool for predicting the structures of proteins from their sequences with an accuracy comparable to that obtained by experimental techniques at a fraction of their time and costs, as shown for various biological problems (Evans et al., 2021; Mosalaganti et al., 2021; Tunyasuvunakool et al., 2021; McCoy et al., 2022). Now, in eLife, Diego del Alamo, Davide Sala, Hassane Mchaourab and Jens Meiler report how AF2 can also predict different conformations of membrane proteins (Del Alamo et al., 2022).

To do so, the researchers – based at Vanderbilt University and Leipzig University – used a set of eight membrane proteins representing different structural classes and mechanisms of action. This included five unique transporters (LAT1, ZnT8, MCT1, STP10, and ASCT2), whose structures had been previously experimentally determined in both inward- and outward-facing conformations (Figure 1), and three representative G-protein-coupled receptors (CGRPR, PTH1R, and FZD7), whose structures had been solved experimentally in active and inactive states. None of these proteins were part of the original AF2 training set, which included structures located in the Protein Data Bank (PDB).

Conformational changes of the alanine-serine-cysteine transporter 2 (ASCT2).

An artificial intelligence-based programme, called AF2, can predict the conformational diversity of membrane proteins, such as ASCT2, by modifying the depth of the input multiple sequence alignment. Shown are the cryo-electron microscopy structures of ASCT2 in conformations facing inside (blue) and outside of the cell (yellow). ASCT2 uses an elevator-type alternating access mechanism to transport molecules, which involves a change in the relative orientation of the scaffold (dark tones) and transport domains (light tones) of the protein.

Image credit: inward-facing structure, PDB 6RVX (Garaeva et al., 2019); outward-facing structure, PDB 7BCQ (Garibsingh et al., 2021) (CC BY 4.0).

The researchers then tested the ability of AF2 to model distinct conformational states by varying different parameters of the predictor, such as the number of models generated, and by using known structures of protein homologues as templates. One key feature of AF2 is its ability to generate a multiple sequence alignment (MSA) of evolutionarily related sequences, which is critical for accurate modeling. These MSAs, which can include thousands of sequences, are used by AF2 to identify residues that have co-evolved, thereby highlighting the contacts critical for defining the three-dimensional fold of the protein.

Remarkably, del Alamo et al. demonstrated that by reducing the size of the input MSA or alignment depth from 5,120 to as few as 16 sequences, the conformational diversity explored by AF2 increased, thereby capturing the structures that were experimentally determined in different conformations. This procedure also generated misfolded or outlier models, which were identified by the lack of structural similarity to other models and excluded from further analysis. This provides an important step to distinguish structural models that represent biologically relevant states. Moreover, in some cases using templates as input increased the conformational diversity of the generated models when MSAs with reduced number of aligned sequences (shallow MSAs) were used. Taken together, these results suggest that minor modifications to the input parameters allow AF2 to explore a larger area of the conformational space of proteins to capture distinct, biologically relevant states.

However, the analysis was performed on a relatively small benchmark set of proteins, due to the limited number of membrane protein structures not included in the AF2 training set and resolved in multiple states. Furthermore, del Alamo et al. did not identify a one-size-fits-all protocol that could accurately model the conformational diversity of all the membrane proteins in their benchmark set. A more generalized approach would be useful to study a larger variety of proteins that adopt different conformations, including enzymes and transcription factors. Finally, given the significant role of membrane proteins as drug targets, it will be crucial to assess whether the models generated with the proposed approach can be used for rational drug design, which typically requires accurate modeling of the protein’s amino acid sidechains.

In conclusion, the work by del Alamo et al. extends the scope of AF2 beyond structure prediction of a single state to the exploration of the conformational diversity of proteins. Even though determining the populations of alternative conformations and the interconversion pathways between them still appears to be out of reach, this work represents a crucial step towards describing the dynamic nature of proteins with modern artificial intelligence-based structure predictors.

References

Article and author information

Author details

  1. Avner Schlessinger

    Avner Schlessinger is in the Department of Pharmacological Sciences, Icahn School of Medicine at Mount Sinai, New York, United States

    For correspondence
    avner.schlessinger@mssm.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-4007-7814
  2. Massimiliano Bonomi

    Massimiliano Bonomi is in the Department of Structural Biology and Chemistry, Institut Pasteur, Université Paris Cité, Paris, France

    For correspondence
    mbonomi@pasteur.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7321-0004

Publication history

  1. Version of Record published: April 21, 2022 (version 1)

Copyright

© 2022, Schlessinger and Bonomi

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 4,377
    Page views
  • 402
    Downloads
  • 1
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Avner Schlessinger
  2. Massimiliano Bonomi
(2022)
Artificial Intelligence: Exploring the conformational diversity of proteins
eLife 11:e78549.
https://doi.org/10.7554/eLife.78549
  1. Further reading

Further reading

    1. Biochemistry and Chemical Biology
    2. Structural Biology and Molecular Biophysics
    Jessica M Roberts, James D Beck ... Eric J Hayden
    Research Article Updated

    Self-cleaving ribozymes are RNA molecules that catalyze the cleavage of their own phosphodiester backbones. These ribozymes are found in all domains of life and are also a tool for biotechnical and synthetic biology applications. Self-cleaving ribozymes are also an important model of sequence-to-function relationships for RNA because their small size simplifies synthesis of genetic variants and self-cleaving activity is an accessible readout of the functional consequence of the mutation. Here, we used a high-throughput experimental approach to determine the relative activity for every possible single and double mutant of five self-cleaving ribozymes. From this data, we comprehensively identified non-additive effects between pairs of mutations (epistasis) for all five ribozymes. We analyzed how changes in activity and trends in epistasis map to the ribozyme structures. The variety of structures studied provided opportunities to observe several examples of common structural elements, and the data was collected under identical experimental conditions to enable direct comparison. Heatmap-based visualization of the data revealed patterns indicating structural features of the ribozymes including paired regions, unpaired loops, non-canonical structures, and tertiary structural contacts. The data also revealed signatures of functionally critical nucleotides involved in catalysis. The results demonstrate that the data sets provide structural information similar to chemical or enzymatic probing experiments, but with additional quantitative functional information. The large-scale data sets can be used for models predicting structure and function and for efforts to engineer self-cleaving ribozymes.

    1. Structural Biology and Molecular Biophysics
    Eshwar R Tammineni, Lourdes Figueroa ... Eduardo Rios
    Research Article

    Calcium ion movements between cellular stores and the cytosol govern muscle contraction, the most energy-consuming function in mammals, which confers skeletal myofibers a pivotal role in glycemia regulation. Chronic myoplasmic calcium elevation (“calcium stress”), found in malignant hyperthermia-susceptible (MHS) patients and multiple myopathies, has been suggested to underlie the progression from hyperglycemia to insulin resistance. What drives such progression remains elusive. We find that muscle cells derived from MHS patients have increased content of an activated fragment of GSK3β — a specialized kinase that inhibits glycogen synthase, impairing glucose utilization and delineating a path to hyperglycemia. We also find decreased content of junctophilin1, an essential structural protein that colocalizes in the couplon with the voltage-sensing CaV1.1, the calcium channel RyR1 and calpain1, accompanied by an increase in a 44 kDa junctophilin1 fragment (JPh44) that moves into nuclei. We trace these changes to activated proteolysis by calpain1, secondary to increased myoplasmic calcium. We demonstrate that a JPh44-like construct induces transcriptional changes predictive of increased glucose utilization in myoblasts, including less transcription and translation of GSK3β and decreased transcription of proteins that reduce utilization of glucose. These effects reveal a stress-adaptive response, mediated by the novel regulator of transcription JPh44.