Protein W3
BIOS5003 Biochemistry of Cell Functions
Techniques for Studying Protein Structure
Determination of 3D Structure
Methods Used:
NMR spectroscopy on solution proteins
X-ray crystallography
Electron microscopy
Computer prediction from sequence
Additional Techniques:
Start with approximate structure by comparison with related known structures; refine by energy minimization
Ab initio prediction
Electron Microscopy
Development:
Five super-imposed images of E. coli glutamine synthase (referenced from Voet & Voet, Fig. 7.60)
Previously characterized by low resolution.
Cryo-Electron Microscopy (Cryo-EM):
Described as a massive game-changer in structural biology
Reference: Chung, Jae-Hee & Kim, Homin, High-Resolution Cryo-Electron Microscopy, Applied Microscopy. 47. 218-222.
Nobel Prize in Chemistry awarded in 2017 for developments in cryo-electron microscopy to:
Jacques Dubochet
Joachim Frank
Richard Henderson
Recent Improvements in Cryo-EM:
Contributing Factors:
New cameras
Better sample preparation
New data processing techniques
Resulting in a massive increase in resolution (i.e., detail discernment).
Drug design capability: Knowing structures allows for designing drugs to influence protein functions.
Insulin Receptor Structures
Type:
The insulin receptor is classified as a Tyrosine Kinase Receptor (TKR).
Structural Characteristics:
Uniquely covalently linked as a multimer of type (ab)2.
Upon insulin binding, the beta (b) domains align, activating tyrosine kinase activity.
Reference: Gutmann et al, 2018 J Cell Biol.
Functionality:
Single particle high-resolution EM demonstrates receptor conformational changes:
Converts from U form to T form in response to increasing insulin concentration.

Receptor proteins can be conjugated to either 1 (1U or 1T) or 2 (2U or 2T) nanodiscs.
Analysis of 10,000 particles for categorization.
Structural Comparisons:
U form derived from x-ray crystallography; T form obtained from cryo-EM.
Computational Protein Structure Prediction
Question: What if a protein sequence exists in the database but hasn’t had a resolved structure?
Concept Explanation:
Protein Folding Challenge:
For a protein with 150 amino acids in sequence:
Assuming 3 configurations per peptide bond (referencing one of the 3 main regions on the Ramachandran plot).
Folding time computed as:
Folding takes 1 picosecond (10^{-12} s) to convert between configurations.
It could take $10^{48}$ years to explore all possible conformations of $3^{150}$ (approximately = $10^{68}$ configurations).
Typical folding time for proteins is between 0.1 to 1000 seconds.
This calculation illustrates why comprehensive computation of native structures is impractical.
Historical Context:
Christian Anfinsen’s experiment (1950s):
Demonstrates that protein structure is encoded in its primary sequence.
When denaturants are removed, a denatured protein can refold into its active state.
This phenomenon confirms the sequence coding for structure despite potential refolding resulting randomly in the presence of denaturants.
Bonds (a-S-S-) can reform in 10^5 possible ways but weak interactions dictate structure.
Emerging Techniques:
Ab initio computer prediction for protein structures is rapidly improving.
Notable event: CASP14 (November-December 2020).
Resources:
Video on AlphaFold: https://youtu.be/nGVFbPKrRWQ
Lasker Foundation recognition of AlphaFold: https://laskerfoundation.org/winners/alphafold-a-technology-for-predicting-protein-structures/
Machine Learning in Protein Design
Subfields:
Artificial Intelligence: Machine Learning and Deep Learning subset.
Applications in managing metabolic diseases and their complications, monitoring dietary habits, and delivering dietary interventions through e-coaching.
CASP15 Overview (2022)
Statistical Summary of Protein Evaluation:
GDT_TS scores and RMSD evaluations tracked.
Over time from CASP1 to CASP15, trends suggest:
Significant changes in the percentage of targets under specified Ca RMSD cut-off (in Ångstroms).
Highlight progress in improvement metrics over several CASP events.
AlphaFold’s Performance:
Note: AlphaFold unaided is not the highest-performing model.
Domain and Multimer evaluation, including TMscore comparisons, showcased AlphaFold's capabilities relative to other approaches.
AlphaFold’s Synergy with Structural Biology
Impact on Structure Determination:
Number of protein entries increased substantially by AlphaFold predictions.
Prediction capabilities led to a transformative increase in known structures, detailing:
200 million total proteins,
350,000 for human-related proteins and model organisms.
Collaboration between DeepMind and EMBL-EBI expanded the AlphaFold database significantly over recent years.
Upcoming CASP16 (December 2024)
Event Summary:
Thirteenth Community Wide Experiment on Critical Assessment of Techniques for Protein Structure Prediction.
Modeling Categories:
Consistent categories including single proteins and domains, protein complexes, nucleic acid structures, and macromolecular binding interactions.
Progress and Challenges:
Emphasis on model reliability, improved rankings, and better methods for complex structures especially antibody targets.
Addressing Protein Design Challenges
Current Limitations:
While primary sequences are predictable from codon sequences, the direct relation to 3D structure functionality remains unclear.
Ability to predict function relies extensively on comparisons to resolved structures.
Example in Mechanisms:
The MCT (Monocarboxylate Transporters) family exemplifies nuanced understanding of transport roles:
MCTs 1, 2, 3, & 4 transport lactate.
MCT10 carries Trp, Tyr, and Phe; MCT8 facilitates T3 and T4 transport; MCT9 is responsible for carnitine transport.
Final Note:
Experimental science in protein structure determination remains essential in resolving outstanding questions and achieving practical applications for drug design and functional predictions.