Exercise:AntibodyStructurePrediction ans

Antibody Structure Modelling Answers

Case study: SI135 antibody from chronic lymphocytic leukemia (CLL)

LYRA prediction: View the original modelling job

Part 1 — Antibody modelling with LYRA

Q1. Framework template sequence similarity

Q1: what is the sequence similarity of the best H and L templates for the framework (FW)?

Answer:

Region	Selected template	Sequence identity
Heavy-chain framework (VH)	[TO VERIFY]	[TO VERIFY]
Light-chain framework (VL)	[TO VERIFY]	[TO VERIFY]

The framework templates are chosen according to their sequence compatibility with the target antibody, taking into account the CDR canonical structures.

The framework is relatively conserved among antibodies, and a high sequence identity usually facilitates reliable homology modelling.

LYRA reports both sequence identity and BLOSUM62 similarity scores. These should not be confused: sequence identity measures identical residues, whereas BLOSUM62 also accounts for conservative substitutions.

Q2. Light-chain type

Q2: What is the light chain type? Is it the most common in humans?

Answer:

The SI135 antibody contains a lambda (λ) light chain.

The sequence is consistent with the IGLV3-21 light-chain family, which is associated with a well-characterized subset of CLL antibodies.

Lambda is not the most common light-chain type in humans. Approximately 60% of human antibodies contain kappa (κ) light chains and 40% contain lambda (λ) light chains, although proportions vary between individuals and tissues.

Q3. Sequence differences between target and template

Q3: where are the differences between target and template located? Which are the most drastic that could affect the modeling process (e.g. G->W)?

Answer:

The framework regions are generally more conserved than the CDRs, which are responsible for much of the sequence diversity in antibodies.

The differences observed in the SI135 alignment should be classified according to their physicochemical properties.

Examples of substitutions with potentially important structural effects include:

• G → W: introduces a large aromatic side chain in place of a very small and flexible residue.
• A → D: introduces a negatively charged side chain.
• D → K: reverses the formal side-chain charge.
• Substitutions involving proline: may restrict local backbone conformations.

Conservative substitutions, such as I → V or D → E, are generally less disruptive.

However, the structural consequences depend on the position of the substitution. A change in the buried framework can be more important than a non-conservative substitution in an exposed loop.

SI135-specific substitutions: [TO VERIFY from the LYRA alignment]

Q4. CDR loop lengths

Q4: Do the loops/CDR have the same length in target and template? What does it mean, in terms of expected accuracy?

Answer:

The CDR loop lengths must be compared with those of the selected structural templates.

The SI135 antibody contains the following CDR3 sequences:

Heavy-chain CDR3:
ARDVNTMDV
Length: 9 amino acids

Light-chain CDR3:
QVWDSGSDHPWV
Length: 12 amino acids

If the target and template loops have the same length and similar conformations, the model can generally be built more reliably.

Differences in loop length make modelling more difficult because insertions, deletions, or alternative loop conformations must be accommodated.

LYRA attempts to address this by selecting separate loop templates according to loop length and predicted canonical structure.

SI135 target/template length differences: [TO VERIFY]

Q5. Number and similarity of structural templates

Q5: How many templates were used? Which is the region for which the template is more dissimilar to the target sequence?

Answer:

LYRA uses structural templates for the heavy- and light-chain frameworks and can select additional templates for individual CDR loops.

There are eight modelled regions in total:

• Two framework regions (VH and VL).
• Three CDRs in the heavy chain.
• Three CDRs in the light chain.

However, the number of distinct PDB templates may be smaller than eight because several regions can be modelled using the same structure.

Number of distinct templates for SI135: [TO VERIFY]

Region with the least similar template: [TO VERIFY]

CDR-H3 is frequently a challenging region to model because of its structural diversity, although the relatively long CDR-L3 of SI135 is also of particular interest.

Part 2 — Exploring the antibody structure

Q6. Shape of the antigen-binding site

Q6: What is the general shape of the ABS: flat, a pocket, a mountain or a groove?

Answer:

The antigen-binding site is formed by the six CDR loops of the variable heavy and light chains.

Its surface can be classified qualitatively as flat, pocket-like, groove-like, or protruding (mountain-like).

SI135 classification: [TO VERIFY from the LYRA structure displayed as a surface]

The classification should be based on the overall geometry of the combined CDR surface rather than the shape of a single loop.

Teaching point: The shape of the antigen-binding site can provide clues about molecular recognition, but it does not directly identify the antigen.

Q7. Longest and most protruding CDR loop

Q7: Which loop is the longest and protruding the most? Is this the most common case?

Answer:

SI135 has a relatively short heavy-chain CDR3 and a long light-chain CDR3.

CDR-H3: ARDVNTMDV       (9 residues)
CDR-L3: QVWDSGSDHPWV   (12 residues)

Using the short, structurally defined CDR boundaries, CDR-L3 is the longest CDR in this antibody.

The loop that protrudes the most should be identified by inspecting the three-dimensional structure.

Most protruding loop: [TO VERIFY in PyMOL]

This is an unusual feature compared with the common situation in which CDR-H3 is the most variable and often the most prominent loop.

A long CDR does not necessarily protrude further than a short CDR, because its conformation can fold back towards the antibody framework.

Part 3 — Somatic mutations

Q8. Mutations potentially affecting antigen binding

Q8: Which Mutations are located inside/near the CDRs? Which are the ones that are most likely to have an impact on the binding of the antibody?

Answer:

The somatic mutations must be identified by comparison with the closest germline V and J sequences, using IMGT/DomainGapAlign.

The SI135 sequences are consistent with IGHV3-21 and IGLV3-21 family usage, which is frequently associated with stereotyped B-cell receptors in CLL.

Exact somatic mutations: [TO VERIFY using IMGT]

Mutations within the CDRs or in framework positions supporting the CDRs are particularly interesting.

Mutations are more likely to affect binding when they:

• Change the electrostatic properties of the binding site.
• Introduce or remove aromatic residues.
• Change hydrogen-bonding possibilities.
• Alter loop conformation or flexibility.
• Affect residues exposed towards the antigen.

Mutations affecting the charge or geometry of residues near the centre of the antigen-binding site are plausible candidates for having a functional effect.

However, the impact of a mutation cannot be established confidently from the structure alone.

Teaching point: Mutations in the framework may indirectly alter binding by affecting the orientation or stability of the CDR loops.

Differences introduced during V(D)J recombination, especially in CDR3, should not automatically be interpreted as somatic mutations.

Part 4 — Electrostatic properties

Q9. Charged residues in the CDRs

Q9: Identify the residues in the CDRs that mostly contribute to the charge distribution.

Answer:

The most relevant charged amino acids are:

• Positively charged: arginine (R), lysine (K).
• Negatively charged: aspartate (D), glutamate (E).

For SI135, the CDR3 sequences contain several charged residues.

Heavy-chain CDR3:

A R D V N T M D V
  + -         -

This loop contains:

• One arginine (R), positively charged.
• Two aspartates (D), negatively charged.

Light-chain CDR3:

Q V W D S G S D H P W V
      -       -

This loop contains:

• Two aspartates (D), negatively charged.
• One histidine (H), which may carry a positive charge depending on its environment and pH.

Both CDR3 loops therefore contain negatively charged residues.

Other charged residues in CDR1 and CDR2 also contribute to the overall charge distribution.

The surface distribution must be inspected in PyMOL because residues that are close in sequence are not necessarily close in three-dimensional space.

A local concentration of aspartate residues could produce a negatively charged region at the antigen-binding surface.

Teaching point: Electrostatic potential reflects the spatial distribution of charges, not simply the total number of charged amino acids.

Q10. Effect of introducing a positive charge (optional)

Q10: (optional) Mutate some residues in order to have a positive charge in the middle of the antigen binding site, then recalculate the electrostatic potentials. Can you see the changes?

Answer:

Introducing a positively charged residue, such as lysine (K) or arginine (R), should increase the local positive electrostatic potential.

For example, replacing an exposed aspartate (D) with lysine (K) in a CDR could reverse the local charge.

After recalculating the electrostatic potential, a change in surface coloring should become visible near the mutated residue.

The extent of the change depends on the location and conformation of the new side chain.

A more positive surface does not necessarily imply stronger antigen binding, because the properties of the antigen must also be considered.

Part 5 — Comparison of modelling approaches

Q11. Comparison between LYRA and ABodyBuilder2

Q11: Go to the AbodyBuilder 2 site and make a new structure prediction. Download the PDB file and compare with the model generated by LYRA. Can you see some differences?

Answer:

LYRA and ABodyBuilder2 use different approaches to antibody structure prediction.

LYRA	ABodyBuilder2
Traditional template-based homology modelling	Deep-learning-based structure prediction
Selects framework and CDR templates from known experimental structures	Uses a neural network trained on antibody structures
Uses canonical loop structures and loop grafting	Learns structural patterns from training data
Quality depends strongly on suitable templates	Provides prediction-error estimates for different antibody regions

The overall immunoglobulin fold is expected to be similar in both models.

Differences may occur in:

• CDR loop conformations, particularly CDR-H3 or unusual loops.
• Relative orientations of the heavy and light chains.
• Side-chain conformations.

Specific differences between the SI135 models: [TO VERIFY after structural superposition]

The models can be compared in PyMOL using:

align ABodyBuilder2, LYRA

Replace the object names with those used in the PyMOL session.

The RMSD obtained after structural alignment provides a quantitative measure of structural differences.

Teaching point: Agreement between two predictions can increase confidence in a structural hypothesis, but does not establish experimental accuracy.

Summary

This exercise illustrates several important principles:

• Antibody frameworks are relatively conserved, whereas CDR loops provide sequence and structural diversity.
• Template-based modelling is influenced by sequence similarity, loop length, and canonical structure.
• CDR-H3 is often particularly variable, but other CDRs may also present modelling challenges.
• Somatic mutations may affect antigen recognition by changing the structural or chemical properties of the binding surface.
• Electrostatic properties depend on the three-dimensional arrangement of charged residues.
• Deep-learning tools provide an alternative to traditional homology modelling, but predicted structures must still be evaluated critically.

References

• SI135 LYRA modelling results
• LYRA — DTU Health Tech
• IMGT immunogenetics
• ABodyBuilder2
• Antibody sequences and antigen recognition in CLL