Prediction for Chameleon (IgG binding domain of protein G)

Contact: Burkhard Rost (rost@EMBL-Heidelberg.de)
Date: May, 1996

• Question:
  Can prediction methods be applied to engineered sequences?
• Method:
  Experiment: In the native sequence of IgG binding domain of protein G the deca-peptide AATAEKVFKQY (chameleon 1: residues 23-33) is embedded in a helix, whereas the deca-peptide EWTYDDATKTF (chameleon 2: residues 42-52) forms a beta-strand. When replacing both deca-peptides by the engineered deca-peptide AWTVEKAFKTF the structure is maintained (Minor & Kim, 1996, Nature, 380, 730-734). Thus, the deca-peptide AWTVEKAFKTF switches from helical to strand conformation. (And was therefore dubbed 'chameleon' .)
  Prediction: Prediction method was PHDsec (Rost & Sander, 1994). Secondary structure was predicted for the orginial and the engineered sequences. Predictions were based on single sequence information (PHD no) and on multiple alignment information (PHD 30, for a cut-off in accepting as homologous sequences all those in SWISS-PROT with more than 30% pairwise sequence identity to the guide sequence).
• Result and conclusion:
  Could the experimental result have been predicted? In general, prediction methods such as PHDsec (Rost & Sander, 1994; Abstract) may not be valid if applied to engineered proteins. Interestingly, for the particular case of chameleon, PHDsec was successful. When basing the prediction on a multiple alignment (rather than on single sequences) the peptide AWTVEKAFKTF > was correctly predicted in both conformations.
  Note: This success is, supposedly, rather the exception than the rule for similar situations.

• Figure caption:
  Given are the residue numbers, the sequence in a one-letter code (AA), the observed secondary structure (DSSP), the secondary structure predicted by PHDsec with (PHD 30) and without alignment (PHD no), and the reliability of the prediction (0 = low; 9 = high).