SIFT-ing through sequence space in the PDB

Many of the protein structures in the PDB archive are cross-referenced to other data resources. These cross-references, for example to UniProt, Pfam, CATH and SCOP are displayed on PDB websites like PDBe. But how are they added and kept up to date?

This is done through the SIFTS resource, which is run by the PDBe and UniProt teams at EMBL-EBI. SIFTS maintains up-to-date residue-level mapping between UniProt and PDB entries, and furnishes users with annotations from IntEnz, GO, Pfam, InterPro, SCOP, CATH and PubMed.

The latest version of SIFTS ensures that structures are mapped not only to a UniProt accession, but to the most appropriate isoform within that accession. If a structure is representative of a specific isoform (generated by alternative splicing of the mRNA), you’ll now see that clearly on the PDBe website.

A good example is PDB entry 1loi, the N-terminal splice region of a cyclic AMP-specific phosphodiesterase. But it’s the N-terminal region of isoform 3 only. The sequence is completely different from the N-terminus of isoform 1 (termed the ‘canonical’ sequence).

PDB entry 1loi contains the N-terminal region of UniProt accession P54748. Specifically the sequence of isoform 3, which is not present in the canonical sequence.

SIFTS updates the mapping between structures in the PDB and Pfam, if the whole Pfam domain is present in the structure. This avoids ‘false’ Pfam mappings, like PDB entry 5d9s, which was previously listed as containing the Pfam ubiquitin domain even though it contains only an 11 residue peptide of that domain.

SIFTS also maps to UniRef90, via the SIFTS API. UniRef90 clusters all sequences in UniProt that have 90% or greater sequence identity. This mapping reveals that while the PDB contains fewer than 3000 unique human proteins, there are a further 3300 structures from other organisms that are highly similar, at the sequence level, to human proteins for which structures are not available in the PDB.