Recently, I have been investigating whether collagen can interact with integrin α2 (the product of human gene ITGA2). There is a structure in the protein data bank (PDB ID: 4BJ3) in which an integrin domain is bound to what appears to be a peptide fragment of human collagen type XXVI (gene COL26A1). However, the sequence of this peptide differs from that in UniProt (Q96A83) for this collagen, which raises the questions:

  • Why are the sequences in the PDB file different from the sequences in UniProt?
  • Even though there are differences between the sequences, can I still assume that chain C/D/E in 4BJ3 represents the protein translated from COL26A1 and that COL26A1 can interact with integrin α2?

enter image description here

enter image description here

  • 3
    $\begingroup$ The PDB structure contains 4-hydroxyproline, a post-translational modification of proline commonly found in collagen. This is not a standard amino acid and doesn’t have a one letter code, which is why it is designated as “X” in your alignment. In the PDB file it’s called HYP. In the paper reporting the structure, it’s designated “O”. Note that it aligns with proline in the UniProt entry. $\endgroup$
    – canadianer
    May 8, 2023 at 15:32
  • $\begingroup$ I have edited the title to make it a little shorter and more focussed on the biology. I have revised my answer after having done some work on the problem. The timeline shows how the synthetic peptide evolved from work with larger collagen fragments. In fact the methodology of studying binding was affinity purification, but I cannot locate the paper that mentioned this — too many already. $\endgroup$
    – David
    May 22, 2023 at 16:34

2 Answers 2



The entry for 4BJ3 on the web pages of RCS Protein Data Bank contains information regarding a “collagen-like, GFOGER peptide” (authors’ words) that is at the best misleading and at the worst incorrect. This peptide is totally artificial, designed to have the integrin binding motif, GFOGER, flanked by model collagen GPP repeats so that it adopts a triple helical structure. The web page editors (I presume) have presented it as a mutated form of the α1 chain of collagen type XXVI, presumably because it scored highest in a sequence alignment. However the alignment is clearly on the basis of the GPP repeats, and this collagen has two mismatches in the key integrin GFOGER motif, specifically in the functionally important F and E residues. It is clear from the literature that the peptide was designed to study the interaction occurring in the integrin-binding collagens type I, II and IV — indeed collagen type XXVI was unknown at the time of design.
Collagen type XXVI completely lacks the GFOGER motif or any similar motif that has been shown to bind integrins. Although the possibility cannot be excluded without experimental evidence, the assumption would be that collagen type XXVI does not bind integrin α2.

The GFOGER peptide is not derived from collagen type XXVI

The PDB structure 4BJ3 is described in a file that can be downloaded from the RCS Protein Data Bank, and in the associated publication. According to these sources it consists of domain I of integrin α2 (residues 171–368) in a complex with a triple-helical “collagen-like, GFOGER peptide” of sequence [GPO]2GFOGER[GPO]3, where O represents hydroxyproline.

The web page for this structure at the RCS Protein Data Bank allows one to view the protein, and contains additional information, prepared by the site curators and web editors. This includes a “Protein Feature View” in which the GFOGER peptide is aligned to Uniprot entry Q96A83, the conceptual translation of the α1 chain of collagen type XXVI:

This is the section that the poster appears to have found, and from which he has produced an alignment that includes X, which is actually the hydroproline in the synthetic peptide. Here this is replaced by P.

Spurious sequence alignment at RCS PDB

I have annotated the alignment to emphasize the GFOGER region and the differences (broken circles). Of the three mismatches, two are in the crucial GFOGER region which is the basis of the interaction between collagen and integrin, and both (‘F’ and ‘E’) are at critical positions.

It is clear that the choice of alignment with collagen type XXVI is not based on anything originating from the authors and makes no biological sense. How did it arise?

I have conducted Blast searches with the sequence of the synthetic peptide and found no exact hit. The highest scoring hit is with type XXVI, even though an intelligent scoring system would have demanded perfect alignment with the GFOGER motif. I assume that a person with an informatics background, but unfamiliar with the biology, accepted this.

It is misleading and wrong!

What is the GFOGER peptide derived from?

To answer this question I tracked back through the literature, following the references in the Introduction or Methods sections, starting from the 2013 paper and ending in 1986. To summarize:

  • 1986: Collagens Type I, II and IV bind to a platelet membrane adhesion protein (Cell 46, 913-920) — Subsequently this was designated VLA-2 and later integrin 􏰁α2β􏰂1.
  • 1994: Artificial collagen-like peptide crystalized, confirming the triple-helix that had been proposed in the 1950s (Science 266, 75–81).
  • 1994: Domain I of integrin 􏰁α2β􏰂1 shown to contain binding site(s) for collagen.
  • 1997: Domain I of integrin 􏰁α2β􏰂1 crystalized (J.Biol.Chem. 272, 28512–28517). Molecular modelling suggested that a glutamic acid side chain (E) from collagen could coordinate to the integrin through a metal ion'
  • 2000: Crystal structure of a complex between domain I of integrin α2β􏰂1 and the “GFOGER peptide” (Cell, 101, 47–56). The rationale for the synthesis of this peptide is clearly stated: “…We therefore designed the shortest peptide that would fold into a stable triple helix and that had the glutamate at its center. We synthesized a 21-residue peptide with the sequence [Ac-(GPO)2GFOGER(GPO)3-NH2] and showed that it folds into a triple helix…”

There are further papers that could be cited, but this makes the provenance of the peptide clear, as also its relationship to collagens I, II and IV. The date of 2000 for the description of this peptide is also significant given the fact that collagen XXVI was not reported until 2002 (J.Biol.Chem. 277, 37678–37684).

Nevertheless could collagen XXVI still bind integrins?

I know of no direct experiments on this topic. What I have done is to perform Blast searches with those hexapeptides that have been shown to interact with integrin, with or without activation of the later. The results were:

Peptide Integrin
GFOGER Type I (α1
GFOGER Type II (α1)
GFOGER Type IV (α3,α4,α5)
GFOGER Type VII (α1)
GFOGER Type XI (α1, α2)
GLOGER Type I (α1,α2)
GLOGER Type II (α1)
GLOGER Type VII (α1)
GASGER Type I (α1)
GMOGER Type I (α1)
GMOGER Type II (α1)
GMOGER Type III (α1)
GMOGER Type IV (α6)
GMOGER Type V (α2)
GMOGER Type IX (α3)
GLOGEN Type III (α1)
GLOGEA Type II (α1)

It is evident that collagen type XXVI does not contain any of these motifs. Thus, in answer to the poster’s question:

Although one cannot exclude that it possesses a hitherto unidentified integrin-binding motif, there is currently no justification for assuming that collagen type XXVI will bind integrin.

Advice to anyone using the RCS Protein Data Bank

Read the documentation on the actual structure file, and the paper in which the structure is reported.

  • $\begingroup$ Thank you for sharing your informative answer and your advice. Using genes to refer to proteins is a practice that I learned from some literature 123.This seems likely to impose an extra burden on people unfamiliar with the genetic symbols. I have revised the question accordingly. $\endgroup$
    – Wan NIE
    May 10, 2023 at 3:08
  • $\begingroup$ @WanNIE — Using gene names to refer to proteins is convenient (a) where the name of the gene product is not known (b) when tabulating data with space limitations (c) where their is ambiguity in protein nomenclature (although this may also be true of gene names. Point (b) is probably the major factor in the paper you cite. Otherwise only collagen researchers would know what COL26A1 and COL11A2 are (I would have no idea), and it is bad practice to use these without the names of their products. The authors of the paper make no reference to any gene in their paper. $\endgroup$
    – David
    May 10, 2023 at 10:06
  • $\begingroup$ @WanNIE — Before I amend my answer could you clarify whether you wish to know whether the collagen human collagen type XXVI interacts with integrin, or whether the peptide GPOGPR does? $\endgroup$
    – David
    May 10, 2023 at 16:07
  • $\begingroup$ Sure, I want to know whether the collagen human collagen type XXVI (COL26A1) interacts with integrin and especially with the alpha2 subunit. $\endgroup$
    – Wan NIE
    May 11, 2023 at 3:21
  • $\begingroup$ By the way, this paper states that the Wnt/Beta-catenin pathway can regulate the expression of genes in extracellular matrix (ECM)-receptor interaction, including COL11A1 (Collagen alpha-1(XI)), COL26A1 (Collagen alpha-1(XXVI)), LAMC3 (Laminin, gamma 3), and ITGA2 (Integrin, alpha 2). Does this implicitly suggest that there is a high possibility of an interaction between COL26A1 and ITGA2? $\endgroup$
    – Wan NIE
    May 11, 2023 at 3:21

I checked ENSEMBL and the reason is not due to known alternative splicing but still allows for the possibility that they investigate a different sequence/allele due to mutations.

Importantly, in the (open source) paper, you can read that ..

GFOGER constitutes a high-affinity motif for all the collagen-binding integrins reported to date [12]–[14].

Therefore, I would highly doubt that you can assume that COL26A1, if it does not posses that sequence, would be bound equivalently. Also, this quite frankly indicates that you are wrong with your premise, that the PDB shows COL26A1.

  • 1
    $\begingroup$ I agree, and have given you credit for this. And human collagen type XXVI alpha 1 does not contain this sequence as a sequence comparison or a Blast search with the peptide GFPGER will demonstrate. $\endgroup$
    – David
    May 21, 2023 at 11:20

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .