Description

Although D. discoideum is an excellent model organism for many mammalian processes and disease states, large portions of its genome that could be useful for research remain uncharacterized. For instance, of the 11,088 protein coding genes in the D. discoideum genome that are not annotated as transposable elements or pseudogenes (Fey et al., 2019), the predicted amino acid sequence of 1,514 proteins are not similar enough to any known protein to receive a GO term, although many initiatives for experimental, large scale manual, and electronic annotations exist (Gene Ontology consortium, 2021).

An alternative way to predict the function of an amino acid sequence is to generate a predicted protein structure, and then compare the predicted structure to protein structures generated by X-ray crystallography or NMR. A commonly used program that does this sort of structure prediction and then structure comparison is I-TASSER (Roy et al., 2010; Yang et al., 2015; Zhang et al., 2017). I-TASSER generates secondary and tertiary structural predictions for a given protein, and then compares the predicted tertiary structure to known structures from the Protein Databank (PDB) (Burley et al., 2017). I-TASSER can then provide structure-based annotations for the protein, including gene ontology terms, possible ligand-binding, and enzymatic activity predictions. The 3D structures produced by I-TASSER can also be visualized and compared to Protein Databank proteins to, for instance, compare the active site of a protein in the PDB to a possible active site in the structure predicted by I-TASSER.

As an example of this approach, we previously found a Dictyostelium secreted protein called AprA that had very interesting properties, but unfortunately BLAST searches of the AprA amino acid sequence showed essentially no similarity to any mammalian proteins or possible gene products (comparing the AprA sequence to the amino acid sequences of all mammalian open reading frames) (Phillips and Gomer, 2012). I-TASSER predicted that AprA has similarity to the human secreted peptidase DPPIV, examination of the predicted AprA structure showed a very close match in both the amino acids and position of the amino acid side chains in one domain to a catalytic triad in the DPPIV structure, and DPPIV and AprA turned out to have very similar properties (Herlihy et al., 2013; Herlihy et al., 2017).

Over the course of a year, a group of 10 undergraduates was able to generate I-TASSER-based annotations for ~220 predicted D. discoideum proteins where there was no significant similarity to a known protein by GO or BLAST (although for many, there were similar hypothetical proteins in other organisms). Some of the I-TASSER identifications were to disparate proteins. For instance, the top structural analogs for the protein produced by D. discoideum gene DDB_G027895 included the BC component of the ABC complex from Yersinia entomophaga (TM score: 0.936), TedB2-TccC3 toxin from Photorhabdus luminescens (TM score: 0.872) and teneurin 2 partial extracellular domain (TM score 0.332). Another example of a protein with multiple possible similar proteins is encoded by DDB_G0275975; I-TASSER predicted that this protein has structural similarity to both antifreeze proteins and ABC transporters.

Other predicted proteins showed a single major similarity. For instance, the protein encoded by DDB_G0277585 has a predicted structural similarity to the ligand gated calcium channel inositol-1,4,5-triphosphate receptor (InsP₃Rs), the proteins encoded by DDB_G0276051, DDB_G0277495, DDB_G0270742, and DDB_G0274535 all have predicted structural similarity to the Clostridium difficile toxins A and B, DDB_G027561 encodes a protein with predicted structural similarity to an influenza RNA-dependent RNA polymerase, DDB_G0276755 encodes a protein with predicted structural similarity to a toll like receptor 3 ligand binding domain, DDB_G0276921 encodes a protein with predicted structural similarity to a pilin proteins from the gram-positive bacteria Streptococcus Pyogenes, and DDB_G0276485 encodes a protein with predicted structural similarity to integrin cell adhesion proteins. In addition, some annotations revealed possible gene duplications. For instance, the proteins encoded by DDB_G0278559, DDB_G0278561, DDB_G0278565, and DDB_G0278553 all have predicted structural similarity to the MU2 Adaptin subunit in multiple species, and an examination of the chromosome location of these genes showed that they are all are near each other on chromosome 3.

As exemplified by I-TASSER correctly predicting a similarity between D. discoideum AprA and human DPPIV, we and others have found that structural similarities predicted by I-TASSER can be useful for annotations, especially in the case where sequence-based annotations do not return any useful information. In our initial set of annotations, we have found a wide variety of interesting and in some cases quite unexpected proteins, including what might be toxins and an RNA-dependent RNA polymerase. Additional I-TASSER studies on other proteins of unknown function, in both D. discoideum and other systems, may well yield additional surprises.