Verification of FMAP for bitopic proteins

The FMAP method was developed as a physics-based approach to predict stability and lengths of membrane-bound alpha-helices in peptides and single-spanning TM proteins. FMAP was tested for a large set of peptides studied by solution NMR and other methods in micelles, bicelles and lipid bilayers. It was also tested for a set of 141 different bitopic proteins with known structures (83 PDB entries), as described below.

All TM single-helical segments were identified with average errors in prediction of N- and C-termini of 2.2 and 3.0 residue/per helix end, respectively (Figure 4 A, B). The average accuracy in prediction of hydrophobic thicknesses for single-pass TM-helices was 2.4±2.0 Å (Figure 4 C). For ~80% of bitopic proteins with known 3D structure the difference between experimental and predicted lengths of TM helices did not exceed 7 residues per helix (Figures 5 A-C). The largest differences were observed for bitopic proteins from multiprotein complexes, likely due to the contributions of specific tertiary inter-helical interactions not included in FMAP.

Figure 4. Accuracy of FMAP in predicting ends and hydrophobic thicknesses of TM helices. Distributions of differences in lengths (A: N-termini; B: C-termini) and hydrophobic thickness (C) between FMAP-predicted TM helices and experimental 3D structures of 139 bitopic proteins from 83 PDB entries. Hydrophobic thicknesses of experimental structures were taken from the OPM database.

Figure 5A. Comparison of FMAP-predicted TM helices with experimental structures of bitopic proteins in monomeric state. Experimental structures of bitopic proteins are shown as cartoons colored gray. FMAP-predicted intra-helical residues are highlighted by purple color. PDB IDs are indicated for each structure. Horizontal lines specify hydrophobic membrane boundaries calculated by PPM2.0, which are colored blue for the inner leaflet and colored red for the outer leaflet.



Figure 5B. Comparison of FMAP-predicted TM helices with experimental structures of bitopic proteins in in homo- and hetero-oligomeric states. Experimental structures of bitopic proteins are shown as cartoons colored gray. FMAP-predicted intra-helical residues are highlighted by purple color. PDB IDs are indicated for each structure. Most structures represent homodimers, except two structures for heterodimers (2k9j, 2ks1) and one structure for heterotetramer (2rlf). Horizontal lines specify hydrophobic membrane boundaries calculated by PPM2.0, which are colored blue for the inner leaflet and colored red for the outer leaflet.



Figure 5C. Comparison of FMAP-predicted TM helices with experimental structures of bitopic proteins in multiprotein complexes. Experimental structures of bitopic proteins are shown as cartoons colored gray. FMAP-predicted intra-helical residues are highlighted by purple color. PDB IDs and subunit names are indicated for each structure. Horizontal lines specify hydrophobic membrane boundaries calculated by PPM2.0, which are colored blue for the inner leaflet and colored red for the outer leaflet.



Figure 5C (continued). Comparison of FMAP-predicted TM helices with experimental structures of bitopic proteins in multiprotein complexes. Experimental structures of bitopic proteins are shown as cartoons colored gray. FMAP-predicted intra-helical residues are highlighted by purple color. PDB IDs and subunit names are indicated for each structure. Horizontal lines specify hydrophobic membrane boundaries calculated by PPM2.0, which are colored blue for the inner leaflet and colored red for the outer leaflet.

Comparison of TMH included in Membranome database (FMAP prediction, with subsequent removal of false-positives, see Protein set), with prediction of TMH by top servers. For a set of 4831 bitopic proteins in Membranome, almost 97% proteins were also identified as bitopic by Phobius. Importantly, >86% of the helices showed a nearly perfect sequence overlap: >80% of residues in the overlapped helices coincide. The corresponding fractions for TMHMM, and TopPred were 75% and 70%, respectively (Figure 6, Table 1).



Figure 6. Performance of FMAP vs top prediction servers in identification of TM α-helices.

Table 1.

Comparison of TM α-helices included in Membranome database with predictions of TMH by top bioinformatics tools. Numbers designate protein fractions (from 4831 bitopic proteins) that demonstrate 80-100%, 1-79% and 0% sequence overlap in TM α-helices predicted by FMAP and other servers.

% of overlap in TM α-helix

Phobius

TMHMM

TopPred

80-100

86.1

75.3

70.0

1-79

10.4

10.2

13.4

0

3.5

14.5

16.6



Verification of TMDOCK

TMDOCK web server predicts formation of parallel homodimers by transmembrane (TM) alpha-helices. Antiparallel dimers and heterodimers are not included in this version.

The underlying theoretical method was able to reproduce 26 experimental dimeric structures formed by TM α-helices of 21 single-pass membrane proteins (including 4 mutants) with Cα-atom r.m.s.d. 1.0 to 3.3 Å and native-like models ranked #1 in 70% of cases (Figure below).

Figure 7. Superposition of experimental structures (colored blue) and TMDOCK-generated models (colored yellow). Theoretical model was ranked #1 in all cases, except NMR structures stabilized by surface helices in APP (2loh), ErbB2 (2n2a) and few other proteins: Toll-like receptor, PGFRB, EphA2, FGFR3, and ErbB3. Helix crossing angles and r.m.s.d. between experimental and modeled structures are indicated. Positions of calculated hydrocarbon membrane boundaries (dotted lines) are indicated by red for extracellular membrane side and by blue for cytoplasmic side.
Note: This web server implements a faster version of the method with threading through only four structural templates. Therefore, it misses several low-stability NMR models that could only be detected using a larger set of templates (2k9y, 2l6w, 2l9u and 2mjo).

Figure 8. Correlation of experimental and calculated helix association energies