Identification of a novel class of farnesylation targets by structure-based modeling of binding specificity

Abstract

Farnesylation is an important post-translational modification catalyzed by farnesyltransferase (FTase). Until recently it was believed that a C-terminal CaaX motif is required for farnesylation, but recent experiments have revealed larger substrate diversity. In this study, we propose a general structural modeling scheme to account for peptide binding specificity and recapitulate the experimentally derived selectivity profile of FTase in vitro. In addition to highly accurate recovery of known FTase targets, we also identify a range of novel potential targets in the human genome, including a new substrate class with an acidic C-terminal residue (CxxD/E). In vitro experiments verified farnesylation of 26/29 tested peptides, including both novel human targets, as well as peptides predicted to tightly bind FTase. This study extends the putative range of biological farnesylation substrates. Moreover, it suggests that the ability of a peptide to bind FTase is a main determinant for the farnesylation reaction. Finally, simple adaptation of our approach can contribute to more accurate and complete elucidation of peptide-mediated interactions and modifications in the cell.

Figure 1. Structural overview of the FTase binding pocket.

A top view of the binding pocket of human FTase (orange) in complex with C’ CNIQ peptide in Rap2a (green), and a farnesyl analog (red) (PDB: 1tn6 [9]). Arrows indicate the constraints used during the simulations: the two structurally conserved hydrogen bonds (C’ carboxylate to FTase Q167α and the a₂ backbone carbonyl oxygen to FTase R202β), as well as the sulfur-Zn²⁺ coordination. The figure was created using PyMOL (http://www.pymol.org).

Figure 2. FlexPepBind allows good discrimination between substrate - and non-substrate sequences.

A. ROC-plot of the discrimination between MTO peptide sequences and non-active peptide sequences on the training set with the FlexPepDock based protocol (green), the fast, minimization based protocol (red), an independent test set (blue), and expected random discrimination (black). The Area Under the ROC Curve (AUC) value for the training set is 0.915/0.875 for the FlexPepDock and minimization based protocols, accordingly. Note that the performance of the minimization-based protocol on the test set is even better than on the training set (0.91 vs. 0.875). For the indicated points on the plot, an energy threshold of -0.4 corresponds to a 69% True Positive Rate (TPR) and 8% False Positive Rate (FPR). A more stringent threshold of -1.1 energy units corresponds to a 44% TPR and 2% FPR. Training and test sets are detailed in Dataset S1A&B. B+C. Validation on additional independent test sets shows robust and reliable performance of our modeling protocol. B. The distribution of energies for known FTase substrate sequences. The horizontal line indicates the -0.4 threshold obtained from the training set (see Text). Using this criterion, 85% of the known binders are recovered. Note that this corresponds to a significantly better TPR than the one obtained on the training set. C. Energy distribution for a synthetic library of Ca₁a₂L peptides investigated in Krzysiak et al. . As in B., the horizontal line indicates a threshold of -0.4, which in this case displays 87.5% TP and 12.5% FP rates (i.e., only 3 false negatives and 2 false positives). The peptide sequences and scores can be found in Dataset S1C&D.

Figure 3. Energy distribution of all possible Cxxx sequences, as well as previously characterized peptides (STO, MTO and NON) .

The distributions of known single turnover (STO) and multiple turnover (MTO) peptide sequences overlap, and are both significantly shifted towards low peptide energies, compared to peptide sequences that do not undergo farnesylation (NON). The thresholds obtained for the discrimination of MTO/NON predict 1349 (17%; -1.1 threshold) and 2309 (29%; -0.4 threshold) of the possible tetramer peptide sequences to undergo farnesylation.

Figure 4. A novel class of farnesylation targets.

The sequence logos of different sets of Farnesylation targets are shown for A. 72 known substrates (Dataset S1C); B. 77 MTO peptides from Dataset S1A; C. 1349 (out of 8000) sequences that pass the stringent threshold of -1.1 and are predicted to undergo farnesylation – while position a₂ of the motif is still prominently aliphatic (ILE/VAL/LEU/PHE), positions a₁ and X are much more versatile than expected; D. A subset of C with D/E at C-terminal position (238/1349) constitutes a novel substrate class for FTase (Logos created by http://weblogo.berkeley.edu/).