Computing substrate selectivity in a peptide transporter

Abstract

The human proton-coupled peptide transporter 1 (PepT1) is responsible for the absorption of di-/tri-peptides from the diet and peptide-like drugs. In this issue of Cell Chemical Biology, Samsudin and coworkers use an integrated computational and experimental approach to provide new insights into understanding substrate selectivity of PepT_St, a prokaryotic homolog of the human PepT1.

In humans, there are presently 456 solute carrier (SLC) transporter genes from 52 families that modulate the import and efflux of a large variety of solutes across cell membranes, including amino acids, neurotransmitters, metabolites, ions, and drugs (Cesar-Razquin et al., 2015). PepT1 (SLC15A1) is a proton-coupled transporter of di- and tri-peptides, primarily found in the small intestine, kidney, and pancreas, where it absorbs a broad range of peptides from the diet and peptide-like drugs such as β-lactam antibiotics (Smith et al., 2013). Pept1 is also an emerging drug target. It is upregulated in the colon of patients with inflammatory bowel disease (IBD) where it transports bacterially-derived products into colonic cytosol, thereby triggering an inflammatory response (Charrier and Merlin, 2006; Smith et al., 2013). Potential drugs targeting PepT1 can be inhibitors that block the uptake of harmful peptides of bacterial origin or substrates that efficiently deliver anti-inflammatory prodrugs to the target site.

The structure of the human PepT1 is unknown, however, structures of a prokaryotic homolog, PepT_St from Streptococcus thermophilus have been experimentally determined in complex with di- and tri-peptides (Lyons et al., 2014). PepT_St has a highly similar substrate specificity to that of PepT1 and a conserved binding site, providing a useful tool for understanding PepT1 function. The PepT_St structures revealed that different substrates can have different orientations within the binding site, where a di-peptide binds “horizontally” and a tri-peptide binds “vertically” (Fig. 1).

Fig. 1.

Binding site of PepT_St bound to the di-peptides AlaPhe (left; PDB 4D2C) and LysAla (right). LysAla binding pose was derived from the PepT_St tri-peptide complex structure (PDB 4D2D). The amino-acid determining di-peptides selectivity is colored in red.

Samsudin et al first evaluated the correlation of various binding prediction methods with experimentally determined IC₅₀s of eight substrates (seven di-peptides and one tri-peptide). They assessed methods with increasing computing time and precision, including knowledge-based scoring function (AutoDock Vina), “end-point” free-energy calculations (e.g., the Linear Interaction Energy (LIE)), and a “theoretically exact” method (Thermodynamic Integration (TI)). While the end-point methods were able to differentiate between peptides with low and high IC₅₀ values, it was shown that TI refines initial prediction to differentiate among substrates with similar IC₅₀s for four peptides. The authors concluded that end-point methods provide optimal compromise between speed and precision to evaluate the interaction of PepT_St and di-peptides. A key assumption that was made was that all di- and tri-peptides have similar horizontal and vertical conformations observed in the two structures. Moreover, for the scoring function, the flexibility of the binding site was not considered, which has been previously shown to be critical for successful applications of related approaches (Colas et al., 2015; Schlessinger et al., 2011).

The authors then used the end-point methods to investigate the specificity of PepT_St in binding di-peptides. The estimated ΔG for 400 possible di-peptides indicated that PepT_St binds neutral di-peptides with the highest affinities. Conversely, acidic di-peptides were predicted to have lower affinities for the transporter, which is in agreement with higher IC₅₀ values for acidic peptides (Solcan et al., 2012). The authors then investigated the importance of the position of the amino acids in di-peptide ligands using TI. Interestingly, the computed ΔΔG between the ΔGs of binding of AlaAla and AlaAla in which the N- or C-terminal residues were mutated to Phe, Asp, Glu or Lys, was higher for the N-terminal mutations. This suggested that the N-terminal residue determined the di-peptide specificity. Interestingly, models of PepT_St with various di-peptides indicated that the N-terminal residue of the substrate forms hydrogen bonds with polar residues previously shown critical for binding and transport (e.g, Tyr30, Asn156, Asn328) (Solcan et al., 2012) (Fig. 1), whereas the C-terminal sidechain interacts with a hydrophobic pocket less critical for function.

Next, to confirm the predicted significance of the N-terminal position, a proton-driven competition uptake assay was performed for four charged di-peptides (i.e. AlaAsp, AspAla, AlaLys and LysAla). As expected, the experimental IC₅₀s of the acidic di-peptides correlated with the predicted affinities. Surprisingly, for basic di-peptides the opposite trend was observed, suggesting that some di-peptide’s binding mode is different. The authors therefore hypothesized that di-peptides with a lysine, which consists of a long sidechain, have binding mode similar to that of a tri-peptide (Fig. 1). Indeed, modeling PepT_St with LysAla and AlaLys based on the PepT_St structure in complex with an AlaAlaAla tri-peptide suggested that LysAla forms hydrogen bonds with important polar residues of the binding site (Tyr68, Glu300, or Asn156), supporting this hypothesis (Fig. 1). The ΔΔG predicted using TI based on this model was also in agreement with experimental data. Taken together, these data suggested that longer di-peptides may bind in a vertical orientation similar to that of tri-peptide (Fig. 1).

Finally, the authors investigated the utility of their approach in predicting inhibition for homology models. They modeled PepT_St based on related structures with varying sequence similarity to PepT_St. The closer PepT_St was to the template structure, the higher was the correlation between predicted and observed inhibition. Following a similar trend, the predicted inhibition for a human PepT1 model was as accurate as random. This suggested that the accuracy of the computational approach highly depends on the quality of the protein structure or model. Although the predicted affinity for the human PepT1 did not correlate with the experimental data, the results presented in this study considerably improve our understanding of the human PepT1 substrate specificity, based on its functional and structural similarity to PepT_St. Moreover, the results provide potential explanation for how drugs (and prodrugs) are recognized by PepT1. The PepT1 substrate valacyclovir is a structural analog of di-peptides: the carrier group of the drug may act as a pseudo-N-terminal amino acid – it is small, neutral, and predicted to form hydrogen bonds with key residues; and the active part of the drug is equivalent to the C-terminal of di-peptide ligands - it is bulky and may fit a large hydrophobic cavity. Still, one must appreciate that a drug affinity does not necessarily translate to being a PepT1 substrate that get transported.

Designing drugs targeting SLC transporters requires the description of their structure and dynamics, and mode of interaction with their ligands, including substrates and inhibitors. This study involves iterative application of multiple computational and experimental methods, to successfully explain substrate specificity in an important transport system. Due to rapid increase in the number of SLC structures and improvement in computer power and methods, future studies using related integrated approaches are expected to be applied to other human SLC transporters. Furthermore, this study highlights the need for high quality structures and/or models of the human SLC members. For example, models, or even x-ray structures, can be optimized for protein ligand complementarity (Schlessinger et al., 2011). Finally, SLC transporters such as PepT1 are dynamic molecules that adopt multiple conformations during transport. Visualizing different conformations through computation or experiments can further describe the specificity of this transporter, and ultimately aid in design of unique PepT1 drugs.