Design of Cyclic Peptides Targeting Protein–Protein Interactions Using AlphaFold

3.0 科研～小助 2025-09-01 4 4 6.24MB 17 页 1知币

侵权投诉

Citation: Kosugi, T.; Ohue, M.

Design of Cyclic Peptides Targeting

Protein–Protein Interactions Using

AlphaFold. Int. J. Mol. Sci. 2023,24,

13257. https://doi.org/10.3390/

ijms241713257

Academic Editor: Ho-Jin Lee

Received: 31 July 2023

Revised: 18 August 2023

Accepted: 24 August 2023

Published: 26 August 2023

Licensee MDPI, Basel, Switzerland.

This article is an open access article

distributed under the terms and

conditions of the Creative Commons

Attribution (CC BY) license (https://

creativecommons.org/licenses/by/

4.0/).

International Journal of

Molecular Sciences

Article

Design of Cyclic Peptides Targeting Protein–Protein

Interactions Using AlphaFold

Takatsugu Kosugi and Masahito Ohue *

Department of Computer Science, School of Computing, Tokyo Institute of Technology, G3-56-4259 Nagatsutacho,

Midori-ku, Yokohama City 226-8501, Kanagawa, Japan; kosugi@li.c.titech.ac.jp

*Correspondence: ohue@c.titech.ac.jp; Tel.: +81-45-924-5522

Abstract:

More than 930,000 protein–protein interactions (PPIs) have been identiﬁed in recent years,

but their physicochemical properties differ from conventional drug targets, complicating the use of

conventional small molecules as modalities. Cyclic peptides are a promising modality for targeting

PPIs, but it is difﬁcult to predict the structure of a target protein–cyclic peptide complex or to design

a cyclic peptide sequence that binds to the target protein using computational methods. Recently,

AlphaFold with a cyclic offset has enabled predicting the structure of cyclic peptides, thereby enabling

de novo cyclic peptide designs. We developed a cyclic peptide complex offset to enable the structural

prediction of target proteins and cyclic peptide complexes and found AlphaFold2 with a cyclic

peptide complex offset can predict structures with high accuracy. We also applied the cyclic peptide

complex offset to the binder hallucination protocol of AfDesign, a de novo protein design method

using AlphaFold, and we could design a high predicted local-distance difference test and lower

separated binding energy per unit interface area than the native MDM2/p53 structure. Furthermore,

the method was applied to 12 other protein–peptide complexes and one protein–protein complex.

Our approach shows that it is possible to design putative cyclic peptide sequences targeting PPI.

Keywords: protein–protein interaction (PPI); cyclic peptide; peptide design; AlphaFold; AfDesign

1. Introduction

Protein–protein interactions (PPIs) play a crucial role in numerous biological and bio-

chemical processes, including signal transduction and metabolism in cellular activities [

Over 930,000 human PPIs have been identiﬁed to date [

] (BioGrid 4.4.222, June 2023).

Since the early 2000s, PPIs have garnered signiﬁcant attention as potential drug targets

for human diseases [

–

]. However, the physicochemical properties of PPI drug targets

are distinct from conventional drug targets, presenting challenges in PPI-targeting drug

discovery [

]. Cyclic peptides have emerged as more effective PPI drug target modalities

compared to small-molecule compounds, boasting advantages in stability, structure, and

membrane permeability [10–12].

•

Due to the absence of termini, cyclic peptides are more resistant to digestive enzymes

like peptidases and exoproteases.

•

The constraint of the cyclic structure facilitates more stable folding without relying on

secondary structures.

•

Cyclic structures can also be engineered for permeability across cell membranes, tar-

geting intracellular PPIs that are inaccessible to large molecules or

antibodies [13–16].

There exist several methodologies for the experimental design of PPI-targeting pep-

tides, such as mRNA display and cDNA display, which are used to screen a vast number of

peptides for protein interactions; however, these methods are prohibitively

expensive [17–19].

Thus, a method for computationally designing candidate peptide sequences that engage

with target proteins is much sought after. While there have been instances of designing

Int. J. Mol. Sci. 2023,24, 13257. https://doi.org/10.3390/ijms241713257 https://www.mdpi.com/journal/ijms

Int. J. Mol. Sci. 2023,24, 13257 2 of 17

sequences with a speciﬁc function, like antimicrobial peptide sequences [

], they re-

quire a signiﬁcant amount of experimental data. Acquiring such extensive data on peptide

sequences interacting with speciﬁc PPI-targeting proteins is challenging. Computational

methods exist to generate peptides similar to known PPI inhibitors [

], but there are few

small molecules that can inhibit PPIs [7,23].

High-throughput virtual screening (HTVS) technology is a potent and efﬁcient method

for pinpointing drug candidates from a vast compound library [

]. Thus, conducting

HTVS of peptides with the right virtual screening steps and strategies could feasibly yield

bioactive peptide hit candidates [

]. The primary technology underpinning HTVS is

molecular docking tools, which encompass both commercial and open-source tools. There

are commercial options available, such as the Molecular Operating Environment suite,

the Schrödinger modeling suite, and the OpenEye Scientiﬁc Software suite. Tools like

Schrödinger Glide can perform protein–peptide docking, though they’re essentially crafted

for linear peptides [

]. When these tools are employed for cyclic peptides, the peptide

is treated as a ligand, and protein–ligand docking is undertaken. Glide has introduced a

macrocycle module that accommodates both ﬂexible and rigid docking [

]. Nevertheless,

these tools are optimized primarily for conventional small molecule compounds. A thor-

ough understanding of the parameters for conformation generation and scoring functions

is indispensable when docking larger cyclic peptides as ligands.

Conversely, while numerous open-source tools exist for global protein–peptide dock-

ing, such as CABS-dock [

], ClusPro PeptiDock [

], and PIPER-FlexPepDock [

], and

for local protein–peptide docking, including PEP-FOLD3 [

] and DINC 2.0 [

], they can-

not handle cyclic peptides. Only two tools, HADDOCK 2.4 [

] and AutoDock CrankPep

(ADCP) [

], are speciﬁcally tailored for protein–cyclic peptide docking, with ADCP

being recognized as the state-of-the-art. Despite signiﬁcant advancements in peptide dock-

ing tools, which are crucial for peptide HTVS strategies, the pace of progress in peptide

HTVS remains considerably slower than that for small molecule compounds. Present HTVS

methodologies, even those utilizing cutting-edge techniques, appear insufﬁcient in fully

tapping the potential of peptide libraries. Notably, few reports in the literature highlight

effective use of HTVS, with only a handful of studies showcasing success through both

proprietary and open computational techniques [39].

Given this context, we introduce a novel method in this study to design cyclic pep-

tides targeting PPI. This method harnesses a deep learning (DL) based protein structure

prediction, diverging from the traditional HTVS strategies reliant on molecular docking.

DL techniques, notably AlphaFold2 (AF2) and RoseTTAFold, have made signiﬁcant strides

in the accurate prediction of 3D protein structures from amino acid sequences [

–

These tools have demonstrated capabilities in predicting individual proteins as well as

more complex assemblies like protein complexes and protein–peptide complexes [

–

AF2 can distinguish between favorable and unfavorable conformational templates. This

suggests that AF2 learns the approximate biophysical energy function of energetically

stable protein backbones, and that DL-based structure prediction methods can link the

sequence space and protein structure space to explore protein structures that satisfy a given

design [

]. Deep network hallucinations, a DL-based design method, can address a variety

of protein designs by starting with random and iterative updating of the protein sequence

until the desired properties speciﬁed in the loss function are obtained [

]. To increase

the possibility of generating more accurate predictions, the loss function converges more

quickly using sequence gradients obtained by inverting the structure prediction network

and using continuous logits, rather than discrete one-hot encodings of the sequence repre-

sentation [

–

]. In our previous study, we designed linear peptide binders in AfDesign

using a three-step sequence representation of logits, softmax, and one-hot steps [

]. AfDe-

sign binder hallucination is based on AF2, so in principle it is not possible to design cyclic

peptides. However, if AF2 can predict the structure of protein–cyclic peptide complexes, it

can design sequences that are likely to form complexes with the target protein, which is

quite efﬁcient methods.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

1 知币 4人已下载

立即下载

Design of Cyclic Peptides Targeting Protein–Protein Interactions Using AlphaFold.pdf

共17页,预览6页

还剩页未读，继续阅读

Design of Cyclic Peptides Targeting Protein–Protein Interactions Using AlphaFold

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

推荐作者

热门标签

举报选择: