Papo, Niv

Abstract Deep mutational scanning (DMS) has proven effective for mapping protein–protein interactions (PPIs), but it cannot provide complete coverage of the mutation landscape, particularly for multi‐mutant variants. To address this limitation, we trained machine‐learning (ML) models on previously generated DMS data for a stabilized amyloid precursor protein inhibitor (APPI) binding to either of two serine proteases, mesotrypsin and kallikrein‐6 (KLK6), which are implicated in various human disorders. We combined the models to accurately predict the binding selectivity of APPI variants, including double‐mutant variants, for the two serine proteases. We achieved a Pearson correlation of 0.937 between predicted log 2 selectivity enrichment ratios and DMS‐derived values. We further validated the predictions of our combined model by yeast‐surface‐display measurements and inhibition assays of purified APPI variants and revealed epistatic interactions that shape protease selectivity. Guided by binding selectivity predictions, we identified highly selective APPI variants, including the most selective mesotrypsin inhibitor reported to date. Together, these findings support the use of DMS and ML as a framework for predicting PPI selectivity and prioritizing selective therapeutic protein variants.

Selective inhibitors are essential for targeted therapeutics and for probing enzyme functions in various biological systems. The two main challenges in identifying such protein‐based inhibitors lie in the extensive experimental effort required, including the generation of large libraries, and in tailoring the selectivity of inhibitors to enzymes with homologous structures. To address these challenges, machine learning (ML) is being used to improve protein design by training on targeted libraries and identifying key interface mutations that enhance affinity and specificity. However, such ML‐based methods are limited by inaccurate energy calculations and difficulties in predicting the structural impacts of multiple mutations. Here, we present an ML‐based method that leverages HTS data to streamline the design of selective protease inhibitors. To demonstrate its utility, we applied our new method to find inhibitors of matrix metalloproteinases (MMPs), a family of homologous proteases involved in both physiological and pathological processes. By training ML models on binding data for three MMPs (MMP‐1, MMP‐3, and MMP‐9), we successfully designed a novel N‐TIMP2 variant with a differential specificity profile, namely, high affinity for MMP‐9, moderate affinity for MMP‐3, and low affinity for MMP‐1. Our experimental validation showed that this novel variant exhibited a significant specificity shift and enhanced selectivity compared with wild‐type N‐TIMP2. Through molecular modeling and energy minimization, we obtained structural insights into the variant's enhanced selectivity. Our findings highlight the power of ML‐based methods to reduce experimental workloads, facilitate the rational design of selective inhibitors, and advance the understanding of specific inhibitor–enzyme interactions in homologous enzyme systems.

Abstract Matrix metalloproteinases (MMPs) regulate the degradation of extracellular matrix (ECM) components in biological processes. MMP activity is controlled by natural tissue inhibitors of metalloproteinases (TIMPs) that non-selectively inhibit the function of multiple MMPs via interaction with the MMPs' Zn2+-containing catalytic pocket. Recent studies suggest that TIMPs engineered to confer MMP specificity could be exploited for therapeutic purposes, but obtaining specific TIMP-2 inhibitors has proved to be challenging. Here, in an effort to improve MMP specificity, we incorporated the metal-binding non-canonical amino acids (NCAAs), 3,4-dihydroxyphenylalanine (L-DOPA) and (8-hydroxyquinolin-3-yl)alanine (HqAla), into the MMP-inhibitory N-terminal domain of TIMP2 (N-TIMP2) at selected positions that interact with the catalytic Zn2+ ion (S2, S69, A70, L100) or with a structural Ca2+ ion (Y36). Evaluation of the inhibitory potency of the NCAA-containing variants towards MMP-2, MMP-9 and MMP-14 in vitro revealed that most showed a significant loss of inhibitory activity towards MMP-14, but not towards MMP-2 and MMP-9, resulting in increased specificity towards the latter proteases. Substitutions at S69 conferred the best improvement in selectivity for both L-DOPA and HqAla variants. Molecular modeling revealed how MMP-2 and MMP-9 are better able to accommodate the bulky NCAA substituents at the intermolecular interface with N-TIMP2. The models also showed that, rather than coordinating to Zn2+, the NCAA side chains formed stabilizing polar interactions at the intermolecular interface with MMP-2 and MMP-9. The findings illustrate how incorporation of NCAAs can be used to probe and exploit differential tolerance for substitution within closely related protein-protein complexes to achieve improved specificity.

Serine protease inhibitors of the Kunitz‐bovine pancreatic trypsin inhibitor (BPTI) family, such as tissue factor pathway inhibitor‐1 Kunitz domain 1 ( KD1 TFPI1) and amyloid precursor protein inhibitor (APPI), are biological regulators of proteolysis and serve as important scaffolds for developing therapeutics. Mesotrypsin is a unique human protease with cancer‐promoting properties which rapidly cleaves and inactivates serine protease inhibitors; because of this, it is a challenging therapeutic target and an excellent model for developing therapeutic inhibitors with improved proteolytic resistance. We hypothesize that stabilization of the Kunitz inhibitor binding loop by disulfide bond anchoring to the scaffold can reduce proteolytic susceptibility and improve binding efficacy. Here we have engineered a new disulfide bond into both KD1 TFPI1 and an APPI variant previously engineered for mesotrypsin affinity. The new disulfide bond anchors the binding loop to the scaffold, resulting in reduced mesotrypsin proteolytic cleavage by 17‐fold and 3‐fold, respectively. To gain insight into the structural basis for improved proteolytic stability, we crystallized mutant KD1 TFPI1 and APPI with mesotrypsin and solved these novel structures. We then performed all‐atom molecular dynamics simulations using these mesotrypsin‐mutant inhibitor complexes compared to models lacking the engineered disulfide bond. Our simulation results support our hypothesis that the cause of decreased proteolysis rates is dampened conformational fluctuations in the binding loop and minimized motion at the enzyme‐inhibitor interface. We also performed simulations using unbound inhibitors to show that stabilization of the binding loop to states similar to the bound configuration correlates with stronger binding affinity. Our disulfide engineering approach is potentially a useful tool in biopharmaceutical development for mesotrypsin, other serine proteases, and possibly for other enzyme targets. Support or Funding Information NIH grant R01CA154387 to E.S.R., European Research Council grant 336041 to N.P., US‐Israel Binational Science Foundation grant BSF2015134 to E.S.R. and N.P. Xray diffraction data were measured at the Berkeley Center for Structural Biology at the Advanced Light Source and at the European Synchrotron Radiation Facility. This abstract is from the Experimental Biology 2019 Meeting. There is no full text article associated with this abstract published in The FASEB Journal .

To facilitate investigations of protein–protein interactions (PPIs), we developed a novel platform for quantitative mapping of protein binding specificity landscapes, which combines the multi-target screening of a mutagenesis library into high- and low-affinity populations with sophisticated next-generation sequencing analysis. Importantly, this method generates accurate models to predict affinity and specificity values for any mutation within a protein complex, and requires only a few experimental binding affinity measurements using purified proteins for calibration. We demonstrated the utility of the approach by mapping quantitative landscapes for interactions between the N-terminal domain of the tissue inhibitor of metalloproteinase 2 (N-TIMP2) and three matrix metalloproteinases (MMPs) having homologous structures but different affinities (MMP-1, MMP-3, and MMP-14). The binding landscapes for N-TIMP2/MMP-1 and N-TIMP2/MMP-3 showed the PPIs to be almost fully optimized, with most single mutations giving a loss of affinity. In contrast, the non-optimized PPI for N-TIMP2/MMP-14 was reflected in a wide range of binding affinities, where single mutations exhibited a far more attenuated effect on the PPI. Our new platform reliably and comprehensively identified not only hot- and cold-spot residues, but also specificity-switch mutations that shape target affinity and specificity. Thus, our approach provides a methodology giving an unprecedentedly rich quantitative analysis of the binding specificity landscape, which will broaden the understanding of the mechanisms and evolutionary origins of specific PPIs and facilitate the rational design of specific inhibitors for structurally similar target proteins.

High structural and sequence similarity within protein families can pose significant challenges to the development of selective inhibitors, especially toward proteolytic enzymes. Such enzymes usually belong to large families of closely similar proteases and may also hydrolyze, with different rates, protein- or peptide-based inhibitors. To address this challenge, we employed a combinatorial yeast surface display library approach complemented with a novel pre-equilibrium, competitive screening strategy for facile assessment of the effects of multiple mutations on inhibitor association rates and binding specificity. As a proof of principle for this combined approach, we utilized this strategy to alter inhibitor/protease association rates and to tailor the selectivity of the amyloid β-protein precursor Kunitz protease inhibitor domain (APPI) for inhibition of the oncogenic protease mesotrypsin, in the presence of three competing serine proteases, anionic trypsin, cationic trypsin and kallikrein-6. We generated a variant, designated APPIP13W/M17G/I18F/F34V, with up to 30-fold greater specificity relative to the parental APPIM17G/I18F/F34V protein, and 6500- to 230 000-fold improved specificity relative to the wild-type APPI protein in the presence of the other proteases tested. A series of molecular docking simulations suggested a mechanism of interaction that supported the biochemical results. These simulations predicted that the selectivity and specificity are affected by the interaction of the mutated APPI residues with nonconserved enzyme residues located in or near the binding site. Our strategy will facilitate a better understanding of the binding landscape of multispecific proteins and will pave the way for design of new drugs and diagnostic tools targeting proteases and other proteins.

MMP ‐14 and MMP ‐9 are two well‐established cancer targets for which no specific clinically relevant inhibitor is available. Using a powerful combination of computational design and yeast surface display technology, we engineered such an inhibitor starting from a nonspecific MMP inhibitor, N‐ TIMP 2. The engineered purified N‐ TIMP 2 variants showed enhanced specificity toward MMP ‐14 and MMP ‐9 relative to a panel of off‐target MMP s. MMP ‐specific N‐ TIMP 2 sequence signatures were obtained that could be understood from the structural perspective of MMP /N‐ TIMP 2 interactions. Our MMP ‐9 inhibitor exhibited 1000‐fold preference for MMP ‐9 vs. MMP ‐14, which is likely to translate into significant differences under physiological conditions. Our results provide new insights regarding evolution of promiscuous proteins and optimization strategies for design of inhibitors with single‐target specificities.

Engineered protein therapeutics offer advantages, including strong target affinity, selectivity and low toxicity, but like natural proteins can be susceptible to proteolytic degradation, thereby limiting their effectiveness. A compelling therapeutic target is mesotrypsin, a protease up-regulated with tumour progression, associated with poor prognosis, and implicated in tumour growth and progression of many cancers. However, with its unique capability for cleavage and inactivation of proteinaceous inhibitors, mesotrypsin presents a formidable challenge to the development of biological inhibitors. We used a powerful yeast display platform for directed evolution, employing a novel multi-modal library screening strategy, to engineer the human amyloid precursor protein Kunitz protease inhibitor domain (APPI) simultaneously for increased proteolytic stability, stronger binding affinity and improved selectivity for mesotrypsin inhibition. We identified a triple mutant APPIM17G/I18F/F34V, with a mesotrypsin inhibition constant (Ki) of 89 pM, as the strongest mesotrypsin inhibitor yet reported; this variant displays 1459-fold improved affinity, up to 350 000-fold greater specificity and 83-fold improved proteolytic stability compared with wild-type APPI. We demonstrated that APPIM17G/I18F/F34V acts as a functional inhibitor in cell-based models of mesotrypsin-dependent prostate cancer cellular invasiveness. Additionally, by solving the crystal structure of the APPIM17G/I18F/F34V–mesotrypsin complex, we obtained new insights into the structural and mechanistic basis for improved binding and proteolytic resistance. Our study identifies a promising mesotrypsin inhibitor as a starting point for development of anticancer protein therapeutics and establishes proof-of-principle for a novel library screening approach that will be widely applicable for simultaneously evolving proteolytic stability in tandem with desired functionality for diverse protein scaffolds.