Understanding what drives protein–protein binding and selecting appropriate protein residues for modification to strengthen protein–protein interactions (PPIs) are crucial to designing a protein binder that targets its binding partner.1, 2 Strategies that can efficiently and accurately identify residues to enhance PPIs have broad applications in therapeutics and studies of cell biology. Knowledge-based, physics-based, and data-driven methods have all been developed previously to explore PPIs and to select mutations that enhance them.3, 4, 5, 6, 7, 8, 9 Computational and combinatorial libraries or in vitro evolutionary approaches also represent popular protein engineering strategies to design stable and specific protein binders.10, 11, 12 Importantly, minimizing the number of residues mutated to significantly enhance PPIs lowers the possibility of engineering unstable proteins. Nevertheless, protein engineering remains challenging, as highly integrated molecular modeling and experimental techniques are needed to understand PPIs in order to re-engineer a protein to increase its binding affinity.

PPI networks are extremely complex, so selecting an appropriate target system for experimental modification requires specialized expertise. Here, we chose Ubiquitin (Ub) as our target system, as it plays critical roles in numerous biological functions.13 Ub is a small 76-residue protein associated with post-translational modifications. This regulatory protein canonically binds to its cascade E1-E2-E3 enzymes to drive ubiquitination and Ub chain formation, thereby modifying nearly half of the human proteome.14, 15 Conversely, deubiquitinases (DUBs)16 cleave the covalent isopeptide bonds from Ub chains or substrates to release Ub and its substrates. The interactome of Ub and cellular proteins have been assessed15 revealing that precise Ub network regulation governs cellular fates. Misregulation of the responsible enzymes significantly impacts cellular functions, leading to diseases such as cancers.17 Moreover, viral DUBs have been found to interfere with host antiviral defenses. For example, the Papain-like protease (PLpro) of coronaviruses (CoVs) is classified as a viral DUB specific to Ub and Ub-like ISG15.18, 19 Previous studies have shown that PLpro alters host innate immune responses, which contributes to the rapid spread of CoVs (such as MERS, SARS-CoV, SARS-CoV-2),20, 21, 22, 23, 24 thus causing pandemics, mortality, and perturbing the global economy.25

The PLpro proteins of MERS and SARS-CoV-2 are crucial for viral replication through their role in proteolytic cleavage of viral nonstructural proteins (NSPs). The PLpro domain resides in NSP3, which drives viral genome replication and subgenomic RNA synthesis.26, 27 PLpro recognizes and cleaves the NSP1-2, NSP2-3 and, NSP3-4 junctions after the amino acid sequence LXGG to yield functional viral proteins, as well as to perform deubiquitination and deISGylation.21, 23, 24 Deubiquitination and deISGylation alter host signaling pathways critical to induction of cellular antiviral and pro-inflammatory innate immune responses, ultimately suppressing the host antiviral response.21, 28 Therefore, inhibiting PLpro simultaneously disrupts viral replication and prevents PLpro from impairing the innate immune response. Given both these properties, PLpro represents an ideal antiviral drug target.

Importantly, wild-type Ub (wtUb) exhibits high thermostability (Tm > 90 °C), so it is an ideal template for protein design. Re-engineered Ub also has potential advantages, such as enhanced binding specificity to PLpro and easier synthesis compared to chemical compounds. Screening of phage-displayed Ub variants (UbVs) against cognate enzymes, including MERS PLpro, has previously demonstrated the feasibility of regulating the activities of E3 ligases and DUBs.29, 30, 31 The phage-display screening technique focused on three surface patches in Ub to iteratively mutate and select variants that displayed tight binding. The resulting DUB UbVs proved to be strong inhibitors, exhibiting IC50 values in the range of 1–30 nM.30, 31, 32 As an alternative approach, computational data were used to rationally design a screening library for the identification of tightly binding regulatory UbVs for USP733 and USP21.34 A combined computational and phage-display screening of UbVs targeting USP7 resulted in an equilibrium constant (KD) for the U7Ub25.2540 variant of 56 nM, whereas for wtUb-USP7 it was >200 µM. A pool of 6,000 designed UbVs for USP21 revealed that ∼10% of the variants tightly bound USP21 consistently between experimental and computational screenings. However, in silico screenings of such large UbV-USP21 pairings require intensive computational resources. Such expensive and time-consuming empirical screenings impede rational design of protein-based inhibitors.

Here, we present an integrated computational and experimental approach to identify critical regions for protein–protein binding that display highly correlated dynamic motion. Specifically, we focus on side-chain dihedral angle correlations at the protein–protein contact interface where mutation of highly correlated residues resulted in both local and distal conformational changes. We demonstrate that mutating residues in these regions can efficiently optimize PPIs to create tight and selective protein binders. We show that our designed UbVs hosting two or three mutated residues achieved 3,500-fold inhibitory efficiency and binding affinity relative to wtUb for MERS PLpro (Table 1). MERS PLpro cleaves both K48- and K63-linked Ub chains,18, 23 and it exhibits distinct inhibitor recognition specificity to that of the PLpro of SARS-CoV and SARS-CoV-2.35 We used non-covalent amino acid interaction and side-chain dihedral angle networks of the Ub and MERS PLpro (Ub-PLpro) complex to guide our design of UbVs that enhance UbV–PLpro binding affinity, thereby inhibiting PLpro activity. Initially, we designed two-point mutations for cost efficiency and to retain intact the overall complex structure. Integrating experimental data and computational analyses informed our experimental design to yield more UbVs (Fig. 1). Binding affinity KD and IC50 measurements of our designed UbVs support that more extensively mutated UbV3, UbV4, and UbV5 represent strong inhibitors.